Department of inorganic technology

LABORATORY WORK

Hydrocarbon separation on micro-porous membranes

Single gas permeance measurement

Milan Bernauer, Vlastimil Fíla, Vladimír Navara, Bohumil Bernauer and Petr Šťastný
 Department of inorganic technology: Single Gas Permeace

Contents
1 Introduction 1
1.1 Basic terms in gas separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2 Theoretical part 2
2.1 Knudsen diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Viscous flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Slip flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.4 Surface diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Experimental 3

4 Treatment of data from single gas permeance measurement 4

4.1 Model of Knudsen diffusion and viscous (Poiseuille) flow . . . . . . . . . . . . . . . . . . . . . 6
4.2 Solution of Knudsen and Poiseuille model - obtaining parameters ΠK and ΠP . . . . . . . . . 7
4.3 Calculation procedure and some recommendations . . . . . . . . . . . . . . . . . . . . . . . . 8

A Example of Octave program for nonlinear fit 9

B Matlab program for nonlinear fit 10

1 Introduction
Measuring gas permeation in micro-porous materials is of a great importance for determining the separation
capabilities of membranes and also to evaluate the effect of mass transport phenomena on the chemical
transformations in a micro-porous catalyst pellet. Along with the transport properties the raw information
on structural properties of the porous material can be obtained.
The simple tool of unsteady permeation experiment is perfectly suited to get such information in a fast
and simple manner. Measurement consists in observing the pressure decrease in constant volume vessel
separated by the studied membrane from its surroundings (atmosphere). At the beginning of the experiment
the vessel is filled by a pure gas at relatively low pressure (≈500 kPa) and is allowed to permeate through
the membrane. The time evolution of the pressure inside the vessel is recorded by pressure gauge. From the
dependence of ∆p(t) one can evaluate the permeability (or permeance) and the mean pore diameter. If this
measurement is performed for two pure gases, the ideal separation factor can be calculated from permeances
of this two gases.

1.1 Basic terms in gas separation

Intensity of molar flow (molar flux) [N L−2 T−1 ] This value is defined as the molar flow per unit area
of the membrane
F
N= , (1)
A
where is A is area perpendicular to the flow direction, F is molar flow [N T−1 ] defined as
p
ṅ ≡ F = vA, (2)
RT
where v is mean stream velocity in [L T−1 ].

1
Last change: February 21, 2024 Milan Bernauer
 Department of inorganic technology: Single Gas Permeace

Permeance [N M−1 T L−1 ] This value is defined as the ratio of the intensity of the molar flow and the
pressure difference on both side of the membrane
N
Π= (3)
∆p
In conventional units the permeance is usually expressed in mol m−2 s−1 Pa−1 .

Permeability [N M−1 T L−2 ] is a value of permeance related to the thickness of the membrane
Π
β= (4)
δ

Ideal separation factor is a dimensionless value expressing the ratio of permeances (or permeabilities)
of pure gases
Πi βi
Si,j = = (5)
Πj βj

2 Theoretical part
Permeation of gases through micro-porous solid is a complex phenomena including interactions between the
pore wall and diffusing molecules as well as interactions between diffusing molecules themselves. Mechanisms
governing the transport of gases in micro-porous media depends of physical properties of the material,
experimental conditions and of course on permeating gas.
The micro-porous material can be characterized by the mean pore radius r, porosity ϵ and tortiuosity τ .
The ratio of ϵ/τ is often expressed as ψ.
Experimental conditions influencing the permeation process are mainly temperature and pressure, which
determine the mean free path of the molecule λ
RT
λ= √ , (6)
πNA dm p 2
where dm diameter of permeating molecule and NA
is Avogadro constant. The ratio
500
λ
Kn = , (7) 400
2r Kn 300
is called Knudsen number and its value give us the 200
information about the nature of the mass transport 100
mechanism inside the porous structure. Dependence 0
of Knudsen number on pressure and temperature 100
for N2 in ZSM-5 zeolite is shown in figure 1. For 200
300 600
p/kPa 500
numbers Kn > 1 the Knudsen diffusion is consid- 400 400 T /K
500 300
ered to be the dominant contribution to the total
flow. For Kn < 1 the combination of viscous and Figure 1: Knudsen number in ZSM-5 zeolite (2r =
−10
slip flow are dominantly responsible for the diffusion dp = 5.5 · 10 m) as as function of p and T for N2
−10
process. The surface diffusion occurs along all pre- (dm = 3 · 10 m).
viously transport mechanisms and has to be taken
into account at low temperatures for polar gases.
As evident from above paragraph total flow of the permeating gas is a linear combination of different
types of diffusion processes.

Last change: February 21, 2024 Milan Bernauer

 Department of inorganic technology: Single Gas Permeace

2.1 Knudsen diffusion

Knudsen diffusion take place in system where the pore diameter is comparable to the molecule mean free path
of diffusing gas and thus the collisions of type molecule-wall predominates over molecule-molecule collisions.
The Knudsen regime of flow occurs typically when Kn becomes larger than 10.The flow generated by Knudsen
diffusion can be written as r
2rϵ 8RT dp dp
NK = − = −DK , (8)
3RT πM dx dx
where ϵ is membrane porosity, r the pore radius, M the molar mass of gas and x is axial coordinate in
membrane
p layer. If only the Knudsen regime is assumed two gases will permeate through a membrane at
ratio Mi /Mj .

2.2 Viscous flow

Viscous (Poiseuille) flow occurs when the mean free path of molecules of diffusing gas is smaller than the pore
diameter and the momentum is shared predominately by more often happening molecule-molecule collisions.
Viscous flow is represented by following relationship
r2 ϵ dp dp
NP = − p = −DP p , (9)
8ηRT dx dx
where η is gas dynamic viscosity. As can be seen from equation 9 in the viscous flow regime we observe poor
separation abilities for gases having similar viscosity.

2.3 Slip flow

This kind of transitional transport mechanism is supposed to take place on the boundary between Knudsen
and viscous flow (Kn ≈ 0.1) and is described by following equation

π Kn−1 dp dp
NS = − −1

= −DS , (10)
4 1 + Kn dx dx

The nature of slip flow is a combination of bulk flow (viscous) and of flow of independently behaving
molecules.

2.4 Surface diffusion

Surface diffusion can occur in parallel with Knudsen diffusion. Gas molecules are adsorbed on the pore
walls of the membrane and migrate along the surface. Surface diffusion increases the permeability of the
components adsorbing more strongly to the membrane pores. At the same time, the effective pore diameter
is reduced. Consequently, transport of non adsorbing components is reduced and selectivity is increased.

3 Experimental
The permeation measurement apparatus is shown in figure 2. The single gas permeation measurements will
be carried out at room temperature. The apparatus consist of three different pressure vessels, and the active
volume is set using valves V1, V2 and V3. The three pressure detectors measure the pressure in the interior
of the apparatus (p1 ), at the outlet side of the apparatus (p2 ) and the difference of pressures between p1 and
p1 (∆p). During the measurement the gas passes through the membrane and the decrease of the pressure
(p1 ) will be recorded as function of time.

Last change: February 21, 2024 Milan Bernauer

 Department of inorganic technology: Single Gas Permeace

p2 ∆p p1

v7 v8

v1 v2 v3

v5
v4

v6

Figure 2: Experimental apparatus for unsteady permeation experiments. Valves v1 to v3 are destined to set
the volume of the vessel, valve v8 serve for filling up the apparatus by chosen gas and valve v7 is used to
open the interior volume to the atmosphere (through the membrane).

4 Treatment of data from single gas permeance measurement

In our calculations we will consider only Knudsen diffusion and viscous flow to participate to the total flow
of the permeating gas.
As mentioned above, permeance of gas is defined as ratio of intensity of molar flow N and the pressure
gradient ∆p over the membrane
N
Π= , (11)
∆p
where ∆p = p1 (t) − p2 , p1 (t) is in our case pressure inside the vessel and p2 is atmospheric pressure. In
following text we omit (t) for simplicity. Material balance of permeation apparatus (figure 3) is as follows

dn1
= −AN, (12)
dt
where A is membrane cross section area and N is the intensity of molar flow. For ideal gas we can write
dp1 ART
=− N = QN. (13)
dt V
It is obvious that
dp1 d∆p
= . (14)
dt dt
Data obtained on the experimental apparatus are in the form of time dependence of pressure p1 (see figure
3). To obtain the value of N from equation 13 it is necessary to evaluate the derivative dp1 /dt. This goal
could be achieved by several ways:

4
Last change: February 21, 2024 Milan Bernauer
 Department of inorganic technology: Single Gas Permeace

p1( t) 4 p=p1

d p1( tx)
[T ] d tx
Ni( t) A =F2,i( t)
p1( tx )
[V ]

tx t

Figure 3: A schematic picture of the permeation apparatus (left) and a typical output from permeance
measurement (right).

Numerical differentiation Numerical derivation of obtained data can be done by using following rela-
tionship
dp1 d∆p p1,(t+∆t) − p1,(t)
= ≈ . (15)
dt dt ∆t
This approach often fails because of the limited number precision of pressure gauge. This effect has an more
pronounced impact specially at lower pressure differences or slower pressure decrease (see figure 4). This
inconvenience hampers the use of equation 15.

d p ( tx) t
pi
p d tx ti
p1
p1 <0
t1
p2
>0
p2 t2

t1 t2 t

Figure 4: A schematic picture of the numerical evaluation of dp/dt.

5
Last change: February 21, 2024 Milan Bernauer
 Department of inorganic technology: Single Gas Permeace

Polynomial representation of data The other way is to represent measured data by adequately chosen
polynomial function
np
X
p1 (t) = ai ti−1 , (16)
i=1

where np is the number of parameters. The derivative dp1 /dt is then calculated using following formula
np
dp1 d∆p X
= = (i − 1)ai ti−2 . (17)
dt dt i=1

The advantage of this approach is its simplicity and fast application. The disadvantage is the fact, that
by using an empirical formula (equation 16) we introduce uncertainty1 and by deriving this “uncertain”
approximation (equation 17) we enhance this error by order of magnitude.

Using a representative ("real") model The most effort end time consuming approach is also the best.
We derive a model, which represent the gas transport mechanism in porous media and derived relationship
∆p = f (t) will be used to represent measured data. This approach is described in following section.

4.1 Model of Knudsen diffusion and viscous (Poiseuille) flow

By solving differential equation 8 for membrane of thickness δ we obtain
DK
NK = − (p2 − p1 ) = ΠK ∆p, (18)
δ
where the parameter ΠK = DK /δ. We keep in mind that we choose in the first paragraph ∆p = p1 − p2 and
thus the minus sign in equation 18 disappear.
By integration of equation 9 over a membrane of thickness δ we obtain
DP 2 ΠP 2
NP = − (p2 − p21 ) = (p − p22 ) (19)
2δ 2 1
where the last therm (p21 − p22 ) can be transformed as follows

Πp ∆p2
NP = (p1 − p2 )(p1 + p2 ) = ΠP ( + p2 ∆p) (20)
2 2
This two above mentioned mechanism of gas transport occurs in parallel and their contribution to the
total flow can be represented by their linear combination

∆p2
N = NP + NK = ΠP ( + p2 ∆p) + ΠK ∆p (21)
2
From equation 11 and equation 23 one can evaluate the permeance as function of the pressure difference
1
Π= ΠP ∆p + ΠP p2 + ΠK (22)
2
1 In other words we are not describing the exact nature of the diffusion process.

6
Last change: February 21, 2024 Milan Bernauer
 Department of inorganic technology: Single Gas Permeace

4.2 Solution of Knudsen and Poiseuille model - obtaining parameters ΠK and

ΠP
By substituting N in equation 13 by equation 21 we obtain
d∆p 1
= Q( ΠP ∆p2 + (ΠP p2 + ΠK )∆p.) (23)
dt 2
The solution of differential equation 23 is as follows (after subsequent evaluation of ∆p)
exp{KQt + C1 }
∆p = 2K , (24)
ΠP − ΠP exp{KQt + C1 }
where K = ΠP p2 + ΠK . The adjustable parameters are K, ΠP a C1 . Their estimation can be obtained from
nonlinear regression procedure of experimental data, i.e. by nonlinear fit of ∆p(t) data by equation 24. This
task is quite complicated because of strong correlation between unknowns parameters. To avoid unsuccessful
convergence of nonlinear estimation procedure, we have to rearrange the equation 24 in following way
exp{KQt + C1 }
∆p = L , (25)
1 − exp{KQt + C1 }
where
L = 2(p2 + ΠK /ΠP ) (26)
For nonlinear regression it is primordial to choose "good" initial estimates of fitted parameters. For parameter
L we can calculate the ratio ΠK /ΠP in equation 26 from 8 and 9 as
r
32 η 2RT
ΠK /ΠP = . (27)
3 r πM
For temperature around 298.15 K this ratio is approximately equal to
η
ΠK /ΠP (298 K) = 635.6 √ (28)
r M
This equation provide the initial estimate of parameter L (through equation 26).
The initial estimate of C1 parameter is given by
∆p0
 
C1 = ln , (29)
∆p0 + L
where ∆p0 is the initial pressure difference.
Values of ΠK a ΠP are then computed from parameters obtained from nonlinear regression by following
relationships
 
2p2
ΠK = K 1 − (30)
L
K
ΠP = 2 (31)
L
The flux is then evaluated from equation 21 using parameters 30 and 31
K
N= ∆p2 + K∆p. (32)
L
The permeance as function of pressure difference is then calculated from equation 22
K
Π= ∆p + K. (33)
L

7
Last change: February 21, 2024 Milan Bernauer
 Department of inorganic technology: Single Gas Permeace

4.3 Calculation procedure and some recommendations

1. Fitting of measured data
(a) Find suitable input parameters, such as viscosity, molar mass and membrane area. Physical
properties for some gases can be found in table 1
(b) Estimate initial values of parameters of equation 24 using relationships described in previous
section
(c) Perform a nonlinear fit using one of provided programs (Octave or Matlab), or use your own
program (recommended!)
(d) If the fit is successful (calculated parameters are "reasonable"), then evaluate desired parameters.
If the non-linear procedure fails, try a new fit with different initial estimates (i.e. different initial
estimation of mean pore radius or different value of Πp )
2. Display results
(a) Display relative deviation of fitted values (∆pc ) from experimental values (∆pe )
(b) Display permeance dependence on ∆p, (equation 33)
(c) Display ratio of Si/j = Πi (∆p)/Πj (∆p) (ideal separation factor), where subscripts i and j corre-
spond to two different gases
3. Final remarks for report writing
(a) If you provide a value (result or input), write corresponding units along the number (i.e. for
Boltzmann constant kB = 1.3806503 · 10−23 m2 kg s−2 ). Number without units is . . . is a number
without units (useless)!
(b) Do not make copy of whole text or whole images from un-cited sources (no CTRL+c - CTRL+v).
If you want to copy a whole image, you have to ask the author. Better way is to draw your own
picture and cite the original in the reference section or your report. Also use your own words
instead of "hard-coping" of somebody else’s text. This is a base of scientific work and also of the
scientific "etiquette"

Gas Molar mass/(kg mol−1 ) η(293 K)/(Pa s) η(298 K)/(Pa s)

H2 2.02 ·10−3 9.9843 ·10−6 9.0837 ·10−6
He 4.00 ·10−3 19.526 ·10−6 19.742 ·10−6
CH4 16.04 ·10−3 10.887 ·10−6 11.042 ·10−6
N2 28.01 ·10−3 17.283 ·10−6 17.512 ·10−6
CO2 44.01 ·10−3 14.688 ·10−6 14.923 ·10−6

Table 1: Dynamic viscosity (η) and molar mass of measured gases.

References
[1] Roy, S., Raju, R., Chuang, H., Cruden, B., Meyappan, M., J. Appl. Phys., 93 (2003) 4870
[2] Fott, P. and Petrini, G., Appl. Cat., 2 (1981) 367
[3] Rutherford, S., and Do, D., Ind. Eng. Chem. Res., 38 (1999) 565

8
Last change: February 21, 2024 Milan Bernauer
 Department of inorganic technology: Single Gas Permeace

Measured data Rel. deviation of fit

1000 0.01
He He
H2 H2

800 0.005

600 0
p / kPa

rd
400 -0.005

200 -0.01

0 -0.015
0 200 400 600 800 1000 1200 1400 0 200 400 600 800 1000 1200 1400
t/s t/s

Permenance Separation factor

1.6 0.5354
He S(i/j)
H2
0.5353
1.4
Π 105/(mol-1s-1m-2Pa-1)

0.5352

1.2
0.5351
S (i/j)

1 0.535

0.5349
0.8

0.5348

0.6
0.5347

0.4 0.5346
100 150 200 250 300 350 400 450 500 100 150 200 250 300 350 400 450 500
∆p/kPa ∆ p/kPa

Figure 5: Measured data (∆p(t)), relative deviation of fit (rd = (∆pc − ∆pe )/∆pe ), permeance and ideal
separation factor of CH4 and H2 as function of pressure difference on titanosilicate membrane.

A Example of Octave program for nonlinear fit

%% This program calculates the coefficients p() resulting from
%% non-linear regression fit of data (data.txt) by function F

T = 298.15; % Temperature [K]

R = 8.314; % Univeral gas constant [J/mol/K]
V = 0.544e-3; % Apparatus volume [m3]
A = 0.7274635e-4; % Membrane area (membrane type K) [m2]
eta= 11.042e-6; % Gas dymanic viscosity (CH4) [Pa.s]
Mw = 16e-3; % Molar mass (CH4) [kg/mol]
rp = 1e-7; % Mean pore radius estimate [m]
val= dlmread(’data.txt’); % Read input data from file
y = val(:,2)*1e3-val(:,4)*1e3; % Independent variables
x = val(:,1); % Dependent variable
[npts,trash] = size(x); % Read n. of points [npts]
Q = -R*T*A/V; % Q constant

Last change: February 21, 2024 Milan Bernauer

 Department of inorganic technology: Single Gas Permeace

%% Intial estimates
p2 = sum(val(:,4))*1e3/npts; % Mean atmospheric pressure [Pa]
dp0= (val(1,2) - val(1,4))*1e3; % Initial pressure difference [Pa]
%% L
pin(1) = 2*p2 + 32*eta/(2*rp)*sqrt(2*R*T/(3.1415926*Mw));
%% K
pin(2) = pin(1)*1.0e-11;
%% C
pin(3) = log(dp0/(dp0+pin(1)));
%% --------
ndf = (npts - 3); % N. deg. of freedom
wt = ones(size(x)); % Weigths
niter = 100; % Max. num. of iteration for non-lin. fit
stol = 1e-9; % Tolerance (stop. criterion)
%
%% Function to fit
F = @(x,p)(p(1).*(exp(p(2).*Q.*x(:,1) + p(3))./( 1 - exp(p(2).*Q.*x(:,1) + p(3))) ));
%% This is the Octave fitting routine for explanation read:
%% http://octave.sourceforge.net/optim/function/leasqr.html
%% [f,p,cvg,iter,corp,covp,covr,stdresid,Z,r2]=
%% leasqr(x,y,pin,F,{stol,niter,wt,dp,dFdp,options})
dFdp = ’dfdp’;
dp = [1e-6; 1.e-6; 1.e-6];
[f,p,kvg,iter,corp,covp,covr,stdresid,Z,r2] = leasqr(x, y, pin, F,stol,niter,wt,dp,dFdp);

%% RESULTS
fprintf(’L = %12.4e\n’, p(1));
fprintf(’K = %12.4e\n’, p(2));
fprintf(’C = %12.4e\n’, p(3));
fprintf(’iter = %4i\n’, iter);
fprintf(’ndf = %4i\n’, ndf);
fprintf(’WSSR = %12.4e\n’, sqrt(sum((f(:)-y(:)).^2)./ndf));
fprintf(’PIk = %12.4e\n’, p(2).*(1-2*p2./p(1)));
fprintf(’PIp = %12.4e\n’, p(2)/p(1));
fprintf(’rp = %12.4e\n’, (32/2)/(p(1)-2*p2)*eta*sqrt(2*R*T/3.1415926/Mw));

B Matlab program for nonlinear fit

%% This program calculates the coefficients p() resulting from
%% non-linear regression fit of data (data.txt) by function F

T = 298.15; % Temperature [K]

R = 8.314; % Univeral gas constant [J/mol/K]
V = 0.544e-3; % Apparatus volume [m3]
A = 0.7274635e-4; % Membrane area (membrane type K) [m2]
eta= 11.042e-6; % Gas dymanic viscosity (CH4) [Pa.s]
Mw = 16e-3; % Molar mass (CH4) [kg/mol]

10

Last change: February 21, 2024 Milan Bernauer

 Department of inorganic technology: Single Gas Permeace

rp = 1e-5; % Mean pore radius estimate [m]

val= dlmread(’data.txt’); % Read data from file
y = val(:,2)*1e3-val(:,4)*1e3; % Independent variables
x = val(:,1); % Dependent variable
[npts,trash] = size(x); % Read n. of points [npts]
Q = -R*T*A/V; % Q constant

%%Nonlinear least squares regression

%% Intial estimates
p2 = sum(val(:,4))*1e3/npts; % Mean atmospheric pressure [Pa]
dp0= (val(1,2) - val(1,4))*1e3; % Initial pressure difference [Pa]
%% L
pin(1) = 2*p2 + 32*eta/(2*rp)*sqrt(2*R*T/(3.1415926*Mw));
%% K
pin(2) = pin(1)*1.0e-11;
%% C
pin(3) = log(dp0/(dp0+pin(1)));
%% N. deg. of freedom
ndf = (npts - 3);
stol = 1e-9; % Tolerance (stop. criterion)
%
%% Function to fit, F(<pars. to evaluate>,<independet var.>)
F = @(p, x)(p(1).*(exp(p(2).*Q.*x(:,1) + p(3))./( 1 - exp(p(2).*Q.*x(:,1) + p(3))) ));

%% This is the Matlab fitting routine for explanation read:

%% http://www.mathworks.com/help/toolbox/stats/nlinfit.html
%%
[p,r,J,covp,wssr] = nlinfit(x,y,F,pin);

%% RESULTS
fprintf(’L = %12.4e\n’, p(1));
fprintf(’K = %12.4e\n’, p(2));
fprintf(’C = %12.4e\n’, p(3));
fprintf(’ndf = %4i\n’, ndf);
fprintf(’WSSR = %12.4e\n’, sqrt(wssr));
fprintf(’PIk = %12.4e\n’, p(2).*(1-2*p2./p(1)));
fprintf(’PIp = %12.4e\n’, p(2)/p(1));
fprintf(’rp = %12.4e\n’, (32/2)/(p(1)-2*p2)*eta*sqrt(2*R*T/3.1415926/Mw));

11

Last change: February 21, 2024 Milan Bernauer