Professional Documents
Culture Documents
In-Situ Combustion Model: Intercomp Resource Development and Engineering Inc
In-Situ Combustion Model: Intercomp Resource Development and Engineering Inc
In-Situ Combustion Model: Intercomp Resource Development and Engineering Inc
Abstract
This paper describes a numerical model for The trend in this modeling has been toward more
simulating wet or dry, forward or reverse combustion rigorous treatment of the fluid flow and interphase
in one, two, or three dimensions. The formulation is mass transfer; inclusion of more components, more
considerably more general than any reported to date. comprehensive reaction kinetics, and stoichiometry;
The model allows any number and identities of and more implicit treatment of the finite difference
components. Any component may be distributed in model equations.
any or all of the four phases (water, oil, gas, and The purpose of this work was to extend the
solid or coke). generality of previous models while preserving or
The formulation allows any number of chemical reducing the associated computing-time requirement.
reactions. Any reaction may have any number of The most comprehensive or sophisticated com-
reactants, products, and stoichiometry, identified bustion models described to date appear to be those
through input data. The energy balance accounts for of Crookston et al. and Youngren. Therefore, we
heat loss and conduction, conversion, and radiation compare our model formulation and results here with
within the reservoir. those models.
The model uses no assumptions regarding degree A common objective of different investigators'
of oxygen consumption. The oxygen concentration is efforts in modeling in-situ combustion is develop-
calculated throughout the reservoir in accordance ment of more efficient formulations and methods of
with the calculated fluid flow pattern and reaction solution. This is especially important in the com-
kinetics. The model, therefore, simulates the effects bustion case because of the large number of com-
of oxygen bypassing caused by kinetic-limited ponents and equations involved. For a given number
combustion or conformance factors. of components and reactions, computing time per
We believe the implicit model formulation results grid block per time step will increase rapidly as the
in maximum efficiency (lowest computing cost), and formulation is rendered more implicit. However,
required computing times are reported in the paper. increasing implicitness tends to allow larger time
The paper includes comparisons of model results steps, which in turn reduces overall computing ex-
with reported laboratory adiabatic-tube test results. pense. To pursue the above objective, then, authors
In addition, the paper includes example field-scale, should present as completely as possible the details of
cases, with a sensitivity study showing effects on oil their formulations and the associated computing-
recovery of uncertainties in rock/fluid properties. time requirements.
The thermal model described here simulates wet or
Introduction dry, forward or reverse combustion in one, two, or
Recent papers by Ali,1 Crookston et al., 2 and three dimensions. The formulation allows any
Youngren 3 provide a comprehensive review of earlier number and identities of components and any
work in numerical modeling of the in-situ com- number of chemical reactions, with reactants,
bustion process. products, and stoichiometry specified through input
data. The transient, three-phase flow is represented
0197-7520/8010012-8394$00.25
by Darcy's law, incorporating temperature-
Copyright 1980 Society of Petroleum Engineers of AIME dependent relative permeabilities plus gravity,
DECEMBER 1980 533
viscous, and capillary forces. Vaporization/con- and one energy balance) to four equations by ex-
densation phenomena are handled with allowance for pressing the transmissibilities of Components 4 and 5
any component to be distributed in any or all of the explicitly in mole fractions. All transmissibilities are
four phases. The fourth (solid or coke) phase is explicit in temperature. Both models simulate
immobile and in part provides capability for treating convective and conductive heat flow within the
in-situ coal gasification problems. Oxygen and fuel reservoir.
consumption are calculated in accordance with the The thermal model formulation described here
specified reaction kinetics and calculated fluid flow allows any number and identities of components and
patterns. any number of chemical reactions. Each reaction
The following sections describe the model for- may have any number of reactants and products,
mulation both in general terms and in greater detail, identified through input data. Reaction rates are
with equations given. Several laboratory and field- functions of temperature and reactant concentrations
scale applications are presented together with a set of raised to powers specified through input data.
sensitivity runs indicating the relative importance of Without recourse to different "versions," the
different types of input data. The laboratory-scale formulation applies to single-component (water)
model results are compared with the associated geothermal problems plus two-, three-, and four-
observed or experimental behavior. component (e.g., including inert gas and/or
While the model formulation was developed to distillable hydrocarbons) steam flooding or hot-water
simulate the in-situ combustion case, it applies also flooding problems and N c-component in-situ
to hot-water or steamflooding and geothermal combustion problems.
problems. Although certain features were introduced If a solid (coke) is one of the components, the
in the formulation with in-situ coal gasification in model accounts for four phases, with S4 representing
mind, no applications to that type of problem have the solid-phase saturation. Thus, our saturation
been attempted to date. constraint is Sw + So + Sg + S4 = 1.0. Our ex-
perience indicates that coke (fuel) deposition can
General Comparative Model Description yield S4 values in the 0.04 to 0.08 range. This may
Crookston et al. 2 describe a combustion model using reduce formation permeability, and the formulation
the components oxygen, inert (noncondensable) gas, allows permeability to vary with the S4 value.
light hydrocarbon, heavy hydrocarbon, water, and The model uses no assumptions regarding degree
coke. Their model accounts for the four chemical of oxygen consumption. The oxygen concentration is
reactions representing oxidation of heavy oil, light calculated throughout the reservoir in accordance
oil, and coke and cracking of heavy oil. They treat with the calculated fluid flow pattern and reaction
three-phase fluid flow in one or two dimensions kinetics. Youngren's assumption that combustion
within the reservoir and use a two-dimensional heat rate equals oxygen flux precludes application of his
conduction calculation for heat flow in the over- formulation to reverse combustion problems and
burden. The additional variable introduced by the problems where an oxidation reaction is kinetic
coke mass balance in their formulation is nc Ibm mol rather than reactant limited. Also, as Youngren
coke/ cu ft of formation, and this mass is assumed to points out, his assumption prevents prediction of
occupy negligible volume - i.e., Sw + So + Sg = spontaneous ignition and his neglect of cracking or
1.0 in their formulation. Crookston et al. use the coke prevents prediction of fuellaydown.
mole-fraction constraints for oil and gas phases as In defense of Youngren's assumption regarding
primary equations and, thus, solve simultaneously a oxygen consumption, we must agree that most
total of nine primary equations (six mass balances, combustion problems probably are reactant limited.
one energy balance, and two mole-fraction con- More important, however, we feel the assumption is
straints) using Gaussian elimination. They use a simply unnecessary. That is, the assumption limits
mixture of implicit and explicit treatment for the applicability of a combustion model but yields
coefficients in interblock transmissibilities and little, if any, benefit in reduction of required
reaction terms. arithmetic or computing time.
Youngren 3 describes a three-dimensional com- Unlike both models just discussed, the model
bustion model that treats the five components - presented here is completely implicit. Our experience
water, oxygen, nonvolatile oil, and two arbitrary in steam flood modeling 4-6 has led us to the belief
volatile components - which can distribute between that total computing expense for thermal-model runs
the oil and gas phases. He neglects coke formation decreases with increasing degrees of implicitness.
and oxidation and treats the single chemical reaction The model described here is a set of mass- and
of heavy oil oxidation. Oxygen is assumed to be energy-balance equations and constraint equations
consumed instantly on contact with fuel; therefore, written for each of a number of grid blocks
combustion rate equals oxygen flux. The model representing a reservoir. The equations are written in
simulates three-phase flow, and a one-dimensional, finite difference form and include an energy balance
overburden heat conduction calculation is used for and one mass balance for each of N C components.
heat loss. Like Crookston et al., Youngren uses a The constraint equations express the requirements
mixture of implicit and explicit treatment but, unlike that fluid saturations must sum to unity and the mole
them, he solves only four primary equations fractions of components in each phase must sum to
simultaneously, using Gaussian elimination. He unity. Components are pure or pseudocompounds
reduces his six primary equations (five mass balances such as oxygen, nitrogen, carbon dioxide, water,
534 SOCIETY OF PETROLEUM ENGINEERS JOURNAL
"heavy oil component," "light oil component," and The model described here simulates wet or dry
solid component (coal or coke). forward combustion and reverse combustion in one,
For maximum efficiency, these equations must be two, or three dimensions. The model equations
divided into groups: (1) N c + 1 "primary" or describe mass transport by Darcy flow, incorporating
"flow" equations (the energy balance and N c- gravity, viscous, and capillary forces. The heat
component mass balances) and (2) N p + 1 con- transport includes mechanisms of convection,
straints equations (one saturation constraint and N p radiation, and conduction within the reservoir and
mole-fraction constraints). N p is the number of conductive heat loss to the overburden and un-
phases, three or four. Constraint equations are used derlying strata.
to eliminate Np + 1 of the total Nc + Np + 2 The model allows any component to be distributed
unknowns present in the N c + 1 primary equations. in any or all of four phases (water, oil, gas, and solid)
The result is a set of N c + 1 primary equations in as dictated by read-in pressure- and temperature-
N c + 1 unknowns, which is solved by the reduced dependent component K-values. Each phase density
bandwidth direct solution technique described by is a function of its composition, pressure, and
Price and Coats. 7 temperature. Gas-phase supercompressibility factor s
A major fraction of total model computing time is obtained from the Redlich-Kwong equation of
may be expended in this direct solution of the state. 8
primary equations. The arithmetic required in this The generality in respect to component
direct solution is proportional to the cube of the distribution among phases allows study of the effect
number of primary equations solved. 6 Thus, it is of CO 2 solution in oil with attendant reduction in oil
important to minimize this number by using all viscosity. Also, CO 2 or any component may be
constraint equations to eliminate unknowns. assigned some solubility in the liquid water phase. An
Crookston et al. retain two mole-fraction constraints interesting potential of this generality is examination
in their set of primary equations and, thus, use direct of emulsion effects. Specifying solubility of water in
solution (Gaussian elimination) to solve a set of nine the oil phase or of oil components in the water phase
primary equations for problems involving N c = six and adjusting input-component viscosity data can
components. The formulation described here ac- yield emulsion-type viscosities for the oil or water
tually results in direct solution of a set of only six phase.
primary equations for the case of six components if The formulation achieves this generality of
one of the components is coke. This set of N c rather component distribution by assigning a user-specified
than N c + 1 primary equations is attained through "master" phase to each component with a resulting
use of the coke mass balance as a constraint single set of master mole fractions (Xi J. User-
equation, which in turn is possible because the coke is specified pressure- and temperature-dependent K-
immobile. This is discussed in detail in the values then give the mole fraction of each component
mathematical model description. i in phase j as x Ii = K vijXi . This treatment simplifies
This discussion suggests that Youngren's approach model descriptIOn, coding logic, and variable sub-
of solving only four primary equations, regardless of stitution logic since it avoids the need to deal with
the number of components, may result in less four sets of phase mole fractions.
computing expense than the implicit formulation The inclusion of a fourth (solid) phase in the
described here. His formulation certainly will, or formulation represents a step toward the capability
should, require less arithmetic on a grid-block/time- to simulate in-situ coal gasification. In this case, the
step basis. However, we must consider the penalty initialized Phase 4 saturation would be large (e.g.,
associated with the method by which he attains a 0.8), representing the coal itself. The formulation
primary set of only four equations. The method treats relative permeabilities as functions of nor-
involves use of explicit mole fractions x4' xs, ... , xN. malized saturations (e.g., Sw = Sw!1- S4) to
in the transmissibilities of the Component 4, 5 .. ~ represent more reasonably the case where S4 is large
N c mass balances. A stability analysis of and highly variable. The model reaction kinetics
multi component , single-phase flow shows that this require alteration in form for the gasification
explicit treatment give a time-step limitation, problem since coal gasification reactions generally
I:.t < Sg !:.x/v gx ' reach an equilibrium between reactants and
products.
where !:.x is the grid block dimension in feet and Vgx
is gas-phase interstitial velocity in feet per day. We Mathematical Description of Model
can substitutel:.y/v gy or !:.z/v z for !:.x/v gx • In The model consists of N equations where N = N c +
steam flood or combustion probfems, Vg can reach Np + 2 and N p denotes the number of phases
locally large values (e.g., 10 to 20 ft/D or more), even present. N p = 4 if solid coke is present in addition to
with low Sg values. We feel this stability limit on the water, oil, and gas phases. N p = 3 if no solid
time-step size will cause in many cases a sufficiently phase exists or appears. The model equations are
increased number of required time steps to more than
offset the reduced arithmetic per time step. However, Nc Component mass balance equations
for some sufficiently large number of components, 1 Energy balance equation
the implicit formulation requiring direct solution of 1 Saturation constraint
N c + 1 primary equations will require more total Np Mole fraction constraints
computing time than Youngren's formulation. Nc + Np + 2 Total equations.
DECEMBER 1980 535
The saturation constraint expresses the fact that the through 4 are oSI' oS2' ... , oSN , oX I , oX2 , ... ,
Np phase saturation must sum to unity. The Np oXN ' op, oT. All terms in Eqs. (through 4 can be
mole-fraction constraints express the requirement exprgssed in terms of these unknowns.
that the sum of all components' mole fractions in
each phase must sum to unity.
Treatment of Accumulation (Left-Side) Terms
We refer to phases by the index or subscript J, with
J = 1 for liquid water, J = 2 for oil, J = 3 for gas, The left side of the component mass balance can be
and J = 4 for the solid phase. Components are expressed as
denoted by index or subscript I, with V Np
I Component
- t( E
tJ.t j= I
pjSjKvuXI )
2 Water V Np f
2
3
Heavy-oil component
Gaseous component
= tJ.t [(¢ j'"f pjSjKvuXI )
l
4 Coke if a solid component is present Np V Np 0
4 Gaseous or oil component if solid com-
ponent is not present
- (¢ jE= I
pjSjKvuXI ) ]
n
+-
tJ.t
E [oP-m
m =I
5 - N c Gaseous or oil components. Np f
oPNp OSNp =
oPNp + I = oX I
oP Np +2 oX2 =
J:
=
+Pl k N -a
E [ E
3 (P- )
kr KuUXj
The preceding sections describe the expansion of all
terms in Eqs. 1 through 4 in linearized expansions
m=1 ap m 1=1 11 I consisting of coefficients times iterate changes aPm'
e m = 1, 2, ... , N, where N = N c + N p + 2. Only
. (p-PWb)Lap m , ............... (22) the first N c + 1 of these N equations possess right-
side flow terms that introduce unknowns {aP m} of
where Pwbk is held constant in the differentiation. neighboring grid blocks. These N c + 1 equations are
The enthalpy production rate in Eq. 2 from grid designated "primary" equations. The remaining N p
block k is + 1 equations are constraint equations, and they are
QHK=PI k [ t
1= I
(~kr)HI(p-pWb)r
11 I k
used to eliminate N p + 1 unknowns from the
primary equations. This elimination is possible
because the constraint equations for a given grid
3 block involve only unknowns {aP m} of that grid
E (~k) block.
1= I 11 r I For that moment, we neglect the interblock flow
terms in Eqs. 1 through 4 and consider the ac-
............. (23) cumulation and all other terms. After the expansion
described previously, the N equations can be written
Heat-Loss Treatment as
The implicit calculation of heat loss is calculated as N
described in detail in Ref. 4. The QHL term in Eq. 2 E CijaP I = bi; i = 1,2, ... , N, ...... (27)
is replaced by a term of type <XI - <X2aT. 1=1
where we locally use i as the equation number and j
Functional Dependency
as the column or unknown index. The equations for i
The fluid and rock properties are dependent on the = 1, 2, ... , N c are component mass balances. The
system variables: equation for i = N c + 1 is the energy balance. The
P1,111'1' I f (p,T,X) equation for i = N c + 2 is the saturation constraint.
The equations for i = Nc + 3, Nc + 4, ... , N are
krI f (S, T) the mole-fraction constraints. The coefficients Cij
f
"
k f
(Sg)
(S4)
and b i terms are evaluated at latest iterate (0 values
of the unknowns {p. }.
The last N p of t6ese equations are mole-fraction
Mf f (T) constraints and coefficients Cij are all zero for j = 1,
538 SOCIETY OF PETROLEUM ENGINEERS JOURNAL
2, ... , N p , since unknowns oP I through oP Np are - f+ I
OP) =
- f
oP) + oP) , ................. . (30)
oSI through OSN and the mole-fraction constraints
do not involve fl~id saturations. Coefficients Cij for where a damping factor may be applied to oP) in Eq.
j = Np + 1, Np + 2, ... , N are generally nonzero. 30 if one or more of the iterate changes OP) are
We diagonalize this submatrix of N p equations by excessive.
Gaussian elimination so that the last N p (mole- C, T, and R in Eq. 29 are re-evaluated using the
fraction constraint) equation's matrix is reduced to latest iterate values of {p) J, and Eq. 29 is solved
again. These iterations are terminated when all the
j= Np +1 N p +2 ... 2Np 2Np +1 following are satisfied.
i=N c +3 1 0 0
max
i=Nc+ 4 0 0 j= 1,Np loS) I :::; TOLS (0.01) ,
max
i=1,Nc 10Xi I :::; TOLX (0.01) ,
and
j = ... N
max loT I :::; TOLT (1.0) , ..... (31)
i=Nc +3 v' OPNp + 1 v'
where "max" denotes maximum over all grid blocks
i=Nc +4 v' OP Np + 2 v'
and default closure tolerance values are given in
parentheses.
Variable Substitution
i=NC +Np +2 v' oPN v' ,
The N p + 1 constraint equati~ns. are used to
................................ (28) eliminate N p + 1 unknowns, resultmg m a set of N c
where the check marks indicate nonzero coefficients.
+ 1 (mass- and energy-balance) primary equations in
Nc + 1 remaining unknowns. In the general case,
These diagonalized N p mole-fraction constraints are
where all four saturations are nonzero, the
used to eliminate Columns (unknowns) j = N p + 1, eliminated unknowns are oX I , oX2 , oX3 , oX4 , and
N p + 2, ... , 2Np in the first N C + 1 equations of
OS4' leaving oSI' oS2' oS3' oX5 , ••• , OXNc' op, and
the equation set (Eq. 27).
The saturation constraint (Eq. 24) then is used to
o Tas the N C + 1 "primary" unknown~. .
If any saturation S) is :ero f?r some Iteratl?n, then
eliminate Column N p from the first N C + 1 no mole-fraction constramt eXists for phase J and we
equations of Eq. 27. This now leaves Nc + 1
ostensibly have on: less unknown (o~) and one less
equations (which involve right-side interblock flow
equation. In actuality, however, we still have N C + 1
terms) and Nc + 1 unknowns. We started with Nc
primary equations in N C + 1 unknowns, with one
+ N p + 2 unknowns, but we eliminated N p less constraint equation counterbalanced by one less
unknowns (mole fractions) using the N p mole-
eliminated unknown. That is, oX) is not eliminated
fraction constraints and eliminated one unknown
and oX· takes the place of oS) in the N c + 1
(saturation) using the saturation constraint. .
unknow'ns. The oS) is known a priori as - S),,!, An
The coefficients in the reduced mole-fractlOn-
example of this substitution of oX) for oS) IS the
constraint matrix (Eq. 28) are saved or stored for two
onset of the superheated water state, where SI
reasons: (1) they are needed to eliminate the men-
becomes zero but oX I still requires calculation since
tioned columns from the expansions on the in-
the gas phase will contain a water mole fraction of
terblock flow terms and (2) they are needed after
solution of the N C + 1 equations to compute the K u13 X I · . . ..
One important exceptIOn to thiS deSCrIptIOn of N c
eliminated unknowns. Similarly, oSN ' which was
eliminated, is calculated after solution bf the N C + 1
+ 1 primary equations in N C + 1 primary variables
is introduced by the immobility of coke. If coke is
equations.
one of the N C components, we can reduce the system
The N C + 1 primary equations can be written in
of N equations to only N C rather than N c + 1
matrix form as primary equations. This is important because the
Cr: = ~ + Ll(T~) , ................... (29)
computing time required for simultaneous, direct
solution of the primary equations (1) is proportional
to the cube of the number of primary equations and
where C and Tare (Nc + 1) x (Nc + 1) matrices (2) can constitute 80llJo or more of total computing
and Rand Pare N C + 1 column vectors. These time. 6
equations are solved by the reduced bandwidth di~ect If coke is one component, then the primary
solution technique. 7 Eqs. 28 and the saturatIOn equations (involving right-side flow terms) are the
constraint then are used to calculate the eliminated energy equation and the N C - 1 mass -balances for
unknowns. The cumulative changes over the time the N C - 1 mobile components. The coke mass-
step then are augmented as balance equation becomes a constraint equation. The
DECEMBER 1980 539
mole-fraction constraint for Solid Phase 4 is used to for liquid phases J = 1 or 2, which is the
eliminate OX4 [mole fraction of Component 4 (coke) representation used by Crookston et al. 2
in the solid phase] from all N equations and then the All model runs were performed using midpoint
resulting coke balance is used to eliminate OS4 from convective weighting (8 = 0.5 in Eq. 12).
the primary equations. The saturation constraint is
used to eliminate oSjrnax from the primary equations, Smith and Perkins Experiment
where jmax is less than 4 and is the largest value of j Smith and Perkins 9 conducted wet, forward com-
for which SJis positive. bustion experiments in a vertical, adiabatic com-
bustion tube. They injected a constant-ratio
Miscellaneous
air/water stream into a 700-md, 120°F vertical tube
Model initialization requires specification of initial- 5.75 ft long after preheating the top to 600°F. They
phase mole fractions, temperature, gas/oil and/or reported (1) measured temperature profiles at 2, 4,
water / oil contacts, and pressure at a specified subsea and 6 hours, (2) percent unused oxygen in the
datum. The model initializes pressure either by produced gas phase vs. cumulative air injection, and
capillary-gravitational equilibrium or by setting (3) amount of oil burned. They also described a
pressure to a constant, at the user's option. If a gas numerical model and compared their calculated with
phase is initially present, the model automatically their experimental results.
will recalculate the mole fraction of a specified Smith and Perkins reported two experiments - a
"swing" component to attain a sum to unity of gas- low-pressure test using Crude A and a high-pressure
phase mole fractions. This swing component test using Crude B. Results discussed here are for the
generally is specified as N2 or a N2 + COx 8I5-psia low-pressure run. Crude A is very light with
pseudocomponent. The reason for this internal API gravity and viscosity of 36° and 2.8 cp,
resetting of a gas-phase mole fraction is the difficulty respectively. They presented a temperature profile at
in calculating by hand the initial mole fractions of time zero corresponding to the heating at the tube top
distributed components so that the gas-phase mole- price prior to air injection. They further stated that
fraction sum will exactly equal unity. the air/water injection-stream temperature varied
The material balance at each time step for each from an initially high value (due to preheating of the
component is calculated as upper tube apparatus) to a later value of
1 + mass in place - (initial mass in place (presumably) 120°F. We simply initialized the first
20070 of tube length to 500°F and the remaining 80070
+ cumulative injection - cumulative production to 120°F and injected at a constant 120°F.
Smith and Perkins reported an initial period of air
+ net production by reactions) injection with no water. This air injection was
+ max (cumulative injection, cumulative continued until they noticed vigorous combustion
production) and no oxygen in the effluent gas. They did not
report the time period of this dry air injection. We
+ max (reaction production, reaction con- used 28.8 minutes. The subsequent water/air ratio
sumption) . was constant at 0.18 Ibm/ scf.
Smith and Perkins stated their insulation was
The energy balance is calculated in the same unusually permeable and heat convected outside the
manner with "energy" substituted for "mass." tube from the burning front upward to the trailing
Values of 1.0000 for these balances simply indicate edge. They accounted for this external convection in
that the finite difference equations are solved ac- their numerical model and reported better trailing-
curately. edge agreement than we obtained.
Automatic time-step control is used with time-step We changed Smith and Perkins' six-component oil
size controlled by the previous step size and the analysis to three components and treated the six-
previous step changes (maximum over grid) in component system given in Table 1. We calculated K-
saturation pressure, temperature, and mole fraction. values for each of the original six oil components
from In P v = A - B/ T, with A and B determined
Applications from boiling-point and critical data given in Ref. 10.
In all the model runs described below (S4)rnax and Viscosity coefficients were obtained for the six
heats of vaporization for volatile oil components are components from the same reference. K-value and
zero. Our experience indicates a negligible effect of viscosity parameters for the three pseudo oil com-
heat of vaporization. No temperature dependence of ponents then were calculated by simple mole-fraction
relative permeability or capillary pressure is present weighted averages.
in any of the problems. Units missing from problem Smith and Perkins represented the oil oxidation by
data listed in tables are given in the Nomenclature. B
Partial pressure is used in lieu of concentration for d0 2 -- -cJ
oxygen in all oxidation reactions. dT = A OP0 2 e T , . . . . . . . . . . . . . . . (32)
While the ASTM form for oil-component viscosity
(Eq. A-20) currently is programmed in the model, the where Po2 is partial of oxygen; A o ' B, and C were
model runs described here used the form determine d from experimental data; and f is the
fraction of oil oxidized. We used a reaction
II
rJJ -- ae b / T , proportional to
540 SOCIETY OF PETROLEUM ENGINEERS JOURNAL
~ EXPERIMENTAL DATA
- EXPERIMENTAL DATA
o 0 RUN \,10 BLOCKS
o 0 RUN 1,10 BLOCK:',
- - - RUN 2,20 BLOCKS
--- RUN2,20BLOCKS
6. 6. RUN 3,30 BLOCKS
6 o RUN3,30BLOCKS
600
0,[j.[j
400 400
T ,oF T, OF
200 200
°
0'------''------'--'---'---'---'---'---'---'-----'
o 1.0
x/L x/L
Fig. 1 - Calculated and observed 2·hour temperature Fig. 2 - Calculated and observed 4·hour temperature
profile - Smith·Perkins experiment. profile - Smith·Perkins experiment.
~
.
.
.
L
c
-
.
.
o~ 0
6/ 0 0
0/ 0
/
:t 0
/0
'Y
/ 0
¥
400 /0
/
/~
T,OF ?
,6
200
0 /"
0
TIME,HOURS
100
"if> ;- '" 0
z",
w<t
"'"'
80
"', I
I 0 So
"0
Xw 60
0
°u
... =>
"I
0
ZO "I
w~
Uo.
a:
WZ
0.-
40
20 "
,
" I 0
o
"" " j,
0
0 20
"40
0
60 80
AIR INJECTED,SCF xI l
Fig. 4 - Calculated and observed oxygen breakout times Fig. 6 - Calculated oil saturation profile at 6 hours -
- Smith·Perkins experiment. Smith·Perkins experiment.
B
T
, ................ (33) temperature profiles at 2, 4, and 6 hours, respec-
tively. We attribute the discrepancy along the trailing
where Xi is mole fraction of oil component i in the edge to the external heat convection. These figures
oil phase. This second-order dependence on oil indicate (1) reasonably good agreement between
concentration partially compensates for our observed and calculated (20 grid blocks) profiles at
neglecting Cj. The -value of A ~ was chosen so that all three times, (2) virtually identical calculated
the rates of Eqs. 32 and 33, expressed in consistent results for 20 and 30 grid blocks at each time, and (3)
units, are equal at initial oil-component con- significant but not extreme numerical dispersion in
centrations. the results calculated using 10 blocks.
Reaction stoichiometry and other data listed in Fig. 4 shows calculated and observed effluent
Table 1 were obtained from Ref. 9. Since that oxygen content vs. time. Calculated results using 20
reference gives no capillary-pressure or relative- and 30 grid blocks agree quite well with the observed
permeability data, we used zero Pc and analytical oxygen breakout time, while the results using 10
relative permeability curves with exponents of 2.0. blocks show effects of numerical dispersion.
Our one-dimensional model Runs 1, 2, and 3 were Fig. 5 shows a moderate effect of numerical
made using the data of Table 1 and using 10, 20, and dispersion on calculated oil recovery vs. time for the
30 grid blocks, respectively. Figs. 1, 2, and 3 show case of 10 grid blocks. Smith and Perkins reported an
the agreement between calculated and observed estimated figure of 8.8 wtOJo of the original oil in
542 SOCIETY OF PETROLEUM ENGINEERS JOURNAL
place burned. The calculated figure for the 30-block We feel the data presented by the authors are not
case was 10.38 wt07Q. sufficiently complete to warrant a serious matching
Fig. 6 shows the calculated oil saturation profile at effort. However, we have made a number of model
6 hours for the 10- and 30-,block runs. The sharp runs seeking qualitative agreement with the ex-
jump in oil saturation occurs at the leading edge of perimental temperature profiles reported.
the sharp temperature profile. A sharp jump from We selected Parrish and Craig's 2,900-cp Test 2 for
essentially zero to 0.80 to 0.90 in mole fraction of matching. Five components were represented -
light oil also occurs at this temperature and oil water, heavy (original) oil, N2 + COx' light
saturation" front." (cracked) oil, and oxygen. The two chemical reac-
A favorable aspect of the combustion process tions were oxidation and cracking of heavy oi!. Coke
frequently emphasized in the literature is the in-situ formation and oxidation were neglected.
upgrading of a heavy oil by cracking. Cracking was Their reported Test 2 experimental conditions are
presumed negligible in this experiment due to the given in the first section of Table 2. Injection rates,
relatively low temperature and light oil. However, an time period of initial dry air injection, relative
"upgrading" type of effect is still present through permeability, and kinetic and stoichiometric data
distillation or preferential vaporization and recovery were not reported. Table 2 gives the values used for
of lighter ends. For example, the calculated total oil these data. The authors reported experimental
recovery on a stock-tank-barrel basis was 88.69070. temperature profiles at three times for a "typical"
However, the calculated percentages recoveries for COFCAW test.
C 6 + C 8 + C 12 , C 20 , and C 32 + C 47 were 99.8, 79, We present computed results here for two runs.
and 69070, respectively. The runs differed in respect to the following data.
The reaction-rate e - Cj factor used by Smith and
Perkins resulted in their numerical model's Heat of
calculated reaction rates declining very rapidly with Reaction, Rate
increasing fraction of oil burned in the range off = 0 Oxidation Constraint,
to 0.1. Our second-order reaction gives a much less Run Reaction Cracking
severe decline in reaction rate with increasing Number (Btu/Ibm oil) Reaction
fraction of oil burned. They determined their large 4 16,000 106
rate decrease (large C) from combustion tests in an 5 20,000 0.25 X 10 6
isothermal tube with no water and an initial oil
saturation of 0.15. We feel that that type of test We heated the first 10070 of tube length to 500°F at
might give a misleading e- Cf factor due to the time zero, then injected 20 scflhr of air at 130°F for
wetting, low-value oil saturation occupying the tight 28 hours. Maximum injection pressure was 900 psia,
or small, relatively inaccessible portion of the pore and producing back pressure was 500 psia. The in-
space. This might lead to strongly diffusion- jection well went on injectivity during the period of
controlled reaction rates. In the actual combustion about 17 to 58 hours due to banking of cold (2,900-
tube tests, water saturation of 20 to 50070 might force cp) oil ahead of the combustion front. Starting at 28
the oil into much more accessible (to air) portions of hours and continuing to 96 hours, injection rates
the pore space, resulting in a decrease of diffusion were 0.19713 STn [cold-water equivalent (CWE) per
control. day] of water and 480 sefiD of air - an air/water
All model runs were performed with midpoint ratio of 2,435 scflbb!. The authors stated their tests
convective weighting (8 = 0.5 in Eq. 12). Use of required 30 to 100 hours. Our 4-day run is 96 hours.
upstream convective weighting resulted in a Figs. 7 and 8 indicate grid sensitivity through
significantly greater discrepancy between calculated comparison of Run 4 with 10 and 20 grid blocks. The
results for 10, 20, and 30 grid blocks. 20-block run calculated a lower injectivity due to
Computations for all runs were performed for 18 sharper oil-bank definition; therefore, the curve of
hours or 0.667 days. Model computing efficiency for cumulative air injected vs. time for the 1O-block run
the three runs can be summarized as follows. lies somewhat above that for the 20-block run.
Fig. 7 shows peak temperature in each grid block
Computing vs. grid block position. The 10- and 20-block runs
Number Number Number Time, agree well, with the exception of the blocks in the 0.1
of Grid of Time of CDC 6600 to 0.3 X I L interval, where the 20-block-run peak
Run Blocks Steps Iterations (seconds) temperature is less due to lower injection rate during
1 10 42 160 83 the corresponding time period. The decline in peak
2 20 53 250 251 temperature with distance traveled reflects a gradual
3 30 62 330 500 "extinction" of the dry combustion zone upstream
of the steam plateau.
Parrish and Craig Data Fig. 8 shows reasonably good agreement between
Parrish and Craig II reported results from wet, oil recovery curves computed with 10 and 20 grid
forward combustion tests in a vertical adiabatic tube blocks.
with length and diameter equal to 12 and 1 ft, The effect of a 25070 increase in heat of reaction
respectively. Original oil viscosity for their 11 tests from Run 4 to Run 5 was one of increased peak
reported varies from 2.0 to 2,900 cp at 100°F. temperature and sustaining of the dry combustion
Pressure level was 500 to 600 psia. zone throughout the duration of the run. That is,
DECEMBER 1980 543
- 20 BLOCKS
900 RUN 4 H = (6,000 BTU/LB Oil
IOIP : 341 STB _---:,----,,.-----,,~~-
600
500 o
.2 .3 4 .5 .6 .7 .8 .9 10
CUMULATIVE AIR INJECTION, MCF
X/L
Fig. 7 - Calculated peak temperature vs. distance along Fig. 8 - Calculated oil recovery vs. cumulative air injection
tube - Parrish and Craig COFCAW experiment. - Parrish and Craig COFCAW experiment.
Molecular Cpl
Component / Weight Pc Tc(OR) Cj (31 (Btullbm)
1. H 2O 18 3,206 1,165.4
2. Heavy oil 300 50.7 1,598 10- 5 0.38 x 10- 3 0.5
3. N2 +CO x 32 750 350 0.226
4. Light oil 170.3 264.6 1,184.9 10- 5 0,38x 10- 3 0,5
5. O2 32 730 277.9 0.24
combustion was continuing (So = 0.15) in the 10th Fig. 9 - Calculated and experimental COFCAW tem-
block at the end of the run. Reported and calculated perature profiles.
air/recovered-oil ratios are 6,020 and 5,166 scflbbl,
respectively.
On a volumetric stock-tank-barrel basis, 24070 of
the oil recovery calculated in Run 5 was Light respects, we will not repeat the extended discussion
Component 4. One hundred percent of the light oil and figures.
formed by cracking was produced, since final oil Table 4 summarizes the agreement between the two
saturation was zero and we included no oxidation sets of calculated results. Oil recovery is volumetric
reaction for the light oil. oil recovery expressed as a percentage of initial oil in
Run 5 used 10 grid blocks and required 108 CDC place. Stock-tank oil densities are 45.313 and 23.633
6600 seconds, 61 time steps, and 302 total iterations Ibm/cu ft for heavy oil and light oil, respectively, at
for the 4 days of real time. 72 OF and 14.7 psia. Initial water and oil in place are
5,168 and 12,590 STB, respectively. The light oil
Crookston, Culham, and Chen Problem recovery is expressed as volumetric percentage of
total oil recovery. Heat loss (to overburden) is ex-
Crookston et al. Z simulated oxygen injection into a pressed as a percentage of total heat of reactions.
one-dimensional, 200°F reservoir 164 ft long, 115 ft Combustion-front velocity is calculated as
wide, and 21 ft thick. Reaction and fluid-property 0.7(164)/(19 tz), where ti is time in days of the
data were estimated from values given in the temperature peak in grid block i and 164 ft is the
literature. Their four reactions represented oxidation system length. Tpavg is average peak temperature of
of light and heavy oil components, cracking of heavy Cells 2 through 9, inclusive, peak temperatures.
oil, and oxidation of coke. Water recovery is expressed as a percentage of initial
Table 3 lists our model input data for this problem. water in place. Heat of reaction is total heat of all
Our data are identical to those of Crookston et al. reactions. Computing time is in CDC 6600 CPU
with the following exceptions. Our oil relative seconds.
permeability curve is only similar to theirs. They llsed Table 4 shows fairly good agreement between the
a multiplicative factor of 1 - (nc/ncmaJ5 in their Crookston et al. results and our Run 6 results. Both
cracking reaction-rate expression, with ncmax = we and they calculated a steam zone or plateau
0.07692 Ibm mol/cu ft of formation. We used no throughout the system by the time the first cell
maximum equivalent (S4)max' since our calculated peaked in temperature. The reason for this very rapid
coke laydown did not exceed their n cmax ' steam-zone development is the passage of unreacted
As given in Table 3, the problem includes the six oxygen throughout the system in the first 10 to 20
components water, heavy oil, Nz-CO x ' coke, days prior to first-block spontaneous ignition. This
oxygen, and light oil. Thus, our simulation required oxygen reacts sufficiently throughout the system to
simultaneous solution of six primary equations - raise temperature throughout to about 300°F.
one energy balance plus five component mass Energy balance and component material balances
balances - with the immobile coke mass balance for Run 6 were excellent, ranging only from 0.9999 to
treated as a constraint equation. 1.0001.
Crookston et al. used propane and dodecane Crookston et al. reported a computing time for
properties for light and heavy oils, respectively, with this problem of 1,071 seconds on an IBM 370/168
an increased viscosity for dodecane. At 200°F, the computer. Run 6 required 242 seconds on a CDC
liquid-component viscosities of light and heavy oil 6600 computer. CDC 6600 computer times are about
are 0.089 and 139 cp, respectively. 2.2 times larger than IBM 370/168 times for our
Crookston et al. described their calculated results steam flood model. 6 Much of the difference between
in some detail both verbally and graphically. Since the computer time reported by Crookston et al. and
our results agree very closely with theirs in all that reported here reflects the fact that their for-
DECEMBER 1980 545
TABLE 3 - DATA FOR CROOKSTON ET AL. ONE·DIMENSIONAL FIELD·SCALE PROBLEM
K-Value
Reservoir length, ft 164
Reservoir width, ft 115 K
v23
= (184,900/p)e- S ,7391(T-1S7.1)
Reservoir thickness, ft 21 Kv63 = (130,700/p)e- 3 ,370/(T-45.29)
Permeability, md 4,000
Porosity 0.38 Reaction Data
Rock heat capacity, Btu/cu ft rock· ° F 35
Thermal conductivity, Btu/ft·D· ° F 1.6 Ci denotes component i
Overburden thermal conductivity, Btu/ft·D· ° F 1.6
Overburden heat capacity, Btu/cu ft overburden·oF 35
Ti , of 200 V~2 = 0.0235 cu ftllbm
Pi, psia 65 V~2 = 0.05 cu ftllbm
Initial water, oil, gas saturation =
0.2,0.5,0.3
~4 = 0.0125 cu ftllbm
Initial mole fraction X1 = 1.0, X2 = 1.0, X3 = 0.806, X4 = X5 = Xs = 0 Viscosity Data
Number of grid blocks (one·dimensional)
Number of components
10
6
J1.22 = 0.0003624 e 8 ,4851T
Number of reactions 4 J1.23 = 0.3926 10- 5 T1. 102
X
J1.33 =
0.0002127 702 ro·
Molecular J1.53 =
0.0002196 721 ro·
Component / Weight Pc Tc(OR) Ci (31 CP1 (Btullbm) J1.S2 = 0.02083 e959.SIT
J1.S3 = 0.2166 X 10- 4
1. H 2 O 18 3,206.2 1,165.4
2. Heavy oil 170 264.6 1,184.9 10- 5 0.000382 0.34 + 0.0003547 T
3. CO 2 44 1,073 547.7 0.25
4. Coke 13 0.3
5. O 2 32 730 277.9 0.24
6. Light oil 44 615.9 665.6 2.2 x 10- 4 0.000769 -1.1843 + 0.003452 T
TABLE 4 - COMPARISON OF CROOKSTON ET AL. WITH RUN 6 RESULTS FOR EXAMPLE ONE·DIMENSIONAL
FORWARD COMBUSTION PROBLEM
Crookston
eta/. Run6
Oil recovery, % 97 97.18
Light oil, % 30 27.7
Heat loss, % 27 26 Cell Peak Temperature F (days)] r
Velocity, ftlD 1.2 1.35 1 600(33.3) 534(32.3
Water recovery, % 119 118.2 2 775(44.9) 728(42.1)
849 836 3 855(57.5) 861(52.7)
Tpavg, of
4 850(71.5) 878(65.9)
Heat of reaction, Btu x 10- 10 0.7925 0.7183
5 868(85.9) 924(79.9)
Computing time, seconds 1,071' 242"
6 820(99) 816(92)
Number of time steps 764 122 7 875(113.1) 824(103)
Number of iterations 3,056 530 8 870(126.6) 836(116)
Average M, days 0.1965 0.122 9 880(140.5) 822(127)
10 780(150.2) 653(137)
Time Step Controls • IBM 3701168.
0.1 0.4 "CDC 6600.
Saturation
Mole fraction 0.2 None
Pressure, psi 90 20
Temperature, of 30 80
mulation required about 764 time steps compared where Ki is the equilibrium K-value (Yi = Kixi,xi =
with 122 for Run 6. Their larger computing time per oil-phase mole fraction, and Yi = gas-phase mole
block step reflects, at least in part, their direct fraction) of component i. Thus, increasing the K V4
solution of nine primary equations compared with value decreases the component volatility. Base-case
our six primary equations. values of K V4 for Components 2 and 6 are 6,739 and
We feel the considerably smaller number of time 3,370, respectively.
steps in our Run 6 is primarily due to the completely A number of runs made are unreported in Table 5
implicit formulation, which allows larger steps. We because of insensitivity of results to (1) 20070 increases
have noted very little time truncation error in forcing in component or rock specific heats, (2) 50070
our runs to use smaller time steps. reduction in oil-component thermal expansion
coefficient, and (3) change from 44 to 32 of COx
Sensitivity Runs
gaseous Component 3 molecular weight.
A number of one-dimensional sensitivity runs were Run 8 indicates little sensitivity to permeability in
made using the Crookston et at. data. These runs the range of 2,000 to 4,000 md. Actually, reduced
were made to indicate which data are "important" in permeability may have a pronounced effect if it is
the sense of having a significant influence on reduced to a value low enough to increase reservoir
calculated results. The results serve as a guide for pressure significantly and, hence, reaction rate.
history-matching efforts in that they indicate the Run 9 indicates a pronounced effect of lowered
types of response to changes in various input-data porosity. Reaction rate decreases due to smaller
variables. The utility of the sensitivity-run results is reactant concentrations and oil recovery falls from
limited because they are only applicable to the 93.5 to 74070 as porosity is decreased from 0.38 to
Crookston et at. data and do not reflect two- or 0.24. Also, of course, more rock is associated with a
three-dimensional flow effects. given pore-space volume as porosity is decreased.
The base case (Run 7) is a reservoir 82 ft long, 115 Reduction of overburden thermal conductivity by
ft wide, and 21 ft thick. The 82-ft length is a factor of two reduces the heat loss moderately but
represented by five 16.4-ft grid blocks. A well has little effect on oil recovery and peak temperature
productivity index of 1,328 was used together with (Run 10).
the data listed in Table 3. Runs 11 and 12 show that a 20070 increase in coke-
Base Run 7 results are tabulated in Table 5. Tpav oxidation heat of rection has a greater effect than a
is the average of peak temperatures in Cells 2 throug6 20070 increase in the direct-oxidation heat of reaction.
4. Velocity is 2 X 16.41(t4 - t 2 ), where ti is time of Run 13 shows increased heat loss and peak tem-
peak temperature in Cell 1. Time tabulated in Table 5 perature caused by decreased activation energy of the
is the time at which the last cell (Cell 5) peaked in direct-oxidation Reaction 1. Run 14 shows that a
temperature; oil recovery was complete at some time decreased cracking-reaction activation energy
prior to that time in all cases. The notation in the last doubles the peak temperature to 1,244°F and gives a
column of Table 5 (e.g., E1 = 0.8) means that ac- much lower combustion-zone velocity and higher
tivation energy of Reaction 1 is 0.8 times the base- (volumetric) oil recovery, with light-oil volumetric
case value. K V4 is defined for heavy and light oil fraction in recovered oil rising from 19.9 to 86.1.
Components 2 and 6 by Increasing the coke-oxidation activation energy by
KV4i
54070 (Run 15) moderately increases peak temperature
Ki = (K V2i/P) e T-Kv5i and heat loss. Runs 16 through 18 indicate a greater
DECEMBER 1980 547
o INJECTION WELL
• PRODUCTION WELL
• GRI D BLOCK CENTER
1=2 1 =3 i =4
2 640.3 869.9 1146.8 TEMP,OF
k= 1 (TIME,DAYS)
(480) (1440) (2140)
396.5
k =3 (1440)
5X3GRID k=4
C>X =6y = 27.33 FT.
Fig. 10 - One-eighth of a five-spot grid. Fig. 11 - Peak temperature of a 5 x 4 wedge - Run 25.
where
XIJ = KvIJX I ' . . . . . . . . . . . . . . . . . . . . . (A-21)
where KVIJ is an equilibrium K-value and XI is the w=
mole fraction of component I in component I's
master or. primary phase, J I' That is, K vIJI == 1.0.
As long as Sg is increasing, always ~ (Sg) max' these
ForJ~JI'
equations give Sgr = S c' and kr from Eq. A-26
K vIJ = [K VI (l,J) + K V2 (l,J) / p will follow the "original$, krg vs. ~ curve. Then, as
Sg decreases toward zero, krg will approach zero at
K V4 (/,l) SgrC'
This hysteresis is generally not important and has
+K V3 (l,J)pJe T-Kvs(/,J), . . . . (A-22) an effect (not a large one, from our experience) only
in cyclic steam stimulation runs.
where K VI through K V5 are read as input data. The Gas-phase relative permeability is treated with no
form of Eq. A-22 is that used by Crookston et al. 2 temperature dependence.
For water, K V13 is calculated as Ps/p, where Ps is The temperature-dependent quantities Swir (T),
water vapor pressure, a single-valued function of k rwro (T), etc., all are treated as
temperature. If T exceeds Tc for water, then Ps is
taken equal to water critical pressure. X =X O
+a(T- T i ) , ................ (A-3l)
Relative permeability and capillary pressure data
may be calculated analytically or interpolated from
where XO and the temperature derivatives (a) are
read-in tables. Analytical calculation of relative
read in as data.
permeabilities is as follows.
The normalized saturations in the above equations
5W - S . (T) ] nw are
k = k (T)
[ 1- wlr
rw rwro S ( T) - S . (T) . Sw
orw wlr 5w
........................... (A-23) I-S4
1 - S orw ( T) - 5w ] now So
k
row
=k· (T) [
rOlw 1- S
orw
( T) - S . (T)
wlr
. 50 I-S4
........................... (A-24) and
DECEMBER 1980 553
s~g = ~ SI Metric Conversion Factors
I-S4 °API 141.5/(131.5 + °API) g/m3
Analytical expressions for capillary pressures are atm x 1.013 250* E+02 kPa
bbl x 1.589 873 E-Ol m3
Pcwo(psi) = [PCC(I) + PCC(2) x (1- 5w )]
Btu x 1.055 056 E+OO kJ
+PCC(3) x (1- 5w ) 3] cp x 1.0* E-03 Pa·s
cu ft x 2.831 685 E-02 m3
x [1 - PCC( 4) x (T - Ti ) ] ' .. (A-32) of CF - 32)/1.8 + 273.18 K
and ft x 3.048* E-Ol m
Ibm x 4.535 924 E-Ol kg
PCgo(psi) = [PCC(5) + PCC(6) x 5g Ibm mol x 4,535 924 E-Ol kmol
+ PCC(7) x 5lJ x [I - PCC(8) psi,psia
scf
x
x
6.894757
2.863640
E+OO
E-02
kPa
std m 3
X (T - Ti ) ] .............. (A-33) sq ft x 9.290304* E-02 m2
SPEJ
PCC(I) through PCC(8) are input data. The tem- Original manuscript received in Society of Petroleum Engineers office Sept.
23. 1979. Paper accepted for publication April 21, 1980. Revised manuscript
perature dependence uses the assumption that in- received Aug. 18, 1980. Paper (SPE 8394) first presented at the SPE 54th Annual
terfacial tension is a linear function of temperature. Technical Conference and Exhibition, held in Las Vegas, Sept. 23·26,1979.