12354949

7/31/2019 12354949

1/12

Journal of Membrane Science 174 (2000) 177188

Air separation properties of flat sheet homogeneouspyrolytic carbon membranes

Anshu Singh-Ghosal1, W.J. KorosDepartment of Chemical Engineering, University of Texas at Austin, Austin, TX 78712, USA

Received 12 February 1999; accepted 7 March 2000

Abstract

Carbon molecular sieve (CMS) membranes with attractive separation properties were formed by pyrolysis of a polymericprecursor. Defect-free membranes with oxygen/nitrogen permselectivities three times greater than the polymer precursorwere obtained. Gas separation properties were also measured at intermediate stages during the pyrolysis protocol to study theevolution of entropic selectivity, which distinguishes molecular sieving materials from typical polymeric materials. Initially,permeabilities increase dramatically during the pyrolysis process due to an increase in overall sorption coefficients. In thefinally pyrolyzed membrane, however, permeabilities are three times lower than in the polymer precursor due to significantlylower oxygen diffusion coefficients. Nevertheless, the separation properties of the pyrolyzed membranes are well above theso-called property upper-bound trade-off curve often used to compare conventional polymeric materials. The increase inpermselectivity is entirely due to an increase in mobility selectivity. Entropic selectivity increases are responsible for thehigher mobility selectivity in the finally pyrolyzed membranes; however, energetic contributions were more significant formaterials at the intermediate stage. Significant conclusions about the structure of the evolving molecular matrix can be drawnfrom the gas separation results. 2000 Elsevier Science B.V. All rights reserved.

Keywords: Carbon molecular sieves; Membrane separations; Pyrolysis; Entropic selectivity; Air separation

1. Introduction

Polymers have conventionally been used as mate-rials for solutiondiffusion gas separation membraneswith useful separation properties. For non-polar gasmixtures, the separation is based largely on molecularsize, and for similar-sized molecules like oxygen andnitrogen, limited permselectivities can be achieved us-ing polymer membrane materials. A previous study

Corresponding author. Tel.:+

1-512-471-5238;fax: 1-512-471-7060.1 Present address: General Electric Co. 1, LEXAN Lane, Mount

Vernon, IN 47620, USA.

compared diffusivity selectivities offered by glassypolymeric and molecular sieving materials and iden-tified a term referred to as entropic selectivity [1].This selectivity arises from the ability of molecularsieving media to selectively reduce the rotational de-grees of freedom of nitrogen versus oxygen in the dif-fusion transition state. The significant segmental mo-tions present in polymers prevent exercising this finelevel of control, thereby limiting the performance ofpolymeric materials compared to molecular sieves.

In recent years, researchers have produced carbon

molecular sieving (CMS) membranes by the pyroly-sis of polymeric precursors already processed in theform of membranes. Typically, improved selectivi-

0376-7388/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved.PII: S0 376-7388 (00)0039 2-6

7/31/2019 12354949

2/12

178 A. Singh-Ghosal, W.J. Koros / Journal of Membrane Science 174 (2000) 177188

ties for oxygennitrogen separation were observed.Jones and Koros [2] used hexafluoroisopropylidene(6FDA)-based co-polyimide hollow fibers with O2/N2selectivities of 4, to yield asymmetric carbon mem-branes with selectivities of 1114 after pyrolysis at550C. Koresh and Soffer [3] used hollow fibers ofcellulosic or phenolic resins, pyrolyzed at 800 and950C to yield carbon membranes with selectivitiesof 78. Some post-pyrolysis steps were required toopen the structure of the 950C membranes whichhad sintered pores. Shusen et. al. [4] pyrolyzed flatdense films of phenol formaldehyde resin at temper-atures of 800950C with a post-pyrolysis activationstep to yield membranes with selectivities of 11. Ha-tori et al. [5] made flat homogeneous carbon filmsfrom KaptonTM polyimide pyrolysis at 800C to yieldO2/N2 selectivities of 4.2. Suda and Haraya [6] reporthollow-fiber membranes, also from KaptonTM pyrol-ysis at temperatures up to 800C, with selectivities of

11 and at temperatures up to 1000

C with selectivitiesof 23. Hayashi et. al. [7] produced carbon membranesfrom a polyimide film coated on porous alumina tubespyrolyzed at 700C to yield O2/N2 selectivities of 9.7which could be increased to 14 after a post pyrolysiscarbon deposition step.

In addition to producing carbonized membranesfrom pyrolysis of a polymer precursor, it was theobjective of this work to systematically study thedevelopment of entropic contributions to diffusiv-ity selectivity as the polymer matrix evolved to arigid carbon matrix. Diffusion coefficients and ac-tivation energies for diffusion were calculated from

temperature-dependent steady-state permeation andequilibrium sorption. Polymer precursor, finally py-rolyzed membranes and membranes pyrolyzed at in-termediate steps in the pyrolysis protocol were testedto study the development of gas separation proper-ties as the material progresses from a polymer to acompletely carbonized membrane.

2. Background

Non-porous polymers and molecular sieving me-dia like zeolites and carbon molecular sieves (CMS)

transport gases by a similar sorptiondiffusion mech-anism. The permeability of a penetrant through amembrane is measured as a steady-state flux, nor-

malized by partial pressure difference and membranethickness. Permeability of component A can beexpressed as the product of a kinetic factor, the dif-fusion coefficient (DA), and a thermodynamic factor,the sorption coefficient (SA) [8]:

PA = DASA (1)

When the downstream pressure is negligibly low,the sorption coefficient, SA, equals the secant slopeof the sorption isotherm of A evaluated under the up-stream permeation partial pressure conditions. The av-erage diffusion coefficient, DA, can then be calculatedas the ratio ofPA and SA.

The ideal permselectivity, A/B, characterizes theoverall ability of a membrane to separate penetrants Aand B. This is an inherent property of the material andits molecular morphology. The permselectivity can befactored into a diffusivity selectivity and a sorptionselectivity term viz.

A/B =PA

PB=

DA

DB

SA

SB(2)

The sorption selectivity term for the O2/N2 pair liesin the range of 12 in almost all glassy polymers, andbetween 0.7 and 2 for molecular sieving materials likezeolite 4A and CMS. It is the diffusivity selectivity inall three classes of materials that is responsible for theremarkable differences in their separation properties,as observed in Fig. 1.

Permeation and diffusion in high-selectivitymedia are activated phenomena, and their temperature-dependence can be described by the following Arrhe-nius relationship shown for diffusion:

D = D0 expED

RT

(3)

where, D0 is the pre-exponential factor and ED is theactivation energy for diffusion. The pre-exponentialfactor can be expressed as the following equation [9]:

D0 = e2 kT

hexp

SD

R

(4)

where is the average jump length a penetrantmolecule executes in one diffusive jump, SD is the

activation entropy of diffusion and k and h are theBoltzmanns and Plancks constants, respectively.The diffusive jump length depends slightly on the

7/31/2019 12354949

3/12

A. Singh-Ghosal, W.J. Koros / Journal of Membrane Science 174 (2000) 177188 179

Fig. 1. Permeabilitypermselectivity trade-off curve for O2/N2separation properties of polymeric materials (taken from [1]).

penetrant molecule in polymeric media, while it tendsto be a fixed average length between sorption sites forall molecules in rigid molecular sieving media. In thecase of the O2/N2 pair, where the molecular size of thetwo gases is very similar, the differences in diffusion

jump length can be neglected to a first approximation,so by combining Eqs. (3) and (4), the diffusivity se-lectivity can be factored into a product of an energetic

selectivity and an entropic selectivity term viz.DA

DB= exp

SDA SDB

R

Entropic selectivity

exp(EDA EDB)

RT

Energetic selectivity

(5)

Fig. 2. Co-polyimide precursor for pyrolyzed carbon membranes (AP).

As noted above, although conventional CMS materialsand zeolite 4A have a favorable entropic contributionto overall diffusion selectivity, conventional polymericmembrane materials generally lack this factor.

3. Experimental

3.1. Pyrolysis

Carbon molecular sieves (CMS) are readily madefrom the pyrolysis of suitable polymeric precursors.Carbon membranes were produced in this work bythe pyrolysis of easily processable polyimide pre-cursors. The polyimide (AP) used in this study wasa co-polymer with the structure shown in Fig. 2,supplied by E.I. DuPont de Nemours. This mate-rial has a rigid backbone with CF3 groups and thetrimethyl-substituted meta-linked diamine provid-

ing significant hindrance to inter-segmental rotationabout the chain axis. The polyimide material andthe temperaturetime protocol used for our pyrol-ysis experiments were the same as that used byJones and Koros. In that work, however, asymmetrichollow-fiber polymer membranes were pyrolyzed toCMS membranes. To compute the entropic selectivityof materials, the diffusion coefficient must be factoredout of Eq. (1) by determination ofPA. Homogeneousmembranes are required to unambiguously know themembrane thickness to convert observed fluxes to per-meabilities; hence, symmetric, dense polymeric filmsrather than asymmetric membranes were pyrolyzed

in this study.The polymer was cast from methylene chloride to

produce strong, flexible 2560m films. Films werepyrolyzed under vacuum (

7/31/2019 12354949

4/12


Fig. 3. Schematic of pyrolysis furnace.

of the pyrolysis furnace is shown in Fig. 3. A 2in.ID, 33 in. long quartz tube with a glass end cap was

used for pyrolysis in a Thermcraft

muffle furnaceequipped with four heating elements to minimize theaxial and radial temperature gradients. A distance ofat least 6 in. was maintained between the samples andthe open ends of the furnace. An Omega CN2011Programmable Temperature Controller with a triacoutput and PID control was used to control the precisetemperaturetime protocol. A standardized protocolwas followed carefully to obtain reproducible gasseparation properties for the carbonized membranes.The tube was evacuated to pressures of 0.03 mm Hg,as measured by McLeods gauge on the furnace, priorto starting the heating cycle. At this point, the valve

connecting the vacuum gauge to the tube was closedand the pyrolysis cycle was started.An important objective of this work was to study the

changes in properties of the material as it progressesfrom a low-selectivity flexible polymer to a rigidhigh-selectivity carbon. To study this transformationprogressively, pyrolysis was terminated at interme-diate steps in the protocol as described below. Theproperties of the material were tested for the partiallypyrolyzed membranes to observe how the entropic se-lectivity (Eq. (5)) changes with the structure of the ma-terial along the pyrolysis path. The temperaturetimeprotocol shown in Fig. 4 was developed by trial and

error optimization to yield optimum properties for thefinal carbon membrane made from this specific poly-mer precursor and it could vary for other polymers.

The initial rate of heating employed was 13.3C/min.The heating rate was slowed to 3.85C/min until thetemperature reached 535C (PM535). Some mem-branes were tested for gas separation properties atthis point. This is the initial part of the pyrolysisprotocol. From 535C, the heating rate was slowedto 0.25C/min as it approached an intermediate tem-perature of 550C. The membranes were held at thistemperature for a period of 2 h (PM550). Membraneswere tested at this point for gas separation propertiesand entropic selectivity. From 550C, the heating ratewas again increased to 3.85C/min. The final pyrolysistemperature of 800C was again approached at a rate

Fig. 4. Temperaturetime protocol for pyrolysis of carbon mem-branes.

7/31/2019 12354949

5/12


of 0.25C/min and the membranes were held at thistemperature for another 2 h (PM800). The approachrate for the temperatures of interest were maintainedat these low values to avoid an overshoot in the finaltemperatures. The furnace was allowed to graduallycool to temperatures below 40C before samples wereremoved from the furnace and placed in a dessicator.

3.2. Permeation and sorption

The dense polymeric films and carbon membraneswere tested in a standard pressure rise type cell wherethe membranes were pressurized on the upstream sur-face, while the downstream face was maintained atvacuum. The permeation flux was calculated using theideal gas law by measuring the pressure rise in a stan-dard volume on the downstream side of the membraneusing a 010 Torr MKS Baratron pressure transducer.The equipment has been described in detail elsewhere

[10]. A 5-minute epoxy

resin was applied alongthe periphery of the carbon membranes to seal them tothe metal flange of the permeation cell. After curingof the epoxy, the membranes were outgassed in a vac-uum oven at 150C for at least 8 h before permeationmeasurements to ensure complete removal of adsorbedmolecules in the ultramicropores of the material.

Mixed gas permeation experiments were also per-formed on the carbonized membranes using equip-ment described previously [11]. The membranes weretested primarily for oxygen and nitrogen separationonly. Ultra-high purity grade oxygen and nitrogenwere used. For the mixed gas experiments, dry com-

pressed air was used. Temperature-dependent per-meabilities were also measured for the polymer andcarbonized membranes. The temperature of measure-ment was increased gradually, with the membranesbeing tested at lower temperatures first. After test-ing at the highest temperatures, the membrane wastested once again at 35C to ensure that no defects orchanges had appeared during the testing procedure.

Exposure to some organics can cause an irre-versible loss in flux for the carbonized membranes[12]. Back-diffusion of vacuum pump oil was pre-vented by using a cold trap with liquid nitrogen or anFL20K Foreline trap with activated alumina adsorbent

purchased from Edwards High Vacuum International.High-pressure sorption isotherms were generated

for all polymeric materials, and membranes car-

bonized at 550 and 800C. To determine the heatsof sorption, the isotherms were generated at severaltemperatures using standard pressure decay equip-ment described earlier [10]. As in the permeationexperiments, the lower temperature experiments werecarried out prior to the higher temperatures for sorp-tion studies. This procedure prevents the materialfrom being subjected to undue thermal cycling andreduces history-dependent behavior. The sample wasevacuated for at least 10h at the new temperaturebefore performing the higher-temperature experi-ments. Compressibility factors for the gases wereused at each temperature and pressure to account fornon-ideality of the gas. These were calculated fromreliable PVT data for oxygen and nitrogen [13].

3.3. Investigation of possible asymmetry in carbon

membranes

Extremely high oxygen and nitrogen flux values ob-served for samples pyrolyzed at 550C (PM550) led tothe concern that some asymmetry may have developedin the samples during the pyrolysis process. A thin se-lective surface layer in the pyrolyzed samples couldconceivably lead to an overestimation of permeabilitycoefficients. The membranes were not strong enoughto be tested for separation properties before and afterremoving a thin surface layer from the samples, so gastransport data was used to probe the structural sym-metry of the carbonized materials. Nitrogen sorptionat 77 K and a relative pressure of0.995 are typicallyused to measure the total porous volume [14]. No ob-

servable sorption of nitrogen was observed for PM550and PM800 membrane materials at this temperature.The lack of any observable nitrogen sorption at 77 Kimplies that both surfaces of the membrane, the onein contact with the quartz plate and the one on top,have a similar inaccessible porous structure where dif-fusion is in the activated regime [15]. This eliminatesthe possibility of a selective surface layer at only onesurface in the pyrolyzed samples.

There may still have been a possibility of asymme-try in the bulk of the membrane with thin selectivelayers at either surface. To test this hypothesis, porousvolume of the carbonized samples was measured using

high- and low-pressure CO2 sorption isothermsat 0C.This data can be analyzed by the DubininAstakhovequation to estimate the porous volume in the micro-

7/31/2019 12354949

6/12


pore range (

7/31/2019 12354949

7/12


Fig. 6. Air separation properties of polymer precursor and py-rolyzed membranes with respect to the upper-bound curve (mul-tiple points on graph represent replicate pyrolysis samples).

Table 1 to enable comparison of properties. The mem-branes chosen were good representatives of the prop-erties of the material. These membranes are referredto as PM535.1, PM550.1 and PM800.1. All valuesreported in Table 1 have been calculated from a curve

fit equation obtained from temperature-dependent rawdata for each membrane. Results reported here are

Table 1Air separation properties of polyimide precursor and pyrolyzedmembranes at 35Ca

Property AP PM535 PM550 PM800

PO2 (Barrer) 69 952 239 23PO2/PN2 4.1 4.6 9.9 12.3DO2 (10

8 cm2/s) 35.6 35 1.8DO2/DN2 2.9 7.3 8.5SO2 (ccSTP/cc atm) 1.5 5.2 9.7SO2/SN2 1.4 1.3 1.4

a

Pure gas data for oxygen at 20 psi, nitrogen at 75 psi for AP,PM550.1 and PM800.1; mixed gas data at 30 psi total pressure forPM535.1; all values have been calculated form curve fit equationobtained from raw data on each membrane.

Table 2Mixed and pure gas permeability and permselectivity for PM550pyrolyzed membranesa

PO2 (Barrer) PN2 (Barrer) PO2/PN2

Mixed gas 812 108 7.5Pure gas 826 115 7.2

a Pure gas data for oxygen at 10 psi, nitrogen at 48 psi, mixedgas data at 57psi, total pressure with 19% O2 and 81% N2 forPM550.2 membrane.

from pure gas permeation experiments for AP, PM550and PM800. Only mixed gas permeation runs wereperformed for PM535. Due to the extremely highfluxes observed for this material, membranes weretested with mixed gas and at lower pressures in orderto keep the fluxes measurable by the chart recorderbeing used. In amorphous polymers, pure gas perme-abilities are reasonable approximations of mixed gaspermeation. This was shown to be true for the py-

rolyzed membranes as well. In separate experimentswith PM550, it was confirmed that pure gas perme-ability and permselectivity were within 3% of mixedgas permeability and permselectivity [18]. The puregas and mixed gas permeabilities and permselectivitydata are presented in Table 2. Pressures used in thepure gas runs maintain the same ratio of oxygen andnitrogen partial pressures as those used in the mixedgas runs.

The co-polyimide AP is a high-permeability,low-selectivity polymer compared to current gener-ation polymeric materials. The separation propertiesfor this polymer fall below the upper-bound curve

and outside the rectangular region which enclosesproperties for commercially attractive membrane ma-terials. The polymer has high oxygen permeabilitiescompared to current generation polymers despite itsmoderate fractional free volume (FFV) of 0.158. Thehigh permeabilities are a result of high diffusivitiesand high solubilities as well. The low selectivities ofthis polymer are primarily due to the low mobilityselectivity, as solubility selectivities for amorphousglassy polyimides do not vary significantly aroundunity. The polymer backbone, despite its rigid chainwith hindered inter-segmental motion, is likely tohave a significant degree of intra-segmental mobility

(see Fig. 2). The methyl substituents in the diaminemoiety and the biphenyl linkage in the dianhydridemoiety are likely to display short-range motion use-

7/31/2019 12354949

8/12


ful for the diffusion of both oxygen and nitrogenmolecules. This reduces the ability of the molecu-lar matrix to selectively restrict the mobility of thepenetrants on the basis of size and shape.

Membranes pyrolyzed at 535C (PM535) had ex-tremely high oxygen permeabilities in the range of8001200 Barrer. The selectivities, however, did notchange very significantly from AP. There was at mosta 20% increase in the O2/N2 selectivities which wasnot enough to place the material in the commerciallyattractive rectangular region. The tremendous increasein permeabilities indicates a more open molecular ma-trix in the partially pyrolyzed material than in the poly-mer precursor. Sorption experiments were not carriedout on this material as it failed to show properties inthe commercially attractive region.

Further pyrolysis at 550C (PM550) subjected themembranes not only to a higher temperature but alsoto longer times of pyrolysis. The total cycle time for

PM550 was longer by 3h than in PM535 and re-sulted in a reduction in oxygen permeabilities and anincrease in permselectivities. The permeabilities arelower than for PM535 but are still significantly higherthan those of the original polymer precursor. More-over, the permselectivities are still 100150% higherthan the original polymer precursor. This suggestsa matrix, which is beginning to condense, yieldinggreater size and shape selectivity than in PM535, andleading to separation properties in the commerciallyattractive region well above the upper-bound curve.The order of magnitude increase in permeabilities is aresult of increased sorptivities, which are higher than

those for AP by about a factor of 4. The diffusioncoefficients for the particular membrane mentionedhere, PM550.1, were similar to those of the polymer;however, other membranes with higher permeabili-ties had diffusion coefficients which were higher bya factor of 34. The increase in permselectivity, onthe other hand, resulted from increases in mobility se-lectivity alone. The mobility selectivities for PM550were greater than those of AP by nearly a factor of2. The sorption coefficients increased for both oxygenand nitrogen with practically no change in the sorp-tion selectivities, going from the polymer precursor tothe carbonized membrane.

The finally pyrolyzed membranes at 800C(PM800) show a further increase in permselectivitiesaccompanied by a reduction in permeabilities. Oxygen

permeabilities in PM800 are lower than in PM550 andAP as well. However, PM800 still has a high oxygenpermeability, near 20 Barrer, with overall separationproperties lying well within the commercially attrac-tive rectangular region in Fig. 6. Of the membranescompared in Table 1, the PM800 membrane hadpermeabilities which were a factor of 10 lower thanthe permeability of the PM550 membrane, becauseof significantly lower diffusion coefficients, around20 times lower in PM800 compared to PM550. Thesorption coefficients for PM800 are actually higherthan those for PM550 by a factor of 1.8 and partlycompensate for the lower diffusion coefficients. ThePM800 permselectivities are 30% higher than thoseof PM550, which is a result of a further increase inmobility selectivity. It is interesting to note that thepermselectivity of PM800 is greater by a factor of3, or more, than that of AP. The increase in sorptioncoefficients points to a molecular matrix which has a

large free volume, or porous volume. The decrease indiffusivities and increase in mobility selectivities isconsistent with an accompanying tightening of peri-odic pore mouths believed to exist between the freevolume sorption sites, thereby allowing greater sizeand shape selectivity.

4.1. Temperature dependence and activation energies

The activation energies for permeation were mea-sured for the polymer, AP and the various carbonizedmembranes from temperature-dependent permeationdata. Temperature-dependent sorption was measured

for AP, PM550 and PM800 and effective heats ofsorption were calculated. Activation energies andpre-exponential factors for diffusion were calculatedfor these materials using the permeation activationenergies and the apparent heats of sorption [10]. Theactivation energies and the pre-exponential factorswere also obtained as a range due to variability inthe permeation data. In this case as well, values forPM535.1, PM550.1 and PM800.1 are reported inTables 3 and 4 in order to make specific comparisons.Permeabilities increased with temperature for all ma-terials, for both oxygen and nitrogen as indicated bythe positive activation energies. The activation ener-

gies for permeation increase in general as the level ofpyrolysis increases. Possible interpretations of thesetrends are considered below.

7/31/2019 12354949

9/12


Table 3Permeation activation energies and pre-exponential factors forpolymers and pyrolyzed materialsa

Property AP PM535 PM550 PM800

EPO2(kcal/mol) 0.8 0.7 4.1 5.2

EPN2(kcal/mol) 1.8 3.1 6.1 6.6

EPO2,N2

(kcal/mol)

1.02 2.4 2.0 1.4

P0O2(Barrer) 2.42 E+2 2.8 E+3 1.65 E+5 1.14 E+5

P0N2(Barrer) 3.11 E+2 3.19 E+4 4.09 E+5 8.66 E+5

PO2/PN2 4.1 4.6 9.9 12.3

a Data for oxygen at 20psi, nitrogen at 75 psi for AP, PM550.1and PM800.1 membranes; mixed gas data for PM535.1 at 30 psitotal pressure; all values have been calculated form curve fit equa-tion obtained from raw data on each membrane.

4.2. AP to PM535 changes

The large increase in permeabilities when go-ing from the polymer precursor to PM535, as seen

in Fig. 5, reflects the much higher pre-exponentialfactors in PM535. There was no change in the ac-tivation energies for permeation of oxygen as themembrane evolved from the polymer precursor APto PM535. A small decrease in EPO2 was observedin some PM535 membranes tested. No such variationwas observed for EPN2 , which increases when go-ing from AP to PM535. This indicates an increasingresistance to nitrogen transport, while oxygen trans-port remains practically unaffected. Despite the smallchange in permselectivities there was a large changein EPO2,N2 . The higher EPO2,N2 values for PM535indicate a greater temperature dependence of overall

permselectivity, i.e. the loss in permselectivity withtemperature will be greater for PM535 than for AP.

Table 4Diffusion activation energies and pre-exponential factors for vari-ous materials from pure gas experimentsa

Property AP PM550 PM800

EDO2(kcal/mol) 4.09 6.83 8.01

EDN2(kcal/mol) 4.69 8.56 8.88

EDO2 ,N2(kcal/mol) 0.6 1.74 0.87

D0O2(cm2/s) 2.8 E-4 2.0 E-2 8.6 E-3

D0N2(cm2/s) 2.6E-4 4.6 E-2 4.2 E-3

DO2/DN2 2.9 7.3 8.5

a Data for oxygen at 20psi, nitrogen at 75 psi for AP, PM550.1and PM800.1 membranes; all values have been calculated formcurve fit equation obtained from raw data on each membrane.

4.3. PM535 to PM550 changes

A very large increase in EPO2 takes place as themembranes are pyrolyzed between 535 and 550C,which is accompanied with a substantial change inEPN2

as well. The EPO2,N2 is lower than that forPM535 but significantly higher than that for AP. Thereis a further increase in the pre-exponential factors inPM550 as compared to PM535. This increase some-what compensates for the higher activation energiesin this material and results in only a small loss inpermeabilities for PM550.

4.4. PM550 to PM800 changes

Going to PM800, a further increase in both EPO2and EPN2 is observed; however, the increase in ac-tivation energies for oxygen is twice as much asthe increase for nitrogen. The EPO2,N2 values for

PM800 are lower than in PM550. No further increasein pre-exponential factors is observed in this caseand hence, the higher activation energies lead to thesignificantly lower permeabilities of this material.

4.5. Overall changes in activation energies for

diffusion

Activation energies for permeation are a sum of theactivation energies of diffusion and the effective heatsof sorption for a penetrant in a material. Comparisonof activation energies for diffusion, rather than those

for permeation is more direct for understanding diffu-sional effects. In this regard, subtracting the heats ofsorption from the activation energies for permeationyields the activation energy for diffusion. The valuescompared in Table 4 are for PM550.1 and PM800.1membranes. The activation energies for diffusion in-creased for both oxygen and nitrogen as the level ofpyrolysis increased. The higher activation energies forboth oxygen and nitrogen for PM550, compared to AP,indicate increased resistance to diffusion in the molec-ular matrix. The larger EDO2,N2 for PM550 also in-dicates a greater temperature dependence of mobilityselectivity. The increase in EDN2 was larger than the

increase in EDO2 for PM550, implying that nitrogendiffusion faces more resistance than oxygen diffusion,which is reflected in the higher mobility selectivity

7/31/2019 12354949

10/12


for PM550 than for AP. The mobility of the polymerchains may have become severely hindered after par-tial pyrolysis in PM550, thereby restricting the motionof the larger nitrogen molecule to a greater extent. Ingoing from AP to PM550, O2 diffusion coefficientsdo not decrease despite the large increase in activationenergies because the two-orders of magnitude increasein pre-exponential factors also occur.

From Eq. (4), the large values of D0 could resultfrom increased activation entropy of diffusion for thegas molecules or from an increase in the jump lengthsof oxygen and nitrogen, or a combination of the twoeffects. Increase in activation entropy of diffusion fora gas molecule would mean that the entropy of thepenetrant-matrix system is larger in the activated statethan in the normal sorbed state, or the penetrant gainsentropy when making a diffusive jump. Increase in

jump lengths would be possible by the formation oflarge open regions in the molecular matrix through

which the penetrant would make large diffusive jumps.It is difficult to imagine an order of magnitude increasein jump lengths, considering the increase in mobilityselectivity. Hence, the larger pre-exponential factorsfor diffusion in PM550 are most likely to be a result ofthe increase in activation entropy with some contribu-tions from increase in jump lengths. For a single gas itwill be difficult to separate the contributions from thetwo effects by gas transport measurements alone. For-tunately, by considering differences, EDO2,N2 , someinsights may be possible, since diffusion jump lengthissues should be essentially eliminated for the simi-larly sized oxygen and nitrogen molecules.

In PM800, EDO2 increases further by about 17%while there is only a 4% increase in EDN2 as comparedto PM550. The EDO2,N2 values are lower for PM800than for PM550 indicating a lower temperature depen-

Table 5Energetic and entropic contributions to overall diffusivity selectivity

Property AP PM550 PM800 Commercial CMS pellets

DO2a 35.6 30150 1.52.0 0.07c

DO2/DN2 2.9 5.98.2 7.99.0 2545EDO2

(kcal/mol) 4.09 5.66.8 7.59.6 5.5b

Energetic contribution exp[EDO2,N2 /(RT)] 2.72.9 1730 4.17.0 5.1Entropic contribution exp[SDO2,N2 /(R)] 1.01.1 0.20.4 1.22.1 4.98.8

a 108 cm2/s.b [29].c Based on 4m Bergbau CMS pellets with diffusivity selectivity of 30.

dence of diffusivity selectivity in PM800. Comparedto AP, there is nearly a similar increase in EDO2 andEDN2 values for PM800. The diffusion coefficients forPM800 were much lower than in PM550, indicatingthat at the higher temperature of pyrolysis the mate-rial has developed a greater resistance to diffusion forboth oxygen and nitrogen. The pre-exponential factorsare somewhat lower in PM800 as compared to that inPM550, but there is some variation in these numbers.

4.6. Energetic and entropic contributions to

diffusivity selectivity

The energetic and entropic contributions to overallmobility selectivity were calculated from the activa-tion energies and the pre-exponential factors for diffu-sion using Eq. (6). These are presented in Table 5. Theresults are compared with properties for a commercialCMS pellet used in pressure swing adsorption based

fixed bed systems for air separation. The polymer APhas low mobility selectivities with energetic selectivi-ties providing nearly all of the contribution to overallmobility selectivity. The polymer, with its rigid back-bone and hindered inter-segmental mobility, may stillhave sufficient intra-segmental motion to prevent se-lective restriction of rotational motion of nitrogen inthe activated state, thereby limiting entropic selectiv-ities to only around unity.

The structure evolution between the polymer andPM550 leads to a remarkable increase in the mobil-ity selectivity and oxygen diffusion coefficients. Ac-companying these changes are increases in oxygen

activation energies for diffusion and increased ener-getic selectivity contributions. However, in this case,the entropic selectivities are surprisingly lower thanthose in AP, and the energetic selectivity, which is

7/31/2019 12354949

11/12


largely responsible for the much higher mobility se-lectivities in PM550. The higher oxygen and nitro-gen pre-exponential factors for PM550 compared toAP suggest a gain in entropy in the activated state asdiscussed in the previous section. Relatively flexiblewell-packed glassy polymers have been found to havesimilarly low entropic selectivities [24]. The molecu-lar matrix of PM550 is in transition from flexible poly-mer chains to a rigid molecular sieving environment.We speculate that in the normal sorbed state nitro-gen, which is slightly larger than the oxygen molecule,may be more effectively restrained from rotation inthe tightly packed regions of the hindered mobilitystructure. To make a diffusive jump still requires theopening of a transient gap in the tightly packed re-gion, thereby allowing nitrogen to gain entropy in themore open transition state environment while mak-ing a diffusive jump. The longer nitrogen molecule,therefore, may gain more entropy as it achieves the

ability to rotate freely in the diffusion transition state.On the other hand, if oxygen can already rotate freelyin the sorbed state, it will gain relatively less entropyin the equivalent activated state. Such a hypotheticalsituation would make the PM550 molecular structureentropically selective towards nitrogen, i.e., giving anO2/N2 entropic selectivity less than unity, as is ob-served. The increased activation energies are also in-dicative of a more restricted motion of the molecularmatrix. In a rather well-packed matrix a diffusion stepwould require a greater deformation to allow the dif-fusion of the larger nitrogen molecule again consistentwith the data.

In PM800, there is a further increase in mobilityselectivity accompanied by a large decrease in diffu-sion coefficients. The diffusion activation energies foroxygen are much higher than in PM550, but the dif-ference in activation energies for oxygen and nitrogenis less than it was for PM550. This material has signif-icant contributions from both energetic and entropicselectivities, which are responsible for its increasedmobility selectivities. The energetic contributions forPM800 decrease relative to PM550, but are greaterthan those for the polymer precursor. It is most likelythat the material finally evolves to a rigid molecularmatrix like those of other molecular sieving materials.

As the matrix becomes rigid, the activation energieswould be required only to overcome the repulsionsfrom the walls of the pore, and are likely to be similar

for the non-polar oxygen and nitrogen molecules withsimilar kinetic diameters.

Comparing these data with that for a commercialCMS sample, the energetic contributions for PM800are similar to those for the CMS, however, the entropicselectivity has further room for improvements. It is in-teresting to note, that commercial CMS materials typ-ically involve a post-pyrolysis deposition step whichreduces the pore size distribution in these materialsand increases their mobility selectivities [2528]. Sim-ilar post-pyrolysis steps may allow further increasesin the contributions from entropic factors.

5. Summary and conclusions

Pyrolysis of polymer films yield defect-free car-bonized membranes with oxygennitrogen separationproperties in the regime of commercial interest. The

permselectivities of the material increase as the levelof pyrolysis increases in terms of time and temperatureof pyrolysis. The finally pyrolyzed material(PM800)had selectivities three times higher than the polymerprecursor(AP). Permeabilities show a large increaseinitially with subsequent decrease as the level of py-rolysis increases. The permeability of the partially py-rolyzed membranes(PM550) was 410 times greaterthan that of the polymer precursor; however, perme-ability of the finally pyrolyzed material reduces to athird of the polymer permeability.

The increase in permselectivity was a result of theincrease in mobility selectivity alone for both PM550

and PM800. The thermodynamic sorption selectivityof the materials did not change measurably betweenthe polymer precursor and the carbonized membranes.The increase in permeabilities for PM550 was a re-sult of increase in absolute sorption coefficients, whichcontinue to increase between PM550 and PM800 aswell. The lower permeabilities of PM800 are a resultof significantly lower diffusion coefficients. The pre-ceding changes suggest a material where the total freevolume, or porous volume, increases with a simulta-neous restriction of the access path (pore mouths) tothese free volume regions.

Activation energies for diffusion increased in the

materials with increasing pyrolysis, indicating an in-creasing resistance to diffusion. The PM550 materialhad high energetic selectivities but no contributions

7/31/2019 12354949

12/12


from entropic selectivity. Entropic selectivity favorsnitrogen diffusion over oxygen in PM550, as in thecase of hindered mobility polymers. This may indicatea partially carbonized structure for PM550 with someseverely hindered residual mobility of the chains. Al-though entropic selectivity appears to develop duringstructural evolution from PM550 to PM800 desirableimprovements in entropic selectivity still appear pos-sible based on comparisons with molecular sievingadsorbents.

Acknowledgements

The authors gratefully acknowledge the support ofthe Texas Advanced Research Program and Medal.The authors would also like to thank Dr. Richard Uber-sax at E.I. duPont de Nemours for the SEM and Mr.Ranjan Ghosal at the University of New Mexico forthe pore volume analysis on the carbon samples.

References

[1] A. Singh, W.J. Koros, Significance of entropic selectivity foradvanced gas separation membranes, Ind. Eng. Chem. Res.35 (1996) 1231.

[2] C.W. Jones, W.J. Koros, Carbon molecular sieve gasseparation membranes. I. Preparation and characterizationbased on polyimide precursors, Carbon 32 (1994) 1419.

[3] J.E. Koresh, A. Soffer, Molecular sieve carbon permselectivemembrane. Part 1. Presentation of a new device for gasmixture separation, Sep. Sci. Technol. 18 (1983) 723.

[4] W. Shusen, Z. Meiyun, W. Zhizhong, Asymmetric molecularsieve carbon membranes, J. Membr. Sci. 109 (1996) 267.

[5] H. Hatori, et al., Carbon Molecular Sieve Films fromPolyimide, Carbon 30 (1992) 719.

[6] H. Suda, K. Haraya, Molecular sieving effect of carbonizedKapton polyimide membrane, J. Chem. Soc., Chem.Commun. (1995) 1179.

[7] J. Hayashi et al., Pore size control of carbonizedBPDA-ppODA polyimide membrane by chemical vapordeposition of carbon, J. Membr. Sci. 124 (1997) 243.

[8] W.J. Koros, R.T. Chern, Separation of gaseous mixtures usingpolymer membranes, in: R.W. Rousseau (Ed.), Handbookof Separation Process Technology, Wiley/Interscience, NewYork, 1987.

[9] V. Stannett, Simple gases, in: J. Crank et al. (Ed.), Diffusionin Polymers, Academic Press, New York, 1968.

[10] L.M. Costello, W.J. Koros, Temperature dependence of gas

sorption and transport properties in polymers: measurementand applications, Ind. Eng. Chem. Res. 31 (1992) 2708.

[11] K.C. OBrien et al., A new technique for the measurementof multicomponent gas transport through polymeric films, J.Membr. Sci. 29 (1986) 229.

[12] C.W. Jones, W.J. Koros, Carbon molecular sieve gasseparation membranes. II. Regeneration following organicexposure, Carbon 32 (1994) 1427.

[13] J. Hilsenrath, Tables of Thermal Properties of Gases, USDepartment of Commerce, National Bureau of Standards,

Washington, DC, 1955.[14] S.J. Gregg, K.S.W. Sing, Adsorption, Surface Area and

Porosity, 2nd Edition, Academic Press, New York, 1991.[15] H. Marsh, W.F.K.W. Jones, The surface properties of carbon.

I. The effect of activated diffusion in the determination ofsurface area, Carbon 1 (1964) 269.

[16] M.M. Dubinin, Physical adsorption of gases and vapors inmicropores, in: D.A. Cadenhead et al. (Ed.), Progress inSurface and Membrane Science, Academic Press, New York,1975.

[17] R. Ghosal, D.M. Smith, Micropore characterization using theDubininAstakhov equation to analyze high pressure carbondioxide (273 K) adsorption data, J. Porous Mater. 3 (1996)247.

[18] J.F. Byrne, H. Marsh, Introductory overview, in: J.W. Patrick

(Ed.), Porosity in Carbons: Characterization and Applications,Edward Arnold, London, 1995.[19] H.K. Chagger et al., Kinetics of adsorption and diffusional

characteristics of carbon molecular sieves, Carbon 33 (1995)1405.

[20] R. Srinivasan, S.R. Auvil, J.M. Schork, Mass transfer incarbon molecular sieves an interpretation of Langmuirkinetics, Chem. Eng. J. 57 (1995) 137.

[21] D.M. Ruthven, Diffusion of oxygen and nitrogen in carbonmolecular sieves, Chem. Eng. Sci. 47 (1992) 4305.

[22] C.M. Zimmerman, A. Singh, W.J. Koros, Diffusion in gasseparation membrane materials: a comparison and analysisof experimental characterization techniques, J. Polym. Sci.:Polym. Phys. 36 (1998) 1747.

[23] A. Singh, Membrane materials with enhanced selectivity: anentropic interpretation, Ph.D. dissertation, University of Texasat Austin, Austin, TX, 1997.

[24] A. Singh-Ghosal, W.J. Koros, Energetic and entropic contri-butions to mobility selectivity in glassy polymers for gasseparation membranes, Ind. Eng. Chem. Res. 38 (1999) 3647.

[25] T.A. Braymer et al., Granular carbon molecular sieves, Carbon32 (1994) 445.

[26] A.L. Cabrera et al., Preparation of carbon molecular sieves. 1.Two-step hydrocarbon deposition with a single hydrocarbon,Carbon 31 (1993) 969.

[27] K. Chihara, M. Suzuki, Control of micropore diffusivitiesof molecular sieving carbon by deposition of hydrocarbons,Carbon 17 (1979) 339.

[28] S.K. Verma, P.L. Walker Jr., Preparation of carbon molecularsieves by propylene pyrolysis over microporous carbons,Carbon 30 (1992) 829.

[29] J. Karger, D.M. Ruthven, Diffusion in Zeolites and OtherMicroporous Solids, Wiley/Interscience, New York, 1992.

12354949

Documents