Online decentralized fault diagnosis for robotic system using ANFIS

In this paper a decentralized model-free method for real-time fault diagnosis applicable in robotic systems has proposed. This method can discriminate any predefined fault type as well as novel faults in actuators and robot components; localization and decreasing of computational burden is guaranteed by distribution of diagnostic agents and their interactions with supervisory system. The generalization of this method relies on using distributed diagnostic agents for different parts of the system. Adaptive Neuro-Fuzzy Inference System has been trained to discriminate between normal and faulty condition in both detection and isolation phases. It is possible to diagnosis large number of fault using multiple ANFIS for different type of faults. Simulation results of 3-Link Rigid Planar Manipulator controlled by a model-reference Adaptive Controller are practiced to demonstrate capability of the proposed method toward achieving a fault tolerant controller.