Membranes in ChE Education

A Simple Analysis For

GAS SEPARATION MEMBRANE

EXPERIMENTS
RICHARD A. DAVIS, ORVILLE C. SANDALL*
University of Minnesota Duluth Duluth, MN 55812

embrane applications for gas separations have made rapid advances over the past decde.[1] In some cases, membrane technologies have been used to enhance or replace more traditional
methods of gas purification. The need for educating undergraduate chemical engineering students about membrane-based separations has not gone
unnoticed. Newer editions of popular separations textbooks have added
chapters on membranes with sections on gas permeation.[2-4]

Earlier, Davis and Sandall[5] described an undergraduate laboratory membrane experiment and analysis for separating the components of air. It
remains relevant today as one approach to providing students with handson experience with this important technology. The experimental objectives included an inverse mass transfer analysis of experimental data for
key membrane transport parameters. The original analysis involved solving a set of differential species balances and fitting the results to experimental data by iterative, trial-and-error techniques. They found that the
numerical methods required to implement their analysis were beyond the
scope of the undergraduate chemical engineering laboratory experience.
Consequently, they provided students with True BASIC programs that
were used to solve the model equations. Unfortunately, the programs were
limited to the specific membrane configuration in the laboratory. Students were unable to explore alternative designs using the validated models without modifying the programs. In the meantime, several popular,
modern, computational software applications (such as Excel, Mathcad,
Matlab, or Polymath) have emerged that provide readily accessible tools
for solving complex problems that involve nonlinear algebraic and differential equations. The drawbacks in the original analysis, along with
developments in computational tools, have led to a simpler alternative
analysis described in this paper.

EXPERIMENT
Davis and Sandall[5] provided specific details of the experimental objectives, apparatus, and procedure for a commercial hollow-fiber membrane unit for air separation. The Prism separator developed by Permea
* University of California, Santa Barbara, CA 93106

Figure 1. Prism hollow-fiber membrane

apparatus.
Richard A. Davis is Associate Professor in the Department of
Chemical Engineering at the University of Minnesota Duluth.
He earned his BS in Chemical Engineering from Brigham Young
University and his PhD from the University of California, Santa
Barbara. He teaches a variety of courses in transport phenomena and separations, and his current research interests include
process modeling and optimization.
Orville C. Sandall is Professor of Chemical Engineering at
the University of California, Santa Barbara. He is a graduate of
the University of Alberta (BSc and MSc) and the University of
California, Berkeley (PhD). His teaching and research interests are in the areas of mass transfer and separation processes.

Copyright ChE Division of ASEE 2003

74

Chemical Engineering Education

Membranes in ChE Education

Corporation, shown in Figure 1, consists of four hollow-fiber membrane modules arranged in a series of columns. Each
module is a shell-and-tube arrangement of a bundle of hollow-fiber membranes that are capped at the top. High-pressure feed air is introduced to the shell side of the fibers. The
permeating gas flows through the hollow-fiber bores and is
collected in a manifold at the open end. The pressure drop
across the shell side of the membrane unit was found to be
negligible.[5] The permeate streams are open to the atmosphere.
The pressure at the closed end of the fiber bores is not directly measurable in the current module arrangements. Information about fiber length, fiber inside diameter, and the number of fibers in the Prism separator bundle is not available,
but a conservative estimate of the pressure build-up in the
fiber bore was calculated to increase by less than nine percent above atmospheric pressure for the range of experimental operating conditions. For most of the experiments, the
pressure build-up was estimated to be less than three percent.

Modern gas-separation membrane modules introduce the

high-pressure feed to the bore side of the fibers to eliminate
channeling and maintain a more uniform flow distribution.
High-pressure feed to the fiber bores can result in a significant axial pressure drop in the fibers. Although not required
for this membrane module, the effects of pressure are included
in the analysis for completeness.
As shown in the schematic of Figure 2, the air-flow pattern
consists of alternating countercurrent and cocurrent flow
through the columns. The composition of the retentate and
permeate streams was measured with oxygen analyzers. The
flow rate of the retentate stream was measured with a volumetric flow meter. The feed and permeate flow rates may be
calculated by mass balances.
The membrane separator may be operated as four columns
in series, or as a single column by closing a valve on the tube
connecting the retentate and feed streams between the first
two columns. The first column operates in countercurrent flow
and was used to calibrate the membrane models from a series
of runs performed at various feed-flow rates and pressures.
The calibrated model was confirmed by favorable comparisons of model predictions with experimental results from the
four-column configuration.

THEORY AND ANALYSIS

A differential model of binary gas separation in the membrane experiment was validated by Davis and Sandall and is
summarized next. For the conditions of the experiment, it
can be shown that a simplification to the equations permits
an algebraic solution.
Figure 2. Schematic of single, countercurrent flow
column or four columns with alternating flow patterns.

Figure 3. Ideal cocurrent flow pattern

The mathematical model of membrane gas separation was

based on several key assumptions. First, the temperature was
assumed to be constant. Further, it was assumed that all
streams through the shell and permeate sides of the fibers
were in plug flow. The air fed to the unit was assumed to be a
binary mixture of 79% N2 and 21% O2. All four columns were
assumed to have the same dimensions and specific area for
mass transfer. Finally, axial pressure drop was ignored for
the fiber bore. This assumption is valid for low permeate flow
or large transmembrane pressure differences where small
changes in permeate pressure are negligible relative to the
high feed pressure.

 Differential Model

Figure 4. Ideal countercurrent flow pattern.

Winter 2003

Walawender and Stern[6] derived the differential equations

for a binary gas system in countercurrent and cocurrent plug
flow patterns, shown ideally in Figures 3 and 4. Details of
the derivation are available in several references.[3,5,6] For a
binary gas system, the total mole and O2 species balances
around the separator are
75

Membranes in ChE Education

(1)
(2 )

nF = nR + np
xFnF = xRnR + ypnp

where nF, nR, and np are the molar flow rates of the feed,
retentate, and permeate streams, respectively, and xF, xR, and
xp are the feed, retentate, and permeate O2 mole fractions,
respectively. The species balances around a differential volume element in the membrane give
d( xn ) = Q'O 2 ( xP yp)dA

(3)

d[(1 x)n ] = Q'N 2 (1 x)P (1 y)p dA

x = xR

y = y i at A * = 0
n * = 1

Note the discontinuity in Eq. (6) at x = xR requires

application of lHpitals rule.[6]
The dimensionless cocurrent flow model equations are
KF

xy *
dx
=
(1 x)( xr y) x (1 x)r (1 y)
dA * y x F

(14)
xy *
dy
=
(1 y)( xr y) y (1 x)r (1 y)
*
x xF
dA

where
is the permeance of species j, A is the membrane
surface area, and P and p are the average retentate and permeate side pressures, respectively.

KF

For convenience in the analysis, Eqs. (1) to (4) were combined into the following dimensionless equations for countercurrent flow:

where

dx x y *
=
(1 x)( xr y) x (1 x)r (1 y)
dA * x R y i

]}

(5)

]}
(6 )

KR

dn *
= * ( xr y) + (1 x)r (1 y)
dA *

( 7)

where yi is the mole fraction in the permeate at the closed

end of the fibers. The dimensionless transport parameters are
defined as
A* = A / A m

(8)

r=P/p

( 9)

K R = n R / Q'N 2 A m p

(10)

* = Q'O 2 / Q'N 2

(11)

n = n / nR

(12)

where Am is the total membrane area. The ideal separation factor, * , was assumed constant, but the dimensionless
transport parameter, KR, was defined as a function of the
retentate molar flow rate. The solution to Eq. (7) was used to
check the assumptions leading to the algebraic model of
the next section. The countercurrent flow equations are
integrated from the retentate end of the membrane, subject to the initial conditions
76

]}

(15)

KF =

nF
'
QN2 AMp

(16)

The cocurrent model equations are integrated from the feed

end, subject to the initial conditions
x = xF
*
at A = 0
y = yi

dy x y *
KR
=
(1 y)( xr y) y (1 x)r (1 y)
dA * x R x

]}

( 4)

Q'j

KR

(13)

(17)

The permeate composition at the capped end of the hollow

fibers is calculated from the ratio of Eqs. (3) and (4)
* [ xr y i ]
yi
=
1 yi
(1 x)r (1 y i )

(18)

where, for countercurrent flow, x = xR. For cocurrent flow, x

= xF. Equation (18) is quadratic in yi. Note that there is an
error in the denominators of Eqs. (17) and (22) of the paper
by Davis and Sandall.[5] The correct solution to the quadratic
equation is

(* 1)(xr +1) + r [(* 1)(xr +1) + r] 4(* 1)*xr

yi =
2( * 1)
2

(19)
Davis and Sandall successfully used the differential model
in their analysis of O2/N2 separation in the membrane module. At the time, they found that the background required to
solve the model equations for * and KR was beyond the
scope of an undergraduate student in their laboratory course.
Consequently, they developed True BASIC programs that
were provided to the students to solve the model equations.
Since then, advances in computational software (such as
Mathcad) have simplified the process of solving the model
Chemical Engineering Education

Membranes in ChE Education

equations. Undergraduate students are now able to develop
their own solutions using standard numerical methods for
solving systems of nonlinear equations or differential equations that are readily available in these computer tools.
Nevertheless, students are still required to set up a standard method such as Eulers or Runge-Kutta for the initialvalue problems in order to find the values for * and KR by
inverse analysis of the first column in countercurrent flow.
For example, Mathcad and Polymath do not permit their intrinsic capabilities for solving systems of first-order differential equations to be treated as part of another function. An
example of programming required in Mathcad for the inverse
mass transfer is shown in Figure 5. This type of solution may
be intimidating for undergraduate students, depending on their
level of experience. This realization, along with the observation that the composition profiles along the membrane were
approximately linear, led to the following alternative analysis that avoids the initial-value problem solution requirements
entirely.

 Algebraic Model
Boucif, et al.,[7] presented a series solution to the binary
component differential model Eqs. (5), (6), (14), and (15)
that requires a numerical solution to a pair of third-order polynomial equations. The solution to the series equations agrees
with numerical solutions to the differential model when the
cut is less than 50%. The series solution does not include
axial pressure effects in the feed or permeate gas, however.
Hundyil and Koros[8] presented a more complete analysis of
hollow-fiber membrane modules for multicomponent gas
separation that includes pressure effects. Their approach is
based on a finite-volume element model that requires iterative solutions to a large system of nonlinear algebraic equations. The finite-element approach is recommended when detailed information of pressure, temperature, and composition
effects is required.
A simpler, alternative analysis of the membrane unit described
here was developed that involves only the solution to a small
system of nonlinear algebraic equations and includes pressure effects when necessary. The simpler-model equations
are analogous to the shell-and-tube heat-exchanger design
equations that are familiar to undergradute chemical engineering students. The following analysis assumes laminar
flow and constant species permeances that are independent
of the pressure and composition of the feed or permeate gas.
The Hagen-Poisseuille equation is commonly used to calculate axial pressure effects[9]
dp 128 RT n
=
dz
pd f4 N f

(20)

where R is the ideal gas constant, T is the gas temperature,

is the gas viscosity, n is the variable molar flow rate of permeate gas, df is the inside fiber bore diameter, and Nf is the
number of fibers in a bundle. Other expressions derived from
the Hagen-Poisseuille equation have been developed to account for compressibility and flow in porous channels when
necessary.[8,10-12]
It has been observed that when the change in the feed mole
fraction of oxygen is less than 50%, the differential balances
may be replaced with algebraic expressions involving the
logarithmic mean of the transmembrane partial-pressure difference.[13] In Eq. (3), let
= xP yp

(21)

The driving force for diffusion across the membrane, , is

assumed to be a linear function of the change in the molar
flow on the feed side of the membrane

Figure 5. Example of Mathcad programming for inverse

mass transfer analysis for * and KR.
Winter 2003

d( xn ) ( xn ) R ( xn ) F
=
d
R F

(22)

Combine eqs. (2), (3), and (22), separate variables and inte77

Membranes in ChE Education

grate
R

ypnp

pc = p2 +

d
= Q'O 2 ( R F ) dA

(23)

or
ypnp =

Q'O 2

(xP yp)lm A m

(24)

where the log-mean difference in O2 partial pressure across

the membrane is defined as

(xP yp)lm

(xP yp) R (xP yp) F

ln[( xP yp) R / ( xP yp) F ]

(25)

A similar result is found for a N2 flux expression

(1 y p )n p = Q'N [(1 x)P (1 y)p]lm A m

2

(26)

The steady-state binary-gas membrane equations can be written in dimensionless form using the average pressures
x F = x R (1 ) + y p

(27)

y p K R = (1 ) * ( xr y) lm

(28)

(1 y p )K R = (1 )[(1 x)r (1 y)]lm

(29)

where the cut is defined here as the ratio of permeate-to-feed

flow rates

(30)

= np / nf

Alternative forms of Eqs. (28) and (29) in terms of KF are

y p K F = * ( xr y) lm

(31)

(1 y p )K F = [(1 x)r (1 y)]lm

(32)

The permeate composition at the closed end of the hollowfiber membranes is calculated from Eq. (19).
The experimental separation factor was calculated from the
measured compositions of the permeate and retentate streams
=

y p (1 x R )

xR 1 yp

(33)

Under conditions where the change in the feed composition exceeds 50%, the log-mean model can be applied two or
more times as necessary across a module such that each cut
does not exceed a 50% change in xF from the previous step.
The pressure at the closed end of the fiber bore can be calculated by assuming that the permeate flow rate is a linear function of distance along the fiber
n=

n pz
L

(34)

where L is the fiber length. Equation (20) can be integrated

with substitution from Eq. (34) to give an estimate for the
permeate pressure at the closed end of the fibers[9]
78

128 RTLn p

(35)

d f4 N f

 Solution Method
The algebraic model Eqs. (19) and (27-29) represent a system with four degrees of freedom, or four equations in eight
variables: xF, xR, yp, yi, , * , KR, and r. The model was initially calibrated by fixing xF and r and measuring xR and yp,
leaving yi, , * , and KR as unknowns in the solution.
The solution of the system of nonlinear algebraic equations requires an iterative, trial-and-error technique, such as
Newtons method. The log-mean approximation of the partial-pressure driving force is notoriously difficult to converge
under these circumstances. Fortunately, there are good approximations to the log-mean that avoid problems of divergence in the solution. The following form of the Chen approximation was used:[14]

( 1 + 2 )
2 1
1 2

ln( 2 / 1 )
2

1/ 3

(36)

Floudas noted that the Chen approximation to the log-mean

has the advantage that it becomes zero if either the feed or
exit partial-pressure driving forces become zero.[15]
The four-column configuration requires sequential solution
to the countercurrent and cocurrent models. Note that n2F =
n1R and K2F = K1R between the first and second columns, and
that n4F = n3R and K4F = K3R between the third and fourth columns. The feed flow rates to each column are calculated from
the cut for the previous column.

RESULTS AND DISCUSSION

The experimental data of Davis and Sandall[5] were used to
illustrate the analysis procedure. The assumption of Eq. (22)
for the log-mean approximation was evaluated by plotting a
representative numerical solution to Eq. (7), shown in Figure
6. A linear least-squares regression of the numerical results
Figure 6.
Numerical
results to
support the
logarithmicmean
assumptions for
xF=0.21,
xR=0.16,
yp=0.48
r=6.47
* =5.9
and KR=4900.
Chemical Engineering Education

Membranes in ChE Education

shows that the assumption of a linear function for is valid
for the conditions of this laboratory experiment.
A sample calculation of the single countercurrent flow
model calibration using Mathcad is shown in Figure 7. The
experimental data and results of the algebraic model are compared with the results from the differential model in Table 1
for * and KR. There are no significant differences in the
results between these models.

A linear relationship between the retentate flow rate and

KR is calculated for use in the remaining three column predictions. The linear function is plotted with the results in Figure 5. The result of a linear least-squares regression gives
K R = 4.0 10 3 n R

(37)

The average value of was calculated to be 5.9 assuming

atmospheric pressure in the fiber bore. An increase in fiberbore pressure would cause the experimentally determined
species permeances to decrease. The axial pressure drop has
been found to vary linearly with flow rate, however.[16] Thus,
the slope in Eq. (37) is not affected by the small pressure
build-up in the permeate stream.
Separation factors for the four-column configuration were
predicted from the sequential calculations of the model for a
range of feed pressures and flow rates. The results plotted in
Figure 9 show good agreement with the experimental values
calculated from the data of Davis and Sandall.
All of these results lend confidence in the algebraic model.
Students are able to quickly design alternative configurations
and explore the potential performance of competing designs.

Figure 8.

Figure 7. Example of Mathcad calculation for inverse

mass transfer analysis using the log-mean model.

TABLE 1
Calibration Data[5] and Results for Single Countercurrent Column
Experimental Data
nRx102
P(kPa) (gmol/s) xR
yp

Differential Model

Algebraic Model

KR

KR

377
377
377
377
377

0.73
0.74
1.03
1.32
2.54

0.18
0.18
0.19
0.19
0.20

0.43
0.43
0.44
0.44
0.44

5.81
5.81
5.98
5.98
5.71

31.1
31.1
49.6
49.6
98.7

5.82
5.82
5.98
5.98
5.71

31.1
31.1
49.6
49.6
98.6

515
515
515
515
515

0.62
0.73
0.95
1.51
2.25

0.15
0.16
0.17
0.18
0.19

0.45
0.46
0.47
0.47
0.48

5.93
6.02
6.12
5.85
5.96

26.2
33.3
43.9
58.2
92.1

5.97
6.05
6.14
5.86
5.96

26.1
33.2
43.9
58.2
92.1

653
653
653
653
653

0.74
0.95
1.32
2.18
3.44

0.14
0.15
0.16
0.18
0.19

0.46
0.47
0.48
0.49
0.5

5.78
5.84
5.90
5.73
5.81

31.5
38.8
49.1
85.7
135

Average
Winter 2003

5.88

Correlation of
KR
with nr
for a
single
countercurrent
flow
column

5.84 31.4
5.88 38.7
5.93 49.0
5.74 85.6
5.81 135
5.90

Figure 9.
Comparison
of
predictions
with
experimental
results for air
separation
in the fourcolumn
configuration:
=377kPa,
=515 kPa,
=653 kPa.
79

Membranes in ChE Education

For example, students usually start by comparing the performance of cocurrent and countercurrent flow. This leads to a
design for one column operating in countercurrent flow with
the same membrane surface area as the four columns. The
single column design gives a predicted increase of 10% N2
recovery when compared to the modular design. Students may
use the models to predict a dimensionless membrane area,
1/K, to recover a desired fraction of oxygen fed to the
permeator. Other designs include four columns operating in
parallel with countercurrent flow or four columns with the
feed side in series and the permeate side in parallel.
The Mathcad files used in the analysis are available at
<www.d.umn.edu/~rdavis/cee>

CONCLUSIONS
A membrane experiment for investigating gas separation
has been in use for over ten years in the undergraduate laboratory at the University of California, Santa Barbara. A simple
analysis method was presented that requires only the solution to a system of four algebraic equations. The simpler analysis is equally applicable to newer membrane configurations
that introduce the high-pressure feed to the fiber bores in order to maintain better flow patterns in the membrane module. The experimental apparatus was designed to permit
single- and four-column investigations of air separation. The
single column was used to calibrate the models for binary
gas separation. Comparing results for the four-column operation validated the calibrated model. Good model and experimental agreement lend confidence in the model and validate the model assumptions. Students are then able to use the
model to develop competing designs for gas separation and
optimize their designs for maximizing efficiency of separation. The advantages of the simpler approach are that students can readily set up and solve the model equations
without complicated programming. Students are also able
to explore alternative designs by building models and
comparing the results.

NOMENCLATURE
A
d
K
L
n
N
p
P
Q
R
T
x
y
80

membrane area, m2
diameter, m
dimensionless membrane transport parameter
fiber length, m
molar flow rate, gmol/s
number of fibers in a bundle
permeate side pressure, kPa
feed side pressure, kPa
permeance, gmol/(s-kPa-m2)
ideal gas constant, kPa-m3/gmol-K
temperature, K
feed stream mole fraction of oxygen
permeate stream mole fraction of oxygen

z
variable fiber length, m
Greek Symbols
experimental separation factor

transmembrane partial pressure, kPa

viscosity, N-m/s
cut of feed to permeate stream

Subscripts/Superscripts
c
closed end of fiber bore
e
experimental
f
fiber
F
feed
i
closed end of permeate stream
lm
log-mean result
m
membrane
N2
nitrogen
O2
oxygen
p
predicted
R
retentate
*
dimensionless or ideal parameter

REFERENCES
1. Pandey, P., and R.S. Chauhan, Membranes for Gas Separation, Prog.
Polym. Sci., 26(6), 853 (2001)
2. McCabe, W.L., J.C. Smith, and P. Harriott, Unit Operations of Chemical Engineering, 5th ed., McGraw-Hill, New York, NY (1993)
3. Geankoplis, C., Transport Processes and Unit Operations, 3rd ed.,
Prentice-Hall, Englewood Cliffs, NJ (1993)
4. Seader, J.D., and E.J. Henley, Separation Process Principles, John
Wiley and Sons, New York, NY (1998)
5. Davis, R.A., and O.C. Sandall, A Membrane Gas Separation Experiment for the Undergraduate Laboratory, Chem. Eng. Ed., 25(1), 10
(1991)
6. Walawender, W.P., and S.A. Stern, Analysis of Membrane Separation Parameters. II: Counter-Current and Cocurrent Flow in a Single
Permeation Stage, Sep. Sci., 7(5), 553 (1972)
7. Boucif, N., S. Majumdar, and K.K. Sirkar, Series Solutions for a Gas
Permeator with Countercurrent and Cocurrent Flow, Ind. Eng. Chem.
Fund., 23, 470 (1984)
8. Thundyil, M.J., and W.J. Koros, Mathematical Modeling of Gas Separation Permeators: For Radial Crossflow, Countercurrent, and
Cocurrent Hollow Fiber Membrane Modules, J. Mem. Sci., 125, 275
(1997)
9. Zolandz, R.R., and G.K. Fleming, Design of Gas Permeation Systems, in Membrane Handbook, W.S. Ho and K.K. Sirkar, eds., Van
Nostrand Reinhold, New York, NY, p. 66 (1992)
10. Bruining, W.J., A General Description of Flows and Pressures in
Hollow Fiber Membrane Modules, Chem Eng. Sci., 44(6), 1441 (1989)
11. Federspiel, W.J., J.L. Williams, and B.G. Hattler, Gas Flow Dynamics in Hollow-Fiber Membranes, AIChE J., 42(7), 2094 (1996)
12. Lim, S.P., X. Tan, and K. Li, Gas/Vapour Separation Using Membranes: Effect of Pressure Drop in Lumen of Hollow Fibers, Chem.
Eng. Sci., 55, 2641 (2000)
13. Mulder, M., Basic Principles of Membrane Technology, Kluwer Academic Publishers, Dordrecht, Netherlands (1991)
14. Chen. J.J.J., Comments on Improvements on a Replacement for the
Logarithmic Mean, Chem. Eng. Sci., 42(10), 2488 (1987)
15. Floudas, C.A., Nonlinear and Mixed-Integer Optimization, Oxford
University Press, Oxford (1995)
16. Vladisavljevic, G.T., and M.V. Mitrovic, Pressure Drops and Hydraulic Resistances in a Three-Phase Hollow Fiber Membrane Contactor
with Frame Elements, Chem. Eng. Proc., 40, 3 (2001)
Chemical Engineering Education