Recursive kernel PCA-based GLRT for fault detection: Application to an air quality monitoring network

This paper aims to improve the use of generalized likelihood ratio test (GLRT) method for fault detection. To achieve this objective, nonlinear fault detection method will be developed. Kernel principal component analysis (kPCA) models have been widely used to represent nonlinear systems. KPCA models rely of transforming the data in a linear form to a higher dimensional spacee. Unfortunately, kPCA models are batch, i.e., they require the availability of the process data before constructing the model. In most situations, however, fault detection is needed online, i.e., as the data are collected from the process. Therefore, recursive kPCA fault detection technique will be developed in order to extend the advantages of the GLRT to online processes. The fault detection performances of the recursive kPCA-based GLRT technique are shown using air quality monitoring network (AQMN). The results showed the effectiveness of the developed algorithm over conventional method.