Uncertain fault estimation using bicausal bond graph: Application to intelligent autonomous vehicle

This article addresses the fault detection and isolation problem of uncertain systems using the bond graph model–based approach. The latter provides through its causal and structural properties an automatic analytical redundancy relations generation. The numerical evaluation of analytical redundancy relations yields residuals, which are used to verify the coherence between the real system and reference behaviors describing the nominal operation. In fact, the residual is compared to its thresholds to detect the fault. In addition, the comparison between all fault signatures allows making a decision on fault isolation. Moreover, to isolate the faults that activate the same set of residuals, an additional residual must be calculated for each fault. This additional residual is the comparison between two estimations of the considered fault obtained using the sensitivity relations. However, due to the presence of uncertainties, errors can occur in the fault estimation allowing false decisions on fault isolation. The novelties and innovative interests in the present work are (1) to improve the fault estimation procedure based on the uncertainties modeling and bicausality notion, in order to overcome the problem related to errors in the estimated fault and (2) to suitably generate the isolation thresholds in a systematic way using the uncertain fault estimation procedure proposed in this article so that fault can be isolated successfully. The proposed methodology is studied under various scenarios via simulations over an electromechanical traction system corresponding to a quarter of intelligent autonomous vehicle, named RobuCar.