The efficiency of statistical and artificial neural network techniques on evaluating and predicting compressive strength of sustainable SCC with basalt fibers

Abstract

This work presents the power of employing statistical and Artificial Intelligence (AI) approaches in analyzing and predicting the effect of basalt fibers on the mechanical behavior of high-strength sustainable self-consolidating concrete (SCC). To alleviate the greenhouse gas emission from the Portland cement manufacturing process, an experimental program with a concrete mix design of up to 80% of the cement content is replaced with a combination of different supplementary materials. The used cementitious materials (i.e., ground granulated blast furnace slag (GGBS), fly ash (FA), and silica fume (SF)) were employed in the design mix as binary, ternary, and quaternary blends with different proportions of cement replacement. Additionally, basalt fibers were integrated into the concrete mixtures with three dosages from the total cementitious materials weight, i.e., 0.5%, 0.75%, and 1.00%. The aim of introducing those fibers was to investigate their ability in enhancing or regaining the mechanical strength compromised with cement replacement. As the compressive strength results showed some ambiguous trends with basalt fiber mixes, analysis of variance (ANOVA) was used to determine the controlling parameter on the compressive strength behavior. The analysis elucidated that the highest contributor to the strength was cement with 58%, followed by the GGBS with 21%, while the remaining factors were 12%, 6%, and 3% for FA, SF, and basalt fibers, respectively. Further, all the experimental data were utilized in an Artificial Neural Networks (ANN) tool to predict the compressive strength of the developed SCC mixes. For maximum prediction accuracy, different network configurations for one-hidden and two-hidden layers networks were exploited. Results demonstrated that high prediction accuracy was attained through 6-28-1 and 6-25-20-1 networks with a high confidence level limited to a ±10% error margin. It was concluded that statistical, and ANN techniques to analyze, capture, and predict the effect of different additives on the concrete compressive strength are efficient despite having ambiguous behavior with the experimental results.