Turbine rotors are subjected to long-term transient thermo-mechanical coupled loads under complex operating conditions, making the accurate and efficient prediction of its thermal-mechanical responses critical for ensuring operational safety. This study focuses on predictive modeling of the rotor's stress and temperature fields under varying operating scenarios. To this end, a Hybrid-DeepONet is proposed to achieve high-fidelity reduced-order modeling and prediction of stress and temperature distributions. The proposed model employs an LSTM-based trunk network to extract temporal features and an MLP-based branch network to encode physical parameters. An adaptive time-masking mechanism driven by physical state variations is introduced to enhance the model’s capability in capturing dynamic patterns under complex transient conditions. First, a time-evolving multiphysics dataset is constructed via finite element simulations. The data are then preprocessed through sliding window segmentation and normalization before being used to train the Hybrid-DeepONet model, forming a predictive framework adaptable to diverse operating conditions. Finally, model performance is comprehensively evaluated by comparing full-field and critical location predictions with multiple baseline methods. Experimental results demonstrate that Hybrid-DeepONet outperforms existing approaches in both accuracy and stability. Additionally, the Shapley Additive Explanations (SHAP) algorithm is applied to quantify the contribution of each input feature, thereby improving the interpretability and transparency of the model. Consequently, the proposed method ensures high prediction accuracy and improved efficiency, providing a reliable solution for turbine rotor monitoring and life assessment.
Yan et al. (Sun,) studied this question.