A novel efficient control chart to monitor the mean vector and variance-covariance matrix using support vector machine and tuned kernel function

Abstract

Control chart is a popular statistical tool for continuous process monitoring and control which has been extensively extended to be applied. Hence current research addresses to a specific control chart based on one of the most important data-driven models, support vector machine. Such multivariate control chart is a useful learning system based on constrained optimisation theory that uses induction of structural error minimisation principle and results a general optimised answer. The novel recommended control chart gives the ability to detect out of control variable according to the appropriate decision functions. Computational results are conducted using Hotelling T2 chart employing average run length to determine the superiority of the proposed method. Implementing different shifts on process mean and variance in two and three variable processes reveals that the proposed chart has superior run length comparing by traditional Hotelling T2 chart. Furthermore, detecting the source of out of control variable is assessed.