Using combination of lifting wavelet and multiclass SVM based on global optimization class strategy for fault pattern identification

This paper presents a new method based on lifting wavelet for obtaining a fast multiclass SVM classification based on global optimization class strategy for fault diagnosis of roller bearing. Decision making was performed in two stages: feature extraction by computing the lifting wavelet coefficients and classification using the multiclass SVM classifiers trained on the extracted features. Experiments demonstrate that in comparison to discrete wavelet transform the lifting wavelet feature extraction can speed up the identification phase as well as achieve higher accuracy of multiclass SVM that is based on global optimization class strategy. Experimental results also reveal that the proposed multiclass SVM of global optimization is better than strategy of one against one and DAGSVM.