AI-driven modeling for predicting compressive strength of recycled aggregate concrete under thermal conditions for sustainable construction

Abstract

This research utilizes sophisticated artificial intelligence (AI) methodologies to forecast the compressive strength of recycled aggregate concrete (RAC) under different temperature scenarios, marking a notable advancement in sustainable construction methodologies. Two distinct models employing Artificial Neural Networks (ANN) and Gene Expression Programming (GEP) were created based on an extensive dataset that includes 157 experimental samples from eight reputable studies conducted between 2014 and 2024. The ANN models underwent optimization via Random Search Hyper-Parameter Tuning, resulting in high prediction accuracy, with correlation coefficients (R 2 ) surpassing 0.9. To prevent overfitting, dropout techniques and L1 & L2 regularization were applied, ensuring strong generalizability. The explicit mathematical equations generated through GEP offer practical applications for engineers involved in thermal design. For each algorithm, two complementary models were developed: one for predicting compressive strength at ambient temperature and another for estimating residual strength following exposure to elevated temperatures. A detailed comparative analysis revealed that ANN models outperformed GEP in terms of predictive accuracy, while GEP models offered interpretable equations for practical engineering use. The study also conducted a comprehensive evaluation against existing standards, demonstrating the superior reliability of the developed AI-driven models in predicting RAC performance at elevated temperatures. Furthermore, a rigorous sensitivity analysis identified key influencing parameters, particularly the water-to-cement ratio and recycled aggregate content, offering valuable insights into the thermal and mechanical behavior of RAC. The findings of this research contribute significantly to sustainable construction by providing a robust AI-based predictive framework for optimizing RAC mix designs, guiding the development of thermal-resistant concrete formulations, and informing future structural design standards for recycled materials in high-temperature applications. • Dual AI models (ANN & GEP) predict RAC compressive strength across various temperatures. • ANN optimized via Random Search Tuning, achieving high accuracy (R² > 0.9). • Dropout, L1 & L2 regularization prevent overfitting, boosting model generalization. • Mathematical formulas provided for RAC strength prediction at different temperatures. • Model results outperform existing codes; detailed ANN vs. GEP comparison included.

Topics & Concepts