Improved Incipient Fault Detection Using Jensen-Shannon Divergence and KPCA

In most of Statistical Process Control applications, multivariate information is available to monitor a system and evaluate its health. The use of Principal component analysis (PCA) as a dimension reduction and feature extraction technique in fault diagnosis has shown its advantages. However, its feature extraction capability is not sufficient enough for the nonlinear cases. In our paper, Kernel principal component analysis is applied for improving the feature extraction performance of PCA. To analyse these features, Jensen-Shannon divergence (JSD) which is known to be efficient for slight changes diagnosis is used here for incipient fault detection using KPCA. Our proposal's detection capabilities for incipient faults are evaluated and validated by comparison with the JSD ones in the PCA framework through an auto-regressive (AR) system. Then, the proposed method is addressed to the incipient fault diagnosis in Tennessee Eastman Process (TEP). Its superior detection performances are proved by comparing them with those obtained using other methods already published in the literature.