Cumulative Damage Simulation and Self-learning Response Surface-Based Non-intrusive Online Monitoring of Random Wear Degradation for Shaft Products

Non-intrusive online evaluation of mechanical products is essential for the health monitoring of long-term operating systems. For shaft components, damage simulation provides a direct means to assess online wear degradation without the need for additional sensors. However, simulating randomly accumulated wear is a challenging and time-consuming task, which poses a contradiction for real-time online applications. To address this issue, this paper proposes an online uncertainty quantification method for random wear based on cumulative damage simulation and a self-learning response surface. The accumulated wear damage is first modeled through high-cost model-updating simulations, after which a self-learning response surface—constructed using Generative Adversarial Networks (GAN) and Long Short-Term Memory (LSTM) networks—is developed as a low-cost surrogate to facilitate online wear uncertainty quantification. Finally, the proposed method is validated through a tracked-vehicle drive shaft case study, demonstrating its accuracy, feasibility, and timeliness in monitoring random degradation.