基于信息融合和SA-CNN的轴承故障诊断  

Bearing Fault Diagnosis Method Based on Information Fusion and Self-attention Convolutional Neural Network

作　　者：王云 徐彦伟[1,2] 何可承 颉潭成 王军华[1,2] 蔡海潮 WANG Yun;XU Yanwei;HE Kecheng;XIE Tancheng;WANG Junhua;CAI Haichao(School of Mechatronics Engineering,Henan University of Science and Technology,Luoyang 471003,China;Intelligent CNC Equipment Henan Provincial Engineering Laboratory,Luoyang 471003,China)

机构地区：[1]河南科技大学机电工程学院,河南洛阳471003 [2]智能数控装备河南省工程实验室,河南洛阳471003

出　　处：《机械与电子》2024年第7期3-9,共7页Machinery & Electronics

基　　金：国家自然科学基金资助项目(51805151);河南省高等学校重点科研项目(21B460004)。

摘　　要：针对轴承故障特征提取困难、输入信号单一及故障识别率低等问题,提出基于多头注意力机制信息融合和自注意力机制卷积神经网络的轴承故障诊断方法。首先,预制地铁牵引电机轴承故障,搭建变工况轴承实验台并设计实验方案,采集轴承振动信号和声发射信号;其次,利用多头注意力机制将轴承的振动信号和声发射信号进行融合;最后,将融合后的信号输入自注意力机制卷积神经网络中进行故障诊断。实验结果表明,基于多头注意力机制信息融合和SA-CNN的轴承故障智能诊断方法,可以有效关注到轴承故障特征信号,提升变工况下轴承故障诊断的准确率。Aiming at the problems of difficulty in bearing fault feature extraction,single input signal and low fault recognition rate,a bearing fault diagnosis method based on multi-head attention information fusion and self attention convolutional neural network(SA-CNN)was proposed.Firstly,the bearing failure of metro traction motor was pre-made.The bearing test stand with variable working conditions was built and the experimental scheme was designed to collect the bearing vibration signal and sound emission signal.Next,the multi-head attention mechanism is employed to fuse the vibration fault signals and acoustic emission signals of the bearings.Finally,the fused signals are put into a self-attentive mechanism convolutional neural network for fault diagnosis.The final results show that based on multi-head attention information fusion and SA-CNN can effectively pay attention to bearing fault characteristic signals,and improve the accuracy of bearing fault diagnosis under varying working conditions.

关 键 词：轴承故障诊断 多头注意力机制 信息融合 自注意力机制 CNN 

分 类 号：TH133.33[机械工程—机械制造及自动化]