You are on page 1of 10

Subscriber access provided by UNIV AUTO DEL ESTADO DE HIDALGO UAEH

Environmental Science & Technology is published by the American Chemical Society.


1155 Sixteenth Street N.W., Washington, DC 20036
Article
Temperature Effects on the Morphology of Porous
Thin Film Composite Nanofiltration Membranes
Ramesh R. Sharma, and Shankararaman Chellam
Environ. Sci. Technol., 2005, 39 (13), 5022-5030 DOI: 10.1021/es0501363 Publication Date (Web): 19 May 2005
Downloaded from http://pubs.acs.org on April 24, 2009
More About This Article
Additional resources and features associated with this article are available within the HTML version:
Supporting Information
Access to high resolution figures
Links to articles and content related to this article
Copyright permission to reproduce figures and/or text from this article
Temperature Effects on the
Morphology of Porous Thin Film
Composite Nanofiltration
Membranes
R A M E S H R . S H A R M A

A N D
S H A N K A R A R A M A N C H E L L A M *
, ,
Department of Civil and Environmental Engineering,
4800 Calhoun Road, University of Houston,
Houston, Texas 77204-4003, and Department of Chemical
Engineering, University of Houston,
Houston, Texas 77204-4004
Even though polymeric nanofiltration (NF) and reverse
osmosis (RO) membranes often operate on surface waters
and surficial groundwaters whose temperature varies
over time and with season, very little detailed mechanistic
information on temperature effects on membrane
selectivity is available to date. Hence, a study was
undertakentoinvestigatetheeffects of operatingtemperature
(5-41 C) on the morphology and structure of two
commercially available thin film composite NF membranes.
Application of hydrodynamic models to experimental
rejection of dilute solutions of hydrophilic neutral alcohols,
sugars, and poly(ethylene glycol)s revealed changes in
both the sieving coefficient and permeability of solutes
below the membrane glass transition temperature. The vast
majority of pores were smaller than 2 nm for both
membranes (network pores) even though evidence for a
small fraction of larger aggregate pores (30 nm) was also
obtained for one membrane. Increasing temperature
appears to cause structural changes in network pores by
increasing its pore size while simultaneously decreasing
pore density. These increases in pore sizes partially explain
reported reductions in contaminant (e.g. arsenic, salts,
natural organic matter, hardness, etc.) removal by NF and
RO membranes with increasing temperature.
Even though nanofiltration (NF) membranes achieve high
removals of several drinking water contaminants including
natural organicmatter anddisinfectionbyproduct precursors
(1, 2), endocrine disrupting compounds (3, 4), pesticides (5,
6), nitrogenous compounds (7), arsenic (8), etc. the factors
governing their separation have not yet been comprehen-
sively elucidated even at room temperature. Additionally,
wide temperature variations (1-35 C) of NF and reverse
osmosis (RO) membrane feedwaters (9-11) necessitate
mechanistic studies on temperature effects on membrane
selectivity.
Most investigators have simply reported that increases in
water temperature increased passage of salts (9, 10, 12),
hardness (13), natural organic matter (14), and arsenic (8)
across polymeric membranes. Paradoxically, very little
detailed mechanistic information on temperature effects on
membrane selectivity is available to date.
The pore size distribution (PSD), pore density, and
morphology of nanofilters are essential tohindereddiffusion
andconvectioncalculations that arecentral tounderstanding
their intrinsic selectivity properties (4, 7, 15-17). In this
manuscript we relate changes inselectivity withtemperature
tovariations inthe structure andmorphology of the polymer
(including PSD) constituting the active layer of the mem-
brane.
Recently, we reportedthat employing a single log-normal
distribution for the PSD systematically over predicted
experimental sieving data for neutral solutes larger than 0.6
nmfor onenanofilter (18). Anadditional focus of this research
is to improve our recent work (18) and derive a single PSD
density function(potentially bimodal) toaccurately describe
membrane sieving over a wide range of solute sizes.
The principal objective of this workis todeduce variations
in skin layer morphology of polymeric thin film composite
NF membranes (PSD, pore connectivity, and pore number
density) with operating temperature. Rejection of several
neutral solutes (that served as molecular tracers) by two
commercially available nanofilters was measured in the
environmentally significant range 5-41 C. PSDs at different
temperatures were derived from solute sieving data using
hindered convection considerations. Also, diffusive perme-
abilities of solutes withdifferent molecular dimensions were
evaluated using a modified porous pathway model account-
ing for PSD, pore density, and solute dependent tortuosity.
Theoretical Work
Transport Equations. For porous or loose nanofilters, in
the absence of specific chemical and electrostatic interac-
tions, thelocal unchargedsoluteflux(JS) throughamembrane
pore is due to contributions from diffusive and convective
flows (4, 7, 15, 19)
where c is the radially averaged solute concentration in the
pore, Jv is the radially averaged solution velocity in the pore,
D is the bulk solution diffusivity, and Kd and Kc are the
hindrance factors for diffusionandconvection, respectively.
To obtain the solute sieving coefficient (Si), eq 1 has been
integrated over the effective membrane thickness (x) with
appropriate boundary conditions on the feed and permeate
side and correcting for concentration polarization effects
using film theory on membrane feed side as (19)
where C
p and Cb are the permeate and bulk phase solute
concentrations, and k is the feed side mass transfer coef-
ficient. All our experiments were designed to minimize
concentration polarization effects (k f ) allowing the
membrane phase concentration to be closely approximated
as the bulk concentration (see Experimental Work section
and ref 18). S, the asymptotic sieving coefficient attained as
Peclet number (Pe) f is a measure of solute transport
solely due to convection and is equal to the product of
equilibriumpartitioningcoefficient, andKc. Themembrane
Peclet number is defined as (19)
* Correspondingauthor phone: (713)743-4265; fax: (713)743-4260;
e-mail: chellam@uh.edu.

Department of Civil and Environmental Engineering.

Department of Chemical Engineering.


J
S
) -K
d
D

dc
dx
+ K
c
cJ
v
(1)
S
i
)
C
p
C
b
)
S

exp
(
J
v
k
)
(1 - exp(-Pe))(1 - S

) + S

exp
(
J
v
k
)
(2)
Environ. Sci. Technol. 2005, 39, 5022-5030
5022 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 13, 2005 10.1021/es0501363 CCC: $30.25 2005 American Chemical Society
Published on Web 05/19/2005
where Ak is membrane surface porosity, x is the effective
distance traveledby solute or solvent throughthe membrane
pore(accountingfor tortuosity), andPMis solutepermeability
coefficient, which is the measure of solute transport due to
diffusion. Membrane phenomenological parameters (S and
PM) andfeedside mass transfer coefficient (k) were obtained
by fitting eq 2 to experimental measurements of Cp over a
wide range of volumetric fluxes (Jv) at a constant feed flow
rate and feed concentration (18). Further, hydrodynamic
models of hindered convection in cylindrical pores (20) and
slit pores (15) were used to relate S and pore radius or half
pore width.
Membrane Pore Size and Pore Size Distribution Ob-
tained from(Hindered) Convection Calculations. Because
the log-normal distribution may be most appropriate for
membranes (18, 21, 22) it was used as the basis for PSD
calculations. A linear combination of a maximum of 2 log-
normal distributions was used to quantify the entire range
of pore sizes that were encountered:
The coefficients h
i are non-negative and ihi ) 1, fi(r)
represents the ith log-normal distribution based on pore
number with mean r ji and standard deviation SPi and r is the
pore radius. hi represents the sieving contribution of the ith
distribution.
First, using a semiempirical approach adopted by several
authors (18, 21-24), we assume that solute rejectionis purely
due to geometric considerations. This approach, wherein a
solute of radius r* permeates through all the pores whose
radii are larger thanitself (r* to) provides consistent results
(23) even though it explicitly ignores any additional drag the
solute may experience in the confined environment of the
membrane pore. The sieving coefficient according to this
purely sieving model is
Next, we have employeda mechanistic approach(19) that
accounts for the hydrodynamic lag and steric effects by
normalizing the convective hindrance coefficient for each
type of pore on the basis of volumetric flow rate through
cylindrical and slit shape pores to calculate an average
asymptotic sieving coefficient, S

where r is used interchangeably to represent pore radius for


cylindrical pores (n ) 4) and or half width for slit pores (n
) 3). Membrane pores were conceptualized to have these
shapes because closed form analytical expressions for the
hindrance factors are available for these pore geometries
(15, 20). Because slit and cylindrical pore models are both
only model conceptualizations of the complex pore network
characteristic of NF membranes, one of themcannot be said
tobe better thanthe other. Thus, bothapproaches (semiem-
pirical eq 5 and mechanistic eq 6) were used to investigate
general trends inPSDwithtemperature. ThePSDparameters
(r ji, and SPi) were then calculated by fitting eqs 5 and 6 to the
experimental asymptotic sieving coefficient using nonlinear
least squares regression. Integration was performed using
Simpsons rule witha narrowstepsize anda truncationerror
of 0.01%.
Analysis of Hindered Diffusion through Membrane
Pores. We have incorporated steric hindrances to diffusion
by introducing the hindrance factor to represent solute
permeability (25) in cylindrical pores as
The free diffusion coefficient of solutes employed in this
work was obtained fromthe literature (26-28), and the pore
size distribution has already been determined using eqs 5
and 6 as described in the previous section. Thus, the only
unknown in eq 7 is the termN0/x, which was calculated by
fitting the experimental permeability data to the respective
equations.
Membrane pores can be expected to be tortuous and
potentially interconnectedwithsome surface pores not even
penetrating the entire thickness of the active layer (29, 30).
Hence, the effective membrane thickness x is given by the
product of the tortuosity (p) and membrane thickness (L):
Recently, a closed form expression for tortuosity as a
function of pore and solute radius (p(r*)) was derived for a
cubical network of cylindrical pores (31). We have modified
their equation to account for hydrodynamic lag in the pores
by introducing an additional parameter in the form of a
hindrance factor for diffusion (Kd) as
Due to the cubical geometry of pore network, pores can
either be parallel or normal to the membrane surface. The
distance between the two adjacent pores oriented parallel
to the membrane surface is and the distance between two
normal pores is , which is related to pore density as
and N
0 were estimated by fitting eqs 7-10 to experi-
mental permeability data after obtaining L from scanning
electron microscopy.
Solute Size Calculations. In this study, the Stokes radius,
molecular width, and mean molecular radius (MMR) of
solutes were all used to characterize NF membranes.
Experimental reports of diffusion coefficients at 25 C for all
solutes were obtained (26-28, 32) to calculate the Stokes
radius. Bulkdiffusivities of solutes at temperatures other than
25 C were estimated using temperature and viscosity
corrections (27).
To better account for shape effects on the removal the
molecular width and mean molecular radius were also
calculated for the solutes employed. Giddings (33) theoreti-
cally analyzed a variety of molecule shapes and showed that
purely steric partitioningis dominatedby theextremegroups
of the molecule, whereas internal shielded groups have no
influence on it. Recognizing this result, we derive a mean
molecular radius that represents the mean length of projec-
tion of the molecule along the X, Y, and Z axes using a
Pe )
K
c
K
d
D

A
k
x
J
v
)
K
c
P
M
J
v
(3)
f(r) )

i)1
2
h
i
f
i
(r) )

i)1
2
h
i
S
Pi
r2
exp
(
-
(ln(r) - ln(r j
i
))
2
2S
Pi
2 )
(4)
S

(r*) )

i)1
2
h
i
r*
1
S
Pi
r2
exp
(
-
(ln(r) - ln(r j
i
))
2
2S
Pi
2 )
dr
(5)
S

(K
c
) f(r) r
n
dr

f(r) r
n
dr
(6)
P
M
) D

N
0
x

K
d
f(r)r
2
dr (7)
x )
p
L (8)

p
(r*) ) 1 +
4
5
(

n)1

n
(
3
4
)
n-1
(

r*

K
d
f(r)dr)
n
(9)
)

1
N
0
(10)
VOL. 39, NO. 13, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 5023
commercially available software (Chem 3D, Cambridgesoft
Corp., v. 5). Details of MMR estimation and a comparison
of solute size parameters are given in the Supporting
Information.
Experimental Work
Membranes. Two commercially available, polyamide thin
film composite membranes denoted as DL (Osmonics,
Minnetonka, MN) and TFCS (Koch Fluid Systems, San
Diego, CA) were employed. The manufacturers provided no
information regarding possible surface chemical modifica-
tions. Operationally, sodium chloride rejection for DL and
TFCS membrane (at 1 mequiv/L concentrationand 345 kPa)
were52%and77%, respectively. Theselowrejections indicate
that the DL and TFCS membranes should be considered as
porous or loose nanofilters.
Solutes Employed. Tracers such as sugars and alcohols
which have well-defined transport properties were chosen
because of the need to quantify membrane morphology
rather thantheir presence inwater supplies. All experiments
were conducted using reagent grade neutral organic solutes
viz. methanol, ethanol, ethylene glycol, tert-butyl alcohol,
dextrose, and sucrose (EM Science Gibbstown, NJ), xylose,
glycerol, raffinose 5-hydrate, and R-cyclodextrin as well as
poly(ethylene glycol)s (PEGs) of 20 kDa, 35 kDa, and100 kDa
(Sigma Aldrich Company, St. Louis, MO). These alcohols,
sugars, and PEGs are often employed in studies of hindered
transport across membranes (16, 18, 30, 32). Similar to other
studies (4, 34), the target feedwater concentration for all
solutes was set at 20 mg/L total organic carbon (TOC). This
value facilitated easy, accurate, and precise measurements
of both feed and permeate concentrations necessary to
calculate rejection. The concentrations of all solutes were
analyzed using a TOC analyzer (TOC5050A, Shimadzu,
Columbia, MD). All results reported are an average of four
or five injections with a coefficient of variation < 2%.
Bench-Scale Cross-Flow Experimental Apparatus. All
experiments were conducted at constant cross-flowvelocity
but varying flux using a pressurized cell (Sepa CF cell,
Osmonics, Minnetonka, MN) that accommodates a 19 cm
14 cmflat membrane sheet (effective filtrationarea 155 cm
2
).
The retentate and permeate streams were recycled to the
feedtank containing 20Lof water witha single organic solute
maintaining a constant feed concentration 20 mg/L TOC.
A positive displacement gear pump (model 74011-11, Cole-
Parmer, Chicago, IL) minimized pressure fluctuations. The
cross-flow velocity was maintained constant at 9.6 and 19.2
cm/s for DL and TFCS membrane, respectively. Inert
materials such as Teflon or stainless steel were used for all
tubing, connections, pump-heads, and the membrane cell.
Filtrationpressureandtemperatureweremonitoredusing
analogue transducers (PX303-200G5VandTJ120 CPSS 116G,
OmegaEngineeringCompany, Stamford, CT). For purewater
permeability measurements, filteredwater was continuously
collected on a digital weighing balance (Ohaus Navigator
N1H110, Fisher Scientific, Houston, TX). The built-in RS232
port in the balance was directly connected to a computer to
obtain the digital signal corresponding to the weight. These
analogue and digital signals were acquired at a rate of 1 per
minute using LabVIEW (Version 5.1, National Instruments,
Austin, TX). Permeateandretentateflows werealsomanually
measured.
The complete cross-flowfiltration apparatus was housed
in a chamber whose temperature was adjusted to 5, 15, 23,
35, and 41 C. The entire apparatus including the water was
equilibrated to the chosentemperature for a minimumof 30
h prior to the commencement of experiments.
Conduct of Experiments. Freshmembranecoupons were
first soakedfacedowninultrapurewater that was replenished
3 or 4 times over a 24 h period. This coupon was then placed
in the membrane holder, and ultrapure water was passed
through the entire apparatus at 750-800 kPa pressure for
24 h. Following this membrane-setting period, steady-state
fluxes were measured for pressures in the range 100-760
kPa.
All experiments wereconductedat lowfeedwater recovery
(<1%) inconjunctionwithhighcross-flowvelocity toreduce
changes inbulkconcentration. Thepermeateconcentrations
were measured as a function of volumetric flux after the
attainment of steady-state by varying pressure in random
order to reduce systematic errors in our experimental
measurements. Five tosevendifferent pressures inthe range
100-760 kPa were used for each combination of solute,
membrane, and temperature. This pressure range resulted
ininitial fluxes between0.77 and10 m/s at 23 C. After both
permeate flux and solute rejection were maintained at their
steady-state values for at least 5-10 h, five separate 10 mL
samples of permeate water were collected for TOC analysis.
Thus, eachphenomenological coefficient corresponds to25-
35 measurements of solute rejection allowing increased
precision of parameter estimates. The retentate flow rate
and concentration were also measured in order to conduct
mass balances.
More details onthe materials andexperimental protocols
can be found in ref 18.
Results and Discussion
Pores in nanofilters can be conceived as the polymer free
(void) spaces in the membrane matrix through which solute
and water transport occurs (35). Such a system can be
characterized by calculating the number and diameter of
cylindrical or slit pores, which would have the same total
volume as well as flowand transport properties as the actual
membrane itself (24, 30). Hence, membrane morphological
parameters including pore size distribution, pore density,
ratio of pore density to effective thickness, tortuosity, and
pore spacing calculated using hydrodynamic transport
models (eqs 5-10) should only be interpreted as being
effective values. In other words, the exact values of
membrane morphological parameters obtainedby fittingthe
transport model to experimental data should be cautiously
interpreted, and only the trends with changing temperature
are emphasized in this paper.
Pore Size Distributions Incorporating Hindered Con-
vection. Experimental sieving coefficients for all solutes
decreasedwithincreasingflux(pressure) at eachtemperature.
For example, Figure 1 depicts sieving coefficients and model
(eq 2) fits at 23 C. As observed, excellent fits of the
phenomenological model were obtained for all solutes in
the flux range employed. Similar results were obtainedat the
other temperatures investigated. Further, eq 2 predicts that
at very high fluxes (Pe f ) S
i f S corresponding to the
y-intercept in Figure 1. As shown in Figure 1, 25-35
experimental measurements of the sieving coefficient ob-
tained over a range of fluxes at each temperature were used
to calculate the asymptotic value (S) for each solute using
eq 2, which were used next to calculate membrane pore size
distributions.
Thediscretepoints inFigure2correspondtotherespective
S values for each solute at room temperature. The curves
in Figure 2 correspond to calculations of the S as a function
of the mean molecular radius using hydrodynamic models
of cylindrical (Figure 2a) and slit (Figure 2b) pore geometry
using eq 6. Correlation coefficients for PSDs employing all
solutes for both pore geometries were not excellent (R
2

0.60) andhada bias whereinhinderedconvectionfactors for


the smallest solutes (methanol, ethanol, andethylene glycol)
were underpredicted. Much better fits of eq 6 were obtained
when these solutes were excluded from this analysis (R
2

0.80). This could be caused by the slippage of water on the


5024 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 13, 2005
surface of solutes that are approximately the same size as
itself (36), which leads to overprediction of drag by hydro-
dynamic models. Additionally, the polar nature of these
compounds further enhances their passage through the
membrane pore (37). Therefore, methanol, ethanol, and
ethylene glycol were not included in PSD calculations, but
their experimental data are superposed in Figure 2 only for
the sake of comparison.
As observed, experimental sieving coefficients at 23 C
for the DL membrane decreased monotonically with solute
MMR suggesting a monomodal pore size distribution. As
expected, a single log-normal distribution was sufficient to
accurately fit eq 6 to experimental data. In contrast, sieving
coefficients for theTFCSmembranedecreasedmonotonically
initially until reaching a constant value near 0.1 for solute
MMR in the range 0.33-6 nm, beyond which it again
decreased for the largest solute investigated (PEG 100K)
suggesting a multimodal distribution. Abimodal log-normal
PSD incorporating two types of cylindrical pores (type 1: r j1
)0.48 nm and SP1 )0.005 nm and type 2: r j2 )16.0 nm and
SP2 ) 0.1 nm), closely fit the entire range of experimental
sieving coefficients.
The presence of two distinct types of pores in the TFCS
membrane canbe interpretedinterms of network pores and
aggregate pores reported for some NF and RO membranes
(35, 38, 39). The network pores arise fromthe spaces between
the segments constituting the polymer network within each
supermolecular polymer aggregate in the film casting solu-
tion, whereas the spaces between each neighboring super-
molecular polymer aggregates themselves are referred to as
aggregatepores (39). Consistent withtheseprevious findings,
thenetworkpores dominatedtheporesizes inthemembrane
skin (h1/h2 ) O(10
8
)). Although, the area fraction of nonse-
lective large pores is only 0.01% they dramatically increase
the breadth of the sieving curve (30-fold) as seen in Figure
2 and control 10% of water permeability (see Supporting
Information). Bimodal PSDs indicate the need to determine
the entire sieving curve for each membrane rather than only
specifying the molecular weight cutoff.
The DLandTFCS membranes were manufacturedby two
different corporations, which likely employed different
materials for the respective skin layer on each polymeric
membrane, resulting in widely different pore size distribu-
tions. Good fits between experimental data and theoretical
values obtained with eq 6 obtained for both membranes (R
2
0.8) demonstrates the applicability of hinderedconvection
models evenfor NFmembranes. Similar toprevious findings
(24), even better fits (R
2
>0.95) were obtained for the purely
sieving model (see Supporting Information Figure S2).
However, its use is limited because it explicitly ignores steric
partitioning and hydrodynamic lag in the pore making it a
strictly empirical model. Nevertheless, both eqs 5 and 6 can
FIGURE 1. Estimation of phenomenological transport coefficients at 23 C using neutral solutes of different molecular dimensions for the
DL (a) and TFCS (b) membranes.
FIGURE 2. Comparison of pore size distributions for two NF membranes obtained by fitting the hindered convection models of cylindrical
pore (a) and slit pore (b) to experimental asymptotic sieving data of various solutes at 23 C. Note that small size solutes such as methanol,
ethanol, and ethylene glycol have not been included in PSD calculation. They are superposed in this figure only for the sake of comparison.
VOL. 39, NO. 13, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 5025
be used to accurately represent pore size distributions. In
thenext section, weusethesemodels toquantifytemperature
effects on network and aggregate pore sizes.
Effect of Temperature on Pore Size Distributions.
Experimental sieving coefficients of various solutes obtained
at 5, 15, 23, 35, and41 Cfor bothmembranes are superposed
on theoretical predictions neglecting hindered convection
(eq5) inFigure 3. For the DLmembrane, whichonly posseses
network pores, the entire PSD was shifted to higher values
of pores sizes (Figure 3a), whereas for TFCS membranes,
only the network pores were shifted to higher pore sizes
(Figure 3b and its inset). In other words, temperature had a
stronger influence on the network pores compared to
aggregate pores.
Figure 4depicts the dependence of the asymptotic sieving
coefficient on the ratio of solute size to pore size corre-
sponding to S )0.5 (denoted as rp,S)0.5 in Figure 4) at each
temperature as suggested by Michaels (22). Pore radius for
50% sieving (S ) 0.5) at 23 C refers to the x-coordinate of
the point on the solid curve in Figure 2 corresponding to the
y-coordinate of 0.5. This exercise was repeated at all
temperatures for both membranes. The insets of Figure 4
shows that the pore radius corresponding to S
) 0.5 for
both membranes and pore geometries increased with tem-
perature. The appropriate value from the inset was used for
normalization at each temperature, which collapsed all
experimental sieving coefficients in the range 0.1 <S
<1.0
to a single curve. Further, theoretical calculations using eq
6 for all five temperatures overlapped in Figure 4, appearing
as a single curve. Therefore, this normalization allowed
simultaneous comparisonof sievingcausedbynetworkpores
of both membranes in the entire temperature range inves-
tigated. Inother words, monomodal andbimodal log-normal
distributions are difficult todistinguishexperimentally using
larger values of the sieving coefficient. The choice of
coordinate scaling suppresses the differences between two
FIGURE 3. Influence of temperature on experimental sieving data using purely sieving log-normal model. Network pore sizes shift to higher
values as temperature increases for both polymeric membranes employed in this study.
FIGURE 4. Dependence of asymptotic sieving coefficient on normalized solute radius (ratio of mean molecular radius and pore radius
corresponding to 50% sieving, (rp,S
) 0.5)) for both NF membranes. Insets show the increasing trend for pore radius corresponding to
50% sieving with temperature for both pore geometries. Experimental sieving coefficients for the TFCS membrane are depicted as empty
symbols and those obtained for the DL membrane as filled symbols. The solid lines depict theoretical predictions of sieving coefficients
using hindered transport models (eq 6) for both membranes at all five temperatures.
5026 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 13, 2005
membranestructures tosuchanextent that only significantly
different membrane structural characteristics are visible. For
example, bimodal distribution of the TFCS membrane
distinguishes from the monomodal structure of the DL
membrane (both experimentally and theoretically) by the
presence of the second mode for S <0.1 (aggregate pores),
while the first mode overlaps the monomodal distribution
for 0.1 < S < 1.0 (network pores). This behavior is similar
to the one reported by Wendt and Klein (40), where an
exponential formfor PSDwas used. Further, assuming either
cylindrical or slit pore shapes only weakly influencedsieving
in Figure 4a,b supporting the current understanding that
this nondimensionalization cannot distinguish pore shapes
(40).
Therefore, despitethedifferences inchemical composition
and the subsequent response to temperature, the solute to
membrane pore radius is the primary factor in determining
membrane selectivity toward the alcohols and sugars em-
ployed in this study. Further, the quantitative agreement
betweentheoretical predictions andexperimental datashown
in Figures 2 and 4 suggests that available hydrodynamic
models (15, 20) can provide important insights into solute
transport through the tortuous pores of NF membranes.
Network pores shrank in size as temperature decreased
whether convective hindrances were included or not (see
Table S2 and associated text and Figure 3). Hence, deterio-
rations inmembrane selectivity withincreasing temperature
can be attributed primarily to the increase in network pore
sizes. Such increases in pore sizes can also partially explain
the decreases in arsenic (8), salt (9, 10, 12), hardness (13),
and natural organic matter (14) rejection reported in the
literature as well as the nonviscous contributions toactivated
transport of water (12, 18, 41).
Temperature Effects on Pore Density to Membrane
Thickness Ratio (N0/x). The value of N0/x at each
temperature was obtained by using it as an adjustable
parameter to fit eq 7 to experimental permeabilities (Figure
5). The hindrance factors for diffusion are not strictly
applicable for solutes that have dimensions comparable to
that of thewater moleculeitself (15). Additionally, thepassage
of methanol, ethanol, and ethylene glycol across the mem-
branes are enhancedby their polarity (37). Therefore, similar
tothe hinderedconvectionanalysis, the smallest solutes with
MMRe0.27nm(viz. methanol, ethanol, andethylene glycol)
have not been included in N0/x calculations. As seen in
Figure 5, N0/x values are of the same order of magnitude
(10
23
pores m
-3
) as those reported at roomtemperature for
cellulose acetate RO membranes (42). This is not surprising
given that the RO membranes employed were calculated to
have apore size distributionsimilar tothat depictedinFigure
2. Figure 5 also shows a monotonic decrease in N0/x with
temperature for both membranes, which is probably caused
by changes in the active layer polymeric structure and
network pore sizes (Figure 4 and Table S2).
Incorporating Tortuosity Factor in Pore Density Cal-
culations. Membrane properties including pore density, size
distribution, and tortuosity as well as the thickness of the
skin layer influences solute permeability (see eqs 7 and 8).
Scanning electron microscopy of cross sections of both DL
and TFCS membranes revealed a very thin top layer (200
nm) supported by multiple layers of coarse granular porous
material, which is approximately in the previously reported
range for other thin film composite RO and NF membranes
(21). Similar to the analysis of Koros and Woods (43), the
skin layer thickness (200 nm) was assumed to remain
invariant with temperature. The pore density (N
0) and pore
spacing ( and ) were then estimated by using them as
fitting parameters to minimize the sum of squared error
between experimental and theoretical solute permeability
using eqs 7-10 at each temperature using the pore size
distributions derived earlier in Figures 2 and 4.
Using this procedure, the hindered transport model
revealed pore densities O(10
15
-10
16
) pores/m
2
for DL and
TFCS membranes at 23 C, which are in a similar range as
the reported values for other NF and RO membranes (35,
44). Additionally, a power law decrease in pore number
densities of the DL and TFCS membranes was calculated.
Decreases in pore density and increasing pore sizes with
increasingtemperaturemanifests as if pores arebeingpushed
apart. Hence, thenormal () andlateral distances () between
twopores calculatedassuming a cubical geometry increased
with temperature as depicted in Figure 6a,b.
Once the membrane morphological parameters (PSD, ,
) were obtained, the solute dependent tortuosity was
calculated using eq 9 at each temperature. Figure 7 depicts
solute dependent tortuosity values for both membranes in
the same range as nanoporous dialysis membranes at room
temperature (45). Figure 7 alsoreveals a monotonic decrease
in tortuosity with increasing solute mean molecular radius
at any given temperature. These results are similar to
transport across the upper layer of skin (stratum corneum),
where smaller hydrophilic drugs have been shown to be
capable of entering into a higher fraction of pores (31),
allowing them to experience a more tortuous path while
penetrating the skin. Further, for a given solute size, the
tortuosity increased with temperature indicating a greater
influence of pore size distribution and on tortuosity as
compared to the value (see eq 9).
Interpreting Temperature-Induced Changes in Skin
Layer Morphology with Implications for Permeability.
Solute transport measurements in conjunction with pore
flowmodels have been successfully used to elucidate effects
of several operational andmanufacturingparameters onPSD
and pore number density of polymeric RO and NF mem-
branes (35, 38, 44, 46). Analogous to these studies, mor-
phological changes with temperature can be interpreted as
the consequence of reorientationandmovement of polymer
chains constituting the membrane matrix thereby changing
the available void spaces or pores. The calculated trends in
membrane morphological parameters with operating tem-
perature shown in Figures 3-7 and S3 are manifestations of
equivalent changes in void spaces due to polymer chain
movement. To our knowledge, this is the first rigorous
FIGURE 5. Decreasing N0/x with increasing temperature for both
membranes.
VOL. 39, NO. 13, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 5027
investigation of the effects of feedwater temperature on skin
layer morphology of NF membranes.
Trends shown in Figures 3-7 and S3 suggest that
increasing temperature results in structural and morpho-
logical changes in network pores of the polymeric skin layer
by increasing pore sizes while simultaneously decreasing
pore density. This increase in the dimensions of the
permeation pathways at higher temperatures can explain
the reported reductions in contaminant rejection (8, 9, 12).
Simultaneously, thermal expansion of the network pores
decreases the number of permeation pathways to weaken
the dependence of solute permeation on temperature.
Movement of polymer chains with increasing temperature
thereby fusing together several network pores consequently
reducing pore density while simultaneously shifting the pore
size distribution to higher values can explain these results.
These structural changes influences solute permeation and
correspondingactivationenergies of permeationas discussed
next.
Substituting eq 8 in eq 7 reveals that solute permeation
is proportional to the hindered diffusivity within membrane
pores, the pore density and inversely proportional to the
tortuosity. Hence, theactivationenergyof solutepermeation,
Ep, is proportional to ED + EN0 - Ep, where ED, EN0, and Ep
are the contributions of effective diffusion, pore number,
andporeconnectivity, respectively. Wehaverecentlyreported
increasing pore sizes resulted in 20 < ED < 200 kJ/mol for
these solutes and membranes (18). Decreasing N0/x in
Figure 5 indicates a net negative activationenergy associated
with pore number and connectivity. This is confirmed by
decreasing N0 (EN0 < 0) and increasing p (Ep > 0) with
temperature inFigures 6and7, respectively. The net negative
contribution of pore density and connectivity to activated
transport (EN0 -Ep <0) weakens theinfluenceof temperature
on permeability compared to the hindered diffusivity.
Changes in the permeation pathways in the membrane
skin layer with temperature will result in the activated
transport of contaminants. However, the negative enthalpy
associated with adsorption of natural organic matter (17)
and hormones (4) will further serve to weaken the temper-
ature dependence of their permeability.
Nomenclature
A
k
membrane porosity
c solute concentration in pore (ML
-3
)
C
p
permeate solute concentration (ML
-3
)
C
m
membrane phase solute concentration (ML
-3
)
D

free solute diffusion coefficient (L


2
T
-1
)
E
P
activationenergyof permeation(ML
2
T
-2
mol
-1
)
E
D
activation energy of diffusion (ML
2
T
-2
mol
-1
)
E
N0
activation energy of pore disappearance (ML
2
T
-2
mol
-1
)
E
P
activation energy of pore connectivity (ML
2
T
-2
mol
-1
)
h
i
fraction of pores corresponding to ith pore size
distribution
J
v
volumetric pure water flux (LT
-1
)
J
s
solute flux (ML
2
T
-1
)
K
c
convective hindrance factor
K
d
diffusive hindrance factor
L membrane thickness (L)
L
p
pure water permeability (M
-1
L
2
T)
N
0
total number of pores per unit areaof membrane
(L
-2
)
Pe Peclet number
P
M
solute permeability (LT
-1
)
P transmembrane pressure (ML
-1
T
-2
)
r pore radius (L)
r j
i
geometric mean pore radius of the ith distribu-
tion (L)
FIGURE 6. Modifications in the membrane skin structure due to
temperature changes depicted as an increase in the normal and
lateral distances between pores for both membranes.
FIGURE 7. Increasing tortuosity with decreasing solute mean
molecular radius and increasing temperature for both polymeric
membranes.
5028 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 13, 2005
r* upper limit of solute radius in eq 5
S
i
solute sieving coefficient
S

asymptotic sieving coefficient


S
Pi
geometric standard deviation of the ith distri-
bution (L)
T temperature (K)
Greek letters
lateral distance between two pores (L)
x effective distance traveled by solute within the
membrane matrix (L)
solute equilibrium partitioning coefficient
normal distance between two pores (L)

P
solute dependent tortuosity
water viscosity (ML
-1
T
-1
)
Acknowledgments
David Paulson and Peter Eriksson of Osmonics Inc. and
Randolph Truby and Tom Stocker of Koch Membrane
Systems Inc. generously donatedmembranesamples. Prasad
Taranekar and Rigoberto Advincula fromthe Department of
Chemistry assisted with molecular mechanics simulations.
The detailed comments generated during anonymous peer
review of this manuscript are greatly appreciated. This
research has been funded by a grant from the National
Science Foundation CAREER program (BES-0134301). The
contents do not necessarily reflect the views and policies of
the sponsors nor does the mention of trade names or
commercial products constitute endorsement or recom-
mendation for use.
Supporting Information Available
Estimation of molecular width and mean molecular radius,
choice of appropriate solute size parameter, experimental
reproducibility and quality assurance protocols, effect of
temperatureonPSD, andtemperatureeffects onN0/xusing
water permeability measurements (Tables S1 and S2 and
Figures S1-S3). This material is available free of charge via
the Internet at http://pubs.acs.org.
Literature Cited
(1) Tan, L.; Amy, G. L. Comparing Ozonation and Membrane
Separation for Color Removal and Disinfection By-Product
Control. J. Am. Water Works Assoc. 1991, 83, 74-79.
(2) Nilson, J. A.; DiGiano, F. A. Influence of NOM Composition on
Nanofiltration. J. Am. Water Works Assoc. 1996, 88, 53-66.
(3) Kimura, K.; Toshima, S.; Amy, G.; Watanabe, Y. Rejection of
Neutral Endocrine Disrupting Compounds (EDCs) and Phar-
maceutically Active Compounds (PhACs) by ROMembranes. J.
Membr. Sci. 2004, 245, 71-78.
(4) Nghiem, L. D.; Schafer, A. I.; Elimelech, M. Removal of theNatural
Hormones by Nanofiltration Membranes: Measurement, Mod-
eling, and Mechanisms. Environ. Sci. Technol. 2004, 38, 1888-
1896.
(5) Kiso, Y.; Kon, T.; Kitao, T.; Nishimura, K. Rejection Properties
of Alkyl Phthalates with Nanofiltration Membranes. J. Membr.
Sci. 2001, 182, 205-214.
(6) Devitt, E. C.; Ducellier, F.; Cote, P.; Wiesner, M. R. Effects of
Natural Organic Matter and the Raw Water Matrix on the
Rejection of Atrazine by Pressure-Driven Membranes. Water
Res. 1998, 32, 2563-2568.
(7) Lee, S.; Lueptow, R. M. Membrane Rejection of Nitrogen
Compounds. Environ. Sci. Technol. 2001, 35, 3008-3018.
(8) Waypa, J. J.; Elimelech, M.; Hering, J. G. Arsenic Removal by RO
and NF Membranes. J. Am. Water Works Assoc. 1997, 89, 102-
114.
(9) Ventresque, C.; Turner, G.; Bablon, G. Nanofiltration: from
Prototype to Full Scale. J. Am. Water Works Assoc. 1997, 89,
65-76.
(10) Taniguchi, M.; Kimura, S. Estimation of Transport Parameters
of ROMembranes for Seawater Desalination. AIChEJ. 2000, 46,
1967-1973.
(11) Abdel-Jawad, M.; El-Sayed, E. E. F.; Ebrahim, S.; Al-Saffar, A.;
Safar, M.; Tabtaei, M.; Al-Nuwaibit, G. Fifteen Years of R and
DProgramin Seawater Desalination at KISR Part II. ROSystem
Performance. Desal. 2001, 135, 155-167.
(12) Mehdizadeh, H.; Dickson, J. M.; Eriksson, P. K. Temperature
Effects onthe Performance of Thin-FilmComposite, Aromatic
Polyamide Membranes. Ind. Eng. Chem. Res. 1989, 28, 814-
824.
(13) Schaep, J.; Van der Bruggen, B.; Uytterhoven, S.; Croux, R.;
Vandecasteele, C.; Wilms, D.; Van Houtte, E.; Vanlerberghe, F.
Removal of Hardness from Groundwater by Nanofiltration.
Desal. 1998, 119, 295-302.
(14) Her, N.; Amy, G.; Jarusutthirak, C. Seasonal Variations of
Nanofiltration(NF) Foulants: IdentificationandControl. Desal.
2000, 132, 143-160.
(15) Deen, W. M. Hindered Transport of Large Molecules in Liquid-
Filled Pores. AIChE J. 1987, 33, 1409-1424.
(16) Aimar, P.; Meireles, M.; Sanchez, V. A Contribution to the
Translationof RetentionCurves intoPore Size Distributions for
Sieving Membranes. J. Membr. Sci. 1990, 54, 321-338.
(17) Combe, C.; Molis, E.; Lucas, P.; Riley, R.; Clark, M. M. The Effect
of CA Membrane Properties on Adsorptive Fouling by Humic
Acid. J. Membr. Sci. 1999, 154, 73-87.
(18) Sharma, R. R.; Agrawal, R.; Chellam, S. Temperature Effects on
SievingCharacteristics of Thin-FilmCompositeNanofiltration
Membranes: PoreSizeDistributions andTransport Parameters.
J. Membr. Sci. 2003, 223, 69-87.
(19) Mochizuki, S.; Zydney, A. L. Theoretical Analysis of Pore Size
Distribution Effects on Membrane Transport. J. Membr. Sci.
1993, 82, 211-227.
(20) Bungay, P. M.; Brenner, H. TheMotionof aCloselyFittingSphere
in a Fluid Filled Tube. Int. J. Multiphase Flow 1973, 1, 25-56.
(21) Singh, S.; Khulbe, K. C.; Matsuura, T.; Ramamurthy, P. Membrane
Characterization by Solute Transport and Atomic Microscopy.
J. Membr. Sci. 1998, 142, 111-127.
(22) Michaels, A. S. Analysis and Prediction of Sieving Curves for
Ultrafiltration Membranes: A Universal Correlation? Sep. Sci.
Technol. 1980, 15, 1305-1322.
(23) Kassotis, J.; Shmidt, J.; Hodgins, L. T.; Gregor, H. P. Modeling
of the Pore Size Distribution of Ultrafiltration Membranes. J.
Membr. Sci. 1985, 22, 61-76.
(24) Van der Bruggen, B.; Schaep, J.; Dirk, W.; Vandecasteele, C. A
Comparison of Models to Describe the Maximal Retention of
Organic Molecules in Nanofiltration. Sep. Sci. Technol. 2000,
35, 169-182.
(25) Tezel, A.; Sens, A.; Mitragotri, S. Description of Transdermal
Transport of Hydrophilic Solutes during Low-Frequency Sono-
phoresis basedona ModifiedPorous Pathway Model. J. Pharm.
Sci. 2003, 92, 381-393.
(26) International Critical Tables of Numerical Data, Physics, Chem-
istry and Technology, 1st ed.; Washburn, E. W., Ed.; McGraw-
Hill Book Company, Inc.: New York and London, 1929.
(27) Hayduk, W.; Laudie, H. Prediction of Diffusion Coefficients for
Non-Electrolytes in Dilute Aqueous Solutions. AIChE J. 1974,
20, 611-615.
(28) Wang, X.-L.; Tsuru, T.; Togoh, M.; Nakao, S.-I.; Kimura, S.
Evaluation of Pore Structure and Electrical Properties of
Nanofiltration Membranes. J. Chem. Eng. Jpn. 1995, 28, 186-
192.
(29) Nakao, S.-i. Determination of Pore Size and Pore Size Distribu-
tion 3. Filtration Membranes. J. Membr. Sci. 1994, 96, 131-165.
(30) Bowen, W. R.; Mohammad, A. W.; Hilal, N. Characterization of
Nanofiltration Membranes for Predictive Purposes - Use of
Salts, Uncharged Solutes and Atomic Force Microscopy. J.
Membr. Sci. 1997, 126, 91-105.
(31) Tezel, A.; Mitragotri, S. On the Origin of Size-Dependent
Tortuosity for Permeation of Hydrophilic Solutes across the
Stratum Corneum. J. Controlled Release 2003, 86, 183-186.
(32) Tam, C. M.; Tremblay, A. Y. Membrane Pore Characterization
- Comparison Between Single and Multicomponent Solute
Probe Techniques. J. Membr. Sci. 1991, 57, 271-287.
VOL. 39, NO. 13, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 5029
(33) Giddings, J. C.; Kucera, E.; Russel, C.; Myers, M. N. Statistical
Theory for the Equilibrium Distribution of Rigid Molecules in
Inert Porous Networks: Exclusion Chromatography. J. Phys.
Chem. 1968, 72, 4397-4408.
(34) Seidel, A.; Waypa, J.; Elimelech, M. Role of Charge (Donnan)
Exclusion in Removal of Arsenic from Water by a Negatively
Charged Porous Nanofiltration Membrane. Environ. Eng. Sci.
2001, 18, 105-113.
(35) Kosutic, K.; Kastelan-Kunst, L.; Kunst, B. Porosity of Some
Commercial Reverse Osmosis and Nanofiltration Polyamide
Thin Film Composite Membranes. J. Membr. Sci. 2000, 168,
101-108.
(36) Marcus, Y. IonSolvation; JohnWiley &Sons Limited: NewYork,
1985.
(37) Van der Bruggen, B.; Schaep, J.; Wilms, D.; Vandecasteele, C.
Influenceof Molecular size, PolarityandChargeontheRetention
of Organic Molecules by Nanofiltration. J. Membr. Sci. 1999,
156, 29-41.
(38) Chan, K.; Matsuura, T.; Sourirajan, S. Effect of EvaporationTime
on Pore Size and Pore Size Distribution of Aromatic Polyami-
dohydrazide RO/UF Membranes. Ind. Eng. Chem. Prod. Res.
Dev. 1984, 23, 492-500.
(39) Nguyen, T. D.; Chan, K.; Matsuura, T.; Sourirajan, S. Effect of
Shrinkage on Pore Size and Pore Size Distribution of Different
Cellulosic Reverse Osmosis Membranes. Ind. Eng. Chem. Prod.
Res. Dev. 1984, 23, 501-508.
(40) Wendt, R. P.; Klein, E. Membrane Hetroporosity and the
Probability Function Correlation. J. Membr. Sci. 1984, 17, 161-
171.
(41) Chen, J. Y.; Nomura, H.; Pusch, W. Temperature Dependence
of Membrane Transport Parameters in Hyperfiltration. Desal.
1983, 46, 437-446.
(42) Chan, K.; Tinghui, L.; Matsuura, T.; Sourirajan, S. Effect of
Shrinkage on Pore Size and Pore Size Distribution of Cellulose
Acetate Reverse Osmosis Membranes. Ind. Eng. Chem. Prod.
Res. Dev. 1984, 23, 124-133.
(43) Koros, W. J.; Woods, D. G. Elevated Temperature Application
of Polymer Hollow-Fiber Membranes. J. Membr. Sci. 2001, 181,
157-166.
(44) Kosutic, K.; Kunst, B. Effect of Hydrolysis onPorosityof Cellulose
Acetate Reverse Osmosis Membranes. J. Appl. Polym. Sci. 2001,
81, 1768-1775.
(45) Kokubo, K.; Sakai, K. Evaluation of Dialysis Membranes Using
a Tortuous Pore Model. AIChE J. 1998, 44, 2607-2619.
(46) Van der Bruggen, B.; Geens, J.; Vandecasteele, C. Influence of
Organic Solvents on the Performance of Polymeric Nanofil-
tration Membranes. Sep. Sci. Technol. 2002, 37, 783-797.
Received for review January 20, 2005. Revised manuscript
received April 14, 2005. Accepted April 26, 2005.
ES0501363
5030 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 13, 2005

You might also like