Professional Documents
Culture Documents
Determination of Gas-Diffusion and Inter PDF
Determination of Gas-Diffusion and Inter PDF
Determination of Gas-Diffusion and Inter PDF
General Formulation
Copyright © 2006 Society of Petroleum Engineers
Following Zhang et al. (2000), consider the schematic experimen-
This paper (SPE 84072) was first presented at the 2003 SPE Annual Technical Conference tal setup, shown in Fig. 1, operating at isothermal conditions. We
and Exhibition, Denver, 5–8 October, and revised for publication. Original manuscript re-
ceived for review 11 February 2004. Revised manuscript received 24 August 2005. Paper
consider a nonvolatile quiescent liquid brought into contact with a
peer approved 11 September 2005. high-pressure pure gas instantly at isothermal conditions. The gas
Note that Eq. 7 is different from that of Zhang et al. (2000) be-
cause they assumed a constant real-gas deviation factor value
Z⳱Zo at an initial po gas pressure.
An alternative expression to Eq. 6 for Q can be derived. Let
c(x,t) be the concentration (mass per unit volume) of the gas that
is dissolved in the liquid phase, and let co be its initial value. Then
Q can be expressed as
where L is the height of the liquid column which varies with time
when the gas/liquid solution swells by gas dissolution [i.e.,
L⳱L(t)]. The function Q(t) can be interpreted as the average gas
accumulation in the liquid-phase region, measured per unit area
of cross section.
冉 冊
Mgp
g = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) 1 ⭸
ZRT =− . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11)
⭸w T
where Mg denotes the molecular weight; p and T are the pressure
and temperature of the gas, respectively; R is the universal gas Hill and Lacey (1934) corrected the diffusion-coefficient value
constant; and Z⳱Z(p,T) represents the real-gas deviation factor, for the effect of liquid expansion by swelling caused by gas dis-
correlated empirically as a function of pressure and temperature. solution using the empirical relation
The volume V and mass m of the gas phase are given by
D = Do共1 − X兲2, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12)
V = AH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
where Do is the uncorrected apparent diffusivity, and D is the
m = gV, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3) diffusivity value corrected for expansion of volume during gas/
liquid solution swelling.
where A and H denote the cross-sectional area of the test tank and Riazi (1996) and Sachs (1997, 1998) expressed the diffusion
the length of the gas column, respectively. The thickness of the gas coefficient for nonideal mixtures as a function of the species con-
phase varies with time when the gas/liquid solution swells by gas tent in the liquid solution according to
dissolution [i.e., H⳱H(t) ]. If J denotes the mass flux of the gas
diffusing from the gas phase into the liquid phase at the gas/liquid
interface, the gas-phase mass balance is given by D = Do 1 + 冉 ⭸ ln
⭸ ln X 冊
, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (13)
dm
= −A J|x=0 , t ⬎ 0, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4) where denotes the fugacity coefficient of the dissolving gas in
dt
the solution.
where t is time. The initial gas mass is given by The equation of continuity for the liquid/gas solution undergo-
m = mo, t = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5) ing swelling caused by gas dissolution is given by
where Q denotes the cumulative mass of gas dissolved in the liquid ⭸c ⭸共uc兲 ⭸J
+ + = 0, 0 ⱕ x ⱕ L共t兲, t ⬎ 0 . . . . . . . . . . . . . (15)
phase per unit cross-sectional area of the gas/liquid interface. ⭸t ⭸x ⭸x
冋 册
Mathematical Model. The gas/liquid mixture is very dilute in
typical laboratory tests. Therefore, we assume an ideal solution MH po p
− = QD共tD兲. . . . . . . . . . . . . . . . . . . . . . . (39)
and omit the nonideal solution correction term appearing in Eq. 13. RTL共c* − co兲 Zo Z
Consequently, we neglect the swelling effect, assume an incom-
pressible gas/liquid solution, and take a constant diffusion coeffi- Next, Eq. 39 is applied at any conveniently selected reference
cient (i.e., D≅constant). Then, Eqs. 11 through 14 are no longer pressure value pr measured at a dimensionless time tDr to obtain
needed, and Eqs. 15 and 16 can be manipulated to eliminate J to
obtain
MH po pr
−
RTL共c* − co兲 Zo Zr 冋 册
= QD共tDr兲. . . . . . . . . . . . . . . . . . . . . . (40)
through 34 for short-time (semi-infinite) or Eq. 35 for long-time QDLT共tD兲 = 1 − Q1共1兲e−1tD, . . . . . . . . . . . . . . . . . . . . . . . . . . . . (50)
(finite-length) for the equilibrium gas transport in the gas/liquid where
solution with the results tested against typical experimental data.
Short-Time (Semi-Infinite) Solution (L→ⴥ). Civan (1997, 4 sin2 1
Q1共1兲 ≡ . . . . . . . . . . . . . . . . . . . . . . . . . . . (51)
1999, 2000a, 2000b) applied the analytical solutions for semi- 1关21 + sin共21兲兴
infinite regions given by Crank (1956) successfully for determi-
nation of the best estimates of the diffusion coefficients in a variety This is known as the long-time approximation. The parameter 1
of problems. These analytical solutions can be applied for the is the smallest positive root of the transcendental equation 1 tan
present case as follows: 1⳱kD. The upper bound 1⳱/2 corresponds to kD→⬁, which in
冋 冑册
turn corresponds to the case for the surface Dirichlet-type bound-
1 tD ary condition of Eq. 27.
QDST共tD兲 = exp共kD2 tD兲erfc共kD公tD兲 − 1 + 2kD . . . (43) It is interesting and useful to note that taking the natural loga-
kD
rithm of Eq. 50 leads to the result
As the mass-transfer Biot number kD→⬁, Eq. 43 simplifies to
ln关1 − QDLT共tD兲兴 = ln关Q1共1兲兴 − 21tD. . . . . . . . . . . . . . . . . . . . . (52)
QDST共tD兲 = 2 冑 tD
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (44) Thus, plotting the natural logarithm of [1-QD LT
(tD)] against tD
produces a straight-line curve. This suggests a method for extract-
This is the result that would have been obtained if the Dirichlet- ing the two unknown coefficients D and k associated with a given
type boundary condition (Eq. 27) had been imposed at the interface. set of pressure-decline data.
Strictly speaking, the length L cancels out of the problem for
Test-Data Analysis and
semi-infinite regions. Thus, these results are more appropriately
Interpretation Methodology
regarded as small-time approximations. These special solutions
can be used when the diffusion coefficient is sufficiently low and/ The objective is to interpret experimental data and infer from it the
or the test time is sufficiently short that the diffusing gas cannot values of the diffusion coefficient D and the interface-mass-
effectively reach the bottom of the test tank. The result by Eq. 43 transfer coefficient k. When a given set of data is plotted on one
can be obtained formally from the small-time inversion of the coordinate vs. another, the nature of the coordinates can make a
Laplace-transform analysis for the finite-length gas/liquid column difference in the insight that the plot provides.
(Civan and Rasmussen 2001, 2002, 2003). We wish to use the short-time approximation by Eq. 43 to-
The short-time solution (Eq. 43) itself has two time ranges gether with the long-time approximations, either in the form of Eq.
embedded within it that are of interest. For very small times, 50 or in its counterpart logarithmic form of Eq. 52, to analyze data
Eq. 43 can be expanded for small z ⳱ kD√tD to obtain generated over the whole time range in the form of QD vs. time t
冋 册
stemming from the pressure-decline relation (Eq. 42). The reason
1 2 4 z4 8 z6 this is possible is that, whereas the short-time approximation is
QDST共tD兲 ≅ z − z3 + − z5 + + . . . , z → 0 basically valid for short times but fails for very large times, and
kD 3公 2 15公 6
because the long-time approximation is basically valid for long
. . . . . . . . . . . . . . . . . . . . . . . . . . (45) times but fails for very small times, they share a common inter-
Thus, for finite kD, when QD is plotted against √tD, the initial slope mediate-time overlap region in which they are both valid (Civan
is always zero. and Rasmussen 2001, 2002, 2003). (See Figs. 2 through 5.) In QD
On the other hand, when tD is very large, the short-time ap- vs. √tD, this overlap region is asymptotic to, and nearly coincides
proximation by Eq. 43 approaches the asymptote with, the straight line given by Eq. 46. Plots of QD vs. √tD em-
冑
phasize the short time. On the other hand, plots of ln(1-QD) vs. tD
tD 1 [or log(1-QD) vs. tD] emphasize the long time. The long-time
QDST共tD兲 ≅ 2 − , kD公tD → ⬁. . . . . . . . . . . . . . . . . . (46) approximation (Eq. 52) is characterized by a straight line on the
kD
semilog plots.
In practice, the large-time behavior of the short-time approxima- An important factor to consider is that the experimental data
tion applies to some midrange of the overall time variation. Thus, are plotted as a function of the real time t and not the non-
when QD is plotted against √tD,we might expect some midrange dimensional time tD. Thus, in the theoretical formulas we must
portion of the curve to be nearly parallel to, or coincident with, the make the substitution
straight-line asymptote by Eq. 46.
Long-Time (Finite-Length) Solution. When the length L is t
tD = , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (53)
finite, a Fourier series analytical solution can be obtained (Walas C
C = L2 Ⲑ D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (54) 21
D 兲 = ln关Q1共1 兲兴 −
ln共1 − QLT t. . . . . . . . . . . . . . . . . . . . . . . . . (57)
Thus the theoretical formulas cannot be plotted against the experi- C
mental data unless C is specified arbitrarily or developed by some
systematic means. Therefore, the short-time and long-time ap- As a result, these functions depend on two parameters, C and 1.
proximations Eqs. 43, 50, and 52 are more appropriately expressed Now, suppose that the data are given at n discrete points des-
functionally as ignated by i going from 1 to n: QDi vs. ti. After plotting the data
vs. √t and on the semilog plots, pick a point im that separates the
D = QD
QST ST
冉 冊t
; . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (55)
C 1
short-time data from the long-time data (e.g., a point near the
crossover or the intersection of the short- and long-time curves),
and form the following squared-error functions:
兺冋 冉 冊册
n
ti 2 The value of kD is calculated by Eq. 48:
E LT共C, 1兲 = QDi − QDLT , . . . . . . . . . . . . . . . . . (59) kD = 1 tan 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (71)
i=im C 1
Finally, with kD and D now known, the interface-mass-transfer
E 共C, 1兲 = E ST共C, 1兲 + E LT共C, 1兲. . . . . . . . . . . . . . . . . . . . . (60) film coefficient k can be determined:
Minimizing the function E determines the best estimate values of D
the two parameters C and 1 for the least-squared error over the k= k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (72)
combined short- and long-time behaviors. L D
aST =
1
,b =
kD ST L
2
冑 D
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (62)
bLT =
4
bST. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (73)
o 共1 兲兴 = −log
aLT = −log关QLT 再 4 sin2共1兲
1 关21 + sin共21兲兴
冎 . . . . . . . . (66)
time. However, extrapolation of experimental data beyond the
measured range may introduce significant error. Nevertheless, this
problem can be alleviated readily either by applying Eq. 41 to use
D only the values measured during a diffusion experiment terminated
bLT = 0.434321 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (67) without waiting until equilibrium or Eq. 42 to use a directly mea-
L2 sured equilibrium pressure value when a diffusion test is continued
Eq. 66 allows 1, and thus kD by virtue of the relation 1 tan until equilibrium.
1⳱kD given by Eq. 48, to be determined. Consequently, the
mass-transfer Biot number kD is determined entirely by the y-axis Digitizing Reported Data. The numerical values of the experi-
intercept. The maximum value of a occurs for the limiting case mental data analyzed in this study were obtained by digitizing the
kD→⬁ and 1⳱/2. Thus, we have measurements reported by various studies in the form of plots.
Therefore, the data analyzed here might inherently involve some
amax = −log 冋册 8
2
= 0.0912 . . . . . . . . . . . . . . . . . . . . . . . . . . . . (68)
digitizing errors.
冋册
are noisy, the straight-line intercept for long time might not be
8 correctly located to meet this limit condition.
amax = −ln = 0.2099. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (69)
2
Long-Time Data Measurements. Errors made in obtaining cu-
With 1 now determined, the diffusion coefficient D can be solved mulative gas dissolution data are more significant at long times
for from Eq. 67: than for short times. Suppose that the error made in obtaining the
2
= 1 − Qoe− tD 1 − 冋 冉 Qo
2
e tD
−1 冊册 . . . . . . . . . . . . . . . (76)
This function shows the effect of error. For example, for the rep-
resentative values of Qo⳱0.7 and 2⳱2.5, zero error corresponds
to ⳱0, and ⳱±0.05 corresponds to ±5% error. Data points
measured to within ±5% error would lie between the upper and the
lower curves (Civan and Rasmussen 2002, 2003). Under these
circumstances, the scatter in the data increases as time increases.
Thus, every effort should be taken to obtain data with special
accuracy at long times.
Subscripts
D ⳱ dimensionless
g ⳱ gas