变转速下基于改进ConvLSTM的滚动轴承故障诊断  

Fault Diagnosis of Rolling Bearings Based on Improved Convolutional LSTM Networks under Variable Speed Conditions

作　　者：黄金鹏 吴国新[1] 刘秀丽[1] HUANG Jinpeng;WU Guoxin;LIU Xiuli(Key Laboratory of Modern Measurement and Control Technology,State Ministry of Education,Beijing Information Science and Technology University,Beijing 100192,China)

机构地区：[1]北京信息科技大学现代测控技术教育部重点实验室,北京100192

出　　处：《噪声与振动控制》2025年第2期76-81,共6页Noise and Vibration Control

基　　金：国家重点研发计划资助项目(2020YFB1713203);北京信息科技大学勤信人才资助项目(QXTCPC202120);机电系统测控北京市重点实验室开放课题资助项目(KF20222223201)。

摘　　要：针对在多传感器下变转速且带有不同程度噪声的工况下故障特征被淹没的问题,提出一种基于改进卷积长短时记忆网络(Convolutional LSTM, ConvLSTM)的故障诊断方法:首先将多个传感器采集的一维振动信号切分为二维矩阵序列;再利用由改进ConvLSTM单元构成的特征提取层提取信号内的时间特征和空间特征,改进ConvLSTM单元是将传统ConvLSTM单元输入门中的普通卷积换成膨胀卷积,在相同的卷积核其有更大的感受野读取输入信息;最后通过由卷积层和全局平均池化(Global Average Pooling,GAP)构造的分类输出层得到诊断结果。试验使用CWRU滚动轴承数据集和XJTU-SY滚动轴承数据集进行验证。试验结果表明,与其他对比模型相比,改进ConvLSTM模型在变转速且带有不同程度噪声下达到较高的精确率并且受样本量的影响更小。Aiming at the problem of fault features being overwhelmed under variable speed operating conditions with different noise levels from multiple sensors,a fault diagnosis method based on an improved Convolutional Long Short-Term Memory network(ConvLSTM)was proposed.Firstly,the one-dimensional vibration signals collected from multiple sensors were decomposed into two-dimensional matrix sequences.Then,an improved ConvLSTM unit consisting of the feature extraction layer was utilized to extract both temporal and spatial features within the signals,where the improvement means to replace the regular convolution in a traditional ConvLSTM input gate with dilated convolution,so that it has a larger receptive field to read input information under the same convolution kernel.Finally,the classification output layer was constructed with convolutional layer and Global Average Pooling(GAP)and the diagnosis results were obtained.The method was validated by using the CWRU rolling bearing dataset and XJTU-SY rolling bearing dataset.Experiment demonstrates that compared to other benchmark models,the improved ConvLSTM model can attain higher accuracy under variable speeds with different noise levels and is less affected by sample size.

关 键 词：故障诊断 滚动轴承 变转速工况 深度学习 ConvLSTM 

分 类 号：TH133.33[机械工程—机械制造及自动化]