Simulation of a Manufacturing Assembly Line Based on WITNESS

2011 Third International Conference on Computational Intelligence, Communication Systems and Networks Simulation of a Manufacturing Assembly Line Based on WITNESS SeyedAli Mirzapourrezaei mirzapour.ali66@gmail.com Ahmad Dargi Dargi77@gmail.com Morteza Lalmazloumian mortezamazloumian@gmail.com Kuan Yew Wong wongky@fkm.utm.my Abstract Nowadays, simulation models have been used to evaluate various aspects of manufacturing systems. This paper introduces a manufacturing assembly line of starter production as a case study and the basic application of the WITNESS software. Adequate modeling but inadequate software experimentation may lead to poor decisions and can be detrimental, particularly when financial investment is involved. The objective of the study is to escalate the productivity and efficiency of the line by using precise simulation. This paper has been divided into three parts. At first, by analyzing the results of the model, the bottlenecks of the production system and the causes of the problems were identified. Second, the results of the model were identified and validated. Finally, some modifications in the model structure to improve the assembly line were set, so it would minimize the inventory of products and improve the total output. Keywords: Simulation, WITNESS, Assembly line, Bottleneck I. INTRODUCTION There are a lot of purposes in manufacturing for which simulation has been used including strategic capacity planning, automation systems design, manufacturing process validation, and evaluation of various manufacturing execution scenarios. Simulation can be used to analyze how system performance is affected by the layout configuration, the number of material handling resources used, the resource operating policies, and the usage of different types of material handling systems. A large number of industrial organizations in developing countries use simulation in their manufacturing systems so that practical production problems relating to their daily operations would be solved. Growing intricacy of modern and competitive manufacturing systems, bitter experiences of system performance and unrealized expectation of system output have forced many companies to carry out detailed simulation to thoroughly test alternative manufacturing possibilities and key manufacturing decisions before any implementation is done. Conducting experiments on the actual manufacturing system for its improvement and change in most of the cases is unwise, or not possible, since this exercise would be expensive and involve unbearable risk to the company. As a contemporary, flexible and stylish management analysis tool, simulation is being widely used in manufacturing and process industries [1]. Kochhar [2] discussed the progress made and progressive use of computer simulation of manufacturing systems since the sixties. There is an obvious growing increase in the use of simulation for the analysis of manufacturing systems. This is strengthened by reduction in computing costs, improvements in simulation software and through a greater emphasis on developing and using automated manufacturing systems to improve productivity and reduce costs. II. ROLE OF SIMULATION IN MANUFACTURING Recently, there is a constant rise in using simulation in manufacturing environments. Evidences on the use of simulation in manufacturing can be found in the literature. There are many publications that provide surveys of simulation being used in manufacturing systems. Hlupic and Paul [3] have analyzed the problems to be solved, the software tools used, and the main results obtained in a number of simulation studies focused on flexible manufacturing systems. Kochar and Ma [4] have described the major characteristics of simulation studies carried out in order to facilitate decision-making for solving production management problems. Kiran and Smith [5] have reported on numerous simulation studies carried out in the area of production scheduling. It is usually not necessary to have detailed level of simulation. Details are always available in companies without any difficulty such as breakdown patterns, rework rates, operator performance variations, etc. Removing the need for such data (at least initially) is an obvious attraction for the use of hierarchical modeling [6]. Libraries of components of a system can be used to develop models quickly. The use of the hierarchical approach will allow 978-0-7695-4482-3/11 $26.00 2011 IEEE DOI 10.1109/CICSyN.2011.38 132

higher level elements to be established which can be used for fast model building [7]. Simulation models are inherently complex. Fishwick [8] discussed whilst the relationships between different elements are simple to understand in detail, an overview of the whole model taken from the overall likely behavior is difficult to ascertain. Therefore, the use of hierarchical techniques is beneficial for an overall reduction in the number of components, interactions and the overall behavioral complexity. While there is an alternative approach in which the detail is hidden in progressive levels, only the former approach actually simplifies the model structure [9]. Singhal et al. [10] have discussed how models can play a major role in the design and control of complex automated manufacturing systems, and provided a review of several studies where off-line simulation studies were used to assist in system design, production planning, scheduling and control. Ramasesh [11] provided a state-of-the-art survey of simulation based research on the dynamic job shop scheduling problem. Similarly, Ballakur and Steudel [12] described simulation studies carried out to address problems such as scheduling and sequencing, workload balancing, work flow structure analysis and job shop capacity evaluation. In the context of current and future issues concerning the scheduling of flexible manufacturing systems, Hutchinson [13] discussed several simulation studies which were carried out in order to improve the performance of flexible manufacturing systems. III. MODEL BUILDING This section explains the execution of the simulation study carried out at a starter assembly line which is located in Iran. As mentioned before, the main purpose is to increase the productivity and efficiency of the line. A. The assembly line Assembly line has been generally used in a mass production system such as in starter production. The conceptual model has been shown in Figure 1. The starter line includes a number of workstations arranged in a sequence and connected by a conveyor (see Figure 2). The workstations are semi-automated, where an operation is done by a machine and operator. Starters are assembled on particular jigs and fixtures and conveyed along the conveyor. At each workstation, a task is done according to a standardized processing time which is called cycle time. Then the operator will release the semiassembled part to the next workstation. The whole sequence of this process is repeated along the assembly line and leads to the production of a complete starter. Two workstations (St401 and St402) are decoupled by a buffer to ensure the balance of the line because they prepare two semi-assembled parts which have less cycle time and can be buffered. Furthermore, the conveyor has a capacity which is determined by the length of the conveyor between two workstations, and this corresponds to the amount of work-inprogress (WIP). Figure 1: Conceptual model of assembly line Figure 2: The assembly line IV. DATA COLLECTION AND ANALYSIS This task is usually the most frustrating and time consuming, and yet it is also the most important issue because it will set the scene for the whole project. The duration for developing the simulation model can very much depend on the amount of quantifiable and quality data that has been collected from the client. If client data are not collected effectively, they can be a major hindrance in trying to establish an accurate model [2]. Banks and Carson [14] argued that "even if the model structure is valid, if the input data are incorrectly collected, inappropriately analyzed, or not representative of the environment, the simulation output data will be misleading and possibly damaging." 133

Duration input (the most critical form of input) to a simulation experiment is classically approached by fitting a statistical distribution to a collected sample of observations. A simulator can fit any of the classical statistical distributions to the sample of observations. In any case, a check for goodness-of-fit should be performed. This is often done in the form of statistical goodness-of-fit tests like the chi-square test, Anderson Darling (AD) test, Kolmogorov- Smirnov (K-S) test, q-q plots, and visual inspection of the quality of the fit of the empirical cumulative density function (CDF) and the fitted (theoretical) CDF. One can also consider the visual inspection of the theoretical probability density function (PDF) and the histogram of the sample data. The steps normally followed in input modeling are shown in the flowchart given in Figure 3 [15]. In this project, the stopwatch tool has been used for gathering data in the production line. The cycle time of each station has been captured as a sample of about 40 times; then, the distribution of each cycle time has been defined using the Minitab software. The statistical goodness-of-fit test which was used for this model is Anderson Darling (AD) and an example is provided in Table 1. Furthermore, the results of distribution of cycle time for each station are summarized in Table 2. For example, in the fifth station (St405), the distribution of cycle time is Normal; the amount of AD is equal to 0.285, which is acceptable with regard to our confidence level (see Table 1). However the amount of AD for Lognormal distribution is also appropriate, it is better to choose Normal distribution for the man-machine processing time in a continuous manufacturing line. Table 1: Goodness-of-fit test (cycle time for station St405) Distribution Anderson Darling(AD) Normal 0.285 Lognormal 0.286 Exponential 18.175 Weibull 0.654 Gamma 0.296 Loglogistic 0.312 Figure 3: Flowchart of checking goodness-of-fit Table 2: Distributions of cycle time (seconds) for each station The name of Station Distribution 401 Normal(16.144, 0.339) 402 Normal (16.164,0.490) 403 Normal (27.673, 0.237) 404 Normal (29.142, 0.332) 405 Normal (47.352, 0.282) 406 Normal (47.289,0.491) 407 Normal(29.926,0.0283) 408 Normal (19.313, 0.262) V. SIMULATION MODEL A. Run initial model and data analysis: The simulation model developed using the WITNESS software is shown in Figure 4. The model was run for 3360 simulation time unit; using the report tools, the statistics for each model element are shown in Table 3. This table shows the busy and idle time for each part of the assembly line along with the number of operations. In this model, St405 which is being fed by St401 and St404, is a bottleneck and paces the assembly line. Since St401 and St402 have relatively less cycle time, semiassembled parts are blocked in St403 and St404. Moreover, there is more idle time in subsequent stations St407 and St408 which are located after the bottleneck because they have less cycle time. 134

Figure 4: Simulation model Table 3: Simulation report (machine statistics) B. Model verification For verification, some elements in the model have been checked by comparing their number of operations. Based on the busy time of each machine, the number of operations was computed, and then it was compared with the number of operations obtained from simulation. Working hours in two shifts are 840 minutes (2*[480-60)]) in a day. All the stations have been tested while the model was developed step by step to ensure that the model worked appropriately. As a sample, two machines have been selected to show the verification results: For station 405: Busy time: 95.40% Mean of distribution of machine s cycle time: 0.789 (Min) Working time for 4 days: 840*4= 3360 Number of operations= (0.9540* 3360)/ 0.789 = 4062 units The number of operations generated by the simulation model is 4062, which is exactly similar to the result computed manually. For station 404: Busy time: 73.51% Mean of distribution of machine s cycle time: 0.4857 (Min) Number of operations= (0.7351*3360)/0.4857 = 5085 units The number of operations which is obtained from the simulation model is 5084. The percentage of deviation from the calculated number of operations is only 0.02%. C. Validation Validation is the process of determining the degree to which a model is an accurate representation of the real world from the perspective of the intended use of the model [16]. The validation process is made up of two different steps: Evaluating the difference between the simulation output and actual production output. Validation is accomplished through the Mann- Whitney test. 1) Comparing the actual and simulation output Based on the data collected from the company, the actual production output is 3212. The simulation output produced by the model is 3247. 2) Mann-Whitney Testing To conduct the Mann-Whitney test, the production outputs of the real system and simulation model were compared using the SPSS software. For this purpose, the hypotheses applied to check the conformity between the real system and built model were: H 0 = the number of production outputs during 3360 minutes, Y, is equal to 3247 (Y=3247) H 1 = the number of production outputs during 3360 minutes, Y, is not equal to 3247 (Y 3247) The level of significance, α chosen was 0.05. As Table 4 shows the p-value (2-tailed) is 0.105 which is greater than 0.05, so the null hypothesis is not rejected and the built model is validated. Table 4: Test statistics Mann-Whitney U 9.000 Z -1.621 Asymp. Sig. (2-tailed).105 Exact Sig. (1-tailed Sig.).180 D. Determining the number of replications The number of simulation runs to produce the desired level of accuracy, was estimated. The amount of replications was computed as follows: Number of replications = (t α/2,n-1 *s)/(h*μ) (1) where t α/2,n-1 : t-value based on the confidence level and the number of runs 135

s: standard deviation of production output from six runs μ: mean of production output from six runs h: allowable error The confidence level was 95% and allowable error was 0.05. Table 5: Results of running model No. of 1 2 3 4 5 6 runs No. of 3247 3248 3250 3246 3247 3247 Products So, the number of replications was computed as follows: (2.571* 0.563)/ (0.05* 3247.5) = 0.009 E. Determining the warm-up period The company operates a continuous production facility, where we are interested in the steady state behavior of the model. Since the model starts out empty, it usually takes some time to reach the steady state. In the steady state condition, the response variables in the system reveal statistical consistency. There is a turbulent behavior in the time of reaching the steady state which is a function of the activity time; however, this is not the actual system behavior. Determining the warm-up period is crucial to avoid this unusual behavior of the system. We will wait until after the warm up period before we start gathering any statistics. In this way, we eliminate any bias result from observations taken during the transient state of the model. Warm-up period varies for different models. In this situation, we have chosen production rate as the response variable. When the model reaches the steady state, which is 1470 minutes in this case, the production rates become steady. The warm-up period diagram of the assembly line is shown in Figure 5. and efficiency and at the same time, reduces the work-inprogress (WIP) in the whole system. Added station Added station Figure 6: Simulation of improved model Table 6: Simulation report of improved model (machine statistics) Figure 5: The warm-up period VI. IMPROVEMENT Since the aim of this study is to improve the throughput and efficiency of the system through reducing the idle time and improving the product flow, we came to the conclusion that by developing one parallel station in St405 and St406 which are the bottlenecks of the assembly line, the efficiency of the whole system would be increased. Having a duplicate station in St405 and St406 causes the total output of the model to increase from 3247 to 4167 parts. As shown in Figure 6 and Table 6, adding parallel stations in the assembly line increases the production rate VII. CONCLUSION This case study presented the details of a production system, designed using the WITNESS simulation software. It analyzed the simulation output and compared the performance with the existing system. The current manufacturing facility was running inefficiently and was unable to meet the demand. A better design of the production system at the company was proposed. This was done by increasing the quantity of machines in the stations which were the bottlenecks of the assembly line. This case study illustrated the methods of modeling and designing a production system so that others can do the same. It will be 136

interesting in the future to use simulation software like ARENA, SHOW FLOW etc, and compare its result with the outcome obtained from WITNESS. REFERENCES [1] Neelamkavil F. Computer simulation and modeling. New York: Wiley, 1987. [2] Kochhar AK. Computer simulation of manufacturing systems Ð 3 decades of progress. In: Proceedings for the 3rd. European Simulation Congress, Edinburgh, UK. 1989. p. 3±9. [3] V. Hlupic and R.J. Paul, A review of simulation research in FMS, in: Proceedings of the 13th International Conference Information Technology Interfaces (Cavtat, June 1991), Croatia, University of Zagreb Computer Centre, Zagreb (1991). [4] A.K. Kochar and X. Ma, Use of computer simulation: Aids for solving production management problems, in: Proceedings of the 3rd European Simulation Congress, Edinburgh (1989) 516-522. [5] A.S. Kiran and M.L. Smith, Simulation studies in job shop scheduling: A survey, in: Proceedings of the Conference on Simulation in Inventory and Production Control, San Diego (1983) 46-51. [6] A.S. Carrie, Hierarchical modelling-principles and review, Proc. WORKSIMS Conf., Singapore, 1994. [7] G.M. Lane, J.M. Fegan, Hierarchical simulation of a flow line for printed circuit board fabrication using ISI-the Intelligent Simulation Interface, Proc. Eur. Simulation Symp., Society for Computer Simulation (SCS), Ghent, Belgium, 1990, pp. 215 219. [8] P.A. Fishwick, The role of process abstraction in simulation, IEEE Trans. Systems, Man Cybern. 18 (1) (1988) 18 39. [9] V. Ceric, Hierarchical abilities of diagrammatic representations of discrete event simulation models, Proc. Winter Simulation Conf., IEEE, Lake Buena Vista, FL, 11 14 December, 1994, pp. 589 594. [10] K. Singhal, C.H. Fine, J.R. Meredith and R. Suri, Research and models for automated manufacturing, Interfaces 17 (6) (1987) 5-14. [11] R. Ramasesh, Dynamic job shop scheduling: A survey of simulation research, OMEGA Internat. J. Management Sci. 18 (1) (1990) 43-57. [12] Ballakur A., Steudel H.J. (1987), A within-cell utilization based heuristic for designing cellular manufacturing systems; International Journal of Production Research 25; 639 665. [13] J. Hutchinson, Current and future issues concerning FMS scheduling, OMEGA Internat..I. Management Sci. 19 (6) (1991) 5299537. [14] Banks, J., and Carson, J. S. (1984). Discrete-event system simulation. Prentice-Hall, Inc., Englewood Cliffs, N.J. [15] AbouRizk, S. M. (1989). BetaFit user's guide. Division of Construction Engineering and Management, Purdue Univ., West Lafayette, Ind. [16] Balci, O. 1998. Verification, Validation and Testing. In Handbook of Simulation, J. Banks eds., pp.335-393, Georgia: John Wiley & Sons, Inc. 137