Computational fluid dynamics -informed virtual safety assessment of steel-framed structure with fire-induced ductile failure

• Coupled CFD-FEA analysis for the ductile failure analysis of steel structure. • Considering spatiotemporal dependent temperature field in real fire incidents. • AI-driven safety assessment for fire-exposed structures in random conditions. • Efficiency of proposed CI-VSA scheme is verified through two practical studies. This paper proposes a Computational Fluid Dynamics-Informed (CI) Virtual Safety Assessment (VSA) framework for predicting the time-dependent ductile failure of steel-framed buildings during fire incidents. By incorporating a CI-based physical model, the spatiotemporally nonlinear temperature field in real fire scenarios can be reproduced and used as thermal boundary conditions for sequential thermal-elastoplastic analysis, enabling the assessment of fire-induced structural responses. Additionally, non-deterministic material properties caused by manufacturing imperfections are considered to analyze their impacts on uncertain high-temperature structural ductile deformation. To achieve rapid assessment, a Virtual Modeling (VM) technique is introduced to capture the nonlinear relationship between physical input parameters and corresponding structural responses. The proposed CI-VSA framework is applied to two real steel structures, a steel-framed factory and a transmission tower, to verify its efficiency and accuracy. The results demonstrate that, compared to traditional simulation-based prediction methods, the proposed CI-VSA framework reduces computational resource consumption by 99% and achieves highly accurate predictions for most sample points, with relative errors below 1%, under a training sample size of 1,000. In practice, the CI-VSA framework enables continuous prediction of spatiotemporal structural responses through the analysis of fire-thermal-structural interactions, achieves real-time updates of structural safety statuses, and ultimately provides early-stage safety warnings.