Strain energy-based gear mesh stiffness modeling and synthetic data generation for AI-driven fault diagnosis in smart manufacturing

Early fault diagnosis of transmission systems is critical for Smart Manufacturing, but it is challenging due to the scarcity of real-world fault data. This paper addresses the issue by proposing a strain energy-based method to accurately model the time-varying mesh stiffness of a spur gear with a tooth root crack. This model accounts for bending, axial, shear, and tooth root foundation deflections, along with crack factors such as depth and propagation. Based on this stiffness formulation, a six-degree-of-freedom lumped-parameter dynamic model was developed to simulate the system’s vibration response. Simulation results show that statistical features like RMS and Kurtosis, along with the appearance of sidebands in the frequency spectrum, clearly reflect the severity of the crack. These fault features are ideal inputs for AI/ML/DL models, helping to overcome the lack of data for training and optimizing fault diagnosis algorithms in Smart Manufacturing.