A hybrid method for estimating the process change point using support vector machine and fuzzy statistical clustering

Abstract

Control charts are the most popular process monitoring techniques designed to determine whether a process is in a state of statistical control or not. When a process change occurs, the control chart exhibits an out-of-control signal. In most cases, the signal is followed by a substantial amount of delay. To address this drawback, supplementary techniques have been considered to be employed along with the control charts to identify the exact time of the process change. This paper presents a hybrid method for estimating the change point on x¯ chart when neither the change type nor its magnitude are known. For this purpose, two sets of features with the most discriminatory power between x¯ chart patterns are selected. Then the feature vectors are extracted from the control chart patterns (CCPs) and served as an input to a classification scheme which is comprised of several support vector machine (SVM) classifiers. After parameters tuning, the classifiers are built and used for classifying the CCPs and identifying the change type. Once the change type is determined, the fuzzy statistical clustering (FSC) and the maximum likelihood (ML) estimators are employed to identify the process change point. The performance of selected features, the classification scheme, and the change point estimators are evaluated by conducting several simulation studies. Empirical results show that in identifying the change types, the proposed classification scheme is more accurate than two recent CCP classification methods. The results also confirm that the proposed hybrid method offers an accurate estimate of the process change point, as compared to the most recent methods developed for the change point estimation.