Short-Term Probabilistic Wear Prediction in Shaft-Misaligned Spline Pairs Using Simulation and Bayesian Neural Networks

In modern aerospace applications, spline pairs serve as critical mechanical interfaces, especially in aircraft engines and transmission systems, where they transmit rotational motion and torque. However, shaft misalignment frequently impacts their performance and longevity, accelerating wear and leading to significant maintenance costs and downtime. This study proposes a short-term wear prediction method for shaft-misaligned spline pairs by integrating dynamic finite element simulations with Bayesian Neural Networks (BNNs). We developed a wear model under misalignment conditions, obtained cumulative wear data through simulations, and utilized BNN-based Long Short-Term Memory (LSTM) network for probabilistic wear state prediction. The results demonstrate that our method effectively predicts the wear depth, accounting for uncertainties, and provides crucial decision support for maintenance and life management of aerospace spline pairs.