Assessing the impact of pozzolanic materials on the mechanical characteristics of UHPC: analysis, and modeling study
Diar Fatah Abdulrahman Askari, Sardam Salam Shkur Shkur, Azad Abdwlqader Mohammed, Hersh F. Mahmood, Brwa Hamah Saeed Hamah Ali, Soran Abdrahman Ahmad
Abstract
Due to the required time for the trail mix for the concrete production, for obtaining required strength or finding the effect of changing the rate of the composed materials on the concrete behavior, with high rate of the cost when the concrete mix be high strength concrete or high performance concrete, or ultra high performance, the availability of the equation that capable to show the effect of each independent parameters on the produced property of the mix become necessary. This article reviews previous experimental work on the use of various pozzolanic materials, including Silica Fume (SF), Fly Ash (FA), Ground Granulated Blast Furnace Slag (GGBS), Quartz Powder (QP), Rice Husk Ash (RHA), and Metakaolin (MT), in the production of Ultra-High Performance Concrete (UHPC).In addition to gathering previous experimental data on the use of various pozzolanic materials to predict the compressive strength of UHPC, the collected data are modeled using several statistical techniques, including Support Vector Machine (SVM), Ensemble Boosting Tree (EBT), and Artificial Neural Networks (ANN). The data are divided into two groups: 70% for training and 30% for testing. The performance of the proposed models is then evaluated using statistical tools such as the Coefficient of Determination (R2), Scatter Index (SI), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) to identify the most reliable method for predicting the compressive strength of UHPC. The results indicate that replacing cement with silica fume is more effective in enhancing both the compressive and flexural strengths of concrete compared to other materials tested. Among the statistical models, the Ensemble Boosting Tree (EBT) method demonstrates superior accuracy, as evidenced by its lower Scatter Index (SI) which is 0.119 and lower than the scatter index value for the support vector machine, and ANN by 24.6, 32.7% and higher Coefficient of Determination (R2) which is 0.79 and higher than the R2 for SVM, and ANN model by 24.4, 46.2% respectively.