Data-driven and intelligent design of high-entropy alloys: A review of machine learning across composition, microstructure, and properties
Year: 2026
Type: article
Abstract: High-entropy alloys (HEAs), with their unique multi-principal element design philosophy, have opened up an exceptionally vast compositional space, breaking through the performance limitations of conventional alloys. As an emerging class of energy materials, HEAs have demonstrated tremendous potential in extreme environment service and structural engineering applications. However, the huge combinatorial space and the complex nonlinear coupling among composition, processing, microstructure, and property make traditional trial-and-error approaches inefficient for discovering high-performance alloys. As a data-driven paradigm, machine learning (ML) provides an effective solution to the "combinatorial explosion" problem in HEAs design by enabling efficient exploration of high-dimensional spaces and revealing hidden structure-property relationships. As HEAs design shifts from experience-driven exploration to data-driven discovery, this review systematically examines the challenges, emerging trends, and future directions of ML in composition screening, microstructure evolution prediction, and mechanical property optimization. For the complexity of multi-principal element systems, ML models based on physics-informed feature engineering and phase stability prediction have successfully identified feasible compositional regions. Combined with active learning and multi-objective optimization, ML has promoted a shift from single-point prediction to closed-loop discovery. At the microscale, surrogate models integrating multimodal characterization and processing parameters offer new ways to quantify processing-microstructure relationships and inversely design thermomechanical pathways. Moreover, by introducing physics-based constraints and uncertainty quantification, ML has significantly improved the interpretability and extrapolation of predictions for strength, ductility, and fracture behavior. With the deep integration of multiscale simulation and large language model technologies, the data-driven paradigm is evolving from an auxiliary tool into a core engine for precise HEAs design and engineering implementation.
Source: Metals Advances
Institution: Northwestern Polytechnical University
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: diamond
Awards: G4041317445, G4437552218, G6839843811
OpenAlex ID: W7171870023