MULTIVARIATE STATISTICAL PROCESS MONITORING USING MULTI-SCALE KERNEL PRINCIPAL COMPONENT ANALYSIS

A monitoring method based on multi-scale kernel principal component analysis (MSKPCA) is proposed for nonlinear dynamic processes, by combining wavelet analysis with nonlinear transformation using kernel principal component analysis. Wavelet analysis is used to analyze dynamic characteristic of process data, while the kernel principal component analysis is to capture nonlinear principal components by kernel functions. This method can simultaneously extract cross correlation, auto correlation, and nonlinearities from the data. Furthermore, a multi-scale principal component analysis similarity factor is proposed for fault identification. Simulation of CSTR process shows that the proposed method outperforms the traditional PCA method in fault detection and diagnosis.