基于小波包与支持向量机的碰摩故障识别方法研究

Research on Rub and Impact Fault Identification Method Based on Wavelet Packet and Support Vector Machine

为了很好地识别旋转机械的转静件碰摩故障,提出了基于小波包和支持向量机(SVM,support vec-tormachine)的碰摩故障识别方法。采用小波包对信号进行特征向量的提取,利用基于"一对多"和"一对一"的改进算法构建多类故障分类器,对多种碰摩故障进行识别。同时,以双盘悬臂转子-轴承系统的碰摩故障为例,应用该方法进行故障识别,试验结果表明,RBF核SVM故障平均识别率达到97.25%。可见,基于小波包与支持向量机分类器诊断方法的识别率明显优于传统的BP神经网络和RBF神经网络分类器,且鲁棒性好,并具有良好的泛化推广能力。

To identify the rub and impact fault of rotating machinery, a new identification method based on wavelet packet transformation and support vector machine （SVM） is proposed. The maximal singular values of wavelet packet decomposition coefficients are extracted as robust feature vectors. A new multi-class SVM classifier is designed in improved algorithm, which can recognize several patterns of faults after being trained. Meanwhile the method is used on the rub and impact fault identification of dual-disk over-hung rotor-bearing system. Experimental results show that the fault patterns can be well identified after training by SVM based on RBF kernel and its average identification rate has reached 97.25%, more effectively and accurately than conventional BP, and RBF neural networks, and has high robustness, good generalization ability as well.