Artificial intelligence develops high-performance alloys from scrap metal
Recycling metals into usable alloys presents considerable challenges. An international research team has developed a strategy based on artificial intelligence (AI) and machine learning to specifically utilise impurities in scrap metal as alloying elements.
Key points at a glance:
- Challenge: Conventional metal production consumes 10 per cent of the world’s energy and accounts for around 8 per cent of all CO₂ emissions. Recycling reduces these environmental impacts by up to 95 per cent by keeping metals in the recycling loop for as long as possible, thereby reducing the need to extract new material from ores. However, fluctuating scrap compositions and unwanted impurities can compromise the quality of recycled alloys and thus limit their use.
- A new approach: using artificial intelligence and machine learning, we are finding ways to specifically utilise unwanted impurities in scrap metal as alloying elements: the basis for new, higher-performance recycled alloys.
- Result: 14 new cast iron alloys made from scrap metal, offering improved tensile strength and elongation at break whilst reducing environmental impact and production costs.
The challenge of recycling
Around two billion tonnes of metal are produced worldwide each year to meet the needs of industry, infrastructure and the energy sector. A large proportion of this is still extracted from ores. Conventional metal production accounts for approximately 40 per cent of industrial greenhouse gas emissions and is responsible for around 10 per cent of global energy consumption. Increased use of recycled metals could significantly reduce the environmental impact of primary production by keeping metals in the recycling loop for longer, whilst simultaneously reducing dependence on imports of scarce and strategically important ores.
However, recycling metals into usable alloys presents significant challenges: the chemical composition of scrap metal often varies and contains undesirable impurities. These can compromise the quality of recycled alloys, rendering them unsuitable for demanding industrial applications.
An international research team has now developed a strategy based on artificial intelligence (AI) and machine learning to specifically utilise impurities in scrap metal as alloying elements. The resulting so-called ‘dirty alloys’ not only tolerate variations in scrap composition but also exhibit improved mechanical properties and lower production costs. The findings were published in the journal *Science Advances*.
The study was carried out as part of the EU ‘CircEUlar’ project and was published in the journal *Nature Communications*. The research team analysed the stocks of 22 different materials contained within today’s fossil-based infrastructure. The aim was to better assess the potential of this so-called ‘urban mine’.
Empa researcher Hauke Schlesier, the study’s lead author, describes the starting point of the investigation as follows: “To understand the potential of this ‘urban mine’, we first need to know which materials are present in it.”
Artificial intelligence links alloy composition directly to material properties
“Until now, impurities in scrap metal have usually been diluted by adding primary metal. However, this undermines the environmental benefits of recycling, as the production of primary metals requires large amounts of energy and causes significant CO₂ emissions. Our new workflow uses AI and machine learning to identify suitable impurities and use them to develop alloys that not only tolerate impurities but actually benefit from them,” explains Dr Liuliu Han, head of a partner group at the Max Planck Institute for Sustainable Materials and one of the corresponding authors of the publication.
What makes the new workflow unique is that it directly links an alloy’s chemical composition and manufacturing conditions to the desired material properties, without requiring extensive information about the alloy’s microstructure, but instead making targeted use of the range of elements available from scrap to improve the product. Starting from already known alloys, a search algorithm generates new alloy candidates. To do this, it varies both the chemical composition and process parameters – such as the temperature and duration of heat treatment – within specified limits.
Artificial neural networks then predict the tensile strength and elongation at break of the respective alloy candidates. This enables the algorithm to search specifically for compositions that offer a better balance between strength and formability than previously known materials. In doing so, it optimises particularly promising alloys whilst also investigating compositions that have been little studied to date. A second model, trained using a larger dataset on alloy compositions and material properties, evaluates the pre-selected candidates. The scientists then produce the most promising alloys and test them.
The results of these experiments are then fed back into the database. This updates the predictive model and guides the next round of optimisation. Through this repeated comparison between computer-aided predictions and experimental results, the selection of suitable alloys is gradually improved, accelerating the development of new materials from recycled metals.
Impurities improve mechanical properties
The researchers used ductile iron, also known as spheroidal graphite cast iron, as their model material. This material is known for its high tolerance to impurities and is of great industrial importance: around 120 million tonnes of it are produced worldwide each year.
The research team investigated 17 alloy variants proposed by the AI model. Of these, 14 achieved particularly good combinations of strength and ductility compared with the recycled iron materials previously recorded in the database.
With comparable elongation at break, the tensile strength could be increased by up to 15.2 per cent. At comparable tensile strength, the elongation at break increased by as much as 73.2 per cent. At the same time, production costs were around 8 per cent lower than those of the conventional process, in which scrap metal is mixed with virgin iron.
The scientists also plan to extend their AI strategy to other material groups, such as recycled steels and aluminium and copper alloys. In addition, several AI agents are to take on different tasks and coordinate with one another – from the development of new alloy compositions and experimental design to process control and the evaluation of results. This would accelerate the development of new alloys whilst enabling flexible adaptation to fluctuating scrap compositions.
Researchers from Shanghai Jiao Tong University, the City University of Hong Kong, the East China University of Science and Technology and Central South University in China, as well as the Max Planck Institute for Sustainable Materials in Germany, were involved in the work.
Source: MPI-SusMat
Original publication
C. Yang, B. Peng, B. Sun, Q. Tao, Y. Zeng, L. Zhang, D. Shu, J. Lu, D. Raabe, L. Han, Y. Wie: From scrap to prime: designing high-performance alloys from dirty feedstock with machine learning. *Science Advances* 12 (2026), 42. DOI: 10.1126/sciadv.aeg9183