Experimental and machine learning-guided optimization of tensile strength of concrete incorporating silica fume, waste foundry sand, and polypropylene fibers

Abstract

Concrete has inherently low tensile strength, which restricts the performance and durability of a concrete structure. This study investigates the effect of Silica Fume (SF), Waste Foundry Sand (WFS), and Polypropylene Fiber (PPF) on Split Tensile Strength (TS) of M40 grade concrete. Experimental mixes were formulated by systematically replacing the SF with WFS (5–40%) and the WFS with PPF (0.5–2%) and cylindrical specimens were cast and cured under controlled conditions for 3–90 days. The results presented that the WFS (30%) and SF (15%) mixes had 90-day tensile strengths of 5.65 MPa and 6.11 MPa, respectively, with 10.6% and 8.0% increase compared with control, and the mix having ternary composition with PPF (1.0%) had the highest tensile strength of 6.11 MPa, which was 5.2% greater than the binary mix of SF and WFS. Machine learning Multilayer Perceptron (MLP), Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) models were developed to predict tensile strength based on material proportions, curing period (CP), and temperature (TEMP). The optimized SVM model showed the highest accuracy (R 2 = 0.9871), root mean square error (RMSE) (0.1097 MPa), mean absolute error MAE (0.0724 MPa), and standard deviation ratio (RSR) (0.1135), The SHAP analysis showed that temperature (45%) and curing period (24%) were the most important factors, followed by WFS, Natural fine aggregate (NFA), PPF, SF, and cement (C). The study confirms the effectiveness of the application of SF, WFS and PPF to produce sustainable, high-performance concrete and offers a reliable basis for the design of mixes with improved tensile strength, crack resistance and durability.