Inverse Design of High-Entropy Superalloys Using Machine Learning and Generative Artificial Intelligence

We present a structure-agnostic inverse-design workflow to discover high-entropy superalloys (HESAs) for high-temperature service. From a curated multi-source dataset of materials described by their composition and various associated physical quantities, with engineer physics-informed, composition-derived descriptors and train calibrated supervised predictive models (surrogates) for key targets (Larson–Miller creep parameter, high-temperature oxidation parabolic constant, melting point, density, Pugh ratio, formation enthalpy, valence electron per atom, distortion parameter, stoichiometric entropy, Ω-criterion for solid-solution formation). Predicted properties are mapped to normalized objectives and filtered via a uniform feasibility floor – each must reach a “minimum score” – and multi-objective non-domination – no other alloy outperforms all its scores; Ward-medoid compression yields a frontier visualized on Ashby-style charts. A conditional variational autoencoder expands exploration under the same admissibility rules – uniform floor and non-domination –, populating lightly sampled by the original HEA set, potentially high-performance regions. Across panels, the compressed Pareto set extends the property envelope of legacy Ni-based superalloys while preserving interpretability and reproducibility. The framework exposes elemental roles, shortlists candidates for targeted experiments, and generalizes to other materials classes; non-physical constraints (cost, supply security, environmental footprint) are addressed in an integrated analysis at the alloy level – combining the contribution of the different elements. Our main contribution is an end-to-end, reproducible workflow that couples data curation, descriptor engineering, surrogate modelling, multi-objective selection and a controlled conditional variational autoencoder to generate feasible, interpretable HESA candidates that extend the property envelope of legacy Ni-based superalloys. A second contribution is an alloy-level analysis of cost, supply security and environmental footprint that links computational design to industrial and societal constraints. Together, these elements provide ranked shortlists of candidates and a general blueprint for sustainable inverse design in other materials families.