(664g) Digital Twin for Mixing in Ultrafiltration/Diafiltration Using Computational Fluid Dynamics

During batch ultrafiltration and diafiltration (UFDF), it is generally desired to keep the material in the retentate vessel homogenous. Non-homogenous material could impact ability to effectively exchange buffer and could potentially impact product quality through increased heterogeneity of pump/filter passes. The UFDF process can involve a large dynamic range of volume for a single process which is compounded when considering operational fit for many products/processes. This leads to challenging decisions for vessel design between facility fit and operational performance. High concentration products are particularly challenging due to higher viscosities and larger volume turndowns during ultrafiltration. Furthermore, mixing challenges often arise during scale up and are difficult to assess using scale down experimental models. Performing experiments at manufacturing scale is expensive, and if mixing challenges arise, experiment-first approaches can threaten timelines of critical business programs.

Computational fluid dynamics (CFD) can be used to simulate the mixing within a computer representation of the manufacturing scale vessel to assess process performance before implementing a UFDF process or as a tool during investigations. Current state of the art CFD tools use a snapshot approach for UFDF mixing to assess short circuiting [1], which reveals the flow and mixing patterns (including short circuiting). Typical time scales simulated are seconds-minutes. This approach is directionally useful but cannot directly predict the success/failure of a given operation. Particularly for ultrafiltration, the snapshot approach cannot simulate a gradually depleting tank and any concentration hotspots that may emerge over longer durations.

We have developed a fully dynamic CFD model that is capable of predicting the success/failure of an ultrafiltration process in the manufacturing scale vessel by simulating the entirety of the UFDF operation (true “digital twin”). Use of cutting-edge lattice Boltzmann models using MStar CFD enabled simulating the entire process timeline, which can be hours of mixing. The failure mode with respect to mixing relates to a dynamic stratification of material which is not predicted by the snapshot approach. The fully dynamic model was validated against at scale manufacturing runs with multiple processes in multiple vessels. While mechanistic models are generally useful for making process decisions, models that can directly predict the outcome of a process can have more direct impact on business decisions. This is an essential part of a digital twin, and the advantages of such models will be discussed. The digital twin approach has been utilized for a model-first decision making process, eliminating the need to wait for costly engineering batches.

[1] Wutz et al. Biochem. Eng. J. 2020. Computational fluid dynamics (CFD) as a tool for industrial UF/DF tank optimization