Statistical features-MLP neural network for recognizing bivariate SPC chart patterns

Abstract

Statistical process control (SPC) charting is an important tool in monitoring, controlling and minimizing unnatural variation in manufacturing process. Nevertheless, its application has become more challenging when involves two correlated quality variables (bivariate) since the conventional multivariate SPC charting scheme is only efficient for triggering unnatural variation but unable to interpret the source of disturbance. Therefore, SPC pattern reStatistical process control (SPC) charting is an important tool in monitoring, controlling and minimizing unnatural variation in manufacturing process. Nevertheless, its application has become more challenging when involves two correlated quality variables (bivariate) since the conventional multivariate SPC charting scheme is only efficient for triggering unnatural variation but unable to interpret the source of disturbance. Therefore, SPC pattern recognition technique has been proposed for solving this issue. In this study, a new SPC pattern recognition scheme is proposed for monitoring and diagnosing nine category of unnatural variation in bivariate cases. Based on single-stage monitoring and diagnosis approach, a generalized multilayer’s perceptron model with statistical features input representation was applied as the pattern recognizer. The proposed scheme has provided rapid triggering capability and high accuracy in interpreting the sources of unnatural variation, especially in dealing with moderate and large magnitude of unnatural variation. The selection of the statistical features as the input representation is very important in determining the effectiveness of the recognizer. This study has opened a new perspective in quality controlcognition technique has been proposed for solving this issue. In this study, a new SPC pattern recognition scheme is proposed for monitoring and diagnosing nine category of unnatural variation in bivariate cases. Based on single-stage monitoring and diagnosis approach, a generalized multilayer’s perceptron model with statistical features input representation was applied as the pattern recognizer. The proposed scheme has provided rapid triggering capability and high accuracy in interpreting the sources of unnatural variation, especially in dealing with moderate and large magnitude of unnatural variation. The selection of the statistical features as the input representation is very important in determining the effectiveness of the recognizer. This study has opened a new perspective in quality control.