International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD 2022)

The gear fault features are drowned by strong noise, which makes difficult to detect early faults. Additionally, gear fault features are often adjusted into several frequency bands, which weakens the effect of fault identification. Aiming these problems, this paper proposes a variable-scale filtering method, named as variable-scale modal sparse filtering, to enhance the early weak fault features. First, the collected vibration signals are adaptively decomposed into a series of component signals with limited bandwidth, where the timing modes can be conduct according to the learned center frequency and bandwidth of the component signals. Then, a series of learned modes are further processed by shift-invariant sparse representation to realize the modal filtering for the original signals. Finally, a series of filtered signal is obtained by the proposed variable scale mode filtering, and those periodic impacts would be more obvious in the time domain, and the characteristics are also significantly enhanced in the envelope spectrum. Simulation and experimental results show that the method is effective. And diagnosis results and comparison further expose the ability of the method in gearbox early weak fault feature enhancement.