基于K-L散度的VMD瞬时能量与PNN的滚动轴承故障诊断

Rolling bearing fault diagnosis using VMD energy feature andPNN based on Kullback-Leibler divergence

针对滚动轴承失效模式的有效识别,提出了基于K-L散度的变分模态分解(VMD)的瞬时能量与概率神经网络(PNN)的滚动故障轴承故障诊断方法。首先,利用VMD将滚动轴承信号进行分解成若干个本征模态函数(IMF);然后,分别计算原始信号和每个IMF分量的K-L值,并选择具有较小的K-L值的两个IMF分量以计算其瞬时能量并组成特征向量;最后,将特征向量输入到PNN实现故障模式识别。通过对滚动轴承故障诊断实验对该方法进行验证,结果表明,基于K-L散度的VMD瞬时能量与PNN的滚动故障轴承故障诊断准确率高达100%,将所提的方法与通过峭度准则选择VMD分量的瞬时能量和通过K-L散度值选择的集合经验模态分解(EEMD)分量与PNN网络相结合的诊断方法作对比,则明显高于其他两种方法,证明了所提方法的可行性。

According to the effective identification of failure modes in rolling bearings,a fault diagnosis method is proposed to identify rolling bearings based on K-L divergence variational mode decomposition(VMD)instantaneous energy and probability neural network(PNN).Firstly,the rolling bearing signals are decomposed into several intrinsic mode functions(IMF)by VMD.Then,the K-L values of the original signal and each IMF component are calculated separately,and two IMF components with the smaller K-L value are selected to calculate their instantaneous energy and form the eigenvector.Finally,the eigen vector is input into PNN to identify obstacle pattern recognition.The method is validated by rolling bearing fault diagnosis experiments.The results show that the accuracy of fault diagnosis of rolling bearing based on K-L divergence and PNN is up to 100%.The proposed method is compared with instantaneous energy of VMD component selected by kurtosis criterion and ensemble empirical mode decompositi(EEMD)component selected by K-L divergence and PNN network.Compared with the other two methods,the combined diagnosis method is obviously higher,which proves the feasibility of the proposed method.