基于K-L散度与PSO-SVM的齿轮故障诊断

摘要

针对表征齿轮故障信号特征难提取及支持向量机结构参数基于经验选取,致使故障状态识别精度差的问题,提出了一种基于K-L散度与PSO-SVM的齿轮故障诊断方法.首先,用经验模式分解(EMD)将齿轮振动信号筛分为多个本征模式分量(IMF);然后,选取包含有信号主要特征的IMF并求其与无故障原信号的K-L散度值;其次,利用粒子群算法(PSO)优化支持向量机(SVM)的惩罚系数和高斯核宽度系数两个结构参数,在此基础上建立齿轮故障分类模型;并利用实验齿轮数据验证方法的有效性,结果表明,与TF-SVM、TF-PSO-SVM、K-L-SVM方法相比,基于K-L散度与PSO-SVM的齿轮故障诊断方法具有更高的精度.%For the problem that the characterization of the gear fault signal feature is difficult to extract and the structure parameters selection of Support Vector Machine (SVM) are based on experience leads the poor precision of fault state recognition,proposes a K-L divergence and PSO-SVM based method of gear fault diagnosis.First of all,the gear vibration signal is divided by EMD into several Intrinsic Mode Functions (IMF).Then,it selects IMF that contains main characteristics of signal and calculates their K-L divergence with the original signal value.Second,the Particle Swarm Optimization (PSO) was used to optimize the punish coefficient of Support Vector Machine (SVM) and the structural parameters of Gaussian kernel width coefficient.The gear fault classification model is buiThe effectiveness of the method was validated by the experimental data of gear.The experimental result shows that compared with the TF-SVM,TF-PSO-SVM,gear fault diagnosis method based on K-L divergence and PSO-SVM has higher precision.