Statistical features-ANN recognizer for bivariate process mean shift pattern recognition

Artificial neural network (ANN)-based recognizers have been developed for monitoring and diagnosis bivariate process mean shift in multivariate statistical process control (MSPC). They have better average run lengths (ARLs) performance in monitoring process mean shifts and gave an useful diagnosis information compared to the traditional MSPC schemes such as T², multivariate cumulative sum (MCUSUM) and multivariate exponentially weighted moving average (MEWMA). The existing recognizers are raw data-based, whereby raw data input representation were applied into ANN. This approach required in a large network size, more computational effort and training time consuming. In this paper, the statistical features input representation was investigated, whereby the raw data were transformed into exponentially weighted moving average, multiplication of mean with standard deviation and multiplication of mean with mean-square value. The statistical features-ANN recognizer resulted in smaller network size, fast training time, better ARLs for monitoring process mean shifts and comparable recognition accuracy for diagnosing the source variable(s) compared to the raw data-ANN recognizer.