Fuel 196 (2017) 47–58

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Full Length Article

Determining the diffusion coefficient of gas diffusion in coal:

Development of numerical solution
Gongda Wang a,b,⇑, Ting Ren b, Qingxin Qi a, Jia Lin b, Qingquan Liu c, Jian Zhang b
a
Mine Safety Technology Branch, China Coal Research Institute, 100013 Beijing, China
b
School of Civil, Mining and Environmental Engineering, University of Wollongong, Wollongong, New South Wales 2500, Australia
c
Key Laboratory of Coal Methane and Fire Control, Ministry of Education, China University of Mining & Technology, Xuzhou 221116, China

a r t i c l e i n f o a b s t r a c t

Article history: Determining gas diffusion coefficient from experimental data is a key step of reproducing and predicting
Received 10 July 2016 the diffusion process in coal. Previously analytical solution, including the unipore diffusion model and the
Received in revised form 18 January 2017 bidisperse diffusion model, has been used extensively to estimate the gas diffusion coefficient(s) in coal.
Accepted 23 January 2017
The utilization of analytical solution is convenient, however, there are some defects which may affect the
Available online 3 February 2017
accuracy of the results. For example, it is not suitable for fitting manometric sorption data, and the
assumption of linear adsorption isotherm is not true.
Keywords:
In this paper, we present a numerical solution to determine the gas diffusion coefficients. Three models
Diffusion coefficient
Coal
were developed based on different assumptions of pore system and diffusion forms, i.e., unipore model
Numerical solution assuming one kind of pore and diffusion, bidisperse model I (BM I) assuming independent macropore dif-
Unipore model fusion and micropore diffusion, bidisperse model II (BM II) assuming dependent macropore diffusion and
Bidisperse model micropore diffusion. Nitrogen diffusion experiment was conducted and the adsorption isotherm was
measured. Sphere geometry was built for numerical simulation and the proposed models were used to
fit the experimental data to determine the diffusion coefficients. Results show that neither the analytical
unipore model nor the numerical unipore model can describe the diffusion process perfectly. By giving
the same diffusion coefficient, the modelled fractional uptake ratio of numerical unipore model is smaller
than the result of analytical unipore modeling at early stage while greater at later stage, which is due to
the different assumptions of adsorption isotherm. Both BM I and BM II can describe the diffusion process
well. The determined macropore diffusion coefficients of the two models are similar, while the micropore
diffusion coefficient and the macropore adsorption ratio of BM II are greater than that of BM I. These can
be explained by the different roles of the macropore diffusion in the two models. The gas pressure change
at the center of the coal sphere was examined, from the modeling result of BM I, the macropore pressure
increases sharply and then drops along with the external gas pressure, while the initial increasing rate of
macropore pressure of BM II is much smaller and tends to be stable at later stage. No apparent impacts of
initial gas pressures on diffusion coefficients can be observed, the change of diffusion coefficients and
macropore adsorption ratio are actually small with increasing gas pressure.
The numerical solution of determining gas diffusion coefficients can easily relax the assumptions and
restrictions of the analytical solution. It can be used to test different kinds of coal samples, investigate
different diffusion mechanisms and match all kinds of experimental data. The measured sorption iso-
therm and coal properties can also be incorporated into the modeling, which makes the determined dif-
fusion coefficients more reliable. This paper is a preliminary attempt and we hope it can bring the
researchers some new ideas about studying the gas diffusion characteristics in coal.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction

⇑ Corresponding author at: School of Civil, Mining and Environmental Engineer-
ing, University of Wollongong, Wollongong, New South Wales 2500, Australia.
E-mail address: wgdcumt@gmail.com (G. Wang).
For coalbed methane (CBM) production and coal mine gas man-
agement, gas diffusion in coal refers to the gas exchange between
coal cleat and coal matrix. It plays an important role in controlling
the gas production rate together with gas sorption capacity and

http://dx.doi.org/10.1016/j.fuel.2017.01.077
0016-2361/Ó 2017 Elsevier Ltd. All rights reserved.
48 G. Wang et al. / Fuel 196 (2017) 47–58

coal permeability. A diffusion process may be related to three found to fit better. Generally the idea of bidisperse model divides
mechanisms, i.e., molecular diffusion, Knudsen diffusion and sur- the diffusion into a macropore diffusion stage and a micropore dif-
face diffusion [1,2]. Different mechanisms prevail in different scale fusion stage (or rapid/slow diffusion stages in the strict sense), tak-
of pores, for example: the molecular diffusion dominates the gas ing the constant external gas concentration boundary as an
flow in macropores where the pore diameter is way greater than instance:
the mean free path of gas molecules; the Knudsen diffusion is  
ma 6 X
1
1 Da
apparent when the mean free path is approaching the pore diam- ¼1 2 exp n2 p2 t 2 ð5Þ
eter; gas flow in micropores (<2 nm) is mainly controlled by sur- ma1 p n¼1 n2 ra
face flow. As a heterogeneous material, coal has a very
 
complicated pore system which includes all kind of pores from mi 6 X
1
1 Di
¼1 2 exp n2 p2 t 2 ð6Þ
micro- to macro- and this has been confirmed by various research- mi1 p n¼1 n2 ri
ers [3–5].
Different laboratory methods have been used to measure the where the subscripts a and i represent the corresponding parame-
gas diffusion process in coal, such as sorption method, steady state ters of rapid macropore diffusion and slow micropore diffusion,
Da Di
method and inverse diffusion method [6–8], among them the sorp- respectively. and can also be written as the effective diffusion
r 2a r 2i
tion method was most extensively used due to the simpleness and
coefficients Dae and Die (1=s), respectively. The overall uptake is
reliability of the experimental setup. The way of studying diffusion
sum of the two stages:
phenomenon through sorption experiment is also called sorption
kinetic study, as it can be measured along with determining mt ma þ mi ma mi
¼ ¼b þ ð1  bÞ ð7Þ
adsorption capacity of gas in coal. A key problem for reproducing m1 ma1 þ mi1 ma1 mi1
and predicting the diffusion process is to estimate the effective dif-
where b ¼ ma1mþm
a1
is the ratio of macropore adsorption to the total
fusion coefficient/coefficients. During the past decades, the influ- i1

ences of gas type [9–11], coal properties [3,12,13], sample size adsorption (1). Through adjusting De in unipore model, Dae , Die
[9,14,15], gas pressure [16,17], moisture [18] and temperature and b in bidisperse model, the best fitting curve of mm1t versus time
[19,20] on diffusion coefficient(s) have been discussed widely. is selected by comparing to the experimental results, and the diffu-
Even though the nature of heterogeneous coal structure makes it sion coefficient/coefficients can thus be determined.
appear to be difficult to fit diffusion data with a simple model, a The above attempts to model diffusion experimental data can
unipore model with single diffusion coefficient was indeed found be classified as analytical solution (mathematical solution), and all
to be sufficient in some experiments [21–23], while bidisperse of them are in the form of infinite series. Although this method
model with two diffusion coefficients was found to fit well in other has been extensively utilized since decades ago and the fitting
experiments [1,11,14,24]. results appeared to be good in most cases, there are still some
The unipore model assumes coal particles used in the experi- defects which may affect the accuracy and applicability of analyt-
ment are homogeneous spheres with uniform radius r, the frac- ical solution:
tional uptake of adsorbed gas mass can be written as:
   Both unipore model and bidisperse model are based on the
mt 6 X
1
1 D
¼1 2 exp n2 p2 t 2 ð1Þ assumption of linear adsorption, while the adsorption of gas
m1 p n¼1 n2 r in coal is typical Langmuir-type rather than linear-type. This
flaw is expected to be more obvious at high pressure stage as
where mt is the amount of adsorbed gas at time t (g), m1 is the final
been pointed out before [14].
adsorbed amount after equilibrium (g), D is the diffusion coefficient
 From the expressions of bidisperse model it can be seen that the
(m2 =s) and t is the sorption time (s). The value of rD2 can be written as
rapid macropore diffusion stage and the slow micropore diffu-
an effective diffusion coefficient De (1=s). Note Eq. (1) is under the sion stage are actually assumed to be independent. However,
assumption that the external/surface gas concentration is constant, the independence of the two diffusion stages is related to the
thus it is suitable for volumetric and gravimetric sorption method pore structure of coal, and this assumption cannot be always
but theoretically it cannot be used for manometric sorption method, true.
where the external gas concentration changes with time. The solu-  The analytical solution is not suitable for modeling manometric
tion of fitting manometric sorption experimental data can also be sorption data. Even though modified solution of unipore model
found from Crank [25]. By assuming the initial gas concentration was proposed (Eqs. (2)–(4)), it has the assumption that the ini-
in coal is 0, the fractional uptake of adsorbed mass can be expressed tial gas concentration in coal is 0, but this is not true for most
as: sorption steps. Moreover, to apply bidisperse model to fit mano-
 
mt X1
6aða þ 1Þ exp  rD2 q2n t metric sorption data, two sets of rD2 , a, qn , V, m1 need to be
¼1 ð2Þ
m1 n¼1
9 þ 9a þ q2n a2 adjusted separately for macropore diffusion and micropore dif-
fusion, which will easily lead to overparameteration.
Here, qn are the non-zero roots of:  The above models restricted to a specific geometry of coal,
3qn which are all spheres.
tan qn ¼ ð3Þ
3 þ aq2n
Comparing with analytical solution, numerical solution is free
where a is the ratio of the void volume Vvoid to the volume of the from the above restrictions. The basic assumption can be improved
solid spheres (1). The parameter a is calculated from the final frac- by developing different numerical models, and different coal
tional uptake of gas by coals: geometries can be easily established. Modern simulation software
m1 1 provides a fast model resolving speed and thus increases the appli-
¼ ð4Þ cability of this method. Preliminary studies on numerical modeling
Vvoid C 0 1 þ a
of gas diffusion in coal can be found from Cui et al. [11] and Shi and
where C 0 is the initial gas concentration of void space (mol=m3 ). Durucan [1] who discussed numerical unipore model and bidis-
Unipore model may be adequate for some bright coals but for most perse model. However, afterwards the related research is rare. It
coals with a multimodal pore distribution, bidisperse model was should be noted that Both Cui model and Shi & Durucan model
 G. Wang et al. / Fuel 196 (2017) 47–58 49

were built in polar coordinate which is limited to the sphere geom-
etry, and meanwhile a questionable assumption was made by
assuming gas adsorption takes place within micropores only.
In the present paper, we propose one numerical unipore model
and two numerical bidisperse models of gas diffusion in three
dimensional Cartesian coordinate. Both independent and depen-
dent macropore and micropore diffusions are discussed for the
bidisperse models. Nitrogen diffusion experiment was conducted
to examine the influences of different models on the determined
diffusion coefficients.
It should be pointed it out that unlike the interactive coal cleat
and matrix system which has been accepted extensively [26,27],
there is no evidence showing that whether the macropore system
and micropore system of coal matrix are independent or not. From
the previous concept of dual pore systems as shown in Fig. 2, the
gas diffusion goes through macropores into micropores, which is
similar to the assumption of BM II.
The gas concentration can be expressed by real gas law:
nf p
C¼ ¼ ð8Þ
V f ZRT

where C is the free gas concentration (mol=m3 ), nf is free gas molar

2. Model developments
volume (mol), Vf is free gas volume (m3 ), p is free gas pressure
which is equal to the adsorbed gas pressure (Pa), Z is free gas com-
All the model developments obey the following general
pressibility factor (1), R is gas constant (J=K  mol) and T is the sys-
assumptions: the diffusion system is isothermal; coal and pore sys-
tem temperature (K). By using Eq. (8), the gas concentration can be
tem is isotropic and homogeneous; the pores are incompressible
transformed to gas pressure. There are several gas adsorption mod-
and the void volume is constant with time, i.e., the adsorption
els for describing the interaction between coal and gas [27], and
induced coal swelling or desorption induced coal shrinkage is not
here we adopt the commonly used Langmuir model as discussed
considered here; gas adsorption occurs instantaneously, i.e., the
in the general assumption:
free phase gas pressure is equal to the adsorbed gas pressure in
an individual unit of coal; gas diffusion in pores is controlled by V 0p
Va ¼ ð9Þ
Fick’s second law; for Bidisperse models, the gas adsorptions in PL þ p
both macropore system and micropore system obey Langmuir
where V a is the adsorbed gas volume (m3 =kg), V 0 is the Langmuir
model.
The above general assumptions of numerical solution are actu- gas volume (m3 =kg) and PL is the Langmuir gas pressure (Pa).
ally the same as that of the previous analytical solution, apart from
Langmuir adsorption isotherm rather than linear adsorption iso- 2.1. Development of unipore model
therm is adopted. Three numerical models are developed based
on different concepts as shown in Fig. 1: For a unit of coal, the mass of gas includes free phase gas and
adsorbed gas, which can be written as:
(1) Unipore Model (UM): coal contains only one kind of pore mm ¼ Va qc qsg þ /pM=ZRT ð10Þ
and diffusion, which is driven by the concentration gradient
between exterior and interior of the coal particle. where mm is the total gas mass in a unit of coal (g), qc is the density
(2) Bidisperse Model I (BM I): coal contains independent macro- of coal (kg=m3 ), qsg is the density of gas at the standard condition
pore system and micropore system, diffusions in the two (kg=m3 ), M is gas molar mass (kg=mol) and / is coal porosity (1).
systems are controlled by different diffusion coefficients Because gas diffusion is driven by the concentration gradient
and the two diffusions are driven separately by the concen- between exterior and interior, by applying the mass conservation
tration gradients between exterior and interior of the coal law and Fick’s second law to the coal unit, we can obtain:
particle.
@mm
(3) Bidisperse Model II (BM II): coal contains continuous macro- þ r  ðDu MrCÞ ¼ 0 ð11Þ
@t
pore system and discontinuous micropore system, diffusions
in the two pore systems are controlled by different diffusion where Du is the diffusion coefficient of Unipore model (m2 =s). Sub-
coefficients and the mechanisms are not the same. Macrop- stituting Eqs. (8) and (10) into Eq. (11) and neglecting the change of
ore diffusion is driven by the concentration gradient coal porosity with respect to time, we can get:
between exterior and interior, while micropore diffusion is  
@Va /M @p Du M
driven by the concentration gradient between macropore qc qsg þ þr  rp ¼ 0 ð12Þ
@t ZRT @t ZRT
and micropore.

Fig. 1. Concepts of gas diffusions of Unipore Model, Bidisperse Model I and Bidisperse Model II.
50 G. Wang et al. / Fuel 196 (2017) 47–58

@ma
þ r  ðDa2 MrC a Þ ¼ Q e ð18Þ
@t

@mi
¼ Qe ð19Þ
@t
where Da2 is the diffusion coefficient of macropore diffusion (m2 =s),
C a is gas concentration in macropores (mol=m3 ), Q e is gas exchange
between macropores and micropores which can be expressed as:
Q e ¼ Di2 rc MðC a  C i Þ ð20Þ
where Di2 is the diffusion coefficient of micropore diffusion (m2 =s),
C i is the gas concentration in micropores (mol=m3 ), rc (1=m2 ) is the
shape factor which is related to the cleat spacing L [28]:
3p2
rc ¼ ð21Þ
L2
Because coal particles are studied in this paper and the cleat
Fig. 2. Schematic model of sphere coal containing a binder phase and crystals [1].
spacing is hardly to estimate, Di2 rc is defined as the effective gas
diffusion coefficient Dei2 , (1=s). Substituting Eqs. (14) and (15) into
Eqs. (18) and (19), the Bidisperse Model II can be deduced as:
Combining Eqs. (9) and (12), the UM model can be expressed as:  
  b2 V 0 PL qc qsg @pa b2 /M @pa Da2 M
V 0 PL qc qsg @p /M @p Du M þ þ r   r pa
þ þr  rp ¼ 0 ð13Þ ðPL þ pa Þ2 @t RT @t ZRT
ðPL þ pÞ2 @t ZRT @t ZRT
¼ Dei2 Mðpa  pi Þ=ZRT ð22Þ

2.2. Development of bidisperse model I ð1  b2 ÞV 0 PL qc qsg @pi ð1  b2 Þ/M @pi

þ
ðPL þ pi Þ2 @t ZRT @t
Assuming the adsorbed volume is in proportion to the porosity
and b1 is the ratio of macropore adsorption volume to the total
¼ Dei2 Mðpa  pi Þ=ZRT ð23Þ
adsorption volume for BM I, the mass of gas in macropores ma
and micropores mi in a unit of coal can be written as:
3. Gas diffusion experiment
V0 pa
ma ¼ b1 mm ¼ b1 q q þ b1 /pa M=ZRT ð14Þ
PL þ pa c sg To examine the influences of different models on the determi-
nation of diffusion coefficients, nitrogen diffusion (sorption
mi ¼ ð1  b1 Þmm kinetic) experiment was conducted at University of Wollongong,
Australia. High pressure nitrogen up to 4 MPa was used in the
V0 pi
¼ ð1  b1 Þ q q þ ð1  b1 Þ/pi M=ZRT ð15Þ experiment. Bulk coal sample from Bulli coal seam, Sydney Basin
PL þ Pi c sg
of Australia was collected, crushed and sieved to 2.36–3.35 mm
where pa and pi denote the gas pressure in macropores and micro- particles. The reason of choosing this sample size is to avoid
pores (Pa), respectively. As the macropore diffusion and micropore destroying the macropore system of coal and meanwhile obtain a
diffusion are independent, the mass conversation equations are reasonable adsorption equilibrium time. A modified manometric
similar to UM. The expressions of Bidisperse Model I can be written method was adopted, detailed description of the apparatus can
as: be found from our previous study [29]. Approximately 200 g sieved
  coal sample was placed into a sample cell with a pressure trans-
b1 V 0 PL qc qsg @pa b1 /M @pa Da1 M
þ þ r   r pa ¼0 ð16Þ ducer, which was connected to a data logger and logging computer
ðPL þ pa Þ2 @t ZRT @t ZRT
during the test. The gas pressure in the sample cell was measured
  in real time by the pressure transducer with accuracy of 0.001 MPa,
ð1  b1 ÞV 0 PL qc qsg @pi ð1  b1 Þ/M @pi Di1 M and the reading intervals were from small to large (up to 300 s).
þ þ r   r pi ¼0
ðPL þ pi Þ2 @t ZRT @t ZRT Comparing with conventional manometric method, there is no ref-
ð17Þ erence cell and the injected gas mass is measured by a high accu-
rate balance (accurate to 0.001 g). The experimental temperature
where Da1 and Di1 are the diffusion coefficient of macropore diffu- of this study was 303 K. Helium was used to measure the void vol-
sion and micropore diffusion (m2 =s) in Bidisperse Model I, ume before adsorbing nitrogen and the following values were
respectively. recorded: the weight of sample cell m0 ; the amount weight of sam-
ple cell and coal sample m1 after 48 h evacuation at 333 K; the
2.3. Development of bidisperse model II amount weight of sample cell, coal sample and helium m2 ; the
pressure of helium in sample cell pHe . The mass of and Helium is:
Assuming b2 is the ratio of macropore adsorption volume to the
total adsorption volume for BM II, Eqs. (14) and (15) can still rep- mHe ¼ m2  m1 ð24Þ
resent the mass of gas in macropores ma and micropores mi in a The void volume in sample cell can be calculated from the SRK
unit of coal. From the assumption of BM II, micropore diffusion is equation:
actually the gas exchange between macropore and micropore.
ZSRK RTnHe ZHe RT mHe
For a unit of coal, applying the mass conversation law to Vvoid ¼ ¼  ð25Þ
macropores and micropores separately: pHe pHe MHe
 G. Wang et al. / Fuel 196 (2017) 47–58
where nHe is the molar volume of Helium in sample cell (mol), pHe is
the pressure of Helium measured by the pressure transducer (Pa),
ZHe is the compressibility of Helium at experimental pressure (1),
MHe is the molar weight of Helium (g=mol). The sample cell was
vacuumed again for another 2 h to remove helium, and a certain
pressure of nitrogen was injected into the sample cell. The following
values were measured to determine the diffusion process: the pres-
sure change with time, pt ; the amount weight of sample cell, coal
sample and nitrogen, mi . After last step of adsorption reached equi-
librium, the initial pressure was stepwise increased, about 0.5 MPa
of each step and up to 4 MPa.
The mass of free nitrogen in sample cell is:
Mpt Vvoid
mfree ¼
RTZ
The mass of total adsorbed nitrogen mti at time t of step i can
thus be calculated as:
mti ¼ mi  m1  mfree
The total excess sorption volume of nitrogen after equilibrium
can be expressed as:
V1 22:4m1
V excess ¼ ¼ ð28Þ
mcoal Mðm1  m0 Þ
where m1i is the total adsorbed nitrogen mass of step i after equi-
librium (g). By calculating the excess sorption volumes at different
pressure steps, the sorption isotherm was determined as shown in
Fig. 3 and fitted by the Langmuir model. The calculated V 0 and PL
are 16.52e3 m3/kg and 4.44 MPa, respectively. The correlation
coefficient R2 is 0.999, indicating a perfect fitting of Langmuir model
with the measured adsorption data.
When analysing the diffusion data, it should be noted that only
the first step started from a vacuum state, and the rest diffusions
processes started from a non-zero pressure. The calculated mti
and m1i include the adsorbed gas mass of previous step/steps
and thus cannot be directly used to calculate mm1t . To calculate the
adsorbed gas mass during step i:
mt ¼ mti  m1ði1Þ
m1 ¼ m1i  m1ði1Þ
Fig. 3. Nitrogen adsorption isotherm on Sydney Basin coal and the corresponding fitting line of Langmuir model.
51
where subscript i and ði  1Þ represent the step i and its previous
step. In order to obtain a general idea of the diffusion processes at
different initial pressures, the ratios between mt and m1 with
pffiffi
respect to the square root of time t are shown in Fig. 4.
It can be seen that the gas uptake ratios of different initial pres-
sure steps appears to be following a similar trend and the total
time taken for equilibrium does not vary significantly either. The
slope of the curves at early time stage is not very steep, indicating
the nitrogen diffusivity may not be very high.
4. Modeling and analyses
4.1. Comparison between unipore analytical solution and numerical
ð26Þ
solution
In this section, the modeling results of unipore analytical solu-
tion and unipore numerical solution are compared. As the analyti-
ð27Þ
cal solution for fitting manometric sorption data (Eqs. (2)–(4))
assumes the initial gas concentration to be 0 [25], only the first
step of diffusion data is used. The analytical bidisperse model is
not analysed because it is believed to be unsuitable for fitting
manometric sorption data, as discussed in the Introduction. The
parameters required for Eqs. (2)–(4) is calculated from the experi-
mental results as shown in Table 1 and the values of qn are gained
from Crank [25] based on the measured value of a.
Three diffusion coefficients were attempted to model the nitro-
gen fractional uptake ratio and the results are compared to the
measured data as shown in Fig. 5. As plotted, the value of
8.0e12 m2 =s obviously overestimates the diffusion speed and in
contrast, the value of 1.0e12 m2 =s underestimates the diffusion
speed. By carefully adjusting the value of diffusion coefficient, a
relatively good fitting result is gained (2.5e12 m2 =s) but some
deviations can still be seen as shown in Fig. 5. In order to obtain
a perfect fitting line, the diffusion coefficient should be greater at
the initial stage while smaller at the later stage. This conflict indi-
cates although the unipore analytical diffusion model can yield a
first-order approximation of the experimental results, it cannot
ð29Þ
describe the diffusion process perfectly.
The value of 2.5e12 m2 =s is then used to compare the differ-
ð30Þ
ence between the two solutions. Eq. (13) is the governing equation
52 G. Wang et al. / Fuel 196 (2017) 47–58

pffiffi
Fig. 4. Fractional uptake ratio with respect to t at different initial pressures.

Fig. 5. Fractional uptake ratios of measured data and unipore analytical modeling
with different diffusion coefficients.
of the numerical solution and it is solved by COMSOL Multiphysics
5.2. Mercury intrusion was carried out to determine the porosity of
Bulli coal, detailed result is given in Appendix. The parameters Table 2
used in the numerical modeling are listed in Table 2. A sphere with Parameters used in Unipore numerical modeling (measured values).
1.43e3 m radius is built as the model geometry as shown in Fig. 6 Parameters Value
and it consists of 23,617 elements. It should be pointed out that
Density of coal qc , kg=m3 1440
although we still adopt sphere geometry similar to the analytical
Porosity of coal /, dimensionless 0.049
solution, which is because the coal particles were used in the test, Langmuir volume V 0 , m3 =kg 16.52
it does not represent the numerical solution will be restricted to Langmuir pressure P L , MPa 4.44
the sphere geometry. If other kinds of coal shapes like cuboid, disc Temperature T, °C 303
and cylinder are prepared and tested for determining the diffusiv- Simulation time, s 232,270
ity, the numerical solution can provide the means of changing coal
geometry while keeping the same diffusion mechanism and model.
The boundary condition of the numerical model is the gas pres-
sure change on the external surface of the sphere, which is equal to
the gas pressure change in free volume of the sample cell. A second
order exponential decay function (Eq. (31)) is selected to represent
the pressure change curve. As can be seen from Fig. 7, a satisfactory
fitting result is obtained.
   
t t
p ¼ a1 exp  þ a2 exp  þ a3 ð31Þ
b1 b2
where a1 , a2 , a3 , b1 and b2 are fitting parameters of the second order
exponential decay function.
The numerical simulation is then performed using above model,
parameters and boundary condition. The amount of gas adsorbed
in the coal particle at time t is evaluated from the volume integra-
tion of the sphere:
ZZZ 
qc qsg V0 p
mt ¼ ð32Þ
PL þ p t
Theoretically the only difference of the modeling results
between two solutions is from the different assumptions of
adsorption isotherm. The analytical solution assumes linear
adsorption isotherm while the numerical solution follows the Fig. 6. Model geometry: sphere of coal particle, 1.43e3 m in radius.
experimental observation – the Langmuir-type adsorption iso-
therm. Fig. 8 shows how this difference impacts the modelled frac- nitrogen diffusion process. Both curves are in the similar trend,
tional uptake ratio mm1t . It can be seen that neither the analytical the modelled fractional uptake ratios are smaller than the
model nor the numerical model can accurately describe the measured data at the early stage while greater at the later stage.

Table 1
Measured parameters used in analytical unipore modeling.

Initial gas concentration C0 (g/m3) Void volume Vvoid (m3) Final adsorbed mass m1 (g) Ratio parameter a (1) Average sphere radius r (m)
4.537e3 1.788e4 0.527 0.537 1.43e3
 G. Wang et al. / Fuel 196 (2017) 47–58
Fig. 7. Measured gas pressure change in sample cell and the corresponding fitting line.
Fig. 8. Fractional uptake ratios of measured data, unipore analytical modeling and
unipore numerical modeling.
Comparing to the result of analytical solution, this trend is more
apparent for the result of numerical solution. The difference may
be explained by the feature of Langmuir-type adsorption isotherm,
of which the amount of adsorbed gas is becoming smaller with
increasing gas pressure. Because linear adsorption is like a compro-
mise between the larger adsorption volume of initial stage and the
smaller adsorption volume of later stage, the initial gas uptake
ratio may be overestimated and the later gas uptake ratio may
be underestimated by the analytical solution.
Aside from adopting the measured adsorption isotherm,
another advantage of the numerical solution is that it provides
means of examining the gas pressure/concentration change at dif-
ferent locations inside the coal particle. For example, the gas pres-
sure distribution on five slices of the coal particle at different times
(10 mins, 1 h, 5 h and 10 h) are shown in Fig. 9. It can be seen that
nitrogen gradually diffuses from the external surface to the middle
of the sphere. Even after 10 h of diffusion, the gas pressure in the
middle of the particle is still much lower than that of the external
locations.
53
4.2. Comparison between BM I and BM II
As discussed in model developments, Bidisperse Model I and
Bidisperse Model II are based on different assumptions of macrop-
ore and micropore systems. In order to illustrate how this differ-
ence impacts the determination of diffusion coefficient, the first
step of diffusion data is again used to calculate the corresponding
diffusion coefficients of these two models.
Eqs. (16) and (17) are the governing equations of BM I, Da1 , Di1
and b1 are adjusted to fit the experimental data. Eqs. (22) and (23)
are the governing equations of BM II, Da2 , Dei2 and b2 are adjusted
to fit the experimental data. For BM I, the boundary conditions of
pa and pi are both the measured pressure change as plotted in
Fig. 7. For BM II, the boundary condition of pa is the measured pres-
sure change, whilst the boundary condition of pi is zero flux bound-
ary according to the model assumption. Since Di1 is the diffusion
coefficient (m2 =s) while Dei2 is the effective diffusion coefficient
(1=s), the effective micropore diffusion coefficient of BM I is calcu-
lated for comparison:
Di1
Dei1 ¼ ð33Þ
r2
For both models, the total adsorbed amount of gas at time t is
evaluated from the sum of volume integrations of macropore
adsorption and micropore adsorption mass:
ZZZ  
qc qsg V0 pa qc qsg V0 pi 
mt ¼ b þ ð1  bÞ ð34Þ
PL þ pa PL þ pi t
where is b equal to b1 for BM I, and b2 for BM II. The modeling
results of BM I and BM II are shown in Fig. 10. It can be seen that
for both models, a satisfactory result is obtained. The macropore
diffusion coefficients of the two models are the same, which is
9e12 m2 =s. However, the micropore effective diffusion coefficients
and the macropore adsorption ratios vary greatly. The micropore
effective diffusion coefficient of BM II is about 5 times of BM I.
The macropore adsorption ratio of BM II is about half of the total
adsorption, while for BM I, it only accounts for 18% of the total
adsorption.
The modelled macropore and micropore pressure changes of
BM I and BM II at the center of the sphere are examined as shown
54 G. Wang et al. / Fuel 196 (2017) 47–58

Fig. 9. Gas pressure distribution inside the sphere coal particle at different times: (a) 10 mins, (b) 1 h, (c) 5 h, (d) 10 h.

in Fig. 11. The external surface pressure (boundary pressure) is also
plotted for comparison. It can be seen that the external surface
pressure keeps dropping, which is due to the gas diffusion from
free space into the coal particles. Macropore pressure change of
BM I at the center of the sphere experiences a sharply increase
and then an apparent decrease, indicating adsorption and desorp-
tion occur in turns in the whole macropore system. Although the
macropore diffusion coefficients of the two models are the same,
the initial increasing rate of macropore pressure of BM II is much
smaller than that of BM I and tends to be stable at the later stage,
no pressure drop can be observed. These can be explained by the
different roles of macropore diffusion in two models. For BM I,
all the migrated gas driven by macropore diffusion are adsorbed
by the macropore system, however, according to the assumption
of BM II, part of the migrated gas driven by macropore diffusion
provides the source of micropore diffusion, which induces the
increase of macropore pressure of BM II to be slower than that of
BM I. The micropore pressures of both BM I and BM II keep increas-
ing along the whole process and approach the macropore pressure
at the end of the diffusion. In comparison, the pressure increase of
BM I occurs earlier, while the increase rate of BM II is greater.
4.3. Impact of gas pressure on diffusion coefficients
Gas pressure has been identified as an important influencing
factor of the gas diffusion coefficient [11,20,30], however these
conclusions were derived from the analytical solutions, which
may be misled by the incorrect assumption of linear isotherm. It
is valuable to examine the impact of gas pressure on the diffusion
coefficients using numerical solutions. Diffusion coefficients of BM
I and BM II are adjusted to fit the gas uptake ratios at different
initial pressures. The changes of macropore diffusion coefficient
 G. Wang et al. / Fuel 196 (2017) 47–58 55

Da , micropore effective coefficient Die and macropore adsorption
ratio b versus pressure are shown in Figs. 12–14, respectively.
The relationship between gas pressure and diffusion coefficient
is controversial from the past studies. For example, Cui et al. [11]
suggested both macropore and micropore diffusion coefficients of
nitrogen decrease with pressure, while Zhang et al. [20] found
the effective diffusivity increases with the increasing nitrogen
pressure at 0.63–10.05 MPa. Based on our experimental data and
numerical simulation results as shown in Figs. 12 and 13, no appar-
ent trend between diffusion coefficients and pressure can be found
for both BM I and BM II. The changes of Da and Die are actually tiny
comparing with previous study results [11]. The largest values of
Da of BM I and BM II are about 2 times of the smallest values,
and the largest values of Dei of BM I and BM II are about 3 times
of the smallest values. The first two sorption steps have smaller
values of b as can be seen from Fig. 14, while the values keep
consistent for the last 5 sorption steps. Generally, the diffusion
parameters of both BM I and BM II do not change significantly with
respect to gas pressure. By comparing the differences between the
two models, it can be seen that the values of Da are similar, while
the values of Die and b of BM II are greater than that of BM I.
4.4. Discussion
From above analyses it can be seen that the application of
numerical solution requires the measurement of some coal proper-
ties such as coal density and porosity, and most importantly, the
gas sorption isotherm. If a manometric sorption method is used,
the change of gas pressure in sample cell also needs to be modelled
to represent the boundary condition. These works makes the appli-
cation of numerical solution complex comparing with the analyti-
cal solution, but it is worthwhile as the true coal properties and

a) b)
Fig. 10. Comparison between the measured data and modeling results of BM I (a) and BM II (b).

pffiffi
Fig. 11. Macropore and micropore pressure changes at the center of the sphere verses t.
56 G. Wang et al. / Fuel 196 (2017) 47–58

Fig. 12. Macropore diffusion coefficients of BM I and BM II versus initial gas pressure.

Fig. 13. Micropore effective diffusion coefficients of BM I and BM II versus initial gas pressure.

sorption isotherm are considered into the modeling and the results
are thus more reliable.
A problem of developing bidisperse models is how to evaluate
the relationship between the macropore adsorption ratio and
macropore volume ratio (to the total porosity). Apparently for a
certain volume of pores, macropores have smaller surface area
and adsorption volume, hence the macropore volume ratio should
be greater than macropore adsorption ratio to obtain the same
adsorption volume. However this proportion is hardly to be esti-
mated due to the complexity of multimodal pore distribution of
coal. In the proposed bidisperse models, we assume the adsorbed
volume is in proportion to the porosity. We think it is acceptable,
although not perfect, as the free gas mass is way smaller than
the adsorbed gas mass, and the small change of volume ratio
should not have great impact on the determined diffusion
coefficients.
Another question raised for the bidisperse models is about the
assumptions of BM I and BM II, whose is more realistic, in other
words, are the macropore and micropore systems dependent or
independent? For the real pore distribution of coal, both cases
 G. Wang et al. / Fuel 196 (2017) 47–58
Fig. 14. Macropore adsorption ratio of BM I and BM II versus initial gas pressure.
should exist. The gas may either diffuse from exterior to interior
through the continued micropores, or diffuse through the
macropore rapidly into the middle part of the coal particle and
then come into the adjacent micropores. However, we are inclined
to agree with the independent pore system assumption of BM I for
the present coal sample, as the ratio of macropore adsorption
seems to be too large from the simulation results of BM II. It should
be pointed out that the dependence of pore systems are not exam-
ined in this study, we recommend to use liquid nitrogen adsorption
or small angle X-ray scattering methods to further study the
assumptions in the future.
Only nitrogen is used in this study and the influence of gas type
on diffusion coefficients is not the focus of the present work. Fur-
ther studies on different gases are still recommended.
5. Conclusions
The flaws of the analytical solution of determining the gas dif-
fusion coefficients in coal were discussed, and three numerical dif-
fusion models were proposed based on different assumptions of
pore system of coal. Nitrogen diffusion experiment was conducted
and the results were used to determine the diffusion coefficients
through numerical simulation. The main conclusions are as
follows:
(1) Neither the analytical unipore model nor the numerical uni-
pore model can accurately describe the nitrogen diffusion
process. Both results are in a similar trend, the modelled
fractional uptake ratio is smaller than the measured data
at the early stage while greater at the later stage. By giving
the same diffusion coefficient, the difference between the
modelled result and measured data is more apparent for
numerical solution. It is due to the feature of Langmuir-
type adsorption isotherm.
(2) Both BM I and BM II can describe the diffusion process well.
The macropore diffusion coefficients of the two models are
similar. However, the micropore effective diffusion coeffi-
57
cients and the macropore adsorption ratios vary greatly.
Taking the first diffusion step as an instance, the micropore
effective diffusion coefficient of BM II is about 5 times of BM
I. The macropore adsorption ratio of BM II is about half of the
total adsorption, while for BM I, it only accounts for 18% of
the total adsorption.
(3) Macropore pressure change of BM I at the center of the
sphere experiences a sharply increase and then an apparent
decrease, while the initial increasing rate of macropore pres-
sure of BM II is much smaller than that of BM I and tends to
be stable at the later stage, no pressure drop can be
observed. The micropore pressures of both BM I and BM II
keep increasing along the whole process and approach the
macropore pressure at the end of the diffusion. The pressure
increase of BM I occurs earlier, while the increase rate of BM
II is greater.
(4) No apparent trend between diffusion coefficients and pres-
sure can be found for both BM I and BM II. The changes of
Da and Die are actually tiny, the largest values of Da of BM I
and BM II are about 2 times of the smallest values, and the
largest values of Dei of BM I and BM II are about 3 times of
the smallest values. Generally, the diffusion parameters of
both BM I and BM II do not change significantly with respect
to gas pressure.
Acknowledgements
This work was supported by National Natural Science Founda-
tion of China 51604153, Beijing Natural Science Foundation
2164057, China Postdoctoral Science Foundation (No.
2016M590519, 2016M600982) and The Australian Coal Industry’s
Research Program (ACARP C24019).
Appendix A. Mercury intrusion test result
See (Fig. A1 and Table A1).
58 G. Wang et al. / Fuel 196 (2017) 47–58

Fig. A1. Raw data of mercury volume versus mercury pressure of Bulli sample.

Table A1 [13] Krevelen DW. Coal: typology, physics, chemistry, constitution. 3rd completely
Porosity summary. revised edition. Elsevier; 1993.
[14] Busch A, Gensterblum Y, Krooss BM, Littke R. Methane and carbon dioxide
Sample (bulk) volume 0.7794 [cc] adsorption–diffusion experiments on coal: upscaling and modelling. Int J Coal
Intruded volume (interparticle) 0.0039 [cc] Geol 2004;60:151–68.
Total interparticle porosity 0.4954 [%] [15] Zhang J. Experimental study and modeling for CO2 diffusion in coals with
Intruded volume (intraparticle) 0.0341 [cc] different particle sizes: based on gas absorption (imbibition) and pore
Total intraparticle porosity 4.3700 [%] structure. Energy Fuels 2016;30:531–43.
Total intruded volume 0.0379 [cc] [16] Clarkson CR, Bustin RM. The effect of pore structure and gas pressure upon the
Total porosity 4.8654 [%] transport properties of coal: a laboratory and modeling study. 2. Adsorption
rate modelling. Fuel 1999;78:1345–62.
[17] Staib G, Richard S, Gray EMA. A pressure and concentration dependence of CO2
References diffusion in two Australian bituminous coals. Int J Coal Geol 2013;116–117:106–16.
[18] Pan Z, Connell LD, Camilleri M, Connelly L. Effects of matrix moisture on gas
[1] Shi JQ, Durucan S. A bidisperse pore diffusion model for methane displacement diffusion and flow in coal. Fuel 2010;89:3207–17.
desorption in coal by CO2 injection. Fuel 2003;82:1219–29. [19] Charrière D, Pokryszka Z, Behra P. Effect of pressure and temperature on
[2] Yang R. Gas separation by adsorption processes. London: Imperial College diffusion of CO2 and CH4 into coal from the Lorraine basin (France). Int J Coal
Press; 1997 [chapter 4]. Geol 2010;81:373–80.
[3] Karacan CÖ. Heterogeneous sorption and swelling in a confined and stressed [20] Zhang Y, Xing W, Liu S, Liu Y, Yang M, Zhao J, et al. Pure methane, carbon
coal during CO2 injection. Energy Fuels 2003;17:1595–608. dioxide, and nitrogen adsorption on anthracite from China over a wide range
[4] Radovic LR, Menon VC, Leon CA, Kyotani T, Danner RP, Anderson S, et al. of pressures and temperatures: experiments and modeling. RSC Adv
Evidence for an interconnected but constricted micropore system and 2015;5:52612–23.
implications for coalbed methane recovery. Adsorption 1997;3:221–32. [21] Jian X, Guan P, Zhang W. Carbon dioxide sorption and diffusion in coals:
[5] Wang G, Wang K, Ren T. Improved analytic methods for coal surface area and experimental investigation and modeling. Sci China Earth Sci 2012;55:633–43.
pore size distribution determination using 77 K nitrogen adsorption [22] Pone JDN, Halleck PM, Mathews JP. Sorption capacity and sorption kinetic
experiment. Int J Min Sci Technol 2014;24:329–34. measurements of CO2 and CH4 in confined and unconfined bituminous coal.
[6] Li Y, Xue S, Wang J, Wang Y, Xie J. Gas diffusion in a cylindrical coal sample – a Energy Fuels 2009;23:4688–95.
general solution, approximation and error analyses. Int J Min Sci Technol [23] Švábová M, Weishauptová Z, Přibyl O. The effect of moisture on the sorption
2014;24:69–73. process of CO2 on coal. Fuel 2012;92:187–96.
[7] Smith DM, Williams FL. Diffusion models for gas production from coals: [24] Siemons N, Wolf KHAA, Bruining J. Interpretation of carbon dioxide diffusion
application to methane content determination. Fuel 1984;63:251–5. behavior in coals. Int J Coal Geol 2007;72:315–24.
[8] Xu H, Tang D, Zhao J, Li S, Tao S. A new laboratory method for accurate [25] Crank J. The mathematics of diffusion. Clarendon Press; 1975.
measurement of the methane diffusion coefficient and its influencing factors in [26] Liu J, Wang J, Chen Z, Wang S, Elsworth D, Jiang Y. Impact of transition from
the coal matrix. Fuel 2015;158:239–47. local swelling to macro swelling on the evolution of coal permeability. Int J
[9] Han F, Busch A, Krooss BM, Liu Z, Yang J. CH4 and CO2 sorption isotherms and Coal Geol 2011;88:31–40.
kinetics for different size fractions of two coals. Fuel 2013;108:137–42. [27] Wang G, Ren T, Wang K, Zhou A. Improved apparent permeability models of
[10] Li D, Liu Q, Weniger P, Gensterblum Y, Busch A, Krooss BM. High-pressure gas flow in coal with Klinkenberg effect. Fuel 2014;128:53–61.
sorption isotherms and sorption kinetics of CH4 and CO2 on coals. Fuel [28] Lim KT, Aziz K. Matrix–fracture transfer shape factors for dual-porosity
2010;89:569–80. simulators. J Pet Sci Eng 1995;13(3):169–78.
[11] Cui X, Bustin RM, Dipple G. Selective transport of CO2, CH4, and N2 in coals: [29] Wang G, Ren T, Wang K, Wu Y. Influence of maximum pressure on the path of
insights from modeling of experimental gas adsorption data. Fuel CO2 desorption isotherm on coal. Energy Fuels 2014;28:7093–6.
2004;83:293–303. [30] Pillalamarry M, Harpalani S, Liu S. Gas diffusion behavior of coal and its impact
[12] Crosdale PJ, Beamish B, Valix M. Coalbed methane sorption related to coal on production from coalbed methane reservoirs. Int J Coal Geol
composition. Int J Coal Geol 1998;35:147–58. 2011;86:342–8.