Flow and mass transfer prediction in anisotropic TPMS-structures as extracorporeal oxygenator membranes using reduced order modeling

Currently, hollow fiber membranes are the standard technology for extracorporeal membrane oxygenators. Apart from the inevitable contact of the circulating blood with the artificial material, a suboptimal flow distribution within the oxygenator favors thrombus formation which leads to a rapid loss of gas exchange capacity. The current advancement in additive manufacturing allows the design of three-dimensional membrane structures, based on triply periodic minimal surfaces (TPMS). One of their unique advantages is local geometry variation to manipulate the flow distribution. But how this anisotropy influences the overall device performance is non-trivial and requires numerical simulation. The aim of this study was to develop a reduced order model (ROM) that is able to efficiently predict three-dimensional flow distribution and gas transfer inside TPMSstructures. We performed a parametric study using a validated micro scale computational fluid dynamics (CFD) model. Afterwards, two different modeling approaches of Sherwood-correlations and artificial neural networks (NN) were compared to characterize flow and mass transfer from the simulated data. To create the ROM, the NN modeling strategy was then implemented into a porous medium CFD model. The developed ROM was also validated. Finally, an anisotropic TPMS-membrane-structure was compared numerically with an isotropic predicate. The NN fitting strategy showed superior accuracy over the Sherwood-correlations for characterizing mass transfer in TPMS-structures. With the ROM, the gas transfer rates of oxygen and carbon dioxide from the micro CFD model could be predicted within relative root mean squared errors (RRMSE) of 2.7% and 5.1%. The pressure drop in the experiments was predicted with an RRMSE accuracy of 7.6%. In comparison with the isotropic TPMS-structure, the anisotropic structure showed a homogenized flow distribution and increased gas transfer rates of 4% to 5% for oxygen and 2.3% to 7.4% for carbon dioxide at the simulated flow rates.

Automated summary

This study aimed to develop a simplified model to predict flow distribution and gas transfer inside TPMS structures. Through a parametric study and model comparison, it was found that using artificial neural networks (NN) as a modeling strategy can accurately characterize mass transfer in TPMS structures. The developed reduced order model (ROM) demonstrated high accuracy in predicting gas transfer rates and pressure drop. The anisotropic TPMS structure showed advantages in flow distribution and gas transfer rates compared to isotropic structures.