From scrap to prime: Designing high-performance alloys from dirty feedstocks with machine learning.

Producing alloys from recycled metals via secondary synthesis can markedly cut energy use, greenhouse-gas emissions, and dependence on scarce, strategically important elements relative to primary routes. The central obstacle is the feedstock itself: Scrap is chemically heterogeneous and unpredictable, a "dirty" mix whose impurities typically embrittle materials. Conventional strategies address this challenge by diluting impurities through the addition of virgin material, thereby restricting scrap utilization, limiting microstructural design freedom, and undermining cost and sustainability benefits. Here, we introduce a fundamentally different, data-driven paradigm that deliberately reverses a five-millennia-old metallurgical injunction: Rather than purifying the alloy, we design the alloy to accommodate and exploit its contaminants. Using a model ductile iron representative of 120 million metric tons of heavily contaminated scrap, we couple experiments with deep-neural-network optimization to identify chemically robust, impurity-tolerant processing windows. The resulting "dirty" irons retain ductility and strength by integrating impurity effects into a purposeful microstructure, rather removing them. The approach achieves complete recyclability, an 8.1% reduction in manufacturing cost, and mechanical gains of up to 15.2% in ultimate tensile strength and 73.2% in elongation at fracture. Crucially, it overcomes the canonical strength-ductility trade-off while enabling far greater scrap utilization and material circularity. By endowing alloys with a practical "gene of scrap compatibility," this strategy reframes alloy design for a resource-constrained future and advances a mass-conserving circular economy for metals.