Applying machine learning in nondestructive evaluating the subsurface tensile strength of cementitious flooring.

Abstract

This study investigates the potential of selected machine learning algorithms to predict the subsurface tensile strength of eco-friendly cementitious floor composites containing fly ash, ground granulated blast furnace slag, and granite processing waste. The experimental database, built on 23 mixtures with up to 30% SCM replacement, was completed and analysed statistically. Destructive testing was used to obtain reference values of subsurface tensile strength. Three machine learning algorithms k-Nearest Neighbors (kNN), Adaptive Boosting (AdaBoost), and Support Vector Machines (SVM) were trained using 5-fold cross-validation. The kNN model with Manhattan distance and distance-based weighting achieved the highest accuracy (R = 0.862, MAPE = 6.81%), outperforming the other models. The findings demonstrate that appropriately calibrated machine learning models can serve as reliable tools for non-destructive prediction of tensile strength in sustainable cement composites, reducing time and material losses in quality control.