High dimensional data driven statistical mechanics.

Area of Science:

• Materials Science
• Computational Materials Science
• Data Science

Background:

• 3D microstructure images contain both metric and topological features crucial for accurate property prediction.
• Traditional methods for microstructure analysis and property prediction are often limited by data complexity and accessibility.

Purpose of the Study:

• To highlight the core categories of 3D4D materials science: image acquisition, processing, analysis, modeling, and data sharing.
• To emphasize the conversion of 3D microstructure data into numerical features (materials genome) for advanced property prediction.
• To showcase advancements in automated 3D image acquisition and data archiving for materials design.

Main Methods:

• Utilizing 3D image data-based modeling (e.g., finite element methods) and machine learning approaches (e.g., artificial neural networks) for property prediction.
• Extracting numerical microstructural features such as grain size, particle connectivity, and stacking degree as "materials genome" descriptors.
• Developing and employing automated serial sectioning 3D optical microscopy (Genus_3D) for efficient image acquisition.

Main Results:

• The development of the "Genus_3D" microscope significantly accelerates 3D image acquisition, reducing it from months to hours.
• Identification of key microstructural features ("descriptors") that dominate material properties.
• Establishment of the "Materials Genome Archive" for comprehensive data sharing and facilitating high-throughput materials design.

Conclusions:

• 3D4D materials science offers a robust framework for advanced materials design by integrating diverse data types and analytical techniques.
• Automated imaging and data sharing are critical enablers for accelerating materials discovery and innovation.
• The "materials genome" concept, combined with machine learning, holds significant potential for predicting material properties with high accuracy.