New Method for Multivariate Statistical Process Monitoring

A new method using discriminant analysis and control charts is proposed for monitoring multivariate process operations more reliably. Fisher discriminant analysis (FDA) is used to derive a feature discriminant direction (FDD) between each normal and fault operations, and each FDD thus decided constructs the feature space of each fault operation. Individuals control charts (XmR charts) are used to monitor multivariate processes using the process data projected onto feature spaces. Upper control limit (UCL) and lower control limit (LCL) on each feature space from normal process operation are calculated for XmR charts, and are used to distinguish fault from normal. A variation trend on an XmR chart reveals the type of relevant fault operation. Applications to Tennessee Eastman simulation processes show that this proposed method can result in better monitoring performance than principal component analysis (PCA)-based methods and can better identify step type faults on XmR charts.