小波·分形·神经网络故障诊断法及其在风机诊断中的应用研究

研究了分形理论、小波变换与人工神经网络相结合进行故障诊断的机理与方法。利用小波包可进行多维多分辨率的特性,对振动信号进行分解与重构,提取频带能量特征分析。选用分形理论中的离散信号分形维数计算方法,提取分形维数的特征。以K-L变换作特征降维,然后用基于梯度符号变化的局部学习率自适应误差反传算法的小波神经网络对故障状态进行分类识别。并利用这种方法本文对风机转子故障进行了诊断,结果表明这种诊断方法是完全行之有效的。

Fault Diagnosis Method Based on Wavelet Packet Transform, Fractal Theory and Neural Network

Based on the wavelet packet transform characteristic of multi- resolving power, the energy feature vectors were extracted after the vibration signal is decomposed and reconstructed. Based on the distributed signal fractal dimension, the technique of fractal theory was calculated, and the fractal dimension feature vectors were extracted. K-L transform was used to reduce the dimension of the feature vectors. Then a wavelet neural network with a new arithmetic was used to recognize the mechanical faults . At last, the fault diagnosis of rotor took as the example to prove the feasibility of the method.