Intelligent prediction of comprehensive mechanical properties of recycled aggregate concrete with supplementary cementitious materials using hybrid machine learning algorithms

Abstract

Considering the wide utilization of multi-component supplementary cementitious materials in recycled aggregate concrete, an intelligent prediction of comprehensive mechanical properties of this sustainable low-carbon concrete was proposed in this study using hybrid machine learning algorithms, rather than conducting tremendous experiments to establish a statistical empirical model previously. The database consisting of 980 datasets was used to train, validate and test the machine learning algorithms with various evaluation indicators. SHapley Additive exPlanations (SHAP) values were used for interpretability analysis of the predictive results of algorithms to quantify the contribution of each input feature. The results show that the hybrid algorithms possess higher prediction accuracy in terms of compressive strength, flexural strength, splitting tensile strength and elastic modulus compared to the single algorithms, among which Sparrow Search Algorithm-Extreme Gradient Boosting (SSA-XGB) stands out as the most promising algorithm model with satisfactory stability and generalization performance. For all mechanical properties, the effective water-binder ratio, cement content, and superplasticizer dosage are extremely important influencing factors. This study provides a systematic assessment framework for predicting the comprehensive mechanical properties of recycled aggregate concrete with supplementary cementitious materials, beneficial for the data-driven intelligent mix proportions design of sustainable low-carbon concrete in construction sector.

Keywords

• Recycled aggregate concrete
• Supplementary cementitious materials
• Mechanical properties
• Machine learning algorithms
• Interpretability analysis