A study on the coupled mechanisms of surface thermal loss and explainable data-driven modeling for marine equipment components in polar environments

In polar operational environments, marine engineering equipment is frequently exposed to extreme low temperatures and strong winds, which can lead to rapid heat loss in critical components or media, resulting in functional failure or ice accretion that threatens operational safety. To enable accurate prediction of heat loss in key areas, this study proposes a data-driven framework for coupled heat transfer prediction tailored to polar equipment. Using a constant heat flux cylinder as the representative object, a multiphysics model incorporating convection, conduction, and radiation was established. Variable-condition experiments involving wind speed and heating power were conducted in a low-temperature wind tunnel to validate the numerical model, yielding average Nusselt number errors of 7.25 % and 1.5 %, respectively. A parameterized simulation platform integrating ANSYS and Python was developed, generating 30,000 data samples through Latin Hypercube Sampling and dimensionless preprocessing. A hybrid predictive model combining attention mechanisms with the Kolmogorov–Arnold Network was constructed, and the Shapley Additive Explanations (SHAP) method was employed for interpretability analysis. The model demonstrated excellent performance across multiple metrics (MSE = 0.009, MAE = 0.050, R² = 0.991). SHAP analysis identified wind speed, thermal conductivity, and emissivity as the most influential factors, revealing a nonlinear enhancement effect due to the coupling of radiation and natural convection under low wind speed conditions. This approach effectively overcomes the high computational cost and long experimental duration associated with traditional simulations, providing a robust tool for thermal design optimization of polar equipment.