The vast majority of polymeric ultrafiltration membranes for water filtration are made from either Polyvinyldifluoride (PVDF) or Polyethersulphone (PES). Both membrane types can be used for a wide range of filtration applications, from drinking water to municipal tertiary wastewater treatment, with differences in operating conditions. The PVDF membranes are usually operated by feeding the raw water to the outside of the fibers (out-in filtration), whereas the PES membranes are operated with raw water entering the lumen side of the capillaries (in-out filtration). Flux rates vary from 50 to 100 L/(m²h) in typical applications, depending on the feed water characteristics. The in-out membranes are typically backwashed with a flux of around 230 L/(m²h). Up to recently, the membranes by inge GmbH-Dupont were produced from PES and consisted of seven capillaries, the mentioned Multibore™ membrane. These membranes usually are operated in dead-end/direct flow mode. In a previous study this membrane has been analyzed by MRI (Schuhmann et al., 2019), showing that the flow distribution during filtration as well as during backwashing is even. The average pore size of these membranes is around 20 nm and the molecular weight cut-off has been measured at 100 kDa (both based on manufacturers internal methods). The novel polymeric membranes used in the present study have the same base ingredients with PES as the main polymer, the surface characteristics and average pore size, however they exhibit 19 feed capillaries. This membrane is called the Multibore™ PRO. The inner diameter of the channels amounts to 2*r ₀ = 0.7 mm, as compared to 0.9 mm for the Multibore™ membrane with seven capillaries. As evident from the photograph (Figure 1), the channels are sorted in two rings of 6 and 12 channels around the central channel with corresponding radii of about 1 mm and 2.2 mm. The benefits of this fiber are easier handling during manufacturing, but as mentioned and more importantly, it is possible to effectively pack more active membrane surface area in a module. This increase amounts to around 18%. As a consequence, less modules need to be transported and filtration plants can be built with a reduced physical footprint.

The Multibore™ PRO membrane fibers were glued by silicone into an MRI compatible filtration module, which allowed in-situ measurements under filtration conditions (under flow) as well as under static conditions.

Fresh 19-channel Multibore™ PRO membranes were used to study wettability and flow fields in the diverse feed channels. The latter give an indication about the incoming flow and the transition from a 6 mm feed tube to the 19 channels of the membrane. Doubly-deionized water was used in these studies ensuring an independence on common water impurities.

The specifications of Multibore™ PRO defined the pressure ranges of filtration. Thus, transmembrane pressures p _TMP in the range p _TMP ϵ [0.1, 0.5] bar were used in the filtration and backwashing experiments. Filtration was performed over 1–2 h in the constant pressure mode, while backwashing at p _TMP = 0.5 bar was done on a time scale of [20, 30]s. In each study, one filtration cycle was performed.

While using pure water as feed to determine the water equivalent or flux reference, MR images revealed structural changes with increasing “filtration” time. The flux J amounted to J = 82 L/(m²h) at p _TMP = 0.12 bar, while J = 325 L/(m²h) at p _TMP = 0.51 bar.

As feed in the filtration experiments, the well-known aqueous sodium alginate solution, for example (Arndt et al., 2017; Schork et al., 2021), was applied with the following composition: c _NaAl = 200 mg/L, c _Ca2+ = 2 mmol/L, and the concentration of MRI contrast agent “Nanomag-D-spio MagAlg” was c _CA = 1 mmol/L. The sodium alginate is known to form gels mediated by Ca²⁺ ions which is illustrated in the egg-box model (Grant et al., 1973). Above a certain concentration and on a time scale compatible with the MRI experiments the build-up of deposit can thus be monitored and analyzed. The nano-particulate contrast agent used is chemically compatible with the sodium alginate and was shown to remain in the gel. It therefore provides a well-established possibility to improve the MR image contrast while keeping the MRI measurement time comparatively short in a time range <1 min. The characteristics of the contrast agent, which are most relevant for the MRI experiments apart from possible sedimentation and agglomeration, are the relaxivities, i.e., the paramagnetic relaxation enhancement. The relaxivities of Nanomag-D-spio MagAlg amount to r ₁ = 1.1 1/(s mol) and r ₂ = 231.2 1/(s mol) at a ¹H Larmor frequency of 200 MHz, in the range of previous values (Arndt et al., 2016; Arndt et al., 2017; Schuhmann et al., 2018; Schork et al., 2021). The transverse relaxivity r ₂ is dominant as usually observed in the case of nano-particulate contrast agents, for example (Vuong et al., 2012). This dominance of r ₂ leads to a reduced intensity in the MR images when the contrast agent is present in significant concentrations above the lower limit of detection. The deposits of alginate gel on the inner surface of the channels will therefore appear “darker” than the liquid feed and permeate in the usually acquired relaxation weighted images.

Apart from the mentioned concentrations, the following filtration parameters were used in the experiments: Filtration was performed as in-out filtration with quasi-constant pressure, namely p _TMP = 0.5 bar for pure water filtration (water equivalent) and during filtration of aqueous sodium alginate, respectively. The backwashing was also performed at p _TMP = 0.5 bar. The filtration setup was the same as described for example in Schork et al.( 2021). Also, the procedures for filtration and backwashing were similar to the previously described ones.

The polymeric 19-channel membrane Multibore™ PRO was glued into an MRI compatible single-fiber module as mentioned (Figure 1), the fluidic connections and valves for controlling the fluidic paths of feed, concentrate, and permeate were realized as described e.g., in (Schuhmann et al., 2019; Schork et al., 2021). In-situ MRI experiments were performed to get time and spatially resolved insight into filtration and backwashing properties via structural imaging and flow imaging. All experiments were performed on an NMR Bruker Avance III 200 MHz spectrometer equipped with the software Paravision 6. An actively shielded Micro2.5 gradient provided the gradient pulses for three-dimensional spatial resolution, while a MIC-WB40 birdcage was used with an inner diameter of 20 mm as radio-frequency coil. The raw data were processed within self-written MATLAB scripts to obtain the MR images of the 19-channel membranes as well as the radial distributions of MR intensity as a function of filtration time. The module was constructed such that it could be repositioned in the magnet/birdcage during the processes.

Diverse pulse sequences are known in MRI (Callaghan, 1991; Kimmich, 1997), three of them were used to study the new membrane and were established in previous investigations (Schuhmann et al., 2019; Schork et al., 2021): Shortly, image contrast is created by NMR relaxation times in the case of the used RARE and FLASH sequences. These relaxation times are often specific for the materials under investigation. Depending on the chosen pulse sequence parameters, mainly repetition time T _R and echo time τ_E, the MRI signal of a given volume element is weighted accordingly. Thereby and roughly, T _R determines the weighting by the longitudinal relaxation time T ₁, whereas τ_E determines the weighting by the transverse relaxation time T ₂.

The advantage of “rapid acquisition with relaxation enhancement” RARE is founded in its lower sensitivity towards magnetic susceptibility and its relative stability against image artifacts when compared to gradient echo-based sequences like “fast low angle shot” FLASH or “echo planar imaging” EPI, while RARE requires often larger measurement times. As the magnetic susceptibility differences between feed, permeate, and the polymeric material of the membrane are relatively small in the present case, the images especially during filtration were acquired using a FLASH sequence based on gradient refocusing. The flow fields were measured by flow_map, which allows the measurement of relative velocities in the channels in all spatial directions, i.e, v _x, v _y, v _z, while knowing about the limitations in the actual implementation. Flow_map is also based on gradient refocusing, therefore sensitive to T ₂*, which is an effective transverse relaxation rate taking into account the locally slightly different magnetic fields in the sample under investigation. The essential experimental MRI parameters are summarized in Table 1.

Filtration was performed on a fiber with a length of about 38 cm, while the observation by MRI was realized at z = 32.5 cm (Figures 3, 4). The filtration mode was in-out in dead-end/direct flow which allowed the detailed observation of deposit formation in each of the 19 channels.

With increasing filtration time, the deposit at z = 32.5 cm forms on the inner surface of the feed channels. All 19 channels qualitatively show a similar behavior as a function of filtration time: the black rings increase in effective thickness, while the structure of the membrane does not change. The air-filled areas which were observed also in the experiments with pure water are visible in all the measured images. Permeate is hardly visible in these voids, they appear to be filled with water only in the case of reverse pressure difference.

The MR images were processed to quantify the kinetics of deposit formation exemplarily in three channels, one in each ring (not shown). Each of the three channels was regarded as an area in which the signal intensity was averaged over the angle, see also (Schuhmann et al., 2018; Schork et al., 2019; Schuhmann et al., 2019; Schork et al., 2020), leading to a radial profile of the MR intensity as a function of the distance to the center, the radius coordinate r′ of the channel under investigation (Figure 1). For an easy comparison, the radial distance r′ was normalized to the radius of the feed channels r ₀. The thus calculated intensity profiles I(r′/r ₀) of the center channel and of the channel in the inner ring were modelled by a sigmoidal Boltzmann function, which reveals the tangential slope m, the turning point x ₀ apart from intensities as fit parameters: I t ,   r ′ r 0 − I t = 0 ,   r ′ r 0 I r e f ,   H 2 O t = 0 ,   r ′ r 0 = 1 − 1 1 + exp ⁡ ⁡ m r ′ r 0 − x 0

It was also noted that the radial dependence of the MR intensity in the channel located in the outer ring followed an exponential behavior which is indicative for slower deposit formation and gelling.

Especially x ₀ reveals the progress of deposit formation and can be regarded as a measure for the deposit height. It allows the quantification of the deposit formation kinetics at the given temperature, pressure, and feed composition. x ₀ of all three channels can be described by exponential decay functions (Figure 5, on the right) of filtration time t, while the time parameter amounts to 31 min for the center channel, 33 min for the ring at r = 1.1 mm and to 90 min for the channel in the outer ring at r = 2.2 mm. Few, not very reliable data points were excluded from the fit at small t in case of r = 1.1 mm.

Apart from deposit formation, the cleanability of the feed channels and especially of their active layer is essential for an efficient and long-term use. Experiments were therefore performed which show these properties in the present case of the alginate gel deposit (Figure 6).

One procedure to remove deposits is backwashing. It was applied to the 19-channel membrane after filtration and flow-through. An MR image was acquired at z = 27.7 cm, which reveals that the gel layer was removed in all but four channels (Figure 7). These channels show the diversity of residues: a complete but swollen gel layer was observed as well as disrupted parts and partial washing-out. The majority of the channels, however, is cleaned up completely from the perspective of the MRI with the used spatial resolution of 70 µm. It is worth to note that the MR-intense circular regions at the surface of the channels disappeared, and the polymer shows a uniform MR intensity. The dark regions as well disappeared in the image. These facts indicate that the backwashing with p _TMP = .5 bar leads to structural changes and water redistribution in the fiber.