The cement sector contributes about 8% of the world’s CO₂ emissions, surpassing the entire aviation industry. Scientists at the Paul Scherrer Institute (PSI) have created an AI tool to fast-track the development of new cement mixes that can achieve similar quality while reducing carbon emissions.

The high-temperature rotary kilns used in cement plants operate at extreme temperatures of 1,400°C to transform ground limestone into clinker, which is the main ingredient of cement. Reaching such heat primarily involves energy-consuming combustion, resulting in high CO₂ emissions.

Interestingly, less than half of the emissions stem from this combustion; the bulk is released from the raw materials. Specifically, CO₂ that is chemically bound in limestone gets released when it turns into clinker in these hot kilns.

To tackle emissions, researchers are exploring ways to adjust cement recipes, using alternative materials to replace some of the clinker. An interdisciplinary team at PSI’s Laboratory for Waste Management is moving beyond lengthy traditional experiments by employing machine learning to optimize the process.

“Our approach enables us to simulate and refine cement recipes to cut down on CO₂ emissions without compromising mechanical performance,” says Romana Boiger, the study’s lead mathematician. “Instead of testing many variations manually, our model can propose options in seconds, much like a digital recipe book for eco-friendly cement.”

This new technique allows the researchers to rapidly narrow down effective concrete mixtures based on defined criteria. “The possibilities regarding material combinations are vast,” mentions Nikolaos Prasianakis, the head of PSI’s Transport Mechanisms Research Group and a co-author of the research.

“This method greatly speeds up the development process by zeroing in on the most promising candidates for further lab testing.” The study’s results were published in the journal Materials and Structures.

High Demand for Cement

Cement is essential for constructing our modern infrastructure. When mixed with sand, gravel, and water, it forms concrete, which is versatile and robust. The volume of cement used is staggering—approximately 1.5 kilograms per person daily, outpacing global food consumption, according to John Provis, a cement expert at PSI. “If we enhanced emissions by a few percent, we’d reduce CO₂ levels equivalent to thousands of cars,” he adds.

The Right Mix

Currently, by-products like steel slag and fly ash are already used to partially replace clinker, reducing CO₂ emissions. However, the immense global cement demand means these materials alone won’t suffice. “Finding the right mix of plentiful materials to ensure high-quality cement is crucial,” Provis notes.

Finding those combinations isn’t straightforward: “Cement acts as a mineral binder, essentially allowing us to create minerals that hold everything together, similar to a fast-tracked geology project,” Provis explains.

This intricate geology, with its myriad physical processes, makes computer modeling quite resource-intensive. Hence, the researchers have turned to artificial intelligence.

AI as a Fast-Track Tool

Artificial neural networks are computer systems that learn from existing data to help speed up complex calculations. By adjusting internal connections based on known data, these networks can swiftly predict similar outcomes, providing a more efficient alternative to traditional intensive modeling methods.

At PSI, researchers harnessed this technology, generating the training data through a thermodynamic modeling tool called GEMS. They studied which minerals form during the hardening process of various cement types and the geochemical reactions involved, says Prasianakis.

By merging these findings with experimental data, they developed a dependable indicator for assessing the mechanical properties and quality of the cement. They also incorporated CO₂ emission factors for each material, allowing them to assess total emissions linkage. “This complex modeling was challenging but essential,” he adds.

Yet, the effort was fruitful—thanks to the generated data, the AI system now computes mechanical properties for any cement mix in milliseconds, about a thousand times faster than conventional methods, Boiger explains.

Reversing the Workflow

How can this AI be utilized to identify ideal cement mixtures with minimal CO₂ emissions and top-notch quality? Instead of testing formulations independently, researchers approach the problem inversely by identifying compositions that inherently meet their emission and quality targets.

Both mechanical properties and CO₂ outputs are directly tied to the composition of the mixture. “Mathematically speaking, changing the recipe will alter both variables,” explains Boiger.

To find the optimal formulation, they set it up as a mathematical optimization challenge, searching for combinations that maximize strength while minimizing emissions. “We’re looking to balance both goals to find the best formulation,” she elaborates.

The team also applied genetic algorithms, inspired by natural selection, in their workflow. This approach allowed them to pinpoint combinations that meet their dual objectives.

By reversing the exploration method, they no longer need to trial endless recipes but can directly seek those meeting desired specifications—maximizing performance while cutting CO₂ emissions.

Interdisciplinary Potential

Among the promising recipes uncovered, several stand out as viable options. “These mixes not only show potential for reduced emissions but also practical manufacturing applicability,” Provis remarks.

However, further testing in the lab is necessary. “We’ll make sure to verify these before scaling up,” Prasianakis chuckles.

This study serves as proof-of-concept, demonstrating that viable formulations can emerge purely through mathematical analysis. “We can expand our AI tools as needed, including factors like raw material availability or usage in different environments,” says Boiger.

Looking ahead, Prasianakis is optimistic: “This is just the beginning. The time efficiency this workflow provides is immense, making it a very exciting approach for developing various materials and systems.”

The project’s success relied heavily on the diverse expertise of the researchers involved. “We needed specialists in cement chemistry, thermodynamics, AI—bringing everyone together was key,” Prasianakis notes. The collaboration with other research institutions, such as EMPA, through the SCENE project was also invaluable.

SCENE (the Swiss Centre of Excellence on Net Zero Emissions) is a multidisciplinary research initiative focused on scientifically viable solutions for drastically cutting greenhouse gas emissions in the industrial and energy sectors. This study was conducted as part of that larger project.