Nonlinear Optimization via Explicit NRTL Model Solubility Prediction for Antisolvent Mixture Selection in Artemisinin Crystallization

Abstract

Continuous Pharmaceutical Manufacturing (CPM) has the potential to revolutionize the pharmaceutical industry, with many expected benefits in terms of cost, efficiency, and quality. Process modeling and optimization are valuable methodologies for comparative technoeconomic evaluations: this paper pursues total upstream CPM cost minimization via two nonlinear optimization methods. An explicit NRTL solubility estimation method has been used in order to analyze the potential performance of three binary antisolvent mixtures for the crystallization of artemisinin. This Active Pharmaceutical Ingredient (API) is a key antimalarial substance and has been the focus of several CPM studies, including continuous chemistry and separation. Ethanol, acetone, and ethyl acetate are the three antisolvents studied. Used as a binary antisolvent with toluene as the solvent, the composition of the binary antisolvent, the quantity of the binary antisolvent with respect to the process solvent, and the temperature of the overall mixture is cooled to in the crystallization process are formulated as the three key variables in a nonlinear optimization problem. Two solvers are used, NOMAD and NLOPT-BOBYQA. Results show that they have very similar precision (<1% difference), but that the latter is up to 66% faster; NLOPT-BOBYQA was used for the majority of optimization cases. Results show that the size of the temperature gradient of the crystallization has a much stronger effect on the total cost than the quantity of antisolvent used. Nearly pure antisolvent use is favored for yielding lowest total costs for a single crystallizer (pure ethyl acetate is most favored, followed by ethanol, with acetone last). When considering multiple, identically-sized crystallizers in series, results indicate that higher API recovery, E-factor, and Operating Expenditure (OpEx) benefits can be achieved when using two crystallizers, but these metrics worsen with three or more; Capital Expenditure (CapEx) follows the same trend. For sequential crystallizers of varying size, increasing the number of crystallizers also benefits OpEx, E-factor, and API recovery, but CapEx continues to increase, thereby promising technical but no economic benefits.