Abstract:
Utilization of cementitious compounds, such as concrete, mortar, and cement paste is predominated in the construction industry. Hence, rising CO2 emissions, overexploitation of natural resources, and elevated energy consumption for production of raw materials have become serious environmental challenges in the contemporary world. The application of supplementary cementitious materials (SCMs) is considered the paramount solution to this dilemma. However, traditional methods require a substantial amount of time and cost to assess the engineering properties of cement compounds for application of alternative cementitious materials. This study reviews the applications of SCMs in cement compounds and use of the competence of machine learning (ML) techniques to predict engineering properties. Among reviewed literature for concrete, 46%, and 27% studies utilized Fly ash and rice husk ash (RHA), respectively. For mortar, 30%, and 10% studies utilized Fly ash and RHA, respectively. Despite eggshell powder (ESP) being a promising calcium source, either utilization or prediction of engineering properties of eggshell powder-based cementitious compounds has not been previously focused. Among studies combining ML models with cementitious compounds, over 50% focused on concrete, whereas about 40% focused on mortar. Prediction of compressive strength was the main focus in previous studies. Flexural strength, shear strength, split tensile strength, and elastic modulus were not focused even though these are promising engineering parameters to examine the quality of cementitious compounds. ML algorithms, such as Support Vector Machine, Artificial Neural Network (ANN), Decision Tree (DT), and evolutionary algorithms, were mostly applied for predicting the compressive strength of concrete. For predicting the compressive strength of mortar, regression techniques, including linear and ridge regression with boosting methods, were more frequently employed. Recently, a diverse range of ML techniques has emerged, shifting from ANN to DT-based methods and boosting techniques. Ensemble approaches, which integrate multiple ML models to improve prediction accuracy has been less studied.
