You are on page 1of 14

Journal of Membrane Science 585 (2019) 67–80

Contents lists available at ScienceDirect

Journal of Membrane Science


journal homepage: www.elsevier.com/locate/memsci

Comparative study of nanofiltration membrane characterization devices of T


different dimension and configuration (cross flow and dead end)
Alessandra Imbrogno∗, Andrea I. Schäfer
Membrane Technology Department, Institute of Functional Interfaces (IFG-MT), Karlsruhe Institute of Technology (KIT), Hermann-von-Helmholtz-Platz 1, 76344
Eggenstein-Leopoldshafen, Germany

ABSTRACT

Comparison of nanofiltration characterization data from literature is challenging due to different hydrodynamics and system designs, which affect membrane
retention. In this study, stirred cell (SC), micro and macro cross flow systems (micro and macro CF) with different configuration were used to measure salt and
organic tracer retention. Minimal concentration polarization conditions were applied in order to:
1) evaluate comparability of the systems for characterization of membrane pore radius and molecular-weight-cut-off,
2) understand the impact of system configuration and operation on mass transfer,
3) compare salt retention at laboratory scale with the retention of spiral wound module.
Results indicated that system dimension was the most important parameter to affect the mass transfer and the concentration of both salt and organic tracers on the
membrane surface (Cm) in the macro CF system. Indeed, the higher channel length to width ratio of macro CF was related to reduced mass transfer and higher Cm.
However, a comparability of the three systems was observed by operating at low flux (below 80 L/m2h) and higher cross flow velocity (above 0.4 m/s), where the
lowest concentration polarization conditions were maintained. This is valid taking into account the variation of hydrodynamics (e.g. relation of Sherwood number
and Reynolds range), which is intrinsically related to the different operation modes of dead end and cross flow. This study gives a novel contribution to improve the
accuracy of membrane characterization methodologies and to identify suitable operative conditions for testing new membrane materials at small scale.

1. Introduction Mass transfer (Km) can be estimated experimentally either by the


velocity variation or osmotic pressure method [6] or can be calculated
Characterization of membrane retention properties often requires from the dimensionless Sherwood number (Sh), as represented in Eq.
the determination of molecular weight cut off (MWCO) that can be used (1) [6,7]:
to estimate pore size. MWCO is determined experimentally by mea- Km D h
suring organic tracer retention with molecules of different molecular Sh = = aRebScc
D (1)
weights that are assumed to do not interact with each other and the
membrane over a range of different operating conditions (such as where Dh is the hydraulic diameter (m), D is the diffusion coefficient
pressure, cross flow velocity or stirrer speed, feed composition) [1,2]. (m2/s), Re and Sc are the dimensionless Reynolds and Schmidt number
However, solute retention is affected by the hydrodynamics of the (Sc = /D), respectively; a, b, c are adjustable dimensionless para-
system (especially when diffusion is involved), and hence system de- meters, that change based on system geometry and laminar or turbulent
sign. This makes the comparability of different system configurations at conditions [8]. There are no unique relations for Sh number and the
laboratory scale with the realistic performance of cross flow system mass transfer, due to changing of operative conditions and system
applied in industrial plant (e.g. spiral wound module, SWM) difficult. configuration. Therefore, the choice of a relation that describes accu-
Since mass transport through a membrane is dependent on the feed rately the system is a challenging task [6]. An overview of the most
flow, a comprehensive understanding of the hydrodynamics in various used Sh equations defined for laminar and turbulent conditions as well
design configurations is required [3]. as stirred cell and cross flow geometries is reported in Table 1. Solute
The stagnant film model is used commonly to describe bulk mass retention is affected by a combined effect of intrinsic membrane
transfer in pressure-driven membrane systems, like nanofiltration (NF) properties (e.g. pore size, membrane porosity, membrane thickness,
[4,5]. A representation of the building-up of a film boundary layer in tortuosity) and concentration polarization [9–11]. The configuration of
both dead end and cross flow NF is reported in Fig. 1 (A and B). the filtration device as well as operational parameters need to be di-
rected towards decreasing the boundary layer thickness and increasing


Corresponding author.
E-mail address: alessandra.imbrogno@kit.edu (A. Imbrogno).

https://doi.org/10.1016/j.memsci.2019.04.035
Received 24 October 2018; Received in revised form 13 April 2019; Accepted 17 April 2019
Available online 25 April 2019
0376-7388/ © 2019 Elsevier B.V. All rights reserved.
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Fig. 1. Schematic representation of boundary layer thickness and velocity profile in dead end (A) and cross flow (B) filtration system.

Table 1
Overview of equations for calculation of Reynolds (Re) and Sherwood (Sh) numbers in cross flow and stirred cell configuration.
Dimensionless number Regime Equation Condition Ref.

Cross flow
Sherwood Laminar (L > L*)a [6,7]
(
Sh=1.86 ReSc )
Dh 0.33
L
a
(L < L*) [7,8]
Sh=0.664Re0.5 Sc ( )
Dh 0.33
L

( )
0.33 Dh
Sh=1.62 ReSc
Dh (100 < ReSc < 5000)b
L L
Turbulent Sh = 0.023Re0.875Sc0.25 1 Sc 1000 [8,14]
Sh = 0.0096Re0.91Sc0.35 Sc > 1000
Reynolds Re =
UDh [6,18]

Uds
Re = (SWM)
Stirred cell
Sherwood Laminar Sh = 0.285Re0.567Sc0.33 Re < 3·104 [14–16]
Turbulent Sh = 0.044Re0.75Sc0.33 (32·103 < Re > 82·103) [15,16]
Reynolds r2 [15,16]
Re =

SWM= Spiral Wound Module. Symbol explanation is reported in list of symbols. Turbulent condition is defined for Re > 104 and laminar condition for Re < 104.
a
Velocity profile is completely developed at a distance from the channel inlet (L*), the concentration profile is developing along the length of the channel [8].
b
Parabolic velocity profile is fully developed at the channel entrance whereas concentration profile is developing along the length of the channel [8].

Km, that is reducing the concentration polarization [8]. filtration channel as represented in the boundary layer thickness in
In cross flow configuration, a velocity profile gradient is developed Fig. 1B. Previous studies reported that a decrease of the layer thickness
along the channel length as reported in Fig. 1B. At the entrance of the (hence an increase of Km) can be obtained by increasing the feed flow
channel, a velocity boundary layer is present, where the velocity profile rate (that is the axial velocity) or reducing the pressure and hence flux
is created, and a fully developed region follows, where the velocity [10,13].
profile is established and remains unchanged. Local variation of Km In dead end configuration, lower rotational speeds correspond to
occurs along the channel length and a concentration layer grows gra- lower mass transfer, thus lower rejection due to the increased con-
dually. As a consequence, a local variation of Km along the channel centration at the membrane surface [11,14]. Colton et al. [15,16] car-
length occurs, as reported by Bhattacharya et al. [12], and a mean Km ried out comprehensive studies on Km in an agitated vessel and an
value can be estimated over the entire channel. Considering this local important, yet unsurprising, conclusion of these studies was that Km is
variation of Km, the concentration layer grows gradually along the not distributed uniformly over the bottom of the cell. Indeed, Km

68
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

decreased continuously as the distance from the sidewall increased to Table 2. The dimension of the cross flow membrane cells is designed
reach nearly zero at the center of the device. As a consequence, the with different channel length (L) to width (W) ratio; the ratio L/W is 2
polarization boundary layer grows from the sidewall towards the cen- and 8 in micro CF and macro CF, respectively. The height of the rec-
tral axis, as observed from an experimental study carried out by Zydney tangular feed channels of micro CF and macro CF (0.7 mm) corresponds
et al. [17], that is represented in Fig. 1A. The stirrer or paddle shape as to a realistic value of a thickness spacer channel in SWM (range be-
well as the distance from the membrane will further influence the mass tween 0.7 and 0.9 mm) [14,30].
transfer due to different mixing efficiency and accumulation of mole- The systems are characterized by different design and functional-
cules in the boundary layer [11]. This indicates that different type of ities as described below.
stirred cells are bound to produce variable membrane characterization Macro cross flow system (macro CF): The system was designed to
results. operate in a cross-flow mode with the recirculation of retentate and
The apparent pore radius of NF membranes is estimated by applying permeate in the feed tank (schematic in Fig. 2A). The cross flow
the steric hindrance pore model (SHP) [19–21]. This is considered to be membrane cell was purchased from MMS Membrane Systems (Swit-
the simplest model for membrane pore radius characterization due to zerland). A diaphragm pump Hydra-Cell P200 (P200MSXSSA05C,
the assumption of the membrane as a porous support of straight uni- Verder, Germany) coupled with a DC-motor (VP3428D, Baldor, Ger-
formly distributed cylindrical pores, in which uncharged solute spheres many) was used. A pulsation damper filled with nitrogen gas
are transported through the pores. However, this assumption is far to be (92100.5581, Speck Triplex Pumpen, Germany) was installed in the
correct for most NF membranes, due to the molecular scale dimension system in order to obtain stable pressure in a range 0–20 bar. Feed
and tortuous structure of the pore as well as the transport mechanism temperature was controlled by a thermostatic circulator (LKB 2219
[22,23]. In addition, solute-solute interactions, solute pore interactions MultiTemp II, Bromma, Germany). Feed and retentate pressure was
and hydration of the molecule are not considered in the model, even measured by using a pressure sensor (Type A-10, WIKA, Germany). A
though they are involved in the transport through NF membrane pressure relief valve (SS-4R3A, Swagelok, Germany) was installed for
[24,25]. For this reason, various extensions of the basic hydrodynamic safety, in case the pressure exceeded accidently 20 bar. Feed flow was
model are progressing with theories including pore shapes, electrostatic measured by using a turbine wheel low volume flow meter (PEL-L045-
interactions, non-spherical solutes and hydration [26–29]. LMXF, Kobold Messring, Germany). The mass of permeate was mea-
The aim of this study was to determine the impact of different sured by using a balance (Ohaus AV2102, Germany) and converted to
system geometry and operation mode on retention of salts and organic volume by using water density (998 and 997 kg/m3 for a range of
solutes by NF membranes, in order to evaluate the comparability of the temperature 20–25 °C) [31]. Retentate and permeate were recirculated
systems for characterization of membrane pore size, MWCO and salt to the feed tank and the permeate sample was collected after 30 min.
retention. Three systems, stirred cell (SC), micro and macro cross flow LabVIEW software (version 2014, NI, Germany) was used to set op-
systems (micro CF and macro CF) with a different configuration (dead erative conditions (pump flow rate, pressure and temperature), to open
end and cross flow) and dimension were used for the filtration tests. The the pressure control valve (Badger RC200, Germany) in the retentate
different dimension is referred to the membrane area (e.g. in order of side as well as to monitor the mass of permeate. For salt filtration ex-
membrane area micro CF < SC < macro CF) and the different periment conductivity of feed, permeate and retentate was measured
channel length (L) to width (W) ratio in macro CF and micro CF sys- after 30 min of filtration by using a conductivity meter (pH/cond3320,
tems, (L/W = 8 and 2, respectively). Salt retention and organic solutes WTW, Germany).
retention were investigated as a function of permeate flux, cross flow Dead end stirred cell (SC): The stainless steel cell (schematic in
velocity and stirrer speed in order to identify suitable operative con- Fig. 2B) was designed in-house and built at the Technik-Haus (TEC-KIT,
ditions for comparability. This study makes a novel contribution to- Germany) to operate in dead end mode [32]. The pressure was applied
wards the accuracy of membrane characterization methodologies as from the top by using synthetic compressed air (20.5% O2 in N2, Al-
well as to identify key operative conditions to take into account for the phagaz 1 Luft, Germany). The mass of permeate was collected at the
determination of MWCO and pore radius of a membrane with filtration bottom of the cell and it was measured by using a balance (Ohaus
devices. This is of great interest for the researchers working in the AV2102, Germany) and converted to volume as reported above. Tem-
manufacturing of new membrane materials, which requires a compre- perature and feed pressure were measured within the cell by using a
hensive understanding of membrane properties without artifacts related thermocouple (TJ2-CPSS-M60U-250-SB, Omega, UK) and a pressure
to the characterization methodology. sensor (PA-21Y, Keller, Germany) connected to the top of the cell. A
pressure relief valve (HPRVS4A-V-KG-300, Parker, Germany) was in-
2. Material and methods stalled for safety, in case the pressure exceeded accidently 10 bar. Mass
of permeate, feed pressure and temperature were monitored by using
2.1. Membrane filtration equipment LabVIEW software (version 2014, NI, Germany). The bottom of the cell
included a stainless steel flow channel covered by a 1.5 mm porous
The filtration experiments were performed with 3 different systems: stainless steel disc, which was used as mechanical support for the
membrane. A magnetic stirrer (15254AM, Millipore, UK) was in-
• Macro CF (macro cross flow system) with a channel length of 19 cm, corporated into the cell in order to reduce concentration polarization
height of 0.7 mm and an active membrane area of 47.5 cm2. The during the experiment. A magnetic stirrer table (8400SCPKIT, Milli-
system can be operated in a pressure range 0–20 bar and flow rate of pore, Germany) was used for stirring. For salt filtration experiment
the pump 0–6300 mL/min. conductivity of permeate was measured in line in a stainless steel flow
• SC (dead end stirred cell) with a capacity of 900 mL and an internal cell built-in-house connected to a conductivity meter (pH/cond3320,
diameter of 7 cm, corresponding to an active membrane area of WTW, Germany). Conductivity data were registered in Excel by using
38.5 cm2. The system can be operated in a pressure range 0–10 bar. MultiLab importer (version v1.09, WTW, Germany).
• Micro CF (micro crossflow system) with a channel length of 2 cm, Micro cross flow system (micro CF): The system (schematic in
height of 0.7 mm and an active membrane area of 2.0 cm2. The Fig. 2C) was used in this study in a cross flow mode. The membrane cell
system can be operated in a pressure range 0–20 bar and flow rate of was designed in-house and built at the Institute for Micro Process En-
the pump 0–500 mL/min. gineering (IMVT-KIT, Germany). The scale dimension of the active
membrane area and the channel length was reduced by 24 times and 10
A schematic representation of each system is reported in Fig. 2 and times, respectively, compared to the macro CF system. The down-scale
detailed information about the system characteristics is reported in approach of the membrane cell allowed to test retention properties of

69
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Fig. 2. System configuration and membrane holder dimensions of micro CF, macro CF, SC.

Table 2 as reported above. For salt filtration experiment, the conductivity of


Membrane module characteristics of SC, macro CF and micro CF. permeate was measured in line using a conductivity sensor on the
Membrane module dimension Unit SC Micro CF Macro CF permeate side (model ET131, eDAQ, Germany). LabVIEW software
(version 2014, NI, Germany) was used to set operative conditions
Dh m 0.07 0.0013 0.0014 (pressure and pump flow rate) and to monitor temperature and mass of
Am m2 0.00384 0.0002 0.00475
permeate.
Channel height m 0.0007 0.0007
Channel length m 0.02 0.19
Channel width m 0.01 0.025
2.2. Nanofiltration membranes

small membrane samples at operative conditions similar to SWM (e.g. Two commercial NF membranes, NF270 and NF90 (provided as flat
pressure, cross-flow velocity, Reynolds number) as reported in details in sheet samples by Dow Chemical Company, Germany) were used in this
paragraph 2.4. The feed solution was pumped along the circuit by using study. NF270 and NF90 consist of semi-aromatic piperazine-based
an HPLC pump (Blue Shadow 80P, Knauer, Germany), that offered the polyamide and fully aromatic polyamide active layers, respectively
advantage to work at constant, stable and low pulsated flow-rate. The [33]. NF270 and NF90 are loose and tight NF membranes, respectively.
LabVIEW driver for the pump was developed in-house. Feed flow rate Different characteristics concerning pore radius, molecular weight cut-
was set by using LabVIEW software (version 2014, NI, Germany). off (MWCO), salt retention and permeability are reported in Table 3.
Pressure sensors in the feed and retentate side (model Type A-10, WIKA The range of pore radius mentioned in the table includes theoretical
Alexander Wiegand SE & Co. KG, Germany) were used to measure the values reported in literature and estimated by applying the SHP model.
pressure within the cell. The pressure was adjusted by using a pressure In this context, the pore radius terminology should not be considered
control valve (Swagelok SS-RL3S4-EP, Germany) displaced at the re- with a physical meaning, but as a hypothetical value based on the
tentate side. The temperature was measured by using a thermocouple simplification of the NF membrane structure with uniform and cylind-
measurement device (NI USB-TC01, NI, Germany). In the permeate rical pores. This is not realistic due to the atomic scale dimension and
side, a 16 port switching valve (Azura V2.1S, Knauer, Germany) was tortuous pore structure of the real NF membrane. NF90 and NF270
installed in order to collect numerous and variable volumes of samples were selected as representative case of tight and loose NF membranes,
automatically via drop count or weight, by controlling the switching respectively, providing a wide range of permeability and retention
valve with LabVIEW software. The mass of permeate was measured by characteristics for the evaluation of characterization protocols.
using a balance (Ohaus AX/622E, Germany) and converted to volume

Table 3
Characteristics of NF90 and NF270 membranes used for characterization.
Membrane MWCOa (Da) IEP Lp (L/hm2bar) Salt retention Pore radius (nm)a

NF90 (tight) 90–180 [34–37] 4.5–5.5 [38,39] 8±2 MgSO4 ≥97%b 0.31–0.38 [35,36,40]
NaCl [35–37,41]
85–95%
NF270 (loose) 150–340 [34–36,42] 3.4–4 [38,43] 14 ± 2 MgSO4 ≥97%b 0.36–0.44 [35,36,44]
NaCl [35–37,41]
40–60%

Isoelectric point (IEP), Permeability (Lp).


a
MWCO was determined by filtration of organic solutes (poly-ethylene-glycol and sugars), pore radius was estimated by applying the SHP model.
b
From Dow chemical data sheet of SWM 4040, the retention was measured using a feed concentration of 2 g/L, 4.8 bar, 25 °C and 15% recovery [45,46].

70
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Table 4
Chemical properties of salts and organic tracers.
Characteristics Sodium Chloride Magnesium Sulfate 1,4 dioxane (%C = 54) Meso-erythritol (%C = 39) D-xylose (%C = 40) D-glucose (%C = 40)

Formula NaCl MgSO4 C4H8O2 C4H10O4 C5H10O5 C6H12O6


MW (g/ mol) 58 120 88 120 150 180
Πfeed at 22°C (bar)a 1.67 0.79 0.01 0.01 0.01 0.01
D (‧10−10 m2/s)b 16.1 8.5 9.8 8.3 7.4 6.7
Stokes radius (nm)c 0.13 0.25 0.23 0.26 0.29 0.32

a b
Molecular weight (MW), Feed osmotic pressure (Πfeed), Diffusion coefficient (D); Calculated using the Van't Off equation [48], Wilke and Chang equation [49],
c
Stokes−Einstein equation [26].

2.3. Chemicals Table 5


Operative and hydrodynamic condition in SC, micro CF and macro CF system.
Sodium chloride (NaCl, Merck Millipore, purity ≥ 99.9%, Germany) Operative condition Unit SC Micro CF Macro CF
and magnesium sulfate (MgSO4 anhydrous, Alfa Aesar, purity ≥ 99.5%,
Germany) were used at 2 g/L to measure the salt retention of NF Salt concentration g/L 2 2 2
membranes in the three systems. Organic tracers used to characterize Organic tracer mgC/L 20 20 20
concentration
the pore radius and MWCO of NF membranes were 1,4 dioxane Vf L 0.4 1 4
(purity ≥ 99.5%), D-xylose (purity ≥ 98%), purchased from Merck Transmembrane bar 4.8–9.6 3.7–10 3.7–10
Millipore (Germany), meso-erythritol (purity ≥99%) and D-glucose or pressure
dextrose (purity ≥ 99%), purchased from Sigma Aldrich (Germany). Stirrer speed rpm 100–400
Cross flow velocity m/s 0.4–1.5 0.1–0.6 0.1–0.8
These were chosen due to the low molecular weight, uncharged and
Temperature °C 23.1 ( ± 0.5) 23.5 ( ± 0.8) 20.2 ( ± 0.3)
non-adsorbing properties [47]. A feed solution containing 20 mgC/L of Feed flow rate mL/min 30–320 108–880
each organic tracer was used for the filtration experiment. The type of Recovery % 15 0.05–0.5 0.6–4.5
tracers and the concentration were selected according to the method Hydrodynamic condition
published by Nghiem et al. [47]. MilliQ water (MilliQ A+ system, Reynolds number 104-5·104 100–800 100–1000
Sherwood number 400–1700 30–80 15–40
Millipore, Germany) was used to prepare the feed solution. Molecular Schmidt number 600–1500 600–1500 600–1500
weight, the osmotic pressure of the feed solution, diffusion coefficient
and Stokes radius of salts and organic tracers are summarized in
Table 4. as reported in literature (3–20 bar [52] and 0.3-0-4 m/s [53,54]). The
cross flow velocities applied in macro CF and micro CF correspond to
2.4. Filtration protocols Reynolds number (100–1000) comparable with Reynolds applied in
numerical studies of spacer-filled channels of SWM (50–1000) [18,53].
A new membrane coupon was used for each experiment when the The range of stirrer speed (0–400 rpm) applied in SC system corre-
operative variable was changed (e.g. pressure, cross flow velocity or sponds to Reynolds (104–5·104) and velocity (0.4–1.5 m/s), which was
stirrer speed, system device); the coupon was soaked in NaCl 10 mM for higher compared to macro CF and micro CF. In SC the filtration was
1 h prior to use, then compacted by filtering MilliQ water at 9.6 bar for carried out until a recovery of 15% was reached (corresponding to a
1 h. The pressure used for the compaction was in accordance with the permeate volume of 60 mL). When organic tracers were filtered, 3
highest pressure used in the experiment (9.6–10 bar). Soaking in NaCl permeate samples of 20 mL were collected during each test for TOC
enhances opening of the pores and swelling of the active layer due to analysis to measure organic solute concentration. When the salt solu-
the interaction of electrolytes with the polyamide layer [50]. The pure tion was filtered, the permeate conductivity was measured in-line
water permeability (Lp, units: L/h m2bar) was measured at steady-state during filtration. In micro CF one permeate sample of 10 mL was col-
for 30 min, before the filtration of solutes. Filtration conditions (e.g. lected for glucose and xylose and 5 mL for erythritol and dioxane after
concentration of solutes, feed volume, the range of pressure, cross flow pressure was stable. 10 mL was the minimum volume required for TOC
velocity and stirrer speed, temperature) as well as hydrodynamic con- analysis when the sample was not diluted. When the salt solution was
ditions (Reynolds, Sherwood and Schmidt number) are reported for filtered, the conductivity was measured in-line during filtration in SC
each system in Table 5. The experiments were carried out at a tem- system and micro CF. In macro CF, one permeate sample (volume
perature variable between 19.9 °C and 23.9 °C, which corresponds to a 10 mL) was collected after pressure was stable in both cases of organic
variation of the dynamic water viscosity between 1.00 mPa‧s and tracer and salt solution. Relatively low recovery in a range of 0.6–4.5%
0.87 mPa‧s [51]. The actual operating temperature at the starting and and 0.05–0.5% in macro CF and micro CF, respectively, were chosen in
ending of the salt filtration experiments in the three systems is reported order to evaluate salt and organic tracer retention at minimal con-
in Table S11. The temperature variation was ≤2.5 °C during the fil- centration polarization conditions. These conditions allowed to keep
tration experiment resulting in variation of 15% of the permeation flux the concentrate concentration (calculated from the mass balance re-
(for the worst condition of 2.5 °C of temperature variation). This var- ported in paragraph 2 of SI) at values very similar to the feed con-
iation of the permeation flux was not considered significant to affect the centration. Operating variables (e.g. permeate flow rate and recovery
accuracy of the data. applied in each system at each pressure and cross flow velocity) are
Filtration of water containing salts or organic tracer was performed reported in more details in Table S1.
after permeability test, using a feed volume of 400 mL, 1L and 4 L for The observed retention (Robs) was calculated by using Eq. (2):
SC, micro CF and macro CF, respectively. Filtration was carried out at
different trans-membrane pressure (4.8–9.6 bar in SC, 3.7–10 bar in Cp,i
R obs (%) = 1 100
macro CF and micro CF), different cross flow velocity (0.1–0.8 m/s in Cb (2)
macro CF and micro CF) and different stirrer speed in SC
(100–400 rpm). In SC the stirrer speed was constant at 400 rpm when where Cp is the salt/organic concentration (g/L for salt and mgC/L for
the pressure was varied. The range of pressure and velocity used in organic, respectively) in the permeate sample i and Cb is the salt/or-
micro CF and macro CF were comparable with values applied in SWM ganic concentration (g/L for salt and mgC/L for organic, respectively)

71
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

in the bulk solution. For the stirred cell system, Cb was calculated as a solutes is different compared to the bulk solution [55], as represented in
function of the permeate volume according to Eq. (3): Fig. S8. The presence of this concentration polarization layer involves
the definition of two retentions: 1) observed retention (Robs, defined in
CFD VFD C p,i Vp,i
Cb = Eq. (2)) is the experimental retention calculated by taking into account
VR (3) the feed and permeate concentration, and 2) real retention (Rreal) is a
where CFD is the initial salt (g/L) or TOC concentration (mgC/L) in the theoretical value calculated by using Eq. (4), where the concentration
feed solution, Cp,i is salt (g/L) or TOC concentration (mgC/L) in the on the membrane surface (Cm) is considered [5]:
permeate sample i, VR (L) is the retentate volume (L), VFD (L) is the Cp, i
initial feed volume, and Vp,i (L) is the permeate volume of sample i. Rreal (%) = 1 100
Cm (4)

2.5. TOC analysis Since this concentration cannot be measured directly, the con-
centration polarization film model was used to correlate the observed
Total organic carbon (TOC) concentration of organic tracers was and real retention as reported in Eq. (5) [5]:
measured in feed, permeate and retentate samples with a GE Sievers M9
1 R obs 1 Rreal Jv
TOC Analyser (GE Analytical Instruments, UK). Feed and retentate ln = ln + )
samples were diluted 2 times to ensure that concentrations were within
R obs Rreal Km (5)
the range of the calibration curve (0–10 mgC/L). Permeate samples where Jv is the volumetric flux (m/s) and km is the mass transfer
containing dioxane and meso-erythritol were diluted twice to be within coefficient (m/s). Jv was measured experimentally during filtration and
the range of the calibration curve. Permeate samples containing glucose Km was calculated using Sherwood (Sh) correlation for stirred cell, Eq.
and xylose were not diluted due to the high retention. The samples and (6) [7] and cross flow, Eq. (7) [6]:
the calibration were analyzed using an acid (H2SO4) flow rate of 0.5 μL/
Sh = 0.044Re0.75Sc0.33 for 32 103 < Re > 82 103 (6)
min, oxidizer (ammonium persulfate) flow rate of 1 μL/min and in-
organic carbon off. Dh 0.33
Sh=1.86 ReSc for Re < 10 4 and ( L> L )
L (7)
2.6. Conductivity analysis
where Re is Reynolds number, Sc is Schmidt number, Dh is the hydraulic
diameter (m), L is the channel length (m), L* (m) is the distance from
The conductivity of salt solution was measured as described above
the channel inlet where the velocity profile is completely developed (=
for each system. The salt concentration was calculated from con-
0.029‧Dh‧Re). Eqs. (6) and (7) were selected based on the range of
ductivity measurements as reported in paragraph 1 in SI.
Reynolds number (104 < Re < 5·104 for the SC system and Re < 104
for the micro CF and macro CF systems) as well as the channel length
2.7. Error estimation (L) of micro CF and macro CF systems. In this context, Eqs. (6) and (7)
should be considered valid taking into account the operating variables
The error calculation is reported in paragraph 8 in SI. (range of Reynolds) and the channel design as boundary conditions. Cm
(mgC/L) was calculated using Eq. (8) that was obtained by rearranging
2.8. Calculation of real retention and determination of pore radius by SHP Eq. (5):
model
Jv
Cm = Cb (1–R obs) + R obse km
In the boundary layer of the membrane surface, the concentration of (8)

Fig. 3. Robs of NaCl and MgSO4 by NF270


and NF90 as a function of Jv. 2 g/L,
3.7–10 bar, 400 rpm (SC) and cross-flow
velocity 0.4 m/s (micro CF, macro CF),
temperature 23.1 ( ± 0.5 °C, SC), 20.2
( ± 0.3 °C, macro CF), 23.5 ( ± 0.8 °C, micro
CF). SWM star is the salt retention reported
in the datasheets of Dow Chemical [45,46].
Points in brackets are repeated experiments.

72
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Characterization of membrane pore radius and MWCO was per- Cp Kc


Rreal = 1 =1
formed by measuring Robs of organic tracers with different molecular Cm 1 exp( Pe)(1 Kc) (9)
weight as a function of permeate flux. A selectivity curve was obtained
by plotting Rreal of the tracers versus the molecular weight. The MWCO where Kc is the convective hindrance factor, is the steric partition
was obtained by the selectivity curve as the molecular weight of the coefficient and Pe is the Peclet number. Pe and are defined by Eq. (10)
tracer that is 90% retained by the membrane [56]. Rreal was correlated and Eq. (11) respectively [20,58]:
with the pore radius by applying the steric hindrance pore model (SHP) K c Jv L
[21,26]. SHP model was applied in NF to explain the relation between Pe =
Kd D (10)
driving force and solute flux and it was mainly correlated to the re-
tention of uncharged solutes, like sugars [28]. Even though SHP model =(1 )2 (11)
presents some drawbacks related to the unrealistic assumption of the
membrane as support of straight and uniform cylindrical pores (as where L (m) is the pore length (membrane thickness), is the mem-
discussed in the Introduction), it was used in this study for the esti- brane porosity, Kd is the solute hindrance factor for diffusion, D (m2/s)
mation of the pore radius. This is because of its simplicity and well- is the diffusion coefficient, is the ratio of the Stokes solute radius to
established model for characterization of NF membrane by filtration of pore radius (= rs/rp). To estimate the pore radius of NF membranes the
uncharged solutes [35,36,40,41,44,57]. The estimated pore radius re- real retention (measured at different Jv) was fitted with Eq. (9) by using
ported in this study corresponds to a hypothetical value of NF mem- Solver function (Microsoft Excel) in order to obtain the parameters Pe,
brane with uniform and cylindrical pores. Therefore, it is a simplifica- Kc and . The pore radius was calculated subsequently from for each
tion of the real NF membrane structure and the value should not be solute and membrane.
interpreted as an effective dimension with direct physical meaning. In
SHP model, the real retention is correlated with the pore radius by Eq. 3. Results and discussion
(9) [58]:
3.1. Comparison of salt retention in laboratory filtration systems and in
spiral wound module (SWM)
Table 6
Km of NaCl and MgSO4 in micro CF, macro CF and SC. NaCl and MgSO4 retention were investigated as a function of
System Km (·10−5 a
m/ s) permeate flux in stirrer cell (SC) and cross flow systems (micro CF and
macro CF) in order to compare with the published retention of SWM.
NaCl MgSO4 The permeate flux was increased by increasing the transmembrane
pressure (3.7–10 bar) in a range similar to realistic conditions applied
Micro CF 6.1 4.0
SC 5.7 3.8
for SWM [52]. Robs of NaCl and MgSO4 by NF90 and NF270 as a
Macro CF 2.8 1.9 function of permeate flux and the different systems is reported in Fig. 3.
Robs of MgSO4 measured in the three systems was between 91 and
a
Km was calculated for DSC using Eq. (6), stirrer speed 400 rpm, for macro 98%, which was comparable with the retention reported for SWM
CF and micro CF Eq. (7), cross flow velocity 0.4 m/s. ( 97%) at the same pressure and salt concentration (4.8 bar, 2 g/L

Fig. 4. Robs of NaCl and MgSO4 by NF90 and NF270 as a function of cross flow velocity in micro CF, macro CF and SC. 2 g/L, 4.8 bar, temperature 20.2 ( ± 0.3 °C,
macro CF), 22.5 ( ± 1 °C, micro CF), 23.1 ( ± 0.5 °C, SC).

73
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Table 7 consequence, the permeate solute concentration is lower. This was


Km of NaCl and MgSO4 as a function of cross flow velocity in micro CF, macro more pronounced for monovalent salts (NaCl) because of the higher
CF and SC (calculated using Eq. (7) for micro CF and macro CF and Eq. (6) for diffusion coefficient and no size exclusion compared with divalent salt
SC). (MgSO4). Other studies showed that multivalent ions (e.g. sulfate ions)
Micro CF Macro CF SC are retained in NF mainly by size exclusion and charge repulsion
compared with monovalent ions (e.g. chloride ions) which are retained
−5 −5
Vcross- Km (·10 m/ s) Vcross- Km (·10 m/ s) Vcross- Km (·10−5 m/ s) less [61,62].
flow
(m/s) NaCl MgSO4
flow
(m/s) NaCl MgSO4
flow
(m/s) NaCl MgSO4
Robs of NaCl measured in the SC system as well as in the macro CF
system was lower compared with the retention reported for the SWM at
0.07 3.5 2.3 0.11 1.9 1.2 0.37 2.0 1.3 the same transmembrane pressure. In order to explain the difference in
0.15 4.5 2.9 0.31 2.6 1.7 0.73 3.4 2.2 NaCl retention, the fluid flow conditions, as well as the different di-
0.31 5.7 3.7 0.40 2.8 1.9 1.10 4.6 3.0
mension of the systems, were considered. Km values for each system and
0.44 6.3 4.1 0.58 3.2 2.1 1.28 5.2 3.4
0.79 8.9 5.8 0.81 3.6 2.3 1.47 5.8 3.7 each salt are reported in Table 6, concentration on the membrane
surface (Cm) and solute flux (Js) are reported in Fig. S2 and Fig. S4,
respectively.
MgSO4) [45,46]. No Jv dependent variation of MgSO4 retention was Even though Km in macro CF was about 2 times lower than micro CF
measured for both membranes (NF270 and NF90) in the three systems. and SC, Cm was not significantly different compared with SC and micro
NaCl retention was affected by the change of permeate flux as well CF. This was the case especially when the systems were operated at flux
as the system configuration for both membrane types. Robs of NaCl in- below 80 and 60 L/m2h for NF270 and NF90, respectively, which can
creased from 60 to 87% (NF90) and 40–69% (NF270) with the increase be attributed to the minimal concentration polarization conditions used
of the flux. The increase of retention can be attributed to the “dilution in this study. At higher flux an increase of Cm was more significant
effect” [59,60], which occurs when the permeate water flux increases especially in macro CF and for the loose membrane (NF270), where Cm
by increasing the pressure, while the ion flux remains unchanged. As a at the outlet of the cell was double compared with the concentration in
the bulk (Cb). The channel length is an important parameter that affects

Fig. 5. Rreal and Robs of organic tracers by NF90 as a function of Jv. 20 mgC/L, 3.7–10 bar, 168 mL/min, 0.4 m/s (micro CF, macro CF), 4.8–9.6 bar, 400 rpm (SC).

74
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Fig. 6. Robs and Rreal of organic tracers by NF270 as a function of Jv. 20 mgC/L, 3.7–10 bar, 168 mL/min, 0.4 m/s (micro CF, macro CF), 4.8–9.6 bar, 400 rpm (SC).
Points in brackets are repeat experiments.

Table 8 increased by keeping constant the channel length and increasing the
Km of organic tracers in micro CF, macro CF and SC (calculated using Eq. (7) for channel width and the number of feed channels [63].
micro CF and macro CF and Eq. (6) for SC). In the case of the stirred cell (SC), the mass transfer as well as the
System Km (·10−5 m/ s) concentration on the membrane surface were very similar to micro CF.
However, these parameters were calculated as average values and the
D-glucose Xylose Erythritol Dioxane local variation at the bottom of the stirred cell were not distinguished.
In a stirred cell, the mass transfer decreased continuously with the
Micro CF 3.4 3.7 3.9 4.4
SC 3.2 3.4 3.7 4.1 distance from the sidewall to nearly zero at the center of the device
Macro CF 1.6 1.7 1.8 2.0 [17]. Therefore, the flow conditions in the center of the membrane are
similar to conditions with no stirring and the build-up of a polarization
boundary layer is likely to occur in this region. This might explain the
the local variation of concentration in the filtration channel, hence the lower NaCl retention measured in the stirred cell. This hypothesis is in
ion transport through the membrane and salt rejection [10,25]. In accordance with a study published by Zydney et al. [17], where the
macro CF the proportion of the channel length to width is 4 times retention of dextran by UF membrane was compared in two stirrer cells
higher than in the micro CF. Therefore, the increase of Cm along the with different dimensions. At the same stirring condition, the cell with
channel length in macro CF is expected to be correlated with the lower largest diameter was less sensitive to changing of flux condition due to
retention. A profile of Cm and Km for NaCl along the channel length of smaller degree of concentration polarization. It should be noted that
macro CF is reported in Fig. S9. According to this profile, Km and Cm cell dimensions and stirrer shape will alter characterization perfor-
were in a range similar to the micro CF at the beginning of the channel mance and hence this must be evaluated in depth for different systems
up to a length of about 0.05 m. This result suggests that the channel before meaningful conclusions can be drawn from characterization re-
length should be maintained constant in order to keep similar hydro- sults. In addition to the different contribution of Km over the membrane
dynamic conditions when the scale of the membrane cell is increased. surface, Tansel et al. [64] claimed that the shear force related to the
This is consistent with a previous study where the membrane area was system configuration was a further parameter which might affect the

75
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

Fig. 7. Rreal and Robs of organic solutes by NF90 at different cross flow velocity in the cross flow system (20 mCg/L, 4.8 bar).

Fig. 8. Rreal and Robs of organic tracers by NF270 at different cross-flow velocity in macro CF and micro CF. 20 mCg/L, 4.8 bar.

flow conditions. The size of the hydrated ions and the mobility me- was determined in all the three system for both membranes. On the
chanism through the membrane was subsequently correlated with the contrary, the relatively low NaCl retention was affected more by
rearrangement of the hydration shell during the filtration based on the system design due to different flow conditions especially in macro CF
flow pattern in cross flow and dead end. The contribution of dehydra- and SC, when the system was operated at the similar conditions of
tion of monovalent ions (e.g. Cl−) with a weaker hydration shell in the the SWM.
transport through the pore was claimed to be minor in the cross flow
compared with the stirred cell due to different shear force. Partial de- 3.2. Impact of system dimension on salt retention and mass transfer
hydration of ions when entering in the membrane pore and under
transmembrane pressure conditions was reported to be involved in the Robs of NaCl and MgSO4 by NF270 and NF90 was measured in the
ion transport through NF membranes [65–67]. However, the effect of three systems as a function of cross flow velocity in order to evaluate
the hydrodynamics on the hydration is not fully evidenced mechan- comparability. Km and Cm were calculated in order to evaluate the
istically. extent of concentration polarization in the range of cross-flow velocity.
In summary, the reproducibility of SWM retention of divalent salt The increase of cross flow velocity from 0.1 m/s to 0.8 m/s in micro CF

76
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

and macro CF and from 0.4 to 1.5 m/s in the SC system (corresponding 3.3. Organic tracer retention as a function of permeate flux (trans-
to an increase from 100 rpm to 400 rpm) resulted in a different range of membrane pressure) for the determination of MWCO and pore radius
Re and Sh applied in the three systems. Indeed, Re and Sh were higher in
the SC system (104 < Re < 5·104, 400 < Sh < 2000) and lower in According to the method of Nghiem et al. [47], NF membrane pore
the micro CF (100 < Re < 800, 30 < Sh < 80) and macro CF sys- radius and MWCO were determined by measuring Robs of organic tra-
tems (102 < Re < 103, 15 < Sh < 40). This was related to the dif- cers as a function of permeate flux. Subsequently, Rreal for each tracer
ferent operation mode of the systems which limited the comparability was calculated using Eq. (4). Robs and Rreal of organic tracers as a
in the same range of dimensionless number. Robs of NaCl and MgSO4 by function of permeate flux by NF90 and NF270 are reported in Fig. 5 and
NF90 and NF270 measured in the three systems as a function of cross Fig. 6, respectively. The aim was to evaluate the reproducibility of
flow velocity is reported in Fig. 4. tracer retention in the three systems and the operative conditions for
In the three systems, Robs of MgSO4 and NaCl by the tight membrane comparability.
(NF90) was constant at 98% and 80%, respectively, indicating no sig- As reported in Fig. 5, overall data showed similar Robs in the three
nificant influence of cross flow velocity. The same result was observed systems for both membranes.
for the loose membrane (NF270). Robs of MgSO4 and NaCl was constant However, Km (reported in Table 8) and the trend of Cm (reported in
at 90% and 40–60%, respectively. The trend of retention was consistent Fig. S6) were clearly different in macro CF compared with the other two
with Js that is reported in Fig. S5. Indeed, no significant variation of Js systems. Indeed, Km in macro CF was 2 times lower compared with SC
was calculated when the cross flow velocity was increased in all the and micro CF. As discussed previously, the difference of channel length
three systems. between the two cross flow systems plays an important role to explain
Cm as a function of cross flow velocity in the three systems is re- the lower mass transfer measured in macro CF. Km over the channel
ported in Fig. S3. Even though no significant impact on retention was length was calculated for each tracer and it is reported in Fig. S10. Km
observed, the trend of Cm revealed a reduction of the concentration on reduced progressively by increasing the channel length and, as a con-
the membrane surface when the cross flow velocity was increased. At sequence, Cm at the outlet of the channel increased in a range 2–3 times
velocity above 0.4 m/s (which corresponds to Re > 12000 and compared with Cb as reported in Fig. S10. However, at flux below 40
rpm > 100 for SC, Re > 500 for macro CF and micro CF systems), Cm and 60 L/m2h for NF90 and NF270, respectively, Cm was comparable
reached values similar to the concentration in the bulk solution in all with Cb in all the three systems and for all the tracers. In the case of SC
the three systems, indicating negligible concentration polarization. This and micro CF, the increase of Cm was less significant and better com-
result indicated that minimal concentration polarization condition can parability among the three systems was achieved by operating at lower
be maintained in all the three systems at higher cross flow velocity flux (≤60 L/m2h).
when the concentration on the membrane surface converged to a si-
milar value of the bulk solution. The increase of cross flow rate in 3.4. Comparability of organic tracer mass transfer in micro CF and macro
macro CF counterbalanced the longer channel length which caused an CF
increase of concentration polarization along the channel. Km as a
function of cross velocity in the three systems was calculated and is Robs and Km of organic tracers were investigated as a function of cross
reported in Table 7. flow velocity in order to identify operative conditions for comparability
As discussed above, Km in macro CF was lower than that in micro CF of the cross flow systems. Results are reported in Fig. 7 and Fig. 8.
and stirred cell. In this case Km was about 3 times lower compared with Contrary to expectations, the results showed no significant increase
micro CF, even though the range of cross flow velocities were the same. of Robs independently from system and membrane type. However, Km of
In order to explain the different Km in macro CF and micro CF, the cell organic tracers increased about 1.5 times in macro CF as reported in
design was considered. Km was calculated using Eq. (1) and the Sher- Table 9. As a consequence, a reduction of Cm from 60 to 30% was
wood correlation (Eq. (7)) where Km was related to operating variables obtained with the increase of cross flow velocity in macro CF (see Fig.
(Reynolds number) as well as channel design (hydraulic diameter and S7). On the contrary, no significant variation of Cm was observed in
channel length). Re number was slightly higher in macro CF system micro CF when the cross flow velocity was increased.
(range 102-103) compared with micro CF system (range 100–800). In
addition, macro CF and micro CF are characterized by the same hy- Table 9
draulic diameter but different channel length. This difference explains Km of each organic tracer as a function of cross flow velocity in micro CF and
the lower Km values achievable in macro CF at the same cross flow macro CF (calculated using Eq. (7)).
velocity. This is because the mass transfer decreases by increasing the Micro CF
channel length as reported in Fig. S9. Km in macro CF was comparable
with micro CF at the beginning of the channel length where the highest vcross-flow (m/s) Km (·10−5
m/ s)
Km for NaCl (6‧10−5 m/s) was calculated.
Glucose Xylose Erythritol Dioxane
In conclusion, the proportion of the filtration channel (channel
length to width ratio) in a cross flow system should be considered 0.07 1.9 2.1 2.2 2.5
when the dimension of the cross flow cell is increased. This concept 0.13 2.3 2.5 2.6 3.0
0.24 2.7 2.9 3.2 3.5
was demonstrated for the feed spacer geometry in SWM modules,
0.44 3.4 3.7 3.9 4.4
which is a crucial parameter to affect mass transfer. Computational 0.59 3.9 4.2 4.5 5.0
fluid dynamics (CFD) simulations revealed that mass transfer
reached a maximum when the mesh length to height ratio of the Macro CF
spacer was decreased from 12 to 2. This was correlated with the
vcross-flow (m/s) Km (·10−5 m/ s)
formation of recirculation region between sequential filaments
where a local mass transfer increase was observed. A decrease of Glucose Xylose Erythritol Dioxane
mesh length means that the number of filaments in the spacer
structure increased and consequently a greater membrane area was 0.16 1.2 1.2 1.4 1.5
0.26 1.4 1.5 1.6 1.8
obscured by the filaments [68].
0.39 1.6 1.7 1.8 2.0
0.48 1.7 1.8 1.9 2.2
0.57 1.8 1.9 2.1 2.3

77
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

In conclusion, high cross flow velocity ( 0.4 m/s) is recommended Table 10


in macro CF to compensate the impact of solute concentration along the Pore radius of NF270 and NF90 estimated from organic tracer retention and
channel length, which occurred mostly when the system was operated SHP model.
with a velocity below 0.4 m/s. Tracer rs (nm) λ rp, SC λ rp, micro CF λ rp, macro CF
(nm) (nm) (nm)

3.5. Comparison of pore radius and MWCO of NF membranes by SC, micro NF270
CF and macro CF D-glucose 0.32 0.78 0.42 0.89 0.36 0.83 0.39
D-xylose 0.29 0.90 0.32 0.93 0.31 0.94 0.31
The comparison of MWCO and pore radius is shown in order to Erythritol 0.26 0.75 0.35 0.74 0.36 0.87 0.30
Dioxane 0.23 0.64 0.36 0.56 0.43 0.66 0.35
evaluate the consistency and reproducibility of the characterization
Average 0.36 ± 0.37 ± 0.06 0.34 ± 0.04
data in the three different systems. Membrane MWCO is determined by 0.04
the selectivity curve where the transport characteristics of the mem- NF90
brane as a function of molecular weight is shown. The selectivity curve D-glucose 0.32 0.96 0.34 0.99 0.32 0.99 0.32
D-xylose 0.29 0.98 0.29 0.99 0.29 0.99 0.29
of NF270 and NF90 for the three systems is reported in Fig. 9. The pore
Erythritol 0.26 0.98 0.27 0.94 0.28 0.99 0.27
radius (rp) (estimated by SHP model) and the ratio of the Stokes solute Dioxane 0.23 0.93 0.25 0.91 0.26 0.85 0.27
radius to pore radius (λ) of NF270 and NF90 are reported in Table 10. Average 0.30 ± 0.29 ± 0.04 0.29 ± 0.02
As expected from literature data, Rreal above 90% was obtained in 0.04
the range of molecular weight 150–180 Da and 120 Da for NF270 and
NF90, respectively [1,36,69]. Average pore radius estimated from the
organic tracer retention (Rreal) was 0.36 ± 0.06 and 0.29 ± 0.04 for Table 11
NF270 and NF90, respectively. The values calculated in this study Operative conditions suggested to test salt and organic tracer retention of new
membrane materials by using the SC, micro CF and macro CF systems.
confirmed previous literature data where the SHP model was used
[35,36,41,69–72]. System Recovery (%) TMP (bar) Jv (L/m2h)a Cross flow velocity (m/s)b
Overall results showed good reproducibility of MWCO as well as
Macro CF ≤1 ≤6 (loose) ≤60 ≥0.4
pore radius for all the three systems, while the small variations were ≤8 (tight)
within experimental error. These results are in accordance with the Micro CF ≤0.2 ≤6 (loose) ≤60 ≥0.4
membrane characteristics, the loose membrane has a larger pore radius ≤8 (tight)
and MWCO than the tight membrane. Rreal (which was used in the se- SC 15 ≤6 (loose) ≤60 ≥0.8 (≥200 rpm)
≤8 (tight)
lectivity curve and in the SHP model) was calculated by applying the
concentration polarization film model. In the latter case, Cm was taken a
At constant cross flow velocity of 0.4 m/s in macro CF and micro CF and
into account at conditions where concentration polarization was 400 rpm in SC.
minimal in all the three systems (cross flow velocity 0.4 m/s and b
At constant TMP of 4.8–5 bar in micro CF. Macro CF and SC.
4.8 bar, corresponding to Jv ≤ 60 L/m2h). This highlights that re-
producible characterization data (MWCO and pore radius) can be ob- mass transfer and higher Cm. This was correlated with lower NaCl re-
tained in different systems at similar boundary layer conditions. tention of the loose membrane (NF270) compared with the retention
reported for the SWM at the same trans-membrane pressure. However,
4. Conclusions good comparability of Cm of salt solution was found in the three systems
at flux below 80 and 60 L/m2h for NF270 and NF90, respectively.
The aim of this study was to evaluate three different laboratory Therefore, these flux values were considered in the suitable range to
nanofiltration membrane characterization systems in terms of impact of test salt retention keeping minimal concentration polarization condi-
system hydrodynamics and dimension on the retention of salt and or- tions. The same trend was observed for the filtration of organic tracers,
ganic tracers. This is important to provide suitable operative conditions where Cm increased significantly for the larger tracers at higher flux and
and to allow careful interpretation of the membrane characteristics. lower cross flow velocities. On the contrary, at flux below 60 L/m2h and
This is especially relevant when new membrane materials with un- velocity above 0.4 m/s, Cm reached values similar to the concentration
known properties are characterized as very small samples. The di- in the bulk solution in all the three systems. In conclusion, the operative
mensions of the membrane cell in the macro CF system (e.g. the longer conditions suggested to test salt and organic tracer retention of new
channel length to channel width ratio) was the most important para- membrane materials in the three systems are summarized in Table 11.
meter to affect mass transfer and concentration of both salt and organic Good reproducibility of MWCO and pore radius was observed for
tracers on the membrane surface (Cm). The higher channel length to the three systems. This is because Cm (that is used to calculate Rreal) was
width ratio of macro CF compared with micro CF resulted in reduced considered at minimal concentration polarization conditions (cross flow

Fig. 9. Selectivity curve of NF270 and NF90 determined by using DSC, MICCF and MACCF (20 mgC/L, TMP 4.8 bar, cross flow velocity 0.4 m/s, 400 rpm).

78
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

velocity 0.4 m/s and 4.8 bar, corresponding to Jv ≤ 60 L/m2h). This Re Reynolds number (−)
highlights that reproducible characterization data can be obtained in Sc Schmidt number (−)
different systems at similar boundary layer conditions. Km mass transfer coefficient (D/δ, m/s)
The results reported in this study should be considered valid in the δ boundary layer thickness (m)
context of the variation of hydrodynamics (e.g. relation of Sherwood
number and Reynolds range) in the three systems, which is intrinsically References
related to the different operation mode of dead end and cross flow. The
different range of Reynolds (104 < Re < 5·104 for the SC system and [1] C. Bellona, J.E. Drewes, P. Xu, G. Amy, Factors affecting the rejection of organic
Re < 104 for the micro CF and macro CF systems) as well as the solutes during NF/RO treatment—a literature review, Water Res. 38 (2004)
2795–2809.
channel length (L) of micro CF and macro CF systems are the boundary [2] B. Van der Bruggen, A. Verliefde, L. Braeken, E.R. Cornelissen, K. Moons,
conditions of the Sherwood correlations selected to describe the sys- J.Q. Verberk, H.J. van Dijk, G. Amy, Assessment of a semi‐quantitative method for
tems. This limits the validity of the operative conditions reported in estimation of the rejection of organic compounds in aqueous solution in nanofil-
tration, J. Appl. Chem. Biotechnol. 81 (2006) 1166–1176.
Table 11 to the design of the three systems and the operation mode. [3] G. Belfort, Membrane modules: comparison of different configurations using fluid
Therefore, they are not applicable to cross flow and dead end NF system mechanics, J. Membr. Sci. 35 (1988) 245–270.
of different configurations as the variation of the boundary condition on [4] A.L. Zydney, Stagnant film model for concentration polarization in membrane
systems, J. Membr. Sci. 130 (1997) 275–281.
the membrane surface will affect the membrane performance. More-
[5] G. Jonsson, C. Boesen, Polarization Phenomena in Membrane Processes, Harcourt
over, membrane properties, in particular permeability, will affect the Bracc Jovanovich, Troy, New York, 1984.
choice of operating conditions and results reported here are limited to [6] G. Van den Berg, I. Racz, C. Smolders, Mass transfer coefficients in cross-flow ul-
trafiltration, J. Membr. Sci. 47 (1989) 25–51.
typical NF. In view of the mentioned limitations, the most important
[7] A.I. Schäfer, A. Fane, T.D. Waite, Nanofiltration: Principles and Applications,
finding of this study was to identify a comparability of systems with Elsevier, 2005 Chap. 4.
different operation mode. This is significant in the area of membrane [8] M.C. Porter, Concentration polarization with membrane ultrafiltration, Ind. Eng.
characterization where devices with different dimension and huge Chem. Prod. Res. Dev. 11 (1972) 234–248.
[9] S. Kim, E.M.V. Hoek, Modeling concentration polarization in reverse osmosis pro-
amount of operative conditions are usually applied. The results high- cesses, Desalination 186 (2005) 111–128.
light that meaningful and comparable characterization of membranes [10] L. Song, M. Elimelech, Theory of concentration polarization in crossflow filtration,
as small as 2 cm2 can be obtained when the filtration systems and op- J. Chem. Soc. Faraday. Trans. 91 (1995) 3389–3398.
[11] D.L. Oatley-Radcliffe, S.R. Williams, C. Lee, P.M. Williams, Characterisation of mass
erating protocols are very carefully designed. transfer in frontal nanofiltration equipment and development of a simple correla-
tion, J. Membr. Sep.Tech. 4 (2015) 149.
Acknowledgements [12] S. De, P.K. Bhattacharya, Prediction of mass-transfer coefficient with suction in the
applications of reverse osmosis and ultrafiltration, J. Membr. Sci. 128 (1997)
119–131.
The Helmholtz Association Recruitment Initiative is thanked for [13] S. Déon, P. Dutournié, P. Fievet, L. Limousy, P. Bourseau, Concentration polariza-
funding the research at IFG-MT. The DOW Chemical Company kindly tion phenomenon during the nanofiltration of multi-ionic solutions: influence of the
filtrated solution and operating conditions, Water Res. 47 (2013) 2260–2272.
supplied the membrane samples. Manuel Kulmus is thanked for his
[14] C.P. Koutsou, A.J. Karabelas, Shear stresses and mass transfer at the base of a stirred
contributions to the macro CF system design. Dr. Heinz Lambach filtration cell and corresponding conditions in narrow channels with spacers, J.
(IMVT-KIT) was responsible for design and manufacture of the micro CF Membr. Sci. 399 (2012) 60–72.
[15] K.A. Smith, C.K. Colton, Mass transfer to a rotating fluid: Part I. Transport from a
membrane cell while Dr. Daniel Kuntz (TEC-KIT) built the SC cell.
stationary disk to a fluid in bödewadt flow, AIChE J. 18 (1972) 949–958.
Benjamin Chatillon, Cristina Onorato and Tobias Berger are thanked for [16] C.K. Colton, K.A. Smith, Mass transfer to a rotating fluid. Part II. Transport from the
their contributions to the micro CF system design, Labview program- base of an agitated cylindrical tank, AIChE J. 18 (1972) 958–967.
ming and troubleshooting. Isaac Owusu-Agyeman and Cristina Onorato [17] A.L. Zydney, A. Xenopoulos, Improving dextran tests for ultrafiltration membranes:
effect of device format, J. Membr. Sci. 291 (2007) 180–190.
for salt retention experiments and salt analysis with the macro CF and [18] G. Schock, A. Miquel, Mass transfer and pressure loss in spiral wound modules,
micro CF systems, respectively. Jin Peng Liu and Minh Nguyen for or- Desalination 64 (1987) 339–352.
ganic tracer experiments with micro CF and macro CF, respectively. [19] C. Combe, C. Guizard, P. Aimar, V. Sanchez, Experimental determination of four
characteristics used to predict the retention of a ceramic nanofiltration membrane,
Prantik Samanta for salt retention experiments and salt analysis in SC at J. Membr. Sci. 129 (1997) 147–160.
different stirrer speeds. Matteo Tagliavini and Tobias Berger are [20] J.D. Ferry, Statistical evaluation of sieve constants in ultrafiltration, J. Gen. Physiol.
thanked for proof reading and Minh Nguyen for provision of the SWM 20 (1936) 95–104.
[21] J.L. Anderson, J.A. Quinn, Restricted transport in small pores: a model for steric
schematic. exclusion and hindered particle motion, Biophys. J. 14 (1974) 130–150.
[22] E. Dražević, K. Košutić, V. Kolev, V. Freger, Does hindered transport theory apply to
Appendix A. Supplementary data desalination membranes? Environ. Sci. Technol. 48 (2014) 11471–11478.
[23] V. Kolev, V. Freger, Hydration, porosity and water dynamics in the polyamide layer
of reverse osmosis membranes: a molecular dynamics study, Polymer 55 (2014)
Supplementary data to this article can be found online at https:// 1420–1426.
doi.org/10.1016/j.memsci.2019.04.035. [24] S. Bason, Y. Kaufman, V. Freger, Analysis of ion transport in nanofiltration using
phenomenological coefficients and structural characteristics, J. Phys. Chem. B 114
(2010) 3510–3517.
List of symbols [25] S. Bhattacharjee, J.C. Chen, M. Elimelech, Coupled model of concentration polar-
ization and pore transport in crossflow nanofiltration, AIChE J. 47 (2001)
Dh hydraulic diameter (m) 2733–2745.
[26] W. Deen, Hindered transport of large molecules in liquid‐filled pores, AIChE J. 33
ds spacer filament diameter (m) (1987) 1409–1425.
L channel length or spacer mesh length in spiral wound module [27] Y. Kiso, K. Muroshige, T. Oguchi, T. Yamada, M. Hhirose, T. Ohara, T. Shintani,
(m) Effect of molecular shape on rejection of uncharged organic compounds by nano-
filtration membranes and on calculated pore radii, J. Membr. Sci. 358 (2010)
L* distance from the channel inlet where the velocity profile is 101–113.
completely developed (=0.029·Dh·Re) [28] S. Déon, P. Dutournié, P. Bourseau, Modeling nanofiltration with Nernst-Planck
U cross-flow velocity (m/s) approach and polarization layer, AIChE J. 53 (2007) 1952–1969.
[29] B. Van Der Bruggen, J. Schaep, D. Wilms, C. Vandecasteele, A comparison of models
ν kinematic viscosity = μ ⁄ ρ (μ is the fluid viscosity m2/s, ρ is to describe the maximal retention of organic molecules in nanofiltration, Sep. Sci.
the fluid density, kg/m3) Technol. 35 (2000) 169–182.
r radius of the stirred cell base (m) [30] C. Koutsou, A. Karabelas, M. Kostoglou, Membrane desalination under constant
water recovery–The effect of module design parameters on system performance,
D diffusivity or diffusion coefficient of retained species defined Sep. Purif. Technol. 147 (2015) 90–113.
by the Stokes-Einstein relationship (m2/s) [31] G.S. Kell, Density, thermal expansivity, and compressibility of liquid water from 0.
ω agitator rotational speed (rad/s) deg. to 150. deg. Correlations and tables for atmospheric pressure and saturation

79
A. Imbrogno and A.I. Schäfer Journal of Membrane Science 585 (2019) 67–80

reviewed and expressed on 1968 temperature scale, J. Chem. Eng. Data 20 (1975) [51] A.I. Schäfer, Natural Organics Removal Using Membranes, UNESCO Centre for
97–105. Membrane Science and Technology, 2002, p. 233.
[32] J. Shen, A.I. Schäfer, Factors affecting fluoride and natural organic matter (NOM) [52] T. Thorsen, H. Fløgstad, D. Techneau (Ed.), Nanofiltration in Drinking Water
removal from natural waters in Tanzania by nanofiltration/reverse osmosis, Sci. Treatment, Literature Review, 2006, p. 5.
Total Environ. 527–528 (2015) 520–529. [53] C. Koutsou, S. Yiantsios, A. Karabelas, Direct numerical simulation of flow in
[33] M. Mänttäri, T. Pekuri, M. Nyström, NF270, a new membrane having promising spacer-filled channels: effect of spacer geometrical characteristics, J. Membr. Sci.
characteristics and being suitable for treatment of dilute effluents from the paper 291 (2007) 53–69.
industry, J. Membr. Sci. 242 (2004) 107–116. [54] C.P. Koutsou, S.G. Yiantsios, A.J. Karabelas, A numerical and experimental study of
[34] E. Sjöman, M. Mänttäri, M. Nyström, H. Koivikko, H. Heikkilä, Separation of xylose mass transfer in spacer-filled channels: effects of spacer geometrical characteristics
from glucose by nanofiltration from concentrated monosaccharide solutions, J. and Schmidt number, J. Membr. Sci. 326 (2009) 234–251.
Membr. Sci. 292 (2007) 106–115. [55] S.S. Sablani, M.F.A. Goosen, R. Al-Belushi, M. Wilf, Concentration polarization in
[35] L.D. Nghiem, A.I. Schäfer, M. Elimelech, Removal of natural hormones by nano- ultrafiltration and reverse osmosis: a critical review, Desalination 141 (2001)
filtration membranes: measurement, modeling, and mechanisms, Environ. Sci. 269–289.
Technol. 38 (2004) 1888–1896. [56] A.R. Cooper, D.S. Van Derveer, Characterization of ultrafiltration membranes by
[36] M.J. López-Muñoz, A. Sotto, J.M. Arsuaga, B. Van der Bruggen, Influence of polymer transport measurements, Sep. Sci. Technol. 14 (1979) 551–556.
membrane, solute and solution properties on the retention of phenolic compounds [57] N. García-Martín, V. Silva, F. Carmona, L. Palacio, A. Hernández, P. Prádanos, Pore
in aqueous solution by nanofiltration membranes, Sep. Purif. Technol. 66 (2009) size analysis from retention of neutral solutes through nanofiltration membranes.
194–201. The contribution of concentration–polarization, Desalination 344 (2014) 1–11.
[37] K. Boussu, Y. Zhang, J. Cocquyt, P. Van Der Meeren, A. Volodin, C. Van [58] V. Silva, P. Prádanos, L. Palacio, A. Hernández, Alternative pore hindrance factors:
Haesendonck, J. Martens, B. Van der Bruggen, Characterization of polymeric na- what one should be used for nanofiltration modelization? Desalination 245 (2009)
nofiltration membranes for systematic analysis of membrane performance, J. 606–613.
Membr. Sci. 278 (2006) 418–427. [59] J. Luo, Y. Wan, Effects of pH and salt on nanofiltration—a critical review, J. Membr.
[38] C.Y. Tang, Y.-N. Kwon, J.O. Leckie, Characterization of humic acid fouled reverse Sci. 438 (2013) 18–28.
osmosis and nanofiltration membranes by transmission electron microscopy and [60] A. Seidel, J.J. Waypa, M. Elimelech, Role of charge (Donnan) exclusion in removal
streaming potential measurements, Environ. Sci. Technol. 41 (2007) 942–949. of arsenic from water by a negatively charged porous nanofiltration membrane,
[39] E.I. Mouhoumed, A. Szymczyk, A. Schäfer, L. Paugam, Y.-H. La, Physico-chemical Environ. Eng. Sci. 18 (2001) 105–113.
characterization of polyamide NF/RO membranes: insight from streaming current [61] H. Kelewou, A. Lhassani, M. Merzouki, P. Drogui, B. Sellamuthu, Salts retention by
measurements, J. Membr. Sci. 461 (2014) 130–138. nanofiltration membranes: physicochemical and hydrodynamic approaches and
[40] A. Simon, J.A. McDonald, S.J. Khan, W.E. Price, L.D. Nghiem, Effects of caustic modeling, Desalination 277 (2011) 106–112.
cleaning on pore size of nanofiltration membranes and their rejection of trace or- [62] H. Al-Zoubi, N. Hilal, N.A. Darwish, A.W. Mohammad, Rejection and modelling of
ganic chemicals, J. Membr. Sci. 447 (2013) 153–162. sulphate and potassium salts by nanofiltration membranes: neural network and
[41] G. Bargeman, J.B. Westerink, C.F.H. Manuhutu, A.t. Kate, The effect of membrane Spiegler–Kedem model, Desalination 206 (2007) 42–60.
characteristics on nanofiltration membrane performance during processing of [63] R. van Reis, E.M. Goodrich, C.L. Yson, L.N. Frautschy, S. Dzengeleski, H. Lutz,
practically saturated salt solutions, J. Membr. Sci. 485 (2015) 112–122. Linear scale ultrafiltration, Biotechnol. Bioeng. 55 (1997) 737–746.
[42] K. Boussu, J. De Baerdemaeker, C. Dauwe, M. Weber, K.G. Lynn, D. Depla, S. Aldea, [64] B. Tansel, J. Sager, T. Rector, J. Garland, R.F. Strayer, L. Levine, M. Roberts,
I.F. Vankelecom, C. Vandecasteele, B. Van der Bruggen, Physico‐chemical char- M. Hummerick, J. Bauer, Significance of hydrated radius and hydration shells on
acterization of nanofiltration membranes, ChemPhysChem 8 (2007) 370–379. ionic permeability during nanofiltration in dead end and cross flow modes, Sep.
[43] I. Owusu-Agyeman, A. Jeihanipour, T. Luxbacher, A.I. Schäfer, Implications of Purif. Technol. 51 (2006) 40–47.
humic acid, inorganic carbon and speciation on fluoride retention mechanisms in [65] B. Tansel, Significance of thermodynamic and physical characteristics on permea-
nanofiltration and reverse osmosis, J. Membr. Sci. 528 (2017) 82–94. tion of ions during membrane separation: hydrated radius, hydration free energy
[44] H.Q. Dang, W.E. Price, L.D. Nghiem, The effects of feed solution temperature on and viscous effects, Sep. Purif. Technol. 86 (2012) 119–126.
pore size and trace organic contaminant rejection by the nanofiltration membrane [66] R. Epsztein, E. Shaulsky, N. Dizge, D.M. Warsinger, M. Elimelech, Role of ionic
NF270, Sep. Purif. Technol. 125 (2014) 43–51. charge density in Donnan exclusion of monovalent anions by nanofiltration,
[45] Dow Chemical Company, Product specification NF90, Filmtec membranes, http:// Environ. Sci. Technol. 52 (2018) 4108–4116.
msdssearch.dow.com/PublishedLiteratureDOWCOM/dh_082d/ [67] L.A. Richards, A.I. Schäfer, B.S. Richards, B. Corry, The importance of dehydration
0901b8038082d59b.pdf?filepath=liquidseps/pdfs/noreg/609-00378.pdf& in determining ion transport in narrow pores, Small 8 (2012) 1701–1709.
fromPage=GetDoc , Accessed date: 20 November 2017. [68] J. Schwinge, D. Wiley, D. Fletcher, Simulation of the flow around spacer filaments
[46] Dow Chemical Company, Product specification NF270, Filmtec membranes, http:// between channel walls. 2. Mass-transfer enhancement, Ind. Eng. Chem. Res. 41
msdssearch.dow.com/PublishedLiteratureDOWCOM/dh_0074/ (2002) 4879–4888.
0901b803800749e1.pdf?filepath=liquidseps/pdfs/noreg/609-00519.pdf& [69] L. Nghiem, A. Manis, K. Soldenhoff, A. Schäfer, Estrogenic hormone removal from
fromPage=GetDoc , Accessed date: 20 November 2017. wastewater using NF/RO membranes, J. Membr. Sci. 242 (2004) 37–45.
[47] L.D. Nghiem, A.I. Schäfer, M. Elimelech, Removal of natural hormones by nano- [70] A.J. Semião, A.I. Schäfer, Removal of adsorbing estrogenic micropollutants by na-
filtration Membranes: measurement, modeling, and mechanisms, Environ. Sci. nofiltration membranes. Part A—experimental evidence, J. Membr. Sci. 431 (2013)
Technol. 38 (2004) 1888–1896. 244–256.
[48] D.W.G. Robert H. Perry, Perry's Chemical Engineers' Handbook, seventh ed. ed., Mc [71] L.D. Nghiem, S. Hawkes, Effects of membrane fouling on the nanofiltration of
Graw Hill, New York, 1999. pharmaceutically active compounds (PhACs): mechanisms and role of membrane
[49] C. Geankoplis, Drying of Process Materials, Transport processes and unit operations, pore size, Sep. Purif. Technol. 57 (2007) 176–184.
1993, pp. 520–583. [72] L.A. Richards, M. Vuachère, A.I. Schäfer, Impact of pH on the removal of fluoride,
[50] M. Nilsson, G. Trägårdh, K. Östergren, The influence of sodium chloride on mass nitrate and boron by nanofiltration/reverse osmosis, Desalination 261 (2010)
transfer in a polyamide nanofiltration membrane at elevated temperatures, J. 331–337.
Membr. Sci. 280 (2006) 928–936.

80

You might also like