Process Fault Detection and Diagnosis Based on Principal Component Analysis

Conventional process fault detection and diagnosis technique need an in-depth comprehension and mastery in process mechanism models, which have to obtain very particular process transcendental knowledge and various physical and chemical parameters. It is very time-consuming and difficult for actual production processes. A novel process fault detection and diagnosis technique based on principal component analysis (PCA) is presented and discussed. The proposed method reduces the dimensionality of the original data set by the projection of the data set onto a smaller subspace defined by the principal components through PCA, and the multivariate statistical process control charts, for example, Hotelling T², Q and contribution charts are used to detect and diagnose the process faults. The monitoring performance of the proposed method to a typical continuous production process indicates that the fault diagnosis model constructed by PCA can efficiently be used to extract the main variable information of original data set independent of the process mechanism, and detect the abnormal change of the process