4, Fleaible Manufacturing Systems Simulation 7 IV. A Model of a Flexible Manufacturing System The example presented in this section is designed to illustrate some of the concepts involved in modeling modem FMSs, A. STATEMENT OF OBJECTIVES Company XYZ is considering investing in a manufacturing system. The decision whether fo invest in the new system depends on the system’s expected performance level, As a result, management has authorized a simulation study to predict the following statistics: 1. The mean number of jobs in the queue and the mean delay in the queue for each workstation 2. The mean time in the system for each job type 3. The proportion of time that machines are processing jobs, blocked, and idle for each machine group 4. The utilization of the transporter 3S. The mean time a job waits for transporter pick-up These statistics will be used to help management decide whether or not to invest in the system, B, SYSTEM DESCRIPTION ‘A manufacturing shop consists of a receiving/shipping station and five workstations. workstations 1, 2, .. ., 5 consist of 3, 3, 4, 4, and 1 identical machines, respectively. The distances (in feet) between the six workstations (the receiving/shipping station is station 6) are shown in Table 1 ‘The time between arrival of jobs is an exponential random variable with 1 mean of 0.25 h. Jobs are of type 1, 2, or 3 with probabilities 0.3, 0.5, and TABLE 1 ‘The Distance Matrix ‘Workstation t 2 3 4 5 6 1 o 0 100 180 200 70 2 °0 ° 100 200 80 x0 3 100 100 ° 100 100 0 4 180 20 100 ° 0 100 5 200 180 100 %0 o 100 6 27 270 10 100 100 a98 Sadashi Adige and MagedDessuty TABLE 2 The Job Routings Job ype Workstation routing 0.2, respectively. Each job enters the system at the receiving station, travels to ‘workstations on its rouie, and then leaves the system at the shipping station, ‘Table 2 shows the job routings. ‘The processing time of a job at a workstation is a normal random variable ‘whose mean depends on the job type and the workstation. The standard deviation of the processing time is 30% of the mean. The mean service time, in hours, of each job type at each workstation on its route is contained in Table 3, For example, job-type 2 has a mean service time of 1.10 h at workstation 4. ‘The dispatching rule at each workstation is FIFO. When a machine finishes processing a job, the job blocks the machine (the machine cannot process another job) until the job is removed by the transporter, A job moves from one workstation to another by a transporter (e.g., an ‘automatic guided vehicle). There is one transporter in the shop and it moves at a speed of 5 fulsec. The transporter processes requests on a FIFO basis. When the transporter finishes moving a job to a workstation, it remains at that work- station if there are no pending job requests. ‘The shop runs one shift a day. Each shift is 6.5 manufacturing hours. €, MODEL DEVELOPMENT ‘The simulation model is developed using SLAM II (see Appendix A for the model code). The entities of the system ate the jobs. The workstations and transporter are modeled as resources. To verify and validate the model, a trace of the simulation events is created TABLES "Mean Service Time of Each Job at Bach Workstation ob ype Mean service time 1 0.50, 060, 0.85, 0.50 2 1.10, 0.80, 075 3 £20, 0.25, 0.70, 0.9, 1.004, Fleible Manufacturing Systems Simaation 99 wo 20030000) so 700-800 900 1000 Figure 3. average ber size versus time (workstation 1), (see Appendix B for a listing ofthe trae). Examination of the trace shows that, the simulation model behaves according to the system description To determine the length of the initialization bias, the average number of jobs waiting at workstation 1 is plotted over time. Figure 3 shows this plot From the diagram, the initialization bias is determined to be S00 h. Stats ae cleared after 500 [As suggested in Section Il, eight simulation runs are made. Each simu- lation run is 200 days (1300 h) after statistics are cleared, D, DISCUSSION OF RESULTS ‘The job statistics are contained in Table 4. Type-2 jobs have the highest average daily production rate of 13.10 jobs with a standard deviation of 0.21 TABLES Job Statistics sob type State 1 2 3 “Average production rate (ty) 199 B.0 sat Standard deviation production rte (iy) ous 021 0.08, Average time in sssem (h) 820 10 to.00 Standard deviation time i system (b) 130 130 130100 Sadashv Adiga and Maged Dessouky TABLES Workstation Stat ‘Workstation number Statistic ' 2 3 4 5 Percentage busy 9586 6590 TRIS THIS 78.20, Percentage blocked age Or) oases: (an) Average queue size 19508803808 Standard deviation ueue size Sine 00s = los 0 | 028) ‘Average waiting time (b) ast oz on om on) Standard devition waiting time (h) 132 002.010.0802 jobs. Type-3 jobs stay in the system the longest with an average time in the system of 10 h and a standard deviation of 1.3 h. A 95% confidence interval of time in the system from the eight simulation runs for type-3 jobs is (8.94, 11.06) 1h. The upper limit ean he used ta set delivery due dates for type-3 jobs The transporter is busy 29% of the time, and requests wait on the average 0.61 min for the transporter. These results indicate that one transporter is suf- ficient to meet the current demand. Table 5 contains the workstation statistics. The results show that work- station 1 is the bottleneck station. It is busy processing jobs 95.86% of the time. All other workstations are busy around 70% of the time. The proportion of time any workstation is blocked is low because the transporter is underutilized Workstation | has on the average 17.96 jobs waiting to be processed with a standard deviation of 5.42 jobs. The jobs on the average wait 4.45 h at workstation 1, The other workstations have much lower queue sizes. The overall average time in the system is 8.26 h Waiting for workstation 1 constitutes about 54% of the time a job is in the system. This result suggests that another machine may be needed at workstation 1. If another machine is added, machine utilization at workstation | decreases to 72%. The average queue size and waiting time reduce to 0.563 jobs and 0.138 bh, respectively. Thus, the overall average time in the system reduces to 3.93 h V. Simulation Software Discrete event simulation can be implemented using most programming lan- guages. But today most simulation models are written using special purpose languages that contain basic utilities needed to perform most simulations.4. Pleible Manufacturing Stems Simulation 101 ‘A. GENERAL-PURPOSE COMPUTER LANGUAGES FORTRAN was the most popular language for simulation modeling in the 1970s because of its ability to handle numeric computations conveniently and efficiently. In recent years, PASCAL and C have emerged as competitors 10 FORTRAN. These relatively newer general-purpose languages provide data structures that make the abstraction process easier. ‘An advantage in using a general-purpose language is the ability to analyze 4 large number of problems since the user defines and develops all the model constructs. However, @ major disadvantage is that a significant amount of com- puter code must be written to manage the simulation mode! and collect statistics General-purpose simulation languages are designed to reduce the amount of user programming. Basic utilities are provided to manage the simulation model and collect statistics. B, GENERAL-PURPOSE SIMULATION LANGUAGES General-purpose simulation languages aid users hy providing utilities that handle repetitive tasks such as initialization and time advancement. Some ex- amples of general-purpose simulation languages are GPSS (Gordon, 1975), SLAM II @ritsker, 1987), SIMSCRIPT (CACI, 1983), and SIMAN (Pegden, 1985) ‘These simulation languages contain collections of FORTRAN subroutines with set of control statement used to invoke them. The system-provided subroutines reduce the amount of code a modeler has 10 write when building a model Utilities offered by some of the popular general-purpose simulation lan- guages include 1, Automatic queue management, including generation of statistics on average waiting time, maximum waiting time, and average queue length 2. A generator of uniformly distributed pseudorandom numbers used to support samples from probability distributions such as Normal and Poisson 3. Provisions to control the number of runs, the length of a simulation run, and time to clear statistics 4, Automatic generation of entity arrivals to the system 5, Collection of statistics (i.e., mean, standard deviation, minimum, max- imam, etc.) for chosen variables 6, An optional animation capability Most simulation languages have the basic constructs needed to represent ‘common production facilities. Simple examples can be modeled easily in these languages but large applications involve complex development. As Grant (1987) points out, simulation models of production facilities are at times expensive 10102 Sadashiv Adiga and Meged Destouby build and cumbersome to use on a daily basis. For this reason, research into the use of alternative approaches to simulation system design is being pursued. Knowledge-based systems (Shannon et al., 1985; Ben-Arich, 1986) and object- oriented programming systems (Adiga, 1989; Olumolade er al., 1990) are ex- amples of the new approaches. C. SPECIAL-PURPOSE SIMULATION PACKAGES, Special-purpose simulation packages have been developed for modeling ‘manufacturing systems. MAPI1 (Wortman and Miner, 1986) and SIMFACTORY (Klein, 1986) are two such packages. These packages use high-level building blocks to provide a user-friendly interface. XCELL (Jones and Maxwell, 1986) has a graphics interface. Building simulation models from these packages is easy and usually faster than using a general-purpose simulation language. However, these packages arc designed to analyze only manufacturing systems and are limited to certain decision rules. Since the packages are fixed in the view of the world they take, using the package to mode! a different view may be difficult D. SELECTION OF SIMULATION SOFTWARE, Personal computers (PCs) have recently become attractive platforms for running FMS simulations due to the availability of large memory and 16-bit (or more) microprocessors. Most of the general-purpose simulation packages dis- cussed earlier have PC versions of their mainframe software. These versions usually have additional interactive and graphics utilities ‘A comprehensive catalog of simulation software appears in the October issue of Simulaion, published by the Society of Computer Simulation. Industrial Engineering magazine, a publication ofthe Institute of Industrial Engineer, also publishes special issues devoted to simulation, ‘The selection ofa language should be guided by the actual needs of users and the application under consideration, Some of the desirable features in sim- ulation software are discussed by Haider and Banks (1986). VI. Limitations of Simulation Simulation modeling is not without its limitations. Since any model is an ab- straction of realty, the modeler may ignore some of the details and focus only ‘on those most relevant to the problem at hand. The results should be evaluated in the context of the assumptions made (or the details ignored). ‘Not all questions can be addressed by simulation. While it isa good tool for evaluating a number of alternative system designs, the modeler must generate4. Plerible Manufacturing Systems Simulation 103 the data for the alternatives (or come up with alternatives). Unlike optimization techniques, simulation does not pick an optimum solution to a problem. ‘The simulation study is a time-consuming process and requires the modeler to have skills in the following areas: data analysis, statistics, programming, and project management. Consequently, simulation is appropriate as an analysis tool ‘when the system is too complex to be studied by performing a data analysis or by building an analytical model VII. Conclusion ‘During the design phase of an FMS, a simulation analysis can aid in evaluating the design. The results ofthe simulation model can aid in determining the required ‘number and location of machines, the location and size of the butlers, and the appropriate type of material handling system. ‘After the FMS has been installed into the manufacturing environment, continuous improvement of the system is needed to fully utilize its resources land meet the production goals. For example, a simulation model can be used to analyze different scheduling strategies without disturbing system operation. ‘A wide range of simulation software tools are available to make the pro- ‘gramming task easier. Selection of the appropriate software tool depends on the application. In building the simulation model, careful consideration must be given to model validity and statistical issues. Appendix A: SLAM II Simulation Model Code sen, dessouky, ins example, 10°31/89.8. 1, rontr, clear, 500, inst, (1) workstation 2,3) 51