基于声信号的滚动轴承故障诊断研究  

Fault diagnosis of rolling bearings based on acoustic signals

作　　者：陈剑[1,2] 徐庭亮 黄志 孙太华 李雪原 季磊 杨惠杰 CHEN Jian;XU Tingliang;HUANG Zhi;SUN Taihua;LI Xueyuan;JI Lei;YANG Huijie(Institute of Sound and Vibration Research,Hefei University of Technology,Hefei 230009,China;Automotive NVH Engineering&Technology Research Center Anhui Province,Hefei 230009,China)

机构地区：[1]合肥工业大学噪声振动研究所,合肥230009 [2]安徽省汽车NVH技术研究中心,合肥230009

出　　处：《振动与冲击》2023年第21期237-244,共8页Journal of Vibration and Shock

基　　金：安徽省科技重大专项(17030901049)。

摘　　要：结合小波包短时能量散布熵、回溯搜索算法以及学习矢量神经网络,提出一种基于声信号的滚动轴承故障诊断新方法。首先利用小波包分解结合短时能量对声信号进行脉冲能量提取,突出与故障相关的时频子空间的能量分布,再通过计算各子空间短时能量序列的散布熵,构造特征矩阵。利用t-分布随机邻域嵌入方法对所获特征进行降维聚类,显示所提取的特征具有较好的聚类性能。然后采用回溯搜索算法优化学习矢量量化建立神经网络故障诊断模型,对轴承故障进行识别,并与多种诊断方法进行比较,试验结果表明,加入短时能量散布熵后,本模型提升了声信号的能量特性,优化了特征矩阵,诊断性能最佳。Here,combining wavelet packet short-term energy dispersion entropy(STE-DE),backtracking search algorithm(BSA) and learning vector quantization(LVQ) neural network,a new method for rolling bearing fault diagnosis based on acoustic signals was proposed.Firstly,wavelet packet decomposition was combined with STE to extract pulse energy of acoustic signals,highlight energy distribution of time-frequency subspace correlated to faults,and then feature matrix was constructed by calculating DE of STE sequence of each subspace.The t-distribution random neighborhood embedding method was used to perform dimensionality reduction clustering for the extracted features.It was shown that the extracted features have better clustering performance.Then,BSA was used to optimize LVQ,and establish a neural network fault diagnosis model.The established model was used to identify bearing faults,and it was compared with various diagnostic methods.The experimental results showed that after adding STE-DE,this model can improve energy characteristics of acoustic signals,optimize the feature matrix,and have the optimal diagnostic performance.

关 键 词：轴承故障诊断 声信号 短时能量散布熵 学习矢量量化 回溯搜索算法 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]