基于AMCNN-BiLSTM-CatBoost的滚动轴承故障诊断模型研究  

Fault Diagnosis Model of Rolling Bearings Based on AMCNN-BiLSTM-CatBoost

作　　者：袁建华 邵星[1] 王翠香[1] 皋军[1] YUAN Jianhua;SHAO Xing;WANG Cuixiang;GAO Jun(College of Information Engineering,Yancheng Institute of Technology,Yancheng 224051,Jiangsu,China;College of Mechanical Engineering,Yancheng Institute of Technology,Yancheng 224051,Jiangsu,China)

机构地区：[1]盐城工学院信息工程学院,江苏盐城224051 [2]盐城工学院机械工程学院,江苏盐城224051

出　　处：《噪声与振动控制》2025年第2期82-89,共8页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(62076215);教育部新一代信息技术创新资助项目(2020ITA02057);盐城工学院研究生科研与实践创新计划资助项目(SJCX22_XZ035、SJCX22_XY061)。

摘　　要：针对现有的轴承故障诊断模型存在的分类精度差、运算效率不高的问题,提出一种基于注意力机制-卷积神经网络-双向长短期记忆网络-CatBoost(AMCNN-BiLSTM-CatBoost)的滚动轴承故障诊断模型。首先,对原始振动信号进行下采样技术处理,然后将经过下采样后的振动信号作为模型输入,通过3个不同的卷积模块提取特征,并使用通道注意力模块对提取的特征进行加权融合,然后将经过加权融合后的数据输入到双向长短期记忆网络中进一步地提取时序特征信息,最后输入到CatBoost中进行故障分类。经过实验表明,该模型不仅能够保证故障诊断的高准确率,还可以大大缩短网络的训练时间。Aiming at the problems of poor classification accuracy and low operational efficiency of existing bearing fault diagnosis models,a rolling bearing fault diagnosis model based on attention mechanism,convolutional neural network(AMCNN),bidirectional long short-term memory(BiLSTM)network and CatBoost was proposed.Firstly,the original vibration signal was processed by down-sampling technique,and then the down-sampled vibration signal was used as the model input to extract features through three different convolution modules.Then,the channel attention module was used to carry out weighted fusion of the extracted features,and then the weighted fusion data was input into the BiLSTM network to further extract the timing feature information.Finally,the timing feature information was input into CatBoost for fault classification.Experiments show that this model can not only guarantee the high accuracy of fault diagnosis,but also greatly shorten the training time of the network.

关 键 词：故障诊断 卷积神经网络 双向长短期记忆网络 注意力机制 CatBoost 轴承 

分 类 号：TH133.33[机械工程—机械制造及自动化]