Freeze–Thaw Resistance of Sustainable Concrete: Experimental Investigation, Damage Mechanisms, and AI/ML-Based Performance Prediction

Abstract

The growing need for sustainable construction materials has led to a need for concrete mixtures that reduce environmental impact and have long term durability. Freeze–thaw exposure is a major deterioration mechanism that can cause internal microcracking, stiffness degradation, surface scaling and loss of mechanical strength in concrete. In this paper freeze–thaw resistance of sustainable concrete (SCM, RCA and fiber reinforced) is investigated along with artificial intelligence and machine learning (AI/ML) based prediction framework. Sustainable concrete mixtures were developed by varying SCM replacement, RCA content, fibre dosage, water-to-binder ratio and other mixture parameters. Specimens were tested under repeated freeze–thaw cycles and compressive strength, relative dynamic modulus and mass loss were considered as main indicators of deterioration. With increasing number of freeze–thaw cycles compressive strength and relative dynamic modulus decrease gradually while mass loss increases. Higher RCA replacement generally makes concrete more susceptible to freeze–thaw deterioration because of its greater water absorption and more old-mortar interfaces. Appropriate SCM replacement and fibre reinforcement can improve retention of mechanical and durability properties. The simulated results show that mixtures with moderate SCM replacement, controlled RCA content, lower W/B ratio and higher fiber dosage are comparatively better resistant to freeze–thaw damage. Of all the studied mixtures M7 and M11 showed overall good performance (especially high retention of compressive strength and relative dynamic modulus and low mass loss). Besides experimental evaluation, ANN, RF, SVR, Gradient Boosting and XGBoost models for prediction of freeze–thaw performance from mixture and exposure parameters were proposed in this study. R², RMSE, MAE and MAPE values as well as explainable AI techniques (SHAP) can be used to evaluate models (along with experiment results) and understand the relative influence of RCA, SCMs, fibre content, water-to-binder ratio, curing age and freeze–thaw cycles. The integrated experimental and AI/ML framework proposed here is promising for understanding deterioration mechanisms, predicting freeze–thaw performance and sustainable concrete mixtures design for durable and resource efficient construction. The numerical results in this study are illustrative/simulated and should be experimentally validated before practical or journal-submission use.