Early Detection of Signal Transients Using a Hybrid Signal Processing Method for Gearbox Fault Diagnosis

Fault diagnosis of gearbox is important to keep the security and reliability of rotating machinery. In actual working condition, it is difficult to detect signal transients caused by early damage, for example, a slightly worn gear. To address this issue, a hybrid signal processing method is proposed to adaptively detect the signal transients generated by the worn gear in the gearbox. In this method, the improved local mean decomposition is firstly used to decompose the collected signal into some simpler signals, and only part of them is useful for fault diagnosis. Second, the critical signals are selected based on statistical indices and ensemble learning. Finally, the signal transients are detected by the adaptive band-pass filter based on energy window merging, so that appropriate parameters of the filter can be determined for successfully extracting transients. The experimental vibration signal collected from the gearbox is used to demonstrate the effectiveness of the proposed method.