Fault Diagnosis of Rolling Element Bearing Using a Mesh of Continuous Wavelet Transform and Visual Geometry Group 19 (VGG-19)

Health management of machines is a very crucial part for any industry. The reason being, a slight ignorance in this field can lead to huge financial losses, and in some cases can even cause causalities. Talking about industrial machines like lathe, milling, laser cutting, and CNCs; every structure has a common mechanical part i.e. bearing. So, fault diagnosis of bearing needs the most attention. This article describes an approach for fault diagnosis of bearing using a non-conventional three step procedure (data acquisition, signal processing, and data classification). For data acquisition, ceramic shear piezoelectric accelerometer having a sensitivity of 10.2 mV/(m/s2), has been used. To convert time domain vibration data into time-frequency domain image, signal processing is performed using continuous wavelet transform (CWT). Unlike conventional approaches, in this method, feature extraction and data classification are performed using a single technique i.e. visual geometry group 19 (vgg-19). For the testing and validation of proposed method, four types of bearing samples (different health conditions) i.e. normal bearing, inner race faulty bearing (fault 1), outer race faulty bearing (fault 2), and rolling element faulty bearing (fault 3) have been used. The results using the proposed approach are very encouraging, and a classification accuracy of 100% was achieved using 5 epochs. In future, this strategy can be used for investigation and condition monitoring of different mechanical parts including gear, pulley, rotor, and reducer.