Surrogate Model-Driven Estimation of Adiabatic Surface Temperature of Fire Exposed Suspension Bridge Towers

Abstract

Evaluating adiabatic surface temperature (AST) as the thermal response of fire-exposed bridge elements is a complex and time-consuming task. Correspondingly, this study streamlined fire dynamic simulator (FDS) and machine learning (ML) in a surrogate model to predict the AST of suspension bridge tower. For this, various FDS simulations were conducted for suspension bridge tower exposed to different vehicular fire conditions incorporating factors such as vehicle type, exposure duration, and wind conditions to generate a diverse bridge fire dataset for training of ML algorithms. Eight ML models were evaluated using performance metrics, whereby the random forest model demonstrated exceptional consistency and reliability in a fivefold cross-validation, maintaining a high R² value of 0.99 across all tests and showing stable MAE and MSE metrics, confirming its superior performance and robustness in predictive accuracy. The proposed surrogate model offers a robust and efficient tool for enhancing the resilience of bridge fire evaluations by providing a time-efficient solution that adapts quickly to a range of fire conditions.