摘要
针对现有柔性薄壁轴承的细粒度故障识别及其现有诊断方法特征提取不足等问题,提出一种基于牛顿-拉夫逊优化(Newton-Raphson-Based Optimizer,NRBO)算法优化的Transformer编码层与双向长短期记忆(Bi-directional Long Short-Term Memory,BiLSTM)神经网络解码层相结合的故障诊断方法。该方法首先使用时变滤波经验模态分解(Time-Varying Filter Empirical Mode Decomposition,TVFEMD)柔性薄壁轴承振动加速度信号,同时采用NRBO算法对其时变滤波带宽和B样条阶数等参数进行优化,依据互相关系数准则筛选主要本征模态分量(Intrinsic Mode Function,IMF),计算其时频域特征,结合振动加速度信号的时域、频域特征构建多域特征数据集,并按比例划分为训练集和测试集;其次将训练集输入模型,通过NRBO算法对模型初始学习率、BiLSTM神经网络的隐藏层节点数以及Transformer模型的正则化系数进行优化;通过测试集对优化模型进行测试,并与其他细粒度故障诊断模型对比。结果表明,所提方法准确率达99.60%,高于其他模型。该方法可为柔性薄壁轴承细粒度的智能诊断提供一种新的研究思路,对其他相关领域的智能化健康管理亦可提供有益借鉴。
To address the challenges of insufficient feature extraction and fine-grained fault identification in existing diagnostic methods for flexible thin-walled bearings,it proposes a fault diagnosis approach based on a Newton-Raphson-Based Optimizer(NRBO)algorithm-optimized Transformer encoder and Bi-directional Long Short-Term Memory(BiLSTM)decoder.First,vibration signals are decomposed using Time-Varying Filter Empirical Mode Decomposition(TVFEMD),with its time-varying filtering bandwidth and B-spline order optimized by NRBO algorithm.Principal Intrinsic Mode Function(IMF)are selected via cross-correlation coefficients,and their time-frequency features are combined with time-domain and frequency-domain characteristics to build a multi-domain dataset.The NRBO algorithm further optimizes the model′s initial learning rate,BiLSTM hidden layer nodes,and Transformer regularization coefficients.Experimental results show a 99.60%accuracy,outperforming other fine-grained diagnostic models.It provides a novel solution for intelligent fault diagnosis of flexible thin-walled bearings and references for intelligent health management in related fields.
作者
郭明军
陈昕昀
石淇
李鑫
赵学智
GUO Mingjun;CHEN Xinyun;SHI Qi;LI Xin;ZHAO Xuezhi(School of Mechanical&Automotive Engineering,Guangxi University of Science and Technology,Liuzhou 545616,China;School of Mechanical&Automotive Engineering,South China University of Technology,Guangzhou 510640,China)
出处
《现代制造工程》
2025年第11期136-145,34,共11页
Modern Manufacturing Engineering
基金
国家自然科学基金项目(52375538)
广西自然科学基金资助项目(2025GXNSFHA069192)
广西高校中青年教师科研基础能力提升项目(2025KY0346)
广西研究生教育创新计划项目(YCSW2025574)
广西科技大学博士基金项目(校科博21z59)。
关键词
柔性薄壁轴承
故障诊断
牛顿-拉夫逊优化算法
双向长短期记忆网络
Transformer模型
时变滤波经验模态分解
flexible thin-walled bearing
fault diagnosis
Newton-Raphson-Based Optimizer(NRBO)algorithm
Bi-directional Long Short-Term Memory(BiLSTM)network
Transformer models
Time-Varying Filter Empirical Mode Decomposition(TVFEMD)
作者简介
郭明军,博士,硕士生导师,主要研究方向为机械系统故障预测与健康管理。E-mail:scutgmj@163.com。
