CDS&E: A Deep Learning Framework for Material Parameters Estimation in Multi-principal Element Alloys

Nontechnical Description: Advanced materials enable many technologies that society depends on, including energy systems, transportation, electronics, and manufacturing. However, discovering new materials remains a slow and expensive process. Computational materials modeling has emerged as a powerful tool for accelerating materials discovery because it allows researchers to predict how materials will behave before they are manufactured. A major challenge is that these models require numerous material parameters that are often difficult, time-consuming, or sometimes impossible to determine experimentally, particularly for complex alloys containing multiple principal elements. The principal investigator (PI) aims to develop an artificial intelligence (AI) framework that can learn these material parameters directly from microstructure images that contain information on how elements are distributed within a material. Artificial intelligence combined with the fundamental physics governing microstructure evolution can make computational materials models more accurate, trustworthy, and easier to apply to new materials systems. The resulting tools can accelerate materials discovery by improving the connection between experiments, simulations, and data. Educational activities are intended to prepare students to join an AI-ready engineering workforce by incorporating AI-assisted computational problem-solving modules into engineering courses and by providing faculty workshops that support the integration of AI into STEM education. Codes, datasets, and educational resources developed through the project will be made publicly available to benefit the broader research and education communities. Technical Description: This award supports research to use artificial intelligence to accelerate and improve the accuracy of computational materials research and associated education activities to infuse artificial intelligence advances in the next generation workforce. The PI aims to develop a Physics-Informed Graph Transformer framework for parameter estimation in partial-differential-equation-constrained microstructure models. Material parameter estimation in multicomponent alloys presents a scientific challenge because it is an ill-posed inverse problem and conventional approaches are computationally expensive, difficult to scale, and often require extensive trial-and-error calibration. The framework combines graph-based representations of two-dimensional and three-dimensional microstructures, Transformer self-attention mechanisms that capture long-range spatiotemporal correlations, and physics-informed loss functions based on the governing equations of microstructure evolution. The PI aims to develop and validate the framework for forward prediction and inverse parameter estimation, targeting thermodynamic, kinetic, and interfacial parameters. The framework will include identifiability quantification using sensitivity and Sobol analyses, and uncertainty quantification through Monte Carlo sampling and neural-network-based ensemble approaches. Expected outcomes include new algorithms, open-source software, benchmark datasets, and a general framework for accelerating parameter estimation and model calibration in Integrated Computational Materials Engineering. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.