Multi-objective optimization of oxygen transfer and hydrodynamic shear in dual-impeller stirred tanks via integrated computational fluid dynamics and machine learning framework.

Optimizing the structure of stirred bioreactors is crucial to improve the efficiency of biomanufacturing. This study proposes an intelligent design framework integrating computational fluid dynamics (CFD), machine learning (ML), and the non-dominated sorting genetic algorithm II (NSGA-II) to co-optimize the oxygenation and shear performance of a dual-impeller stirred tank. Initially, a parameterized CFD model was employed to systematically investigate the impacts of impeller inclination angles, dimensions, and blade numbers on the volumetric oxygen transfer coefficient (kLa) and shear strain rate. Subsequently, six ML algorithms were evaluated using a dataset of 837 CFD-generated samples, where the artificial neural network (ANN) demonstrated superior predictive accuracy (R2 > 0.97) and was selected as the surrogate model. Utilizing the NSGA-II algorithm coupled with the ideal point decision-making method, an optimal asymmetric impeller configuration was identified from the Pareto-optimal frontier. The results reveal that the optimized design enhanced kLa by 98.4% (reaching 312.5 h-1) while maintaining a shear intensity comparable to the baseline. Finally, fermentation experiments with Blakeslea trispora confirmed that the optimized configuration led to a 30.9% increase in β-carotene yield (attaining 3193 mg/L). This work provides a robust and efficient methodology for the rational design of bioreactors and offers a new pathway for the intensification of shear-sensitive bioprocesses.