RESPONSE OF COMPOSITE BRIDGE GIRDERS EXPOSED TO REALISTIC FIRE SCENARIOS USING A DEEP LEARNING BASED APPROACH

Abstract

This paper presents an accurate and predictive framework for evaluating composite bridge performance under realistic fire scenarios by exploiting CFD-FE coupling and advanced ANN modelling capabilities. A series of realistic fire scenarios are developed by conducting CFD simulations. Various parameters affecting the fire behaviour are included in CFD model such as bridge soffit height, fire intensity, and fire location along the bridge span. The time temperature history obtained from CFD simulations is coupled with Finite Element (FE) model to analyse the structural performance of composite bridges with different thicknesses of fire protective coatings. While performing CFD and FE simulations is a tedious and computationally expensive task, an ANN model is trained on simulation data to provide rapid predictions of structural fire response of composite bridges. The ANN model proposed in this study predicts crucial parameters such as failure time, failure temperature and vertical displacement behaviour for assessing structural performance and safety of composite bridges during fire events. This enables fire safety engineers to quickly evaluate the structural response under numerous fire scenarios and corresponding design alternatives, resulting in an efficient decision-making tool in bridge fire design and fire safety assessment.