Monitoring and Characterizing the Process Mean Shifts by Artificial Neural Networks

• Main Authors: Wei-Chun Wan, 萬維君
• Other Authors: Chuen-Sheng Cheng
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/69339239068340394824

碩士 === 元智大學 === 工業工程研究所 === 89 === Control charts are the most commonly used tools for monitoring process mean shifts in manufacturing and service industries. Besides the use of control charts in monitoring process and identifying assignable causes, quality practitioners frequently need to adjust processes based on the magnitude of change. Once the control chart issues an out-of-control signal, an immediate step to identify assignable causes further is to estimate the magnitude of a specific process change. A process control method will be more effective if the change magnitude can be estimated when a process mean has changed. In this study, we propose a neural network-based approach to monitor the process mean shifts and to predict the magnitudes of shifts. Two models have been developed and compared. In the first model, a back-propagation neural network is used to detect process mean shifts and to estimate the magnitude of shifts. The second model is a two-stage approach. At the first stage, a back-propagation neural network will be used to detect process mean shifts. At the second stage, a second back-propagation neural network is used to estimate the magnitude of shifts as soon as the first neural network signals an out-of-control situation. The performances of neural networks were evaluated by estimating the average run lengths (ARL''s) and mean absolute percent errors using simulation. The results obtained with simulated data suggest that neural networks outperform CUSUM charts in terms of ARLs and estimation capabilities. The results also indicate that the second model has better estimation accuracy. The unique feature of this research is a novel approach to generate the training data set. Some numerical examples are presented which demonstrate that the proposed neural networks are applicable to practical manufacturing processes.