Real-time fault detection and isolation with supervised training

Abstract

This chapter focuses on fault detection and isolation (FDI) in a stochastic nonlinear system affected by noise. The substance of the presented Bayesian identification based FDI methodology is a probabilistic model determined by the fault probability table (FTP). The methodology consists in probabilistic mapping of the measured data into a fault variable. Necessary computations are very simple, and heuristic knowledge about the fault can also be easily included in real time. The practical aspects of the proposed FDI algorithm are tested in real time on a laboratory heating system. The results obtained proved suitability of the designed algorithm for fault detection and isolation in real plants.