Deep machine learning approaches for evaluating the Comprehensive Strength of Ultra-high Performance Concrete

Abstract

Ultra-high performance Concrete (UHPC) C.S.S. is determined by the kind, characteristics, and composition of its material ingredients. Using intelligent techniques, such as the ANN, to create a prediction framework that adjusts within an implementation dataset is often necessary to empirically capture this link. However, scientists cannot quantitatively describe its contents because of its opaque character. This research uses two deep machine learning approaches to determine the crucial material components that impact the Artificial Neural Network: Neural Interpretation Diagrams (N.I.D.s) and Sequential Feature Selections (S.F.S.). 110 Ultra-High Performance Concrete C.S.S. tests with different material amounts were combined into a database to train the ANN. Thus, Four material components were chosen: cement, fly ash, silica fume, and water. After that, the ANN was used with these material components to make more accurate predictions than those made by the model that included all eight material constituents (r2 = 21.5% and NMSE = 0.035) (r2 = 80.1% and 0.012). Ultimately, parametric research was carried out, and a nonlinear regression framework was created depending on the four chosen material elements. The use of Artificial Neural Networks with Sequential Feature Selections and Neural Interpretation Diagrams was shown to significantly increase the model's accuracy and provide insightful information about the ANN CS forecasts for various Ultra-High Performance Concrete combinations.