(564e) Bayesian Optimization of a Continuous Polymer Precipitation Process

Polymer purification is a crucial process in chemical engineering because impurities can greatly affect polymer properties. In the semiconductor industry, polymer precipitation is a common technique for polymer purification. Compared to batch polymer precipitation, continuous polymer precipitation promises easy scale-up of polymer purification and significant energy and cost savings. We have designed a continuous polymer precipitation system that utilizes static mixers, peristaltic pumps, and a phase separator. In this system, an antisolvent precipitation method is used. A secondary solvent known as antisolvent, or precipitant, is added to the solution, resulting in the reduction of the solubility of the solute in the original solvent and consequently generating a supersaturation driving force. After the polymer chains collapse, aggregate, and the precipitation reaction takes place, the liquid waste is continuously isolated from the solid polymer, and the solid polymer is completely dried through a phase separation unit. The purified polymer is then re-diluted to the same conditions as the original polymer solution.

The development of an effective continuous polymer precipitation process requires careful thought of process parameters. For antisolvent precipitation processes, these include the ratio of antisolvent to polymer solution, the flow rate of polymer solution, and the retention rating of the phase-separation component. This study continues the 2023 AIChE presentation (“An Efficient, Cost-Effective, Continuous Polymer Purification Method”), focusing on a self-optimization machine learning method, i.e., Bayesian optimization (BO), that expedites the optimization of experimental parameters’ design space. In this study, BO is used to predict the combination of process parameters that achieve desired outcome measures, including impurity reduction, reduction in polydispersity index (PDI), etc. BO designs a series of experiments to gradually locate the desired optimum. Results obtained demonstrate that the sequence of generated experiments allows pinpointing operating conditions for target outcome measures. The implementation of BO led to reduced experimental effort and increased efficiency in process development. The transition to continuous polymer precipitation combined with the Bayesian optimization technique advances manufacturing process development to the next level.