Effect of sulfate freeze-thaw on the stress-strain relationship of recycled coarse aggregate self-compacting concrete: Experimental and machine learning algorithms

Abstract

Based on experimental studies, this paper proposes two efficient machine learning models to evaluate the macroscopic mechanical properties of recycled coarse aggregate self-compacting concrete (RCASCC) after sulfate freeze-thaw action. Initially, the stress-strain curves of RCASCC after sulfate freeze-thaw action were measured, yielding the peak stress (σ_c), peak strain (ε_c), and elastic modulus (E) for each group of RCASCC. Among these, a strong linear correlation was observed between the elastic modulus and the peak stress. An RCASCC uniaxial compression behavior model considering the effects of sulfate freeze-thaw cycles has been established. This model is used to predict the stress-strain characteristics of RCASCC under uniaxial compression after exposure to sulfate freeze-thaw cycles. Using the stress-strain data of uniaxial compression test, a machine learning model for RCASCC after sulfate freeze-thaw cycle was developed by using MATLAB. Eight different machine learning algorithms are used to train and test the model, and six performance indicators are used to measure its generalization performance. The three models, RF, ET and GB, exhibit the highest prediction accuracy compared to other machine learning models. The relative importance of strain and Na₂SO₄ mass fraction is largest and smallest in the three models, RF, ET and GB, respectively. Based on RF, ET and GB models with good predictive performance, we plot the stress-strain curves of the predicted models. The fit is better for the ascending and descending segments of the curves in each group, and worse for the curves near the peak. RF and ET can better predict the macroscopic mechanical properties of RCASCCC under different conditions.