In-Loop X-Ray Phase Fraction and Structure Analysis for High-Rate Closed-Loop Autonomous Experimentation

Recent advances in autonomous closed-loop experimentation have created new opportunities in high-throughput material discovery research. However, quantitative phase and phase fraction identification from x-ray scattering at speeds commensurate with synchrotron-based measurements remains challenging and limits full exploitation of autonomous searches. To address this challenge, we have developed a probabilistically quantitative, multi-phase labeling algorithm that provides in-loop full structural information for AI agents controlling high-throughput autonomous experiments; data include quantitative lattice strains, peak broadening and grain size effects, and probability estimates for likely multi-phase combinations. In conjunction with non-equilibrium laser spike annealing of compositional libraries to explore composition and processing spaces, we demonstrate the ability of the autonomous agents to utilize the additional data from x-ray scattering to implement a two-stage objective function, switching between <i>exploration</i> and <i>exploitation</i> modes based on uncertainty of the whole material processing space with structurally labeled phase fields. By providing in-loop data at rates commensurate with the laser spike annealing and synchrotron measurements, we allow active learning AI agents to explore the phase space efficiently and to identify optimal conditions to synthesize high phase-purity structures. With further design of targeted acquisition functions and incorporating additional property measurements, we expect to extend this method for on-the-fly modeling of structure-property relationships and enable practical multi-objective optimization.