Machine learning enabled design of sustainable alloys considering real world constraints

Abstract

Designing sustainable alloys must reconcile high performance with real-world constraints, including feedstock scarcity and variability, processing and energy requirements, regulatory thresholds, and economic limitations. Addressing these constraints reaches far beyond today's conventional performance-focused frameworks. It needs to consider not only complex property-processing-performance trade-offs but also sustainable production pathways. Here, we propose a physics-informed ensemble (PIE) learning framework that integrates fundamental structure-property relationships with processing routes and sustainability criteria for mechanism-based alloy design. It predicts properties from composition-microstructure relationships and infers processing routes under sustainability constraints, establishing an interdisciplinary approach to identifying sustainable alloy solutions. We validate the approach for Mg alloys, a lightweight yet energy-intensive metallic material class, considering performance, cost, and carbon emissions. Trained on a benchmark dataset, the PIE model systematically explores plausible alloy compositions and microstructures to construct a digital library of millions of candidate alloys. From this space, we identify two sustainable Mg alloy solutions that overcome the conventional performance-cost-emission trade-off: a rare-earth-containing alloy increases yield and tensile strength by 72.8 percent and 61.1 percent, respectively, over alloys from the same family at comparable cost and emissions; and a rare-earth-free alloy reduces cost and carbon emissions by 40.4 percent and 88.9 percent, respectively, relative to its primary-feedstock counterpart, without compromising mechanical performance. This work introduces a general, physics-informed, and quantifiable framework for designing sustainable alloys under real-world constraints.