Multi-component optimization and hydration mechanism of recycled concrete performance based on machine learning

To improve the utilization rate of recycled aggregates and enhance the performance of recycled concrete through mixture proportion adjustment, this study investigates the synergistic regulation of the fifth and sixth constituents, namely, fly ash and superplasticizer, by combining physical experiments and machine learning analysis. The objective is to predict the extent to which the performance of recycled concrete can be optimized and to elucidate its hydration mechanism. The results show that, based on machine learning and experimental data, the replacement ratio of recycled aggregates is the key factor influencing the performance of recycled concrete. When the recycled aggregate replacement ratio ranges from 30% to 50%, the slump of the fresh mix can be increased by up to 315% by adjusting the contents of fly ash and superplasticizer, with the superplasticizer dosage fixed at 0.15%. When the fly ash content ranges from 20% to 40%, the 28-day compressive strength of recycled concrete increases by 10.2%–11.0%. In addition, a recycled concrete performance prediction model was established using a random forest regression algorithm, and the predicted values agreed well with the experimental results, with a coefficient of determination of R2 > 0.94. Feature importance analysis further quantified the contribution weights of recycled coarse aggregate, water-reducing agent, and fly ash content to the performance indices, confirming the effectiveness of multi-component optimization. Microscopic observations revealed that fly ash, due to its relatively low reactivity, significantly reduced the amount of Ca(OH)2 after incorporation, whereas the addition of superplasticizer markedly increased the amounts of AFt and C–S–H in the cementitious matrix.