Interacting multiple particle filters for fault diagnosis of non-linear stochastic systems

In this paper, an approach to fault diagnosis in a nonlinear stochastic dynamic system is proposed using the interacting multiple particle filtering (IMPF) algorithm. The fault diagnostic approach is formulated as a hybrid multiple-model estimation scheme. The proposed diagnostic approach provides an integrated framework to estimate the system's current operational or faulty mode, as well as the unmeasured state variables in the system. Particle filtering algorithm is used to statistically model the underlying dynamics of a nonlinear/non-Gaussian stochastic system. A set of models is assumed to present the possible system behavior pattern or modes. A bank of particle filters runs in parallel, each based on a particular mode, to obtain mode-conditional estimates according to the probabitisticafly weighting scheme. The interaction among particle filters allows estimation from multiple filters to be fused in a principled manner. The simulation results on a highly nonlinear system are provided which demonstrate the effectiveness of the proposed method by comparing it with other nonlinear estimation techniques (extended Kalman filter (EKF) and unscented Kalman filter (UKF)-based).