一种新的机器学习方法——PSVM应用于数控磨床智能诊断的研究

A study of new machine learning method- PSVM and its application in intelligent fault diagnosis for CNC grinding machine

针对经典SVM对于样本噪声敏感的局限性问题,在经典SVM基础上,引入主成分分析(PCA)方法,提出一种改进SVM新方法-PSVM。这种新的机器学习方法,既利用了PCA降噪的特性,又具有经典SVM泛化能力强、分类快的特性。改善了经典SVM的鲁棒性。应用小波包分析对信号进行预处理,直接得到特征矢量,并作为PSVM的输入,提出满足实时性要求的分类模型WPSVM。通过实例分析,证明这种方法在分类正确率、分类速度以及适用的样本规模等方面都表现出了一定的优越性。

In this paper, SVM based classification to obtain the knowledge of fault diagnosis is introduced. The sensitivity oftypical SVM is analyzed. And in orderto solve this limitations problem, on the basis oftypical SVM,with the introduction of PCA (PrincipalComponents Analysis), a improved machine learning method-PSVM(Primary component analysis Sup- portVector Machine)is proposed in this paper.This new method has both the obvious de-noising ability from PCA and the generalization from SVM. Thus the robusticity ofSVM is improved.The waveletpacket analysis is introduced for translation of the signal into char- acter vector directly. And the translation result is the input of PSVM. Therefore a quick classification modelthat meets the real-time request is proposed.Finally, a detailed experi- ment result proves the veracity , short computational time and good generalization ability for sample setscale.