Assessment of Different Signal Processing Techniques for Classifying Induction Motor Faults Using PCA Features: A Comparative Analysis

Identification of faults in electrical machines is an integral part for their continuous and reliable operation. In this study, we created a technique for locating faults in the induction motors (3 phase) under three different loading conditions. We have used cross correlation method followed by Principal component analysis (PCA) on the motor fault signals and analysed the PCA score features using nearest neighbourhood criterion for authentication of faults. We have also extracted features directly from the fault current signals, in the domain of time–frequency and frequency by Discrete Wavelet Transform and Fast Fourier transform, respectively, followed by analysing the features using PCA. Comparison of results revealed that cross correlation produced the least distance from the true class in most cases, providing highest authentication of faults. The sensitivities of wavelet decomposition in different level have also been verified and it is found that the results are almost insensitive towards the level of decomposition. Besides, in several cases, wavelet analysis of the faults signals were also fund less effective compared to other methods of analysis. Results have also been validated using several other faults.