Fault Bearing Recognition Based on MCSA Approach Using Principal Component Analysis and Power Spectrum Density

Early fault classification using the Motor Current Signature Analysis approach is extremely challenging for many reasons: prior knowledge of the descriptive parameters of the acquired stator current signal to be analyzed (harmonics number), analysis of this signal in the time domain does not offer satisfactory detection performance, and a low signal-to-noise ratio (SNR) has a dramatic effect on detection quality. This leads to considerable impairments in signal features. In this study, principal component analysis (PCA) combined with signal processing spectral methods (PSD and SPSD) is proposed as a new feature extraction technique for the efficient extraction of the power spectra and square spectrum features of the stator current signal in the presence of Gaussian noise to distinguish the healthy or faulty state of electromechanical machine. The use of PCA enables the extraction of principal components associated with different harmonic appearances, characterizing mechanical defects in these machines as the input vector of the classifier. This makes detection easier, even in poor SNR conditions, because PCA allows the removal of Gaussian noise. The simulation was performed using MATLAB software with various stator-current signals containing different harmonics describing these faults. The results obtained showed that such data from this combination would enable faults in electric induction machines to be distinguished and classified with a high degree of accuracy, irrespective of the number of harmonics and noise. Keywords. Bearing fault, MCSA, Spectrum feature, Power spectrum density, Classification, Principal component analysis