Gearboxes are widely used in modern power transmission systems such as wind turbines, helicopters, and railway vehicles. Due to inappropriate operation conditions or fatigue degradation, faults may develop in gears. If early faults cannot be detected and their severity assessed in a timely manner, faults will grow, and gearbox systems will fail eventually, which can lead to major economic loss or catastrophic accidents. Detection and the severity assessment of faults prior to the failure of gearbox systems can enable condition-based maintenance scheduling and thus prevent the sudden failure of gearbox systems and reduce maintenance costs. Therefore, it is of great significance to detect faults and assess their severity. Vibration-based signal analysis is a good option for the detection and severity assessment of gear tooth crack thanks to its advantages, including being easy to collect and sensitive to the tooth crack fault. To this end, the objective of this PhD research is to develop advanced vibration signal analyzing and processing techniques for tooth crack detection and severity assessment.

Specifically, the research objective is divided into three sub-objectives based on the operating condition of gearboxes. First, an improved singular value decomposition-based method is proposed for the tooth crack detection and severity assessment when the rotating speed of gearboxes is constant. The proposed method is more useful for the extraction of tooth crack induced periodic impulses than existing methods. Second, a sparse functional pooled autoregression model is proposed for more accurate modeling of nonstationary baseline vibration from a gearbox under variable speed condition. Last, a time series model-based method is developed for the tooth crack detection and severity assessment under random speed variation.

The outcome of this research can help us better detect the gear tooth crack fault and assess its severity under either constant or variable speed conditions. Condition-based maintenance can then be better scheduled to prevent the sudden failure of gearbox systems and reduce maintenance costs. This research work has assumed only a single channel of vibration data is available for the fault detection and severity assessment. Future work will address multichannel scenarios. The fault detection and severity assessment under variable load conditions also deserve to be investigated in the future.