|M.Sc Student||Roitman Hernan Andres|
|Subject||Release and Dispatching Policies for Large Reentrant Job|
|Department||Department of Industrial Engineering and Management||Supervisors||Mr. Michael Masin|
|Mr. David Sinreich (Deceased)|
|Full Thesis text|
In this research we develop new Production Control Systems (PCSs) for large job shops with reentrant flows and stochastic operations. The model captures the essential features of many production facilities from which we consider in particular semiconductor manufacturing. The PCS in study focuses on the following features: (1) integration of the Release (“starts”) strategy and the Dispatching rule, (2) scalability to large production systems and (3) self-regularity and robustness to random events. A multi-objective approach is applied for minimizing Mean and Variance of Cycle (flow) Time subject to a required throughput constraint in a stochastic reentrant job shop environment. We propose generic workload release policies and develop an optimization procedure for determining their parameters. We compare the new policies to the main release controls proposed in the research: Constant time between arrivals, CONWIP, CONTIP (constant time in process) and Workload regulation. As well, we suggest a new dispatching rule called Load and Delay Balance (LDB), and test it against twelve rules: Benchmark priority rules, Fluid based rules and dispatching rules recently proposed for specifically for semiconductor facilities in order to improve the cycle time performance. A simulation experiment is conducted on different multi-product line configurations. In particular, we test a real-size semiconductor facility model. We apply an efficient frontier approach to compare the different PCS and we analyze the results statistically. The research shows that release and dispatching policies have a distinct effect on the performance. The effect of release policies is usually higher than the effect of dispatching. The degree of potential improvement over standard control policies depends on the line features. The proposed release policies significantly outperform the existing controls in all the lines and performance measures tested. They can improve the other releases by tens of percents reduction in the cycle time mean and standard deviation without affecting the throughput. There is no single scheduling rule that is optimal for all line configurations and the performance of the dispatching rules is usually very sensitive to the line characteristics. However, LDB shows the more robust results as is the only rule that outperforms FIFO in all cases and it belongs to the (statistically indifferent) group of the best or next to the best rules regarding the mean cycle time objective. Also, in almost all the lines LDB significantly outperforms the other rules on respect to the cycle time variance.