Robotic Workflow for Autonomous Glass Discovery

Today's glass development must meet increasingly complex property profiles within an ever-shorter time-to-market. However, the disruptive progress required in glass development to overcome this challenge is hampered by manual workflows, empirical trial-and-error approaches, and a lack of process documentation. Moreover, computer-aided glass formulation tools such as databases or property prediction tools currently suffer from insufficient database consistency, variable data fidelity, and a significant lack of property and process data. To this end, a cross-laboratory infrastructure has been established that combines computer-aided glass formulation and robot-assisted glass synthesis. This infrastructure provides a steadily expandable web-based glass data space, machine learning-driven data mining and property modeling tools, as well as a robotic glass melting system equipped with inline sensors and semi-autonomous intervention tools for process optimization and system protection. An underlying backbone ontology defines the machine-readable glass terminology and semantic relations. Logging and codifying even subconscious process expertise and human intervention, this robot-assisted infrastructure guarantees exact reproducibility and builds up an ever-growing expert knowledge system over time. This article aims to illustrate the significant progress made with this approach while also highlighting ongoing needs on the way to autonomous glass development.