Probabilistic Fault Diagnosis for Nonlinear Stochastic Systems

The paper focuses on fault detection and isolation (FDI) in stochastic nonlinear systems using Bayesian statistics. The main concept of this approach leans on a probabilistic model, which estimates the occurrences of faults by probabilities. The methodology consists in probabilistic mapping of the measured data into a fault variable, which acts as an indicator of considered modes of the systems. One of the main advantages of the proposed FDI approach is the ability to perform realtime supervised training by which the parameters of the fault probability table are updated in real time. The suggested methodology is very simple for numerical calculations and enables one to include heuristic knowledge about the faults. The practical aspects of the proposed FDI algorithm were successfully tested in real time using a laboratory heating system.