AI based surrogate prediction model of fire endurance time for RC frame structure

HyunKyoung Kim , Ju-young Hwang

Abstract

Accurate prediction of fire-damaged RC structural behavior is essential for safety assessment; however, traditional numerical analysis becomes computationally intensive due to complex thermo-mechanical responses including non-mechanical strains at elevated temperatures. This study develops ML surrogate models (NN, XGB, LGBM) to predict RC member fire endurance time using a 4.37 million-point dataset generated from P-M diagrams obtained via high-fidelity numerical analyses. Tree-based ensembles outperformed NN in large-data regimes (test error 96%), providing superior interpretability through feature importance and monotonic constraints. Specifically, by defining the input conditions through a flexible 7-variable framework (B, H, BN, HN, M, P, R), this study enables the direct generation of P-M interaction diagrams under any arbitrary loading scenarios. Frame-level validation comparing predictive model and numerical analysis results on a 1-bay, 1-story RC frame confirmed applicability of proposed model with consistent column failure patterns, enabling very rapid predictions showing similar trends.

Key Words

extreme gradient boosting (XGB); fire-damaged RC; fire analysis; fire endurance time; neural network (NN)

Address

HyunKyoung Kim — Department of Civil and Environmental Engineering, Korean Advanced Institute for Science and Technology, Daejeon 34141, Republic of Korea
Ju-young Hwang — Department of Civil Engineering, Dong-Eui University, Busan 47340, Republic of Korea

PDF Viewer

Preview is limited to the first 3 pages. Sign in to access the full PDF.

← Back to Volume 37, No. 5