Meta for artificial intelligence researching new ways to mix concrete

Meta's artificial intelligence (AI) research team has found a new concrete mixing formula that is harder and more environmentally friendly.

Artificial intelligence systems have shown great potential in the field of new compound discovery, whether sifting through vast amounts of data to find compounds or developing new recipes using Use flavor profiles of ingredients.

Meta's AI research team has developed an artificial intelligence system to find a new, more environmentally friendly concrete mix recipe.

Recently, a team from Meta AI worked with researchers at the University of Illinois, Urbana-Champaign, and created an AI that can generate and refine formulations for low-carbon concrete with stiffness even taller than traditional concrete, according to Engadget.

The traditional methods of creating concrete, which we produce billions of tons each year, are not environmentally friendly. In fact, they produce about 8% of total global carbon dioxide emissions annually. Much progress has been made in recent years to reduce the carbon footprint of the concrete industry as well as make the material stronger, more resilient and even capable of EV charging. However, concrete production is still among the most carbon-intensive industries in modern construction.

To reduce the amount of carbon entering concrete, simple methods such as changing the main ingredients can be used. The material is made of four basic ingredients: cement, reinforcement, water and additives with cement being the most carbon-intensive ingredient. So, to reduce the amount of cement needed, the team will add lower-carbon materials such as ash, slag or ground glass.

Similarly, reinforcement materials such as gravel, crushed stone, sand can be replaced with recycled concrete. The point is that there are dozens of potential component materials that can be used, and how they interact with each other will affect the structure of the concrete. In short, there are many combinations for researchers to experiment with, choose from, and refine. With such a large number of tests, researchers from Meta trained an AI to do it at a much faster rate.

Working with Professor Lav Varshney, department of electrical and computer engineering, and Professor Nishant Garg, faculty of civil engineering, both University of Illinois at Urbana-Champaign, Meta's team trained the AI ​​model This is done using the Concrete Compressive Strength dataset. This dataset includes more than 1,000 concrete formulations as well as their structural properties. The team determined the carbon footprint of the resulting concrete mix using the Cement Sustainability Initiative's Environmental Product Claims (EPD) tool.

Among the potential formulations generated, the team selected the five most promising and refined them several times until they met the requirements. The refining process took only a few weeks and resulted in a satisfactory concrete formulation replacing up to 50% of the required cement with ash and slag. Meta then teamed up with concrete company Ozinga, which built Meta's newest data center in Illinois, to further refine the formula and conduct real-world testing.

Going forward, the Meta team hopes to further improve this formula and better understand how it withstands different weather conditions such as wind or high humidity.