Physics-Informed Neural Networks for Modelling Reinforced Concrete Beams Under Fire Exposure

The advancement of performance-based fire safety design in reinforced concrete (RC) structurescritically relies on the availability of precise numerical simulation tools capable of accurately capturing thebehaviour of RC elements under fire exposure. This paper develops a three-dimensional numericalsimulation based on Physics-Informed Neural Network (PINN), which is designed to meticulously predictboth the thermal and mechanical responses of RC beams subjected to fire conditions. A key focus of thePINN model is to represent the interfacial bond-slip phenomena occurring between the reinforcing steeland concrete—a nuanced aspect often overlooked in prior numerical investigations. The accuracy of theproposed PINN model is rigorously assessed through comparative analysis with established experimentaldata. This study demonstrates that integrating the intricacies of steel-to-concrete interfacial behaviourenhances the model's predictive capability, particularly in forecasting the deflection of RC beams under firescenarios. Moreover, the predictive insights afforded by the PINN model enable a detailed examination ofstress distribution and evolution within both the reinforcing steel and concrete domains, facilitating a deepercomprehension of local responses in RC beams exposed to fire. Beyond its validation, the PINN modelpresented herein holds significant practical utility; it stands poised for direct application in performancebasedfire safety design of RC beams, offering a refined toolset for engineers and designers. Finally, itsversatility extends to facilitating parametric investigations aimed at refining and establishing simplifieddesign guidelines—a testament to its potential for advancing the state-of-the-art in fire-resistant RCstructural design.

کلیدواژه ها:

Deep Learning ، Physics-Informed Neural Network ، Reinforced Concrete Beam ، FireExposure ، Thermo-Mechanical Simulation