Gaussian mixture models-based control chart pattern recognition

Abstract

Abnormal patterns exhibited in control charts can be associated with certain assignable causes for process variation. Hence, accurate and fast control chart pattern recognition (CCPR) is essential for significantly narrowing down the scope of possible causes that must be investigated, and speeds up the troubleshooting process. This study proposes a Gaussian mixture models (GMM)-based CCPR model that employs a collection of several GMMs constructed for CCPR. By using statistical features and wavelet energy features as the input features, the proposed CCPR model provides a more simple and effective training procedure and better generalisation performance than using a single CCPR recogniser, and hence is easier to be used by quality engineers and operators. Furthermore, the proposed model is capable of adapting novel control chart patterns (CCPs) by applying a dynamic modelling scheme. The experimental results indicate that the GMM-based CCPR model shows good detection and recognition performance for current CCPs and adapts further novel CCPs effectively. Moreover, the proposed model provides a promising way for the on-line recognition of CCPs because of its efficient computation and good pattern recognition performance. Analysis from this study provides guidelines for developing GMM-based statistical process control (SPC) recognition systems.