Cement is crucial for modern infrastructure, but its manufacturing process releases much carbon, contributing heavily to environmental concerns.
So, how do we make cement greener? One promising answer lies in changing the recipe.
An interdisciplinary team at Paul Scherrer Institute (PSI), led by mathematician Romana Boiger, has found a new way to do just that – not through endless lab experiments, but with the power of machine learning.
“It’s like having a digital cookbook for climate-friendly cement,” said Boiger, the first author.
“This allows us to simulate and optimise cement formulations so that they emit significantly less CO₂ while maintaining the same high level of mechanical performance,” Boiger added.
Using AI to find recipes
Cement serves as a mineral binding agent. When mixed with water and gravel to create concrete, it artificially forms new minerals that effectively bind all the components together, giving the material strength and cohesion.
Making cement involves heating ground limestone to 1,400 degrees Celsius to produce clinker, which releases significant carbon dioxide.
Surprisingly, most of these emissions don’t come from the fuel used for heating, but from the CO₂ chemically bound within the limestone itself, which is liberated during its transformation in the kilns.
Currently, by-products like iron slag and coal fly ash replace some clinkers. But global cement demand is so massive that we need more options.
“What we need is the right combination of materials that are available in large quantities and from which high-quality, reliable cement can be produced,” said John Provis, head of the Cement Systems Research Group at PSI and co-author of the study.
That’s where Artificial Intelligence steps in.
The PSI researchers developed an artificial neural network trained with vast data. The data was produced by combining PSI’s GEMS thermodynamic modeling software and experimental results.
“We calculated – for various cement formulations – which minerals form during hardening and which geochemical processes take place,” said Nikolaos Prasianakis.
The trained neural network dramatically speeds up calculations, determining a cement recipe’s mechanical properties in milliseconds—about 1,000 times faster than conventional modeling.
Long road ahead
But the innovation doesn’t stop there. Instead of trying countless variations, they flipped the script.
The team used another AI technique, genetic algorithms, inspired by natural selection, to ask: “Which cement composition meets the desired specifications regarding CO₂ balance and material quality?”
“Some of these formulations have real potential. Not only in terms of CO₂ reduction and quality, but also in terms of practical feasibility in production,” said John Provis.
Much must be done before these new cement recipes can be fully adopted, including extensive laboratory testing.
This study mainly functions as a proof of concept, demonstrating that promising cement formulations can be identified through purely mathematical calculations.
The new study is just the beginning. But this AI-powered “digital cookbook” for cement could rapidly advance the creation of sustainable building materials.
Reducing carbon emissions in cement production is the need of the hour. This research might take years to become a reality.
But new developments are already underway. For instance, in one of the recent developments, experts turned clay into a cement material. In another, a 3D printing technique was used to make strong concrete using graphene.