Industrial Process Fault Detection Using Singular Spectrum Analysis and Kernel Principal Component Analysis

A nonlinear fault detection method based on Singular spectrum analysis (SSA) and Kernel principal component analysis (KPCA) is proposed for detecting faults in nonlinear dynamic processes. In the proposed method KPCA is used for the nonlinear transformation of multiscale signals obtained through SSA decomposition. SSA is a data-adaptive multimodal method based on singular value decomposition, which is a promising tool to capture the dynamic characteristics of process data, while the KPCA is used to extract nonlinear features in the process using kernel functions. Thus the proposed method takes cross correlation, auto correlation and nonlinearities of the data into account. Simulation of the continuous stirred tank reactor (CSTR) process shows that the proposed method outperforms the conventional SSA method in fault detection.