本文研究基于神经网络的故障诊断系统，利用其自学习、联想记忆和客俗等良好的性能，使所建系统具有较高的推理速度和容错性，试验证明能够较好地应用于故障的在线诊断当中。 In this paper, fault diagnosis system based on neural network is studied. By using its good performance such as self-learning, associative memory and custom, the proposed system has high reasoning speed and fault tolerance. The experiment proves that it can be applied to fault online Diagnosis.