For several weeks, the whiteboard in the laboratory was filled with annotations, sketches, and chemical equations. A research group from the Olivetti Group and the MIT Concrete Sustainability Hub (CSHub) was diligently focused on a crucial issue: How can we minimize the quantity of cement in concrete to reduce costs and emissions?
This inquiry was certainly not unprecedented; substances like fly ash, a byproduct of coal production, and slag, a byproduct of steel manufacturing, have been utilized for some time to substitute part of the cement in concrete mixtures. However, the demand for these materials is surpassing supply as industries strive to lessen their environmental impact by broadening their utilization, thus intensifying the urgency to find alternatives. The challenge that the team encountered was not a dearth of options; rather, there were too many to sift through.
On May 17, the group, spearheaded by postdoc Soroush Mahjoubi, released an open-access paper in Nature’s Communications Materials detailing their resolution. “We recognized that AI was pivotal for progress,” states Mahjoubi. “There is a vast amount of data regarding potential materials — hundreds of thousands of pages of research articles. Navigating through them would have consumed countless lifetimes of work, by which time more materials would likely have been identified!”
Using large language models similar to the chatbots many of us utilize daily, the team developed a machine-learning framework that assesses and categorizes candidate materials based on their physical and chemical characteristics.
“Primarily, there is hydraulic reactivity. The reason concrete possesses strength is that cement — the ‘adhesive’ that binds it — hardens upon contact with water. Therefore, if we replace this adhesive, we must ensure the substitute reacts in a similar fashion,” clarifies Mahjoubi. “Secondly, there is pozzolanicity. This occurs when a substance reacts with calcium hydroxide, a byproduct produced when cement interacts with water, to increase the concrete’s hardness and strength over time. We need to achieve a balance between the hydraulic and pozzolanic materials in the mixture so the concrete can perform optimally.”
By examining scientific literature and over 1 million rock samples, the team utilized the framework to categorize candidate materials into 19 different types, including biomass, mining byproducts, and recycled construction materials. Mahjoubi and his colleagues discovered that suitable substances were accessible worldwide — and notably, many could be integrated into concrete mixtures simply by grinding them. This indicates that it’s feasible to achieve emission reductions and cost efficiency with minimal additional processing.
“Some of the most intriguing materials that could replace a fraction of cement are ceramics,” remarks Mahjoubi. “Old tiles, bricks, pottery — all these materials may exhibit high reactivity. That’s something we’ve noticed in ancient Roman concrete, where ceramics were included to enhance waterproofing. I’ve had many engaging discussions with Professor Admir Masic, who spearheads numerous studies on ancient concrete here at MIT.”
The prospects of commonplace materials like ceramics and industrial materials such as mine tailings highlight how substances like concrete can facilitate a circular economy. By recognizing and repurposing materials that would typically be discarded in landfills, researchers and industry can contribute to giving these materials a renewed purpose within our buildings and infrastructure.
Looking forward, the research team aims to enhance the framework to evaluate even more materials while experimentally confirming some of the top candidates. “AI tools have propelled this research forward rapidly, and we are eager to see how the latest advancements in large language models facilitate the upcoming phases,” states Professor Elsa Olivetti, the senior author on the project and a member of the MIT Department of Materials Science and Engineering. She serves as a mission director for the MIT Climate Project, a principal investigator for CSHub, and the head of the Olivetti Group.
“Concrete is the foundation of the built environment,” asserts Randolph Kirchain, co-author and CSHub director. “By incorporating data science and AI tools into material design, we aspire to bolster industry initiatives aimed at constructing more sustainably, without sacrificing strength, safety, or durability.”
Alongside Mahjoubi, Olivetti, and Kirchain, co-authors on the project include MIT postdoc Vineeth Venugopal, Ipek Bensu Manav SM ’21, PhD ’24; and CSHub Deputy Director Hessam AzariJafari.