Rapid Temperature-Field Prediction and Equivalent Interface Heat-Transfer Parameter Identification for Double-Steel-Plate Concrete Structures

Abstract

Abstract

Fire-induced temperature fields in double-steel-plate concrete structures are strongly governed by steel–concrete interfaces, where global temperature error alone is insufficient to characterize interface heat-transfer behavior. This study presents an interface-aware, surface-temperature-driven framework for rapid prediction of temperature fields and interpretation of interface behavior in fire-exposed composite structures. Full-field temperatures are reconstructed using surface-temperature histories, spatial coordinates, and material-region labels. To explicitly evaluate interface behavior, paired steel-side and concrete-side temperatures are extracted to compute temperature jumps and heat fluxes based on Fourier’s law. On this basis, an equivalent interface heat-transfer parameter heff is introduced as a physically interpretable descriptor that links interface temperature discontinuity and contact heat flux through qcontact=heff(Ts−Tc). This formulation enables direct assessment of interface consistency from the predicted temperature field, rather than indirect inference from global error metrics. An FEM-generated dataset comprising 150 two-dimensional transient fire-heating cases is used, covering perfect-bond, constant-conductance, and temperature-dependent interface conditions. The proposed model achieves an overall validation RMSE of 7.73 °C. Local RMSEs are 11.91 °C, 14.86 °C, and 2.00 °C at the fire-exposed surface, front interface, and back interface, respectively. The predicted heff shows strong agreement with reference values, with a correlation coefficient of 0.956. In addition, the normalized contact heat-flux error decreases from 0.31 to 0.26. A locked checkpoint held-out test further confirms robustness, yielding an overall RMSE of 7.71 °C and an heff correlation of 0.946. The proposed framework is applicable to rapid thermal analysis of interface-dominated composite structures under fire exposure when surface-temperature histories, material-region labels, and paired interface samples are available. Overall, the results indicate that the method improves interface interpretability while maintaining accurate and computationally efficient temperature-field prediction.