What question did this study set out to answer?

This evaluation aims to identify an optimal sustainable concrete mix using silica fume and manufactured sand to improve mechanical properties.

March 13, 2026Open Access

Integration of machine learning and microstructural characterization for strength forecasting with silica fume and M-sand for sustainable concrete

Key Points

This evaluation aims to identify an optimal sustainable concrete mix using silica fume and manufactured sand to improve mechanical properties.
Evaluated concrete mixes with 10% fly ash and varying silica fume levels (0% to 24%)
Conducted microstructural analysis using scanning electron microscopy (SEM) and energy-dispersive spectroscopy (EDS)
Performed thermogravimetric analysis (TGA)
Applied various machine learning models, including Gradient Boosting, to predict compressive strength
Measured strengths at 7, 28, and 90 days
The optimal mix of 10% fly ash and 12% silica fume showed significant strength increases at various intervals
Microstructural analysis indicated a denser C-S-H gel structure and reduced porosity
Gradient Boosting model achieved an R^2 of 0.9929, ensuring accurate strength predictions

Abstract

Incorporating industrial by-products into concrete reduces the environmental impact ofcement production. This study evaluates sustainable ternary concrete mixes containing 10% fly ash, varying silica fume levels (0%, 6%, 12%, 18%, 24%), and 100% manufactured sand as fine aggregate to identify the optimal mix for enhanced mechanical and microstructural properties using scanning electron microscopy (SEM),energy-dispersive spectroscopy (EDS), thermogravimetric analysis (TGA), and machine learning (ML) assessment were done to streamline the experimental process. Compressive, split tensile, and flexural strengths, as well as ultrasonic pulse velocity, were measured at 7, 28, and 90 days. The mix with 10% fly ash, 12% silica fume, and 100% manufactured sand demonstrated the highest performance, with compressive strength increases of 18.61%, 16.85%, and 19.83% at each interval. Microstructural analysis revealed a dense C-S-H gel and uniform matrix, indicating improved hydrationand reduced porosity. Machine learning models (LASSO, Random Forest, Gradient Boosting, XGBoost, AdaBoost, and ANN) were applied to predict compressive strengthand to minimise the number of experimental trials. Gradient Boosting achieved the mostaccurate predictions, with an R2 of 0.9929 and minimal error, even with limited data. Both laboratory and machine-learning results confirm that concrete with 10% fly ash, 12% silica fume, and 100% manufactured sand provides a durable, high-performance solution for structural applications.

Bookmark

View Full Paper