Classification of Bearing Faults using Discrete Wavelet Packet Analysis and Support Vector Machine

The automation of condition monitoring strategies of industrial sectors is on the rise with the current industrial revolution. The automatic detection and classification of bearing faults using machine learning is also one of the application domains. A lot has been investigated into the use of different classifiers for bearing fault classification and of them, Support Vector Machine (SVM) has emerged out as a simple and confident classifier. Sufficient studies are available on selecting the features from raw vibration data and then training them using SVM; however, there is a dearth in investing the role of denoised signal for feature extraction. This work presents a comparative study of extracting the features from raw signal and signal processed through Discrete Wavelet Packet Analysis (DWPA). For this the opensource data from Case Western Reserve University has been used and the compressed and denoised signals are obtained using Symlet5 wavelet. The features obtained from raw, compressed and denoised signals are trained using different SVM classifier and their classification accuracies are compared. The classification accuracy is enhanced by using the denoised signal for fault investigation.