Professional Documents
Culture Documents
The Mechanisms of Ground Surface Subsidence Above Compacting Multiphase Reservoirs and Their Analysis by The Finite Element Method
The Mechanisms of Ground Surface Subsidence Above Compacting Multiphase Reservoirs and Their Analysis by The Finite Element Method
method
K. Morgan, R. W. Lewis and I. R. White
Department of Civil Engineering, Swansea SA2 8PP, UK (Received January 1980) University College of Swansea, Singleton Park,
An analysis of multiphase flow through a compacting porous medium is investigated in this paper. The nonlinear partial differential equations governing the flow regimes are derived and a method of calculating the effective stress gradients is given. This method involves a local averaging technique to model the interaction of two immiscible fluid phases on a small volume of the deforming porous medium. The combined stress and flow problem is then solved using the finite element method where the flow is confined to a submesh of the whole region. A simulation is made of a typical vertical cross-section of the Forties Field.
Introduction
With the continually increasing world demand for energy the fastest possible withdrawal of hydrocarbons from underground reservoirs has become an economic necessity. One of the attendant dangers when withdrawing reservoir fluid is the large pressure drop within the reservoir itself. This change in pressure can cause significant alterations in the effective stress pattern within the reservoir thus bringing about settlement at the ground surface. Geological studies of these phenomena have been carried out in the cases of both gas1 and oil* production. Significant settlements at the ground surface have been observed that have produced damage in urban areas. The Wilmington field, Long Beach, California, has created a subsidence bowl with a dip of over 9 m at its centre3 and other fields have observed smaller settlements: oil. production from around the Bolivar coast, Venezuela,* and water production from beneath the Venice area of Italy4 are two examples. The subsidence in the province of Gronningen, Holland, has been studied extensively11 s where natural gas is being withdrawn from the reservoir under its own pressure. This case of single phase withdrawal can, in a simple geometry, be solved analytically6 where a uniform pressure drop is assumed to have taken place in a disc-shaped reservoir of constant thickness. However, in realistic geometries an analytic treatment is normally impossible and numerical methods must be used. The authors are unaware of any
0307-904X/80/030217-08/$02.00 0 1980lPCBusinessPress
analytic treatments of the compaction of multiphase reservoirs where oil is being flushed out by injecting water. This water-flooding procedure is common in the oil industry and this paper examines the displacement mechanisms of the fluid flow and their effect on the reservoir and its overburden. The deformation in the reservoir and overburden is examined by considering the change in effective stress field u$ caused by the fluid flow. The commonly used expression: oii = oij - 6.. ZIP where aii is the total stress tensor, p is the fluid pressure and 6ij the Kronecker delta is only valid for inviscid single phase flows so a generalization of this expression is derived for two-phase viscous flows. The finite element method is used to discretize the differential equations to produce a numerical solution where irregular geometry and material heterogeneity may be accounted for. The method is then applied to an EastWest vertical cross-section of the North Sea Forties Field to demonstrate its validity.
The domains
of solution
Naturally occurring hydrocarbon deposits are normally to be found in subsurface rock strata where they are trapped by the surrounding impervious rocks. Thus the flow of the
217
Ground surface subsidence: K. Morgan et at. hydrocarbons only takes place within these porous strata, whereas the stress variation occurs both in the reservoir and its overburden (Figure 1). For this reason two domains will be considered: one comprising the combined reservoir and overburden and one being the reservoir alone where the stress and flow analyses will be carried out respectively. Let Q be the N-dimensional region representing the combined reservoir and overburden and let !$ C a be the subregion of fi where fluid flow takes place. It is in this subregion of s2 that the combined pressure of the fluids will play a significant role in the stress pattern prevailing in a. Thus the equations governing the multiphase flow will be confined to $. Now let58 be a discretization of R into finite elements7 and let: fi?h = d;2hd I be such that the area of symmetric difference a - ah is sufficiently small.sh must be chosen so that there exists 93 C gh which also leaves Szf - $ acceptably small where and x1(+). Observe that .% ~93 and that for any u, v E .x1 we have: (u,v)= s
i-2
uv+Vu.Vvda
where V is the Laplacian operator. The numerical solution of the equilibrium problem will not be sought throughout the whole of %(a) but will be confined to a subspace Sh whose basis is a suitably chosen set of polynomials with small compact support. In addition the flow equations will be solved inside the restriction of Sh to the subregion a$ of s2 where the basis of Shl+ will be assumed to be the canonical sub-basis of Sh produced by this restriction.
Furthermore, the members of9 must be disjoint and have edges that are polynomial functions of order p = 1,2,3 in the space variables. Isoparametric quadrilaterals, Lagrangian quadrilaterals and triangular elements7 are all permitted in the same @. This allows the user to mix mesh types and thus handle some of the awkward geometries of these problems more easily. Having carried out the discretization of Q the finite element analysis of the equilibrium problem will be carried out in ah and the multiphase flow in !$. In this study the equilibrium equations and the flow equations are solved in different regions and this fact leads to the construction of different solution spaces. Consider the Hilbert space H(R) consisting of all functions f: R + R with square integrable mth derivatives over a. The scalar product and norm on this space are given by:
ao13
I
0
-t y+b(x).j=
~t-t--p(x).k=o
ay
I
u1313 u23/3
where i, j and k are the unit vectors in the x1, x2 and x3 directions respectively and x E !A. Similarly if T(X) is the boundary traction at x and if I,, l2 and 13 are the direction cosines of the outward facing normal we have: 7(x)-i
7(x)-j
= urrzr + or212 +
= 02lll-b 02,/,+
(u,v),=
ol<m
0.
T(X).k = ~3rZr + ~32Z3+ ~3313 which must hold together with (1) to maintain equilibrium. If ur, u2, u3 are the displacements in the x1, x2 and x3 directions respectively then the strains resulting from these displacements are given by: k,Z= l,... and the dilatation is denoted by: ,N
D% . D% dR
IIUIlkI =(u,u),
where: Du = and fOraiEN with any u, v, E31Pm(S1). In this study the spaces X0(a) and X(R) will be used along with the spaces %(af)
Sea-bed mounted
aall
ax:1 ax;' ...a~.~
=ks, ig
Making the assumption that the sum of the moments of the surface tractions on any small element of the region with respect to each axis vanishes we have the statement of symmetry for the stress tensor:
(Jkl (7lk =
iv auk
k,Z= l,...,N
okI = 2G ekl + 6kle 1 - 2v 1 where G is the bulk shear modulus and v is Poissons ratio. For the purposes of clarity in the finite element discretization the stress and strain tensors are rearranged into
Figure
Typical
cross-section
of sub-sea reservoir
218
Appl.
Math.
Modelling,
1980,
Vol 4, June
vector form u and E respectively.7 Thus if u is a displacement field in fi producing strains E we have: u=L?e where 9 is the appropriate differential operator. Further if uh is a vector of members of Sh which represents a displacement field producing strains eh we have: eh =B.& where uh = ZcFNi
i
F,
= s r
[NIT7 dI-
Fb = s n FE0 =
[N] Tb dR
BTDco dSZ s n
aNi
F,o = -
a,O
Bi = 0 aNi -aNi
av
aNi
So from equation (5) u may be calculated from a knowledge of the initial stresses. boundary conditions and bodv forces within Qh. _
ay
ax
Multiphase flow
The subregion L?f C s2 is the domain of the multiphase flow analysis and is taken to be a porous medium. Its near infinite geometrical complexity makes rigorous analysis of the flow impossible. However, regular pore arrangements have been studied by Graton and Fraser.9 In order to model the flow the fictitious seepage velocity (otherwise known as bulk or Darcy velocity) is defined by: q=-
for N = 2. When the region fi is undergoing deformation the sum of the work done by the external forces and the internal strains must vanish. Consequently the application of the principle of virtual work8 gives: JP.Tdr+ r jP.bdQ=/e.odlt rr 52
where Us is the virtual displacement field. Now if we choose [zP] to have members in Sh where Sh is spanned by the Ni(i=l,...,n)wehave:
1KV(p f -yh)
P
dQ = /[eh]?$ 52
da
where [@I and uh are the strains and stresses produced the displacements [u]. Consequently we may write:
by
Now the stresses u may vary from an initial state of stress u and initial strains e by: a =D(e - 8) t u0 (4)
where D is the appropriate elasticity matrix obtained by re-writing (2). Thus, observing that the virtual displacements [UU]are arbitrary, equation (3) becomes:
=( SBTD~d~)u-SRTDEodntrBT7d~ n n ;2
where p is the fluid pressure, y the specific weight, i? the permeability matrix, p is the dynamic viscosity, and h the height above an arbitrary datum. This formulation is a generalization of an experimental result carried out by DarcylO where a tube of sand of finite length was used. Scheideggerrr discusses the limiting case of the experimental result where the change in head per unit length of tube is replaced by V(p t -yh). He decided in favour of the form of the law used here where the permeability terms remain outside the differential operator. This formulation is also supported by Hubbert. l2 When considering a multiphase flow regime, however, the permeability of the medium to each individual fluid is different and in addition may vary according to the saturation of a particular fluid at a given point. Thus, a relative permeability function k,,(&) needs to be introduced to modify the E matrix in (6). Although analytic expressions do exist for k,,(S,) based on the properties of ideal materials, the values of these functions are usually obtained by experiment. In a two-phase flow where one fluid is assumed to wet the pore space the relative permeabilities are typically those shown in Figure 2. Defining the fluid potential @* by: @n=Pn+-rnh for the rr-phase we have: qn = -&V@, (7)
where u are the nodal displacements required to maintain equilibrium. This may be rewritten as: F,+F,+F,o+F,o=[K]u where : 0)
WI =
s
fl
where qn is the velocity of the n-phase and I?, = k,,E. When considering a single fluid phase II the continuity equation becomes: - divq, = @& (8)
BTDB da
where $Jis the porosity of the medium and the raised dot
Appl.
Math.
Modelling,
1980,
Vol 4, June
219
Ground surface subsidence: K. Morgan et al. and by using Greens theorem and (7) we have:
vq,.ri dr Wz WP
I
= 6, s s2
6(x)-$ v(+,
c
- 6,) da
(11)
Observe that v and Q.n are members of .%(slf) and are thus only required to have Co continuity (i.e. v and @ must be continuous whereas Vv and o@, are only required to be square integrable). Now let v = N!(X) for x E fl2$ and let:
I_
S
Figure 2 Relative permeability curves
- &I
(12)
represents differentiation with respect to time. In a twophase flow the capillary pressure at a point is given by: Pc=Pn-Pw and is linked by experimental results to the saturation, S, of the wetting phase. Thus: S=S(p,)= 1 -S, (9)
(Pn =
hwT%2,...?$4#-
4 rri =
-1
+1
This is the pair of nonlinear partial differential equations that will be solved within LQ with a discretization based on $. The boundary of Qf, denoted I,, will be divided into three disjoint portions r, Wr, IVp where r is the impervious part of the boundary along which (I,, . ri = 0 (n = n, W) where ri is the outward facing normal. Wetting phase fluid will be injected on WI and a mobility boundary conditionr3 will be employed on W, where the nonwetting phase or a mixture of the two phases is produced from s2f.
When calculating values of m, and dS/dp, it is necessary to have an accurate knowledge of the saturation S. This is obtained from an experimental curve a typical example of which is shown in Figure 3 and is usually fed piecewise into the program. However, the values of pc that are required are sometimes subject to ill-conditioning since they are calculated from the difference @, - Qw of two nearly equal quantities. Consequently the equations (12) are added and subtracted respectively to give: O=MP+NR+Q, 4Ck =MR +NP+Q, (13) (14)
The spatial discretization of (10) is carried out by the Galerkin method where the trial and test functions are both taken to lie within Sh. Multiplying (10) by v E .8(R) and integrating gives:
Qd =Q, -Qw
The sequence of initial value problems (13) and (14) are highly nonlinear. However, it should be noted that the matrices M, N and C depend solely on a knowledge of R
220
Appl.
Math.
Modelling,
1980,
Vol 4, June
Ground surface subsidence: K. Morgan et al. type require highly stable timestepping methods such as &,-stable methods.r4 This predictor-corrector pair is Aestable for 19< :.I5
Calculation
When considering the deformation of porous materials under the influence of mutliphase flow regimes, careful consideration needs to be given to the precise nature of the effective stress tensor Ub. It is generally accepted161 l7 that the effective stress in the case of a single phase inviscid flow is given by: Uij =
Uij
-Gijp(X,
t)
where p(x, t) is the space and time varying pressure. If, however, the medium is deforming simultaneously with the flow the stress tensor may be written in the form: u;j = ujj - $j(l
O-
- @p(x, t)
(18) to its
where /3is the ratio of the rock matrix compressibility bulk compressibility cb where: 3(1 - 2V) cb = 2G( 1 + V)
Figure 3
Saturation/capillary
pressure curve
and may thus be rewritten in the form: R where: A(R) = ; C_[M - NM-N] and V(R) = ; C-[Qd - NM-Q,] The equations (1.5) are discretized in time using a two-level predictor-corrector methodr3: Predictor: R z+l- R = At[(l tAtI Corrector: R ;+; -R= At[(l - B)A(R;)R;f; + BA(R;)R] (17) - B)A(Rn)R,+ + &4(R) R] (16) =A(R)R t V(R)
after Ceertsmals but this formulation is limited to inviscid single phase flows. Using this expression for the effective stress the problem of a disc-shaped reservoir of small thickness within which a uniform pressure drop has taken place can be solved analytically.6 However, no exact solutions to the multiphase case are known to the authors. The expression (18) must now be generalized to viscous multiphase flow regimes. By considering the flow on a microscopic level Gray and ONeill19 have reduced the interstitial momentum balance equation to a form of Darcys Law by using an averaging technique developed by Whitaker.20-22 This averaging process is used to link the individual pore velocities of the fluids to the fictitious seepage velocity originally introduced by Darcy. WhitakerZo has shown that valid results are obtained if:
where d is a characteristic microscopic length over which significant variations in the space dependent quantities occur, I is the length of the averaging volume V (see Figure 4) and L is a characteristic macroscopic length. Provided that the averaging volume Vis independent of time we may define the phase average of some quantity < in the n-phase by: *
Denoting an application of the predictor and corrector by P and C respectively and an evaluation of new matrices by E the method is used in the mode PE(CE)m where m is the smallest integer that satisfies: IIRk+: -R;+:, II_ <e convergence factor.
where E is a user-specified
where v is the volume of the region I/ and {,, is the value of t within the n-phase which is taken to be zero in all other phases.* Observe also that (<{,)) is an average associated with a given point in the flow domain which may not lie within the rr:phase but its value may still be non-zero even if that point lies inside the n:phase where n# rr. The averaging region V may be divided into regions V, occupied by each of the phases 71and having volume v,. Thus:
Stability analysis
It can be shown that the equations (15) are numerically stiff with a condition number O(h-) where h is a measure of the mesh size. Systems of initial value problems of this
* For the purposes of this averaging the subregion solid is treated as a phase in its own right. of V occupied by
221
It follows that the nodal loading corresponding tive stresses u is given by: P r Fur= J BTodC2 + J (1 -S)BTmp,da cl 52
to the effec-
[l, LOIT
on-wetting Figure 4 Typical averaging volume for a multiphase phase flow
N=2 N=3
[l,l,l,O,O,OIT
By applying the loading term (23) in equation (5) the displacements U(X, t) at time t may be evaluated from:
and we may additionally define the region VI to be the region occupied by the liquids. Now, if the pore space is occupied by two immiscible fluids as shown in Figure I, with one wetting (denoted w) and one non-wetting (denoted n) we may consider the phase average of the total stress tensor and express it as: C(uij)>=
settlements
S(X, t) = u(x, t) - 24(x,0) for xE I$ is the subset of rh corresponding surface. to the ground
s .I __
Uij
dV
Application
=((Ofj>)
k
w
o;dVt~
I v
o;dV
((Ufj))
S~((U~)) + SV,((U~>>
where Sz is the fraction of the pore space V, u VW occupied by the n-phase (normally named the saturation of the n-phase). This is, however, not yet in a form which may be used in the computer for two reasons. First, no information is available concerning the shear stresses in the fluid phases and second, the divergence of the stress tensor is required to calculate the force terms in the equilibrium equation (5). This divergence may be expressed as:
g((Oij)) = lJ<(O$)) + SEO<COn)) + S~~((U~))
(20)
since S, is space independent within V. Now summing (20) over all volumes V in af we have:
v'Ub=
o.Uij-S,~U$-Sw~.U~
(21)
Bear23 observes that in the case of slow, or Darcy flow the gradients of the shear stresses in the fluid phases are small enough to be neglected and SO:
The method described above has been applied to a vertical cross-section of the Forties Field in the North Sea. The Forties Field is located mainly in the UK licence block 2 l/10 about 180 km ENE of Aberdeen. The hydrocarbon bearing rock is mainly Paleocene sandstone in an anticline at about 2100m subsea. An assessment of its geological structure was given by Hillier et ab24 who also outlined the reservoir history and a possible development plan to maximize the recovery up to 1985. The present study considers a vertical cross-section of the reservoir shown in Figure 5 which is located beneath 2100m of overburden. The reservoir itself consists of two main oil producing areas, the main sands and the Charlie sands in which the rock and fluid properties are similar. These are separated by an interchannel and margin with lower values of porosity and permeability. Four wells penetrate this east-west cross-section as shown in the figure. Wells FC26 and FD34 are taken to be iniection wells and FC32 and FC62 are wells where oil and water are being produced. This serves to model the peripheral injection envisaged by Hillier et al. and also to provide a reasonable test of the algorithm in the irregular geometries so frequently encountered in practise. In contrast to other studies69 2s the assumption that reservoir thickness is small compared with the depth of burial is not made in this work
with 3~ being the identity tensor of order N and pa the fluid pressure in the n-phase. Thus (2 1) becomes:
222
Appl.
Math.
Modelling,
1980,Vol
4, June
Ground surface subsidence: K. Morgan et al. as there is no restriction on the size of ,Q$ in comparison to R. Geological data indicates that in fields such as the one being studied there is not a distinct oil/water interface but a continuous variation in saturation between the oil-rich zones and the watered-out zones. Initially, since no flow is taking place, the values of pC are constant so the saturation variation with respect to height is obtained from Figure 3 by taking: h= %-%-PC -Yn-Yw
x, ( km) 0 1-o
directly from the definition of pC. Relative permeabilities of the form k, = 1 - S, k, = S are used to validate the method. Water is injected into the oil below the nominal oil-water contact line on wells FC26 and FC34 and the equations (13)-(14) are solved with the time discretization (16)-( 17) to give the fluid pressures and then establish the saturation profiles.
Development
of subsidence bowls
The development of the subsidence bowl above the section of the Forties Field shown in Figure 6 was analysed under a variety of withdrawal rates. The fluid withdrawal which was increased over a period of five days to a constant rate makes significant differences to the pressure field within the reservoir. Thus, a larger withdrawal rate may be expected to produce an increased pressure drop resulting in greater settlements. The behaviour of the S = 0.5 saturation contour within s2, is shown in Figure 7 where the rise in the water level and subsequent breakthrough of water in the well FC62 both occur as expected. The depth of the subsidence bowl for given material properties varies both with withdrawal rate and reservoir depth. The variation in reservoir depth can be predicted qualitatively by considering the equations of poro-elasticity.63 26 The Boussinesq-Papkowitch solution shows that the poro-elastic displacement potential gradients at a point
Figure 8
Development
bed
----_--
Overburden
Rh
CY
-3
Reservoir t
Figure 6
are reduced as the point moves further from the reservoir. Thus, smaller surface settlements may be expected for greater reservoir depth. Variations in the total fluid withdrawal rate bring about changes in the potential gradient V(Q, + a,+,) within SIQ.These increases in potential gradient cause a corresponding increase in the difference between the terms of the total and effective stress tensors which in turn cause greater surface settlements. These graphs of Figure 8 demonstrate this in the Forties Field example where significantly deeper subsidence bowls are obtained with the higher flow rates. A problem of practical importance is that of the prediction of the differential settlement at the surface. In the North Sea, where tall sea-bed mounted structures are located, a differential settlement can induce an overturning moment. The case of the Forties Field was examined under a variety of withdrawal rates and the development of the subsidence bowls can be seen from Figure 8. With current withdrawal rates the surface settlement is only of the order of millimetres because of the depth of the reservoir. Consequently differential settlements can be considered as insignificant in this case. However, the results verify the validity of the method for establishing such potentially dangerous ground movements.
FC32
CharlIe /
Sands
FC62
FD34 1
Conclusions
The problem of ground surface settlement above compacting multiphase reservoirs has been analysed by considering the displacement and flow problems separately. The partial differential equations of the flow problem have been derived in a subregion fly of R to model the case usually prevailing underground where the hydrocarbon reservoir is trapped between impervious strata.
Figure 7
Development
of S = 0.5 saturation
line
Appt. Math.
Modelling,
1980.
Vol 4, June
223
K, Morgan et al.
5 6 7 8 9 10 11 12 13 Geertsma, J. Pet. Tech. 1973, 226,734 Geertsma, J. Verhandelingen Kon. Ned. Geol. Mynbouwk Gen. 1973,28,43 Zienkiewicz, 0. C. The finite element method (3rd Ed.), McGraw-Hill, London, 1977 Timoshenko, S. P. and Goodier, J. N. Theory of elasticity (3rd Ed.), McGraw-Hill, London, 1970 Graton, L. C. and Fraser, H. J. J. Geol. 1935,43. (8), 785 Darcy, H. Les Fontaines Publiques de la Ville de Dijon, Dijon, 1856 Scheidegger, A. E. The physics of flow through porous media, Univ. ofToronto Press, i957 Hubbert. M. K. J. Geol. 1940.47.785 Lewis, R. W. et al. Int. J. NuM. keth. in Eng. 1978, 12, (2), 319 Cryer, C. W. BIT 1973,13,153 White, I. R. PhD Thesis, University College of Swansea (1978) Lambe, T. W. and Whitman, R. V. Soil mechanics, Wiley, Chichester, 1969 Terzaghi, K. Theoretical soil mechanics, Wiley, Chichester, 1943 Geertsma, J. Trans. AZME 1957,210, 331 Gray, W. G. and ONeill, K. WQterResour. Res. 1976, 12, (2), 148 Whitaker, S. AZChE J. 1967, 13,420 Whitaker, S.Znd. Eng. Chem. 1969,61,14 Whitaker, S. Chem. Eng. Sci. 1973, 28, 139 Bear, J. Dynamics of fluids in porous media, American Elsevier, N.Y. 1972 Hillier, G. R. K. et al. EUR 98, European Offshore Petroleum Conf, London, October 1978 Final, A. and Farouq Ali, S. M. Sot. Petrol. Eng. J. 1975, 15, 411 Biot, M. A. J. Appl. Math. 1956, 23,91
A new representation of the macroscopic effective stress tensor has been derived which applies to the viscous multiphase flow case under consideration. This representation is then used in the final settlement analysis. The problem is solved numerically by a finite element discretization in space and a predictor-corrector method in time. Finally, the technique is tested on an east -west cross-section of part of the North Sea Forties Field where the sea-bed settlements are predicted as the water-flood of the reservoir proceeds.
Acknowledgement
The research described in this paper is part of a project supported by the Department of Energy to develop new finite element software for settlement analysis and graphic representation of the results. Their financial support is gratefully acknowledged.
14 15 16 17 18 19 20 21 22 23 24 25 26
References
Schoonbeek, J. B. Sot. Pet. Eng. AIME, paper SPE 5751,1976 Van der Knapp, W. Proc. Seventh World Petrol. Gong., Mexico City, 1970, Galliard, UK Allen, D. R. The mechanics of compaction and rebound, Wilmington Oil Field, Long Beach, California, Dept. of Oil Properties, City of Long Beach, Calif., 1969. Allen, D. R. et al. En&. Del/a. Sci. E. Della. Tech. Mond. 1971,19,281
224
Appl.
Math.
Modelling,
1980,
Vol 4, June