A decision support system for simulation modeling

A decision support system for simulation modeling

Journal of Manufacturing Systems Volume 10/No. 6 A Decision Support System for Simulation Modeling Jorge Haddock, Rensselaer Polytechnic Institute, T...

694KB Sizes 3 Downloads 156 Views

Journal of Manufacturing Systems Volume 10/No. 6

A Decision Support System for Simulation Modeling Jorge Haddock, Rensselaer Polytechnic Institute, Troy, NY Nitin Seshadri, Seshadri Consulting Group, Madras, India V.R. Srivatsan, Oracle Corporation, Redwood City, CA

Abstract

the issues involved in the design of the cell are facilitated by effective use of the decision support system.

This paper presents a decision support system for the simulation of a flexible manufacturing cell. The design, construction, development, and structure of the system will be discussed, and a few of its features described. This paper also presents an overview of some popular methods used to develop userindependent simulations and discusses the developed modeling methodologies in this context. Different system configurations are compared using relevant performance measures. Following a comprehensive analysis of the results, we make specific recommendations regarding appropriate courses of action.

Developing UserIndependent Simulations The increasing popularity and use of simulation as a technique for modeling and analyzing complex systems has aroused interest in the development of easy-to-use analysis systems that could function and be implemented without excessive human intervention. Some of the efforts in this field include research into the development of simulation generators, and the use of simulation support systems.

Keywords: Simulation Modeling, Simulation Analysis, Decision Support Systems, Manufacturing Systems Modeling.

Introduction

Simulation Generators and Their Applications

The use of decision support systems in the area of simulation support is bound to receive a lot of attention in the years to come. These decision support systems greatly facilitate model building and simulation output analysis, and provide an easy-to-use interface between the end user and the simulation program. This paper deals with the design, construction, and development of a decision support system for a typical simulation, i.e., modeling a flexible manufacturing cell. The structure of the support system will be explained, and a few of its features discussed. This paper also discusses the generalizability of the modeling methodologies presented, and presents an overview of some popular methocts used to develop user-independent simulations. The development of the decision support system is discussed with reference to a particular flexible manufacturing cell modeling problem. Solutions to

A simulation model of a flexible manufacturing system (FMS) must be designed to effectively model the stochastic and time dependent nature of the system. Increasingly, simulation generators can be used to develop, improve, and implement simulation models of FMS alternatives. 3'1° A simulation generator has been defined by Mathewson s as "an interactive software tool that translates the logic of a model described in a relatively general symbolism into the code of a simulation language and so enables a computer to mimic model behavior." Examples of simulation languages are GPSS 9 and SIMAN. 7 To model the behavior of an FMS, it is necessary to define the parameters and characteristics that can be represented by a general symbolism. Thereafter, it is possible to transform the input parameters supplied by the user to a simulation code capable of modeling the behavior of the FMS.

484

Journal of Manufacturing Systems Volume 10/No. 6

The development of a simulation generator could facilitate the analysis and evaluation of a wide number of alternatives. The creation of simulation models and their coding and debugging would not be as tedious and time-consuming. Several advantages resulting from the use of simulation generators have been summarized by Haddock and Davis z as follows: • Facilitate Model Building and Data Input: The simulation generator makes it unnecessary for the user to be proficient in coding and high-level simulation languages. The dialogue could be made more user-friendly by using a microcomputer as a data-input device. • Facilitate Experimentation: Simulation generators emphasize the development of model characteristics. By incorporating an editor, it is possible to modify the model to analyze and model different scenarios. This eliminates the necessity of redefining data that does not change. • Incorporate Model Refinements More Rapidly: It is obvious that the more detail a simulation generator captures, the more complex the logical description of the system becomes. This, in turn, makes the coding even more difficult. • Reinforces Evaluation of Alternative Design/ Control Scenarios: By aiding the process of model-building, the simulation generator facilitates spending more time on the study of alternatives and the analysis of the implications of each of the alternatives.

• The user can be prompted for initializations of key simulation variables, thereby permitting the change of run time variables; • Little or no knowledge of the simulation language is necessary to model the system and analyze the results obtained; • Comparative analysis of competing systems can be easily done; and • The powerful features of a statistical/graphics package can be employed to carry out output comparisons and analyses and to pictorially depict the differences in system performance.

System to be Modeled The system selected for modeling was a flexible manufacturing cell proposed to be installed at a local plant. The cell was to consist of horizontal machining centers (HMCs) equipped with either palletized carrousels or automatic vehicles that would facilitate linear transfer. Five different varieties of tiller transmissions were to be made on this system. Each type of transmission had its own processing time and fixturing capability. Another factor that was of importance in this problem was the seasonal nature of the demand for products. The sequence of operations involved in the cell are as follows: • Loading pallet loads of the respective job types onto the corresponding carrousels or guided vehicles; • Waiting for the HMCs to become available; • Indexing pallet carrousels, or transporting pallet loads using a guided vehicle to their destination machine; • Exchanging pallets; • Computing machining; • Indexing pallet carrousels or transporting pallet loads using a guided vehicle to the unloading point; and • Unloading the completed parts from the carrousel or guided vehicle.

Simulation Support Systems as a Simulation Aid The use of simulation support systems is another powerful tool for simulation analysis. A simulation support system provides support to the simulation model by automating data collection facilities, providing more efficient and effective data analysis, and enhancing the quality of data presentation. A support system typically consists of a support database. Data supplied by the user can be stored in the database and called by the source simulation program when needed. The database also serves as a repository of simulation output data, which can be referenced easily by the user through a userassigned name. s There are several advantages to using a support shell:

Decision Variables of Interest The system had to be built so that the decision maker could study the effect of several different loading and processing configurations on the performance of the system. Of particular interest to the

485

Journal of Manufacturing Systems Volume 10/No. 6

Input Phase

decision making process was data concerning the following: • The total number of units of each type that would be produced in a typical year with a typical loading configuration; • The utilization of the work force in the plant; and • The number of fixtures that would have to be provided to make the system viable. In view of these requirements, it was clear that the model had to be built to permit the user to easily interact with the system from time to time. The modeling philosophy, therefore, had to be developed with this end in mind.

In the input phase, a database is created to store user-supplied data in a form that would facilitate easy retrieval. Further, the database is created so as to clearly distinguish between the different scenarios being simulated.

Model Implementation Phase The functions of the support system during the model implementation phase include formatting stored data in the database to facilitate easy reading by the simulation source code. Further, the system must make arrangements for actually transferring the data to the source program without active user intervention.

Alternatives Considered Three mutually exclusive options were debated as possible system alternatives. The options included a dedicated palletized carrousel for each machine in the cell, a carrousel capable of supplying two or more machines at the same time, and a linear transfer line system with a number of unmanned automatic guided vehicles (AGVs). Decisions needed to be made not only with regard to the best system among the three that would effectively meet the needs of the user, but also with regard to the capacity of pallet carrousels or number of guided vehicles required.

Output Phase The support system plays a number of key roles during the output phase. It is responsible for the transfer of simulation output data to the storage database and to format the data into an easy-to-read format for the output analysis software. Further, in this phase, the transfer of data from the storage database to the graphical analysis package must be effected. The structure of the support system is depicted in the block diagram (Figure 1).

Development of the Model The core models for the system alternatives being considered were developed using SIMAN, which incorporates a powerful set of blocks that can effectively model the behavior of manufacturing systems. FORTRAN routines were used in conjunction with the core models to aid in data transfer. All core models and routines are fairly generalizable and can be readily applied to other similar systems with minor modifications, n The basic pieces of input data required are the annualized demands for each job type, the number of parts of each part type per pallet load, and the processing times per pallet load for each job type. The first step toward the creation of a simulation support system is to plan and structure the basic system modules and select appropriate software programs to perform the functions expected of the various modules.

Selecting Software Packages The next step in the development of the support system is to select the appropriate software packages to effectively perform the functions of the different phases. An important consideration in the selection of these packages was user friendliness. dBASE III was chosen to perform the functions of the database system, dBASE III facilitates quick and efficient data storage and manipulation and is able to create command files containing a batched set of programming commands. dBASE III's programming language has certain features that make it appropriate for interfacing with other programs and languages. These features include the following: • The ability to execute operating system commands from within the command file;

486

Journal of Manufacturing Systems Volume 10/No. 6

• Formatting the user-supplied information in the form of an ASCII file to facilitate easy use by the FORTRAN subroutine running the simulation program; • Transferring the simulation results saved in an ASCII-formatted FORTRAN file to a dBASE record; • Transforming the results into an appropriate format and effecting a transfer to a suitable output analysis package; and • Returning control back to the command program after output analysis so that further simulations could run, if desired. Lotus 1-2-3 was selected to graphically analyze the output. The macro programming capabilities of Lotus 1-2-3 were exploited to automatically output different types of histograms and graphs in color. These enable the end user to visually compare the three different systems in terms of various performance parameters of interest. The use of a spreadsheet package is indicated because dBASE III facilitates the easy transformation of data stored in the form of fields and records to a standard spreadsheet-type row and column format.

User selections

~

-Base Ill shell~ Control Transferred

UI)roIItinesJ~

~

source programJ

Control Returned

~

-Base Ill Shell~ l

Control sent to Spreadsheet

LOTUS 123 Macro R e t u r n control to d-Base program

Implementing and Integrating the System

Figure 1

In this stage, the data formats are made compatible with the software system. The database organization is scrutinized to ensure that data from the different scenarios being evaluated is stored in distinct storage addresses. The system can be preconfigured using appropriate configuring files to further simplify the operation of the system.

Software Modules and Flow of Control

• The ability to read ASCII and text files and format an output file either using ASCII or text format; and • The ability to convert a spreadsheet of rows and columns into sequential records in a database, or to format database records as spreadsheet files that can be read by standard spreadsheet packages. The programming capability of dBASE III's command files helped to build a user-friendly shell. The shell functions as a link between the end user and the simulation source program. The following functions were performed by the dBASE III command program: • Prompting the user for decisions regarding the choice of the system(s) to be simulated, initializations of some of the simulation variables, and the method of output analysis of the simulation;

Analysis of Output Results and Conclusions Analyzing simulation output results consists of first identifying the measures of merit that could be used to draw conclusions about the system. Based on the values obtained for these measures of merit, the systems could be compared and the inherent trade-offs could then be estimated. L4 The best decision at this stage depends on the priorities of the decision maker.

487

Journal of Manufacturing Systems Volume 10/No. 6

Recommendations

Methods of Merit Considered Three key measures of merit were considered while evaluating the three proposed systems-weighted fixture utilization (140, normalized job count (J), and workeffAGV utilization. These measures of merit were selected to reflect the decision maker's priorities. Further, another key consideration was the ease with which these measures of merit could be evaluated. One factor that governed the selection of measures of merit was the different processing times associated with each job type. Each measure of merit was designed to ensure accuracy and fairness in comparisons.

Throughout this section, the terms System I, System II, and System III will be understood to refer to the following systems: • System I will refer to a cell containing two machines, each equipped with independent pallet carrousels. • System H will refer to a cell containing three machines, one equipped with an independent carrousel, while two are configured with one pallet carrousel being shared between the two systems. • System III will refer to a linear transfer line configuration wherein the pallet loads are transported to the desired machines by guided vehicles. Conclusions about the relative performance of one system versus another can be arrived at by using two comparison approaches--comparing performances of different configurations of the same system, and comparing the performance of different systems. To compare the performance of different configurations of the same system, confidence intervals developed for the different measures of performance are utilized. The data regarding confidence intervals for each of the different configurations for Systems I, II, and III are summarized in Tables 1-11.

Weighted Fixture Utilization The weighted fixture utilization (W) is defined as: ( F l x N 1 ) + (F2xN2) + (F3xN3) + (F4xN4) + (F5xN5) W = (N1 + N2 + N3 + N4 + N5) In this equation, F1, F2, F3, F4, and F5 represent the fixture utilizations of fixtures of job types 1 through 5, respectively. N1, N2, N3, N4, and N5 refer to the number of fixtures of each job type available for use. This measure compensates for the impact of differing quantities of available fixtures of each job type.

Table 1 System I With Carrousel Capacities 2 and 4

Normalized Job Count The normalized job count was used to standardize the number of jobs produced with different systems. For instance, if P(k) is the number of parts per pallet load for job type k, and if the total machining time associated with the production of one pallet load of each job type is T(k), then the normalizing ratio R(k) is given by:

Parameter

Confidence Interval

Normalized Job Totals

21568 4- 508

Weighted Fixture Utilization

0.6 + 0.0063

Worker Utilization

0.1136 4- 0.052

R(k) = T(k) / P(k) Table 2 System I With Carrousel Capacities 2 and 6

Define C(k) = number of parts of job type k produced in a specific time period. The normalized job count is given by the equation: [R(I)xC(1)] + [R(2)xC(2)] + [R(3) x C(3)] + [8(4) x C(4)] j = + [8(5) x C(5)] R(1) + R(2) + R(3) + R(4) + R(5)

488

Parameter

Confidence Interval

Normalized Job Totals

21593 + 1199

Weighted Fixture Utilization

0.6485 + 0.1168

Worker Utilization

0.1076 + 0.0224

Journal of Manufacturing Systems Volume 10/No. 6

Table 3 System I With Carrousel Capacities 2 and 8

Table 7 System III With 2 Machines and 1 AGV

Parameter

Confidence Interval

Parameter

Normalized Job Totals

21367 + 3106

Normalized Job Totals

14442 _+ 2648

Weighted Fixture Utilization

0.722 + 0.7056

Weighted Fixture Utilization

0.5945 _+ 0.0663

Worker Utilization

0.0946 + 0.0549

AGV Utilization

0.10 _+ 0.0044

Table 4 System II With Carrousel Capacities 2 and 4

Confidence Interval

Table 8 System III With 2 Machines and 2 AGVs

Parameter

Confidence Interval

Parameter

Confidence Interval

Normalized Job Totals

28163 + 2632

Normalized Job Totals

21732 + 227

Weighted Fixture Utilization

0.3 + 0.0045

Weighted Fixture Utilization

0.2956 + 0.3308

0.0396 + 0.0378

AGV Utilization

0.20 + 0.0055

Worker Utilization

Table 5 System II With Carrousel Capacities 2 and 6

Parameter

Table 9 System lII With 3 Machines and 1 AGV

Confidence Interval

Parameter

Confidence Interval

Normalized Job Totals

27687 + 580

Normalized Job Totals

18305 + 13770

Weighted Fixture Utilization

0.399 + 0.0063

Weighted Fixture Utilization

0.606 ___0.0309

Worker Utilization

0.4045 _+ 0.0158

AGV Utilization

0.10 + 0.0039

Table 10 System III With 3 Machines and 2 AGVs

Table 6

System II With Carrousel Capacities 2 and 8

Parameter

Confidence Interval

Parameter

Confidence Interval

Normalized Job Totals

27866 + 3251

Normalized Job Totals

28505 -+ 10220

Weighted Fixture Utilization

0.484 + 0.0631

Weighted Fixture Utilization

0.4619 + 0.1856

Worker Utilization

0.398 + 0.0631

AGV Utilization

0.20 + 0.0063

System I Evaluation

means of the fixture utilization statistics suggest that this is the case, it is not reflected in the confidence estimate. This is probably because the number of runs was inadequate to get a better estimate of the confidence interval. No significant conclusions about the performance of the three configurations studied can be drawn from the analysis of the normalized job count. In fact, the data seems to suggest that there is very little statistical difference between the job through-

The performance of System I was evaluated based on a comparison of carrousel capacities of 4, 6, and 8. Based on the weighted fixture utilization, it can be concluded that the greater the capacity of the carrousel in terms of the pallet loads it can carry, the greater the fixture utilization. This stands to reason because a larger pallet carrousel will require more fixtures for the extra pallet positions. Although the

489

Journal of Manufacturing Systems Volume 10/No. 6

Table 11 System HI With 3 Machines and 3 AGVs

Parameter

Confidence Interval

Normalized Job Totals

33263 + 18474

Weighted Fixture Utilization

0.4294 +_ 0.3043

AGV Utilization

0.254 + 0.2778

required in this system so that a completed part may be off-loaded without necessarily waiting for that part to undergo machining on the other machine. Once again, as in the case of System I, it does not appear that any significant advantage will occur as a result of operating with a paUetized carrousel of a larger capacity. A four-pallet carrousel meets the demands of the system.

put in each of the three cases. This is probably due to the fact that the indexing times of the carrousels are of the order of seconds, whereas the machining times of jobs are of the order of several minutes. As far as worker utilization is concerned, a definite trend is evident. The worker utilization decreases as the pallet capacity of the carrousel increases. This could be explained by the fact that the crane switching time for changing fixtures is greater if there are a larger number of pallet loads to be changed on the carrousel. Consequently, as the simulation is run for a fixed interval of time, the worker is utilized less. Here, too, the number of simulation runs is probably insufficient to draw definitive conclusions. On the whole, it would be reasonable to infer that a four-pallet carrousel would perform as well as a carrousel of a larger capacity, at a much lower cost. Therefore, this is the ideal configuration to select.

System III Evaluation The evaluation of System III is carried out by systematically varying the number of machines and the number of AGVs available. As far as the normalized job count is concerned, it is seen that it is dependent on the number of machines and AGVs available. The greater the number of machines and vehicles available, the greater the throughput in terms of production. Further, the data suggests that for a given number of machines, it would be better to have at least as many vehicles as the number of machines. The AGV utilization figures suggest that utilization decreases for a larger number of vehicles. This supports the earlier conclusion about job throughput. The AGVs' decreased utilization can only be explained by the fact that too few vehicles are unable to respond to simultaneous demands and are forced to keep jobs waiting at the load/unload points. As far as the fixture utilization figures are concerned, it is seen that the fixture utilizations drop dramatically as the number of AGVs approaches the number of machines available. This is easily explained by the fact that a fewer number of AGVs would force a number of jobs to continue waiting for a vehicle, and thus would keep fixtures busy for a longer period of time. As far as System III is concerned, it seems safe to conclude that a combination of three machines with three AGV vehicles would produce excellent results in all respects.

System II Evaluation The analysis of System II is based on the pallet carrousel capacity of the carrousel serving two machines. Simulations were conducted with carrousel capacities of 4, 6, and 8. The fixture utilization statistic reflects that the fixture utilization increases as the number of pallets on the carrousel increases. The reason for this phenomenon has been explained already. In this case, the confidence intervals developed also support the aforementioned conclusion. The normalized job count is not very different statistically from configuration to configuration. The worker utilization remains fairly constant and appears to be independent of the pallet capacity of the carrousel. It appears that the impact of the crane change-over times has been offset by a differential change in the time required for indexing. It may, be noted that frequent indexing is

Comparison Between Systems Having evaluated the possible configurations of the three systems, comparisons are now made to select the best system.

490

Journal of Manufacturing Systems

Volume 10/No. 6

Conclusion

T h e c o m p a r i s o n across systems will be m a d e primarily b e t w e e n S y s t e m s I and II, because these are the systems that can be directly e v a l u a t e d in terms o f similar capital e q u i p m e n t . Rather than m a k e a categorical j u d g e m e n t o f the suitability o f one s y s t e m o v e r the other, an analysis o f the trade-offs i n v o l v e d will be made. It must first be u n d e r s t o o d that f r o m this point on S y s t e m I will refer to S y s t e m I with a four pallet carrousel, and S y s t e m II will refer to the configuration o f S y s t e m II with a carrousel o f f o u r pallet capacity supplying two machines. W h i l e S y s t e m II results in a larger throughput in terms o f the n o r m a l i z e d j o b count, this increase is at the e x p e n s e o f adding an extra HMC to the cell. T h e advantage o f p r o d u c i n g a greater n u m b e r o f jobs must be balanced against the additional capital e x p e n d i t u r e required. As a logical extension, if S y s t e m I were to be m o d e l e d assuming another m a c h i n e present in the cell, it w o u l d be o b v i o u s that S y s t e m I would result in a larger j o b c o u n t than S y s t e m II. In this case, h o w e v e r , it must be n o t e d that S y s t e m I would involve the purchase o f one additional pallet carrousel, so the t r a d e - o f f applies. T h e fixture utilizations for S y s t e m II reflect the fact that they are practically the same as the corresponding figures for S y s t e m I. This agrees with logic since at any given time the pallet carrousels are fully l o a d e d in the case o f both systems. T h e w o r k e r utilization figures suggest that the workers are less utilized in S y s t e m I. This is also logically consistent, as clearly the s y s t e m p r o d u c i n g f e w e r jobs would also end up with a l o w e r w o r k e r utilization. It is interesting to note, though, that the sharing o f one w o r k e r a m o n g m o r e than o n e m a c h i n e c o u l d have p r o f o u n d implications on the w o r k e r utilization figures. As far as S y s t e m III is c o n c e r n e d , the c o m p a r i s o n o f this s y s t e m to S y s t e m s I and II is difficult because the capital cost o f an AGV is c o n s i d e r a b l y h i g h e r than that o f the palletized carrousel. It is debatable, therefore, w h e t h e r this s y s t e m can be c o n s i d e r e d preferable to Systems I and II, e v e n though the increased job c o u n t is certainly a factor. It is possible that a larger n u m b e r o f HMCs w o u l d result in the e c o n o m i c calculation favoring S y s t e m III.

A decision support s y s t e m was d e v e l o p e d to facilitate m o d e l i n g and analysis o f a typical flexible m a n u f a c t u r i n g cell. Suitable measures o f merit were d e v e l o p e d to m a k e effective comparisons b e t w e e n the different configurations. Specific r e c o m m e n d a tions regarding appropriate courses o f action were m a d e in light o f the results obtained f r o m the simulation and analysis o f these results.

References 1. J. Banks and J.S. Carson, Discrete-Event System Simulation, Prentice-Hall, 1984 2. J. Haddock and R.P. Davis, "Building a Simulation Generator for Manufacturing Cell Design and Control," Proceedings: 1985 Annual International Industrial Engineering Conference, 1985, pp. 237-44. 3. K&T's WorMof Advanced Manufacturing Technologies, Kearney & Trecker Corporation, WI, 1980. 4. A.M. Law and W.D. Kelton, Simulation Modeling and Analysis, 2nd edition, McGraw-Hill, 1991. 5. S.C, Matbewson, "The Application of Program Generator Software and its Extensions to Discrete Event Simulation Modeling," liE Transactions, Vol. 16, No. 1, 1984, pp. 3-18. 6. T.H. Naylor and J.M. Finger, "Verification of Computer Simulation Models," Management Science, Vol. 14, 1967, pp. 92-101. 7. C.D. Pegden, R.E. Shannon, and R.P. Sadowski, Introduction to Simulation Using SIMAN, McGraw-Hill, 1990. 8. L.J. Rolston, "Equipment Requirements Planning Using a Simulation Support System," ball Industrial Engineering Conference Proceedings, 1985. 9. T.J. Schriber,An Introduction to Simulation Using GPSS/H, John Wiley and Sons, Inc., New York, 1991. 10. Examples of Types of Computer-Controlled Flexible Manufacturing Systems, Werk Fritz Werner, Berlin, 1987.

Authors' Biographies Jorge Haddock is an associate professor of Industrial Engineering and Operations Research in the Departmentof Decision Sciences and Engineering Systems at Rensselaer Polytechnic Institute. He holds a BSCE from the University of Puerto Rico, a MSMgtE from Rensselaer, and a PhD in industrial engineering from Purdue University. Professor Haddock's primary research interests involve modeling of manufacturing/productionand inventory control systems, as well as the design and implementation of simulation modeling and analysis tools. He has numerous technical publications. His research has been funded by the US and New York State Governments, Alcoa, GM, GE, Kodak, and RCA. Professor Haddock has also been a consultant to several companies including Baxter-Travenol, Citicorp, Michelin, Bendix, and Jiffy Lube. He received the prestigious Outstanding Young Industrial Engineer Award from IIE in 1990. Nitin Seshadri is presentlydevelopinghis own consultingcompany in Madras, India. He holds a Bachelor of Engineering in mechanical engineering from Anna University in Madras, India and an MS in industrial and managementengineering from Rensselaer. His professional experience includes a position as an industrial engineer with Webco Industries in Oklahoma. V.R. Srivatsan is presently a technical analyst supporting manufacturing applications at Oracle Corporationin California. He holds a Bachelor of Technology in mechanical engineering from the Indian Institute of Technologyin Madras, India and an MS in industrial and management engineering from Rensselaer.

491