Prediction of Compressive Strength Loss of Normal Concrete after Exposure to High Temperature

Abstract

In recent years, there has been an increasing number of fires in buildings. The methods for detecting residual properties of buildings after fires are commonly destructive and subjective. In this context, property prediction based on mathematical modeling has exhibited its potential. Backpropagation (BP), particle swarm algorithms optimized-BP (PSO-BP) and random forest (RF) models were established in this paper using 1803 sets of data from the literature. Material and relevant heating parameters, as well as compressive strength loss percentage, were used as input and output parameters, respectively. Experimental work was also carried out to evaluate the feasibility of the models for prediction. The accuracy of all the models was sufficiently high, and they were also much more feasible for prediction. Moreover, based on the RF model, the importance of the inputting parameters was ranked as well. Such prediction has provided a new perspective to non-destructively and objectively assess the post-fire properties of concrete. Additionally, this model could be used to guide performance-based design for fire-resistant concrete.