DSmT-Based Fusion System for Fault Detection and Identification in Industrial Process

Multi-Classifier fusion strategies have been widely applied in Fault Detection and Identification (FDI) problems. However, the outputs obtained by multi-classifier systems are often uncertain, imprecision or even conflicting. In this paper, we use the Dezert-Smarandache Theory (DSmT) to propose a novel fusion framework for FDI in industrial process, which can effectively improve the accuracy of detection and identification. Specifically, the training data of each feature are used to build a Kernel Density Estimation (KDE)-based model for each class. Then, the most discriminative model is selected and regarded as the corresponding template for each class. After that, a structure of Basic Belief Assignment (BBA) can be constructed, using the relationship between the test data and the selected KDE models of all considered classes. In order to eliminate the conflict between the obtained BBAs, Proportional Conflict Redistribution rule-5 (PCR5) is applied to fuse these acquired BBAs. Performances of our proposed method in this paper are evaluated through the Tennessee Eastman Process (TEP).