Bayesian Optimization for Multi-Step Organic Flow Battery Batch Synthesis

Traditional methods for materials discovery are not fast enough to respond to the rapid progress of climate change. Therefore, faster methods have become a rising interest. This has prompted the integration of robotics and machine learning in materials research, which has shown promising results. However, challenges remain in synchronizing the physical hardware and digital algorithms where current Python libraries often do not fit chemical workflow needs. In this work, we have developed custom batch Bayesian optimization for a high-throughput robotic platform to optimize redox molecules for flow batteries. We designed three models that adapt to the redox molecules' multi-step chemical process where formulation and heating are separate and differ in hardware availability. Our work showcases an approach for generating formulation and heating conditions using clustering and mixed variable batch Bayesian optimization. Leveraging this method for exploring and exploiting the space, we have selected optimal conditions that maximize yield.