Application of Artificial Neural Networks for Developing Temperature-Dependent Fragility Curves for Vulnerability Assessment of I-Girder Bridges

• Published in: Numerical Methods in Civil Engineering Vol. 9; no. 4; pp. 24 - 40
• Main Authors: Saeed Sabori Ghomi, Saman Shiravand, Mohammad Mahdi Zare Zardeyni, Najib Rabiee
• Format: Journal Article
• Language: English
• Published: K. N. Toosi University of Technology 01.06.2025
• Subjects: artificial neural networks; grid search; high temperatures; multi-hazard; post-fire seismic performance; steel i-girder bridges
• ISSN: 2345-4296; 2783-3941
• DOI: 10.61186/NMCE.2504.1087

Although previous investigations have shown that a bridge’s overall capacity may remain largely intact after a fire, its seismic performance in the aftermath of such events remains poorly understood. The primary objective of this study is to investigate the influence of fire on the seismic performance of a multi-span simply supported steel I-girder (MSSSS-IG). To achieve this, an artificial neural network (ANN) model was used to develop a multivariate probabilistic seismic demand model (MPSDM) and fragility curve. A total of 1,600 three-dimensional MSSSS-IG bridge models were generated using the OpenSees tool, incorporating material and geometric variability through Latin Hypercube Sampling (LHS). A set of 1,600 ISO 834 fires featuring peak temperatures varying between 200 °C to 1,000 °C was developed. The maximum temperature in the column was determined through heat transfer analysis. Accordingly, column reduction factors were computed via Eurocode provisions. Bridge, reduction factors, and input ground motion records were randomly paired using nonlinear response history analysis (NRHA). The XGBoost technique and grid search were employed to identify the important features and calibration hyperparameters of ANNs, respectively. It can be pointed out that the proposed ANN algorithm accurately estimates the component demands. Moreover, fragility findings demonstrate that local fire exposure in the column, ranging from 12.30 % and 22.30 %, increases the probability of system-level bridge failure.