An integral approach for testing and computational analysis of glass powder in cementitious composites

Abstract

In this research, test methods followed by supervised machine learning (SML) modeling were used to examine the effect of glass powder as a partial sand and cement substitute on the water absorption capability of cement mortar. This study also utilized SHapley Additive exPlanations (SHAP) analysis to determine the significance of raw materials. The dataset was derived from experiments and was used to build SML-based prediction models, including support vector machine, bagging, and AdaBoost. According to the test results, using glass powder reduced the water absorption capacity of cement mortar, and the optimum glass powder contents were noted to be 10 % as cement and 15 % as a sand substitute. Additionally, the built SML models indicated good accord with test outcomes and could be applied to calculate the water absorption of cement mortar comprising glass powder. Moreover, the AdaBoost model was found to be more accurate (R2 = 0.98) than the support vector machine (R2 = 0.95) and bagging regressor (R2 = 0.95) models in terms of R2, the difference between test and estimated findings, and the assessment of the errors.