摘要
针对噪声导致轴承振动信号有效退化信息难以提取的问题,采用离散小波变换对信号进行分解得到细节分量和近似分量,提取多种敏感特征输入变分自编码器进行融合降维来构建综合性能退化指标,从而有效抑制信号中的噪声分量,获得更有单调趋势性的退化指标;引入经过超参数优化的长短时记忆网络构建滚动轴承剩余寿命预测模型,采用分层抽样方法划分数据集并输入预测模型进行试验验证,结果表明:基于离散小波变换和变分自编码器所得深层退化特征能有效表征轴承的退化信息,获得更精准的轴承剩余使用寿命预测结果。
Aimed at difficult extraction of effective degradation information of bearing vibration signal caused by noise, the discrete wavelet transform is used to decompose the signal, the detailed components and approximate components are obtained, and a variety of sensitive features are extracted to input into variational auto-encoders for fusion dimension reduction to construct the comprehensive performance degradation indexes, so as to effectively suppress the noise components in signal and obtain the degradation indexes with a more monotonic trend;a long-short-term memory network optimized by hyperparameters is introduced to build a prediction model for remaining life of rolling bearings, and the stratified sampling method is used to divide the data set and input into prediction model for experimental verification. The results show that the deep degradation features obtained based on discrete wavelet transform and variational auto-encoder can effectively characterize the degradation information of the bearings, and obtain more accurate prediction results of remaining service life of the bearings.
作者
孟祥龙
丁华
吕彦宝
施瑞
MENG Xianglong;DING Hua;LYU Yanbao;SHI Rui(College of Mechanical and Vehicle Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Shanxi Key Laboratory of Fully Mechanized Coal Mining Equipment,Taiyuan 030024,China)
出处
《轴承》
北大核心
2022年第8期55-63,共9页
Bearing
基金
国家自然科学基金面上项目资助项目(52174148)
山西省重点研发项目资助项目(201903D121064)
山西省科技合作交流专项项目(202104041101003)。
关键词
滚动轴承
剩余寿命
寿命预测
小波变换
编码器
深度学习
分层抽样
rolling bearing
residual life
life prediction
wavelet transform
encoder
deep learning
stratified sampling