利用類神經網路於多重輸入多重輸出之製程管制系統

Construct MIMO Process Control System by Using Neural Network

孫任東(Jen-Tong Sun)

指導教授 ： 江瑞清

中原大學/工學院/工業工程研究所/碩士(2003年)

https://doi.org/10.6840/cycu200300246

摘要

本研究結合整合SPC與EPC、回饋控制、以及類神經網路（Artificial Neural Network）的概念及方法，建構一多重輸入多重輸出（Multiple-Input Multiple-Output；MIMO）之製程管制系統，其涵蓋MIMO製程子系統、量測子系統、偏移偵測與分析子系統、類神經網路製程輸出預測子系統、以及參數調整子系統等五個部分。其中製程輸出預測子系統及參數調整子系統的核心分別為BPNN-1與BPNN-2兩個倒傳遞網路模式，而類神經網路之最適網路參數設定則以田口式品質工程方法（Taguchi Method）決定。

本研究以BPNN-1預測製程下一時間點之多重輸出值，將預測結果輸入BPNN-2以獲得製程之多重輸入變數調整量，透過EPC的技術對MIMO製程進行回饋調整。並以整合SPC與EPC之方式，同時監控製程是否產生可歸屬變異，可有效掌控製程變異及避免EPC的過度控制情形發生。本研究之驗證部分以晶圓製程中之化學機械研磨（Chemical Mechanical Polishing；CMP）為驗證對象，結果證明所建構之MIMO製程管制系統，在製程受偏移干擾下，能使製程之多重品質特性輸出值有效維持於目標值並縮小變異。

關鍵字

多重輸入多重輸出 ； 倒傳遞網路 ； 製程管制系統

並列摘要

In this research, a multiple-input multiple-output (MIMO) process quality control system is constructed by applying several techniques, such as engineering process control (EPC), statistical process control (SPC), feedback control and artificial neural network (ANN). This MIMO process control system includes five parts: MIMO manufacturing process subsystem, measurement subsystem, disturbance detecting and analyzing subsystem, process multiple-output forecasting subsystem and parameter adjustment subsystem. The back-propagation neural networks, BPNN-1 and BPNN-2, are the basic models of process multiple-output forecasting subsystem and parameter adjustment subsystem. In this study, we apply Taguchi methods to choose the best parameters of neural networks.

In the MIMO Process control system, we use BPNN-1 to forecast the multiple-output values of the next lot. Once the forecasts are got, we estimate the multiple-input parameters of MIMO process system by using BPNN-2. Consequently, we give feedback to adjust the process parameters by applying the theory of EPC. SPC is applied simultaneously to detect if assignable causes occurs and avoid the over control in EPC. Finally, in this study, we create the simulated process data of chemical mechanical polishing (CMP) process and use the data to validate and verify this process control system. The result shows that the MIMO process control system can reduce the variation from target.

並列關鍵字

Multiple-Input Multiple-Output ； Back-propagation Neural Network ； Process Control System

參考文獻

31. 林怡彣，2002，「建構多重輸入、多重輸出之EPC模式」，中原大學工業工程研究所碩士論文。

1. Box, George E. P., and Tim Kramer, 1992, “Statistical Process Monitoring and Feedback Adjustment - A Discussion,” Technometrics, pp.251-267

2. Box, George E.P., 1993, “Process Adjustment and Quality Control,” Total Quality Management , 4, pp.215-227

4. Cheng, C. S., 1995, “A Multi-layer Neural Network Model for Detecting Changes in the Process Mean,” Computers and Industrial Engineering, Volume: 28, Issue: 1, pp. 51-61

5. Collani, Elart Von; 1997, “A Discussion on Statistically based Process Monitor and Control-Individual Contributions,” Journal of Quality Technology, Vol. 29, No. 2, pp. 148-156