A Particle Filtering-based Framework for Real-time Fault Diagnosis of Autonomous Vehicles

Abstract


 
 
 Reliability and supervision of autonomous systems and their corresponding subsystems, can be significantly improved using advanced methods of anomaly detection and fault diagnosis. A reliable fault detection module can improve the autonomy of the vehicle itself, by leading to efficient fault tolerant control designs and mission scheduling. This paper presents a fault detection framework for incipient faults that take place on the actuators of an autonomous-small scale hovercraft. The enabling technologies borrow from the fields of modeling, data processing, Bayesian estimation theory and in particular a technique called particle filtering. Condition indicators or features are derived based on first principles modeling of the actuator’s and the vehicle’s dynamics. The fault detection algorithm fuses information obtained from different subsystems of the vehicle that use distinct sets of sensors, producing a robust degree of confidence. In addition, the algorithm is decoupled from the control system. This achieves the goal of minimizing the fault masking, or the compensation of fault-induced navigational errors by the control system, and allows for early detection even for small fault conditions. The efficacy of the diagnostic approach is demonstrated via simulation results. The proposed fault detection methodology can be easily leveraged to other types of autonomous vehicles.