基于低阶加权与卷积稀疏学习的齿轮箱故障诊断  

Gearbox Fault Diagnosis Based on Low-Order Weighting and Convolution Sparse Learning

作　　者：王帅旗 张焕可 陈会涛[2] WANG Shuai-qi;ZHANG Huan-ke;CHEN Hui-tao(Department of Mechanical and Electrical Engineering,Xuchang Vocational College of Electrical Engineering,He'nan Xuchang 461000,China;School of Mechanical and Power Engineering,HeJnan University of Technology,He'nan Jiaozuo 454003,China)

机构地区：[1]许昌电气职业学院机电工程系,河南许昌461000 [2]河南理工大学机械与动力工程学院,河南焦作454003

出　　处：《机械设计与制造》2023年第5期41-50,共10页Machinery Design & Manufacture

基　　金：2018年度河南省重点研发与推广专项(182102310793)。

摘　　要：为了提升在噪声与复杂传递路径调制下齿轮箱故障诊断的精度,提出了一种基于低阶加权与卷积稀疏学习的齿轮箱两阶段源特征恢复方法。首先利用源特征的周期性自相似性结构,设计了一种低阶加权模型,当两种波形耦合在同一频带内时,可以有效地区分调制波和干扰波。然后采用卷积滤波器直接描述传输路径的调制过程,保证了脉冲源包络的可靠恢复。同时,通过非负有界稀疏先验保证了反褶积能力。最后数值仿真与风力发电机组实验结果证明了低阶模型主能够分离聚焦特征波形,卷积稀疏学习能够突出脉冲源特征,从而有效提升齿轮箱的故障诊断精度。In order to improve the accuracy of gearbox fault diagnosis under noise and complex transfer path modulation,a two-stage source feature recovery method based on low-order weighting and convolution sparse learning was proposed.Firstly,a low-order weighting model was designed by using the periodic self similarity structure of source characteristics.When the two wave forms were coupled in the same frequency band,the modulated wave and the interference wave could be distinguished effectively.Then the corwolution filter was used to describe the modulation process of the transmission path directly to ensure the reliable recovery of the envelope of the pulse source.At the same time,the ability of decoiwolution was guaranteed by non negative bound sparse prior.Finally,numerical simulation and wind turbine experiment results show that the low-order model can mainly separate the focus feature waveform,and convolution sparse learning can highlight the pulse source characteristics 9 so as to effectively improve the fault diagnosis accuracy of gearbox.

关 键 词：齿轮箱 故障诊断 卷积稀疏学习 低阶加权模型 

分 类 号：TH16[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程] TH132.41[自动化与计算机技术—控制科学与工程]