A mini-module with built-in spacers for high-throughput ultrafiltration

Ultrafiltration membrane modules suffer from a permeate flow decrease arising during filtration and caused by concentration polarization and fouling in, e.g., fermentation broth purification. Such performance losses are frequently mitigated by manipulating the hydrodynamic conditions at the membrane-fluid interface using, e.g., mesh spacers acting as static mixers. This additional element increases manufacturing complexity while improving mass transport in general, yet accepting their known disadvantages such as less transport in dead zones. However, the shape of such spacers is limited to the design of commercially available spacer geometries. Here, we present a methodology to design an industrially relevant mini-module with an optimized built-in 3D spacer structure in a flat-sheet ultrafiltration membrane module to eliminate the spacer as a separate part. Therefore, the built-in structures have been conceptually implemented through an in-silico design in compliance with the specifications for an injection molding process. Ten built-in structures were investigated in a digital twin of the mini-module by 3D-CFD simulations to select two options, which were then compared to the empty feed channel regarding mass transfer. Subsequently, the simulated flux increase was experimentally verified during bovine serum albumin (BSA) filtration. The new built-in sinusoidal corrugation outperforms conventional mesh spacer inlays by up to 30% higher permeation rates. The origin of these improvements is correlated to the flow characteristics inside the mini-module as visualized online and in-situ by low-field and high-field magnetic resonance imaging velocimetry (flow-MRI) during pure water permeation.


Introduction
Ultrafiltration processes play an essential role in the biotechnological and pharmaceutical industries [1][2][3]. The key function of ultrafiltration (UF) membranes is removing harmful It tends to accumulate at the membrane surface, establishing a concentration polarization layer, and forming a complex deposit film on top of the membrane [6]. This fouling leads to a decrease in the solvent flux through the membranes [7]. Matthiasson and Sivik have already stated in the 1980s that concentration polarization occurring in membrane filtration processes leads to mass transfer inhibition [8]. Various issues have been taken to address this issue, including static mixers inserted into membrane modules to increase mixing and, subsequently, mass transfer [9].
Obstacles in the flow field reduce the critical Reynolds number to initiate oscillations, which highly depend on the fluid velocity and spacer geometry [10][11][12]. Static mixers function as such obstacles disrupting the laminar flow profiles in the boundary layer, causing vortices in the flow field [13], effectively reducing concentration polarization at low Reynolds numbers.
Qamar et al. approached the gap of a comprehensive study on the transition from steady to unsteady flow [12]. Obstacle-induced mixing also leads to higher velocities and shear rates at the membrane surface. The locally increased shear rate has, in turn, a positive effect on reducing fouling [14,15] induced by biological or colloidal matter. However, there is a maximum shear rate, above which biological matter is likely to be destroyed and can favor biofouling [12,15]. Computational fluid dynamics (CFD) simulations of mesh-type spacers show a better mixing (high shear rate, low drag coefficient) at the expense of an increase in axial pressure drop along the feed channel [12,[15][16][17]. This leads to a constant trade-off between flux increase and the correlated rise in energy input due to higher pressure loss [18,19].
Studies on net-or diamond-shaped spacers and meshes are available in large numbers and are discussed in several reviews [19][20][21][22][23]. The literature on experimental and simulation-based investigations of mesh spacers proves reduced biofouling and prolonged stable membranemodule performance [15,18,[24][25][26][27][28]. On a simple basis, the literature describes such mesh spacers or static mixers using 2D structures like triangles, squares, or circles and focuses 2 on their position and dimension [13,29,30]. Such studies support the understanding of flow-dynamic correlations as a pre-step on the way to new innovative design solutions, yet simplifications sometimes are too rigorous and correct representation of the genuine spacers details is questionable [15].
One of the less considered downsides of mesh-type spacers is that they require several assembling steps during in-line module assembly. Here, incorporation into the module housing is proposed with an anticipated single-step production by injection molding, accelerating the production process and reducing operating costs. Before establishing such a production process, we propose to follow a methodology of in-silico design combined with 3D-printingbased rapid prototyping. Removing such mesh spacers from a module architecture has been addressed by incorporating spacers as turbulence promoters into the membrane as so-called membrane-con-spacer geometries for various membrane-surface structures [31][32][33][34][35].
A patterned membrane surface acts as a spacer structure and induces vortices [34]. Typical geometries include flat, line and groove patterns, pyramids, rectangular and circular pillars over the whole channel width , or individual elevations. Detailed studies on sinusoidal membrane surfaces investigate the effect of sinusoidal structures and the resulting hydrodynamics on fouling resistance [33,34].
Implementation of commercial mesh spacers used as built-in spacer structures exists only for simple geometries such as squares [36] or ribs [37]. Flow-aligned, bar-like filaments in a feed channel showed that flow is affected more by transverse than by longitudinal alignments [38]. A similar structure, a ladder-type spacer within a frame fabricated by 3D printing, was studied in ultrafiltration experiments [39] and through experiments and simulations of nanofiltration and reverse osmosis (RO) processes [10]. Both studies showed that an increasing inter-filament distance reduces the critical Reynolds number for the transition from laminar to turbulent flow. Here, the spacer elements direct the fluid flow to the upper and lower parts of the membrane module but contain undercuts, which is not suitable for injection molding. Other experiments show positive effects only in the entrance region for mesh-type spacers and embedded ribbon-type spacers [11]. Another built-in spacer type suitable for ultrafiltration is a zigzag spacer, which outperforms mesh-type spacers in terms of pressure 3 drop [40]. Staggered herringbone structures are a good example for built-in spacer structures but are suitable primarily for microfluidic applications [41]. Shrivastava et al. [42] examined asymmetric herringbone, helix, and ladder-type structures, intending to provide a guide for developing reasonable spacers. Positive effects in the mass transfer were highest for helices, followed by herringbones and ladder-type spacers [42].
While much work tries to understand flow distribution and hydrodynamics, some studies also address the corresponding simultaneous mass transfer phenomena quantitatively [15,18,37,[43][44][45][46]. At the membrane, high shear stress and the formation of recirculation zones and vortices are desired, enhancing the intermixing of material adjacent to the membrane [43].
Mass transfer enhancement is accompanied by an increase in axial pressure drop [45]. This results in an increase in mass transfer coefficients with increasing shear rates [43,44]. Mass transfer investigations were considered for simple geometries only [44,45], but recently our group has addressed membranes with spacers [31,47,48] and 3D-printed novel spacer and mixer geometries [49][50][51]. A general overview is given by the Virtual Issue of the Journal [52].
Some studies investigate 3D-printing-aided membrane modules due to the great potential of 3D printing for fast and cheap module improvement [23,53]. Module design and material selection must be considered in 3D printing, but research on 3D printing of membraneprocess-related questions progresses rather slow [21,23].
So far, fundamental research was done focused on fluid flow simulations, simple 2D structures, or inlay mesh-type spacers to better understand correlations between feed channel obstacles and fouling behavior. 3D simulations are preferable to 2D simulations because the latter underestimate the parameters examined and the three-dimensional effects of intermixing [15,22,24].
In this study, we built on all this preliminary work and used CFD as a widely established tool for evaluating flow and concentration behavior in a digital twin. The novelty of this work consists in the methodology to (a) combine CFD simulations within a digital twin, (b) protein fouling experiments with a 3D printed copy of the digital twin, and (c) MRI investigations to prove the presence of flow patterns as predicted by the CFD simulations. We performed 3D-4 CFD simulations to compare ten potential built-in structures regarding shear rate distribution and pressure loss. Subsequently, the simulations of the two structures with the most promising shear rate-to-pressure loss ratio were extended by mass transfer correlations. To verify the simulated permeation increase, these two structures were implemented in a high-throughput membrane module fabricated by 3D printing for fast prototyping and tested in bovine serum albumin (BSA) filtration experiments. The positive effect of the chosen structure in the final product was correlated to hydrodynamic effects visualized by online and in-situ low-field (LF) as well as high-field (HF) flow-MRI measurements. We complement CFD simulations with both fouling experiments and flow-MRI measurements for a qualitative comparison and to better understand prevailing hydrodynamic phenomena.

Additive manufactured membrane modules
The permeation rate for different built-in spacer geometries was experimentally evaluated in an additively manufactured membrane module that was fabricated using polyjet 3D printing (Stratasys, Objet Eden 260V). The membrane module shown in Figure 1 was printed layer by layer with a transparent photopolymer (Stratasys, RGD810). During printing, the internal fluid channels need supporting structures (Stratasys, SUP705), which were later removed using a high-pressure washer (Krumm-tec, RK Top 5). Any remaining material was subsequently dissolved in a stirred bath of 1 mol L −1 sodium hydroxide for approximately 24 hours.

Design of the built-in spacer
For the conventional implementation, a mesh spacer, usually a fabric, would be placed in the empty feed channel, which also functions as an additional element to protect the membrane from damage. The advantage of replacing mesh spacers with built-in spacers is the possibility of single-step production that requires less time, e.g., injection molding, without the need for a spacer placement step. During injection molding, polymer granulate is molten and transferred into a metal shape injected into a mold cavity via a nozzle. For the design of a spacer built-in in the membrane module, any ramifications or undercuts need to be avoided in order to facilitate removing the molded part. Simple superficial cavities ensure a uniform distribution of the mold mass in the tool and a smooth surface of the module resulting in consistent product quality [54]. Built-in spacer structures were designed following injection molding restrictions, thus avoiding ramifications or undercuts and highly detailed structures.
The module investigated in this work is a small-scale crossflow module consisting of two parts. Figure 2 shows the two module parts with the membrane in between (Figure 2  The adjacent sides of the basic element display sinus functions, which differ by half a period.
As the sinus functions are orthogonal, they form a cross-shaped pattern. A complete sinus wave measures 2 mm and, like the herringbone structure, has a distance to the membrane of 32 µm. These dimensions result in 38 peaks for the feed channel length. As a variation of the sinusoidal corrugation, the number of peaks is halved. phenomena [27]. This study used specifications for a reverse osmosis membrane and its permeability for the chlorophenol concentration studied by Al-Obaidi and Mujtaba [55], called RO-approach in the following. The calculated results of Al-Obaidi and Mujtaba were validated by experiments [56].

Fluid dynamics simulation
Combining laminar flow with transport of diluted species prolongs the simulation time significantly as it combines convective flow phenomena with diffusive transport. For this study, the following assumptions were made: • membrane rejection of 100 % for solutes; • membrane modeled as rigid element without deformation due to pressure; • constant ambient pressure and pure water on permeate side; • constant feed composition with a fixed solute concentration; • and density, viscosity, and diffusion coefficients are constant and concentration-independent. 10 Transport of diluted species is dominated by convective flow according to the Navier Stokes equations. Concentration gradients cause species transport in the form of diffusion.
In addition to flow tangential to the laminar streamlines, diffusive mass transport can also occur normal to the flow direction. In this study, mass was neither produced nor consumed nor changed over time. Thus, the total fluid movement was dominated by convective and diffusive flow.
To enable permeate volume flow, trans-membrane pressures (TMP) were applied by setting p Outlet as outlet condition in the laminar flow node and to later calculate the TMP according to: The outlet pressure on retentate side was set between 0 bar and 12.5 bar in steps of 2.5 bar.
The model evaluated the corresponding inlet feed pressure, which resulted to be slightly higher than the outlet pressure. The pressure on permeate side was at ambient pressure, which equals an overpressure of zero in this model. Thus, the local pressure p corresponds to the pressure difference between the feed and permeate side, acting as driving force.
The addition of a diluted species demanded a further characterization of the mass transport at the membrane. The model provided a permeate flux as a second outlet condition as function of the pressure difference and the concentration at the membrane. It was described as the orthogonal outlet velocity U p , where the characteristics of the membrane are summarized in the permeability or solvent transport coefficient A w : The membrane flux increases with higher pressure difference p and decreases with increasing osmotic pressure difference π = R · T · c due to higher local concentration (see equation (2)). To compare the different structures, the permeate volume flow was determined by integrating the orthogonal outlet velocity over the membrane surface for the different TMP. A sketch of the geometry with the applied boundary conditions can be found in the supporting information (see Supplement Figure 1).
Flow conditions and concentration-related assumptions taken from [56] are summarized in Table 1, where c 0 stands for the entry concentration, c p for the permeate concentration, and A w for the membrane permeability. Table 1: System conditions for simulation with concentration polarization using material properties according to [56].

Parameter Value Unit
Laminar flow

Laminar flow simulations visualize hydrodynamics
In addition to vortex formation, shear stress is also considered to favor mass transport [43]. Shear stress is caused by a tangential force generated through distortion of fluid and can reduce particle deposition on membrane and module surfaces. Thus, high shear rates are considered to increase the permeation rate of an ultrafiltration membrane [30,34].
The study below compares spacer designs in order to identify high shear rate distributions averaged over the membrane. Figure 3 shows the development of shear rate and concentration over the whole module (inlet on the left; outlet on the right) for the empty channel, herringbone structure and sinusoidal corrugation. In both Figure 3 a)   with CFD simulations of mesh-type spacers and similar inlet velocities [16]. In general, the trade-off for the increased wall shear stress is an increase in pressure drop along the feed channel [13,14,18,62]. However, in this case, the sinusoidal corrugation does not elevate the pressure loss (see Supplement Figure 5).
In the literature, wall shear stress is often considered. In this study (Re = 18, TMP 12.5 bar), a value of 1.7 Pa is obtained, which correlates with simulated values of Mazinani et al. [33] and is in the same order of magnitude compared to recommendations for optimal operation of common UF applications (2 -6 Pa, see [63]).
The shear rate highly depends on the local velocity. The spacer structures occupy different amounts of space in the empty channel, consequently narrowing the cross-section. This leads to a higher flow velocity and thus to an increase in shear rate. The void volume of the empty channel measures 332 mm 3 , which is a lot larger than that of the herringbone structure (306 mm 3 ) and the sinusoidal corrugation (223 mm 3 ) and leads to the observed shear rate increase. The inserted structures were not normalized regarding the remaining void space but instead kept a fixed membrane distance in order to maintain similar conditions for the shear rate effects. The proposed membrane distance of 32 µm resulted from previous investigations.

Diluted species retention simulations
To investigate the correlation between shear rate and concentration, 2D plots for the concentration on the membrane surface were compared (see Figure 3 b)). Low concentrations at the membrane are desired as they result in a small osmotic pressure, which, in turn, increases the permeate flux (see the function for the permeate flux in equation (2)). In the legend in Figure 3   Thus, permeate fluxes, velocity and pressure profiles develop according to the respective structure. The permeate flux rises for increasing TMP according to equation (2). As this study uses RO as a replacement system, the permeate flux is small compared to an ultrafiltration process, which complicates visualizing differences in the respective permeate fluxes clearly. Thus, the TMP is increased up to 12.5 bar, which is also in accordance with pressures used by Al Obaidi et al. [55].   The critical Reynolds number, usually an indicator for the transition from steady to unsteady flow, can be identified for complex, realistic spacers [12] or calculated for rather simple structures [13,38], and lies between 100 and 400, which exceeds the here obtained in [62]). An exponent of 1.73 indicates turbulent flow, but mass transport can also develop laminarly despite the reported exponents in mesh-type spacers [18].
High pressure drop and high volume flow both affect pumping performance in the later application, especially when scaling up to larger membrane modules. A substantial increase in pressure drop is undesirable because it leads to higher operating costs, e.g., due to pumping capacity [18]. Given that the module is implemented as a single-use product for ultrafiltration processes, peripherals (vessels, hoses) are also designed to be disposable and often limited to maximum operating pressures of 3 -4 bar. Although higher volume flow usually has a positive effect, turbulence can also negatively affect friction losses and heat input and lead to higher operating costs (especially regarding the pumping capacity at high volume flows and high viscosities, as in UF processes). Turbulence is especially undesired in the main flow due to friction losses and temperature increase, in addition to the almost quadratic pressure drop mentioned above. The temperature increase is equivalent to energy dissipation and can negatively affect the products present (proteins, enzymes). For this reason, a low-pressure  This results in large permeate-flux differences between the two systems.

Experimental validation from BSA experiments
The BSA experiments demonstrate that all three spacer structures outperform the empty channel, which is in accordance with literature [13,64]. Sinusoidal corrugations outperform the other spacer structures in agreement with the 3D-CFD simulations. Still, 3D-CFD simulations serve the qualitative comparison of the emerging trends rather than an exact match of the conducted experiments. In literature, mesh-type spacers are discussed to negatively influence biofouling due to their unfavorable shear stress distribution behind mesh-strand intersections [65], and thus biofouling being rather an inevitable feed spacer problem [66]. Here, the built-in spacer structures are implemented as a superficial shape and therefore offer a positive effect compared to conventional mesh-type spacers. Although, in-situ biofilm formation has not been studied here, we assume that initial nucleation is unlikely to occur on the elevations of the sinusoidal corrugation but rather on the membrane.  Figure 6 reveal the inner structure of the membrane module, consisting of sixteen permeate channels, a membrane lying on top of the channels, and the feed channel with the built-in sinusoidal corrugation. The brighter the color in the magnetic resonance image, the higher the proton density in the respective pixel in the image, and thus, the water content is higher. The reason for permeate channels not being visible is the absence of water in the measuring area (see the fifth permeate channel in Figure 6 a) or the third and fifth permeate channel in Figure 6 b)). In the MRIs, the benefits of a high-field tomograph become obvious. LF imaging needs a bigger slice thickness to achieve high signal-to-noise 21 ratios, which are desired. However, small structural details can no longer be distinguished, as the signal is averaged over the whole slice thickness. On the other hand, HF imaging can capture a higher level of detail with a smaller slice thickness. Only the combination of non-invasive measurements with bigger (low-field NMR imaging) and smaller (high-field NMR imaging) slice thickness reveal and characterize all hydrodynamic effects in small filtration devices completely.

Conclusion
The comprehensive methodology to design fluid compartments minimizing concentration polarization resulted in a novel high-throughput membrane module optimized for flexible use in ultrafiltration screenings. The investigated built-in structures are suitable for one-step production, reducing manufacturing complexity significantly.
Simulations focusing on mass transfer and permeate-flux modeling visualize that the analyzed built-in structures influence velocity gradients and thus the shear rate distribution at the membrane. They show good agreement between regions of high shear rates and low concentrations on the membrane surface and explain the observed permeate flux enhancement.
The results suggest that the cross-sectional narrowing and the alternating structures not only increase shear rates at the membrane but also favor diffusion and thus mass transfer.
Both, simulations and experiments prove the highest permeate volume flux for the sinusoidal corrugation. This structure outperforms the conventional net-shaped spacer in experiments 23 with BSA by 10 % to 30 % higher permeate fluxes; followed by the herringbone structure and the empty channel. The reasons for the increased permeate flux were confirmed and elucidated using high-and low-field magnetic resonance imaging. Low-field measurements showed a mixing effect when averaging over several sinus waves as fluid flows around each sinus wave. High-field measurements revealed the flow pointing towards and away from the membrane in one single sinus wave. These two overlapping effects favor the filtration performance inside the membrane module. An extensive parameter study could further optimize the sinusoidal geometry.
Where other studies focus on simplifications and fundamental research of fluid flow, this study handles realistic built-in structures, which can directly be implemented in a commercial membrane module. The proposed CFD simulation can detect concentration accumulation and dead zones in membrane modules in advance and supports module design.
The proposed species-retention model is an acceptable first-order approach and elucidates the interaction of hydrodynamics and membrane transport during membrane filtration affected by local hydrodynamic conditions. For a more representative ultrafiltration model, the concentration-dependent properties such as diffusion coefficient, viscosity, and density should be implemented. Future studies could focus on in-situ biofilm formation to further correlate hydrodynamics with shear stress and time-dependent fouling development.
The new module design with built-in structures improves the filtration process due to mass transfer enhancement and makes conventional net-shaped inlays obsolete. Due to the simplicity of the structure and the practical manufacturing step of injection molding, an enlargement of the module to more membrane area would be well conceivable. In the future, spacer structures could also be modified according to specific applications (e.g., high or low viscous concentration profiles, shear sensitive products like cells).