AI-Powered Concrete: MIT Study Uses Machine Learning to Revolutionize Construction Recipes
Researchers at MIT have successfully leveraged artificial intelligence to optimize concrete production, potentially leading to significant cost savings and reduced environmental impact. A collaborative effort between the Olivetti Group and the MIT Concrete Sustainability Hub (CSHub) has resulted in a machine-learning framework capable of identifying alternative materials to replace cement in concrete mixes.
The core issue addressed by the study was the over-reliance on cement, a key ingredient in concrete, the production of which is both expensive and carbon-intensive. While materials like fly ash and slag are already used as cement replacements, their supply is limited. The MIT team sought to broaden the range of suitable replacement materials using AI.
Soroush Mahjoubi, a postdoc leading the research, explains, “We realized that AI was the key to moving forward. There is so much data out there on potential materials — hundreds of thousands of pages of scientific literature. Sorting through them would have taken many lifetimes of work, by which time more materials would have been discovered!” Their findings were published in Nature’s Communications Materials.
The team developed a machine-learning framework using large language models to evaluate materials based on hydraulic reactivity and pozzolanicity. Hydraulic reactivity refers to the material’s ability to harden when exposed to water, similar to cement. Pozzolanicity describes the material’s reaction with calcium hydroxide, a cement byproduct, enhancing concrete strength over time.
By analyzing over one million rock samples and vast amounts of scientific literature, the AI framework categorized candidate materials into 19 types, including biomass, mining byproducts, and demolished construction materials. The study revealed that numerous globally available materials could be directly incorporated into concrete mixes simply by grinding them, reducing the need for extensive additional processing.
One particularly promising category of cement replacement materials is ceramics. “Some of the most interesting materials that could replace a portion of cement are ceramics,” Mahjoubi notes. “Old tiles, bricks, pottery — all these materials may have high reactivity. That’s something we’ve observed in ancient Roman concrete, where ceramics were added to help waterproof structures.”
The research highlights the potential for concrete to contribute to a circular economy by repurposing materials that would otherwise end up in landfills. By identifying and utilizing these materials, the construction industry can significantly reduce its environmental footprint.
Looking ahead, the research team plans to enhance the AI framework and experimentally validate the most promising candidate materials. Professor Elsa Olivetti, senior author of the study, anticipates further advancements through the latest developments in large language models. Randolph Kirchain, co-author and CSHub director, emphasizes the importance of applying data science and AI to material design to support sustainable building practices without compromising structural integrity.
Other co-authors include MIT postdoc Vineeth Venugopal, Ipek Bensu Manav, and CSHub Deputy Director Hessam AzariJafari.
The research was supported by the MIT Concrete Sustainability Hub and the MIT-IBM Watson AI Lab.
