New Pca-based Methodology for Sensor Fault Detection and Localization

This work proposes a new methodology for sensor fault detection and localization using principal component analysis (PCA). A new index is proposed in order to detect simple and multiple faults affecting the dependent and independent process variables. A new iterative selection method of principal component number is presented. This method determines a model allowing the detection of faults without a priori knowledge of their natures. The fault localization is carried out using hierarchical contribution plots applied to the proposed detection index. The performance of this approach becomes poor for a bad partitioning of the variables into blocs. A new partitioning method is proposed to identify correctly all the faults affecting the process. The whole proposed results were applied to a non linear noisy system subjected to simple and multiple faults.