Study On The Performance Factor And Performance Evaluation Of Multi-product Re-entrant Manufacture Systems

Re-entrant manufacturing systems is a typical representative of the semiconductor manufacturing system.With the increasingly fierce competition in semiconductor market,customer demand for personalized is becoming increasingly obvious.Semiconductor manufacturing factories are faced the transformation from “small classes,mass production” mode to “multiple classes,small scale production “mode.The factories need to provide personalized products to meet the needs of customers as soon as possible.In order to respond to the changes,a more comprehensive and systematic understanding of the performance of multi-product re-entrant production systems is required.The establishing of local or global analysis model can be performed to assist the short-term control decisions and timely adjust configuration parameters to control.The purposes of this thesis is to analyze the factors of performance and evaluate the performance of the multi-product re-entrant systems.By researching single-product influence indicators of the performance,new indicators previously possessed rarely by multi-product manufacturing systems are acquired.Through training the model with simulation data collected from a typical multi-product systems and historical data by Random Forest,we determine the extent to which these indicators impact the performance of multi-product manufacturing systems and explore the relevant rules and mechanisms.The proportion of different products and different processing times for different products are found to significantly impact the throughput in addition to the feeding strategy and dispatch rules.As our primary goal is to build a highly effective evaluation model of the actual multiproduct systems.We divide the target system into open feeding and closed feeding systems according to different control objectives and nature of systems;we then establish performance prediction models for the respective subsystems.In the open system,the rate of the output is equal to the rate of the input at a steady state,so truncating the cycle time is consistent with reducing the number of products in the system and throughput can be predicted accordingly.The output and input of the system are not the same when the system is in an unsteady state,however.The throughput boundary can be estimated according to this paper.In the closed system,the number of lots is fixed under a certain limitation so there is no need to consider the stability of the system.We use an improved Mean Value Analysis method to estimate system performance in a parallel machine environment with the cycle time and output rate as the control target.By comparing our results against Park's simulation results,we are able to verify that the modified MVA is effective.In this study,we first introduce the factors that only exist in the multi-product manufacturing system and estimate the importance of those indicators via machine learning on a large set of simulation data,to this effect,the results presented here may extend the theoretical application of multi-product re-entrant manufacturing systems.We then compare those importance indictors to establish their rules and mechanisms as they work within the multiproduct manufacturing system.Finally,we establish local or global analytical models that can be used to facilitate rapid analysis of the system to set reasonable parameters.The results may provide useful insight for personnel tasked with managing the operations of multi-product systems.