参数优化变分模态分解与多域流形学习的行星齿轮箱故障诊断  

Fault diagnosis of planetary gearbox based on parameter optimized VMD and multi-domain manifold learning

作　　者：王振亚 姚立纲[1] 戚晓利[2] 张俊[1] 郑近德[2] WANG Zhenya;YAO Ligang;QI Xiaoli;ZHANG Jun;ZHENG Jinde(School of Mechanical Engineering and Automation,Fuzhou University,Fuzhou 350116,China;School of Mechanical Engineering,Anhui University of Technology,Ma’anshan 243032,China)

机构地区：[1]福州大学机械工程及自动化学院,福州350116 [2]安徽工业大学机械工程学院,马鞍山243032

出　　处：《振动与冲击》2021年第1期110-118,126,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金项目(51775114);国家自然科学基金项目(51275092);福建省工业机器人基础部件技术重大研发平台(2014H21010011)。

摘　　要：针对行星齿轮箱特征提取困难的问题,提出一种基于参数优化变分模态分解与多域流形学习的故障诊断方法。首先,利用樽海鞘群优化变分模态分解(SSO-VMD)对信号进行分解与重构,降低噪声干扰;然后,从多域提取故障特征,并采用改进监督型自组织增量学习神经网络界标点等度规映射(ISSL-Isomap)算法进行降维处理,获取低维故障特征;最后,运用人工蜂群优化支持向量机(ABC-SVM)多故障分类器进行诊断识别。将SSO-VMD与经验模态分解进行对比,仿真信号分析结果验证SSO-VMD的优越性。将所提故障诊断方法应用于行星齿轮箱故障诊断实验分析中,结果表明:多域特征提取效果优于时域、频域和尺度域等单域特征提取效果;ISSL-Isomap降维效果优于等度规映射,t-分布邻域嵌入,线性判别分析,加权等度规映射和监督等度规映射等算法;所提方法故障识别率达到100%,能够有效识别出行星齿轮箱各工况类型。Aiming at the problem of features extraction of planetary gearbox being difficult,a fault diagnosis method based on parameter optimized variational mode decomposition(VMD)and multi-domain manifold learning was proposed.Firstly,the salp swarm optimization variational mode decomposition(SSO-VMD)was utilized to decompose and reconstruct signals to reduce noise interference.Then,fault features were extracted from multi-domain,and the improved supervised self-organizing incremental learning neural network boundary punctuation isometric mapping(ISSL-Isomap)algorithm was used to reduce dimension,and acquire low-dimensional fault features.Finally,the artificial bee colony support vector machine(ABC-SVM)multi-fault classifier was used to do diagnosis and identification.The SSO-VMD was compared with the empirical mode decomposition(EMD),and the superiority of SSO-VMD was verified with simulation signal analysis results.The proposed fault diagnosis method was applied in planetary gearbox fault diagnosis test analysis.Results showed that the multi-domain feature extraction is better than the feature extraction in single-domain including time domain,frequency one and scale one;the dimension reduction effect of ISSL-Isomap is better than those of Isomap,t-distributed stochastic neighborhood embedding,linear discriminant analysis,weighted Isomap and supervised Isomap;the fault recognition rate of the proposed method reaches 100%,and it can effectively recognize various types working conditions of planetary gearbox.

关 键 词：变分模态分解 等度规映射 流形学习 支持向量机 行星齿轮箱 故障诊断 

分 类 号：TH165.3[机械工程—机械制造及自动化] TN911.7[电子电信—通信与信息系统]