Rao-Blackwellised particle filtering for fault diagnosis

Abstract

We tackle the fault diagnosis problem using conditionally Gaussian state space models and an efficient Monte Carlo method known as Rao-Blackwellised particle filtering. In this setting, there is one different linear-Gaussian state space model for each possible discrete state of operation. The task of diagnosis is to identify the discrete state of operation using the continuous measurements corrupted by Gaussian noise. The method is applied to the diagnosis of faults in planetary rovers.