Reprint from ABB Review 2/2004 New levels of performance for the cement industry The cement industry, like most other industries, is under pressure to increase profit and margins while ensuring sus- tainable and environmentally friendly use of natural resources. This puts the onus on plant owners to develop new strat- egies that will support a quick, optimized response to changing conditions, often involving complex scenarios with con- flicting goals. ABB has developed new modules and algorithms aimed at solving these crucial customer issues.
Transcript
Reprint from ABB Review 2/2004
New levels of performance for the cement industry
The cement industry, like most other
industries, is under pressure to increase
profit and margins while ensuring sus-
tainable and environmentally friendly use
of natural resources. This puts the onus
on plant owners to develop new strat-
egies that will support a quick, optimized
response to changing conditions, often
involving complex scenarios with con-
flicting goals. ABB has developed new
modules and algorithms aimed at solving
these crucial customer issues.
2
Data flow envisioned for economic process optimization 1
Market data
(fuels, alternative fuelsand raw meal costs,
clinker price, emissions costs,
customer orders, etc.)
Online plant data
(energy requirements,feed rate constraints, material availability,
etc.)
Plant targets
(clinker chemistry,emission limits, other process parameters)
Fuels and raw meal properties
(chemical composition, ash content,
calorific value, etc.)
Optimal fuel mix
(maximum profit, strict constraints satisfaction, process stability)
Mathematical models +
Optimization techniques
To implement effective optimizationstrategies, tools are needed that
enable cement plants to perform at theiroptimum economic level within thegiven technological, environmental andcontractual constraints. They must workin real time, and the information theyprocess must be consistent and alwayscorrect. It goes almost without sayingthat an efficient information manage-ment system is a precondition for this.
Two areas for which ABB has alreadydeveloped such tools are thermal energymanagement and electrical energy man-agement. The first module, which isbased on dedicated mathematical mod-els and state-of-the art optimization tech-niques, computes the lowest-cost fuelmix that satisfies the current process andmarket constraints. It uses real-timeinformation from sources that includelaboratory analyses, market prices, fore-casts of alternative fuel availability, envi-ronmental constraints and process con-ditions. The second module is a sched-uling tool that enables cement mills andsilos to be operated in such a way thatthe production goals are reached at thelowest possible energy cost.
The software modules can be used indecision supportmode or inclosed-loop con-trol mode. Cus-tomers benefitnot only froman immediatereduction intheir energy billsbut also frommore stable operation, leading to higherquality and lower maintenance costs.
The vision. . .Intelligent, flexible process controlsystems, constantly performing at thehighest level and able to coordinatetheir actions, are the key to successfulstrategies for meeting profitability andsustainability goals. A number of tech-nological advances have made suchsystems possible today:
The superior performance offered byrelatively low-cost hardware (PCs,digital buses, etc).
New software that permits seamlessinterconnection and operation ofdiverse systems.Advanced control technology. Com-bined with the computational powernow available, this has opened upcompletely new possibilities in plantprocess control, scheduling and plan-
ning, leading totrue economicprocess opti-mization.
ABB’s Indus-trialIT initiativeis instrumentalin the realiza-tion of this
vision. Industrial IT enabled solutionsnot only control and optimize thedifferent parts of the process but alsomake it possible for these parts tocommunicate and cooperate with eachother for overall performance opti-mization.
. . .and its realizationIndustrial IT (IIT) is not only a newtechnology; it is a new way of doingbusiness. IIT seamlessly integrates acompany’s distributed industrial andbusiness assets to make them more ver-satile, more efficient and more prof-
itable. Two outstanding characteristicscontribute to this:
Real-time integration of informationsystems and automation systemsacross the enterprise.Economic process optimization, madepossible by integration of the enter-prise systems.
Having process-wide, real-time informa-tion available lets companies:
Monitor all parts and their specificcharacteristics.Know how parts interact with eachother. Diagnose problems instantly.Prevent serious downtime.Track productivity.Trouble-shoot.Optimize processes.Continuously coordinate every part ofthe process according to key per-formance indicators (KPIs).
While this might seem like a ‘wish list’,it is absolutely realistic. A closer look athow IIT is being instantiated in thecement industry shows why.
From raw data to useful informationTo provide the most complete picture ofa cement plant’s performance, data frommany different sources have to be col-
Optimization strategiesdepend on tools thatenable plants to perform at their optimum economiclevel within the givenconstraints.
3
lected and evaluated. However, herethere is a problem: The raw data fromthese sources are often inconsistent be-cause of the limited accuracy of the in-struments used. And some key data maynot be available at all if it has not beenpossible in the past to justify the cost ofinstalling the necessary sensors.
The problem can be overcome by tak-ing an integrated approach to datamanagement. The data flow betweendepartments and data sources, etc, is
then automated, data inconsistenciesare corrected (mathematical methodsplay a central role here), and all infor-mation is made available, in real time,to authorized users.
An integrated data management systemrelies on data from different sources
being available. The system can usuallycollect these data directly from the sub-systems already installed in many plantsfor recording process data. An ideal datamanagement system would therefore beable to take data straight off sensors,process control systems and other datacollectors, like historians. Obviously, thekey here is to have built-in flexibility, sothat the systems already installed can beused.
ABB has had a data management systemwith these characteristics on the marketfor some time [1]. It combines all thenecessary technical features and archi-tecture with ease of use, thin client im-plementation and Microsoft WindowsTM
‘look and feel.’
From information to decisionsEconomic process optimization will beone of the key factors driving the cementindustry in the future, largely because ofhow production efficiency can be in-creased through the timely, optimal useof resources.
Model-based strategies could not be usedextensively in the past for on-line com-putation of process targets. This was dueto the inherent complexity of industrialprocesses, which resulted in mathemati-
Kiln model inputs and outputs2
Ambient conditions
TF/AF/RM rates
TF/AF/RM chemistry
TF/AF/RM physical properties
Mathematicalmodelof kiln
Volatiles recirculation
Clinker chemistry
Emission rates
Heat release
Kiln temperature profile
AF Alternative fuelsRM Raw mealTF Traditional fuels
IndustrialIT seam-lessly integrates acompany's distrib-uted assets to makethem more versatile,more efficient andmore profitable.
cal problems that could not be solved ina reasonable amount of time. Poormodel accuracy was another problem,and meant that the value of the setpointswas also open to question. Obviously,these problems have to be solved first.
The way forward
Three recent technological develop-ments have made it possible to over-come these restrictions:
Modern mathematical modeling tech-niques provide a uniform frameworkfor efficient modeling of complexindustrial systems.Advanced control and optimizationtechniques have become matureenough to be used to drive processand business decisions.Mixed-integer and non-linear pro-gramming technology offers an effi-cient and robust means of solving thedescribed problems.
shows a data flow diagram for thekind of solution envisioned here.
There are several ways in which optimalsolutions of the described problems canbe approximated. One interesting, andwidely adopted, approach to solvingcontrol problems involving systemswhich are subject to input and output
1
4
constraints is model predictive control(MPC) [2].
MPC is based on the so-called recedinghorizon philosophy, ie, a sequence offuture optimal control actions is chosenaccording to a prediction of the short- tomedium-term evolution of the systemduring a given time. When measure-ments or new information become avail-able, a new sequence is computed whichthen replaces the previous one. The ob-jectives of each new sequence run arethe optimization of performance andprotection of the system from constraintviolations. The latter objective introducesan important issue – the efficient han-dling and satisfaction of the problemconstraints. To this end, it is convenientto model the plant using the MixedLogical Dynamic (MLD) framework [3].This does, however, involve relativelycomplex mathematical techniques.
The issue of model tuning and adapta-tion also has to be solved. Indeed,model-based control relies on the abil-ity of the models to represent the realplant to a certain degree of accuracy.ABB ensures the correctness of thisassumption through the use of parame-ter identification techniques. Subspaceidentification, Kalman filtering, andneural networks are used extensively toidentify model characteristics (para-meters, etc), estimate non-measurablephysical values (volatility coefficients,etc), and forecast boundary conditionssuch as power prices, fuel availabilityand market constraints .
Together, these mathematical techniquesprovide a comprehensive and flexibletoolbox for tackling the overall eco-nomic optimization of industrial plants.
ABB’s IndustrialIT vision for thecement industry
Thermal energy management:
kiln fuel mix optimization
ABB already offers the cement industrya state-of-the art product for stabilizingand optimizing kiln operation. CalledExpert Optimizer, it builds on the excel-lence of its predecessor, LinkmanGraphics. The Expert Optimizer system
2
Implementation of the new algorithms for kiln fuel mix optimization3
1000
900
800
700
600
500
400
Tons
0 20 40 60 80 100 120 140 160 180
Time [h]
Silo 1
Cement grinding plant scheduling: silo levels4
1000
900
800
700
600
500
400
Tons
0 20 40 60 80 100 120 140 160 180
Time [h]
Silo 2
Reference state Measured state
5
combines rules-based control with mod-ern tools like neural networks and fuzzycontrol.
In addition to Expert Optimizer, ABB’scement portfolio is now being enhancedwith an Alternative Fuels Optimizationmodule. Developed to meet the indus-try’s need for a tool that will allow opti-mal management of alternative as wellas traditional kiln fuels, this module cansignificantly enhance the economic per-formance of kilns.
The module uses data gathered by thevarious information management sys-tems to calculate online the lowest-costfuel mix able to satisfy the process andbusiness constraints. These can benumerous, and may include the heatbalance, excess oxygen level, clinkerchemistry, volatiles concentration, emis-sion limits, actuator speed change, oper-ative constraints on fuel consumption,and contractual conditions.
A dedicated mathematical model, devel-oped in Matlab/SimulinkTM, is used toimplement the (model predictive) con-troller. This model can estimate cooler,flame, burning zone, back-end and pre-heater temperatures, kiln energy require-ments, emission and volatiles levels, etc.The model parameters are tuned using acombination of neural networks andKalman filtering techniques. The opti-mization algorithms are able to copewith both hard and soft constraints,which considerably enhances the relia-bility of the optimization process.
The input data are updated at samplingtimes of 15 to 30 minutes, and new pro-cess setpoints are computed and passedto the Expert Optimizer strategy for im-plementation. Between samplings, the‘standard’ Expert Optimizer strategy guar-antees process stability and highest per-formance. In particular, this strategy en-forces an economically optimal responseto changing conditions in fuel, waste andraw meal quality, as well as ensuringstrict satisfaction of the environmental,contractual and technical constraints.
A prototype implementation of thisalgorithm is currently being tested at a
Swiss cement plant, where it is improv-ing operating profits .
Electrical energy management:
cement grinding plant scheduling
In the final stage of cement manufactur-ing, clinker is ground with additives.The differentcement types, orgrades, are de-fined by theirchemical com-position andparticle size.Grinding takesplace in hugemills, where ro-tating steel balls crush the material untilthe required grain size distribution isreached. The produced cement is then
3
conveyed to different silos according tograde before being packaged andshipped to customers.
Cement mill scheduling, ie decidingwhen to produce a certain cement gradeand in which mill, is currently performed
manually, us-ing heuristicrules and rely-ing on opera-tor experience.However, thenumerousmills, gradesand silos, plusthe various
operating constraints, make the problema complex one. Too often, the operator’schoices are far from optimal.
This solution provides a powerful tool for monitoringand adapting organizations to changing micro and macro needs.
Inner-loop MPC response: on/off control sequence5
0 20 40 60 80 100 120 140 160
Hour
G1 G2 G3 G4 G5
a)
b)
0 20 40 60 80 100 120 140 160
Hour
G1 G2 G3 G5
c)
d)
Mill 1 Mill 2a) MPC weekly scheduling c) MPC weekly scheduling b) Reference scheduling d) Reference scheduling
6
Grinding plant scheduler – graphic user interface7
Based on customer orders andenergy price forecasts, anouter-loop MPC is executed atleast once a week and its out-put used as a reference sched-ule for mill operation. Here,the cost functional representscosts associated with electric-ity consumption and theamount of low-grade cementproduced (cement producedduring the switch from onegrade to another). Electricitycost reduction is achieved bycommitting the production totime periods when the tariffsare lower, and by managingthe mills in such a way thatcontracted thresholds of maxi-mum electrical power are notexceeded. Reductions in low-grade cement are obtained bypenalizing the number of pro-duction switches. The costfunctional also includes com-ponents related to soft constraints.
Unplanned events, such as componentfailures or unexpected sales are, how-ever, frequent, and an inner-loop MPCis used to react to these disturbances. In this phase, the state variables are thesilo levels, and the control variables arethe switching commands to the mills.The cost functional is a weighted sumof deviations from the values given bythe outer-loop MPC reference schedule.The typical sampling time is one hour.
Apart from the physical constraints im-posed by the silo capacity and millavailability, there are several other con-straints to consider:
Transition time: Grade changes cancause delays, during which the mill’soutput is conveyed to a special silo.Order fulfillment: The optimizationalgorithm requires sales forecasts forevery grade as input. If the sales fore-cast cannot be completely fulfilled,the algorithm will choose the gradeto be produced first according to agiven ranking.Conveyor belts: Possible constraintshere include the possibility that theremight be three mills but just two inde-pendent conveyor belts. (Multiple
mills can simultaneously discharge thesame cement grade to the same con-veyor belt. However, belts can serveonly one silo at a time and silos canbe served by only one belt at a time.)
Silo content: Only one ce-ment grade can be stored in agiven silo.
An example of reschedulingcapability is shown in . Theblue line represents the pre-computed reference levels forsilo 1 and silo 2, while the redline shows the actual silo levelsmeasured by online sensors.The deviation from the refer-ence in silo 1 at the 57-hourmark is caused by the sale ofcement exceeding the forecast.
The inner-loop MPC responsefor mills 1 and 2 is shown in
. It can be seen how theinner-loop MPC reacts imme-diately to the deviation, com-mitting both mills at time t = 58 to the production ofgrade-one cement.
Two more things need to be specified tocomplete the schedule: the actual silo towhich the cement grade is to be trans-ported and the actual belt to be used. Inaddition, it has to be decided from
5
41400
1200
1000
800
600
400
200
0
Tons
0 5 10 15 20 25
Time [h]
FLUVIO 5 - Silo levels
Silo levels for the one-day scheduling scenario shownin the table opposite
6
Silo 1 Silo 8 Silo 10Silo 13 Silo 17
7
References
[1] C. Colbert, M. Mound: Seamless quality and process information integration with collaborative laboratory capability.
International Cement Review, November 2003.
[2] E. Camacho, C. Bordons: : Model Predictive Control. Publ: Springer Verlag, 2000.
[3] A. Bemporad, M. Morari: Control of systems integrating logic, dynamics and constraints.
Automatica: Special issue on Hybrid Systems, vol 35, no 3, 407–427.
[4] E. Gallestey, et al: Using model predictive control and hybrid systems for optimal scheduling of industrial processes.
Automatisierungstechnik, 2003, vol 51, no 6, p 285.
One-day schedule (simulated results for a cement plant in Switzerland)
Mill 1 Mill 2 Mill 3Time
Cement Belt Silo
FLU5 B 1FLU5 B 8FLU5 A 8FLU5 B 10FLU5 C 13FLU5 B 17FLU5 B 17FLU5 B 17FLU5 B 17FLU5 B 17FLU5 B 17FLU5 B 17FLU5 B 17FLU5 B 1FLU5 B 1FLU5 B 1FLU5 B 1FLU5 B 1FLU5 b 1FLU5 B 1FLU5 B 1FLU5 B 1FLU5 B 1FLU5 B 1
Cement Belt Silo
FLU5 B 1FLU5 B 8FLU5 B 10FLU5 B 10FLU5 B 10FLU5 B 17FLU5 B 17NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 11NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 2NOR4 C 5NOR4 C 5NOR4 C 5NOR4 C 5
Cement Belt Silo
– – –– – –– – –– – –– – –
FOR5 A 3– – –– – –– – –– – –– – –– – –– – –– – –
FOR5 A 3FOR5 A 3
– – –FOR5 A 3FOR5 A 3FOR5 A 3FOR5 A 3FOR5 A 3FOR5 A 3
– – –
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
which silo the cement will be taken tofulfill a given customer order. The table(this page) shows, as an example, thesimulated results for a cement plant inSwitzerland. As can be seen, mill 3 islargely idle. The silos are ranked in theorder in which they are to be filled andemptied. In the case of grade FLU5, forexample, silo 1 has the highest ranking,followed by silos 8, 10, 13 and 17. Fromthe table and the silo levels shown in ,it can be seen that FLU5 is first con-veyed to silo 1. This silo is filled to itslimit (640 tons) at time t = 1, after whichthe product is diverted to other siloswith a lower priority. Sales, which beginat t = 10, result in silo 1 being emptiedfirst, followed by silo 8. Immediatelyafterwards, the conveyors begin trans-porting the cement again to the highest-ranked silo – silo 1.
shows the main graphic user inter-face for controlling the schedulingmodules.
Where customers benefitThe direct benefits of the described so-lution are estimated to be a saving of upto 5% in thermal and electrical energy,a reduction in low-grade cement, morestable operation, consistent clinker qual-
7
6
ity, strict emissions control, and lowermaintenance costs.
Other less tangible, although not lessimportant, benefits are the total integra-tion of vital process data for bothclosed-loop and open-loop optimiza-tion, active KPI monitoring, the singleprocess control interface, and the foun-dation laid down for developing value-added applications that address cus-tomers’ needs.
This solution provides cement plantmanagers with a powerful tool, capableof monitoring their organizations andadapting to changing micro and macroneeds. By implementing it, customers
can expect lower costs, a consistentlyhigher quality, environmentally soundprocesses, fast payback and a largerreturn on investments.
![Cement Industry[1]](https://static.documents.pub/doc/80x56/577d34d91a28ab3a6b8f02d9/cement-industry1.jpg)
