Kolmogorov–Arnold Network modeling of recycled aggregate concrete under elevated temperatures: Predicting residual strength and elastic modulus through temperature-based segmentation

The global demand for sustainable construction materials has accelerated the development of recycled aggregate concrete (RAC) as an environmentally friendly alternative to conventional concrete. Despite its ecological advantages, the thermal degradation behavior of RAC remains insufficiently understood due to the complex interactions among recycled aggregate content, microstructural heterogeneity, and temperature‑dependent properties. This study introduces a novel modeling framework that, for the first time, employs Kolmogorov–Arnold Networks (KANs) for predicting thermo-mechanical properties of RAC under elevated temperatures (150–1000°C). Unlike conventional machine learning approaches that treat the entire temperature spectrum uniformly, this research implements intelligent temperature-based segmentation combined with KAN’s superior nonlinear approximation capabilities to capture the distinct degradation mechanisms across different thermal regimes, addressing critical gaps in existing literature where advanced interpretable algorithms and range-specific modeling have been largely overlooked. A curated database of 260 experimental results collected from 15 international studies was systematically classified into three temperature intervals for compressive strength (150–300°C, 300–600°C, and 600–1000°C) and two ranges for elastic modulus (150–500°C and 500–800°C). The models employed five key predictors: water‑to‑powder ratio, pozzolanic materials‑to‑aggregate ratio, recycled aggregate percentage, ambient compressive strength, and the ratio of test to ambient temperature. The optimal KAN models achieved outstanding performance with determination coefficients (R²) of 0.935, 0.964, and 0.962 for the three compressive strength groups, and 0.908 and 0.971 for the two elastic modulus models. On average, residual compressive strength declined from approximately 29 MPa in the 150–300°C range to 7.2 MPa in the 600–1000°C range, while the elastic modulus decreased by about 63% beyond 500°C. Parametric analysis confirmed that temperature ratio and recycled aggregate content were the dominant factors influencing thermal degradation. The proposed KAN‑based approach offers a high‑fidelity, interpretable, and computationally efficient tool for predicting the thermo‑mechanical behavior of RAC under elevated temperatures, providing valuable insights for the design of safer and more sustainable fire‑resistant concrete structures.