基于小波一阶灰度矩向量和神经网络的旋转机械故障诊断的研究

The Research on Fault Diagnosis for Rotating Machinery Based on Wavelet Gray Moment Vector and Probability Neural Networks

在旋转机械轴系振动故障模拟试验的基础上,对大量故障模拟试验数据进行计算,建立了典型故障的小波一阶灰度矩向量样本,将其作为概率神经网络的输入进行故障诊断研究。结果表明,基于一阶灰度矩向量的概率神经网络可实现对训练样本100%的正确识别率,对"陌生"样本的正确识别率也超过75%。可见,概率神经网络综合了Bayes分类器和神经网络的优势,利用概率神经网络融合信号的一阶灰度矩向量特征实现旋转机械轴系故障模式识别是一种可行有效的方法。

Based on the faulty simulation experiment on vibration from our rotor test rig, the large amount of faulty simulation experiment data were calculated, and the sample of wavelet gray moment vector of typical fault was built. As its input to the Probability Neural Networks （PNN）, the fault diagnosis was researched. The result shows that the classification accuracy of 100% for training data is realized by the PNN, has good classification ability of typical vibration faults of rotating machinery, and with 75% for fresh data based on the wavelet gray moment vector. So it can be deduced that PNN has integrated advantages of neural networks and Bayes classifier, and it is a practical diagnosis method for typical fault identification of rotating machinery by using integrated signals of PNN with characteristics of the wavelet gray moment vector.