Accurate assessment of cumulative fatigue damage in container gantry cranes under long-term cyclic loading is hindered by the inability of traditional single-model methods to capture real-time structural conditions. This paper proposes a digital twin-driven framework that fuses multi-source data for dynamic fatigue life prediction. The framework’s core is an improved Extended Hyper-Heuristic Neural Network (EHH-NN), which incorporates regularization optimization, a split node structure, and ANOVA-based function decomposition to model complex stress responses under limited training data. The improved model achieves a goodness of fit (R2) of 0.942 and a mean relative error of 4.4%, outperforming standard BP and LSTM models while maintaining a prediction time of 42 ms. A closed-loop correction mechanism driven by measured stress feedback is designed to dynamically adjust model outputs, and a prototype system integrating PLC-based data acquisition with Unity3D 2022.3 visualization is developed to demonstrate engineering applicability.
Zhu et al. (Mon,) studied this question.