Gear tooth root fatigue test monitoring with continuous acoustic emission: Advanced signal processing techniques for detection of incipient failure

The phenomenon of fatigue in gears at the tooth root can be a cause of catastrophic failure if not detected in time. Where traditional low-frequency vibration may help in detecting a well-developed crack or a completely failed tooth, a system for early detection of the nucleation and initial propagation of a fatigue crack can be of great use in condition monitoring. Acoustic emission is a potentially suitable technique, as it is sensitive to the higher frequencies generated by crack propagation and is not affected by low-frequency noise. In this article, a static gear pair is tested where a crack was initiated at a tooth root. Continuous acoustic emission was periodically recorded throughout the test. Data were processed in multiple ways to support the early detection of crack initiation. Initially, traditional feature–based acoustic emission was employed. This showed qualitative results indicating fracture initiation around 8000 cycles. A rolling cross-correlation was then employed to compare two given system states, showing a sensitivity to large changes towards the final phases of crack propagation. A banded fast Fourier transform approach showed that the 110- to 120-kHz band was sensitive to the observed crack initiation at 8000 cycles, and to the later larger propagation events at 22,000 cycles. Two advanced data processing techniques were then used to further support these observations. First, a technique based on Chebyshev polynomial decomposition was used to reduce each wavestream data to a vector of 25 descriptors; these were used to track the system deviation from a baseline state and confirmed the previously observed deviations with a higher sensitivity. Further confirmation came from the analysis of wavestream entropy content, providing support from multiple data analysis techniques on the feasibility of system state tracking using continuous acoustic emission.