Microstructure-Aware Bayesian Materials Design

• DOI: 10.48550/arxiv.2502.03727
• Publication Date: 2025-02-05
• Authors: Danial Khatamsaz; Vahid Attari; Raymundo Arróyave
• Keywords: Condensed Matter - Materials Science; Materials Science (cond-mat.mtrl-sci); FOS: Physical sciences

In this study, we propose a novel microstructure-sensitive Bayesian optimization (BO) framework designed to enhance the efficiency of materials discovery by explicitly incorporating microstructural information. Traditional design approaches often focus exclusively on direct chemistry-process-property relationships, overlooking critical role microstructures. To address limitation, our integrates descriptors as latent variables, enabling construction comprehensive process-structure-property mapping that improves both predictive accuracy and outcomes. By employing active subspace method for dimensionality reduction, identify most influential features, thereby reducing computational complexity while maintaining high in process. This approach also enhances probabilistic modeling capabilities Gaussian processes, accelerating convergence optimal material configurations with fewer iterations experimental observations. We demonstrate efficacy through synthetic real-world case studies, including Mg$_2$Sn$_x$Si$_{1-x}$ thermoelectric energy conversion. Our results underscore microstructures linking processing conditions properties, highlighting potential microstructure-aware paradigm revolutionize discovery. Furthermore, work suggests since microstructure awareness discovery, characterization stages should be integral automated -- eventually autonomous platforms development.