摘要
针对现行轴承故障诊断方法难以提取故障信号的细微特征值并且准确率较低的问题,提出基于小波变换和改进LeNet-5网络的轴承故障诊断方法。用小波变换将振动信号转换为二维时频图,并将网络模型中的卷积层构建成两个相对独立的且卷积核数不同的卷积网络支路,通过特征值融合提高提取特征值的效率,提升区分不同故障时频图的能力;选择ReLU作为激活函数,避免梯度消失;添加Dropout层,提高神经网络的泛化性。实验验证结果表明,相对于典型的卷积神经网络,所提方法用于轴承故障分类可以减少所需的迭代次数并提高准确率。
In view of the fact that bearing fault diagnosis is difficult to extract the subtle eigenvalues of the fault signal and the accuracy is low,this paper proposes a fault diagnosis method based on wavelet transform and improved LeNet-5 network,and uses wavelet transform to convert the vibration signal into a two-dimensional time-frequency graph.The convolutional layer in the network model is constructed into two relatively independent convolutional network branches with different convolution kernels,and the eigenvalue fusion is used to improve the efficiency of extracting eigenvalues and improve the distinction between different fault time-frequency maps ability.This paper chooses ReLU as the activation function to avoid the disappearance of the gradient,and adds a dropout layer to improve the generalization of the neural network.Experimental verification shows that,compared with typical convolutional neural networks,this method can reduce the number of iterations required for bearing fault classification and has higher accuracy.
作者
亓海征
殷海双
Qi Haizheng;Yin Haishuang(School of Electrical Information Engineering,Northeast Petroleum University,Heilongjiang Daqing,163318,China)
出处
《机械设计与制造工程》
2021年第10期54-58,共5页
Machine Design and Manufacturing Engineering
基金
黑龙江省自然科学基金资助项目(E2016013)。
