Multiclass Support Vector Machines For Diagnosis Of Broken Rotor Bar Faults Using Advanced Spectral Descriptors

This paper proposes an original combination of Power Spectrum Density estimation (PSD) and Support Vector Machines (SVMs) to diagnose and detect broken-rotor-bar faults in induction motor under different levels of load The PSD estimate is used for feature extraction and aim to identify, the descriptors that show high variability between different classes and thus would help distinguish between them This paper analyses three different spectral decomposition methods applied to induction machine stator current namely Welch Burg and Multiple Signal Classification (MUSIC) Based on different power spectral estimation concepts, the frequency resolution, variance and detection capability are different for each method according to the set of parameters used In this paper the strategy of multiclass SVMs-based classification is applied to perform the faults recognition The proposed methodology aims to determine which spectral estimate method best suited for implementation in SVMs Also the classification process performance due to the choice of kernel function is presented to show the excellent characteristic kernel function Various scenarios are examined using data sets of stator current signals from experiments under different load, and the results are compared to obtain the best performance of classification process Copyright (C) 2010 Praise Worthy Prize S r l. - All rights reserved.