基于LSTM-CNN的双路径滚动轴承故障诊断
Fault Diagnosis of Dual-path Rolling Bearing Based on LSTM-CNN
景斯桐 1吴东升1
作者信息
- 1. 沈阳理工大学 自动化与电气工程学院,沈阳 110159
- 折叠
摘要
针对轴承故障诊断中特征提取对人工依赖性强、特征提取不准确、对振动信号中的时间信息不敏感的问题,提出一种基于长短时记忆网络(LSTM)和卷积神经网络(CNN)结合的双路径递归神经网络方法,该方法对振动等原始信号进行处理,捕获时间序列数据中的远距离依赖关系,并引入注意力机制,抑制输入信号的高频噪声,使模型关注关键信息,提高模型训练效率.在凯斯西储大学轴承故障数据集上的实验结果表明,该方法能有效提升轴承故障识别率,具有良好的故障分类能力.
Abstract
Aiming at the problems of strong manual dependence of feature extraction,inaccurate feature extraction and insensitivity to temporal information in vibration signals in fault diagnosis of bearing,a dual-path recurrent neural network method based on the combination of long and short-term memory network(LSTM)and convolutional neural network(CNN)is proposed,which proces-ses the original signals such as vibration,captures the long-range dependence in time series data and introduces attention mechanism to suppress high-frequency noise of the input signal,so that the model focuses on key information and improves model training efficiency.The result of experiment on the bearing fault dataset of Case Western Reserve University shows that this method can effec-tively improve the accuracy of bearing fault identification and has good fault classification ability.
关键词
故障诊断/注意力机制/长短时记忆网络/卷积神经网络Key words
fault diagnosis/attention mechanism/long short-term memory network/convolutional neural networks引用本文复制引用
基金项目
辽宁省教育厅高等学校基本科研项目(LJKMZ20220618)
出版年
2024
NETL
NSTL
万方数据