Improved signal processing for bearing fault diagnosis in noisy environments using signal denoising, time-frequency transform, and deep learning

Vibration signal processing is a crucial task in machine fault diagnosis. Several signal processing methods in the past relied on more conventional approaches to diagnose bearing defects. Recent advances in artificial intelligence have sparked the development of deep learning-based signal processing methods for bearing fault diagnosis. Many of the proposed methods do not apply to signals heavily polluted with noise. In this paper, a new method for bearing fault diagnosis is presented, which combines variational mode decomposition and wavelet thresholding to denoise vibration signal, time-frequency continuous wavelet transform to generate scalogram images, and deep learning structure to classify and diagnose bearing fault. The proposed method incorporates signal denoising, time-frequency transform, and deep learning to process and diagnose machine faults in the presence of background noise. Data collected from a bearing test rig are used to validate the accuracy of the proposed method, effectiveness, and robustness. The results show that the proposed method can extract "clean" vibration signals from noisy signals and accurately diagnose the fault.