A Blind Denoising Method for Noisy Rotating Machinery Vibration Signals

The quality of signals plays a critical role in the diagnosis of faults in rotating machinery. It significantly impacts the accuracy of fault Identification and the precision of predicting the remaining useful life. During the signal collection process, it is inevitable that noise is acquired from the surrounding environment or other components within the rotating machinery. This noise can pose a significant challenge when attempting to extract meaningful features from vibration signals. Therefore, there is a pressing need to create a denoising model that can effectively remove noise from collected vibration signals, even in the absence of pristine, noise-free reference signals. In this study, we first developed a self-supervised denoising method for vibration signals that differs markedly from the traditional supervised learning denoising models in the field of rotating machinery, which rely on noise-clean signal pairs for training.