A mechanistic approach for predicting mass transfer in bioreactors. (29th June 2021)

• Record Type: Journal Article
• Authors: Thomas, John A.; Liu, Xiaoming; DeVincentis, Brian; Hua, Helen; Yao, Grace; Borys, Michael C.; Aron, Kathryn; Pendse, Girish
• Is Part Of: Chemical engineering science. Volume 237(2021)
• Publication Date: 2021-06-29
• Subjects: Mass transfer -- Bioreactor -- Mixing -- Multiphase -- GPU; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; Chemical engineering; Periodicals; Electronic journals; 660
• DOI: 10.1016/j.ces.2021.116538
• Languages: English
• ISSNs: 0009-2509

Abstract

Graphical abstract: Highlights: Oxygen transfer in bioreactors is governed by complex fluid mechanics. These complexities make predictive mathematical modeling slow and difficult. Here we develop a general, physics-based approach to modeling oxygen transfer. We purposely design the model to execute on fast-running GPU-based computers. The models run fast and correctly predict oxygen transfer at multiple vessel scales. Abstract: The biomanufacturing processes that produce biologic drugs has become an extremely important area of study within the pharmaceutical industry. Within such processes, the drug substance is typically produced by living organisms within stirred tank bioreactors that require a continuous supply of sparged oxygen. The overall oxygen transfer rate to the fluid is a nonlinear convolution of the gas bubble size distribution, fluid properties, local fluid energy dissipation rates, and local dissolved oxygen concentrations. The complexity of this process presents challenges to process scale-up and intensification. In this work, we propose, implement, and validate a mechanistic transport model for predicting oxygen transfer rates within stirred tank bioreactors. To begin, we describe the relevant conservation laws and key principles from turbulence theory that govern mass transfer. Next, we present a physics-based modeling approach for solving these equations in tandem and in real-time. We then systematically validate the model against experimental data at operating scales ranging from 5 L to 2000 L. By running the algorithm on graphics processing units (GPUs), the approach is shown to solve at timescales practical for industrial application.