基于LMD排列熵和BP神经网络的行星齿轮箱故障诊断方法

Fault Diagnosis Method of Planetary Gearboxes Based on LMD Permutation Entropy and BP Neural Network

针对行星齿轮箱故障诊断过程中的故障特征向量区分度差、诊断成功率不够高等问题,提出了一种基于局部均值分解(Local mean decomposition,LMD)排列熵和BP神经网络结合的方法。对原始信号进行LMD,获得包含主要信息的PF分量,计算排列熵值,构造特征向量,利用提取的特征向量训练BP神经网络,完成故障模式识别。以EMD排列熵方法和无量纲分析方法作为对比组,实验验证说明,提出方法提取到的不同工况的特征向量区分度更强,故障诊断效果更好;且当训练组数发生变化时,提出方法的综合表现更优秀。

In view of the problems of poor discrimination of fault feature vectors extracted in the process of fault diagnosis of planetary gearboxes and insufficient diagnosis success rate,a method based on Local Mean Decomposition(LMD)permutation entropy and BP neural network is proposed.Through the LMD decomposition of the original signal,the PF components containing the main information are obtained,and the permutation entropy values are calculated to construct the feature vector.The extracted feature vectors are used to train the BP neural network and complete the failure pattern recognition test.Taking the EMD permutation entropy method and the non-dimensional analysis method as the comparison groups,the experiment proves that the feature vectors extracted from different working conditions with this method are more distinguishable,and the fault diagnosis effect is better.Moreover,this method shows better comprehensive performance when the number of training groups changes.