分析了旋转机械振动故障的特性，在此基础上对基于ＢＰ算法的诊断方法进行了研究，建立了振动故障诊断的神经网络模型，对影响诊断网络的参数进行了详细分析，获得了用于振动故障诊断的最佳网络模型。针对ＢＰ网络收敛速度慢的缺点，提出了改进算法。该诊断模型在模拟转子实验台上进行了实验验证取得了良好的效果。 Based on the analysis of the diagnosis method based on BP algorithm, the neural network model of vibration fault diagnosis is established, and the parameters affecting the diagnosis network are analyzed in detail. The best network model for fault diagnosis. Aiming at the disadvantage of slow convergence of BP network, an improved algorithm is proposed. The diagnostic model is validated experimentally on a simulated rotor test bench.