Sensor and Actuator Fault Diagnosis for a Multi-Robot System Based on the Kullback-Leibler Divergence

Abstract

In this paper, a fault diagnosis method based on information theory is proposed to detect and isolate faulty sensors and actuators in a multi mobile-robot system. The proposed diagnosis method is an observer-based residual generation approach, where residuals are generated using the Kullback-Leibler Divergence (KLD), which is a dissimilarity measure between two probability distributions. Information form of the Kalman Filter is used for multisensor data fusion. Odometers are used to obtain the predicted estimation, while the remaining sensors give the updated estimation. For fault detection and isolation, a bank of filters is used, and residuals are obtained by evaluating the KLD betweeen the predicted and updated estimations of each filter. In order to isolate actuator and odometer faults, closed loop control signals are designed to feed an additional bank of filters. The combination of the two banks of filters within an informational framework allows a better isolation of sensor, actuator and odometer faults. A simulation is performed to demonstrate the efficiency of the proposed method.