Experimental and Machine Learning-Based Investigation of Coarse Aggregate Characteristics Impact on Mechanical Properties of Concrete

This research investigates the impact of coarse aggregate (CA) type, shape, and specimen size on the compressive behavior of concrete, aiming to better understand how these factors affect its mechanical performance. Eight concrete mixtures were designed according to four different concrete mix design (CMD) codes using two types of coarse aggregates: crushed basalt and naturally rounded, both with a 15 mm size. A total of 96 concrete samples were tested to evaluate their failure mode, compressive strength (CS), energy accumulation (GA), and post-peak fracture energy (GF). The results show that concrete made with basalt CA offered significantly higher CS (by 7% to 40%), GA (by 34% to 57%), and GF (10% to 48%) compared to concrete made with natural CA across different CMD codes and specimen dimensions. Larger cylinders demonstrated higher CS than smaller cylinders, ranging from 7% to 19%. The incorporation of basalt CA enhanced the toughness and ductility of concrete, leading to superior post-peak behavior. In addition to the experimental program, four machine learning algorithms, i.e., Extreme Gradient Boosting (XGB), Gradient-Enhanced Regression Tree (GBR), Random Forest (RF), and Support Vector Regression (SVR), were employed to forecast the concrete’s CS. RF (R2 = 0.93) and gradient boosting models (R2 = 0.92) showed remarkable accuracy, whereas SVR underperformed. The feature importance and SHAP analysis identified cement content and CA type as the primary determinants of CS, while the water–cement ratio served as a crucial regulator. Moreover, a graphical user interface tool was developed to practically allow engineers to rapidly estimate concrete CS, bridging the gap between experimental validation and practical use.