基于改进VMD-MCKD和深度残差网络的风机齿轮箱故障诊断  

Fault Diagnosis of Fan Gearbox Based on Improved VMD-MCKD and Deep Residual Network

作　　者：蔡昌春[1,3] 何捷 承敏钢 张能文 王全凯 CAI Changchun;HE Jie;CHENG Mingang;ZHANG Nengwen;WANG Quankai(College of Artificial Intelligence and Automation,Hohai University,Changzhou 213022,China;College of Information Science and Engineering,Hohai University,Changzhou 213022,China;Jiangsu Key Laboratory of Power Transmission&Distribution Equipment Technology,Hohai University,Changzhou 213022,China;Jiangsu Xindaoge Automatic Control Technology Co.,Ltd.,Wuxi 214433,China)

机构地区：[1]河海大学人工智能与自动化学院,江苏常州213022 [2]河海大学信息科学与工程学院,江苏常州213022 [3]江苏省输配电装备技术重点实验室,江苏常州213022 [4]江苏新道格自控科技有限公司,江苏无锡214433

出　　处：《山东电力技术》2024年第2期67-78,共12页Shandong Electric Power

基　　金：国家自然科学基金项目(51607057);常州市应用基础研究计划项目(CJ20220245);江苏省输配电重点实验室开放基金项目(2021JSSPD07)。

摘　　要：行星齿轮箱是风电机组传动系统中的重要部件,其运行工况复杂,背景噪声大,导致齿轮早期故障信号微弱且极易受背景噪声的影响。针对风电机组齿轮箱早期故障特征难以有效提取,齿轮故障难以识别的问题,提出一种风机齿轮箱故障诊断方法。首先,通过变分模态分解算法(variational mode decomposition,VMD)分解风机齿轮箱原始振动信号,获得振动信号故障的最优模态分量;接着,利用最大相关峭度解卷积算法(maximum correlated kurtosis decnvolution,MCKD)通过解卷积重构最优模态分量,削弱背景噪声增强故障冲击成分,获得故障特征;同时利用麻雀搜索算法(sparrow search algorithm,SSA)优化惩罚因子α、模态分解个数K、滤波器阶数L和反褶积周期T等参数,提升振动信号故障特征提取的准确度;最后,构建基于深度残差网络(deep residual network,ResNet)的齿轮箱故障诊断模型,建立齿轮箱故障特征与类别的非线性映射关系,实现风机齿轮箱故障分类识别。实验结果表明,所提风机齿轮箱故障诊断方法的准确率达到97.48%,相较其他方法在信号特征提取和故障诊断效率方面有明显提高。Planetary gear box is an important part in the transmission system of wind turbine.Its operation condition is complex and the background noise is high,and thus the weak early fault signals of gear are susceptible to background noise.Aiming at the problems that early fault features of wind turbine gearbox are difficult to extract effectively and gear faults are difficult to identify,a fault diagnosis method of wind turbine gearbox was proposed in this paper.Firstly,the original gear vibration signal undergone a preliminary decomposition using variational mode decomposition algorithm(VMD)to obtain the optimal modal component of vibration signal fault;Then,maximum correlation kurtosis decnvolution(MCKD)algorithm was employed to further reconstruct optimal modal components by deconvolution,which can weaken the background noise and highlight the fault impact components,so as to obtain fault characteristics;Meanwhile,sparrow search algorithm(SSA)was used to optimize the penalty factor,the number of mode decomposition,filter order and deconvolution period to improve the accuracy of vibration signal fault feature extraction.Finally,the gearbox fault diagnosis model based on the deep residual network(ResNet)was constructed,and the nonlinear mapping relationship between gearbox fault features and categories was established to realize the classification and identification of gearbox faults.Experimental results show that the accuracy of gearbox fault diagnosis by the proposed method reaches 97.48%,and the proposed method has a distincly better performance on signal feature extraction and fault diagnosis efficiency.in comparion other methods.

关 键 词：齿轮故障诊断 变分模态分解 最大相关峭度解卷积 深度残差网络 麻雀搜索算法 

分 类 号：TM315[电气工程—电机]