Diffusion models for gas production from

coal
Determination of diffusion parameters

Douglas M. Smith’ andFrank L. Williams

Department of Chemical and Nuclear Engineering, University of New Mexico,
Albuquerque. NM 8713 i,
USA (Received 2 July f962)

A method for the laboratory measurement of transient diffusivity parameters for methane in coal has
been developed. Previously described chromatographic techniques for measurement of single
diffusivity parameters have been modified for the bi-disperse pore diffusion model. This model, which
is necessary to describe methane diffusion and desorption in coal for both short and long times,
requires the determination of three parameters to characterize a particular coal sample. The
experimental technique uses a pulse of methane gas which is introduced into a helium stream
flowing through a long tube containing the coal sample. The three desired parameters are
determined by matching the experimental and theoretical elution curves. Further refinement of the
technique accounts for tracer pulse duration thus allowing operation over a greater flow rate and
sample size range. This technique has been applied
to anthracite, bituminous and subbituminous coals. For all three coal types, the macropore diffusivity
5
( 10 ” s) was about two orders of ma9nitude higher than the micropore diffusivity. The anthracite
adsorbs more methane in the micropores than the other coals. These results showed the expected
trends taking the known change of pore structure with rank into account.

(Keywords: coal: methane; diffusion model)

Improving economics for the production of natural gas

from coal-beds has resulted in renewed interest in the
study of methane diffusion in coal. The production of
this coal-bed methane is generally considered to be a
three- step process. First, after a reduction in seam
pressure, methane desorbs from the internal surfaces of
coal. Next, desorbed gas molecules diffuse through the
coal pore structure to the coal’s naturally occurring crack
network. This step is followed by the simultaneous
laminar flow of water and gas through these cracks to
the production well. This same mechanism also
describes the migration of methane into underground
coal mines.
The modelling of the methane transport process
outlined above is important for several reasons. For the
commercial production of coal-bed methane, modelling
is necessary for assessing the economic viability of a
particular well and as a tool for optimizing drilling
strategy. For mine safety, modelling is helpful for
predicting ventilation requirements and studying the
effects of coal drainage before mining.
For a production model to be useful, the relative rates
of all three steps outlined above must be considered.
Since methane is physically adsorbed' on coal, the
desorption step is much faster than the two subsequent
steps and, thus, may be ignored in the overall
production model. The relative magnitude of the
dNusion and laminar flow steps has been studied by
Kissell’ and Kuuskra et at.'. However, both these studies
ignored the coupling of the diffusion and laminar flow
mechanisms. Ancell et nf.* have
• Present address: Department of Chemical Engineering, Montana
State University, MT, 59717, USA.
0016—2361/84/020256W$3.00
Qc 1984 Butterworth & Co. (Publishers) Ltd
256 FUEL, 1984, Vol 63, February
correctly recognized that both mass transport
mechanisms are necessary. The gas from the coal
simulator developed in that work used a ‘unipore’
diffusion model coupled to Darcy law equations for gas
and water flow through the cracks. This ‘unipore’ model
assumes that all the coal pores are of the same diameter
and has been utilized by several investigators of
methane diffusion in coal’ '. For dNusion effects to be
correctly included in a production model, the diffusion
model must be satisfactory for describing desorption
from a coal particle for both short and long times. It has
been reported previously by Smith and Williams"' that
the ‘unipore’ model is only applicable for describing
desorption for approximately the first 50% f gas
desorbed. The ‘unipore’ model was found to
underestimate the time required for complete desorption
by an order of magnitude.
The failure of the unipore model to describe methane
diffusion in coal may be attributed to the restrictive
assumption that all pores are the same diameter. Gan et
at.'1 have shown that many coals exhibit bi-disperse pore
distributions. To accurately describe the desorption of
methane from coal particles, a dNusion model must
include this bi-disperse pore structure. A bi-disperse pore
model for coal has been presented previously"' 2 which
accurately describes methane desorption rates over the
entire timescale of desorption. Three parameters are
required to describe diffusion with this model. These
parameters are the micropore effective diffusivity, D;, a
macropore dNusivity, D., and a constant, A, which is
proportional to the micropore/macropore gas distribu-
tion. At equilibrium, this distribution is equal to b/3D;.
The three bi-disperse pore model diffusion parameters
were initially determined’ by measuring the rate of
 Diffusion models for gas production from coal: D. M. Smith and F. L. Williams
desorption from coal particles. However, this technique is
time consuming(w 3 weeks per sample) and the results are
sensitive to the type of curve matching used. Therefore, an
experimental technique has been developed which allows
independent the rapid determination of these three parameters with the
greater than the minimum of experimental and computational effort. With pulse residence time.
these parameters determined for a particular coal, the bi-
disperse pore model may be combined with a laminar flow
due model for more realistic and accurate production to the number of samples necessary to characterize a
given modelling.
Pulse tracer analysis
In recent years, increased use of pulse
tracer(chromato- graphic) experiments for the
determination of transport coefficients has been reported
in the catalysis literature. Various investigations using
l
these techniques are reviewed by Satterfield ' and
Smith'*. By passing a pulse of tracer gas through either a
packed bed, pellet string reactor or a single particle and
then relating the shape of the elution curve to a model
describing the system, the transport coefficients of
interest may be obtained. Haynes
and Sarma i have studied dNusion in bi-disperse
catalysts using a packed bed and transforming the
governing equations into the Laplace time domain. The
desired transport coefficients were subsequently
extracted via moment analysis. However, the accurate
numerical determination of the second central moment is
difficult due to the large degree of tailing of the elution
curve. Thus, the accuracy of the moment analysis
technique for either packed beds or a pellet string reactor
16
is questionable. Gangwal et at. have utilized a Fourier
series to approximate the elution curve and have thus
avoided the problems inherent to moment analysis at the
expense of increased computational effort. After
normalizing the elution curve, the outlet concentration as
a function of time may be described by:
k t kw
bz
C(t,1 - Z °• sin T)
cOS (1)
) +
The experimental coefficients, O@ p and Qp„, are related
to the normalized elution curve by:
kit
C(t, L) sin )dt,
T 8
kit
S dt, (2)
T 8
Similarly, transport model coefficients, o,qp and Qqp,
may be related to the system parameters by
transforming the governing equations into the
frequency time domain (see Appendix). The desired
transport coefficients are then obtained by minimizing:
(3)
where:
Az —— T( a2 + bJ),- Q tan ' (@/az)
for either experimental or model coefficients. The major
advantage of Fourier analysis for this problem is that the
number of experimental and theoretical model points
which must be matched, ñ—total, is equal to the number
of
unknown transport coefficients. If the matching of the
experimental and theoretical elutions curves is attempted
in real time, many more points are required to
characterize the curves. The above analysis is
of the period, F, if Tim taken to be much
Due to the heterogeneous nature of coal, the single
particle technique, although simpler, is inappropriate
coal. The pellet string reactor (PSR) is a cylinder filled
with a large number of pieces of coal of approximately
equal diameter such that 1.1 < Dtube /D < 1.4. Scott et
at.' have shown that when the number of particles is of
the order of 50, the behaviour (i.e., axial dispersion,
pressure drop) of a packed bed is approached. Thus, the
PSR offers the advantage of averaging over many pieces
of a given coal while minimizing the total amount of
coal required.
EXPERIMENTAL
In principle, the PSR used in this work resembles a gas
chromatograph with gas supply/control, column,
temperature control, detection and data reduction zones.
A schematic diagram of the experiment is shown in
Figure
/. Sample injection of the methane tracer pulse into the
helium carrier is aocomplished using a Valco 8-port
valve equipped with a 1 cm' sampling loop. The reactor
tube (i.d.=0.457 cm) is coupled directly to the sampling
valve to minimize entrance effects. Tube length is varied
to keep the length/pellet diameter ratio constant. Gas
pressures are determined with a Senso-Metrics Inc. SP-
975-D strain gauge pressure transducer(0—350 KPa).
System operating pressure is maintained between 85 and
105 KPa. The reactor tube, detector and sampling valve
are all contained in a constant temperature air bath (303
K + 0.2).
The methane concentration in the gas stream leaving
Flomneter
Thermal
Pallet Tube
O-7 MPo
Air
Tr aducer
Figure I PSR schematic diagram
FUEL, 1984, Vol 63, February 257
Diffusion models for gas production from coal: D. M. Smith and F. L. Williams
the reactor tube is determined with a thermal
conductivity detector coupled directly to the tube exit. spheres is 0.694. Table 2 indicates tortuosity factors
The detection system consists of a Gow-Mac thermistor from several investigations of axial dispersion in both
detector and power supply/controller. The detector uses packed beds and PSRs, as well as the results of this
two 8K I) thermistors contained in a cell with internal work. The tortuosity factor for this work was 0.753. The
volume of value is somewhat above theoretical because the
0.025 cm' thus minimizing time lag due to cell mixing. particles are not spheres thus resulting in a packing
The power supply analog output, which is proportional configuration which is less dense than the theoretical
to CHP concentration, is digitized with a Keithley No. value.
192 multimeter (60 readings s ') and subsequently
transferred to a HP-85A microcomputer via an IEEE- PSR results/coal
488 instrument bus. The integrations required by With the applicability of the PSR/frequency analysis
Equation (2) for the evaluation of the experimental technique demonstrated for solid pellets, the
Fourier coefficients are performed with the HP-85A in investigation was extended to the coal/methane system.
real time using Simpson’s rule. Three coals are used in this investigation: an anthracite
With the experimental Fourier coefficients coal from Madrid, New Mexico; a bituminous coal from
determined, the desired diffusion parameters may be the Pittsburgh coal seam; and a subbituminous coal from
extracted by minimizing Equation (3). This nonlinear the Fruitland formation of northwest New Mexico. The
least squares minimization problem is solved using a tube length was varied for different samples to maintain
Levenberg—Marquart algorithm adapted by More et the optimum tube length/particle diameter ratio.
at.' '. Since this algorithm is weakly convergent to the Sample preparation before testing consists of several
global minimum, a range of initial guesses for the steps. First, the coal is ground to the desired mesh size
dNusion parameters are used to ensure that the global and particles which exhibit a regular geometry are
minimum is attained. removed. The size of these particles is determined
before drying at 378-383 K for 3 h. After drying, the
PSR resultsJsolid pellets samples are immediately loaded into the PSR tube and
the tube is reattached to the apparatus. Helium flows
Before undertaking the study of methane dNusion in through the tube at 323 K, overnight, to purge the
coal, the performance of the PSR was studied using solid system of all extraneous gases. After thermal
pellets. The use of solid pellets reduces the number of equilibrium is attained at the operating temperature of
unknown transport coefficients to two, the Peclet 303 K and the helium carrier flow rate is stabilized at its
number and axial dispersion coefficient, Dz. Thus, this desired value, experimental runs are conducted.
system represents a good starting point for analysing the
accuracy and reproducibility of both the experimental
apparatus and frequency analysis method. Since axial Table I Reproducibility of calculated transport coefficients for
dispersion in packed beds has been well studied CH4 dispersion in solid pellets
experimentally and theoretically' 7"’ 2', significant data
Velocity
are available for comparison. Run No. Peclet No. (cm s —’ L (cmm s—’)
The measurement of axial dispersion for methane 0.533
1 0.654 1.138
pulses in solid pellets was undertaken by packing the 2 0.643 1.143 0.544
pellet string reactor described previously with 303 3 0.658 1.145 0.533
plastic spheres of diameter 0.3063 + 0.0045 cm yielding 4 0.651 1.153 0.543
a particle/tube diameter ratio of 0.67. It should be noted 5 0.669 1.J 55 0.536
Mean 0.655 1.147 0.536
that previous investigators using frequency analysis standard devia- 0.009 0.006 0.006
assumed the input pulse to be a Dirac delta function and tion calculated 1.118
did not account for the actual pulse period. By
accounting for the actual pulse period (see Appendix,
Equation (A.8)), it was possible to operate at lower flow
rates and shorter
tube lengths. Pulse periods in this work are calculated Table 2 Tortuosity factors obtained from axial dispersion data for
packed beds and pellet string reactors
from the measured gas flow rate and known system
pressure. Investigator System
To demonstrate the reproducibility of the
experimental apparatus, runs No. 1—5 were conducted Theory 0.694
with the same operating conditions (Re= 0.427, CHP 19 Packed bed 0.73
tracer, period = 1000 s). The mean values (n= 1—10) of Airergon
the calculated transport coefficients are presented in 20 Packed bed 0.73
Table 1. Repro- ducibility between runs for the
apparatus used in this work is significantly better than 07 — He—Nd 0.753
that reported by previous investigators. This may be
PSR
attributed to the use of a fast response thermistor 17 He—N$ 0.64
detector coupled with digital signal processing as well as
automatic control of all aspects of PSR
N2- He
the experiment. 21 17 PSR 0.57
Previous investigators' 7" have shown that the He —
N2
limiting value of DL at low Reynolds number may be
This work PSR 0.692 0.753
related to the bimolecular diffusion coefficient by a He—CHP (303 K)
constant normally called the tortuosity factor, y. On
theoretical grounds, the value of y for an assemblage of 258 FUEL, 1984, Vol 63, February
^ See ref. 22
 Diffusion models for gas production from coal: D. M. Smith and F. L. Williams

After one experimental elution curve was obtained and
reduced to the required amplitude ratio, A., and phase
lag, $., the minimization software was tested. The
experimental and theoretical values of $ and A, for
this run are shown in Fof›fe 3. The good agreement
between the experimental elution curve and the model
is also presented in Figure 2, a reproduction of an
experimental elution curve. For comparison, the
theoretical elution curve using the ‘best-fit’ transport
coefficients as well as the theoretical curve for the
case when no diffusion into the coal is occurring are
presented. The technique is clearly sensitive to the
internal diffusion of methane in the pores of the coal.
For the coal/methane study, a total of 12
experimental runs were conducted. Since the
reproducibility of the experimental apparatus is
demonstrated in preceding sections, a large number of
runs were not deemed necessary. Table 4 is a
summary of the calculated mean
r,
CONCENTRATION
values of D„ D., d and A/(3D;) for the three coals
studied.
Although only three coals are studied, several qualitative
conclusions about the bi-disperse pore model and coal
rank may be ascertained. It has been shown
previously'' that anthracites possess smaller mean pore
diameters and have a greater fraction of micropore
volume as compared with lower rank coals. This face is
reflected by the smaller values of effective micropore
and macropore dNusivity for the Madrid coal
compared with the bituminous (Pittsburgh) and
subbituminous (Fruitland) coals. Also, the fraction of
gas contained in the micropores is greater for the
Madrid coal. Reflecting the general increase in pore size
with decreasing rank, both D, and D. are greater for
the Fruitland coal compared with the Pittsburgh sample.
These values for the transport parameters compare
well with the previous values’. For example, the average
values of D;, D, and A for six Fruitland coals were 2.6
x 10**, 2.8 x 10* and 4.0 x 10*6 s '. Differences
between the results of the previous work’ and this
technique may be partly attributed to moisture content as
this work used dry coals but the previous investigation
used coals saturated with water. It is planned to extend
the use of the PSR to wet coals to allow the direct
comparison with past work.

I
SUMMARY
0-3 - I“ A laboratory technique and attendant analysis scheme
has been developed for determining transport parameters
governing methane diffusion in coal. The technique is
shown to be sensitive to the transient diffusion of
methane within the pore structure of coal. Results from
three ranks of coal show expected trends when the
known variation of coal pore structure with rank is
TIME (sec) considered.
Figure 2 Typical PSR elution curve for methane difiusion in
coal. @. Experimental: . model (best fit): ———. model
(no internal diffusion) ACKNOWLEDGEMENT
The authors would like to thank the US Department
of Energy for financial support under Contract No.
DE-

rune 3 Comparison of experimental and theoretical A T and g n for coal run No.10

Amplitude ratio Phase lag

Frequency number Experimental Calculated Experimental Calculated
1 0.99767 0.99846 1.24705 1.24932
2 0.55144 0.99388 0.92471 0.92791
3 0.98275 0.98628 0.60388 0.60663
4 0.97224 0.97575 0.28367 0.28556
0.95953 0.96239 —0.03657 —0.03524
6 0.94420 0.94634 —0.35661 —0.35571
7 0.92646 0.92774 —0.67599 —0.67577
8 0.50674 0.90677 —0.95474 —0.99536
9 0.88515 0.se364 —1.31305 —1.31443
AO 0.86164 0.85854 1.51072 J.50870

Table 4 Bi-disperse pore model parameters as measured with the PSR

Sample Rank Ot x J0’ Is—’) A x 10’ Is—* I

Madrid Anthracite 1.0 1.2 3.7 J .23
Pittsburgh Bituminous 4.4 5.2 5.9 0.45
Fruitland Subbituminous 11.0 5.6 16.0 0.48
FUEL, t984. Vol 63. February 259
Diffusion models for gas production from coal: D. M. Smith and F. L. Williams
ACF1-81MC16015. Also, the help of Dr C. Moler and
the UNM Computer Science Department in 18 More, J. J., Garbow, B. S. and Hillstrom, K. E. Argonnc
National Laboratory, ANL-80-74, National Technical
implementing the MINPACK software is appreciated. Information Ser- vice, Springfield, VA, 1980
19 Edwards, M. F. and Richardson, J. F. Chem. Eng. Sci. 1968,
23, i09
20 Edwards, M. F. and Richardson, J. F. Can. J.Chem. £np.
NOMENCLATURE 1970,48,
A —— amplitude ratio 466
21 Urban, J. C. and Gomezplata, A. Can. J. Chem. £ng. 1969, 47,
o, b —— Fourier series coefficients 353
C —— concentration 22 Hirschfelder, J. 0., Curtiss, C. F. and Bird, R. B. 'Molecular
D —— effective diffusivity Theory of Gases and Liquids’, Wiley, 1954
0 —— axial dispersion coeEcient
F -— minimization function defined by Equation (3)
APPENDIX
H —— Henry’s law coefficient
'' Bi-disperse pore pulse tracer model
L= tube length To relate the experimental elution curve to the desired
Pe —— Peclet number, UR/Dz transport coeficients, a model describing mass transport
A = particle radius in the PSR is required. The three continuity equations,
S —— surface area written in dimensionless form, necessary for this model
t —— time are:
t, —— pulse period
T-— period Micropore
(A.1)
U -- velocity D, b
y = tortuosity factor Macropore +A
' "‘' (A.2)

PSR 2
•
Pe fi( fi( U
s = porosity
(= dimensionless axial position (A.3)
q = dimensionless radial position with initial and boundary conditions
8= dimensionless concentration
Q = phase lag
8;(0, y,)= (t, 0)=0 8;(r, 1)= 8,(r, y,)

6.(0, q,)= (t,0)=0 8 (t, 1)= 8,(r, {) (A.4)

ni=frequency
8 (0, {)= 8,(r, a)= 0 8,(0 fi t ñ t„ 0)= 1
Subscripts
a= macropore 8b('’ t, 0) = 0
b=tube 2155
i= micropore

REFERENCES 260 FUEL. 1984, Vol 63, February

1 Ruppel, T. C., Grein, C. T. and Bienstock, D. Fuel 1972, 51,
247
2 Kissell, F. N. US Bureau of Mines No. 7649, Pittsburgh, PA,
1972
3 Kuuskraa, V. A., Hammershaimb, E. C. and Doscher, T. M.
‘Proc. of the 2nd Methane Recovery from Coalbeds
Symposium’, 1979, pp. 27-44
4 Ancell, K. L., Lambert, S. L. and Johnson, F. S. ‘Proceedings
of the Ist Symposium on Unconventional Gas Recovery’,
1980, pp. 355—369
5 Thimons, E. D. and Kissell, F. N. Fuel 1973, 52, 274
6 Sevenster, P. G. Fuel 1959, 38, 403
7 Nandi, S. P. and Walker, P. L. Fuel 1964, 43, 385
8 Airey, E. M. Int. J. Rosk Mech. Mining Sci. 1968, 5, 475
9 Smith, D. M. and Williams, F. L. fuel 1984, 63, 251
10 Smith, D. M. and Williams, F. L. ‘Proceedings of the Uncon-
ventional Gas Recovery Symposium’, Society of Petroleum
Engineers, 1982, pp. 261—273
11 Gan, H., Nandi, S. P. and Walker, P. L. Fuel 1972, 51, 272
12 Smith, D. M. Ph.D. Thesis, University of New Mexico,
1982
13 Satterfield, C. N. ‘Mass Transfer in Heterogeneous Catalysis’,
MIT Press, Boston, MA, 1970
14 Smith, J. M. ’Chemical Kinetics’, McGraw-Hill, 2nd
Edition, 1970
15 Haynes, H. W. and Sarma, P. N. AfC8£ J. 1973, 19, 1043
16 Gangwal, S. K., Hudgins, R. R., Bryeon, A. W. and Silverston,
P.
L. Con. J. Chem. Eng. 1971, d9, 113
17 Scott, D. S., Lee, W. and Papa, J. Chem. Enp. Sci. 1974, 29,
Equations (A.1—4) may be transformed into the
frequency time domain using a finite Fourier where:
transform:
u sinh(a) +sin(a)
2 cosh(o) — cos(o) —
(cu)= 8(t) e“"°’ d7

(A.5)
Transforming A.1 and solving the resultant ordinary
differential equation (ODE), the concentration gradient
at the microscphere surface is obtained: Likewise, the concentration gradient at the coal particle
surface is given by:
(A.6)
(A.7)
Diffusion models for gas production from coal. D. M. Smith and F. L. Williams
where:
Y$ —— Pe ‹PYE -F

x, _ 422 cos tan '(Yr/+ )/2)

422
cos [tan“’(Y2/ Xy)/2§ Y6 sin [tan '(Yr/+ )/2)
d""**^"
sin [tan '(Y2/ 2)/2) 7' [sin(Y$)(I — cos(— !b))
X sinh(2A ) + Y$ sin(2Y ) — cos(Y ) sin( — mib)j
cosh(2X ) —cos(2Y ) d^ •J
sinh(2N ) — X sin(2Y Y7 = [cos(Y )(cos(— ‹ut b ) — 1)
) cosh(2X,) — cos(2Y ) — sin(Y ) sin(— nit ))
The concentration distribution within the PSR is If X — X, and Y$ - Y7 are evaluated with (= L/fi„ then
obtained by transforming Equation (A.3) and the theoretical amplitude ratio and phase lag are given
substituting by:
22
(A.7) into the resultant ODE. The solution is given by: (A.9)

Ag
(A.8) Qp = tan '(YQ/ 7) (A.10)
Thus, by matching Equations (A.9) and (A.10) to
Equation (3), the desired diffusion parameters (D;, D„
A) may be extracted.
FUEL, 1984, Vol 63, February 261