Research on Multi-faults Classification of Hoister Based on Improved LMD and Multi-class SVM

According to the nonlinear, non-stationary and modulated by high frequency characteristics of vibration signals of hoister, a method of multi-faults classification is studied based on Improved Local Mean Decomposition (LMD) and Multi-class Support Vector Machines (MSVM). Firstly, based on prolongation of SVR, the improved LMD algorithm is used to remove the end effects of LMD. The feature sets of the amplitude domain, time domain, frequency-domain, approximate entropy and power spectral entropy are calculated in detail synthetically. The feature sets, which are used to research feature selection and classification algorithm. SVM-REF is used to reduce the dimension and rearranged the order of them. Finally, the multi-class classification model is proposed based on the MSVM for fault classification. The research result shows that the model realizes the classification of multiple-faults and feature sets reduction. The model is characterized by higher accuracy and the better effectiveness and practicability.