Journal of Membrane Science 163 (1999) 75–91

Thermal effects in osmotic distillation

Carlo Gostoli
Dipartimento di Ingegneria Chimica, Mineraria e delle Tecnologie Ambientali, Università degli Studi di Bologna,
Viale Risorgimento 2, I-40136 Bologna, Italy
Received 30 September 1998; accepted 16 April 1999

Abstract
Osmotic distillation (OD) is a concentration technique for aqueous mixtures based on porous hydrophobic membranes in
contact on both sides with liquid phases at pressure lower than the pressure needed to displace the gas phase in the pores. The
driving force for the water vapour diffusion through the gas phase immobilised within the membrane pores is sustained by an
activity difference by using a hypertonic solution, typically concentrated brines, downstream the membrane. The mass transfer
causes a cooling down of the feed and a warm up of the brine, as a consequence a temperature difference is created which
reduces the effective driving force for mass transfer. This ‘thermal effect’ is investigated both theoretically and experimentally,
it is shown that the effect on the flux is substantial. ©1999 Elsevier Science B.V. All rights reserved.
Keywords: Concentration; Hydrophobic membranes; Osmotic distillation; Membrane distillation; Fruit juices

1. Introduction brane pores, so that a thin gas layer (essentially air)

is entrapped within the pores forming the so called
Osmotic distillation (OD) is a concentration tech- gas membrane. Water vapour diffuses through the
nique regarded with great interest in the processing gas membrane, the driving force being the vapour
of liquid foods such as fruit or vegetable juices. It pressure difference between the membrane sides. The
does not require high temperatures, thus preserving process has been referred as osmotic distillation with
the quality of the fresh juices; unlike reverse osmosis, reference to the mass transfer mechanism, which im-
OD does not suffer of osmotic pressure limitations, so plies evaporation of water at the feed side, diffusion
that concentration levels close to the values currently across the gas immobilised within the pores, and
obtained by evaporation can be reached. condensation at the brine side. The mechanism is
Osmotic distillation is a sort of direct osmosis, in the same as in membrane distillation, but the driving
that water is removed from the feed by a hypertonic force for water transport is sustained by a composi-
solution (typically concentrated brine) flowing down- tion difference instead of a temperature difference,
stream a membrane. The difference with respect to owing to this strict analogy the process has also been
the usual direct osmosis relays on the particular mem- referred as ‘isothermal membrane distillation’ [1].
brane employed, which is porous and hydrophobic, Comprehensive reviews recently appeared on both
typically made of PTFE or polypropylene. osmotic distillation [2] and membrane distillation [3].
Due to the hydrophobicity of the membrane All the published papers emphasised the role of heat
material neither the juice nor the brine enter the mem- transfer in membrane distillation, which is very

0376-7388/99/$ – see front matter ©1999 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 6 - 7 3 8 8 ( 9 9 ) 0 0 1 5 7 - X
76 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

important indeed; osmotic distillation, on the contrary, which can be described by the molecular diffusion
was considered essentially a mass transfer operation. model, Knudsen diffusion model or by a combination
The aim of the present work is to investigate both of the two mechanisms.
theoretically and experimentally the thermal effects Poiseuille flow model has also been cited for de-
associated to the mass transfer in OD. Due to the evap- scribing mass transfer in MD. Actually the convec-
oration at the feed side and condensation at the brine tive Poiseuille flow would require a total pressure dif-
side, a temperature difference is created through the ference across the membrane. The situation can be
membrane even if the bulk temperatures of the two realised eliminating the air from the pores by either
liquids are equal. Of course this thermal effect reduces deaerating the liquids before entering the membrane
the driving force for water transport. module, or operating under vacuum. This has been
It is expected that the temperature difference will widely discussed by Schofield at al.[5]. In the situ-
be quite small, owing to the low membrane thickness. ations considered here on the contrary, air is always
However, it is known from membrane distillation stud- present in the pores at pressure values close to the
ies [4] that a small temperature difference cancels out atmospheric pressure.
a large concentration difference. The effect of the 1T For pore size quite large with respect to the mean
on the flux can thus be substantial. molecular free path of water vapour, molecular diffu-
sion is the controlling mechanism and the water mass
flux can be written as [6]:
2. Theory
PA2 Pwl − Pw2
ν = KD ln = KD (3)
PA1 PAlm
2.1. Mass transfer through the membrane
in which PAlm is the log mean of the partial pressure
The mass transfer mechanism in OD as well as in of the air within the pores and KD represents a sort of
MD has been widely discussed: at both pore entrances membrane permeability depending on the membrane
vapour–liquid equilibria are established giving rise to properties, temperature and pressure:
a partial pressure difference of water vapour which εM P D
diffuses through the gas immobilised within the pores. KD = (4)
Xδ RT
The water vapour pressures at the pore mouths are
related to the temperature and activities prevailing in For a given gas pair the dependence of D on temper-
the liquids facing the membrane by: ature and pressure is given by [6]:
∗ PD
Pw1 = Pw1 aw1 (1) = constant (5)
Tb
∗
Pw2 = Pw2 aw2 (2) In the case of air–water vapour the exponent b
in which Pw∗ represents the vapour pressure of pure was given the value of b = 1.75 after Fuller [7] and
water and aw the water activity in the solutions. b = 2.33 according to the Bird-Slattery equation [6].
In OD the driving force (Pw1 − Pw2 ) for water Since small temperature variations are considered here
transport is sustained by an activity difference i.e. we can assume b = 2 as was made in MD modelling
aw2 < aw1 . In membrane distillation, on the contrary, [8]. As a consequence the membrane permeability lin-
the activity difference represents a counter driving early increases with temperature.
force to be overcome by a temperature difference If the membrane pore size is lower than the mean
across the membrane, i.e. in that case aw1 < aw2 , molecular free path the vapour transport is limited by
∗ > P ∗ . Of course situations are possible in
Pwl the collision against the pore wall, rather than against
w2
which both temperature and activity differences act the molecules of the gas present in the pores and the
synergistically. Knudsen model is the convenient one:
In any case, whatever the driving force may be, tem- v = KK (Pw1 − Pw2 ) (6)
perature, activity difference or both, we have a diffu-
sion of water vapour through the stagnant gas phase in which [9]:
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 77
r
1 εdp 8M to know with accuracy all the parameters, as the void
KK = (7)
3 Xδ πRT fraction and tortuosity, needed for the calculation.
The pore size itself is an ambiguous quantity, which
The mean molecular free path for water vapour at
depends on the measuring method adopted [5]. As an
25◦ C and atmospheric pressure if 0.13 ␮m [9], com-
example the TF200 membrane with nominal pore size
parable with the typical pore size of membranes used
0.2 ␮m used in refs. [8,10] has a penetration pressure
in OD and MD, both molecular and Knudsen mecha-
of 2.8 bar [8]. Based on the Cantor–Laplace equation:
nisms are thus expected to play a role.
In other words the diffusing molecules of water 4σ cos θ
Pp = (10)
vapour are slowed down by the molecules of air as dp
well as by the pore wall. In order to combine the two
the pore diameter would be dp = 0.38 ␮m. Use has
resistances we have to assume the same driving force
been made of θ = 112◦ [11] for the contact angle of
for the two processes, either (Pw1 − Pw2 )/PAlm , as de-
water–PTFE.
manded by Eq. (3), or (Pw1 − Pw2 ), as demanded by
With these limitations in mind we compared the
Eq. (6).
prediction of Eq. (9) with many experimental data of
The choice seems of limited practical interest, at
our own or from literature
25◦ C and atmospheric pressure, as an example, Pw
For pore size dp = 0.2 ␮m Eq. (9) predicts a 40%
is in the order of 20–30 mbar and PAlm , in the order
contribution of Knudsen resistance to the overall mass
of 990 mbar changes very little with Pw , i.e. with the
transfer resistance. Nevertheless in early studies on
water activities. The question may be more important
MD [10] through TF200 membrane the transport was
at high temperature or at reduced pressure.
successfully modelled as a molecular diffusion pro-
Some authors assumed (Pw1 − Pw2 ) as driving force
cess: the flux showed to be linear with ln (PA2 /PA1 )
and the ratio KD /PAlm in Eq. (3) as a constant mem-
as predicted by Eq. (3) and the slope was nearly
brane parameter. More correctly Schofield et al. [5]
Km = 300 kg/m2 h. This value is in good agreement
assumed (Pw1 − Pw2 ) as driving force, but taken into
with the value KD = 344 kg/m2 h, calculated from Eq.
account of the average air pressure within the pores in-
(4) for  = 0.6, X = 2, d = 60 ␮m, T = 25◦ C. The same
corporating it in the expression of the combined mem-
agreement has been observed in this study, as will be
brane permeability.
shown later in Section 2.
We prefer the use of a flux equation formally equal
Mengual et al. [12] reported the fluxes observed in
to Eq. (3):
OD experiments with three PTFE membranes (Gel-
Pw1 − Pw2 man) with the same nominal characteristics but the
ν = Km (8) pore size. The authors assumed a linear relationship,
PAlm
similar to Eq. (6), between the flux and the water
in which Km is the combined membrane permeability vapour pressure difference and called ‘osmotic distil-
taking into account of both transport mechanisms: lation coefficient’, C, the proportionality factor. The
1 1 1 membrane with 0.45 ␮m pore size showed an osmotic
= + (9) distillation coefficient lower than the membrane with
Km KD PAlm KK
0.2 ␮m pore size; whereas the values observed with the
In the conditions considered in the present work (at- 1 ␮m pore size membrane was only 40% larger. The
mospheric pressure and low pore size) the mass trans- same authors [13] reported MD data with the same
fer model represented by Eqs. (8) and (9) is substan- membranes: in all the conditions inspected the flux
tially equivalent to that of Schofield [5]. appeared not strongly dependent on the pore size as
Eq. (9) is not useful for predictive purposes: any would occur in the Knudsen mechanism.
membrane has a more or less wide pore size distribu- Schofield et al. [5] recognised that molecular dif-
tion of irregular shape, calculation based on the nomi- fusion is the controlling mechanism at atmospheric
nal pore size can give only an estimation of the actual pressure, Knudsen (or Knudsen–Poiseuille) mecha-
membrane permeability, which has to be experimen- nism being important only in deareated systems. Only
tally measured. On the other hand it is quite difficult the paper of Pena et al. [14] suggested the Knudesen
78 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

mechanism for the water transport across the mem- water transport implies evaporation at the feed side
brane. This statement seems unrealistic for membranes and condensation at the extract side. As a consequence
with 1 ␮m pore size used in their experiments. a temperature difference is created which reduces the
In the following we will use Eq. (8) to describe the vapour pressure difference across the membrane, i.e.
mass transfer through the membrane considering the the driving force for the water transport.
permeability Km as an experimental parameter. This In this section we will present an estimation of the
does not mean that we assume the transport mecha- thermal effect based on theoretical calculation. In Sec-
nism is the molecular diffusion in any case, but sim- tion 3 of the paper we will present a series of runs per-
ply Eq. (8) showed to be satisfactory, at least for MD formed to quantify exactly the thermal effect and the
at atmospheric pressure through membranes with pore consequence on the mass transfer rate. The concentra-
size 0.2 ␮m or more. tion polarisation will be neglected for the present, its
role will be discussed in the next section.
2.2. Effect of temperature Steady state heat transfer considerations accounting
for the phase change at both membrane surfaces lead
The flux will increase with temperature essentially to the following expression for the heat flux, q, through
as the vapour pressure of water. A more correct cal- the system:
culation can be made deriving Eq. (8) with respect to
q=h1 (T1 −T1I )=vλ−hm (T2I −T1I )=h2 (T2I −T2 ) (13)
T at aw s constant, which roughly means at constant
concentrations. Assuming Km linear with T, as occurs In which T1 and T2 are the bulk temperatures of the
in molecular diffusion mechanism, we have: feed and extract, respectively, h1 , h2 , the heat transfer
1 ∂v 1 PAlm P 1 dPw∗ coefficients of the liquid boundary layers and hm the
= + (11) heat transfer coefficient of the membrane which can be
v ∂T T PA1 PA2 Pw∗ dT
evaluated through the conductivities of the polymer,
Replacing the logarithmic mean air pressure, PAlm kS , and of the air, kA as:
with the geometric mean air pressure, PAm , in the
(1 − ε)kS + εkA
pores Eq. (11) reduces to: hm = (14)
δ
1 ∂v 1 P 1 dPw∗
= + (12) Two situations will be considered here: (a) stirred cell
v ∂T T PAm Pw∗ dT
with fixed bulk temperatures, (b) apparatus with fixed
The first term on the right side (1/T) represents the inlet temperatures.
effect of temperature on Km , the second term the effect
on the driving force. Later one is the effect on the 2.3.1. Stirred cell
vapour pressure of pure water corrected by the ratio The situation (a) is the case of many published
P/PAm , which at room temperature is close to unity. works [12] in which the membrane is placed between
That means that the effect of temperature on the flux two stirred cells kept at the same temperature. Assum-
is essentially the effect of temperature on the vapour ing h1 = h2 = hL , from Eq. (13) we have:
pressure of pure water, as stated above.
vλ
For example, with reference to pure water feed and 1T = (15)
NaCl 24% downstream the membrane, at 25◦ C Eq. hm + hL /2
(12) gives a flux variation of 6.4% per ◦ C. For com- In which 1T represents the temperature difference
parison the variation of vapour pressure of pure water across the membrane, see Fig. 1.
from the Clausius–Klapeyron equation is 5.94% per The effect on the driving force can be estimated by
◦ C.
the Clausius–Klapeyron equation:

2.3. Heat transfer λMPw∗ awm

Pw1 − Pw2 = 1Pwb − 1T (16)
RT 2
Notwithstanding OD is essentially a mass transfer in which 1Pwb is the water vapour pressure difference
process, also heat transfer is concerned. Indeed the calculated at the bulk conditions, i.e. at the temper-
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 79

2.3.2. Co-current flow

This situation is typical of industrial apparatuses in
which the inlet temperatures of the two streams are
fixed (equal or not) while the bulk values evolve along
the module according to the energy balance equations.
For the sake of simplicity the case of co-current flow
will be analysed here, in that case we have:
dT1
m1 cp1 = −q (19)
dA
Fig. 1. Temperature profile in a stirred cell. dT2
m2 cp2 =q (20)
dA
ature T1 = T2 and awm is the average water activity, Near the entry section we have a temperature profile
(aw1 + aw2 )/2. similar to that represented in Fig. 1 for the symmetric
From Eqs. (8), (15) and (16) we finally have: case (equal inlet temperatures and heat capacities), but
the feed (stream 1) cools down and the extract (stream
1Pwb 2) warms up according to Eqs. (19), (20) and (13). As
v = ηKm (17)
PAlm a consequence the heat flux from the feed to the mem-
brane and from the membrane to the extract reduces
in which: and the temperature difference 1T across the mem-
brane increases. An asymptotic value 1T∞ is finally
 −1
λ2 MPw∗ awm Km reached at which the convective heat flux through the
h= 1+ (18) membrane (νλ) is exactly balanced by the conductive
RT PAlm hm + hL /2
2
heat back-flux (hm 1T):
is a sort of efficiency coefficient describing the ther- vλ
mal effect in OD, in that it represents the fraction of 1T∞ = (21)
hm
the driving force really effective for the mass trans-
fer through the membrane. There is some analogy be- The situation is represented in Fig. 2 which reports
tween the coefficient η and the temperature polarisa- the qualitative temperature profiles along the module
tion coefficient, τ , used in membrane distillation [15].
Experiments with stirred cells are generally used to
measure the Km value as the slope of the flux ver-
sus the driving force in experiments in which the feed
is pure water and the extractant NaCl (or one other
salt) solutions at different concentrations. In this con-
ditions η is nearly constant and the flux will appear
linear with the driving force, but the slope is ηKm , and
depends on the hL , i.e. on the stirring rate. Let us as-
sume for example Km in the order of 300 kg/m2 h and
hm = 670 W/m2 K , as for the TF200 membrane used in
ref. [8]. Let us assume in addition hL = 2000 W/m2 K
as a typical value for ‘well stirred’ cell. In that case at
25◦ C η ∼= 0.85 and the membrane permeability would
appear 15% lower than the true value. This example
shows that the thermal effect is substantial and very Fig. 2. Qualitative temperature profiles along the module in
difficult to cancel. The effect makes difficult to char- co-current flow in the case of equal inlet temperatures and heat
acterise the membrane by simple experiments. capacities.
80 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

for the symmetric case. The different couples of lines The evaluation of the length of the thermal entry re-
correspond to different values of the flow rates. Clearly gion would require the numerical solution of the mass
the asymptotic values of T1 and T2 are the same for all and energy balance equations governing the system.
the lines, but the module length needed to approach Nevertheless it is useful an estimation of the order of
that value, which can be called ‘thermal entry region’, magnitude of the entry region which can be used to
increases as the flow rates increase. check the above assumption of constant concentration
The effect of 1T∞ on the driving force for water and to establish a priori if the module works near to
transport can be evaluated by the Clausius–Klapeyron or far from the asymptotic conditions.
equation, as previously made in the analysis of stirred Since we are interested only in the order of magni-
cell. We finally have: tude some rough approximations can be accepted. Let
1Pwm us thus consider the symmetric case, i.e. equal inlet
v∞ = h∞ Km (22) temperatures and heat capacities for the two streams,
PAlm
and assume the mass flux, ν, as a constant.From Eq.
in which: (13) we have:
 −1
λ2 MPw∗ awm Km q = U [1T∞ − (T2 − T1 )] (24)
η∞ = 1 + (23)
RT 2 PAlm hm
in which U is the overall heat transfer coefficient de-
and 1Pwm represents the vapour pressure difference
fined as usual:
evaluated at the average temperature, which in the
symmetric case is equal to the inlet temperature. 1 1 1 1
= + + (25)
The following example gives an idea of the amount U h1 h2 hm
of the effect: in the extraction of pure water with NaCl
solutions at 25◦ C through the TF200 membrane we By simple integration of Eq. (19) or Eq. (20) we
have η∞ = 0.7, that means the asymptotic flux, ν ∞ , is have:
30% lower than the ideal value calculated ignoring the 1T∞ − 1Tb UA
temperature difference through the membrane. ln = −2 (26)
1T∞ mcp
It is interesting to note that since both Km and hm
are inversely proportional to the membrane thickness, The temperature difference between the two streams,
η∞ is independent of it. The only way to reduce the 1Tb , approaches the asymptotic value 1T∞ , as the
asymptotic temperature difference across the mem- membrane area, A, goes to infinity, as by definition.
brane is the use of material with high thermal conduc- However, for A > mcp /U the difference between 1Tb
tivity ks , nevertheless high void fraction are of course and 1T∞ becomes quite small. Thus we can use the
desirable, so that the main thermal membrane resis- quantity:
tance is represented by the air within the pores. mcp
In the above discussion the flow rates and the con- A∞ = (27)
U
centrations of the two streams have been tacitly as-
sumed constant along the module, this is not true, of as a measure of the thermal entry region.
course. The modelling of the OD apparatus would re- It is worth noting that this length is independent of
quire the mass balance equations coupled with the en- the flux, which of course determines the value of the
ergy balance Eqs. (19), (20). But the flux ν through asymptotic temperatures.
the membrane is generally very small with respect to For a given membrane and flow rate the minimum
the flow rates m1 and m2 . It seems thus reasonable to length of the thermal entry region is obtained for h1 ,
assume that the two streams reaches the asymptotic h2 → ∞; in real systems it may be substantially larger
temperatures in a section at which the concentrations because of the heat transfer resistances in the boundary
are still very close to the inlet value. In other words layers.
the discussion above neglects the concentration change The analysis can be easily modified for different
through the thermal entry region. This assumption will situations, for example for different values of the flow
be checked now. rates of the two streams.
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 81

Now we are in the conditions to check the assump- Pw∗ (aw1

I − aI )
v = KmApp w2
(31)
tion of constant concentration in the thermal entry re- PAlm
gion. We have:
In which KmApp , equal to either ηKm or η∞ Km de-
1m vA∞ vcp pending on the situation considered, takes into account
= = (28) of the thermal effects and Pw∗ is calculated at the av-
m m U
erage temperature.
By using typical values for the variables involved, the Formally Eqs. (29–31) represent a purely mass
variation of flow rates, and thus of concentrations of transfer problem which can be analysed by the resis-
the two streams can be estimated in the order of 1%. tance in series concept:

2.4. Mass transfer through the boundary layers v = K0 Pw∗ (aw1 − aw2 ) (32)
The overall mass transfer coefficient is defined as:
As in all the membrane processes, the flux through
1 γw1 x1ml γw2 x2ml PAlm
the membrane can be limited by the concentration = + + ∗ (33)
polarisation phenomenon. Since only water passes K0 ρ1 kL1 ρ2 kL2 Pw KmApp
though the membrane, the solute concentration near Let us consider the special case in which the stream
the membrane is larger than the bulk value in the feed 1 is pure water and stream 2 a brine, this is the case of
side, and lower than the bulk value in the extract side. many published works. In this case the first term on
According to the well known film theory model we the second side of Eq. (33) vanishes and we have to
have: compare only the mass transfer resistance in the side
x1I aw1 − aw1
I 2 and through the membrane. Since x2ml ≤ x2 , one can
v = kL1 ρ1 ln ≈ kL1 ρ1 (29) say that the concentration polarisation is negligible if:
x1 γw1 x1ml
γw2 x2 Pw∗ KmApp
x2 a I − aw2 kL2  (34)
v = kL2 ρ2 ln I ≈ kL2 ρ2 w2 (30) ρ2 PAlm
x2 γw2 x2ml
For example for the TF200 membrane at 25◦ C Eq. (34)
In which the water activity coefficients γ w1 and γ w2 gives kL2  4 × 10−7 m/s, value which is very small
have been assumed constant in the boundary layers. indeed. Of course any case has to be analysed consid-
These equations, together with the previous ones ering the particular situation, but, generally speaking,
describing the mass transfer through the membrane, the concentration polarisation is expected to play a mi-
Eq. (8), and the heat transfer through the membrane nor role in osmotic distillation experiments in which
and through the boundary layers, Eq. (13), allow the the feed is pure water and the extractant a salt solution.
calculation of the interfacial compositions, x1I , x2I , and Of course the situation may be completely different if
the flux ν. the feed is a viscous fluid, as concentrated fruit juices.
Now we will present an estimation of the concentra-
tion polarisation effects in some typical conditions to
establish a priori if these effects are important or not. 3. Experimental
Reference will be made to the symmetric cases con-
sidered before, i.e. the stirred cell with the same bulk A series of experiments were performed to check
temperatures T1 = T2 and to the co-current apparatus the theory presented above, concerning the role of heat
with equal inlet temperatures and heat capacities of the transfer in osmotic distillation. Capillary membranes
two streams. In this latter case we will restrict the anal- in shell-tube configuration were preferred to flat sheet
ysis to the asymptotic conditions, at which the tem- membranes for this purpose, indeed the analysis is
peratures of the two streams have reached the asymp- simpler in capillary membranes, which have a well
totic values. The two situations can be analysed as a defined geometry and does not require any support; the
‘pseudo isothermal’ case in which the flux through the heat and mass transfer coefficients can be calculated
membrane is given by: by standard well known relationships.
82 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

Table 1
at various concentration as extractant. The inlet tem-
The relevant characteristics of the modules used in the experiments
peratures of both streams were 25◦ C. The flow rate of
Module LM 2P 06 MD 020 CP 2N the extractant was kept quite large (93 l/h) in order to
Membrane Accurel PP Q3/2 Accurel PP S6/2 minimise possible concentration polarisation effects.
Pore size 0.2 ␮m 0.2 ␮mm The flow rate of the water stream, on the contrary, was
Penetration pressure 3.1 bar 3.5 bar considered not important and was not recorded. The
Void fraction 70% 70%
measured flux values were then reported as a func-
Capillary diameter
Inner 0.6 mm 1.8 mm tion of the driving force, (Pw1 − Pw2 )/PAlm , obtain-
Outer 1.0 mm 2.6 mm ing a straight line of slope 92 kg/m2 h. After that we
Active length 250 mm 460 mm passed to experiments in which the feed was a mix-
Shell diameter 18 mm 25 mm ture of glucose/sucrose in the weight ratio 2 : 1, which
N. of capillaries 85 40
roughly simulate the juice. Surprisingly the observed
Inside area 0.04 m2 0.104 m2
flux in these runs was larger than the values already
measured with pure water feed. The date seemed in
Two capillary modules supplied by Microdyn accordance with a Km value of nearly 100 kg/m2 h. The
GmbH (Germany) manufactured with Accurel first though was that same pore was obstructed in the
polypropylene membranes (Enka) were used in the never used membrane by something, perhaps a preser-
experiments. The relevant characteristics of the two vative, which was progressively removed during the
modules are reported in Table 1. experiments.
The experimental set up is shown schematically in The module was thus taken in operation for a day
Fig. 3. The feed and the extractant were circulated with distilled water on both sides, then washed with
co-currently through the module and the respective ethanol and dried with a clean warm air stream. The
reservoirs by two gear pumps with variable speed. The runs with water and sodium cloride were then repeated
volume of the extractant was quite large in order to several times. Each run begun with NaCl at high con-
maintain a nearly constant concentration during the centration, say 25%, which was diluted several times
experiments. adding distilled water directly in the salt reservoir; for
The inlet temperatures of the two streams were con- each concentration value the flux was measured as de-
trolled by two coils immersed in thermostated baths. scribed above. All the variables but the salt concen-
The module was wrapped up by insulating material. tration were rigorously constant in each run.
Inlet and outlet temperatures were recorded by four Apparently in each run the flux was a linear function
thermocouples. The flow rates were controlled by the of the driving force, but the slope changed randomly.
revolution speed of the pumps. No flow meters was Finally we realised that the flux depends on the flow
installed, the values of the flow rates were calculated rate of the feed stream, notwithstanding the feed was
from the rpm by the calibration curve periodically con- pure water, and recognised the crucial role played by
trolled measuring the time to fill a graduated bottle. the heat transfer rate.
The instantaneous flux was measured by the decrease Only at this point we installed the two thermocou-
in level in the burette; for this purpose the feed reser- ples at the exit of the module, in addition to the ones
voir was temporarily cut off by closing the valves V1 already present at the inlet and repeated the exper-
and V2 and opening valve V3 . iments with water–sodium cloride but recording ex-
Most experiments were performed with pure water actly the flow rates of both streams.
feed and sodium cloride solutions as extractant. Same The first runs were performed keeping equal each
results with magnesium cloride are also presented. other the inlet temperatures of the two streams. The
It may be interesting to explain how the experimen- salt solution flowed shell side with a flow rate of 93 l/h,
tal program was born. At the beginning of the work its concentration was kept constant at 25% wt. The
we performed experiments with the LM 2P 06 module water, on the contrary, was tube side with flow rate
to characterise the membrane, i.e. to calculate the per- changing from 3 to 30 l/h.
meability Km defined by Eq. (8). To this end we mea- The exit temperature of the salt solution (not re-
sured the flux for pure water feed and sodium cloride ported) showed to be just a bit larger than the inlet
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 83

Fig. 3. Experimental stet-up for osmotic distillation in capillary modules.

Fig. 4. Water exit temperatures vs. the reciprocal of the feed flow rate for different inlet temperatures. Module: LM2P06; Feed: pure water,
tube side; Extractant: NaCl 25 wt.%., shell side, 93 l/h

value, a temperature rise of 0.1–0.2◦ C was observed Apparently the exit temperature of the feed stream
in some cases. Apparently the warming up was quite was always lower than the inlet value and decreased
small owing to the large flow rate. with the flow rate: for the LM2P06 module the maxi-
Figs. 4 and 6 show the exit water temperatures and mum cooling of the feed stream, corresponding to the
Figs. 5 and 7 the transmenbrane fluxes as a function minimum flow rate inspected, was 0.9◦ C in the run at
of the reciprocal of the feed flow rate for the two mod- 25◦ C and 2.7◦ C in the run at 50◦ C. Notwithstanding
ules and for different values of the inlet temperature. the lower flux, the thermal effects were a bit larger
The use of 1/Q, which is proportional to the residence in the MD020CP2N module: the maximum cooling
time, instead of the flow rate Q makes more clear and of the feed stream, was 1.3◦ C in the run at 25◦ C and
significant the figures. 3.2◦ C in the run at 50◦ C. Also the flux decreased with
84 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

Fig. 5. Flux vs. the reciprocal of the feed flow rate for different inlet temperatures. Module: LM2P06; Feed: pure water, tube side;
Extractant: NaCl 25 wt.%., shell side, 93 l/h.

the feed flow rate, the behaviour was quite similar for runs allowed to found that the exact value is 47.2◦ C,
the two modules: in both cases a lowering of the feed indeed keeping the inlet water temperature equal
flow rate from 30 l /h to 3 l/h produced a flux decrease to this value no temperature change was observed
of nearly 8% in the run at 25◦ C and nearly 18% in the and the flux was independent of the feed flow rate.
run at 50◦ C. The flux value measured in these conditions is thus
Figs. 4 and 6 suggest that the water exit tempera- the asymptotic value ν ∞ , defined in the theoretical
tures approach asymptotic values as the flow rate de- section.
creases, i.e. as the residence time goes to infinity. The Some results for the module MD020CP2N are re-
values of the asymptotes can only be estimated from ported in Fig. 10, only the two cases of Twi = Tsi and
Figs. 4 and 6. The exact values were searched for by Twi = Tsi − 5 are represented for simplicity sake. It
the following trial and error procedure: keeping con- seems that in this case the asymptotic temperatures
stant the flow rate and the inlet temperature, TSi , of are really reached at the lowest flow rate, this is due to
the salt solution, a series of runs was made for feed the larger membrane area of the module with respect
inlet temperature decreasing from TSi to TSi − 5 with to the previous one.
intervals of 1◦ C. For each inlet temperature the outlet The procedure outlined above allowed to obtain the
temperature and flux were measured at feed flow rate asymptotic temperature differences, 1T∞ , across the
changing from 30 to 3 l/h. membrane and the corresponding fluxes, ν ∞ , for all
Figs. 8 and 9, which refer to the LM2P06 mod- the cases inspected. The membrane heat transfer co-
ule and TSi = 50◦ C show an example of the results efficient, hm , was then calculated by Eq. (21) and the
obtained. Apparently the feed stream approaches the membrane permeability, Km , by Eq. (8) assuming neg-
same exit temperature, as the flow rate decreases, by ligible concentration polarisation.
either cooling down or warming up depending on The results for the two modules and all the condi-
the inlet temperature. The flux behaviour is perfectly tions inspected are summarised in Table 2.
parallel. Some experiments were performed also with MgCl2
Apparently the asymptotic water temperature lies 30% wt. as extractant and pure water feed. In these
between 47◦ C and 47.5◦ C in this case. Few further runs the inlet temperatures of the two streams were
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 85

Fig. 6. Water exit temperatures vs. the reciprocal of the feed flow rate for different inlet temperatures. Module: MD020CP2N; Feed: pure
water, tube side; Extractant: NaCl 25 wt.%., shell side, 93 l/h

Fig. 7. Flux vs. the reciprocal of the feed flow rate for different inlet temperatures. Module:MD020CP2N; Feed: pure water, tube side;
Extractant: NaCl 25 wt.%., shell side, 93 l/h.

always kept equal to each other and the salt flowed effects are larger than in the extraction with NaCl, the
shell side with flow rate 93 l/h. The exit temperatures warming up of the salt solution was not negligible in
as well as the fluxes were measured for different val- this case, temperature rises as large as 0.6◦ C were of-
ues of the feed flow rate. The results are reported in ten observed. The results were thus reported in Fig.
Figs. 11 and 12. Since the fluxes and thus the thermal 11 as temperature difference between the two streams
86 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

Fig. 8. Water exit temperatures vs. the reciprocal of the feed flow rate for different feed inlet temperatures. Module: LM2P06; Feed: pure
water, tube side; Extractant: NaCl 25 wt.%., shell side, 93 l/h, inlet temperature 50◦ C.

Fig. 9. Flux vs. the reciprocal of the feed flow rate for different feed inlet temperatures. Module: LM2P06; Feed: pure water, tube side;
Extractant: NaCl 25 wt.%., shell side, 93 l/h, inlet temperature 50◦ C.

instead of temperature of the water stream, as in the a consequence the thermal effects were more visible:
previous cases. a temperature difference at the module exit as large
Due to the larger driving force, the fluxes observed as 5.5◦ C (in the run at 50◦ C) was observed. How-
in the extraction with MgCl2 were nearly 2.5 times the ever, the effects on the flux were quite similar: as the
values obtained with NaCl in the same conditions. As flow rate decreases from 30 to 3 l/h a flux decrease
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 87

Fig. 10. Water exit temperatures vs. the reciprocal of the feed flow rate for three different brine temperatures. The feed inlet temperature
was either equal to the brine temperature (open symbols) or 5◦ C lower (black symbols). Module: MD020CP2N; Feed: pure water, tube
side; Extractant: NaCl 25 wt.%., shell side, 93 l/h.

Table 2
Asymptotic temperature differences between the two streams and asymptotic fluxes as a function of the salt inlet temperature for the two
modulesa
Module TSi (◦ C) 1T∞ (◦ C) ν ∞ (kg/m2 h) hm (W/m2 ) Km (kg/m2 h)
LM2P06 40 2.0 1.330 440.6 134.1
45 2.4 1.624 450.1 140.1
50 2.8 1.877 443.7 137.9
MD020CP2N 25 1.2 0.295 166.8 54.1
30 1.6 0.377 159.1 59.5
35 2.1 0.466 149.1 66.4
40 2.4 0.553 154.0 65.7
45 2.8 0.641 152.3 66.1
50 3.2 0.736 152.2 65.5
a Feed: pure water, tube side; Extractant: NaCl 25 wt.%., shell side.
88 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

Fig. 11. Temperature differences between salt and water at the module exit vs. the reciprocal of the feed flow rate for different inlet
temperatures. Module: MD020CP2N; Feed: pure water, tube side; Extractant: MgCl2 30 wt.%, shell side, 93 l/h.

Fig. 12. Flux vs. the reciprocal of the feed flow rate for different inlet temperatures. Module: MD020CP2N; Feed: pure water, tube side;
Extractant: MgCl2 30 wt.%, shell side, 93 l/h.
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 89

Table 3
Comparison between the experimental and theoretical values of the membrane heat transfer coefficient (hm , W/m2 K) and membrane
permeability (Km , kg/m2 h)a
Module hmExperimental hmTheoretical KmExperimental KmTheoretical
 = 0.6  = 0.7  = 0.6  = 0.7

LM2P06 446.1 466.1 391.3 137.4 135.0 157.0

MD020CP2N 155.6 216.3 182.0 62.9 62.5 73.0
a Average values are between 25◦ C and 50◦ C.

of 20–30%, depending on the inlet temperature, was Table 4

observed. Efficiency coefficients for different feed flow rates and
temperaturesa
Module 25◦ C 50◦ C
4. Discussion and conclusions
LM2P06 30 l/h 0.72 0.55
3 l/h 0.69 0.45
The experimental program was expressly designed η∞ 0.43
to quantify the thermal effect in osmotic distillation. MD020CP2N 30 l/h 0.71 0.58
The brine flow rate and concentration were kept con- 3 l/h 0.61 0.40
stant in all the runs, so that concentration polarisation, η∞ 0.61 0.37
if any, had always the same weight. a Feed: pure water, extractant: NaCl 25 wt.%.
In order to estimate the role of the concentration po-
larisation the brine side mass transfer coefficients kL2 Actually for the MD020CP2N the agreement is per-
were calculated following the method reported in ref. fect as regard to Km , whereas the thermal conductivity
[16]; we obtained kL2 = 2.4 10−5 m/s for the LM2P06 of the membrane appears a bit lower than the theoret-
module and kL2 = 1.5 10−5 m/s for the MD020CP2N ical value. There may be an explanation for that: the
module. Both calculations were made for NaCl solu- porosity  entering the mass transfer may be different
tions 25 wt.% flowing shell side through the modules from the  entering the heat transfer, especially for
at 93 l/h flow rate. The kL values are quite large with thick membrane, which can contain some non-passing
respect to the values given by Eq. (34), The conclusion pores.
is drawn that the experimental data reported above are The efficiency coefficients η have been calculated
not affected by the concentration polarisation effects. as the ratio between the experimental fluxes and the
The experimental results allowed the calculation of ideal fluxes calculated by Eq. (8) ignoring the ther-
the relevant membrane properties hm and Km reported mal effects. The results for the lowest and the largest
in Table 2. For comparison theoretical values of hm temperatures and flow rates are reported in Table 4 to-
and Km were also calculated by Eqs. (14) and (4), re- gether with η∞ , which represents the limiting value
spectively, assuming molecular diffusion mechanism for Q → 0.
for mass transfer, X = 2, ks = 0.14 W/mK [9] and cor- Apparently the heat effects are relevant in all the
recting the membrane thickness d by the ratio di /dlm conditions. Larger η values are observed at larger feed
owing to the cylindrical shape. The calculation was flow rate, since as the flow rate increases the thermal
made for both  = 70% (the nominal porosity given by entry region increases according to Eq. (27), however,
the producer) and  = 60%. The results are reported flow rates quite larger than the values inspected are
in Table 3, apparently the agreement between experi- needed to approach unit efficiency.
mental and theoretical values is fairly good assuming The efficiency strongly depends on the temperature,
 = 60%, of course this does not necessarily mean that as a consequence the observed flux dependence on
this is the true membrane porosity, good agreement temperature is weaker than that foreseen by Eq. (12).
can be obtained also adjusting other parameters, as the Interestingly the two modules show nearly the same
tortuosity and thickness. efficiency. On the other hand the two membranes have
90 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91

the same structure, but the thickness, which according m average value
to the theory, do not influence the heat effects. s salt solution
1 feed side
2 extract side
5. List of symbols * pure water
∞ asymptotic value
A membrane area
aw water activity
Cp specific heat
D diffusion coefficient Acknowledgements
dp pore diameter
Pw water vapour pressure I wish to thank Dr Sidney Loeb for stimulating my
m mass flow rate interest in the field and for fruitful discussions. This
M molecular weight work was supported by the Italian Ministry for Uni-
P pressure versity and Scientific Research (MURST).
PAlm log mean pressure of air
R gas constant
h heat transfer coefficient
References
hm membrane heat transfer coefficient
Km membrane permeability, Eq. (8) [1] A.G. Fane, M. Costello, P.A. Hogan, R.W. Schofield,
k thermal conductivity Membrane distillation and osmotic (isothermal membrane)
kL mass transfer coefficient distillation: factors for enhanced performance, Workshop on
Q volume flow rate Membrane distillation, osmotic distillation and membrane
q heat flux contactors, Cetraro, Italy, 2–4 July, 1998.
[2] W. Kunz, A. Benhabiles, R. Ben-Aim, Osmotic evaporation
T temperature through macroporous hydrophobic membranes: a survey of
U overall heat transfer coefficient current research, osmotic evaporation through macroporous
ν water mass flux hydrophobic membranes: a survey of current research and
x mole fraction applications, J. Membr. Sci. 121 (1996) 25–36.
γ activity coefficient [3] K.W. Lawson, D.R. Lloyd, Membrane distillation, J. Membr.
Sci. 124 (1997) 1–25.
η efficiency coefficient
[4] G.C. Sarti, C. Gostoli, Use of hydrophobic membranes in
δ membrane thickness thermal separation of liquid mixtures: theory and experiments,
 membrane porosity in: E. Drioli, M. Nakagaki (Eds.), Membranes and Membrane
λ latent heat of vaporisation Processes, Plenum Press, New York, 1986, pp. 349–360.
χ tortuosity factor [5] R.W. Schofield, A.G. Fane, C.J.D. Fell, Gas and
vapour transport through microporous membranes I.
σ surface tension
Knudsen–Poiseuille transition, J. Membr. Sci. 53 (1990) 159–
θ contact angle 171.
ρ density [6] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport
Phenomena, Wiley, New York, 1960.
[7] E.N. Fuller, P.D. Shettler, J. Giddrings, Ind. Eng. Chem. 58
5.1. Subscripts (1966) .18.
[8] G.C. Sarti, C. Gostoli, S. Matulli, Low energy cost
desalination processes using hydrophobic membranes,
A air Desalination 56 (1985) 277–286.
b bulk [9] R.H. Perry, C.H. Chilton, Chemical Engineers’ Handbook,
e exit McGraw-Hill, New York, 1973.
[10] C. Gostoli, G.C. Sarti, S. Matulli, Low temperature distillation
I interface
through hydrophobic membranes, Separation Sci. Technol. 22
i inlet (1987) 855–872.
L liquid phase [11] R.J. Duham, M.H. Nguyen, Hydrophobic membrane
lm logarithmic mean evaluation, hydrophobic membrane evaluation and cleaning
 C. Gostoli / Journal of Membrane Science 163 (1999) 75–91 91

for osmotic distillation of tomato puree, J. Membr. Sci. 87 [14] L. Pena, M. Paz Godino, J.I. Mengual, A method to evaluate
(1994) 181–189. the net membrane distillation coefficient, J. Membr. Sci. 143
[12] J.I. Mengual, J. Ortizde Zarate, L. Pena, A. Velazquez, (1998) 219–233.
Osmotic distillation through porous hydrophobic membranes, [15] R.W. Schofield, A.G. Fane, C.J.D. Fell, Heat and mass transfer
J. Membr. Sci. 82 (1993) 129–140. in membrane distillation, J. Membr. Sci. 33 (1987) 299–313.
[13] A. Velazquez, J.I. Mengual, Temperature polarisation in [16] C. Gostoli, A. Gatta, Mass transfer in a hollow fiber dialyzer,
membrane distillation, Ind. Eng. Chem. Res. 34 (1995) 585– J. Membr. Sci. 6 (1980) 133–148.
590.