Analysis of Models to Predict Mechanical Properties of High-Performance and Ultra-High-Performance Concrete Using Machine Learning

Abstract

High-Performance Concrete (HPC) and Ultra-High-Performance Concrete (UHPC) have many applications in civil engineering industries. These two types of concrete have as many similarities as they have differences with each other, such as the mix design and additive powders like silica fume, metakaolin, and various fibers, however, the optimal percentages of the mixture design properties of each element of these concretes are completely different. This study investigated the differences and similarities between these two types of concrete to find better mechanical behavior through mixture design and parameters of each concrete. In addition, this paper studied the correlation matrix through the machine learning method to predict the mechanical properties and find the relationship between the concrete mix design elements and the mechanical properties. In this way, Linear, Ridge, Lasso, Random Forest, K-Nearest Neighbors (KNN), Decision tree, and Partial least squares (PLS) regressions have been chosen to find the best regression types. To find the accuracy, the coefficient of determination (R2), mean absolute error (MAE), and root-mean-square error (RMSE) were selected. Finally, PLS, Linear, and Lasso regressions had better results than other regressions, with R2 greater than 93%, 92%, and 92%, respectively. In general, the present study shows that HPC and UHPC have different mix designs and mechanical properties. In addition, PLS, Linear, and Lasso regressions are the best regressions for predicting mechanical properties.