基于连续小波变换和注意力机制堆叠的电机滚动轴承故障诊断方法  

Failure Diagnosis Method of Motor Rolling Bearing Based on Continuous Wavelet Transform and Attention Mechanism Stacking

作　　者：谢奕尘 李振璧 XIE Yichen;LI Zhenbi(Department of Electronics and Information Engineering,Bozhou University,Bozhou Anhui 236800,China)

机构地区：[1]亳州学院电子与信息工程系,安徽亳州236800

出　　处：《辽宁科技学院学报》2024年第4期48-52,共5页Journal of Liaoning Institute of Science and Technology

基　　金：安徽省教育厅高校自然科学重点研究项目“基于机器视觉的不透明瓶装白酒内可见异物检测的研究”(KJ2019A1307);亳州学院自然科学研究一般项目“基于CAN总线控制的原酒自动摘酒系统研究”(BYZ2018C01).

摘　　要：文章针对基于卷积神经网络的电机滚动轴承故障诊断模型在训练时不能将特征按重要程度进行区分、故障类别诊断精度较差等缺点,构建了基于注意力机制堆叠的卷积神经网络,该网络可对特征图片中的重要特征进行凸显,通过连续小波变换将一维故障信号按照特定长度进行截取后转换为二维的时频图,该图作为卷积神经网络的输入。在注意力机制中引入深度可分离卷积以在凸显特征的同时减少参数量,在Inception模块中引入注意力机制对特征图像进行多尺度特征提取的同时凸显特征,并通过实验探讨在该模块中注意力机制堆叠的次数以及网络结构中Inception块堆叠次数对网络故障诊断性能的影响,在诊断精度表现最优的网络结构中设定不同的Batch_Size,进一步探究最佳模型。实验结果表明,改进后的网络结构在Inception模块中堆叠两层注意力机制、网络中堆叠一次Inception块时,性能达到最优,诊断准确率达到100%。The rolling bearing failure diagnosis model based on convolutional neural network cannot distinguish the features according to their importance during training and has poor failure category diagnosis accuracy.To address these shortcomings,a convolutional neural network based on attention mechanism stacking is built to highlight the important features in the feature pictures.Firstly,the one-dimensional failure signal is intercepted by continuous wavelet transform according to a specific length and is then converted into a two-dimensional time-frequency map as the input of the convolutional neural network.After this,the depth-separable convolution is introduced in the attention mechanism to reduce the number of parameters while highlighting the features.Experiments are conducted to investigate the impact of the number of attention mechanism stacks in the Inception module and the number of Inception block stacks in the network structure on the performance of network failure diagnosis,and different Batch_Size is set in the network structure with the best performance in diagnosis accuracy to further explore the optimal model.The experimental results show that the improved network structure achieves optimal performance when two attention mechanisms are stacked in the Inception module and one Inception block is stacked in the network with the diagnostic accuracy being 100%.

关 键 词：轴承故障诊断 连续小波变换 注意力机制 

分 类 号：TH133.33[机械工程—机械制造及自动化]