Computers and Chemical Engineering 149 (2021) 107313

Contents lists available at ScienceDirect

Computers and Chemical Engineering

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

Water desalination by forward osmosis: Dynamic performance

assessment and experimental validation using MgCl2 and NaCl as
draw solutes
Jaouad Eddouibi a,b,∗, Souad Abderafi b, Sébastien Vaudreuil a, Tijani Bounahmidi a
a
Euro-Med Research Institute, Euro-Mediterranean University of Fes (UEMF), Rond-Point Bensouda Route Nationale de Meknès, BP 51 Fès, Morocco
b
MOSEM2PI, Ecole Mohammadia d’Ingénieurs, Université Mohamed V de Rabat, Avenue Ibn Sina B.P 765, Agdal-Rabat, Morocco

a r t i c l e i n f o a b s t r a c t

Article history: Although the FO process steady state performance has been extensively investigated, there is a huge gap
Received 4 January 2021 in forward osmosis dynamic studies especially for water desalination applications. In this paper, an ap-
Revised 27 February 2021
propriate dynamic model was developed based on the phenomena involved in the FO process. For its
Accepted 27 March 2021
validation, an experimental study was carried out using a FO test bench and flat-sheet membrane for
Available online 1 April 2021
two draw solutes (MgCl2 and NaCl), to measure the water flux, Jw, and the reverse solute flux (RSF), Js .
Keywords: The time evolution of these two experimentally obtained fluxes were compared to those calculated using
FO desalination a dynamic model developed using Python language for both draw solute. This comparison shows that the
Experiment mean absolute error did not exceed 5.25% for water flux and 5.63% for RSF. The dynamic model can thus
Dynamic model be considered reliable when considering experimental errors. The resulting model can thus analyze the
Performance influence of the initial draw solute (DS) concentration on both water flux and reverse solute flux dynam-
MgCl2 draw solute
ics, and subsequently compare the osmotic performances of MgCl2 and NaCl based on the specific RSF
Python
ratio. The water flux obtained with MgCl2 is on average 8.23% lower than for NaCl. MgCl2 nevertheless
performed better in terms of reverse solute flux and has been demonstrated to be 28.87% lower than
NaCl. The calculated specific RSF ratio is shown to be independent of both time and DS concentration.
© 2021 Elsevier Ltd. All rights reserved.

1. Introduction hand, a good membrane for industrial FO applications should be

highly selective for salts, highly permeable to water, have a min-
Among the various membrane desalination technologies un- imum reverse solute flux (RSF) and insensitive to pH. All these
der development, forward osmosis (FO) has emerged as an impor- criteria can be summarized into the structural parameter S of the
tant technology for desalination of brackish water (Shaffer et al., membrane. This parameter depends in fact on the thickness of the
2015). This is mainly due to its promising low energy consumption membrane support layer dL, its tortuosity τ and its porosity ε, and
(Zhou et al., 2015), interesting water flux (Chakrabortty et al., 2015; is given by Deshmukh et al. (2015) as S = τ dL/ε . A forward osmo-
Lee et al., 2018) and low fouling rate (Lee et al., 2020). It is also sis membrane is thus considered more efficient when its structural
considered in the food processing industry as a promising tech- parameter is low. This can be attained through a low support layer
nology for concentrating products (Chen et al., 2019). Many con- thickness, less tortuosity and high permeability to water while
straints are however related to the performances of the membrane having a low permeability to reverse solute. Another challenge for
and the draw solute, the main components of this process. These FO is the lack of a suitable nontoxic and thermally separable draw
constraints limit the large-scale industrialization of this technology solution. This also hinders the development and commercialization
(Su et al., 2013; Cai and Hu, 2016). A good draw solution should of forward osmosis separation units (Cai and Hu, 2016; Field and
exhibit a high osmotic pressure, be easily recoverable with low en- Wu, 2013; Qasim et al., 2017).
ergy consumed and be nontoxic (Johnson et al., 2018). On the other The dilutive internal concentration polarization (ICP) and con-
centrative external concentration polarization (ECP) are major fac-
tors reducing the osmotic pressure difference between the two
∗
Corresponding author. sides of the membrane (Field and Wu, 2018; Nagy, 2019; Tan and
E-mail addresses: [email protected] (J. Eddouibi), abderafi@emi.ac.ma Ng, 2008). As this driving force is reduced, a significant decrease
(S. Abderafi), [email protected] (S. Vaudreuil), [email protected] in the water flux is observed (McCutcheon and Elimelech, 2006).
(T. Bounahmidi).

https://doi.org/10.1016/j.compchemeng.2021.107313
0098-1354/© 2021 Elsevier Ltd. All rights reserved.
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Nomenclature w Water
s Salt
FO Forward osmosis i Time index
RSF Reverse solute flux
ICP Dilutive internal concentration polarization
ECP Concentrative external concentration polarization In order to predict the effect of the concentration polarization and
DS Draw solute attenuate its effect on water flux, many studies have been carried
Jw Water flux L m−2 h−1 out (Tan and Ng, 2008; Gao et al., 2014; Arjmandi et al., 2020). The
Js Reverse solute flux g m−2 h−1 FO process performance has been significantly improved through
A Water permeability coefficient of the membrane optimizing the thickness of the support layer of the membrane
L m−2 h−1 bar −1 based on different membrane manufacturing methods and new
B Salt permeability coefficient of the membrane improved DS.
L m−2 h−1 Many modeling and simulation studies of forward osmosis pro-
S Membrane structural parameter μm cess have been carried out to assess the water flux and reverse
dL Thickness of the membrane support layer μm solute flux. Simulations were performed, using steady state mod-
D bulk solute diffusion coefficient cm2 s−1 els based on transport phenomena (Su et al., 2013; Li et al., 2017;
k Mass transfer coefficient cm s−1 Kim et al., 2017). These models were mainly focused on the three
CF Feed solution salt concentration M aforementioned parameters of the membrane. In most modeling
CD Draw solution salt concentration M work, the three parameters are determined through experiment
πF Osmotic pressure of the feed solution bar using NaCl or NH4 Cl as draw solution, with the help of non-linear
πD Osmotic pressure of the draw solution bar least-square algorithm to estimate values of the water permeabil-
ns Number of salt ions dissolved in water - ity coefficient A, the salt back-diffusion coefficient B and the mem-
Rg gas constant Lbar ◦ K −1 mol −1 brane structural parameter S while minimizing the error between
T Temperature ◦ K the measured and predicted water flux. Among the commonly
Dfs Feed side bulk solute diffusion coefficient cm2 s−1 used models, the one developed by Tiraferri et al. (2013) seems
Dds Draw side bulk solute diffusion coefficient cm2 s−1 the most realistic one. Although the FO steady state performances
kfs Feed side ass transfer coefficient cm s−1 have been extensively investigated, only a few studies have been
conducted to investigate water and solute transport dynamic be-
kds Draw side ass transfer coefficient cm s−1
havior. Assessing the impact of draw solution type on the FO per-
Sh f s Feed side Sherwood number -
formance over time could be of great importance for enhancing
Shds Draw side Sherwood number -
water quantity and quality production using FO. To extend the use
Sc f s Feed side Schmidt number -
of FO technology to water desalination, the influence on the water
Scds Draw side Schmidt number -
and reverse solute flux dynamics of the draw solution performance
Re f s Feed side Reynolds number -
(i.e. osmotic pressure) and of its concentration and temperature on
Reds Draw side Reynolds number -
both sides of the membrane is highly critical. This should thus be
ufs Feed solution velocity m s−1
carefully taken into account in order to drive FO application to a
uds Feed solution velocity m s−1
mature stage in water desalination.
F Circulation pump volumetric flowrate L min−1
In the present work, a brief literature review of the FO process
F1 Feed solution flowrate at the feed channel outlet of
dynamic studies is carried out, followed by the development of a
the membrane cell L min−1
FO dynamic model for water desalination applications based on
F2 Draw solution flowrate at the draw channel outlet
existing models. The dynamic model reliability is assessed using
of the membrane cell L min−1
data obtained during 30 h of operation for experiments performed
Am The effective membrane Area cm2
with two DS, namely MgCl2 and NaCl. The aforementioned model
Wm Width of the FO cell channel m
is used to investigate the influence of DS concentration on both
Dm Depth of the FO cell channel m
water flux and RSF through sensitivity analysis study, and compar-
dh Hydraulic diameter of the FO cell channels m
ing MgCl2 and NaCl in terms of osmotic dynamic performances.
Vcal,i Calculated volume at time index i Liter
Vexp,i Experimental volume at time index i Liter
t Time step min or h 2. Existing FO dynamic models analysis

Greek letters Continuous or batch operation of a FO desalination process

τ Membrane tortuosity - requires dynamic tracking of the water flux and RSF for con-
ε Membrane porosity - trol purposes and optimal performances. To better understand the
φs Osmotic coefficient of salt solutions - FO dynamic behavior, a review of all FO dynamic models adapt-
σ Membrane reflectance coefficient - able to water desalination was performed to the best of our
ρw Pure water density kg m−3 knowledge.
ρds Draw solute density kg m−3 Good control in continuous FO process helps maintain drinking
μw Water viscosity mPa.s water production and water quality through various means. While
μds Draw solute viscosity mPa.s water flux can be maintained by preventing a reduction in the os-
motic pressure difference induced by the ICP and ECP, water qual-
Superscript ity is ensured through control of the residual salt concentration.
fs Feed solution Maintaining the water flux can be achieved by the continuous di-
ds Draw solute lution of the draw solution to change its concentration according
m Membrane to a very precise time interval, brine evacuation and regular mem-
n Number of values brane cleaning. These control strategies require a reliable dynamic
model of the forward osmosis process. Only a few studies have

2
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

been identified that focused on FO dynamic models development

to cover the transient behavior of FO process. Ryu et al. (2020) and
Gadêlha et al. (2015) conducted studies of FO dynamics operating
in batch mode and taking into account the non-linear relation-
ship between the solute back-diffusion flux and water flux. The
water and solute flux model developed in both studies has been
experimentally validated using sodium chloride as draw solution.
Gulied et al. (2020) have developed a dynamic model for FO using
fertilizers as draw solution and simulated its transient behavior us-
ing Aspen plus® and Matlab®. Ruprakobkit et al. (2019, 2017, 2016)
established a dynamic model for acid concentration using FO. The
model has been compared against experimental data for an operat-
ing period of 30 h and different acids. In contrast to the aforemen-
tioned studies in which a plate and frame membrane configuration
was used, Munubarthi et al. (2020) have focused on the transient
behavior of concentration and volume for both sides of a FO cell
based on hollow fiber membrane.
In a manner similar to static FO models, development of these
dynamic models has been based on mass transport phenomena
Fig. 1. Illustration of fluxes through a semi-permeable membrane and concentra-
through the semi-permeable membrane. Each model was built
tions gradients in FO desalination process.
with its own level of details according to the considered assump-
tions and accuracy of its parameters. The models developed by
Ryu et al. (2020) and Gadêlha et al. (2015) were used to pre-
3.1. Steady-state FO flux model
dict water flux in FO desalination using glucose and sodium chlo-
ride (NaCl) respectively. These two models have been tested un-
In steady state, the model of FO water desalination only takes
der batch mode operation and do not take into account the back-
into account the water flux and reverse solute flux as depicted
diffusion phenomena of all the ions involved in the DS. The third
in Fig. 1. Concentration gradient between the feed and draw so-
model developed by Ruprakobkit et al. (2016) presents a detailed
lutions induce a gradient of osmotic pressure which drives water
explanation of all the phenomena taking place in FO process.
molecules from the low concentration solution (low osmotic pres-
It was validated in continuous operating mode with Ammonium
sure) to the concentrated DS (high osmotic pressure) through a
chloride (NH4 HCO3 ) as draw solution. Given that the sought model
semi-permeable membrane. During batch FO operation, the feed
will serve to predict water production under continuous operation
solution becomes more and more concentrated as time evolves
mode, the approach developed by Ruprakobkit et al. (2016) for car-
while the draw solution gets diluted. This process continues un-
boxylic acid concentration using FO could be adapted for dynamic
til osmotic equilibrium is reached, where the difference in osmotic
simulation of FO desalination. These FO dynamic models were ori-
pressure between both membrane sides becomes zero. This can
ented towards performance calculation in product concentration
be expressed by incorporating the water and reverse solute fluxes
applications such as acid and juice purification. This leaves a huge
Eqs. (1) and (2) as proposed by (Tiraferri et al., 2013) into the mass
gap in FO process dynamics investigation, especially for water de-
balance of water and DS on each side of the FO cell:
salination applications. Even though some existing few studies in- ⎡     ⎤
volved FO dynamics, only NaCl or NH4 HCO3 was considered as πD exp −Jw S/D − πF exp Jw/k
draw solute. Jw (t ) = A⎣      ⎦ (1)
In the present work, the dynamic model development for FO 1+ B
Jw
exp Jw/k − exp −Jw S/D
process is tested using MgCl2 as a draw solution and its per-
formance will be compared to NaCl. The two DS considered ⎡     ⎤
CD exp −Jw S/D − CF exp Jw/k
have been chosen for their promising osmotic pressure and low
Js (t ) = B⎣      ⎦ (2)
cost (Johnson et al., 2018). Another benefit is their low toxic-
1+ B
exp Jw/k − exp −Jw S/D
ity for human consumption at low and moderated concentrations Jw
(MENEELY and DAHL, 1961).
These equations are obtained by considering certain simplify-
ing assumptions and taking into account the concentration polar-
ization phenomena, as detailed by Tiraferri et al. (2013). The val-
3. Dynamic Model development for FO process
ues of the osmotic pressures of the DS (πD ) and feed solution (πF )
are considered to be linearly dependent to the solute concentration
The dynamic model is based on phenomena involved in the
through the Van’t Hoff equation given by relations (3) and (4):
forward osmosis process, leading to diffusion and material bal-
ance equations. Model equations are written to account for the πD = ns φsCD Rg T (3)
substances in the mixture and their associated variables during
the operation of forward osmosis process (such as viscosity dif-
πF = ns φsCF Rg T (4)
fusivity, density…, of each substance). The dynamic model of
Ruprakobkit et al. (2016) was developed for the operation of for- Where: φs is the DS osmotic coefficient considered equal to
ward osmosis process and simulated a low concentration of sin- unity; ns is the number of salt ions dissolved in water; Rg the gas
gle carboxylic acid feed solution with a 1 M NH4 Cl draw solution. constant and T the temperature in K; CD and CF are respectively
For our case, which is water desalination, the dynamic model in- the salt concentration in the draw solution side and the feed so-
cludes flux equations for water and 1.0 M NaCl/MgCl2 draw solu- lution side in M, and S is the membrane structural parameter in
tion. The equations corresponding to their thermo-chemical prop- μm.
erties are also considered. Both Steady state and dynamic perfor- The water permeability coefficient A, salt back-diffusion coef-
mance model equations of the FO process are presented. ficient B, and the structural parameter of the membrane S have

3
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Table 1 The terms Sh f s and Shds in the two Eqs. (6) and (7) represent
Constants used in the reverse salt and water flux model (Ruprakobkit et al., 2016).
respectively the Sherwood number on the feed and draw solution
Variable Value Unit side. They evolve according to the flow regime expressed by the
DC l − ,298 2.032 × 10−5 cm2 /sec Reynolds number (Re f s , Reds ) and the diffusive forces expressed by
DH + ,298 9.311 × 10−5 cm2 /sec
the Schmidt number (S c f s , Scds ). The salt back-diffusion observed
DN a+ ,298 1.334 × 10−5 cm2 /sec
DMg2+ 0.71 × 10−5 cm2 /sec in FO lead to a progressive increase in the NaCl ions concentra-
ns
298
2 − tion on the feed side, while a decrease is observed on the draw
Rg 0.08314 Lbar ◦ K −1 mol −1 side. As the same ions are found on both sides of the membrane,
σ 1.0 the same correlations have been used for the Sherwood number
φs 1.0 −
on both sides of the membrane. As the membrane configuration is
considered to be vertical, an appropriate correlation Tanaka et al.,
2020) is used to determine Sherwood number as given by the set
been estimated from experimental data generated in reverse os- of Eqs. (12) to ((15).
mosis mode using the Sterlitech Corporation bench test with CTA
membrane and NaCl as DS according to the methodology proposed Sh f s = 2.0 + 0.6Re1f /s3 Sc1f /s 3 (12)
by Tiraferri et al. (2013).
The salt diffusion coefficient D(cm2 /s ) is evaluated through the Shds = 2.0 + 0.6Re1ds/2 Scds
1/3
(13)
following equation:

D = D f s = Dds =
2
(5)
μw
Sc f s = (14)
1
DNa+ ,T
+ 1
DC l − ,T ρw D f s
The mass transfer coefficients of feed solution k f s and draw so- μd
Scds = (15)
lution kds expressed in (dm/min ) are evaluated through Eqs. (6) and ρds Dds
(7). Table 1 summarizes all the constants used in the reverse Where Re f s and Reds represent the Reynolds numbers charac-
salt and water flow model (supplementary data adopted from terizing the flow of fluids on both sides of the membrane, i.e. feed
Ruprakobkit et al., 2016 and Cussler, 2007). solution and draw solution, respectively. Their values are estimated
Sh f s D f s using respectively Eqs. (16) and (17).
kfs = (6)
dh L u f s ρw
Re f s = (16)
Shds Dds
μw,T
kds = (7)
dh Luds ρds
Reds = (17)
Even though the Van’t Hoff equation does not performs well at μds
higher concentrations (Van Gauwbergen et al., 1997; Wilson and Where, L represents the length of the membrane cell of the test
Stewart, 2013), it still applicable in the case where the concentra- bench FO; u f s and uds are the feed side and draw side average
tion at the interface between the support and active layer is signif- flow velocities inside the membrane channels which are given by
icantly much lower than the bulk concentration. This assumption Eqs. (18) and (19), respectively:
may lead to inaccurate results at the beginning of the process. As
the water flux is however higher at the beginning, the DS dilution F − (Jw − Js )Am
ufs = (18)
rate is much faster. This consequently results in osmotic pressure Wm Dm
prediction using Van’t Hoff equation becoming more accurate as
time evolve. F + (Jw − Js )Am
uds = (19)
The diffusion coefficients of the ions involved in the two con- Wm Dm
sidered DS, namely DMg2+ , DN a+ ,T , and DCl − are highly influenced Density of MgCl2 and NaCl DS as a function of molar con-
T T
by temperature and the dynamic viscosity of the solution in which centrations Cds and absolute temperature T are determined using
they are dissolved. To take this impact into account, the diffusion Eqs. (20) and (21) adapted from (Novotný and Söhnel, 1988), re-
coefficients of the three ions were expressed as a function of tem- spectively.
perature and dynamic viscosity as given by Eqs. (8) and (9):
ρMgCl2 = ρw + 0.8099 × 102Cd,s − 1.887 × 10−1Cd,s (T − 273)
 T  μ
w,298
DNa+T = DNa+ ,298 (8) + 2.315 × 10−3Cds (T − 273 )2 − 6.029Cds
1. 5
+ 7.449
298 μw,T
 T  μ × 10−2Cds
1. 5
(T − 273 ) − 8.305 × 10−4Cds
1. 5
(T − 273 )2 (20)
w,298
DClT− = DCl298
− (9)
298 μw,T
ρNaCl = ρw + 0.4485 × 102Cds − 9.634 × 10−2Cds (T − 273)
The ion i diffusion coefficient at 298 K Di− is given in
Table 1 and the water dynamic viscosity, μw,T , are calculated from
298
+ 0.6136 × 10−3Cds (T − 273 )2 − 2.712Cds
1. 5
+ 1.009
these equations (Haynes, 2014): × 10−2Cds
1. 5
(T − 273 ) (21)
 T  μ
DMg2+ = DMg2+
w,298
(10) Where ρw represents the water density calculated by
T 298 298 μw,T Eq. (22) as proposed by (Novotný and Söhnel, 1988), and Cds
is the molar concentration of the considered DS in (mol.L−1 ):
(T − 273.15) + 246.0 ρw = 999.65 + 2.043810−1 (T − 273.15) − 6.17410−2 (T − 273.15)3/2
μw,T =
(0.05594(T − 273.15) + 5.2842)(T − 273 ) + 137.37 (22)
(11)

4
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Fig. 2. Mass transfer and dynamic material balances involved in FO desalination process (Ruprakobkit et al., 2016).

The dynamic viscosity μds of both MgCl2 and NaCl is estimated Eqs. (28) and (29):
by taking into account the impact of molar concentration, density  
and absolute temperature, using Eqs. (23) and (24) as suggested by Cds,0Vd,0 + Cds,2 F2 t − Cds F t = Cds Vd,0 + F2 t − F t (28)
Laliberté (2007) :
⎛   2.2694 ⎞
24.032 MWds Cds/ + 3.7108
⎜ ρds ⎟
μMgCl2 = exp⎜
⎝   0.14474 ⎟
⎠ (23)
 −2
 Cd,s/
2.185310 (T − 273.15) + 1 −1.1236 MWds ρds +1

⎛   1.3229 ⎞
16.222 MWds Cds/ + 1.4849
⎜ ρds ⎟
μNaCl = exp⎜
⎝   2.0583 ⎟
⎠ (24)
 
7.469110−3 (T − 273.15) + 1 30.78 MWds Cds/ρ +1
ds

Where MWds represents the molecular weight of the concerned

DS and ρds its density.
Cds F t − Js Am t = Cds, 2 F2 t (29)
3.2. FO dynamic mass balances
Where F2 represents the DS flowrate at the outlet of the draw
solution side of the membrane. Its value is assessed with the help
The system consists of two hydraulic pumps delivering constant
of Eq. (30):
flow of feed and draw solution to the FO cell in which the two
streams are separated by the FO membrane. Both flows are then F2 = F + Jw Am (30)
returned to their respective storage tanks, as illustrated in Fig. 2. A
molar balance is performed on the salt in the feed solution tank
and the feed side of the membrane cell, yielding corresponding 4. Methodology
balance Eqs. (25) and (26) as a function of time:
  Reliability and accuracy of the dynamic model must be eval-
C f s F1 t − C f s,i F t = C f s, i V f,0 + F1 t − F t (25) uated before its use. This is performed through an experimental
study combined with simulation.

C f s,i F t + Js Am t = C f s F1 t (26)
4.1. Experimental study
Where F1 denotes the feed solution volumetric flowrate at the
outlet of the membrane cell feed side. This flowrate is derived from 4.1.1. Experimental equipment
the instantaneous water flux Jw such as: The experimental study is carried out using a forward osmo-
sis test bench to measure the water flow and the reverse so-
F1 = F − Jw Am (27)
lute flow. The experimental device represented in Fig. 3 is used.
Where F is the feed pump flowrate and considered equal to the It is a lab-scale FO test skid from Sterlitech Corporation© (Kent,
draw solution flowrate. Am is the membrane area. USA) and include a FO plate-and-frame type cell (Sepa CF type)
The molar balance applied to the draw solution tank and membrane with a 140 cm² area (14.7 cm length and 9.53 cm
the draw solution side of the membrane cell leads to balance width). Three flat-sheet cellulose triacetate (CTA) membranes from

5
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Fig. 3. Forward osmosis test bench used to perform the experimental study.

FTSH2O© have been used for experiments. Their intrinsic parame- Table 2
The uncertainty introduced in the flux measurements, for different concentration of
ters, namely water permeability coefficient A, reverse solute coef-
NaCl.
ficient B and structural parameter were determined based on RO
mode experiments as suggested by (Tiraferri et al., 2013). The FO Concentration (M) JW (L/m2 h) STDW (L/m2 h) Js (g/m²h) STDs (g/m²h)
test cell is designed with symmetric channels with a 0.19 cm depth 0.5 9.731 0.742 3.497 0.766
on both membrane sides. Flowrate is provided by two diaphragm 1.0 20.313 1.011 8.298 0.729
pumps (M03S Hydra-cell©) to continuously circulate the feed and 2.0 26.822 1.391 11.901 0.758

draw solutions on both sides of the FO membrane. Flow rate of the

feed solution was maintained equal to the draw solution flowrate Table 3
which is maintained at 1.5 ± 0.2 L/min at 25±0.25 °C. Constant Fixed variables and independent variables considered for simulations and experi-
ments.
temperature of both solutions in their respective 30 kg tank is
achieved by LS-Series Compact Chiller (PolyScience Corporation). MgCl2 NaCl
The system is equipped with conductivity, temperature, pressure Fixed Variables:
and flow sensors to monitor any changes in the solution overtime. Membrane surface area Am (cm2 ) 140 140
Water flux was measured through weight change of the draw so- Depth of the FO cell channel Dm (mm ) 1.9 1.9
Width of the FO cell channel Wm (mm ) 95.3 95.3
lution. This is achieved with ±0.001 g precision through a Mettler
Length of the FO cell channel dh (mm ) 146 146
Toledo scale placed under the tank. All relevant data is collected
Feed and draw volumetric flow F (L/min ) 1.2 1.3
via an integrated data logger. All experiments have been conducted Feed and draw tanks temperature T (◦C ) 28.0 28.0
in FO mode (active layer of the membrane facing the feed solu- Independents variables:
tion). Feed solution initial volume V f,0 (L ) 20.0 20.0
Draw solution initial volume Vd,0 (L ) 7.0 7.0
Water permeability coefficient A (L/m2 h.bar ) 1.21 1.23
4.1.2. Procedure and data reliability Membrane salt permeability coefficient B(L/m2 h ) 0.24 0.47
To assess model accuracy and compare the dynamic perfor- Structural parameter S (μm ) 328 328
mances of Magnesium Chloride (MgCl2 ) against Sodium chloride
(NaCl), two experiments have been performed using both DS. The
first test was carried out using a 1.0 M MgCl2 DS while the sec- of the mean values of water flows and RSF is found in Table 2,
ond was using 1 M NaCl DS. The 1.0 M concentration is within the as measured for different NaCl concentrations. This shows very
validity range of the Van’t Hoff equation which has been demon- little changes in standard deviation as function of concentration,
strated to be accurate only for diluted solution (Van Gauwbergen both for water flow and RSF. The maximum value does not exceed
et al., 1997; Wilson and Stewart, 2013; Yokozeki, 2006). 1.4 L/m2 h and 0.77 g/m²h respectively.
Prior to recording measurements, the desired flow rates and The experiments’ reliability suggests that the test bench allows
pressures on both sides of the membrane were adjusted through to reproduce almost the same water and reverse solute fluxes un-
minute changes with control valves. No weights nor electrical con- der the same experimental conditions.
ductivities values were considered during this phase because of
rapid fluctuations. Data logging was started once flow rates and 4.2. Model implementation in python
pressures became stable, indicating the beginning of the test.
Accuracy of experimental data was evaluated in terms of repro- In order to calculate the time profile of the water flux cross-
ducibility, using 0.5, 1.0 and 2.0 M NaCl. Each 10 h experiment ing the membrane Jw , the reverse solute flux Jw and concentra-
was triplicated to evaluate the random experimental error of the tions C f s and Cds , the dynamic model described by Eqs. (1) to
method. Fig. 4 presents the standard deviation corresponding to (30) was implemented in Python. This model used constant
the mean value of each triplicates measurement point, for water variables (Table 1) along with fixed and independent variables
flux and reverse solute flux of 1.0 M NaCl. This figure shows sat- reported in Table 3. The Levenberg-Marquardt (LM) algorithm
isfactory errors for the measured fluxes. The standard deviation integrated in scipy.optimize framework (Matott et al., 2011) is

6
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Fig. 4. Standard deviation corresponding to measurements of water flux and RSF, for 1.0 M NaCl during 10 h of operation.

adapted to solve iteratively Eq. (1). This algorithm included within

Scipy.optimize is just a wrapper around the original algorithm
available from the Fortran MINPACK package (Community, 2013).
Similar performance is therefore achieved when using LM from
Scipy.optimize package to solve the water flux equation. Some dif-
ferences in the convergence time were pointed out when the LM
algorithm is used to solve the flux equation with different initial
guess values. When the model is being solved, iteration occurs over
time points and the LM tries at each point to solve the flux equa-
tion iteratively as well. If an inappropriate initial guess value of
the water flux is given as argument to the LM, it will take more
time to converge to the expected solution at each time step. Since
the LM uses the same initial guess value at each time point, the
simulation time may increase considerably as its value is equiv-
alent to the sum of the convergence time corresponding to each
time point. To avoid this situation, the water flux equation has
been solved with an initial guess value equal to the maximum ex-
perimental water flux attained by the membrane used in the ex-
perimental study. By considering this approach, the water flux dy-
namic simulation with the LM algorithm only takes between 0.02
and 0.035 s.
The problem solving process consists in calculating at first
densities, viscosities, osmotic pressures and other derived prop-
erties, taking into account concentration, flowrate, and tempera-
ture. The parameters involved in the model, namely Reynolds and
Schmidt numbers, are then evaluated before calculating the Sher-
wood number on both sides of the FO membrane. These param-
eters are used to establish the mass transfer coefficient and diffu-
Fig. 5. Algorithm for solving the set of Eqs. (1)–(30) governing the FO process dy-
sion coefficient of solute ions. The respective values of these coeffi-
namics.
cients are subsequently introduced into the water flux and the RSF
expressions. Once the parameters initialization step is finished, the
LM algorithm is then run to solve the Eq. (1) for water flux. The
calculated Jw value is subsequently used in Eq. (2) to evaluate the 5. Results and discussion
reverse solute flux Js . The calculated values of Jw and Js are then
fed-back into the material balance equations in order to evaluate 5.1. Experimental results
the new values of flowrates and concentrations for both sides of
the membrane. All these calculation steps are repeated within a The measured variables are feed and draw solutions weights
For Loop indexed by time step, with the properties and parame- and their electrical conductivities, flowrates and temperatures. To
ters updated at each iteration according to the concentrations and convert electrical conductivities to concentrations for both mem-
flowrates trends. The overall algorithm structure is illustrated in brane cell sides, two correlations are obtained using data adopted
Fig. 5. After obtaining water flux value at each time step, the re- from (Haynes, 2014) with an R² equal to 0.9264 and 0.9986 for
verse salt is deduced using Eq. (2). The two fluxes are then in- MgCl2 and NaCl, respectively. The resulting equations for concen-
troduced into balance Eqs. (25 to 30). Solving these set of equa- tration expressed in weight percentage (%wt.) of the solution are:
tions allows the evaluation of concentrations and flowrates at each
time step. The obtained values are then reintroduced into the os- CMgCl2 (%wt ) = 0.0193 σ 1.421 (31)
motic pressure equation, mass transfer coefficient and diffusion co-
efficient expressions to establish their instantaneous values in the
water flux equation. CNaCl (%wt ) = 2 10−6 σ 3 − 4 10−4 σ 2 + 0.1004 σ − 0.4809 (32)

7
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Fig. 6. Experimental water fluxes (a) and reverse solute flux (b) using MgCl2 as draw solute.

Fig. 7. Experimental water fluxes (a) and reverse solute flux (b) using NaCl as draw solute.

Where σ is the measured electrical conductivity expressed in the draw to the feed side) exclusively by diffusion phenomena,
(mS/cm) and C is the corresponding mass concentration expressed resulting in an increasing salt concentration in the feed solution.
in weight percentage (%wt ). These two phenomena, known in forward osmosis as dilutive and
The concentration values calculated with Eqs. (31) and (32) are concentrative concentration polarizations, tend to reduce the salt
used to calculate the experimental water flux and reverse solute concentration difference between feed and draw side of the mem-
flux during the operating period. The water flux Jw is deduced from brane. This results in less osmotic pressure difference and subse-
the mass change of the DS tank over time. Its value is calculated quently less water flux and RSF.
using Eq. (33):
m 5.2. Model validation for both MgCl2 and NaCl
Jw = (33)
ρds Am t
Where m is the DS mass change over the test period t, Am is To evaluate the performance of the dynamic model, a compari-
the membrane area and ρd is the feed solution density as a func- son between predicted and experimental values was performed for
tion of salt concentration. a period of thirty hours of operation. This comparison, presented
The reverse solute flux Js is measured through the electrical in Fig. 8, is based on predicted and experimental values of water
conductivity variation overtime of the feed solution. It is evaluated flux and RSF for the MgCl2 and NaCl draw solution. These graphs
with the help of the following equation (Tiraferri et al., 2013) : show a good accordance between calculated and experimental val-
ues, with correlation coefficients (R2 ) close to unity. This indicate
CF, t+1 (VF, t+1 − Jw,t Am t ) − CF,t VF, t
Js = (34) a satisfactory model correlation.
Am t The performance of the dynamic model was also tested by plot-
Where CF,t and VF,t are respectively the initial and final feed ting the residual distribution as a function of each predicted values
concentration and volume, Jw,t is the water flux at time t; Am is of water flux and RSF for both MgCl2 and NaCl draw solution. Fig. 9
the membrane area while t is the time step. Since both water illustrate the four comparisons, showing that points are distributed
flux and RSF have a very slow dynamics and their values are eval- randomly around the zero axis. Errors are found between [−2, 2]
uated hourly, the time step is taken equal to one hour. and [−1.5, 1.5] for the water flux of MgCl2 and NaCl, respectively,
The time evolution of the water flux and RSF obtained experi- and between [−0.8, 0.8] for the RSF of both DS. Fig. 8 also shows
mentally for the MgCl2 and NaCl DS are reported in Figs. 6 and 7, that values are found on both sides of the normal, with resid-
respectively. These figures show that both water flux and RSF de- ual distribution confirming the performance of the dynamic model.
crease according to the osmotic difference trend across the mem- The accuracy of this dynamic model will be assessed and analyzed
brane. As time evolves, the water flux crossing the semi-permeable through the most commonly used criterion for each comparison
membrane causes DS dilution. At the same time, a small quan- which are the Root Mean Square Error (RMSE) and the Mean Ab-
tity of salt crosses the membrane in the reverse direction (from solute Error (MAE). These are defined, respectively by Eqs. (35)

8
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Fig. 8. Comparison between predicted and experimental values of water flux (a) and RSF (b) for the two draw solution.

Fig. 9. Residual distribution, for the two-draw solutes.

9
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Table 4 estimate the osmotic pressure of both draw and feed solutions. As
Reliability of dynamic model, based on R2 , MAE, ME and RMSE, in the case of 1.0 M
this model is accurate only at lower concentration, this lead to in-
for the two draw solution.
creasingly larger error. Another reason which contributes to accu-
MgCl2 NaCl racy loss is the impact that the draw solute concentration has on
Jw (L/m²h) Js (g/m²h) Jw (L/m²h) Js (g/m²h)
membrane parameters.
R2 0.958 0.937 0.977 0.945 The model performance depends mainly on the osmotic pres-
MAE (%) 5.247 5.635 4.245 5.045 sure generated by the draw solute. Its value should thus be esti-
ME (%) 11.938 11.486 10.618 12.342
mated adequately by the Van’t Hoff equation. The bulk concerta-
RMSE (L/m² h, g/m²h) 0.774 0.334 0.808 0.397
tion of both draw and feed side solutions of the FO cell has been
considered homogeneous. The impact of the draw solute concen-
tration on membrane parameters was neglected within the con-
and (36).
centration range up to 2.0 M. In fact, the model prediction errors

n 
 2 shown to be acceptable up to this concentrations despite these re-
1
MRSE = Vexp,i−Vcal,i (35) sults.
n
i=1
  6. Performance of the two draw solutions
1  Vexp,i−Vcal,i 
n
MAE = (36)
n Vexp,i 6.1. Dynamic performance of MgCl2 and NaCl
i=1

Where, n is the number of data points in the entire data set; The performance analysis of the two draw solution for FO de-
Vcal and Vexp are respectively the calculated and experimental val- salination process must be conducted using a reliable dynamic
ues. model. The DS osmotic performances have been extensively an-
Table 4 summarize for both DS the MRSE, MAE, Maximum Er- alyzed in the literature in steady state (Cornelissen et al., 2011;
ror (ME) and R2 obtained. These results demonstrate the dynamic Hancock and Cath, 2009; Ge et al., 2013), but dynamic perfor-
model’s reliability to predict water and RSF fluxes for these two mances have not been studied. In order to evaluate and compare
DS. The MAE of water flux over the whole simulation period is the dynamic osmotic performance of both MgCl2 and NaCl, the
estimated at around 5.25% for the two draw solution while their specific reverse salt flux ( Js /Jw ) and its evolution with time as well
ME does not exceed 11.94%. The RSF, its MAE for MgCl2 and NaCl as with concentrations considered as a key criterion.
is around 5.63% while the ME exceeds 12.34%. These errors, while For the sake of simplification and clarity, the water flux and RSF
slightly greater than the water flux error, remain within acceptable dynamics of both MgCl2 and NaCl are simulated only for three con-
range. Many error sources contribute to the global difference be- centration values, i.e. 0.5 M, 1.0 M and 1.5 M. The impact of the
tween model predictions and experimental results, including the DS on Jw and Js will be analyzed further. The results for the three
experiment itself, the correlation for converting conductivity to aforementioned concentrations are depicted in Figs. 10.a and 10.b
concentration, the error incurred when using the Van’t Hoof equa- for water flux and reverse solute flux respectively. These figures
tion and the error induced by the experimental setup used to es- show that an increased DS concentrations yield higher water and
tablish the membrane intrinsic parameters. It can be concluded RS fluxes. Both fluxes decrease however with time as the FO driv-
from the statistical analysis that this dynamic model gives very ing force decreases according to the dilution rate of the DS. Salt
satisfactory results in terms of precision and reliability. back-diffusion, which is mostly driven by diffusion from the draw
side to the feed side, depends on DS concentration. So as the DS
5.3. Impact of the draw concentration becomes progressively diluted, the salt back-diffusion tends to be
less intense over time.
To evaluate the impact of the draw solute’s initial concentra- Throughout the considered period, the water flux obtained with
tion on the model performance, NaCl was tested at 2.0 M and MgCl2 is less important than when using NaCl (Fig. 10.a). MgCl2
3.0 M. The experimental values were then compared to predic- seems however to be more suitable in terms of RSF (Fig. 10.b).
tions. Table 5 summarize the MAE and maximum error. Comparing As NaCl has a smaller molecular size, its ionic diffusion capacity
measured water flux and RSF shows that as the initial draw solute is higher, resulting in a more significant salt back-diffusion rate
concentration increases, the difference between the experimental than MgCl2 over the whole simulation period and for all consid-
fluxes (Water flux and RSF) become increasingly significant. When ered concentrations (Fig. 10.b).
the draw solute concentration goes from 1.0 M to 2.0 M, the MAE Simulation results show that the water flux with NaCl DS is
of water flux increases from 4.245% to 9.755% while RSF increases on average slightly higher than with MgCl2 . The DS cannot how-
from 5.045% to 8.314%. The maximum error for water flux shows ever be compared only through the achieved water flux, as this
an increase tendency from 10.618% to 14.272% and a change from is influenced by several other parameters which include RSF and
12.342% to 14.269% is observed in terms of RSF. At 3.0 M, the MAE molecular size. So even though NaCl allows for higher water flux,
is comparable to its value at 2.0 M; however, the maximum error its RSF is higher as well. This is undesirable in FO because it sig-
becomes very significant. nificantly impacts process performance through the ECP. In the
The model loses its accuracy with increasing draw solute con- present work, this is demonstrated for a transient operation of the
centration, mostly because of the use of the Van’t Hoff model to FO process and three initial DS concentration values. Throughout
the whole simulation period, MgCl2 perform better with respect to
water flux. For the whole concentration range considered, MgCl2
Table 5
Impact of the initial draw solute (NaCl) concentration on the model performance. presents a lower salt back-diffusion rate. The achieved time aver-
aged water flux value for the three NaCl DS initial concentrations is
2.0 M NaCl 3.0 M NaCl
respectively 7.23%, 8.59% and 9.28% greater than with MgCl2 . NaCl
Jw (L/m²h) Js (g/m²h) Jw (L/m²h) Js (g/m²h)
presents however a greater RSF due to the capability of its ions
MAE (%) 9.755 8.314 9.757 9.583 to diffuse easily. Simulations conducted for the three DS concen-
Max Error (%) 14.272 14.269 27.492 23.937
RMSE (L/m²h, g/m²h) 1.997 0.701 3.153 1.477
trations of MgCl2 and NaCl show that RSF values obtained with
MgCl2 are 21.59%, 22.68% and 23.28% less than for NaCl. The spe-

10
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Fig. 10. Model simulated water flux (a) and RSF (b) of the two draw solutes and for three concentrations, 0.5 M, 1.0 M, and 1.5 M.

Fig. 11. Evolution of the time averaged water flux and the RSF with draw solute concentration (a), and the variation of the MRD with concentration for both Jw and Js (b).

cific RSF ratio was evaluated for both MgCl2 and NaCl and was found to be independent of the concentration. Its value is thus the
found to be constant during the entire simulation period regard- same as in the previous section.
less of time. The specific RSF is equal to 0.377 g/L and 0.446 g/L It can be concluded from these results on specific RSF ratio that
for MgCl2 and NaCl, respectively. This implies that both water and the osmotic performances of MgCl2 are more suitable to drive wa-
RSF flux decrease at the same rate with time. ter flux in both steady state and dynamic evolution. This despite
that MgCl2 exhibits a lesser water flux compared to NaCl. But as
6.2. Evolution of MgCl2 and NaCl performance with concentration its RSF within the range 0 to 2.0 M is lower than NaCl by a large
margin during the entire simulation period, this implies that the
To get a deeper insight into the behavior of the osmotic perfor- perceived water flux-increasing propensity is more significant than
mance of both MgCl2 and NaCl, their water flux and RSF have been that of the RSF in the case of MgCl2 compared to NaCl.
compared for thirteen DS concentration within the range 0.15 M to
2.0 M, with a step change of 0.15 M. Evolution of the time averaged 7. Conclusion
water flux and RSF is shown in Fig. 11.a. For the entire DS con-
centration range, NaCl exhibits a higher water flux and RSF than From a literature survey, few studies have considered the inves-
MgCl2 . As shown in Fig. 11.a, the water flux obtained with MgCl2 tigation of FO dynamics, especially for water desalination applica-
for the entire considered concentrations range is on average 8.23% tions. In this work, a simple yet robust FO dynamic model has been
lower than when NaCl is used. The average reverse salt flux for the derived from an existing model selected among those available
same concentration range of MgCl2 is lower by 28.87% than with on the basis that it was experimentally validated. This model was
NaCl. These differences in water flux and RSF between MgCl2 and subsequently improved and adapted to water desalination applica-
NaCl increase with higher concentration, as shown by the evolu- tion, before been validated through experiments with both MgCl2
tion of Jw and Js with concentration (Fig. 11.b). Their values are and NaCl. The derived model involves the time evolution of water
quantified through the mean relative differences (MRD) at each flux as well as reverse solute fluxes, taking into account the inter-
concentration point. The difference in water flux increases from nal and external concentration polarizations of asymmetric mem-
4.84% to 9.70% over the considered range, while RSF increases from branes. The transient fluxes model, in combination with all the
24.27% to 30.95% for the same concentration interval. appropriate equations and correlations, was solved using LM algo-
Comparing MgCl2 to NaCl by using separately water flux, RSF rithm for two DS. Validation of the FO model dynamics was con-
and their evolution with concentration does not allows for proper ducted for each DS using two experiments carried out with 1 M.
decision on which is more suitable to drive water permeation The predicted water and RSF fluxes are validated against experi-
across the FO membrane. The specific RSF (Js /Jw ) is used as key mental data obtained using a Sterlitech SEPA test Bench and CTA
criterion to distinguish between the two DS in terms of osmotic flat sheet membrane provided by FTSH2O. This validation was per-
performances for each concentration value. A larger specific RSF formed for a 30 hours’ period of operation. The difference in water
ratio reflects a poor membrane selectivity for a given DS. The spe- flux between model predictions and experimental data for MgCl2
cific RSF value has been calculated for each concentration and was and NaCl was respectively found to be 5.27% and 4.95%, expressed

11
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

in terms of MAE. Such difference was deemed to be reasonably ac- Deshmukh, A., Yip, N.Y., Lin, S., Elimelech, M., 2015. Desalination by forward osmo-
ceptable for both DS as the magnitude of MAE does not exceeds sis: identifying performance limiting parameters through module-scale model-
ing. J. Memb. Sci. 491, 159–167. doi:10.1016/j.memsci.2015.03.080.
6% for NaCl, and 5% for MgCl2 . The model is therefore considered Field, R.W., Wu, J.J., 2013. Mass transfer limitations in forward osmosis: are some
suitable for predicting both water flux and RSF dynamics. Its preci- potential applications overhyped? Desalination 318, 118–124. doi:10.1016/j.desal.
sion is however lost as the DS concentration increases. The model 2013.01.025.
Field, R.W., Wu, J.J., 2018. On boundary layers and the attenuation of driving forces
prediction error for both water flux and RSF becomes increasingly in forward osmosis and other membrane processes. Desalination 429, 167–174.
important with higher DS concentration because of rising inaccu- doi:10.1016/j.desal.2017.12.001.
racy in osmotic pressure prediction using Van’t Hoff equation. Gadêlha, G., Gadêlha, H., Hankins, N., 2015. Water flux dynamics in closed-loop,
batch-mode forward osmosis systems. J. Memb. Sci. 476, 457–468. doi:10.1016/
The model developed has been used to perform sensitivity anal-
j.memsci.2014.11.056.
ysis of the DS concentration on water flux and RSF. The model Gao, Y., Wang, Y.N., Li, W., Tang, C.Y., Characterization of internal and external con-
was also used to compare osmotic performances of MgCl2 against centration polarizations during forward osmosis processes, Desalination. 338
(2014) 65–73. https://doi.org/10.1016/j.desal.2014.01.021.
NaCl. For an initial DS ranging from 0.15 M to 2.0 M, NaCl yields
Ge, Q., Ling, M., Chung, T.S., 2013. Draw solutions for forward osmosis processes:
a higher water flux than MgCl2 , with a relative difference in water developments, challenges, and prospects for the future. J. Memb. Sci. 442, 225–
flux between MgCl2 and NaCl varying from 4.84% to 9.7% (8.23% 237. doi:10.1016/j.memsci.2013.03.046.
on average). RSF when using MgCl2 is, on the other hand, much Gulied, M., Al Nouss, A., Khraisheh, M., AlMomani, F., 2020. Modeling and simu-
lation of fertilizer drawn forward osmosis process using Aspen Plus-MATLAB
lower than NaCl. This relative difference in RSF goes from 24.27% to model. Sci. Total Environ. 700, 134461. doi:10.1016/j.scitotenv.2019.134461.
30.95%, averaging 28.87%. The specific RSF ratio (Js /Jw ) was found Hancock, N.T., Cath, T.Y., 2009. Solute coupled diffusion in osmotically driven mem-
to be independent of both time and concentration, with a value brane processes. Environ. Sci. Technol. 43, 6769–6775. doi:10.1021/es901132x.
Haynes, W.M., 2014. Handbook of Chemistry and Physics, 95th ed. CRC Press.
estimated at 0.377 g/L for MgCl2 and at 0.446 g/L for NaCl. As the Johnson, D.J., Suwaileh, W.A., Mohammed, A.W., Hilal, N., 2018. Osmotic’s potential:
specific RSF of NaCl is higher than MgCl2 , the effects of water flux an overview of draw solutes for forward osmosis. Desalination 434, 100–120.
increase on the RSF are more noticeable with NaCl than MgCl2 . doi:10.1016/j.desal.2017.09.017.
Kim, B., Gwak, G., Hong, S., 2017. Review on methodology for determining forward
This work analyzed in a dynamic manner the osmotic perfor- osmosis (FO) membrane characteristics: water permeability (A), solute perme-
mances of both MgCl2 and NaCl at various concentration. Results ability (B), and structural parameter (S). Desalination 422, 5–16. doi:10.1016/j.
have shown that MgCl2 performs in overall better than NaCl, even desal.2017.08.006.
Laliberté, M., 2007. Model for calculating the viscosity of aqueous solutions. J. Chem.
though NaCl exhibits a higher water flux. The specific RSF ratio,
Eng. Data 52, 321–335. doi:10.1021/je0604075.
which was shown to be stable with time and concentration vari- Lee, W., Kang, P.K., Kim, A.S., Lee, S., 2018. Impact of surface porosity on water flux
ation, indicates that the tradeoff between water flux and RSF is and structural parameter in forward osmosis. Desalination 439, 46–57. doi:10.
1016/j.desal.2018.03.027.
more advantageous when using MgCl2 as DS to drive water per-
Lee, W.J., Ng, Z.C., Hubadillah, S.K., Goh, P.S., Lau, W.J., Othman, M.H.D., Ismail, A.F.,
meation across the FO membrane as the NaCl RSF is higher than Hilal, N., 2020. Fouling mitigation in forward osmosis and membrane distillation
for MgCl2 . for desalination. Desalination 480. doi:10.1016/j.desal.2020.114338.
Li, X., He, T., Dou, P., Zhao, S., 2017. 2.5 Forward osmosis and forward osmosis mem-
branes. Compr. Membr. Sci. Eng. 2, 95–123. doi:10.1016/b978- 0- 12- 409547- 2.
Declaration of Competing Interest 12266-8.
Matott, L.S., Leung, K., Sim, J., 2011. Application of MATLAB and python optimiz-
ers to two case studies involving groundwater flow and contaminant transport
The authors declare that they have no known competing finan-
modeling. Comput. Geosci. 37, 1894–1899. doi:10.1016/j.cageo.2011.03.017.
cial interests or personal relationships that could have appeared to McCutcheon, J.R., Elimelech, M., 2006. Influence of concentrative and dilutive inter-
influence the work reported in this paper. nal concentration polarization on flux behavior in forward osmosis. J. Memb.
Sci. 284, 237–247. doi:10.1016/j.memsci.2006.07.049.
MENEELY, G.R., DAHL, L.K., 1961. Electrolytes in hypertension: the effects of sodium
Acknowledgements chloride. The evidence from animal and human studies. Med. Clin. North Am.
45, 271–283. doi:10.1016/S0025-7125(16)33891-3.
The authors would like to thank IRESEN, the Research Institute Munubarthi, K.K., Gautam, D.K., Anki Reddy, K., Subbiah, S., Distributed param-
eter system modeling approach for the characterization of a high flux hol-
for Solar Energy and New Energies for the continuous funding. low fiber forward osmosis (HFFO) membrane, Desalination. 496 (2020) 114706.
https://doi.org/10.1016/j.desal.2020.114706.
Nagy, E., 2019. Forward osmosis, basic equations mass. Transp. Through Membr.
Funding
Layer 447–456. doi:10.1016/b978- 0- 12- 813722- 2.0 0 017-0.
Novotný, P., Söhnel, O., 1988. Densities of binary aqueous solutions of 306 inorganic
This work was supported by the Institut de Recherche en En- substances. J. Chem. Eng. Data 33, 49–55. doi:10.1021/je0 0 051a018.
ergie Solaire et Energies Nouvelles (IRESEN), Rabat, under InnoPro- Qasim, M., Mohammed, F., Aidan, A., Darwish, N.A., 2017. Forward osmosis desali-
nation using ferric sulfate draw solute. Desalination 423, 12–20. doi:10.1016/j.
jet Series. desal.2017.08.019.
Ruprakobkit, T., Ruprakobkit, L., Ratanatamskul, C., 2016. Carboxylic acid concentra-
References tion by forward osmosis processes: dynamic modeling, experimental validation
and simulation. Chem. Eng. J. 306, 538–549. doi:10.1016/j.cej.2016.07.091.
Arjmandi, M., Peyravi, M., Altaee, A., Arjmandi, A., Pourafshari Chenar, M., Jahan- Ruprakobkit, T., Ruprakobkit, L., Ratanatamskul, C., 2017. Dynamic modelling of
shahi, M., Binaeian, E., 2020. A state-of-the-art protocol to minimize the internal carboxylic acid filtration in forward osmosis process: the role of membrane
concentration polarization in forward osmosis membranes. Desalination, 114355 CO2permeability. Comput. Chem. Eng. 98, 100–112. doi:10.1016/j.compchemeng.
doi:10.1016/j.desal.2020.114355. 2016.10.017.
Cai, Y., Hu, X.M., 2016. A critical review on draw solutes development for forward Ruprakobkit, T., Ruprakobkit, L., Ratanatamskul, C., 2019. Sensitivity analysis tech-
osmosis. Desalination 391, 16–29. doi:10.1016/j.desal.2016.03.021. niques for the optimal system design of forward osmosis in organic acid recov-
Chakrabortty, S., Pal, M., Roy, M., Pal, P., 2015. Water treatment in a new flux- ery. Comput. Chem. Eng. 123, 34–48. doi:10.1016/j.compchemeng.2018.12.024.
enhancing, continuous forward osmosis design: transport modelling and eco- Ryu, H., Mushtaq, A., Park, E., Kim, K., Chang, Y.K., Han, J.I., 2020. Dynamical mod-
nomic evaluation towards scale up. Desalination 365, 329–342. doi:10.1016/j. eling of water flux in forward osmosis with multistage operation and sensitiv-
desal.2015.03.020. ity analysis of model parameters. Water (Switzerland) 12, 1–20. doi:10.3390/
Chen, G.Q., Artemi, A., Lee, J., Gras, S.L., Kentish, S.E., 2019. A pilot scale study on w12010031.
the concentration of milk and whey by forward osmosis. Sep. Purif. Technol. Shaffer, D.L., Werber, J.R., Jaramillo, H., Lin, S., Elimelech, M., 2015. Forward osmosis:
215, 652–659. doi:10.1016/j.seppur.2019.01.050. where are we now? Desalination 356, 271–284. doi:10.1016/j.desal.2014.10.031.
Community, SPy, 2013. SciPy Reference Guide 0.7 1229. https://docs.scipy.org/doc/ Su, J., Chung, T.S., Helmer, B.J., de Wit, J.S., Understanding of low osmotic efficiency
scipy/reference/generated/scipy.optimize.least_squares.htm. in forward osmosis: experiments and modeling, Desalination. 313 (2013) 156–
Cornelissen, E.R., Harmsen, D.J.H., Beerendonk, E.F., Qin, J.J., Kappelhof, J.W.M.N., 165. https://doi.org/10.1016/j.desal.2012.12.022.
2011. Effect of draw solution type and operational mode of forward osmosis Tan, C.H., Ng, H.Y., 2008. Modified models to predict flux behavior in forward os-
with laboratory-scale membranes and a spiral wound membrane module. J. Wa- mosis in consideration of external and internal concentration polarizations. J.
ter Reuse Desalin. 1, 133–140. doi:10.2166/wrd.2011.042. Memb. Sci. 324, 209–219. doi:10.1016/j.memsci.2008.07.020.
Cussler, E., 2007. Diffusion, Mass Transfer in Fluid Systems, Third Edit. Cambridge Tanaka, S., Kastens, S., Fujioka, S., Schlüter, M., Terasaka, K., 2020. Mass transfer from
University Press, New York www.cambridge.org/9780521871211. freely rising microbubbles in aqueous solutions of surfactant or salt. Chem. Eng.
J. 387, 121246. doi:10.1016/j.cej.2019.03.122.

12
J. Eddouibi, S. Abderafi, S. Vaudreuil et al. Computers and Chemical Engineering 149 (2021) 107313

Tiraferri, A., Yip, N.Y., Straub, A.P., Romero-Vargas Castrillon, S., Elimelech, M., 2013. Yokozeki, A., 2006. Osmotic pressures studied using a simple equation-of-state and
A method for the simultaneous determination of transport and structural pa- its applications. Appl. Energy 83, 15–41. doi:10.1016/j.apenergy.2004.10.015.
rameters of forward osmosis membranes. J. Memb. Sci. 444, 523–538. doi:10. Zhou, X., Gingerich, D.B., Mauter, M.S., 2015. Water treatment capacity of forward-
1016/j.memsci.2013.05.023. osmosis systems utilizing power-plant waste heat. Ind. Eng. Chem. Res. 54,
Van Gauwbergen, D., Baeyens, J., Creemers, C., 1997. Modeling osmotic pressures for 6378–6389. doi:10.1021/acs.iecr.5b00460.
aqueous solutions for 2-1 and 2-2 electrolytes. Desalination 109, 57–65. doi:10.
1016/S0011-9164(97)0 0 056-8.
Wilson, A.D., Stewart, F.F., 2013. Deriving osmotic pressures of draw solutes used
in osmotically driven membrane processes. J. Memb. Sci. 431, 205–211. doi:10.
1016/j.memsci.2012.12.042.

13