A multidimensional continuous conditional generative adversarial network to link composition-process-microstructure-properties in Mg alloys

This investigation presents a novel multidimensional continuous conditional generative adversarial network that is able to create alloy microstructures directly from composition-process variables in a single forward pass. The model is applied to discover ternary Mg-Al-Sn alloys manufactured by casting with high ultimate tensile strength and low corrosion rate. From the target performance indices, the model provides the candidate alloy compositions together with their corresponding process parameters, and the resulting microstructures. Experimental validation reveals deviations of only 1.4 % for the strength and 7.5 % for the corrosion rate with respect to the predictions while the errors in the fraction of each phase are below 0.4 %. By coupling continuous microstructure generation with multi-objective optimization, this work provides a scalable artificial intelligence-based inverse-design paradigm for magnesium and broader engineering alloys.