Self-driving laboratory for emulsion polymerization

• Bespoke multi-geometry flow reactor for aqueous free radical emulsion polymerization. • Online dynamic light scattering facilitates programmed design-of-experiment (DoE) or closed-loop optimization campaigns. • Tailored application of machine learning algorithm can target or map feasible particle sizes. • Algorithms tested in-silico using a response surface generated from experimental DoE data. Modern approaches to chemical product discovery are exploiting the benefits of flow-chemistry, online characterization, and smart automation to rapidly screen and optimize chemical transformations. The present work describes the development and application of an automated continuous-flow reactor platform for the rapid prototyping of latexes prepared via seeded free-radical emulsion polymerization. Using a multi-reactor system comprising a cascade of miniature continuous stirred-tank reactors (CSTRs) followed by a sonicated tubular reactor (STR) with five pumps for reagent delivery, the capability to explore a four-dimensional parameter space of surfactant concentration, seed fraction, monomer ratio, and feed-rate is demonstrated. With user-defined boundary conditions, a one-factor-at-a-time (OFAAT) approach first illustrates the capability to prepare products with unique and tuneable properties. Subsequently, an experimental design is constructed to explore a three-dimensional parameter space, with 16 reactions completed in under three days of platform time. This rapid generation of product prototypes allowed features of the polymer system to be evaluated on a timescale much shorter than traditional methods with a significant reduction in manual effort and human-chemical interaction. The resulting response surface model was applied for in silico optimization using the Thompson-sampling efficient multi-objective (TSEMO) optimization algorithm. Finally, online dynamic light scattering (DLS) was applied with the physical platform which enabled self-optimization of the polymerization, identifying the attainable particle sizes whilst minimizing the amounts of seed and surfactant used. Closing the loop resulted in a fully operational self-driving laboratory.