基于Fisher标签字典学习的齿轮箱故障诊断方法研究  

Research on Gearbox Fault Diagnosis Method Based on Fisher Label Dictionary Learning

作　　者：赵伟恒 陈曦晖 杨冠雄 吕铖跃 Zhao Weiheng;Chen Xihui;Yang Guanxiong;Lyu Chengyue(College of Mechanical and Electrical Engineering,Hohai University,Changzhou 213200,China)

机构地区：[1]河海大学机电工程学院,江苏常州213200

出　　处：《煤矿机械》2024年第4期169-172,共4页Coal Mine Machinery

基　　金：江苏省研究生科研与实践创新计划项目(SJCX22_0166)。

摘　　要：齿轮箱结构复杂,工作环境恶劣,导致难以有效诊断其故障。传统稀疏表示方法缺乏对字典原子和稀疏编码的约束,不能体现出信号的结构特性。提出Fisher标签字典学习算法。首先构建Fisher约束项,学习样本的类内和类间特征;然后引入样本标签信息,构建标签约束项,将样本标签与每个字典原子相关联;最后采用最优方向法实现误差最小化,字典与稀疏编码交替更新,提高字典学习的识别率。通过实验数据对比分析,该方法在稀疏编码、原子聚类等方面均优于传统稀疏表示方法,可对齿轮箱故障进行有效诊断。The complex structure and harsh working environment of gearboxes make it difficult to effectively diagnose faults.The traditional sparse representation methods lack constraints on dictionary atoms and sparse coding,and cannot reflect the structural characteristics of the signals.The Fisher labeled dictionary learning algorithm was proposed.Firstly constructs Fisher constraint term to learn the within-class and between-class features of the samples.Then the sample label information is introduced and the label constraint term is constructed to associate the sample label with each dictionary atom.Finally,uses the optimal directions method to minimize the error,the dictionary is alternately updated with sparse coding to improve the recognition rate of dictionary learning.Through the comparative analysis of experimental data,the method is better than the traditional sparse representation method in terms of sparse coding and atom clustering,and can effectively diagnose gearbox fault.

关 键 词：故障诊断 稀疏表示 字典学习 FISHER准则 

分 类 号：TH132.46[机械工程—机械制造及自动化]