Development of an Automated Laboratory for High-Throughput Characterization of Structural Materials for Extreme Environments

Machine learning and AI-driven approaches to materials exploration and discovery are being driven by advances in automated laboratories, which can generate statistically-significant quantities of materials data. In many cases, automated materials research laboratories emphasize functional properties of materials, which are relatively insensitive to microstructure. But mechanical properties and behaviors of bulk structural materials, particularly under extreme loading conditions such as shock, are inherently linked to microstructure and microstructural length scales. This necessitates the development of automated laboratories[DE1] with unique capabilities.

The AI for Materials Design Laboratory (AIMD-L) at Johns Hopkins University is an automated high-throughput facility for characterizing microstructure and mechanical behavior of structural metals and ceramics intended for application in extreme environments, including high pressure, high strain rate, and high temperature. AIMD-L currently comprises three main characterization capabilities: (i) A custom x-ray instrument for measuring microstructure and composition simultaneously using transmission high-energy x-ray diffraction and x-ray fluorescence; (ii) A laser-driven microflyer impact system for assessing behavior under shock loading, including spall strength, equation of state, and Hugoniot elastic limit; and (iii) Nanoindentation for measuring quasi-static mechanical properties including hardness, stiffness, and strain-rate sensitivity. All three instruments are fully automated and capable of making hundreds to thousands of individual measurements per day.

The instruments in AIMD-L are linked by an automated robotic sample transfer system under centralized control, currently human-directed but ultimately to be fully autonomous. To achieve autonomy, AIMD-L will seamlessly integrate data from three sources: The instruments, AI/ML decision-making models, and the automated control system. Integrating these three elements requires a whole-system approach enabled by automated data flow and cross-task data contextualization created by a unifying semantic model.