A new AI-powered model is helping scientists design low-carbon concrete by accurately forecasting emissions from green mixes, cutting experimental time and boosting sustainability in construction.
Study: AI-based modeling of CO2 footprint in geopolymer concrete production containing GGBFS as a by-product from the iron industry. Image Credit: chayanuphol/Shutterstock.com
A recent study published in Scientific Reports presents an artificial intelligence (AI)-based approach for modeling the carbon dioxide (CO2) footprint of geopolymer concrete made with ground granulated blast-furnace slag (GGBFS), a by-product of the iron industry. This material is used as a more sustainable alternative to ordinary Portland cement (OPC), a major source of global carbon emissions.
Why GGBFS Matters
OPC production is responsible for roughly 7 % of global CO2 emissions. To reduce this impact, researchers have been exploring partial or full replacements of OPC with supplementary cementitious materials (SCMs) like GGBFS. These materials not only help lower emissions but also enhance concrete's strength and workability.
Geopolymer concrete is produced when SCMs rich in silicates and aluminates react with alkaline activators. GGBFS, in particular, offers multiple advantages: it reduces the heat of hydration (helping to prevent thermal cracking), improves workability, and enhances durability by resisting chloride penetration, sulfate attack, and alkali–silica reaction. These benefits make it an ideal candidate for use in marine infrastructure, precast products, and other demanding applications.
While GGBFS-based geopolymer concrete holds strong potential for reducing emissions, identifying optimal mix designs through conventional lab testing can be costly and time-intensive, especially with so many design variables in play. That’s where AI comes in.
The Study
In this study, the researchers developed a hybrid AI model combining an artificial neural network (ANN) with biogeography-based optimization (BBO) (called ANN-BBO) to forecast the carbon footprint of GGBFS-based geopolymer concrete.
The model considered 25 key variables, including mix design factors, aggregate and activator characteristics, chemical composition, and curing conditions. By training on a dataset of 122 records drawn from 19 previously validated studies, the model could predict the total CO2emissions based on material inputs and curing temperatures.
To prepare the dataset, all features were normalized using Z-score scaling to avoid bias and ensure consistency. The data was then split into training, validation, and test sets to promote accuracy and generalizability. The Levenberg–Marquardt algorithm was chosen for training due to its proven performance, and the hyperbolic tangent function was used as the model’s activation function.
Two models were built for comparison: a standalone ANN and the hybrid ANN-BBO. The goal was to evaluate how much the BBO algorithm improved predictive performance.
Key Findings
The hybrid ANN-BBO model significantly outperformed the standalone ANN in forecasting emissions:
- The ANN-BBO achieved an objective function (OBJ) value of 2.393 (closer to the ideal value of zero) compared to 6.156 for the ANN.
- 89 % of the ANN-BBO’s predictions fell within a 5 % error margin, compared to 62 % for the ANN, showing a 27 % improvement in accuracy.
- The hybrid model also delivered better generalizability and reliability.
Among the input variables, coarse aggregate had the most significant negative impact on emissions, while superplasticizer had the most positive. The optimal GGBFS content was identified as 400 kg/m3 for achieving the highest compressive strength with a lower carbon footprint.
Final Thoughts
This study shows that AI-based modeling, particularly hybrid approaches like ANN-BBO, can be an effective tool in designing sustainable concrete mixes.
By minimizing the need for repeated lab testing and streamlining the mix design process, AI frameworks like this one could accelerate the shift toward low-carbon construction materials.
Journal Reference
Kazemi, R., Bashtani, A., Mirjalili, S. (2025). AI-based modeling of CO2 footprint in geopolymer concrete production containing GGBFS as a by-product from the iron industry. Scientific Reports, 15, 43436. DOI: 10.1038/s41598-025-27361-7, https://www.nature.com/articles/s41598-025-27361-7
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.