Multi-Objective Bayesian Optimization of Continuous Purifications with Automated Phase Separation for On-Demand Manufacture of DEHiBA

The optimization of purifications has received little attention in an era of machine-learning driven optimization technologies that focus on synthesis, despite purifications being equally challenging and critical. This work utilizes lab-scale continuous purification equipment to automate the mixing and separation of phases for the purification of N,N-Di-2-ethylhexylisobutyramide (DEHiBA), a specialized ligand in demand for advanced nuclear reprocessing. Bayesian optimization drove the purifications via feedback from HPLC and GC-FID quantitative analysis to maximize purity and product recovery via a weighted single objective. Batch purification screening found removal of N,N-Di-2-ethylhexylamine (DiEHA) to be problematic with aqueous only extractions, adding complexity to the purification. Three purification routes were optimized in continuous flow and compared for their efficacy after a single extraction stage. Optimization of both product purity and recovery process metrics was crucial to identify optimum Pareto conditions. Product purities >95% were attainable for all routes, but the target of >99.9% was eluded after a single extraction in continuous flow. Product loss to the aqueous phase could be limited to <5%, but at the expense of product purity for all routes. Ultimately, a two-step process was devised from this work, employing a combination of water or 0.2 M nitric acid and acetonitrile to remove DiEHA and ∼90% isobutyric acid, subsequent sodium bicarbonate extraction yielded >99.9% purity.