基于改进LSTM的滚动轴承性能退化趋势预测  

Performance Degradation Trend Prediction of Rolling Bearings Based on Improved LSTM

作　　者：吕明珠[1,2] LYU Mingzhu(School of Automatic Control,Liaoning Equipment Manufacturing Vocational and Technology College,Shenyang 110161,China;Liaoning Radio and TV University,Shenyang 110034,China)

机构地区：[1]辽宁装备制造职业技术学院自控学院,沈阳110161 [2]辽宁广播电视大学,沈阳110034

出　　处：《轴承》2022年第4期70-76,共7页Bearing

基　　金：国家自然科学基金资助项目(51675350);辽宁省高等学校基本科研项目(重点项目)(LJKZ1286);辽宁省高等职业教育协同创新科研项目(2021360-191)。

摘　　要：针对传统LSTM方法不能合理利用在线数据的问题,提出了参数实时更新的改进LSTM方法并建立了有效的退化趋势预测模型。首先,依据获取的历史资料离线生成LSTM预测模型;然后,在采集在线观测数据时用已有模型前向计算方式得到预测值;最后,将新增的观测数据做为前一个采样阶段时刻的真实值,将预测值与真实值之间的偏差累积到一个整体的误差中并使用误差最小化计算方法不断地修正和更新模型参数。试验结果表明,改进LSTM方法可以准确、高效地对小样本数据的轴承退化趋势进行预测,预测准确度和模型训练时间比传统的BPNN,SVR,LSTM方法更具优越性。Aimed at the problem that the traditional LSTM method can't reasonably use the online data,an improved LSTM method with real-time parameter updating is proposed,and an effective degradation trend prediction model is established.Firstly,the LSTM prediction model is generated offline based on known historical data.Secondly,when collecting online observation data,the prediction value is obtained by using forward calculation method of existing model.Finally,the newly added observation data is taken as real value at the time of previous sampling stage,and the deviation between predicted value and real value is accumulated into an overall error,and the error minimization calculation method is used to continuously modify and update the model parameters.The test results show that the improved LSTM method can accurately and efficiently predict the bearing degradation trend of small sample data.Compared with traditional BPNN,SVR and LSTM methods,this method has more advantages in prediction accuracy and model training time.

关 键 词：滚动轴承 性能退化 寿命预测 风力发电机组 LSTM 

分 类 号：TH133.33[机械工程—机械制造及自动化] TM315[电气工程—电机]