Detection and identification of actuator faults in robotic systems based on multiple-model nonlinear state estimation

Modern robotic systems perform elaborate tasks in a complicated environment and have close interactions with humans. Therefore fault detection and isolation (FDI) systems must be carefully designed and implemented on robots in order to guarantee safe and reliable operations. In addition, many high performance robotic controllers require full state feedback; hence it is essential to implement state estimators whenever not all state variables are measurable. Moreover, the state estimator must work properly despite the presence of faults so that the robot is fault tolerable. In this paper, we propose an algorithm for state estimation, fault detection, and fault identification of a robotic system. All faults in consideration are associated with a set of exclusive fault modes. Then a multiple-model nonlinear state estimator is applied to estimate not only the state but also the fault mode of the robot at each time step. Furthermore all fault modes are organized in a hierarchical structure to alleviate the computation load. Simulations show that state estimation is accurate even in the event of actuator faults, and that the occurrence of faults is detected immediately. The computational advantage of the proposed hierarchical structure is also demonstrated by the simulations.