Partial kernel PCA-based GLRT for fault diagnosis of nonlinear processes

In this paper, a novel fault detection and isolation (FDI) framework based on kernel PCA (KPCA) and generalized likelihood ratio test (GLRT) that is capable of detecting and identifying faults is developed. Specifically, three main objectives are addressed. First, system model identification and residuals generation are addressed using KPCA model. Second, KPCA-based GLRT method is proposed to detect different types of faults in the systems. Third, partial KPCA (PKPCA)-based GLRT is developed for fault isolation. The proposed approach aims to apply a structured PKPCA-based GLRT to a set of sub-models. The fault detection and isolation performances using PKPCA-based GLRT are illustrated through two examples: A simulated continuous stirred tank reactor (CSTR) data and an air quality monitoring network data. The obtained results demonstrate the effectiveness of the partial KPCA-based GLRT method over the partial PCA-based GLRT method.