基于多信息融合与GRU的轴承剩余寿命预测  

Prediction of Bearing Residual Life Based on Multi Information Fusion and GRU

作　　者：曹胜博 徐彦伟[1,2] 颉潭成 王浏洋[1] CAO Shengbo;XU Yanwei;XIE Tancheng;WANG Liuyang(School of Mechatronics Engineering,Henan University of Science and Technology,Luoyang Henan 471003,China;Intelligent Numerical Control Equipment Engineering Laboratory of Henan Province,Luoyang Henan 471003,China)

机构地区：[1]河南科技大学机电工程学院,河南洛阳471003 [2]智能数控装备河南省工程实验室,河南洛阳471003

出　　处：《机床与液压》2023年第24期164-168,196,共6页Machine Tool & Hydraulics

基　　金：国家自然科学基金青年科学基金项目(51805151);河南省高等学校重点科研项目(14B460007);河南省机械装备先进制造协同创新中心资助项目。

摘　　要：为解决单一传感器信号易受干扰且能提取的退化信息有限,导致轴承剩余寿命预测精度低的问题,提出一种基于双通道信息融合与门控单元(GRU)神经网络的轴承剩余寿命预测方法。进行轴承寿命试验时,在振动传感器采集信号的基础上增加声发射传感器,弥补单一信号易受干扰的缺点;使用卷积神经网络自动挖掘出包含轴承退化信息的特征,避免传统算法过分依赖专家判断的弊端;通过归一化处理对信息进行融合;最后使用这些数据训练GRU神经网络,利用训练好的门控单元神经网络预测高铁牵引电机轴承的剩余寿命。结果表明:相比单通道数据,双通道数据训练出的门控神经网络模型的预测结果更为准确;门控单元神经网络相比长短时记忆神经网络有更高的轴承寿命预测精确度。In order to solve the problem that single sensor signal is easy to be interfered and the degradation information that can be extracted is limited,which leads to low accuracy of bearing residual life prediction,a bearing residual life prediction method based on dual channel information fusion and gated recurrent unit(GRU)neural network was proposed.In the bearing life test,acoustic emission sensor was added as a supplement on the basis of using vibration sensor to collect signals,which could make up for the disadvantage that single signal was easily interfered.After that,the convolutional neural network was used to automatically mine the features containing bearing degradation information,which could avoid the disadvantage of traditional algorithms that rely too much on expert judgment.Then the information was fused through normalization.Finally,these data were used to train the GRU neural network,then the trained GRU was used to predict the remaining life of high-speed railway traction motor bearings.The results show that the neural network model trained with dual channel data are more accurate;the GRU neural network has higher bearing life prediction accuracy than the long and short-term memory neural network.

关 键 词：退化特征 信息融合 剩余寿命预测 门控单元神经网络 

分 类 号：TH133.33[机械工程—机械制造及自动化]