Comparative Analysis of Continuous Wavelet Transforms on Vibration signal in Bearing Fault Diagnosis of Induction Motor

Bearing failure is considered as one of the major problems in induction motor, which can result a huge mechanical damage if is not monitored from the initial stage. A complete fault classification method is presented in this paper by combining wavelet-based signal processing technique and deep learning method for fault classification. Vibration signal for eight different bearing conditions has been considered and initially, Hillbert Transform, and envelope spectrum are applied to extract the fault frequency information. With the fault frequency band, three different continuous wavelet transforms (Morse, Morlet, and Bump) are employed to generate the 2-D image of each bearing conditions. Each RGB image for different faulty conditions reflects the distinguishable pattern. Finally, a convolution neural network (CNN) is applied to classify the bearing faults by learning the discriminative features of the 2-D images. The CNN model is trained and test individually with the three different wavelet images and the classification accuracy is found more than 98% for each case. Additionally, two predefined CNN models (LeNet-5 and AlexNet) also applied to validate our proposed method for image classification. The experimental results indicates that the conversion of the 1-D vibration time domain signal into 2-D time-frequency image through wavelet and incorporate CNN model to classify the images can be an effective approach in the field of fault diagnosis.