基于改进鲸鱼算法优化LSTM的滚动轴承故障诊断  

Rolling bearing fault diagnosis based on IWOA-LSTM

作　　者：郑直 张华钦 潘月 ZHENG Zhi;ZHANG Huaqin;PAN Yue(College of Mechanical Engineering,North China University of Science and Technology,Tangshan 063210,China;HUIDA Sanitary Ware Co.,Ltd.,Tangshan Hebei 063000,China;School of Mechatronical Engineering,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]华北理工大学机械工程学院,河北唐山063210 [2]惠达卫浴股份有限公司,河北唐山063000 [3]北京理工大学机电学院,北京100081

出　　处：《振动与冲击》2021年第7期274-280,共7页Journal of Vibration and Shock

基　　金：河北省博士后科学基金(B2020003033);河北省省属高等学校基本科研业务费研究项目(JQN20190004);唐山市应用基础研究计划项目(20130211b);华北理工大学博士科研启动基金(28412499);国家重点研发计划项目(2017YFB1302501)。

摘　　要：适应性动量(Adam)估计优化器易使深度长短时记忆神经网络(long short-term memory,LSTM)陷入局部极小值,导致故障诊断精度过低;鲸鱼算法(whale optimization algorithm,WOA)的寻优区域过大,导致寻优效率过低。针对上述两问题,将WOA进行改进(improved whale optimization algorithm,IWOA),并优化LSTM,提出IWOA-LSTM新方法。所提方法将WOA赋予动量驱动功能,继承了LSTM中的Adam优化器动量项,可优化细胞权值的搜索区域,进而提高权值寻优效率;将其与Adam优化器联合优化更新权值矩阵,以跳出局部最小值,提高故障诊断精度。此外,还系统地分析了学习效率和迭代次数对IWOA-LSTM的诊断精度影响,实现高效的故障诊断分析。通过分析实测滚动轴承内圈、外圈和滚动体三种故障可知,IWOA-LSTM的故障诊断效率分别较浅层BP神经网络(BPNN)、深度卷积神经网络(convolutional neural network,CNN)、深度门限循环单元神经网络(gated recurrent unit,GRU)、LSTM、WOA优化的LSTM(WOA-LSTM)高出了47.60%,38.06%,37.62%,26.82%,22.71%,且实现高达97%的诊断精度。The adaptive moment(Adam)estimation optimizer tends to make the long short-term memory(LSTM)neural network fall into local minimum value,which leads to excessive low accuracy of rolling bearing fault diagnosis.The searching area of the whale optimization algorithm(WOA)is too large,which leads to excessive low optimization efficiency.Here,aiming at the above two problems,WOA was improved,and LSTM was optimized to propose a new IWOA-LSTM method.It was shown that the proposed method endows WOA with momentum driven function and inherits the momentum term of Adam optimizer in LSTM to optimize the search area of cell weight value and improve the efficiency of weight value optimization;then,WOA is combined with Adam optimizer to update the weight value matrix to jump out of the local minimum value and improve the accuracy of fault diagnosis;in addition,effects of learning efficiency and iteration times on the diagnosis accuracy of IWOA-LSTM is systematically analyzed to realize highly effective fault diagnosis analysis.Three kinds of faults of inner ring,outer ring and rolling element of actual bearing data measured were analyzed,the results showed that the fault diagnosis efficiency of IWOA-LSTM is 47.60%,38.06%,37.62%,26.82%and 22.71%higher than those of shallow BP neural network(BPNN),deep convolution neural network(CNN),deep gated recurrent unit neural network(GRU),LSTM and WOA-LSTM,respectively;the diagnosis accuracy is up to 97%.

关 键 词：滚动轴承 深度学习 鲸鱼算法(WOA) 长短时记忆(LSTM) 

分 类 号：TH137[机械工程—机械制造及自动化] TP277[自动化与计算机技术—检测技术与自动化装置]