Diagnosis of nonlinear systems using reduced kernel principal component analysis

Abstract

In this article, we consider a sensor fault detection and identification procedure based on reduced kernel principal component analysis (KPCA). The fault detection based on reduced KPCA method consists first to reduce the amount of training data using K-means clustering in the input space while conserving the structure of the data in the feature space. Then, it consists to built KPCA model and use it for fault detection. The proposed fault identification based on reduced KPCA uses reconstruction-based contributions to identify and estimate the fault using the reduced KPCA model. The proposed fault detection and identification methods are tested with a simulated CSTR process. The simulation results show that the proposed fault detection and identification methods are effective for KPCA.