A Hybrid Method for Wheel Hub Stress Field Prediction Under Non-Stationary Excitation

Wheel lifespan directly affects vehicle safety and economic efficiency. Uneven road surfaces impose non-stationary excitation on wheel hubs during vehicle operation, which brings difficulties to the accurate prediction of wheel hub failures. The service life of wheel hubs has a strong correlation with the magnitude and dynamic variation of their internal stresses. Accordingly, time-series prediction of stress fields for wheel hubs is of great practical significance. However, non-stationary excitation leads to non-periodic and nonlinear stresses on wheel hubs, and it is generally challenging to acquire real-world stress data of wheel hubs. To tackle these issues, this manuscript proposes a hybrid prediction method combining node stress prediction and stress field reconstruction for wheel hubs. After dimensionality reduction and key node extraction of simulation data, the CNN–BiLSTM–Attention network is applied to predict nonlinear and non-periodic node stresses. Proper Orthogonal Decomposition is further used for field reconstruction based on predicted node data, which addresses the problem of scarce measuring points in practical engineering. The normalized RMSE and MAE of the predicted stress field are 1.15% and 0.68%, with a MAPE of 4.402%, indicating favorable prediction accuracy. Comparative analyses with peer methods validate the superiority of the proposed method.