Development of high-efficiency 3D models for simulation of building structures in fires

Abstract

The fire safety evaluation of isolated structural members cannot address the changing boundary conditions due to the interaction between heated members and the rest of the structural system. Since conducting full-scale fire tests is expensive, complex, and time-consuming, it is essential to develop 3D models for numerical simulation to analyse structures subjected to various fires. The first challenge is to include slab models in numerical simulations without significantly compromising computational efficiency. The second challenge is to capture the complex local behaviour of structural members in a fire in a structural model. This thesis aims to develop numerical solutions to resolve these challenges, providing a numerical tool to incorporate computational complexity into built-in models and enable high efficiency and convenience of use. Based on previous studies of modelling reinforced concrete slabs, modelling steel-concrete composite slabs has been enabled in OpenSees for fire with the work in this thesis. Using 3D structural models with slabs, the load redistribution and failure patterns due to localized column failure and multi-floor fires have been studied. Moreover, to simulate the column behaviour in a structural system, a modular AI model to account for local buckling, global buckling, and thermally induced degradation has been developed. This model can co-work with the 3D structural system model to overcome the limitations of traditional beam-column elements and maintain the feature of fast response. Based on the work presented in this thesis, the following outcomes can be summarized: (1) Steel-concrete composite slabs can be modelled using ribbed sections and flat sections of a unified reference plane supported by the new development in thermo-mechanical shell elements in OpenSees for fire; (2) The ribbed section and flat section of composite slabs can be further integrated to simplify the modelling process similar as modelling flat slabs, which employs a novel distribution scheme developed to allocate the overall deformation to component sections with analytical base and experimental validation; (3) The load redistribution mechanisms in a 3D mid-rise steel-concrete composite structure are revealed using the 3D simulation models, which show that the ‘pull-up’ effect from those non-fire floor systems can mitigate the deflection after local column failure except in a multi-floor fire scenario; (4) AI column models trained with SVR method and ANN method can accurately predict the fire performance of steel columns subjected to fires to address the local behaviour and to ensure model efficiency. (5) An AI-FE hybrid simulation framework is developed to exploit the 3D models with slabs and the modular AI models, which would become a promising tool for ‘structures in fire’ simulation in future to ensure structural safety in fires.