Dynamic compartment models

Abstract

Anticipating the occurrence and effects of mass transport limitations during fermentation scale-up is essential for commercialization, as heterogeneities might affect microorganisms. Tools like Computational Fluid Dynamics (CFD) aid this analysis but are computationally intensive, limiting design space exploration and consequently, fermentation optimization. Compartment models (CMs) based on CFD simulations offer an affordable alternative but require CFD recalibration with changing geometries or operating conditions, restricting their usage in optimization.
In this work, we introduce a hybrid machine-learning-aided compartment model (ML-CM) that accounts for flow pattern dynamics upon changes in both volume and stirring speed in a stirred tank bioreactor. The ML-aided dynamic compartment model (dyn-CM) enabled the spatiotemporal study of a process in 1/500th of the fermentation simulation time, maintaining reasonable accuracy. This method facilitates fed-batch fermentation modeling, process optimization, and scale-up effect analysis with modest computational resources, supporting reactor design and operational improvements within a defined operating space.