PCS: Property-composition-structure chain in Mg-Nd alloys through integrating sigmoid fitting and conditional generative adversarial network modeling

Abstract

Here, we report a unified framework to establish the property-composition-structure (P-C-S) relationship, utilizing a property-oriented generative model that incorporates composition as a continuous conditional label within a conditional generative adversarial network (CGAN). This approach enables arbitrary microstructure prediction across a specified composition range. Using as-cast Mg-Nd alloys as a case study, the model synergistically integrates sigmoid fitting and CGAN modeling. Sigmoid fitting effectively bridges the relationship between alloy hardness and composition, while CGAN generates microstructures corresponding to specified compositions. The phase fraction of the alloys serves as a metric for assessing the similarity and accuracy of the generated microstructures. By elucidating the intricate interdependence between property, composition, and structure, this framework offers a scalable and systematic approach to property-driven material design. Beyond enhancing our understanding of the fundamental mechanisms governing material properties, it provides a powerful computational tool for tailoring material properties for specific applications.