Experimental and numerical validations of predictive models for compressive strength of pervious concrete

Abstract

Pervious concrete (PC) is a key material in sustainable urban development, but its design is complicated by the inverse relationship between permeability and compressive strength. To optimize its use, various predictive models—analytical, numerical, and data-driven—have been developed. However, a comparative validation of these diverse approaches on a consistent experimental dataset is lacking. This study aims to validate and compare the performance of three fundamental predictive frameworks: an analytical model based on micromechanics, a numerical simulation using the phase-field method, and an Artificial Intelligence (AI) approach, specifically a "white-box" symbolic regression model. An experimental case study involving the fabrication and testing of PC specimens was conducted to provide a new, independent dataset for validation. The results show that the symbolic regression, finite element method, and micromechanical models demonstrate strong agreement with the experimental data, achieving a high coefficient of determination. While black-box AI models often offer high accuracy, this study highlights that simpler, interpretable models provide a compelling balance of precision and practical applicability for engineering design.