Multivariate Change Point Estimation in Covariance Matrix Using ANN

In statistical process control, change point estimation is an essential requirement for diagnosing the source of a deviation when a process is out of control. In this study, an ANN- based method is proposed to estimate the change point in the multivariate normal process which is subjected to covariance variation. Since in a physical system parameter is correlated, correlation is kept constant to obtain realistic simulated data. Employing statistical simulation, different out of control scenarios are simulated and statistics are calculated for each scenario. This study is to predict the change point in the control chart using the simulated set and corresponding statistical sets, an ANN is adopted. The resulting model achieved a high accuracy of 90% in training and 80% testing with a reasonable level of confidence in the prediction. Also, results show that Bayesian reaches a higher accuracy than Levenberg in ANN training.