An Improved PCA-based Fault Detection Method for non-Gaussian Systems Using SIP Criterion

Abstract

In the past, many statistical process monitoring methods based on principal component analysis (PCA) have been drawn up and applied to various chemical processes. However, these methods were almost proposed under the assumption that the monitored process data obeys gaussian distribution. In this paper, aiming at the non-gaussian characteristics of the system, an improved PCA-based fault detection method is presented, where the survival information potential (SIP) is used to characterize the non-Gaussian randomness of the process data. Firstly, an improved PCA method, called SIP-PCA is proposed by using SIP to minimize the reconstruction error from the perspective of both randomness and magnitude. Then, the SIP-PCA based fault detection strategy is presented by using the kernel density estimation method to determine the control limit of fault detection indicators. Finally, the proposed method is applied to a continuous stirred tank reactor (CSTR) process with non-Gaussian disturbances. The comparative monitoring results illustrate that the proposed method is more reliable and effective than the traditional PCA based fault detection methods.