MODELING SUSTAINABLE CONCRETE BEHAVIOR USING FOUNDRY SAND, GENE EXPRESSION PROGRAMMING, AND MACHINE LEARNING METHODS

Abstract

Sustainable materials are increasingly used in the construction industry to reduce the environmental impact. One promising approach is the use of residual foundry sand (RFS), an industrial byproduct, as a partial replacement of natural sand in concrete. This helps to divert industrial waste away from landfills, conserve natural resources and encourage environmentally sustainable construction. The mechanical properties of RFS based concrete were predicted using advanced computational techniques like Genetic Programming (GP) and Machine Learning (ML). Predictive models for compressive strength (f’c), elastic modulus (Ec) and split tensile strength (fst) have been developed based on material properties, mix proportions and environmental factors. Four mathematical models were established on the basis of experimental data. Linear Model #1 gave an excellent prediction of the compressive strength, R2 = 0.99. Exponential Model #3 yielded R2 = 0.985. The linear model #2 was found to fit the elastic modulus very well (R² = 1.00) and the nonlinear model #4 was the most accurate (R² = 0.998). The predicted ranges were 20-50 MPa, 15-30 GPa and 2-5 MPa for the compressive strength, elastic modulus and split tensile strength respectively. The results indicate that GP and ML can successfully describe the complex behavior of RFS concrete, allowing for reliable property prediction and supporting sustainable concrete design and increased use of recycled industrial materials.