Theoretical Modeling and Simulation PDF
Theoretical Modeling and Simulation PDF
Theoretical Modeling and Simulation PDF
Authors Im, Baek-Gyu; Lee, Jung Gil; Kim, Young-Deuk; Kim, Woo-Seung
Citation Im B-G, Lee J-G, Kim Y-D, Kim W-S (2018) Theoretical modeling
and simulation of AGMD and LGMD desalination processes using
a composite membrane. Journal of Membrane Science 565: 14–
24. Available: http://dx.doi.org/10.1016/j.memsci.2018.08.006.
DOI 10.1016/j.memsci.2018.08.006
Publisher Elsevier BV
Rights NOTICE: this is the author’s version of a work that was accepted
for publication in Journal of Membrane Science. Changes
resulting from the publishing process, such as peer review,
editing, corrections, structural formatting, and other quality
control mechanisms may not be reflected in this document.
Changes may have been made to this work since it was submitted
for publication. A definitive version was subsequently published
in Journal of Membrane Science, [, , (2018-08-07)] DOI: 10.1016/
j.memsci.2018.08.006. © 2018. This manuscript version is
made available under the CC-BY-NC-ND 4.0 license http://
creativecommons.org/licenses/by-nc-nd/4.0/
membrane
a
Department of Mechanical Engineering, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763,
Republic of Korea
b
King Abdullah University of Science and Technology (KAUST), Water Desalination and Reuse Center (WDRC),
Biological and Environmental Science and Engineering Division (BESE), Thuwal 23955-6900, Saudi Arabia
c
Department of Mechanical Engineering, Hanyang University, 55 Hanyangdaehak-ro, Sangnok-gu, Ansan,
Corresponding author. E-mail address: [email protected] (W.-S. Kim); Tel.: +82-31-400-5248; Fax: +82-
31-418-0153
Abstract
Most previous studies of air- and liquid-gap membrane distillation (AGMD and LGMD) processes using a
composite membrane have been focused on an experimental approach. In this paper, rigorous theoretical
investigations of the AGMD and LGMD processes were performed with a flat sheet type module using a
composite membrane comprised of a polytetrafluoroethylene (PTFE) active layer and a polypropylene (PP)
support layer. The model predictions were verified by comparing with measured data, where good agreement
between the prediction results and experimental data was obtained. It was observed that as the gap size increased
the AGMD permeate flux decreased exponentially with increased diffusion resistance. On the other hand, the
LGMD permeate flux decreased exponentially and then increased asymptotically after attaining a minimum at a
certain liquid-gap size (5 − 7 mm). This phenomenon was due to the onset and enhancement of a natural
convection, resulting in an improvement in heat and mass transfer in the liquid gap.
Keyword: Air-gap membrane distillation, Liquid-gap membrane distillation, Desalination, Heat and mass
Membrane distillation (MD) is a thermally-driven separation process in which only water vapors or other
volatile molecules move through porous hydrophobic membranes. Compared to conventional desalination
processes such as multi-stage flash (MSF), multi-effect distillation (MED), and reverse osmosis (RO), several
core advantages of the MD process are; (i) low sensitivity to salt concentration, (ii) theoretically 100% salt
rejection, and (iii) low operating temperature and pressure [1,2]. Such a MD process can be classified into: (i)
direct-contact membrane distillation (DCMD), (ii) air-gap membrane distillation (AGMD), (iii) vacuum
Among the MD configurations, DCMD has been the most widely studied due to its simple configuration, the
capability to produce a relatively high flux, and it has no additional equipment such as an external condenser to
condense water vapor, which is required in VMD and SGMD [3]. However, the DCMD process has the
drawback of a relatively higher heat loss over other MD configurations [4]. The heat loss is mainly caused by the
heat conduction through a single hydrophobic membrane as well as the heat losses to the ambient through the
module structure. Reduced heat loss in the MD process can be achieved by (i) increasing the pore size, porosity
and thickness of the membrane, and (ii) decreasing the thermal conductivity of the membrane. However, the
latter cannot be fulfilled as most hydrophobic polymer materials have very similar thermal conductivities.
Meanwhile, in the former case, there exists a conflict between low conductive heat loss (with a high membrane
thickness) and high mass transfer (with a low membrane thickness). In conclusion, it is difficult for a single
hydrophobic membrane to meet the requirements of an ideal membrane for the MD process, i.e., high mass
transfer and low conduction heat loss, with a single hydrophobic membrane [5]. Therefore, in order to avoid the
membrane that consists of a thin active layer for high mass transfer and a thick support layer for low conductive
heat loss can be employed. Most studies on composite membranes have been experimentally carried out [5‒11],
while a few papers based on theoretical approaches have been reported [5,10,11]. Furthermore, the
aforementioned theoretical researches were mainly focused on the DCMD process using a composite membrane
and the developed models were based on a length-averaged lumped model, which disregards spatial variations of
The DCMD module configuration consists of two stream flows separated by a thin membrane with a
hydrophobic nature. However, this configuration is vulnerable to conductive heat losses. Among all of the
possible MD processes, one of the alternatives to cope with conductive heat loss is the AGMD, which constitutes
an air gap between the membrane and cooling plate, resulting in lower conductive heat loss and a higher thermal
efficiency. In addition, if a low temperature feed is used as a cooling stream, the latent heat can be recovered
through the condensation of the vapors on the cooling plate [3]. On the contrary, AGMD has a lower permeate
flux compared to DCMD due to the high mass transfer resistance in the air gap [13]. In this regard, liquid-gap
membrane distillation (LGMD) has been suggested to overcome the lower permeate flux and higher heat loss in
AGMD and DCMD, respectively. LGMD utilizes a liquid gap between the membrane and cooling plate, which
is typically filled with a liquid. Recently, Francis et al. [14] attempted to fill the gap with different types of
materials, such as a sponge (polyurethane), DI water, sand, and polypropylene mesh. It was found that the
permeate flux of LGMD is higher than that of AGMD at the same operating conditions. Furthermore, an
improved LGMD permeate flux was achieved due to an enhanced natural convection in the liquid gap with
increasing liquid gap thickness from 9 mm to 13 mm. Essalhi et al. [15] also carried out an experimental
investigation of AGMD and LGMD using a composite membrane. The permeate flux of LGMD was slightly
higher (2.2 − 6.5%) than that of AGMD at the feed inlet temperatures of 35 – 80 °C and an air/liquid gap of
4.532 mm. Therefore, there is a dearth of literature for both comprehensive experimental and theoretical studies
In this study, theoretical investigations were performed to attain more comprehensive and systematic
understanding of the heat and mass transfer phenomena in the AGMD and LGMD processes using a commercial
hydrophobic microporous composite membrane with an active layer of polytetrafluoroethylene (PTFE) and a
scrim-backing support layer of polypropylene (PP). The main objective of this study was to develop the
integrated numerical analysis model of AGMD and LGMD processes, which consists of mass, momentum,
species, and energy balances of both the bulk feed and coolant flows coupled with the heat and mass transfer of
water vapor through the membrane and air/liquid gap. The theoretical model developed in this study was
validated with experimental data obtained from the literature [14,16]. Further studies were carried out to identify
the effects of the feed and coolant temperatures, the feed and coolant flow rates, surface porosity, and natural
2. Theoretical models
The mathematical models for the AGMD and LGMD processes were formulated to include heat and mass
transfers. In both processes, a flat-sheet composite membrane is in direct contact with the feed stream, and an air
or liquid gap is interposed between the membrane and a condensation surface while a coolant flows in the
direction countercurrent to the feed. The mass and heat transfer mechanisms in the AGMD and LGMD processes
are illustrated in Fig. 1. For the simulation, the following assumptions are employed: (i) steady-state
incompressible flow, (ii) negligible heat loss to the ambient environment, (iii) only water vapor permeation
through membrane pores, (iv) atmospheric pressure in the air gap, and (v) unidirectional Newtonian fluid flow in
feed and coolant side. The SEM images of the composite membrane used in the present work are shown in Fig. 2
[12,14,16,41]. It is shown that the knot-fibril net structured PTFE active layer (dark gray) was partially covered
Fig. 1. Schematic diagram of the heat and mass transfer mechanisms in (a) AGMD and (b) LGMD.
Fig. 2. SEM micrographs of a commercial PTFE/PP composite membrane. Clockwise from top left: 100×, 500×,
With the assumptions given above, the momentum, mass, species, and energy balances for the feed side can
be expressed in terms of pressure (Pf), velocity (vf), concentration (xf), and temperature (Tf) [1,4,17]. The
dPf 3 f
2
vf (1)
dz hch
The overall molar balance and the molar species balance for the salt species yield Eqs. (2) and (3) in terms of
1 dv f vf M M dx f J
2 s w (2)
X f dz X f s w dz M f sp hch
x f dv f v f M w dx f
0 (3)
X f dz w X 2f dz
The energy balance on the feed side yields the following equation on simplification:
d f v f c p, f T f Qf
(4)
dz sp hch
For the coolant flow, the momentum and energy balances can be expressed in terms of the pressure (Pc) and
d c vc c p ,cTc Qc
(6)
dz sp hch
Heat transfer in AGMD includes the heat transferred across the feed boundary layer (Qf), composite
membrane (Qm), air gap (Qag), condensate film (Qcf), cooling plate (Qcp), and coolant boundary layer (Qc), as
shown in Fig. 3.
The convective heat transfer across the feed boundary layer is written as follows [4,16,18]:
Q f h f T f T f ,m (7)
where Tf is the bulk feed temperature and Tf,m is the interface temperature between the feed and membrane. As
turbulence promoter and membrane supporter, a non-woven net spacer is used to enhance the heat and mass
transfer in the feed channel. In order to incorporate an enhancement of heat and mass transfer by spacers, the
spacer factor (kdc) is employed by considering the structural parameters of spacer, as given in Eqs. (9) and (10).
Therefore, the convective heat transfer coefficient, hf, in the spacer-filled feed channel can be determined by the
modified Dittus-Boelter’s correlation by incorporating the spacer factor on the flat surface of a fully-developed
h f dh
Nu 0.023kdc Re0.8 Pr 0.33 (8)
kf
0.039 0.086
d spf 0.75
kdc 1.654 sp sin (9)
hst 2
d spf
2
sp 1 (10)
2lm sp sin
Here, kf is the thermal conductivity of the feed solution, Nu is the Nusselt number, dh is the hydraulic diameter,
defined as 4 sp / 2 wch hch / wch hch 4 1 / d f , Re is the Reynolds number, Pr is the Prandtl number, df
is the filament diameter of spacer, δsp is the spacer thickness, εsp is the spacer porosity, lm is the mesh size, θ is
The overall heat transfer through the composite membrane can be calculated as follows [12]:
where Qm,a and Qm,a‒s are the heat transfers across the active-only and active/support layers, respectively. Here,
the heat transfer through the active/support layer consists only of conductive heat transfer, as the mass transfer
across the active/support layer is negligible [12]. Therefore, the heat transfers can be described as follows:
sf
Qm, a J agmd H m, a
Rm, a
T f ,m Ta (12)
1 sf
Qm,a-s
Rm,a s
T f ,m Ts (13)
where Ta and Ts are the interface temperatures between the active layer and the air gap and between the support
layer and the air gap, respectively, Jagmd is the mass flux, ∆Hm,a is the enthalpy of evaporation, and εsf is the
surface porosity, defined as the surface area of the PTFE active layer exposed to the permeate side divided by
the total membrane surface area. The thermal resistance, Rm, in Eqs. (12) and (13) can be described as follows
[12]:
a
Rm, a (14)
km, a
Rm,a s a s (15)
k m , a km , s
where km,i m,i kair 1 m,i ki , i = a, s. km,i is the effective thermal conductivity of the composite membrane
[12], kair is the thermal conductivity of air in the pores, and ki is the thermal conductivity of the membrane
material.
The heat transfers across the air gap are estimated as follows:
Qag,a-s 1 sf hag Ts Tcf (17)
where Tcf is the interface temperature between the air gap and the condensate film and the heat transfer
kag Nu
hag (18)
ag
where Nu can be calculated by considering natural convection, Eq. (19), or conduction, Eq. (20), depending on
1/9 11 L / ag 42
L
Nu L
0.197 Ra1/4
, 0.5 Pr 2 (19)
ag
2 103 RaL 2 105
with RaL
g Ts Tcf ag
3
Pr (21)
2
In the above equations, RaL is the Rayleigh number, g is the gravitational constant, β is the volume expansion
Thus, the overall heat transfer through the air gap is given as shown below.
Also, the heat transfer across the condensate film is determined by the following equation [4,13,16,24]:
Qcf hcf Tcf Tcp (23)
where Tcp is the interface temperature between the condensate film and the cooling plate, and the heat transfer
1/4
g cf2 Hkcf3
hcf 0.943 (24)
Lcf Tcf Tcp
The heat transfer through the cooling plate is given by the equation shown below [4,13,16,24]:
Tcp Tp
kcp
Qcp (25)
cp
where Tp is the interface temperature between the cooling plate and the coolant, kcp is the thermal conductivity of
cooling plate material, and δcp is the thickness of the cooling plate.
The heat transfer across the coolant boundary layer is expressed as follows [4,13,16,24]:
Qc hc (Tp Tc ) (26)
where Tc is the coolant temperature and hc is the heat transfer coefficient at the spacer-filled coolant channel,
which is same as that defined in Eq. (8) at the spacer-filled feed channel.
The overall heat transfer in the AGMD process at steady state can be established as follows [4].
J a sf Ca Pf ,m Pa (29)
As shown in Fig. 4, the relevant transport mechanisms for water vapor in AGMD are Knudsen diffusion and
molecular diffusion. The membrane distillation coefficient through the active-only layer can be estimated by the
1
1 1
Ca (30)
Ckn Cmol
where Ckn is the Knudsen diffusion coefficient and Cmol is the molecular diffusion coefficient. The equations for
these coefficients are given as follows [4,16,18]:
4 d p a Mw
Ckn (31)
3 a a 2 RT
Mw a DP
Cmol (32)
P Pv a a RT
where εa is the active layer membrane porosity, D is the diffusion coefficient, P is the total pressure, dp is the
membrane pore diameter, τa is the active layer membrane tortuosity, δa is the active layer membrane thickness,
and R is the gas constant. The binary diffusion coefficient D is calculated by the Fuller-Shelltler-Giddings (FSG)
T 1.75
D 1.19 104 m (33)
P
m
where Tm and Pm are the mean temperature and mean water vapor pressure, defined as Tm (T f ,m Ta ) / 2 and
Pm ( Pf ,m Pa ) / 2 , respectively.
In Eq. (32), the water vapor pressure without dissolved species in the water can be calculated using the
3816.44
Pv exp 23.1964
T 46.13
(34)
Due to the existence of dissolved species with molar concentrations at the feed side, the reduction of the vapor
pressure can be determined using Raoult’s law by assuming as an ideal solution for simplicity [4,12,28]:
Pf ,m Pv (1 x f ,s ) (35)
where xf,s is the molar fraction of the solution and Pv is the water vapor pressure calculated by Eq. (34).
Also, the mass flux across the air gap is written as follows [4,18,29]:
J ag Cag Pa Pcf (36)
where Pa is the interface pressure between the active layer and the air gap and Pcf is the interface pressure
For the stagnant non-condensable gases trapped within the membrane pores and in the air gap between the
membrane and the condensing plate, the dominant transport mechanism is molecular diffusion expressed as
follows [4,18,29].
Mw DP
Cag (37)
P Pv ag RT
Both the permeate fluxes across the active-only layer (Ja) and the air gap (Jag) are assumed to be the same and
J agmd J a J ag (38)
Based on Eq. (38), Eqs. (29) and (36) can be rearranged, respectively, as follows.
J agmd Ja J agmd J ag
Pf , m Pa , Pa Pcf (39)
sf Ca sf Ca Cag Cag
Combining the above equations yields the local permeate flux as follows.
1
1 1
J agmd
C C P f ,m Pcf (40)
sf a ag
The mean permeate flux over the membrane length L is given by the following equation [1,12,30].
L
1
Jm J agmd dz (41)
L 0
Theoretical models of LGMD are similar to those of AGMD described in the previous section. In particular,
the momentum, mass, and energy balance equations for the bulk feed and coolant flows in LGMD are the same
as those of AGMD, as defined in Eqs. (1)–(6). The main differences compared to AGMD models are the heat
and mass transfers through the composite membrane and the liquid gap, while the transport mechanisms for the
Fig. 5 illustrates the heat transfer mechanism in LGMD, which includes heat transferred across the feed
boundary layer (Qf), composite membrane (Qm), liquid gap (Qlg), condensate plate (Qcp), and coolant (Qc). In this
section, the heat transfers through the feed, condensate plate, and coolant are not described, as these are the same
The heat transfer through the composite membrane consisting of active-only and active/support layers can be
sf
Qm,a J lgmd H a
Rm,a
T f ,m Ta (42)
1 sf
Qm, a s
Rm, a s
T f ,m Ts (43)
where Qm,a and Qm,a‒s are the heat transfers across the active-only and active/support layers, respectively. Here,
Rm,a and Rm,a‒s are calculated using Eqs. (14) and (15), respectively. The total heat transfer is then given as
follows [12].
Additionally, in order to determine Ta in Eq. (42), the conductive heat transfer is considered as follows [12].
k
Qs w Ta Ts (45)
s
The heat flux across the liquid gap is written as follows [31].
The heat transfer coefficient, hlg, inside the gap filled with a liquid can be shown as follows:
klg Nu
hlg (47)
lg
where Nu can be obtained by considering natural convection or conduction, respectively, as follows [14,32,33]:
1 L / lg 40
Nu 0.046 Ra1/3
L , 1 Pr 20 (48)
10 RaL 10
6 9
The convective heat transfer coefficient in the range of 2×103 ≤ RaL < 106 is estimated by a linear
The overall heat transfer through the LGMD process at steady state can be established as follows:
Qm,a Qs (51)
The mass transfer of LGMD through the membrane has transport mechanisms such as Knudsen diffusion,
molecular diffusion, and viscous flow, as shown in Fig. 6. This phenomenon is the same as the mass transfer of
DCMD. The permeate flux through the composite membrane can be calculated as shown below [12].
J lgmd sf Clg Pf ,m Pa (52)
The membrane distillation coefficient for the active layer is calculated using the following equation [12,34]:
1
1 1
Clg C pf (53)
Ckn Cmol
where Ckn and Cmol are equal to the values calculated using Eqs. (31) and (32), respectively. Cpf is the viscous
where rp is the membrane pore radius and μ is the dynamic viscosity of water vapor.
In both the AGMD and LGMD processes shown in Fig. 1, the bulk feed and coolant streams are circulated in
a countercurrent-flow manner. Thus, the velocities, concentrations, and temperatures are known at the inlets of
the feed and coolant solutions (i.e., z = 0 and L, respectively), while the pressures are known at their outlets (i.e.,
z = L and 0, respectively). For the simulations, the boundary conditions for the velocity, pressure, concentration,
V f ,in
v f (0) v f ,in , Pf ( L) P0 , x f (0) x f ,in , T f (0) T f ,in (55)
sp hch wch
where P0 depicts the ambient atmospheric pressure at the outlets of the feed and coolant channels, and the
On the other hand, the boundary conditions for the velocity, pressure, and temperature at the coolant side are
as follows:
Vc ,in
vc ( L) vc ,in , Pc (0) P0 , Tc ( L) Tc,in (56)
sp hch wch
In this study, the thermophysical properties of water and seawater were employed from [35]. The set of
coupled ordinary differential equations (ODEs) of the AGMD and LGMD processes, i.e., Eqs. (1)‒(6), including
both bulk feed and coolant flows and the characteristics of the membrane for heat and mass transfer, were
solved simultaneously with the boundary conditions given in Eqs. (55) and (56). The unknowns (i.e., Pf (0), vc
(0), and Tc (0) at z = 0) are guessed, which converts the boundary-value problem into an initial-value problem.
In this study, Broyden’s method, which is a quasi-Newton method for finding solutions of nonlinear equations,
was employed to solve the system of ODEs [1,12,17,37,38]. The solution for each control volume proceeds
forward until z = L, and then the presumed values are updated. Thus, this procedure is iterated until the
boundary conditions, Eqs. (55) and (56), are satisfied with an absolute error below 10-6.
The numerical models developed for the AGMD and LGMD processes were evaluated for inlet feed
temperatures in the range of 40 °C to 80 °C while maintaining a constant coolant temperature and inlet feed and
coolant flow rates of 20 °C and 1.5 l/min, respectively. The feed salt concentration was kept constant at 4.2 wt%.
Fig. 7 shows the comparison of the experimental and predicted results of the permeate flux. All the experiments
were repeated to demonstrate the reproducibility of permeate fluxes measured at different feed water
temperatures, and the maximum deviations in the measured permeate fluxes were found to be less than 6% for
LGMD and 2% for AGMD, respectively [14,16]. It was shown that both the predicted and measured permeate
fluxes increased exponentially as the feed temperature increased. The model predictions were in good agreement
with the experimental results over the entire temperature range, with a maximum relative deviation of 10%.
Tables 1 and 2 show the specifications and characteristics of the spacer and composite membrane employed in
the present work, respectively. The surface porosity was confirmed to be 42% using CAD software based on the
Fig. 2. The membrane module specifications are listed in Table 3. The numerical model was assessed at various
operating conditions including the temperatures and flow rates of the feed and coolant streams and the
membrane surface porosity. Here, the feed salt concentration was assumed to be 4.2 wt%.
Table 1 Specifications of the spacer.
Material PP
Thickness, δ (m) 20 80
Porosity, (%) 70 34
The results in Fig. 8 imply that the permeate flux increases with increasing feed temperature at a coolant
temperature of 25 °C for both AGMD and LGMD processes. A significant augmentation of the MD permeate
flux was achieved at the higher feed temperatures because of the enhanced transmembrane temperature
difference, i.e., MD driving force, as demonstrated previously [14,15,36]. In the AGMD processes considered in
this study, the Nusselt number (Nu) is always unity as the Rayleigh number (Ra) calculated by Eq. (21) is less
than 2×103. As a result, only conduction heat transfer will occur in the air gap, and based on Fourier’s law of
heat conduction, the larger the air gap thickness, the higher the heat transfer resistance. Therefore, the heat loss
through the air gap in the AGMD process can be ignored, mainly due to the very low thermal conductivity of air.
Meanwhile, the AGMD permeate flux decreases greatly with increasing air gap thickness, which is attributed to
increased diffusion resistance through the air gap [13,24,25,27]. It is also shown that the LGMD permeate fluxes
are significantly higher than those of AGMD at a given gap size, regardless of the feed temperature. This is due
to the nonexistence of additional diffusion resistance through the air gap and the higher transmembrane
temperature difference in LGMD. Similar to the trends of the AGMD permeate flux, as the liquid-gap size
increases from 1 mm to 5 or 7 mm, the LGMD permeate flux also decreases over the entire temperature range.
However, the further increase of liquid gap results in an asymptotic increase of the permeate flux. This
phenomenon occurs primarily due to the enhancement of natural convection, which results in a relatively higher
heat transfer across the boundary layer in the liquid gap leading to lower temperature polarization and thus, a
regime (Ra ≥ 106) with a higher liquid-gap size, variations in the permeate flux tend to be negligible because the
influence of gap size on the permeate flux diminishes as Ra number increases [37,38]. These results agree well
with the experimental results [14] illustrated in Fig. 7, showing that the larger the liquid-gap size, the slightly
higher the permeate flux, but at higher feed temperatures above 70 °C the variance in the permeate flux is nearly
negligible.
Fig. 8. Effect of the feed temperature (Tf) in the range of 40 − 70 °C on (a) AGMD and (b) LGMD permeate
Fig. 9 illustrates the effect of the feed flow rate ranging from 0.5 l/min to 5.0 l/min on the permeate flux of
both AGMD and LGMD with an inlet feed temperature of 70 °C, while keeping the other operating conditions
constant, as before. It is shown that the LGMD permeate fluxes are always greater than those of AGMD and
both MD permeate fluxes increase asymptotically with increasing feed flow rate, especially for all of the
LGMDs and AGMD with a 1 mm air-gap size. At lower feed flow rates, the boundary layer resistance on the
feed side is representative of the overall mass transfer resistance and thus, the permeate flux was largely
enhanced with increasing flow rate. This behavior can be attributed to the heat transfer augmentation through the
feed stream, resulting in a decrease of the temperature gradient across the feed-side boundary layer yet lower
temperature polarization [1,39‒41]. However, a further increase of the feed flow rate leads to a plateau in the
MD permeate flux profile as the overall mass transfer resistance is dominated by the diffusion (or heat transfer)
resistance in the air (or liquid) gap [42]. As the flow rate increases at the air and liquid gaps of 1 mm, the mass
transfer resistances in the gaps increase from 50% to 65% and from 46% to 77%, respectively, of the total mass
transfer resistance, whereas those across the feed-side boundary layers decrease from 30% to 9% and from 46%
to 12%, respectively. Meanwhile, at higher air-gap sizes (≥ 3 mm), the diffusion resistance in the air gap highly
limits the mass transfer for the AGMD process and thus, the effect of the feed flow rate on the permeate flux is
negligibly small.
Fig. 9. Effect of the feed flow rate (Vf) in the range of 0.5 − 5.0 l/min on (a) AGMD and (b) LGMD permeate
The influence of coolant temperature in the range of 10 ‒ 30 °C on both AGMD and LGMD permeate fluxes
while the other operating conditions mentioned previously were fixed is presented in Fig. 10. It is observed that
as the coolant temperature increases, the LGMD permeate flux decreases greatly as compared to that of AGMD.
This is because the interface temperature between the active layer and liquid gap, Ta, increases proportionally
with the coolant temperature due to the higher heat transfer through the liquid gap compared to the AGMD case,
resulting in a deterioration of the transmembrane driving force obtained from Eq. (52). In the AGMD process, on
the other hand, the permeate flux is not influenced by the coolant temperature because the heat and mass transfer
resistances in the air gap are predominant yet the interface pressure between the air gap and condensate film, Pcf,
calculated by Eq. (34) is not sensitive in the low temperature range from 10 °C to 30 °C.
Fig. 10. Effect of the coolant temperature (Tc) in the range of 10 − 30 °C on (a) AGMD and (b) LGMD permeate
Fig. 11 depicts the permeate fluxes of the AGMD and LGMD processes at different coolant flow rates. In
general, at a given coolant temperature, a sufficient cooling energy should be supplied to remove the latent heat
of condensation of the permeate vapor in both processes. It should be noted that a further supply of coolant to
both membrane modules can lead to a higher pumping power as a consequence of the higher pressure drop.
However, insufficient cooling and enhanced heat transfer resistance through the coolant-side boundary layer,
induced by a low coolant flow rate, may cause an increase of the interface temperatures between the air gap and
condensate film (Tcf) and between the active layer and liquid gap (Ta), respectively, for both the AGMD and
LGMD processes. Hence, the permeate flux will decrease. Above all, the most dominant factor in the permeate
fluxes of both AGMD and LGMD is the heat and mass transfer resistances through the air and liquid gaps.
Therefore, as shown in Fig. 11, the permeate flux variation of the AGMD process with the high gap resistances
is imperceptible to the coolant flow rate considered in this study whilst the LGMD permeate flux is dependent
Fig. 11. Effect of the coolant flow rate (Vc) in the range of 0.5 − 5.0 l/min on (a) AGMD and (b) LGMD
Fig. 12 presents the permeate flux profiles with respect to the membrane surface porosity in the range of 0.1 –
0.7. As shown in Eqs. (40) and (52), the membrane surface porosity (εsf) is an important factor in determining
the MD permeate flux. It was found that with an increase of the membrane surface porosity, the permeate fluxes
increase asymptotically in both the AGMD and LGMD processes, which has already been demonstrated in the
DCMD process [12]. At larger air-gap sizes (≥ 3 mm), however, the higher mass transfer resistance through the
air gap is significantly more dominant than that of the membrane and thus, the permeate flux variations are
relatively insensitive even as the membrane surface porosity was increased [14,42]. In all cases of larger air-gap
sizes, the air-gap mass transfer resistances account for more than about 66% of the total mass transfer resistance,
while those across the membrane exhibit less than about 29%. On the other hand, the mass transfer at a 1 mm air
gap was significantly enhanced compared to that of larger air gaps, yet its magnitude becomes comparable to
that of the membrane and hence, the AGMD permeate flux is significantly influenced by the membrane surface
porosity. It is shown that as the surface porosity increases from 0.1 to 0.7 at the air gap of 1 mm, the mass
transfer resistances through the membrane decrease asymptotically from 51% to 13% of total mass transfer
resistance, whereas those in the air gap increase asymptotically from 39% to 69%.
Fig. 12. Effect of the surface porosity (εs) in the range of 0.1 − 0.7 on (a) AGMD and (b) LGMD permeate fluxes
4. Conclusions
In the present work, rigorous theoretical models of AGMD and LGMD using a hydrophobic PTFE/PP
composite membrane were developed. The developed models involve mass, momentum, species and energy
balances in the feed and coolant streams, the mass and heat transfers through the membrane and air or liquid gap,
and the physical characteristics of the composite membrane. The effects of the feed and coolant temperatures,
the feed and coolant flow rates, and the membrane surface porosity on the MD performance were evaluated at
It was noted that as the liquid gap increased the LGMD permeate flux decreased exponentially up to its
minimum at a certain liquid-gap size and then increased asymptotically. It was because the natural convection
that brings about an improvement of heat and mass transfer occurred at a certain liquid gap, and then was
enhanced with a further increase of the gap size. In general, the temperature polarization in the feed boundary
layer diminishes with an increase of the feed flow rate, leading to an increased interface temperature between the
feed and membrane and hence, the permeate flux. With larger air gaps, however, the AGMD permeate flux is
restricted by the high diffusion resistance in the air gap. The AGMD permeate flux is less affected by the coolant
temperature compared to the LGMD permeate flux. The negligible effect of the coolant temperature in AGMD is
due to the low sensitivity of the interface pressure between the air gap and condensate film at the lower
temperatures and also the heat and mass transfer resistances, which are strongly dominant in the air gap. The
effect of the coolant flow rate was marginal on the AGMD permeate flux because of the heat and mass transfer
resistances in the air gap, which are the most influential factors in the overall heat and mass transfers, although
the heat transfer resistance through the coolant-side boundary layer decreases as the coolant flow rate increases.
On the other hand, its effect on the LGMD permeate flux is relatively more significant at lower coolant flow
rates. As compared to the mass transfer resistance at larger air gaps, its magnitude at a 1 mm air gap is
comparable to that of the composite membrane. Therefore, the membrane surface porosity has a relatively large
In all cases of the AGMD process, the permeate flux exhibits a maximum at an air gap of 1 mm. As expected,
the larger permeate flux is achieved with a smaller air gap with lower mass transfer resistance. Similarly, the
LGMD permeate flux at liquid gaps of less than 5 mm reveals a maximum at a gap of 1 mm, whereas above 5
mm the permeate flux increases asymptotically, leading to a maximum at a gap size of about 9 mm to 11 mm.
Acknowledgments
This work was supported by the Korea Institute of Energy Technology Evaluation and Planning (KETEP)
funded by the Ministry of Trade, Industry & Energy (MOTIE) of the Republic of Korea (No. 20153010130460
Nomenclature
g Gravity [m/s2]
T Temperature [°C]
v Velocity [m/s]
Dimensionless numbers
Greek letters
δ Thickness [mm]
ε Porosity [-]
ρ Density [kg/m3]
τ Tortuosity [-]
Subscripts
a Active layer
ag Air gap
air Air
c Coolant
f Feed
g Gas
lg Liquid gap
p Membrane pore
s Support layer
sf Surface
sp Spacer
v Vapor
w Water
Reference
[1] Y.-D. Kim, K. Thu, N. Ghaffour, K.C. Ng, Performance investigation of a solar-assisted direct contact
[2] K.W. Lawson, D.R. Lloyd, Membrane distillation, J. Membr. Sci., 124 (1997) 1–25.
[3] L.M. Camacho, L. Dumée, J. Zhang, J.-D. Li, M. Duke, J. Gomez, S. Gray, Advances in membrane
distillation for water desalination and purification applications, Water, 5 (2013) 94–196.
[4] L.-H. Cheng, P.-C. Wu, J. Chen, Numerical simulation and optimal design of AGMD-based hollow fiber
[5] M. Khayet, J.I. Mengual, T. Matsuura, Porous hydrophobic/hydrophilic composite membranes: Application
in desalination using direct contact membrane distillation, J. Membr. Sci., 252 (2005) 101–113.
[6] P. Peng, A.G. Fane, X. Li, Desalination by membrane distillation adopting a hydrophilic membrane,
[7] C. Feng, R. Wang, B. Shi, G. Li, Y. Wu, Factors affecting pore structure and performance of poly(vinylidene
fluoride-co-hexafluoro propylene) asymmetric porous membrane, J. Membr. Sci., 277 (2006) 55–64.
[8] R. Huo, Z. Gu, K. Zuo, G. Zhao, Preparation and properties of PVDF-fabric composite membrane for
membrane distillation, Desalination, 249 (2009) 910–913.
[9] J.A. Prince, V. Anbharasi, T.S. Shanmugasundaram, G. Singh, Preparation and characterization of novel
triple layer hydrophilic–hydrophobic composite membrane for desalination using air gap membrane distillation,
[10] M. Qtaishat, M. Khayet, T. Matsuura, Guidelines for preparation of higher flux hydrophobic/hydrophilic
composite membranes for membrane distillation, J. Membr. Sci., 329 (2009) 193–200.
Effect of backing structures and membrane properties under different operating conditions, Desalination, 323
(2013) 120–133.
[12] J.-G. Lee, Y.-D. Kim, W.-S. Kim, L. Francis, G. Amy, N. Ghaffour, Performance modeling of direct contact
membrane distillation (DCMD) seawater desalination process using a commercial composite membrane, J.
[13] G.L. Liu, C. Zhu, C.S. Cheung, C.W. Leung, Theoretical and experimental studies on air gap membrane
[14] L. Francis, N. Ghaffour, A.A. Alsaadi, G.L. Amy, Material gap membrane distillation: A new design for
desalination by air gap and liquid gap membrane distillation: A comparative study, Sep. Purif. Technol., 133
(2014) 176–186.
[16] A.S. Alsaadi, N. Ghaffour, J.D. Li, S. Gray, L. Francis, H. Maab, G.L. Amy, Modeling of air-gap membrane
distillation process: A theoretical and experimental study, J. Membr. Sci., 445 (2013) 53–65.
[17] J.-G. Lee, W.-S. Kim, Numerical study on multi-stage vacuum membrane distillation with economic
[18] A. Cipollina, M.G. Di Sparti, A. Tamburini, G. Micale, Development of a membrane distillation module for
solar energy seawater desalination, Chem. Eng. Res. Des., 90 (2012) 2101–2121.
[19] A.R. Da Costa, A.G. Fane, D.E. Wiley, Spacer characterization and pressure drop modelling in spacer-filled
[20] R.W. Schofield, A.G. Fane, C.J.D. Fell, Heat and mass transfer in membrane distillation, J. Membr. Sci., 33
(1987) 299–313.
[21] S. Al-Sharif, M. Albeirutty, A. Cipollina, G. Micale, Modelling flow and heat transfer in spacer-filled
membrane distillation channels using open source CFD code, Desalination, 311 (2013) 103–112.
[22] J. Phattaranawik, R. Jiraratananon, A.G. Fane, C. Halim, Mass flux enhancement using spacer filled
channels in direct contact membrane distillation, J. Membr. Sci., 187 (2001) 193–201.
[23] J. Phattaranawik, R. Jiraratananon, A.G. Fane, Effects of net-type spacers on heat and mass transfer in direct
contact membrane distillation and comparison with ultrafiltration studies, J. Membr. Sci., 217 (2003) 193–206.
[25] A.M. Alklaibi, N. Lior, Transport analysis of air-gap membrane distillation, J. Membr. Sci., 255 (2005)
239–253.
[26] F.A. Banat, J. Simandl, Membrane distillation for dilute ethanol: Separation from aqueous streams, J.
[27] S. Kimura, S.I. Nakao, S.I. Shimatani, Transport phenomena in membrane distillation, J. Membr. Sci., 33
(1987) 285–298.
[28] S.M. Shim, J.G. Lee, W.S. Kim, Performance simulation of a multi-VMD desalination process including the
[29] M. Khayet, Membranes and theoretical modeling of membrane distillation: A review, Adv. Colloid Interface
[30] J.-G. Lee, W.-S. Kim, Numerical modeling of the vacuum membrane distillation process, Desalination, 331
(2013) 46–55.
[31] V.V. Ugrozov, I.B. Elkina, V.N. Nikulin, L.I. Kataeva, Theoretical and experimental research of liquid-gap
[32] Y.A. Cengel, Heat transfer: A practical approach, 2nd ed., McGraw Hill, 2002.
[33] R.K. MacGregor, A.F. Emery, Free convection through vertical plane layers – Moderate and high Prandtl
[34] L.-H. Cheng, P.-C. Wu, J. Chen, Modeling and optimization of hollow fiber DCMD module for desalination,
[35] M.H. Sharqawy, J.H. Lienhard, S.M. Zubair, Thermophysical properties of seawater: a review of existing
hydrophobic/hydrophilic membrane preparation and application in air gap and direct contact membrane
[38] K.J. King, Turbulent natural convection in rectangular air cavities, in: Queen Mary College, University of
[39] Y.-D. Kim, K. Thu, K.C. Ng, G.L. Amy, N. Ghaffour, A novel integrated thermal-/membrane-based solar
energy-driven hybrid desalination system: Concept description and simulation results, Water Res., 100 (2016) 7–
19.
[40] Y.-D. Kim, K. Thu, S.-H. Choi, Solar-assisted multi-stage vacuum membrane distillation system with heat
[41] J.-G. Lee, W.-S. Kim, J.-S. Choi, N. Ghaffour, Y.-D. Kim, A novel multi-stage direct contact membrane
distillation module: Design, experimental and theoretical approaches, Water Res., 107 (2016) 47–56.
[42] A.M. Alklaibi, N. Lior, Heat and mass transfer resistance analysis of membrane distillation, J. Membr. Sci.,
Table captions
Figure captions
Fig. 1. Schematic diagram of the heat and mass transfer mechanisms in (a) AGMD and (b) LGMD.
Fig. 2. SEM micrographs of a commercial PTFE/PP composite membrane. Clockwise from top left: 100×, 500×,
Fig. 7. Experimental (symbol) [14,16] and simulated (line) permeate fluxes in AGMD and LGMD with different
gap sizes (δgap = 9, 13 mm) at Tf = 40−80 °C, Tc = 20 °C, and Vf = Vc = 1.5 l/min.
Fig. 8. Effect of the feed temperature (Tf) in the range of 40−70 °C on (a) AGMD and (b) LGMD permeate
Fig. 9. Effect of the feed flow rate (Vf) in the range of 0.5−5.0 l/min on (a) AGMD and (b) LGMD permeate
Fig. 10. Effect of the coolant temperature (Tc) in the range of 10−30 °C on (a) AGMD and (b) LGMD permeate
Fig. 11. Effect of the coolant flow rate (Vc) in the range of 0.5−5.0 l/min on (a) AGMD and (b) LGMD permeate
Fig. 12. Effect of the surface porosity (εs) in the range of 0.1−0.7 on (a) AGMD and (b) LGMD permeate fluxes
Material PP
Thickness, δ (m) 20 80
Porosity, (%) 70 34