Control chart pattern recognition of multivariate auto-correlated processes using artificial neural network

Abstract

The objective of this study is to model a control chart pattern recognition method for multivariate auto-correlated processes. The model development process uses a multi-layer feed forward Artificial Neural Network (ANN) architecture governed by back-propagation learning rule to identify and classify a set of sub-classes of abnormal patterns. Network training was conducted using simulated Control Chart Patterns (CCP) with Monte-Carlo simulation technique. A total of 3500 CCP examples (500 CCPs for each type of pattern) were generated and all CCPs data are normalized (standardized) before being employed as input to the neural network for better performance of the network. With this the study proposes a model for control chart pattern recognition of multivariate autocorrelated statistical process control to identify and classify seven types of typical control charts patterns: i.e. normal, downward shift, upward shift, decreasing trend, increasing trend, cyclic, and systematic patterns. The proposed framework is effective in control chart pattern recognition of multivariate autocorrelated processes with 94.9% recognition accuracy. Furthermore, the Control Chart Pattern Recognition (CCPR) model is validated by the data which is obtained from in-control process of a factory producing Alcohol and Liquor which demonstrates the accuracy of the CCPR. Pattern recognition for multivariate processes is common in the literature. But, this study proposed a new model for multivariate auto-correlated processes.