AI stirs up concrete formula in MIT study | MIT News

For weeks, whiteboards in the lab were packed with graffiti, charts and chemical formulas. The Olivetti Group and the MIT Center for Concrete Sustainability (CSHUB) research team is working to solve a key question: How do we reduce the amount of concrete cement to save costs and emissions?

This question is certainly not new. Materials such as by-products of coal production and slag (ores) have long been used to replace certain cements in concrete mixtures. However, demand for these products exceeds supply, as the industry is eager to find alternatives by expanding its use to reduce its climate impact. The challenge the team found was not the lack of candidates. The problem is that there is too much to sort out.

On May 17, a team led by postdoctoral Soroush Mahjoubi published an open paper on nature Communication materials Overview of their solutions. “We realized that AI is the key to moving forward,” Mahjoubi noted. “There is a lot of data on potential materials out there – hundreds of thousands of pages of scientific literature. Organizing them will take up a lot of work and by then you’ll find more materials!”

With large language models, like chatbots, which many of us use every day, the team has built a machine learning framework that evaluates and classifies candidate materials based on their physical and chemical properties.

“First of all, there is hydraulic reactiveness. The reason why concrete is strong is that cement – the glue that holds it together – hardens when exposed to water. So if we replace this glue, we need to make sure that the alternative reacts similarly,” explains Mahjoubi. “Secondly, there is volcanic ash. This is a material that reacts with the byproducts produced when cement encounters water to make the concrete harder and harder. We need to balance the hydraulic and limestone materials in the mixture in the mixture so that the concrete performs well at its best.”

The team analyzed scientific literature and over 1 million rock samples, classifying the framework into 19 species of candidate materials, from biomass to mining by-products to demolished building materials. Mahjoubi and his team found that the right materials are available worldwide – and what is more impressive is that many materials can be incorporated into concrete mixtures by grinding alone. This means that emissions and cost savings can be extracted without much extra treatment.

“Some of the most interesting materials can replace part of the cement,” Mahjoubi noted. “Old tiles, bricks, pottery – all of these materials can be highly reactive. This is something we observed in ancient Roman concrete, where ceramics were added to help with waterproof structures. I had many interesting conversations at Professor MASIC, who led many ancient concrete studies at MIT.”

The potential of everyday materials such as ceramics and industrial materials, such as mine tailings, is an example of how materials such as concrete can help achieve a circular economy. By identifying and reusing these materials that would otherwise end up in landfills, researchers and the industry can help make these materials part of our buildings and infrastructure.

Going forward, the research team plans to upgrade the framework so that more materials can be evaluated while experimentally validating some of the best candidates. “AI tools have obtained this study in a very short time and we are excited to see how the latest developments in the big language model can achieve the next step,” said Professor Elsa Olivetti, a worker and member of the Department of Materials Science and Engineering at MIT. She serves as mission director for the MIT climate program, principal investigator at CSHUB and head of the Olivetti Group.

“Concrete is the backbone of the built environment,” said co-author and CSHUB director Randolph Kirchain. “By applying data science and AI tools to material design, we want to support the industry in a more sustainable effort without damaging strength, security or durability.

In addition to Mahjoubi, Olivetti and Kirchain, co-authors include MIT Postdoc Vineeth Venugopal, Ipek Bensu Manav SM ’21, PhD ’24; and CSHUB Deputy Director Hessam Azarijafari.