Sustainable additive manufacturing of polysulfone membranes for liquid separations

Membranes serve as important components for modern manufacturing and purification processes but are conventionally associated with excessive solvent usage. Here, for the first time, a procedure for fabricating large area polysulfone membranes is demonstrated via the combination of direct ink writing (DIW) with non-solvent induced phase inversion (NIPS). The superior control and precision of this process allows for complete utilization of the polymer dope solution during membrane fabrication, thus enabling a significant reduction in material usage. Compared to doctor blade fabrication, a 63% reduction in dope solution volume was achieved using the DIW technique for fabricating similarly sized membranes. Cross flow filtration analysis revealed that, independent of the manufacturing method (DIW vs. doctor blade), the membranes exhibited near identical separation properties. The separation properties were assessed in terms of bovine serum albumin (BSA) rejection and permeances (pressure normalized flux) of pure water and BSA solution. This new manufacturing strategy allows for the reduction of material and solvent usage while providing a large toolkit of tunable parameters which can aid in advancing membrane technology.


Introduction
Membranes for liquid, vapor, and gas separation serve as important components for many modern manufacturing and purification processes, with applications including water treatment, carbon capture and organic solvent separation [1].Membranes typically consume significantly less energy compared to other separation processes and are poised to play an important role in industrial decarbonization efforts.In addition, the emergence of new contaminants, such as perfluoroalkyl substances, require specifically tailored membranes [2,3].Emerging applications thus provide opportunities to not only design new membrane materials, but also to develop sustainable membrane manufacturing strategies.
Traditionally, flat sheet water treatment membranes are fabricated via film casting of polymer dope solutions as the precursor [4], the underlying mechanism being non-solvent induced phase separation (NIPS).Typically, these fabrication processes require large volumes of solvents [5].Therefore, investigation into approaches that require lower solvent volumes is of interest to the membrane community.
Additive manufacturing-based techniques, such as direct ink writing (DIW), are being increasingly considered for membrane fabrication since they facilitate precise control to the membrane fabrication process [6][7][8].DIW can be classified into three categories: continuous droplet-based, energy-assisted, and extrusion-based DIW [9].Extrusion-based DIW was chosen as the primary method for this work because of the flexibility that the technique provides for manufacturing processes of liquid-based inks.Extrusion DIW is able to accommodate a wide range of viscosities and viscoelastic behavior suitable for polymer dope solutions, while requiring less specific equipment compared to the other mentioned techniques, such as nozzles for droplet based techniques [10].DIW involves the robotic deposition of a prepared solution (i.e. an ink) via the application of a driving force, commonly in the form of pneumatic pressure or mechanical devices such as screws.The deposited ink may have a diverse composition as well as a wide range of viscoelastic properties, with a reported upper viscosity limit being ∼10 6 mPa s at a shear rate of 0.1 s −1 [11].Following the DIW process, solvent evaporation begins to occur which results in the precipitation of the solute molecule dissolved in the ink [9].By controlling the conditions which drive the solvent/solute separation, DIW/additive manufacturing can be implemented for membrane fabrication.
In this paper, an innovative additive manufacturing-based process for the fabrication of porous polysulfone (PSf) membranes as an alternative to traditional manufacturing methods is introduced.The fabrication method described here synergizes the DIW technique with the NIPS mechanism using a simple binary (polymer in solvent) ink.PSf was chosen since it is one of the most common membrane materials for liquid and gas separations [12].A critical goal of this work was to benchmark the separation performance of these DIW membranes against membranes fabricated using the doctor blade technology, in terms of water (or solution) permeance and solute rejection.An additional focus was to establish the scalability, repeatability, and long-term stability of these membranes.The DIW technique offers a range of tunable parameters, beyond the ones described in this paper, which could be used to create advanced membranes with minimal material usage.Overall, this technology could be transformative for membrane manufacturing and be used for a variety of challenging separations.

Formulation of membrane dope solution for DIW and doctor blade
PSf pellets were first dried at 110 • C for 24 h under vacuum.The dried PSf was added to DMAc at a concentration of 17 wt% and mixed at 230 RPM for 72 h at 65 • C (figure 1(a)) [13].The PSf-DMAc solution was then allowed to cool prior to use.Similarly, dope solutions used for the doctor blading method were prepared at a concentration of 17 wt.%PSf with NMP as the solvent; the solutions were mixed for 72 h at 230 RPM and 40 • C before being cooled for use.Prior work on doctor blading by Lu et al [14], was done with NMP as the solvent and it was decided to adopt it in this work as well.

PSf DIW membrane making
For membrane fabrication, the prepared PSf-DMAc solution was pipetted into a 5 cc syringe barrel fitted with a Nordson Luer-lock needle (Nordson EFD Precision Tips, Westlake, Ohio, USA) with an inner diameter of 200 µm and a length of 6.35 mm.The filled syringe was then loaded into the holder arm of a Nordson Janome JR 2304N Robotic Printer (Chicago, Illinois, USA) (figure 1(b)).Pressure controllers (Nordson EFD, Westlake, Ohio, USA) were used to set the pneumatic back pressure applied during printing.An ultra-flat silicon wafer was used as the printing substrate (cleaned with ethanol and isopropanol prior to use), with the separation distance between the needle and substrate set optically using a Dinocam digital microscope prior to the start of the printing process.Optimized printing parameters are presented in table 1.A schematic of the needle path used during membrane printing is provided in the supporting information (figure S1).
The ambient relative humidity (RH) of the membrane fabrication environment was controlled by constructing a cover composed of two glove bags that fully encapsulated the 3D printer (supporting information figure S2).Humidified air from a mister (Ultrasonic humidifier, Guangdong, China) was used to set the desired humidity level within the enclosure and the humidity level was monitored using a benchtop meter.The temperature inside the enclosure was similarly monitored.Since the printing process is typically slow, the ambient humidity causes the phase inversion to be initiated before the film is immersed in the quench bath.This was apparent in the film losing some of its optical transparency as compared to the just-printed film (figure 2).Details of the fabrication process can be found in the video included in the supplementary materials.To ensure film uniformity and reproducibility, the films were allowed to undergo a full vapor phase inversion in the ambient humidity prior to quench bath submersion.The time to achieve full phase inversion was ∼9 min with the film considered being fully phase inverted once the entirety of the printed ink became white and opaque.DI water at ambient temperature was used for the quenching step.The fabricated membranes were kept immersed in DI water, with the water exchanged several times during the initial 24 h period prior to testing.These exchanges were done three times on the day of fabrication and three additional times the following day.

PSf doctor blade membrane making
The procedure used to fabricate PSf membrane sheets via doctor blading was adapted from previous studies [14 -16].These membranes were fabricated at the University of Kentucky, with all subsequent characterizations done at West Virginia University.Briefly, for each sample, approximately 3-5 ml of dope solution (3 ml generates a 13 × 7 cm membrane size and was the dimension used in solvent reduction analysis for this work) was poured onto a glass plate (Gardner Glass Products, North Wilkesboro, NC, USA) and cast across the surface using a doctor blade (Micrometer Adjustable Film Applicator-250 mm, MTI Corp, Richmond, CA, USA).This casting process was performed using an automatic bench-top flat sheet casting machine (Model: BTFS-TC, PMI, Ithaca, NY, USA) set at a casting speed of 500 cm min −1 .An automated doctor blading setup was used to avoid variations in membrane film thickness that could be present in manual casting.To match the approximate duration of the DIW process, the film was exposed to ambient air for 8-9 min (RH: 40 ± 2%) before being immersed in a water quench bath.Once immersed in the quench bath for 1 min, the samples were stored in DI water.

Membrane permeance testing protocol
All fabricated membranes underwent a compaction step with pure DI water utilizing a CF 042 standard lab grade cross flow system (Sterlitech, Kent, Washington, USA) with a membrane surface area of 42 cm 2 (figure 3).A positive displacement pump was used to feed water from the feed tank (∼19 l) to the membrane cell.A bypass needle valve was utilized to recycle part of the feed stream back to the feed tank.
A PolyScience chiller (Niles, IL, USA) was utilized to maintain the desired water temperature during the testing process.Transmembrane pressure across the membrane module was controlled by adjusting the control valve connected to the retentate site as well as the bypass needle valve connected to the feed site.The pressure was monitored using a pressure gauge.The retentate flow rate was monitored using a Site Read Panel Mount Flowmeter (Tampa, FL, USA).All pure water tests were carried out at a temperature of 25 • C, a 0.8 l min −1 retentate rate, and a 5 bar transmembrane pressure.The permeate flux was measured gravimetrically by weighing the volume of water collected over a 5 min period.The permeate was recycled back to the feed tank, except during sample collection periods, to avoid concentration polarization.The BSA rejection tests were started once the membrane reached its steady state pure water permeance value (typically ∼6 LMH bar −1 ).Once this steady state was reached, the cross flow filtration was stopped, and all DI water was removed and replaced with 50 mg l −1 BSA solution.The BSA filtration experiments were performed using a retentate flow rate of 2 l min −1 .This increase in tangential flow was done to minimize concentration polarization [18].Once the filtration was restarted, an immediate sample was taken followed by three more samples taken on the hour.The BSA solution permeance was measured in the same way as the pure water permeance.The feed tank and the remaining system was cleaned in between consecutive experiments to prevent membrane fouling.Membrane permeance (reported in L m 2 .h.bar , LMH bar −1 ) was measured using the following relation (equation ( 1 S3).Membrane rejection (%R) was calculated using equation ( 2) (2)

Scanning electron microscopy sample preparation
All membrane films were prepared for scanning electron microscopy (SEM) imaging (JSM-7600F, JEOL, Tokyo, Japan) by fracturing under liquid nitrogen to help preserve the internal morphology.The membrane sections were then sputter coated for 30 s in a Cressington Sputter Coater (Argon environment) using a mixed Au/Pd metal target (60/40 ratio, Ted Pella, Watford, UK) and a sputter gas chamber pressure of ∼0.1 mbar.An acceleration voltage of 5 kV and a working distance of approximately 15 mm was used.

Substrate surface roughness measurement
The surface roughness of the silicon and glass substrates were determined using optical profilometry (Contour GT KO Optical Profiler, Bruker, Billerica, MA, USA).The profilometer was operated in the vertical scanning interferometry measurement mode with a green light source for the measurement of rough surfaces, while the phase shift interferometry measurement mode with a white light source was used for significantly smoother surface.

Results and discussions
Prior studies have primarily focused on fabricating membranes suitable for dead-end filtration setups (<15 cm 2 ) and membrane spacers [19][20][21][22][23].In contrast, the goal of this work was to produce defect-free PSf membrane samples of >40 cm 2 area so they can be tested in lab-scale cross flow setups for prolonged periods.

Preliminary optimizations 3.1.1. Substrate surface roughness effects
An often-overlooked aspect of membrane casting is the role of the underlying substrate which is typically a glass slide in most cases.For fabrication membranes using DIW, the surface roughness of the substrate was found to play a significant role.The glass surface did allow for some membranes to be fabricated, however, the frequency and area of defects (i.e.holes and gaps in the printed films) produced in most of the prints did not allow for consistent and reproducible membrane fabrication.Upon focusing on the defects, it was observed that defect occurrence appeared with a certain degree of spatial consistency, which we correlated to surface roughness of the underlying substrate (figure 4).The surface roughness hypothesis was further confirmed by briefly sanding the glass slide which led to a slightly lower overall surface roughness (figure 4).This resulted in some defects disappearing as well as an increase in the frequency of fabricating defect free membranes.Despite these initially promising results, consistent fabrication of membranes on the glass substrate still proved challenging, leading to the use of an ultrasmooth, polished silicon wafer, with roughness measuring in the nanometer range.The frequency of defect formation was drastically reduced for the silicon wafer, compared to the glass slide (with and without sanding).The results agree with the hypothesis regarding the substrate surface roughness affecting the film formation, since the mean surface roughness of the silicon wafer was ∼2 orders of magnitude lower than that For the comparatively rough substrates such as the initially chosen glass slide, defects (i.e.holes) in the printed films occurred with regularity at specific positions on the substrate.While sanding the glass substrate reduced the surface roughness and mildly decreased the frequency of defect formation, transitioning to a polished silicon wafer (∼2 orders of magnitude lower roughness) significantly eliminated the occurrence of these defects in the printed films. of glass slide.All subsequent experiments were performed with the silicon wafer as the substrate.It must be noted that such subtle effects may not have been apparent if smaller dead-end filtration samples were printed instead, which highlights the need to fabricate samples of reasonable dimensions while developing new manufacturing strategies.

Relative humidity (RH) effects
The ambient RH is another important factor, which plays an important role in the phase inversion process.Since water acts as a non-solvent, the amount of moisture present in the ambient air could impact the phase separation behavior of the polymer membrane.Some preliminary assessments of the RH effects revealed that 40% RH provided optimum permeances for the DIW membranes.Initial pure water permeances of DIW membranes, as a function of the RH, can be found in supporting information figure S4.The RH level, the DIW membranes were fabricated in, had a dramatic effect on the substructure morphology of the membranes.As seen in supporting information figure S5, a 20% RH environment produces finger like pore structure while 40% RH and above produces a honeycomb like pore structure (figure 5).The 20% RH membrane was fabricated and then almost instantaneously introduced into a DI water quench bath post-fabrication to begin the NIPS process.The DIW membrane took ∼8-9 min to be printed.In the case of the 40% RH membrane, the exposure to humidity during this long print time inadvertently initiated some vapor-induced phase separation (VIPS).The supplementary video provided along with this paper clearly shows this fact.Partial phase inversion, as shown in figure 2(a), could lead to non-repeatability in the membrane performance, therefore it was decided to wait for this process to be complete for all membranes.To allow for the phase inversion process to be completed, an additional ∼0.5 min were required and during this time, the 40% RH DIW membranes were left in the humidified glove bag.The actual fabrication time for doctor blade membranes was significantly shorter (3-5 s).However, to ensure a fair comparison, these membranes were also left exposed to the humid environment for 8-9 min to initiate the VIPS process in them.
As noted in prior studies, the honeycomb structure has been commonly observed for a membrane fabricated using high humidity conditions using the VIPS process [24].This was also seen for the doctor bladed membranes (discussed later), when fabricated under equivalent RH conditions.It should be noted that typically the finger like pores generate a larger pure water permeance, but in this instance, it was found to be the opposite [25].This is presumably because the fabrication process is a combination of partial VIPS and NIPS and hence some properties are unique to this case.Changes to printing parameters like higher print speeds and nozzles with larger diameters could reduce the printing time, which will essentially eliminate the VIPS-like effects.

Morphology of DIW and doctor bladed membranes
Defect-free DIW PSf membranes were fabricated on silicon wafers at 40% RH conditions.The performance and morphologies of these membranes were compared against doctor bladed membranes fabricated under similar conditions.To investigate the morphologies of doctor bladed and DIW membranes, SEM imaging of the surface and cross section of the membranes were performed both before and after pure water permeance testing (figure 5).
Owing to differences in lab safety regulations between the two universities, the chosen solvents used in doctor bladed and DIW membrane fabrication were NMP and DMAc, respectively.This solvent difference, having been explored in literature, has been noted to result in nearly identical membranes, with DMAc based membranes having slightly greater porosity compared to NMP, and thus a slightly higher pure water permeance [26].As shown in supplementary figures S11 and S12, the performance and morphology of DIW membranes fabricated with NMP showed minimal differences with the ones fabricated with DMAc.The morphology, as shown in figure S12, was not exactly similar to DMAc-membranes, however, honeycomb-like pores were observed in this case as well.The starting fluxes were significantly higher for the NMP-fabricated membranes; however, the final steady state values leveled out at similar values as DMAc-fabricated membranes.Thus this data shows that the solvent effect, while not negligible, is not particularly significant either.
The cross-section images of the membranes fabricated using the two manufacturing processes are almost identical, with the doctor bladed membrane having a slightly thicker selective layer than the DIW printed membranes.Both membranes had similar thickness values (∼120 µm) as well.The cross-sections of both the membranes were quite symmetric as opposed to typical NIPS membranes, and this can again be attributed to the slightly long exposure to 40% RH conditions.Such symmetric morphologies have been observed in prior work [27,28], for membranes prepared under high RH values.SEM images were taken before and after testing, with the cross-section of both membranes remaining unchanged in terms of thickness and porosity.An interesting observation was made for the doctor bladed membrane with regards to the formation of microvoids.Multiple SEM images were taken to confirm that this observation was indeed true, and that these voids were not simply a result of sample preparation artifacts (supporting information figure S6).The microvoids are hypothesized to be due the motion of the doctor blade over the stationary glass support, which may lead to some air being trapped between the substrate and the dope during casting.An analysis of the size distribution of the pores in the cross-sectional SEM images as well as the corresponding BET data can be found in supporting information sections 7 (figure S7) and 8 (figure S8) respectively.The average pore size of the membrane cross section was ∼5.4 µm for DIW membranes and ∼6.4 µm for doctor bladed membranes.For this range of pore-sizes (i.e.>300 nm), the BET analysis was not particularly useful and no changes were apparent in the pore size distribution of just-prepared and tested membranes.
Unlike the cross section, few key differences were observed between the two membrane types in terms of the surface SEM images.Interestingly, the side that is open to the atmosphere has larger pores (∼3.4 µm), compared to the side fabricated on the printing substrate (∼1.4 µm).In contrast, the selective layer of the doctor bladed membrane was formed at the surface facing the atmosphere, with pores of ∼1.7 µm.It is the difference in printing substrates (ultra-smooth silicon wafer as opposed to a glass slide) that influences the selective layer formation.When membranes were manufactured on the glass slide, the membrane is released off the slide surface almost instantaneously when it is quenched in water, whereas the silicon wafer fabricated membranes would release in ∼45 s after submergence.The contrast in surface roughness (figure 4) between the glass slide and the silicon wafer is the hypothesized reason for the membrane sticking longer to the substrate.It is this increased membrane dwell time on the substrate that allows for a smaller pore size distribution to be formed on the bottom of the DIW membrane rather than in the middle like in most pure VIPS process [27].These trends can be seen in the feed and permeate pore sizes found in supporting information section 6. Bearing these differences in mind, the surface with the smaller pore size was chosen as the 'feed' side in both cases.Another interesting observation was the dilation of pores in both the feed and permeate sides of the DIW membrane.It is hypothesized that the diffusion of the solvent (i.e.DMAc) was incomplete and the application of transmembrane pressure caused the residual DMAc to diffuse across the membrane, therefore dilating the pores on both surfaces.In both membrane types, complete densification of the selective layer did not occur, as previously observed [29], due to the exposure to moisture for slightly longer time periods.

Separation performance of DIW and doctor bladed membranes
The combined VIPS and NIPS process not only influences the morphology, but also the resulting performance.A tight cross-section was formed due to slow non-solvent induced demixing kinetics, and this resulted in lower water permeances than the prior studies reported using PSf membranes with finger-like pores [30].While the cross-section was very porous, the pores lacked inter-connectivity and led to a more tortuous diffusion pathway for the water and solute molecules.
The steady state pure water permeance of the DIW membranes (120 h) and doctor bladed membranes (80 h) was determined in a cross flow system (figure 5).Such a system allowed testing the membrane samples in presence of a tangential retentate stream which further validates their mechanical integrity under realistic conditions.
Both doctor bladed and printed membranes showed similar characteristics in their pure water permeance behavior (figure 6).Despite an initial variability in starting initial pure water permeances, the final pure water permeances attained by both DIW and doctor bladed membranes were the similar (∼6 LMH bar −1 ).The doctor bladed membranes showed a slightly higher initial permeance of 110 LMH bar −1 compared to 96 LMH bar −1 for the DIW membranes.The decline in pure water permeance over time for both membrane types was a noted concern for this study.This decline was hypothesized to occur either due to membrane compaction or simply due to collapse in the porous structure during drying.To investigate the porous structure collapse phenomenon, a method, known as solvent exchange, was borrowed from hollow fiber spinning [31].This process involves slowly displacing the water within the pores of the membrane using volatile solvents like methanol and hexane.For each solvent the membrane is allowed to dwell for 20 min before the solvent is drained and replaced with fresh solvent.This was repeated three times for each solvent.Following the final hexane wash, the membrane was allowed to dry overnight before being dried in a vacuum oven at 75 • C for 2 h.Once the membrane was allowed to cool it was immediately tested in the cross flow system.Unfortunately, this solvent exchange did not prevent the pure water permeance decay with the subsequent results of that experiment shown in supporting information figure S9.This showed that the cause of pure water permeance decline was primarily due to membrane compaction.The solvent exchange procedure was discontinued for subsequent experiments.
The two types of membranes were also compared in terms of BSA solution permeance and rejection (figure 7).Since BSA fouling was expected, a high retentate flow rate of ∼2 l min −1 was used for these experiments.A decline in % rejection and BSA permeance was observed with time, despite this high retentate flow rate, and this could be attributed to fouling-induced concentration polarization.The BSA rejection was ∼60%-80% initially and dropped to ∼30% in case of both the membranes.These values are lower than previously reported values [4] for PSf membranes, the primary reason being the high (40% RH) humidity conditions used here.As noted in the context of the surface morphology, the presence of moisture prevents the complete densification of the selective layer and this affects both membranes almost equally.On the other hand, the 'closed pore' structure of the cross section provides a highly tortuous diffusion pathway for the solute molecules.In this case of high solute diffusion through the selective layer, the tortuous cross section presumably contributes more to the rejection value.
The BSA water permeance was slightly higher in the DIW membranes (5.7 LMH bar −1 ) compared to the doctor bladed membranes (4.5 LMH bar −1 ).However, after filtering BSA solution for 3 h the membrane permeance had switched, with the doctor bladed membranes finishing with a higher permeance (2.3 LMH bar −1 ) as opposed to the DIW membranes (1.7 LMH bar −1 ).Nevertheless, it is interesting to note that no significant differences were observed in terms of the morphology and performance for the DIW membranes from the doctor bladed membranes.This is encouraging, since the DIW process required only ∼37% of solvent for fabricating an equivalent size of membrane sample compared to doctor blading (supporting information section 10).In the current stage of development, the permeance and BSA rejection values are not at par with other state-of-the-art liquid separation membranes, however, several processing parameters (e.g.RH) and printing parameters (e.g.print speed, printing pattern) can be engineered to improve the separation performance.With respect to the fabrication time to create the membranes, parallelization through using multiple nozzles or nozzle arrays may help scale up the manufacturing throughput by reducing the printing time to be on the timescale of doctor blading.Employing nozzles with larger tip diameters can contribute to print time reduction by allowing for higher flowrates while the use of nozzles with unique tip geometries may provide additional dimensions for engineering the local shear flow conditions during deposition.Finally, the capacity for tuning the local flow rate of the dope solution through a combination of nozzle travel speed (print speed) and the driving pneumatic pressure may allow for locally controlling the membrane thickness, thus influencing the formation of the internal porosity across the membrane film.The large toolkit available for the DIW process provides opportunities to further engineer the structures and properties of these membranes and achieve better separation performance.Such optimization strategies will be pursued in our future work.

Summary and conclusions
This work establishes DIW as a novel and sustainable manufacturing technology for PSf membranes.When compared to doctor blading, the DIW membranes achieved virtually indistinguishable morphologies as well as similar values of steady state permeances and BSA rejections.Substrate surface roughness and ambient humidity played a significant role in regulating the morphology and performance of these membranes.Importantly, this technique only requires 37% of the solvent used per area of membrane compared to doctor blading.The DIW membranes are easy to scale and show stable performance when tested for prolonged time periods under cross flow conditions.In this work the primary focus was to develop a prototype and benchmark these membranes against doctor bladed membranes.Beyond the parameters discussed, the DIW process involves a large toolkit of tunable parameters, that can aid in advancing this technology and improving the separation performance of these membranes.It is envisioned that this technology could be transformative for scalable membrane manufacturing, especially for emerging applications.With impending regulations on the use of toxic solvents and the increasing focus on sustainable manufacturing [15], this DIW strategy allows for equivalent membrane area production with a fraction of material and solvent usage compared to traditional processes [32].
While PSf based liquid separation membranes were the focus here, future advancements in this technology could enable non-porous membranes for gas and vapor separations.The technology is not limited to PSf and can be extended to other glassy and rubbery polymers used for membrane manufacturing.The technique can possibly be adapted to produce advanced configurations like hollow fibers wherein multiple parallel nozzles can be used to simultaneously extrude the 'dope' and 'bore' solutions.The ability to fabricate such advanced configurations will expand the applicability of this novel technology to make membranes for varied separation processes.

Figure 1 .
Figure 1.Direct ink writing fabrication process for polysulfone (PSf) membrane films.A dope solution of PSf and DMAc is mixed (a) and loaded into a syringe barrel for patterning of the polymer solution on substrate (b).Following printing, the dope solution and substrate are submerged into water to quench and solidify the membrane through phase inversion (total printing area of 13 × 7 cm) (c).

Figure 2 .
Figure 2. In-situ phase inversion process during membrane fabrication via printing.(a) Partial humidity induced phase inversion of a printed polysulfone-DMAc film.The fully phase inverted sections are opaque and white in color while the non-phase inverted portions are transparent.(b) Example of fully formed membrane after non-solvent induced phase inversion (NIPS) process, distinguished by total opaqueness and uniform white color.

Figure 3 .
Figure 3. Cross flow testing setup for determining pure water permeance for the fabricated membrane films.Reprinted from [17], Copyright (2015), with permission from Elsevier.

Figure 4 .
Figure 4. Impact of the substrate roughness on the occurrence and repeatability of defect formation in the printed polysulfone membranes.For the comparatively rough substrates such as the initially chosen glass slide, defects (i.e.holes) in the printed films occurred with regularity at specific positions on the substrate.While sanding the glass substrate reduced the surface roughness and mildly decreased the frequency of defect formation, transitioning to a polished silicon wafer (∼2 orders of magnitude lower roughness) significantly eliminated the occurrence of these defects in the printed films.

Figure 5 .
Figure 5. Cross sectional and surface SEM images of the printed and doctor bladed membranes both before and after pure water permeance testing.The scale bar in each image is 50 µm.The tested DIW membranes have a width of 125 µm and the tested doctor bladed membrane is 119 µm.

Figure 6 .
Figure 6.Pure water flux of (a) doctor bladed membranes and (b) DIW printing/direct ink written membranes.For the doctor bladed films, four samples (Dr-M1-M4) prepared under identical conditions were measured to determine the variability in the permeance performance.Similarly, six membranes (DIW-M1-M6) prepared via printing under identical conditions were used for the pure water permeance measurements.

Table 1 .
Optimized printing parameters used during the membrane DIW process.
UV-VIS Genesys 10 S Thermo Electron Corporation, Madison, WI, USA) was used to quantify BSA concentrations in the feed (C f ), retentate (C r ) and permeate (C p ) solutions.BSA concentration was determined using the absorbance at a wavelength of 280 nm.A BSA solution standard was created and utilized for all BSA rejection experiments.The resulting figure, with a regression coefficient of 0.998, can be found in the supporting information (figure