Proceedings of IDETC/CIE 2006 ASME 2006 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference September 10-13, 2006, Philadelphia, Pennsylvania USA DETC2006-99163 FLEXIBLE PRODUCT PLATFORMS: FRAMEWORK AND CASE STUDY Olivier L. de Weck ∗ Center for Innovation in Product Development Aeronautics & Astronautics, Engineering Systems Division Massachusetts Institute of Technology Cambridge, Massachusetts 02139 Email: [email protected]Eun Suk Suh Printing Platforms & Systems Engineering Xerox Innovation Group Xerox Corporation Webster , New York 14580 Email: [email protected]ABSTRACT Customization and market uncertainty require in- creased functional and physical bandwidth in product plat- forms. This paper presents a platform design process in response to such future uncertainty. The process consists of seven iterative steps and is applied to an automotive body- in-white (BIW) where 10 out of 21 components are identi- fied as potential candidates for embedding flexibility. The paper shows how to systematically pinpoint and value flex- ible elements in platforms. This allows increased product family profit despite uncertain variant demand, and speci- fication changes. We show how embedding flexibility sup- presses change propagation and lowers switch costs, despite an increase of 34% in initial investment for equipment and tooling. Monte Carlo simulation results of 12 future sce- narios reveal that as the degree of uncertainty increases, the value of embedding flexibility also increases. 1 INTRODUCTION Mass customization emerged as a paradigm in the late 1980s (Pine 1993) and focuses on serving the needs of indi- vidual customers through high product variety. This demanded a corresponding decrease in development time (Sanderson and Uzumeri 1997). Manufacturers were forced to seek more effi- cient and flexible product design and manufacturing strategies. Two of the more successful strategies were the lean man- ufacturing strategy (Womack et al. 1991) and the product plat- form strategy (Meyer and Lehnerd 1997; Bremmer 1999). The lean manufacturing strategy attempts to reduce manufacturing costs by eliminating inefficiencies in the supply chain, as well as in fabrication and assembly processes. The product plat- form strategy attempts to save costs by sharing core elements among different products in the product family. (Simpson et al. ∗ Address all correspondence to this author. 2006). Both strategies have received significant attention in the literature, but opportunities for further research still abound. This is mainly so because new situations arise that are not han- dled by the traditional approaches. Lean supply chains have been shown to be excessively vulnerable due to unexpected dis- ruptions such as terrorism and natural disasters (Sheffi 2005). Product platforms often turn out to be overly constraining in a dynamic market environment. Figure 1 illustrates this last point by showing the percent change in aggregate demand for various types of automobiles in the United States from 2003 to 2004. While small sports utility vehicles (SUV) and crossover wagons gained in pop- ularity, traditional large cars (sedans) and pickup trucks suf- fered significant losses. These market dynamics are caused by a multitude of factors such as the price of fuel, changing demo- graphics, international competition and shifting customer pref- erences in terms of styling and favored functional attributes. Figure 1. Changes in aggregate unit sales from 2003 to 2004 for U.S. car and truck market (Simmons 2005) When new products are designed in response to or antici- pation of such changes, the manufacturing firm has essentially two alternatives: design a full up new product or derive a prod- uct by modifying an existing product to suit the changed re- 1 Copyright c 2006 by ASME
Transcript
Proceedings of IDETC/CIE 2006ASME 2006 International Design Engineering Technical Conferences &
Computers and Information in Engineering ConferenceSeptember 10-13, 2006, Philadelphia, Pennsylvania USA
DETC2006-99163
FLEXIBLE PRODUCT PLATFORMS: FRAMEWORK AND CASE STUDY
Olivier L. de Weck∗
Center for Innovation in Product DevelopmentAeronautics & Astronautics, Engineering Systems Division
Massachusetts Institute of TechnologyCambridge, Massachusetts 02139
ABSTRACTCustomization and market uncertainty require in-
creased functional and physical bandwidth in product plat-forms. This paper presents a platform design process inresponse to such future uncertainty. The process consists ofseven iterative steps and is applied to an automotive body-in-white (BIW) where 10 out of 21 components are identi-fied as potential candidates for embedding flexibility. Thepaper shows how to systematically pinpoint and value flex-ible elements in platforms. This allows increased productfamily profit despite uncertain variant demand, and speci-fication changes. We show how embedding flexibility sup-presses change propagation and lowers switch costs, despitean increase of 34% in initial investment for equipment andtooling. Monte Carlo simulation results of 12 future sce-narios reveal that as the degree of uncertainty increases,the value of embedding flexibility also increases.
1 INTRODUCTIONMass customization emerged as a paradigm in the late
1980s (Pine 1993) and focuses on serving the needs of indi-vidual customers through high product variety. This demandeda corresponding decrease in development time (Sanderson andUzumeri 1997). Manufacturers were forced to seek more effi-cient and flexible product design and manufacturing strategies.
Two of the more successful strategies were the lean man-ufacturing strategy (Womack et al. 1991) and the product plat-form strategy (Meyer and Lehnerd 1997; Bremmer 1999). Thelean manufacturing strategy attempts to reduce manufacturingcosts by eliminating inefficiencies in the supply chain, as wellas in fabrication and assembly processes. The product plat-form strategy attempts to save costs by sharing core elementsamong different products in the product family. (Simpson etal.
∗Address all correspondence to this author.
2006). Both strategies have received significant attentionin theliterature, but opportunities for further research still abound.This is mainly so because new situations arise that are not han-dled by the traditional approaches. Lean supply chains havebeen shown to be excessively vulnerable due to unexpected dis-ruptions such as terrorism and natural disasters (Sheffi 2005).Product platforms often turn out to be overly constraining in adynamic market environment.
Figure 1 illustrates this last point by showing the percentchange in aggregate demand for various types of automobilesin the United States from 2003 to 2004. While small sportsutility vehicles (SUV) and crossover wagons gained in pop-ularity, traditional large cars (sedans) and pickup truckssuf-fered significant losses. These market dynamics are caused bya multitude of factors such as the price of fuel, changing demo-graphics, international competition and shifting customer pref-erences in terms of styling and favored functional attributes.
Type of Model2004
unit sales% change from 2003
Small SUV 269,851 18.5Sport wagon/crossover 1,734,622 13Large pickup trucks 2,456,656 7.8Large van 344,693 6.6Luxury cars 1,496,753 4.4Minivan 1,110,817 3.4Mid-size SUV 1,742,463 0.7Luxury SUV 237,065 -0.6Mide-size car 3,500,065 -1.1Small car 2,194,148 -1.8Large SUV 759,157 -6.1Small pickup truck 653,823 -10.6Large car 306,257 -25
Figure 1. Changes in aggregate unit sales from 2003 to 2004 for U.S.
car and truck market (Simmons 2005)
When new products are designed in response to or antici-pation of such changes, the manufacturing firm has essentiallytwo alternatives: design a full up new product or derive a prod-uct by modifying an existing product to suit the changed re-
quirements. If these modifications are done in a systematicway with sharing of common elements across multiple vari-ants we call this a platform strategy. Figure 2 shows an exam-ple of a new vehicle derived from an existing platform. Theexploded view highlights new and unique components (darkgray), carryover-modified (light grey) and carryover-common(medium grey) components. Only the last category of elementsis reused without modifications and is traditionally considereda part of the platform.
Figure 2. Decomposition of new automotive product (courtesy: Gen-
eral Motors, 2004)
The tension between wanting to reuse as much as possiblefrom previous products, i.e. having the platform comprise alarge percentage of the product, and the desire for distinctive-ness, innovation and new styling requiring many new-uniquecomponents is well documented in practice and in the acad-emic literature (Simpson et al. 2006). What has not receiveda lot of attention is the second category of components in Fig-ure 2. The components labeled as “carryover-modified” arethose that are very similar to existing components, but not ex-actly the same. These components are generated by redesign ofexisting components, and such redesigns are most often donein an expensive reactive mode. The degree of change varies,but oftentimes these components require substantial redesignas well as tooling and equipment changes in manufacturing.The purpose of this paper is to develop and demonstrate a sys-tematic design process for treating such elements as “flexibleelements” and to consider them as part of an expanded productplatform. The hypothesis is that if the right subset of elementsis designed with flexibility, that a platform will be more nimblein the future, therefore avoiding expensive redesigns and man-ufacturing switch(ing) costs.We strive to (1) demonstrate howto select flexible elements by projecting exogenous uncertaintyinto the platform and (2) to quantify both the additional up-front investment required to achieve this flexibility as well asthe downstream benefits resulting from the investment.
After a brief literature review in Section 2, we present anormative flexible product platform design process in Section3. This flexible platform design process takes exogenous un-certainties into account and incorporates the concept of flexi-ble elements. Flexibility is defined as “the property of a systemthat is capable of undergoing specified classes of changes withrelative ease (Moses 2002).” In Section 4 we demonstrate theprocess in a real world case study where three car variants are
to be built from a common, but flexible platform. How muchflexibility is needed? How much will flexibility cost? Whatare the future benefits of flexibility? We will attempt to answerthese questions in Section 5.
2 PREVIOUS WORKThe state of the art in product family and platform design
research has been recently summarized and broadly reviewedby Simpson, Siddique and Jiao (2006). Instead of repeatingsuch a broad review here, we will focus our discussion on fivepapers that are most closely related to product platform designunder uncertainty. They are papers by Simpson et al. (2001),Martin and Ishii (2002), Li and Azarm (2002), and Gonzalez-Zugasti, Otto et al. (2000, 2001).
Simpson et al. (2001) proposed the Product Platform Con-cept Exploration Method (PPCEM). In the paper, the authorsstate that PPCEM is a “formal method that facilitates the syn-thesis and exploration of a common product platform conceptthat can be scaled into an appropriate family of products.” Themethod applies to scalable product platforms and families,andconsists of five steps: 1) market segmentation grid creation,2) factor and range classification, 3) meta-model creation andvalidation, 4) product platform specifications aggregation, and5) product platform and family development. The method wasdemonstrated through a universal motor case study, in whichafamily of ten motors is designed by varying the stack length.
Martin and Ishii (2002) developed the Design for Variety(DFV) method, to develop modularized product platforms. Theauthors used the Generational Variety Index (GVI) and Cou-pling Index (CI) to design platforms that can be easily changedin the future. In the paper, GVI is defined as an “indicator ofthe amount of redesign required for a component to meet thefuture market requirements.” The CI “indicates the strength ofcoupling between the components in a product. The strongerthe coupling between components, the more likely a change inone will require a change in the other.” The method is demon-strated through a water cooler example, in which the GVI andCI for seven major components are calculated. Then, for com-ponents with high GVI and CI, flexible designs are generatedto reduce GVI and CI, thus lowering future redesign cost.
Li and Azarm (2002) developed a design process for aproduct line (family) design under uncertainty and competi-tion. The design process is divided into the design alternativegeneration stage and the design evaluation stage. During thedesign alternative generation stage, each design alternative isoptimized through multiobjective optimization. In the designevaluation stage, each design alternative is optimized andeval-uated using a Multi-Objective Genetic Algorithm (Narayananand Azarm 1999), due to the combinatorial nature of the for-mulated optimization problem. In the end, the best productline (family) is chosen using a selection rule, which takes intoaccount the designer’s utility of the product line as a whole.The proposed design process was demonstrated through a casestudy in which a cordless screw driver family is designed. Ofthe three major components (motor, gear, battery), the motorwas designated as the platform componenta priori. Throughoptimization of the other components, the authors identifiedbest designs for several different uncertain scenarios.
Finally, Gonzalez-Zugasti, Otto and Baker introduced aquantitative method to design product platforms (Gonzalez-Zugasti et al. 2000) and a framework to assess value of theproduct platform-based family using a real options approach(Gonzalez-Zugasti et al. 2001). In the first paper, the pro-posed method was implemented for an interplanetary space-craft family in which three candidate platform designs basedon various telecommunications technologies and bandwidths(X-band, Ka-band, optical) were optimized for mass, cost, andlaunch margin, given a pre-determined set of future NASA mis-sions. In the second paper, the interplanetary spacecraft familywas evaluated under uncertain future mission requirementsandplatform development investments were valued using the realoptions approach.
This previous research covers several areas of product plat-form design which inspired our work. However, of all previ-ously published methods, none deal with an end-to-end designprocess in which the uncertainty is systematically mapped to(functional) product attributes, design variables, physical com-ponents, flexible designs, and then to relevant costs for eco-nomic evaluation. Second, in most processes, the notion of“flexible elements” is not explicitly apparent. This is crucialsince system architects want to knowwhere and how muchflexibility to embed in a system in a specific design-under-uncertainty context. In the methods proposed by Li and Azarm,and by Gonzalez-Zugasti et al., the focus of the process wasto identify common and unique elements for maximum per-formance and/or profit but they offered no mention of flexibleelements. In the work published by Martin and Ishii, flexi-ble design alternatives were presented in the case study, butthe economic consequences and subsequent uncertainty analy-sis were not developed. Work by Simpson et al. deals withscalable (“flexible”) universal motors, but only optimizesthemfor current needs. Finally, most of the previous work deals withrather simple examples, thus not fully capturing the intricacy ofdesigning complex products. The main difficulty in going fromsimple to complex products is that the product architectureisnot trivial and that the effects of change propagation (Eckertand Clarkson 2004) must be captured. In the next section, thesteps and logic of a normative flexible platform design processare presented.
3 FLEXIBLE PLATFORM DESIGN PROCESS3.1 Overview
Figure 3 outlines the framework for designing flexibleproduct platforms. This process assumes that preliminary de-signs of a product platform (set of related common compo-nents) and variants (platform + unique components) have beendeveloped by experience or by using one of the formal methodsproposed in the literature (Simpson et al. 2006). At the outset,however, uncertainty has not yet been considered. The processin Figure 3 resulted from numerous interviews and iterationswith car designers and vehicle architects at a large automotivemanufacturer. However, we believe the process to be generalenough to be applied to other types of physical products.
The process begins by identifying target market segments,product variants, and critical uncertainties that the product plat-form must be able to accommodate (Step I). Subsequently,
functional product attributes impacted by uncertainty andre-lated system-level design variables are identified (Step II). Theidentified set of design variables for each product variant in thefamily is optimized to yield maximum product family revenue(Step III). In this way the required bandwidth for key prod-uct design variables in the product family is determined. Giventhe requirement to achieve bandwidth for uncertainty-impacteddesign variables, a critical set of physical elements, affected bythe design variable change(s), is determined via change prop-agation analysis (Step IV). Using the identified physical el-ements and given bandwidth requirements, flexible platformdesign alternatives are generated (Step V). Initial investment,variable costs, and switch costs for the design alternatives arecalculated in Step VI. The final step in the framework consistsof uncertainty analysis (Step VII), wherein the benefit of eachplatform design alternative is estimated under future scenar-ios with varying degrees of uncertainty. Finally, the best flexi-ble platform design alternative is selected, or one enters aloopback to Step I or Step V if a satisfactory solution has not beenfound. For each of the steps in the platform design process avariety of methods and tools may be used, see Table 1.
The subsections that follow present the generic formula-tion and give explanations for each step of the process.
3.2 Step I: Identify Market, Variants, and Uncertain-ties
The first step of the process is to identify target mar-ket segmentsM = [M1,M2, ...], desired product variantsP =[
p1, p2, ..., pnp
]
assigned to those segments, and a set of uncer-taintiesU = [u1,u2, ...] that are related toM andP . Here weassume that all product variants in a product familyP will bederived from a common product platform. A graphical repre-sentation of the assignment of the set of variantsP to the set ofmarket segments inM is shown in Figure 4.
Product Platform
60
90
120
150
180
$10,000 $20,000 $30,000 $40,000
Pass
enge
r Vol
ume [
cft]
SMLMIDLRGVAN
Market Segmentation for Cars (2002 U.S. Autopro Database)
Price 2002 [$]
p1 p2 p3
SMLMID
LRG
VAN
Figure 4. Market Segmentation based on Clustering
A set of market segmentsM for a specific type of prod-uct is typically defined through clustering analysis (Jajuga et al.2002). Figure 4 shows how small, midsize and large sedans aswell as vans cluster in terms of passenger volume [cft] versusprice [2002 $]. We can see that products in the same marketsegment are grouped together in terms of similar product at-
Determine Uncertainty Related Key Attributes and Design Variables
Optimize Product Family and Platform Bandwidth
Identify Critical Platform Elements
Determine Costs of Design Alternatives
Create Flexible Plat- form Design Alternatives
Uncertainty Analysis
IIIII IV
VVI
Identify Market, Variants and Uncertainties
I
VII
Satisfactory Solution?
Yes
Exit
No
Go Back to Step V
Start
Figure 3. Flexible Product Platform Design Process. Multiple arrows indicate that several alternatives could be carried along.
Table 1. Methodologies and Tools for Individual Design Steps
Step Available Methodologies and ToolsStep I Clustering Analysis (Jajuga et. al 2002), Conjoint Analysis (1992)Step II Principal Components Analysis (Dunteman 1999), QFD (Hauser 1988), Response Surfaces (Myers 2002)Step III Gradient-based Optimization (Papalambros 2000), Heuristic Optimization (Goldberg 1989, Kirckpatrick 1983)Step IV Change Propagation Analysis (Clarkson,Eckert 2004), Engineering Expertise (Bahl and Beitz 1996), QFDStep V Brainstorming (Pahl and Beitz 1996), Concept Screening andScoring Matrix (Ulrich and Eppinger 1999)Step VI Parametric Cost Modeling (Kirchain 2004)Step VII Decision Trees (Clemen 1996), NPV Analysis (de Neufville etal. 2004), Real Options (Trigeorgis 1996)
tributes. Figure 4 assumes that three variants:p1 a small sedan,p2 a midsize sedan andp3, a large sedan, are built from thesame platform and differentiated via price and other attributesnot shown on the plot. An individual product variant can beexpressed as a vector of specific product attributes (JA) andprice (P), i.e. for the i-th product variant letpi = [JA,i Pi ]
T .Therefore, the product familyP can be expressed as a matrixof specific product attributes’ values and prices, as in (1):
P =
[
JA,1 JA,2 ...
P1 P2 ...
]
. (1)
The last item to be defined in this step is a set of uncertaintiesU that might impact the product platform in significant ways.Figure 1 demonstrated that demand can change significantlyfor different market segments from year to year, whereby suchfluctuations can be amplified for individual product variants.To illustrate this point, Table 2 summarizes how various quan-tities have evolved dynamically in the sports utility vehicle(SUV) market in North America in the period from 1999-2003.
The data suggests that the aggregate SUV market hasgrown at an average rate of 10% per year and that these typesof vehicles have grown larger, more powerful and yet slightlymore fuel efficient over the same 5 year period. The data showsthat exogenous uncertainties and future trends cannot be ig-nored in engineering design of product platforms that have longlifecycles (>10 years). The main issue addressed by the flex-ible platform design process (Fig. 3) is that product platformsoften have a lifecycle that exceeds that of the variants built fromit and that market and technological trends are difficult or im-possible to predict accurately over such planning horizons. Aplatform must therefore be designed to accommodate severalproduct variants at its point of inception, as well as be flexibleto respond to future uncertainties.
Table 2. Dynamic evolution of SUV market in North America. 1999-
3.3 Step II: Determine Uncertainty-Related KeyFunctional Attributes and Design Variables
In the previous step, we identified market segmentsM ,product variantsP , and uncertaintiesU . Each market segmentM j can be expressed as a range of customer-preferred attributevalues and price, within which a specific product variant’sJA,i
and price Pi must fall (see dashed boxes in Fig. 4)1:
Depending on the number and position of competing prod-ucts and the firms own current product attribute values in aspecific market segmentM j , the firm needs to set theirith prod-uct’s JA,i and Pi values within the established range ofM j inorder to gain market share and a competitive position.JA isa function of a system-level design variable vectorXA. Ex-amples of system-level design variablesxA,i ∈ XA are height,
wheelbase and engine horsepower rating (see Table 2) in au-tomobiles. These are design variables because customers don’tvalue these variables directly and designers can choose their in-stantiations freely (within bounds and subject to physicalcon-straints). The term “system-level” implies that these variablesare not directly associated with individual components such asthe ones shown in Figure 2, but instead they describe the prod-uct at an aggregate level. We write:
P =
[
JA,1 (XA,1) JA,2(XA,2) ...
P1 P2 ...
]
(3)
Even though there can be many different product attributeswithin JA, the ones that are of special interest are product at-tributes that are related to the set of uncertainties,U . A prod-uct attribute vector, related to a set of uncertaintiesU , can beexpressed asJU , whereJU ⊆ JA. These attributes are signifi-cantly affected by the uncertainties identified in Step I andmustbe mapped to system-level design variables. The next step isto establish the relationship between the uncertainty-specificproduct attributesJU and the related system-level design vec-tor XU , whereXU ⊆ XA. This is expressed as
JU = f (XU ) (4)
Given the target market segmentM j assigned for eachpi , theupper and lower bounds of the uncertainty specific system-leveldesign variables vectorXU ,i for a product variantpi must bewithin the limits ofM j .
3.4 Step III: Optimize Product Family and PlatformBandwidth
For eachpi , defined as a function ofXU ,i and its estab-lished upper and lower bounds, allpi in the product variant setP need to be positioned within their respective market segmentto generate maximum revenue as a product family. This can bestated as:
maximizenp
∑i=1
Rpi (Ju,i (Xu,i) ,Pi)
s.t. h(Ju,i ,Xu,i) = 0g(Ju,i ,Xu,i) < 0
(5)
whereRpi is the total revenue generated by theith product vari-ant, andh and g are inequality and equality constraints thatmust be satisfied. Individual product variant revenueRpi is fur-ther explained in Equation (6):
Rpi = msi, j (Ju,i (Xu,i) ,Pi)PiDT, j (6)
wheremsi, j is the market share for theith product variant in itsassigned market segmentM j (see Fig. 4), andDT, j is the totaldemand in market segmentM j . Market share is a function ofproduct attribute valuesJA and variant priceP.
Estimating a reliable market share for given values ofJA,i
and Pi is, in itself, uncertain and a large research field. It can be
accomplished through conjoint analysis (1992), in which com-panies estimate customers’ preference sensitivities for particu-lar products by systematically changing the product’s attributevalues. Once the maximum revenue generating solution for Eq.(5) is obtained through optimization, the values ofXU ,i andJU ,i for each product variant are determined, thus defining thebandwidthof the product platform in both the system-level de-sign variable space and the customer-preferred attribute space.Figure 5 shows bandwidths of a hypothetical product platformin design variable and attribute space (grey shaded area).
System-Level Design Variable Bandwidth
020406080
100x1
x2
x3x4
x5
Product Attribute Bandwidth
020406080
100J1
J2
J3J4
J5
Variant 1
Variant 2
Variant 3
System-Level Design Variable Bandwidth
020406080
100x1
x2
x3x4
x5
Product Attribute Bandwidth
020406080
100J1
J2
J3J4
J5
Variant 1
Variant 2
Variant 3
shaded area=platform bandwidth
Figure 5. Platform Bandwidth in Design Variable and Attribute Space
3.5 Step IV: Identify Critical Elements for FlexibilityAfter establishing the platform bandwidth, eachxi ⊆ XU ,
must be mapped to a set of specific physical elements. This isan important prelude to identifying critical platform elementsthat must be flexible enough to achieve the desired design vari-able bandwidth, as dictated by the result of variant optimizationin Step III. This step will be explained using a generic exam-ple. Figure 6 shows both a graph (network) and Design Struc-
AC
F G
HB
E
D
Product System
AC
F G
HB
E
D
DSM View
A B C D E F G HA 1 1 1B 1 1 1 1C 1 1 1D 1 1 1E 1 1 1F 1 1 1G 1 1H 1 1 1
Graph (Network) View
Figure 6. Graph and DSM Representation of a Generic System
ture Matrix (DSM) representation of a generic product system.Within the system, there are eight physical elements (A – H)connected to each other. Elements can be connected physi-cally (e.g. welded together), or through information (e.g.sig-nals), energy (e.g. electrical power) or fluid flow. The DSMrepresents the system using a matrix format with 1’s indicat-ing connectivity between elements, see Eppinger et al. (1998).This is useful because (i) when a system-level design variableis required to be flexible, the designer needs to identify systemelements affected by such change; and (ii) when the identifiedelements are changing, the system designer must observe thechange propagation to other elements (which may not be di-rectly related toXU ) to estimate the effects of change.
Next, for everyxi in XU that must have a non-zero band-width (Fig. 5), the designer must observe how a change in∆xpropagates throughout the system. We refer to Clarkson andEckert’s (2004) seminal work on change propagation in thiscontext. There are four sources of changes for product plat-forms:
1. Non-zero bandwidth of design variables is required by theinitial revenue optimization (Step III).
2. Product family revenue or market share might be very sen-sitive to some design variables and might benefit from flex-ibility in the future, even if the initially required bandwidthis zero or very small.
3. Changes might be required in response to changes in othercoupled elements of the system .
4. New product variants might be added in the future.
Figure 7 shows how a hypothetical change∆x can propa-gate through the system. This figure represents the final sys-tem configuration after the change (due to∆x) has been imple-mented, showing the direction of change propagation.
Multiplier: A, C Carrier: B, D, F, G Absorber/Constant: E, H
x A
F G
B
E
C
HD
Product System
Rece
iving
Cha
nge
∆
∆
100
50
75
70
25
65 55
65
Figure 7. Change Propagation Due to ∆x
The termsmultiplier, carrier, absorber, andconstanthavebeen defined by Eckert et al. (2004) to classify elements thatreact to changes. We find this nomenclature compelling andadopt it here. Multipliers are elements that “generate morechanges than they absorb.”Carriersare elements that “absorb asimilar number of changes to those that they cause themselves.”Absorbersare elements that “can absorb more change than theythemselves cause.” Finally,constantsare elements “that areunaffected by change.” In Figure 7, each element is classified,with multipliers indicated as circled elements. Then the ques-tions are: how can these classes of elements be identified quan-titatively, and how does identification of such elements providea guide for better (flexible) product platform design?
To measure the degree of change propagation for a singleelement,i, we introduce the Change Propagation Index (CPI)measuring the degree of physical change propagation causedby this element when the change is imposed on the element.Equation (7) is shown below:
CPIi =n
∑j=1
∆E j ,i −n
∑k=1
∆Ei,k = ∆Eout,i −∆Ein,i (7)
where∆E is a binary change propagation matrix. In Equa-tion (7), n is the number of elements in the system; and∆Ei, j
is a binary number (0,1) indicating whether theith element ischanged because of elementj. CPI helps classify elements andmeasures physical change propagation to other elements.
However, simply measuring the degree and number ofphysical change propagation instances is not enough. One mustalso consider the cost impact caused by∆x on the system viaits affected elements. For each changed element, the change-related investment cost (switch costKswitch) needs to be iden-tified. This provides the system designer with two quantitativemeasures for each element and each type of change∆x: one in-dicating the degree of physical change propagation (CPI), andthe other indicating the economic consequence of such change(Kswitch).
In Figure 7, the final state of change propagation is shownfor a system after it has been altered due to the design variablechange∆x. This final state can be expressed in matrix formshown in Figure 7 (bottom). The column sum indicates the to-tal number of changes going outward from a specific element(∑∆Eout). The row sum indicates the total number of changescoming into a specific element (∑∆Ein). Subtracting∑∆Ein
from ∑∆Eout yields the CPI value for each specific element.Depending on the value of the CPI, an element can be classi-fied according to the terms previously defined. A positive CPIindicates that the element is amultiplier (classM ); a zero CPI(with ∆Eout,i = ∆Ein,i 6= 0) indicates that the element is acar-rier (classCa); a negative number indicates that the element isanabsorber(classA) and an element with∆Eout,i = ∆Ein,i = 0is aconstant.
The numbers added above each component in Figure 7(top) show the relevant switch cost,Kswitch (hypothetical) dueto change propagation. Note that for element A (the change ini-tiating element), total incoming change is set to 0 since thereis no component sending changes to that particular component.The switch cost is the engineering cost of design changes and
additional fabrication and assembly tooling and equipmentin-vestment to implement the changes. Based on the CPI andswitch cost incurred for each element, the following recom-mendations can be made for selecting critical elements and(re)designing them to be flexible:
1. Multiplier elements are prime candidates for incorporatingflexibility. These are elements that, as more changes areadded, make the system harder to change.
2. One must investigate elements connected to multiplier ele-ments to understand the nature of change. These elementsmight require flexibility (a “buffer” to absorb the change,as Eckert (2004) calls them) to reduce or even eliminatechange propagation.
3. Carrier elements must be examined as well. For example,a carrier element might receive changes from five elementsand send out five changes, making it more expensive thana multiplier element that receives change from only oneelement and sends it out to two elements.
4. Elements with high switch costs, even though they maynot be multipliers, also require special attention. Theseelements, through high switch costs, make it financiallyunattractive to change the system in the future.
5. Physical suppression of future change propagation and in-vestment in flexibility must be carefully balanced. In somecases, physical propagation may be entirely eliminated,but it may require prohibitive investment to do so.
One practical example for the last point comes from the au-tomotive industry. When engineers design a front motor com-partment (see Section 4), they may have the option to design thecompartment to accommodate a future V8 engine, even thoughit may only require a V6 engine initially. This will incur extraupfront investment, but when a future situation requires imple-mentation of the V8 engine configuration, the built-in optioncan reduce or eliminate the change propagation to other majorparts of the vehicle.
3.6 Step V: Create Flexible Design AlternativesWith target elements for embedding flexibility identified2
in Step IV, the system designer must consider (re)designingtheelements so that they propagate a smaller degree of changeand/or require lower switch cost than for the inflexible de-sign. This is accomplished by embedding flexibility into keyelements - the ones that have the greatest impact. Accordingto Hull (1993) and de Neufville (2004), such flexibility is areal option, in which we can either avoid downside risks orexploit upside opportunities.” However, this flexibility will of-ten incur additional upfront investment and might result inad-ditional system complexity. This raises important questions:How much flexibility is needed? How should it be embeddedinto these elements?
To answer the first question, we examine the platformbandwidth obtained by the revenue optimization in Step III(Fig. 5). The upper and lower limit values ofXA, establishedthrough Equation (5), set the range within which the platformmust be flexible. Additionally, sensitive system design vari-ables inXU need to be examined (see case study below).
2change multipliers withCPI > 0 and/or elements withKswitch >> 0
Addressing the second question the system architect mustconsider several factors related to the identified elementssuchas the demand distribution among variants, the types of phys-ical changes required and the frequency with which thosechanges are expected in terms of future product releases. Em-bedded flexibility should be biased towards a particularpi inP to yield favorable overall cost expenditure to amortize in-vestments in flexible parts and tooling. After considering allfactors discussed, the product architect can generate a setofdifferent platform design alternatives. One of the challenges inthis step is the non-uniqueness of the design space. For a givenrequirement of achieving platform bandwidth, multiple flexibledesigns can be generated (Pahl and Beitz (1996)). After flex-ible elements have been generated, the system is divided intotwo portions: (1) the product platform that consists of commonelements, and flexible elements that, with minor modifications,can be used for multiple product variants, and (2) the uniqueportion of the product that is customized for each variant.
3.7 Step VI: Determine Costs of Design AlternativesAt the end of Step V, one or more flexible product plat-
forms are defined. At a minimum we need to be able to com-pare two platform alternatives (rigid, flexible), but couldin-clude platforms with varying degrees of flexibility. To deter-mine whether the generated platform design alternatives areflexible to change, accurate cost estimates for each alternativeneed to be generated. Costs are divided as follows:
1. Initial investment costKinit , which includes fabrication andassembly equipment and corresponding tooling;
2. Variable costCtotal, which is the unit cost of each productmultiplied by the number of products produced;
3. Switch related capital investment costKswitch, which con-sist of investment costs caused by design changes.
To verify that the generated design alternatives are moreflexible than the original design, CPIs and switch costs forthe same set of changes (identified in Step IV) are calculated.For a particular change, one design is more flexible if it incurslower switch costs than other design. However, one must con-sider the extra “price,” paid upfront, to make the system flexi-ble. Whether the upfront investment is worthwhile depends onwhether the flexibility (the option) is truly needed and can beamortized over the course of the product platform life cycle.
3.8 Step VII: Uncertainty AnalysisOnce all costs are identified, design alternatives must be
evaluated under scenarios with varying degrees of uncertaintyto determine their economic performance. The underlying hy-pothesis is that flexibility has more value as the degree of uncer-tainty grows. For each design alternative, the expected futurebenefit expressed in terms of the expected net present value canbe generically stated as:
E [NPV]i = f (RT,i ,Kinit,i ,Ctotal,i ,Kswitch,i ,U ) (8)
where the total expected benefitE[NPV] for theith design alter-native, is a function of the total product family revenueRT,i , the
initial capital investmentKinit,i , the total variable costCtotal,i ,and the switch costKswitch,i incurred due toU , as defined inStep I. After evaluating the proposed platform design alterna-tives under several scenarios, the system designer(s) can selectthe best platform design (maximum expectedNPV) for a givenuncertainty setU . In the next section, the flexible platformprocess is demonstrated through a case study in which a ve-hicle platform is designed with flexibility to respond to futureuncertainties in demand, length and styling.
4 CASE STUDY: AUTOMOTIVE PLATFORM4.1 Background
An automotive company is planning to add a new prod-uct platform to its portfolio. The new platform will accommo-date three vehicle variants that were not previously built froma common platform. All three variants are midsize to largepassenger sedans in different market segments (see Fig.4) andhave different requirements in terms of styling, production vol-ume, and system-level design variables. Two variants will havea short wheelbase and one variant will be a stretched vehicle(long wheelbase). The new platform must be flexible enough toaccommodate the initial vehicle variant specifications, aswellas uncertain changes in the future. To achieve these objectives,we identify a critical subset of vehicle elements, incorporateflexibility into these platform elements, and then evaluatetheflexible design under various uncertain scenarios. This casestudy investigates in detail the Body in White (BIW), which isassumed to be the (structural) product platform. At the end,the common, flexible and unique BIW platform elements aredefined along with recommendations on when to implement aflexible BIW platform versus a traditional (rigid) BIW.
4.2 Step I: Identify Variants, and Uncertainties4.2.1 Market Segments For this case study, the ve-
hicle sedan market segment is divided into smaller segmentsaccording to vehicle size and price (see Fig. 4).
4.2.2 Product Variants We define the sedan vehiclefamily Pveh as: Pveh = [p1, p2, p3] where eachpi in setPveh isdescribed by specific values ofJA and P, according to Equa-tion (1). Detailed explanations ofJA and P for this automotivemarket are presented in the next section. The three variantsarepositioned in the following market segments:
Table 3. Market Segment Designation for each Vehicle Variant pi
Variant Vehicle Market Segmentp1 Mid Size Sedanp2 Large Sedanp3 Large Luxury Sedan
4.2.3 Uncertainties In this case study, the followingset of uncertaintiesU veh is defined:
U veh =[
DPveh (t) SPveh(t)]
. (9)
DPveh is the future demand of the vehicle family as a functionof time t, andSPveh is a discrete sequence of required stylingchanges of the vehicle family as a function of timet. Note thatthe three variants are currently produced on different platformsand that their initial yearly demand is known:DPveh(t = 0) =[280,000 125,000 60,000].
4.3 Step II: Critical Attributes and Design Variables4.3.1 Key Attributes For automobiles, the
customer-preferred attributes setJA has several attributes(Cook 1997). Some of these attributes are fuel economy,acceleration, reliability, towing capacity, and workmanshipquality, to name a few. From these attributes, four attributesrelated to the uncertaintiesU veh, were identified throughinterviews. They are
JU veh = [RM, IE,FE,AC50−70] (10)
RM is customer-perceived vehicle roominess,IE is theease of front ingress/egress,FE is fuel economy, andAC50−70
is the acceleration time interval from 50 to 70 mph.RM andIE are scores between 0 and 100 and represent the percentageof customers who are either “very satisfied” or “satisfied” witha specific vehicle. These scores are derived from past data ob-tained through a market survey of customers who owned theirvehicle for six months or less.RM andIE are selected as at-tributes which are related to one of the uncertainties identified:styling. Vehicle styling is mostly influenced by the shape ofBIW. Similarly, RM and IE are attributes which are also in-fluenced by the BIW shape and key dimensions. The reasonfor selecting these attributes are that: (i) these four attributesare among the most important attributes for market segmentswherePveh is targeted; and (ii)FE and AC50−70 are vehicleperformance attributes affected by the vehicle size, and thus arecoupled withRM andIE. Other attribute values, not includedin JU veh, are treated as constants in the case study.
4.3.2 Design Variables for Key Attributes Oncethe set of uncertain attributesJU veh is identified, the next stepis to establish the mapping relationship between the attributespace and the system-level design variable space, as describedby the system-level design variable setXU veh. For many en-gineering performance attributes, mapping from the attributespace to the system-level design space can be straightforwardand analytical. However, in this case study, the two attributesRM andIE are customer perceived attributes, and so establish-ing the analytical relationship between the two spaces is nottrivial. In order to translate customer judgments in terms of RMandIE into the vehicle-level design variables, various vehicledimensions were decomposed into uncorrelated factors by ap-plying principal component analysis (PCA) (Dunteman 1989).Then, the perceived customer preferences for vehicleRM andIE from a marketing survey were regressed onto those uncorre-lated BIW dimensions. Shown in Figure 8 are relevant systemlevel design variables, identified for each attribute. Dimensionsin the figure are designated using standard SAE nomenclature(SAE 2001).
Figure 8. System Level Design Variables for IE and RM
We assume that there exists a set of design variables thatinfluence people’s perception of vehicle roominess (RM) andease of front ingress/egress (IE) more than others. The first stepis to gather relevant data for different vehicles.RM scores anddimensions for 94 vehicles, produced between 1997 to 2001,were collected for the analysis. ForIE scores and dimensions,57 vehicles, produced between 1995 to 2000, were used (Suh2005). For each vehicle we collected the following data set:
whereX is the mean of the sample data andΣX is the diagonalmatrix of standard deviations. Using the collected standardizeddata, we identified the principal components through singularvalue decomposition:
[U,S,V] = svd(Xs) whereXs = USVT (13)
The principal component matrixV was obtained and we re-tained the first four principal components. Using the principalcomponents analysis,RM andIE scores could be estimated asfunctions of the design variables in Equation 11. Details canbe found in Suh (2005). The following design variables wereselected as a result of the PCA as independent design variablesfor eachpi in Pveh, and they will be used for optimization inStep III:
XU veh,i = [L48i,W3i ,W20i ,H5i ,H50i] (14)
• L48: Second row knee clearance relates toRM and wheel-base differentiation.
• W3: Width W3 is one of the most sensitive dimensionsthat affectsRM.
• W20: Head to centerline width affectsRM strongly.• H5: Distance from ground to seat for ease ofIE.• H50: Overall BIW height affects bothRM andIE.
From the dimensions shown in Figure 8, several depen-dent variables are expressed as functions of independent de-sign variables defined in Equation (14). The dependent vari-ables areH11, H30, H31, H61, H63, andS97. The last taskis to identify constants, which are either common or unique foreach vehicle variant. They areL18, W27, H112, H115, andH122. The constantsL18, H112, andH115 are the same forall vehicle variants. The variablesW27 andH122 are variant-unique for styling differentiation of the “greenhouse”, i.e. thepart of the vehicle above the belt line, see Fig. 8 lower right.This is further discussed below.
4.4 Step III: Optimize Product Platform Bandwidth4.4.1 Product Family Optimization The ultimate
goal of the product platform is to maximize profit of the prod-uct family built from it through product variety increase andcost reduction. To begin the process of maximizing profit, thefirst task is to position each vehicle variant inPveh within thecorresponding vehicle market segment to generate maximumrevenue as a product family. Using relationships defined previ-ously, the revenue maximization problem for the vehicle vari-ant setPveh can be formulated as shown in Equation (15).
maximize3∑
i=1Rpi ;Rpi = msi
(
JU veh,i ,Ppi
)
Ppi DT
w.r.t.{
XU veh,i ,Ppi
}
s.t.h(
JU veh,i ,XU veh,i)
= 0, g(
JU veh,i ,XU veh,i)
< 0
(15)
In the equation, the individual vehicle market sharemsi is acritical value that is difficult to estimate. In our case study thisinformation was obtained through a market simulation softwarefor the North American automotive market for the 2002 modelyear, as a function of aforementioned vehicle attributes. Figure9 shows the simulation and optimization framework for prod-uct family revenue optimization in step III. Coupling equationscapture the effect ofRM andIE on fuel economyFE and ac-celerationAC (Suh 2005). Generally, an increase in vehicledimensions leads to poorer fuel economy due to increased dragas well as longer 50-70 [mph] acceleration times due to largerstructural mass, assuming a given powertrain.
Once all optimized attribute values and design variablevaluesXU veh are determined, the vehicle platform bandwidthof the product familyPveh is determined, both in the designspace and attribute space. Tables 4 and 5 list optimized values(normalized) ofXU veh andJU veh. They are normalized with re-spect to the maximum value of each variable among the threevehicle variants.
or no bandwidth is required for those design variables. Threeindependent variables -H5, L48, and Pw - require significantbandwidths. Variable Pw (weighted price across trim levels)will be used during the uncertainty analysis (Step VII) to cal-culate the overall product family profit3. The next task is toperform a sensitivity analysis of the optimum solution, whichwill identify additional design variables that might benefit fromflexibility, even if their initially required bandwidth is small.
4.4.2 Sensitivity Analysis The normalized sensi-tivity of variant p1’s revenue with respect to the product designvariable setXU veh is shown in Figure 10.
3Actual MSRP and transaction prices may not reflect this ‘optimal’ pricedue to discounts and other factors.
The chart shows the percent change in the revenue of ve-hicle variantp1 as a function of percent change in each designvariable. Note, that with the exception ofL48, the design vari-ables have negative sensitivity value. This means that whenthese design variable values increase, revenue decreases.Thereason is that as vehicle size increases to improveIE andRM, itdegradesFE andAC50−70 values, resulting in decreased mar-ket share and variant revenue. Analysis results show that thevehicle price P is the most sensitive parameter, representingthe price elasticity of demand. The most sensitive geometricaldesign variable isH50, the upper body opening to ground di-mension. It has a significant effect on total revenue, especiallyfor p1. Even though this particular dimension does not requireany differentiation initially (Table 4), incorporating flexibilityfor this particular dimension may be advantageous in the fu-ture. When the customers’ preferences change in the future(e.g., they want roomier cars), the firm may want to respond tothis uncertainty with greater ease.
4.5 Step IV: Identify Critical Elements
4.5.1 Selecting Flexibility Drivers Fig-ure 11 shows the independent design variables(L48,W3,W20,H5,H50) and differentiating constants(H122,W27). The figure shows two variants,p1 and p2 thatfeature differences in these values, which are most pronouncedin the geometry of the body (“greenhouse”) above the belt lineand the difference in wheelbase.
The upper variant in Fig. 11 shows a short wheelbase sedan(related toL48) with a box-like greenhouse (W27,H122) andhigher (easier) ingress/egress, while the second variant fea-tures a longer wheelbase, sportier look and lower ingress/egresspoint. H50 was shown to be a very sensitive dimension, and ifmade flexible, can potentially add value under future uncer-tainty. We select the four design variablesL48, W27, H122,andH50 as those that are the primary drivers of flexibility inthe BIW. Some of the flexibility is dictated by the initial band-width between variants, while forH50 the need for flexibilityarises out of sensitivity to potential future requirements.
4.5.2 Bounding the Physical Domain The chal-lenge of this step is the non-uniqueness of achieving flexibilityin the domain of physical elements. The identified design vari-ables can be mapped to the physical elements space in manyways, generating many non-unique solutions. To address thisproblem, the system architect must decompose the physicalsystem to bound the element space, thus constraining the phys-ical space under consideration.
Motor Compartment(Common)
Passenger Compartment(Flexible)
Rear Compartment(Common)
L48
H50H122
W27
DLDL
DLDL
Figure 12. Body in White (BIW) of a Passenger Sedan
The BIW of a passenger car is shown in Figure 12 witha high level system decomposition (motor compartment, pas-senger compartment, and cargo compartment). Since the keycustomer-preferred attributes,RM and IE, are attributes thatare directly related to the passenger compartment in additionto the styling aspect, the system designer must focus on thepassenger compartment to identify critical elements as candi-dates for incorporating flexibility. The motor compartmentandcargo compartment are assumed to be common.
Once the boundary of the “flexible” domain is established,components in the BIW structure need to be identified. In thisstudy, the BIW is decomposed down to its individual compo-nent level, at which individual components are end-items sup-plied to the BIW assembly line directly. The architecture ofthe steel body is a Body Frame Integral (BFI) structure with 21components that are spot welded together. These componentsare part of the passenger and rear cargo compartments (but donot include motor compartment components). Next, the con-nective relationship between individual components is estab-lished and expressed in design structure matrix (DSM) format
or as a network graph (Figure 14) based on the DSM.To achieve flexibility in the variables
L48, W27, H50,andH122, the product designer must (i)identify components that need to change, and (ii) determinehow such changes propagate through the BIW. The linksrepresent physical connections, where each component is con-nected to another by spot welding. There are four system-leveldesign variables,xi , mentioned in Section 4.4, that requiredifferentiation for each vehicle variant. Additionally, stylinguncertainty is a key factor that causes body-in-white changes tooccur. For each specified design variable change,∆xi, one mustidentify multipliers and carriers that send out changes to othercomponents when they themselves are changed. Once thesecomponents are identified, the system designer can (re)designthe system (BIW platform) to reduce the degree of changepropagation or switch costs by incorporating flexibility intothe multiplier/carrier components directly, or into componentsinto which secondary changes propagate.
4.5.3 Change Propagation Analysis: LengthChange As a result of the revenue optimization in Step III,it was determined that the vehicle platform must achieve band-width for length, represented byL48. Such BIW changes arerequired because of varying needs forRM, IE or styling asshown in Figure 11. We need to investigate cases in whichlength and styling requirements change in the future withintheoptimizedL48 bandwidth.
The change originates from the body outer panel, the out-ermost body component that is perceived by the customer andthe most important component for vehicle styling. The changesubsequently propagates throughout the BIW, and the finalchange propagation state is shown in the change propagationmatrix ∆E in Figure 13. This matrix and method of quantify-ing change propagation was explained in Section 3.5.
The change propagation matrix shows CPI values for allcomponents affected by the length change and classifies eachcomponent into four pre-defined classes, depending on thevalue of CPI. Ten components are affected by the lengthwisedirection change, initiated by a change inL48. We refer to thechange as∆L. Once these components are identified, then theswitch costs for making such a change needs to be calculated.Switch related investment costs for all components were calcu-lated using a process based cost model (Kirchain 2004). Theinvestment cost consists of blanking tool investment, stampingtool investment, and welding tool investment cost. Table 6 liststhe initial investment cost and BIW length related switch costsfor the ten identified components. The assumption is that - ini-tially - these BIW components are customized for each vehi-cle variant. This corresponds to the components designatedas“new-unique” in Figure 2. Costs in Table 6 are normalized withrespect to the initial investment cost of the body outer panel.
Figure 14 summarizes all change propagation related in-formation into a graphical network format. Above the name ofa particular component, its component class (for this particularchange) and related switch cost are displayed. Change propa-gation paths are shown as thicker arrows, components affectedby (this) change are shaded.
Once all critical BIW components and relevant switch
Figure 13. Change Propagation Matrix ∆E for BIW Length Change
Body Outer Panel (LH)
Body Outer Panel (RH)
Body Inner Panel (RH)Body Inner
Panel (LH)
Rocker Inner Panel (LH)
Rocker Inner Panel (RH)
Rear Wheel Housing (LH)
Rear Wheel Housing (RH)
FBHP (LH) FBHP (RH)
Center Pillar Support (LH)
Center Pillar Support (RH)
Dash Panel
Floor Pan
Rear Reinforcement A
Plenum Panel
Rear Reinforcement B
Roof Panel
Front Roof Support
Rear Roof Support
Rear Floor Pan
L L
M 100.0
M 134.3
Ca45.9
A 120.5
Ca 39.9
A 3.5
A 3.5
M: MultiplierCa: CarrierA: AbsorberC: Constant
M 100.0
M 134.3
Ca45.9
Legend
Figure 14. Change Propagation Network for BIW Length Change
costs are identified, this information is used to generate flex-ible BIW design alternatives in Step V. Other change scenarios(e.g. restyling of the greenhouse only) as in Suh (2005) canalso be analyzed in this fashion.
that are affected by the specified uncertainties and attributesthrough change propagation analysis. The task is to reduce
the magnitude of change propagation through flexible compo-nent design and in turn to reduce the economic impact of futurechanges on the system (platform).
4.6.1 Passenger Compartment DecompositionStrategy We developed the following decomposition strat-egy to make the BIW flexible to change. The passengercompartment is decomposed into three sub-compartments, asshown by the dashed decomposition lines (DL) in Figure 12.
Front Roof Support 3.5 3.5Rear Roof Support 3.5 3.5
The lower front passenger compartment remains common forall three vehicle variants. The lower rear passenger compart-ment must be flexible in order to accommodate the design vari-able bandwidth forL48. The upper passenger compartment,also known as the “greenhouse,” will be either unique or flex-ible for differentiation inW27, H122, and the overall vehiclestyling.
4.6.2 Single Component Decomposition: BodyOuter and Inner Panels The body outer panel is a criti-cal component that is visible to customers. It probably is themost sensitive component to styling changes. Figure 15 (top)shows how the component can be decomposed to incorporateflexibility.
Body Outer Panel � Lower (Common)
Body Outer Panel � Upper (Unique)
L48
H50H122
W27
Decomposition Line
Figure 15. Body Outer (top) and Inner (bottom) Panel Decomposition
for Flexibility
The lower body outer panel is made common for all threevehicle variants. The upper body outer panel is customizedfor each vehicle variant for styling differentiation, as well asfor the critical design variables differentiation as shown. Com-mon and unique portions of the body outer panel are weldedtogether to create the body outer panel for each vehicle vari-ant. The welding interfaces for all three vehicle variants are
also common. The proposed decomposition will incur extra in-vestment in blanking, stamping, and welding tools, but whenthe design changes, those changes will result in lower switchcosts. The body inner panel is also a multiplier and incurs highswitch cost whenever a change occurs. To reduce the impact ofchange, it is decomposed into three different pieces as shownin Figure 15 (bottom). These pieces must be designed to meetthe L48 bandwidth requirement while meeting the manufac-turing, strength, crashworthiness and quality requirements aswell. One way to achieve these requirements is to design theflexible piece to meet the long vehicle specification, and to trimthe end (where it is welded to the common piece) to producethe short wheel base variant. A cursory analysis of teardowninspections of newer vehicles from various manufacturers re-vealed that flexible BIW panel buildup starting from smallersub-panels is starting to occur in practice, particularly for innerpanel assemblies where welding lines are less of an issue.
4.6.3 Other Components Of the remaining sixcomponents identified as critical in Fig. 14 - the rocker innerpanel (RH & LH), floor pan, roof panel, front roof support andrear roof support - the roof panel is the only component thatmust be designed uniquely for each variant every time the de-sign changes since it must comply with the styling restrictionsimposed by the particular design change. In this case study,flexible designs used for the rocker inner panel, floor pan, andfront and rear roof support use the trimming strategy in whichthese components are designed for longer length specifications,and then trimmed down to meet shorter specifications (Suh etal. 2005b).
4.6.4 Flexible Assembly Process Assembly re-lated investment is perhaps the biggest cost driver during theinitial investment phase. In order to accommodate the flexiblecomponent designs proposed in the previous section, the BIWassembly line must also be flexible. Shown in Figure 16 is thesimplified BIW assembly process (based on the actual process)and the proposed areas to incorporate flexibility (shaded).
The motor compartment is common for all vehicle vari-ants. However, remaining downstream processes do requireflexibility in assembly tooling to accommodate different vehi-cle variants.
4.6.5 Vehicle Platform Element Selection As aresult of the system decomposition strategy, several compo-nents and assembly processes became “flexible” elements, asparts of the vehicle platform. Table 7 shows a platform ele-ment comparison between the inflexible BIW design and theflexible BIW design.
Note that in the inflexible BIW design, components andprocesses are divided into either common or unique elements.In the flexible BIW design, several unique elements are re-designed to become flexible elements as part of the platform.
4.7 Step VI: Determine Costs of Platform Alterna-tives
The next step is to determine the cost of the flexible BIWplatform design. For this case study, the process-based costmodel developed by (Busch and Field 1988; and Han et al.1993) is used to determine the capital investment and the unit(variable) cost of each vehicle. As mentioned in the previous
Unique Rocker Inner Panels Outer Panels - UpperElements Floor Pan Body Inr Pnl - Upper
Roof Panel Roof PanelRoof Supports
BIW Assembly Line
section, the architecture of the BIW is Body Frame Integral(BFI), using spot welded steel sheets as its material. Company-specific cost parameters were used for accurate cost calcula-tion, and numbers are reproduced here in a normalized format.
4.7.1 Investment and Unit Cost of Critical BIWComponents In Section 4.6, we generated a flexible BIWdesign (other alternatives could be generated). The costs of theflexible design and the original inflexible design, customizedfor each vehicle, need to be determined in order to comparebenefits and costs under future uncertainty. Table 8 shows,for each vehicle variant, the initial estimated annual produc-tion volume, expected volume trend and volatility, maximumexpected production volume during the life of the vehicle plat-form, and the number of required BIW assembly lines per par-ticular vehicle variant.
The number of required assembly lines is based on themaximum expected production volume during the lifetime ofeach vehicle variant built from the platform (15 years). In each
Maximum Demand 650,000 125,000 60,000BIW Lines Required 3 1 1
assembly line, a maximum of 225,000 BIW units can be assem-bled per year. Assembly lines with fixed tooling are dedicatedto only one vehicle variant while assembly lines with flexibletooling can accommodate all vehicle variants. The followingassumptions are made for determining relevant costs:
• The life of the vehicle platform is 15 years (three genera-tions of variants).
• From the analysis in Step IV and V, only ten componentsrequire differentiation while the other components remaincommon. For this study, only costs related to these tencomponents are calculated.
• Two design alternatives are considered: the inflexibleBIW design, in which ten differentiating components areuniquely customized for each vehicle variant and the flex-ible BIW according to Table 7. The assembly process forinflexible BIW design assumes fixed tooling, while theprocess for flexible design utilizes flexible tooling in theflexible (shaded) assembly sequences, as shown in Fig. 16.
• Fabrication and assembly tools are refurbished every fiveyears, at 25% of the new tooling cost.
• Once the initial investment costs and unit costs are deter-mined, they are assumed to be fixed for the remainder ofthe platform life.
For each design alternative, the initial capital investmentcost, refurbishing cost, and switch cost are calculated. Table 9lists normalized values of initial investment cost, refurbishingcost, and switch cost of inflexible and flexible BIW platformdesigns. Values are normalized to the initial investment cost ofcustomized (rigid) BIW designs.
The numbers indicate that the flexible BIW design, withflexible parts fabrication and assembly, requires approximately
34% more upfront investment than the inflexible BIW design.The inflexible BIW design is also more cost-efficient in termsof refurbishing costs. However, the flexible BIW design, withflexible assembly lines, outperforms the inflexible design interms of switch cost when the styling and the length of theBIW need to be changed (within the pre-defined bandwidth)in the future. This shows the costs and benefits of the flexibleBIW platform design; extra investment is required initially, butsubsequent changes can be accommodated with lower invest-ment costs. It is clear that the flexible BIW design is more ex-pensive to implement initially, but has the potential to performmore economically when the frequency of styling changes in-creases. Step VII of the flexible platform design process (recallFigure 3) - the uncertainty analysis - will help determine thosecases in which adding flexibility in the platform is worthwhileand those in which it is not.
4.7.2 BIW Unit Costs The total unit (variable) costsof BIW components for each vehicle variant were calculated.Total unit costs are calculated as a function of annual produc-tion volume, component mass (for fabrication), and the numberof spot welding points required (for assembly). Table 10 liststhe normalized unit BIW cost (ten components only) of eachvehicle variant for two different platform design alternativesbeing compared. Unit costs are normalized with respect to theunit cost ofp1 for the customized BIW.
Table 10. BIW Unit Cost of Vehicle Variants for Different Platform De-
Note that forp1 andp2, the unit cost for the flexible BIWdesign is higher, as expected, due to the higher investment costto amortize, and the additional welding costs for the flexiblecomponents. However, the unit BIW cost ofp3 for the flexibleBIW design is lower than the cost of the inflexible BIW de-sign. This is due to the effect of common component sharing inwhich the flexible BIW shares more common components withsmaller variants, thus lowering the unit cost through economiesof scale. Whether or not the additional cost of flexibility isben-eficial has not yet been determined.
4.8 Step VII: Uncertainty Analysis4.8.1 Problem Formulation In Step VI, all relevant
costs for the inflexible and flexible BIW platform design were
calculated. Costs include initial investment (Kinit), refurbishingcost (Kref), switch cost (Kswitch), and BIW unit cost. Using theidentified costs, uncertainty analysis can be performed to eval-uate the economic performance (profit) of each platform undervarious degrees of uncertainty. The following assumptionsaremade prior to uncertainty analysis:
• All costs are normalized to the initial investment cost ofthe inflexible BIW design (see Section 4.7).
• The time horizon is 15 years, corresponding to three gen-erations ofnominalvehicle variant redesigns.
• Fabrication and assembly tools are refurbished every fiveyears unless they are being replaced.
• Geometric Brownian Motion (GBM) is used for future de-mand prediction (α,σV ) (de Weck et al. 2004).
• The demand for individual vehicle variants is equal to theirproduction volume.
• The demand for individual vehicle variants cannot exceedthe maximum assembly line capacity set by the number ofassembly lines designated for each variant (for inflexibleBIW platform design).
• Flexible BIW platform manufacturing is also capacity lim-ited, even though the flexible tooling in all assembly linesenables flexible capacity utilization in that case.
• Styling changes and length changes occur within the de-sign variable bandwidths defined from the results of rev-enue optimization in Step III.
• When the styling changes, it is assumed that all three ve-hicle variants change together.
• To calculate the total vehicle family lifetime profit for eachdesign alternative, the net present value (discounted cashflow) method is used with an annual discount rate of 6%.
Table 8 lists demand forecast parameters for each variant,whereα is the demand trend coefficient, andσv is the demandvolatility coefficient. These parameters are calculated from ac-tual vehicle sales data (annual) between 1997 and 2003. Withinthe boundaries of the pre-stated assumptions, expected demandtrends, and volatility, the two BIW design alternatives areeval-uated and compared under several future scenarios (Table 11).
Scenarios I through IV are scenarios with varying degreesof uncertainty. Uncertainty analysis starts with investigation ofscenarios where only the production volume of the variants isuncertain. Styling change uncertainty is added to increasethedegree of uncertainty in scenarios III and IV, in addition toan-nual production volume uncertainty. Scenarios V through VIIIinvestigate instances where styling is changing above the ve-hicle belt line only but with a change frequency thatexceedsthe nominal 5 year life of individual variants, and under un-certain future demand. Scenarios IX through XII investigateinstances where the styling is changing in the lengthwise di-rection with increasing frequency but within theL48 bandwidthdefined from the optimization in Step III. Length changes re-sult in higher switch costs since more component changes arerequired.
The expected net present valueE[NPV] for the total prod-uct family is used to measure and compare the economic per-formance of each platform design alternative. The net present
NPV is the total sum of time discounted cash flow overa period of 15 years;CFt is the total cash flow at timet; r isthe discount rate;Ri,t is the revenue generated by sales of theith vehicle variant at timet; Ctotal,i,t is the total variable costincurred to produce theith variant;KInit,t is the investment thatoccurs at timet; Kref,t is the refurbishing related investment thatoccurs at timet; Kswitch,t is the switch-related investment thatoccurs at timet; Di,t is the demand for theith vehicle variantat timet; Pw,i is the weighted average price of theith vehiclevariant, obtained from Step III; andcveh,i is the unit cost ofthe ith vehicle variant. In this case study, since only the BIWof the vehicle is investigated, the unit cost of the BIW will beused as the unit costcveh. This is acceptable since our goalis not absolut profit forecasting, but a relative comparisonoftwo platform designs, one with embedded flexibility and onewithout.
4.8.2 Scenario I - XII Results Monte Carlo simu-lation is conducted to determine the range of future vehicle
demand and revenue. For each scenario (with the exceptionof Scenario I, in which no uncertainty is present), simulationconsists of 25,000 runs to represent a full range of outcomes.First, future demand uncertainty is simulated using GeometricBrownian Motion (GBM) and the initial demand, mean trendand volatility from Table 8 are used for the variants inPveh.Figure 17 shows a particular instantiation of a demand scenariofor variantp2.
Trend
Demand vs. Time
0
75000
100000
125000
150000
175000
1 9 17 25 33 41 49 57 65 73 81 89 97Time Period
Dem
and
E[Dt]GBM[Dt]
Figure 17. Geometric Brownian Motion (GBM) Simulation of Future
Demand for p2 (1 out of 25,000 runs).
Depending on the scenario (I-XII) the simulation capturesthe effects of implementing changes to vehicle variants at vary-ing time intervals. Scenarios III and V-VIII capture demandun-certainty (Fig. 17) plus styling changes above the belt lineonlyat an increasing frequency. In scenario III the styling changeis done every five years, while in scenario VIII we assume thatthe variant redesigns take place every year.
The second group of scenarios assumes that both vehi-cle length changes and styling changes occur together (seeFig. 14). In scenario IV this more invasive change is assumedto occur only every five years. In scenarios IX-XII the changefrequency is increased by one year at a time. Under what sce-narios will the inflexible or the flexible BIW yield a higherE[NPV]?
To answer this, an NPV analysis is conducted for eachchange scenario, BIW architecture and demand scenario. Fig-ure 18 shows the results of the Monte Carlo simulation forScenario VI in terms of the expected difference in NPV be-tween the flexible and inflexible BIW, E[∆NPV]. The distribu-tion of 25,000 runs converges to a probability density functionwith lower and upper bounds of 5.00 and 14.00 and a mean ofE[∆NPV]=9.1. This means that the flexible BIW platform issuperior to the rigid BIW platform in this scenario.
Table 12 summarizes the normalized profit difference forall the scenarios described in Table 11. Table 12 also shows thetotal expected lifetime profit asNPV for each design over thelife of the product platform4.
In Scenarios I - IV, the inflexible BIW design performedbetter than the flexible BIW platform. Even for Scenario IV,where the uncertainty is greatest among the first four scenar-ios, the inflexible BIW design outperformed the flexible BIWdesign. Results suggest that under these circumstances, this
4Note that the revenue for the entire vehicle is taken into account in the NPVcalculations, but that the costs only capture the components and assembly ofthe BIW.
Figure 18. Monte Carlo Simulation Frequency Histogram based on a
normalized NPV
particular flexible BIW platform should not be implemented asthe higher investment in flexible elements and assembly equip-ment (Table 9) is never amortized over the 15 year time hori-zon. However, when the frequency of styling change increases,the results are different.
In Scenarios V - VIII, styling for all vehicle variants ischanged more frequently. In these scenarios, styling is changedabove the vehicle belt line only (no length change). The ratio-nale for increasing styling change frequency is that there mightbe a situation in which, to maintain current demand levelsunder competition, the company must change vehicle stylingmore frequently to refresh the product family. Mean lifetimeNPV for each platform design alternative is calculated, basedon the Monte Carlo simulation described above. Results areshown in the middle of Table 12. As the frequency of stylingchange increases, the profit difference between the inflexibleBIW design and the flexible BIW design initially decreases.The crossover point occurs when the styling change frequencyincreases from every four years to every three years. Whenthe styling changes every three years (Scenario VI) or morefrequently, the flexible BIW design outperforms the inflexibleBIW design in terms of totalNPV. This is due to the switchcost incurred every time the styling of the variants - built fromthe shared platform - changes. TotalNPV for the flexible BIWdesign does not decrease as rapidly as that for the inflexibleBIW design as changes are made more frequently. This is dueto the lower switch cost of the flexible BIW design (Table 9),making it more profitable under increasing uncertainty due tomarket dynamics (see also Fig.1).
Scenarios IX through XII evaluate situations in whichstyling changes also require a vehicle length change withintheestablishedL48 bandwidth from the optimization in Step III.Since there are more components and tooling that require mod-ifications when the vehicle length changes (10 versus 7 com-
ponents when the ‘greenhouse’ only changes), switch costs forboth designs are higher. However, due to its significantly lowerswitch cost, the flexible BIW has better economic performanceonce styling and length changes occur every four years (Sce-nario IX) or faster.
4.9 DiscussionEvaluating two different BIW platform designs under
scenarios of varying uncertainty produced interesting results.When uncertainty was not present, or very small, the inflexi-ble BIW design performed better. However, as the degree ofuncertainty increased, the expectedNPV difference betweenthe two designs decreased, and at a certain point, the flexibleBIW platform design started to show higher expected NPV. Thereason is that the magnitude of switch costs for the inflexibleBIW design is much higher than for the flexible BIW design,and when the frequency of design change increased, the flex-ible BIW design outperformed the inflexible platform design.The results suggest that, under uncertain styling change fre-quency and uncertain vehicle family demand, it is beneficialto implement the flexible BIW platform design if styling mustchange every three years or less, or if styling and length changetogether every four years or less. While the actual particular-ities of future geometry changes are subject to styling trendsand represent exogenous uncertainty (since vehicles are not de-signed ten years ahead of their release), the frequency of thestyling change is a controlled decision variable that can bede-cided by the firm’s management. Given this situation, Table 12offers decision makers a useful quantitative guideline formak-ing a decision on whether or not flexibility should be embeddedinto the BIW platform.
5 SUMMARY AND DISCUSSION5.1 Summary
This paper introduces an end-to-end design process forflexible product platforms (Fig. 3). The framework is appliedto a vehicle platform case study. In the case study, the platformis designed to accommodate three vehicle variants while beingflexible to uncertain future demand and specification changes.
The process is general and consists of seven steps, usinga combination of quantitative analysis and expert engineeringknowledge for each step. First, uncertainties are identified (stepI) and mapped to quantifiable vehicle attributes (step II), then tocritical system-level design variables which require bandwidthand/or are sensitive to the aforementioned attributes (step III).Once the bandwidths for the system-level design variables aredetermined through product family revenue optimization, crit-ical product platform components are identified using change
propagation analysis (step IV). Flexible design alternatives aregenerated for critical components (step V) in order to reducechange propagation and lower switch costs. The cost of flexi-ble design, both in component fabrication and in the assemblyprocesses, are calculated using a process based cost model (stepVI). Uncertainty analysis is performed to determine the eco-nomic performance of both the inflexible and flexible platformalternatives (step VII) under a set of uncertain future scenarios.
We believe that step IV, identifying critical platform ele-ments as candidates for embedding flexibility, is the most criti-cal and difficult step. This is were the main contribution of thispaper lies. Much of the literature on platform design and opti-mization is satisfied with setting the values of design variablesxi to be common (up to and including step III). If, however theproduct is complex and the common design variables do notmap to the same physical elements for reuse, such purely para-metric commonality is of little real benefit. By decomposingthe physical product in depth and performing change propaga-tion analysis we reveal those components that act as changemultipliers (CPI > 0) or whose switch cost is significant. Em-bedding flexibility in those components requires initial invest-ment in design, tooling and assembly equipment and is akin totaking out a “real option” on the platform design.
A cost comparison in our case study showed that the flexi-ble platform design will cost 34% more to implement initially,but will incur significantly lower switch cost when the vehicledesign changes. Higher investment also affects BIW unit costs,resulting in higher unit cost for flexible BIW design for somevariants. However, other vehicle variants benefited from com-mon component sharing, resulting in lower unit cost relative tothe inflexible BIW design based variants.
5.2 Critical DiscussionThe case study demonstrated that a critical subset of flex-
ible BIW platform components allowed the whole BIW to be-come flexible to respond to the uncertainties defined at the be-ginning of the design process. Ten (10) out of twentyone (21)BIW components together with the flexible assembly process(Fig. 16) made the BIW flexible to future styling and lengthchanges, while remaining economically profitable in terms oftotal expectedNPV.
Another important outcome of the paper is that the resultsquantitatively demonstrate the increasing value of flexibility asuncertainty increases. This fact is already well known in op-tions analysis, but the main contribution of the paper lies in amethodology to identify (i) where in a complex, coupled sys-tem or product to embed flexibility and (ii) how to value thatflexibility in the context of realistic scenarios.
To do so required a number of simplifying assumptionsthat had a significant impact on the results:
• In step I we choseIE andRM as well as variant demand asthe driving uncertainties for which flexibility was embed-ded. We substantiated this choice by empirical considera-tion of demand variations across market segments (Fig. 1)as well as historical trends of key product attributes (Ta-ble 2). As shown in this paper, flexibility can only be ben-eficial if the “right” uncertainties were selected in the firstplace and we acknowledge that the method developed here
does not help product platforms deal with wholly unknownuncertainties or provide insurance against all future even-tualities.
• In step III we established the “optimal” bandwidth of theproduct platform across a set of variantspi . In order to doso required setting lower and upper bounds of the system-level variables. The setting of these bounds was informedby the minimum and maximum occurrences in the con-stituent market segments (recall Fig. 4). However, marketsegment boundaries are always arbitrary and fluid so thatsetting different bounds might lead to different platformbandwidths. This relates to the question of platform extentand when it is beneficial to split a single large platforminto two or more smaller platforms; a question which hasbeen discussed elsewhere (Seepersad et al. 2001, de Weck2005).
• A critical analysis of step VI reveals that the BIW onlyencompasses a relatively small percentage of total vehiclecost (typically< 20%). So, there remains vast potential forplatforming and embedding flexibility in other parts of thesystem such as the powertrain, or the electrical equipmentand software. The cost contribution of the latter subsys-tems has been steadily increasing in recent years.
• Throughout the paper we have tacitely assumed that abasic product platform architecture already exists (seeFig. 12) and that flexibility is achieved by redesigning thatexisting platform. There might be other, more efficientways of deriving a flexible platform architecturede novo.
• In the change propagation analysis (Section 4.5) we as-sumed that the CPI and switch costs were independent ofthe amount of change (e.g.∆L48 could be 1”, 5”, 10”,...)as long as the change was within the allowed bandwidth.In reality the number of affected components and switchcosts will not only depend on the type of change, but alsoon its magnitude.
5.3 Future Work
Throughout this work we assumed that all variants wouldbe built from a single common platform. However, this maynot be true in some cases where the differences between vari-ants are too great. In those situations, multiple platformsmaybe required, see work on this topic by Seepersad et al. (2000,2001) and more recently by Suh et al. (2003) and de Weck(2005). Future work will include determination of “splitting”criteria which will tell system designers in which cases a plat-form has been “stretched” too far and should therefore be splitinto separate platforms. One of the difficulties with this inprac-tice is that the true bandwidth of a platform can often only beestablished via testing of physical prototypes in the field.
Change propagation analysis was presented for an auto-motive body-in-white in this paper. While involved in its ownright, the change propagation was relatively straightforward,given that only changes in geometry were taken into account.Future work will include analysis of change propagation incomplex products were changes can potentially “jump” acrosssubsystem boundaries and are not simply transmitted to di-rectly neighboring components.
ACKNOWLEDGMENTThis research was supported by General Motors under
R&D grant 6894615. Dr. David Chang, Chief Scientist MathTools, served as project technical monitor. Mrs. Jennifer Craigand Mr. Michael Mack at MIT helped in proofing the manu-script. Dr. Sangdon Lee at GM provided the data and prin-cipal components analysis to extract the key dimensions forIngress/Egress and Roominess. Moreover, Dr. Zhihong Zhang,Dr. Doo-Yearn Jo, Ty Bollinger, Randy Urbance, and Dr. PeterFenyes at General Motors provided specific data and generalguidance for the automotive platform case study. Dr. DanielWhitney, as well as Prof. Christoper Magee and Prof. DavidWallace provided input and served on Dr. Suh’s doctoral com-mittee. The assistance of all individuals named and unnamedwho supported this work is gratefully acknowledged.
REFERENCESConjoint Analysis: A Guide for Designing and Interpreting
Conjoint Studies (1992) American Marketing AssociationMotor Vehicle Dimensions (2001) Society of Automotive En-
gineersBremmer R (1999) Cutting-Edge Platforms, Financial Times
Automotive World, September, pp 30-38Busch J, Field III F (1988) Blow Molding Handbook, Hansr
Price, Product and Organization. Chapman & Hallde Neufville, R et al.(2004) Uncertainty Management for En-
gineering Systems Planning and Design, The Second Engi-neering Systems Symposium, MIT, Cambridge, MA, March
de Weck O, de Neufville R, Chaize M (2004) Staged Deploy-ment of Communications Satellite Constellations in LowEarth Orbit, Journal of Aerospace Computing, Information,and Communication, 1, 119-136, March
Dunteman G (1989) Principal Component Analysis, Sage Pub-lications
Eckert C, Clarkson P, Zanker W (2004) Change and Customi-sation in Complex Engineering Domains, Research in Engi-neering Design, 15(1):1-21
Eppinger S, Whitney D, Smith R (1994) A Model-BasedMethod for Organizing Tasks in Product Development, Re-search in Engineering Design, 6(1):1-13.
Gonzalez-Zugasti J, Otto K, Baker J (2000) A Method for Ar-chitecting Product Platforms, Research in Engineering De-sign, 12(2):61-72
Gonzalez-Zugasti J, Otto K, Baker J (2001) Assessing Value inPlatformed Product Family Design, Research in EngineeringDesign, 13(1):30-41
Han H, Chen A, Clark J, Field III, F (1993) Material DesignSensitive Costing of the Body-in-White, Proceeding of theInternational Body Engineering, Detroit, MI, September 21-23
Hauser J. R. and Clausing D. (1988), The house of quality. Har-vard Business Review, 66(3):63(73)
Hull, J (2002) Options, Futures, and Other Derivative Securi-ties, Prentice Hall
Jajuga, K, Sokolowski, A, Bock, H (2002) Classification, Clus-tering and Data Analysis, Springer
Kirchain R (2004) Cost Modeling of Materials and Manufac-
turing Processes, Encyclopedia of Materials: Science andTechnology, pp. 1718-1727
Li H, Azarm S (2002) An Approach for Product Line De-sign Selection under Uncertainty and Competition, Journalof Mechanical Design 124(3):385-392
Martin M, Ishii K (2002) Design for Variety: Developing Stan-dardized and Modularized Product Platform Architecture,Research in Engineering Design 13(4):213-235.
Meyer M, Lehnerd A (1997) The Power of Product Platforms:Building Value and Cost Leadership, Free Press, New York
Moses, J. et al., ESD Committee (2002) ESD SymposiumCommittee Overview, The ESD Internal Symposium, MIT,Cambridge, MA
Narayanan S, Azarm S (1999) On Improving MultiobjectiveGenetic Algorithms for Design Optimization, Structural Op-timization, 18:146-155
Pahl G, Beitz W (1996) Engineering Design: A Systematic Ap-proach, 2nd Edition, Springer-Verlag
Papalambros P, Wilde D, Principles of Optimal Design: Mod-eling and Computation, Cambridge University Press, 2000
Pine J (1993) Mass Customization: The New Frontier in Busi-ness Competition, Harvard Business School Press
Sanderson S, Uzumeri M. (1997) The Innovation Imperative:Strategies for Manufacturing Product Models and Families,Irwin Professional Publisher
Seepersad C, Hernandez G, Allen J (2000) A Quantitative Ap-proach for Determining Product Platform Extent, ASME In-ternational Design Engineering Technical Conference, Bal-timore, Maryland, September, DETC2000/DAC-14288
Seepersad C, Mistree F, Allen J (2002) A Quantitative Ap-proach for Designing Multiple Product Platforms for anEvolving Portfolio of Products, ASME International DesignEngineering Technical Conference, Montreal, Canada, Sep-tember, DETC2002/DAC-34096
Sheffi J (2005) The Resilient Enterprise: Overcoming Vulner-ability for Competitive Advantage, MIT Press
Simmons, G., ”U.S. Auto Market Splinters”, The DetroitNews, January 31, 2005
Simpson T, Maier J, Mistree F (2001) Product Platform De-sign: Method and Application, Research in Engineering De-sign 13(1):2-22
Simpson, T. W., Siddique, Z. and Jiao, J., Eds., (2006) ProductPlatform and Product Family Design: Methods and Applica-tions, Springer, New York
Suh, E. S., (2005) Flexible Product Platforms, PhD. Disser-tation, Massachusetts Institute of Technology, EngineeringSystems Division
Suh, E.S., de Weck O.L., Kim I.Y. and Chang D., FlexiblePlatform Component Design under Uncertainty, Journal ofIntelligent Manufacturing , Special Issue on Product FamilyDesign and Development, in Press
Ulrich, K, Eppinger, K (1999) Product Design and Develop-ment, McGraw-Hill
Womack J, Jones D, Roos D (1991) The Machine ThatChanged World, Harper-Collins Publishers
![SYSTEMATIC DESIGN METHOD FOR INFORMATION …eislab.gatech.edu/pubs/conferences/2003-asme-detc-zeng-pb/DETC... · Chamberlain and coauthors [4]. Our course project is in part the motivation](https://static.documents.pub/doc/80x56/5b5d160b7f8b9ac6028d8da6/systematic-design-method-for-information-chamberlain-and-coauthors-4-our.jpg)