분류와 예측에 기반한 자기상관 공정 모니터링 절차의 성능 비교

초록

There has been extensive research on the procedures for monitoring autocorrelated processes. Among them, the most commonly used approach is to forecast the next observation based on a fitted model, calculate residuals, and apply control charting procedures to the residual data. In this paper, we propose a process monitoring procedure based on a recurrent neural network (RNN) to classify whether the process is in control or out of control. The performance of this procedure is compared with the forecasting procedure based on a RNN and the traditional residual control charting procedure through simulation study. The results show that the RNN-based classification procedure quickly detects changes in the process level, and the RNN-based forecasting procedure quickly detects changes in the process variance. Additionally, unlike the traditional monitoring procedure, the RNN-based procedures have the advantage that they do not require accurate model fitting for process data.

키워드