Robust Fault Diagnosis using a Data-based Approach and Structural Analysis

Abstract

This paper presents a fault diagnosis approach that combines structural and data-driven techniques. The proposed method involves two phases. As a first step, the residuals structure is obtained from the structural model of the system by using structural analysis without considering mathematical models (only the component description of the system). Secondly, the analytical expressions for residuals are derived from available historical data using a robust identification approach. Through adaptive nets, residuals are adjusted by determining an interval model that takes into account the uncertainties and noises affecting the system. In the diagnosis part, residuals are tracked and evaluated. The presence of inconsistent residuals can be regarded as a fault, therefore thresholds for each residual are introduced. In addition to detecting faulty scenarios, it is also possible to determine which is the most likely fault that occurred in the system. To accomplish such classification, the proposed approach implements a Bayesian reasoning that uses the FSM (Fault Signature Matrix) that is obtained from the structural analysis of the system and residual activation signals. A brushless DC motor (BLDC) is used as a case study to illustrate the proposed approach. Simulation experiments illustrate the overall performance.