Please copy and paste this embed script to where you want to embed
A Simulation-Based Approach for a Solar Panel Production System Ange Lionel Toba Department of Modeling and Simulation Old Dominion University Norfolk, VA 23529 Leonardo Bedoya-Valencia Department of Engineering Colorado State University – Pueblo Pueblo, CO 81001 Abstract Assembly Line Balancing Problems (ALBP) involve production planning in order to improve an existing assembly process or create a new one. The ultimate goal, in regard to high efficiency is to maximize the facility throughput, in function of time, capital investment, resource utilization, customer demands and material handling. In this paper, an approach for balancing a Concentrated Photovoltaic (CPV) solar panel assembly line is proposed, using Discrete Event Simulation (DES). It aims to develop a more efficient CPV solar panel assembly line, which will allow the manufacturing facility to meet the demand, while reducing congestion between workstations within the production process. Several possible scenarios are developed for the deterministic case. These scenarios are then used as benchmarks for their corresponding stochastic scenarios analyzed by a simulation model, considering changing demands and different production capacity. With this proposed approach, the decision maker will be able to select the best assembly line in order to satisfy customers’ demand while using time and resources required efficiently.
Keywords Simulation, Assembly Liner Balancing, Production Planning, Solar Panels
1. Introduction Concentrated photovoltaic (CPV) is one of the newest forms of solar energy technology on the market today. A CPV system converts light energy into electrical energy like conventional photovoltaic technology does. The difference in these technologies lies in the addition of an optical system that focuses a large area of sunlight onto each cell, called multi-junction solar cell. This type of cell generally utilizes three different photovoltaic materials: Gallium Indium Phosphide (GaInP), Gallium Arsenide (GaAs), and Germanium (Ge) P-N junctions in a single cell, which extract more energy from the range of wavelengths in sunlight. This system enables the cells to produce a consistent increasing amount in voltage, while allowing only minor loss in energy. This paper is intended to develop a DES model in order to analyze alternative production systems of CPV solar panels. The basic idea is to simulate scenarios for the different production systems, analyze them and define the best system according to defined measures of performance. What quantity and quality of the personnel is needed for a specific job? How much idle time between tasks is affordable? How flexible is the process assembly and how easy is it to adapt it to increasing demand at some point in time? Those questions are answered by using a DES model, which allow for a more appropriate design of experiments and more accurate results, given the “randomness”, or the inclusion of uncertainty. Unlike deterministic models, which are just used for results-confirmation purposes, stochastic models are able to handle uncertainty.
Toba and Bedoya-Valencia The following section presents the literature review. Section 2 shows the development of the simulation model and the conducted experiments. In section 4, the results are analyzed and finally, section 5 presents the conclusion of the experiments.
2. Literature Review The assembly line balancing problem (ALBP) deals with the amount of the work, in terms of time, which has to be performed at each workstation, given a defined structure of precedence requirements. Assembly line balancing is used to determine optimal allocation of operations at the workstations in order to minimize the number of workstations needed for a given cycle time (Type I problem) by equalizing the loads on the workstations or to determine optimal allocation of operations at the workstations in order to minimize the cycle time of the line for a given or fixed number of workstations (Type II problem). The objectives of both types of problems are similar in nature in terms of minimizing idle time and consequently improving the assembly line production efficiency . Extensive research in assembly line balancing has been done in the last 50 years. The first published analytical statement of the ALBP was made by Salveson  and followed by Jackson , Bowman , Supnik and Solinger , White , and Hu . Since then, the topic of line balancing has been of great interest to academicians. Although extensive research has been done in the area, the problem has consistently resisted the development of efficient algorithms for obtaining optimal solutions . The performance of assembly systems is very much dependent on the throughput of the system, on the utilization of the transportation systems (robots, AGVS, etc.) and on the reuse of these transportation systems when different configuration of the assembly line are required to meet customers’ demand. Framing this problem as an optimization problem can be cumbersome in most cases therefore different approaches have been developed. Among others, simulation has recently become an essential and effective tool for designing and managing manufacturing systems. Patel  discussed the methodology of modeling and studying the final process system of the automobile manufacturing process in order to develop an effective and efficient process to ensure the system throughput. Choi  discussed initial efforts to implement simulation modeling as a visual management and analysis tool at an automotive foundry plant manufacturing engine blocks. The optimum performances were identified through the use of scenarios by varying the number of assembly machines and processing times. In their work, Ali et al.  stated that simulation has been commonly used to study behavior of real world manufacturing system to gain better understanding of underlying problems and to provide recommendations to improve the systems. Most recently, Bukchin and Rubinovitz  studied the problem of assembly line design, focusing on the station paralleling and equipment selection. They discussed two problem formulations, minimizing the number of stations and the total cost. Masood  showed increased throughput and higher machine utilization in an automotive plant as a result of line balancing. Templemeir  has provided an overview of research on line balancing and its application to the real world. He made the observation that real world systems have stations with non-identical mean processing times. Since many algorithms for the evaluation of stochastic flow production systems make the assumption of equal processing times, they are not well suited to these types of systems. For a line with all buffers equal to zero, Vidalis et al.  found that the optimal workload allocation followed a bowl pattern (Assigning less work to the workstations located near the center of the assembly line, and more work to the workstations near the beginning and end of the line resulted in optimal productivity, when the number of service phases at each station was equal as defined by Hillier and Boling ). Shaaban and McNamara  used simulation to compare bowl, inverted bowl, and monotonically increasing and decreasing arrangements with equal buffer sizes (of at least one per station), and different degrees of imbalance. They found that the bowl arrangement resulted in the least idle time, whereas the decreasing workload arrangement (with bottleneck at the beginning) resulted in the lowest average buffer levels. They also found that buffer capacity had a higher impact on both these measures than the degree of station imbalance.
3. Development of the Simulation Model The approach chosen for this project is to use a DES model implemented in the software tool (SIMIO). This approach helps to find a high performance configuration. The simulation model provides control over key components of the model, such as assembly workstations, customer orders and requirements, work shifts and flow control. Figure 1 displays the assembly process with its corresponding precedence constraints. Next, the stages are briefly explained:
Toba and Bedoya-Valencia • • •
• • • • • • •
Cells: In this stage, the multi-junction solar cells are assembled. Cell_Box: In this stage, the multi-junction solar cells are assembled onto the big box. Wire_Cell_Box: In this stage, electric wires are assembled to the previous assembly. This is to wire the cells to each other and create a circuit in series and allow the same amount of current to flow through the cells. Wire_Yoke: Wires are inserted inside of the yoke. Motor1_Sensor: The sub-assembly consisting of one motor, two sensors and four screws is put together. Motor1_Yoke: The previous sub-assembly is assembled to the yoke (with the wires already inserted). CB_Motor1_Yoke: The command box is added to the previous sub-assembly. Box_Yoke: This stage requires one of each sub-assembly (yoke + box) obtained in previous process. Panels: In this stage, the lenses are added to yoke + box sub-assembly. Tack_Free: This stage does not require any work, but instead it requires time for the sealant attaching the lenses to cool off enough to allow the panels to be moved someplace else. After this stage, the assembly is considered to be complete. Cure Time: It takes place in a warehouse, where all the assembled panels rest, in order to give time for the thermo paste to cure completely. Once this time has elapsed, the CPV panels are ready for delivery.
1 Cells Assembly
2 Wire_Yoke assembly
4 Cell_Box Assembly
6 Wire_Cell_Box Assembly 8 Yoke_Box Assembly
5 Motor1_Yoke Assembly
7 Motor1_Yoke _CB Assembly
3 Motor1_Sensor Assembly
11 Cure Time
9 Panel Assembly
10 Tack Free
Figure 1: Schema of the assembly system In this research, an ALBP is considered with two variations from the traditional problem formulation. The first one is the consideration of stochastic processing times and the second one deal with the objective. Here, based on the precedence constraints given in Figure 1 and using the average value of the processing times, the minimum expected cycle time is found. Then a Discrete Event Simulation (DES) model is used to estimate, by experimentation, the cycle time under different scenarios of demand and level of resources. Note that the minimum expected cycle time is 540 seconds (9 minutes), given by the panel assembly operation. According to Nahmias , for a cycle time C, the minimum number of workstations is given by W = ?T/C? where ?? indicates that the ratio is rounded up to the next larger integer and T is the total work content associated with the production of a solar panel. Table 1 shows the expected values for the processing times of the different operations to assemble the solar panel. It is important to notice that the processing times presented in this table correspond to manual assembly operations performed by operators with the required tools. Moreover, since there is not actually any real production process
Toba and Bedoya-Valencia taking place, the processing times were obtained by conducting a time study using “Maynard Operation Sequence Technique” (MOST) as described by Niebel and Freivalds , with the assumption that the operations are performed by experienced and skilled employees when analyzing the assembly of a CPV solar panel prototype. MOST decomposes assembly operations into three basic activities and their processing times can be obtained by using MOST tables. Table 1: Expected Processing Times Operation Processing Time (seconds) Cells Assembly Cell_Box Assembly Wire_Cell_Box Assembly Wire_Yoke Assembly Motor1_Yoke Assembly Motor1_Yoke_CB Assembly Motor1_Sensors Assembly Yoke_Box Assembly Panel Assembly
133 170 27 28 40 10 19 108 540
In this case, T = 1075, C = 540 so the minimum number of workstations W is equal to 2. Notice that the curing time was not included in the calculation as this operation is performed at the same station where the panel assembly is performed and it does not require any resource other than space for the glue to fully cure. A special consideration was made regarding the balancing of the assembly line when this last station has work content not requiring resources but blocking the station. The expected curing time is almost twice the minimum cycle time and in the worst case almost three times the cycle time. Therefore, in order to avoid stoppage of the assembly line, three stations are required in the worst case. A sub-assembly is produced out of station 1 in average every 540 seconds (9 minutes) after 24 minutes from the starting time of the 8 hour shift. Station 2 requires a work content of 9 minutes in average plus the curing time (15 minutes). The curing time will block station 2 for 15 minutes on average, and 20 in the worst case, so in order to keep station 1 producing subassemblies and assuming the worst case for the curing time (20 minutes) three stations 2s are required. As mentioned before, there is no real assembly process yet being performed so not much is known about the distribution of the outcome. The data distribution fit could not be performed the traditional way. However, the time studies performed using MOST allowed to estimate/approximate the most likely value of the processing times in each workstation, and the worst and best case as a deviation of 5% above and below from this value respectively. Therefore, the triangular distribution is typically used as a subjective description, given the scarcity of data, for this process. The validation step was performed by comparing the average processing times on each workstation obtained from the simulation model with its corresponding value obtained by the time study .
4. Experiments This section discusses the results obtained after running different scenarios analyzing a balanced assembly line. The measures of performance to be used are: the cycle time, workstation utilization percentage, the workload distribution among workstations, and the number of shifts required to meet the demand. The transportation time was willingly left out, due again, to absence of data. We could not have access to a real assembly process so including some transportation time could seep more errors in the model. It would also be relevant to notice that this study does not take any layout study and dimensioning into account. However, these times can easily be incorporated into the model if the methods of transportation are correctly specified. Details just have to be provided as to how far the stations are, to each other.
Toba and Bedoya-Valencia 4.1 Scenario Analysis In this set of scenarios, eight (Cell, Cell_Box, Wire_Cell_Box, M1_Sensors, M1_Yoke, CB_M1_Yoke, Wire_Yoke and Yoke_Box) out of the nine stations are put together to form a single station named W1. The work at this single station will be performed by a single worker, respecting of course, the rules of precedence. Referring to Figure 2, the worker performs, in order, task 1-2-3-4-5-6-7-8. Alternative orders are possible, but it will not make any difference, since the makespan in W1 will be the same. After all the operations at this station are completed, a second worker, assigned at the Panel station named W2, can start performing task 9, see Figure 2. In the base case, there is one production order arrival every 18 minute demanding one panel. Also, three Panel and TackFree stations are assumed in W2. For the scenario 1, another worker is added to the W1. Two workers are now taking care of W1 with the following assignment: worker 1 works on Cell, Wire_Cell_Box and Yoke_Box and worker 2 works on Cell_Box, M1_Sensors, M1_Yoke, Wire_Yoke and CB_M1_Yoke. This task allocation is optimal as we are trying to reduce the idle time of the workers. Based on the processing times given above, worker 2 would spend 97 time units (28+19 +40+10) on stations M1_Sensors, M1_Yoke, Wire_Yoke and CB_M1_Yoke, while in the mean time time, worker 1 would still be at the Cell station (133 time units). After worker 2 is done, he can then move to Cell_Box station and start working (worker 1 would have not finished yet but will surely have done enough for worker 2 to start his job at the next station). Once worker 1 is done, he can move to Wire_Cell_box station to start (Here too, worker 2 would still be busy but would have done enough for worker 1 to pick up), knowing that output of CB_M1_Yoke is already ready. The last step would be the Yoke_Box station, which will obviously be taken care of by worker 1, since worker 2 would still be finishing up at Wire_Cell_Box station. Any other assignment would see an increased idle time of either worker between tasks, which is unwanted. The scenario 2 is built upon the previous one, with the same environment, but the demand is doubled (2 arrivals every 18 minutes). Scenario 3 suggests the presence of an additional worker (worker 4) at station W2 may be necessary. The demand is reduced back to the initial quantity, with 1 arrival per 18 minutes. Finally in Scenario 4, the environment is still the same, with 4 workers, 3 Panel stations and 2 arrivals at 18-minutes intervals.
1 Cells Assembly
2 Wire_Yoke Assembly
4 Cell_Box Assembly
6 Wire_Cell_Box Assembly 8 Yoke_Box Assembly
W1 5 Motor1_Yoke Assembly
7 Motor1_Yoke _CB Assembly
3 Motor1_Sensor Assembly
9 Panel Assembly
W2 12 Delivery
11 Cure Time
10 Tack Free
Figure 2: Task Assignment to Work Stations 4.2 Results and Interpretation The base case provides an average cycle time for each assembly of 32.90 minutes. The presence of an additional Panel station will be enough to allow each arrival to be worked on, immediately at W2. This worker will then be available to perform his task after worker 1 has released the assembly. Also, this causes the worker 1 to be active,
Toba and Bedoya-Valencia approximately 49.53 % of the time while worker 2 will slightly have more work to do, with 49.93 % of busy-time. In the deterministic case, by the end of the shift, a total of 24 panels can be assembled. The simulation model shows a cycle time of 34.02 minutes for the base case scenario, with 26 panels assembled. Worker 1 is busy 51.23 % of the time and worker 2 is less busy than his co-worker, with 48.97%. This result does not push for the suppression of the additional Panel station, because it will create a waiting queue at the exit of the Yoke_Box station. The cycle time will dramatically increase, with an average of 87.76 minutes and only 18 panels assembled. The scenario 1 offers a cycle time is slightly reduced, 33.30 minutes. The busy-time percentages of the workers at W1 are 25.68% and 25.26%, which is approximately half the amount of work to do at this station, considering the fact that with one worker the percentage of work processing was 51.23%. As for W2, the worker in this case is almost as busy (49.13%) as in the base case. The throughput is 26 panels assembled. In scenario 2 the workers are twice as busy (See Figure 3). Worker 1 is working 51.48 % of the time, worker 2, 49.09 % and the worker at W2, 96.11%. Those numbers make sense, since there is twice as much work to do, and it is expected to see the two workers at W1 have their busy-time going double. The throughput is of 45 panels, with a cycle time of 78.11 minutes. This time is obviously too long, and it is due to panels queuing up at the entrance of W2. The three Panel stations are obviously enough to host the coming panels and avoid any queue, but the availability of the worker 3 is the issue. This worker is getting overwhelmed, and as indicated by his busy-time percentage, it appears to be more and more difficult for him to do his tasks on time.
Figure 3: Percentage of busy-time of workers at Scenarios 1 and 2 Scenario 3 appears to be of no interest, as an additional worker (worker 4) is added at station W2 when the demand is reduced back to 1 arrival per 18 minutes. The exact same numbers as in the Scenario 1 are found again. The model shows that the fourth worker added is of no use, because worker 3 is doing the entire job. The same statement can be made for one of the Panels and TackFree stations. In Scenario 4, the environment is still the same, with 4 workers, 3 Panel stations but now with 2 panel arrival at 18-minutes intervals. As suspected, the busy-times at Stations W1 and W2 will stay roughly the same, with 51.00% and 49.27% respectively for workers 1 & 2 as in scenario 2. The two workers at the three Panels stations are performing tasks on availability, meaning that whoever is idle takes the next job. Even with this assignment, they turn out to be quite busy; with 80.97% and 66.34% as shown in Figure 4. The cycle time is 51.97 minutes with a throughput of 48 panels. Judging by the fact that the cycle time still high, it would be reasonable to assume that there are still panels piling up, at the entrance of W2. The results of these scenarios are summarized in Table 1 below.
Toba and Bedoya-Valencia
Figure 4: Percentage of busy-time of workers at Scenarios 2 and 4
Scenarios Base Case Scenario 1 Scenario 2 Scenario 3 Scenario 4
Table 2: Expected Processing Times Cycle Time Arrival Mode Resources (Minutes) 1 Arrival /18 minute 1 Worker at W1 34.02 1 Panel /Arrival 1 Worker at W2 1 Arrival /18 minute 2 Workers at W1 33.3 1 Panel /Arrival 1 Workers at W2 1 Arrival /18 minute 2 Workers at W1 78.11 2 Panels /Arrival 1 Worker at W2 1 Arrival /18 minute 2 Workers at W1 33.3 1 Panel /Arrival 2 Workers at W2 1 Arrival /18 minute 2 Workers at W1 51.97 2 Panel /Arrival 2 Workers at W2
Throughput (Panels) 26 26 45 26 48
5. Conclusions The basic model is a simulating representation of a hypothetical assembly process. The different scenarios proposed involving eventual/potential changes in the system were designed to test the system strength, measure the stochastic influences and try to achieve the goal originally set. Considering the results obtained in the previous section, the use of DES appears to be a very good approach, as far as flexibility and variability of the system is concerned. It can test the limits of the systems and eventually try to extend them by adding or removing additional requirements. Assembly Line Balancing (ALB) was also applied to deal with the resource management. The resource utilization was modified for testing purposes, so as to save money. Idle time, workers utilization percentage and throughput are the measures of performance addressed, to select the best option. This simulation model appears to be an efficient tool for helping to spot bottlenecks and preventing any unwanted situations by analyzing several process performances. It provides several options to go over, in a timely manner, and gives the opportunity to design appropriate solutions that will allow a CPV panel production company to be more competitive. In this study, ALB was used to take care of the workforce and activities management. Different levels of human resources were analyzed by using the simulation model and the results show that we typically obtain a better performance measures in Scenario 4 of ALB.
Toba and Bedoya-Valencia
2. 3. 4. 5. 6. 7. 8. 9. 10.
11. 12. 13. 14. 15.
16. 17. 18. 19. 20.
Das B., Sanchez-Rivas J. M., Garcia-Diaz A., and McDonald C. A., 2010, “A computer simulation approach to evaluating assembly line balancing with variable operation times,” Journal of Manufacturing Technology Management, 21(7), 872-887. Salveson, M. E., 1955, “The assembly line balancing problem,” Journal of Industrial Engineering, 6(1), 1825. Jackson, J.R., 1956, “A computing procedure for a line balancing problem,” Management Science, 2(3), 261-271. Bowman, E. H., 1960, “Assembly-line balancing by linear programming,” Operations Research, 8(3), 385389. Supnick, F, and Solinger, J., 1960, “An external production line problem,” Operations Research, 8(3), 381384. White, W. W., 1961, “Comments on a paper by Bowman,” Operations Research, 9(2), 247-277. Hu, T. C., 1961, “Parallel sequencing and assembly line problems,” Operations Research, 9(6), 841-848. Gutjahr, A. L. and Nemhauser, G. L., 1964, “An algorithm for the line balancing problem,” Management Science, 11(2), 308-315. Patel, V., Ashby, J. and Ma, J., 2002, “Discrete event simulation in automotive Final Process System,” Proc. of the 2002 Winter Simulation Conference, December 8-11, San Diego, California, 1030-1034. Choi, S. D., Kumar, A.R. and Houshyar A., 2002, “A simulation study of an automotive foundry plant manufacturing engine blocks,” Proc. of the 2002 Winter Simulation Conference, December 8-11, San Diego, California, 1035-1040. Ali, S. A., Seifoddini H. and Sun H., 2005, “Intelligent modeling and simulation of flexible assembly systems,” Proc. of the 2005 Winter Simulation Conference, December 4-7, Orlando, Florida, 1350-1358. Bukchin, J. and Rubinovitz, J., 2003, “A weighted approach for assembly line design with station paralleling and equipment selection,” IIE Transactions, 35(1), 73-85. Masood, S., 2006, “Line balancing and simulation of an automated production transfer line,” Assembly Automation, 26(1), 69-74. Templemeir, H., 2003, “Practical considerations in the optimization of flow production systems,” International Journal of Production Research, 41(1), 149-170. Vidalis, M. J., Papadopoulos, C. T. and Heavey, C., 2005, “On the workload and ‘phaseload’ allocation problems of short reliable production lines with finite buffers,” Computers & Industrial Engineering, 48(4), 825-837. Hillier, F. S. and Boling, R. W., 1966, “The effect of some design factors on the efficiency of production lines with variable operation times,” Journal of Industrial Engineering, 17(12), 651-658. Shaaban, S. and McNamara, T., 2009, “Improving the efficiency of unpaced production lines by unbalancing service time means,” International Journal of Operation Research, 4(3), 346-361. Nahmias H., 2009, Production and Operation Analysis, 6th Edition, McGraw-Hill, Boston. Niebel, B.W. and Freivalds, A., 2003, Methods, Standards and Work Design, 11th Edition, McGraw-Hill, Boston. Sargent R. G., 2007, “Verification & Validation of Simulation Models,” Proc. of the 2007 Winter Simulation Conference , December 9-12, Washington D.C., 124-137.
Please copy and paste this embed script to where you want to embed