Generalized fragility curves for concrete columns exposed to fire through surrogate modelling

Abstract

Common structural fire design relies on recommendations from design codes, or (a single or small set of) more advanced numerical analyses. When applying such procedures to the design of structures under normal loading conditions, adequate safety is ensured through calibrated safety factors and ample experience with structural failures. This is however not the case when considering accidental fire loading, where the stochasticity in the structural fire behaviour is rarely fully acknowledged. Therefore, a significant interest in the use of probabilistic approaches to evaluate structural fire performance, which take into account the uncertainty associated with model parameters, can be observed among researchers, with a special focus on the development of fragility curves. The calculation of fragility curves is, however, a laborious task, demanding huge computational expense, mainly attributed to the adoption of advanced calculation procedures and the need for a large number of model evaluations. The present study contributes to addressing the limitations imposed by these computational requirements through the development of surrogate models for fire exposed structural members. To achieve this, a framework for carrying out probabilistic studies of structures under fire through the use of surrogate modelling is presented. The framework is applied to a concrete column subjected to a standard fire and proves efficiency and accurateness for the selected simple example. Future studies will investigate the applicability of the framework to structural assemblies under physically-based fires.

Regression-Based Surrogate Models for the Probabilistic Study of Fire Exposed Composite Structures Considering Tensile Membrane Action

“…In [8], the ML model developed based on ANN could predict the fire resistance of timber with an accuracy of 99 %. Adopting a ML algorithm, Chaudhary et al [9] carried out a probabilistic study for the evaluation of the load capacity of a reinforced concrete (RC) column under fire. The regression-based surrogate model showed a promising performance in approximating the non-linear finite element (FE) model for the RC columns (load capacity at 1 % quantile predicted with an error of less than 5 %).…”