Fault Diagnosis of Spur Gearbox by Image Classification Using Deep CNN

Fault diagnosis of two stages gearboxes using deep learning techniques has gained significant attention in the last decade. Spur gearboxes are widely used for power transmission in industries which makes the fault diagnosis study of spur gearboxes very crucial. A significant amount of work has been done to diagnose faults in spur gearbox at higher load and high speeds. However, fewer studies are performed to diagnose faults in spur gearbox running at comparatively low speeds and lower loads. In this paper, fault diagnosis of spur gearbox has been performed at relatively low speed and loading conditions using advanced signal processing techniques combined with deep convolution neural network. Four different image datasets are prepared by converting vibration signals into images using time-series plots, frequency-amplitude plot, Hilbert-Huang Transform (HHT) spectrum and Continuous Wavelet Transmission (CWT) spectrum. The deep convolutional neural network (DCNN) has been trained by using these four different image datasets. Their accuracy and performance for identifying multiple faults in spur gear has been compared. Among these four methods, CWT-DCNN has been found to be more effective.