Reduced-complexity Rao-Blackwellised Particle Filtering for fault diagnosis

We explore an approach for complexity reduction for fault diagnosis problems. The underlying algorihtm is a sequential Monte Carlo method known as the Rao-Blackwellised Particle Filter or RBPF. In this paper, we show that the complexity of the RBPF algorithm can be reduced by applying the Kalman updating step to only one representative particle of a group particles gathering in a particular state. The time consumption for the algorithm to complete computation is substantially reduced especially for systems employing large number of particles. Simulation results reveal that the performance of the proposed reduced-complexity algorithm or RC-RBPF in terms of percentage estimation errors is identical to that of the standard RBPF which in turn much better than that of the particle filtering.