Prediction of Fire-induced Spalling in Fiber-reinforced Concrete by Artificial Neural Networks

Abstract

Predicting fire-induced spalling in fiber-reinforced concrete (FRC) incorporating polypropylene (PP) and steel fibers under elevated temperatures presents significant challenges due to the complex interactions within the concrete matrix. Conventional methods have limitations in accurately capturing these complex behaviors. To address this, the study proposes two artificial neural network (ANN) models: the first (ANN1) for evaluating concrete mix compositions and the second (ANN2) for assessing compressive strength characteristics. A detailed dataset of 318 and 321 test samples from existing reference was applied to train ANN1 and ANN2, respectively. Validation was performed using 24 distinct concrete mixtures, totaling 96 tests, which included seven plain concrete (PC) mixes, four PP fiber-reinforced high-performance concrete (HPC) mixes, three PP fiber-reinforced ultra-high-performance concrete (UHPC) mixes, and ten hybrid fiber-reinforced UHPC mixes containing both PP and steel fibers. Results show that ANN1 achieved a predictive accuracy of 89.6%, while ANN2 reached 84.4%, demonstrating the reliability of the proposed ANN models in assessing the risk of explosive spalling in fiber-reinforced concrete under high-temperature conditions.