基于WP-ICA及SVM的齿轮故障诊断研究

针对旋转机械设备齿轮故障诊断问题,为全面提取反映齿轮运行状态的特征信息,提出了基于WP(小波包)与ICA(独立成分分析)相融合的特征提取及SVM(支持向量机)相适配的故障诊断方法。用小波包对信号进行分析并提取其能量特征,采用独立成分分析方法对提取的能量特征进一步优化,进而得到反映齿轮运行状态的特征向量。最后采用支持向量机对齿轮运行状态的四种类型(正常、轻微故障、中等故障、断齿故障)进行诊断评估。通过纵向比较和横向比较研究表明,所提特征提取方法较单一的小波包特征提取方法更能全面反映齿轮状态信息。采用SVM方法进行齿轮故障模式诊断,较其它方法具有更高的分类准确率,达到了很好的诊断效果。

Study of Gear Fault Diagnosis Based on Wavelet Packet and Independent Component Analysis Combining SVM

A feature extraction and fault diagnosis matching method was proposed for gear fault diagnosis of rota-ting machinery equipments,which was based on wavelet packet(WP)and independent component analysis(ICA)in order to effectively extract the features of running gear and match the support vector machine(SVM)classifier.First-ly,the signal was analyzed by WP and its energy features were extracted.Secondly,the extracted features were fur-ther optimized by using the ICA method,and then the feature vector reflecting the running state of the gear was ac-quired.Finally,the SVM was adopted to evaluate the state of gear running conditions of normal,minor fault,medium fault,or broken tooth fault.Through comparative investigation with different views,it is shown that the WP-ICA fea-ture extraction method can reflect the gear running condition more effectively than the wavelet packet feature extraction method.Meanwhile,the SVM method presents higher classification accuracy than other methods in gear fault diagno-sis and achieves very good diagnostic performances.