Cf

LSa

b

a

ARRAA

KCBMDD

1

ha

duiotivsl

tasn

A

(

0d

Journal of Membrane Science 351 (2010) 36–43

Contents lists available at ScienceDirect

Journal of Membrane Science

journa l homepage: www.e lsev ier .com/ locate /memsci

haracterization and evaluation of carbon nanotube Bucky-Paper membranesor direct contact membrane distillation

udovic F. Duméea,b,∗, Kallista Searsa,∗∗, Jürg Schütza, Niall Finna, Chi Huynha,tephen Hawkinsa, Mikel Dukeb, Stephen Grayb

CSIRO Materials Science and Engineering, Bayview Ave, Clayton, Vic 3168, AustraliaVictoria University Werribee Campus, Hoppers Lane, Werribee PO Box 14428, Melbourne, Vic 8001, Australia

r t i c l e i n f o

rticle history:eceived 8 September 2009eceived in revised form 4 January 2010ccepted 12 January 2010

a b s t r a c t

Self-supporting carbon nanotube (CNT) Bucky-Papers have unique structural and surface propertieswhich can be utilised in many applications. In this work we characterised pure self-supporting CNT mem-branes, where CNTs were held together only by Van der Waals forces, and evaluated their potential andperformance in direct contact membrane distillation. The membranes were found to be highly hydropho-

vailable online 20 January 2010

eywords:arbon nanotubeucky-paperembrane

bic (contact angle of 113◦), highly porous (90%), and to exhibit a thermal conductivity of 2.7 kW/m2 h.We demonstrate, as a proof of concept, that self-supporting CNT Bucky-Paper membranes can be usedfor desalination in a direct contact membrane distillation setup with 99% salt rejection and a flux rateof ∼12 kg/m2 h at a water vapour partial pressure difference of 22.7 kPa. Ageing of the membranes bydelamination is a main factor limiting their performance and work is currently under way to address this

posi

irect contact membrane distillationesalination

issue by investigating com

. Introduction

Growing concern over our finite and stressed water resourcesas led to renewed interest into membrane distillation (MD) as anlternative method for water purification and desalination.

MD is a thermally driven process based on a vapour pressure gra-ient across a hydrophobic membrane [1,2]. The most commonlysed configuration is direct contact membrane distillation (DCMD)

n which a hydrophobic membrane acts as a barrier between a feedf hot seawater or brackish water and a permeate of cold freshwa-er (Fig. 1). While liquid cannot cross the membrane, water vapours able to travel through the pores driven by the difference in waterapour partial pressure. Previous studies have shown that evenmall temperature differences between the feed and permeate canead to significant fluxes [3].

In contrast to other desalination techniques such as nanofiltra-

ion and reverse osmosis, DCMD offers a potentially low energynd high rejection route to the desalination of highly dirty oralty waters. However, to become competitive with other desali-ation technologies, the DCMD process needs to be improved and

∗ Corresponding author at: CSIRO Materials Science and Engineering, Bayviewve, Clayton, Vic 3168, Australia. Tel.: +61 3 9545 2107; fax: +61 3 9545 2363.∗∗ Corresponding author. Tel.: +61 3 9545 2107; fax: +61 3 9545 2363.

E-mail addresses: ludovic.dumee@csiro.au (L.F. Dumée), kallista.sears@csiro.auK. Sears).

376-7388/$ – see front matter © 2010 Elsevier B.V. All rights reserved.oi:10.1016/j.memsci.2010.01.025

te material structures.© 2010 Elsevier B.V. All rights reserved.

membranes tailored specifically for this application. Two mainmembrane properties are critical for DCMD. Firstly, in order toachieve efficient vapour transport from the hot to the cold side, themembrane needs to be highly porous and as thin as possible [4,5].Secondly, the pores need to be large enough to facilitate vapourtransport, while having sufficiently small dimensions to avoidmembrane wetting and formation of a direct liquid bridge betweenthe feed and the permeate sides [2–6]. Most previous studieshave investigated membranes formed from poly(propylene) (PP),poly(vinylidene-fluoride) (PVDF) and poly(tetrafluoroethylene)(PTFE) [7]. While these materials are highly hydrophobic, they arealso expensive and difficult to process. It is consequently importantto investigate other alternatives as well as techniques for improv-ing the process efficiency by modifying the membrane propertiesand structure.

One possibility is to explore carbon nanotube (CNT) based mem-branes. CNTs are nano-scale cylinders of rolled up graphene withinner diameters as small as 0.7 nm (Fig. 2). They have exceptionalmechanical, electrical and thermal properties, and it was recentlyrevealed that they exhibit fast fluid flow through their interior. Thishas stimulated intense interest in CNT based membranes of variousstructures [8–11]. In this work we focus on CNT Bucky-Paper (BP)

membranes, which is a paper-like structure of CNTs processed byvacuum filtration [12]. We then apply CNT BP membranes to thenovel application of MD.

For the application of MD, the properties surrounding the nan-otubes are of key interest. Here we utilise various microscopy,

L.F. Dumée et al. / Journal of Membra

F

putctNst

2

ast

f

k

wbmvbv

F

ig. 1. Schematic showing the concept behind direct contact membrane distillation.

orosimetry, surface and physical property analytical devices toncover valuable features of the CNT BP membranes and relatehis to performance in MD. The CNT BP membranes have beenharacterised using various techniques including scanning elec-ron microscopy (SEM), transmission electron microscopy (TEM),itrogen adsorption, contact angle measurements, mechanical ten-

ile tests, and thermal conductivity measurements. They were alsoested against commercial PTFE membranes.

. Theory

Theoretical derivations around MD mass and energy transportre well described [13]. Here we present the relevant theory for ourtudy. Two properties, permeance, f, and permeability, k, are usedo quantify mass transport across a membrane and are defined as

= J

A × �P(1)

= J × L

A × �P(2)

here J (kg/s) is the flux through the membrane, L (m) the mem-2

rane thickness, A (m ) the membrane area and, in the case of

embrane distillation, and �P (Pa) is the difference in waterapour partial pressure across the membrane. The permeance cane determined by taking the gradient of a plot of flux against theapour pressure difference, and dividing by the membrane area.

ig. 2. TEM image of a multi-walled CNT showing a number of concentric walls.

ne Science 351 (2010) 36–43 37

The permeability is simply the permeance multiplied by the mem-brane thickness.

If the temperatures of the hot and cold streams are known, thewater vapour partial pressure can be estimated based on Antoine’sequation:

P = e(23.328−(3841/(T−45))) (3)

where the vapour partial pressure P is in Pa and the temperature Tin Kelvin (K).

For direct contact membrane distillation it is generally acceptedthat mass transport through the membrane can operate in one ofthree regimes: (i) Knüdsen flow, (ii) molecular flow or (iii) a com-bination of both known as transition flow. The dominating regimeis indicated by the Knüdsen number (Kn), defined as the ratio of themean free path (�) to the pore diameter (d) [14–16]. The mean freepath is defined as

� = RT√2�d2NAP

(4)

where R is the Boltzmann constant, T the temperature in Kelvin, dthe average particle diameter, NA the Avogadro number and P is theworking pressure.

For Knüdsen numbers, Kn > 1, the molecule–pore wall collisionsdominate leading to Knüdsen type flow: For isoporous membraneswith circular pores the permeance is given by [7–17]

fK = 23

×√

8M

�RTa×

(rε

�ı

)(5)

where R represents the gas constant (J/(K mol)), M the molar mass(g/mol), Ta the average temperature (K), r the average pore radius, ıthe membrane thickness (m), �, the tortuosity and ε, the membraneporosity.

At the other extreme, for Kn < 0.01, molecule-molecule collisionsdominate leading to molecular flow through the stagnant air filmtrapped in the membrane. The permeance is then given by

fM = PT DM

RTa

1Pln

ε

�ı(6)

where D is the diffusivity of water vapour in air, PT is the total pres-sure (air and water vapour) within the membrane and is assumedto be 101.325 kPa in this work, Pln is the mean log average of theair pressure in the membrane defined as

Pln = Pair2 − Pair1

ln(Pair1/Pair2)(7)

where Pair1 and Pair2 represent the air partial pressure at the hotand cold side of the membrane, respectively.

At intermediate Knüdsen numbers 1 > Kn > 0.01, a combinationof both transport mechanisms can occur. The total resistivity of thesystem is then equal to the sum of both Knüdsen and molecularresistivity. This was modelled as resistances in series in previousstudies [15], leading to the following equation for permeance:

fT = 1(1/fK ) + (1/fM)

(8)

Temperature polarisation is an important effect that candegrade MD performance. One method to estimate its extent isthrough the calculation of the temperature polarisation coefficientdefined as

� = Tm1 − Tm

2

Tb − Tb(9)

1 2

where the subscripts 1 and 2 refer to the hot and the cold side,respectively, Tm stands for the temperature of the membrane sur-face and Tb is the temperature in the bulk solution, which is takenas the average of the inlet and outlet temperatures.

3 embra

3

3

ti(6∼

Csv4((f

Mcp

3

3

pStpp(caycmafaded

db(bp1smtbd

pst

3

foe

8 L.F. Dumée et al. / Journal of M

. Experimental details

.1. Material and processing

The CNTs were grown by chemical vapor deposition. A 1–5 nmhick iron catalyst film was deposited onto a silicon substrate bear-ng a thin silicon dioxide layer. A mixture of helium (95%)–acetylene5%) was used as the carbon feedstock and heated to between50 and 750 ◦C. The CNTs typically have an outer diameter of10–15 nm (Fig. 2) and length of 150–300 �m [18].

The CNT BP membranes were processed by vacuum filtration ofNTs dispersed in 99.8% pure propan-2-ol [19]. Well-dispersed CNTuspensions were obtained by repeated sonication for 15 min inter-als at a power of 150 W. Vacuum filtration was performed with a7 mm diameter Millipore filtration unit with house line vacuumdP = −95 kPa). The CNTs were filtered onto a poly(ethersulphone)PES) 0.2-�m pore size Millipore membrane and then pealed off toorm a self-supporting membrane.

PTFE membranes with 0.22 �m pore size were purchased fromillipore and tested for comparison. The PTFE membranes were

omposed of a thin (∼30 �m) layer of pure stretched PTFE on aoly-ethylene grid support.

.2. Membrane characterisation techniques

.2.1. Pore sizeThree complementary techniques were used to characterise

ore size. Firstly a surface apparent pore size was determined fromEM images taken with a Philips FEG SEM at 2 kV. Each image wasaken under the same conditions to reduce errors when comparingore sizes between membranes. At least three images per sam-le were taken and the scans were performed at low frequencies120 mS) to increase the pixel content. The CNTs were sufficientlyonductive to not require metal coating which also avoided anydditional error in the pore size estimation. The images were anal-sed using the software, ISIS and the edge-pores were not taken inonsideration for the calculations. The minimum, mean and maxi-um average pore size distributions were calculated using Feret’s

pproach [13–20]. In this method, three diameters are calculatedor each pore, corresponding namely to the minimum, the meannd the maximum pore size, which gives information on pore shapeistribution. However, this method only presents a surface appar-nt pore size and other tests were required to confirm the pore sizeistribution.

The second method relied on particle size exclusion tests toetermine the particle size cut-off. These tests were performedy filtrating bovine serum albumin (BSA) and nano-particlespurchased from Sigma–Aldrich) through the Bucky-Paper mem-ranes. BSA has an average diameter of 5 nm while the nanoarticle diameters were 50 nm (polystyrene-amine modified),00 nm (polystyrene-sulphate modified) and 500 nm (polystyrene-ulphate modified). The particle diameter distribution waseasured using a Zeta Sizer and agreed well with the manufac-

urer’s specifications. The solute concentration was analysed eithery UV–vis spectrophotometry or fluorescence spectrophotometryepending on the reactivity of the components.

Finally, porometer measurements were made on a capillary floworometer from Porous Material Inc., NY USA. In addition to poreize distribution this also provides a bubble point measurement forhe membranes.

.2.2. Membrane porosityThe membrane porosity was measured using a 2 mL pycnometer

ollowing the procedure outlined by Smolders et al. [5]. A minimumf three samples was used to calculate the average porosity. Thexperiment involved measuring the mass of sample in two differ-

ne Science 351 (2010) 36–43

ent liquids. The first liquid (propan-2-ol) wets the membrane andenters the pores, while the second liquid (deionised water) doesnot wet or fill the pores. Once immersed, the sample was left for20 min and then sonicated for 2 min at low power (150 W and 20 ◦C)to remove any trapped air bubbles. Furthermore, prior to changingliquids the sample was dried in a vacuum oven for 20 min. Theporosity was then calculated according to

ε = 1 − �b

�a× (M1 + Ma

2 − Ma3)

(M1 + Mb2 − Mb

3)(10)

where ε is the porosity, the superscripts a and b correspond topropan-2-ol and water, respectively, � is the liquid density, M1 isthe mass of the dry sample, M2 is the mass of the bottle filled withliquid and M3 is the mass of the bottle plus sample filled with liquid.

3.2.3. Surface areaAn average BET surface area was determined by N2 adsorp-

tion on a Micromeretics Tristar 3000 [21]. The samples were firstdegassed for 70 h and then analysed at 77 K.

3.2.4. Thickness and cross-sectionAn FEI Nova nanolab 200 Dual Beam focused ion beam (FIB) was

used to form cross-sections of the BP membranes. Milling was per-formed with a 1 nA, 30 kV Ga ion beam, followed by 0.3 nA cleaningsteps.

3.2.5. Thermal conductivityThermal conductivity tests were preformed on 100 �m thick

BP using an Armfield Limited Thermal conductivity meter HT1-A.The samples were placed between two copper cylinders. A corre-lation between the power provided to the circuit and the thermalconductivity can be calculated according to

�th = ˘ × ı

Acontact(11)

where � is the thermal conductivity (W/(m2 K)), ˘ the input power(W), ı the thickness (m) and A is the area of contact (m2) of the sam-ple with the copper cylinders. The power used was 10 W and thecooling water temperature was 12 ◦C. Stable measurements wereachieved within 5 min for each sample.

3.2.6. Temperature polarisationTo determine the temperature polarisation coefficient, thermo-

couples were placed at the inlets and outlets of the membranemodule for both cold and hot streams. Two extra thermocoupleswere placed in contact with either side of the membrane surface toestimate their temperature. Those thermocouples were placed tosandwich the membrane without damaging it and the tests wereperformed at six different feed temperatures ranging from 25 to85 ◦C.

3.2.7. Mechanical tensile behaviourTensile mechanical tests of the self-standing BPs were per-

formed on an Instron with a 10 N load cell and a 2.5 mm/min rate ofextension. The sample strips were 10 mm in length and 2–2.5 mmin width.

3.2.8. Contact angleA pocket goniometer PG-3 from Fibro Systems was used to

determine contact angle. The tests were performed in dynamicmode at 20 ◦C with 4 �L drops of deionised water [22] and over

10 min. Each type of membrane was tested at least 3 times.

3.2.9. Limit entry pressure (LEP) and bubble pointLEP tests were performed in an Amicon cell. The pressure was

increased in 5 psi increments and the samples maintained at each

L.F. Dumée et al. / Journal of Membrane Science 351 (2010) 36–43 39

pmt

3

(hCrtlwdoahsoitocmvp

4

4

SsTCCpwfCu

B

Fig. 3. Schematic of the direct contact membrane distillation test rig.

ressure for 10 min. Bubble point measurements were done withethanol following ASTM F316. The bubble point was also charac-

erised with a Porometer (cf. Section 3.2.1).

.3. Direct contact membrane distillation setup

A schematic of the test rig is shown in Fig. 3. A peristaltic pumpCole Palmer Masterflex, Model 7521-25) fitted with two coupledeads (Easyload II, Model 77200-60) was used. The heater (CS Lauda6) and cooler (Thermo Scientific Neslab RTE-7) enabled a wideange of operating temperatures. The electrical conductivity andemperature of the hot and cold electrolytes as well as the waterevel transferred to the cold side were monitored over time. Tests

ere performed in a PTFE module in counter current flow, witheionised water, on the cold side, and synthetic seawater on the hotne (35 g/L NaCl solutions). The cell was cylindrical, with a diameternd height of 25 mm and 2 mm, respectively. The flow rate for eachead was kept at 200 mL/min, while the temperatures of the watertreams were kept constant. The membrane tested area was a discf 25.4 mm of diameter. No major temperature drop between thenlet and the outlet of the module were recorded and the tempera-ure was considered constant over the membrane. For calculationf the partial pressure differences the temperatures on the hot andold reservoirs were used. As such they may differ from the hueembrane temperature. Consequently, the water vapour pressure

alue calculated based on these are referred to as global partialressures.

. Results

.1. Membrane characterisation

The structure of the BP membrane was clearly discerned fromEM images of the surface, such as in Fig. 4A and B, and of cross-ections milled by Focused Ion Beam (FIB), as presented in Fig. 4C.he BP membrane was essentially a mat of randomly entangledNTs. Furthermore the cross-sectional images showed that theNTs tend to form a layered structure with very few CNTs alignederpendicular to the surface. The surface was relatively smoothith a roughness ranging from a few hundred nanometres to a

ew micrometers, as visible in Fig. 4A. Thanks to the well-dispersedNT suspensions, the membrane surface and cross-section wereniform and free from large CNT-bundles [8].

Three techniques were used to characterise the pore size of theP membranes. SEM images yielded to a surface apparent pore size

Fig. 4. SEM images of (A and B) the BP membrane surface (tilt 52◦ and 0◦ , respec-tively), and (C) a cross-section milled using the focused ion beam (tilt 52◦).

distribution as shown in Fig. 5. This gave an average pore size of∼22 nm with ∼95% of the pores falling below 70 nm. This valuecompared well with the particle size exclusion test, where a cut-offof >50 nm was determined by filtration of latex beads as presentedin Fig. 6.

One of the most important membrane parameters is its ther-mal conductivity which directly impacts the process efficiency.Most of the heat transferred across the membrane should be car-

ried with the vapour and heat losses due to conduction throughthe membrane material should be minimised. Using the thermalconductivity meter, the thermal conductivity of the BPs was mea-sured to be ∼2.7 W/(m2 K). This is low in comparison to the highthermal conductivities (∼700 W/(m2 K)) that have been predicted

40 L.F. Dumée et al. / Journal of Membra

Fig. 5. Pore size distribution of the BP membrane as determined by SEM imagingof the surface. Minimum, mean and maximum Feret pore diameters were definedas the average values over the surface of the studied SEM images. Each pore wasdefined by a combination of three vectors.

atomtC

TS

a

b

c

d

Fig. 6. Pore size selectivity determined from particle exclusion tests.

nd reported for individual multi-walled CNTs [23–25]. For MD,

his smaller conductivity is preferred from a performance pointf view. The difference is likely due to the high porosity of theembrane structure combined with the BP’s layer structure and

he relatively poor heat conductivity at interconnects betweenNTs. Other experiments are under way to fully characterize the

able 1ummary of characterisation results for the BP and PTFE membranes.

Property Units

Pore size (nm) nmSEM minimuma

SEM meana

SEM maximuma

Porometer meanCut-offc

Porosity %BET surface area m2/gThermal conductivity W/(m2 K)Density kg/m3

Mechanical tensile specific modulusd GPa m3/gContact angle ◦

Bubble point kPaLiquid entry pressure kPa

Value given in brackets is the standard deviation.Thickness of test was 5 �m (±0.2 �m).For 90% rejected particles.Data taken from Millipore specification data sheet.

ne Science 351 (2010) 36–43

thermal behaviour of the BP structures. For comparison, thermalconductivities given in the literature for other polymeric mem-branes commonly used for MD lie between 0.25 W/m K (PTFE) and4 W/m K (PVDF). Our BP thermal conductivity was experimentallyshown by direct thermal conductivity measurement to lie betweenthe two mainly used materials.

Various other techniques were also used to characterise themembranes and determine their suitability for membrane distil-lation. These results are summarised in Table 1 and where possibleare also compared with values for a typical PTFE membrane. Table 1shows that the BP membrane properties are generally comparableto those of the PTFE membrane making them a promising candi-date for MD. For example, pycnometer tests indicated a porosity of90%, exceeding the 70% measured for PTFE. The BP membrane alsoexhibited a high contact angle of ∼113◦ close to that of PTFE (120◦).

4.2. DCMD

4.2.1. Flux and permeabilityA number of PTFE and BP membranes were tested in a DCMD

setup. The test conditions and membrane details are given inTable 2.

Fig. 7 illustrates how the flux varies over time for representa-tive PTFE and BP membranes. PTFE membranes are amongst thebest available for MD [23] and as expected, the flux reaches steadystate after an initial settling period of ∼15–20 min. The BPs on theother hand behaved differently. The flux initially increased over thefirst 15 min before stabilising. However, this stabilisation is oftenfollowed by a slow decline until the membrane breaks. The origin ofthis declining flux is still uncertain. This could be due to a numberof factors. It may be related to temperature polarisation as the BPmembranes have a thermal conductivity 10 times higher than thatof PTFE although lower than PVDF. However, preliminary experi-ments seem to indicate that temperature polarisation (Eq. (9)) is nota severe effect. For all feed temperatures investigated, the tempera-ture polarisation coefficients were measured for both the PTFE andBP membranes and were always greater than 0.9 (Fig. 8). This seemsto indicate that temperature polarisation is not a dominant effect.However further work is still under way to more thoroughly char-acterise the effects of temperature polarisation and to determineits dependency on temperature and membrane structure.

Instead we suspect that the decline in BP membrane flux withtime is due to physical ageing of the membrane including micro-crack formation, which is discussed later in greater detail. Thedeclining flux of the BP membranes may also be linked to a pro-gressive compaction of the membrane due to pressure from the hot

BP PTFE

20.69 (10.27) –27.28 (16.7) 220b

32.37 (22.5) –44 54850–100 500d

90 70197 21

2.3 0.25232.5 293.8

5.27 –113.3 120.4

37.92 99.9b

55.15 344.7

L.F. Dumée et al. / Journal of Membrane Science 351 (2010) 36–43 41

Table 2Membrane properties, test conditions and rejection efficiency.

Membrane Mass (mg) Thickness (�m) Tave (◦C) �T (◦C) �P (kPa) Rejection (%)

PTFE20.9(8.61)

205(30)a

41 29.1 12.7 10032 46.8 15.02 10039 59.5 28.81 10045 65.9 42.29 100

BP4.75 45 27 41.2 10.89 99.996.65 55 45 63.8 40.43 88.467.98 42 36 53.2 22.73 94.08

a The thickness of the complete structure is ∼205 �m while the active layer is ∼30 �m thick.

Fig. 7. Representative curves for BP and PTFE membranes showing the variation influx with time. The data was acquired for a cold and hot-side flow rate of 300 mL/h,ai27

atb

patpt

Ft

Fig. 9. Dependence of flux on global vapour pressure difference, �P, for the BP (black

nd an initial hot-side conductivity of ∼10 �S/cm. The working temperatures usedn the experiments corresponded to various global vapour pressure differences: at2 kPa (Hot: 65 ◦C; Cold: 5 ◦C); at 28 kPa (Hot: 70 ◦C; Cold: 5 ◦C) and at 44 kPa (Hot:7 ◦C; Cold: 5 ◦C).

nd cold water streams [24,25]. A previous study has shown thathe permeate pressure has an impact on the flux and membraneehaviour [3].

The flux for the BP and PTFE membranes after 2 h of operation islotted in Fig. 9 as a function of the global partial pressure difference

cross the membrane (bottom axis). The top axis, corresponding tohe temperature of the hot stream is given for reference. The globalartial pressure differences were calculated using the tempera-ure in the bulk feed and permeate in the reservoirs and applying

ig. 8. Temperature polarisation as a function of feed temperature. The test condi-ions were kept similar as a classical DCMD test.

curve, left Y axis) and PTFE (grey curve, right Y axis) membranes. The temperatureof the hot-side corresponding to the calculated dP is given on the top X axis as aguideline.

Antoine’s law (Eq. (3)). As a consequence, the measured tempera-ture may still differ from the true temperature at the membranesurface due to temperature polarisation effects and potential heatlosses. To distinguish this effect we refer to a “global” partial pres-sure difference.

The flux through the PTFE membranes, depending on the oper-ating temperatures, is 5–12 times greater than that for the BPmembranes (Fig. 9). As expected, the PTFE membranes showa linear dependence on the global partial pressure difference.The slope of a linear fit to the data was calculated and multi-plied by the membrane thickness to give a global permeability of∼21.1 × 10−12 kg/(m s Pa). However, in the case of the BP mem-branes, where a stable and steady flux is not always observed, itis difficult to accurately determine the dependence on global par-tial pressure difference. A linear fit to the data in Fig. 9 was usedto estimate a permeability on the order of 0.8 × 10−12 kg/(m s Pa),which is ∼25 times lower than the PTFE membranes.

A number of factors may explain the lower fluxes and perme-ability measured for the BP membranes.

First of all, calculated Knudsen number for both the BP and PTFEmembranes within the range of working temperatures were calcu-lated to be respectively between (5.35; 6.80) and (0.48; 0.61) whichplaces them in a Knüdsen/transition flow regime, with a dominantKnüdsen mode. In the Knüdsen transport regime, the permeabilityis directly proportional to the pore radius, making this an impor-tant parameter. Our measurements indicate that the BP membrane

has pores which are an order of magnitude smaller than those ofthe PTFE membrane and this could therefore be an important limit-ing factor in BP performance. This can be addressed in the future bytailoring the pore size distribution of the BP membranes. Work pre-

42 L.F. Dumée et al. / Journal of Membrane Science 351 (2010) 36–43

Fa

vwwt

pmpasciaitp

Bdwaics

4

mbtcsorBcarmb

tbcbM

Fig. 11. SEM images at 5 keV and 7.5 mm w.d. showing a cross-section through a

ig. 10. Representative conductivity curves recorded for the cold permeate side asfunction of time. The conductivity of the hot stream was ∼11 mS.

iously done by us showed that mixing CNTs of different diameterould lead to different structures and average pore size [26] and itas also shown that porosity could be tailored by using polymeric

emplates inside the structure of BPs [27].A second possible reason for the difference in flux rate and

ermeability may be the distinctly different structure of the twoembranes. While the PTFE membrane has more conventional

ores, which traverse the entire membrane thickness, the BP poresre defined by an interconnected network of CNTs (Fig. 4). The BPtructure is therefore closer to that of a non-woven fabric and theoncept of a cylindrical pore not an appropriate simplification. Thisnterconnected CNT network also results in a very high surfacerea, ∼20 times greater than that for PTFE. This could lead to anncreased number of molecular-wall collisions and hence a moreortuous path for water vapour through the BP membrane reducingermeability.

The most important outcome of this work is that self-supportingP membranes were successfully operated in a DCMD setup toesalinate synthetic seawater. A flux on the order of 5–25 kg/m2 has measured for “global” vapour pressure differences between 10

nd 40 kPa. This is encouraging given that the BP membranes weren no way supported and that no binding agents were used in theironstruction. The CNTs within the BP membrane are held togetherolely by Van der Waals interactions.

.2.2. Membrane integrity for salt rejectionMembrane integrity is an important parameter affecting both

embrane lifetime and performance. The integrity of the BP mem-ranes was investigated in two ways. Firstly, the conductivity ofhe cold reservoir was monitored with time. If only water vapourrosses the membrane, then the electrical conductivity of the coldolution should progressively decrease in proportion to the amountf water transferred. PTFE membranes show this exact behaviour,epresenting close to 100% salt rejection (Fig. 10). In contrast, mostP membranes showed a very slight but progressive increase inonductivity corresponding to desalination efficiencies between 88nd 99% as shown in Table 2. Furthermore, after 4–7 h of testing aapid increase in conductivity was typically observed for the BPembranes indicating that liquid water was breaching the mem-

rane allowing salt passage from the hot to the cold stream.Secondly, the membranes were characterised after testing in

he MD setup by milling sections through the BP with a focused ioneam milling followed by SEM imaging. As shown in Fig. 11, micro-racks were present and seem to represent regions of delaminationetween the CNT layers. These micro-cracks likely develop duringD testing and eventually form a direct bridge from the feed to

BP membrane after testing. The section was milled with a focused ion beam andthe image was taken at a 52◦ tilt. The left inset shows EDX curves taken inside andwell away from the crack. The right inset shows a low magnification view of thecross-section.

permeate, leading to the rapid increase in permeate conductivitymentioned above. This is further supported through EDX analy-sis which confirmed the presence of sodium and chloride withinthe micro-cracks, while none was detected in regions away fromthe cracks. It is suspected that the ageing of these membranes isaccelerated due to the MD setup which incorporated two pulsat-ing peristaltic pumps. This causes an oscillation of the membranes,and potentially delamination of sheets nanotubes as they are heldtogether solely through Van der Waals forces [28,29]. It is likelythat the gradual formation of micro-cracks contributes to the lossof flux observed during operation for the BP membranes. Variousapproaches, such as composite structures and the addition of poly-mers are currently being investigated to avoid the formation ofcracks.

5. Conclusions

CNT BP membranes were fabricated by vacuum filtration andcharacterised to determine their suitability for MD. They werefound to have many properties favourable for MD. For example theyexhibit a high contact angle (113 degrees), high porosity (90%), andrelatively low thermal conductivity of 2.7 kW/m2 h. These proper-ties are better than most of polymeric membranes conventionallyused in MD and compare well with PTFE membranes.

Most importantly desalination was successfully demonstratedusing CNT BP membranes in a DCMD setup. The best results gave99% salt rejection and a flux rate of 12 kg/m2 h at a water vapour par-tial pressure difference of 22.7 kPa. Some issues were encounteredduring DCMD testing such as a decline in flux with time and delam-ination of BP membranes due to the formation of micro-cracks.Lifespan of the BP membranes is still inferior to PTFE membranes.However it is important to stress that these BP membranes werenot supported and were composed solely by CNTs. Work is cur-rently underway to optimise the BP membranes with an emphasison supported and composite structures for improved lifetime andpermeability. The pore size of BP membrane can be tailored andwork is currently done to have engineered pore size distribution

while keeping high porosity membranes.

In summary, carbon nanotube based membranes have manyphysical and surface properties beneficial for MD. This work was aproof of concept that CNT structures are suitable for water desalina-tion. Our results to date indicate that BP membranes are a promising

embra

ai

A

IEVeGfXNfmUt

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

L.F. Dumée et al. / Journal of M

lternative to current polymeric membranes although further works needed to fully understand and optimise their properties for MD.

cknowledgments

The author would like to thank the Commonwealth Scientificndustrial and Research Organization (CSIRO) Materials Science andngineering and the Institute for Sustainability and Innovation atictoria University for financial support. We also wish to acknowl-dge the expert advice and assistance of Dr. John Ward and Markreaves on SEM, and Dr. Sergey Rubanov and Dr. Kenneth Goldie

or focused ion beam milling. We would also like to thank Zongliie and Lisa Wong for their help with BET measurements, Dr. Julieigro for her advice with the fluorescence tests, Dr. Shane Cox,

rom UNSW – Australia, for his assistance with Porometer measure-ents and finally, Dr. Jun De-Li and Jianhua Zhang, from Victorianiversity – Australia, for their assistance with heat conductivity

ests.

eferences

[1] M.S. El-Bourawi, Z. Ding, R. Ma, M. Khayet, A framework for better understand-ing membrane distillation separation process, Journal of Membrane Science285 (1–2) (2006) 4–29.

[2] K.W. Lawson, D.R. Lloyd, Membrane distillation. II. Direct contact MD, Journalof Membrane Science 120 (1) (1996) 123–133.

[3] K.W. Lawson, D.R. Lloyd, Membrane distillation, Journal of Membrane Science124 (1) (1997) 1–25.

[4] M. Khayet, T. Matsuura, Preparation and characterization of polyvinylidenefluoride membranes for membrane distillation, Industrial & Engineering Chem-istry Research 40 (24) (2001) 5710–5718.

[5] K. Smolders, A.C.M. Franken, Terminology for membrane distillation, Desalina-tion 72 (3) (1989) 249–262.

[6] M. Gryta, M. Tomaszewska, K. Karakulski, Wastewater treatment by membranedistillation, Desalination 198 (1–3) (2006) 67–73.

[7] D. Basmadjian, Mass Transfer: Principles and Applications, 2004. CRC Press,ISBN: 0-8493r-r2238-1.

[8] B. Corry, Designing carbon nanotube membranes for efficient water desalina-tion, The Journal of Physical Chemistry B 112 (5) (2008) 1427–1434.

[9] F. Fornasiero, H.G. Park, J.K. Holt, M. Stadermann, C.P. Grigoropoulos, A. Noy, O.Bakajin, Ion exclusion by sub-2-nm carbon nanotube pores, Proceedings of theNational Academy of Sciences of the United States of America 105 (45) (2008)17250–17255.

10] B.J. Hinds, N. Chopra, T. Rantell, R. Andrews, V. Gavalas, L.G. Bachas, Alignedmultiwalled carbon nanotube membranes, Science 303 (5654) (2004) 62–65.

[

[

[

ne Science 351 (2010) 36–43 43

11] J.K. Holt, H.G. Park, Y.M. Wang, M. Stadermann, A.B. Artyukhin, C.P. Grigoropou-los, A. Noy, O. Bakajin, Fast mass transport through sub-2-nanometer carbonnanotubes, Science 312 (5776) (2006) 1034–1037.

12] H. Muramatsu, T. Hayashi, Y.A. Kim, D. Shimamoto, Y.J. Kim, K. Tantrakarn,M. Endo, M. Terrones, M.S. Dresselhaus, Pore structure and oxidation stabil-ity of double-walled carbon nanotube-derived Bucky-Paper, Chemical PhysicsLetters 414 (4–6) (2005) 444–448.

13] A. Bessieres, M. Meireles, R. Coratger, J. Beauvillain, V. Sanchez, Investigations ofsurface properties of polymeric membranes by near field microscopy, Journalof Membrane Science 109 (2) (1996) p13.

14] A. Burgoyne, M.M. Vahdati, Direct contact membrane distillation, SeparationScience and Technology 35 (8) (2000) 1257–1284.

15] J. Phattaranawik, R. Jiraratananon, A.G. Fane, Effect of pore size distribution andair flux on mass transport in direct contact membrane distillation, Journal ofMembrane Science 215 (1–2) (2003) 75–85.

16] R.W. Schofield, A.G. Fane, C.J.D. Fell, R. Macoun, Factors affecting flux in mem-brane distillation, Desalination 77 (1–3) (1990) 279–294.

17] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, rev. 2nd ed.,John Wiley & Sons Inc., New Yor/Chichester/Weinheim/Brisbane/Singapore/Toronto, 2006.

18] C. Huynh, S. Hawkins, Understanding the synthesis of directly spinnable carbonnanotube forests, Carbon, submitted to publication in Carbon.

19] J. Gou, Z.Y. Liang, B. Wang, Experimental design and optimization of dispersionprocess for single-walled carbon nanotube Bucky-Paper, International Journalof Nanoscience 3 (3) (2004) p14.

20] A. Hernández, J.I. Calvo, P. Prádanos, F. Tejerina, Pore size distributions in micro-porous membranes. A critical analysis of the bubble point extended method,Journal of Membrane Science 112 (1) (1996) 1–12.

21] S.M. Cooper, H.F. Chuang, M. Cinke, B.A. Cruden, M. Meyyappan, Gas perme-ability of a Bucky-Paper membrane, Nano Letters 3 (2) (2003) 189–192.

22] S. Nuriel, L. Liu, A.H. Barber, H.D. Wagner, Direct measurement of multi-wall nanotube surface tension, Chemical Physics Letters 404 (4–6) (2005)263–266.

23] C. Cabassud, D. Wirth, Membrane distillation for water desalination: how tochose an appropriate membrane? in: Conference on Desalination and the Envi-ronment – Fresh Water for All, Elsevier Science B.V., Malta, Italy, 2003.

24] F. Laganà, G. Barbieri, E. Drioli, Direct contact membrane distillation: modellingand concentration experiments, Journal of Membrane Science 166 (1) (2000)1–11.

25] K.W. Lawson, M.S. Hall, D.R. Lloyd, Compaction of microporous membranesused in membrane distillation. I. Effect on gas permeability, Journal of Mem-brane Science 101 (1–2) (1995) 99–108.

26] L. Dumée, K. Sears, J. Schütz, N. Finn, M. Duke, S. Gray, Design and Characterisa-tion of Carbon Nanotube Bucky-Paper Membranes for Membrane Distillation,in: ICOM08, Honolulu - Hawaii, USA, 2008.

27] R.K. Das, B. Liu, J.R. Reynolds, A.G. Rinzler, Engineered macroporosity in single-wall carbon nanotube films, Nano Letters 9 (2) (2009) 677–683.

28] D. Chandler, Interfaces and the driving force of hydrophobic assembly, Nature437 (7059) (2005) 640–647.

29] V.A. Parsegian, Van der Waals Forces: A Handbook for Biologists, Chemists,Engineers, and Physicists, Cambridge University Press, 2005.