Decentralized Particle Filtering Based Fault Diagnosis for Nonlinear Distributed and Interconnected Systems

The problem of decentralized fault diagnosis for nonlinear, non-Gaussian systems composed of spatially distributed and physically interconnected subsystems, is considered. Normal and faulty behaviours are modelled as two hypotheses with two different models. Accordingly, the general system is modeled as a hybrid system with two faulty and healhty modes. Then, it is decomposed into hybrid interconnected subsystems. Particle filtering approach is employed for nonlinear state estimation as well as non-Gaussian probability distribution estimation. Faulty or health modes are estimated using Maximum A Poserior Estimators (MAP). The effectiveness of the proposed distributed algorithm is demonstrated through applying to a quadruple tank process.