About 2 billion tons of metal are produced worldwide each year, and a significant share of this volume still comes from primary raw materials. Such production accounts for roughly 40% of industrial greenhouse gas emissions and consumes 10% of global energy. Greater use of recycled metal could reduce the burden on the environment and lessen dependence on imports of scarce strategic raw materials.
However, recycling faces an obstacle: the chemical composition of scrap metal varies widely, and unwanted impurities degrade the mechanical properties of alloys. To obtain quality material, manufacturers usually dilute scrap with pure metal, which reduces the environmental benefit.
An international team of researchers has proposed a different approach. Instead of treating impurities as undesirable, the scientists decided to use them as alloying elements - additives that give an alloy the desired properties. The resulting "dirty alloys" better tolerate fluctuations in scrap composition and demonstrate improved mechanical characteristics at lower cost. The results were published in the journal Science Advances.
The method is based on machine learning, that is, algorithms that find patterns in data and make predictions. The new strategy links alloy composition and processing conditions to its future properties, relying almost not at all on microstructure. A tree search algorithm, by iterating through options, generates new compositions and processing parameters such as temperature and time. Neural networks predict for each candidate the tensile strength and elongation - indicators of how strong the material is and how far it can stretch without breaking. Then a second model, trained on a larger dataset, selects the best options, which are manufactured and tested. The experimental results are returned to the database, and the cycle repeats, gradually improving the recommendations.
As a model material, the scientists took malleable cast iron, which is notable for its high tolerance to impurities and is widely used in industry - its global production is 120 million tons per year. Of the 17 compositions proposed by the algorithm, 14 showed the best combinations of strength and ductility among all known samples of recycled cast iron. Compared with the reference, tensile strength rose to 15.2% at comparable elongation, or elongation at break increased to 73.2% at the same strength. In addition, production costs were 8.1% lower than with the traditional mixing of scrap with pure iron.
The authors emphasize: alloys made entirely from scrap can outperform primary-production materials and cause less harm to the environment. Going forward, the approach can be extended to recycled steels, aluminum and copper alloys. The researchers see the next step as combining several AI agents with automated experimental platforms to accelerate development and adaptation to changing scrap composition. It is important to understand that this concerns laboratory tests and modeling, not a ready industrial technology.