Prediction of the effects of silica fume activation on the performance of recycled micro-powder mortar based on machine learning

In this study, silica fume was used as an activator to address the issue of mortar performance degradation caused by high-dosage recycled micro-powder, with the aim of enhancing its reactivity and utilization rate. A Box–Behnken Design–based Response Surface Methodology (RSM) was employed to optimize the experimental design involving silica fume, recycled micro-powder, and superplasticizer. Based on all experimental results, machine learning models such as Gradient Boosting Decision Trees (GBDT) were introduced for performance prediction and feature importance analysis. Finally, Scanning Electron Microscopy (SEM) and X-ray Diffraction (XRD) were used to analyze the microstructure and hydration products of silica fume–recycled micro-powder mortar. The results showed that increasing the silica fume dosage alone led to improvements in 28-day compressive strength and flexural strength by 3 MPa and 0.58 MPa, respectively. XRD and SEM analyses indicated that increasing silica fume content from 10% to 30% resulted in the formation of more C–S–H gel, enhancing strength, while the cumulative pore volume decreased from 0.09297 mL/g to 0.08209 mL/g, with an increase in harmless and less harmful pores. In contrast, increasing recycled micro-powder dosage from 10% to 20% raised the cumulative pore volume from 0.08947 mL/g to 0.10267 mL/g, and the proportion of more harmful pores increased to 28.6%. The RSM model achieved a fitting accuracy with R² > 0.95 and relative error ≤ 3.7%. The decision tree model demonstrated superior prediction accuracy, with both mean absolute error (MAE) and root mean square error (RMSE) below 3, and a coefficient of determination R² greater than 0.95. This model enables high-precision optimization of mortar mix proportions for silica fume–activated recycled micro-powder systems. • Silica fume chemically activates recycled powder, enhancing mortar hydration and interface. • RSM quantifies synergistic effects of silica fume, recycled powder, and water reducer. • ML model enables mix design optimization for silica fume–activated recycled powder mortar.