Efficient Process Optimization of Hybrid Cooling/Antisolvent Crystallization for 2-tert-Amyl Anthraquinone via Bayesian Optimization

Purification of 2-tert-Amyl Anthraquinone (2-AAQ), a key fine chemical in hydrogen peroxide production, is essential due to the various byproducts generated in the synthesis process. Hybrid cooling/antisolvent crystallization offers an effective purification route for 2-AAQ, but its design and optimization are challenged by numerous input variables. In this work, a constrained Bayesian optimization (cBO) framework was applied to this crystallization process to maximize product yield while maintaining purity above 98%. Within this application, two constrained acquisition functions, constrained Expected Improvement (cEI) and constrained Probability Improvement (cPI), were systematically compared under the same experimental budget. The results showed that cEI outperformed cPI, achieving a yield of 87.65% versus 82.69%. To further improve optimization efficiency under limited experiments, a practical Hybrid Decision Making (HDM) strategy was implemented by switching from cEI in the early stage to cPI in the late stage. This strategy delivered the highest yield of 88.48% while satisfying the purity constraint. These results demonstrate that cBO is an effective tool for constrained optimization of hybrid crystallization processes and provide a practical workflow for process development in fine chemical manufacturing.