ML-based structural response forecasting and early warning system for RC structures under fire conditions

Abstract

Structural response forecasting and early warnings during fire events are crucial for enhancing structural safety and supporting effective fire rescue operations. This study proposes an integrated finite element (FE)-based machine learning (ML) framework for forecasting structural responses and establishing an early warning system (EWS) for reinforced concrete (RC) frame structures subjected to fire. A Long Short-Term Memory (LSTM) network is trained using a comprehensive FE simulation dataset generated through a macro-modeling strategy in the GiD–OpenSees interface, with stochastic input parameters to account for uncertainties in fire exposure, material properties, and applied loading. The framework is demonstrated on a three-story, three-bay RC frame, where structural displacements and reinforcement temperatures are forecasted using limited inputs consisting of compartment gas temperatures and joint displacements at peripheral structural locations, over an initial time window. The trained ML model shows high predictive accuracy, with mean absolute error ratios below 5% and coefficient of determination ($R^2$) $\ge 0.95$. An EWS configured from the forecasted response achieves an 85% recall efficiency relative to FE-based failure predictions. The findings highlight the potential of FE-informed ML models to enable structural response forecasting and graded collapse warnings, thereby providing a decision-support framework for fire rescue operations.