A New Machine Learning Method based on PCA and SVM

In fault pattern recognition field, the real-time online fault diagnosis is a new requirement especially from the high-speed machines, and also the magnificent researching direction. The precision and speed of the classification are important research issues in fault pattern recognition for this kind of intelligent fault diagnosis. Although many improved ANN (artificial neural net) methods have been proposed for this purpose, most approaches focus only on the classification precision, instead of the computing speed. In this paper, a SVM model for fault diagnosis is introduced and analyzed about its limitation from sensibility to noisy data. To address this problem, we introduce the PCA (principal component analysis) method to reduce the dimension of the sample set and de-noise data sampled from the machine. Furthermore, for the whole real-time data processing and fault pattern recognition, a kind of wavelet packet analysis is applied to translate the field sensor signal in time-spectrum into the energy value in frequency-spectrum corresponding frequency segments real-timely. Therefore we present a new machine learning method: PSVM (primary component analysis support vector machine) method based on wavelet packet analysis in the fault diagnosis field. And this improved method particularly betters the precision and computing speed of the classification then other typical networks. And also the whole processing workflow is illustrated in details. Finally the diagnosis result of the CNC grinding machine demonstrates this method with both high classification precision and quick computational speed. And the whole integration of wavelet packet analysis, PCA, and SVM is proved a new, effective and practical approach especially for the real-time online fault diagnosis exactly