On-Board Real Time Failure Detection and Diagnosis of Automotive Systems

This paper presents a technique for failure detection and diagnosis (FDD) for on-board vehicle applications. Failures are represented by abrupt changes in the characteristic parameters of a system that occur at random instants of time. This representation facilitates the formulation of the FDD problem as a combined state and parameter estimation problem of a system with uncertain parameters. This non-linear estimation problem is solved using a Bayesian approach and implementing a suboptimal filter. The performance of the FDD system is demonstrated in simulations of an active suspension subjected to sensor failures. In the simulations, the system diagnosed failures correctly within .05 seconds from their occurrence.