Experimental and Data-Driven analysis on compressive strength of steel fibre reinforced high strength concrete and mortar at elevated temperature

Abstract

The compressive strength of concrete and mortar under thermal exposure can be determined using the method of elevated-temperature strength test or residual strength test after cooling from heating. Based on the experimental findings reported in this study, the results obtained by these two methods should not be used interchangeably because the tested high strength mortar exhibits distinctively different trend in the change of its compressive strength and elastic modulus when tested under these two conditions. The addition of steel microfibres is found to improve the compressive strength, elastic modulus and post-peak ductility of high strength mortar at elevated temperatures, but such beneficial effect becomes insignificant in residual strength tests after cooling. A database containing 674 concrete and mortar specimens tested at elevated temperatures is established. Using this database, a prediction model based on XGBoost algorithm is developed for the compressive strength retention factors of steel fibre reinforced concrete and mortar at elevated temperatures. This model is shown to improve the accuracy of prediction by 21% compared to the tabulated data in EN1992-1–2 and outperform the other seven machine learning methods evaluated in this study, including Random Forests, Artificial Neural Network, AdaBoost, k-Nearest Neighbours Regression, Multivariate Adaptive Regression Splines, Support Vector Regression and Linear Regression. The XGBoost model shows that the top four most important factors that determine the compressive strength of concrete and mortar at elevated temperatures are heating temperature, compressive strength of concrete and mortar at room temperature, content of steel fibre and aspect ratio of specimen. Based on this finding, less important factors are eliminated from the input features to improve the computational efficiency of the prediction model.