旋转机械故障诊断的神经网络方法研究

BP神经网络具有较好的非线性映射能力,可以描述频率特征和故障之间的关系,而概率神经网络学习规则简单、训练速度快、避免局部极小和反复训练的问题。根据两种神经网络的原理选择合适的参数建立两个旋转机械故障诊断模型,并利用模型对某旋转机械的故障数据进行处理,结果显示两种网络在故障诊断方面的实用价值。通过对故障数据的结果对比可以看到PNN网络比BP网络具有更好的容错能力。

Research on the Neural Network in Rotating Machinery Fault Diagnosis

BP neural network is effective for dealing with non-liner mapping which could describe the relations between frequency characters and faults.Probabilistic neural network(PNN) is simple in learning rules and rapid in training,which could void the problems of the local optimization and the repeating training.Two models for rotation machinery fault diagnosis are established by using appropriate parameters according to the theory of two neural networks.The fault data of some rotation machineries is dealt with by using the models,and the result shows the applied value of the neural networks in the fault diagnosis.Comparing the result of the fault data,PNN neural network is better than BP neural network in the fault-tolerant capability.