In this paper, we presents a novel Neural Network (NN) based Reinforcement Learning (RL) strategy for accurately estimating the faults in nonlinear systems. More specifically, this NN-RL fault estimation strategy takes advantage of the remarkable generalization and function approximation capabilities of the NN and exceptional optimal decision-making and strong learning capabilities of the RL method. The NN-RL performances are comprehensively evaluated by applying them to fault estimation problems in nonlinear manipulator systems. In addition, a comparative analysis of the NN-RL approach is also made to a pure RL fault estimation strategy. Performance comparison on the fault estimation problems of manipulator indicate that the NN-RL method results in better fault estimation as evidenced by high accuracy and overshoot rejection.

fault estimation,neural network,reinforcement learning,nonlinearity