Reading List for the Literature Exam 1. D. R. Dennis and J. R. Meredith. An analysis of process industry production and inventory manage- ment systems. Journal of Operations Management, 18(6):683–699, 2000 2. J. C. Fransoo and W. G. Rutten. A typology of production control situations in process industries. International Journal of Operations & Production Management, 14(12):47–57, 1994 3. M. Ketokivi and M. Jokinen. Strategy, uncertainty and the focused factory in international process manufacturing. Journal of Operations Management, 24(3):250–270, 2006 4. R. Akkerman, D. Van Der Meer, and D. P. Van Donk. Make to stock and mix to order: choosing intermediate products in the food-processing industry. International Journal of Production Research, 48(12):3475–3492, 2010 5. A. C. Lyons, K. Vidamour, R. Jain, and M. Sutherland. Developing an understanding of lean thinking in process industries. Production Planning & Control, 24(6):475–494, 2013 6. A. Pool, J. Wijngaard, and D.-J. Van der Zee. Lean planning in the semi-process industry, a case study. International Journal of Production Economics, 131(1):194–203, 2011 7. K. Rajaram, R. Jaikumar, F. Behlau, F. Van Esch, C. Heynen, R. Kaiser, A. Kuttner, and I. Van De Wege. Robust process control at cerestar’s refineries. Interfaces, 29(1):30–48, 1999 8. L. Strijbosch, R. Heuts, and M. Luijten. Cyclical packaging planning at a pharmaceutical company. International Journal of Operations & Production Management, 22(5):549–564, 2002 9. M. L¨ utke Entrup, H.-O. G¨ unther, P. Van Beek, M. Grunow, and T. Seiler. Mixed-integer linear programming approaches to shelf-life-integrated planning and scheduling in yoghurt production. In- ternational Journal of Production Research, 43(23):5071–5100, 2005 10. D. P. Van Donk. Make to stock or make to order: The decoupling point in the food processing industries. International Journal of Production Economics, 69(3):297–306, 2001 11. D. P. Van Donk, R. Akkerman, and T. Van der Vaart. Opportunities and realities of supply chain integration: the case of food manufacturers. British Food Journal, 110(2):218–235, 2008 12. W. Van Wezel, D. P. Van Donk, and G. Gaalman. The planning flexibility bottleneck in food processing industries. Journal of Operations Management, 24(3):287–300, 2006 13. J. G. Wacker. A definition of theory: research guidelines for different theory-building research methods in operations management. Journal of Operations Management, 16(4):361–385, 1998 1
Transcript
Reading List for the Literature Exam
1. D. R. Dennis and J. R. Meredith. An analysis of process industry production and inventory manage-
ment systems. Journal of Operations Management, 18(6):683–699, 2000
2. J. C. Fransoo and W. G. Rutten. A typology of production control situations in process industries.
International Journal of Operations & Production Management, 14(12):47–57, 1994
3. M. Ketokivi and M. Jokinen. Strategy, uncertainty and the focused factory in international process
manufacturing. Journal of Operations Management, 24(3):250–270, 2006
4. R. Akkerman, D. Van Der Meer, and D. P. Van Donk. Make to stock and mix to order: choosing
intermediate products in the food-processing industry. International Journal of Production Research,
48(12):3475–3492, 2010
5. A. C. Lyons, K. Vidamour, R. Jain, and M. Sutherland. Developing an understanding of lean thinking
in process industries. Production Planning & Control, 24(6):475–494, 2013
6. A. Pool, J. Wijngaard, and D.-J. Van der Zee. Lean planning in the semi-process industry, a case
study. International Journal of Production Economics, 131(1):194–203, 2011
7. K. Rajaram, R. Jaikumar, F. Behlau, F. Van Esch, C. Heynen, R. Kaiser, A. Kuttner, and I. Van
De Wege. Robust process control at cerestar’s refineries. Interfaces, 29(1):30–48, 1999
8. L. Strijbosch, R. Heuts, and M. Luijten. Cyclical packaging planning at a pharmaceutical company.
International Journal of Operations & Production Management, 22(5):549–564, 2002
9. M. Lutke Entrup, H.-O. Gunther, P. Van Beek, M. Grunow, and T. Seiler. Mixed-integer linear
programming approaches to shelf-life-integrated planning and scheduling in yoghurt production. In-
ternational Journal of Production Research, 43(23):5071–5100, 2005
10. D. P. Van Donk. Make to stock or make to order: The decoupling point in the food processing
industries. International Journal of Production Economics, 69(3):297–306, 2001
11. D. P. Van Donk, R. Akkerman, and T. Van der Vaart. Opportunities and realities of supply chain
integration: the case of food manufacturers. British Food Journal, 110(2):218–235, 2008
12. W. Van Wezel, D. P. Van Donk, and G. Gaalman. The planning flexibility bottleneck in food processing
industries. Journal of Operations Management, 24(3):287–300, 2006
13. J. G. Wacker. A definition of theory: research guidelines for different theory-building research methods
in operations management. Journal of Operations Management, 16(4):361–385, 1998
1
Ž .Journal of Operations Management 18 2000 683–699www.elsevier.comrlocaterdsw
An analysis of process industry production and inventorymanagement systems
Daina R. Dennis a, Jack R. Meredith b,)
a Department of Management, Miami UniÕersity, Oxford, OH, USAb Babcock Graduate School of Management, Wake Forest UniÕersity, P.O. Box 7659, Winston-Salem, NC 27109, USA
Abstract
The process industries — those firms that add value by mixing, separating, forming andror chemical reactions by eitherbatch or continuous mode — continue to lag behind the discrete industries in the identification and implementation of
Ž .effective production and inventory management P&IM techniques. A contributing factor is that the process industries havetraditionally been lumped together and contrasted from the discrete industries as a whole, thus leading to misunderstandingsregarding individual process industries. From site interviews and the literature, we identified four critical dimensions —
Ž .planning resource requirements for materials and capacity , tracking resource consumption, control of work-in-processŽ .WIP , and degree of computerization — represented by seven variables by which to contrast and analyze process industries.Based on in-depth field studies of 19 diverse process plants, we find that there exist at least four distinct types of process
Keywords: Production and inventory management; Process industries
1. Introduction and background
Identifying the right kind of production and inven-Ž .tory management P&IM system for a manufactur-
ing firm can be a difficult and complex task. Sincethe investment in a P&IM system is large andremains fixed over a considerable length of time, thecorrect system choice is critical to both a firm’s shortand long-term profitability. Research on the selectionand implementation of P&IM systems for differentmanufacturing environments has been extensive. The
majority of the research, however, has been forindustries handling discrete units that are fabricatedandror assembled during manufacturing. This re-search has resulted in numerous successful develop-ments in taxonomies, P&IM systems, and imple-mentation strategies for the discrete industries.
The process industries, consisting of firms thatAadd value by mixing, separating, forming androrchemical reactions by either batch or continuous
w xmodeB Wallace, 1992 , continue to have difficultyrealizing the benefits of many of the managementsystem developments in the discrete industries. ATheprocess sector, although more automated than dis-crete industries at the process control level, lagsdiscrete when it comes to overall manufacturing
0272-6963r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S0272-6963 00 00039-5
Ž .management systems toolsB Crow, 1992 . The rea-sons for this are many and varied. A brief review ofpast and current research sheds some light on theessence of this problem.
Unfortunately, as shown in the left column ofTable 1, the great majority of process industry re-search simply reports general characteristics of theindustry without reference to their P&IM systems.There are three subgroups of research in this column.The research listed in item 1 focuses on the uniquecharacteristics of the process industries and com-pares them as a whole to discrete industries. Thisgroup of research nevertheless provides some back-ground understanding of the problems unique to theprocess industries that are caused by the handling ofnon-discrete materials involving variability, sequenc-ing, co-products, shelf-life, and so on. A smallersegment of research, item 2, addresses the generalproblems that process industries encounter regardingP&IM. A very small segment, item 3, describes thevarying degrees of success that specific processfirms have had with material requirements planningŽ .MRP implementations.
A smaller overall group of research pertains toP&IM systems in process industries. The referencesin item 4 address specific P&IM problems that
process firms have encountered and their solutions.Item 5 captures much of the current schedulingresearch on P&IM systems focused primarily onhigh volume process flow manufacturers and illus-trates the application of a technique called process
Ž .flow scheduling PFS . PFS has been described boththeoretically and analytically, and shown in certaincircumstances to be successful in practice. However,the full extent of applicability of the PFS logic hasnot yet been established.
Item 6 includes three references, each of whichprovides a different perspective for creating a basictaxonomy of process industry P&IM systems. Thesethree perspectives may be considered to constitutethe entirety of current theory about different types ofprocess industry P&IM systems. Robinson and Tay-
Ž .lor 1986 classify P&IM systems by the primaryresource that must be managed: material dominated,capacity dominated, or a combination of both.Knowing which resource is primary provides valu-able P&IM information, particularly about schedul-ing priorities and resource constraints. In contrast,
Ž .Vollmann et al. 1997 in their well-known manufac-turing planning and control book categorize P&IMsystems by distinguishing between time-phased andrate-based systems. The third categorization by Finch
Table 1Process industry literature
General process industry characteristics Process industry P&IM systems
1 — process industry uniqueness 4 — specific P&IM problems and solutionsAdelberg, 1984; Aiello, 1982; Allen, 1980; Allen and Schuster, 1994; Allen and Schuster, 1994; Baumeister, 1997; Daniels, 1983;Baumeister, 1997; Clark, 1983; Cohen, 1980; Cokins, 1988; Dayvolt and Symonds, 1994; Eads, 1989; Gerchak et. al., 1996;Covey, 1984; Dayvolt and Symonds, 1994; Doganaksoy and Hahn, Katayama, 1996; Parker, 1997; Thompson, 19911996; Duncan, 1981; Eads and Undheim, 1984; Fransoo, 1992;Fransoo and Rutten 1994; Haxthausen, 1995; Katayama, 1996; 5 — high-Õolume P&IM schedulingNelson, 1983; Nichols and Ricketts, 1994; Parker, 1997; Bolander and Taylor, 1990; Bolander and Taylor, 1993;Rice and Norback, 1987; Rutten, 1993; Shelley, 1995; Swann, Hubbard et al., 1992; Taylor and Bolander, 1990; Taylor and1984; Tiber, 1981 Bolander, 1991; Taylor and Bolander, 1993; Taylor and Bolander,
1995; Taylor and Bolander, 1997
2 — general P&IM problems 6 — basic P&IM taxonomiesBurt, 1980; Crow, 1992; Dennis, 1993; Finch, 1986; Finch and Cox, Robinson and Taylor, 1986; Volmann et al., 1997;1987; Fransoo, 1992; Fransoo and Rutten 1994; Taylor, 1979 Finch and Cox, 1988
3 — specific MRP successesCokins, 1988; Dibono, 1997; McKaskill, 1992
Ž .and Cox 1988 looks at whether the P&IM systemsare driven by make-to-stock, assemble-to-order, ormake-to-order requirements.
A primary goal of this research study was to helpbuild operations management theory in this area byidentifying the range of P&IM systems being usedat individual plants in the process industries, the unitof analysis in this study. Since theory testing shouldnot be conducted using the same data that built the
Žtheory Green, 1978, p. 432; Meredith et al., 1989,.pp. 301–303 , no attempt was made here with this
limited data set to also test the theory. Nor was itused to try to identify the AbestB P&IM systemsunder various circumstances, such tests and applica-tions thus being deferred for future studies.
A detailed analysis of the P&IM systems of 19diverse processing plants was carried out. The focuswas on both the similarities and differences betweenthe P&IM systems. Overall, this paper addresses thefollowing three questions:
1. What are the relevant variables by which processindustry P&IM systems can be differentiated?
2. How do these variables differ between differentP&IM systems?
3. Can the different P&IM systems be put intomanageable subgroups based on their P&IM vari-ables and, if so, what are the subgroups and theircharacteristics?
The following sections provide the backgroundfor answering these three questions. In the nextsection, the choice of variables, the research sample,and the analysis process will be addressed.
2. Identifying the key P&IM system variables
In order to effectively compare the P&IM sys-tems, the characteristics that best differentiate be-tween different P&IM systems had to be identified.The challenge was to isolate those characteristicsthat would provide a parsimonious yet thoroughdescription of the sites’ P&IM systems. Researchsuggests that this description must distinguish be-tween materials and capacity. Many researchersŽPlossl and Wight, 1967; Abraham et al., 1985;
.Dietrich, 1987; Vollman et al., 1997 recognize that
the basic elements of any P&IM system includeattention to both materials and capacity. Robinson
Ž .and Taylor 1986 expand on this by suggesting thatP&IM systems may be classified by the primaryresource that must be managed: materials, capacity,or a combination of both. Knowing which resource isprimary provides valuable information aboutscheduling priorities and resource constraints. Assuch, the variables used in this research alwaysconsidered materials and capacity separately.
Any description of a P&IM system must alsoaddress both the planning and control of resources.
ŽSeveral researchers Abraham et al., 1985; Dietrich,.1987; Vollman et al., 1997 illustrate the relationship
between planning and control within P&IM systemsusing hierarchical frameworks. As one moves downthese hierarchies, the required input and output infor-mation shifts from plans for resource requirementsto the actual tracking of resources consumed. Theinformation also become more detailed and compre-hensive as the time horizon changes from long rangeto short range.
These two sets of research thus indicated thenecessity for variables that measure the planning of
Ž .material requirements MAT-REQUIREMENTS ,Žthe planning of capacity requirements CAP-RE-
.QUIREMENTS , the tracking of actual materialsŽ .consumed MAT-CONSUMPTION , and the track-
Žing of actual capacity consumed CAP-CONSUMP-.TION .
Ž .Plossl and Wight 1967 describe work-in-processŽ .WIP as one of the most significant inventoriesbecause of its direct effect on manufacturing leadtimes. They also indicate that tracking WIP can be
Ž .difficult and complicated. Vollman et al. 1997 statethat WIP can be determined by using detailed WIPsystems based on shop order transactions at oneextreme. At the other extreme, it may be simplycalculated by exploding the bills of materials forwhatever has been delivered into finished goods lesswhatever is left in raw materials. Because of theimportance of WIP and the varied methods that maybe applied to monitor it, the variable WIP-CON-TROL was included to describe the manner in whichWIP was tracked.
The final source was preliminary in-depth inter-views with two separate firms. This informationindicated that in addition to the variables identified
from the above sources, the fraction of materials andcapacity planned and controlled by computerizationis also an important characteristic because highlydetailed and complex P&IM systems require com-puterization. As one plant manager stated: AWe hadto computerize this process to get good information.It was too complicated to keep track of manually.BAnother manager said: AWe need input from somany different areas that we had to develop aninformation system to manage it.B Both managersindicated that degree of computerization was animportant factor in distinguishing P&IM systems.Thus, the variables MAT-COMPUTERIZATION andCAP-COMPUTERIZATION were added.
However, this characteristic was intentionally lim-ited to P&IM elements; it was not meant to includegeneral firm sophistication in computers, decision
support, information systems, and other non-P&IMaspects. Moreover, given the quickly evolving natureof the computer area, this characteristic was inten-tionally kept as straightforward as possible. A great
Ž .number of papers e.g., Boyer et al., 1996, 1997have investigated the impact of computerized tech-nology on performance but again, that was not thepurpose here. The aim here was to identify the mostsimple and straightforward characteristics that woulddistinguish between P&IM systems, and fraction ofcomputerized control, again of both materials andcapacity, was such a characteristic.
Combining all the above information resulted inthe seven composite variables described in Table 2that were believed to differentiate process industries’P&IM systems from each other. These seven vari-
Ž .ables are: 1 planning of material requirements
Table 2P&IM system variables
Ž .MAT-REQUIREMENTS the level of detail and frequency used to generate material requirements1. knowledge of basic historical usage and visual observations of inventory levels; steady flow usage2. use of fixed interval reorder system; weekly physical to update inventory records3. use of fixed interval reorder system; daily physical updates of inventory records4. use of explosions to reflect gross requirements in combination with multi-criteria ABC analysis5. use of cumulative MRP with time phasing and net requirements in combination with multi-criteria ABC analysis
Ž .CAP-REQUIREMENTS the level of detail and frequency used to generate capacity requirements1. capacity is fixed and known; steady flow usage2. infinite load on an informal basis; orders are pushed into production3. capacity planning using overall factors
Ž .4. capacity planning using overall factors in combination with general capacity bills no routingsŽ .5. capacity planning using overall factors and loading product mix is not considered
Ž .MAT-CONSUMPTION the level of detail and frequency used to track material consumption1. monthly physical inventories2. weekly physical inventories3. standard consumption amounts are recorded for batches; weekly adjustments are made from physical inventories4. actual consumption is recorded at the completion of a batch5. standard consumption is recorded at the time of consumption; actual consumption is recorded at time of completion
Ž .CAP-CONSUMPTION the level of detail and frequency used to track capacity consumption1. not formally tracked2. track standard labor hours by backflushing3. track standard labor and volume of output over time4. track actual labor hours andror machine hours for each product5. track actual and standard machine andror labor hours for each product; make comparisons to update standards
Ž .WIP-CONTROL control of work-in-process1. no formal system; walk floor or make phone calls2. physical inspections are supported by daily meetings and use of shift notes
Ž .3. can back into WIP information from other inventory information backflushing ; use of more detailed shift notes4. batched starts are tracked at regular intervals.5. starts and completions are tracked on a perpetual basis
MAT-COMPUTERIZATION — the percent of material control P&IM tasks that are computerizedCAP-COMPUTERIZATION — the percent of capacity control P&IM tasks that are computerized
. Ž . Ž . Ž .TION , 5 tracking WIP WIP-CONTROL , 6computerization used to plan and control materialsŽ . Ž .MAT-COMPUTERIZATION , and 7 computeri-
Žzaused to plan and control capacity CAP-COM-.PUTERIZATION .
3. The research sample
Three important issues with respect to the selec-tion of the research sample are: deciding what consti-tutes a true process firm, determining how to selectthe firms, and deriving a procedure to determinewhat constitutes a separate type of P&IM system.
3.1. What constitutes a process firm
Approximately 50% of all firms consider them-Žselves to be in the process industries Novitsky,
.1983 . Many of these firms are actually hybrids dueto the fact that their nondiscrete units become dis-crete at some point during the manufacturing pro-cess. How far into a manufacturing process a productbecomes discrete can vary widely.
To eliminate most of the confusion that thesehybrids can cause, this research was limited to firmswith products that only become discrete at either thepoint of containerization or during the last process
Žimmediately prior to containerization e.g., baked.goods, tablets, capsules, and blended meat products .
The materials of the firms chosen for this researchare in the forms of gases, liquids, slurries, pulps,crystals, powders, pellets, films, andror semi-solidsand can only be tracked by weight or volume. Be-cause of the similarities in the physical nature of thematerials used, the firms share similar problemsin the storing, tracking, and transporting of theirmaterials.
At this early stage of theory building for identify-ing different kinds of process industry P&IM sys-tems, it was decided that a limited but heterogeneoussample of firms would be necessary. The sampleneeded to be limited because the data was to becollected through a labor-intensive field study but
heterogeneity was needed to be able to distinguishbetween different types of P&IM systems. In partic-ular, heterogeneity was desired across firm size,process industries, and product types. In addition,firm size heterogeneity would likely include build-to-stock as well as build-to-order and assemble-to-order heterogeneity.
3.2. Selecting the firms
This issue was addressed in several steps. First, alist of firms that could potentially be considered forthe research was compiled. Second, each firm wascontacted to determine if it met the criteria forconstituting a process firm described above. Andthird, the firm had to give its permission to beincluded in this research.
Several sources were solicited to assist in compil-ing a list of potential firms to consider for thisresearch. The primary source was the American Pro-
Ž .duction and Inventory Control Society APICS .Ž .APICS has a special interest group SIG for process
industries whose membership list is available uponspecial request. This source accounted for over halfof the contact list. Another important source was thelocal Business-to-Business directory. Firms listed un-der products typically considered part of the processindustries were added to the contact list. Finally,some participants provided referrals and contacts forother firms.
The initial contact with each of 62 potential firmswas made by telephone. The primary purpose of
Ž .these initial telephone contacts was to: 1 determineŽ .a willingness to participate, 2 find out what process
Ž .industry the firm belonged to, 3 determine if theŽ .firm met the process criteria outlined above, 4
obtain some preliminary ideas on the kind of manu-Ž .facturing systems used by the firm, 5 establish the
Ž .existence of a functioning P&IM system, and 6schedule an onsite meeting. It took at least two andas many as 12 telephone calls to obtain this informa-tion. About a third of the firms required some writtenexplanation of the research before they would givethe request further consideration.
These efforts resulted in onsite meetings with 14different process firms. Of the 48 non-participatingfirms, about half were disqualified quickly becausethey were distribution centers or the production facil-
Žities were too far away more than a 2-hour driving.radius . The rest of the non-participants were elimi-
nated because they either did not want to shareinformation they considered to be confidential orthey simply did not want to invest the time requiredto participate.
To help ensure that appropriate and well-managedP&IM systems were being used, the P&IM systems
Ž .were evaluated using Landel’s 1982 audit. The useof this audit provided information on whether theconditions related to the P&IM system were excel-lent, good, average, very poor, or unacceptable. Anyfirm that rated below average was disqualified fromparticipain the research study. After the results ofthis audit were considered, 13 firms qualified to beincluded in the research study.
To test for possible non-respondent bias, a two-Žwave test was conducted Armstrong and Overton
Ž ..1977 since we had no data at all on the firms thatdeclined to participate in the study. Four sites werevery late holdouts in joining the study and thus usedas our second wave, considered to be typical of thenon-respondents. The overall MANOVA test was not
Ž .significant ps0.81 so the two waves were treatedas one group in the study.
3.3. What constitutes a separate type of P&IMsystem?
A particular firm may have one or more differenttypes of P&IM systems in use in their plant. In somecompanies, the different P&IM systems are locatedin separate plants, the unit of analysis, making thedistinction clear. In other companies, the differentP&IM systems shared the same plant site. In thesesituations, considering only entirely different productlines, different types of processes, separate profitcenters, and independent P&IM systems resulted ina clear separation. In each of the firms, the separa-tion between the different P&IM systems was thusvery apparent.
The 13 firms provided data on 19 different P&IMsystems. In the firms where only one site existed, thesite and the firm were considered to be equivalentresearch units. Six of the firms, however, providedtwo separate research sites resulting in two units ofanalysis each. To test for possible correlation be-tween these six pairs of sites, Spearman non-para-
metric rank correlation coefficients were computedfor each of the seven variables. The highest correla-tion obtained was 0.55 and none were significant.Moreover, in the cluster analysis described later,three of the pairs clustered in different clusters,almost what would be expected by chance. Thus, all12 sites were treated as having independent P&IMsystems.
The 19 research sites represent a wide variety ofŽ .different process industries see Table 3 . The 19
sites vary in size, number of employees, availableproducts, transformation system, and so on. All ofthese sites, however, share one very important char-acteristic described earlier — each of the sites usenon-discrete materials throughout the majority of themanufacturing process. In all cases, the products donot become discrete until either containerization orthe step immediately prior to it. For a complete
Ž .description of the sites, refer to Dennis 1993 .
3.4. Data collection
Each plant visit was extensively planned andcombined interviews with physical tours, therebyresulting in the desired extensive documentation.Anywhere from 6 to 16 h were spent on-site askingquestions. Between one and eight people were inter-viewed at each site. In addition, approximately 1 to3 h were spent on the telephone for follow-up orclarification after each on-site visit. To minimize theuse of everyone’s time, preliminary information aboutthe company was obtained prior to the on-site visit.Information was acquired from the firm’s personneldepartment, stockholders’ reports, the library, andother such sources. Also, pre-set schedules wereused to structure the day so that the sequence ofinformation was obtained in the desired order.
Interviewees were chosen carefully and the firmsthemselves were heavily involved in the selectionprocess. Titles were avoided, basing instead the in-terviewee selection on the types of information thatwould be gathered to determine who would be mostknowledgeable in each area. Final interviewees in-cluded owners, manufacturing VPs, plant managers,production supervisors, information systems person-nel, material managers, production and inventorycontrol managers, schedulers, warehouse foremen,and plant foremen.
a Identified by this descriptor through the remainder of this paper.
Triangulation was used to ensure reliability byobtaining the same piece of information from threedifferent sources: oral statements from knowledge-able people, documentation, and visual observation.It was possible to obtain information from at leasttwo of these sources for almost all of the data. Threesources, however, were possible for only a few ofthe characteristics.
Referring back to Table 2, it was possible toobtain direct and continuous measures for bothMAT-COMPUTERIZATION and CAP-COM-PUTERIZATION simply by estimating the percentof P&IM tasks that are performed manually andthose that are performed by the computer systems.There are no direct measures, however, for the re-quirements and consumption variables MAT-RE-QUIREMENTS, CAP-REQUIREMENTS, MAT-CONSUMPTION, and CAP-CONSUMPTION, aswell as WIP-CONTROL. Thus, it was necessary toestablish the relative indirect measures shown in thetable for these five variables. Since a cluster analysiswas to be conducted on the variables, the indirect
measures were scaled so that the site with the leastamount of detail on each variable received the lowestscore and the site with the greatest detail receivedthe highest score. Each P&IM system was evaluatedand then assigned a score between one and fivewhere a 1 represents the least amount of detail and a5 the most.
For example, the lowest score for MAT-CON-SUMPTION, 1.0, is given when only monthly orless frequent physical inventories are taken. Since a2.0 represents weekly inventories, biweekly couldreceive a 1.5. However, to get more than a 2.0requires more rigorous attention to recording real-time data, either standard andror actuals. In addi-tion, the extensiveness of this attention is also impor-tant in terms of whether all materials receive real-timerecording or only some of them. Thus, based on thisinformation and the fact that inconsistencies in pro-cedures are common in manufacturing, the finalscore could range anywhere between the integers aswell as anywhere between the upper and lower val-ues of the range.
The data for the P&IM system variables wereexamined in several ways. First, a non-statisticalgeneral overview of the raw data was performed.
Ž .Next, Ward and Jennings’ 1973 clustering methodwas used to cluster the sites based on the P&IMsystem variables. Finally, an analysis of the means,standard deviations, coefficients of variation, medi-ans, and ranges of the variables was conducted toprovide information for within-group comparisonsand between-group comparisons for each of the indi-vidual clusters.
4.1. OÕerÕiew of the P&IM system data
Some interesting insights into the similarities anddifferences between the 19 individual process indus-try P&IM systems were gained from the raw data
Ž .concerning the seven P&IM variables see Table 4 .Although our conclusions about the types of processindustry P&IM systems can only be considered validfor the industries we investigated, because of thewide diversity of process firms included here, webelieve that the sample provides a good cross-sectionof the process industries.
First, an overall inspection of Table 4 illustratesthe wide range of data collected. Some variablesclump in the middle, others congregate at the endsand are sparse in the middle, others are relativelyevenly dispersed, and still others are somewhat ran-
Ždomly spread along the scale. The two computeriza-tion variables were transformed from their fractionalrange of 0 to 1, to match the other variables with a
.range of 1 to 5. Also of interest is that no site scoredconsistently across all measures at the same values,although we tend to see some that stay closer to oneend than the other, such as 12 near the bottom and11 near the top. Some general insights are describedbelow. Those variables that tended to be similaracross the sites are discussed first, followed by thosethat tended to differ among the sites.
4.1.1. MAT-REQUIREMENTSThe inter-site diversity in the level of detail re-
quired for the generation of raw material require-Žments is large but most sites are at the middle score
. Žof 3 . The simplest systems sites 1, 3, and 12 in.Table 4 use methods of generating requirements that
are based on knowledge of historical usage com-bined with visual observation of what is on hand.
This can be done effectively because of steady us-age.
The most detailed systems use a combination ofmethods including a form of cumulative MRP andmultiple-criteria ABC analysis. In these systems,regular bills of materials are used but the timebuckets are monthly. Material requirements are metby maintaining appropriate days of on-hand levels. Ifamounts fall below the desired days of on-handlevels, materials are ordered regardless of what themonthly MRP explosion calls for. Multiple-criteriaABC classifications influence the decisions on whatamounts of raw material inventories to maintain.This method cannot function as a regular MRP sys-tem because the computerized inventory tracking isnot perpetual. Inventory changes to the computer arebatched and therefore not timely enough to rely onfor ordering purposes. Even though the range forMAT-REQUIREMENTS is large, over half of the
Ž .sites received a score of three out of five for thisvariable reflecting their use of some sort of fixedinterval reorder system with set minrmax inventorylevels. Within these P&IM systems, the inventorylevels are updated from either physical inventories ora batched method and are not in real time.
4.1.2. CAP-REQUIREMENTSThe distribution on this variable is very similar to
that on MAT-REQUIREMENTS. At three of theŽ .sites 3, 12, and 14 the capacities are fixed and
known and production is run at basically the samecapacity all the time.
Eleven of the sites use capacity determinationmethods that are so similar that there were no signif-icant differences noted for the variable CAP-RE-QUIREMENTS. At these eleven sites the capacitiesare not fixed and are quite difficult to measure. Ineach, the capacity varies with the product mix andthe rough-cut capacity planning technique called ca-
Ž .pacity planning using overall factors CPOF is used.Individual work centers are scheduled by knowl-edgeable people using historical information. Thecapacity scheduling process is informal and manual;problems due to insufficient capacity are dealt withas they arise.
Ž .The most sophisticated methods sites 4 and 11for determining capacity requirements go through themechanics of finite loading. The results, however,
provide only approximate figures because the effectsof product mix and sequencing are not taken intoconsideration. In these systems, forecasted informa-tion is used to plan overtime or inventory buildups.However, once the aggregate capacity is set at theselevels, actual orders are infinite loaded at the disag-gregate level. Again, problems due to insufficientcapacity are dealt with as they arise.
4.1.3. MAT-CONSUMPTIONThe inter-site diversity in the tracking of raw
material consumption is quite large. Also, the distri-bution across this variable from the site using the
Ž .simplest method 19 to the site using the mostsophisticated raw material consumption tracking
Ž .method 6 is fairly even. In the simplest methods,consumption is not even tracked on a batch basis; amonthly inventory is conducted primarily for ac-counting purposes. Most of the firms that fall in themiddle of the distribution use a backflushing tech-nique to calculate estimated raw material consump-tion based on yields.
The most sophisticated method tracks inventoryon almost a perpetual basis. The standard amounts ofraw materials consumed are sent to the computer atthe time of consumption and actual amounts arereported when the batch is completed. The lead timesare short so the information is fairly timely. It isinteresting to note that even though this system issupported by the most accurate and timely inventoryrecords, it still does not use MRP. The raw materialsare ordered based on reorder points.
4.1.4. CAP-CONSUMPTIONThree of the sites have P&IM systems that em-
ploy very simple methods for tracking capacity con-sumption and do not measure consumption directly.Production is run full out all the time and only theaggregate volumes of materials produced over timeare measured. The majority of the sites have P&IMsystems that use a tracking method where laborhours are tracked and the exact quantities of materi-als produced over time are recorded. Some of thesemethods measure labor at standard while others mea-sure it at actual. The most sophisticated methodtracks actual on a batch basis for both labor andmachine hours and compares them to standard. Nosite has a P&IM system that employs a real timecapacity consumption tracking method.
4.1.5. WIP-CONTROLSix of the sites were ranked at the lowest level for
the tracking of WIP. These sites have P&IM sys-tems that do not track WIP on a formal basis.Walking the floor or making telephone calls are theprimary sources of information for WIP at these
Ž .sites. Only one site 6 was ranked at the highestlevel of detail for tracking WIP. This site tracks WIPon a board that is kept in production where the exactlocation of all batches is recorded. All movement,stages, and locations are tracked on a real time basis.It is interesting that this most sophisticated WIPtracking method is a manual one.
4.1.6. MAT-COMPUTERIZATIONThe fraction of computerization of the materials
Žcontrol methods MAT-REQUIREMENTS, MAT-.CONSUMPTION, and WIP-CONTROL varies
widely. Two sites use methods for the control ofmaterials that are completely manual. Eleven of theP&IM systems use methods that are less than halfcomputerized. The most computerized systems haveabout three-quarters of their material control meth-ods computerized.
4.1.7. CAP-COMPUTERIZATIONThree sites have not computerized any of the
methods used for capacity control in their P&IMsystems. Ten sites are less than half computerized.The most computerized systems with respect to ca-pacity control methods are almost fully computer-ized.
5. P&IM system subgroups
The third research question is answered in thissection: can the different P&IM systems be put intomanageable subgroups based on their P&IM vari-ables and, if so, what are the subgroups and theircharacteristics?
5.1. Clustering the 19 P&IM systems
The full clustering sequence for the 19 P&IMsystems based on the seven P&IM variables is shownin the dendrogram in Fig. 1. Ward and Jennings’Ž .1973 hierarchical clustering method was used toidentify the P&IM system clusters. In this method,
the distance between two clusters is the sum of thesquares between the two clusters summed over allthe variables. This method was chosen primarily on
Ž .the recommendation of Anderberg 1973 , and be-cause it is hierarchical.
Since Ward’s method uses the sum of the squaresof distances, it is not necessary for the data to havenormality. However, for this same reason, it wasnecessary to standardize the data to keep certainvariables from influencing the results more heavilythan others. All the variables were thus standardizedto a mean of zero and a standard deviation of one.
Ž .SAS 1988 was the statistical package used to per-form Ward’s method.
After Ward’s method was used on the data, it wasnecessary to determine how many different clusterswere actually formed. Deciding on a method to use
Žfor this determination is not straightforward Everitt,. Ž .1979, 1980 . SAS Institute 1988 , however, sug-
gests that when cluster analysis is used for thepurpose of dissection, the R2 value provides anappropriate cut-off point. Thus, the criterion chosento determine the final number of clusters in thisanalysis was the R2 value. The R2 values in Fig. 2show the amount of explained variance that is lost byreducing the number of clusters. The plot of thenumber of clusters versus the R2 values in Fig. 2shows that a substantial drop-off in the R2 values
Žoccurs below four clusters only 0.088 from 5 to 4.versus 0.111 from 4 to 3 . Consequently, the number
of clusters should not be reduced beyond four. Thedendrogram in Fig. 1 shows the final four P&IM
Žsystem clusters on the left Cluster 1 through Cluster.4 that were formed by using this cutoff point.
5.2. Four types of P&IM systems
It is apparent that a large amount of diversity wascaptured in the sample of 19 process industry sites.In fact, four distinct types of P&IM systems, de-scribed in detail below, were identified within the19-site sample of this research. Thus, it can beconcluded that there are at least this many differenttypes of P&IM systems in the entire population ofprocess industries. The major contribution of thisresearch is the development of this taxonomy ofprocess industry P&IM systems based on the sevendistinguishing variables.
Ž .Fig. 1. Dendrogram for P&IM systems Clusters 1 through 4 .
Table 5 gives the actual values and standarddeviations of each of the seven variables. AnANOVA test was performed on the clusters for each
Fig. 2. Determining the number of P&IM system clusters.
variable and the p-values are listed at the right sideof Table 5, along with the clusters that were identi-fied as significantly different on that variable from apost-hoc Scheffe test. As can be seen, there were´highly significant differences on all the variablesexcept capacity requirements, which was only signif-icant at the 0.10 level rather than 0.05 as used in this
ŽScheffe test. If the 0.10 level had been considered´significant, the corresponding different clusters would
.have been 3 and 4. As might be expected, Clusters 3Ž . Žthe AcomputerizedB cluster and 4 the AsimpleB
.cluster were found to be the most different, differingsignificantly on five of the seven variables. Clusters2 and 3 were also quite different, differing on four ofthe seven variables. Considering all the variables atonce, all of the clusters differed significantly fromeach of the other clusters, often on multiple vari-
ables. For example, Cluster 3, the computerized clus-ter, was found to differ more than all the otherclusters, differing eleven times from the other clus-ters among the seven variables. Cluster 4, the simplecluster, was next most different, differing nine times.Not surprisingly, Cluster 1, the AcommonB cluster,differs the least from the other clusters, being signifi-cantly different only five times.
Cluster 4 was termed the ASimpleB cluster be-cause it had the lowest scores on almost all the sevenvariables, only once slightly exceeding another clus-
Ž .ter Cluster 2 on CAP-CONSUMPTION . Cluster 3
was termed the AComputerizationB cluster because ithad the highest scores on both computerization vari-ables. As might be expected due to the demonstratedneed for computerization, this cluster also scoredhigh on both measures of requirements, as well asconsumption. Cluster 2 was termed the AWIP-CON-TROLB cluster due to its exceedingly high score onthat variable whereas its other scores were close tothose of the Simple cluster. Finally, Cluster 1 wastermed the ACommonB cluster due to its having moresites than any other cluster and scoring in the middleof all the variables.
Given the difficulty of seeing the patterns inTable 5, the relative positions of the clusters aremore clearly displayed in terms of the seven vari-ables in the graph of Fig. 3. Here, each cluster isplotted against the seven variables at the bottom. Thefigure more clearly illustrates the points made in theprevious explanation such as the Simple cluster hav-ing the lowest scores, the WIP-Controlled clusterbeing substantially higher in WIP-CONTROL thanthe other clusters, and so on.
5.3. Validating the clusters
To validate the clusters, two methods suggestedŽ .by Jobson Jobson, 1992, p. 564 were employed:
validating the cluster results by conducting otherforms of cluster analyses to see if they form thesame clusters, and running a split-half cluster analy-sis to see if both halves form the same set of clusters.The two additional clustering analyses to validateWard’s method were centroid heirarchical clusteranalysis and average linkage cluster analysis. Bothmethods resulted in the same identical four clustersas Ward’s method. We also used Ward’s method torun the split-half analysis, both halves again resultedin the same four clusters.
In addition, the observations themselves tend tovalidate the clusters in the following sense. Threesets of variables in Table 5 have two items each thatmight be expected to run somewhat similarly withineach cluster: material and capacity requirements, ma-terial and capacity consumption, and material andcapacity computerization. Although these pairs neednot be highly correlated within a cluster, it would besurprising if one variable of a pair was quite highwhile the other was quite low. As it happens, in allfour clusters, all three pairs of variables tend to benear the same values and the same rank among thefour clusters. For example, cluster 1 is ranked secondon material requirements and also second on capac-ity requirements, it is ranked third on material com-puterization and also third on capacity computeriza-tion, and finally, it is ranked second on materialconsumption and second on capacity consumption.The same can be seen in Fig. 3 for the other clusters.
5.4. Interpretation of the clusters
The interpretation of the characteristics of each ofthe clusters is next described in more detail.
5.4.1. Cluster 1: commonThe site members of this cluster are COLORS,
COATINGS, ORGANICS, FEED-BLEND, MEATS,and PAINT. The means of the planning and controlvariables — MAT-REQUIREMENTS, MAT-CON-SUMPTION, CAP-REQUIREMENTS, CAP-CON-SUMPTION — all rank second indicating that thiscluster is the second most detailed with respect to thegeneration of materials and capacity requirementsand the tracking of materials and capacity consump-tion. The specific characteristics implied by the
Ž .means of these four variables are: a a fixed intervalreorder point method is used to generate raw mate-rial requirements; updates to records are made daily;Ž .b material consumption is tracked by recordingstandard consumptions with periodic actual adjust-
Ž .ments; c the capacity requirements are generatedŽ .by CPOF; and d actual labor hours andror ma-
chine hours are tracked for each product.The mean score for the variable WIP-CONTROL
also ranks second and suggests that WIP is trackedby physical observations supplemented with detailedshift notes. The computerization variables MAT-COMPUTERIZATION and CAP-COMPUTERIZA-TION both rank third and indicate that somewhatless than half of their materials and capacity controlsystems are computerized.
5.4.2. Cluster 2: WIP-controlledThe members of this cluster include: FLEXPAC-
EXTRUDE, FLEXPAC-PRESS, BREW, and IN-DUST CLEAN. The means of the planning andcontrol variables suggest the following character-
Ž .istics: a raw material requirements are generated bythe use of a fixed interval reorder point system withset minrmax inventory levels; material updates oc-
Žcur somewhere between weekly and daily less fre-. Ž .quently than for cluster 1 ; b raw material con-
sumption is tracked by a combination of physicalinventories and standard consumption recording at
Ž .the start of each batch; c the generation of capacityrequirements falls between informal infinite load touse of capacity planning using overall factorsŽ . Ž .CPOF ; and d the tracking of capacity consump-tion falls between not being tracked at all and keep-ing track of some standard labor hours. Cluster 2 hasthe highest mean score for WIP-CONTROL of thefive clusters. The mean score of 4.13 implies that the
starts of batches are tracked and that some of thetracking is actually done by the processing equip-ment. The variables CAP-COMPUTERIZATION andMAT-COMPUTERIZATION rank second, with ma-terials control being somewhat over half computer-ized and capacity being somewhat less than halfcomputerized.
5.4.3. Cluster 3: computerizedThe members of Cluster 3 are PLASTIC,
PHARM-LIQUIDS, PHARM-TABLETS, and ICECREAM. All the planning and control variable means— MAT-REQUIREMENTS, MAT-CONSUMP-TION, CAP-REQUIREMENTS, CAP-CONSUMP-TION — for Cluster 3 rank first, implying that thiscluster uses the greatest amount of detail for generat-ing materials and capacity requirements and trackingmaterials and capacity consumption. This amount ofdetail is obviously what generates the need for thehigh degree of computerization. The characteristics,
Ž .as implied by the means, are: a raw material re-quirements are generated by a combination of meth-ods including some use of explosions for gross re-quirements, multiple-criteria ABC analysis, and fixedreorder point methods; materials are updated fre-
Ž . Ž .quently almost real time ; b the actual consump-tion of raw materials is tracked at the completion of
Ž .a batch; c capacity requirements are generated byŽ .finite loading using approximate values; and d
capacity consumption is tracked by recording actuallabor hours and machine hours for each product. Themean for the WIP-CONTROL variable is exception-ally low and suggests that WIP is tracked verysimplistically by the use of physical inspections,meetings, and shift notes. Both of the means for thecomputerization variables rank first, of course, andtheir control is highly computerized.
5.4.4. Cluster 4: simpleThe members of Cluster 4 are RESINS, BEVER-
AGES, FINISHES, BAKE, and FEED-YEAST. Theplanning and control variable means generally rank
Ž .last and suggest the following: a the methods usedfor the generation of raw material requirements fallbetween knowing the steady flow usage and theoccasional use of fixed interval reorder points with
Ž .semi-weekly updates; b raw material consumptionis tracked by a combination of taking weekly physi-
cal inventories and recording standard consumptionsat the completion of a batch for AkeyB raw materials;Ž .c capacity requirements are either known and fixed
Ž .or generated by using CPOF; and d the methodsused to track the consumption of capacity are either
Ža combination of informal tracking e.g., meetings,.shiftnotes plus tracking of standard labor hours or
are not tracked formally at all. The mean for theWIP-CONTROL variable also ranks last implyingthe WIP is tracked by walking the floors, makingphone calls, or through informal discussions in pro-duction meetings. Understandably, this cluster alsoranks last on both computerization variables.
5.5. Managerial interpretation of the four clusters
In managerial terms, these four basic types ofP&IM systems generally make sense for processindustries, although they cannot be confidently ex-trapolated beyond the industries investigated here.There is a Simple P&IM system for those processindustries that are relatively straightforward and aCommon system for those that are more normallydemanding, especially in terms of their materials andcapacity requirements and consumption. However,many firms that have the most complex and demand-ing production processes have turned to computer-ized technology to more precisely monitor and con-trol those processes; these more technologicallysophisticated P&IM systems are thus called Com-puterized. However, there is yet another type ofP&IM system that does not have especially highdemands for generating and tracking the consump-tion of materials and capacity but very closely moni-tors WIP, usually through the processing equipmentitself. Their goal is often to maintain short leadtimes; this P&IM system is thus called WIP-Con-trolled. As might be expected, the level of computer-ization of this system is relatively high.
In addition to the variables that distinguishedbetween the clusters, it is interesting to note thatneither size nor product gives any indication of thesimilarity of P&IM systems. For example, the largestplants, pharmaceuticals, clustered with ice cream,one of the smallest. And the smallest, finishes, clus-tered with the second largest, resins. Equally interest-ing is the unexpected similarities of firms withineach cluster and the kinds of differences across
clusters. For example, beverage P&IM systems havemore in common with baking systems due to their
Žsimplicity than they do with brewing close control. Ž .of WIP , ice cream computer controlled , or paints
Ž .common demands , all of which are liquids likebeverages. Similarly, the P&IM system for paints is
Ž .more like that for meats common demands than itŽ .is for other chemicals like finishes simple demandsŽor industrial cleaning chemicals close control of
.WIP . What this means for the manager is that someanalysis of the seven variables identified here isrequired to determine which P&IM systems wouldbe best to benchmark, or what colleagues to talk toabout similar problems in their production process.Based on this research, what might seem like anindustry with a similar P&IM system may lead amanager into taking precisely the wrong actions.
6. Conclusion
As stated in the Introduction, the goal of ourresearch here was to identify dimensions that coulddiscriminate between process industry P&IM sys-tems, determine how these systems differ from oneanother, and identify major subgroups of such sys-tems and their characteristics. An essential step re-quired to accomplish this goal has been offered here:the detailed analysis and categorization of 19 processindustry P&IM systems. The systems were shown todiffer significantly on six of the seven identifiedvariables: materials requirements, materials and ca-pacity consumption, WIP control, and degree ofmaterials and capacity computerization. The catego-rization resulted in four distinct groups of P&IM
Ž . Ž . Ž .systems: 1 common, 2 WIP-controlled, 3 com-Ž .puterized, and 4 simple. Since these 19 sites exhib-
ited four clearly understandable P&IM systems, theyappear to be relatively generic to the process indus-tries, though we cannot confidently extrapolate be-yond our sample. A larger sample might, of course,
Žuncover more unexpected types similar to the largely.unexpected WIP-controlled cluster here but those
identified here would probably surface also. Forexample, a AsimpleB cluster and a AcomputerizedBcluster would certainly represent two expected ex-tremes. However, there may be more subclasses ofwhat we have identified here as the AcommonB
cluster. Future research should be directed towardtesting the results of this study, extending the resultswith larger and more varied groups of sites, andeventually determining the AbestB P&IM systemunder various processing circumstances.
In addition to the categorization, this phase ofthe research provided other important insights, par-ticularly in the realm of existing theory. Some
Žresearchers Bolander, 1981a,b, 1983; Taylor and.Bolander, 1991 have suggested that process indus-
Žtries need to focus primarily on capacity based on.the assumption that they tend to be capital intensive
and that materials should be considered secondarily.In this sample, only 53% of the firms consideredthemselves to be capital intensive and only one siteŽ .FEED-YEASTS considered maximizing the use ofcapacity to be a particularly important goal. Also, inthis sample, the level of detail exercised for capacityplanning and control was generally less than thatrequired for materials planning and control. In fact, itwas secondary to material control at most sites.Finally, an increased level of sophistication in capac-ity control was associated with an increased level ofsophistication in materials control, implying that asmaterials are more closely controlled, so are capaci-ties. That is, as firms invest in their P&IM system,they tend to invest in both capacity and materialcontrol.
In general, dividing process industry P&IM sys-tems between material-dominated versus capacity-dominated, or time-phased versus rate-based, or evenmake-to-stock versus make-to-order versus assem-ble-to-order, as existing theories advocate, providesinsufficient guidance for managers to identify P&IMsystems similar to their own. As indicated above,selecting another plant that also has a rate-based
Ž .system or capacity-dominated or make-to-stock maywell result in selecting a system from a totallydifferent cluster that is not at all comparable to one’sown. Thus, existing theory is not only insufficient; itmay be wrong. That is, if you have a rate-basedsystem and find another to compare yours to, it is notjust that not all rate-based systems will be appropri-ate, the most appropriate comparison may well be toa time-based system.
In summary, the categorization developed in thisresearch study provides an improved understandingof process industries’ P&IM systems. It is hoped
that this understanding provide firms with an en-hanced ability to share P&IM system accomplish-ments with other similar types of process firms andfoster additional study in this critical and largelyunder-researched area.
Acknowledgements
We would like to acknowledge the statistical helprendered us by Mark Weaver and Clinton Dart, andthe SAS data analysis by Alex Wilson.
References
Abraham, C., Dietrich, B., Graves, G.W., Maxwell, W.L., Yano,C., 1985. Discrete Manufacturing Taxonomy, working docu-ment. IBM Research, Yorktown Heights, NY.
Adelberg, A.H., 1984. An improved analysis of production-mixvariances. Production and Inventory Management Journal 25Ž .4 , 35–41.
Aiello, J.L., 1982. Production cost measurement and control in theprocess industry. APICS 25th Annual International ConferenceProceedings, Falls Church, VA. pp. 542–544.
Allen, R.L., 1980. The applicability of standard manufacturingsoftware to primary metals processing. APICS 23rd AnnualConference Proceedings, Falls Church, VA. pp. 85–89.
Allen, S.J., Schuster, E.W., 1994. Practical production schedulingwith capacity constraints and dynamic demand: family plan-ning and disaggregation. Production and Inventory Manage-
bias in mail surveys. Journal of Marketing Research 4, 362–402.
Baumeister, W., 1997. Building an infrastructure. ManufacturingŽ .Systems 15 1 , 34–36.
Bolander, S.F., 1981a. Capacity planning through forwardscheduling. APICS Process Industry Seminar Proceedings,Falls Church, VA. pp. 73–79.
Bolander, S.F., 1981b. Capacity planning with TPFS. APICS 24thAnnual International Conference Proceedings, Falls Church,VA. pp. 207–209.
Bolander, S.F., 1983. Time-phased forward scheduling: a capacitydominated scheduling technique. Production and Inventory
Ž .Management Journal 24 1 , 83–97.Bolander, S.F., Taylor, S.G., 1993. System framework for process
flow industries. Production and Inventory Management Jour-Ž .nal 34 4 , 12–17.
Bolander, S.F., Taylor, S.G., 1990. Process flow scheduling:mixed-flow cases. Production and Inventory Management
Ž .Journal 31 4 , 1–6.
Boyer, K.K., Ward, P.T., Leong, G.K. Approaches to the factoryof the future: an empirical taxonomy. Journal of OperationsManagement, Vol. 14, No. 4, 297–314.
Clark, F.T., 1983. Sperry and the process industry. APICS ProcessIndustries Seminar Proceedings, Falls Church, VA. pp. 50–63.
Cohen, R.L., 1980. Production and inventory control in the pro-cess industry. APICS 23rd Annual Conference Proceedings,Falls Church, VA. pp. 153–154.
Cokins, G.M., 1988. Control systems for process industries. Man-ufacturing Systems, 79–83, May.
Covey, R.E., 1984. MRP for process manufacturers. ProductionŽ .and Inventory Management Review and APICS News 4 3 ,
44–48.Crow, D.A., 1992. Process industries need a new generation of
system solutions. APICS — The Performance Advantage 2Ž .4 , 29–32.
Daniels, V.L., 1983. The lot sizing and sequencing problem in theprocess industries. Ph.D. dissertation. Ohio State University,Columbus, OH.
Dayvolt, B.H., Symonds, F.P., 1994. Automate batching in aŽ .multi-product facility. Chemical Engineering 10 7 , 151.
Dennis, D.R., 1993. Defining Production and Inventory Manage-ment Systems for Process Industries. Ph.D. dissertation. Uni-versity of Cincinnati, Cincinnati, OH.
Dietrich, B.L., 1987. A Taxonomy of Discrete ManufacturingSystems. Manufacturing Research. IBM Watson Research,Yorktown Heights, NY, pp. 1–15.
Doganaksoy, N., Hahn, G.J., 1996. Evaluating the potential im-pact of blending on product consistency. Journal of Quality
Ž .Technology 28 1 , 51–60.Duncan, R.M., 1981. The by-product bill of material. APICS 26th
Annual International Conference Proceedings, Falls Church,VA. pp. 288–292.
Eads, G., 1989. Manufacturing systems for the process and repeti-tive industry. Production and Inventory Management Reviewand APICS News, Vol. 9, No. 2, 32, 34.
Eads, G., Undheim, H., 1984. Inventory control with lot traceabil-ity and quality assurance. Readings in Production & InventoryControl and Planning. APICS 27th Annual International Con-ference, Falls Church, VA. pp. 57–61.
Everitt, B.S., 1979. Unresolved problems in cluster analysis.Biometrics 35, 169–181.
Finch, B.J., 1986. Production planning and control for processmanufacturing. Ph.D. dissertation. University of Georgia,Athens, GA.
Finch, B.J., Cox, J.F., 1987. Production planning and control:principles and cases for process manufacturing. AmericanProduction and Inventory Control Society, Falls Church, VA.
Finch, B.J., Cox, J.F., 1988. Process-oriented production planning
Haxthausen, N., 1995. The painkiller for batch control headaches.Ž .Chemical Engineering 2 10 , 118–124.
Hubbard, D.T., Taylor, S.G., Bolander, S.F., 1992. Process flowscheduling in a high-volume repetitive manufacturing environ-
Ž .ment. Production and Inventory Management Journal 33 4 ,21–26.
Jobson, J.D., 1992. Applied Multivariate Data Analysis.Springer-Verlag, NY.
Katayama, H., 1996. On a two-stage hierarchical production plan-ning system for process industries. International Journal of
Ž .Production Economics 44 1 , 63–72.Landel, R.D., 1982. Taking the pulse of a business: a plant-tour
guide for assessing management effectiveness. OperationsManagement Review, 44–48.
McKaskill, T., 1992. Applying the process production model todiscrete manufacturing. Manufacturing Systems, 70–72, Octo-ber.
Meredith, J.R., Raturi, A., Amoako-Gyampah, K., Kaplan, B.,1989. Alternative research paradigms for operations. Journalof Operations Management 8, 297–326.
Nelson, N.S., 1983. MRP and inventory and production control inprocess industries. Production and Inventory Management 24Ž .4 , 23–32.
Nichols, W.E., Ricketts, J.D., 1994. Identification, tracking, con-Ž .trol key to minimizing mill liquor loss. Pulp and Paper 68 3 ,
111–121.Novitsky, M.P., 1983. Process industries — where are you?
Ž .Production and Inventory Management Journal 24 1 , 118–121.
Parker, K., 1997. A configurator for process industries. Manufac-Ž .turing Systems 5 9 , 30–32.
Plossl, G.W., Wight, O.W., 1967. Production and inventory con-
trol, Principles and Techniques. Prentice-Hall, EnglewoodCliffs, NJ.
Rice, J.W., Norback, J.P., 1987. Process industries productionplanning using matrix data structures. Production and Inven-
Ž .tory Management Journal 28 2 , 15–23.Robinson, W.E., Taylor, S.G., 1986. Alternatives for planning and
scheduling production: matching tools and tasks. Workingpaper.
Rutten, W.G.M., 1993. Hierarchical mathematical programmingfor operational planning in a process industry. European Jour-
Ž .nal of Operational Research 64 3 , 363–369.SAS Institute, 1988. Statistical Analysis System. Cary, NC.Shelley, S., 1995. On-line device automates conductivity measure-
Ž .ments. Chemical Engineering 2 3 , 151.Swann, D., 1984. MIS for process industries. Production and
Ž .Inventory Management Review and APICS News 4 6 , 65–70.Taylor, S., 1979. Production and inventory management in the
process industries: a state of the art survey. Production andŽ .Inventory Management 20 1 , 1–16.
Taylor, S., Bolander, S., 1990. Process flow scheduling: basicŽ .cases Part 1 . Production and Inventory Management Journal
Ž .Production and Inventory Management Journal 32 1 , 67–71.Taylor, S., Bolander, S., 1993. Process flow scheduling calcula-
Ž .tions. Production and Inventory Management Journal 34 1 ,58–64.
Taylor, S., Bolander, S., 1995. A framework for flow manufactur-Ž .ing scheduling systems. Industrial Management 37 6 , 1–4.
Taylor, S., Bolander, S., 1997. Process flow scheduling: past,present, and future. Production and Inventory Management
Ž .Journal 38 2 , 21–25.Thompson, O., 1991. Manufacturing in the process industries.
Production and Inventory Management and APICS News 11Ž .2 , 38–39.
Tiber, J.S., 1981. Use of the dynamic bill in process industries.APICS Process Industries Seminar Proceedings, Falls Church,VA.
Vollman, T.E., Berry, W.L., Whybark, D.C., 1997. ManufacturingPlanning and Control Systems. IrwinrMcGraw Hill, BurrRidge, IL, pp. 4–7.
Ž .Wallace, T.F. Ed. , APICS Dictionary. 7th edn. American Pro-duction and Inventory Control Society, Falls Church, VA.
Ward, J.H. Jr., Jennings, E., 1973. Introduction to Linear Models.Prentice-Hall, Englewood Cliffs, NJ.
International Journal of Operations & Production ManagementA Typology of Production Control Situations in Process IndustriesJan C. Fransoo Werner G.M.M. Rutten
Article information:To cite this document:Jan C. Fransoo Werner G.M.M. Rutten, (1994),"A Typology of Production Control Situations in Process Industries",International Journal of Operations & Production Management, Vol. 14 Iss 12 pp. 47 - 57Permanent link to this document:http://dx.doi.org/10.1108/01443579410072382
Downloaded on: 14 November 2016, At: 12:18 (PT)References: this document contains references to 31 other documents.To copy this document: [email protected] fulltext of this document has been downloaded 1988 times since 2006*
Users who downloaded this article also downloaded:(2002),"Cyclical packaging planning at a pharmaceutical company", International Journal of Operations & ProductionManagement, Vol. 22 Iss 5 pp. 549-564 http://dx.doi.org/10.1108/01443570210425174(1995),"Production control in the food processing industry: The need for flexibility in operations scheduling", InternationalJournal of Operations & Production Management, Vol. 15 Iss 8 pp. 73-88 http://dx.doi.org/10.1108/01443579510094107
Access to this document was granted through an Emerald subscription provided by emerald-srm:211952 []
For AuthorsIf you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors serviceinformation about how to choose which publication to write for and submission guidelines are available for all. Pleasevisit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.comEmerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio ofmore than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of onlineproducts and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on PublicationEthics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation.
*Related content and download information correct at time of download.
A Typology of ProductionControl Situations inProcess Industries
Jan C. Fransoo Eindhoven University of Technology, The Netherlands, and
Werner G.M.M. RuttenWageningen Agricultural University, The Netherlands
IntroductionDuring the last decade, various articles have appeared in professional andscientific journals regarding production control in process industries. The vastmajority have focused on the typical characteristics of process industriesproduction control vis-à-vis the more traditional approaches of productioncontrol for discrete manufacturing systems. In this body of literature, twoschools of thought can be distinguished. The first advocates the applicability oftraditional MRP (manufacturing requirements planning) concepts and systemsin process industries[1-3]. The researchers and practitioners in this schoolconcentrate on the specific characteristics that may occur in process industriesand try to find solutions to be able to implement MRP. The second schoolstresses the differences between discrete and process manufacturers and comeswith new or adapted techniques and concepts for production control in thesesituations[4]. Very seldom is the variety of production systems within processindustries discussed. Some articles do address the problem of variety (or theopportunities this offers), but the consequences for production control are notworked out in more detail.
In this article, we will present a simple, though useful typology of processindustries, which recognizes two extreme production systems on a continuum.The typology is in line with the APICS (American Production and InventoryControl Society) definitions on process/flow and batch/mix[5]. APICS definesbatch/mix as:
A process business which primarily schedules short production runs of products.
Process/flow is defined as:A manufacturer who produces with minimal interruptions in any one production run orbetween production runs of products which exhibit process characteristics such as liquids,fibres, powders, gases.
It will appear that these definitions are very useful in characterizing themanufacturing systems in view of the requirements for production control.Most research so far has been focused on the process/flow systems; and what is
International Journal of Operations& Production Management, Vol. 14
The authors are indebted to Professor J.W.M. Bertrand, whose suggestions considerablyincreased the quality of this article.
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
IJOPM14,12
48
being done on batch/mix systems mainly presents detailed scheduling/programming approaches. Additional research to obtain a production controlframework for batch/mix businesses is therefore required.
After presenting the characteristics of process industries as recognized in theliterature and organizing these characteristics for each of the two extremes, wewill address the differences in production control in more detail. In the lastsection of the article we will draw some conclusions.
Literature on Process IndustriesThe general characteristics of process industries are well represented in theAPICS definition:
Process industries are businesses that add value to materials by mixing, separating. forming,or chemical reactions. Processes may be either continuous or batch and generally require rigidprocess control and high capital investment[6].
The definition indicates that the type of manufacturing process performed isone of the most important characteristics. Mixing, separating, forming andchemical reactions are operations that are usually performed on non-discreteproducts and materials. These processes can only be performed efficientlyusing large installations, which tend to be very expensive. If large quantities aredemanded, this justifies continuous production (thus higher investment). Ifdemand is low, the investment into a large installation is not worthwhile, andbatchwise production is used. Also, these processes are difficult to controlwhich often results in typical symptoms as variable yield and returning flowsof material.
In the literature, many characteristics are mentioned as being “typical” ofprocess industries. Though these characteristics can be found in processindustries, they are not general, in a sense that virtually all process industriesare characterized by these issues. On the other hand, they are discriminating inthat they will predominantly be found in process industries and not in discreteindustries. In this section, we will provide an overview of these characteristics.Production scheduling in process industries is often complicated by a variableyield, due to the nature of the process, even if it is statistically under control[7].In process/flow businesses, the yield can change as a function of processingdecisions[8]. Burt and Kraemer[9] present two ways to deal with variable yieldin a production control system: (1) use a mean yield in the bill of materials(BOM) and (2) create a safety stock of raw materials which have the mostvariable yield. In a later paper however, Burt[10] states that variable yieldshould be controlled by creating safety time instead of safety stock.
Process industries often obtain their raw materials from mining oragricultural industries. These raw materials have natural variations in quality.For example, crude oils from different oil fields have different sulphur contentsand different proportions of naphtha, distillates, and fuel oils. Oil refinerydesigns, production plans and operating schedules must account for thisvariability[4]. Another aspect of materials variability associated with naturalraw materials, is that the yield or potency is usually not known or measured
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
ProductionControl
Situations
49
until the process is started[11]. The variability in raw materials quality oftendetermines which products will be produced[12]. Kochalka[1] advises to plan atthe average quality or yield of the raw materials. If you get a different quality, itmay mean reorder and recycle. This can result in shortages, but if the safetystocks are established giving consideration to the frequency of theseoccurrences, the stock-out impact can be minimized.
Variations in raw material quality often lead to variations in bills of material(recipes)[13]. For example, variations in the moisture contents, acidity, colour,viscosity or concentration of active ingredient in raw materials may causevariations in the ingredient proportions required to make finished productquality specifications[4]. Another factor which causes variations in bills ofmaterial is the price of alternative ingredients[4]. For example, a pet food mayhave specifications for the minimum amount of proteins, carbohydrates andfats per pound of pet food; however, the proportions of various ingredients maybe varied depending on their current price and availability. In processindustries, intermediate products are quality-measured and the results candictate formula-sensitive processing steps requiring varying, not fixed,“quantities per…” and alternative or additive compounds. Seasonalconsiderations, the availability of raw materials, or even the unique vessel, tankor line availability can govern the best recipe (BOM) for production[14].
Process industries often initiate their flows with only a few raw materials andsubsequently process a variety of blending and resplitting operations[14]. Inother words, many products are produced from a few kinds of raw material,compared to the usual bill in discrete manufacturing in which end items containmany different components[12]. Figure 1 exhibits the differences betweenprocess and discrete manufacturers.
The divergence in the product flow sometimes is not voluntary because by-products are being produced at certain processes[15]. It is important tostructure the appropriate BOM to recognize the yield of by-products.Theseitems in the BOM may be included by giving them a “negative quantity per”,equal to the standard amount of the by-product yield[1,11]. When therequirements are exploded, these items will show as negative, or in other words,as an inventory gain. Duncan[2] developed a by-product BOM because the“negative quantity per” can cause “netting” being confused with “planning”
Figure 1.Process and Discrete
Manufacturing;ConceptualDifferences
Outputs
InputsDiscrete Discrete
with productoptions
Process Processwith product
options
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
IJOPM14,12
50
and because it can cause the shop floor control system to expect a negativereceipt of the output into stores. The by-product BOM connects processes aswell as components and every process can have numerous outputs independentof the number of inputs. Duncan defines a task-item which is a process. Everytask-item (process) gets input from processed items (the components) or fromother task-items. The output of a task-item is one or more processed items(components). The features of the by-product BOM are that the bill can handlemultiple outputs and multiple inputs, and that many levels of the bill can be tiedtogether through the process task-items.
A common problem is the unit of measure (“quantity per”). Themanufacturing BOM shows component quantities per batch of parent (e.g.litres) and the product BOM, as used for forecasting etc., shows componentquantities per unit of parent (e.g. bottles)[16]. This problem can be solved byfinding a common denominator[1]. Furthermore, the per unit BOM needs toaccommodate many decimal places because of the unit of measure relationshipin the BOM between stocking units. For example, the active ingredient of apain-killing tablet is stock in kilograms, but the standard tablet contains0.00325 kg (325 milligrams)[11]. Rice and Norback[12] use matrix datastructures to solve the unit of measure problem. They build matrices of theproduction schedule and the product structure, with which they can allocate the
Table I.CharacteristicsMentioned in theLiterature
Characteristic Literature Example of industry
Variable yield Sepheri et al.[7] Chemical industryHaglund et al.[8]Burt and Kraemer[9]Burt[10]May[11]
Variable quality Taylor et al.[4] Oil forest productsRice and Norback[12]Kochalka[1]
Variable quantity/availability Cokins[14] Coffee, agricultural industry
Variable recipe Taylor et al.[4] Oil (animal) food industry, paperRutten[13]
Divergent flow Fransoo[15] Glass
Price of raw materials Taylor et al.[4] Agricultural
Divergent BOM/by-products Cokins[14] Beef cutting, forestryRice and Norback[12]May[11]Duncan[2]
Unit of measure/batch Appoo[16] Fine chemicals, drugsproblem Kochalka[1]
May[11]Rice and Norback[12]Nelson[3]
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
ProductionControl
Situations
51
costs of capacities and materials to the products. Another effect of batchproduction in process industries, is that usually the total batch must bescrapped when the quality is poor. In discrete batch manufacturing a portion ofthe total batch might be rejected, but it is unlikely that the entire batch would berejected[3].
The characteristics found in the literature, are summarized in Table I. Asmentioned above, a lot of these typical characteristics have been tackled interms of data registration. However, in order to address the control problem,process industries will be characterized from a different point of view. This willbe clarified in the development of a typology in the next section.
TypologySamuel Taylor and his research group published an innovative series of articlesin the first half of the 1980s on production control in process industries. In oneof their first articles[4], they discuss a typology of industries in general intowhich they fit all kinds of process industries. The two dimensions they use are:degree of product differentiation and material flow complexity. The degree ofproduct differentiation refers to the marketing environment of the business; thematerial flow complexity refers to the way the production process is organized.Taylor already notes that some fabrication (i.e. discrete) industries tend towardsthe flow shop/commodity type, while some process industry groups (e.g.speciality chemicals) are in the centre of the matrix. So both process anddiscrete manufacturers are spread over the matrix.
As appears from Figure 2, which depicts this typology, these two axes are infact one: the more an industry appears to be a job shop, the more its productsare customer specific. Therefore, we propose to only use one axis with twoextremes: job shop/custom specific and flow shop/commodity. Only processindustries will be included in this typology. Industries producing discreteproducts are excluded. Products are not discrete if individual items are
Figure 2.Taylor’s Typology
(1981)
AerospaceIndustrial machinery apparel Machine tools Drugs Speciality chemicals Electrical and electronics Automobile Tyre and rubber Steel products Major chemicals Paper Containers Brewers Oil Steel Forest products
CustomLow volumedifferentiated
High volumedifferentiated Commodity
Jobshop
Flowshop
Pro
cess
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
IJOPM14,12
52
indistinguishable from each other (like oil, chemicals) or if the products aresimple and produced in very large quantities such that it does not make sense todistinguish them individually (like glass bottles, aluminum cans). Thischaracterization refers to a single-phase process. In case of a multi-phaseproduction system, it is obvious that this refers to the most important step inthe production process (which creates the majority of the added value). Ourtypology is presented in Figure 3.
The APICS process industry definition already discriminates these two typesof process industries, stating “…Processes may be either continuous orbatch…”. We use the names and definitions provided by the APICS ProcessIndustry Thesaurus[5]. As mentioned above, batch/mix is defined as:
A process business which primarily schedules short production runs of products.
Process/flow is defined as:A manufacturer who produces with minimal interruptions in any one production run orbetween production runs of products which exhibit process characteristics such as liquids,fibres, powders, gases.
The discriminating characteristics of each type are presented in Table II.In process/flow businesses, the lead time is mainly determined by the cycle
time, i.e. the time between two consecutive runs of the same product. The actualprocessing time per unit is very small, but due to the high change-over timesand the high production speed, the production orders are large. The number ofdifferent products is not only limited, but there is also relatively little varietybetween the products. Little variety, low product complexity and the smallnumber of production steps cause all products to have the same routing. Sincethe total market demand for the relatively small number of products is high,investments in specialized single-purpose equipment are economicallyjustifiable. The use of single-purpose equipment simplifies the determination ofavailable capacity: usually the installations are used continuously (round-theclock production). The added value in general is quite low. Since the productionspeed is very high, the material costs usually account for 60-70 per cent of thecost price. The characteristics of process/flow businesses are summarized in theleft-hand column of Table II.
In batch/mix businesses, on the other hand, the number of process steps islarger and the level of product complexity is higher[17]. In fine chemicals
Figure 3.One-dimensionalTypology for ProcessIndustries
Batch/mix Process/flow
Drugs
Specia
lity ch
emica
ls
Rubbe
r
Majo
r che
mica
ls
Paper
Brewer
s
Steel
Oil
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
ProductionControl
Situations
53
production, for instance, sometimes more than ten different production stepscan be distinguished. Since the large variety of products requires the use of thesame – general type – of equipment, routings are much more complex. In somecases, even the process configurations are adapted: series of installations arerebuilt and reconnected to make a certain type of process possible (retrofitting).Consequently, lead times are longer and the work in process is higher;intermediate storage is more common than in process/flow businesses.Additionally, it is very difficult to make a good estimate of the availablecapacity. Lot sizes are predominantly determined by the technical batch sizerequirement instead of the changeover times. As a result of the increasedproduct complexity compared to process/flow businesses, the share of rawmaterials in the cost price is lower than in process/flow businesses and theadded value is higher. The characteristics of batch/mix businesses aresummarized in the right-hand column of Table II.
The production control structure to be used in process industries isdependent upon the position of the business on the axis in Figure 3. In the nextsection, we will discuss the typical production control aspects for each of theextremes on the axis.
Production and Inventory ControlIn the planning, scheduling and control literature, an explicit distinctionbetween the process/flow environments and the batch/mix production systemshas not been made. The concepts and approaches offered, however, each focuson one of the two extremes. In this section, we will classify the planning,scheduling and control literature which is relevant to process industries. Wewill first discuss the process/flow businesses, and then the batch/mixindustries.
Table II.Characteristics of
Process/Flow versusBatch/Mix Businesses
Process/flow businesses are Batch/mix businesses arecharacterized by characterized by
● High production speed, short ● Long lead time, much work inthroughput time process
● Clear determination of capacity, ● Capacity is not well-defined (differentone routing for all products, configurations, complex routings)no volume flexibility
● More complex products● Low product complexity
● High added value● Low added value
● Less impact of changeover times● Strong impact of changeover times
● Large number of production/● Small number of production steps process steps
● Limited number of products ● Large number of products
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
IJOPM14,12
54
Production and Inventory Control in Process/Flow BusinessesThe research applicable to process/flow industries, may be classified into thefollowing categories:
● general production control concepts and structures;
● scheduling approaches and heuristics;
● integrated production control and scheduling approaches.
The APICS process industry work groups have focused on the development ofa general production control concept. A production control concept is thedescription of and relations between all decision functions regarding themanagement of materials flow and capacity resources. The APICS processindustries “planning system framework” is presented in Bolander et al.[18]. Theframework strongly resembles the MRP II framework with a more dominantposition for the resource requirements planning and production schedulingfunctions. The framework does not present an integrated approach as far astechniques go, but it is assumed that each decision function can be equippedwith readily available or newly developed techniques. The interaction betweenthe different techniques is established using a detailed flow of informationbetween the various decision functions.
Scheduling approaches have been developed around the single machinemultiproduct lot-sizing and scheduling problem. A vast body of literature haspaid attention to this problem, especially the deterministic problem (EconomicLot Scheduling Problem) (ELSP). An excellent overview of the ELSP ispresented by Elmaghraby[19]. Later, the problem with stochastic demand hasbeen analysed. The first researchers to study this problem in detail wereLeachman and Gascon[20]. In an original paper they investigate theapplicability of deterministic models in stochastic situations, and present aheuristic to deal better with the uncertainty.
The well-known Massachusetts Institute of Technology hierarchicalproduction planning systems, integrating a control concept and detailedscheduling decisions[21], have been applied in process industries and single-stage systems as well[22]. Hax and Meal use the aggregation of products tofamilies, and from families to types, to make more aggregate decisions on alonger-term horizon. In this way, the planning and scheduling is more detailedif the horizon is shorter, and more aggregate if the horizon is longer. Since theirapproach is general and not restricted to process industries, they do not discussissues like high change-over times, as a dominant control parameter.
This dominance of the long cycle times as an important parameter is theprinciple of the conceptual aggregation model developed at EindhovenUniversity of Technology[23]. This approach has been worked out in moredetail by Fransoo[24,25]. In this model, the cycle times are not only determinedby considering cost, but also by considering capacity consequences. This leadsto a two-tiered model, in which at the higher level the cycle times aredetermined, and at the lower level the actual on-line scheduling takes place.
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
ProductionControl
Situations
55
Production and Inventory Control in Batch/Mix BusinessesThe planning scheduling and control literature in the batch/mix area ofresearch is differently oriented. This is mainly caused by the fact that thisresearch has been developed in the chemical engineering area, while theprocess/flow production control literature has developed in the operationsresearch and management science area.
Since chemical engineers do not limit themselves to the operational planning,scheduling and control of the system, but also include process design andprocess control, these aspects are sometimes integrated into the planning andscheduling issues.
We can classify the literature according to the decision functions that areaddressed:
(1) design of the production system (grass roots);(2) redesign of the production system (retrofitting);(3) planning/scheduling of the production system (off-line);(4) control of the production system (on-line scheduling);(5) process control.
In this article, we focus on the third and fourth decision functions mentionedabove. An excellent overview of the state of the art in grass roots design andretrofitting can be found in Reklaitis[26]. In the batch/mix literature, a cleardistinction is made between off-line scheduling and online scheduling[27]. Off-line scheduling is the creation of a predetermined schedule, assumingdeterministic demand and production. Online scheduling is the continuousadaptation of the off-line schedule, reacting to changes in demand andproduction. This distinction is similar to the distinction in deterministic andstochastic scheduling rules in the single-machine multi-product problemsdiscussed above. However, the emphasis placed on the processing ofinformation is much higher in the batch/mix literature[26,28]. This is probablydue to the higher complexity as compared to process/flow businesses.
The exchange of information between different control levels, includingprocess control, is illustrated in Cott and Macchietto[29]. Also the integration ofoff-line scheduling and process design is discussed in the literature. Usually, inthe design process, some assumptions about demand are made, andsimultaneously to the design, an off-line schedule is created[30]. Sometimes, thepossibility to physically partially reorganize the plant equipment still exists.One could think of combining different reactors by pipes in various ways. If thislimited equipment design is combined to a scheduling problem, this is called aretrofitting problem[31].
From the above description, it is clear that in the batch/mix problem, there aremore degrees of freedom, and also that there are more interrelationshipsbetween production units and material flows. Both production flow complexityand material complexity are high. The high number of production steps, thepresence of intermediate storage, and the divergent materials flow enables
Dow
nloa
ded
by U
nive
rsity
of
Gro
ning
en A
t 12:
18 1
4 N
ovem
ber
2016
(PT
)
IJOPM14,12
56
postponing the scheduling decision until the latest possible moment, i.e. when adecision needs to be taken in which unit a specific batch is going to be produced.This approach is a flexible scheduling approach, and should be designed insuch a way that each decision leaves maximum flexibility to the followingscheduling decisions. This seems a promising avenue for further research anddeserves increased attention in Production and Operations Management (POM)research.
ConclusionsThe objective of this article was to present a reference model or typology forresearch in production planning and scheduling in process industries. Thetypology was supported by classifying illustratively a number of papersaccording to this typology. It was concluded that a clear distinction needs to bemade between the research in process/flow and in batch/mix businesses. Inprocess/flow industries, the concept of Leachman and Gascon[20] has proven tobe of considerable value for uncapacitated problems. The concept ofFransoo[25] could be used as a basic model for capacitated situations. Theseconcepts can be worked out in detail for company-specific situations. Especiallythe distinction between make-to-stock and make-to-order companies may leadto different varieties of the respective concepts. A concept for process/flowbusinesses in make-to-order situations using the same basic ideas asFransoo[25] can be found in Bertrand, et al.[23]. In batch/mix industries,detailed scheduling and design procedures have been developed by chemicalengineers. A more general framework for this situation, involving flexiblescheduling procedures within a decision framework, is however lacking andshould receive increased research attention. The excellent work done by thechemical engineering research community should however be incorporated inthis model.
References1. Kochalka, T.S., “MRP in a Process Industry – Why Wait?”, Production and Inventory
Management Journal, Vol. 19 No. 4, 1978, pp. 17-20.2. Duncan, R.M., “The By-product Bill of Material”, APICS Conference Proceedings, 1983,
pp. 288-92.3. Nelson, N.S., “MRP and Inventory and Production Control in Process Industries”,
Production and Inventory Management Journal, Vol. 24 No. 4, 1983, pp. 15-22.4. Taylor, S.G., Sewart, S.M. and Bolander, S.F., “Why the Process Industries Are Different”,
Production and Inventory Management Journal, Vol. 22 No. 4, 1981, pp. 9-24.5. Connor, S.J., Process Industry Thesaurus, American Production and Inventory Control
Society, Falls Church, VA, 1986.6. Wallace, T.F. (Ed.), APICS Dictionary, 5th ed., American Production and Inventory Control
Society, Falls Church, VA, 1984.7. Sepehri, M., Silver, E.A. and New, C., “A Heuristic for Multiple Lot Sizing for an Order
under Variable Yield”, IIE Transactions, Vol. 16 No. 1, 1986, pp. 63-9.8. Haglund, E.A., “Primary Metals – We’re Different”, Production and Inventory Management
Journal, Vol. 22 No. 3, 1981, pp. 21-6.9. Burt, J. and Kraemer, B., “Integrating Material and Product Control Systems into a Tightly
Regulated Processing Environment”, APICS Conference Proceedings, 1979, pp. 32-4.
10. Burt, J., “Master Production Scheduling Considerations for Process Industries”, APICSConference Proceedings, 1980, pp. 417-18.
11. May, N.P., “The Functions Needed for Process-oriented Systems”, APICS ConferenceProceedings, 1984, pp. 29-32.
12. Rice, J.W. and Norback, J.P., “Process Industries Production Planning Using Matrix DataStructures”, Production and Inventory Management Journal, Vol. 28 No. 2, 1987, pp. 15-23.
13. Rutten, W.G.M.M., “Hierarchical Mathematical Programming for Operational Planning ina Process Industry”, European Journal of Operational Research, Vol. 64 No. 3, 1993,pp. 363-9.
14. Cokins, G.M., “Control Systems for Process Industries”, Manufacturing Systems, Vol. 6No. 5, 1988, pp. 79-83.
15. Fransoo, J.C., Process Industries Logistics Characteristics , research reportTUE/BDK/KBS/89-01, Eindhoven University of Technology, 1989.
16. Appoo, P.M., “A Practical Method of Implementing the Batch BOM in the ProcessIndustries”, Production and Inventory Management Journal, Vol. 28 No. 2, 1987, pp. 79-84.
17. Rippin, D.W.T., “Batch Process Planning”, Chemical Engineering, Vol. 98 No. 5, 1991,pp. 100-7.
18. Bolander, S.F., Heard, R.C., Seward, S.M. and Taylor, S.G., Manufacturing Planning andControl in Process Industries, American Production and Inventory Control Society, FallsChurch, VA, 1981.
19. Elmaghraby, S.E., “The Economic Lot Scheduling Problem (ELSP): Review andExtensions”, Management Science, Vol. 24 No. 6, 1978, pp. 587-98.
20. Leachman, R.C. and Gascon, A., “A Heuristic Scheduling Policy for Multi-item, Single-machine Production Systems with Time-varying, Stochastic Demands”, ManagementScience, Vol. 34 No. 3, 1988, pp. 377-90.
21. Bitran, G.R. and Hax, A.C, “On the Design of Hierarchical Production Planning Systems”,Decision Sciences, Vol. 8 No. 1, 1977, pp. 28-55.
22. Bitran, G.R., Haas, E.A. and Hax, A.C., “Hierarchical Production Planning: A Single StageSystem”, Operations Research, Vol. 29 No. 4, 1981, pp. 717-43.
23. Bertrand, J.W.M., Wortmann, J.C. and Wijngaard, J., Production Control: A Structural andDesign Oriented Approach, Elsevier, Amsterdam, 1990.
24. Fransoo, J.C., “Demand Management and Production Control in Process Industries”,International Journal of Operations & Production Management, Vol. 12 No. 7/8, 1992,pp. 187-96.
25. Fransoo, J.C., Production Control and Demand Management in Capacitated Flow ProcessIndustries, PhD Thesis, Eindhoven University of Technology, 1993.
26. Reklaitis, G.V., “Progress and Issues in Computer-aided Batch Process Design”, in Siirola,J.J., Grossman, I.E. and Stephanopoulos, G. (Eds), Foundations of Computer-aided ProcessDesign, Elsevier, Amsterdam, 1990, pp. 241-75.
27. Cott, B.J. and Machietto, S., “A General Completion Time Determination Algorithm UsingOn-line Schedule Modification”, Computers and Chemical Engineering, Vol. 13 Nos. 1/2,1989, pp. 105-13.
28. Jänicke, W., “Computergestützte Apparatebelegung für Mehrzweckanlagen”, Chemie-Ingenieur-Technik, Vol. 64 No. 4, 1992, pp. 368-70 (in German).
29. Cott, B.J. and Machietto, S., “An Integrated Approach to Computer-aided Operation ofBatch Chemical Plants”, Computers and Chemical Engineering, Vol. 13 Nos. 11/12, 1989,pp. 1263-71.
30. Mauderli, A. and Rippin, D.W.T., “Production Planning and Scheduling for Multi-purposeBatch Chemical Plants”, Computers and Chemical Engineering, Vol. 3, 1979, pp. 199-206.
31. Espuna, A. and Puigjaner, L., “On the Solution of the Retrofitting Problem for MultiproductBatch/Semicontinuous Chemical Plants”, Computers and Chemical Engineering, Vol. 13Nos. 4/5, 1989, pp. 483-90.
1. Sayeh Noroozi, Joakim WiknerS&OP Related Key Performance Measures with Integration of Sustainability: A DecouplingPoint Based and Modularized View on Supply Chains 197-233. [CrossRef]
2. Joakim Wikner, Sayeh Noroozi. 2016. A modularised typology for flow design based on decoupling points – a holistic viewon process industries and discrete manufacturing industries. Production Planning & Control 27:16, 1344-1355. [CrossRef]
3. Peter Samuelsson Department of Material Sciences, Royal Institute of Technology, KTH, Stockholm, Sweden Per StormDepartment of Material Sciences, Royal Institute of Technology, KTH, Stockholm, Sweden Thomas Lager School ofIndustrial Management, EMINES, Université Mohammed VI Polytechnique, Ben Guerir, Morocco . 2016. Profilingcompany-generic production capabilities in the process industries and strategic implications. Journal of ManufacturingTechnology Management 27:5, 662-691. [Abstract] [Full Text] [PDF]
4. Sayeh Noroozi, Joakim Wikner. 2016. A modularized framework for sales and operations planning with focus on processindustries. Production & Manufacturing Research 4:1, 65-89. [CrossRef]
5. Zhen He, Panpan Zhou, Min Zhang, Thong Ngee Goh. 2015. A Review of Analysis of Dynamic Response in Design ofExperiments. Quality and Reliability Engineering International 31:4, 535-542. [CrossRef]
6. Avinash Panwar, Bimal P. Nepal, Rakesh Jain, Ajay Pal Singh Rathore. 2015. On the adoption of lean manufacturing principlesin process industries. Production Planning & Control 26:7, 564-587. [CrossRef]
7. Avinash Panwar Department of Mechanical Engineering, Government R C Khaitan Polytechnic College, Jaipur, IndiaRakesh Jain Department of Mechanical Engineering, Malviya National Institute of Technology (MNIT), Jaipur, India A.P.S.Rathore Department of Management Studies, Malviya National Institute of Technology (MNIT), Jaipur, Indi . 2015. Leanimplementation in Indian process industries – some empirical evidence. Journal of Manufacturing Technology Management26:1, 131-160. [Abstract] [Full Text] [PDF]
8. Dennis Stindt, Ramin Sahamie. 2014. Review of research on closed loop supply chain management in the process industry.Flexible Services and Manufacturing Journal 26:1-2, 268-293. [CrossRef]
9. Eldon A. Gunn, Corinne A. MacDonald, Andrea Friars, Glen Caissie. 2014. Scotsburn Dairy Group Uses a HierarchicalProduction Scheduling and Inventory Management System to Control Its Ice Cream Production. Interfaces 44:3, 253-268.[CrossRef]
10. Tim Van Kampen, Dirk Pieter Van Donk. 2014. Coping with product variety in the food processing industry: the effect ofform postponement. International Journal of Production Research 52:2, 353-367. [CrossRef]
11. Andrew Charles Lyons, Keith Vidamour, Rakesh Jain, Michael Sutherland. 2013. Developing an understanding of leanthinking in process industries. Production Planning & Control 24:6, 475-494. [CrossRef]
12. A.J.J. Braaksma, W. Klingenberg, J. Veldman. 2013. Failure mode and effect analysis in asset maintenance: a multiple casestudy in the process industry. International Journal of Production Research 51:4, 1055-1071. [CrossRef]
13. Tim J. van KampenDepartment of Operations, University of Groningen, Groningen, The Netherlands RenzoAkkermanDepartment of Management Engineering, Technical University of Denmark, Kgs Lyngby (Copenhagen), DenmarkDirk Pieter van DonkDepartment of Operations, University of Groningen, Groningen, The Netherlands. 2012. SKUclassification: a literature review and conceptual framework. International Journal of Operations & Production Management32:7, 850-876. [Abstract] [Full Text] [PDF]
14. Stephan M. Wagner, Nikrouz Neshat. 2012. A comparison of supply chain vulnerability indices for different categories offirms. International Journal of Production Research 50:11, 2877-2891. [CrossRef]
15. Javid KoochakiDepartment of Business and ICT, Faculty of Economics and Business, University of Groningen, Groningen,The Netherlands Jos BokhorstDepartment of Operations, Faculty of Economics and Business, University of Groningen,Groningen, The Netherlands Hans WortmannDepartment of Business and ICT, Faculty of Economics and Business,University of Groningen, Groningen, The Netherlands Warse KlingenbergDepartment of Operations, Faculty of Economicsand Business, University of Groningen, Groningen, The Netherlands. 2011. Evaluating condition based maintenanceeffectiveness for two processes in series. Journal of Quality in Maintenance Engineering 17:4, 398-414. [Abstract] [Full Text][PDF]
16. Björn Kvarnström, Bjarne Bergquist, Kerstin Vännman. 2011. RFID to Improve Traceability in Continuous Granular Flows—An Experimental Case Study. Quality Engineering 23:4, 343-357. [CrossRef]
17. A.J.J. (Jan) Braaksma, W. (Warse) Klingenberg, P.W.H.M. (Paul) van Exel. 2011. A review of the use of asset informationstandards for collaboration in the process industry. Computers in Industry 62:3, 337-350. [CrossRef]
18. Dong-Qing Yao. 2011. Application of the Theory of Constraints (TOC) to Batch Scheduling in Process Industry.International Journal of Applied Industrial Engineering 1:1, 10-22. [CrossRef]
19. Helmut Haberleitner, Herbert Meyr, Alfred Taudes. 2010. Implementation of a demand planning system using advance orderinformation. International Journal of Production Economics 128:2, 518-526. [CrossRef]
20. Aiying Rong, Martin Grunow. 2010. A methodology for controlling dispersion in food production and distribution. ORSpectrum 32:4, 957-978. [CrossRef]
21. Onur A. Kilic, Dirk Pieter van Donk, Jacob Wijngaard, S. Armagan Tarim. 2010. Order acceptance in food processing systemswith random raw material requirements. OR Spectrum 32:4, 905-925. [CrossRef]
22. R. Andriansyah, T. Van Woensel, F.R.B. Cruz, L. Duczmal. 2010. Performance optimization of open zero-buffer multi-serverqueueing networks. Computers & Operations Research 37:8, 1472-1487. [CrossRef]
23. Erik Vanhatalo. 2010. Multivariate process monitoring of an experimental blast furnace. Quality and Reliability EngineeringInternational 26:5, 495-508. [CrossRef]
24. Renzo Akkerman, Dirk van der Meer, Dirk Pieter van Donk. 2010. Make to stock and mix to order: choosing intermediateproducts in the food-processing industry. International Journal of Production Research 48:12, 3475-3492. [CrossRef]
25. T. Van Woensel, R. Andriansyah, F. R. B. Cruz, J. MacGregor Smith, L. Kerbache. 2010. Buffer and server allocation ingeneral multi-server queueing networks. International Transactions in Operational Research 17:2, 257-286. [CrossRef]
26. Björn KvarnströmLuleå University of Technology, Luleå, Sweden Erik VanhataloLuleå University of Technology, Luleå,Sweden. 2009. Using RFID to improve traceability in process industry. Journal of Manufacturing Technology Management21:1, 139-154. [Abstract] [Full Text] [PDF]
27. Renzo Akkerman, Dirk Pieter van Donk. 2009. Product mix variability with correlated demand in two-stage foodmanufacturing with intermediate storage. International Journal of Production Economics 121:2, 313-322. [CrossRef]
28. Renzo Akkerman, Dirk Pieter van Donk. 2009. Analyzing scheduling in the food-processing industry: structure and tasks.Cognition, Technology & Work 11:3, 215-226. [CrossRef]
29. Erik Vanhatalo, Kerstin Vännman. 2008. Using factorial design and multivariate analysis when experimenting in a continuousprocess. Quality and Reliability Engineering International 24:8, 983-995. [CrossRef]
30. James A. Rappold, Keenan D. Yoho. 2008. A model for level-loading production in the process industries when demand isstochastic. Production Planning & Control 19:7, 686-701. [CrossRef]
31. Björn Kvarnström, Pejman Oghazi. 2008. Methods for traceability in continuous processes–Experience from an iron orerefinement process. Minerals Engineering 21:10, 720-730. [CrossRef]
32. Alex HillOperations Research Group, Kingston Business School, Kingston University, Kingston upon Thames, UK SteveBrownSchool of Business and Economics, University of Exeter, Exeter, UK. 2007. Strategic profiling. International Journalof Operations & Production Management 27:12, 1333-1361. [Abstract] [Full Text] [PDF]
33. Renzo Akkerman, Dirk Pieter van Donk. 2007. Product prioritization in a two-stage food production system with intermediatestorage. International Journal of Production Economics 108:1-2, 43-53. [CrossRef]
34. F. David, H. Pierreval, C. Caux. 2006. Advanced planning and scheduling systems in aluminium conversion industry.International Journal of Computer Integrated Manufacturing 19:7, 705-715. [CrossRef]
35. Dirk Pieter Van Donk, Jan C. Fransoo. 2006. Operations management research in process industries. Journal of OperationsManagement 24:3, 211-214. [CrossRef]
36. Mikko Ketokivi, Mikko Jokinen. 2006. Strategy, uncertainty and the focused factory in international process manufacturing.Journal of Operations Management 24:3, 250-270. [CrossRef]
37. Wout Van Wezel, Dirk Pieter Van Donk, Gerard Gaalman. 2006. The planning flexibility bottleneck in food processingindustries. Journal of Operations Management 24:3, 287-300. [CrossRef]
38. Bradley Z. HullDepartment of Management Marketing and Logistics, John Carroll University, University Heights, Ohio,USA. 2005. Are supply (driven) chains forgotten?. The International Journal of Logistics Management 16:2, 218-236. [Abstract][Full Text] [PDF]
39. F. David, H. Pierreval *, C. Caux. 2005. Enterprise resource planning systems in the aluminium conversion industry. ProductionPlanning & Control 16:8, 785-795. [CrossRef]
40. Simme Douwe P. Flapper, Jan C. Fransoo, Rob A. C. M. Broekmeulen, Karl Inderfurth. 2002. Planning and control of reworkin the process industries: A review. Production Planning & Control 13:1, 26-34. [CrossRef]
41. Wenny H.M. Raaymakers, J Will M. Bertrand, Jan C. Fransoo. 2001. Makespan estimation in batch process industries usingaggregate resource and job set characteristics. International Journal of Production Economics 70:2, 145-161. [CrossRef]
42. Dirk Pieter van Donk. 2001. Make to stock or make to order: The decoupling point in the food processing industries.International Journal of Production Economics 69:3, 297-306. [CrossRef]
43. Dirk Pieter van DonkFaculty of Management & Organisation, University of Groningen, Groningen, The Netherlands. 2000.Customer‐driven manufacturing in the food processing industry. British Food Journal 102:10, 739-747. [Abstract] [Full Text][PDF]
44. Daina R Dennis, Jack R Meredith. 2000. An analysis of process industry production and inventory management systems.Journal of Operations Management 18:6, 683-699. [CrossRef]
45. Daina Dennis, Jack Meredith. 2000. An Empirical Analysis of Process Industry Transformation Systems. Management Science46:8, 1085-1099. [CrossRef]
46. J.W.M Bertrand, W.G.M.M Rutten. 1999. Evaluation of three production planning procedures for the use of recipe flexibility.European Journal of Operational Research 115:1, 179-194. [CrossRef]
47. W.G.M.M. Rutten, J.W.M. Bertrand. 1998. Balancing stocks, flexible recipe costs and high service level requirements in abatch process industry: A study of a small scale model. European Journal of Operational Research 110:3, 626-642. [CrossRef]
Journal of Operations Management 24 (2006) 250–270
Strategy, uncertainty and the focused factory in
international process manufacturing
Mikko Ketokivi a,*, Mikko Jokinen b
a Department of Industrial Engineering and Management, Helsinki University of Technology, P.O. Box 5500, FI-02015 Hut, Finlandb Consolidated Metals Corporation (pseudonym), Executive School of Industrial Management, Helsinki University of Technology,
P.O. Box 5500, FI-02015 Hut, Finland
Received 11 September 2003; received in revised form 3 July 2004; accepted 15 July 2004
Available online 16 February 2005
Abstract
The extant literature on the focused factory has not explored the contingencies associated with the de facto adoption and use
of focused factory principles: Why are some plants focused while others are not? Is focus—or unfocus—a strategic choice, best
practice or perhaps a reflection of an environmental constraint? In his pioneering work, Skinner [W. Skinner, 1974. The focused
factory. Harvard Business Review 52 (3), 113–121] prescribes companies to ensure that the manufacturing task of their
manufacturing units is simple and focused, for instance, by assigning a narrow product mix for each factory or concentrating on a
narrow mix of production technologies. Especially in the absence of compelling empirical evidence on the effectiveness of the
focused factory approach, we argue that we still do not understand why some plants may remain unfocused.
We observe that in the international process industry case examined in this paper, some factories are unfocused and their
manufacturing tasks are all but simple. Yet, some of them appear to be high performers. This presents an opportunity to seek
empirical insight on the questions raised above. Specifically, we examine why manufacturing companies in the process
industries may or may not follow the focused factory strategy. Our results suggest that in certain operating environments and
with certain competitive strategies, choosing not to focus the manufacturing task should be viewed as a viable alternative
manufacturing strategy, perhaps even a constraint imposed by the operating environment. We develop four contingency
propositions to explain why focused manufacturing strategy may not be desirable or even possible.
# 2005 Elsevier B.V. All rights reserved.
Keywords: Manufacturing strategy; International manufacturing; Focused factory; Case study; Process industry
1. Introduction
The contention in this paper is that even though
we have talked about the focused factory for 35 years
tingency and institutional explanations in the adoption of inno-
vative manufacturing practices. Journal of Operations
Management 22 (1), 63–89.
Lawrence, P.R., Lorsch, J.W., 1967. Organization and Environment:
Managing Differentiation and Integration. Harvard University,
Boston, MA.
Meredith, J.R., 1998. Building operations management theory
through case and field research. Journal of Operations Manage-
ment 16 (4), 441–454.
New, C.C., Szwejczewski, M., 1995. Performance measurement and
the focused factory: empirical evidence. International Journal of
Operations and Production Management 15 (4), 63–79.
Panzar, J.C., Willig, R.D., 1981. Economies of scope. American
Economic Review 71 (2), 268–272.
Pesch, M.J., 1996. Defining and understanding the focused factory: a
Delphi survey. Production and Inventory Management Journal
38 (2), 32–36.
Pesch, M.J., Schroeder, R.G., 1996. Measuring factory focus: an
empirical study. Production and Operations Management 5 (3),
234–254.
Porter, M.E., 1980. Competitive Strategy: Techniques for Analyzing
Industries and Competitors. Free Press, New York.
Priem, R.L., Butler, J.E., 2001. Is the resource-based ‘‘view’’ a
useful perspective for strategic management research? Acad-
emy of Management Review 26 (1), 22–40.
Rice, J.W., Norback, J.P., 1987. Process industries production
planning using matrix data structures. Production and Inventory
Management Journal 28 (2), 15–23.
Safizadeh, M.H., Ritzman, L.P., Sharma, D., Wood, C., 1996. An
empirical analysis of the product-process matrix. Management
Science 42 (11), 1576–1591.
Schmenner, R.W., 1983. Every factory has a life cycle. Harvard
Business Review 61 (2), 121–129.
Sharman, G., 1984. The rediscovery of logistics. Harvard Business
Review 62 (5), 71–79.
Skinner, W., 1969. Manufacturing—missing link in corporate strat-
egy. Harvard Business Review 50 (3), 136–145.
Skinner, W., 1974. The focused factory. Harvard Business Review
52 (3), 113–121.
Skinner, W., 1996. Manufacturing strategy on the ‘‘S’’ curve.
Production and Operations Management 5 (1), 3–14.
Stinchcombe, A., 1965. Social structure and organization. In:
March, J. (Ed.), Handbook of Organizations. Rand McNally,
Chicago, IL, pp. 142–193.
Taylor, S.G., Sewart, S.M., Bolander, S.F., 1981. Why the process
industries are different. Production and Inventory Management
Journal 22 (4), 9–24.
Teece, D.J., 1982. Towards an economic theory of the multiproduct
firm. Journal of Economic Behavior and Organization 3 (1), 39–
63.
van der Vorst, J.G.A.J., van Dijk, S.J., Beulens, A.J.M., 2001. Supply
chain design in the food industry. International Journal of
Logistics Management 12 (2), 73–85.
van Donk, D.P., van der Vaart, D.A., 2005. A case of shared
resources, uncertainty and supply chain integration in the pro-
cess industry. International Journal of Production Economics
96 (1), 97–108.
Vokurka, R.J., Davis, R.A., 2000. Focused factories: empirical study
of structural and performance differences. Production and Inven-
tory Management Journal 41 (1), 44–55.
Vollmann, T.E., 1996. The Transformation Imperative. Harvard
Business School Press, Boston, MA.
Wheelwright, S.C., Bowen, H.K., 1996. The challenge of manu-
facturing advantage. Production and Operations Management
5 (1), 59–77.
Whetten, D.A., 1989. What constitutes a theoretical contribution?
Academy of Management Review 14 (4), 490–495.
Williamson, O.E., 1985. The Economic Institutions of Capitalism.
Free Press, New York.
Winter, S.G., Szulanski, G., 2001. Replication as strategy. Organi-
zation Science 12 (6), 730–743.
Woodward, J., 1994 [1965]. Industrial Organization: Theory and
Practice, second ed. Oxford University Press, London.
Yin, R.K., 1989 [1984]. Case Study Research: Design and Methods.
Sage Publications, Thousand Oaks, CA (Revised ed.).
Zaidman, B., 1994. Market vulnerability in the process industries.
International Journal of Production Economics 34 (1), 57–
64.
International Journal of Production ResearchVol. 48, No. 12, 15 June 2010, 3475–3492
Make to stock and mix to order: choosing intermediate
products in the food-processing industry
Renzo Akkermana*, Dirk van der Meerb and Dirk Pieter van Donkb
aDepartment of Management Engineering, Technical University of Denmark, Produktionstorvet424, 2800 Kgs. Lyngby (Copenhagen), Denmark; bFaculty of Economics and Business,
University of Groningen, P.O. Box 800, 9700 AV Groningen, The Netherlands
(Received 29 September 2008; final version received 21 January 2009)
In contrast to discrete manufacturers, food-processing companies can sometimesproduce the same end products in different ways: either mix first and thenprocess, or process first and mix later. Moreover, a final product can be mixedfrom different raw materials or intermediates. That adds a new dimension topostponement and decoupling point theory as choices need to be made notonly with regard to where to locate inventory, but also which products to store.That aspect has not been covered so far. This paper explores this problem fora typical two-stage food production situation in a flour mill. The number andcomposition of intermediate products in the decoupling point is determined usinga stepwise solution approach supported by mathematical programming models.The procedure facilitates decision-making for the management of the millregarding how many and what intermediates to store. Extensions of the modelspresented might be helpful to solve related problems such as determiningthe number of intermediate storage tanks required.
Keywords: food industry; decoupling point; postponement; intermediate storage;case study
1. Introduction
In the food processing industry, a typical production plant produces a multitude ofintermediate products in an even wider range of packages. Often, the output variety isbased on a relatively small number of (agricultural) raw materials. In general, sucha divergent product structure is typical for the process industries (Fransoo and Rutten1994). Differences in products can be associated with customer-specific products, either inpackaging form, size, or print, labelling, or (more fundamental) product recipe. Recipesnormally differ with respect to a bundle of product quality attributes. Often a recipe can becharacterised in terms of minimal requirements for each attribute. Different options mightexist in relation to where in the process a recipe is made specific: either adding specificingredients early or late in the process.
A number of factors and trade-offs determine at what stage a product will becomecustomer-specific. Generally, product specifications are assumed to be known for eachstage of the production process. In other words, the final product determinesthe requirements and specification of all intermediates univocally. However, in
food-processing companies – specifically those that mix and blend – recipes of final
products can sometimes be manufactured in different ways. Two extreme possibilities are
either starting with a customer-specific recipe and process the resulting relatively small
batches and store final product waiting for orders, or process raw materials in largebatches, store these and mix them to recipe as the last step as customer orders arrive.
It might be clear that there is an endless number of variants in between in terms of the
amount and nature of possible intermediate recipes that can be stored, and deciding how
to organise this is an essential factor in determining the operational performance of
many mix or blend production systems. Finally, given the nature of recipes, one might
decide to deliver products having a too high quality for one or two requirements.In general, each of the alternatives has implications for the processing costs, but,
additionally, it might also affect the costs of the raw materials used. This problem has
some similarities with binning, where quality grades are determined and subsequently
downgrade possibilities are used to satisfy demand with higher quality levels to save
setups (see Lyon et al. 2001). However, their problem is operational (how to satisfyorders), whereas we focus on a tactical decision (what are the recipes for a certain period
of time). Moreover, in discrete manufacturing far less possibilities exist for the product
structure (recipe).So far, the literature has not addressed the problem of jointly determining (i) when to
specify a product to customer specifications, (ii) which recipes to use for flexible
products, i.e. intermediate products that can still be used to mix various customerorders, and (iii) how many intermediate products to use. This paper develops an
approach to decide on the number and composition of intermediate products in two-
stage food production systems, supported by quantitative modelling. More specifically,
we consider the case of a flour mill, facing the problem of how much and which
intermediate recipes to store and to determine the consequences of different options.
These consequences range from operational issues such as the costs of ingredients tomore tactical and strategic decisions such as the required number of intermediate silos.
More generally, the aim of this paper is also to contribute to the understanding of
product specification in the food industry if end products can be made to stock, mixed
to order, or made to order.The paper is structured as follows. In the next section, the theoretical background of
the studied problem is discussed. Then, the case study is introduced, also outlining thequality parameters of the products in question. The following section then presents the
solution procedure for determining the number and composition of the intermediate
products. Subsequently, the results of the application of the model in the case study are
discussed. Finally, the last section will present our conclusions and a discussion on the
paper’s contribution and the future research opportunities.
2. Theoretical background
Resulting from increasingly powerful retailers (Dobson et al. 2001), food manufacturers
find themselves in a situation where the downstream supply chain requires more and more
flexibility, whereas the characteristics of the production systems involved often do notsupport this (Van Donk et al. 2008). To improve their flexibility, while still trying to
produce in efficient volumes, food manufacturers often try to postpone the diverging
of their product mix as long as possible (e.g. Van Donk 2001, Soman et al. 2004).
3476 R. Akkerman et al.
This late specification allows for shorter lead times, while keeping production efficiencybefore the specification on an acceptable level. Early product specification would lead toan early diverging of the number of different products and would result in increasedrequirement of storage facilities (as each product requires its own tank or silo), andnumerous additional changeovers and/or cleaning activities.
This concept of ‘postponement’ (or delayed differentiation) has been extensivelystudied in the literature. In manufacturing situations, postponement aims to retainproducts in a neutral and non-committed status as long as possible (Yang et al. 2004b).An overview of the research on postponement can be found in Van Hoek (2001), whichwas updated and extended by Boone et al. (2007). They propose several new challengesfor further research. Among others, they conclude that the application of postponementwas not as widespread as was expected based on the attention in the literature. Recently,Forza et al. (2008) developed a typology that identifies three types of postponement.They stress that it is important to clarify which type is studied, to be able to carefullyassess the resulting operational performance. The first type they identify deals withpostponement of product specification from the forecast-driven production stages to theorder-driven production stages. In our work, when deciding on when to specifyproducts, and what flexible recipes to use to mix final products, this is the kind ofsituation we are dealing with. However, the specific role of recipes in the productstructure encountered in most process industries is not explicitly considered by Forzaet al. (2008).
Next to his 2001 review, Van Hoek also wrote specifically about postponement inthe food industry (Van Hoek 1999). Here, he noted that, in comparison with otherindustries, the application of postponement in the food industry is fairly low. Food-specific characteristics, e.g. perishability and short lead times, limit the applicabilityof postponement. Van Hoek, however, sees a strong focus on product standardisationin the food industry, as this is the only way to increase capacity utilisation of capitalintensive production technology. The problem here is that standardisation is oftendifficult to achieve, as customers nowadays demand more and more product variety. Forthis reason, Abukhader and Jonson (2007) encourage food companies to analyse theirproduct mix and product development process in detail, to see where postponement canbe applied.
However, the divergence in the product mix does not have to happen at the firstproduction stage, but can often be placed later in the production process. Here, theconcept of postponement is strongly connected to the decoupling point concept (Hoekstraand Romme 1992, Olhager 2003). Van Donk (2001) elaborates on this concept for thefood-processing industry. One of the shortcomings he mentions is its qualitative natureand its focus on individual products as opposed to the entire product portfolio. Anotherissue is that the decoupling point concept assumes that specifications of products andintermediate products are known and fixed. However, as explained earlier, that is often notthe case in process industries and food processing.
A number of (process) industries face the situation where once it is decided to locatethe decoupling point at the intermediate product level, the set of intermediate products stillhas to be determined. That generally applies to the specification and/or the number ofintermediates. For instance, in steel mills, an extensive range of slab lengths is demandedby customers. To reduce inventory costs (and often also to increase processing efficiency),a limited number of those slab lengths are held in storage, and are cut to customerspecification when final orders arrive. Next to reducing inventory costs, this introduces
International Journal of Production Research 3477
waste costs due to cutting losses. In the literature we find several approaches dealing with
this specific case. Recent examples are the work by Caux et al. (2006), who develop
a mathematical model to evaluate the cost trade-off and the graphical approach presented
by Kerkkanen (2007). A similar example can be found in the cardboard industry, where
a limited number of large sheet sizes are subsequently cut into a wide variety of small
sheets (Wanders et al. 2004).In the food industry, determining the set of intermediate products mostly deals with
products that are to be used in combinations to comply with the requirements on the final
product (e.g. Rajaram et al. 1999). Once the intermediates are determined, a certain stock
level will be maintained from which the intermediates are blended to order to create the
final wide range of products. However, in theory, almost any possible set of intermediates
could be used, but intermediate storage possibilities are often limited (Akkerman et al.
2007), which causes a very complex decision problem. The increasing importance of food
safety and food quality (e.g. Griffith 2006) even adds to this complexity, due to additional
constraints.The composition (or recipe) of the intermediate products has to be derived from the
required specifications of the final products and the given attributes or qualities of the
raw materials. As the composition of raw materials is often variable for food products,
the composition of the intermediate products either has to be robust to these changes or
should be changed on a regular basis. In related work by Rutten and Bertrand (1998) the
concept of recipe flexibility is introduced to deal with (i) variations in raw material
quality, (ii) cost minimisation by raw material selection, and (iii) substitutions of raw
materials in recipes due to unavailability (see also Rutten 1995). In Rutten and Bertrand
(1998), only a single mixing stage is present, and determining recipes is an operational
problem that can be solved for each production order. This also holds for the case
of chemical fertilisers studied by Ashayeri et al. (1994). Similarly, Lyon et al. (2001)
present an operational method in discrete manufacturing that matches different quality
grades to orders by using higher qualities than required to prevent set-ups. However,
in two-stage production systems, that would imply that both processes can produce
order-driven, which is often not the case and certainly not in the case of flour milling
we present here.In this paper, specification (and hence order-driven production) starts at the
intermediate storage stage, and the aim is to design several intermediate products that
can be used in a mix-to-order strategy to fulfil future demand. This means the
determination of recipes takes place not on an operational level, but on a strategic or
tactical level, and is only performed once in a while to achieve cost minimisation.
To support this, we develop a stepwise approach supported by mathematical program-
ming models, which takes a wide variety of typical food-related quality aspects into
account. The resulting model can be used to design recipes for intermediate products,
which can be used in a cost-efficient mix-to-order strategy. We feel that the approach
presented in this paper will support the decision-making on postponement and the
decoupling point in practice, taking into account several operational characteristics that
are typical for the food industry, and addressing the inherent trade-offs between product
flexibility and material costs. Furthermore, it contributes to the theory on postponement
and the decoupling point, by providing a structured approach based on quantitative
modelling, balancing several cost factors and including the whole product range in the
decision making process.
3478 R. Akkerman et al.
3. Case description
We study a medium-sized flour manufacturer, supplying flour products to bakeries and
industrial manufacturers. Figure 1 shows an outline of the production process. Grains and
additional ingredients are pre-processed, blended, and milled to obtain a selection of
intermediate products, which can already be blends of various grains. These products are
then mixed into a wide range of flour products, transported in bulk to industrial customers
and large bakeries, or packaged in bags for smaller traditional bakeries.
3.1 Problem description
Although flour mills seem to be relatively simple processes, the above-sketched situation
fully applies: a small amount of raw materials (around 10 types of grain) and an increasing
amount of partly customer-specific end products (currently around 50). The two extreme
options are either to mix first and then mill, or to mill all raw grains separately and mix
them to recipe. Both these extremes are not reasonable due to the fact that it would either
lead to a few intermediates, which would always have to be mixed, or many intermediates
which would already be customer-specific. The first option would lead to a lot of mixing
operations, whereas the latter situation would require a huge storage capacity. Taking this
into account, the challenge is to find a solution in-between these extremes. Therefore, it is
logical to store a limited number of (possibly blended) intermediate products and use these
to mix end products. In terms of the decoupling point, we can characterise this as mix to
order. It is worth stressing that some intermediates can be used directly as end products
without mixing and after mixing in other products. An important reason for the mix-
to-order strategy is the required short delivery time. Operating this system incurs all types
of operational problems. Here, we address a more tactical oriented design problem:
determining the amount and composition of intermediate products. On the one hand a low
number of intermediates will increase the efficiency and quality of the milling operation by
enabling milling in larger batches, as well as limiting the number of intermediate storage
silos. On the other hand, a low number has two downsides. First, to be able to mix final
products from a lower number of intermediate products, more expensive, high-quality raw
materials have to be used, leading to an increase in material costs. Secondly, to mix from
a smaller number the wide variety of end products, more mixing operations would be
necessary, increasing operational cost of mixing.Now, the main problem is to determine the number and composition of intermediate
products, increasing operational efficiency, product quality, and flexibility on the one
hand, while on the other hand keeping material costs and operational costs from
exploding. Currently, this design is based upon intuition, craftsmanship, and habit, which
are not supported by an evaluation of costs incurred. Currently, management envisages an
increase in volume, which will mainly be realised by adding customer specific recipes, but is
Figure 1. Outline of the production process of the flour mill.
International Journal of Production Research 3479
afraid that more intermediate products will be needed. The current project aims at
benchmarking current practice, developing a sound understanding of the trade-off
between different performance measures and costs, and to support future decisions with
respect to investments in silos and mixing capacity. The aim is to develop a model to
support this important design and selection decision.
3.2 Quality parameters
For the final products produced by the company, a variety of quality parameters are
important. These parameters also need to be considered in the design of intermediate
products. Wherever possible, we will not go into details on the specific details of the
chemical properties or the units in which they are measured, as this would only distract
from the problem discussed in this paper. There are nine quality parameters relevant in
determining the intermediate products (and final products). It concerns:
(1) Protein percentage.(2) Water absorption ability.(3) Dough extensibility, parameter 1, length of the extensometer curve.(4) Dough extensibility, parameter 2, height of the extensometer curve.(5) Deoxynivalenon (DON) level, a mycotoxin that can affect health when present at
high levels.(6) A product usability index for use in bread, to cover usage parameters that are
difficult to measure, assigned by the baker from the test bakery.(7) A second (similar) product usability index, this one for use in biscuits.(8) Falling number, which indicates the sprout damage of the grain used.(9) Bread volume, which is determined in a test bakery.
In creating different intermediate products or different end products, these parameters
have to be considered. Most of these parameters lead to simple linear relationships in
a blending process (i.e. weighted averages):
Qpk ¼Xi2I
fipQik 8k 2 1, . . . , 8f g, 8p 2 P, ð1Þ
where fip is the blending fraction of raw material i for intermediate product p, Qpk quality
parameter k for intermediate product p (or raw material i), I the set of raw materials, P the
set of intermediate products, and K the set of quality parameters involved. Only the last
(ninth) quality parameter, bread volume (Qp,9), acts differently, and is calculated as
follows:
Qp,9 ¼ 1000� lnXi2I
fipeQi,9=1000
!8p 2 P ð2Þ
When relating quality parameter Qp,9 to minimum and maximum values Qminp,9 and Qmax
p,9 ,
we would normally get a nonlinear constraint like:
Qminp,9 � 1000� ln
Xi2I
fipeQi,9=1000
!� Qmax
p,9 8p 2 P ð3Þ
3480 R. Akkerman et al.
For use in the MILP models in the remainder of the paper, the nonlinearity of this
restriction on fip is undesirable. However, such a constraint can be linearised in fip by
‘moving’ the nonlinearity into the quality parameters, as follows:
eQminp,9=1000�Xi2I
fipeQi,9=1000 � eQ
maxp,9=1000
8p 2 P ð4Þ
4. Solution approach
The main aim of the models presented in this paper is to support the current organisational
decision-making processes. The current process basically consists of three steps:
(1) Consider all possible intermediates.(2) Find the most economical way to make them from raw materials.(3) Select a limited number to be actually used, implicitly aiming at balancing material
and mixing costs.
As said, this selection process is done a few times a year. Here we aim at supporting the
second and third step and leave the first one to the experience of the people involved.
However, given the numerical support a wider set of intermediates can be considered.
Given the restrictions, we develop two mixed-integer linear programming (MILP) models.
The first one simply calculates the optimal composition of an intermediate, while the
second one chooses the best intermediate if we restrict the number to be used. This
stepwise approach is illustrated in Figure 2.First, a set of potential intermediate products is defined. Secondly, the optimal
composition of these potential intermediate products is determined using a first MILP
model. Third, and finally, a selection of the potential intermediate products is made to
compose the final products using a second MILP model.Next to the correspondence with the current decision-making process, this stepwise
approach also fits well with the decoupling point concept, as all three steps are centred on
the decoupling point. After defining potential products to be stored at the decoupling
point, one model aims to optimise the upstream process (blending and milling the
Figure 2. Schematic representation of the solution approach.
International Journal of Production Research 3481
intermediate products), while the second model aims to optimise the downstream process
(selecting and mixing the intermediate products).In the following paragraphs, the three stages will be further elaborated upon.
4.1 Defining potential intermediate products
Finding the right potential intermediate products is a difficult task. Even for relatively
small-scale problems, the number of combinations is enormous. Including nine different
quality parameters (as presented in Section 3) also means that care should be taken not to
design potentially infeasible recipes.What we learn from this is the need for careful consideration in the process of defining
or designing intermediate products. Given the importance and difficulty, this step is
executed by the experts from the recipe management and quality management department.
The process of designing intermediates was supported by available recipe information (on
raw materials, current intermediate products, and final products) from management
information systems to validate the obtained information.Finally, this resulted in the design of a set of over 70 possible intermediate products,
covering a wide range for the quality parameters. Next to newly designed ‘flexible’ recipes,
this set also includes recipes for final products (on customer specification), as this allows
the solution approach to store this product on the intermediate storage level, thereby
reducing further specification (i.e. mixing) costs. In this way, the approach can select
a combination of final recipes and flexible recipes. It is worth noting that both can be used
in mixing.Initially, the model was validated with the current set of intermediate products as
potential intermediate products. Allowing all of them to be used in the final solution (i.e.
not setting a maximum on the number of intermediate products), and calculating the
compositions and costs involved, the case company could get familiar with the model, and
grew more confident in its potential.
4.2 Optimal composition of the potential intermediate products
In this stage, the composition of the potential intermediate products is determined,
based on a fairly simple MILP model (labelled MILP 1) for each of the intermediates
p2P. The objective function minimises the raw material costs:
MinXi2I
fipci, ð5Þ
where ci is the unit cost for raw material i. The quality constraints for the intermediate
products are determined according to Equations (1) to (4)
Qminpk �
Xi2I
fipQik � Qmaxpk 8k 2 1, . . . , 8f g, ð6Þ
eQminp,9=1000�Xi2I
fipeQi,9=1000 � eQ
maxp,9=1000, ð7Þ
Xi2I
fip ¼ 1: ð8Þ
3482 R. Akkerman et al.
Constraint (8) deals with the material balance for the intermediate product.For modelling purposes, it was chosen to formulate the quality constraints inEquation (6) in a uniform way, always including a maximum and a minimum. Forsome of the constraints, we do however not have both. For instance, the DON level(parameter 5) only has a maximum, and the falling number (parameter 8) only hasa minimum (as higher falling numbers mean less enzyme activity which is related to lesssprout damage). In these cases, the minimum or maximum values are set to zeroor infinity, respectively.
For quality parameter 2, the water absorption ability, there is another constraintin addition to the one in Equation (6). If a certain intermediate product is already inuse, the water absorption ability cannot change too much. This leads to a constraintrelating Qp,2 to its current value Q0
p,2:
1�l
100
� �Q0
p,2 �Xi2I
fipQi,2 � 1þu
100
� �Q0
p,2, ð9Þ
where l and u are percentages that the product can decrease or increase in waterabsorption.
To be able to set constraints on the fractions fip, we introduce a set of indicatorvariables. These are necessary to distinguish between raw materials that are used and rawmaterials that are not used (only the first are constrained):
�ip ¼1, if raw material i is used in intermediate product p,
0, otherwise:
�ð10Þ
Now, we can define minimum ( fminip ) and maximum ( fmax
ip ) fractions in case a raw materialis used in the intermediate products:
�ipfminip � fip � �ipf
maxip 8i 2 I: ð11Þ
Finally, certain customers require that their product contains at least a certain percentage�p of a certain type of raw material (R� I), which we therefore also require from a subsetof the potential intermediate products PS�P:X
i2R�I
fip � �p: ð12Þ
Solving model MILP 1, as defined by Equations (5) to (12), for each of the potentialintermediate products will result in the optimal composition, in terms of minimal materialcosts.
4.3 Selection of intermediate products
In the third stage, the resulting compositions of potential intermediate products are used ina selection process to determine which intermediate products will eventually be used increation of the final products. Again, this results in an MILP model (MILP 2). This timethe costs to minimise consist of two parts, mixing operation costs and material costs:
MinXj2J
�jDjc� þXp2P
fpjcpDj
!, ð13Þ
International Journal of Production Research 3483
where c� the mixing cost per unit of final product, which is multiplied by the demand
for product j, Dj. Whether or not mixing is necessary for a certain product is included
through:
�j ¼1, if mixing is necessary for product j,
0, otherwise:
�ð14Þ
Material cost is based on fpj, the fractions of intermediate products p in final
product j, and cp is the material cost for intermediate product p, which are both outcomes
of MILP 1.Similar to MILP 1, we again include quality constraints for all final products, based
on the quality parameter values of the intermediate products Qpk, and the blending
fractions fpj.
Qminjk �
Xp2P
fpjQpk � Qmaxjk 8k 2 1, . . . , 8f g, 8j 2 J,
ð15Þ
eQminj,9=1000�Xp2P
fpjeQp,9=1000 � eQ
maxj,9=1000
8j 2 J, ð16Þ
Xp2P
fpj ¼ sj 8j 2 J:ð17Þ
As can be seen, the fractions do not sum to 1 in this model, but to a parameter sj which is
often still 1, but sometimes has a value smaller than 1 (typically between 0.85 and 1). This
is the case because some final products have additional ingredients that are added in the
final stages. This is left out of the model presented in this paper, because it does not affect
the final quality of the product.The additional constraint for the water absorption ability changes is similar to its
MILP 1 version:
1�l
100
� �Q0
j,2 �Xp2P
fpjQp,2 � 1þu
100
� �Q0
j,2 8j 2 J: ð18Þ
To make sure the customer requirement of at least a certain percentage of a certain
type of raw material (Is� I) is continued from the intermediate products, we use:
Xp2Ps�P
Xi2Is�I
fð1Þip
!fpj � �j 8j 2 Js � J: ð19Þ
For reasons of operational simplicity, the management of the company would like to
use a maximum number of intermediate products per final product, which can be
formulated as: Xp2P
�pj �M 8j 2 J,ð20Þ
where
�pj ¼1, if intermediate product p is used in product j,
0, otherwise,
�ð21Þ
3484 R. Akkerman et al.
and M is the maximum number of intermediate products used per final product. To makesure the binary variables �pj have the right values, we use the following constraint:
fpj � �pj 8p 2 P, 8j 2 J: ð22Þ
As the mixing stage has a limited capacity, the following constraint assures that we do notuse more than the existing mixing capacity L.X
j2J
�jDj � L:ð23Þ
Again, an additional constraint is introduced to set the binary variable used:Xp2P
�pj � 1 �M � �j 8j 2 J:ð24Þ
One of the most essential constraints in this model is the maximum number ofintermediate products we allow the model to choose. This is formulated as:X
p2P
�p � P�,ð25Þ
where the binary variable �p is defined as:
�p ¼1, if intermediate product p is used,
0, otherwise,
�ð26Þ
and this gets the correct value using the following constraint:Xj2J
fpj �M � �p 8p 2 P:ð27Þ
Furthermore, there are additional constraints due to the fact that some of theintermediates are not allowed to be used in some final products, due to certaincharacteristics. This leads to a distinction between products that are used to produce breadand products that are not. For this, we define a binary parameter as follows:
�pj ¼1, if intermediate p can be used in product j,
0, otherwise,
�ð28Þ
which leads to the following constraints:
fpj � �pj 8p 2 P, 8j 2 J: ð29Þ
Solving model MILP 2, as defined by Equations (13) to (29), will result in the selection ofa preset number of intermediate products. This stage can be repeated for different numbersof resulting intermediate products.
5. Results
Due to confidentiality requirements, the actual costs are not reported in this section.We do, however, report the cost differences between various scenarios for the numberof intermediate products by using indexed results. To give an indication of theperformance of the model, we can compare the solution with 15 intermediate
International Journal of Production Research 3485
products with the current situation at the company, which would lead to a cost decreaseof 0.6%. Although this is fairly limited in terms of the relative difference, theabsolute difference still makes the use of the model interesting, as food manufacturersoften work with fairly low profit margins. More importantly, the model hasproven valuable in tactical and strategic discussions, which we will come back to in themanagerial implications in Section 5.3.
The model was implemented in ILOG’s OPL software and solved with CPLEX.The calculation times for the results presented in this section range from a few seconds toa few minutes.
5.1 Number of intermediate products
The main aspect the company was interested in was an overview of how changing thenumber of intermediate products (P*) would affect costs, and how large the costincreases were that they expected to see with a decrease in the number of intermediateproducts (assuming that this would require more flexible, higher quality, and moreexpensive raw materials). This decrease was expected to simplify the operations and theefficiency of the milling process. We set the cost (including raw material cost andmixing cost) found for the solution with 15 intermediate products as 100, as this is thecurrent number of intermediate products. This presentation format facilitates theunderstanding of what happens to the costs when changing the number of intermediateproducts.
Figure 3 shows that decreasing the number of intermediate products below the currentnumber of 15 increases costs, in what seems to be an exponential curve. Based on theseresults, it seems that the number of intermediate products can be decreased down to 10while only marginally increasing costs. Reducing the number of intermediate productsfurther has a larger impact on the costs. It seems that more expensive intermediateproducts are necessary to provide the flexibility to mix the required end products. This isreflected in, for example, the set of intermediates used in the case of eight intermediatesthat is not totally part of the larger sets of intermediates. It should be noted that for
Figure 3. Cost results (indexed) for several scenarios for the number of intermediate products.
3486 R. Akkerman et al.
numbers of intermediates below seven, it is not possible anymore to design a set of
intermediates that can be mixed to create all required final products, and therefore no
results can be generated for these scenarios.
5.2 Choosing the decoupling point: deliver-from-stock or mix-to-order
As both final products and flexible (still to be mixed) products are considered for
intermediate products, deciding on which ones to use also decides on the decoupling point
for all final products. Essentially, the choice is between deliver-from-stock (when final
products are selected as intermediates) and mix-to-order (when flexible recipes are stored
and subsequently mixed when customer orders arrive). It should be noted that in case
a final product is selected as an intermediate, it can then also be used as an ingredient in
mix-to-order products.In the scenario with 15 intermediate products, we can see that nine existing final
products are selected as intermediates, in addition to six flexible products. Furthermore,
due to overlapping intervals in quality specifications, some of the selected final products
have such specifications that they can also be substituted for other final products, thereby
actually increasing the number of deliver-from-stock products from 9 to 14, and further
reducing mixing operations.Figure 4 illustrates how the split between flexible and final products changes for fewer
or more intermediate products. For final products, a distinction is made between final
products that are also used as ingredients for other final products (multi-use), and final
products that are not (single-use). We can see that, for fewer intermediate products, all of
them are used in mixing other final products. When the number of intermediate products
increases, there is an increasing share of single-use final products. Furthermore, these
results illustrate that the final products that are chosen to be stored as intermediates often
have quality specifications that make them useful as ingredients in other products. Only if
Figure 4. Split between flexible and final products for several scenarios for the number ofintermediate products.
International Journal of Production Research 3487
we allow a significant increase in the number of intermediate products, we will seea decrease in the use of intermediates as ingredients for other products.
Comparing the final products that still have to be mixed with the ones that can now bedelivered directly from storage, we see that the products that can be delivered withoutmixing have – on average – a larger demand volume; out of the top-10 products in terms ofdemand volume, only four products still need to be mixed. Based on a typical month ofdemand data, Table 1 illustrates the difference in demand volumes. Like in the previoussection, the data is again indexed, here by setting the average demand per product in thisspecific month to 100.
These results correspond to what one might expect based on decoupling point theory:for products with large demand volumes earlier specification seems sensible. Although itwas not considered in our model, the predictability of demand for these products mightalso be higher, which would also argue for an upstream effect on the decoupling point(Van Donk 2001).
To be able to use the six flexible intermediate products (from the scenario with 15products) to mix the remaining final products, it would be sensible to choose a set ofproducts that cover a wide quality spectrum. To see how the model results relate to this,we analysed the quality specifications of the six selected products, in relation to the 30potential intermediate products designed by the quality management department of thecase company.
As it is not practical to illustrate all nine dimensions of the quality spectrum, weillustrate the results using the parameters protein content and bread volume (qualityparameter 1 and 9 as described in Section 3.2). Figure 5 shows the values for theseparameters for all 30 flexible intermediate products, and highlights the six selectedproducts. Again, results were indexed; here by setting the average value for the specificparameter to 100. We can see that the model indeed selected a set of products that coverthe quality spectrum to a large extent. It is worth noting that the six products shown havefairly similar values for the other quality parameters, with the exception of the threeproducts clusters around a protein index of 105–110. These products differ significantly interms of the fifth quality parameter: DON level, which also explains why these threeproducts are quite close to each other in this illustration. Further outcomes for otherquality parameters are not presented in this paper, but show similar results.
5.3 Managerial implications
Based on the results presented in Section 5.1 and 5.2, the research team andmanagement along with people directly involved, have discussed the approach
Table 1. Comparison of final product demand volumes (indexed) for different products.
Product type Number of products Average demand volume (þSD) Share of demand
and outcomes of the model. All agreed that well-informed decisions can be made regardingthe number of intermediate products, their compositions, and whether they are made tostock or mixed to order. The company acknowledges the value of the outcomes and theirvalidity. The outcomes of the model confirmed the viability of the company’s policy ofhaving as little as possible intermediates, while keeping all costs low as well. This alsoconfirmed the market approach of the company to provide a broad range of customer-specific products and to further increase that range in the future. The model illustratedthat this would not have to be matched by an increase in intermediate products. Further,the model’s outcomes facilitate discussion on the overall production strategy, investmentsin processing equipment, intermediate storage silos and mixing capacity, based on factsand figures. Although the results are consistent with the current way of working and areaccepted by the company, the detailed outcomes of the model are not used yet, for reasonsof product integrity. Moreover, running the mill involves daily small changes in the recipesto accommodate for small differences in lots of ingredients. Below we discuss some of theabove issues in more depth.
When making the final decision on the number of intermediate products, the amountof intermediate storage tanks and the possible cost of investment in additionalintermediate storage tanks could play a deciding role. For example, additional costsavings on raw materials and operating costs might not be high enough to warrant theinvestment in additional storage tanks.
Also, factors like required customer lead times could influence the final decision, as it isvery well possible that certain final products have to be available on the intermediatestorage level (delivered from stock) to ensure immediate delivery or packaging withoutfurther processing operations. If such preferred customer treatment should exist, this couldeasily be formulated in the model to see if and how that would change the selection of theremaining intermediate products.
By using a limited number of flexible intermediate products (six in the scenariodiscussed in Section 5.2), postponement of specification for a large range of final products(here 31) can be achieved. This downstream effect in terms of the decoupling point can beused to control demand uncertainty, both in terms of which products will be ordered andwhen they will be ordered (Yang et al. 2004a).
Figure 5. Illustration of the quality parameter range covered by the six selected flexible intermediateproducts.
International Journal of Production Research 3489
Finally, it is worth noting that, following Graman and Magazine (2006), successful
implementations of postponement not only build on selecting the right products, but also
on how the results are translated into other organisational functions such as operations
scheduling, where different types of products might have to be dealt with in different ways.
Graman and Magazine also stress that care should be taken related to possible changes in
product integrity. As our case study concerns an application in the food industry, this
relates to food quality and safety, which is indeed something that should be considered
very carefully.All in all, based on the experiences gathered in our case study, we are sure that the
model can provide a valuable tool to support decision making at several levels. As with all
models it will not replace decision making.
6. Conclusion and discussion
In this paper, we develop a method to design intermediate products in a two-stage food
production process, taking into account a wide variety of quality-related attributes. The
method consists of a two-stage mathematical programming model that aims to find
a balance between material cost and operating cost. The model is developed and applied in
a case study concerning a medium-sized flour manufacturer.The results show that the number of intermediate products can be reduced, simplifying
operations and increasing efficiency of the milling process, while only leading to
a marginal cost increase. Several scenarios have been studied for operational costs and raw
material availability to validate the results, also illustrating the usefulness of the model as
a decision support tool for the make or mix to order decision. It also shows that the costs
increase if the number of intermediate products decreases, although cost changes are
marginal above a certain number of intermediates. The model is designed to be used for
what-if analysis and should be solved on a regular basis, to make sure the current market
environment (including developments in the raw material supply and the customer
demand) is reflected in the composition of the intermediate products. Even more
important is that the results and scenarios of the model can be used to discuss and develop
the production strategy, explore consequences in terms of investments needed, etc., based
on objective figures and comparisons, in addition to experience, traditions and tacit
knowledge.Although the paper largely describes one single case study, the model developed can
easily be adapted for other food production systems, as many of these can be seen as two-
stage production systems with intermediate storage and a divergent product flow.
Furthermore, a lot of the quality attributes modelled in the case study are typical for the
food industry, which suggests numerous possible applications of the approach and the
model.Scientifically, this paper contributes in several ways. We incorporate the design of
intermediate products to ‘decoupling point theory’, whereas such intermediates are
normally given. We help in developing quantitative, but also integral decision tools to that
body of knowledge. It is clear that the model can help managers to make better-informed
decisions regarding what products to store as intermediates.Future research could follow a number of directions. One possible way is to
incorporate more elements into our quantitative model: constraints relating to raw
material usage or availability, or capacity limitations such as storage or batch sizes.
3490 R. Akkerman et al.
Related to these additional constraints, it might be beneficial to create an integratedmodel instead of the decomposed approach presented in this paper. This would allowfor better ways to include, for example, constraints on raw material usage. Also, in anintegrated model, we would not need to limit the list of potential flexible intermediateproducts, as we could let the model design these, based on the available raw materialsand the required final products. An integrated model would be more complex froma computational viewpoint, but methods that could be used to solve such a problem inan efficient way are available in the literature (e.g. Benders 1962). However, for theapproach presented in this paper, we aimed at supporting the current organisationaldecision-making process and did not pursue an integrated modelling approach. Alsothe results of the current model are therefore accepted more easily. Finally, moreresearch is also needed to further explore how a tool like the one developed here,should and can be used in the organisational context. Here, we need to decide howoften the models have to be solved with updated information, and identify whichdecisions (be they strategic, tactical or operational) the tool could also support, next tothe issues discussed in this paper.
Acknowledgements
The authors would like to thank the company involved for their willingness to participate in thisresearch and for their suggestions and remarks during discussion sessions. Also, the first authorwould like to acknowledge support from a H.C. Ørsted postdoctoral fellowship from the TechnicalUniversity of Denmark.
References
Abukhader, S.M. and Jonson, G., 2007. Postponement strategies in food supply chains – an
empirical investigation. International Journal of Logistics Systems and Management, 3 (2),
215–234.Akkerman, R., Van Donk, D.P., and Gaalman, G., 2007. The influence of capacity- and time-
constrained intermediate storage in two-stage food production systems. International Journal
of Production Research, 45 (13), 2955–2973.
Ashayeri, J., Van Eijs, A.G.M., and Nederstigt, P., 1994. Blending modelling in
a process manufacturing: A case study. European Journal of Operational Research, 72 (3),
460–468.
Benders, J.F., 1962. Partitioning procedures for solving mixed-variables programming problems.
Numerische Mathematik, 4, 238–252.Boone, C.A., Craighead, C.W., and Hanna, J.B., 2007. Postponement: An evolving supply chain
concept. International Journal of Physical Distribution & Logistics Management, 37 (8),
594–611.Caux, C., David, F., and Pierreval, H., 2006. Implementation of delayed differentiation in batch
process industries: A standardisation problem. International Journal of Production Research,
44 (16), 3243–3255.
Dobson, P.W., Clarke, R., Davies, S., and Waterson, M., 2001. Buyer power and its impact on
competition in the food retail distribution sector of the European Union. Journal of Industry,
Competition and Trade, 1 (3), 247–281.
Forza, C., Salvador, F., and Trentin, A., 2008. Form postponement effects on operational
performance: a typological theory. International Journal of Operations & Production
Management, 28 (11), 1067–1094.
International Journal of Production Research 3491
Fransoo, J.C. and Rutten, W.G.M.M., 1994. A typology of production control situations inprocess industries. International Journal of Operations & Production Management, 14 (12),47–57.
Graman, G.A. and Magazine, M.J., 2006. Implementation issues influencing the decision to adopt
postponement. International Journal of Operations & Production Management, 26 (10),1068–1083.
Griffith, C.J., 2006. Food safety: Where from and where to? British Food Journal, 108 (1),
6–15.Hoekstra, S. and Romme, J.H.J.M., eds, 1992. Integral logistic structures: Developing customer-
oriented goods flows. London: McGraw-Hill.
Kerkkanen, A., 2007. Determining semi-finished products to be stocked when changingthe MTS-MTO policy: Case of a steel mill. International Journal of Production Economics,108 (1/2), 111–118.
Lyon, P., Milne, R.J., Orzell, R., and Rice, R., 2001. Matching assets with demand in supply-chainmanagement at IBM Microelectronics. Interfaces, 31 (1), 108–124.
Olhager, J., 2003. Strategic positioning of the order penetration point. International Journal ofProduction Economics, 85 (3), 319–329.
Rajaram, K., Jaikumar, R., Behlau, F., Van Esch, F., Heynen, C., Kaiser, R., Kuttner, A., andVan de Wege, I., 1999. Robust process control at Cerestar’s refineries. Interfaces, 29 (1),30–48.
Rutten, W.G.M.M., 1995. The use of recipe flexibility in production planning and inventory control,PhD Thesis. Eindhoven University of Technology, The Netherlands.
Rutten, W.G.M.M. and Bertrand, J.W.M., 1998. Balancing stocks, flexible recipe costs and high
service level requirements in a batch process industry: A study of a small scale model.European Journal of Operational Research, 110 (3), 626–642.
Soman, C.A., Van Donk, D.P., and Gaalman, G., 2004. Combined make-to-order and make-to-stock in a food production system. International Journal of Production Economics, 90 (2),
223–235.Van Donk, D.P., 2001. Make to stock or make to order: The decoupling point in the food processing
industries. International Journal of Production Economics, 69 (3), 297–306.
Van Donk, D.P., Akkerman, R., and Van der Vaart, J.T., 2008. Opportunities and realities ofsupply chain integration: The case of food manufacturers. British Food Journal, 110 (2),218–235.
Van Hoek, R.I., 1999. Postponement and the reconfiguration challenge for food supply chains.Supply Chain Management: An International Journal, 4 (1), 18–34.
Van Hoek, R.I., 2001. The rediscovery of postponement: A literature review and directions for
research. Journal of Operations Management, 19 (2), 161–184.Wanders, H.L.T., Gaalman, G.J.C., and Sierksma, G., 2004. The composition of
semi-finished inventories at a solid board plant. European Journal of Operational Research,155 (1), 96–111.
Yang, B., Burns, N.D., and Backhouse, C.J., 2004a. Management of uncertainty throughpostponement. International Journal of Production Research, 42 (6), 1049–1064.
Yang, B., Burns, N.D., and Backhouse, C.J., 2004b. Postponement: a review and
an integrated framework. International Journal of Operations & Production Management, 24(5), 468–487.
3492 R. Akkerman et al.
Copyright of International Journal of Production Research is the property of Taylor & Francis Ltd and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.
Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=tppc20
Download by: [University of Groningen] Date: 15 November 2016, At: 01:25
Production Planning & ControlThe Management of Operations
Developing an understanding of lean thinking inprocess industries
Andrew Charles Lyons , Keith Vidamour , Rakesh Jain & Michael Sutherland
To cite this article: Andrew Charles Lyons , Keith Vidamour , Rakesh Jain & Michael Sutherland(2013) Developing an understanding of lean thinking in process industries, ProductionPlanning & Control, 24:6, 475-494, DOI: 10.1080/09537287.2011.633576
To link to this article: http://dx.doi.org/10.1080/09537287.2011.633576
Production Planning & ControlVol. 24, No. 6, June 2013, 475–494
Developing an understanding of lean thinking in process industries
Andrew Charles Lyonsa*, Keith Vidamoura, Rakesh Jainb and Michael Sutherlandc
aManagement School, University of Liverpool, Chatham Building, Liverpool L69 7ZH, UK; bDepartment of MechanicalEngineering, Malaviya National Institute of Technology, Jawahar Lal Nehru Marg, Jaipur 302017, Rajasthan, India;
cStatistical Consulting Center, University of Massachusetts, Amherst, MA 01003, USA
(Received 26 July 2010; final version received 15 October 2011)
The research described in this article has set out to determine the extent to which lean thinking is being adoptedas a manufacturing philosophy by process industries. It concerns the application and examination of key leanmanufacturing principles, namely, the alignment of production with demand, the elimination of waste, theintegration of suppliers (IS) and the creative involvement of the workforce in improvement activities, to a rangeof process industry types based on Dennis and Meredith’s taxonomy of process industry transformation systems[Dennis, D. and Meredith, J., 2000a. An empirical analysis of process industry transformation systems.Management Science, 46 (8), 1085–1099]. Seventy-nine process industry product streams across 62 sites werestudied. In addition, a five-site investigative field study was also undertaken. The findings demonstrate that leanpractices associated with the elimination of waste are consistently used for improving manufacturingperformance throughout the taxonomy of process industries but practices associated with other lean principlesare inconsistently applied. In addition, explanations are provided on the appropriateness of lean thinking as amanufacturing philosophy and a strategy for improving manufacturing performance in different process industrytypes, and on the extent to which lean principles and practices are dependent on the characteristics of processindustry transformation systems.
Keywords: process industries; lean manufacturing; survey; field study; multi-method
1. Introduction
The aim of this research is to attempt to supportorganisational decision-making in process industriesby determining the extent to which lean thinking canbe employed as an appropriate manufacturing philos-ophy for practising, process industry managers. Thepurpose of this study is not solely to examine theappropriateness of specific lean practices such as 5S,value steam mapping and kanbans in process indus-tries, as has been tried by Abdulmalek et al. (2006), butalso to explore the suitability of the underlyingprinciples that drive the implementation of lean.
Lean thinking has been extensively and rigorouslystudied. It originated in discrete manufacturing but atits heart is a set of core, complementary principles thatenhance customer value. Do these key tenets of leanthinking apply equally to discrete and process indus-tries? If not, which tenets are applicable, to what extentand to which types of process industries do they apply?These are the questions the authors have posed andhave set out to explore in this research.
This article has the following structure. In the nextsection, a review of empirical examples of process
industry lean applications is provided followed by a
justification of the lean thinking framework employed.
Discrete industries are then compared with process
industries in order to characterise and contextualise the
original and archetypal lean applications’ domain
(discrete) and contrast this with the characteristics of
a process industry setting. Such a contrast provides a
simple exposition and characteristic signature of both
discrete and process environments in order to high-
light, in simple terms, those organisational and trans-
formational variances a lean system would be required
to address in a process industry implementation. The
process industry, however, is not one industry but a
collection of industry types. A corollary to this is
shown in the next section, where the process industry
domain is subjected to a simple sub-division in order to
articulate the diversity of types. The heterogeneity of
process industries is further highlighted through the
description of published process industry classification
systems. This is followed by a description and expla-
nation of the research methodology and justification of
the process industry classification system employed.
The results of the survey of process industries and
detailed field studies are then discussed followed by theconclusions from the research, an explanation of theresearch limitations and recommendations for futureresearch.
2. Lean thinking in process industries
The diffusion of lean thinking, particularly since thepublication of the landmark text, The machine thatchanged the world (Womack et al. 1990), has fuelled adebate in both the practitioner and academic commu-nities concerning the applicability of the lean approachoutside discrete, repetitive industries (Kenney andFlorida 1993, James-Moore and Gibbons 1997,Cooney 2002). Womack et al. (1990) stated that notonly is lean manufacturing a superior way for humanbeings to make things, but that its principles can beapplied in every industry. Such a sentiment is shared byShah and Ward (2003) who suggested that leanpractices are prevalent in all industries and shouldnot be limited to discrete manufacturing industries.These are bold claims, yet there is clear evidence thatlean thinking is being tried and tested in a range ofdifferent industrial contexts from low volume, highlycomplex (Crute et al. 2003, Slomp et al. 2009) to highvolume, highly standardised (Dhandapani et al. 2004)manufacturing environments through to services(Piercy and Rich 2009) and construction (Ballardet al. 2003).
Several studies have been undertaken in processindustries. Billesbach (1994) described the applicationof lean thinking at the DuPont Textile company. Apull system was used using a kanban-like approach.The results indicated a 96% reduction in work inprocess, working capital reduction of $2 million andproduct quality improvement of 10% (Billesbach1994). Hodge et al. (2011) found visual methods tobe the most-frequently used lean technique in a studyof textile companies. The Dow Chemical Company inNorth America and one of its supply chain partnersdecreased the lead time variability by 50% in acontinuous process chemical manufacturing environ-ment by applying lean concepts (Cook and Rogowski1996). Houghton and Portugal (1997) analysed single-stage batch manufacturing as well as multi-stage batchmanufacturing scenarios that were based on Just-In-Time inventory (JIT) production in the food industry.Roy and Guin (1999) described the JIT purchasing at asteel plant in India and Dhandapani et al. (2004)applied some aspects of lean thinking to a steelcompany and demonstrated a reduction in the pro-duction cost by 8% and a lead-time reduction of 50%.He and Hayya (2002) examined the JIT manufacturing
and its positive impact on the quality in the food
industry. Jones and Clark (2002) reported a case study
of a contract processor, which bottled and canned
several different branded soft drinks, which used a lean
thinking approach to compress its value stream.
Simons and Zokaei (2005) introduced some leanpractices, specifically takt time and standardised
work, to a meat processing plant to improve the
productivity and quality.These empirical studies demonstrate how the
improvement of operational performance can be
achieved through the adoption of lean practices; yet,the studies reviewed have not explicitly attempted to
link specific process industry characteristics and the
structure of process industry operations to specific lean
principles. One research team that has attempted to
make this connection is Abdulmalek et al. (2006). Theyused the case of a steel mill to illustrate how lean tools
could be applied in a hybrid continuous flow/batch
process environment. They concluded that 5S, value
stream mapping and visual systems were universally
applicable. Set-up reduction, just-in-time, production
smoothing and total productive maintenance (TPM)were partially applicable and cellular manufacturing
was ‘very difficult’ to implement. The researchers offer
a limited explanation but the implication here is that
the structure and system design of the steel mill
support the use of certain practices but conspireagainst the use of others.
3. Proposed lean thinking framework
Capturing and codifying the essence of lean or
decoding its DNA (Spear and Bowen 1999), remains
a challenge. This has been promulgated and exacer-
bated by the speciation of lean into agility, leagility andlean six sigma. There is even confusion as to what can
be regarded as lean and what cannot (Hines et al.
2004). Researchers have used different terms, such as
Toyota Production System, Just-in-Time, Lean
Production and World-Class Manufacturing to express
a lean approach to manufacturing (Oliver et al. 1994).There have been recurrent attempts at fusing together
the elements of lean and a number of researchers have
proposed lean frameworks (Ramarapu et al. 1995,
Oliver et al. 1996, James-Moore and Gibbons 1997,
Karlsson and Ahlstrom 1997, Panizzolo 1998, Sanchez
and Perez 2001, Hines et al. 2004, Hodge et al. 2011),but the multiple degrees of freedom present in the
application of lean, the emphasis on practice andthe proliferation of interpretative approaches thwart
the neat crystallisation of lean thinking and
476 A.C. Lyons et al.
consequently there has been no unanimous agreement
on a single, unified structure.The intention in this research, therefore, was not to
provide the definitive lean thinking framework but
posit broadly independent dimensions of lean that are
both contemporary and conspicuous in the literature.
Four wide-ranging ambitions were postulated and
deemed suitable to portray the principles of lean
thinking:
(1) alignment of production with demand;(2) elimination of waste;(3) IS; and(4) creative involvement of the workforce in
process improvement activities.
The four principles and some of their contributory
lean practices are depicted in Figure 1. The principles
are distinct yet mutually supportive ideas. The align-
ment of production with demand is predicated on the
notion of principle 2, waste elimination. The elimina-
tion of waste helps to reduce cost and organise the
required, value-creating production activities into an
efficient system design that facilitates smooth
production flow with minimal interruptions, delays
and variations. Such a process is essential to facilitate
principle 1, the alignment of production with customer
demand. A business imperative is the ability of compa-
nies to minimise organisational inertia and configure
their operations to respond to customer’s requirements.
In lean thinking terms, the means to achieve such
responsiveness is by producing at the pace of customer
demand and facilitating flow through the systematic
elimination of non-value-adding activities. Spear and
Bowen (1999) described the apparent paradox of this
relationship at Toyota in creating ‘rigidly scripted
activities and production flows yet having operations
that are enormously flexible and adaptable’. A prereq-
uisite to the alignment concept is principle 3, the notion
of supplier integration. Dassbach (1994) noted that
‘Where the organisational logic of Fordism has been to
internalise, the organisational logic of lean production
has been to externalise. Instead of creating a giant,
integrated enterprise each lean producer has created
their own co-operating network’. Without establishing
a close relationship with the suppliers, the concept of
aligning production with demand cannot be achieved.
Figure 1. Outline of the lean thinking framework.
Production Planning & Control 477
Suppliers must deliver parts and materials frequently insmall quantities directly to the point of use withminimalreceiving inspection (Womack and Jones 1994, McIvor2001). The inculcation of process improvement activi-ties into the minds of the workforce, principle 4, andknown as ‘soikufu’ in Japanese manufacturing, isessential for the elimination of waste. Total employeeinvolvement through attempting to release the talentsand creativity of people is a ubiquitous lean driver andambition. Forza (1996) suggested that lean manufactur-ing plants are characterised by the use of small-teamproblem solving, worker suggestions for improvement,decentralisation of authority and the use of multi-functional employees.
The hierarchical nature of the lean frameworkrepresented by Figure 1 conveys the notion of policydeployment and the need for organisational goals to bein concert with lean principles and practices. Thisnotion of devolving strategic intent throughout anenterprise and decision-making responsibility to theworkforce is then reflected in principle 4 (the creativeinvolvement of the workforce in process improvementactivities).
The framework was not developed in order to makean unnecessary addition to the existing lean thinkingframework-set. Rather, it provides a lean architecturethat is not only suitable for this study by allowing theadoption of lean principles and practices to be readilyestablished but also it provides a coherent, uncompli-cated amalgam of those goals, principles and practicesthat are evident in the most authoritative lean thinkingresearch. For example, Ramarapu et al. (1995) identi-fied the critical elements of JIT implementation as beingthe elimination of waste and quality improvement(together equivalent to the elimination of waste principlein the proposed framework), production strategy(includes the alignment of production with demandprinciple), management commitment and employeeparticipation (analogous to the creative involvement ofthe workforce principle) and supplier participation(equivalent to the IS principle). Spear and Bowen’s(1999) much-cited research on the DNA of the ToyotaProduction System includes a description of Toyota’snotion of the output of an ideal production system thatis defect free without wasting resources (elimination ofwaste and IS), can be delivered immediately one requestat a time (alignment of production with demand and IS)and can be produced in a work environment that isprofessionally supportive for every employee (creativeinvolvement of the workforce).
There is unanimous agreement amongst researchersand practitioners that waste elimination is an integralpart of lean. Some key advocates include Sanchez andPerez (2001), Shah and Ward (2003), Wood (2004),
Shah and Ward (2007) and Pettersen (2009). Researchsupporting the ‘alignment of production with demand’principle includes Naylor et al. (1999), Lewis (2000),Crute et al. (2003) and Shah and Ward (2003). The ‘IS’principle is supported by research that includes McIvor(2001), Bhasin and Burcher (2006), Shah and Ward(2007) and Pettersen (2009), and the ‘creative involve-ment of the workforce’ principle is evident in the workof many researchers including Campbell (1987),Panizzolo (1998), Lewis (2000), Sanchez and Perez(2001) and Kojima and Kaplinsky (2004). Tables 1–4provide a summary of key references that are supportiveof each of the four lean principles.
Table 1. Key references associated with aligning productionwith demand.
Lean-relatedreference
Alignment of production with demanddescriptor
Monden (1983) Alignment of production system todemand fluctuation
Shingo (1989) Stockless production by aligningproduction to orders usingsingle-minute exchange of die
Womack et al.(1990)
Pull production with smootheddemand, improving flow of materialand information
Womack andJones (1994)
Continuous flow and alignment of allproduction activities
Mould and King(1995)
Alignment of flow with demand ratefor finished products
Forza (1996) Synchronised flow, every workingactivity must be supplied with thenecessary components at the neces-sary time and in the necessaryquantity
Karlsson andAhlstrom(1997)
Every process should be provided withone part at a time exactly when thatpart is needed
Levy (1997) Responsiveness to changing demandJames-Moore and
Gibbons (1997)Flexibility through quick set-up times
and build to customer orderPanizzolo (1998) Levelled production, mixed model
scheduling, pull flow control, set-upreduction and small lots
Lewis (2000) Emphasis on customer pull rather thanorganisation push
Sanchez and Perez(2001)
Production and delivery just in time
Shah and Ward(2003)
Produce at the pace of customerdemand, continuous flowproduction
Crute et al. (2003) Production is pulled in response tocustomers
Shah and Ward(2007)
Establish mechanisms that enable andease the continuous flow of products
478 A.C. Lyons et al.
Despite capturing the sentiments of the publishedlean thinking literature, the proposed framework is notexhaustive. For example, it could be argued that theframework does not adequately incorporate the long-term thinking and strategising embedded in a number
Table 3. Key references associated with the integration ofsuppliers.
Lean-relatedreference Integration of supplier descriptor
Womack et al.(1990)
Close integration with suppliers
Golhar andStamm (1991)
Supplier participation through long-term contracts, supplier training,single sourcing, communicationwith suppliers and suppliernetworks
Oliver et al. (1994) Close relations with suppliersRamarapu et al.
(1995)Supplier participation implemented
through small lot size, communi-cation with suppliers, long-termcontracts, supplier training andsingle source supply
Lamming (1996) Lean supply to make the entire flow,from raw materials to consumer,an integrated whole
Oliver et al. (1996) Active information exchange withsuppliers, long-term relationships
Levy (1997) Close relationship with suppliers onquality and design for manufac-ture issues
Kinnie et al.(1998)
Involvement of all suppliers in acontinuous process to improveproducts
Panizzolo (1998) Early information exchange on pro-duction plans with suppliers, sup-plier involvement in qualityimprovement, long term contracts
Naylor et al.(1999)
Integrated supply chain to remove allboundaries
Sanchez and Perez(2001)
Supplier integration, supplierinvolvement in information andcomponent design
McIvor (2001) Concept of lean supply, supplierinvolvement in customer designactivities and joint buyer–suppliercost reduction
Bhasin andBurcher (2006)
Actively develop links with suppliersand attempt to reduce the numberof suppliers
Matsui (2007) Just-in-time delivery by suppliersShah and Ward
(2007)Creating a dependable and involved
supplier base that consists of a fewkey suppliers with long-term con-tracts, supplier feedback anddevelopment
Pettersen (2009) Long-term relationships with suppli-ers and supplier development
Table 2. Key references associated with the elimination ofwaste.
Lean-relatedreference Elimination of waste descriptor
Monden (1983) Elimination of non-value-addingactivities
Ohno (1988) Elimination of seven types of wasteShingo (1989) Total elimination of waste; strive
for stockless productionWomack et al.
(1990)The same output with less workers,
less inventory and less floorspace through the elimination ofwaste
Elimination of waste throughreduced lot sizes and lead times
Katayama andBennett (1996)
Minimise unnecessary time, mate-rial and effort in the productionprocess
Karlsson andAhlstrom(1996)
Elimination of waste – everythingthat does not add value to theproduct
James-Moore andGibbons (1997)
Waste elimination implementedthrough low inventory levels,short distance travelled, no overproduction, high yield, highproductivity and short leadtimes
Hallihan et al.(1997)
Elimination of the seven wastes
Kinnie et al.(1998)
Elimination of waste in terms ofmaterial and human resources
Naylor et al.(1999)
All non-value-adding activities, ormuda, must be eliminated
Sanchez and Perez(2001)
Elimination of everything that doesnot add value to the product orservice
Shah and Ward(2003)
Continuously reducing, and ulti-mately eliminating all forms ofwaste
Wood (2004) Systematically drive out waste bydesigning better ways of working
Seth and Gupta(2005)
Goal of lean manufacturing is toreduce waste in human effort,inventory and time to market tobecome responsive
Simons andZokaei (2005)
Waste elimination through stan-dardised work, visual signalsand 5S
Narasimhan et al.(2006)
Minimal waste due to unneededoperations, inefficient opera-tions or excessive buffering inoperations
Shah and Ward(2007)
Eliminate waste by concurrentlyreducing or minimising supplier,customer and internal variability
Engeuland et al.(2008)
Continuous focus on value-addingactivities through the elimina-tion of waste
Pettersen (2009) Elimination of the seven wastes
Production Planning & Control 479
of approaches such as Liker’s (2004) ‘4P’ (philosophy,
process, people and partners and problem solving)
model of the Toyota Way derived from Toyota’s own
four high-level principles of genchi genbutsu, kaizen,
respect and teamwork, and challenge. The notion of
basing management decisions on a long-term philoso-phy is not absent from the proposed framework as the
notion of change, kaizen and innovation are included
in the principle associated with the creative involve-
ment of the workforce but long-term thinking is clearly
more conspicuous in other approaches. Nevertheless,
the framework is fit for purpose; it is not intended to bedefinitive but a representative lean model that can be
utilised in a practical manner for determining the
adoption of lean thinking principles and practices.
4. Discrete versus process industry characteristics
The type of primary process employed within a
manufacturing facility is a fundamental characteristicof the facility’s operation and is the foundation of its
opposed to countable), indistinguishable products and
is identifiable in the job shop, batch and continuous
flow types of primary process. A much-cited definition
of process manufacturing has been provided by the
APICS dictionary. The 12th edition (Blackstone 2008)
offers the following: ‘Production that adds value bymixing, separating, forming and/or performing chem-
ical reactions. It may be done in either batch or
continuous mode’. The batch and continuous flow
process types are those that are most often associated
with process manufacturing. Job shop manufacturing
has been excluded from the APICS definition but isalso a feasible choice, albeit a relatively rare industrial
occurrence. Abdulmalek et al. (2006) highlight organic
dyes as being an example job shop product; others
include speciality foods and personalised medicine.
Table 4. Key references associated with the creativeinvolvement of the workforce.
Lean-relatedreference
Creative involvement of the workforcedescriptor
Campbell(1987)
Vitalised workforce, a key to success;workers used creatively and givenresponsibility
Womack et al.(1990)
Transferring the task and responsibilitiesto front-line workers
Golhar andStamm(1991)
Employee participation in decisionmaking through cross-training, groupdecisions and employee suggestions
Ramarapuet al. (1995)
Employee participation in decisionmaking through cross-training/education, team decision making andemployee suggestions
Boyer (1996) Lean production through training of theworkforce, worker empowerment andthe use of small teams for problemsolving
Forza (1996) Worker involvement through small teamproblem solving, employee sugges-tions, decentralisation of authority,worker autonomy and commitment tocontinuous quality improvement
Oliver et al.(1996)
High commitment human resourcepolicies; team-based workorganisation
Hallihan et al.(1997)
Employee involvement through problemsolving techniques and dedicated timefor improvements
Karlsson andAhlstrom(1997)
Continuous improvement, multi-functional teams and decentralisedresponsibilities
James-Mooreand Gibbons(1997)
People utilisation, employee contribu-tion, teamwork, empowerment andrespect for humanity
Kinnie et al.(1998)
Involvement of all employees in acontinuous process to improve prod-uct and job design
Panizzolo(1998)
Expansion of autonomy, workerinvolvement in quality improvementand team decision making
Lewis (2000) Commitment to continuousimprovement enabled by peopledevelopment
Sanchez andPerez (2001)
Involvement of all productionemployees for continuousimprovement
Kojima andKaplinsky(2003)
Greater employee involvement,participation through the use ofcontinuous improvement practices
Simons andZokaei(2005)
Kaizen, enhanced problem solvingability of employees and enhancedemployee participation
Matsui (2007) Employee suggestions and small groupproblem solving
Shah and Ward(2007)
Employee involvement in problemsolving
Engeulandet al. (2008)
Involvement of everyone in the contin-uous process of driving out wastefulprocedures
480 A.C. Lyons et al.
The Institute of Operations Management (IOM2009a) provides a list of characteristic processmanufacturing industries that includes chemicals, bio-technology, food and beverages, paper and board,textiles, glass, rubber and plastics, semi-conductorsand primary metals. An equivalent list for discretemanufacturing includes automotive, domestic appli-ances, electronics, telecommunications equipment,machinery and capital equipment (IOM 2009b). Atsome point during the course of production in manyprocess manufacturing environments, the final productbecomes discrete. This co-existence is highlighted byBillesbach (1994) and Abdulmalek et al. (2006).Abdulamalek et al. (2006) refer to the position in theproduction process, where process flow becomes dis-crete flow as the point of discretisation. The IOM(2009a) identifies pharmaceuticals, cosmetics and con-fectionery as examples of these hybrid manufacturingenvironments containing both process and discretephases.
A number of researchers have sought to clarify thedifferences between discrete and process manufactur-ing. Notable contributions have included researchundertaken by Fransoo and Rutten (1994), Ashayeriet al. (1996), Crama et al. (2001), Flapper et al. (2002)and Abdulmalek et al. (2006), and other, more implicitdifferences have been identified by Dennis andMeredith (2000b) and Akkerman et al. (2010).Table 5 shows an attempt to elucidate the differences
between discrete and process environments through asimple aggregation of differentiating factors. There aremany exceptions but the table highlights the keydiscriminating and distinctive features of processmanufacturing environments.
Other commonly referred to differences betweenprocess and discrete industries such as product varietyand capital intensity have been omitted from Table 5 assuch contrasts have consistently been based on theimplicit assumption that process industries follow asingle, homogeneous approach to product productionthat is characterised by continuous flow industries.
5. Process industry classification systems: typologies
and taxonomies
Some researchers have attempted to classify processindustries. In the following review of process industryclassification systems, the distinction between taxo-nomical and typological systems suggested by Boyeret al. (2000) is used: ‘taxonomies provide comprehen-sive classification systems (including ‘‘good’’ and‘‘bad’’ phenomena) while typologies only describeideal types’. Fransoo and Rutten (1994) offered atypology for process industries based on a continuumfrom batch to continuous flow manufacturing. Theyconflated Taylor et al.’s (1981) two-dimensional prod-uct–process matrix into a one-dimensional, eight-
Table 5. Some key differences between discrete and process manufacturing.
FactorDiscrete
(project, job shop, batch and repetitive)Process
(job shop, batch and continuous)
Product and productstructure
Solid Solid, liquid, or gasDeep product structure Shallow product structureAssembled bill-of-materials Blended formula or recipePrimarily convergent product flow Primarily divergent product flowCountable and distinguishable Measurable and indistinguishableMany input raw materials/components Few input raw materialsLimited shelf-life constraints Frequent shelf-life constraints
Manufacturingprocesses
Fabrication-based Fabrication-freePredictable material grade Variable material gradeProcess sequence precedence constraints Flexible process plans with fewer precedence
constraintsMinimal regulatory involvement Changes may be governed by regulatory
constraints
Production planningand control
Item tracking and control Lot tracking and controlPlanning of residual products unnecessary Residual products regularly produced as part
of the production processPredictable yield expected Often variable yieldPost-process equipment cleaning unnecessary Equipment cleaning requirements accounted
for in planningHigh degree of process control automation not
necessaryOften highly automated process control
Production Planning & Control 481
position perspective of process industry delineation.Fransoo and Rutten (1994) refer to continuous flowbusinesses as process/flow businesses in their typology.They describe how such environments have highproduction speeds, short throughput times, one rout-ing for all products, low product complexity, lowadded value, long changeover times, a small number ofproduction steps and a limited number of products.Batch/mix businesses (batch-driven environments),according to Fransoo and Rutten (1994), are char-acterised by long lead times, complex routings, productcomplexity that is higher than process/flow businessesand changeover times that have less impact thanprocess/flow businesses, high added value, a largenumber of production steps and a large number ofproducts. The typology is simple and does not definethe attributes of those types of process industriestowards the centre of the continuum, and therefore,furthest away from the more prescriptively charac-terised industries at the continuum extremities.However, the typology has an enduring appeal andthe research has provided motivation for further workin the field.
Abdulmalek et al.’s (2006) process industry typol-ogy consists of three parts. The first part is a two-dimensional approach based on material variety andproduct volume. It provides a general system ofclassification and distinguishes between customisedand commoditised process environments. The secondpart of the typology is comparable to the Fransoo andRutten (1994) typology and has the same generalsentiments but with the additional, explicit inclusion ofa job shop environment. The third part offers a novelperspective that accounts for the change in processcontinuity when process units become discrete units.Abdulmalek et al. (2006) use the three-part typology toexamine the appropriateness of a range of leanpractices to the highlighted process industry positionson each part of the typology.
There is greater heterogeneity in process industriesthan a simple volume/variety sub-division into jobshop, batch and continuous flow types of processmanufacturing or the typologies proposed by Fransooand Rutten (1994) and Abdulmalek et al. (2006) wouldsuggest. Dennis and Meredith (2000a) deduced ataxonomy of seven distinct, process industry sub-groups: process job shop, custom blending, fast batch,custom hybrid, stock hybrid, multistage continuousand rigid continuous (Table 6) from a study of 19process industry sites. Sub-group differentiation wasbased on an aggregation of four production dimen-sions: materials diversity, equipment, materials move-ment and run time. There are 16 characteristicsassociated with these four dimensions.
The seven sub-groups provide a delineation that is
based on a range of manufacturing characteristics. It is
too simplistic to analogise intermittent, hybrid and
continuous groups with the job shop, batch andcontinuous groups of Abdulmalek et al. (2006) and
in the case of the latter two, with Fransoo and Rutten’s
(1994) typology. Despite being founded on a small
empirical sample, the specificity of this taxonomy is itsstrength. Process industries can be assigned to a sub-
group analytically based on material diversity, equip-
ment, material movement and run-time characteristics.
This process is aided by the identification of example
industries but is less reliant upon an example matchthan in the other approaches. Exact matches are more
difficult using the qualitative thinking, where assign-
ment is based on a general profile as in the cases of
Fransoo and Rutten (1994) and Abdulmalek et al.(2006), rather than a numerical analysis.
6. Research hypotheses and methodology
The research set out to address the applicability of lean
thinking principles and practices in different types of
process industry. Three hypotheses were formulated:
H1: The fundamental principles of lean thinking are
being adopted by process industries.
Table 6. Process industry transformation systems (Dennisand Meredith 2000a).
Group Sub-group Example products
Intermittent Process job shop Speciality organicchemicals
H2: Lean principles are adopted unevenly in differenttypes of process industry.
H3: Different types of process industry can bediscriminated by their lean practices.
The main tasks that constituted the multi-methodapproach to the research are depicted in Figure 2. TheDennis and Meredith (2000a) taxonomy (Table 6) wasselected as a suitable process industry classificationsystem. This taxonomy provides a more nuanced,comprehensive and analytical picture of the differencesbetween process industry sub-groups than the moreimplicit, continuum-driven approaches of Fransoo andRutten (1994) and Abdulmalek et al. (2006).
7. Analysis and discussion of process industry
product streams
7.1. Data collection
A postal questionnaire was employed as the primarymethod of data collection. The questionnaire followedDillman’s (2007) seven stages of development.A preliminary questionnaire was pre-tested at threeprocess industry sites by interviewing senior produc-tion staff. Interviewees were asked to critique thequestionnaire, after which several presentational andlanguage changes were made. A list of over 400 process
industry sites was compiled from various sources in theUK, representing a broad Standard IndustrialClassification (SIC) cross-section and sent the finalversion of the questionnaire over a 3-month period.Questionnaires were sent to production directors andmanagers. All sample sites were required to satisfy aprimary UK SIC code within the manufacturingranges of DA-DI in order to qualify for this study.Three questionnaires were returned incomplete andeliminated from this study. Questionnaires from 62sites representing 79 product streams were returnedfully completed and usable. Sixty of the productstreams belonged to large enterprises, classified on aone-dimensional basis for having more than 500employees, the remaining 19 product streams wereidentified as belonging to small and medium-sizedenterprises (SMEs) with less than 500 employees. Nobias was found when a comparison was made betweenearly and late returned responses. For clarification ofresponses, additional information was obtainedthrough direct observation at 18 of the 62 sites andfrom face-to-face interviews with process industrymanagers at each of these sites. These interviews alsofacilitated a ‘test–retest’ approach to the assessment ofthe reliability of the questionnaire.
There were three sections in the questionnaire:background information, a process audit and alean audit. The first section concerned capturing
Figure 2. Research methodology.
Production Planning & Control 483
background information on the respondent’s plant,including its employee numbers, production volumes,product variety, the number of product streams andthe types of manufacturing and supply chain initiativesemployed. The second section concerned an audit ofprocess characteristics in order to provide the means toclassify the respondent’s chosen product stream(s)according to Dennis and Meredith’s (2000a) processindustry taxonomy. Consequently, questions con-cerned run times, numbers of work centres, inventorypoints, formulation complexity, routings, productvariety, equipment flexibility, lead-times and therange of raw material ingredients. The final sectionconcerned a lean audit based on the lean principles andpractices depicted in Figure 1. The content of the leanaudit consisted of a series of statements that respon-dents were asked the extent to which they agreed within relation to a particular product stream using a six-
point Likert scale ranging from ‘strongly disagree’ to‘strongly agree’. The lean audit statements have beentranscribed in Table 7.
7.2. Descriptive statistics
The results from the process audit section of thequestionnaire allowed the respondents’ productstreams to be classified against the Dennis andMeredith (2000a) taxonomy. This was facilitated bythe construction of a signature template for each of theprocess types based on the taxonomy variables. Forty-nine (63%) were classified as rigid continuous, 12(15%) as multistage continuous, 8 (10%) as stockhybrid, 5 (6%) as custom hybrid, 1 (1%) as fast batchand 4 (5%) as custom blending. None were classified asprocess job shops. A high degree of customisation with
Table 7. The lean audit statements.
Alignment of production with demand1 Production is ‘pulled’ based upon downstream customer demand2 Production is undertaken based on an instruction from a downstream process3 Production is regarded as ‘make-to-order’ rather than ‘make-to-stock’4 Production is paced to a customer demand rate or takt time (takt time is the customer demand rate and is calculated
from the ratio of time available/day to demand/day)5 Production rates vary in line with customer demand rates6 Production is mixed on the same processes and facilities7 Changes in demand volume and mix can be easily accommodated8 There is a commitment to reduce production run lengths and utilise a minimum economic batch size
Integration of suppliers9 Suppliers are actively supported in resolving their problems and improving performance10 Deliveries are based upon production requirements, are not excessive and arrive just before being used11 The notion of being a part of a complete (supply chain) value stream is understood and accepted12 Suppliers receive schedules that are stable and predictable without unexpected changes13 Raw materials and ingredients are single sourced14 Suppliers have flexible processes that can easily accommodate demand changes15 Deliveries are made directly to the point-of-use (rather than to a remote storage area)16 Supply inventory buffers are planned and set at minimum acceptable levels
Elimination of waste17 There is a real commitment to eliminate or minimise all non-value-adding activities18 Standard operating procedures are systematically used to provide work instructions19 Abnormal process behaviour is recognised and controlled by the workforce20 Visual displays are extensively used to support the standardisation and defect-free execution of the production process21 Quality systems and procedures are in place to prevent defects from moving to downstream operations22 The 5S system of workplace organisation is implemented and embraced by the workforce23 There is a commitment to reducing process set-up and changeover times24 TPM is well established and response to equipment breakdowns is systematised
Creative involvement of the workforce25 The workforce is actively involved in improvement activities and is empowered to make changes26 The work environment is organised, so that most work is undertaken in teams27 Individual and team-based improvement ideas are regularly received from the workforce28 A structured programme of employee training is in place and adhered to29 Management devolve work-related decisions to the workforce30 The workforce is multi-skilled and cross-trained and a system of job rotation is employed31 The work culture means that change is readily accepted and regarded as the norm32 Kaizen and constant, incremental improvement and innovation are embraced and practised by the workforce
484 A.C. Lyons et al.
long run times, characteristics of Dennis andMeredith’s (2000a) process job shops, was not acombination recognised within the sample. Table 8provides a full classification of the product streams.The table elucidates the unreliability of product-drivenprocess industry classification systems. Brewing andfood, for example, appear in both the rigid continuousand custom blending process types, and, therefore, theleast and most flexible process environments studied.Furthermore, sites of different sizes are found in eachprocess type. All further analyses were undertakenusing the respondents’ responses to the 32 lean practicestatements from the lean audit section of thequestionnaire.
Figure 3 reveals inter- and intra-sub-group com-parisons via a series of ladder diagrams that contrastthe four lean principles (ordinate axis) with five processindustry types (abscissa). The fast batch data set hasbeen omitted from this analysis as a single examplecannot be relied upon to be representative of the sub-group. Each of the lean principles has eight gradua-tions corresponding to the eight statements in the leanaudit (Table 7). Graduations have been shaded indiagram (a) for responses with a median� 4.00,in diagram (b) for responses with a median� 4.50, indiagram (c) for responses with a median� 5.00 andin diagram (d) for responses with a median� 5.50. Ineach case, the relative adoption of the lean thinkingprinciples in each of the five process types can bevisually compared. In addition, the results for a singlediscrete exemplar (‘Auto.’) have also been included.This exemplar is a UK-based, medium-to-high volume,automotive assembly plant. The plant has a maturelean thinking programme and is recognised, and wonawards, for its lean endeavours and achievements.
Interviews were conducted with four production andengineering managers in order to obtain the responsesto the lean audit. The exemplar has been included notto provide a direct comparison but demonstrate theextent to which the chosen lean principles and practicescan feasibly be adopted in an industrial environment.
Figure 3(a), with a median threshold of 4 corre-sponding to ‘somewhat agree’ on the lean audit Likertscale, provides a general impression that lean principlesare being adopted throughout the continuum ofprocess industry types. An exception, where the‘white space’ is conspicuous, concerns the IS principlefor the rigid continuous type. Rigid continuous com-panies are characterised by the production of a narrowrange of different finished products based on a low,average number of ingredients per product. Run timesare typically measured in days and there are few workcentres and inventory points. Economies of scale aretypically regarded as necessary to drive efficientproduction. Forward buying obviates the need fordeliveries to be based upon immediate shop floor need,and raw materials are typically not single sourced.However, the more prosaic management practice,working with suppliers to improve performance,scored highly. The lean principle that most promi-nently features in Figure 3(a) concerns the eliminationof waste. Almost all practices associated with thisprinciple achieved a median rating of 4.
Figure 3(b), with a median threshold of 4.5, depictsa gradual upward trend in the adoption of leanprinciples from the rigid continuous to the customblending types. It also provides an indication of thesensitivity of the data as the overall picture differsmarkedly from Figure 3(a). The elimination of wasteis still the most-widely adopted lean principle.
Table 8. Classification of product streams.
Dennis andMeredithclassification
Number ofproduct streams Product classification
Rigid continuous 49 (40 LEs and 9 SMEs) Paints, coatings and coating applications (21), other chemicals (bulkand speciality) (7), resins and plastics inc. packaging (6), soaps anddetergents (5), pharmaceuticals (3), food (3), brewing (2), buildingmaterials (1) and ceramics and glassware (1)
Multistage continuous 12 (9 LEs and 3 SMEs) Other chemicals (bulk and speciality) (5), ceramics and glassware (2),paints, coatings and coating applications (1), resins and plasticsinc. packaging (1), pharmaceuticals (1), food (1) and oil (1)
Stock hybrid 8 (5 LEs and 3 SMEs) Pharmaceuticals (3), other chemicals (bulk and speciality) (2), resinsand plastics inc. packaging (2), paper (1)
Custom hybrid 5 (4 LEs and 1 SME) Resins and plastics inc. packaging (4) and pharmaceuticals (1)Fast batch 1 (SME) Other chemicals (bulk and speciality) (1)Custom blending 4 (2 LEs and 2 SMEs) Food (2), brewing (1) and building materials (1)
Note: LE, large enterprise.
Production Planning & Control 485
The creative involvement of the workforce principlefeatures notably less in the rigid continuous type thanin the other process industry types. Workforce involve-ment in shop-floor decision-making is adopted less in arigidly designed environment rather than one, wheremarket satisfaction requires process flexibility.Furthermore, multi-skilling, cross-training and rou-tinely dealing with change are required less in a rigid,limited variety process environment. The supplierintegration principle is considerably less populated inthe Figure 3(b) version of the ladder diagram thanFigure 3(a). Custom blending is populated the mostbut there is no discernible pattern across thecontinuum.
Figure 3(c), with a median threshold of 5.0 corre-sponding to ‘agree’ on the lean audit Likert scale,discerns no trend across the continuum. The elimina-tion of waste is still the most-widely adopted leanprinciple. Figure 3(d), with a median threshold of 5.5,is appreciably different to Figure 3(c). Very few leanpractices are apparent at this very high level of leanadoption.
The automotive exemplar provides a conspicuouscontrast with the process industry types. Even at the
5.5 median level (Figure 3d), a majority of leanpractices across each of the four principles are present.This reflects the maturity of lean adoption generally inthe high-volume automotive sector and specificallywithin this best-practice automotive plant.
7.3. Discriminant analysis of survey data
Linear discriminant analysis (LDA) is a statisticalpattern recognition technique for classifying items intomutually exclusive groups. An LDA was undertaken inorder to determine if the responses to all 32 statementsfrom the lean audit could be used to distinguishprocess industry types. The single fast batch data setwas included in the custom blending process industrytype for the purpose of this analysis. The classificationresults are shown in Table 9. The cross-validation ofresults compensates for an optimistic error rate.Overall, 74 out of 79, or 93.7%, were correctlyplaced. This demonstrates that, when taken in anaggregate form, the lean practices included within thelean audit differ across Dennis and Meredith’s (2000a)process industry types.
Figure 3. Lean thinking ladder diagrams.
486 A.C. Lyons et al.
An analysis of the ‘squared distance betweengroups’ revealed the shortest distance to be betweenthe rigid continuous and multistage continuous processtypes, indicating their relationship to be closest interms of the adoption of lean practices, and the longestdistance to be between multistage continuous andstock hybrid, indicating their lean practices to be themost dissimilar.
The Dennis and Meredith (2000a) sub-groups canbe aggregated into continuous (rigid and multistage),hybrid (stock and custom) and intermittent (fast batch,custom blending and process job shop) groups(Table 6). These groups were the subject of anotherLDA across the 32 lean audit statements. The resultsare shown in Table 10.
Overall, 73 out of 79 data sets were classifiedcorrectly. This allows for a small sample of intermittentdata sets which are likely to have affected the resultsnegatively. The lean audit statements with the highestdiscriminant functions (regression coefficients) and,therefore, contributed most to the level of discrimina-tion, in both LDAs were statements #3 (‘Production isregarded as make-to-order rather than make-to-stock’), #26 (‘The work environment is organised so
that most work is undertaken in teams’) and #27(‘Individual and team-based improvement ideas areregularly received from the workforce’).
7.4. Field studies
Field studies were undertaken at five sites (rigidcontinuous, multistage continuous, stock hybrid,custom hybrid and custom blending) in order tocorroborate, or otherwise, the findings from thesurvey and explore not only the implementation oflean thinking principles and practices in each of theprocess industry sub-groups but also their feasible levelof adoption and implementation. The single fast batchsite was discarded as a field study candidate. At eachsite, interviews with between four and eight staffformed one-third of a triangulation-based approach toknowledge acquisition which also included directobservation of the manufacturing, supply and distri-bution systems and a review of relevant documenta-tion, including manufacturing improvement strategies,annual reports and manufacturing and supply chainperformance analyses. Table 11 summarises the keycharacteristics of the field study environments.
The F1 site provides coating solutions for paperand imaging products. Production run times average 5days, equipment flexibility is limited and flow is uni-directional. Manufacturing is supported by structuredquality and process systems and procedures includingISO9000. The production process involves dispersingactive components into solutions in either aqueous ororganic media. These are then applied by a variety ofmethods onto a plastic or paper substrate in acontinuous process. Product formulations are com-plex. F1 production is not pulled or paced to a takttime. Manufacturing equipment is not dedicated, soproduction is mixed on the same line. There are few
Table 9. Summary of classification results with cross-validation.
ingredients and inventory points. Quantity discountsmotivate forward buying, resulting in large-lot deliv-eries to remote storage areas. This policy is beingreviewed to look at volume-based quantity discountsas an alternative in order to consider total purchasesover an annual period. Most ingredients are not single-sourced. The need to facilitate flow through theelimination of waste is familiar and understood bythe F1 workforce. Standard operating procedures aremature but although TPM is practised, it is not wellestablished. The workforce is involved in improvementactivities but does not do so autonomously. Despitebeing in a market environment where economies ofscale have previously been highly desirable, productlifecycles are progressively shortening. Insufficientattention is being paid to responding to this need byreducing production run lengths, and improvementactivities are not focused in this area.
F2 is part of a global pharmaceutical concern witha strategic focus on infectious diseases. The F2 site hasFood and Drug Administration (FDA, USA), MHRA(UK) and World Health Organisation regulatoryapprovals. Production of a narrow range of productsis based on campaigns which can last for manymonths. Flow is uni-directional but some parallelequipment options are present. Despite the site’s well-established continuous improvement initiatives andstaff development programmes, regulatory compliancecreates infrastructural inertia which can inhibit andslow change. There are many production steps but they
are generally physically connected and so result inrelatively few inventory points. The practices associ-ated with the alignment of production with demandgenerally rated poorly. Production is make-to-stockand the notions of ‘pull’ and takt time are notrecognisable within the facility. Production is mixedon the same facilities and despite there being noperceived need to reduce run lengths, production isbased on a series of consecutive steps from which amodified pull system in the form of a series of loops isfeasible. Overall, supplier integration rated poorly atthe F2 site but one ingredient has a short shelf-life andshould be JIT-driven. A key business imperative is toensure a high level of product availability but yield isunpredictable, so deliveries are made in large lots andbuffers are set cautiously. Ingredients are multi-sourced. Work instructions are conveyed via standardprocedures, quality assurance mechanisms are firmlyembedded and production areas are meticulouslyorganised, yet production downtime is excessive andTPM is only at an introductory stage. The productionenvironment is team-based, the workforce is creativelyinvolved in process improvement activities through thegeneration and execution of new ideas.
The F3 site is a contract manufacturing pharma-ceutical plant. It possesses FDA and MHRAapprovals. Production run times differ but are typicallyof 1 day’s duration. There are multiple work centresmost of which are non-dedicated; so, production ismixed using the same equipment and processes.
are evident throughout the facility with emphasis on
eliminating sources of variability. A structured pro-
gramme of employee training is in place to support a
multi-skilled workforce. Kaizen is at an embryonic
stage.Table 12 depicts a qualitative summary of the
adoption of lean principles within each of the field
study sites and identified opportunities for lean
improvement.
8. Discussion of results
The fundamental principles of lean thinking are beingadopted by process industries. This hypothesis, (H1), ispartially supported from the studies undertaken. The
second hypothesis, H2: lean principles are adoptedunevenly in different types of process industry, issupported. The results of the LDAs support the thirdhypothesis (H3): different types of process industry can
be discriminated by their lean practices. The LDAresults indicate that, when taken in an aggregate form,the lean practices included within the lean audit differacross Dennis and Meredith’s (2000a) process industrygroups and sub-groups.
The survey and the field studies demonstrate theenthusiasm all process industry types have for lean
practices associated with the principle of waste elim-ination. The unevenness of adoption arises with theremaining three principles. The survey suggests that thealignment of production with demand is adopted
moderately albeit with different practice profiles, butthe field studies indicate that the adoption of thisprinciple was considerably less in the rigid andmultistage continuous types. The survey and fieldstudies demonstrate that supplier integration is more
evident in multistage continuous, stock hybrid andcustom blending than in rigid continuous and customhybrid types. The creative involvement of the work-force in improvement activities is shown to have ahealthy adoption across all process types but is less in
the rigid continuous type than the others.However, to put lean adoption into perspective, the
automotive exemplar provided a conspicuous contrastwith the process industry types and demonstrates theextent to which lean principles can be adopted andpractices can be implemented. The lean performance in
the process industry types surveyed was affected by thevery limited adoption of a range of lean practicesacross the whole taxonomy. These were #2(‘Production is undertaken based on an instructionfrom a downstream process’), #10 (‘Deliveries are
based upon production requirements, are not excessiveand arrive just before being used’), #13 (‘Raw materialsand ingredients are single sourced’), #14 (‘Suppliershave flexible processes that can easily accommodatedemand changes’) and #29 (‘Management devolve
work-related decisions to the workforce’).Certain practices were found to have generic
applicability across the process types. 5S and visualsystems are broadly independent of structure and fallinto this category. In a ‘never stop producing’ (Melton2005) continuous process environment, the need for
effective maintenance practice is acutely felt, motivat-ing the use of TPM particularly in the rigid and
Production Planning & Control 489
Table
12.Theadoptionofleanprincipleswithin
thefieldstudysites.
Field
study
F1
(rigid
continuous)
F2
(multistage
continuous)
F3(stock
hybrid)
F4
(custom
hybrid)
F5
(custom
blending)
Alignmentof
productionwith
dem
and(A
PD)
Currentstatus
Low
Low
Moderate
High
Moderate
Leanim
provem
ent
Shorter
production
runs
Modifiedpull
system
loops
Productionlevelling
–Modifiedpull
system
loops
Integrationof
suppliers(IS)
Currentstatus
Low
Moderate
Low
Low
Moderate
Leanim
provem
ent
Volume-based
quantity
discounts
withsuppliersfor
smaller
batch
deliveries
Short
shelf-life
ingredients
tobe
JIT-driven
Supplier
relationship
managem
entfor
bettercontrolof
inventory
buffers,
vendor-managed
inventory
Contract
developmentwith
multiple
commodity
suppliers
Improved
flexibilityof
suppliers’
processes
Eliminationof
waste(EW)
Currentstatus
Moderate
Moderate
High
High
High
Leanim
provem
ent
Set-upreduction,
extendTPM
Fullintroduction
ofTPM
––
–
Creative
involvem
ent
ofthework-
force(C
W)
Currentstatus
Moderate
High
Moderate
High
Moderate
Leanim
provem
ent
Kaizen
activities
andteams
–Kaizen
activitiesand
teams
–Kaizen
activities
andteams
490 A.C. Lyons et al.
multistage continuous types. Standard operationsrequire a repetitive environment in order to be takenfull advantage of and so are not as appropriate in thecustom hybrid process type where repetition is mini-mal. Set-up reduction initiatives are justified whereregular changeovers are advantageous and were foundto be a significant contributor to manufacturingperformance in all except the multistage continuous(F2) field study, where campaigning characterisedproduction flow. Production pull requires a sequenceof consecutive and predictable steps. This can berealised in the discrete segment of flow, downstream ofthe point of discretisation, but not in the continuoussegment, where there are no intermediate stages fromwhich to facilitate pull. However, decoupling invento-ries in replenishment-based product streams createopportunities to use postponement-based techniquesand provide production loops to make pull systemsfeasible. Such a scenario exists in multistage continu-ous, stock hybrid and fast batch process types. Pullquantities need to be set conservatively in order toaccommodate yield variability. Production smoothing,or levelling, requires quick changeovers and was foundto have a particular opportunity for adoption in thestock hybrid process type.
9. Conclusions
Lean thinking is not necessarily better thinking but noindustry is immune to the relentless need to respond tocompetition by continuously innovating and improv-ing. In recent years, innovation and improvement inmanufacturing industries have been more conspicu-ously driven by lean thinking than any other initiative.This study set out to determine the extent to which leanthinking is being adopted as a manufacturing philos-ophy in different types of process industry. Research ofthis kind, undertaken from a taxonomical perspectivethat recognises the wide variety of process industries,begins to fill a gap in the process industry manufactur-ing and lean literature. The approach is based on adescriptive and quantitative analysis of a survey of 79process industry product streams supported by 18 sitevisits plus a qualitative 5-site field study.
Lean thinking is much more than the elimination ofwaste, and in this research, the notion of lean isarticulated by the alignment of production withdemand, the integration of suppliers, the creativeinvolvement of the workforce in improvement activi-ties in addition to the elimination of waste. Leanthinking, expressed by these four principles, is beingpartially adopted by process industries and, in overallterms, the inter-type diversity does lead to uneven
lean adoption. The elimination of waste is mostapparent across the process types followed by thecreative involvement of the workforce. These twoprinciples are essentially ‘managerial’ in nature, havean intuitive appeal, almost universal practicability andare less susceptible to issues associated with thephysical nature of the transformation system, regula-tory constraints, shelf-life and yield variability.However, they still require managerial commitmentand recognition of the value of the practices associatedwith these principles. The alignment of productionwith demand and the integration of suppliers are‘structural’ lean principles and less apparent across theprocess types. These two principles are more directlyinfluenced by the physical nature of the transformationsystem, regulatory compliance, shelf-life and yieldvariability, and their adoption is more immaturethroughout the process types. However, even forobdurate ‘structural’ practices, the field studies dem-onstrated that different process industry transforma-tion systems can be leaner and that yield, regulatoryand shelf-life constraints necessarily influencemanufacturing policy but should not be used asreasons for not judiciously adopting the alignment ofproduction with demand and supplier integration leanprinciples. Process industries have the potential to gainsignificantly more by adopting a more holisticapproach to lean thinking.
10. Limitations and future research
A methodological aim of the study was to identify anappropriate process industry classification system. Thisresulted in the use of the Dennis and Meredith (2000a)taxonomy. Only one fast batch process type was foundand no process job shops were found. Seventy-ninedata sets is a relatively small sample but one that wasconsidered sufficient for furthering insight, developingan understanding of lean thinking in process industriesand providing guidance and agenda-setting for furtherstudy, yet further work is necessary to confirm, orsuitably amend, the sub-group propositions within thetaxonomy.
The lean framework is not definitive and the leanprinciples and their associated practices were examinedon a like-for-like basis. The four principles wereconsidered to have equal weighting as were each ofthe 32 practices. Despite acknowledging the fact thatstructural differences in process industry sites arenecessary to support competitive priorities and order-winning criteria, no attempt was made to attach anyform of differential weighting to the principles andpractices embodied in the research methodology.
Production Planning & Control 491
The research provides an assessment of theincumbent situation for each product stream andfound that lean thinking concepts are widely diffusedthroughout the different process industry types.However, this study does not conclusively show thatlean, as defined by the four principles, is being used asa strategic value proposition. Focused surveys, longi-tudinal cases and action research are necessary tofurther explore how lean thinking can be fully tested ineach process industry type. The conclusions from thisresearch have provided a set of yardsticks as to theappropriateness of lean principles and practices, butlongitudinal cases can be directed to provide anassessment of the feasible lean limits in each type.Such an approach would allow performance andimprovement, concomitant with the introduction orextension of lean principles and practices, to bemeasured and managed.
Some evidence was apparent in both the survey andfield studies that innovative practices such as the use ofgeneric kanbans with a focus on flow paths rather thanparts are being used in hybrid process environmentsand innovative inventory management techniques arebeing used in continuous process environments.Further research is needed to explore if and how aninterpretative form of production pull can be reliablyachieved.
Notes on contributors
Andrew Lyons is a Reader inOperations Management at theUniversity of Liverpool ManagementSchool. Dr Lyons has wide-rangingresearch interests and undertakenresearch, consultative and teachingprojects in a number of industrialsectors in areas ranging from opera-tions strategy development to the
introduction of new supply network designs. He haspublished his work in a range of journals including theInternational Journal of Operations and ProductionManagement, Transportation Research Part E, ProductionPlanning and Control and the International Journal ofProduction Research.
Keith Vidamour is a CharteredEngineer and an EngineeringManager in the process industry, anindustrial sector. He has a passion forand has worked in most of his pro-fessional career. His professionalinterests concern manufacturingstrategy development using leanthinking principles, total productive
maintenance and innovative use of kanban and pull-basedmanufacturing techniques.
Rakesh Jain is a Professor in theMechanical Engineering Departmentat the Malaviya National Institute ofTechnology, Jaipur, India. Hisresearch primarily concerns processimprovement methods in processindustries with a particular emphasison the food sector, but he also hasmature research projects concerned
with collaborative approaches to new product developmentand in six sigma methods.
Michael Sutherland has a PhD inStatistics from Harvard Universityand is the former Director of theStatistical Consulting Center at theUniversity of Massachusetts. He hasdecades of experience as both a con-sultant and professor in the use ofmathematics and optimisation. He is asix sigma master black belt, has an
extensive list of professional clients and has providedstatistical input to a number of research projects that havebeen published in leading journals including the Journal ofHuman Evolution, Ecology, the American Journal of PhysicalAnthropology and Quality Engineering.
References
Abdulmalek, F.A., Rajgopal, J., and Needy, K.L., 2006. A
classification scheme for the process industry to guide the
implementation of lean. Engineering Management Journal,
18 (2), 15–25.Akkerman, R., van der Meer, D., and van Donk, D.P., 2010.
Make to stock and mix to order: choosing inter-
mediate products in the food-processing industry.
International Journal of Production Research, 48 (12),
3475–3492.Ashayeri, J., Teelen, A., and Selen, W., 1996. Computer-
integrated manufacturing in the chemical industry.
Production and Inventory Management Journal, 37 (1),
52–57.
Ballard, G., Harper, N., and Zabelle, T., 2003. Learning to
see work flow: an application of lean concepts to precast
concrete fabrication. Engineering, Construction and
Architectural Management, 10 (1), 6–14.Bhasin, S. and Burcher, P., 2006. Lean viewed as a
philosophy. Journal of Manufacturing Technology
Management, 17 (1), 56–72.Billesbach, T.J., 1994. Applying lean production principles to
a process facility. Production and Inventory Management
Oliver, N., Delbridge, R., and Lowe, J., 1996. Lean
production practices: international comparisons in the
auto components industry. British Journal of Management,
7, 29–44.Oliver, N., et al., 1994. World-class manufacturing: further
evidence in the lean production debate. British Journal of
Management, 5, 53–63.
Panizzolo, R., 1998. Applying the lessons learned from 27
lean manufacturers: the relevance of relationships manage-
ment. International Journal of Production Economics,
55 (3), 223–240.Pettersen, J., 2009. Defining lean production: some con-
ceptual and practical issues. The TQM Journal, 21 (2),
127–142.Piercy, N. and Rich, N., 2009. Lean transformation in the
pure service environment: the case of the call service
centre. International Journal of Operations and ProductionManagement, 29 (1), 54–76.
Ramarapu, N.K., Mehra, S., and Frolick, M.N., 1995. Acomparative analysis and review of JIT implementationresearch. International Journal of Operations andProduction Management, 15 (1), 38–49.
Roy, R.N. and Guin, K.K., 1999. A proposed model of JITpurchasing in an integrated steel plant. InternationalJournal of Production Economics, 59 (1), 179–187.
Sanchez, A.M. and Perez, M.P., 2001. Lean indicators andmanufacturing strategies. International Journal ofOperations and Production Management, 21 (11),
1433–1451.Seth, D. and Gupta, V., 2005. Application of value streammapping for lean operations and cycle time reduction: anIndian case study. Production Planning and Control, 16 (1),
44–59.Shah, R. and Ward, P.T., 2003. Lean manufacturing:context, practice bundles, and performance. Journal of
Operations Management, 21 (2), 129–149.Shah, R. and Ward, P.T., 2007. Defining and developingmeasures of lean production. Journal of Operations
Management, 25 (4), 785–805.Shingo, S., 1989. A study of the Toyota production systemfrom an industrial engineering viewpoint. Portland, OR:
Productivity Press.Simons, D. and Zokaei, K., 2005. Application of leanparadigm in red meat processing. British Food Journal,107 (4), 192–209.
Slomp, J., Bokhorst, J.A.C., and Germs, R., 2009. A leanproduction control system for high-variety/low-volumeenvironments: a case study implementation. Production
Planning and Control, 20 (7), 586–595.Spear, S.J. and Bowen, H.K., 1999. Decoding the DNA ofthe Toyota production system. Harvard Business Review,
September/October, 97–106.Taylor, S.G., Seward, S.M., and Bolander, S.F., 1981. Whythe process industries are different. Production and
Inventory Management Journal, 22 (4), 9–24.Womack, J.P. and Jones, D.T., 1994. From lean productionto the lean enterprise. Harvard Business Review, 72 (2),93–104.
Womack, J.P., Jones, D.T., and Roos, D., 1990. The machinethat changed the world. New York: Rawson Associates.
Wood, N., 2004. Lean thinking: what it is & what it isn’t.
Management Services, 48, 8–10.
494 A.C. Lyons et al.
Int. J. Production Economics 131 (2011) 194–203
Contents lists available at ScienceDirect
Int. J. Production Economics
0925-52
doi:10.1
n Corr
E-m
journal homepage: www.elsevier.com/locate/ijpe
Lean planning in the semi-process industry, a case study
Arnout Pool, Jacob Wijngaard n, Durk-Jouke van der Zee
Department of Operations, University of Groningen, P.O. Box 517, 9700 AV Groningen, The Netherlands
a r t i c l e i n f o
Article history:
Received 1 July 2008
Accepted 28 April 2010Available online 8 June 2010
The lean approach is an idealizing improvement approach that has an enormous impact in the field of
operations management. It started in the automotive industry and has since been widely applied in
discrete manufacturing. However, extensions to the (semi-) process industry have been much slower.
Resource characteristics of the (semi-) process industry obstruct a straightforward application. The
notion of the point of discretization for the (semi-) process industry is helpful here. This notion builds
on the fact that in most (semi-) process industries there is a point in production where process
production turns into discrete production. Downstream of this point lean principles are applicable in a
straightforward manner, while upstream lean needs to be interpreted in a more liberal way. In this
article we address this issue by a case study. The study considers how the principles of ‘flow’ and ‘pull’
production – suggesting a regular, demand-driven product flow – may be implemented for the (semi-)
process industry by introducing cyclic schedules. The conjectures guiding the case study are: (i) Cyclic
schedules fit in a lean improvement approach for the semi-process industry, (ii) Cyclic schedules help to
improve production quality and supply-chain coordination and (iii) Discrete event simulation is a useful
tool in facilitating a participative design of a cyclic schedule. The case study is extensively described to
be able to judge how the context of the changes and the intervention process contribute to the results of
the intervention.
& 2010 Elsevier B.V. All rights reserved.
1. Introduction
Lean starts from the refusal to accept waste (Womack et al.,1990). Different categories of waste are distinguished, such asinadequate processing, unnecessary transportation, excess mo-tion, less than 100% quality, waiting, overproduction andinventory (Ohno, 1988; Shingo, 1989). A whole framework ofprinciples, methods and tools has been developed to fight waste,see references Womack and Jones (2003), Hines and Rich (1997),Shah and Ward (2003), and Monden (1993).
The lean movement started in the automotive industry(Womack et al., 1990) and has since been widely applied indiscrete manufacturing. However, extensions to the (semi-)process industry have been much slower (Abdulmalek et al.,2006; Melton, 2005). Abdulmalek et al. (2006) explain this bypointing at product and/or process characteristics, which mayhinder a straightforward application of lean. For example,production efficiencies related to large product volumes mayhinder JIT production, whereas process flexibility determinesrelevance of techniques like Kanbans. Resource complexity of the(semi-) process industry obstructs a straightforward application
ll rights reserved.
.
of SMED and total productive maintenance (TPM) (Van Donk andVan Dam, 1996).
As an answer to the aforementioned ‘‘obstacles’’, Abdulmaleket al. (2006) suggest the notion of the point of discretization for the(semi-) process industry. This notion builds on the fact that inmost process industries there is a point in production whereprocess production turns into discrete production. Downstream ofthis point lean principles are applicable in a straightforwardmanner, while upstream lean needs to be interpreted in a moreliberal way. In this article we address this issue by a case study.The study considers how the principles of ‘flow’ and ‘pull’production – suggesting a regular, demand-driven product flow– may be implemented for the (semi-) process industry byintroducing cyclic schedules.
Cyclic schedules (‘‘Heijunka’’) have proven to be an effectivemethod to synchronize subsequent non-discrete productionstages (Glenday, 2006; King, 2009). According to Hall (1988) therepetition of intrinsic to cyclic schedules offers advantages bothfor shop floor activities and for planning. It may help to detectdisturbances earlier and reduce set up time and costs. Operatorsget a better hold on their processes, which fosters continuousimprovement activities. Cyclic schedules also enhance chaincoordination, as planners and operators as well as suppliers andcustomers get acquainted to the fixed schedule. In turn this savestime for coordination and enables an anticipatory attitude.
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203 195
The re-engineering of planning systems in process industries mayheavily rely on domain knowledge from various fields, especiallyproduction planning and control, and process engineering andoperation. Hence, a participative approach should underlie planningsystems redesign. This is in line with the lean concept of staffinvolvement in solution engineering (McLachlin, 1997; Warneke andHuser, 1995; Shah and Ward, 2003). Success of the approach largelyrelies on the use of adequate engineering, visualization, andsimulation tools, for both creating and communicating solutions(Detty and Yingling, 2000; Abdulmalek and Rajgopal, 2007).
The case study concerns the main Sara Lee plant for liquidcoffee. The company is considering a redesign of its planningsystem. Reasons for this are in improvements in the underlyingshop floor operations, perceived potential for a higher perfor-mance on customer service, and low efficiency of its planningprocedures and organization. The case study is meant to: (i)supply an illustrative example of implementing lean for the(semi-) process industry, (ii) gain an understanding of the waycyclic schedules may enhance efficiency and effectiveness ofplanning and (iii) highlight relevance of quantitative methods andprocess visualization for the success of a participative improve-ment approach. Basic conjectures underlying the study are:
�
Cyclic schedules of the process (continuous) part of productionfit in a lean improvement approach for the semi-processindustry. � Cyclic schedules help to improve production quality and
supply-chain coordination.
� Discrete event simulation is a useful tool to facilitate a
participative design of a cyclic schedule.
This article is organized as follows. The next section addressescompany characteristics before the intervention. Details of theintervention are discussed in Section 3 by considering projectbackground, objectives, improvement approach, organization, andoutcomes, i.e. changes to the current planning system. Next, Section4 describes the performance improvements as they resulted fromthe new planning system. Starting from the conjectures introducedabove, Section 5 discusses the causal mechanisms underlying theseresults. Finally, Section 6 summarizes main conclusions, andsuggests issues for further research.
2. Pre-intervention situation
This section introduces the case study plant, by considering itsproducts and markets, production process and current planningsystem.
2.1. Products and markets
The plant is the main Sara Lee plant for ‘‘liquids’’, i.e., coffeeextract used in coffee machines in the ‘‘out-of-home’’ market.Customers are twenty Sara Lee operating companies (Opcos) allover the world. Liquids are produced in about ten different blends.Three blends may be characterized as ‘‘general’’, addressing awide array of countries. Seven blends are more adjusted toregional preferences. Product volumes are dominated by a singleblend, being responsible for about 65% of the turnover. Packagingis tailored towards specific customers—see below. Two types ofplastic bags containing the liquids (1.25, 2 l) and many customerspecific carton boxes increase product variety drastically to about100 stock-keeping units (skus), see Table 1.
The largest Opco realizes 40% of the turnover, whereas sales forthe next largest Opco account for 10% of the turnover. So-called‘‘collaborating’’ Opcos allow the plant to view their stock levels, as
an input to the supply network planning (SNP). The remainderOpcos are characterized by their relatively small and rather lumpydemand. For these Opcos liquids are produced to order. Whereasdemand for specific skus may be volatile, aggregate demands perblend tend to be rather stable.
2.2. Process description
The production process can be distinguished into twosuccessive stages; process and discrete production, see Fig. 1.
The initial production step concerns the roasting and grindingof green beans, as they arrive from around the world. Roastedbeans are stored in large silos. Processing and storage capacity forthis step guarantee a steady supply for the next step, i.e.,extraction.
Extraction is a rather complicated process meant to absorb thecoffee extracts from the roasted and ground beans by usingboiling water. It is a continuous process in which a battery with12 cylinders is used according to a rotation scheme. The schemeassumes the periodic filling of cylinders with new ‘‘fresh’’ coffeebeans. Extraction is considered the most critical step in processproduction, due to the costs involved and its impact on blendquality. Blend quality is dependent on a careful tuning of processparameters. Costs of changeovers, i.e., product loss, and qualityissues force a tendency towards long runs of the same blend.However, due to volume restrictions in successive steps, runs aresubdivided in batches of 8, 12 or 16 cylinders. This furthercomplicates matters as – for reasons of traceability – it is notallowed to mix different batches of the same blend.
Evaporation and centrifugation are used as a next step forfurther concentrating liquid coffee in a batch wise manner. Therespective processes are decoupled by storage tanks. There isample evaporation and centrifugation capacity. However, mutualcoordination of these steps, and their tuning to extraction andpackaging, is nevertheless complex. This is due to restrictions onstorage capacity, the need for daily cleaning stops, and qualityconsiderations expressed in terms of time windows for inter-mediate storage.
Discrete production steps concern the filling of plastic bagswith liquid coffee and next the packaging of bags in customerdedicated printed carton board boxes. In principle, the speed ofthe packaging line, in terms of liters processed per time unit, ishigher than the speed of the preceding extraction process.Exceptions to this rule are certain high-yield blends, to beproduced in 1.25 l bags. Single long runs for these products resultin capacity shortages of the output tanks for the extractionprocess, as well as quality problems—due to the length of stay inthe tanks. Typically, suchlike runs are avoided by a packagingschedule, which alternates between both types of bags, taking duenotice of change-over times for the packaging line at the sametime (Table 2).
Roasting&Grinding
Extraction &Concentration Packaging Freezing &
Testing
< 1 week 2-3 weeks < 8 weeks
Transport
Process production Discrete production
Fig. 1. Production stages.
Table 2Change-over times packaging line.
Size of bags Blend Carton board
Mixing of subsequent blends allowed Mixing of subsequent blends not allowed
25 min 5 min 15 min 0 min
Demand plan
Shop floor planning(every week)
Shop floor monitoring andcontrol
(every day)
Shop floor
Deployment planning(every week)
Supply network planning(every week)
Fig. 2. Levels of the current planning system.
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203196
Packaged goods are transported pallet wise to a third partycompany in a nearby village. Here products are frozen for a week.After completing this period a sample is taken from eachproduction batch for final quality control. Defrosting and analysisof each sample take about one week for normal blends and twoweeks for high-quality blends.
Finally, finished and approved goods are transported to therespective Opcos. Transportation time varies from 1 day to 8weeks, depending on transportation distance and means oftransport.
2.3. Current planning system
The current planning system is based on MRP logic (Vollmannet al., 2005) and consists of four different levels of planning, seeFig. 2. The top level, i.e., supply network planning, concerns the
assessment of what needs to be produced in the upcoming weeksto fulfill demand in the supply chain. Shop floor planning and shop
floor monitoring and control determine when and how therespective product volumes will be produced. Finally,deployment planning is used to decide on distributing products tocustomers.
The Demand Plan serves as the main input for the supply
network planning. It reports estimated demands per Opco, at sku-level, for two years ahead. Planners use the ERP system SAP R3TM
on a weekly basis to generate an updated supply network plan(SNP) starting from the demand plan and information about stocklevels for collaborative Opcos. The SNP coordinates procurement,production and distribution on a week-by-week basis and has arolling horizon of 40 weeks. Given expected lead times, demandestimates for the first weeks of the SNP are fixed, i.e., not open forchanges any more. This period equals 5 or 6 weeks, depending onthe time required for final quality control, see process description.Effectively this means that production can be fixed for the next 3weeks, assuming quality control and transportation being re-sponsible for the remainder weeks. In week t, the SNP is used tomake a packaging plan for week t+4, which serves as input forshop floor planning. Based on the SNP, green coffee beans fromaround the world are ordered, well in advance of production.Moreover, the SNP is shared with the supplier of packagingmaterials to guarantee timely delivery.
Shop floor planning starts from the packaging plan to work outa detailed production plan for week t+1. This boils down toconverting planned output on sku-level into production orders foreach production step, see Fig. 1. Planning activities start bydetermining production orders for the extraction process bysearching for optimal run-lengths, batch sizes, sequences, andstarting times. Next, production orders for roasting and packagingare derived from these orders. Planning is supported by theadvanced planning system SAP APOTM in building a schedule.However, at the same time, scheduling heavily relies on theplanners’ in-depth knowledge of the production processes.Scheduling of the packaging line builds on a home-made MSExcelTM application. Among others, this application supports anadequate alternation of runs concerning 1.25 and 2 l bags.
Shop floor monitoring and control addresses the organization oftransport and supportive processes. It is meant to ensure asmooth flow of production orders by, for example, controllingtimeliness of transports to the third party company responsiblefor the freezing process and guaranteeing availability of packagingmaterial. Furthermore, direct communication with customers isused to deal with disruptions like product rejections and delayeddeliveries.
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203 197
At the end of each week, deployment planning decides on whichproducts are sent to which Opcos. This refers to products whichpassed quality control, see process description. The distributionplan is supported by SAP APOTM and accounts for all recentchanges in demand. In principle, orders placed by non-collabor-ating Opcos are prioritized. Furthermore, the transportation planforesees in efficient deliveries by considering the possibilities offull truck loads for larger Opcos and optimal combinations ofdeliveries for smaller Opcos.
Planning activities are executed by a team of nine planners.They are split into three different groups, i.e., the supply-chaincoordination group (SCCG) responsible for supply network anddeployment planning, the detailed planning and scheduling group(DPSG) responsible for shop floor planning, and the operationalcoordination group (OCG) responsible for shop floor monitoringand control. All three groups report to the logistics manager, whois a member of the management team of the plant.
In principle, the planning hierarchy is organized in astraightforward manner. In practice however, planning activitiesinclude several iterations involving much communication be-tween planners, both within groups and among groups. Essentialcauses of the respective (un)planned meetings are the uncertain-ties in customer demand, production processes, production yields,and quality. Uncertainties in production relate mainly to varia-tions in extraction times and machine breakdowns. Detection ofquality problems after freezing may have a big impact onproduction yield. Furthermore, customers, i.e., the Opcos, havegreat difficulty in making good estimates on their demand for fiveor six weeks ahead.
3. Intervention
3.1. Background
Following from a company-wide shift to lean thinking a ‘‘leanteam’’ is set up for the coffee plant. Its initial activities concern theimprovement of shop floor operations. The team realizesimprovements both with respect to logistics’ performance andproduction quality. An important measure is to fix productroutings, i.e., production steps are executed by a pre-specifiedset of manufacturing resources. Reducing routing flexibility in thisway both simplifies processes and guarantees more constantproduction quality. The changes with respect to shop flooroperations both enable and stimulate a different approach toproduction planning. Moreover, a thorough analysis of the currentplanning system reveals several shortcomings (Van der Zee et al.,2008):
�
Performance: High inventory costs due to excessive stocks forspecific products. Low service level for other products (out ofstock), for which (safety) stocks are insufficient for meetingcustomer demand. Further, customer delivery times areconsidered long (several weeks). � Planning logic: Planners and operators experience a high level
of system nervousness, caused by ongoing rescheduling andreplanning activities to respond to changes in the productionenvironment (machine breakdowns, varying yield, rejections,new orders etc.).
� Staff organization: The planning system tends to be labour-
intensive, involving many people and setting high require-ments to a correct tuning of their activities.
� Supportive systems: Next to the ERP system the company
maintains a poorly organized set of databases and spreadsheetapplications.
Main reasons behind these shortcomings seem to be themodular and complex structure of the current planning system.A new planning approach that links demand planning moredirectly with production is sought-after. Given the fact thatdemand for blends is rather stable (see Section 2), the idea ofcyclical scheduling seems promising (De Smet and Gelders, 1998).This should result in a more straightforward planning logic and amore robust production plan.
3.2. Cyclical scheduling
The idea of developing a cyclical planning system for theproduction of liquid coffee is supported by Glenday (2006). Hestresses that such an approach can be successful in the semi-process industry and in food production. He clarifies how ‘‘leveledproduction’’ and ‘‘EPEC’’ (¼every product every cycle) areimportant means to realize ‘‘flow’’ and ‘‘pull’’.
The causal mechanisms associated with the introduction ofcyclical scheduling – as foreseen at the start of the project – areillustrated in Fig. 3. Cyclical scheduling creates both productionregularity and coordination simplicity. This enables operationalstaff to better grasp their role in the total process, facilitating theirpartaking in process improvement (for example, optimization ofchangeovers and standardization of work-in-progress times).
Production regularity also increases reliability of the produc-tion process: product quality will be more constant, whereasoutput volumes and their timing are more predictable. In turn,higher reliability allows for a reduction of safety stocks.
Lower stock levels are expected to result from coordinationsimplicity as well. After all, planners can plan more accurately in aplanning environment, which is more simple and more transpar-ent. At the same time customer service will improve, as lowerstock levels imply higher product quality in terms of their bestbefore dates. Moreover, customers can better tune their internalprocesses with product deliveries, building on their knowledge ofthe cycle pattern.
Coordination simplicity also improves the efficiency of plan-ning. Instead of building plans from scratch each week, planningbecomes an exercise of filling in the blanks of the cyclicalschedule. Many repetitive tasks, like decisions about batch-sizingand sequencing, become superfluous. In addition, the cyclicalschedule can be used as a tool to coordinate planning tasks withinthe planning department by serving as a common format fordetailing production orders on various time horizons. Planningefficiency will improve beyond the planning department as well.Once operators on the shop floor are acquainted with the cyclicalschedule, there is less need for coordination with the planningdepartment. It is expected that a cyclical schedule does not onlymake production more lean, but planning itself as well.
3.3. Improvement approach
The development of a new planning system is a complex task,which heavily relies on the distributed skills and domain knowl-edge of managers, planners, and operators. This sets highdemands on project organization and supportive engineeringtools. Existing contacts with the operations management group ofthe University of Groningen (UofG) are renewed to receivesupport for this task.
Two teams are set up to develop the new cyclical planningsystem. The design team addresses the design of the planningsystem, whereas main focus of the second team is on itsimplementation.
The design team consists of the logistics manager of the plant;a representative of the lean team; the head of the supply-chain
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203198
coordination group (SCCG); the head of the detailed planning andscheduling group (DPSG); and three researchers of the UofG: oneexpert in production planning, one expert in discrete eventsimulation, and one junior researcher. The logistics manager is theproject owner, while the junior researcher acts as the projectleader, doing most of the preparatory and executive work,including simulation modeling and experimentation. He alsoserves as the linking pin for the implementation team. This teamconsists of foremen, planners, process engineers, and the juniorresearcher.
Next to more common means, like spread sheets, discreteevent simulation is heavily used to facilitate participative designof the planning system. This is motivated by its possibilities forvisualizing and testing of alternative set ups of the planningsystem, ahead of their actual introduction. Simulation useconcerns both the coded model and the underlying conceptualmodel—capturing scope and relevant detail for the system understudy (Law and Kelton, 2000). Graphical and textual visualizationsof conceptual models are used to facilitate joint solutionengineering in initial project phases. Coded models furtherimprove insight among project team members and stakeholderson system workings by visualizing its dynamics and allowing forits quantitative evaluation. Both conceptual model and codedmodel build on a clearly defined reference architecture, concern-ing a comprehensive set of building blocks for representingplanners, planning tasks and their interaction (Van der Zee andvan der Vorst, 2005; Van der Zee, 2006; Van der Zee, 2007).
3.4. Project overview
The project starts in March 2007. The time frame set for theproject assumes that effective use of the planning system shouldstart by September 1, 2007. The course of the project can beroughly divided into five successive phases (Fig. 4).
The initial project phase concerns an analysis of planning,production, and customer demand. This serves as input for theiterative development of a skeleton model for the new planningsystem, starting from the concept of cyclical scheduling. The
Cyclic schedules
Productionregularity
Coordinationsimplicity
Fig. 3. Expected causal mechan
Analyse currentsituation
Develop skeletonmodel Detail modelDesign
team
Implementationteam
March April May
Fig. 4. Projec
model visualizes and describes a planning hierarchy consisting ofplanning levels, supportive systems, and their interactions. Modeldefinitions build on a reference architecture for manufacturingsimulation (Van der Zee et al., 2008), see above. The notion of askeleton model supports the frequent meetings of the multi-disciplinary design team, by offering a starting point, focus andoverview for discussion and solution engineering.
The next step, i.e., ‘‘detail model’’, concerns the detailing of thelevels within the aforementioned planning hierarchy in terms oftasks, organization, information systems, and planning rules. Notehow model detailing boils down to ‘‘filling’’ the skeleton model.For example, basic decisions on cycle length and productsequencing are made during this phase. The overview offered bythe skeleton model is used to structure discussion and solutionengineering by focusing on specific model elements. In turn this isused to facilitate an ordered execution of design tasks. As a neteffect of detailing activities an initial design for the new planningsystem is defined. This design is tested for logic and completenessby a structured walk through of all planning system elements.
June marks the start of activities for the project implementa-tion team. Initial activities of the team concern the validation ofthe planning system design. Among other things, this resulted inminor adaptations of the blend sequence in the cyclical schedule.Moreover, a major task of this team concerns making operators atthe shop floor familiar with the new system and fostering theirinvolvement (McLachlin, 1997). Team discussions, both amongteam members and with other company staff, are enabled andfacilitated by displays of the design.
During the last two months of the project both teams work onseparate tasks. The project implementation team continues toprepare for the implementation of the new planning system. Thisincludes communication of changes in shop floor control tooperators and reorganizing the planning department. Meanwhilethe design team seeks to further refine the initial planningsystems design, being supported by a coded simulation model.Simulation is used to study the logistics performance of the newplanning system. Scenarios considered include alternative de-mand patterns and process parameters (for example product
Processimprovement
Stock reduction &Customer service
improvement
Planning efficiencyimprovement
Operational
in vo lvement
Ope ra tio na la nd
cust omerinvolvem
ent
isms of cyclical scheduling.
Discuss planningsystem design Fine tune new planning system
Implement new planning systemDiscuss planningsystem design
June July - august
t phases.
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203 199
yield, machine down times, and extraction processing times). Inthis way the team is able to estimate effects of the cyclicalschedule on stock levels, throughput times, and work-in-progresswaiting times. Moreover, different rules for the timing ofevaporation and centrifugation and the alternation of bag sizeson the packaging line are tested. Finally, on September 1, 2007,the new planning system is launched successfully.
3.5. The new planning system—main changes
The new planning system builds on the notion and relevanceof the point of discretization in the production process(Abdulmalek et al., 2006). Fig. 5 summarizes the maincharacteristics of production stages at either side of thedicretization point. The respective characteristics underlie thechoice of planning concepts for the new planning system.Planning of extraction, which is characterized as processproduction, follows the concept of cyclical scheduling, whileplanning of the packaging activities, which are discrete by nature,is based on run-out times on sku-level. This is meant to create‘‘regular flow’’ and ‘‘demand pull’’, respectively. Below, bothconcepts are detailed. Next their implementation is discussed bylinking planning tasks to a planning hierarchy. Finally, planningorganization is considered.
3.5.1. Planning process production—cyclical scheduling
Following the concept of ‘‘Heijunka’’ planning activities forprocess production start from a generic schedule for a given timeperiod (Huttmeir et al., 2009), see Fig. 6. The schedule has a fixedduration, i.e., one week, and consists of two parts, i.e., a fixed partconcerning cyclical blends and a variable part concerningremainder blends. The choice of cycle length follows from theway the supportive ERP system is implemented and the need totune activities within Sara Lee and with supply-chain parties. Thedominant planning and communication cycle is one week. In linewith the EPEC principle it is strived to include as many blends aspossible in the fixed part of the schedule. Nevertheless, someblends need to be produced in the variable part following fromtheir demand pattern (irregular or small turnover). Next to theseblends the variable part accounts for experimental products,weekly equipment cleaning operations and extra runs of cyclicalblends (due to short-term deviations in estimated demand).
Cycle execution foresees in fixed start and completionmoments for the overall cycle as well as its parts. It is hopedthat such ‘‘forced regularity’’ contributes to process improve-ments. The price to pay is that in case of delays productionoutputs are not in agreement with the schedule. Becausevariations in extraction processing times, and occasional break-downs or stops may hinder completing the fixed part of theschedule in time, a so-called ‘‘breathing blend’’ has been defined.Run length of this blend may be adapted at the shop floor toguarantee a timely completion of the fixed part.
Processproduction
Small number of blendsStable demand per blendFixed batch sizesLong runsSequence restrictionsWork in process time andvolume restrictions
Fig. 5. Main characteristics on bot
3.5.2. Planning discrete production—run-out times
Whereas mix flexibility for the process production is small, it isrelatively large for the discrete production, given ample capacityand relatively small change over times. This allows to closely linkdiscrete production and customer demand, and – hence – create ademand pull. Relevance of mix flexibility especially relates to thecollaborating Opcos, as they are supplied from stock. Note howorders from non-collaborating orders are covered in a direct wayby including them in the variable part of the cycle.
Packaging orders referring to collaborating Opcos are based onthe ‘‘run-out time’’ concept. Run-out time for an sku is defined asthe estimated period demand may be covered from the netavailable stock, i.e., the actual stock level minus the re-order level.According to the run-out time concept, skus with zero or negativerun-out times are prioritized. Re-order levels are defined in termsof average demand for a pre-specified number of weeks. Choicesmade reflect (uncertainties) in replenishment lead times andcustomer preferences.
The run-out time concept is implemented in terms of threesuccessive activities. First, output of the fixed part of theupcoming cycle is allocated to skus in such a way, that run-outtimes of skus are leveled per blend, starting from the sku with theshortest run-out time. Next run-out times for all skus arecalculated. This allows planning orders for the variable part ofthe cycle, starting with the blend that has the sku with theshortest run-out time. Order assignment continues until produc-tion capacity of the variable part of the cycle has been filled orwhen all skus indicate positive run-out times.
For implementing both planning concepts a three levelplanning hierarchy has been defined. It distinguishes betweenmid-term planning, short-term planning, and the shop floor, seeFig. 7. Below elementary planning tasks and interactions will berelated to each level.
Essentially, the task of mid-term planning concerns matchingproduction and demand on a three monthly basis. This periodlength is related to foreseen quality of demand estimates perblend and maintenance policies—suggesting periodic down times,due to execution of major maintenance activities.
Key activities concern (i) the determination of the cyclicschedule and (ii) the choice of stock re-order levels for all skusrelating to collaborating Opcos. Relevant cycle parameters includethe choice of cyclical blends, their sequencing, their run-lengths(number of batches), and the starting time of the cycle (which dayof the week). Re-order levels for stock items, i.e., skus relating tocollaborating Opcos are determined by considering estimateddemand for the next three months. This may involve consultingcustomers on service levels and demands and process engineerson production quality issues.
Short-term planning is executed on a weekly basis. It concernsthree main tasks, i.e., (1) scheduling the variable part of the cycle,(2) relating blend output to sku’s, and (3) scheduling of roasting
Discreteproduction
High number of skusVariable demand per skuVariable batch sizesShort runs
h sides of discretization point.
FixedFixedVariableFixed
Fixed part(cyclical blends)
(bre
ath i
ngbl
end)
Variablepart
1 week
Time
Duration
Production
Fig. 6. Layout of the cyclical schedule.
Mid-term planning(every 3 months)
Short-term planning(every week)
Shop floor(every day)
Demand plan
Actual demand
- Fixed part of cycle- Re-order levels
- Variable part of cycle- Packaging plan
Fig. 7. Levels of the new planning system.
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203200
and packaging activities. The first two tasks follow from the run-out times concept and orders from non-collaborating Opcos, seeabove.
Planning activities are executed shortly before the start of anew cycle. At that moment all required information (stock levels,work-in-process, orders, estimated demand, and quality controlassessments) are downloaded from SAPTM. An ExcelTM application(LP-solver) has been developed to support the choice of blends tobe produced during the variable part of the cycle and, next toallocate output to packaging orders, i.e., sku’s. Next, blends to beproduced within the variable part of the cycle are sequenced,according to preset guidelines for minimizing sequence depen-dent product loss.
The schedules for roasting and packaging are created when theextraction schedule for the complete cycle is finished. Withrespect to packaging, attention is given to an adequate alternationof 1.25 and 2 l bags during production of blends with a high yield.
At the shop floor level, operators in process production areresponsible for process control in terms of determining theprecise sequencing of evaporation and centrifugation activities,and their timing. Furthermore, they have to be responsive todisturbances within the extraction process. Finally, they have tocontrol the run length for the breathing blend, see above.Operators in discrete production have some freedom in tuningthe packaging step with the output of extraction. This includesdecisions on the start and speed of packaging activities.
3.5.4. Planning organization
The redesign of the planning system also requires a newplanning organization. Mid-term planning concerns a jointactivity of the head of the planning department, a member ofthe SCCG group, a process engineer and a representative from theshop floor. The main short-term planning task, i.e., determiningproduction of the variable part of the cycle and output on
sku-level, is accomplished by a single person from the SCCGgroup. Next, constructing the production schedules for extraction,packaging and roasting is done by a single person from theDPSG group. Executing planning tasks on the shop floor is theresponsibility of process operators.
4. Results
This section reports the results of the three months pilot of thenew planning system from September to November in 2007. Thefollowing elements will be discussed (cf. Fig. 3):
�
Degree to which cyclic schedules are enforced, � Changes in production regularity, � Process improvement, � Changes in coordination simplicity, � Stock reduction and customer service improvement, � Efficiency improvement in planning and control.
4.1. Enforcement of cyclic schedules
During the pilot this cyclic schedule is kept rather precisely;each week it starts on time with the fixed sequence and run-length of cyclical blends. Run-time variations during extractionare compensated with run-size variations of the breathingblend of up to 10% of its planned run-length. As a result thecompletion time of the fixed part of the cycle does not vary morethan 15 min.
Note how enforcement of the schedule addresses timing andsequencing of production activities according to a pre-specifiedpattern. This leaves the possibility of alternative patterns. In fact,two alternative week patterns are implemented, implying adifferent choice of blends for the fixed part of the cycle. Weekpatterns are executed alternately. This freedom of choice allowsto reflect demand characteristics (for example, blends with stablebut moderate turnover) and technological considerations (forexample, sequence restrictions and minimum batch sizes).
4.2. Changes in production regularity
Production regularity mainly depends on two factors: reg-ularity in sequence of blends and regularity of in-process times fordifferent batches of the same blend. Maintaining a cyclic scheduleincreases sequence regularity; it is fixed during the fixed part ofthe cycle and for the variable part of the cycle preset guidelinesare agreed to keep sequence as constant as possible.
In-process times with an important quality impact are timebetween evaporation and centrifugation and time betweencentrifugation and packaging. The standard deviation of both in-process times is reduced with about 2 h. Based on the cyclicalschedule it is possible to establish a better routine for evaporationand centrifugation. Moreover, operators are better enabled toanticipate. As the production rate of packaging (measured in time
Table 3Summary of results.
Lean elements Results
Cyclic schedule Cyclic schedules are enforced.
Production regularity More regularity in both sequencing and in-
process times.
Process improvement Yield, quality and energy use seem to be
improved.
Coordination simplicity Better understanding of the planning system of
both planners and operators.
Stock reduction & customer
service improvement
Potential not realized as customers are not yet
fully integrated.
Planning efficiency
improvement
Less planners, meetings, and iterations needed
to set up plans.
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203 201
per batch) depends on blend specific yield and bag size, focusingon more constant in-process times demands high flexibility inpackaging. This requires a different attitude of operators at thepackaging line. Instead of packaging as quick as possible, thespeed of the line should now match the speed of extraction. Formost blends this implies that the packaging line is working atmoderate speed, allowing for short stops between successivebatches. While this may conflict with the perception of thepackaging operators, it does not lead to quality problems, as longas the stops do not last for more than half an hour.
4.3. Process improvement
Increased regularity in blend sequencing results in closermonitoring of changeovers at extraction. Production realizes asmall increase in yield and a substantial reduction of variation inyield. Moreover, energy costs are declining. Operators tend tosubscribe these changes to the introduction of cyclical production,but the pilot has been too short to draw final conclusions here.The same holds for product quality. Significant effects are not yetquantifiable, but it appears that product rejection rates due toprocess variations have decreased.
4.4. Changes in coordination simplicity
Coordination simplicity refers to the degree to which planning asa process is more simple within the planning department and easierto comprehend beyond its borders. The concept of the cyclicschedule makes planning more simple in several ways. Difficultdecisions about sequencing, batch sizes, and re-order levels nolonger need to be made every week, but only once every threemonths. Moreover, short-term planning does not have to start fromscratch every week. It boils down to an exercise of filling in theblanks of the cyclic schedule, starting from the concept of run-outtimes and by using a set of pre-defined guidelines.
The general impression is that both operators and planners aremore aware of the material flow. Operators have a more activerole in production planning and thus have a better understandingof the way their decisions influence the synchronization ofsubsequent production steps. Activities of short-term plannersare linked more closely to execution. Hence they are more awareof their scheduling consequences.
4.5. Stock reduction and customer service improvement
The delivery reliability for non-collaborative Opcos is definedas the percentage of the deployment plan that is realized. Thisfigure has not changed. For collaborative Opcos the local skuinventories are relevant. There is a norm for the safety stock of3–5 weeks (depending on the blend). During the pilot it ispossible to keep the safety stocks right on this norm, while beforethe pilot, the safety stocks varied from 0 to 8 weeks. Thisimproved control of stocks results in a reduction of outdatedstock.
The changes in the new planning system are not discussedexplicitly with the Opcos. This is partly due to the fact that withinSara Lee the contacts with the customers are the end responsi-bility of value stream management (VSM), a unit that does notbelong to plant management. That makes the decision processabout changes in these relationships more complicated.
4.6. Efficiency improvement in planning and control
The new planning system requires less iterations for setting upproduction plans relative to the current system. Before the
production plans were the output of rather mechanisticMRP type calculations. In-attractive scheduling consequenceswere discussed of course, but only if planners of the DPSGgroup asked for adaptations. In fact short-term productionplans can now be constructed by a single person in a straightfor-ward manner, as described in Section 3.5.4. This requires someextra training in scheduling for a SCCG planner. But as DPSGplanners are still required to plan production of the remainderproducts, i.e., so-called ‘‘instants’’, this task is still assignedto them.
The choice to support the sku allocation by an ExcelTM LP-solver model forces the definition of a new decision routine—asthis is not embedded in SAPTM. Downloading, allocation, anduploading of production data necessitate a new business process.Working in cycles with a week pattern that starts on Fridaycomplicates the DPSG task a little. Planners have to upload theproduction runs of two subsequent cycles (Monday to Thursday ofone cycle and Friday to Sunday of the next one) to SAPTM, becauseSAPTM works with calendar weeks.
4.7. Summary of results
The major results as described above are listed in Table 3.
5. Discussion
In this section the conjectures formulated in Section 1 arereconsidered starting from project results. We conclude bydiscussing some further insights following from the specifics ofthe project set up.
5.1. Cyclic schedules for continuous production fit in a lean
improvement approach for the semi-process industry
The guiding principles for the redesign of the planning systemare ‘‘regular flow’’ of the process part of production and ‘‘demandpull’’ for the discrete part of production.
The need for regularity of the non-discrete part of productionis met by the application of cyclic schedules. They reducevariation of two major production factors: change-over patternand work-in-process times. Moreover, cyclical schedules are inline with ongoing lean projects aiming at dedicating resourcesand standardizing working procedures, see Sections 3.1 and 3.2.
In order to realize ‘‘demand pull’’ in the discrete part ofproduction the concept of cyclical scheduling is linked with theconcept of run-out times. By allocating production to specificcustomers as late as possible both planning and production followdemand in a direct way. Moreover, planning and control
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203202
processes become more simple and transparent. In sum, cyclicschedules fit in a lean improvement approach in this semi-processproduction situation.
5.2. Cyclic schedules help to improve production quality and supply-
chain coordination
Cyclic schedules help to improve production quality in at leasttwo ways:
�
The fixed sequence of blends in cyclic schedules causes amore regular change-over pattern in process production.This enables operators to optimize changeovers forextraction, learning from their experiences in each successivecycle. � Fixed time frames associated with the cyclical plan help
operators to control their workload. Consequently, work-in-process variation is reduced, and hence, production quality isbetter (controlled).
Apparently, cyclic schedules improve production quality in anindirect way. Operators play a crucial role as they both have torealize and exploit the regularity of the scheme.
Implementing the concept of cyclical scheduling contributes toa better tuning of supply-chain operations, concerning both intra-firm activities and those of supply-chain partners, i.e. Opcos:
�
Intra-firm: Linking the regularity of the cyclical plan with theresponsiveness of the run-out time policy allows planners touse actual information regarding customer orders and sku-stocks in the allocation of production. They are able to do so,without compromising efficiencies in process production.Improved tuning is reflected by a better control ofwork in process, and stocks for packaged liquids (skus).Obviously, these improvements are not for free – (significant)changes are required relative to the current planning system,see Section 3. � Supply-chain partners: In principle, improved stock control
allows for lower re-order levels. So far, this has not beenrealized due to administrative hurdles. Currently, a separatevalue stream management organization is responsible for thecontacts with the Opcos. As the project is restricted to theplant operations, it is not possible to make these changes atthis stage. Nevertheless, the results of the project can be usedas a firm basis for reconsidering the current agreements withthe Opcos.
5.3. Simulation is a useful tool to facilitate a participative design of a
cyclic schedule
Cyclic schedules reduce manufacturing flexibility by definition.It is important to find ways such that this flexibility reduction hasonly minor effect, while the increased regularity has full effect.This necessitates a thorough participative design process, bringingtogether the firm’s domain experts. Discrete event simulationturns out to be useful here serving as a common means fordisplaying, discussing, creating, and testing alternative solutionsfor their logic and performance. Its utility, however, relies onsound conceptual modeling highlighting key choices in planningsystems design in a structured, transparent, and complete way.The project makes clear how the initial planning systems design ismainly a net outcome of the conceptual modeling for simula-tion—leaving its fine tuning to the coding and experimentalphases.
5.4. Some reflections on project set up
Next to the insights based on the effects of implementingcyclic schedules, insights can also be derived from the project setup itself. The company wide attention for lean planning and EPEC(every product every cycle) contributes to the willingness of theplant management to start the project and to allocate resources.The combination of a design team and an implementation teamworks well to foster participation and communication. Thepartaking of the chairman of the lean team and the logisticsmanager (also management team member) are important torealize changes and attune the project with other ongoingprojects. The participation of the junior researcher from theuniversity of Groningen in both teams is essential to coordinatetheir progress. The aim of building a detailed simulation model todesign, analyze, and support the new control structure helps ingetting the right focus in the project meetings. The planning andcontrol expertise of the university members of the design team ishelpful in deciding to deviate from SAPTM.
6. Conclusions and suggestions for further research
Cyclical scheduling fits well in a lean improvement approachin semi-process production. The distinction between continuousproduction and discrete production is important here (seeAbdulmalek et al., 2006). Cyclical scheduling helps to realizeregularity in the continuous part of production. Its simplicity andtransparency lead to a closer coordination of the planning andcontrol processes with the production processes. Improvementsare conditional on adequate organizational arrangements withrespect to change management (communication and involve-ment) and business process changes (McLachlin, 1997). Thenecessary business process changes include changes in the waythe ERP software is used. This complicates the intervention,certainly in a multi-national company like Sara Lee.
Discrete event simulation is useful in the detailed design of thecyclic planning and control structure. It is necessary to make thisdesign participative and simulation is very suitable in realizingthis. Preparing the simulation and especially the process ofconceptual modeling is just as useful as using the output of thesimulation (Robinson, 2004).
While the case study confirms the conjectures formulated inthe introduction, it raises some new questions for furtherresearch, as well. In the first place regarding the quality and yieldimprovement effect: how do the involvement of operationalpeople and the characteristics of the technology influence theimprovements? In case of a well standardized technology, thepositive effect of more regularity may be conjectured to besmaller than in case of new and not yet standardized technology.A next point that needs more research concerns organizationalefforts required for the development, use and maintenance of ERPwork-arounds: what do comparable experiences with respect todownloading and uploading tell about the complexity of im-plementing and maintaining such procedures? A third point ofresearch is the possibility to realize production lines with variablespeed. The speed of the packaging line has to vary – includingsmall stops – to synchronize it with extraction. This conflicts withthe way of working of the packaging operators. How to keepoperators effective and efficient in such a situation?
The effect of cyclical scheduling cannot be estimated inisolation. It is intertwined with the effects of more employeeinvolvement, more organizational and inter-organizational inte-gration, new decision support systems, etc. Some of the effectscan be estimated beforehand, by analysis or simulation. For someother effects empirical evidence can be found. Results on the just
A. Pool et al. / Int. J. Production Economics 131 (2011) 194–203 203
mentioned research questions may also be added here. But manyof the effects are very situation specific. This case study confirmsthat there is a high idiosyncrasy of the improvement measuresapplied and of the effect of these measures. Case studies, ifsufficiently rich described, give insight in this idiosyncrasy andmake it possible to discuss the improvement measures and theireffects in general terms. That is what we have been trying to dohere as well.
Acknowledgments
The authors would like to thank Ric Oud, Romke Tijsma, Tjarkvan Heuvel, Rimmer van der Hoek, Martin Balt, Dick Winkeler,Gert Bazuin, Ria de Ruiter, Sjoerd van Koten, Derk Tiesma, andBinne Jan Kramer for their involvement in the project andcontributions to the research.
References
Abdulmalek, F.A., Rajgopal, J., Needy, K.L., 2006. A classification scheme for theprocess industry to guide the implementation of lean. Engineering Manage-ment Journal 18 (2), 15–25.
Abdulmalek, F.A., Rajgopal, J., 2007. Analyzing the benefits of lean manufacturingand value stream mapping via simulation: a process sector case study.International Journal of Production Economics 107 (1), 223–236.
De Smet, R., Gelders, L., 1998. Using simulation to evaluate the introduction of aKanban subsystem within an MRP-controlled manufacturing environment.International Journal of Production Economics 56–57 (1), 111–122.
Detty, R.B., Yingling, J.C., 2000. Quantifying benefits of conversion to leanmanufacturing with discrete event simulation: a case study. InternationalJournal of Production Research 38 (2), 429–445.
Glenday, I, 2006. Breaking Through to Flow. Lean Enterprise Academy, Ross-on-Wye, UK.
Hall, R.W., 1988. Cyclic scheduling for improvement. International Journal ofProduction Research. 26 (3), 457–472.
Hines, P., Rich, N., 1997. The seven value stream mapping tools. InternationalJournal of Operations & Production Management 17 (1), 46–64.
Huttmeir, A., Treville, S., Ackere, A., Monnier, L., Prenninger, J., 2009. Trading offbetween heijunka and just-in-sequence. International Journal of ProductionEconomics. 118 (2), 501–507.
King, P.L., 2009. Lean for the Process Industries—Dealing with Complexity.Productivity Press, New York, NY.
Law, A.M, Kelton, W.D., 2000. Simulation Modeling and Analysis. McGrawHill,Boston.
McLachlin, R., 1997. Management initiatives and just-in-time manufacturing.Journal of Operations Management 15 (4), 271–292.
Melton, T., 2005. The benefits of lean manufacturing—what lean thinking has tooffer the process industries. Chemical Engineering Research and Design 83(A6), 662–673.
Monden, Y., 1993. Toyota Production System—An Integrated Approach to Just-In-Time 2nd ed. Industrial Engineering and Management Press, Norcross, GA.
Ohno, T., 1988. The Toyota Production System: Beyond-Large Scale Production.Productivity Press, Portland, OR.
Robinson, S., 2004. Simulation—The Practice of Model Development and Use..Wiley, Chichester, UK.
Shingo, S., 1989. A Study of the Toyota Production System. Productivity Press, NewYork, NY.
Shah, R., Ward, P.T., 2003. Lean Manufacturing: context, practice bundles, andperformance. Journal of Operations Management. 21 (2), 129–149.
Van der Zee, D.J., van der Vorst, J.G.A.J., 2005. A modeling framework for supplychain simulation—opportunities for improved decision-making. DecisionSciences 36 (1), 65–95.
Van der Zee, D.J., 2006. Modeling decision making and control in manufacturingsimulation. International Journal of Production Economics 100 (1), 155–167.
Van der Zee, D.J., 2007. Developing participative simulation models—framingdecomposition principles for joint understanding. Journal of Simulation 1 (3),187–202.
Van der Zee, D.J., Pool, A., Wijngaard, J., 2008. Conceptual Modelling forParticipative Simulation—A Case Study on Lean Planning. Proceedings of theOperational Research Society Simulation Workshop (SW08), 117–127 Birming-ham: Operational Research Society.
Van Donk, D.P., Van Dam, J.P., 1996. Structuring complexity in scheduling: a studyin the food-processing industry. International Journal of Operations andProduction Management 16 (5), 54–63.
Vollmann, T.E., Berry, W.L., Whybark, D.C., Jacobs, F.R., 2005. ManufacturingPlanning and Control for Supply Chain Management 5th ed. McGraw-Hill, NewYork, NY.
Warneke, H.J., Huser, M., 1995. Lean Production. International Journal ofProduction Economics. 41 (1–3), 37–43.
Womack, J.P., Jones, D.T., Roos, D., 1990. The Machine that Changed the World.Free Press, New York, NY.
Womack, J.P., Jones, D.T., 2003. Lean Thinking: Banish Waste and Create Wealth inyour Corporation 2nd ed. Free Press, New York, NY.
Robust Process Control at Cerestar's Refineries
KUMAR RAJARAM Anderson Graduate School of Management, UCLA,Los Angeles, California 90095. USA
R. JAIKUMAR • Graduate School of Business, Harvard University,Cambridge, Massachusetts 02138, USA
FRANZ BEHLAU Cerestar Benelux BV, Nijverheidsstraat 1, 4551
LA, Sas Van Gent, The Netherlands
FRANS VAN ESCH Cerestar Benelux BV ' |
CORRIE H E Y N E N Cerestar Benelux BV
RALPH KAISER Cerestar Deutschland GmbH, Dusseldorfer Strafie
191, D- 47809 Krefeld, Germany
ALBERT KUTTNER Cerestar Deutschiand GmbH \
IZAAK VAN DE W E G E Cerestar Deutschland GmbH
With annual sales of over $2 billion, Cerestar is Europe's lead-ing manufacturer of made-to-order wheat- and corn-basedstarch products. Cerestar relies on refineries that are highly au-tomated and require large fixed investments. Starting in 1993,we developed Robust Process Control (RPC) to increase aver-age throughput and reduce throughput variation by combin-ing engineering principles with OR/MS techniques. RPC in-cludes a mathematical-programming model to reducedowntimes due to product switchovers, models for process op-timization, and dynamic control models for process-flow syn-chronization. Cerestar implemented the resulting decision sup-port system at eight refineries in six countries. It has increasedaverage daily throughputs by 20 percent and reduced averagethroughput variation by 50 percent. Concomitantly, the refiner-ies have reduced their consumption of supplies and utilities. Inaddition to over $11 million in annual benefits, RPC has hadmajor strategic and organizational impact.
Large-scale process-based operationsare prevalent in almost every major
industry. For instance, most processes inthe intermediate-food-processing, pharma-ceutical, paper, and petrochemical indus-tries fall into this category. They are con-structed to reliably produce large volumesof well-established product, using conven-tional and tested technology.
Cerestar is Europe's leading manufac-turer of made-to-order wheat- and corn-based starch products, such as glucose,sorbitol, dextrose, and gluten, with annualsales exceeding $2 billion. From 1927 to1987, it was part of the Corn Products Re-fining Company (CPC), New York. In1987, it was incorporated as a separatecompany under the Italian group Ferruzi,and in 1994, due to corporate restructur-ing, it became a company of the Frenchgroup Eridania Beghin-Say. During thistime, Cerestar has grown to become a ma-jor producer of these products, which areused extensively as components in thefood-processing industries (for example,by breweries, confectioneries, and baker-ies), the consumer-product industries (forexample, cosmetics and toothpaste), andsuch other industries as paper, pharma-ceuticals, textiles, and specialty chemicals.
To produce these products, Cerestar op-erates over 20 different types of industrial-scale processes in 16 plants located in ninecountries. These can be broadly classifiedinto physical processes, such as refining,separation, grinding and extracting, andchemical processes, such as hydrogenatingand modifying starch products. Sincebuilding these processes requires majorcapital investments, it is crucial that theyconstantly produce high volumes of out-
put at the correct quality level To achievethis goal, these processes rely on high de-grees of automation, operate continuously,and usually shut down only a few times ayear for scheduled maintenance. As prod-uct switchovers result in long downtimes,products are produced in long campaigns.
Cerestar has a rich tradition of develop-ing innovative approaches to process man-agement. During the mid-80s, it imple-mented a framework for processimprovement called Social Technical Sys-tems (STSs). STSs have been widely recog-nized as benchmarks for process improve-ment, and they have been adapted forsimilar processes in many industries. Un-der these systems, automation at theseprocesses was standardized and unified ina single centralized control room. This leddirectly to drastic reductions in workforces and increases in productivity. Exist-ing operators were retrained to f>erformnew tasks or reassigned to new processes.
Upside variation indicates apotential for increasingthroughput.
In particular, adoption of STSs increaseddaily average throughputs to levels wellabove the capacities estimated in engineer-ing specifications. However, there was stillsignificant variation in day-to-day outputsat several processes. Downside variation isextremely expensive due to several factors.It requires larger quantities of finishedproduct inventory and typically leads togreater per-unit costs for energy, supplies,and environmental degradation. In addi-tion, such variation causes greater attritionof physical components and lower
January-February 1999 31
RAJARAM ET AL.
customer-service levels, and it impedes thedevelopment of process knowledge asvaluable operator time is spent on "firefighting."
Upside variation indicates a potentialfor increasing throughput. Such increasesare crucial in this industry because prod-ucts are commodities with market-definedprices. Consequently, firms can increasetheir margins only by reducing the unitcost of production. Recent trends in thisindustry include consolidation, increaseddemand for products due to growing mar-kets in east Europe and Asia, and tighterproduct specifications. In light of thesetrends, Cerestar needed to quickly in-crease both the capacity and the capabili-ties of its processes to maintain its domi-nance in the industry. Realizing these
Cerestar operates over 20types of industrial-scaleprocesses in 16 plants in ninecountries.
goals through the traditional strategy ofbuilding new plants was no longer a via-ble option, because the costs of automa-tion, information, and control, a significantproportion of total costs, had greatly in-creased. In addition, demand in these newmarkets was not yet stable, and large in-vestments could be very risky. Finally,Cerestar might lose a large portion of mar-ket share during the time it would take tostart up these new processes. These factorscompelled Cerestar to explore methods ofincreasing the reliability and output of itsexisting process without making signifi-cant investments in new processtechnology.
To achieve these objectives, DavidChallenor, the executive vice president incharge of manufacturing, put together aninternal team in September 1992. Membersagreed that Cerestar's production plan-ning and process control were excessivelycomplex and lacked a scientific and sys-tematic approach to problem solving. Theprocess-automation systems provided datain huge volumes. However, these datawere not being used efficiently to improveprocess productivity because there wereonly a few common standards, insufficientdiffusion of ideas across plants, and lowretention of knowledge and experiencewithin the organization. The team mem-bers thought that these factors contributedto the tremendous variation in outputsacross several processes.
When the team reported its findings inMay 1993, Cerestar decided to collaboratewith Jai Jaikumar, at the Harvard BusinessSchool, and Kumar Rajaram, then at theWharton School. Their research focused onimproving the productivity of large-scaleindustrial processes without significant in-vestment in new process technology. Theybased their approach on combining engi-neering principles with OR/MS-basedtechniques. In June 1993, Cerestar formeda task force comprising the authors to im-prove the performance of these processesby applying these techniques. We focusedon the refining processes because refinedproducts accounted for a large part oftotal profits. We chose the glucose refineryat Sas Van Gent in The Netherlands as ourtest site because it was the flagship refin-ery of the company with the most sophis-ticated process and automation technol-ogy. If we could achieve improvements at
INTERFACES 29:1 32
CERESTAR
this refinery, our techniques could clearlyresult in improvements at all the otherrefineries.The Operating Environment
Glucose is the generic term given to awide variety of sugars produced fromstarch. The glucose-refining process atCerestar converts starch slurry to glucose.The slurry first goes through an incuba-tion stage, where it is treated with en-zymes to convert it to glucose at a certainsugar or dextrose-equivalent (DE) level.The glucose is then purified by filtration,made colorless with decolorizers, madetasteless by polishers, and reduced in wa-ter content through evaporation. Individ-ual refineries at Cerestar vary in the typeof technology they use to achieve this con-version and in the number of parallelstages at each step.
These glucose products are used as in-termediate products in various food-processing industries to produce suchproducts as candy, beer, and soft drinksand also in other industries to producesuch products as paper and pharmaceuti-cals. Customers specify their requirementsby DE level. Rather than produce smallruns for each customer and incur exten-sive downtimes due to switchovers, Ceres-tar developed many years ago a simplebut then revolutionary idea of producing afew types of products across a range of DElevels. It then meets customer require-ments by blending these glucoses, calledbasic grades. This enabled Cerestar to con-duct long production campaigns withoutcompromising its ability to accuratelymeet customer demand. Cerestar producesbasic grades in a predefined sequence,stores them in tanks and draws them from
the tanks and mixes them in blenders tomeet demand. Since switching from onebasic grade to another still causes signifi-cant downtimes, minimizing switchoversis crucial. In minimizing switchovers, Cer-estar had to consider how much storage toprovide, how to allocate storage across thebasic grades, and when to conduct whichcampaign for basic grade production.
Control of the stages of the refinery isautomated through centralized process-control systems. These systems are broadranging and fiexible, allowing the user toset and change many parameters at eachstage. Broadly speaking, based upon theirfunctionality, we can partition these pa-rameters into two groups. The primaryfunction of the parameters in the firstgroup is to provide a stable operating en-vironment for the process stage. These pa-rameters, such as temperature and pres-sure, are set within a well-defined andnarrow range determined by design andsafety considerations. The process-controlsystem uses regulators to automaticallymaintain these settings within this range.The second group of parameters specifythe rate at which each stage should be op-erated. The process-control system usescontrol variables to set these rates. Opera-tors determine the setting of these controlvariables to best match productionrequirements.
The major focus of engineering para-digms of control at any stage is to deter-mine the partition of parameters into regu-lators and control variables and todetermine the optimal setting for the regu-lators. To determine these, we developedscientific models based on the physics,chemistry, and mechanics of that stage to
January-February 1999 33
RAJARAM ET AL.
explain its operation while assuming thecontrolled environment of the laboratory.Under such an assumption, setting thecontrol variables to meet a production rateis straightforward. However, in practice,we may have to alter this scientific modelto account for the interaction effects be-tween the stages forming the process andfor the operating environment at the plant.In this environment, stages typically lag inreacting to changes in the settings of con-trol variables. Capacity imbalances mayexist within a process because the compo-nents in the process may not be made tothe same specifications, and some may re-quire periodic downtimes because of thenature of the chemical processes involved.In addition, operators must act quickly tomeet short-term production requirements.
Since September 1994, theautomated system has run therefinery 95 percent of thetime.
Because of these circumstances, opera-tors have difficulty understanding the pro-cess well enough to choose and set controlvariables to achieve high levels of outputconsistently and to base their choices onscientific and systematic approaches toproblem solving. Instead, they rely onsubjective expertise, may be preoccupiedwhile setting control variables, and pay in-sufficient attention to routine maintenanceand to the manual portions of the opera-tions. These practices may lead to signifi-cant variation in outputs. For instance, atthe Sas refinery, although average dailythroughput was 345 tons, well in excess ofthe 300 tons per day capacity estimated
from engineering specifications, this out-put varied between 445 and 90 tons perday.
Our approach was to blend engineeringprinciples with OR/MS techniques in theoperating environment of the plant. To doso, we developed a framework called Ro-bust Process Control designed to increaseaverage throughputs and reduce through-put variation without major investmentsin new process technology. Cerestar firstimplemented this framework in Sas VanGent between 1993 and 1994 and subse-quently implemented it at seven other re-fining processes in five countries between1994 and 1997.Robust Process Control
Robust Process Control (RPC) consistsof four steps. We first simplify the process,then we stabilize the bottleneck, synchro-ruze the other stages v dth this bottleneck,and standardize procedures to ensure thatwe maintain gains.
Step 1: Process simplification. To sim-plify the process, we reduce its operationalcomplexity by investigating existing pro-cedures and performing tests to reduce re-work, recycles, and transients at individ-ual stages of the process. Reducing thesedisruptions makes the process more pre-dictable and increases the duration of thesteady state. In these processes, it is diffi-cult to switch from one basic grade to an-other because regulator and control-variable settings at each stage must bereconfigured in a coordinated way. Conse-quently, nunimizing the number of switch-overs diminishes operational complexity.To reduce the number of switchovers, weused optimization-based models (de-scribed in the appendix) to determine the
INTERFACES 29:1 34
CERESTAR
SCIENTIFIC MODEL
Use physical principles ofscience to model
manufacturing phenomena
STATISTICAL MODEL
Distinguish information fromnoise, estimate parameters of
the scientific model andvalidate control procedures
CONTROL MODEL
Distinguish regulators and controlvariables and optimize control
variable settings of the scientificmodel
Figure 1: The process simplification step of Robust Process Control consists of a scientificmodel, a statistical model, and a control model. These models are jointly developed in the op-erating environment of the plant.
size of the storage tanks, the allocation ofbasic grades to tanks, and when to switchfrom one grade to another IRajaram 1998].A simplified process is a transparent pro-cess. It is now easier to detect the bottle-neck, and optimization models based onsimple first-order physical relationshipsare sufficient to stabilize it.
Step 2: Process stabilization. To stabiUzethe process, we develop an optimizationmodel to stabilize the bottleneck, translatea production target to the required fiowsat the bottleneck, and calculate the settingof the control variable that best achievesthis flow. The architecture of this optimi-zation model consists of a scientific model,a statistical model, and a control model(Figure 1). The scientific model describesthe operation of a stage based on a com-
prehensive examination of the operatingmechanism at this stage. To develop thismodel, we focus on simplicity and first-order interactions and ensure that we canaccurately measure the variables in themodel. These variables need to be part ofthe existing control system and compatiblewith the safety tolerances of the process.To estimate the parameters for the scien-tific model, we develop a statistical modelbased on data gathered in independentsamples at the correct physical points inthe process. The control model is used tochoose the control variable and its settingsto best achieve the fiows needed to meetthe production target. This model is devel-oped from the validated scientific modelIJaikumar and Rajaram 1996].
Step 3: Process synchronization. To syn-
January-February 1999 35
RAJARAM ET AL.
chronize the stages in the process, we en-sure that the requirements at the bottle-neck are alv 'ays met by stabilizingoperations at the other stages and syn-chronizing the flows at each of thesestages to those at the bottleneck. We dotbis by using a procedure that calculatesthe buffers and the steady state flow re-quired in real time at each stage to com-pensate for regeneration downtimes(appendix).
Step 4: Process standardization. The firstthree steps of this framework ensure asimplified, stabilized process with a fewcontrol variables set to ensure that flowsettings are synchronized to the bottle-neck. To standardize the process, we de-velop a decision support system, whichautomates the computation required in theprevious steps. In effect, during steadystate the process completely runs itself likethe autopilot in an aircraft. The role of theoperator is to maintain a log of activitiesto record abnormalities, to understandwhy they occur, and to deal with extremecontingencies. Operators have more timeto do maintenance, to detect faults insmall nonautomated parts of the process,such as pumps and motors whose fail-ure could be extremely disruptive, andto develop an understanding of the pro- ,cess, key to further technologicalinnovation.Implementation
We first implemented RPC at Cerestar'sglucose refinery at Sas Van Gent. This re-finery produces over 150 types of prod-ucts, blended from six basic grades. Therefinery at Sas is expected to operate con-tinuously, except during the four plannedmaintenance shutdowns each year. Aver-
age throughput was around 345 tons perday, well above the capacity of 300 tonsper day estimated from engineering speci-fications. However, day-to-day variationsin this throughput were around 25 percentof this average. Since demand is quite sta-ble, this variation is largely due to theproduction-planning and process-controltechniques used at this refinery.
The refining process at this plant com-prises 10 stages controlled sequentiallyand concurrently by three operators, eachresponsible for a fixed and prespecifiedgroup of stages (Figure 2). The first opera-tor controls the process through to the firstbuffer tank, the second from the decolori-zation until the second buffer tank, andthe third the remaining portion of this re-finery. In the production-planning andprocess-control architecture prior to imple-mentation of RPC (Figure 3), the produc-tion planner transformed customer de-mand to demand for basic grades anddetermined the duration of the productioncampaigns required by specifyingproduction-planning targets. Each opera-tor then translated this plan into flow re-quirements and chose the control variablesand settings for his or her parts of the pro-cess. The architecture after implementationof our four-step approach (Figure 4) con-sists of production-planning, input, pre-scriptive, and output modules. Each mod-ule was developed during the differentstages of RPC and integrated to form theautomated system, which since September1994 has actually run the refinery 95 per-cent of the time. The remainder of the timerepresents downtimes due to maintenanceshutdowns or mechanical breakdowns.During these nontypical periods, operators
INTERFACES 29:1 36
CERESTAR
FRONTEND
INCUBATION
FILTRATION
iBUFFER TANK
DECOLORIZATION
iPOLISHING
BUFFER TANK 2
iEVAPORATION
STORAGE TANKS
iMIXING
Figure 2: The glucose-refining process at SasVan Gent comprises 10 stages controlled se-quentially and concurrently by tbree opera-tors, each responsible for a fixed and prespeci-fied group of stages. Tbe first operatorcontrols the process through to the first buffertank, tbe second from the decolorization untiltbe second buffer tank, and the third the re-maining portion of this process.
control the process manually.The first step we took in developing this
system was to simplify the process by re-ducing rework at the decolorizers and pol-ishers, eliminating recycles at the filters,and reducing switchovers between basicgrades. Finally, we used the model de-scribed in the appendix to determine thetotal storage volume required at the end ofthe process and its allocation across basicgrades. The portion of this model that dy-namically determines when to switch to adifferent campaign is part of theproduction-planning module (Figure 4).This information is then passed on to theinput module. By simplifying the process,we ensured that disruptions arising fromswitches of products, filters, and ion-exchange regeneration at the decolorizerand polishers are predictable and mini-mized in frequency and duration. With in-creased stability, we identified the second-stage evaporator as the bottleneck.
To stabilize the process, we increasedthe throughputs and reduced throughputvariation in this evaporator based uponthe optimization model described byJaikumar and Rajaram [1996]. Using thismodel, we can translate the productiontarget into flow required at the evaporatorand determine the setting of the best con-trol variable to achieve these flows. Westabilized the decolorizers, polishers, andfilters using similar models and synchro-nized their flows to the evaporator usingthe procedure outlined in the appendix.Jaikumar and Rajaram 11997] discuss thesemodels and the estimation of the timingand duration of regeneration downtimesin detail. These actions reduced the num-ber of control variables from 44 to four.
January-February 1999 37
RAJARAM ET AL.
PRODUCTIONPLANNING:
TARGETSOPERATOR 1
STAGESITO 4
OPERATOR 2STAGES
5 TO 7
\\
Reaction Arc
Information Arc
OPERATOR 3STAGES8 TO 10
PROCESS CONTROL SYSTEM
REFINING PROCESS : STAGES 1 TO 10
Figure 3: In the production-planning and process-control architecture prior to implementationof Robust Process Control at the Sas Van Gent glucose refinery, the production planner trans-formed customer demand to demand for basic grades and determined the duration of the pro-duction campaigns required by specifying production-planning targets. Each operator thentranslated these targets into flow requirements and chose the control variables and settings forhis or her parts of the process.
The control variables eliminated werechanged to regulators and set at their opti-mum levels.
This reduction in control variables had aprofound impact on the operational com-plexity at this refinery. It now employsfixed standards during all shifts and nolonger depends so heavily on the subjec-tive expertise of the operators. The opera-tors can now monitor the remaining con-trol variables more effectively and betterunderstand the cause and effect in theoperation of each stage. This knowledgeand the reduction of control variablesmeans fewer changes during the operationof this process, which drastically reduces
the variation at each stage and at the
output.To standardize and preserve these
changes, we incorporated all the calcula-tions required by these models in the pre-scriptive module of the architecture (Fig-ure 4). The input module collects the datato drive these models from the process-control system based upon a half-hourlysampling frequency. The output modulepasses the flow requirements and the con-trol settings on to the process-control sys-tem. We developed these modules usingthe C ^ programming language andlinked them to Extend [1992], a softwarethat offered a user-friendly interface and
INTERFACES 29:1 38
CERESTAR
the capability to perform off-line simula-tions. This provided us with an opportu-nity to simulate and test this system beforeimplementation.
Since implementation, downtimes dueto unplanned disruptions at tbis refineryhave been reduced by over 70 percent.One can better appreciate this change byobserving the flows at the filters, polishers,and evaporators before and after imple-mentation (Figures 5 to 7) during a ran-domly sampled period of four days. Aver-age daily throughput at the refinery hasincreased by over 18 percent from a baselevel of 345 tons per day and day-to-daythroughput variation has been reduced byaround 60 percent. To understand the im-pact of these changes, compare the daily
throughput during a two-month periodbefore Cerestar initiated this project withthe daily throughput during a two-monthperiod after implementation (Figure 8).
A major challenge we faced during im-plementation was to understand and rede-fine the role of the operator in tbis newsystem. It was crucial that the operatorsbelieved in the system and did not feelthreatened by it. Consequently, duringeach stage of development, we activelysought and incorporated their suggestions.We elected to provide them with tbe flexi-bility to overrule the recommendations ofthe model if they prepared a detailed re-port explaining tbe reasons for their ac-tions. While the operators were initiallyskeptical about the ability of this system,
PRODUCTIONPLANNING
MODULE
(
INPUTMODULE
^ \
PRESCRIPTIVEMODULE
^ - ^ - ^
OUTPUTMODULE
Reaction Arc
Information Arc
PROCESS CONTROL SYSTEM
REHNING PROCESS : STAGES 1 TO 10
Figure 4: The production-planning and process-control architecture after implementation of Ro-bust Process Conlrol at the Sas Van Gent glucose refinery consists of production-planning, in-put, prescriptive, and output modules, which are integrated to form an automated system. SinceSeptember 1994, this system has actually run the refinery 95 percent of the time.
Figure 5: Outflow variation at the filtration stage of the glucose-re fin ing process at Sas VanGent has been significantly reduced after implementation of Robust Process Control.
they embraced this approach once it wasproven in the plant. They were extremelypleased that they no longer needed tospend their time determining settings tothe control variables and fire fighting. Thecurrent role of the operator is to improvethis system, to develop better understand-ing of the process, to spend more timeperforming routine maintenance, and tomonitor the smaller nonautomated por-tions of the process. This strategy has hadhandsome dividends. Disruptions due tomechanical breakdowns have been re-duced by more than 50 percent. Betterprocess understanding has led to furthertechnological innovations, which has low-ered the consumption of supphes and util-ities at several process stages.
In September 1994, the system for RPCbecame operational at the Sas refinery.Since then we have been implementingthis concept in Cerestar's refineries inSpain, France, the United Kingdom, andItaly, and af its three sorbitol refineries in
Germany (Figure 9). By October 1996, Cer-estar was using our system to run all theseprocesses. In carrying our approach to theseven refineries, we observed that theoperations people at these processes hadsimilar engineering backgrounds to thoseof Sas and, like them, had a lot of know-how about the process. Thus, we were notsurprised to find their attitudes toward thenew system comparable to those encoun-tered at Sas. Here again, operators used alot of unproved theories to defend certainconcepts. We came across older and skep-tical process pioneers who argued thesetheories based upon seniority instead ofproven data from the plant. To convincethese people that our approach was valid,we often resorted to describing examplesand showing the specific results fromother projects.
In addition, we found it crucial to un-derstand the organizational differences inthese countries to understand what wecould accomplish and when. In particular.
INTERFACES 29:1 40
CERESTAR
the cultural differences at these plantsplayed an important role in determiningthe character of tbe implementation teams.Broadly speaking, teams ranged from self-standing groups of people completelycommitted to this approach to groups ofpeople who wanted to look for compro-mises, who thus required more guidelines.The biggest burdle in implementation
turned out to be in configuring the soft-ware to account for differences in automa-tion capacity at each plant. We overcametbis problem through the support of tbeboard of directors at the corporate head-quarters in Brussels and of the manufac-turing directors responsible for each plantwho redeployed software and automationengineers from other projects to work on
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
Time (Hours)
Figure 6: Outfiow variation at the decolorization stage of the glucose-refining process at SasVan Gent has been significantly reduced after implementation of Robust Process Control.
Figure 7: Outflow variation at the evaporation stage of the glucose-refining process at Sas VanGent has been significantly reduced after implementation of Robust Process Control.
January-February 1999 41
RAJARAM ET AL,
0 5 10 15 20 25 30 35 40 45 50 55 60
m IFigure 8: Average daily throughput has increased while throughput variation has decreased af-ter implementation of Robust Process Control at the Sas Van Gent glucose refinery.
Glucose Refinery: Sas
Other Four GlucoseRefineries
Sorbitoi Refinery: NonCrystalline Line
Sorbitol Refinery; Crystaland Special Lines
g- > c 15 ^ "5w Z -> S 5 ">
Figure 9: The implementation schedule of Robust Process Control at eight Cerestar refineriesfrom June 1993 to September 1996.
this project. In the final analysis, this levelof support was one of the most importantfactors in overcoming the not-invented-here syndrome; it was the key to achiev-ing the results.Results '
Since October 1996, RPC systems haverun eight refining processes at Cerestar atleast 90 percent of the time. Basic gradeswitches have been reduced by an averageof around 40 percent, resulting in an aver-age reduction of downtime by over 15 per-
cent. The number of control-variable op-tions for the operators has been reducedby an order of magnitude. Disruptionsarising from incorrect specification ofcontrol-variable settings (leading to con-gestion and stops at process stages) orfrom mechanical breakdowtts have de-creased by more than 60 percent across allthese processes. On average, dailythroughput has increased by more than 20percent and throughput variation has di-minished by around 50 percent. Results at
Table 1: The number of basic grade switches and control variables has been nofably reducedafter implementation of Robust Process Control al eight of Cerestar's refineries. This has drasti-cally reduced operational complexity and resulted in dramafic reduction in disruptions.
Table 2: After Robust Process Control was implemented at eight of Cerestar's refineries, aver-age daily throughput has increased significantly, while the coefficient of variation at these eightrefineries has been notably reduced.
January-February 1999 43
RAJARAM ET AL.
has reduced the consumption of supplies,such as enzyn\es, reagents, catalysts, andother chemicals used in operating these re-fineries. In addition, savings in such utili-ties as energy and water have been esti-mated to be around $3.5 million annually.Finished goods inventory has been re-duced by over 30 percent. Service levelshave been increased, and customers haverewarded Cerestar with larger contractsover extended periods. Cerestar is nowapplying the same concepts in other pro-cesses. Implementation at a wheat-grinding process has already increasedyields by over five percent by reducingconsumption of raw materials by around15 percent. This process has also reducedits consumption of water and energy byover 15 percent. These improvements havebeen valued at over $2.5 million annually.Promising results are expected from proj-ects started at four starch-modificationprocesses and five corn-grindingprocesses.
The reduction of variation hasreduced the consumption ofsupplies.
The strategic impact of RPC on Cerestarhas been substantial. RPC has providedCerestar with the ability to produce spe-cialty products at the cost of commodities.It has improved the tolerances of basiccommodity grades to such an extent thatCerestar can now make specialty productseffectively by using existing blending pro-cedures. In addition, because RPC can runexisting processes with higher capacityand better precision, Cerestar has beenable to buy plants and run their processes
more efficiently, a better alternative thanconstructing new, more expensive pro-cesses. This strategy was strongly affirmedduring January 1996. Cerestar purchasedAmerican Maize Products, a company thatruns a network of starch-processingplants. It expects to run the existing pro-cesses at these plants more profitably us-ing this approach. Currently, Cerestar isimplementing RPC in several of the pro-cesses at this company and has alreadyobtained remarkable results. Finally, be-cause RPC minimizes automation, control,and information technology, Cerestar hasundertaken major capacity expansions ofcurrent processes in a cost-efficient man-ner. For example, Cerestar is planning toexpand the capacity of the wheat-grindingprocess at Sas by around 250 percent byinvesting $150 million; building a newprocess to achieve comparable capacitygains would cost around $250 million.
The organizational impact of RPC hasbeen tremendous. In his annual presenta-tion to the board in 1996, David Challenor(the intemational manufacturing director)said, "Robust process control is transform-ing us into a learning organization. Everyprocess problem is now viewed as an op-portunity for learning and process im-provement. We do not just gather data; weconvert these data to valuable informationfor process analysis. Decisions are basedon this information instead of allowingopinions or anecdotal evidence to dictatefuture actions. Standardization of opera-tional procedures is now a priority:proven control concepts are implementedand are not subject to personal interpreta-tion. Now, in all our projects, we strive to-ward simplicity. We would rather be near
INTERFACES 29:1 44
CERESTAR
optimal and stable during a long periodthan try to be optimal always and achieveoptimality for extremely short periods.This said, 1 must emphasize that we arenever satisfied with the status quo and arealways striving for improvement withoutadding complexity. In this effort, we arebeing aided by all our operators. This sys-tem has transformed our operators fromfire fighters to process innovators"IChanenorl996,p.5].
The success of this project has motivatedmanufacturing to look more closely atother processes and other areas, includingproduction planning, product costing, andprocess design. This has fostered its closecooperation with several other areas in theorganization, including marketing, fi-nance, and engineering. As Challenornotes, "In addition to the obvious eco-nomic benefits and the impact on ourthinking, this work has improved thespirit of team work and communication inour multinational organization. We are ex-tremely optimistic about the future of thisproject, not only due to its promise inother types of processes, but to its poten-tial to help us focus on the managerial andstrategic decisions required to guide usthrough the next millennium" [Challenor1996, p. 51.
In summary, RPC has had a major eco-nomic, strategic, and organizational im-pact at this company. Cerestar expects tomaintain the gains we described and to in-crease them continuously several yearsinto the future.Dedication
This paper is dedicated to the memoryof R. Jaikumar, who passed away whilemountain climbing in Ecuador on Febru-
ary 10,1998. Words cannot express oursorrow and deep appreciation of Jai's con-tribution to this work.Acknowledgments
We are indebted to many people for theintellectual contributions, support, and en-couragement they provided during thiswork. Although this list is very long, somemust be mentioned individually. In partic-ular, we offer our deep appreciation to theinternational manufacturing director,David Challenor. He was the person whoconceived this project, provided us withthe resources, and constantly encouragedand supported the ideas presented in thispaper. We also thank J. Massot, manufac-turing director, Cerestar Spain, and M.Natale, manufacturing director, CerestarItaly. We express our deep gratitude to allthe process-development managers andengineers, plant superintendents and oper-ators at these processes who provided uswith valuable process information and ac-cepted and tested our ideas. Without theirsupport, these gains would not have beenpossible. After implementation, they stillcontinue to furnish valuable feedback, cru-cial to the continuous improvement ofthese systems. We thank Stephen Gravesfor his comments on this paper. We be-lieve they greatly increased the clarity ofexposition of our work.APPENDIXModels for Optimizing Product Switches
To minimize the number of productswitches, we calculate the total end-process buffers (tanks) required and deter-mine how to allocate products (basicgrades) to these tanks to minimize basicgrade switches based on a long-run aver-age of demand. We also develop a proce-dure to correct for deviations from this av-
January-February 1999 45
RAJARAM ET AL.
erage and determine when to switch andto which grade to switch while runningthe process in real time. To model theproblem of choosing the total number oftanks and the allocation of basic grades totanks, we define the following variables;
/ ^ ll,.. .m}: Index set of basic grades,/ = {1 «}: Index set of tanks,D,: Long-term average of demand for basicgrade / per time period,Vy Size of tank; in volume units,V: Total available volume,C;,: Cost per unit volume (including space,installation, tank and basic grade holdingcosts),S/: Switchover cost for product /.
The volume selection problem (VSP) isrepresented as follows:
(VSP) Z = Min S,Z(IO + QVV > 0.
In this problem, we trade off the cost ofvolume with the costs due to downtimesbecause of basic grade switches. The num-ber of switches Z{V) is the solution to thevolume allocation problem (VAP) definedas follows:
M
(VAP) Z(IO ^
/ = ]
a^yE {0,11, V/,/.
During each production campaign, wewould ideally produce enough of a basicgrade to fill up the allocated tanks beforeinitiafing a switch. This would ensure thatproduct switches are minimized. In ourapplication, all possible switches betweenbasic grades are feasible, and the switch-ing times between these grades are identi-cal. Violation of this assumption would re-quire us to make significant modificationsto include the effects of sequencing in this
model [Rajaram 19981.It is easy to solve VSP once we solve the
subproblem VAP. However, this problemis highly intractable due to the integervariables that are nonlinear in the objec-tive. Consequently, we elected to developa heuristic to address this problem. Thisheuristic consists of two phases. In thefirst phase, we solve the following contin-uous version of this problem (CVAP).Note that this provides a lower bound onVAP:
(CVAP) r'\'" D
Min 2 T7t = 1 Vi
.=1
V, > 0. . .
This problem is easily solved by settingthe tank allocation to each basic grade as
However, the continuous solution maybe infeasible due to tank-batch-size con-straints enforced by Vy. To derive a feasi-ble solution, we rank order set / in increas-ing order of demand to form set /'. Weconstruct two sets A and 6, which form aparfition on/ ' , and either \A\ - I B I orIAI + 1 = I BI. The basic grades in Ahave lower demand and are thereforemore sensitive to demand variation. Con-sequently, we provide more safety stockfor these grades by rounding up the con-tinuous solution to the next feasible solu-tion. Conversely, we round down the con-tinuous solution for the basic grades in B.This approach performed remarkably wellfor the parameters defined by these nineprocesses. In all these cases, the solutionprovided by this heuristic was within twopercent of the upper bound provided bythe continuous approximation. As Rajaram
INTERFACES 29:1 46
CERESTAR
11996] discussed, this method performs fa-vorably with other randomly generateddata sets.
Once we have determined the volumeand allocation, we correct for deviationsfrom the long-run average demand anddetermine when to switch grades and towhich grade to switch on a real-time basisusing the following method. To develop aprecise definition of this procedure, we de-fine the following:
^i,mA\ (respectively, V,n,jn): The maximum(respectively, minimum) permissible vol-ume for the (th basic grade, defined basedupon the sizes of the allocated tanks,Vi/. The actual total volume of the basicgrade at these tanks,,max (respectively, Djmin): The maximum
(respectively minimum) daily demand forthe ith basic grade, estimated from theshort-term planning horizon,k,: The daily production rate for the ith basicgrade,tj^t'. Time required to start up the ith basicgrade including switchover times (in hours)(this is independent of the current grade un-der production),f,, : Time required to shut down the ithbasic grade under production.
For the /th basic grade, we develop thefollowing disjunctive constraints:
1/ -L
"
' ^/,min/'i
241/ 4- '."wx^i>t ^ 17•' M) "*" ryt — ' i,min*
(1)
(2)
Constraint (1) enforces the conditionthat while this grade is being produced,actual volumes at the tank and the maxi-mum expected buildup during shutdownshould always be lower than its maximumpermissible volume. Conversely, while an-other grade is being produced, the actualvolume of this grade and the maximumexpected volume depletion during itsstart-up should always be greater than its
minimum permissible volume. To makethis procedure operational, we wouldkeep producing the ith basic grade untilConstraint (1) is violated for this grade orConstraint (2) is violated for any othergrade, whichever occurs earlier. At that in-stant, we would switch to that grade forwhich Constraint (2) is violated first. It isimportant to recognize that we minimizethe number of basic grade switches andmaximize the duration of a productioncampaign by initiating a switch only whenthese boundary conditions are violated.Procedures for Flow Synchronization
We consider an H-stage sequential pro-cess. In this process, we consider the ithstage, which operates for a known andconstant duration (steady state) and isthen periodically regenerated for a fixedperiod (transient state). We construct abuffer in front of this stage to ensure suffi-cient storage to keep the bottleneck oper-ating during the transient state. We alsodetermine the flow on a real-time basis atthis stage to keep the bottleneck opera-tional. We term this flow the synchronizedflow. To determine the size of this bufferand the synchronized fiow, we define thefollowing variables:
/ = II,.. .n\: Index the set of stages,f,,: Flow at the /th stage during steady state.Fit. Flow at the /th stage during the transientstate,ti^. Duration of steady state at stage /,f,_,: Duration of the transient state at stage i.
We define f,,., the effective flow at the/th stage, as follows:
The bottleneck is the stage with the low-est effective flow. Without loss of general-ity, let us assume that this is the kt\\ stage.To ensure that the /th stage always meetsthe flow requirements at the bottleneck,we would set F,^ = Ff.^ + (Ff.^ — Fjjjtif/
January-February 1999 47
RAJARAM ET AL.
equation suggests that duringsteady state we would operate this stageat that flow which meets the effective flowat the bottleneck (i.e., f/,.) and also buildsup sufficient volume to account for theshortfall during transience (i.e., (f ^ —F,j)t,t/tjJ. This buildup should be the sizeof the in-process buffer after this stage.
At any instant, we monitor the follow-ing variables from the process:
Vj/. The actual volume at this buffer,F : The required flow at the bottleneck,/,: Time remaining before the next transientat the ith stage.
We determine Fj, the synchronized flow atthe ith stage, as f, = Fj, + (f f,-, - V, )/f;.This equation implies that synchronizedflow is equal to the sum of the requiredbottleneck flow and the buildup requiredin excess of available volume duringsteady state used to compensate for theimpending transience.ReferencesChallenor, David 1996, "Cerestar 2000: Strategy
and actions for competitive success in thenew millennium," presentation to the boardof directors, Cerestar Headquarters, Brussels,Belgium.
Extend 1992, "Performance modeling for deci-sion support," Imagine That, Inc., San Jose,California.
Jaikumar, R. and Rajaram, K. 1996, "Incorporat-ing operator—process interactions in processcontrol: A framework and an application toglucose refining," Working paper #96-070,Harvard Business School.
Jaikumar, R. and Rajaram, K. 1997, "A decisionsupport system for operationally intelligentcontrol," Working paper. Department ofOperations and Technology Management,Anderson School, UCLA.
Rajaram, K. 1998, "Lot sizing in large scale in-dustrial processes: Models, applications andanalysis," Working paper. Department ofOperations and Technology Management,Anderson School, UCLA.
INTERFACES 29:1 48
International Journal of Operations & Production ManagementCyclical packaging planning at a pharmaceutical companyL.W.G. Strijbosch R.M.J. Heuts M.L.J. Luijten
Article information:To cite this document:L.W.G. Strijbosch R.M.J. Heuts M.L.J. Luijten, (2002),"Cyclical packaging planning at a pharmaceuticalcompany", International Journal of Operations & Production Management, Vol. 22 Iss 5 pp. 549 - 564Permanent link to this document:http://dx.doi.org/10.1108/01443570210425174
Downloaded on: 14 November 2016, At: 12:34 (PT)References: this document contains references to 17 other documents.To copy this document: [email protected] fulltext of this document has been downloaded 1445 times since 2006*
Users who downloaded this article also downloaded:(1994),"A Typology of Production Control Situations in Process Industries", InternationalJournal of Operations & Production Management, Vol. 14 Iss 12 pp. 47-57 http://dx.doi.org/10.1108/01443579410072382(2002),"Demand and capacity management decisions in services: How they impact on one another",International Journal of Operations & Production Management, Vol. 22 Iss 5 pp. 527-548 http://dx.doi.org/10.1108/01443570210425165
Access to this document was granted through an Emerald subscription provided by emerald-srm:211952 []
For AuthorsIf you would like to write for this, or any other Emerald publication, then please use our Emerald forAuthors service information about how to choose which publication to write for and submission guidelinesare available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.comEmerald is a global publisher linking research and practice to the benefit of society. The companymanages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well asproviding an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committeeon Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archivepreservation.
*Related content and download information correct at time of download.
1. Stephen M. Disney, Arnold Maltz, Xun Wang, Roger D.H. Warburton. 2016. Inventory managementfor stochastic lead times with order crossovers. European Journal of Operational Research 248:2, 473-486.[CrossRef]
2. Diane P. Bischak, David J. Robb, Edward A. Silver, Joseph D. Blackburn. 2014. Analysis and Managementof Periodic Review, Order-Up-To Level Inventory Systems with Order Crossover. Production andOperations Management 23:5, 762-772. [CrossRef]
3. Sushmita A. Narayana, Rupesh Kumar Pati, Prem Vrat. 2014. Managerial research on the pharmaceuticalsupply chain – A critical review and some insights for future directions. Journal of Purchasing and SupplyManagement 20:1, 18-40. [CrossRef]
4. Tim Van Kampen, Dirk Pieter Van Donk. 2014. Coping with product variety in the food processingindustry: the effect of form postponement. International Journal of Production Research 52:2, 353-367.[CrossRef]
5. George J. Besseris. 2013. Improving quality cost performance with qualimetrical methods: a case from ahigh-speed packaging process. The International Journal of Advanced Manufacturing Technology 69:5-8,1775-1789. [CrossRef]
6. R. Germs, N.D. van Foreest. 2013. Order acceptance and scheduling policies for a make-to-orderenvironment with family-dependent lead and batch setup times. International Journal of Production Research51:3, 940-951. [CrossRef]
7. Lin Wang, Jing He, Desheng Wu, Yu-Rong Zeng. 2012. A novel differential evolution algorithm forjoint replenishment problem under interdependence and its application. International Journal of ProductionEconomics 135:1, 190-198. [CrossRef]
8. Remco Germs, Nicky D. Van Foreest. 2011. Admission policies for the customized stochastic lot schedulingproblem with strict due-dates. European Journal of Operational Research 213:2, 375-383. [CrossRef]
9. George Nenes, Sofia Panagiotidou, George Tagaras. 2010. Inventory management of multiple items withirregular demand: A case study. European Journal of Operational Research 205:2, 313-324. [CrossRef]
10. Niall PiercySchool of Management, University of Bath, Bath, UK Nick RichCardiff Business School,Cardiff, UK. 2009. Lean transformation in the pure service environment: the case of the call service centre.International Journal of Operations & Production Management 29:1, 54-76. [Abstract] [Full Text] [PDF]
11. Anne L. Olsen. 2008. Inventory replenishment with interdependent ordering costs: An evolutionaryalgorithm solution. International Journal of Production Economics 113:1, 359-369. [CrossRef]
12. Edward A. Silver, David J. Robb. 2008. Some insights regarding the optimal reorder period in periodicreview inventory systems. International Journal of Production Economics 112:1, 354-366. [CrossRef]
13. J. Ashayeri, R.J.M. Heuts, H.G.L. Lansdaal, L.W.G. Strijbosch. 2006. Cyclic production–inventoryplanning and control in the pre-Deco industry: A case study. International Journal of Production Economics103:2, 715-725. [CrossRef]
14. Anne L. Olsen. 2005. An evolutionary algorithm to solve the joint replenishment problem using directgrouping. Computers & Industrial Engineering 48:2, 223-235. [CrossRef]
15. Chad Lin, Geoffrey JallehEvaluation of B2B Pharmaceutical Supply Chain in Australia 56-79. [CrossRef]
International Journal of Production Research,Vol. 43, No. 23, 1 December 2005, 5071–5100
Mixed-Integer Linear Programming approaches to shelf-life-integrated
planning and scheduling in yoghurt production
M. LUTKE ENTRUPy, H.-O. GUNTHER*y, P. VAN BEEKz,M. GRUNOWy and T. SEILERy
yDepartment of Production Management, Technical University of Berlin,
Wilmersdorfer Str. 148, D-10585 Berlin, Germany
zOperations Research and Logistics Group, Wageningen University,
Hollandseweg 1, NL-6706 KN Wageningen, the Netherlands
(Received April 2005)
In the production of perishable products such as dairy, meat or bakery goods, theconsideration of shelf life in production planning is of particular importance.Retail customers with relatively low inventory turns can benefit significantlyfrom longer product shelf life as wastage and out-of-stock rates decrease.However, in today’s production planning and control systems, shelf-life issueswith regard to specific products or customers are only seldom accounted for.Therefore, the objective of this paper is to develop Mixed-Integer LinearProgramming (MILP) models that integrate shelf-life issues into production plan-ning and scheduling. The research is based on an industrial case study of yoghurtproduction. Relying on the principle of block planning, three different MILPmodels for weekly production planning are presented that apply a combinationof a discrete and continuous representation of time. Overnight productionand, hence, the necessity for identifying two different shelf-life values for thesame production lot is included in the model formulation. Numericalexperiments show that near-optimal solutions can be obtained within reasonablecomputational time.
With an approximate turnover of E50 billion, fresh food industries (e.g. productionof meat, dairy, fruit, vegetables or bakery products) are major sectors of the Germaneconomy. The dairy industry is the dominating fresh food industry with a turnoverof about E15.5 billion in 2000 (Lebensmittel Zeitung 2001). Within the dairy indus-try, production planning of yoghurt products is certainly one of the most challengingtasks. The planner has, for example, to cope with a high number of variants as wellas sequence-dependent set-up and cleaning times on capital-intensive processingequipment. One of the most distinctive factors to consider in fresh food productionplanning is shelf life. Shelf life restrictions directly influence wastage, out-of-stock
ISSN 0020–7543 print/ISSN 1366–588X online # 2005 Taylor & Francis
http://www.tandf.co.uk/journals
DOI: 10.1080/00207540500161068
rates and inventory levels. The shelf life of a product is defined as the time in whichthe food product will remain safe, be certain to retain the desired sensory, chemical,physical and microbiological characteristics as well as comply with any label declara-tion of nutritional data (Kilcast and Subramaniam 2000, according to the LondonInstitute of Food Science and Technology Guidelines 1993). The possibility to offer ahigher shelf life than its competitors constitutes a pivotal competitive advantage forfresh food producers, making the provision of shelf life functions crucial for modernproduction planning systems.
The paper is organized as follows. The product characteristics, the supply chainand the production of yoghurt are presented in section 2. A literature review ofresearch on scheduling in make-and-pack production as well as on production plan-ning for perishable products is given in section 3. In section 4, three different modelformulations are presented that integrate shelf life into production planning.The models are numerically validated in section 5. Finally, recommendationsare provided concerning the applicability of the models for specific productionenvironments.
2. Case study: yoghurt production
2.1 Product characteristics of yoghurt
Yoghurt is arguably one of the most popular dairy products. It is a semisolidfermented product made from standardized milk mixed with a symbiotic blend ofyoghurt culture organisms (Chandan and Shahani 1993). Several types of yoghurtexist: the two main types are set and stirred yoghurt (Varnam and Sutherland 1994).While set yoghurt is incubated and fermented in the retail cups, stirred yoghurt isfermented before packaging. According to Rosenthal (1991), Chandan and Shahani(1993) and Tamime and Robinson (1999), the success of yoghurt can be attributed toits health-related glamour, the increase of low-fat products, the high variety of tastesand textures, intense marketing activities, and its relatively low costs.
As all fresh products, yoghurt has a relatively short shelf life. The manufacturerusually determines the shelf life of a product. The shelf life of yoghurt producedunder normal conditions is about 8–10 days at <10�C. However, following the trendto concentrate production capacities, to extend the delivered markets and to increasethe product portfolio, many manufacturers have increased the shelf life of theirproducts up to 3 or 4 weeks (Spreer 1995), mainly by means of aseptic packagingtechnology.
2.2 Yoghurt supply chain
The yoghurt supply chain has been subject to major changes throughout the recentyears. With regard to economic developments, a clear trend towards a consolidationcan be observed on all stages in the supply chain. As an example, at the farming stagethe number of dairy farms in the Netherlands decreased from 67 000 in 1980 to36 000 in 1997 (Frouws and Van der Ploeg 2000). With regard to manufacturing,the number of German dairy companies decreased for instance from 859 in 1976to 250 in 2000 (Weindelmaier 2002). The continued consolidation of retailers isperceived as the biggest threat to the manufacturers (KPMG Corporate Finance
5072 M. Lutke Entrup et al.
UK 2000). In the UK, the share of the top five retailers has already achievedover 80% (van Wezel 2001). Due to the strong retail concentration, retailers willincreasingly command the retailer–manufacturer interface (McLaughlin 2002)or even the whole upstream part of the chain. Another characteristic of the yoghurtsupply chain is the growth of the discount channel and of private labels. Whilemanufacturers can still obtain relatively high margins of around 5–7% in thepremium segment (Murmann and Wolfskeil 2004), the margins for private labelsare much lower.
Therefore, differentiation gets increasingly important for manufacturers in orderto avoid cost and margin pressure. On the product level, new product introductionsincreased considerably, especially in the yoghurt segment. The number of productinnovations introduced, for example, in German retail has increased since 1997 by11.1% annually; however, the failure rate reached almost 66% in 2000 (Michael et al.2002). Furthermore, this product proliferation leads to production complexity.Promotional activities are another important differentiating factor. Particularly inGermany, significant revenue shares are generated by promotions (Treeck andSeishoff 2002). Several studies reveal that most promotions are not effective andsuccessful. According to Fairfield (2002), manufacturers estimate that only 35% ofall promotions are profitable. Product freshness is one of the few remaining differ-entiating factors for manufacturers and retailers. For consumers, freshness is one ofthe major quality characteristics of a fresh food product (N.N. 2003). Most foodprocessors and retailers expect that the share of fresh and perishable productsin retail will continue to grow (Grievink et al. 2002), from 45 to 60% of allsupermarket sales after 2005 (McLaughlin 2002). Krampe (2004) stresses thatgrocery retailers need more fresh products to distinguish themselves from thediscount channel. For example, freshness is one of the most important competitiveadvantages for the German retailer ‘Globus’ to attract customers. If the lead-timeof a product exceeds a quarter of its total shelf life, the product will be rejected(N.N. 2003).
The attempt to deliver the yoghurt products as freshly as possible to the retailerhas significant effects on the supply chain. First, increasing the product freshness inthe outlet requires the products to be produced as close as possible to the demanddate; at best, each product is produced daily. However, due to significant set-uptimes in yoghurt production, smaller lot sizes are far more expensive. On the otherhand, fresher products can help to reduce out-of-stock rates in the retail outlet,which represents a severe problem in the current retail environment. Fairfield(2002), for example, mentions out-of-stock rates of 6.5% in terms of revenue and8.2% in terms of articles. In the case of stock-outs, the revenue is largely lost for themanufacturer; only 26% of consumers buy the same product in another store (Seifert2001). As fresher products have a longer remaining shelf life, the inventory level inthe retail outlet can be increased and the out-of-stock rate be reduced accordingly.Finally, the consumer’s choice, which product to buy, is influenced in a positive way.As freshness is a decisive buying criterion, the consumer is likely to be inclined to buythe product with the longer shelf life.
With regard to production planning, the obvious advantages of fresher productsmust be balanced against the higher sterilization and cleaning costs in produc-tion due to smaller lot sizes. Therefore, production planning should consider productfreshness, lot sizing and scheduling simultaneously. As many of the benefits of
5073MILP approaches to shelf-life-integrated planning and scheduling
fresher products are at the retailers side (e.g. less wastage, lower out-of-stock rates),however, almost all additional costs arise at the manufacturer’s side (e.g. highersterilization and cleaning cost), both parties must agree on how to share costs andbenefits. To solve this conflict, one possibility is to integrate a shelf-life-dependentpricing component into the terms and conditions system.
2.3 Production process
This investigation is based on the production of stirred yoghurt. On the one hand,stirred yoghurt comprises most of the fruit yoghurts and hence covers the largestyoghurt sub-segment to a significant extent. On the other hand, the production ofstirred yoghurt has to cope with a very high product variety, making productionplanning for stirred yoghurt more challenging than for other yoghurt types. Theproduction of stirred yoghurt comprises four major steps (figure 1). The followingoutline of the production process is based on Rosenthal (1991), Chandan andShahani (1993), Spreer (1995), Tamime and Robinson (1999) and Walstra et al.(1999).
2.3.1 Raw milk preparation. On reception of the raw milk at the dairy factory, theraw milk is cooled and stored in a silo in which it should not remain longer than oneor two days. Before the raw milk is fermented into yoghurt, it is subject to a numberof preparatory treatments. The entire preparation process is generally highlyautomated. As the raw milk components vary significantly, it is first necessaryto standardize the raw milk in order to meet the compositional standardsfor yoghurt. Milk powder is widely used to fortify liquid milk, but also condensedskim milk or cream. Then, the raw milk is pre-heated in a plate heat exchangerand concentrated in an evaporator. Thereafter, the milk is pressed through a homo-genizer, heated again in a holding tube, cooled and transferred to the fermentationtanks.
2.3.2 Fermentation. Stirred yoghurt is fermented in tanks before filling andpackaging. Starter cultures are added in order to incubate the mix. Dependingon the type of starter concentrate, the kind of product, and the fermentationtemperature, the fermentation time can vary significantly. After incubation, thecoagulum is cooled relatively quickly in a plate cooler and kept in intermediatetanks in which the yoghurt is retained only for a short period.
2.3.3 Flavouring and packaging. The flavouring and packaging step has to copewith a high number of product variants, which is caused by the variety of tastes andpackaging materials. Before the yoghurt is filled into the retail containers, it is mixed
Raw MilkPrepa-ration
Fermen-tation
Flavouringand
Packaging
Storageand
Delivery
Figure 1. Processing steps in stirred yoghurt production.
5074 M. Lutke Entrup et al.
with fruits and other ingredients. For filling and packaging, a variety of differentunits are available which are generally capital-intensive and which can achieve aperformance of up to 100 000 retail containers per hour. These units sterilize andclean the packaging material, fill the cups and close them with a lid. The unitcompartments that get directly in contact with the final product must be cleanedregularly and carefully. A cleaning sequence, including the removal of productremains and the subsequent sterilization, can last up to several hours. In additionto the decrease of the available filling capacity, cleaning also causes a material loss(yoghurt volumes remaining in the tubes or calibration losses when setting up theunit for the new product). However, cleaning and sterilization can be avoided if thepreceding product has no or only a minor negative effect in terms of colour, taste,or microbiology on the following product or product variant.
2.3.4 Storage and delivery. After packaging, the final cooling of the product takesplace in the retail container to a temperature below 10�C since yoghurt organisms(e.g. Streptococcus thermophilus or Lactobacillus delbrueckii) show only limitedgrowth below this temperature. A retaining period of 48 h before the dispatchis advantageous to achieve the final stability of the coagulum. Therefore, manycompanies have introduced quarantine times for the final product.
3. Literature review
3.1 Lot sizing and scheduling in make-and-pack production
Batch production in the chemical industry can be taken as a reference for theproduction of yoghurt as both must consider numerous variants, which are basedon few product types or recipes. The variants are caused by the different types ofpackaging as well as by different tastes. In literature, a production environmentwhich is characterized by a single production stage and a subsequent packagingstage is named ‘make and pack production’ (cf. Mendez and Cerda 2002, Guntherand Neuhaus 2004). Major issues of operational production planning in this envir-onment are lot sizing and scheduling, which can be performed in one single or twoseparate planning steps. As lot sizing usually aims at balancing set-up costs on theone hand and inventory holding costs on the other hand, the set-up costs of eachsingle set-up operation must be known in advance. However, as set-up times andcosts in yoghurt production are sequence-dependent, their exact values can only bedetermined after the sequencing of the orders. On the other hand, sequencingis dependent on lot sizing so that both tasks must be performed simultaneously(cf. Sikora et al. 1996).
The available approaches can be classified according to their representation oftime; a discrete and a continuous time representation can be distinguished. For adiscrete representation of time, the planning horizon is divided into a certain numberof periods that have usually the same length. Examples of problems and relatedmodelling approaches with discrete time representation are the CapacitatedLot size Problem (cf. Gunther 1987), the Discrete Lot-sizing and SchedulingProblem (cf. Fleischmann 1990), the Continuous Set-up and Lot-sizing Problem(Karmarkar and Schrage 1985), the Proportional Lot-sizing and Scheduling
5075MILP approaches to shelf-life-integrated planning and scheduling
Problem (cf. Haase 1994), and the Capacitated Lot-sizing Problem with Sequence-dependent Set-up Costs (cf. Haase 1996). Fleischmann and Meyr (1997) integrate allmentioned models within the General Lot-sizing and Scheduling Problem (GLSP).All models have in common that set-up times can only be considered if they do notexceed the length of a period. However, Koclar and Sural (2005) show that through asimple modification of the GLSP, set-up times exceeding the length of a period canbe incorporated.
Nevertheless, choosing the length of a period becomes a crucial aspect of model-ling. Especially because a high number of relatively small periods are required foran exact representation of the production activities (cf. Meyr 1999), the number ofperiods can considerably increase the size of the model (cf. Mendez and Cerda 2002,Gunther and Neuhaus 2004). Stadtler (2002) emphasizes that particularly sequence-dependent set-up times cannot be represented properly within a model using largeperiods. In addition, the approximation or ‘rounding’ of processing and set-up timesto fit them into the fixed periods can lead to overproduction, idle times or unfeasi-bility (cf. Ierapetritou and Floudas 1998, Gunther and Neuhaus 2004). Due to thehigh complexity of these models, they are less suitable to practical purposes (cf. Meyr1999, Burkard et al. 2002). The application of periods of different length can beuseful to avoid the stated problems. Burkard et al. (2002) introduce the notion of the‘Event-Driven Model’, in which only such points in time are considered, at which aprocess is allowed to start.
Alternatively, it is possible to use a continuous representation of time, whichallows scheduling the start and end of an activity precisely on a continuous timescale. In particular, unfeasibilities due to ‘fitting’ set-up times into a discrete time gridcan be avoided (cf. Ierapetritou and Floudas 1998, Gunther and Neuhaus 2004).Sahinidis and Grossmann (1991) propose a model that uses a continuous representa-tion of time and considers explicitly sequence-dependent set-up times. They applya position-based model and assume constant demand patterns to generate a cyclicschedule. Among the latest publications, the approach of Ierapetritou and Floudas(1998) is to be mentioned that considers sequence-dependent set-up times for bothbatch and continuous production. Mendez and Cerda (2002) suggest a similar modelformulation that is characterized by a lower number of binary variables and hence iscomputationally more efficient. Nevertheless, none of the mentioned models sup-ports lot sizing and it is not possible to integrate demand data of a single finalproduct for every single production day, only for the aggregate demand of theentire planning horizon.
Gunther and Neuhaus (2004) present an approach that is based on the principleof block planning and that simultaneously considers lot sizing and scheduling.By integrating several variants of a product type or recipe into a ‘block’, thecomplexity of the model is significantly reduced. For the determination ofthe sequence of batches within a block, a ‘natural’ sequence of batches oftenexists, for example, from the lower taste to the stronger or from the brightercolour to the darker. In the concept of flexible block planning presented byGunther and Neuhaus (2004), the blocks are assigned to a macro-period.According to this concept, the completion of a block must take place before theend of a macro-period. However, as the start of a block can be in the same or aprevious period, a production lot cannot be assigned to a specific day, which iscritical in order to consider its shelf life.
5076 M. Lutke Entrup et al.
3.2 Production planning for perishable products
Most models assume an unlimited storage of intermediate and finished products.However, yoghurt products have a finite shelf life that has to be respected in pro-duction planning. Many authors concede the necessity to integrate the shelf life ofproducts in production planning and scheduling (e.g. Kallrath 2002, Gunther andNeuhaus 2004); nonetheless shelf life has only been considered explicitly in veryfew models. Blomer (1999) gives an overview of 31 different approaches to batchscheduling. Although most authors respected unlimited or zero storage times, noauthor considered finite shelf life. Within the available approaches that consider theshelf life or the deterioration of products in OR-related literature, two main streamscan be distinguished.
On the one hand, a vast body of literature exists on inventory management forperishable products. Beside perishable food products, perishable inventory theorycovers also the behaviour of radioactive materials, photographic film, prescriptiondrugs, or blood conserves. Nahmias (1982) and Raafat (1991) give a comprehensiveliterature overview and an analysis of proposed inventory models for perishables.Raafat particularly clarifies the difference between continuously deteriorating goodsand products that can be unrestrictedly used before the expiry date. However,yoghurt products show characteristics of both. On the one hand, the value ofyoghurt decreases over time as customers give a higher value to a fresh product;on the other hand, yoghurt is almost worthless after the expiry date. Abad (2003)developed a non-linear single-product model for the maximization of the contribu-tion margin while considering the decay of the products in order to determine thebest-possible sales price.
On the other hand, with regard to the integration of shelf life into productionplanning and scheduling approaches, most research deals with adding a shelf lifeconstraint to the Economic Lot Scheduling Problem (ELSP), which is concernedwith generating a cyclic schedule for several products, based on a single resource andconstant demand rate (cf. Elmaghraby 1978, Cooke et al. 2004). Soman et al. (2004)provide a review of the major contributions: as one of the first authors, Silver (1989)argued that reducing the production rate is more effective than reducing the produc-tion cycle time in order not to violate shelf life constraints. However, he assumedthat reducing the production rate does not cause additional costs. Sarker and Babu(1993) completed the model of Silver by adding production time related costs anddemonstrated that the choice between reducing the cycle time or the production ratedepends on the shelf life of the products, machine and product set-up times, andunit costs. Goyal (1994) and Viswanathan (1995) elaborate on the idea to producea product more than once in a cycle. Silver (1995) as well as Viswanathan and Goyal(1997) aim at optimizing the cycle time and the production rate simultaneously.Viswanathan and Goyal (2000) allow backordering within the ELSP with shelf lifeconsiderations. Among the recent publications, Chowdhury and Sarker (2001),Viswanathan and Goyal (2002) and Sarker and Chowdhury (2002) elaborate onthe three options ‘adjusting the production rate’, ‘adjusting the cycle time’, and‘adjusting production rate and cycle time simultaneously’. Finally, Soman et al.(2004) argue that in case of high capacity utilization, as it can be found in the foodindustry, the production rate cannot be reduced because quality problems can occurif the production rate is changed. Therefore, they use a constant production rate
5077MILP approaches to shelf-life-integrated planning and scheduling
in their model. Furthermore, they do not allow backordering, which reflects thecompetitive environment in food manufacturing.
Almost all mentioned ELSP models include assumptions that are seldom presentin an industrial environment. First, a constant demand rate is frequently supposed,which is not very realistic for fresh food industries with seasonalities and intensepromotional activities. Manna and Chaudhuri (2001) underline, for example, thatthe demand for deteriorating products may be time-, stock- or price-dependent(e.g. end-of-day pricing). Moreover, most models consider only one single facilityand do not account for sequence-dependent set-up times. Finally, the most impor-tant criticism is that ELSP models aim at generating a production cycle whichis repeated in certain intervals and which must not exceed the shortest productshelf life. Hence, product freshness is not considered in the objective function,only as a constraint.
4. Model formulations
4.1 Problem demarcation and model overview
The modelling of the yoghurt production and the integration of shelf life relies on anMILP approach, since the underlying decision problem involves binary decisions(e.g. a block is set up on a specific day or not). The MILP models presented inthis section focus on the flavouring and packaging stage. The comparison of thefermentation and packaging processes shows that the production facilities requiredin order to add fruits and to wrap and seal the products are far more capital intensivethan those needed for fermentation purposes. Due to the use of multi-purpose tanksfor the fermentation (e.g. for other dairy products), the fermentation processes areonly considered by a capacity restriction and by imposing minimum batch sizesfor the packaging lines. Furthermore, the distribution of products is not regardedwithin the models because it is often performed by retail organizations (cf. Funke1990, N.N. 2003).
The developed models are based on a continuous representation of time.To guarantee the compactness and computability of the models, a block planningapproach is chosen. A block covers all products based on the same recipe. Theproduct sequence within a block is determined by increasing colour intensity ofthe products (e.g. banana before cherry taste; shown by decreasing brightness ofthe bars in figure 2). For the consideration of shelf life, it is necessary to employ adiscrete, uniform time grid in addition (cf. Gunther and Neuhaus 2004). A periodrefers to a day as shelf life for yoghurt is usually given in days.
All product variants based on the same recipe form a block. When changing theproduction between two products that are not based upon the same recipe, it isalways necessary to perform the set-up operations. Only when changing the produc-tion between two product variants of the same recipe, the cleaning and sterilizing ofthe production facilities can be neglected. Hence, not only the sequence of productswithin a block is fixed but also the sequence of recipes/blocks can be fixed within theproduction day. In that case, the different recipes are enumerated according to theirposition within the day. In most lot-sizing/scheduling models so-called balance con-straints are used linking inventory at the end of each period to its starting inventory,
5078 M. Lutke Entrup et al.
the production and demand in that period. To handle shelf life considerationsproperly, production lots are assigned to specific demand dates and this removesthe need for the aforementioned balance constraints.
The objective function maximizes the contribution margin by consideringrevenues and variable cost elements. Shelf-life aspects are taken into accountby considering a shelf-life-dependent pricing component that may also includeinventory-holding costs. The time horizon of short term planning for yoghurt man-ufacturing is usually one week (Nakhla 1995). In the discussed example, the planninghorizon is divided into macro-periods (figure 3). The regular production time is fromMonday 0:00 until Friday 24:00, with possible extensions from Sunday (earliest at0:00) till Saturday (latest at 24:00). The planning of the packaging lines is alwaysperformed for one entire week taking into consideration the stored goods from theprevious week. All accumulated demand has to be met from Sunday to Tuesday ofthe following week. The demand data are based on orders as well as on forecasts.Due to the various packaging types for different retailers, it is relatively simple toassign the products to their customers.
4.2 Model with day bounds
The ‘model with day bounds (MDB)’ allows every product to be produced on everyproduction day. The recipe sequence on a line is the same for every production day.
S 1 k+12 S S k+2 k+3 k+n... ... S 2k+1 2k+2
de1,1 de1,3de1,2
day 1 day 2 day 3
.. .
set-up
demand:
fixed production sequence
Figure 2. Block planning approach (based on Gunther and Neuhaus 2004) with dejd indicat-ing demand element for product j assigned to the end of demand period d.
Planning horizon for demand
Scheduling horizon for production
OvertimeRegular
working time
Productionof inventory
Overtime
Su Mo Tu We SaTh Fr Su Mo Tu We Th Fr SaSaFrThWeTuMoSu
Figure 3. Planning horizon.
5079MILP approaches to shelf-life-integrated planning and scheduling
Therefore, it is necessary to clean and sterilize the line between two successiveproduction days. A production of a block or product over midnight is not possible.
Indices:j, k 2 J products,l 2 L packaging lines,s 2 S days,
p 2 P � S production days,d 2 D � S demand days,
r, v 2 R recipes, blocks,j 2 JðrÞ products based on recipe r,
l 2 LRðrÞ packaging lines that can process recipe r,l 2 LJð jÞ packaging lines that can process product j,d 2 DðsÞ demand days (to meet the demand of these days, lots produced on day
s can be considered ),s 2 Sðd Þ production days (the lots produced on these days can be considered to
meet the demand of demand day d ).
Parameters:varcj variable costs for the production of one unit of product j,
slj maximum shelf life of product j (days),B sufficiently large number,
capl capacity of packaging line l (units/day),sterl sterilization time of packaging line l,cll cleaning time of packaging line l,
benj maximum benefit for meeting the maximum shelf life of product j (E),lossr loss of fermented plain yoghurt of recipe r when cleaning the line (kg),clossr costs for the cleaning loss of plain yoghurt of recipe r (E/kg),dejd demand element for product j on demand day d,sjs inventory of product j, produced on day s,cl costs of utilization of packaging line l (E/day),
mb minimum batch size to be processed (kg),psj packaging size of product j (kg/unit),ftr fermentation time for recipe r (h),fc fermentation capacity (kg h/day),crj minimum shelf life of product j required by customers, as a fraction of
maximum shelf life,qj quarantine time of product j,osl overtime supplement for weekend production on packaging line l
(E/day),adj percentage of plain yoghurt contained in one unit of product j,fdp start of the first production day within the week (Sunday),ldp start of the last production day within the week (Saturday).
Decision variables:Srpl¼ 1 if recipe r is produced on production day p on line l (0, otherwise),
Xjpl units of product j produced on packaging line l on production day p,Zjds units of product j produced on production day s that are used to meet
the demand of demand day d,
5080 M. Lutke Entrup et al.
Lrpl duration of recipe/block r on production day p on packaging line l,ENDrpl end time of recipe/block r on production day p on packaging line l,ESTl start time of packaging line l,LFTl end time of packaging line l,SAOl overtime at the end of the planning horizon (Saturday) on packaging
line l,SUOl overtime at the beginning of the planning horizon (Sunday) on
packaging line l.
Objective function:
maxXj2J
Xd2D
Xs2Sðd Þ
Zjds � benj �1� crj� �
� slj� �
� d� sð Þ
1� crj� �
� slj
�Xj2J
Xp2P
Xl2LJð jÞ
Xjpl � varcj
�Xl2L
LFTl � ESTlð Þ � cl �Xl2L
SAOl � SUOlð Þ � osl
�Xp2P
Xr2R
Xl2LRðrÞ
Srpl � lossr � clossr ð1Þ
The objective function aims at maximizing the contribution margin taking intoaccount a shelf life-dependent pricing component. It is supposed that the manufac-turer yields a financial benefit if the products have a longer residual shelf life whenbeing delivered. The shelf life-dependent benefit increases linearly between the mini-mum customer requirement on shelf life (crj� slj) and the maximum possible shelflife (slj) since the benefits for the retailer increase with every additional day ofresidual shelf life (figure 4). As an example, suppose that product j has a total
Shelf Life DependentPricing Component
Shelf Lifecrj*slj slj
benj
slj − (d − s)
( )( ) ( )( ) jj
jj
slcr
sdslcrbenj ⋅−
−−⋅−⋅
1
1
Figure 4. Shelf life and shelf life-dependent pricing component.
5081MILP approaches to shelf-life-integrated planning and scheduling
shelf life of 30 days (slj¼ 30) and that the customers require a minimum residual shelflife when being delivered of 66% of the total shelf life (crj¼ 66%). Suppose furtherthat the shelf life of the product starts on day 6 (s¼ 6), the product is delivered to theretailer on day 10 (d¼ 10) and the maximum benefit for meeting the maximum shelflife of product j, benj is E0.30/kg. In this case, the manufacturer yields a financialbenefit of E0.18/kg of product j (60% of the maximum benefit).
Costs include the variable costs for the input factors, the costs of the utilizationof the packaging lines in regular and overtime mode, and set-up costs for cleaninglosses of plain yoghurt. Due to minimum batch sizes, it might be necessary toproduce more products than demanded. Therefore, Xjpl is not a constant, but avariable.
Constraints to be considered are the following:
Set-up: Xj2JðrÞ
Xjpl �Srpl � B p 2 P; r 2 R; l 2 LRðrÞ ð2Þ
Product j can only be produced on production day p on packaging line l if the line isset up for the corresponding recipe (Srpl¼ 1).
Output quantities:
Srpl � ðsterl þ cllÞ þXj2JðrÞ
Xjpl
capl� Lrpl p 2 P; r 2 R; l 2 LRðrÞ ð3Þ
The duration of block r on line l contains the sterilization and cleaning times as wellas the production time of the products based on this recipe j 2 JðrÞ.
Sequencing:
ENDrpl � Lrpl þ ENDvpl p 2 P; r, v 2 R: v < r; l 2 LRðrÞ & l 2 LVðvÞ ð4Þ
Due to the fixed sequence of recipes on a packaging line within a day, block rmay not start before the end of its predecessor v, thus avoiding an overlappingof blocks on line l. The expression v< r means that block v is a predecessor ofblock r.
Day bounds:
ENDrpl � pþ 1 p 2 P; r 2 R; l 2 LRðrÞ ð5Þ
ENDrpl � Lrpl � p p 2 P; r 2 R; l 2 LRðrÞ ð6Þ
These constraints ensure that every production lot is assigned to one distinct pro-duction day. Constraint (5) makes sure that every block ends before the end ofthe corresponding day, (6) that every block starts after the beginning of this day.The feasible interval for the float variables indicating the end of a block is set by theinterval derived from the integer day numbers. For instance, according to (5) and(6), any block r produced on day p¼ 5 must be completed by 6 and started after 5.Thus, ENDrpl will assume a value between 5.00þLrlp and 6.00.
5082 M. Lutke Entrup et al.
Stock balance:Xl2LJð jÞ
Xjpl �Xd2DðsÞ
Zjds j 2 J; s 2 S; ldp � s � fdp; p 2 P: p ¼ s ð7Þ
sjs �Xd2DðsÞ
Zjds j 2 J; s 2 S: s < fdp ð8Þ
The first constraint allows the lots of product j produced on different packaging linesto be considered in order to meet the demands on different demand days d; thesecond constraint assigns the inventory available at the beginning of the planningperiod to demand days. sjs is the amount of initial inventory of product j producedon day s. It is the inventory built up at the end of the previous week which may beused to satisfy the demand of the actual planning week. This amount may be used tosatisfy a demand of demand day d if maturation and shelf life requirements arerespected ðd 2 DðsÞÞ.
Utilization of packaging lines:
LFTl � ENDrpl � 1� Srpl
� �� B p 2 P; r 2 R; l 2 LRðrÞ ð9Þ
ESTl � ENDrpl � Lrpl þ 1� Srpl
� �� B p 2 P; r 2 R; l 2 LRðrÞ ð10Þ
Constraint (9) sets the end time of a packaging line within the planning horizonequal to the end of the last block produced on this line. Yet, because even thoseblocks that do not have any production output have an end time (which is equal tothe beginning time), it is necessary to ensure that the line was actually set up for theblock. Constraint (10) ensures the same for the start time of a packaging line.
SUOl � fdpþ 1� ESTl l 2 L ð11Þ
The overtime at the beginning of the week (Sunday) can be calculated by subtractingthe start time of the packaging line from the end of the first day ( fdpþ 1).
SAOl � LFTl � ldp l 2 L ð12Þ
The overtime at the end of the week (Saturday) is determined by subtracting theend of the regular working period (ldp) from the end of the last block on thepackaging line.
Meeting demand:
dejd ¼X
s2Sðd Þ: sþqj<d^ðd�sÞ�ð1�crjÞ�slj
Zjds ¼Xs2Sðd Þ
Zjds j 2 J; d 2 D ð13Þ
The demand of product j on demand day d can only be filled using lots that complywith the two requirements concerning shelf life and quarantine time. First, thedemand day d must be strictly larger than the shelf life day s of the product con-sidered plus the quarantine time qj of the product j. Second, the shelf life days lostin the warehouse (d� s) may not exceed a threshold defined by the customer require-ment on shelf life ((1� crj)� slj). As an example, a product with a maximum shelf lifeof 30 days (slj¼ 30) and a customer requirement on shelf life of 66% (crj¼ 66%) maynot spend more than 10 days in the warehouse ((1� 0.66)� 30¼ 10). Otherwise, theseproduction volumes cannot be considered to satisfy this demand. For the validity of
5083MILP approaches to shelf-life-integrated planning and scheduling
the objective function (1), it is necessary that the volume meeting these requirementsis the only one considered to meet the demand on the corresponding demand days.
Minimum batch sizes:Xj2JðrÞ
Xjpl � psj � adj � mb � Srpl r 2 R; p 2 P; l 2 LRðrÞ ð14Þ
Since the production uses fermentation capacity, it is necessary to guarantee acertain filling level of the fermentation tanks. This is ensured using minimumbatch sizes.
Fermentation capacity:
Xr2R
Xj2JðrÞ
Xl2LRðrÞ
Xjpl � psj � adj
!� ftr þ
Xr2R
Xl2LRðrÞ
Srpl � lossr � ftr � fc p 2 P ð15Þ
The daily fermentation volume must not exceed the available fermentation capacity.Due to different fermentation times of the recipes, the fermentation capacity is givenin kg h.
Variable domains:
Srpl 2 0; 1f g p 2 P; r 2 R; l 2 LR rð Þ ð16Þ
Xjpl � 0 j 2 J; p 2 P; l 2 LJ jð Þ ð17Þ
Zjds � 0 j 2 J; s 2 S; d 2 D sð Þ ð18Þ
Lrpl � 0 r 2 R; p 2 P; l 2 LRðrÞ ð19Þ
ldpþ 1 � ENDrpl � fdp r 2 R; p 2 P; l 2 LR rð Þ ð20Þ
ESTl,LFTl,SAOl,SUOl � 0 l 2 L ð21Þ
This model is suitable to integrate shelf life aspects into the planning and schedulingdecisions in yoghurt production, mainly because of the assignment of every lot to aspecific production day. The model formulation does not support a conservationof the set-up state in order to allow production over midnight. It is possible to usethis model for facilities that interrupt production during night-time. However, asyoghurt-packaging lines are highly capital intensive, high system utilization is desir-able. Therefore, the model does not necessarily meet all requirements of yoghurtproduction. Yet, it can be used for estimating the costs, the set-up frequency or theprofit of a planning week.
4.3 Model with set-up conservation
To allow overnight production, model MDB is extended to the ‘model with set-upconservation (MSC)’ by adding a binary variable that conserves the set-up stateover midnight. Additional information about the position of the products within ablock is necessary in order to determine the exact product start and end times forovernight production.
5084 M. Lutke Entrup et al.
Additional variables:Cjpl¼ 1 if product j is produced on packaging line l until the end of production
day p� 1 and at the beginning of day p (0, otherwise).
Additional parameters:bpj position of product j in the corresponding block.
The objective function of model MDB (1) is replaced by the following:
maxXj2J
Xd2D
Xs2Sðd Þ
Zjds � benj �1� crj� �
� slj� �
� d� sð Þ
1� crj� �
� slj�Xj2J
Xp2P
Xl2LJðjÞ
Xjpl � varcj
�Xl2L
LFTl � ESTlð Þ � cl �Xl2L
SAOl þ SUOlð Þ � osl
�Xp2P
Xr2R
Xl2LRðrÞ
Srpl �Xj2JðrÞ
Cjpl
!� lossr � clossr ð22Þ
The conservation of the set-up state for day p has the effect that the cleaning loss isonly applicable for day p� 1. Although the cleaning itself takes place at the end of ablock on a line, the loss is considered at the beginning of each block.
Output quantities:Furthermore, constraint (3) has to be replaced by the following constraints (23)and (24).
Srpl �Xj2JðrÞ
Cjpl
!� sterl þ Srpl �
Xj2JðrÞ
Cj, pþ1, l
!� cll þ
Xj2JðrÞ
Xjpl
capl� Lrpl
p 2 P; p < ldp; r 2 R; l 2 LRðrÞ ð23Þ
If the set-up state of product j 2 JðrÞ is conserved from the previous day(Cjpl¼ 1), the sterilization time can be neglected. Furthermore, cleaning can beneglected, if the set-up state for product j 2 JðrÞ is conserved for the following day(Cj, pþ 1, l ¼ 1):
Srpl �Xj2JðrÞ
Cjpl
!� sterl þ Srpl � cll þ
Xj2JðrÞ
Xjpl
capl� Lrpl
p 2 P: p ¼ ldp; r 2 R; l 2 LR rð Þ ð24Þ
A conservation of the set-up state is not possible for the last day of productionp¼ ldp, i.e. from Saturday to Sunday. The variable Cj, pþ 1, l is not defined forthis case. On that day (Saturday, p¼ ldp), the transfer of the set-up state to thefollowing day (Cj, pþ 1, l¼ 1) is not possible, since this day is not part of the planninghorizon and the variable Cj, pþ 1, l is therefore not defined. Consequently, the pack-aging line has to be cleaned on this day, no matter which recipe was producedlast. However, at the beginning of this day conservation of the set-up state fromthe previous day is possible; in this case (Cjpl¼ 1) the time for sterilization can beneglected.
5085MILP approaches to shelf-life-integrated planning and scheduling
If the set-up state for product j 2 JðrÞ is transferred from production day p to pþ 1,the amount of plain yoghurt of the following day can be added in order to meet theminimum batch size (mb). The right hand side of (25) does not need to be multipliedby the set-up variable Srpl (as, for example, in (26) or (27)), because the equation isonly relevant for
Pj2JðrÞ Cj, pþ1, l ¼ 1.
In that case, Srpl is always equal to one because the set-up state for a product jof block r ð j 2 JðrÞÞ can only be established if the corresponding block r is set up.Since a conservation of the set-up state from the last day of production ( p¼ ldp)to the successive day is not possible, this constraint is only valid for all days ofproduction but the last.X
j2JðrÞ
Xjpl � psj � adj þXj2JðrÞ
Cjpl
!� Bþ
Xj2JðrÞ
Cj, pþ1, l
!� B � mb � Srpl
r 2 R; p 2 P: p < ldp; l 2 LRðrÞ ð26Þ
In case the set-up state of a recipe is neither transferred from day p� 1 to day p norfrom day p to day pþ 1 (Cjpl¼ 0 and Cj, pþ 1, l¼ 0 for all j 2 JðrÞ), the amount of plainyoghurt of this block has to meet or exceed the minimum batch size requirements.X
j2JðrÞ
Xjpl � psj � adj þXj2JðrÞ
Cjpl
!� B � mb � Srpl
r 2 R; p 2 P: p ¼ ldp; l 2 LR rð Þ ð27Þ
Since the conservation of the set-up state from the last day to its successor is notpossible, the amount of plain yoghurt needed for a block has to meet or exceed theminimum batch size, except the set-up state is transferred from the previous day.
Fermentation capacity:As the conservation of the set-up state for day p has the effect that the cleaning lossis only applicable for day p� 1, the fermentation capacity constraint (15) has to bealtered as follows:X
r2R
Xj2JðrÞ
Xl2LRðrÞ
Xjpl � psj � adj
!� ftr
þX
r2R
Xl2LRðrÞ
Srpl�X
j2JðrÞCjpl
� �� lossr � ftr � fc p 2P ð28Þ
The constraints (29)–(37) must be considered in addition to the ones formulatedabove.
Conservation of the set-up state:Xk2JðrÞ: bpk>bpj
Xkpl � 1� Cj, pþ1, l
� �� B r 2 R; j 2 J rð Þ; p 2 P: p < ldp; l 2 LR rð Þ
ð29Þ
5086 M. Lutke Entrup et al.
If product j 2 JðrÞ is produced until the end of production day p and the set-upstate is conserved for the following day (Cj, pþ 1, l¼ 1), the output quantities of thefollowing products within this block have to be zero for day p.
ENDrpl � p � Cj, pþ1, l r 2 R; j 2 J rð Þ; p 2 P: p < ldp; l 2 LR rð Þ ð30Þ
The variable Cj, pþ 1, l can only take the value of 1 if the production of productj 2 JðrÞ on production day p runs right until the end of day p. Production of productj may then continue without interruption on day pþ 1 (following constraints).
Xjpl � Cj, pþ1, l j 2 J; p 2 P: p < ldp; l 2 LJ jð Þ ð31Þ
The conservation variable Cj, pþ 1, l can only take the value of one if at the end ofday p there has been production output of product j.
ENDrpl � Lrpl � p � 1�Xj2JðrÞ
Cjpl r 2 R; p 2 P; l 2 LR rð Þ ð32Þ
If the set-up state is conserved from day p� 1 to day p for product j 2 JðrÞ, theproduction of this block r has to continue directly at the beginning of the day p.The expression ENDrpl�Lrpl� p can only take a value of zero if the start of theproduction (ENDrpl�Lrpl) is equal to the beginning of the production day the p. Forexample, if p represents day 5, then the start of the production of block r must beexactly at 00:00 on day 5 (ENDrpl�Lrpl¼ 5.00) in order to set the left hand side ofthe constraint to zero. This value of zero on the left-hand side of the constraint isnecessary for the binary variable Cjpl to take a value of 1.
Xjpl � 1�X
k2JðrÞ: bpk>bpj
Ckpl
0@
1A � B r 2 R; j 2 J rð Þ; p 2 P; l 2 LR rð Þ ð33Þ
If the set-up state is conserved for product j 2 JðrÞ from production day p� 1 to dayp, production on this day may only continue with the same product or a successorin the same block. For all predecessors the output on production day p has to bezero. The large number B has the function to allow the production of product jin case the set-up state is not conserved for product k with a higher block positionthan product j. X
j2J
Cjpl � 1 p 2 P; l 2 L ð34Þ
For every day and packaging line, the set-up state for only one product can beconserved.
Srpl �Xj2JðrÞ
Cjpl � 0 r 2 R; p 2 P; l 2 LR rð Þ ð35Þ
The set-up state can be conserved for a product j 2 JðrÞ if the line is set up for thecorresponding block r.
Cjpl � 0 j 2 J; l 2 LJ jð Þ; p ¼ fdp ð36Þ
At the beginning of the week ( p¼ fdp), a conservation of the set-up state from thepreceding day is not possible.
5087MILP approaches to shelf-life-integrated planning and scheduling
Variable domains:
Cjpl 2 f0; 1g p 2 P; j 2 J; l 2 LJ jð Þ ð37Þ
This model is capable of conserving the set-up state while meeting the strict assign-ment of production lots and days. Therefore, it is not only possible to produce aproduct overnight but also for more than two successive days. However, the fixedsequence of recipes on every packaging line and every packaging day influences theconservation of the set-up state, as only for the last recipe of a day, the set-up statecan be conserved. Therefore, we remove this constraint in the following position-based model, in which the sequence of recipes within a day is no longer fixed.
4.4 Position-based model
For the ‘position-based model (PBM)’, the planning horizon is split up into con-secutively enumerated positions i 2 I to which a block can be assigned. The begin-ning and the end of every block have to take place on a given day the position isassigned to by the parameter estartli. Only for the first positions of a day ðo 2 O � I Þ,a conservation of the set-up state from the preceding day is possible. Since the modelformulation is considerably different from the previous ones, the entire PBM modelis presented.
Indices:j, k 2 J products,l 2 L packaging lines,s 2 S days,
p 2 P � S production days,d 2 D � S demand days,
r, v 2 R recipes, blocks,i 2 I positions,
i, o 2 O � I positions at the beginning of a day,j 2 JðrÞ products based on recipe r,
l 2 LRðrÞ packaging lines that can process recipe r,l 2 LVðvÞ packaging lines that can process recipe v,l 2 LJð jÞ packaging lines that can process product j,d 2 DðsÞ demand days (to meet the demand of these days, lots produced on
day s can be considered ),s 2 Sðd Þ production days (the lots produced on these days can be considered to
meet the demand of demand day d ).
Parameters:estartli earliest starting time of a block at position i on packaging line l,varcj variable costs for the production of one unit of product j,
slj maximum shelf life of product j,B sufficiently large number,
capl capacity of packaging line l (units/day),sterl sterilization time of packaging line l,cll cleaning time of packaging line l,
benj maximum benefit when meeting the maximum shelf life of productj (E),
5088 M. Lutke Entrup et al.
lossr loss of fermented plain yoghurt of recipe r when cleaning the line (kg),clossr costs for the cleaning loss of plain yoghurt of recipe r (E/kg),dejd demand element for product j on demand day d,sjs inventory of product j, produced on day s,cl costs of utilization of line l (E/day),
mb minimum batch size to be processed (kg),psj packaging size of product j (kg/unit),ftr fermentation time for recipe r (h),fc fermentation capacity (kg h/day),crj minimum shelf life of product j required by customers, as a fraction
of maximum shelf life,bpj position of product j in the corresponding block,qj quarantine time of product j,osl overtime supplement for weekend production on packaging line l
(E/day),adj percentage of plain yoghurt contained in one unit of product j,fdp start of the first production day within the week (Sunday),ldp start of the last production day within the week (Saturday).
Decision variables:Bril¼1 if recipe r is produced at position i on packaging line l (0, otherwise),Cjol¼1 if product j is produced at position o� 1 until the end of the produc-
tion day p� 1 (estartlo¼ p) and at position o on packaging line l(0, otherwise),
Yjil units of product j produced on packaging line l at position i,Zjds units of product j produced on production day s that is used to meet
the demand on demand day d,Lril duration of recipe/block r at position i on line l,
ENDril end time of recipe r at position i on line l,ESTl start time of packaging line l,LFTl end time of packaging line l,SAOl overtime at the end of the planning horizon (Saturday) on packaging
line l,SUOl overtime at the beginning of the planning horizon (Sunday) on
packaging line l.
Objective function:
maxXj2J
Xd2D
Xs2Sðd Þ
Zjds � benj �1� crj� �
� slj� �
� d� sð Þ
1� crj� �
� slj�Xj2J
Xi2I
Xl2LJð jÞ
Yjil � varcj
�Xl2L
LFTl � ESTlð Þ � cl �Xl2L
SAOl þ SUOlð Þ � osl
�X
i2I: i =2O
Xr2R
Xl2LRðrÞ
Bril � lossr � clossr
�Xr2R
Xl2LRðrÞ
Xo2O
Brol �Xj2JðrÞ
Cjol
!� lossr � clossr ð38Þ
5089MILP approaches to shelf-life-integrated planning and scheduling
This objective function is similar to (22), with the days p being replaced by thepositions i.
Constraints to be considered are the following.
Set-up: Xj2JðrÞ
Yjil � Bril � B i 2 I; r 2 R; l 2 LR rð Þð39Þ
Xr2R
Bril � 1 i 2 I; l 2 L ð40Þ
Similar to (2), the set-up constraint for production output has to be met.Additionally, a position may not be assigned to more than one block.
Output quantities:
Bril � ðsterl þ cllÞ þXj2JðrÞ
Yjil
capl� Lril
r 2 R; l 2 LR rð Þ; i 2 I: i =2O & ðiþ 1Þ =2O ð41Þ
For those positions that can neither be scheduled at the beginning nor at the endof a day ði 2 I: i =2O & iþ 1 =2OÞ, the full sterilization and cleaning time must beconsidered.
Brol �Xj2JðrÞ
Cjol
!� sterl þ Brol � cll þ
Xj2JðrÞ
Yjol
capl� Lrol
r 2 R; l 2 LR rð Þ; o 2 O ð42Þ
For a position at the beginning of a day ði 2 OÞ, the set-up state for a productproduced on the previous day can be conserved. Then, the sterilization process atthe beginning of the day does not need to take place.
Bril �Xj2JðrÞ
Cj, iþ1, l
!� cll þ Bril � sterl þ
Xj2JðrÞ
Yjil
capl� Lril
r 2 R; l 2 LR rð Þ; i 2 I: iþ 1ð Þ 2 O ð43Þ
Similarly, the set-up state of a position at the end of a day ðiþ 1 2 OÞ can betransferred to the following day. Then, the cleaning of the packaging line does notneed to be considered.
Sequencing:
ENDr, iþ1, l � Lr, iþ1, l þ ENDvil
i 2 I; iþ 1 2 I; r, v 2 R; l 2 LR rð Þ & l 2 LV vð Þ ð44Þ
Similar to (4), block v scheduled for position i must be finished before block r atposition iþ 1 is allowed to start.
Day bounds:
estartli þ Lril � ENDril � estartli þ 1 r 2 R, l 2 LR rð Þ; i 2 I ð45Þ
5090 M. Lutke Entrup et al.
For any block r assigned to position i, the start (ENDril�Lril) and the end (ENDril)of block r may only take place on the day for which the position i has been defined.For example, a value of estartli¼ 5.00 means that position i is carried out on day 5.ENDril has to take a value between 5.00þLril and 5.00þ 1.
Stock balance:Xi2I: estartli¼s
Xl2LJð jÞ
Yjil �Xd2DðsÞ
Zjds j 2 J; s 2 S, ldp � s � fdp ð46Þ
In analogy to (7), the output of product j on different lines may be added to meet thedemand of different demand days. Furthermore, the production quantities needto be sufficient:
sjs �Xd2DðsÞ
Zjds j 2 J; s 2 S ð47Þ
Inventory can be used to meet the demand of different demand days, yet the demandmay not exceed the volume stored. As this does not account for inventory built up inthe current planning period, the parameter sjs is zero for all days s>fdp.
Utilization of the packaging lines:
LFTl � ENDril � ð1� BrilÞ � B r 2 R; i 2 I; l 2 LR rð Þ ð48Þ
ESTl � ENDril � Lril þ ð1� BrilÞ � B r 2 R; i 2 I; l 2 LR rð Þ ð49Þ
As in (9), constraint (48) sets the value of variable LFTl (the end of productionon line l ) equal to the finishing time of the last block produced on this line.Constraint (49) sets the value of the variable ESTl (the start of production online l ) equal to the starting time of the first block produced on this line in theplanning period.
SUOl � fdpþ 1� ESTl l 2 L ð50Þ
The overtime at the beginning of the week (Sunday) can be calculated by subtractingthe start of the packaging line from the beginning of the regular working time( fdpþ 1).
SAOl � LFTl � ldp l 2 L ð51Þ
The overtime at the end of the week (Saturday) is determined by subtracting the endof the regular working period (ldp) from the ending time of the packaging line.
Meeting demand:
dejd ¼X
s2Sðd Þ: sþqj<d^ðd�sÞ�ð1�crjÞ�slj
Zjds ¼Xs2Sðd Þ
Zjds j 2 J; d 2 D ð52Þ
In analogy to (12), this constraint ensures that the external demand is met. Onlythose lots may be considered that fulfil the requirements of quarantine and minimumshelf life.
Minimum batch size:Xj2JðrÞ
Yjil � psj � adj � mb � Bril r 2 R; l 2 LR rð Þ; i 2 I: i =2O & ðiþ 1Þ =2O ð53Þ
5091MILP approaches to shelf-life-integrated planning and scheduling
For those positions that can neither be scheduled at the beginning, nor at the endof a day ði 2 I: i =2O & iþ 1 =2OÞ, the volume of plain yoghurt required forproduction of the block assigned to that position has to meet or exceed the minimumbatch size.
Xj2JðrÞ
Yjil þ Yj, i�1, l
� �� psj � adj � mb � Bril þ Br, i�1, l �
Xj2JðrÞ
Cjil
!
r 2 R; l 2 LR rð Þ; i 2 I: i 2 O ð54Þ
For the positions at the beginning of a production day, the required quantities ofplain yoghurt may be added to meet the minimum batch size if the set-up state isconserved.
Xj2JðrÞ
Yjol � psj � adj þXj2JðrÞ
Cjol
!� B � mb � Brol r 2 R; l 2 LR rð Þ; o 2 O ð55Þ
If the set-up state is not conserved for a block at a position at the beginning of aproduction day, the amount of plain yoghurt required for this block has to respectthe minimum batch size.
Xj2JðrÞ
Yjil � psj � adj þXj2JðrÞ
Cj, iþ1, l
!� B � mb � Bril
r 2 R; l 2 LR rð Þ; i 2 I: iþ 1ð Þ 2 O ð56Þ
If the set-up state is not conserved for a block at the end of a production dayðiþ 1 2 OÞ, the volume of plain yoghurt required for this block has to meet orexceed the minimum batch size.
Fermentation capacity:
Xr2R
Xj2JðrÞ
Xl2LRðrÞ
Xi2I: estartli¼p
Yjil � psj � adj
!� ftr þ
Xr2R
Xl2LRðrÞ
Xi2I: estarli¼p^i=2O
Bril � lossr � ftr
þXr2R
Xl2LRðrÞ
Xo2O: estarlo¼p
Brol �Xj2JðrÞ
Cjol
!� lossr � ftr � fc p 2 P ð57Þ
The used fermentation capacity must not exceed the installed capacity.
Conservation of the set-up state:
ENDril � Lril þ sterl þX
k2JðrÞ: bpk�bpj
Ykil
capl� estartli � Cj, iþ1, l
r 2 R; j 2 J rð Þ; l 2 LR rð Þ; i 2 I: iþ 1ð Þ 2 O ð58Þ
This constraint ensures that for positions at the end of a day ðiþ 1 2 OÞ, theconservation variable Cjol can only take the value of one for product j 2 JðrÞif this product is produced until the end of that production day.
Yjil � Cj, iþ1, l j 2 J; l 2 LJ jð Þ; o 2 O; i 2 I: iþ 1ð Þ 2 O ð59Þ
5092 M. Lutke Entrup et al.
The set-up state can only be preserved for product j assigned to a position i thatcan be scheduled at the end of a day ðiþ 1 2 OÞ, if the output of that product takes apositive value.
ENDrol � Lrol � estartlo � 1� Cjol r 2 R; j 2 J rð Þ; l 2 LR rð Þ; o 2 O ð60Þ
If the set-up state is conserved for product j 2 JðrÞ from position i� 1 to positioni 2 O, the production of the block the product is based on has to continue directlyafter the beginning of the day the set-up state is transferred to.
Yjol � 1�X
k2JðrÞ: bpk>bpj
Ckol
0@
1A � B r 2 R; j 2 J rð Þ; l 2 LJ jð Þ; o 2 O ð61Þ
In case the set-up state is conserved for product j 2 JðrÞ, production on this day mayonly continue with the same product or a successor in the same block.X
j2J
Cjol � 1 o 2 O; l 2 L ð62Þ
Constraint (62) ensures that the preservation of a set-up state can only take placefor one product for every packaging line l 2 L and every position at the beginning ofa day ðo 2 OÞ.
Brol �Xj2JðrÞ
Cjol � 0 r 2 R; l 2 LR rð Þ; o 2 O ð63Þ
Similar to (35), the set-up state can only be preserved if the line is set up for theblock.
Variable domains:
Bril 2 0; 1f g i 2 I; r 2 R; l 2 LR rð Þ ð64Þ
Cjol 2 0; 1f g j 2 J; o 2 O; l 2 LJ jð Þ ð65Þ
Yjil � 0 j 2 J; i 2 I; l 2 LJ jð Þ ð66Þ
Zjds � 0 j 2 J; s 2 S; d 2 D sð Þ ð67Þ
Lril � 0 r 2 R; i 2 I; l 2 LR rð Þ ð68Þ
ldpþ 1 � ENDril � fdp r 2 R; i 2 I; l 2 LR rð Þ ð69Þ
ESTl,LFTl,SAOl,SUOl � 0 l 2 L ð70Þ
The solvability of this MILP model mainly depends on the number of binary vari-ables and on the number of positions defined. This number must be determinedbefore solving the model. The number of positions can be restricted since the clean-ing and sterilization of a packaging line is always necessary when proceeding withthe sequence of positions except when conserving the set-up state. For example, thenumber of possible positions per day multiplied by the set-up time of the consideredpackaging line should not exceed 24 h. To keep the number of binary variables small,the conservation of the set-up state is only possible for positions at the beginningof a day ðo 2 OÞ. Therefore, it is impossible for models that require more than onepossible position per day to produce one block for more than 48 h. To determinethe number of positions required for a scheduling problem, the system utilization,the number of recipes to be processed on a packaging line, and the set-up time
5093MILP approaches to shelf-life-integrated planning and scheduling
should be considered. Furthermore, it is possible to increase the number of positionsiteratively and stop the process if an improvement of the objective value can nolonger be observed.
5. Numerical investigation
The purpose of the numerical investigation is to assess the suitability of the modelsfor specific planning problems in industry. The different models can then be trans-formed into a toolkit for the planner. To determine an acceptable MIP-Gap, theinterpretability of the objective function value (OV) has to be taken into account.MIP-Gap is the difference (%) between the actual OV and a theoretical upper boundfor the optimal OV, which is obtained from an LP relaxation of the problem. In theirsearch for the optimal solution, branch-and-bound procedures sequentially add themissing integrality constraints to the LP relaxation and thus reduce the MIP-Gap tothe best known feasible solution. Hence, MIP-Gap is the maximum remainingimprovement potential of the OV. Since the sales revenues for consumer goods arevery small (cf. Seifert 2001, Murmann and Wolfskeil 2004), MIP-Gap should be<1%. Therefore. the objective functions of the presented models are altered forbetter interpretation by adding the following term:
þXj2J
Xd2D
djd � revj
where revj is the fixed pricing component per unit of product j when selling it to theretail organizations, which represents the price when the products are provided withthe shortest possible shelf life. By adding this term, the OV can be interpreted asthe contribution margin after subtraction of all variable costs. The data set used todemonstrate the practical applicability of the proposed models consists of 30products based on 11 recipes that can be processed on four packaging lines. Twoof the packaging lines process the same range of products (packaging line type c,lines 3 and 4); lines 1 and 2 serve for packaging a specific range of products each.Table 1 indicates the number of variables and constraints for each of the modelspresented in the previous section.
The numerical investigation was performed on a computer with an AMD XP2600þ CPU and 1GB RAM. The models were implemented using ILOG’s OPLStudio 3.6.1 as a modelling environment and its incorporated standard optimizationsoftware CPLEX 8.1. The described models MDB, MSC and PBM were first exam-ined based on the complete configuration of four packaging lines. The number of
Table 1. Number of variables and constraints for each of the models.
Figures indicated for PBM are based on 21 positions per week, i.e. 3 positions per day.
5094 M. Lutke Entrup et al.
positions used for model PBM was determined in advance to be 21 (three positionsper day).
The performance of each model is assessed along the dimensions OV, MIP-Gapand CPU time (tables 2 and 3). OV is the value of the objective function at themoment at which the optimization run is stopped. The indicated CPU time is thetime limit set for the optimization run or—in case the optimal value has beenobtained (MIP-Gap¼ 0%)—the time required to obtain the OV. To generate thefinal solutions within relatively short computational time (which is an importantrequirement for industry applications), two different time limits have been set,at 300 and 1800 s. In the first investigation, the optimization models consider allpackaging lines simultaneously.
The results of this first investigation show that model MDB performs well regard-ing MIP-Gap and computational time. The additional flexibility of models MSC andPBM does not result in a higher OV within the time limits. To realize this potential,it is necessary to reduce the model complexity. Hence, the different types of pack-aging lines were looked at separately. An independent optimization of the differentpackaging lines is possible, if the fermentation capacity is not limiting. Therefore, thefeasibility of the solution obtained via this decomposition procedure (with respect topackaging lines) has to be checked against the installed fermentation capacity.The optimization took place separately for the different types of packaging lines(type a: line 1; type b: line 2; type c: lines 3 and 4). The number of positions usedfor model PBM is 21 for types a and c, and 14 for type b. The number of positions touse per line type has been derived from the number of recipes and the volume of finalproducts to produce on each line. In our example, line type b has to handle fewerrecipes than the other line types. Hence, the corresponding number of positions islower. Aggregate numerical results obtained from combining the three line-specificoptimization runs are indicated in table 4.
Results of the basic MDB model show that optimization runs are very fastand exact optimal solutions are obtained. In case all four packaging lines areoptimized simultaneously, the optimal solution with an OV of E1,429,498 is deter-mined within 1167 s (table 3). Optimizing the different types of packaging linesseparately leads to a solution with almost the same OV in a much shorter CPU
Table 3. Optimizing all packaging lines simultaneously (t� 1800 s).
MDB MSC PBM
OV (E) 1 429 498 1 416 089 1 428 973MIP-Gap (%) 0.00 2.34 1.78t (s) 1167.14 1800.00 1800.00
Table 2. Optimizing all packaging lines simultaneously (t� 300 s).
MDB MSC PBM
OV (E) 1 429 272 1 416 089 1 420 907MIP-Gap (%) 0.08 2.37 2.38t (s) 300.00 300.00 300.00
5095MILP approaches to shelf-life-integrated planning and scheduling
time of 33 s (table 4). Therefore, decomposing the problem is effective inreducing computational times without impairing the overall quality of the schedule.In addition, the other models benefit from the separate optimization of thepackaging lines.
Still the question is which model is the ‘right choice’ for a specific problem.In particular for packaging lines 3 and 4, a combination of different models isrecommendable. The distinct assignment of recipes to one or both lines can bederived from the results of the optimization run using model MDB (e.g. recipe 1should be processed only on line 3; recipe 2 on line 3 and 4, etc.). Under theseconditions, the OV as shown in table 5 were obtained. For the two models withassignment of recipes to lines, the 32 s of the optimization run of model MDB havebeen added to the 300 s computational time as model MDB must be run beforehand.Particularly the results of model PBM have been improved by the assignment, whilemodel MSC improved only slightly.
A suitable strategy for the decomposition of the entire optimization problem andthe choice of adequate MILP models allows the planner in practice to determine asuitable schedule within reasonable CPU time. Hence, for the case study considered,it is recommended to combine the presented models carefully. In table 6, sucha possible combination is given. The solution is characterized by an MIP-Gap ofabout 0.5%. In this case, the OV is almost 0.4% higher than the best solution listedin table 3.
Table 6. Combination of the different models.
Line Model OV (E) t (s)
1 MSC 384 806 29.482 PBM 531 996 1.473 and 4 MDB and PBM with assignment 518 388 332.20
Total 1 435 190 363.15
Table 5. Optimization for packaging lines 3 and 4 (t� 300 s).
Table 4. Aggregate results obtained from combining three line-specific optimization runs.
MDB MSC PBM
OV (E) 1 429 498 1 431 929 1 432 470MIP-Gap (%) 0.00 0.57 0.76t (s) 32.82 330.10 601.47
5096 M. Lutke Entrup et al.
The approach suggested appears to be very satisfactory both concerning thequality of the solutions as well as computational time. To benefit from the fullpotential of the presented models, practical experience and knowledge aboutthe applicability of each of the models for specific problem settings is of majorimportance (table 7). For instance, model MDB is suitable if only very short com-putational times are allowed or for determining the basic recipe-line assignment.Furthermore, it should be used for more complex problems (e.g. longer planninghorizons or greater variety of feasible recipe-line assignments). Model MSC is parti-cularly appropriate for high-volume production with a limited variety of recipes(bulk), which requires the conservation of the set-up state for several consecutivedays. For increased flexibility, desired for production systems covering a high varietyof products (specialties), model PBM is more suitable, although this approachinvolves higher computational effort.
6. Conclusions and outlook
Three different MILP model formulations for scheduling problems in fresh foodindustry have been presented. They have been shown to be suitable to generatenear-optimal solutions for a planning and scheduling problem from industry. Asthe shelf life of the products has been explicitly considered, the use of the proposedplanning tools promises improved product freshness. With regard to further exten-sions of the models, one possibility could be the extension of model PBM, in order toallow that the production of a particular product may last several days. However,due to the small improvement, which was gained using model PBM instead of modelMSC, this extension was not realized in our investigation. Due to increasingcomputational times, the practicability of this approach is questionable. Anotheroption to extend the proposed models is the integration of the fermentation pro-cesses into the planning procedure. Whether this can take place simultaneously tothe scheduling of the packaging facilities is debatable because of the multifunctionaltanks often employed in industry. A sequential planning procedure, which performsthe planning and scheduling of the fermentation facilities based on the schedulingresults of the packaging lines, seems to be more promising. Finally, the incorporationof uncertainty constitutes an interesting extension of the model, especially withrespect to the use of safety stocks.
However, for the success of shelf-life-integrated planning for real lifeapplications, the value of the shelf-life-dependent pricing component needs to be
Table 7. Suitability of the different models.
Model Suitability
MDB Short computational time.Complex planning problems due to a long planning horizon or many recipes.
MSC Large-scale bulk production (e.g. production of a product over several days).Many products based on very few recipes.
PBM Small-scale specialty production.Planning problems for which computational time is not crucial.
5097MILP approaches to shelf-life-integrated planning and scheduling
determined. The implementation into existing supply chain management conceptsmay prove to be difficult. Developing suitable incentives based on customer satisfac-tion is therefore necessary. Yet, even without taking an integrated perspective ontothe supply chain, this concept offers a suitable tool for a yoghurt manufacturer toconsider freshness as a part of production planning. For markets characterized byintense competition, this provides an additional quality-oriented feature that canconstitute a pivotal competitive advantage. In addition to the dairy industry, theapplicability of these models for similar problems arising in the production of otherfresh foods (e.g. meat, fish, fruits, vegetables or bakery goods) has to be examined.
References
Abad, P.L., Optimal pricing and lot-sizing under conditions of perishability, finite productionand partial backordering and lost sale. Eur. J. Oper. Res., 2003, 144, 677–685.
Blomer, F., Produktionsplanung und -steuerung in der chemischen Industrie, 1999 (Gabler:Wiesbaden) [in German].
Burkard, R.E., Fortuna T. and Hurkens, A.J., Makespan minimization for chemical batchprocesses using non uniform time grids. Comput. Chem. Eng., 2002, 26, 1321–1332.
Chandan, R.C. and Shahani, K.M., Yoghurt. In Dairy Science and Technology Handbook,edited by Y.H. Hui, pp. 1–56, 1993 (VCH: Weinheim).
Chowdhury, M.R. and Sarker, B.R., Manufacturing batch size and ordering policy forproducts with shelf lives. Int. J. Prod. Res., 2001, 39, 1405–1426.
Cooke, D.L., Rohleder, T.R. and Silver, E.A., Finding effective schedules for the economic lotscheduling problem: a simple mixed-integer programming approach. Int. J. Prod. Res.,2004, 42, 21–36.
Elmaghraby, S.E., The economic lot scheduling problem: review and extensions. Manag. Sci.,1978, 24, 587–598.
Fairfield, D.K., Funf Fehlannahmen uber CPFR. In Collaborative Planning, Forecasting andReplenishment, edited by D. Seifert, pp. 205–224, 2002 (Galileo: Bonn) [in German].
Fleischmann, B. and Meyr, H., The general lot sizing and scheduling problem. OR Spektrum,1997, 19, 11–21.
Fleischmann, B., The discrete lot-sizing and scheduling problem. Eur. J. Oper. Res., 1990,44, 337–348.
Frouws, J. and Van der Ploeg, J.D., Issues of quality in Dutch dairying. In Dairy IndustryRestructuring, edited by H.K. Schwarzweller and A.P. Davidson, pp. 211–228, 2000(Elsevier: London).
Funke, H.-P., Chancen und Risiken fur die Distribution von Lebensmitteln. In 7. LogistikDialog, Logistik-Synergien zwischen Handel und Industrie, pp. 57–90, 1990 (TUVRheinland, Cologne) [in German].
Goyal, S.K., A note on ‘Effect of production cost on shelf life’. Int. J. Prod. Res., 1994, 32,2243–2245.
Grievink, J.-W., Josten, L. and Valk, C., State of the Art in Food, 2002 (Elsevier BusinessInformation: Arnhem).
Gunther, H.-O. and Neuhaus, U., Advanced planning and scheduling in the consumer goodsindustry: realizing block planning concepts for make-and-pack production using MILPmodelling and SAP’s APO software. In Operations Management as a Change Agent,Vol. 2, edited by L.N. Van Wassenhove, A. de Meyer, E. Yucesan, E.D. Gunes andL. Muyldermans, pp. 757–766, 2004 (INSEAD: Fontainebleau).
Gunther, H.-O., Planning lot sizes and capacity requirements in a single stage productionsystem. Eur. J. Oper. Res., 1987, 31, 223–231.
Haase, K., Capacitated lot-sizing with sequence dependent setup costs. OR Spektrum, 1996,18, 51–59.
Haase, K., Lotsizing and Scheduling for Production Planning, 1994 (Springer: Berlin).
5098 M. Lutke Entrup et al.
Ierapetritou, M.G. and Floudas, C.A., Effective continuous-time formulation for short-termscheduling—1. Multipurpose batch processes—2. Continuous and semicontinuousprocesses. Ind. Eng. Chem. Res., 1998, 37, 4341–4374.
Kallrath, J., Planning and scheduling in the process industry. OR Spectrum, 2002, 24, 219–250.Karmarkar, U.S. and Schrage, L., The deterministic dynamic product cycling problem.
Oper. Res., 1985, 33, 326–345.Kilcast, D. and Subramaniam, P., The Stability and Shelf Life of Food, 2000 (Woodhead:
Cambridge).Koclar, A. and Sural, H., A note on ‘The general lot sizing and scheduling problem’.
OR Spectrum, 2005, 27, 145–146.KPMG Corporate Finance UK, Europe’s Recipe for Success: Innovate and Consolidate.
Industry Report, 2000 (KPMG Corporate Finance UK: London).Krampe, G., Discounter und kein Ende. Lebensmittel Zeitung, 2004, 1, 20–21 [in German].Lebensmittel Zeitung, LZ Report 2001/2002, 2001 (Verlagsgruppe Deutscher Fachverlag,
Frankfurt am Main) [in German].Manna, S.K. and Chaudhuri, K.S., An economic order quantity model for deteriorating items
with time-dependent deterioration rate, demand rate, unit production cost andshortages. Int. J. Sys. Sci., 2001, 32, 1003–1009.
McLaughlin, E., The food industry in 2005, a forecast. In State of the Art in Food, edited byJ.-W. Grievink, L. Josten and C. Valk, pp. 598–608, 2002 (Elsevier BusinessInformation: Arnhem).
Mendez, C.A. and Cerda, J., An MILP-based approach to the short-term scheduling ofmake-and-pack continuous production plants. OR Spectrum, 2002, 24, 403–429.
Michael, J., Funder, J. and Huppertz, S., Konsum Guter 2010: Handel—Macht—Marken.Industry Report, 2002, Accenture, Frankfurt am Main [in German].
Murmann, C. and Wolfskeil, J., Magere Zeiten fur Molkereien. Lebensmittel Zeitung, 2004,9, 48 [in German].
N. N., Globus plant disposition im Stundentakt. Lebensmittel Zeitung, 2003, 48, 28[in German].
Nahmias, S., Perishable inventory, a review. Oper. Res., 1982, 4, 680–708.Nakhla, M., Production control in the food processing industry. Int. J. Oper. Prod. Manag.,
1995, 15, 73–88.Raafat, F., Survey of literature on continuously deteriorating inventory models. J. Oper. Res.
Soc., 1991, 42, 27–37.Rosenthal, I., Milk and Dairy Products, 1991 (VCH: Weinheim).Sahinidis, N.V. and Grossmann, I.E., MINLP model for cyclic multiproduct scheduling on
continuous parallel lines. Comput. Chem. Eng., 1991, 15, 85–103.Sarker, B.R. and Babu, P.S., Effect of production cost on shelf life. Int. J. Prod. Res., 1993, 31,
1865–1872.Sarker, B.R. and Chowdhury, M.R., Response to technical note on ‘Manufacturing batch
size and ordering policy for products with shelf lives’. Int. J. Prod. Res., 2002, 40,1971–1972.
Seifert, D., Efficient Consumer Response—Strategische Erfolgsfaktoren fur dieWertschopfungspartnerschaft von Industrie und Handel, 2001 (Rainer Hampp: Munich)[in German].
Sikora, R., Chhajed, D. and Shaw, M.J., Integrating the lot-sizing and sequencing decisionsfor scheduling a capacitated flow line. Comput. Ind. Eng., 1996, 30, 659–679.
Silver, E.A., Dealing with a shelf life constraint in cyclic scheduling by adjusting both cycletime and production rate. Int. J. Prod. Res., 1995, 33, 623–629.
Silver, E.A., Shelf life considerations in a family production context. Int. J. Prod. Res., 1989,27, 2021–2026.
Soman, C.A., Van Donk, D.P. and Gaalmann, G.J.C., A basic period approach to theeconomic lot scheduling problem with shelf life considerations. Int. J. Prod. Res.,2004, 42, 1677–1689.
Spreer, E., Technologie der Milchverarbeitung, 1995 (B. Behr’s: Hamburg) (in German).
5099MILP approaches to shelf-life-integrated planning and scheduling
Stadtler, H., Production planning and scheduling. In Supply Chain Management and AdvancedPlanning, 3rd ed., edited by H. Stadtler and C. Kilger, pp. 197–214, 2005 (Springer:Berlin).
Tamime, A.Y. and Robinson, R.K., Yoghurt—Science and Technology, 2nd ed., 1999(Woodhead: Cambridge).
Treeck, S. and Seishoff, M., CPFR in Deutschland, Osterreich und der Schweiz.In Collaborative Planning, Forecasting and Replenishment, edited by D. Seifert,pp. 265–277, 2002 (Galileo: Bonn) (in German).
Van Wezel, W.M.C., Tasks, Hierarchies and Flexibility—Planning in Food ProcessingIndustries, 2001 (Labyrint: Capelle a/d Ijssel).
Varnam, A.H. and Sutherland, J.P., Milk and Milk Products, 1994 (Chapman and Hall:London).
Viswanathan, S., A note on ‘Effect of production cost on shelf life’. Int. J. Prod. Res., 1995,33, 3485–3486.
Viswanathan, S. and Goyal, S.K., Incorporating planned backorders in a family productioncontext with shelf life considerations. Int. J. Prod. Res., 2000, 38, 829–836.
Viswanathan, S. and Goyal, S.K., On ‘Manufacturing batch size and ordering policy forproducts with shelf lives’. Int. J. Prod. Res., 2002, 40, 1965–1970.
Viswanathan, S. and Goyal, S.K., Optimal cycle time and production rate in a family produc-tion context with shelf life considerations. Int. J. Prod. Res., 1997, 35, 1703–1711.
Walstra, P., Geurts, T.J., Noomen, A., Jellema, A. and Van Boekel, M.A.J.S., DairyTechnology, 1999 (Marcel Dekker: New York).
Weindelmaier, H., Perspektiven der Verarbeitung von Lebensmitteln tierischer Herkunft.In Hulsenberger Gesprache, pp. 148–164, 2002 (H. Wilhelm Schaumann Stiftung:Hamburg) (in German).
Make to stock or make to order:The decoupling point in the food processing industries
Dirk Pieter van Donk*
Faculty of Management and Organization, University of Groningen, P.O. Box 800, 9700 AV Groningen, Netherlands
Received 13 April 1999; accepted 29 February 2000
Abstract
Food processing industries experience growing logistical demands, growing variety in products and intense competi-tion. As a reaction companies try to produce more on order. Often managers "nd it di$cult to decide which products tomake to order and which products to stock. This paper develops a frame that is an aid for managers in balancing thefactors and characteristics of market and production process that in#uence such decisions. The frame is based on thegeneral decoupling point concept by Hoekstra and Romme, which is adapted to the speci"c characteristics of the foodprocessing industry. Its usefulness is illustrated in a case study. Some directions for further research are given. ( 2001Elsevier Science B.V. All rights reserved.
For many years it was a common policy for foodprocessing companies to produce in large batchesto keep production costs low and limit the numberof set-ups. This seemed to be a good policy. The lastdecade showed a number of changes, graduallygrowing in signi"cance. The background andcauses of these changes can be summarised underthree main themes.
Firstly, consumers' wishes seem to changein an ever growing rate, causing an increase in
packaging sizes, the number of products as wellas in the number of new products introduced,e.g. [1].
Secondly, many retailers are restructuring theirsupply chain both in a physical and information#ow sense. The aims are reduction in inventories,faster replenishment and shortening of cycle times.The result for food processing industries is thatlogistical performance needs to be improved: fasterand more dependable. There are some examples ofreductions in lead-time from 120 hours in the pastto 48 hours now and still further reductions are tobe expected.
Thirdly, the above-mentioned changes have tobe realised in a market which can be characterisedby low margins in retailing and mergers andacquisitions in retail chains [1]. Both lead toa downward pressure on prices paid to producers.
0925-5273/01/$ - see front matter ( 2001 Elsevier Science B.V. All rights reserved.PII: S 0 9 2 5 - 5 2 7 3 ( 0 0 ) 0 0 0 3 5 - 9
In summary, food processing industries have todeliver a greater variety of products, have to meethigher logistical demands, while keeping costs aslow as possible. These demands are especiallyvisible in those industries that produce for theconsumer-market and their direct suppliers.
This article explores more #exible production asa possibility to cope with these market demands. Assaid above, the industry standard is to produce tostock. Producing more #exible requires that (partof the) customer orders have a direct impact onproduction orders. However, usually there are cer-tain limits and not all products for all customerscan be made to order. Therefore an importantquestion is which products, product families orproduct}market combinations can and should beproduced to stock and which can be made to order.To answer this question the general frame ofHoekstra and Romme [2] is used. Especially theirconcept of decoupling point, which is de"ned as `thepoint that indicates how deeply the customer orderpenetrates into the goods #owa. Hoekstra andRomme distinguish a frame in which a number ofproduct and market characteristics, and processand stock characteristics in#uence the location ofthe decoupling point. This frame will be adapted tothe speci"c characteristics of the food processingindustry.
The main question to be answered in this articleis how managers in food processing industry can besupported in deciding if some of their products canand should be produced to stock or be made toorder and, if there are some, which should beselected. In investigating this, we also add to ourknowledge of the usefulness of the decoupling pointconcept in this particular type of industry. So far,little attention has been paid to this type of ques-tions in the literature.
The article is organised as follows. First, we willintroduce the general frame of the decouplingpoint. Next, the speci"c characteristics of food pro-cessing industries are presented. Then, the "ndingsof these two sections are combined to arrive ata concept for the decoupling point tailored to thefood processing industry. Section 4 will show theusefulness of the concept in a case study. Lastly,conclusions are drawn and some remarks regardingfurther research are made.
2. The decoupling point concept
The background of the concept of decouplingpoint lies in the observation that within productionmanagement and logistical management attentionhas been paid to all kind of separate elements ina production chain, without notice for the need foran integrated framework. In developing the frameof the decoupling point (DP) Hoekstra and Romme[2] intend to furnish a concept for integral control.Integral control in this context means planning andmanagement of the goods #ow from purchasedmaterials to delivery takes place, based on the char-acteristics of the product}market combination,within a suited organisational and control structure(also see, [3]).
An important concern in designing integral con-trol is "nding a balance in the costs of procurement,production, distribution and storage against thecustomer service to be o!ered.
The result of this balancing establishes thedecoupling point for a certain product}marketcombination. The decoupling point separates thepart of the organisation oriented towards activitiesfor customer orders from the part of the organisa-tion based on forecasting and planning [2, p. 6], or,in other words the decoupling point is the pointthat indicates how deeply the customer order pen-etrates into the goods #ow [2, p. 66].
The decoupling point is important for a numberof reasons:
f It separates the order-driven activities from theforecast-driven activities. This is not only impor-tant for the distinction of di!erent types of activ-ities, but also for the related information #owsand the way the goods #ow is planned and con-trolled.
f It is the main stock point from which deliveriesto customers are made and the amount of stockshould be su$cient to satisfy demand in a certainperiod.
f The upstream activities can be optimised in someway, as they are based on forecasts and are moreor less independent from irregular demands inthe market.
According to [2] "ve possible DP positions cor-respond to "ve basic logistical structures: make and
298 D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306
Table 1Determinants of the decoupling point
Product and market characteristics Process and stock characteristics
Required delivery reliability Lead times and costs of steps in the (primary) processRequired delivery time Controllability of manufacturing and procurementPredictability of demand Costs of stock-holding and value added between stock pointsSpeci"city of demand Risk of obsolescence
Fig. 1. Business characteristics classi"ed according to their nature and in#uence on the DP [[2], p. 71].
ship to (local) stock; make to (central) stock; as-semble to order; make to order; purchase and maketo order. As stated the determination of the posi-tion of the DP depends, in general on two sets ofcharacteristics: product and market characteristics,and process and stock characteristics, which aresummarised in Table 1.
Their respective in#uence is depicted in Fig. 1 [2,p. 71]. Fig. 1 should be interpreted as follows. Foreach of the terms mentioned the in#uence on thelocation of the DP is shown. e.g. irregular marketdemand will (if all other things remain stable) havean upstream e!ect on the location of the DP, whileshort delivery times will force the DP more down-wards, towards the make-to-stock position.
The DP concept is a valuable tool in describingand analysing production processes and the goods#ow of organisations. However, from the existingliterature it is hard to derive rules for locating orchanging the position of the DP or procedures forbalancing the relevant characteristics. Also, thee!ects of having more DP's (for di!erent prod-uct}market combinations) in one factory and theconsequences for planning have been largelyneglected. Recently, [4] combines the location ofthe DP and the location of the capacity constraintinto one matrix to derive the characteristics of anintegrated material and capacity-based masterschedule. Further work into this direction seemsa good addition to the DP concept.
D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306 299
Fig. 2. An example of a production process in food processing.
3. Characteristics of food processing industries
This section introduces the food processingindustries from a production management pointof view. For the purpose of this paper it is usefulto distinguish between companies that produce in-termediate products from natural materials andcompanies that process these intermediate prod-ucts further into consumer or industrial products.The "rst category involves mills, abattoirs, sugarre"neries, etc. while the second category entailsproducers of canned meat, bakeries, producers ofchocolates, etc. Of course, this distinction is roughand some of the producers of intermediate prod-ucts, produce some consumer products as well (e.g.a sugar re"nery producing sugar cubes). We willpay attention to the second category. An exampleof a food processing process can be depicted as inFig. 2.
Instead of packaging, other processes mightoccur and sometimes products are pasteurisedor sterilised before packaging. In many cases, stockpoints as depicted can only store temporarily, dueto the perishability of intermediate products orbecause the capacity for storage on the shop#oor islimited.
From the literature [5}9] the following enumer-ation of characteristics of food processing industrycan be compiled [10].(1) Plant characteristics
(a) Expensive and single-purpose capacitycoupled with small product variety andhigh volumes. Usually, the factory showsa #ow shop oriented design.
(b) There are long (sequence-dependent) set-uptimes between di!erent product types.
(2) Product characteristics(a) The nature and source of raw material in
food processing industry often impliesa variable supply, quality, and price due tounstable yield of farmers.
(b) In contrast with discrete manufacturing,volume or weights are used.
(c) Raw material, semi-manufactured prod-ucts, and end products are perishable.
(3) Production process characteristics(a) Processes have a variable yield and process-
ing time.(b) At least one of the processes deals with
homogeneous products.(c) The processing stages are not labor inten-
sive.(d) Production rate is mainly determined by
capacity.(e) Food industries have a divergent product
structure, especially in the packaging stage.(f) Factories that produce consumer goods can
have an extensive, labor-intensive packag-ing phase.
(g) Due to uncertainty in pricing, quality, andsupply of raw material, several recipes areavailable for a product.
In most cases a limited number of these character-istics is present. We use the list above for examin-ing, describing and analysing real-life situations.Each of the factors presented, has to be taken intoaccount for planning and scheduling purposes. E.g.high set-ups and an orientation to use capacity asmuch as possible, cause planning of long produc-tion runs and stocks of end products.
4. Decoupling point in food processing industry
The previous sections elaborated upon the gen-eral in#uences of market and process character-istics on the position of the decoupling point andon the characteristics of the food processing indus-try. This section will relate the two. For each factorthe e!ect on the decoupling point for a certainproduct/market combination will be discussedunder a ceteris paribus clause for other factors. This
300 D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306
Table 2Market characteristics and their in#uence on the decoupling point
Market characteristic Presence/value in food E!ect on DP
Delivery reliability High DownstreamDelivery time Short DownstreamPredictability of demand (rather) unpredictable Upstream through information sharingSpeci"city of demand Great variety end products (with common recipes) Upstream possibilities
analysis is performed from the point of view ofa producer of food products, delivering to retailers(nowadays, usually supermarket chains, or integ-rated combinations of wholesaler with a number ofoutlets).
With respect to the market characteristics we mayconclude that, as was already stated in the intro-duction, delivery times are usually short and cus-tomers (such as retail chains) need a high reliability.Both tend to have a downstream e!ect on thedecoupling point. These short delivery times seemto go hand in hand with a relative unpredictabilityof the demand from the point of view of theproducer. In a number of cases the retailers try topass all the uncertainty in demand onto the pro-ducers by asking instant delivery within very shorttime, without adequate support in forecastingdemand. Nowadays better opportunities emergein forecasting by analysing point-of-sale and scan-ning data. Better co-operation and sharing of thesedata between retailer and producer could bringmutual bene"ts within easy reach and even openpossibilities for a longer lead time. Although actualdemand of the consumers might be still erratic,such joint e!orts in forecasting might improveoverall performance of the supply chain. A lastpoint to pay attention to, is the speci"city of de-mand. As noticed above, food processing industrieshave divergent product structures. Often the diver-sity of products originates from the large number ofpackaging sizes, labels and brands. Speci"city alsocomes into being through the best-before-date. It isinteresting to note that many products have a tech-nical best-before-date which is reasonably long.However, retailers do not accept succeeding delive-ries with identical best-before-dates. The result isthat from a technical point of view products are still
fresh, but from a commercial point of view obsoleteand in fact, extremely perishable. All in all, thefactor speci"city o!ers some potential for an up-stream repositioning of the decoupling point ifcommonality can be used and intermediate storagebetween processing and packaging is technicallypossible. The above discussion is summarisedin Table 2.
With respect to the process and stock character-istics it is clear that part of food processing indus-tries have an uncontrolled process with variableyield and, due to variability in natural materials,variable quality of products. Such factors causea downstream e!ect on the decoupling point,because storing a product after the uncontrolledprocess safeguards undisturbed delivery. Cleaningtimes and set-ups (which are often sequence depen-dent) are an important factor in production in foodprocessing industry. If these are large, the e!ect onthe decoupling point will also be downstream. Anupstream e!ect on the decoupling point might beexpected for the remaining factors: stock levels (andcosts) and risk of obsolescence. Both factors relateto the nature of food: it's perishability. Retailersalways want the most recent `best-beforea andstock might easily become out-of-date due to thebest-before on the product. This might occur even ifthe technical shelf-live is still quite long. The valueof stock is related to this aspect. There are twomajor factors that contribute to the value: the valueadded in production and the value of the materialspurchased. A high value added in production willhave an upstream e!ect on the decoupling point,as it is "nancially bene"cial to store low-valuegoods instead of higher valued end products. Thevalue of the materials is only a relevant factor ifpurchasing can be postponed until actual usage in
D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306 301
Table 3Process characteristics and their in#uence on the decouplingpoint
Process characteristic Presence/value in food E!ect on DP
Lead times & costs Relevant set-up times DownstreamControllability (sometimes) low DownstreamValue added and costs
of stock-holdingUnclear (in general) *
Risk of obsolescence High Upstream
production. In a number of food processing indus-tries e.g. co!ee, co!ee beans of a desired quality canonly be purchased once a year. In that case there isno "nancial reward for changing the decouplingpoint. In factories for canned meat, however, thepurchasing of meat can be postponed until produc-tion starts. In such a case the aggregate level ofinventories can be lower and an upstream move-ment of the decoupling point may have a "nanciale!ect. The above discussion is summarised inTable 3.
There are some special points of interest in thefood processing industries, which may in#uence thepossible location of the decoupling point.
The "rst point relates to the capacity of theproduction system. In many cases there is somekind of a bottleneck capacity: either in the process-ing stage or in the packaging stage. In generala bottleneck capacity is less suited for producing toorder, as more variation can be expected in directcustomer orders then in stock orders. Conse-quently, if the packaging department has limitedcapacity the possibility for locating the decouplingpoint upstream from the packaging stage will belimited. If the processing stage has limited capacitythen locating the decoupling point betweenprocessing and packaging is possible, setting asidepotentially restricting characteristics of products.
The second point is related to regulations andlaws aiming at protecting the safety of consumers.More and more, food processing companies haveto obey to strict regulations, which demand demon-strable safety through HACCP (hazard analysis ofcritical control points) and traceability. This putsnot only a pressure on registration and information
#ows, but might also restrict the possible locationof the decoupling point as intermediate inventoriesare less controllable and traceable than inventoriesof materials and end products.
A third point of interest is the attention given toand the introduction of the concept of ECR (e$-cient consumer response). One of the important socalled improvement concepts within this frame iscontinuous replenishment. For food processing in-dustries this concept can result in two kindsof consequences. On the one hand, retailers candemand short delivery times, without communicat-ing forecasts or information on sales. This results ina downstream e!ect on the decoupling point, asmentioned before. On the other hand, this mightresult in a closer cooperation between retailer andproducer and more intensive exchange of relevantinformation on sales will take place. As a resultmore production to order is possible.
A fourth point, which is interesting for the loca-tion of the decoupling point, makes a small adapta-tion to the assumptions and statements made in theintroduction. In food processing industries produc-tion to stock is usual, but especially in exporting"rms we observe a combination of make-to-stockand make-to-order. Usually the (large) orders forforeign countries are separately dealt with and pro-duction to order is possible due to longer leadtimes. Often production to order is necessary asthese orders arrive with great intervals and speci-"cation and magnitude is not known in advance.For our discussion this observation is importantbecause these companies already cope with di!er-ent decoupling points and combine make-to-stockwith make-to-order.
5. An illustrative case
The case concerns a manufacturer that buildsa new facility. In the old facility production to stockwas the only possibility due to the technologicallimitations. The new facility has the possibility tostore a number of semi-"nished products in silosand the management of the factory aims at produ-cing as much as possible to order. So, the questionwas which products were to be made to stock,which products to semi-"nished product and which
302 D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306
Fig. 3. The production process of the case.
products totally to order. In other words, the ques-tion is where the decoupling point ought to belocated for each product. To answer this questionthe afore-mentioned characteristics were analysed.We will brie#y describe the results.
5.1. Process
The production process has three steps: process-ing, granulation and packaging (see Fig. 3).
The processing stage consists of several steps:mixing of raw materials according to recipe anda number of subsequent processing steps which areexecuted without interruption or intermediate stor-age. Then the semi-manufactured product may bestored in one of the silos or can be granulated,directly. Next, the product is separated into severalfractions that are made of the same recipe but di!erin the size of the granule. In the last step theproduct is put into big bags or smaller bags of5}25 kg with (sometimes) a client-speci"c text on it.Now, the "nished product can either be delivered tothe customer or can be stored. Throughput timesfor each batch are approximately two hours forprocessing and half an hour for granulation. Theproduction rate (in kilogram/hour) of the "rst stepis about half the rate of the second step. Set-up-times are relatively large for both processes andare sequence dependent. The production process isreliable in quality and amount of output.
5.2. Market
The company produces some 200 di!erent prod-ucts, which di!er in recipe (40 di!erent recipes),granule (30 di!erent sizes) and packaging. Demandis stable in an aggregate way but irregular (both inamount and time) and not easy to forecast on
a detailed day-to-day or even week level. Fiverecipes (which are the basis of several "nished prod-ucts) account for about 70% of total demand. Thenumber of customers is high and even the largestcustomers have a share in total volume of less than10%. The lead time for delivery has a standard of5 days, but quite a few customers ask for shorterdelivery. On the other hand, some important cus-tomers order in a regular way with a lead time of2 weeks. Some customers ask the company to keepa certain amount of stock dedicated to them, forimmediate delivery. Customers ask for dependentdelivery, as is usual nowadays.
5.3. Product and stock
The product can be kept for almost half a year.However, the producer has to guarantee his cus-tomers a shelf life of 4 months at least. That meansthat slow-moving products have a risk of becomingobsolete. As said before the management of thecompany aims at lower inventories.
Each of the three storage points has a limitedcapacity: the capacity of the silos is most restrictinghaving a storage capacity of a little more then1 week of average sales.
Due to the nature of the product, the qualityrequired and the way it is produced, there arerestrictions with respect to the minimal batch sizesin the processing stage, the granulation andpackaging stage. These restrictions are incorpor-ated in the production technology and will betreated as limitations that cannot be changed.
5.4. Locating the decoupling point
From the description of the production process itis clear that, theoretically, three possible decoup-
D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306 303
ling points exist: the stock of raw materials, thestock of semi-"nished products (in silos) and thestock of "nished products. The storage space forsemi-"nished products is limited by the numberand capacity of the silos. So, in this case it seemsnatural to investigate "rst, which products shouldbe stored there.
Firstly, we observed that a number of semi-"n-ished products is used in one "nished product,only. That kind of products can be excluded fromstorage in the silos. The underlying logic is thatthese products are client speci"c as soon as the "rstoperation is started. Consequently, intermediatestorage has no advantage and production can beeither to stock or to order, depending on theamount asked for and the lead time needed. Aboutone-third of the semi-"nished products is soexcluded.
A second observation is related to the limitedcapacity in the processing stage. To use capacity ase$ciently as possible the number of set-ups andcleaning times in the processing stage has to berestricted. Thus, it is advantageous to produce inrelatively large bat ches, which is only necessary forproducts having a relatively large aggregate de-mand. The products with su$ciently large demandare subsequently stored in the silos to limit thestock of "nished products and still being able todeliver fast. In applying this rule, we could choosethe right number of products to be stored in thesilos. For the products stored in the silos the dilem-ma between e$ciency and responsiveness seems tobe solved: processing is in large and e$cientbatches, while granulation and packaging isconducted in response to the market.
Some products that we would like to produce toorder (e.g. due to irregular demand) will be produc-ed to stock if the amount asked for is smaller thanthe minimum batch sizes. The minimum batch sizecauses an amount of stock of the "nished product.This stock has a relatively high risk of becomingobsolete. Resolving this problem is not straightfor-ward and a!ects the market strategy. For productswhich are ordered irregularly, but in largeramounts than the minimum batch size there is noproblem and these products are of course producedto order. In other words, the decoupling point is thestock of raw materials.
5.5. Results
As a result of the above location decisions about75% of the number of the "nished products will beproduced on order. An important reason to havestock of the other articles is the very short lead timeasked for (often in combination with the wish ofcustomers to let the producer keep an amount ofstock dedicated to them). Another reason is regu-larity in demand: each week a number of orders fora certain product.
Changing the DP for a number of productsa!ects other performance measures as well.However, it is hard to make a comparison with thepreviously existing situation because a completenew factory is built. Still we can highlight some.First, it is important to note that customer service(in terms of dependability and speed) is improved,largely due to the fact that too many end items werestocked in the old situation, which caused problemswith inventory control and shelf lives of products.The number of obsolete products is thus reduced aswell as the inventory costs. Normally, producing onorder could result in less utilisation of capacity.However, here aggregate demand and capacityneeded is quite stable. Moreover, some orders havea longer lead time which enables a smooth produc-tion plan and high utilisation.
Another result of the decisions with respect tothe decoupling point relates to a better knowledgeof the market and the production capabilities andtheir interrelationship. Gathering informationabout products, the demand and the patterns indemand and orders, and lead times, gives a lot ofinformation not yet available to the company inthat way. This opens up discussion regarding thepro"tability of certain articles or the possibilitiesfor using the same recipes for more products. Soa start is made with discussing the marketing andmarket strategy. In such a discussion productioncapabilities play an important role.
5.6. Discussion of the case
The case teaches us that the frame helps us todetect the relevant factors for locating the decoup-ling point and to decide which products should bemade to order or which made to stock. It is clear
304 D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306
that the case has some speci"c elements in it, suchas the possibility to stock intermediate productsand the way capacities in subsequent stages arebalanced.
In the description of the case no attention hasbeen paid to planning and scheduling. Some com-ments can be made, however. Due to the restric-tions made regarding the minimum batch sizes andthe production in larger quantities of product forintermediate storage in silos, the capacity in theprocessing stage will be used e!ectively. The capa-city in the granulation and packaging stages is suchthat the time needed for set-ups leaves enoughspare capacity. Most important, however is thatdemand is stable in terms of the need for capacity.Still, planning and scheduling will be more impor-tant than it was previously and more interactionbetween production and sales department will benecessary. However, that can only be advantageous.
6. Conclusion
This paper develops a frame for a decision, man-agers face in food processing industries: whichproducts have to be made to stock and whichproducts have to be made to order. To supportsuch decisions the general decoupling point con-cept has been adapted to the speci"c characteristicsof the food processing industry. The frame o!ersa systematic means for food processing companiesto "nd the important in#uencing factors in themarket and in their production system. This resultsin a means for making this kind of decision as isillustrated in the case. An important insight fromthis paper is that, at least in the food processingindustry, the decoupling point theory can betransformed into an applicable decision aid formanagers. This paper also contributes to ourknowledge in applying this theory.
The paper shows some problems in decouplingpoint theory such as the development of generalapplicable rules for (changing) the location of thedecoupling point. In fact, balancing the diversefactors in#uencing the location of the DecouplingPoint (both in a qualitative and quantitative sense)is missing in the original writings of Hoekstra andRomme as well. In this particular case, the frame
supported us to develop appropriate decision rules.Other cases are needed to elaborate these rulesfurther for a decision logic for the positioning of thedecoupling point in the food processing industry.We think that progressing along the lines as putforward in this paper will unveil such a logic whichis, at least in the food processing industries, gener-ally applicable.
Further research should also be directed towardsthe consequences of changing the position of thedecoupling point. These consequences apply to theplanning and scheduling (how to combine maketo-stock and make-to-order) and to the capabilitiesof the production systems (see also [4]). Otherareas of attention are the organisational arrange-ments such as the relation between planning,production and marketing and the #ow of informa-tion. Experiences in case studies suggest thatbesides the problem of "nding a balance betweenthe factors in#uencing the position of the decoup-ling point, a major point of concern in implemen-ting chances is to overcome organisational andcultural barriers.
References
[1] M.T.G. Meulenberg, J. Viaene, Changing food marketingsystems in western countries, in: W.M.F. Jongen, M.T.G.Meulenberg (Eds.), Innovation of Food ProductionSystems: Product Quality and Consumer Acceptance,Wageningen Press, Wageningen, The Netherlands, 1998,pp. 8}36.
[2] S. Hoekstra, J. Romme (Eds.), Integral Logistic Structures:Developing Customer-oriented Goods Flow, McGraw-Hill, London, 1992.
[4] J. Olhager, J. Wikner, A framework for integrated materialand capacity based master scheduling, in: A. Drexl, A.Kimms (Eds.), Beyond Manufacturing Resource Planning(MRP II): Advanced Models and Methods for ProductionPlanning, Springer, Berlin, 1998, pp. 3}20.
[5] S.F. Bolander, Materials management in the process in-dustries, American Production and Inventory Control So-ciety 1980 Conference Proceedings, 1980, pp. 273}275.
[6] S.G. Taylor, S.M. Seward, S.F. Bolander, Why the processindustries are di!erent, Production and InventoryManagement Journal 22 (4) (1981) 9}24.
[7] J.C. Fransoo, W.G.M.M. Rutten, A typology of produc-tion control situations in process industries, International
D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306 305
Journal of Operations & Production Management 14 (12)(1994) 57}72.
[8] P. van Dam, Scheduling packaging lines in the processindustry, Ph.D. Thesis, University of Groningen, Gronin-gen, 1995.
[9] J.P. van Dam, G.J.C. Gaalman, G. Sierksma, Scheduling ofpackaging lines in the process industry: An empiricalinvestigation, International Journal of Production Econ-omics 30}31 (1993) 579}589.
[10] W. van Wezel, D.P. van Donk, Scheduling in food pro-cessing industries: preliminary "ndings of a task orientedapproach, in: J.C. Fransoo, W.G.M.M. Rutten (Eds.), Sec-ond International Conference on Computer IntegratedManufacturing in the Process Industries } Proceedings,BETA, Eindhoven, The Netherlands, 1996, pp. 545}557.
306 D.P. van Donk / Int. J. Production Economics 69 (2001) 297}306
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
1
OPPORTUNITIES AND REALITIES OF SUPPLY CHAIN
INTEGRATION: THE CASE OF FOOD MANUFACTURERS
Dirk Pieter van Donk1,*, Renzo Akkerman2 and Taco van der Vaart1
1 Faculty of Economics and Business, University of Groningen, P.O. Box 800, 9700 AV, Groningen, The Netherlands. ([email protected], [email protected])
2 Department of Manufacturing Engineering and Management, Technical University of Denmark, Produktionstorvet 425, 2800 Kgs. Lyngby (Copenhagen), Denmark. ([email protected])
* Corresponding author ABSTRACT Purpose The purpose of the paper is to investigate the limitations and barriers for supply chain integration that food manufacturers experience and to highlight their planning and scheduling problems. Possible ways to cope with these are offered. Methodology The paper is theoretical/conceptual in nature: the findings are illustrated in an explorative case study. Findings It is often suggested that food supply chains are typical for what can be achieved in supply chain management. This paper challenges this belief by investigating the possibilities and limitations for supply chain integration for food manufacturers. We argue that a combination of typical food characteristics and the use of shared resources limit the possibility for integration, while uncertainties and complex business conditions increase the need for integration. In a case study, the paper explores alternatives to cope with that situation. Limitations/implications The paper is based on previous empirical work, which is applied and further developed in a case-study setting of a consumer product food manufacturer. We argue that the case has several generic characteristics, but further research is needed to test the main ideas in a wider context. Practical implications Production managers and planners in food manufacturing are often aware of the described situation, but general managers, marketing managers, and supply chain managers can learn that there are limits to aligning operations to customers. The paper offers a number of solutions that might assist production managers in better understanding their situation and thinking about improvements. Originality/Value of the paper The paper introduces buyer focus, shared resources and the limitations of supply chain integration into the field of food supply chains. Keywords: supply chain management, supply chain integration, buyer focus, shared resources, food industry
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
2
INTRODUCTION Supply chain integration is often described as the seamless flow of products and information from supplier on to customer. Food supply chains are among the often-quoted examples of reaching this ideal state. For instance, Hill and Scudder (2002) suggest that food supply chains are in the front line with respect to supply chain practices, coordination of the chain, and the use of concepts like EDI, VMI, QR and CPFR. Often referred showcases are in the grocery chain and many articles refer to Wall-Mart and Kmart (Schwartz, 2004) as being benchmarks for supply chain integration. There are also numerous papers suggesting the introduction of quick response (Whiteoak, 1999), CPFR (Fliedner, 2003), Category management (Hutchins, 1997), and other tools and concepts to improve the flow of goods and information in the food supply chain. It is evident that the supply performance of food manufacturers has increased over the last years, largely driven by the restructuring of the food sector (e.g. Duffy et al., 2003; Hendrickson et al., 2001). The initiatives of powerful retailers have resulted in reduction in inventories in their distribution centres while maintaining the same level of customer service. Exchange of information, use of category management, and cross-docking operations are among the most applied practices (Van der Vorst and Beulens, 2001). However, it seems that the reality for food manufacturers is less fortunate than the rhetoric of many papers and popular books suggest. For example, Morgan et al. (2007) state that practice of supplier involvement in category management is low “despite the widespread prescription” (p. 513). They also state that the literature offers few empirical studies on that subject. All in all it seems that the position of food manufacturers is relatively ignored in the literature.
In this paper the aim is to explore the specific problems of food manufacturers seeking supply chain integration. It is our contention that the characteristics of many food manufacturing companies limit the possibilities of, whereas the supply requirements from retailers increase the need for integration. This might be not too new for many practitioners, but, so far, has not been dealt with sufficiently in research. This paper aims at analysing the specific characteristics of and demands placed upon food manufacturers in the context of supply chain integration. In other words, we focus our analysis on the operational problems stemming from supply chain requirements. As a consequence , we will not provide an extensive review of the supply chain management literature. Further, we limit ourselves to food manufacturers that produce for consumer markets, although part of our analysis might be relevant for other food producers. Supply chain integration is defined as the mutual coordination within or across organisational boundaries (Stevens 1989). The main points of our study will be illustrated in a case study.
The paper is organized as follows. The next section will develop the theoretical background of the paper by linking previous work on supply chain integration to the specific characteristics of the food industry. The third section of the paper will elaborate upon the supply chain strategies as mentioned above. Then, we will introduce the case and subsequently, analyse demand, production and planning aspects of this case. The fifth section will pay attention to (re-)design alternatives. In the last section we will formulate our conclusions. SUPPLY CHAIN INTEGRATION AND FOOD CHARACTERISTICS In this section we outline the theoretical background of the research. We explore supply chain integration, food manufacturing, and in the third subsection, their relationship.
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
3
Supply chain integration Over the last decade, different instruments and approaches to enhance supply chain integration have been investigated, mainly looking at the impact of supply chain integration on performance. So far, the influence of information systems (e.g., Vickery et al., 2003), the influence of operational practices (e.g., Frohlich and Westbrook, 2000), and the effect of simplifying the materials flow (e.g., Childerhouse and Towill, 2002) have been investigated. Others stress the development and implementation of specific tools, such as Vendor Managed Inventory (VMI), Collaborative Planning, Forecasting and Replenishment CPFR, radio frequency identification (RFID) or bar-coding in a supply chain context. The above and similar studies add to our knowledge and understanding of what can be achieved and how performance can be improved. Most of the published studies fail to address the business conditions or context of a supply chain (Ho et al., 2002). Ramdas and Spekman (2000) are among the few that investigated the influence of factors such as availability of substitutes, changes in market conditions, changes in technology, market maturity, and product life cycle in order to distinguish between functional and innovative products. As such, they add to the work of Fisher (1997) who argues that innovative products can be associated with high levels of uncertainty and need responsive supply chains, while functional products need efficient supply chains.
So far, specifically the influence of uncertainty in demand on supply chain management and integration has been explored. To Lee (2002), uncertainty is one of the drivers for supply chain integration. Empirical evidence also indicates that the level of uncertainty influences the level of integration (Davis, 1993; Childerhouse and Towill, 2002). Recently, Van Donk and Van der Vaart (2004) measure operational characteristics that influence supply chain integration, labelled as business conditions: the decoupling point (MTO/MTS), time window for delivery, volume-variety characteristics, process type (batch size, set-ups, and routings), and order-winners. In line with Davis (1993), these factors are important indicators for the amount of uncertainty manufacturers are facing in their production planning and delivery schedules. Van Donk and Van der Vaart (2004) distinguish between simple (high volume, low product variety, large batches, make-to-stock, and costs as a major order-winner) and complex (low volume, high product variety, small batches, make-to-order, and flexibility among the main order-winners) business conditions. Complex conditions correspond with a high level of uncertainty within the supply chain. They state and empirically show that only complex business conditions require a high level of supply chain integration. However, they also show that shared resources (capacity used to serve different customers) limit the possibilities to perform integration while buyer focus (singling out capacity for the purpose of serving one customer) is an enabler for supply chain management integration. A combination of uncertainty and shared resources is seen as one of the most difficult ones and it seems that many food manufacturers are exactly in that position. Figure 1 summarises the above relationships (Van Donk & Van der Vaart, 2004). In our view, integration relates to the amount and the level of activities such as vendor-managed inventories, packaging customisation, joint planning and forecasting, dedicated planners, use of inter-organisational planning systems, and use of Point of Sale (POS) data. A high level of integration corresponds with more intense and more activities.
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
4
Food manufacturing Food manufacturing is generally considered as a part of the (semi-) process industry. Process industries in general and food manufacturers in particular have been considered as being large-scale, capital-intensive, mass producers of bulk products in large batches for low costs. This uniform picture of process industries has been challenged by empirical work by Dennis and Meredith (2000), who clearly showed the diversity in production systems in process industries. For many food manufacturers the scenery has changed due to trends in markets and changes in consumer’s preferences. As a result, food manufacturers and specifically those that manufacture consumer products have adapted their product portfolio and production strategy in order to survive. The market for food products is more and more consumer-driven (Kinsey, 2003), and can be characterised by an increase in packaging sizes, products, recipes and product introductions (Meulenberg et al., 1998); higher logistical performance due to restructuring in the supply chain of retailers (e.g. Wall-Mart); and low margins in retailing and thus downwards pressure on prices for the manufacturers (Dobson et al., 2001). As a result, food manufacturers face a dilemma, as on the one hand they have to produce in response to the market, but, on the other hand, they have to produce at the lowest cost. In other words, flexibility and dependability are needed and on the other hand high utilisation. To complicate supply chain management initiatives further, we need to incorporate a number of food specific production characteristics. From previous studies (Van Donk, 2000) we compile the following enumeration:
(2) Product characteristics: variable supply, quality, and price of raw material due to unstable yield; raw material, semi-manufactured products, and end products are perishable;
(3) Production process characteristics: variable yield and processing time; homogeneous products; not labour intensive except for the packaging phase; production rate determined by capacity; divergent product structure especially in the packaging stage.
For many food manufacturers, the above characteristics are not all present and not all characteristics present will be evenly important for managing the process. Moreover,
Shared Resources Buyer Focus
High level of integration, typical practices are close co-operation, daily communication, and joint problem solving
Low level of integration, typical practices aim at efficient information and material flows
Integration is necessary, but limited by the shared resources
Integration is easy to achieve, but there is little need for it
Com
plex Sim
ple
Business conditons
Shared Resources Buyer Focus
High level of integration, typical practices are close co-operation, daily communication, and joint problem solving
Low level of integration, typical practices aim at efficient information and material flows
Integration is necessary, but limited by the shared resources
Integration is easy to achieve, but there is little need for it
Com
plex Sim
ple
Business conditons
Figure 1. Context and supply chain integration (Van Donk and Van de Vaart, 2004).
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
5
only a few of them will be really influential in implementing supply chain management initiatives. Supply chain integration in the food industry If we combine the two above subsections, some interesting first observations can be made. With regard to the type of resources, we can conclude that both types co-exist for many food manufacturers. Shared resources can be recognised in the expensive capacity and high set-ups, but packaging lines are often more dedicated to a limited number of products and buyers. Total output is usually determined by the (limited) capacity of the processing stage. Decoupling of the two stages is normal, but rather limited due to limited storage space and limited shelf life of unpacked products. The flexibility of the packaging stage is normally larger: fluctuations in mix (different packages) can easily be dealt with and fluctuations in volume can be achieved by adapting the amount of labour.
With respect to the business conditions one might be inclined to see food manufacturing as a typical case of functional products and simple business conditions as high volume, low variety, make-to-stock, short time for delivery, and costs as a major order-winner. However, the market requirements ask for smaller batches, more product differentiation and product innovations. The make-to-stock policy is not viable in a number of situations as retailers demand products with the most recent best-before date. Due to the nature of raw materials, processing times and yields in the processing stage can be unpredictable. Also the attuning of the two main stages (processing and packaging only separated by a limited storage capacity) results in delays and waiting times (Akkerman et al., 2007). This last type of uncertainty is, together with the earlier mentioned business conditions typical for the type of uncertainty that has to be dealt with in production planning and control (Davis, 1993).
What are the consequences for supply chain management in the food industry given these observations? Taking Figure 1 as a point of reference, it seems that many food producers are still in the lower-left quadrant (simple business conditions and shared resources) where initiatives to increase efficient flow of information and material between food manufacturers and retailers are appropriate. However, the combination of increased performance requirements, higher variety and specific characteristics of the food industry presses at least part of the industry into more complex business conditions. The conclusion is that supply chain integration is increasingly needed, but hard to reach due to the shared resources and other specific food characteristics. OPTIONS FOR SUPPLY CHAIN INTEGRATION IN THE FOOD INDUSTRY The above section made clear that two types of uncertainty are important for managing the supply chain in the food industry: uncertainty in demand and uncertainty in manufacturing due to typical food characteristics. Moreover, the tuning of the processing stage and the packaging stage adds to the complexity of supply chain management for food manufacturers. Each of the two stages has different characteristics. The processing stage is often flexible with respect to the type of product (e.g. the recipe processed in a tank), given the availability of raw materials but inflexible with respect to volume as capacity is limited (e.g. size of a tank). The packaging stage often is inflexible with respect to type of product as lines are dedicated for one (or a few) type(s) of packaging (e.g. only litres or half-litres), but volume flexibility is often considerable because labour is relatively flexible (e.g. adding an extra
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
6
shift). In summary, the challenge for many food manufacturers is to deal with the external and internal uncertainty while attuning the two stages in their process.
Based upon the previous sections and earlier attempts in the literature, we distinguish four different strategies to achieve the performance required:
The first approach is to single out part of the (shared) resources with the purpose of
satisfying demand for one single buyer. More specific, buyer-focused operations aim at reacting to the changes in mix, volume and timing of demand of a specific buyer. In a number of situations the packaging stage might be buyer-focused already if either the volume of one buyer is large enough to justify such or if the type of packaging is buyer specific. However, as indicated previously, the main problem might be the coordination of the processing stage and the intermediate storage. Singling out part of the capacity can only be achieved in case of different lines or capacities. In some cases the processing stage is one source of capacity e.g. a kettle or integrated process. Then, of course, capacity cannot be singled out for one buyer. In other cases, we might have a number of interchangeable kettles. Then we can single out one kettle to serve the needs of one single buyer. The advantage will be that flexibility in mix and delivery can be totally attuned with the buyer to achieve a high level of supply chain integration, although capacity utilisation is likely to decrease.
The second option is to single out part of the capacity for specific buyers virtually. This might be an option if physically singling out resources is not possible either because of technological or financial reasons. Depending on the situation, capacity is allocated to a certain buyer for a fixed number of days each week or a number of hours each day. There is an analogy in the real-life example of the allocation of capacity of an operating theatre in hospitals. Each specialist medicine is given certain time that can be freely used. In food manufacturing, such allocated capacity can be used to produce the different products of the buyer. On the one hand, supply chain integration will be limited, but the flexibility to change priorities and react to uncertainties will be larger within a more or less fixed volume. It seems that one of the prerequisites is that volume uncertainty is not too large and that the packaging stage can react without taking into account other products (limited or no shared resources). This option is comparable with the approach outlined in Lowson et al (1999) for quick response supply chain relationships in the textile industry, where manufacturing capacity is booked and flexibility maintained in order to adjust to unpredictable market demand.
The third way to manage this type of situation is to organise the planning decisions in a hierarchy. This approach goes back to among others Hax and Meal (1975) and the basic idea is to attune decisions at an aggregate level. Within the boundaries of the aggregate plan decisions at lower levels of aggregation can be decoupled, including processing and packaging stage. Van Dam et al. (1998) design such an approach in a case study of a tobacco company. Basically, the demand of each group of products or customers is balanced against the available capacity over a longer period of time e.g. a week or month. Within the planning horizon, each group receives a part of the available capacity that can be filled without any further attuning with other decisions. Here, the division of capacity at an aggregate level is crucial for the success of the approach. Stability of aggregate demand is a prerequisite.. If uncertainty is mainly related to the
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
7
demand within product families (the mix within product families), this might be the ideal way of dealing with uncertainty.
The last possibility stems from the more classical production and operations management approach. The basic idea is that by using all available information regarding orders to be produced, processing times and sophisticated algorithms and software, the problem of attuning the two stages can be reduced to a scheduling problem of finding the optimal order of producing the required product quantities in time. This usually involves mathematical programming techniques ranging from basic linear programming (LP) models to more advanced models based on mixed integer linear programming (MILP). It might be clear that considerable effort is needed to implement this option, as all basic data with regard to processing need to be known, and food-specific characteristics need to be considered in the algorithms. Product shelf life is one of the most important factors in this industry, and has recently been studied in this context by Lütke Entrup et al. (2005). Another prerequisite is that within the scheduling/planning horizon, the number of changes should be minimal. Rescheduling an integrated schedule/plan will cause a lot of organisational disturbance and confusion (e.g. Van Wezel et al., 2006). Rescheduling might also take too much time. It seems therefore that this option is specifically relevant for situations with relatively low levels of uncertainty within the planning horizon, little production disturbances and relatively low complexity of the process.
We realize that each of the above strategies might be appropriate under the circumstances sketched, but each strategy is probably only applicable if the business context (or both types of uncertainty) is more or less homogeneous for all main buyers, or if the production for each buyer can be dealt with independent from the production for the other buyers. If this is not the case (due to e.g. the shared resources), it is not directly clear if and how different strategies can be mixed or applied next to each other for different buyers. INTRODUCTION TO THE CASE STUDY The food manufacturer in this case study is part of a multinational company that operates a large number of plants across the world and serves both consumer and industrial markets. The specific plant under study is large in this type of industry and mainly produces consumer products for both export and domestic markets. The majority of products is produced as own brand, some of them premium brands, but the plant also produces private label products for large retail chains, as well as a limited number of brand-products for other food companies. The variety in products is extensive: both in recipe and in package sizes and labelling. All production is make-to-order for three buyers that are the commercial business units (BUs) of the parent multinational. These BUs stock and distribute the products to a large number of customers around the world. End products have a shelf life between nine months up to two years. Still, products cannot be stored that long, as buyers will not accept relatively short remaining shelf lives. Data collection In the collection of data, a variety of data-gathering techniques was used: mapping of the processes, interviews with employees, reading reports and manuals (for formal procedures), and analysing data with respect to production and demand from the plant’s ERP system. A substantial part of the material was collected by a student as part of his thesis project, complemented with data collection during a project by one of the authors
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
8
and plant visits by the others. The main focus of this project was to investigate the operation of the plant, but we also took interviews with representatives of the business units. While the use of different methods and sources of data already guarantees the quality and reliability of the findings, we also presented the main results and findings at several occasions to the management of the plant to ensure validity of the data and further triangulate the findings. .All in all, we gathered data on the capabilities and limitations of the production system, demand characteristics (mix, volume, uncertainty), the characteristics of the business units, and the planning procedures and practices. Production and process characteristics The production process is typical for food manufacturers. There are two main stages: processing and packaging. The operations in the processing stage can be subdivided into three main categories. The first stage involves preparation activities like, the receiving of raw (natural) materials and the pre-processing of raw materials to achieve homogeneous materials. The second stage is blending batches of different types of raw materials in tanks and adding additives to have the basic recipes. In the third processing stage products are separated in three different product streams, based on the product type (normal, sweetened, and special products). Each op these categories has its own process routings, mainly concerning heat treatment for condensing or pasteurising the (fluid) product. After processing the products are temporarily stored in a large number of intermediate storage tanks.
The packaging stage consists of three departments that package a specific range of packaging sizes and types (cartons, glass bottles, cans). The operations performed consist mostly of sterilizing, packaging (sometimes in reversed order), labelling, case-packaging and palletising. It is important to note that all product types can be used by all packaging departments. The flow of products is summarised in Figure 2, and is mainly characterised by product types in the first stage and by packaging types in the second stage. This characterisation also causes the intermediate storage tanks to be quite important in the control of the production system.
As indicated above, the plant has three buyers: the commercial BUs of the parent company that are responsible for the contacts with the customers and for inventory control. Each of the BUs has distinct characteristics and different types of customers in diverse markets:
• BU Export delivers products to a number of European, Asian and African Markets. Here a number of well-established brands are delivered to partly independently operating foreign buyers, which distribute and sell the products in their countries. In general, demand is unpredictable. The timing of the deliveries is partly dependent on shipping dates.
• BU Home Market sells and distributes the own well-established brand to all retail chains and some other distributors as well as some retail brands for large retailers. Achieving an almost 100% customer service is one of the main objectives, as well as good cooperation with major buyers to support
preparationBlen-ding
Packaging/sterilizing
Packaging/sterilizing
Packaging/sterilizing
Processing Specials
Processing
Processing Sweet
Figure 2. Flow of goods.
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
9
promotional activities. In general, the demand of the consumers is relatively predictable and stable.
• BU Supply maintains and supports buyers that outsource their production to the focal plant. Supply started as a way to utilise excess capacity, but is now a significant part of the total business. Here, less influence and insight exists with respect to demand and demand patterns, but fluctuations in capacity usage are more or less restricted by agreements with respect to total capacity and number of batches. The forecast accuracy differs between the buyers.
As is the case for most food manufacturers, this plant has experienced a steady increase in the number of SKUs over the years. The number of recipes increased due to the increased pressure for healthy and low-fat food and variety in taste and ingredients, while also the number of packaging sizes and types increased due to demographic reasons (e.g. on average smaller sizes of households), increase in brands and demand for easy-to-use products. The increase in both recipes and packaging types, while total demand is staying the same, naturally reduces batch sizes in both stages of production. Batch sizes are further decreased as a result of stock reductions in the supply chain. To some extent, this is problematic, as the plant was originally developed (as many food manufacturing plants are) to produce large batches.
More and more it is felt that whereas packaging can cope with fluctuations and due dates, the processing department has problems in producing the required amounts. In fact, processing has become the bottleneck of the whole process, whereas it previously could supply the packaging departments without problems. The increase in recipes and the reduced batch sizes cause more set-ups and cleaning time than before. All things considered, it is clear that the plant under consideration is finding itself precisely in the situation sketched earlier. Each of the various factors from the theoretical introduction will be further analysed in the next section. ANALYSIS OF THE CASE STUDY Figure 1 is the starting point for our analysis: assessing the fit between business conditions and the type of resources. More specifically, the aim of this section is to analyse and confront demand (demand uncertainty) and production process characteristics. The third part of our analysis relates to the role of planning in handling demand and its uncertainties. In other words, we explore the fit (or lack of it) between demand and production characteristics and planning. Demand Our analysis shows that the batch sizes are decreasing both at the recipe and the packaging level. This can be illustrated by the total number of SKUs produced (about 590 each month for all packaging departments), which is increasing with a yearly rate of about 3. It can also be illustrated by the difference between expected and actual number of SKUs produced for one particular packaging department: 154 expected and174 produced.. The increase in the number of recipes produced each month is 10%, partly due to the introduction of new recipes.
The demand pattern and uncertainty in demand are rather different between the three BUs:
• BU Export keeps a close contact with their customers and forecasts monthly demand over a horizon of three months, based on forecasts of the customers and a number of important factors. However, the average forecasts suffer from a very low accuracy (about 44% lies outside the preset accuracies). For
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
10
some products the actual demand differs 100% from the forecasts. The knowledge of upstream inventories is weak and delivery lead times of these export products are one month.
• BU Home Market operates in a rather stable market and sells a large variety of different packaging sizes and types. The BU keeps stocks of all products, both own brand and retail brands. Demand is forecasted on a weekly base (with a horizon of 13 weeks) and production orders are based upon demand and stock positions. The main deviations here are caused by promotional activities, which are generally known beforehand.
• BU Supply receives estimated demand for a year of most customers. At the operational level an 8-week rolling forecast is provided. The reliability of the rolling forecast differs among customers: some provide more or less lumpy, hardly forecasted demand, while others have the ability to make reliable forecasts. In general minimum batch sizes are agreed upon.
For the processing department, production is based on type of recipe, and orders for different SKUs can often be combined into one processing order if it involves the same recipe. Still, on the recipe level, a lot of variety exists in the demand patterns; some recipes are produced every week in about the same volume, but a lot of recipes have a more irregular pattern (in volume and order size). This is illustrated in Figure 3, where for all recipes, the average weekly recipe volume is plotted against the average time between two orders for that specific recipe. It should be noted that to ensure confidentiality, the volumes have been multiplied by a constant. Obviously, this does not affect the structure of the graphic. The figure clearly shows the large differences between recipes, both in volume and regularity. Production Just looking at the performed operations, production seems relatively simple as it basically concerns mixing, processing, packaging and preservation. However, the amount of lines and products adds to complexity. The large number of routings possible and the connections and relations between packaging lines and processing stages further increases the complexity. Packaging lines within one department use common
Figure 3. Demand pattern on the recipe level.
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
11
equipment (shared resources); lines in different departments package the same recipe and might get the intermediate product from the same storage tank.
In production, we also see a number of typical food characteristics. Limited shelf life of some (and certainly the main) raw material induces the need for cleaning after a certain time period but also between different recipes. Contamination of different products is usually seen as a large problem both from a quality and hygienic perspective. As production speeds differ and because the processing is batch oriented, the processing and packaging stage are separated by a number of tanks. The three processes in the processing stage and the tanks are all more or less general purpose: serving a broad range of products/recipes. Although there are quite some intermediate storage tanks, the increase in the number of recipes, combined with the decrease in batch sizes, results in extremely high utilisation rates of the tanks, although the average tank content drops. The packaging departments have lines that are more labour intensive and dedicated to one type of packaging (e.g. glass bottle of 0.5 litre). Some packaging lines are even producing for just one end user or one BU. Here, cleaning is also an issue. Most packaging lines operate at high speed, but seem to be vulnerable to breakdowns. The due date performance is, as a result, rather fluctuating. The result is that the intermediate storage is longer occupied and the processing stage and specifically the special products lines that have limited capacity cannot produce at full speed. Planning Planning needs to balance demand and capacity at various levels over time and at the same time assure supply of raw material. Specifically with respect to the main raw material, coordination takes place at various levels and plans are adapted at various moments in time to assure optimal supply. Due to the possibility to balance the supply of this factory with others, supply of raw materials is generally not a problem. As the company as a whole has a policy of being market-oriented and market-driven, the starting point for planning are the packaging departments. In general, their plans form the basis for the plans of the processing department, including planning of the required raw material. Here, a capacity check is made at various levels (monthly, weekly and daily plan). In general, inventories of finished products are hardly considered in the planning process, as stocks are kept by the customers (for BUs Export and Supply) or by the buyers (in case of BU Home Market). Furthermore, it is felt that coordination at a monthly level is insufficient and that too many adaptations have to be made to the more detailed plans. Finally, due to the vulnerability to breakdowns in the packaging departments, a lot of rescheduling is done on the operational level (also affecting the processing stage through the strong interrelationship). Conclusion All in all, the conclusion is that the business conditions are complex due to the unpredictability of demand, the process interactions between departments and the uncertainty in production. In complex business conditions, one would like to have a high level of integration in the chain, but due to the shared resources and some of the specific food characteristics, this seems hard to achieve. It is also clear that there is hardly any difference between the different buyers with respect to integration, although they have quite different characteristics (summarised in Table 1). The next section explores to what extent the four basic strategies are applicable.
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
12
Table 1. Main characteristics in the case study.
Scope Characteristics:
Products Customers Demand Forecast Production Planning
General Increase in number of SKUs and recipes
Decreasing order sizes
Process inter-actions, packaging breakdowns, shared resources
On various levels, market-driven, lot of rescheduling
BU Home Large variety of premium and retail brands
Retail chains and distributors
Relatively stable, some deviation by promotional activities
13 weeks, updated every week, high quality
BU Export Well-established premium brands
Foreign distributors
unpredictable 12 weeks, updated every 4 weeks, low quality
BU Supply Processing and packaging capacity
Food companies Contract agreements
1 year, updated every 8 weeks, quality varies between buyers
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
13
REDESIGNING THE CASE It might be obvious that choosing one of the suggested strategies is the preferred course of action for changing the current situation. Before investigating that choice, it should be stressed that from a SCM point of view, it is not that bad to start planning with actual demand and due date setting, and with the planning of the packaging departments. Due to the flow of goods between the processing and packaging departments, complex interrelationships come into being. The shared resources in the processing stage limit total output both in the long and short run, and output is also partly determined by the type and size of packaging orders and the disturbances in packaging. It seems logical to pay attention to the operational constraints (Bertrand et al., 1990) on the orders accepted and planned for packaging. Bertrand et al. mention batching constraints (e.g. to avoid set-ups), sequence constraints (e.g. to combine work orders), workload constraints (to realize a certain level of utilisation), and capacity constraints (possible adjustments in the short and long run). A second concern is that the three BU's and their customers differ in type and the ability to forecast demand and thus pose different requirements on the production system.
If we consider the analysis of the case and the above conclusions, it seems that each of the proposed alternative strategies is hard to implement as an overall solution (see Table 2).
Each of the alternatives, however, can be used at a lower level of analysis. Our concern is thus to split the overall complexity into relatively manageable parts. For that purpose a number of observations can be made: (i) the BUs differ in level of uncertainty, (ii) some packaging lines are almost dedicated to buyers (or buyer-focused), and (iii) for a selection of products, volumes are large and stable –in Figure 3, these are the products/recipes in the lower-right corner. The relationship with BU Home Market has relatively little uncertainty, a number of specific recipes and packaging types, and there is a low level of integration with the buyer. In principle according to Figure 1, this might be fine. Part of the buyer-focused strategy can be used here. The exchange of more information, better and joint decision making e.g. with regard to inventory and batch sizes could yield some extra flexibility to cope with the uncertainty of the other business units. For a number of products it is beneficial to increase batch sizes if all costs are considered (labour, waste, etc.). Currently, such trade-offs along the supply chain are hardly made. For some of the high-volume recipes it could be beneficial to single out part of the capacity in the processing stage to integrate it with the packaging lines that are already buyer-focused. Here, Figure 3 could be used as a guideline in the selection of recipes, also taking into account their regularity. This separation of capacity can be achieved virtually by reserving capacity each week or on certain days. Actual processing and packaging
Table 2. Applicability of integration strategies in the case Supply chain integration strategy Limitation / barrier Buyer-focused operations Volume uncertainties
Capacity used for several buyers Virtual buyer-focused operations Volume uncertainties Aggregated hierarchical planning Processing and packaging not fully decoupled
Different demand characteristics for buyers Integrated planning and scheduling Uncertainties in the market
Production induces frequent rescheduling High complexity of the plant
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
14
might be postponed until relatively late in the planning process to produce those packaging sizes that are most needed for replenishing (change of capacity constraint). The positive effect can be increased delivery performance if for these products a limited number of fixed tanks is used such that disturbance here do not influence other parts of the process. An integrated scheduling and planning approach could manage this part of the supply chain (last strategy).
The workload and capacity constraints also need to be changed in order to maintain supply chain control. As a result of the market-orientation, batches tend to become smaller (recipes move left in Figure 3). Given the considerable amount of cleaning time, the influence of batch sizes on capacity utilisation is rather large. So far, too much emphasis has been put on overall volume of products, while from a capacity point of view, the number of cleaning and set-up times can be directly incorporated. Specifically for BU Supply this will result in proper agreements with customers to really sell capacity. That implies that within a given volume the number of different recipes (and thus the amount of cleaning time) will be restricted. Here in fact the operational constraints in processing and packaging can result in the adoption of the third strategy of aggregate hierarchical planning at a high level that is detailed in a later stage within the agreed constraints.
For the BU Export it seems that the current efforts paid to forecast demand is a waste of time. The uncertainty in demand is not really a problem as the lead-time is about one month. The preferred course of action is to invest in developing the tools to schedule the orders that come in. Good planning and scheduling to be able to process the orders, fast delivery of supplies and coordination with transport are the main instruments to increase performance. CONCLUSIONS AND DISCUSSION This paper aims at developing a better understanding of the specific problems of food manufacturers that aim at supply chain integration. While it seems that integration is specifically high in food supply chains, we show that the specific nature of food manufacturing companies and specifically the shared resources that are operated in such companies can be barriers for integration. Specific factors are the increase in product variety, smaller batch sizes and uncertainties in demand, combined with limited shelf life of products and processing uncertainties. We discuss four basic strategies to deal with these circumstances: singling out buyer-focused resources, virtual buyer focus, hierarchical planning, and integrated planning and scheduling.
A case study is used to illustrate the concepts and relationships developed. The case clearly shows the problems that have to be dealt with by food manufacturers. The four strategies developed are applicable to improve supply integration and performance, if different types of demand are dealt with separately, linked to specific characteristics and to the structure of the process. Dealing with the shared resources is possible, but they will remain a major factor in supply chain improvements. We realize that the empirical part of this paper concerns only a single case study, but based on existing research on the characteristics of food manufacturing, we feel that this case study represents a typical food manufacturer.
For the longer run it seems that constantly monitoring the product portfolio both in terms of SKUs and recipes and their profitability is needed. A second point related to that is the possibility to change the point at which products become specific. Most products consist basically and for the larger part of the same raw materials and only their relative percentages and some ingredients differ, the specific recipes are mixed
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
15
before the processing stage. Considerable amounts of cleaning time could be saved if mixing could be postponed until just before packaging. It seems that technological progress will allow for that soon. It should be noticed, however, that while such solutions are promising, they do not solve the fundamental problem of shared resources.
REFERENCES Akkerman, R., Van Donk, D.P. and Gaalman, G.J.C. (2007), “The influence of capacity- and
time-constrained intermediate storage in two-stage food production systems”, International Journal of Production Research, Vol 45, No 13, pp. 2955-2973.
Bertrand, J.W.M., Wortmann, J.C. and Wijngaard, J. (1990), Production control: A structural and design oriented approach, Elsevier, Amsterdam.
Childerhouse, P. and Towill, D.R., (2002), “Analysis of the factors affecting the real-world value stream performance”, International Journal of Production Research, Vol 40 No 15, pp. 3499-3518.
Davis, T. (1993), “Effective supply chain management”, Sloan Management Review, Vol 34 No 4, pp. 35-46.
Dennis, D.R. and Meredith, J.R. (2000), “An analysis of process industry production and inventory management systems”, Journal of Operations Management, Vol 18, pp. 683-699.
Dobson, P.W., Clarke, R., Davies, S. and Waterson, M. (2001), “Buyer power and its impact on competition in the food retail distribution sector of the European Union”, Journal of Industry, Competition and Trade, Vol 1, No 3, pp. 247-281.
Duffy, R., Fearne, A. and Hornibrook S. (2003), “Measuring distributive and procedural justice: An exploratory investigation of the fairness of retailer-supplier relationships in the UK food industry”, British Food Journal, Vol 105 No 10, pp. 682-694.
Fisher, M. (1997), “What is the right supply chain for your product?”, Harvard Business Review, March/April, pp. 105-116.
Fliedner, G. (2003), “CPFR: an emerging supply chain tool”, Industrial Management & Data Systems, Vol 103, No 1, pp. 14-21.
Frohlich, M.T. and Westbrook, R. (2001), “Arcs of integration: an international study of supply chain strategies”, Journal of Operations Management, Vol 19, pp. 185-200.
Hax, A.C. and Meal, H.C. (1975), “Hierarchical integration of production planning and scheduling”, in Geisler, M.A. (ed.), Studies in Management Science Vol 1: Logistics, Chapter 1, Elsevier, New York.
Hendrickson, M., Heffernan, W.D., Howard, P.H. and Heffernan J.B. (2001), “Conslolidation in food retailing and dairy”, British Food Journal, Vol 103 No 10, pp. 715-728.
Hill, C. A. and Scudder, G. D. (2002), “The use of electronic data interchange for supply chain coordination in the food industry”, Journal of Operations Management, Vol 20, pp. 375-387.
Hutchins, R (1997), “Category management in the food industry: a research agenda”, British Food Journal, Vol 99, No 5, pp. 177-180.
Ho, D.C.K., Au, K.F. and Newton, E. (2002), “Empirical research on supply chain management: a critical review and recommendations”, International Journal of Production Research, Vol 40 No 17, pp. 4415-4430.
Kinsey, J. (2003), “Emerging trends in the new food economy: Consumers, firms, and science”, in Proceedings of the Conference on Changing Dimensions of the Food Economy: Exploring the Policy Issues, 6-7 February 2003, The Hague, Netherlands.
Lee, H.L. (2002), “Aligning Supply Chain Strategies with Product Uncertainties”, Sloan Management Review, Vol 44 No 3, pp 105-119.
Lowson, B., King, R. and Hunter, A. (1999), Quick Response: Managing the Supply Chain to Meet Customer Demand, John Wiley & Sons, West Sussex, England.
Lütke Entrup, M., Günther, H.-O., Van Beek, P., Grunow, M. and Seiler, T. (2005), “Mixed-integer linear programming approaches to shelf-life-integrated planning and scheduling in
D.P. van Donk, R. Akkerman, and J.T. van der Vaart (2007), Opportunities and realities of supply chain integration: The case of food manufacturers, British Food Journal, Vol. 110, No. 2, pp. 218-235.
16
yoghurt production”, International Journal of Production Research, Vol 43, No 23, pp.5071-5100.
Meulenberg, M.T.G. and Viaene, J. (1998), “Changing food marketing systems in western countries”, in Jongen, W.M.F. and Meulenberg, M.T.G. (Eds.), Innovation of Food Production Systems: Product Quality and Consumer Acceptance, Wageningen Press, Wageningen, pp. 8-36.
Morgan, N.A., Kaleka, A. & Gooner, R.A. (2007), “Focal supplier opportunism in supermarket retailer category management”, Journal of Operations Management, Vol 25, no 2, pp. 512-527.
Ramdas, K. and Spekman, R.E. (2000), “Chain or Shackles: understanding what drives supply chain performance”, Interfaces, Vol 30 No 4, pp. 3-21.
Schwartz, L.B. (2004): “The state of practice in supply-chain management: a research perspective”, in Akçali, E., Geunes, J., Pardoles, P.M., Romeijn, H.E. and Shen, Z.J. (eds.), Applications of Supply Chain Management and E-Commerce Research in Industry, Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 1-36.
Stevens, G.C. (1989): “Integrating the supply chain”, International Journal of Physical Distribution & Material Management, Vol 19, No 8, pp. 3-8.
Van Dam, J.P., Gaalman, G.J.C. and Sierksma, G. (1998), “Designing scheduling systems for packaging lines in process industries: A tobacco company case”, International Journal of Production Economics, Vol 56-57, pp. 649-659.
Van der Vaart, T. and Van Donk, D.P. (2004), “Buyer focus: Evaluation of a new concept for supply chain integration”, International Journal of Production Economics, Vol 92 No 1, pp. 21-30.
Van der Vorst, J. G. A. J. and Beulens, A. J. M. (2002), “Identifying sources of uncertainty to generate supply chain redesign strategies”, International Journal of Physical Distribution & Logistics Management, Vol 32 No 6, pp. 409-430.
Van Donk, D.P. (2000), “Customer Driven Manufacturing in the Food Processing Industry”, British Food Journal, Vol 102 No 10, pp. 739-747.
Van Donk, D.P. and Van der Vaart, T. (2004), “Business conditions, shared resources and integrative practices in the supply chain”, Journal of Purchasing & Supply Management, Vol 10 No 3, pp. 107-116.
Van Wezel, W., Van Donk, D.P. and Gaalman, G. (2006), “The Planning Flexibility Bottleneck in Food Processing Industries”, Journal of Operations Management, Vol 24, No 3, pp. 287-300.
Vickery, S.K., Jayaram, J., Droge, C. and Calantone, R. (2003), “The effects of an integrative supply chain strategy on customer service and financial performance: an analysis of direct versus indirect relationships”, Journal of Operations Management, Vol 21, pp. 532-539.
Whiteoak, P. (1999), “The realities of quick response in the grocery sector; a supplier viewpoint”, International Journal of Physical Distribution and Logistics Management,, Vol 29, No 7/8, pp. 508-519.
All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.
ing Planning and Control Systems. Irwin/McGraw-Hill,
New York.
Whetten, D.A., 1989. What constitutes a theoretical contribution.
Academy of Management Review 14 (4), 490–495.
Ž .Journal of Operations Management 16 1998 361–385
A definition of theory: research guidelines for differenttheory-building research methods in operations management
John G. Wacker )
Department of Management, Iowa State UniÕersity, Ames, IA 50011-2065, USA
Abstract
This study examines the definition of theory and the implications it has for the theory-building research. By definition,theory must have four basic criteria: conceptual definitions, domain limitations, relationship-building, and predictions.Theory-building is important because it provides a framework for analysis, facilitates the efficient development of the field,and is needed for the applicability to practical real world problems. To be good theory, a theory must follow the virtuesŽ .criteria for ‘good’ theory, including uniqueness, parsimony, conservation, generalizability, fecundity, internal consistency,empirical riskiness, and abstraction, which apply to all research methods. Theory-building research seeks to find similaritiesacross many different domains to increase its abstraction level and its importance. The procedure for good theory-buildingresearch follows the definition of theory: it defines the variables, specifies the domain, builds internally consistentrelationships, and makes specific predictions. If operations management theory is to become integrative, the procedure forgood theory-building research should have similar research procedures, regardless of the research methodology used. The
Ž .empirical results from a study of operations management over the last 5 years 1991–1995 indicate imbalances in researchmethodologies for theory-building. The analytical mathematical research methodology is by far the most popular methodol-ogy and appears to be over-researched. On the other hand, the integrative research areas of analytical statistical and theestablishment of causal relationships are under-researched. This leads to the conclusion that theory-building in operationsmanagement is not developing evenly across all methodologies. Last, this study offers specific guidelines for theory-buildersto increase the theory’s level of abstraction and the theory’s significance for operations managers. q 1998 Elsevier ScienceB.V. All rights reserved.
Keywords: Theory; Research guidelines; Operations management
1. Introduction
Although many business professionals, social sci-entists, and other academics have very similar beliefson the definition of theory, there are still somedifferences of opinion in a theory’s exact nature. Forexample, some practitioners and academics believe
that theory and its application are very limited and,therefore, not very useful in the real world of busi-ness. Others feel that very little theory exists in theacademic world. For example, consider the followingdiscussions of theory and scientific investigation:
Theory, for theory’s sake, can easily degenerate intoan uninteresting art form. Yet, practice without the-ory can quickly become a dull and dangerous occu-pation. Unfortunately, the world is a complicatedplace and complicated solutions and processes are
0272-6963r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved.Ž .PII S0272-6963 98 00019-9
( )J.G. WackerrJournal of Operations Management 16 1998 361–385362
often required to make complex organizations run.The ability to live with uncertainty and the insightinto both one’s professional powers and limitations
Žis the sign of a mature management science Shubik,.1987 .
. . . Of all our valid knowledge of the social world,most of it seems to have been the product of layrather than professional inquiry . . . A typical situa-tion in social science is that scientific inquiry onlymodestly raises the validity of a lay proposition by
Ž .qualifying it Lindblom, 1987, p. 517 .
Ž .. . . N o theory of inventory exists unless the theorydictates the manner in which the measurements ofthe parameters of the theory are to be made. Thisamounts to saying that today we have no theory ofinventory. We will only have such a theory when wecan specify the information necessary to test the
Žtheory and justify this specification Churchman,.1961, 132 .
The implication of the first statement is that the-ory does not necessarily require application. Thisstatement means that theory is abstract and does nothave to be applied or tested to be a ‘good’ theory. Ifthis is the case, then theory can remain totally ab-stract and non-applied. The crux of whether theorycan be totally abstract and non-applied depends onthe definition of theory.
The second statement implies that scientific inves-tigation and theory-building are not necessarily use-ful for the development of social science or manage-rial decision-making fields of academic study. Thisstatement implies that ‘good’ theory is ‘found’through trial and error rather than through scientificinvestigation in a systematic matter. Further, thisstatement implies that ‘good’ theory-building in so-cial sciences and managerial decision-making is de-rived from lay investigation rather than by scientificinvestigation. The validity of this statement dependsupon how theory is defined.
The third statement implies that unless theoryexplicitly indicates how it is measured, it is not a
Ž .theory. Churchman 1961 states that since inventorytheories do not indicate how they are to be measured,they are not theories. The statement infers that atheory should offer how it is to be measured for
empirical testing, and without this testing, a theorycannot be ‘good’ theory. Churchman’s conclusionthat no inventory theory exists in management alsodepends upon the definition of theory.
All three of the above statements cause concernsfor academics since they infer three common criti-
Ž .cisms of theory: 1 theory does not have to beŽ .applied, 2 it does not make significant improve-
Ž .ments in the external world, and 3 theory does notexist due to lack of measurement of definitions. Eachconcern has some measure of truth. Unless opera-tions management research addresses relevant practi-cal problems to explain complex phenomena, it can-not develop into a theory-building discipline. Each ofthe above concerns depends upon the definition oftheory and, more importantly, what criteria is used todevelop ‘good’ theory. It does not seem logical tocall any theory a ‘good’ theory without definingtheory or the criteria for ‘good’ theory.
Besides these concerns, there are additional rea-sons why theory is important for researches and
Ž .practitioners: 1 it provides a framework for analy-Ž .sis; 2 it provides an efficient method for field
Ž .development; and 3 it provides clear explanationsfor the pragmatic world. The first reason is that goodtheory provides a framework of analysis for opera-tions management since it provides structure forwhere differences of opinion exist. For example, a‘theory of internationally competitive manufactur-ing’ would provide a structure to evaluate factories’manufacturing international competitiveness. Thistheory enables academics to enumerate the exactconditions to classify firms into degrees of interna-tional competitiveness. Since it is unlikely that allacademics would agree on which factors are mostimportant in a hierarchy of factors, a framework of‘good’ theory procedures provides an understandingof where these differences of opinion lie.
The second reason for developing theory is effi-ciency. Theory development reduces errors in prob-lem-solving by building upon current theory. Build-ing upon current theory is equivalent to incorporat-ing all that is known from the current literatureŽtheoretical, mathematical, empirical, and practi-
.tioner research into a single, integrated consistentbody of knowledge. For researchers, using a singleintegrated body of knowledge for analytical andempirical testing gives the results a deeper theoreti-
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 363
cal meaning by differentiating between the compet-ing theories. An integrated body of knowledge canonly be pursued efficiently if integrated theory isdeveloped through consistent theory-buildingmethodologies.
A third reason why theory is important is itsapplicability. Consider this example: a group of man-agers hires a consultant to improve their firm’s com-petitiveness. The consultant provides a set of sugges-tions. If the consultant does not provide anythingelse, there is little reason to be surprised if themanagers disregard these suggestions, since man-agers need more rationale than a set of rules. Morespecifically, they want to know why this set of rules
Žhold for their specific manufacturing facility what,.who, where, when and how and why these specific
Žrelationships could, would, or should improve pre-.diction claim their manufacturing facility. Although
managers seek advice, it is essential for them to beskeptical of any suggestions since they are directlyresponsible for both good and bad results. Conse-quently, they must logically be convinced that noth-ing important is overlooked. They want the analyti-cal reassurance that these suggestions are logicallyand practically compatible with each other. Addition-ally, they want to know specific instances where
Ž .these rules have been successful empirical support .Therefore, even in the most practical of instances,use of the formal definition of theory is important forall managers wishing to achieve measurable results.For this reason, many academics believe that: ‘‘There
Žis nothing as practical as a good theory’’ Hunt,.1991; Van de Ven, 1989 . ‘‘Good theory is practical
precisely because it advances knowledge in a scien-tific discipline, guides research toward crucial ques-tions, and enlightens the profession of management’’Ž .Van de Ven, 1989 .
Because of the need for ‘good’ theory, the pur-pose of this study is to set guidelines for integratingtheory-building research. In order to achieve thislofty goal, this study first determines what a ‘good’theory is. Next, it closely ties ‘good’ theory tospecific research procedures. After the good-theory-building procedures are outlined, this study catego-rizes different research methodologies by their the-ory-building purpose and depicts how all theory-building procedures are similar. Then this study em-pirically investigates the prevalence of different re-
search methodologies in operations management todetermine where operations management is weak intheory-building research. Last, the article concludeswith the guidelines for theory-building research. Asare all articles suggesting theory-building and re-search investigation, this article is an opinion articlesince ‘‘the value laden nature of theory can never be
Ž .eliminated’’ Bacharach, 1989 . Therefore, the opin-ions expressed here should be evaluated on theirlogical appeal and internal consistency.
2. A definition of theory
Before beginning any discussion on theory, thisstudy must differentiate between the common notionof ‘theory’ and a formal definition of theory. In thisarticle, the term theory is interpreted as following theformal definition and operationalization of theory.This operationalization of the definition of theoryshould directly be tied to the necessary componentsof theory. Generally, academics point to a theory as
Ž . 1being made up of four components, 1 definitionsŽ . 2of terms or variables, 2 a domain where the
Ž .theory applies, 3 a set of relationships of variables,Ž . Ž . Žand 4 specific predictions factual claims Hunt,
.1991; Bunge, 1967; Reynolds, 1971 . Theories care-fully outline the precise definitions in a specificdomain to explain why and how the relationships are
1 Theoretical definitions are not observable as such. Rather,their existence and properties are asserted in order to account forwhat is observable. Theoretical definitions are conceptual in na-ture. Even a relatively simple term such as manufacturing leadtime has conceptual foundations that transcend its measurement.By definition, manufacturing lead time is ‘‘The total time requiredto manufacture an item, exclusive of lower level purchasing time’’Ž .Crawford, 1987 . Conceptually, manufacturing lead time repre-sents the internal time a manufacturing system takes to manufac-ture an item. This concept is not observable because the precisemoments from when all materials become available to when theorder is completed cannot be exactly specified.
2 Domain of a theory: the domain of the theory is the exactsetting or circumstances where the theory can be applied. Forexample, a just-in-time theory domain may be manufacturingfacilities that focus on few products and compete on cost withacceptable quality. The just-in-time theory domain would giveextensive factors that limit the instances of when and where it isapplied.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385364
logically tied so that the theory gives specific predic-tions. Therefore, the precision of good theory causesa theory to be very exacting for all the key compo-
Ž .nents of a theory, or as Poole 1989 and Van de VenŽ .1989 state: ‘‘A good theory is, by definition, alimited and fairly precise picture.’’ A theory’s preci-sion and limitations are founded in the definitions ofterms, the domain of the theory, the explanation ofrelationships, and the specific predictions.
Authors usually agree that the goal of ‘good’theory is a clear explanation of how and why spe-cific relationships lead to specific events. Conse-quently, these explanations of relationships are criti-cal for ‘good’ theory-building. Other authors’ state-ments on theory indicate the importance of relation-ship-building:
Ž .Theory is . . . an ordered set of assertions about ageneric behavior or structure assumed to holdthroughout a significantly broad range of specific
Ž .instances Sutherland, 1976: 9 .
Researchers can define theory as a statement ofrelationships between units observed or approxi-mated in the empirical world. Approximated unitsmean constructs, which by their very nature cannotbe observed directly. . . . A theory may be viewed asa system of constructs and variables in which theconstructs are related to each other by propositionsand the variables are related to each other by hy-
Ž .potheses Bacharach, 1989 .
These statements indicate the importance of rela-tionship-building in explaining how and why specificphenomena will occur. Sometimes how and why andspecific predictions are condensed into the expres-sion ‘adequate explanation’, which implies that un-less an explanation can predict, it is not considered
Ž .adequate Hunt, 1991 .A very important aspect of a theory definition is
phrased in the common questions that researchersrequire to exactly specify a theory. Consider thisstatement: ‘‘The primary goal of a theory is to
Ž .answer the questions of how, when or where , andwhy . . . unlike the goal of description, which is to
Ž . Žanswer the question of what or who ’’ Bacharach,.1989 . In short, any definition of theory should
answer common questions that researchers face. First,theory defines all variables by answering the com-
mon questions of who and what. The domain speci-fies the conditions where the theory is expected tohold by using the common questions of when andwhere. The relationship-building stage specifies thereasoning by explaining how and why variables arerelated. And last, the predictive claims specify thewhether ‘‘Could a specific event occur?’’, ‘‘Should aspecific event occur?’’, and ‘‘Would a specific eventoccur?’’ In short, the definition of theory providesguidelines to answer the common questions thatoccur in natural language. From the pragmatic per-spective of operations managers, the predictive claimsfrom theory answer the could, should, and wouldquestions which are quite critical for managers’ fu-ture success. Consequently, the should, could andwould questions are very important for theory to beconsidered useful to managers.
In summary, the definition of theory suggested bythis study has these four components: definitions,domain, relationships, and predictive claims to an-swer the natural language questions of who, what,when, where, how, why, should, could and would.
3. Virtues of ‘good’ theory
The purpose of this section is to limit the domainof theory-building to the domain of what PopperŽ .1957 calls ‘good’ theory. Theories in the domain of‘good’ theory are superior to other theories becausethey possess what philosophers of science call the
important because researchers must evaluate the rela-tive significance of opposing theories. Although thereis no general agreement between philosophers ofscience concerning the relative importance of eachvirtue of ‘good’ theory, there seems to be a fairlywidespread agreement as to what they are. A com-mon set of virtues of a theory is: uniqueness, parsi-mony, conservatism, generalizability, fecundity, in-ternal consistency, empirical riskiness, and abstrac-
Ž .tion Quine and Ullian, 1980 .Table 1 provides an overview of the virtues of
‘good’ theory. The first five virtues are fairlystraightforward: each theory must be differentiated
Ž .from other theories uniqueness : new theories can-not replace existing theories unless they are better
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 365
Table 1The virtues of ‘good’ theory: key features and why they are important to ‘good’ theory development
Virtue Key feature Why important for ‘good’ theory and for the develop-ment of the field
Uniqueness The uniqueness virtue means that one theory must If two theories are identical, they should be consideredbe differentiated from another. a single theory. Although it applies to all criteria for
theory, this virtue directly applies to definitions sincedefinitions are the most elemental of building blocksfor theory.
Conservatism A current theory cannot be replaced unless the new Therefore, current theory is not rejected for the sake oftheory is superior in its virtues. change. This criteria is needed so that when a new
theory is proposed, there is a good reason to believe allŽother theories are lacking in some virtue Quine and
.Ullian, 1980; Kuhn, 1980; Popper, 1957 .Generalizability The more areas that a theory can be applied to If one theory can be applied to one type of environ-
makes the theory a better theory. ment and another theory can be applied to manyenvironments, then the second theory is a morevirtuous theory since it can be more widely applied.Some authors call this virtue the utility of the theorysince those theories that have wider application havemore importance.
Fecundity A theory which is more fertile in generating new Theories which expand the area of investigation intomodels and hypotheses is better than a theory that new conceptual areas are considered superior to theo-has fewer hypotheses ries which investigate established research areas. This
meansTheory parsimony theory The parsimony virtue states, other things being If two theories are equal in all other aspects, the onesimplicity, theory effi- equal, the fewer the assumptions the better. with fewer assumptions and the fewer definitions isciency Occum’s razor more virtuous. This virtue also includes the notion that
the simpler the explanation, the better the theory. Thisvirtue keeps theories from becoming too complex andincomprehensible.
Internal consistency Internal consistency means the theory has identified Internal consistency refutation means that the theoryall relationships and gives adequate explanation. logically explains the relationships between variables.
The more logically the theory explains the variablesand predicts the subsequent event, the better the theoryis. This internal consistency virtue means that thetheory’s entities and relationships must be internallycompatible using symbolic logic or mathematics. Thisinternal consistency means that the concepts andrelationships are logically compatible with each other.
Empirical riskiness Any empirical test of a theory should be risky. If there are two competing theories, the theory thatRefutation must be very possible if theory is to be predicts the most unlikely event is considered theconsidered a ‘good’ theory. superior theory. In the opposite case, if the theory
predicts a very likely event then it is not seen as beinga very valuable theory. This criteria is sometimes putin a different way: ‘‘Every good theory has at least oneprohibition; it prohibits certain things from happening’’Ž .Popper, 1957 .
Abstraction The abstraction level of theory means it is indepen- The abstraction level means it is better to integratedent of time and space. It achieves this indepen- many relationships and variables into a larger theory. Ifdence by including more relationships. one of two competing theories integrates more inter-
nally consistent concepts, it is more virtuous than atheory that integrates fewer internally consistent rela-tionships.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385366
Žtheories evaluated in the light of their virtues con-.servation : new theories can be introduced that have
Žrich new areas of investigation fecundity virtue,some academics would call this area grounds for a
.paradigm shift ; and last, any theory should not beŽ .overly complex theory parsimony .
However, the remaining virtues are worthy ofadditional discussion since they are the focus of mostof the academic discussion. There are two theoryrefutation virtues: internal consistency and empiricalriskiness. The internal consistency virtue implies thatthe theory is internally consistent using either mathe-
Ž .matics or symbolic logic Bacharach, 1989 . Withoutinternal consistency, each relationship is independentfrom every other relationships causing the theory tonot be integrated within itself. This virtue is differentfrom face validity because each relationship in atheory may be explained individually, and thereforethe relationship may have face validity. However,each relationship may not be consistent with otherrelationships in the theory, thus, causing the entiretheory to be internally inconsistent. Since theories tietogether many concepts, it usually is not easy toidentify internal inconsistency. Therefore, it is neces-sary to develop the theory using mathematics orsymbolic logic.
The criterion of empirical riskiness has been thefocus of most of critical evaluators of ‘good’ theory.Most academics believe that empirical tests of theoryshould be risky so that there is a good chance of thetheory being refuted. On the other hand, if an empiri-cal test of theory supports the occurrence of a likelyevent, then the theory is deemed to be a ‘weak’theory. One empirical riskiness criteria is frequentlycalled the ‘single event refutation’ criteria whichstates that it takes only one sample to question atheory’s legitimacy. Put another way, every legiti-mate empirical test is designed to disprove the theory
Ž .and should be risky Popper, 1957 .Sometimes the two refutation criteria are not com-
patible. The incompatibility between the two refuta-tion criteria typically occurs when empirical evi-dence does not support an existing theory. What arethe deciding criteria to determine which evidence ismore important, empirical or logical? Generally, thelogical evidence is considered superior to the empiri-cal evidence since there is less chance of error. In thephilosophy of science the phrase used is the ‘power
of deduction rules’. The ‘power of deduction rules’provides the underlying structure for theory-building
Ž .research Popper, 1957 . A brief example may proveuseful to clarify why the ‘power of deduction rules’.Suppose a researcher gathered a large set of data ona bank and statistically found that when the arrivalrate is greater than the service rate, the line isshorter. To most operations management academics,there is no amount of empirical evidence that couldconvince them that having greater arrival rates thanservice rates will lead to shorter lines. This conclu-sion is contrary to queuing theory’s deductive con-clusion. In short, the ‘‘power of deduction rules’’.
Yet, most operations management academicswould investigate the specific circumstances to de-termine which queuing model should be applied. Oralternatively, academics would develop a new theoryto explain the empirical evidence. Therefore, basedupon the empirical evidence, operations managementacademics may develop a ‘new’ internally consistent
Ž . Ž .theory model to replace the existing theory model .The last virtue is the theory’s abstraction level,
which is usually classified into three levels: high,middle, and low. High abstraction level theoriesŽ .general or grand theories have an almost unlimitedscope, middle abstraction level theories explain lim-
Žited sets of phenomena which serve as the rawmaterials for the construction of more general theo-
. Žries , and lower level theories called empirical gen-.eralizations of very limited scope serve as simple
Žrelationship identifications Bluedorn and Evered,.1960 . These three levels of abstraction are points on
an abstraction continuum. A low abstraction leveltheory can be directly applied to a specific instancesince a low abstraction level theory completely de-
Ž .fines the variables, domain, relationship s , and thenstates the factual claim for this specific instance.However, the usefulness of low abstraction leveltheories to academics and practitioners is limitedbecause in the majority of instances, the theorycannot be applied since the definitions and domainare so narrow that the theory only applies to a fewinstances. Yet, low abstraction level theories areused to build middle abstraction level theories which,in turn, lead to high abstraction level theories. At thehighest abstraction level, the application of a theoryis the broadest since the domains are broad. Highlevel abstraction theory, therefore, has wide applica-
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 367
tion. Consequently, an important goal of theory-building is to advance lower abstraction level theo-ries to middle level theories, and then to high ab-straction level theories. This leads to the generaliza-tion that as a field progresses, it tends to have theoryin higher abstraction levels. This progression fromlow abstraction to high abstraction is called ladder-
coming evident by several of the empirical general-Ž .izations in just-in-time JIT theories. For example,
the following are three low-level abstraction theo-Ž .ries: 1 shorter set-up times facilitate smaller lotŽ .sizes, 2 smaller lot sizes cause lower work-in-pro-
Ž .cess, and 3 lower work-in-process causes increasedquality. These three different smaller theories arebeing integrated into a large theory through a morecareful explanation as to how they are related to eachother. When addressed individually, each of thesetheories is a low abstraction level theory. Yet, whencombined into single integrated theory, it becomes a
Žmid-range theory Miller and Roth, 1995; Demeyer.and Arnoud, 1989; Wacker, 1987, 1996 . Still, the
integration into high abstraction level theories seemsto be somewhat in the future for operations manage-ment.
Though all of these theory virtues are highlysignificant for theory-building, the relative weightgiven to each virtue becomes important for compar-ing competing theories, since virtues trade off witheach other. For example, a new theory may explain
Ž .phenomena in many domains generalizability , butŽbe relatively complex hampering the parsimony cri-
.teria . Or a theory may be rich in hypotheses devel-Ž .opment fecundity , but can only be applied to a very
Žlimited set of conditions hindering the generalizabil-.ity criteria . Or theories that are generalizable, may
Žnot be internally consistent hindering the internal.consistency virtue . To some degree, each ‘good’
theory weighs each virtue against other virtues. Thisevaluation requires value judgments to decide whichtheory is superior to other theories based on eachtheory’s virtues. Although there are always trade-offsamong virtues, the internal consistency and empiricalriskiness virtues are generally considered the mostimportant virtues since without refutation being pos-sible or likely, the theory becomes tautologicalŽ .Popper, 1957 .
4. When does a theory become a ‘good’ theory?
One of the most difficult questions to answerabout theory is: when does a theory become a ‘good’theory? The answer to that question is inherentlycontroversial since it involves the degree to whichindividuals believe in adhering to the formal defini-tion of theory and follow the virtues of ‘good’theory. For the purpose of this article, a ‘good’theory should meet the definitional criteria of theoryas well as follow the virtues of ‘good’ theory. There-
Žfore, ‘good’ theory must first be a theory i.e., havedefinitions, have a domain, have relationships, and
.make predictions and must meet each ‘good’ theoryvirtue to some degree.
Any theory which adheres to both the definitionof theory and the virtues of a ‘good’ theory is a‘good’ theory. This adherence, however, does notmean that the ‘good’ theory is valid since ‘good’theories can be ‘just plain wrong’. Yet, ‘good’ theo-ries which are wrong are more quickly identified as
Žbeing wrong since they are more easily refuted in-.ternally inconsistent or empirically invalid . Still,
‘good’ incorrect theories serve a very important pur-pose for field development, since ‘good’ theory is abeginning point to determine why the theory iswrong. Therefore, because ‘good’ theory is easilyrefuted and is a beginning point for future investiga-tion, using ‘good’ theory for empirical tests seems tobe a laudable objective for the development of anacademic field.
On the other hand, a theory that violates one ofthe virtues of ‘good’ theory is difficult to refute.Since the virtues of theory apply to the definitionalcriteria of theory, there are innumerable ways a‘bad’ theory can be developed. Consequently, it ispossible to give myriad examples of how a ‘bad’theory hinders the development of an academic fieldby detracting research efforts away from ‘good’ the-ories.
5. The general procedure for ‘good’ theory-build-ing research
The purpose of this section is to demonstrate ageneral research procedure that satisfies the follow-ing specifications: theory’s definition, ‘good’ theory’s
( )J.G. WackerrJournal of Operations Management 16 1998 361–385368
virtues, and answers to the natural language ques-tions of who, what, when, where, why, how, should,could, and would. From the logical perspective, theseguidelines are all necessary conditions for theory-
Ž .building Whetten, 1989; Van de Ven, 1989 . Theprocedure must state the purpose of each step.
The academic literature suggests many differentŽprocedures for theory-building Eisenhardt, 1989;
.Swamidass, 1986; Bacharach, 1989 . Generallyspeaking, these procedures suggest how to opera-tionalize specific types of research projects. Theresearch procedures suggested here are similar tothose operationalizations of research. However, thisstudy identifies a minimal procedure for ensuringthat all guidelines for ‘good’ theory-building are metregardless of the type of research. Each step reflectsa necessary condition to fulfill ‘good’ theory-build-ing guidelines, since all theory-building requires def-initions, domain, relationships, and predictions. Con-sequently, the same procedure is required regardlessof the methodology used for any theory-buildingresearch. Therefore, from the viewpoint of theory’svirtues, this study suggests that the simplest, mostgeneralizable procedure be adopted. The requirementthat theory-building research follows the definitionof theory leads to similarities in theory-building re-search for both analytical and statistical procedures.
More about these two basic research methods arediscussed below.
Although the stages suggested here are listedsequentially, they are not necessarily sequential sincethey interact with each other. For example, when atheory is visualized, a new definition may be neededto identify a new concept. Or when a prediction orfactual claim is made, it may require a differentdomain or a different relationship. In addition, theoperationalization of variables for empirical testsmay require some modifications of selected defini-tions. Consequently, although this procedure is listedin stages, a research project may not necessarilyproceed inexorably sequentially through each stageto a ‘good’ theory.
For all stages of theory-building, the role of theliterature search in the research procedure is ex-tremely important since it provides the accepteddefinitions, domains of where a theory applies, pre-
Žviously identified relationships along with empirical.tests , and specific predictions of other theories.
Therefore, to assure that all theory-building condi-tions are fulfilled, an extensive literature search ofthe academic as well as the practitioner articles isrequired.
Table 2 outlines the general procedures for devel-oping theory. The precise definitions of variables are
Table 2A general procedure for theory-building and the empirical support for theory
Purpose of this step Common question ‘Good’ theory virtuesemphasized
Definitions of variables Defines who and what are included and Who? What? Uniqueness, conserva-what is specifically excluded in the tiondefinition.
Limiting the domain Observes and limits the conditions by When? Where? GeneralizabilityŽ .when antecedent event and where the
subsequent event are expected to occur.Ž .Relationship model Logically assembles the reasoning for Why? How? Parsimony, fecundity,
building each relationship for internal consis- internal consistency,tency. abstractness
Theory predictions and Gives specific predictions. Important for Could the event occur? Should the event Empirical testsempirical support setting conditions where a theory pre- occur? Would the event occur? refutability
dicts. Tests model by criteria to giveempirical verification for the theory.The riskiness of the test is an importantconsideration.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 369
needed to limit the area of investigation by definingwho and what. Generally, the literature provides abase for defining the variables. New definitions mustdemonstrate why current definitions are not ade-quate, otherwise the conservation virtue is violated.Ordinarily, currently defined concepts should be usedto avoid violating the conservation virtue. Althoughthere are many difficulties in precisely defining andmeasuring variables, and the complete discussion ofthe implications are beyond the scope this paper,there are several concerns which bear mentioning.One important difficulty with definitions is what hascome to be known as ‘concept stretching’ whichoccurs:
Ž .. . . W hen concepts are broadened in order toextend their range of application, they may be so
Ž .broadly defined or stretched that they verge onbeing too all-embracing to be meaningful in therealm of empirical observation and professional prac-tice. . . . Because of this broadness, it tends toconfuse more than help the development of the field:. . . Because concept stretching in organizational sci-ence results in amorphous unclear conceptualiza-tions, what appears to have been a gain in exten-
Ž .sional coverage breadth often has been matchedand even surpassed by losses in connotative preci-
Ž . Ž .sion depth Osigweh, 1989 .
One common source of ‘concept stretching’ isnatural language which is too broad for precisemeasurement, causing a research project to build anill-defined theory, which in turn, gives misinforma-
Ž .tion Osigweh, 1989 . Because of the difficultieswith natural language, artificial languages and defini-
Ž .tions sometimes called formal language must bedeveloped to avoid the confusions caused by naturallanguage. Some feel that theory-building researchcannot be effectively achieved unless the academic
Žfield has a precise artificial language Teas and.Palan, 1997 . To develop an artificial language and
avoid ‘concept stretching’, some theorists suggestthat the definitions be examined by the negationprinciple. That is, definitions must be examined bywhat they specifically exclude. Although it is nottrue in the extreme, generally, the more a conceptualdefinition excludes, the more precise it is and the
Ž .more likely it will be unique Osigweh, 1989 .
Once the precise definitions of variables are estab-lished, the domain is established to limit when andwhere the theory hold. The domain of the theorydirectly limits the its generalizability since the morespecific the domain of the theory is, the lower thegeneralizability. Theory-building research extendsdomains to new broader areas by testing the theoryin a new environment or a different time period. Inshort, theory-building research extends the domain ofthe theory.
Typically, after both the definitions are clarifiedŽ .and the domain is specified, relationship model
building begins. This step is necessary to establishwhich variables have logical connections to othervariables. Technically, every variable used shouldexplicitly state how and why it is related or unrelatedto each and every variable in the model. The summa-tion of the related plus the unrelated variables shouldbe equal to the total number of variables in themodel. Hence, in any given theory, the relationshipbetween any two variables in the theory must beexplicitly stated, or else the theory cannot be shownto be internally consistent. The third stage is quite
Žcomplex since the theory and the model by which.the theory is to be tested should be fully developed
and internally consistent. For this step, the academicliterature suggests which relationships are importantfor the development of the theory. The more care-fully a researcher builds the relationships from otherresearch, the more theoretically important the re-search is, since it is integrating theory to raise itsabstraction level.
In the relationship-building step, there are fourtypes of theory-building relationships: those relation-
Žships that are assumed to be true fundamental laws.or axioms ; those laws that are derived from the
Ž .fundamental laws derivative laws or theorems ; thoselaws that span the gap from the theoretical to the
Ž .empirical world bridge laws or guiding hypotheses ,Žand the relationships that are being investigated re-
. Ž .search or theoretical hypotheses Hunt, 1991 . Forexample, research on cost and small lot sizes analyti-cally assumes that the number of orders that areplaced in a year is equal to demand divided by the
Ž .fixed order size a fundamental law . Additionally,the average inventory is equal to maximum plus
Žminimum inventory divided by two a second funda-.mental law . The total cost of inventory is equal to
( )J.G. WackerrJournal of Operations Management 16 1998 361–385370
the order cost times the number of orders plus theŽholding cost times the average inventory a derived
.law or theorem . The cost optimization solution leadsŽ . Žto the economic order quantity EQQ a second
.derived law . A bridge law is necessary to tie theanalytical results to the empirical world. For exam-ple, given the annual demand, annual holding cost,and ordering cost, firms should have a predicted
Žorder quantity the research or theoretical hypothe-.sis . One bridge law that would facilitate this empiri-
cal test is: rational firms act to minimize their totalinventory costs. Note that without a bridge law, theEQQ derived law does not span into empirical world.The area of bridge laws is important for integrationof theory in operations management. It is the lack of
Ž .bridge laws that Churchman 1961 referred to thebeginning of the study. Consequently, Churchman’scriticism cannot be summarily dismissed, since manyanalytical research studies do not give any guidanceas to how they are to be tested in the externalempirical world.
In the above example, fundamental and derivedlaws are not tested, but are assumed to be trueanalytically. If these relationships were tested alongwith other relationships, the statistical complexitymay be too complex for meaningful investigation.Because of this complexity, it is important that re-searchers limit their investigation using the literature,since the parsimony virtue may be violated. Theliterature provides the best guidelines as to whichrelationships are theoretically important for investi-gation and which relationships may be consideredfundamental or derived laws that do not need consid-eration in the investigation. Consequently, theory-building research uses the literature as a guideline todecide which relationships are important for investi-gation.
The last stage is theory prediction. In order for atheory to meet the minimal requirements for thedefinition of theory, it is technically not necessarythat a theory offer evidence to support predictive
Ž .claims only predictions are necessary . Yet, from apractical perspective, it is possible that a huge num-ber of theories could be proposed. Therefore, manyacademics prefer that empirical evidence be pre-sented to verify that a proposed theory has somemerit in the empirical world. Different methodolo-gies use different empirical evidence to verify their
predictive validity. More will be discussed about thispredictive stage below when discussing proceduresfor analytical and empirical research.
Though each stage of theory-building needs tofollow all the virtues of ‘good’ theory, each stage ismore closely related to certain specific virtues. Forexample, in the definition step, uniqueness and con-servation are key virtues since they limit theory-building to developing current concepts before de-veloping new unique concepts. Some researchersargue that many definitions such as total qualitymanagement, continuous improvement, and just-in-
Žtime overlap and are neither unique, nor new West-.brook, 1987; Robinson, 1990 . They argue that these
concepts do not limit the definition and are notnecessarily unique from each other which makesthem malformed thus causing data-gathering difficul-
Ž .ties as well as misinformation Osigweh, 1989 . Inthe domain specification stage, the generalizabilityvirtue is important because the more domains inwhich a theory can be applied, the more importantthe theory is. In the relationship-building step, parsi-mony, fecundity, and abstraction virtues enhance thetheory by using only necessary relationships, offer-ing new areas for investigation, and integrating rela-tionships for a higher abstraction level. Also in thisstage, internal consistency is important to verifywhich relationships are logically compatible witheach other. Generally, as more internally consistentrelationships are integrated into a theory, the theorycan explain more, therefore raising the theory’s ab-straction level. In the theory prediction step, theimportance of internal consistency and empiricalriskiness are both needed for the theory to makepredictions.
Table 2 summarizes the procedure for ‘good’theory-building research. Each of the stages are re-quired for a theory to be considered a ‘good’ theory.Unless a proposed theory has all these stages, it doesnot meet the criteria of the formal definition oftheory. The first column shows the components oftheory. The second column states why these compo-nents are necessary. The third column gives thecommon question that each stage addresses. Whilethe first three columns use the definitional criteriafor theory, the last column gives the most relevantvirtues for that stage to ensure that any theory devel-oped is a ‘good’ theory.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 371
‘Good’ theory-building research requires the ful-fillment of the formal definition of theory. Addition-ally, a ‘good’ theory should not violate any ‘good’theory virtue. The stages of a ‘good’ theory-buildingmirrors the four criteria for theory and meet thevirtues of ‘good’ theory. However, since it is rela-tively easy to propose many ‘good’ theories thatmeet the theory criteria, usually some empirical evi-dence is needed for ‘good’ theory to be ‘good’theory-building. Therefore, empirical evidence usu-ally is included in the prediction stage to support the
Žtheory it could be argued that this is an additional.stage of theory-building . Consequently, ‘good’ the-
ory-building should have empirical evidence to sug-gest why the theory has some legitimacy in theempirical world.
ŽAs a side note, the statements by Shubik that.theory can be totally impractical cannot be summar-
ily dismissed since a ‘good’ theory does not have tobe tested to be ‘good’ theory. In short, it is possiblethat a new ‘good’ theory may not yet have externalempirical evidence to support it. For this reason,many academics believe that empirical evidenceshould be offered before any theory is considered a
Ž‘good’ theory Whetten, 1989; Poole and Van De.Ven, 1989 .
6. The two objectives of research: some importantobservations
The two general objectives of research aretheory-building and fact-finding. The difference be-tween these two objectives is grounded in the pur-pose of the research. ‘Good’ theory-building re-search’s purpose is to build an integrated body ofknowledge to be applied to many instances by ex-plaining who, what, when, where, how and whycertain phenomena will occur. On the other hand,‘good’ fact-finding research’s purpose is to build alexicon of facts that are gathered under specifiedconditions. The division line between the two typesof research is quite fine since both types of researchusually include data gathering and empirical estima-tion. However, the contrast between the two pur-poses comes from two areas: how and when the apriori predictions are made; and how the results areintegrated with other studies.
First, theory-building research carefully definesconcepts, states the domain, explains how and whyrelationships exist, and then predicts the occurrenceof specific phenomena. After the prediction, it typi-cally gathers evidence to see if the phenomena oc-curs. ‘Good’ fact-finding research also carefully de-fines concepts and states domains. Then, ‘good’fact-finding research uses evidence to discover ifrelationships exist. Next, ‘good’ fact-finding researchthen explains how and why specific phenomena oc-curred. In the formal sense of the definition oftheory, fact-finding research is not theory-buildingresearch since the evidence is gathered before therelationships are explained and before the relation-ships are predicted. Therefore, fact-finding researchdoes not meet the conditions of ‘good’ theory-build-ing, since it lacks a priori explanations and predic-tions before the data are gathered.
Even though fact-finding research is not theory-building research, it serves an important purpose byproviding facts that can later be integrated into atheory. ‘Good’ fact-finding research serves to pro-vide fertile ground for subsequent new theory-build-ing. Fact-finding research is not constrained by exist-ing theory-based relationships and, therefore, canmore easily investigate new relationships that canlater be integrated into a theory. However, to be ofvalue for ‘good’ theory-building, ‘good’ fact-findingresearch must follow the first two conditions fortheory, since without precise definitions and well-de-fined domains, facts are not meaningful for anyfuture use. Because fact-finding research investigatesrelationships and then offers explanations as to whythe results occurred, fact-finding research suggestsnew relationships for new theory development. Dueto the freedom from the constraints placed ontheory-building research, fact-finding research fre-quently provides the basis for new theory develop-ment. Consequently, there is a large measure of truth
Ž . Žin the statement of Lindblom 1987 statement made.at the beginning of this study that many new theo-
ries come from fact-finding investigation.A second contrast between theory-building and
fact-finding research is the degree of integration ofthe results with other studies. This degree of integra-tion has two dimensions: the internal integration ofall the empirical data and the integration of theresults with other studies. Empirical evidence used in
( )J.G. WackerrJournal of Operations Management 16 1998 361–385372
theory-building research stresses subtle systematicsimilarities between all data to raise the theory’sabstraction level. These systematic similarities pro-vide additional factors to explain all data in oneintegrated framework applied to diverse environ-ments. By using one theory across many differentenvironments, theory-building research raises the ab-straction level to explain how and why the theorycan be applied to predict events. When the empiricaldata do not support the theory, the reason for thelack of support is important for the further develop-ment of the theory. If the theory is logically inter-nally consistent, then the reason for the lack of
Ž .support must be a missing explanatory factor s . ThisŽ .explanatory factor s is introduced into the theory
and then the theory is examined for internal consis-tency. Once the ‘new’ theory is internally consistent,then the theory can be empirically tested. This se-quence of empirical tests contradicting the theory,
Ž .introducing new factor s , building a ‘new’ internallyconsistent theory, and empirical testing of the ‘new’theory is a common theory-building sequence fortheory building. This sequence has the goal of build-ing a single integrated theory to explain all data. Inshort, when empirical tests do not support a theory,theory-building research examines the theory to inte-grate new factors to raise the abstraction level. Con-sequently, ‘good’ theory-building research is alwaysstriving to find integrating factors to expand thetheory to apply to diverse environments, therebyraising the theory’s abstraction level.
In contrast, fact-finding research typically stressesthe descriptive differences in data. These descriptivedifferences exist because all data, to some degree,are different. Fact-finding research attempts to dis-cover differences in data and explains these differ-ences. Unfortunately, without a theory to be tested inthe empirical world, many times the explanations ofdescriptive differences are not specifically tested.Even more unfortunate, is that they may not beexplanations at all. Consider a case where a re-searcher finds that Japanese manufacturers whencompared to USA manufacturers, have statisticallysignificantly higher manufacturing performance onsome performance measures. The basic fact indicatesthat Japanese manufacturing managers are superiorto USA manufacturing managers on these perfor-mance measures. However, this fact does not offer
Ž .adequate or any explanation of how or why thisresult occurred since the source of the difference
Žcould be better planning forecasting, and resource. Žscheduling ; control forecast control, production ac-. Žtivity control ; advanced technology better informa-
. Žtion technology, CAD, FMS, etc. ; culture work.ethic, managerial behavior, etc. or numerous other
causes; all of which were not specifically tested. Inshort, fact-finding research frequently explains de-scriptive differences, but because the explanation isnot specifically tested, any inferences andror con-clusions are deceptive. For this reason, these expla-nations are deceptive descriptive differences.
One type of integration used in theory-buildingresearch is the identification of subtle systematicsimilarities across different studies. For example,
Ž .Ettlie 1995 noted the empirical similarities betweenseveral new product development research studiesŽMadique and Hayes, 1984; Crawford, 1987; Kekreand Srinivasan, 1990; Klemschmidt and Cooper,
.1991 . The similarities he deduced were between thetwo extremes on product development. At one ex-treme, firms which introduce variants of existingproducts reap significant benefits from increasedmarket share and profits. At the other extreme, henoticed that firms which introduce totally new prod-ucts are also reap significant benefits from increasedmarket share and profits. Yet, firms that attemptedan intermediate strategy were not successful. Hence,
Ž .he concluded, ‘‘ T he relationship between productinnovativeness and success in the market place isu-shaped.’’ His integration of the three studies intoone theory is accomplished by identifying the subtlesimilarities between different studies. This integra-tion across several studies raises the abstraction level,therefore increasing the importance of his study’sfindings.
In short, researchers are faced with whether tofind subtle systematic similarities or to explain de-ceptive descriptive differences between individuals,organizations, businesses, industries, and countries.Fact-finding research focuses on descriptive differ-ences among data, while theory-building researchconcentrates on the underlying factors for similari-ties. Theory-building research raises the abstractionlevel by integrating subtle systematic similaritiesacross the descriptive dimensions of individuals, or-ganizations, businesses, industries, and countries.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 373
Consequently, from the standpoint of ‘good’ theory-building, it seems that systematic similarities aremore important than descriptive differences.
7. Types of theory-building research
The purpose of this section is to classify thetheory-building types of operations management re-search by their theory-building purpose. There are
Žtwo major classifications of research: analytical for-.mal and empirical. These differences are best ex-
Ž .pressed by Sax 1968 :
When an organized body is based primarily on de-ductive rules, it is called formal science, to distin-guish it from those areas of knowledge that dependprimarily upon empiricism and induction . . . calledempirical science. Considered from this point ofview, mathematics, logic, and library science areprimarily formal sciences, whereas chemistry, psy-chology, and education are primarily empirical sci-ences.
Ž .This statement means that analytical formal sci-ences use deductive methods to arrive at theorieswhile empirical sciences use induction methods toarrive at theories.
This section uses these traditional major classifi-cations of analytical and empirical theory-buildingresearch and further divides them into three sub-cate-gories for each major classification. The categoriza-tion of the research types serves two purposes: itillustrates the need for different research methodolo-gies to raise abstraction levels of theory and, itillustrates that the ‘good’ theory-building procedures
Žare applicable to different types of research dis-.cussed in the Section 8 . The first purpose for the
classification is based on the goal of the ‘good’theory-building research. This study argues that thereare six different types of theory-building research,and each type serves to develop and verify theoryusing different research methodology. This theoryverification usually is called triangulation and isextremely important for the final verification of the-
Ž .ory Meredith et al., 1989 . Ideally, if a theory istested by all six methods with supporting results,then these results would be compelling evidence tobelieve that the theory is confirmed.
7.1. Analytical research
The analytical research method uses deductiveŽ .methods to arrive at conclusions Swamidass, 1986 .
Analytical research methods primarily use logical,mathematical, andror mathematical–statistical meth-ods. The three different sub-categories of analyticalresearch have subtle differences in how they uselogic and mathematics for the development of thetheory. Additionally, the analytical sub-categorieshave different theory-building purposes.
7.1.1. Analytical conceptual researchFrom a theory-building perspective, the purpose
of analytical conceptual research is to add new in-sights into traditional problems through logical rela-tionship-building. This research methodology com-prises new insights through logically developing re-lationships between carefully defined concepts intoan internally consistent theory. These studies usuallyuse case study examples to illustrate these conceptu-alizations. There are several types of research in thissub-category. One example is called introspectiveresearch which uses the researcher’s experience toformulate concepts. Consequently, it describes andexplains relationships from past experience to de-velop theory. A second example in this sub-categoryis conceptual modeling. This modeling is where amental model of deduced relationships is posited,which may then be evaluated using a framework thatcaptures the essence of the systems under investiga-tion. A third example in this sub-category ishermeneutics research which deduces facts from what
Ž .is being observed Meredith et al., 1989 .A representative article of analytical conceptual
Ž .research is the Kim and Lee 1993 article where theresearchers logically tied together manufacturingstrategies and production systems. In their study,they logically integrated technical flexibility, techni-cal complexity, and production system type withdifferentiation and cost efficiency strategies. Thistype of research falls into the analytical conceptualsub-category since the methodology used is logicaland does not use empirical data for theory develop-ment.
7.1.2. Analytical mathematical researchThe theory-building purpose of this research sub-
category is to develop sophisticated relationships
( )J.G. WackerrJournal of Operations Management 16 1998 361–385374
between narrowly defined concepts through develop-ing new mathematical relationships to study how themodels behave under different conditions. These arti-cles mathematically develop the relationships andgive numerical examples from their derivations orcomputations. Analytical mathematical research doesnot use external data to test the theory, but insteaduses deterministic or simulated data to draw conclu-sions. The research in this sub-category is sometimescalled ‘operations research’ andror ‘managementscience’.
There are many types of research in this sub-cate-gory: reasonrlogical deductive theorem proving;normative analytical modeling research; descriptiveanalytical modeling; proto-typing and physical mod-eling research methods; experimentation; and mathe-matical simulation. In all these methods, the modelsusually are built using formal logic and tested usingartificially developed data. Since the data is derivedartificially, all these methods are classified under the
Žanalytical mathematical research method Meredith.et al., 1989 .
7.1.3. Analytical statistical researchThis research sub-category integrates
logicalrmathematical models from analytical re-search and statistical models from empirical research
Ž .into a single integrated theory Moorthy, 1993 . Theanalytical statistical research is different from theanalytical mathematical method since its models areexplicitly developed for future empirical statisticaltests. This methodology uses large bodies of knowl-edge and integrates them into a single model forfuture empirical tests. Typically, the variables usedfor investigating relationships have measurement er-rors due to random variability caused by the externalworld. In sum, the purpose of analytical statisticalresearch is to provide larger, more integrated modelsfor empirical statistical testing.
7.2. Empirical research
In the empirical research major classification, themethodology must use data from external organiza-tions or businesses to test if relationships hold in theexternal world. Empirical research methods could beclassified more correctly as ‘real world’ empiricalmethodologies. However, since this is a fairly longphrase, it is abbreviated to empirical methods.
7.2.1. Empirical experimental researchThe theory-building purpose of this sub-category
is to investigate relationships by manipulating con-trolled treatments to determine the exact effect onspecific dependent variables. The research designuses treatment variables that are manipulated to de-
Ž .termine their effect on the dependent variable s .Because direct manipulation of the treatment vari-ables causes direct effects on the dependent vari-ables, this research sub-category comes the closest todemonstrating causality between variables. Thissub-category of empirical research is also called
Ž .‘field experiments’ Meredith et al., 1989 .The empirical experimental research methodology
is difficult to implement in operations managementsince the environment must be closed to contamina-tion effects. In operations management, the systemfrequently is an open system and therefore can besubject to contamination effects. Yet, it seems possi-ble that controls could be placed on some experi-ments to determine if one treatment causes a certainresult. For example, a researcher may find an organi-
Ž .zation that is about to install a just-in-time JITsystem. If the facility has two different lines ormanufacturing areas that are separate, the researchermay experiment with the amount and type of trainingto determine the effect of training on selected perfor-
Ž .mance measures throughput time, defect rates, etc.This type of experiment could offer convincing evi-dence of the necessity or non-necessity for specificamounts or types of training needed for JIT imple-mentation.
This research type is not to be confused with theexperimental design in mathematical simulationfound in the analytical mathematical sub-category
Žabove where the data are developed by the re-.searcher in a closed simulated environment .
7.2.2. Empirical statistical researchThis research sub-category’s purpose is to empiri-
cally verify theoretical relationships in larger sam-ples from actual businesses. Generally, the morecomplex the research issues are, the more likely thestudy will use this methodology. A typical topicwould be: manufacturing strategy’s effect on manu-facturing performance.
There are many types of research that fall into thisŽsub-category. Structured and unstructured elite and
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 375
.intensive interviewing processes which gather datafor statistical analyses are examples of this sub-cate-gory of research. Other representative examples aresurveying and historicalrarchival research, expertpanels, and Delphi techniques. Each of these meth-ods has the goal to statistically analyze data fromrelatively large external samples. Therefore, from atheory-building perspective, this methodology offersempirical support for theoretical relationships in
Ž .larger samples in real world Meredith et al., 1989 .
7.2.3. Empirical case studyThe purpose of this type of research is to develop
insightful relationships within a limited set of com-panies. By limiting the number of companies investi-gated, this research method investigates small sam-ples using a large number of variables to identifynew empirical relationships. Frequently, this researchmethod analyzes organizations across time and pro-vides the dynamic dimension to theory to elevate thetheory’s abstraction level. This sub-category includes
Žfield studies and action research Meredith et al.,.1989 .
A typical research article of this sub-category isŽ .the article of Marucheck et al. 1990 on manufactur-
ing strategic practice, which examines the strategicpractices in six firms. Using the practices of the sixfirms, they found that firms follow the general con-ceptual models developed in the academic literature.
Ž .Additionally, they stated ‘‘.. T he real benefits ofmanufacturing strategy come from implementing the
Ž .manufacturing strategy’’ p. 121 . The research pro-Ž .ject of Marucheck et al. 1990 analyzed data from a
limited number of firms to identify manufacturingstrategy procedures. Since a limited number of firmsare utilized to identify possible theories, this projectis classified as empirical case study. Hence, the keydifference between the empirical case study methodand the analytical conceptual method is that theempirical case study method uses data to form thetheory and the analytical conceptual method usesdeduction to form theories.
7.3. An important conclusion
From the standpoint of good theory, one impor-tant conclusion drawn is that no single researchcategory or sub-category is superior to any other
research category or sub-category. Each methodserves a very different, but important, purpose fortheory development in operations management. Thebasic purpose of all three types of analytical method-ologies is to develop logically internally consistenttheories and models. First, the analytical conceptualresearch type serves as a forum for expressing newconceptual perspectives on theory to better explainand integrate underlying relationships. The sub-cate-gory of analytical mathematical research serves tologically evaluate the internal consistency of com-plex relationships. The analytical statistical researchserves to integrate both the analytical mathematicalresults and the empirical statistical results into alarger theoretical body of knowledge for statisticalestimation. In short, all these analytical researchsub-categories serve to develop internally consistenttheories through logical analyses.
The empirical methodologies provide empiricalverification of models, while offering evidence forthe development of new theory. The empirical exper-imental research uses experimental design to verifythe causality of a specific theory while elevatingrelationships from a testable hypothesis to an empiri-cally verified theory. A verified theory may not needto be tested in future research projects and may beassumed to be a fundamental law. The empiricalstatistical research methodologies verify models fortheir empirical validity in larger populations to re-duce the number of relationships in future research.The empirical case studies provide new conceptualinsights by empirically investigating individual casesfor an in-depth understanding of the complex exter-nal world.
Table 3 provides a summary of the classificationof research along with its relative importance tooperations management. Basically, theories devel-oped using the analytical research methodologies arerefuted by empirical evidence, while theories devel-oped using the empirical research methodologies arerefuted by internal consistency. The last row of thetable states the theory-building purpose of the re-search method. It would be difficult to disagree withthe importance of any of these six sub-categories,since it would be paramount to stating that opera-tions management does not need: new conceptualtheory; internal mathematical consistent theory; inte-grative theory; causal verification of theory; large
()
J.G.W
ackerrJournalof
Operations
Managem
ent161998
361–
385376
Table 3Specific research sub-category, refutation methods, and importance to operations management theory-building
Analytical Empirical
Conceptual Mathematical Statistical Experimental design Statistical sampling Case studies
Types of research in- Futures research sce- Reasonrlogical theo- Mathematical statisti- Empirical experimen- Action research struc- Field studies, casecluded narios, introspective rem proving, norma- cal modeling tal design, descriptive tured and unstruc- studies
Refutation methods Empirical data from Empirical data from Empirical data from Analyticalrlogical in- Analyticalrlogical in- Analyticalrlogical in-empirical methods empirical methods empirical methods consistency consistency consistency
Importance to opera- Develops new logical Explores the mathe- Integrates the other Tests and verifies Tests the theory by Tests and developstions management relationships for con- matical conditions un- five methods into a causal relationships investigating statisti- complex relationshipstheory-building ceptual models of the- derlying the relation- single theory for em- between variables cal relationships to between variables to
ory ships used in theory- pirical investigation verify their existence suggest new theorybuilding in larger populations
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 377
population verification of theory; and new empiricalrelationship exploration for theory development.Consequently, all these methods are extremely im-portant for the complete development of theory-building in operations management.
8. The need for similarity in theory-building re-search procedures
This section suggests that general research proce-dures should apply to all research methodologies iftheory-building is to be important for operationsmanagement. If all research methodologies followcommon definitions, common domains, common re-lationships, and common predictions, the integrationacross the six different sub-categories of theory-building methodologies would be more likely.
Table 4 provides an outline of the procedures forthe different theory-building methodologies. All sixmethodologies have the same four stages in theprocedure, since these stages map directly to thedefinition of theory. In the first stage, it is importantthat the conceptual definitions used in the analyticalmethodologies are the same definitions that are oper-ationalized in the empirical methods. It does notseem logical for analytical methods to use concep-tual definitions for their mathematical convenience ifthese definitions have no hope of ever being opera-tionalized for measurement in empirical studies.Therefore, analytical researchers should give mea-surement guidelines to empirical researchers. Con-versely, it is important for empirical methods not toredefine an existing conceptual definition for thesake of measurability. From the theory-building per-spective, ideally, all six methodologies would use thesame conceptual definitions for the same concept sothat analytical and empirical evidence are buildingon the same theory. Therefore, for theory-building,the first step in theory-building research should beidentical regardless of the sub-category of researchpursued.
In the second stage of theory-building research,the domain of the theory is defined. In this stage, allthree types of analytical methodologies should spec-ify where their theories apply. Generally, thesemethodologies have broader domains than empiricalmethodologies. Yet a common challenge to the ana-
lytical methodologies is a cautious domain specifica-tion to identify exactly where the theory is to beapplied. Without this specification, it is assumed thatall domains are included, which causes these theories
Ž .to be questioned or refuted by a single empiricalobservation from any domain. For empirical method-ologies, the domain usually is carefully specified by
Žor by the specific case studied empirical case study.methodology . From the perspective of theory-build-
ing, ideally, the domains suggested by the analyticalmethods would be investigated by the empiricalmethods to verify relationships to raise the abstrac-tion level.
In the third stage of theory-building research pro-cedures, relationships are suggested for theory-build-ing. In all three of the analytical methodologies,logic is used to determine the relationships. Theanalytical methodologies use fundamental laws todeduce derived laws to suggest which relationships
Ž .are logically compatible internally consistent witheach other. Empirical methodologies should test ana-lytically developed relationships before offering newrelationships since new relationships may not beinternally consistent. Empirical methodologies shouldexhibit great care before offering new relationshipssince some relationships may have already beenidentified using analytical methods. If an empiricalstudy fails to examine the analytically derived rela-tionships and the study results are contrary to theanalytical relationships, the empirical results are con-sidered artifacts, since the ‘power of deduction rules’Ž .Hunt, 1991 .
The last stage of theory-building research is theprediction and verification. ‘Good’ theory-buildingsuggests that operations management theory shouldoffer internally consistent predictions in the ‘realworld’ empirical realm. This suggestion means thatanalytical methodologies should use bridge laws aspart of their methodology to provide the means totest their theories’ empirical predictions in the ‘real’world. This suggestion is not to be interpreted asmathematical models are not needed in operationsmanagement, but rather, these models should offersome hope of improving operations management asit is practiced. To develop practical models, it seemsthat offering ‘how’ these results will be tested in the
()
J.G.W
ackerrJournalof
Operations
Managem
ent161998
361–
385378
Table 4The theory-building procedure for different theory-building types of research
Definitions develop- Conceptual defini- Conceptual defini - Conceptual defini - Conceptual defini - Conceptual defini - Conceptual Defini -ment tions. However, many tions from the litera- tions from the litera- tions. However, fre- tions from the litera- tions from the litera-
times, new definitions ture. ture. quently the method ture. However, many ture. However, manyare offered. may require new, times, new constructs times, new relation-
more measurable con- are developed to rep- ships require new def-cepts. resent the theoretical initions.
concept.Domain limitations Logically developed Mathematically de - Mathematically and Experimental design From analytical statis- Developed from cases
veloped statistically developed with a narrow con- tical models or devel- studied.trolled domain. oped experimentally
Ž .Relationship model Usually, relationships Usually, relationships Usually, other studies Usually developed Usually using other The combining of thebuilding are logically devel- are mathematically are used to develop with limited relation- statistical studies’ relationships discov-
oped. developed without mathematical statisti- ships between vari- suggested theories. ered from the case.stochastic error terms. cal models with error ables.
terms.Theory predictions: Usually, predictions Mathematically de - Mathematically and Prediction from the Prediction from other Supported by caseevidence of predic- come from logical duced predictions. logically derived pre- experimental design. studies. Results from studiestion analyses. Empirical Examples from math- dictions. Uses empiri- Statistical signifi- sample’s statistical
evidence comes from ematical calculations cal evidence from cance of the tests tests for significance.case studies. or simulated results other studies.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 379
empirical realm is a valuable aid for ‘real world’empirical testing.
In summary, for each stage of the theory-buildingprocedure, there is a need for integration of theresults from other research methods to raise theabstraction level of operations management theory.To be a theory-building study, each research projectshould not ignore results from other methodologies,since ignoring other studies lowers the abstractionlevel and reduces the significance of any findings.
9. The current state of theory-building method-ologies in operations management
One question that should be answered is: how arethe different research methods in operation manage-ment being utilized? Table 5 illustrates the currentstate of theory-building in operations management
Ž .over 5 years 1991–1995 inclusively . The top eightacademic and practitioner journals in operationsmanagement provide data for investigating the cur-
Žrent state of operations management research seeŽ ..Barman et al. 1993 . These journals are the Interna-
tional Journal of Operations and Production Manage-Ž . Ž .ment IJOPM ; Decision Sciences DS ; Journal of
Ž .Operations Management JOM ; Management Sci-Ž .ence MS ; Production and Operations Management
Ž .Journal POM ; International Journal of ProductionŽ . Ž .Research IJPR ; Harvard Business Review HBR ;
Ž .and Production Inventory Management PIM . Onecaveat is needed before any further discussion: thisarticle did not investigate the relative degree oftheory-building of each article but rather concen-trated its efforts on classifying each article into thesub-categories of research methodologies.
There are two deliberative ‘judgment calls’ inthese classifications. The ‘first judgment call’ ismade when classifying each article in terms ofwhether its primary contribution is to the operationsmanagement literature, or to another academic field.This study uses a broad perspective to include abroad spectrum of topics from the traditional list of
Žoperations management topics see Hahn et al.Ž ..1982 . Of the 2817 articles reviewed, only 2002 areclassified as operations management. Almost all ofthe articles in the operations management journalsŽInternational Journal of Operations and Production
Ž .Management IJOPM , Journal of Operations Man-Ž .agement JOM , Production and Operations Manage-
Ž .ment POM , International Journal of ProductionŽ .Research IJPR , and Production Inventory Manage-
Ž .ment PIM are designated as contributions to theŽoperations management literature It is not surprising
to operations management academics that DecisionŽ . Ž .Sciences 43% , Management Science 35% , and
Ž .the Harvard Business Review 17% all have signifi-cantly less than 50% of the articles related to tradi-
.tional POM topics. See Table A.1 in Appendix A. .The ‘second judgment call’ is the sorting of arti-
cles into the six major categories. This sorting re-quired considerable judgment since studies fre-quently use multiple research methodologies. In thisstudy, the classifying procedure concentrated on thepredominant methodology used in the study. For
Žexample, the productivity article not included since. Ž .pre-1991 of Hayes and Clark 1989 would be
considered a cross between empirical and analyticalresearch since it uses statistical samples along within-depth case studies to logically derive a conceptualproductivity theory. However, when considered moreclosely, the article’s primary methodology uses sev-
Table 5Overall classification of the articles by research sub-category
( )J.G. WackerrJournal of Operations Management 16 1998 361–385380
eral case studies over time to develop the factoryproductivity theory. Consequently, it would havebeen classified in the empirical case study sub-cate-gory.
Classifying articles into the six research sub-cate-gories addresses the question of: ‘‘How are thearticles distributed into research sub-categories overthe last 5 years?’’ Using Table 5, it is apparent thatover the last 5 years over 55% of all operationsmanagement’s published articles use the analytical
Žmathematical method see Table A.2 in Appendix A.for the breakdown by journal . The next most popu-
lar methodology is analytical conceptual, followedby empirical statistical, and empirical case study.Both empirical experimental and analytical statisticalare not popular methodologies in the operationsmanagement literature. For the academic journals,
Žfive of the journals Management Science, Interna-tional Journal of Production Research, Decision Sci-ences, Production Operations Management Journal
.and the Journal of Operations Management had amajority of articles following the analytical mathe-matical methodology. Only the International Journalof Operations and Production Management has itslargest percent of articles in any other methodologyŽ .analytical conceptual was the most popular . As one
Žmight expect, the broader based journal Harvard.Business Review and the practitioner-oriented jour-
Ž .nal Production and Inventory Management had ahigher percentage of their articles classified intoanalytical conceptual and empirical case studies.However, the fact remains, that in the respectedacademic journals of operations management, thepredominant research methodology is analyticalmathematical.
One important implication of these results is thatempirical experimental design and analytical statisti-cal methodologies are not being used to verifycausality and integrate theory. The implications ofthis result leads to two important constraints fortheory-building in operations management. First, be-cause experimental design is not being used, there isa lack of studying causality in the empirical worldwhich hinders the development of verified relation-ships that are assumed to be true. This lack ofverified relationships means that all relationshipsmust be tested through empirical testing, causingstatistical models to be too complex for meaningful
Ž .investigation Hunt, 1991 . Second, because analyti-cal statistical methodology is under-researched, thereis a lack of an integrated internally consistent theoryacross all methodologies used in operations manage-ment. Since analytical statistical methods’ basic pur-pose is to develop integrated models across all re-search methods, it provides a critical linkage toelevate theory to higher-level abstraction levels. Inshort, the evidence presented here suggests that oper-ations management is not fully utilizing all researchmethods to verify old relationships and to buildintegrated theory.
10. How this article follows the criteria for ‘good’theory-building
This article developed a definition of theory fromthe academic literature. The definition refined theconditions for what constitutes a theory. Next, itlimited the domain to ‘good’ theory using the tradi-tional virtues of ‘good’ theory. Third, it suggested aprocedure for all types of theory-building researchwhich fulfills the definition of theory. Additionally,this study classified research into six categories anddemonstrated that all sub-categories of theory-build-ing research in operations management use the samestages. Finally, it presented evidence that the sixpredominant types of theory-building research in op-erations management are not all equally distributed.This unequal distribution led to the conclusion thatresearch on causality and theory integration are hin-dering the elevation of the abstraction level of opera-tions management theory.
11. Guidelines for theory-builders and conclusions
The purpose of this section is to offer theory-building guidelines for theory-builders. It is impor-tant for the academic field to follow the formaldefinition of theory for theory-building research andincorporate these elements: definitions, domain, rela-tionships, and predictions. Consequently, the follow-ing is a set of questions to provide guidelines fortheory-builders.
Ž .1 For definitions: are the terms used in the studyŽ .standard artificial terms agreed upon by aca-
demics? Are the terms unique? Are new terms care-
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 381
fully differentiated from the standard terms in exis-Ž .tence conservation criteria ? Do all the new and
standard terms avoid ‘concept stretching’? Do all theterms used carefully exclude other concepts? Aredifficulties with measurement of definitions high-lighted for future researchers?
Ž .2 For domains: are the specific conditions ofwhen and where the data were gathered carefullyenumerated? Are conditions for when and where theresults apply prudently articulated? Is the domain ofthe results wide enough to be of value for re-searchers using other research methodologies?
Ž .3 For relationship and model building: are allthe variables listed in the estimates necessaryŽ .parsimony criteria ? Do all the variables used
Žspecifically state how and why they are related or.unrelated to each and every other variable in the
Ž .study internal consistency criteria ? Is the model asŽsmall as possible to explain all the results parsimony
.criteria ? Do the relationships offer new areas forŽ .research fecundity criteria ? Are all the relationships
specified before the data is gathered and before theŽrelationships are estimated internal consistency and
.empirical riskiness criteria ? Does the model elevatethe level of abstraction by adding variables and
Ž .inter-relating theories abstraction criteria ?Ž .4 For theory prediction: does the model used
make specific predictions which could likely be fal-Ž .sified empirical riskiness criteria ? Do the theory
predictions prohibit some events from happening?Does the theory discuss specifically how the theoryis to be used and tested in the external empirical
Žworld i.e., specifying bridge laws which are needed.for empirical riskiness ?
Although all these guidelines are necessary for alltheory-building articles, there are four specific con-cerns for the full development of operations manage-ment theory. First, there is a need for cautiousattention to the measurement of the defined terms,since without exacting definitions, all theory is tenu-ous at best. Second, the specific domain of when andwhere the theory is to be applied is needed for thetheory to be empirically tested. Third, the relation-
ship-building stage is most important for empiricalresearchers since all variables should specificallystate whether they are related or unrelated to eachand every other variable. Consequently, the ‘how’and ‘why’ each variable is related other variablesassures that the empirical models are internally con-sistent. On the other hand, the fourth criteria, predic-tion, is most important for analytical methodologies,since these methodologies should offer how the the-ory can be tested and refuted in the external empiri-
Ž .cal world empirical riskiness virtue .A final word about the above theory-building
guidelines. These guidelines are to be used as sug-gestions rather than hard and fast rules which cannever be violated. Each guideline offered is basedupon ‘good’ theory’s virtues. Since ‘good’ theory’svirtues are weighed against each other, judgment isnecessary to determine the relative importance of
Ž .each virtue and each guideline . Still, theory-builderscan use the above guidelines to increase their re-search’s significance by ‘good’ theory-building andraising their research’s abstraction level. It was inthis hope that this article was written.
Acknowledgements
The author wishes to thank Professor R. KennethTeas for his suggestions, comments, insights andlengthy discussions on this study. The author alsowishes to thank numerous academics for their in-sights and discussion that inspired many of thethoughts presented in this study. In particular, the
Žauthor wishes to thank Richard Van Iten Iowa State. Ž .University , Paul Swamidass Auburn University ,
Ž .and Jack Meredith Wake Forest University for thehelpful discussions. Next, the author wishes to thankthe three anonymous referees and the editors of thisissue. Last, the author wishes to thank MichelleJohnson, Matthew Hanson, and Vicki Bouillon fortheir help with the editing. Naturally, any errors,omissions, or oversights are entirely the responsibil-ity of the author.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385382
App
endi
xA
A.1
.T
able
A1:
The
clas
sifi
cati
onof
arti
cles
inle
adin
gop
erat
ions
man
agem
ent
jour
nals
for
prod
ucti
onr
oper
atio
nsm
anag
emen
tto
pics
IJO
PM
DS
JOM
MS
IJP
RP
OM
HB
RP
IMJ
Tot
alY
ear
PO
MN
on-
PO
MN
on-
PO
MN
on-
PO
MN
on-
PO
MN
on-
PO
MN
on-
PO
MN
on-
PO
MN
on-
PO
MP
OM
PO
MP
OM
PO
MP
OM
PO
M19
9162
629
4519
335
4116
11
757
630
1992
646
3352
4072
175
280
1055
630
1993
655
2121
280
4871
172
116
011
5063
019
9478
519
1922
134
8117
219
015
4763
019
9550
515
2014
042
9915
31
240
951
600
Tot
al31
927
117
157
834
199
364
833
387
052
260
312
081
5)P
erce
nt92
.20%
7.80
%42
.70%
57.3
0%95
.40%
4.60
%35
.35%
64.6
5%99
.64%
0.36
%10
0.00
%0.
00%
16.6
7%83
.33%
100.
00%
0.00
%28
.93%
)
Non
-PO
Mar
ticl
esov
eral
lan
dno
n-P
OM
perc
enta
geof
all
thos
ere
view
ed.
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 383
A.2
.T
able
A.2
:OÕer
all
clas
sifi
cati
onof
all
arti
cles
Jour
nal
Res
earc
hty
pe19
9119
9219
9319
9419
95T
otal
perc
ent
Inte
rnat
iona
lJo
urna
lof
Ope
rati
ons
and
Ana
lyti
cal
conc
eptu
al20
2023
3215
110
34.4
8%Ž
.P
rodu
ctio
nM
anag
emen
tIJ
OP
MA
naly
tica
lm
athe
mat
ical
2716
2621
1910
934
.17%
Ana
lyti
cal
stat
isti
cal
10
00
12
0.63
%E
mpi
rica
lex
peri
men
tal
00
01
12
0.63
%E
mpi
rica
lst
atis
tica
l9
1210
158
5416
.93%
Em
piri
cal
case
stud
y5
166
96
4213
.17%
Ž.
Dec
isio
nS
cien
ces
DS
Ana
lyti
cal
conc
eptu
al2
72
21
1411
.97%
Ana
lyti
cal
mat
hem
atic
al27
2015
125
7967
.52%
Ana
lyti
cal
stat
isti
cal
02
01
25
4.27
%E
mpi
rica
lex
peri
men
tal
01
11
03
2.56
%E
mpi
rica
lst
atis
tica
l0
23
36
1411
.97%
Em
piri
cal
case
stud
y0
10
01
21.
71%
Jour
nal
ofO
pera
tion
sA
naly
tica
lco
ncep
tual
47
70
1821
.69%
Ž.
Man
agem
ent
JOM
Ana
lyti
cal
mat
hem
atic
al9
1711
845
54.2
2%A
naly
tica
lst
atis
tica
l0
00
11
1.20
%E
mpi
rica
lex
peri
men
tal
00
00
00.
00%
Em
piri
cal
stat
isti
cal
43
45
1619
.28%
Em
piri
cal
case
stud
y2
10
03
3.61
%Ž
.M
anag
emen
tS
cien
ceM
SA
naly
tica
lco
ncep
tual
25
94
424
12.0
6%A
naly
tica
lm
athe
mat
ical
3132
3526
3015
478
.97%
Ana
lyti
cal
stat
isti
cal
02
21
38
9.30
%E
mpi
rica
lex
peri
men
tal
00
00
11
1.25
%E
mpi
rica
lst
atis
tica
l2
12
34
1215
.19%
Em
piri
cal
case
stud
y0
00
00
00.
00%
( )J.G. WackerrJournal of Operations Management 16 1998 361–385384
Pro
duct
ion
and
Ope
rati
ons
Ana
lyti
cal
conc
eptu
al9
43
420
22.9
9%Ž
.M
anag
emen
tP
OM
Ana
lyti
cal
mat
hem
atic
al12
1110
1245
51.7
2%A
naly
tica
lst
atis
tica
l0
11
02
2.30
%E
mpi
rica
lex
peri
men
tal
00
00
00.
00%
Em
piri
cal
stat
isti
cal
60
37
1618
.39%
Em
piri
cal
case
stud
y1
02
14
4.60
%In
tern
atio
nal
Jour
nal
ofP
rodu
ctio
nA
naly
tica
lco
ncep
tual
4038
3036
1916
319
.57%
Ž.
Res
earc
hIJ
PR
Ana
lyti
cal
mat
hem
atic
al11
312
313
112
711
961
373
.59%
Ana
lyti
cal
stat
isti
cal
11
10
03
0.36
%E
mpi
rica
lex
peri
men
tal
23
30
08
0.96
%E
mpi
rica
lst
atis
tica
l4
82
511
303.
60%
Em
piri
cal
case
stud
y1
25
44
161.
92%
Ž.
Har
vard
Bus
ines
sR
evie
wH
BR
Ana
lyti
cal
conc
eptu
al3
34
75
2242
.31%
Ana
lyti
cal
mat
hem
atic
al0
00
00
00.
00%
Ana
lyti
cal
stat
isti
cal
00
00
00
0.00
%E
mpi
rica
lex
peri
men
tal
00
00
00
0.00
%E
mpi
rica
lst
atis
tica
l2
01
01
47.
69%
Em
piri
cal
case
stud
y2
76
83
2650
.00%
Pro
duct
ion
and
Inve
ntor
yA
naly
tica
lco
ncep
tual
3228
2123
1511
938
.4%
Ž.
Man
agem
ent
PIM
Ana
lyti
cal
mat
hem
atic
al14
1111
1113
6019
.23%
Ana
lyti
cal
stat
isti
cal
00
10
01
0.32
%E
mpi
rica
lex
peri
men
tal
00
10
01
0.32
%E
mpi
rica
lst
atis
tica
l7
1213
1214
5818
.54%
Em
piri
cal
case
stud
y10
1216
1718
7323
.40%
Tot
al20
02
( )J.G. WackerrJournal of Operations Management 16 1998 361–385 385
Barman, S., Tersine, R.J., Buckley, M.R., 1993. An empiricalassessment of the perceived relevance and quality of POMjournals by academicians. J. Operations Manage.
Bluedorn and Evered, 1960.Bunge, M., 1967. Scientific Research 1: The Search For System.
Springer-Verlag, New York.Churchman, C.W. 1961. Prediction and Optimal Decisions. Pren-
wood, IL.Demeyer and Arnoud, 1989.Eisenhardt, K.M., 1989. Building theories from case study re-
Ž .search. Acad. Manage. Rev. 14 4 , 532–550.Ettlie, J., 1995. Product-process development in manufacturing.
Ž .Manage. Sci. 41 7 , 1224–1237.Hahn, C., Pinto, P., Mabert, V., Biggs, J., 1982. A profile of
undergraduate programs in production and inventory manage-ment. Prod. Inventory Manage. J., Third Quarter, 48–66.
Hayes and Clark, 1989.Hunt, S.D., 1991. Modern Marketing Theory: Critical Issues in
the Philosophy of Marketing Science. Southwestern Publish-ing, Cincinnati, OH.
Kekre, S., Srinivasan, K., 1990. Broader product line: a necessityŽ .to achieve success. Manage. Sci. 36 10 , 1216–1231.
Kim, Y., Lee, J., 1993. Manufacturing strategies and productionsystems: an integrated framework. J. Operations Manage. 11Ž .1 , 3–15.
Klemschmidt, E.J., Cooper, R.G., 1991. The impact of productinnovation on performance. J. Prod. Innovation Manage. 8,240–251.
Kuhn, T.S., 1980. In: Klempe, E.D., Hollinger, R., Kline, A.D.Ž .Eds. , Theory Choice, Introductory Readings in the Philoso-phy of Science. Prometheus Books, Buffalo, New York.
Lindblom, C.E., 1987. Alternative to validity: some thoughtssuggested by Campbell’s guidelines. Knowledge: Creation,Diffusion, Utilization 8, 509–520.
Madique, M., Hayes, R., 1984. The art of high-technology man-Ž .agement. Sloan Manage. Rev. 25 2 , Winter.
Marucheck, A., Pannesi, R., Anderson, C., 1990. An exploratorystudy of the manufacturing strategy process in practice. J.
Ž .Operations Manage. 9 1 , 101–123.
Meredith, J.R., Raturi, A., Amoako-Gyampah, K., Kaplan, B.,1989. Alternative research paradigms in operations. J. Opera-
Ž .tions Manage. 8 4 , 297–326.Miller and Roth, 1995.Moorthy, K.S., 1993. Theoretical modeling in marketing. J. Mar-
keting 57, 92–106.Osigweh, Chimizie, A.B., 1989. Concept fallibility in organiza-
Ž .tional science. Acad. Manage. Rev. 14 4 , 579–594.Poole, 1989.Poole, M.S., Van De Ven, A.H., 1989. Using paradox to build
Popper, S.K., 1957. Philosophy of science: a personal report. In:Ž .Mace, C.A. Ed. , British Philosophy in Mid-Century.
Quine, W.V., Ullian, J.S., 1980. Hypothesis. In: Klempe, E.D.,Ž .Hollinger, R., David Kline, A. Eds. , Introductory Readings
in the Philosophy of Science. Prometheus Books, Buffalo,New York.
Reynolds, P.D., 1971. A Primer In Theory Construction, Bobbs-Merrill Educational Publishing, IN.
Robinson, A., 1990. The Shingo System: Modern Approaches toManufacturing Improvement. Productivity Press, Cambridge.
Sax, 1968.Shubik, M., 1987. What is an application and when is theory a
Ž .waste of time?. Manage. Sci. 33 12 , 1511–1522.Sutherland, 1976.Swamidass, P.M., 1986. Manufacturing strategy: its assessment
Ž .and practice. J. Operations Manage. 6 4 , 471–484.Teas, R.K., Palan, K.M., 1997. The realms of scientific meaning
framework for constructing theoretically meaningful nominalŽ .definitions of marketing concepts. J. Marketing, forthcoming .
Van de Ven, A.H., 1989. Nothing is quite so practical as a goodŽ .theory. Acad. Manage. Rev. 14 4 , 486–489.
Wacker, J.G., 1987. The complementary nature of manufacturinggoals by their relationship to throughput time: a theory ofinternal variability of production systems. J. Operations Man-
Ž .age. 7 2 , 91–106.Wacker, J.G., 1996. A theoretical model of manufacturing lead
times and their relationship to a manufacturing goal hierarchy.Ž .Decision Sci. 27 3 , 483–617.
Westbrook, R., 1987. Time to forget ‘Just-in-Time’?: observationsŽ .on a visit to Japan. Int. J. Prod. Operations Manage. 8 4 ,
5–21.Whetten, D.A., 1989. What constitutes a theoretical contribution?.
