Research on Load Identification and Prediction of Ship Propulsion Shafting Based on Digital–Physical Hybrid Models

Shafting load directly reflects shafting alignment quality and is critical to ship safety and reliability, yet remains difficult to measure directly in engineering practice. To address this, we propose a load identification and prediction method based on a Digital–Physical hybrid model. This approach integrates shafting load inversion with the time-series dependency characteristics of LSTM networks to construct an interpretable framework comprising physical, data, and decision layers. Modal testing calibrates the finite element model, while Tikhonov regularization addresses the ill-posed nature of frequency response function inversion. Additionally, a weight allocation strategy is designed during preprocessing to enhance training data quality. Validation experiments for load identification and prediction are conducted using an optimized dataset fused from measured and simulation data. Results show that, compared with purely physical or purely simulation-based models, the proposed hybrid model reduces prediction errors (RMSE, MAE, MSE) by 32–48.4% and increases the goodness of fit of prediction curves by 4%. This demonstrates superior predictive capability and interpretability, providing a new avenue for the monitoring of shafting conditions and load prediction in complex mechanical structures.