Data Infrastructure for Automated Labs and Framework for Interpretation of Characterization Results

The rapid advancements in autonomous laboratories, combined with sophisticated data-driven methodologies, present new opportunities as well as new challenges for materials innovation. Central to this transformation is the need for robust data storage, management, and visualization systems to meet the complex demands of these environments. In this talk, we present an approach to data infrastructure that stores workflow metadata, characterization outputs, and process log data using a mix of NoSQL databases, local file storage, and ongoing cloud integration. We also introduce an interactive visualization interface that supports multiple interpretations of raw characterization data (SEM-EDS and XRD) and highlighting correlations between synthesis processes and outcomes. This makes presenting data more accessible and improves the planning of future experiments.

Our goal with this interface is to go beyond simply tracking the experimental process. By comparing all recorded experiments, we create a similarity score based on target materials, precursors, heating profiles and characterization outcomes. This serves as a filter for users to easily find and compare similar syntheses. We aim to allow users to also create custom similarity filters to serve their unique interests. We demonstrate the power of this interface in an autonomous laboratory for solid-state synthesis (A-Lab).

Additionally, the large number of characterization results produced by automated labs presents challenges for traditional analysis methods that depend on human interpretation and often on prior chemical knowledge of most likely outcomes. We present our work in developing a framework to incorporate researcher intuition into automated analysis methods, including incorporating information from multiple measurement techniques to achieve a unified hypothesis. The ultimate goal is for the framework to produce interpretations consistent with researcher’s analysis, but such that it can be effectively scaled to analyzing hundreds or thousands of results. As a demonstration of the framework, we present a case study of identifying phases from powder XRD patterns with supplemental data from SEM-EDS.