SPE 106179 Multiscale Mixed Finite Element Modeling of Coupled Wellbore/Near-Well Flow

S. Krogstad, SINTEF ICT, and L. J. Durlofsky, SPE, Stanford University

Copyright 2007, Society of Petroleum Engineers This paper was prepared for presentation at the 2007 SPE Reservoir Simulation Symposium held in Houston, Texas, U.S.A., 2628 February 2007. This paper was selected for presentation by an SPE Program Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as presented, have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material, as presented, does not necessarily reflect any position of the Society of Petroleum Engineers, its officers, or members. Papers presented at SPE meetings are subject to publication review by Editorial Committees of the Society of Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O. Box 833836, Richardson, Texas 75083-3836 U.S.A., fax 01-972-952-9435.

Abstract An accurate well modeling capability is important for both production and reservoir engineering calculations. Ideally, the models used for these two applications should be related in a logical manner. Multiscale methods allow varying degrees of resolution and can therefore provide a natural linkage between production and reservoir models. In this work, we present a coupled wellbore-reservoir flow model that is based on a multiscale mixed finite element formulation for reservoir flow linked to a drift-flux wellbore flow representation. The model is able to capture efficiently the effects of near-well heterogeneity in the reservoir and phase holdup and pressure variation in the wellbore. The formulation presented here is for oil-water systems. The basic reservoir-wellbore linkage is described and validated through comparison to results from an existing simulator. The multiscale methodology is then applied to a heterogeneous reservoir model. Both vertical and deviated wells are considered. Comparisons of the multiscale solution to the fully resolved (fine-scale) solution demonstrate the high degree of accuracy of the method, for both reservoir and wellbore quantities, as well as its efficiency. Introduction The accurate modeling of near-well flow effects is essential for large-scale reservoir simulation and detailed production engineering calculations. Near-well models should ideally include wellbore and reservoir flow effects and should be suitable both as standalone well production applications and as modules in reservoir simulators. Multiscale finite element methods are in concept well-suited for this type of modeling, as they allow varying resolution and provide a systematic procedure for coarsening and refining. Thus they can provide the basis for a modeling framework that maintains consistency between reservoir and production engineering models. To our knowledge, however, multiscale methods have not yet been applied for this problem.

In this work, we develop a multiscale mixed finite element method (MsMFEM) for modeling coupled wellbore and nearwell flow. The MsMFEM solves the pressure equation (in the reservoir domain) on a coarse grid, but captures fine-scale effects through basis functions determined from numerical solutions of local single-phase flow problems on the underlying fine-scale geological grid. We fully resolve the well trajectory on the fine scale using a flexible grid, which is close to radial around the well but logically Cartesian away from the well. The flow within the wellbore is represented using a drift-flux model. This model captures the slip between phases and provides the in-situ phase fractions (holdup), which provide a means for computing the pressure profile within the wellbore. Pressure variation within the wellbore is important in many settings as it impacts the local inflow from the reservoir into the well (in the case of a production well). Multiscale methods for reservoir simulation have been introduced as an alternative to standard upscaling as a tool for more fully incorporating fine-scale features at low computational cost. Among the relevant approaches are the multiscale finite element methods (e.g., Hou and Wu, 1997), the multiscale finite volume method (Jenny et al., 2003) and the multiscale mixed finite element method (Chen and Hou, 2003). The variational multiscale approach of Arbogast (2000), also formulated within a mixed finite element context, represents another related methodology. Here we consider a version of the multiscale mixed finite element method (MsMFEM) introduced by Aarnes (2004) and extended in later publications (Aarnes et al., 2006a, 2006b). Well representations have been incorporated into multiscale formulations by, for example, Chen and Yue (2003) and Wolfsteiner et al. (2006), though these formulations did not include wellbore flow effects. As far as we are aware, multiscale formulations have not been applied previously to detailed near-well models that include both reservoir and wellbore flow effects. Wellbore flow modeling within the context of reservoir simulation is often accomplished using drift-flux (or related) models. These models are well suited for use in reservoir simulation as they are simple, continuous and differentiable. The drift-flux representation (Zuber and Findlay, 1965) includes two basic parameters the profile parameter C0 and the drift velocity Vd. These quantities depend on system parameters such as phase flow rates, well inclination and fluid properties. The general drift-flux model applied here is based on the formulation originally introduced by Holmes (1977) and Holmes et al. (1998). In recent work (Shi et al., 2005a), the model parameters for two-phase flow (gas-liquid and oil-

S. Krogstad and L.J. Durlofsky

SPE 106179

water) were optimized based on large-scale experiments involving a 15 cm diameter inclinable flow loop (see Oddie et al., 2003, for details of the experiments). In the current work, we apply the drift-flux model described in Shi et al. (2005a). We note that the model was subsequently extended to threephase wellbore flow (Shi et al., 2005b), though we do not consider flows of this type here. This paper proceeds as follows. After introducing the model equations, we briefly present three mixed finite element approximation techniques relevant for this paper: the standard Raviart-Thomas approach, mimetic finite differences, and the MsMFEM. The drift-flux model for computation of wellbore in-situ phase fractions and pressure drop is then presented, followed by a description of its coupling to the MsMFEM. Next, we validate the mimetic finite difference and MsMFEM implementations through comparison to an existing simulator. MsMFEM solutions are then compared to reference solutions for a heterogeneous model computed directly on the fine grid and close agreement is observed. We present results involving vertical and inclined wells, where slip between phases and pressure drop due to wellbore friction occur. Numerical Procedures Governing Equations. We consider model equations for twophase immiscible flow in porous media. Although included in the numerical experiments of this paper, we here neglect compressibility and gravity for ease of exposition. Letting denote the domain of interest, the pressure equation reads: u = kp in , . .(1) u = q in . Here u is the Darcy velocity, p is the pressure, is the total mobility (which depends on water saturation S), k is the permeability tensor and q is the source term. We henceforth assume no-flow boundary conditions, u n = 0 on , where n is the outward pointing unit normal. Letting the two phases be oil and water, the conservation of water is expressed via the saturation equation:

of square integrable vector functions having square integrable divergence and zero flow across the boundary. The basic idea in MFEM discretizations is to search for solutions of Eq. (3) in finite dimensional subspaces div U V H 0 () L2 () . That is, find (u, p) U V such that Eq. (3) is satisfied for all (u, p ) U V . Letting { k }
and {i } be bases for U and V, one obtains approximations u=

u
k k

and p =

p
i

i i

, where the vectors U = {u k }

and P = { pi } solve the linear system

B T C

C U 0 = . .(4) O P Q

The coefficients of the sub-matrices are given by

B kl = k (k ) 1 l dx , .......(5)

Cki = i k dx and Q i = i q dx . ....(6)

Let {Ti } be a collection of cells constituting a grid. Then, using the lowest-order approximation spaces, each basis function for pressure i corresponds to a cell Ti , while each flux basis function corresponds to an interface (edge in 2D, face in 3D) between two neighboring cells. For two neighboring cells Ti and T j , we henceforth denote their common interface by ij , and the corresponding flux basis function by ij . Next, we briefly describe three discretization techniques for Eq. (3): Raviart-Thomas, mimetic finite differences and MsMFEM. For all these methods the approximation space V for pressure coincides, and is simply the space of piecewise constant functions over {Ti } . So for a cell Ti , the corresponding pressure basis function i is given by

S + ( f ( S )u) = qw . .....(2) t

Here is the rock porosity, f is the Buckley-Leverett fractional flow function, and qw is the water source term. Mixed Methods. The starting point for mixed finite element methods (MFEM) (Brezzi and Fortin, 1991) is the weak div formulation of Eq. (1): Find ( u, p ) H 0 () L2 () such that

i ( x) =

1 / | Ti | if x Ti , . ..(7) otherwise. 0

u ( k )

u dx

p u dx = 0,

p u d x = qp d x ,

. ...(3)

div for all (u, p) H 0 () L2 () . Here L2 () is the space of div square integrable functions over and H 0 () is the space

We now describe the approximation spaces for velocity. Raviart-Thomas. The lowest-order classical MFEM approximation spaces are the Raviart-Thomas spaces ( RT0 ). For triangular/tetrahedral and regular grids, the space U is characterized by piecewise linear vector functions over {Ti } , having constant normal component across the cell interfaces. For more generally shaped grids, one usually needs mappings (Piola transforms) to reference elements to enable assembly of the linear system (4). For general grids, this is not straightforward. Mimetic Finite Differences. Mimetic finite difference methods applied within the context of subsurface flow modeling were presented by Hyman et al. (1997, 2002). Although motivated by finite differences, recent versions by Brezzi et al. (2005a, 2005b) can be seen as a generalization of the RT0 -based method that eases treatment of more general

SPE 106179

MsMFEM for Coupled Wellbore/Near-well Flow

geometries. In particular, for two neighboring cells Ti and T j with interface ij , one can define the corresponding flux basis function ij as the solution of the problem
1 / | Ti | for x Ti ij = p, ij = , ...(8) 1 / | T j | for x T j with ij n = 0 on the boundary (Ti U T j ) , and
ij n ij = 1 / ij on ij , where n ij is the unit normal vector to ij pointing from Ti to T j . For triangular/tetrahedral and

Figure 1: A fine rectangular grid partitioned into coarse blocks constituting the coarse grid.

regular grids, these basis functions will coincide with those in RT0 , while for general grids Eq. (8) cannot be computed exactly. However, as shown by Brezzi et al. (2005b), one can approximate the resulting integrals (5)-(6) solely by using boundary integrals, so there is no need to have an explicit form of the velocity inside the cells. For further details on the construction of these methods and for convergence and stability results, see Brezzi et al. (2005a, 2005b). Multiscale MFEM. In the current MsMFEM (Aarnes et al., 2006b) we consider two grids as in Figure 1, where the coarse grid consists of blocks containing a connected set of cells from the underlying fine grid. We let a flux basis function ij be a numerical solution of the following local problem: wi ij = kp, ij = w j for x Ti for x T j , ....(9)

MsMFEM for Two-phase Flow Simulations. In two-phase flow simulations, where the pressure and saturation equations (1)-(2) are solved sequentially, the MsMFEM provides computational efficiency because the computed basis functions (9) can be reused through several timesteps. As has also been noted for the multiscale finite volume method (Jenny et al., 2004), at most the updating of basis functions near the front is necessary (this assumes the basis functions are computed locally as in (9) rather than globally). For moderate mobility ratios, the updating of basis functions is generally not required. It follows that the computational cost of using the MsMFEM is comparable to that of a flow-based upscaling method. Thus, the use of MsMFEM for two-phase flow simulation can be described by the following steps:
Initially compute basis functions corresponding to coarse interfaces. for n = 1 to N If desired, update basis functions in blocks having steep saturation gradients Solve coarse system based on current saturation Assemble fine-scale fluxes Advance fine-scale saturation by timestep end

with ij n = 0 on (Ti U T j ) , computed on the fine grid. In

(9), wi is a weight function of unit integral over Ti . In particular, we choose

We note that in the current implementation we use the hybrid formulation (Brezzi and Fortin, 1991) which, though equivalent to the mixed formulation, results in a symmetric positive definite linear system of size equal to the number of grid interfaces. The system (4), by contrast, is symmetric but indefinite. Drift-flux Wellbore Flow Model. The effects of multiphase flow in wellbores can have a strong impact on reservoir performance. Neglecting slip between phases in the wellbore may lead to under-prediction of the pressure drop as the holdup of the heavier phase is not included in the model. As discussed in the Introduction, the in-situ volume fractions are represented here by the drift-flux formulation using the model parameters determined by Shi et al. (2005a). In the numerical experiments presented in this paper, we consider only oilwater flow, and thus present details for the oil-water drift-flux model. The gas-liquid and three-phase cases are similar; refer to Shi et al. (2005a,b) for further details. We denote the average oil and water velocities in the wellbore by Vo and Vw and the in-situ volume fractions of the two phases by o and w , with o + w = 1 . The oil and water superficial velocities (Vso and Vsw) are given by Vso=oVo and Vsw=wVw, respectively. The mixture velocity Vm is then given as the sum of the oil and water superficial velocities:

The above choice for cells with nonzero source q over Ti ensures a conservative approximation on both the fine and coarse scales. In the case when q is zero over Ti , the weight function to some extent reflects the permeability in the block. We remark that this choice improves the overall accuracy of the method compared to using just a constant weight, even for anisotropic tensors (Aarnes et al., 2006a). In principle, any conservative solver can be used for the local fine-grid problems (9), but the RT0 -MFEM and mimetic finite differences are especially well suited because they enable direct and exact evaluation of the coefficients (5)-(6) for the coarse system. In the numerical examples presented in this paper, we apply a mimetic discretization which reduces to RT0 for regular grids and scalar permeability.

S. Krogstad and L.J. Durlofsky

SPE 106179

Vm = Vso + Vsw = oVo + wVw . .(11)

As in the gas-liquid drift-flux model, the relationship between the phase velocities in the oil-water model is given by (Hassan and Kabir, 1999)
Vo = C0Vm + Vd , (12)

terms in Eq. (16) are due to hydrostatic, friction and acceleration effects respectively.
Linkage of Reservoir and Wellbore Flow Models Gridding around the Well. Our focus here is on capturing near-well heterogeneities down to the scale of the wellbore. Therefore, as a starting point we consider a fine grid that completely resolves the wellbore trajectory, including the wellbore segments which are modeled as cylinders. Based on this fine grid we perform a partitioning into a coarse grid. Figure 2 shows an example of such a grid, where the coarse blocks are indicated by coloring. The special choice here is that the well segments are also treated as individual blocks in the coarse grid. Figure 3 shows a cross section of two possible coarse grids around a well. Note that in the coarse grid to the right, the well segment (itself being a coarse block) is completely encapsulated in the surrounding coarse block, and thus the coarse grid here differs considerably from a standard simulation grid. There are two motives that justify this choice. First, it facilitates an easy coupling with the wellbore flow model, and second, it eliminates the need for well indices. The well index is, however, in a sense incorporated into the basis function corresponding to the interface between the segment and the surrounding block.

where C0 is a parameter depending on the phase and velocity profiles over the pipe cross section and Vd is the drift velocity due to buoyancy effects. In general both C0 and Vd depend on o and other quantities. For flow in inclined pipes, the drift velocity in Eq. (12) is scaled by a multiplier m( ) where is the angle of deviation from the vertical, such that
Vd , = m( )Vd .

Based on the large-diameter experiments of Oddie et al. (2003), Shi et al. (2005a) suggested the use of C0 = 1 and Vd = 1.53Vc (1 o ) . Here Vc is the characteristic velocity of an oil droplet rising in stagnant water and is given by
g ( w o ) Vc = ow 2 w
1/ 4

, ...(13)

where ow is the oil/water surface tension, g is the gravitational acceleration and o , w are the phase densities. The corresponding deviation multiplier was determined to be m( ) = 1.07 cos( ) + 3.23 sin(2 ) 2.32 sin(3 ). .(14) Assuming incompressibility, the governing equation for oil-water flow in a wellbore can be expressed by writing the mass balance for oil o Vso = qo . ..(15) t x Here x is the coordinate along the wellbore and qo is the inflow to the well from the reservoir. This term acts to couple the reservoir and wellbore flow equations. The inflow qo depends on the reservoir and wellbore pressures and the oil mobility in the near-well reservoir region. We next describe the determination of the pressure profile in the wellbore.
Pressure Drop Calculations. The pressure drop along the wellbore depends on the in-situ phase fractions and can be expressed as (Ouyang et al., 1998):
2 2 f tp mVm 2 qm mVm p . (16) = m g sin( ) + + x D A

Figure 2: Cutaway view of the fine and coarse grids. Fine grid is radial around the well. Coarse blocks indicated by coloring.

Here D and A are the inner cross-sectional diameter and area, m is the mixture density, q m is the mixture reservoir inflow and f tp is the Fanning friction suggested by Haaland (1981), which depends on the Reynolds number and the absolute roughness of the pipe. All mixture quantities are computed as the weighted average of the individual phase quantities with respect to the in-situ phase fractions as in Eq. (11). The three
Figure 3: Cross sections of two coarse grids over the same fine grid. Including the well segment the grid on the left contains nine coarse blocks and the grid on the right contains two blocks.

SPE 106179

MsMFEM for Coupled Wellbore/Near-well Flow

Sequential Splitting of the Equations. Using the MsMFEM requires solving the reservoir equations (1)-(2) sequentially. Thus, we need a similar splitting for the wellbore flow (driftflux) model. Pressure Equations. The pressure drop relations of Eq. (16) are discretized and coupled to the system (4) holding the current phase fractions fixed. The non-linearity in (16) is here treated by a simple linearization such that a fixed-point iteration is required to obtain an initial pressure profile. This approach leads to fast convergence for the numerical examples included in this paper, but we expect that Newton iterations would be more appropriate for pressure boundary conditions, for which the current method converges quite slowly. We also remark that we treat compressibility and gravity in the MsMFEM only at the coarse level. This compressibility treatment is justified by the fact that we here consider nearly incompressible flow. For highly compressible systems special care must be taken; see Lunati and Jenny (2006) for the treatment of such cases within the multiscale finite volume context. We note further that gravitational effects are often important (if not dominant) in the wellbore for vertical or deviated wells. This can be the case even if gravitational effects in the reservoir are of secondary impact. Through treating the well segments as individual coarse blocks, these effects are taken into account. Saturation/Holdup Equations. Because the MsMFEM provides a fine-scale velocity field, the saturation equation (2) requires a fast solver. However, an explicit solver is not appropriate in this setting due to the strong timestep restrictions caused by the small grid cells around the wellbore. Here we use an implicit upstream finite volume method (Natvig and Lie, 2006) followed by an explicit step to account for gravitational effects. The implicit method is based on an optimal ordering of the cells such that the Newton iteration can be done in sequence for each individual cell (or small clusters of cells), rather than on the whole system. For calculation of holdup, we employ the following implicit finite difference approximation to Eq. (15):
n n o +1 o

Numerical Results Homogenous Model with a Vertical Well. The purpose of this example is to validate the current implementation against GPRS, a general purpose research simulator developed in the Department of Energy Resources Engineering, Stanford University (the initial GPRS implementation was by Cao, 2002). We compare both the mimetic method on the fine grid and the MsMFEM on the coarse grid to the GPRS results. The model is a 4200 ft 4200 ft 600 ft homogenous reservoir (k=200 md, =0.3) with a vertical well situated in the center. The wellbore has an inner diameter of 1/6 ft and the roughness is set to 0.001 ft. The well penetrates the entire reservoir and is modeled with 12 segments (see Figure 4). The reservoir has an initial water saturation of 0.5 everywhere. We set water and oil relative permeabilities as krw= S2 and kro= (1-S)2, viscosities as w=o=1, and densities as w=64.8 lbm/ft3 and o=49.2 lbm/ft3. The oil and water compressibilities are both set to 310-6 psi-1. The reservoir is sealed on all boundaries. The fine grid (Figure 4) for this case consists of 7056 cells, while the coarse grid contains 284 blocks. In addition, the 12 segments are included in both the fine and coarse grid. Each coarse block either consists of 3 3 2 Cartesian cells or, if it encapsulates a well segment, 13 12 2 radial cells. The GPRS grid contains 588 blocks. We run one simulation with prescribed total rate of 1600 STB/d and one with total rate 20000 STB/d. For both the mimetic method and the MsMFEM, the prescribed rate is incorporated as a flux boundary condition on the top face of the top segment.

n 1 n 1 Vso+in Vso+out , ,

= qo , .(17)
Figure 4: Fine grid with radial resolution around the well in the center of the model.

n 1 where t is the timestep, L is the segment length, and Vso+in , , n 1 Vso+out are the inflow and outflow superficial velocities along ,

the segment. Using an upstream weighting (with respect to oil velocity), and assuming oil can flow only from the reservoir into the segment (and not the other way around), we first advance the reservoir saturation to obtain qo , and then solve (17) to obtain o . We note here that the formulation outlined above represents in some sense a prototype implementation. Several enhancements will be required in order to develop a model that is applicable and robust for more general flow scenarios such as three-phase flow. These include a more general treatment of compressibility, the handling of dissolved gas, and increased implicitness in the formulation.

We first consider the case with a production rate of 1600 STB/d. Results for wellbore pressure after one day of production are shown in Figure 5, while those for in-situ oil fraction (o) are shown in Figure 6. For both sets of results, it is apparent that the fine-grid mimetic method and the coarsegrid MsMFEM solutions are in close agreement with the reference GPRS results. The wellbore pressure in this case is almost linear, which is due to the fact that hydrostatic effects dominate. The low velocity, however, results in considerable holdup of water as seen in Figure 6. For the high flow rate case, the pressure drop is considerably larger (Figure 7) due to friction (and to some extent acceleration). Here water and oil flow at close to equal rates as seen in Figure 8. For these simulations we again observe close agreement between MsMFEM, the fine-scale mimetic method, and GPRS. The

S. Krogstad and L.J. Durlofsky

SPE 106179

agreement evident in Figures 5-8 suggests that both the mimetic and MsMFEM implementations (including the linkage of the reservoir to the wellbore) are correct.
In-situ oil fraction

0.600 0.500 0.400 0.300 0.200 0.100 GPRS Fine MsMFEM

5600 Wellbore pressure (psia) 5550 5500 5450 5400 5350 5300 5250 5200 1 2 3 4 5 6 7 8 9 10 11 12 Segment number GPRS Fine MsMFEM

0.000 1 2 3 4 5 6 7 8 9 10 11 12 Segment number

Figure 8: Wellbore in-situ oil fraction for high flow rate case (20000 STB/d).

Figure 5: Wellbore pressure profiles for low flow rate case (1600 STB/d).

0.600 0.500 In-situ oil fraction 0.400 0.300 0.200 0.100 0.000 1 2 3 4 5 6 7 8 9 10 11 12 Segment number GPRS Fine MsMFEM

Figure 6: Wellbore in-situ oil fraction for low flow rate case (1600 STB/d).

Heterogeneous Model with an Inclined Well. In this example we consider a larger heterogeneous model with permeability sampled from the tenth SPE comparative solution project (Christie and Blunt, 2001), with additional variability introduced in the near-well region. The model depicted in Figure 9 has a thickness of about 600 ft and extends 5000 ft 3500 ft in the horizontal directions. The well configuration (Figure 10) consists of one vertical injection well and one long production well with angle of inclination varying from 70o to 80o. The wellbore is modeled using 62 segments. The inner wellbore diameter is 1/3 ft and the roughness is again 0.001 ft. The fine grid has about 85000 cells and the coarse grid consists of 17 24 7 (for a total of 2856) blocks plus the 62 segments. We consider incompressible flow, quadratic relative permeabilities (as above) and viscosities w= 0.3 cp and o= 3 cp. Densities are as in the previous example. The reservoir is initially filled with oil and water is injected at a constant rate. Two flow rates (with balanced injection and total production) are again considered: 4000 STB/d and 60000 STB/d. Results are presented at 0.12 pore volume injected.

5400 Wellbore pressure (psia) 5200 5000 4800 4600 4400 4200 1 2 3 4 5 6 7 8 9 10 11 12 Segment number
Figure 9: Heterogeneous permeability permeability displayed). field (logarithm of

GPRS Fine MsMFEM

Figure 7: Wellbore pressure profiles for high flow rate case (20000 STB/d).

SPE 106179

MsMFEM for Coupled Wellbore/Near-well Flow

5500 5000 Wellbore pressure (psia) 4500 4000 3500 3000 2500 2000 1500 1000 1 6 Fine - low rate MsMFEM - low rate Fine - high rate MsMFEM - high rate 11 16 21 26 31 36 41 46 51 56 61 Segment number
Figure 11: Wellbore pressure profiles for high (60000 STB/d) and low (4000 STB/d) flow rate cases.

Figure 10: Vertical injection well and inclined production well.

Figure 11 displays the wellbore pressures for the two flow rates. For both cases, the pressure drop given by MsMFEM matches the fine-scale solution essentially exactly, but an error in absolute pressure, although not excessive, is apparent for the high flow rate case. In fact, the error in pressure is close to proportional to the total reservoir pressure drop. The corresponding curves for in-situ oil fraction agree closely (Figure 12). We note that the abrupt behavior of the in-situ oil fraction curve towards the toe of the well (for the 4000 STB/d case) occurs because this part of the well penetrates a low permeability region. As a result the wellbore mixture velocity here is very small, and thus small amounts of water may cause large holdup. For this case, there is no water breakthrough at the last segment, so o=1 there. A few segments further towards the heel, however, the water inflow results in a large water holdup (and thus the abrupt drop in o). For the high flow rate case (60000 STB/d) this behavior is not observed. To gauge the variation in accuracy of the MsMFEM solution with increased coarsening, we now consider four different coarse grids for this case. The finest grid is 17 24 21 with an additional partitioning in the near-well region as in Figure 3 (left), giving a total of about 12000 blocks. The coarser grids contain 17 24 7 , 10 12 4 and 5 5 4 blocks, respectively. In Figure 13, the pressure profiles for the 4000 STB/d case are shown. The profiles for the three finest grids are fairly close, while the coarsest grid shows some discrepancies (although the pressure drop along the well continues to match the fine-grid solution). The corresponding in-situ oil fraction is depicted in Figure 14, where all the grids produce reasonable results, though again degradation in accuracy with increased coarsening is observed. For the high flow rate case (not shown), trends similar to those in Figures 13 and 14 are observed. These results indicate that, although the accuracy does degrade with increased coarsening (as would be expected), the method continues to provide sensible results even with very coarse grids.

1 0.9 0.8 In-situ oil fraction 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1

Fine - low rate MsMFEM - low rate Fine - high rate MsMFEM - high rate

6 11 16 21 26 31 36 41 46 51 56 61 Segment number

Figure 12: Wellbore in-situ oil fraction for high and low flow rate cases.

4950 Wellbore pressure (psia) 4900 4850 4800 4750 4700 4650 4600 1

Fine 17x24x21 17x24x7 10x12x4 5x5x4

6 11 16 21 26 31 36 41 46 51 56 61 Segment number

Figure 13: Wellbore pressure profiles for low flow rate case with varying degrees of coarsening.

S. Krogstad and L.J. Durlofsky

SPE 106179

1 0.9 0.8 In-situ oil fraction 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1

Fine 17x24x21 17x24x7 10x12x4 5x5x4

p q S t T trace u V, Vs x, x

6 11 16 21 26 31 36 41 46 51 56 61 Segment number

Figure 14: Wellbore in-situ oil fraction for low flow rate case with varying degrees of coarsening.

Concluding Remarks In this paper we have presented a coupled two-phase wellborereservoir flow model that is based on a multiscale mixed finite element method for reservoir flow linked to an oil-water driftflux model for wellbore flow. By resolving the wellbore trajectory in the fine grid, the methodology successfully incorporates near-well heterogeneity down to the scale of the wellbore and avoids the use of well indices. Through use of multiscale resolution the number of coarse blocks can be kept relatively low, enhancing the efficiency in solving the coupled pressure equations. The implementation was validated against an existing simulator and we presented numerical simulations involving vertical and deviated wells in a heterogeneous reservoir. The results demonstrate that we have successfully coupled wellbore and reservoir flow models within a multiscale finite element context. The MsMFEM model was shown to provide accurate results on significantly coarsened models in cases with substantial wellbore pressure variation. Future efforts should be directed toward increasing the degree of implicitness of the formulation and extending the procedure to three-phase flow. Acknowledgements We are grateful to S. Livescu (Stanford University) for running the comparison simulations in GPRS and to J. Natvig (SINTEF) for providing his code for the discontinuous Galerkin schemes with optimal ordering of elements. This work was funded in part by the Research Council of Norway under grant no. 162606/V30. Nomenclature C0 profile parameter in drift-flux model g gravitational acceleration k, k absolute permeability, scalar and tensor kr relative permeability m deviation multiplier MFEM mixed finite element method MsMFEM multiscale mixed finite element method n normal vector to interface or boundary

pressure source term water saturation time cell or block sum of diagonal entries of matrix Darcy velocity velocity and superficial velocity along wellbore coordinate, in reservoir and along wellbore in-situ wellbore volume fraction porosity pressure basis function interface between cells or blocks total mobility viscosity angle of deviation from vertical density flux basis function computational domain

Subscripts d m i, j ij o w

drift mixture index to cell or block referring to interface between cells/blocks i and j oil water

References
Aarnes, J. E. 2004. On the use of a mixed multiscale finite element method for greater flexibility and increased speed or improved accuracy in reservoir simulation. Multi. Mod. Sim. 2(3): 421439. Aarnes, J. E., Krogstad, S. and Lie, K.-A. 2006a. A hierarchical multiscale method for two-phase flow based upon mixed finite elements and nonuniform coarse grids, Multi. Mod. Sim. 5(2): 337-363. Aarnes, J. E., Krogstad, S. and Lie, K.-A. 2006b. Multiscale mixed/mimetic finite-element methods on corner-point grids. Submitted to Comp. Geosci. Arbogast, T. 2000. Numerical subgrid upscaling of two-phase flow in porous media. Numerical Treatment of Multiphase Flows in Porous Media, eds. Z. Chen et al., Lecture Notes in Physics, 552: 35-49. Brezzi, F. and Fortin, M. 1991. Mixed and hybrid finite element methods. Computational Mathematics, Springer-Verlag, NewYork. Brezzi, F., Lipnikov, K. and Shashkov, M. 2005a. Convergence of mimetic finite difference methods for diffusion problems on polyhedral meshes. SIAM J. Num. Anal. 43: 1872-1895. Brezzi, F., Lipnikov, K. and Simoncini, V. 2005b. A family of mimetic finite difference methods on polygonal and polyhedral meshes. Math. Models Methods Appl. Sci. 15: 1533-1553.

SPE 106179

MsMFEM for Coupled Wellbore/Near-well Flow

Cao, H. 2002. Development of techniques for general purpose simulators. PhD thesis, Stanford University. Chen, Z. and Hou, T. 2003. A mixed multiscale finite element method for elliptic problems with oscillating coefficients. Math. Comp. 72: 541-576. Chen, Z. and Yue, X. 2003. Numerical homogenization of well singularities in the flow transport through heterogeneous porous media. Multi. Mod. Sim. 1: 260-303. Christie, M. A. and Blunt M. J. 2001. Tenth SPE comparative solution project: a comparison of upscaling techniques. SPEREE 4(4): 308-317. Haaland, S. E. 1981. Simple and explicit formula for the friction factor in turbulent pipe flow including natural gas pipelines. IFAG B-131, Technical Report, Division of Aero- and Gas Dynamics, The Norwegian Institute of Technology, Norway. Hasan, A. R. and Kabir, C. S. 1999. A simplified model for oil/water flow in vertical and deviated wellbores. SPEPF (1): 56-62. Holmes, J. A. 1977. Description of the drift flux model in the LOCA Code RELAP-UK. I. Mech. E. paper 206/77, Proceedings of the Conference on Heat and Fluid Flow in Water Reactor Safety, Manchester, UK. Holmes, J. A., Barkve, T. and Lund, O. 1998. Application of a multisegment well model to simulate flow in advanced wells. SPE paper 50646 presented at the European Petroleum Conference, The Hague, 20-22 October. Hou, T. and Wu, X.-H. 1997. A multiscale finite element method for elliptic problems in composite materials and porous media. J. Comput. Phys. 134: 169-189. Hyman, J., Morel, J., Shashkov, M. and Steinberg, S. 2002. Mimetic finite difference methods for diffusion equations. Comp. Geosci. 6: 333-352. Hyman, J., Shashkov, M. and Steinberg, S. 1997. The numerical solution of diffusion problems in strongly heterogeneous nonisotropic materials. J. Comput. Phys. 204: 633-665. Jenny, P., Lee, S. H. and Tchelepi, H. 2003. Multi-scale finitevolume method for elliptic problems in subsurface flow simulation. J. Comput. Phys. 187(1): 47-67. Jenny, P., Lee, S. H. and Tchelepi, H. 2004. Adaptive multiscale finite-volume method for multiphase flow transport in porous media. Multi. Mod. Sim. 3(1): 50-64. Lunati, I. and Jenny, P. 2006. Multiscale finite-volume method for compressible multiphase flow in porous media. J. Comput. Phys. 216: 616-636. Natvig, J. R. and Lie, K.-A. 2006. Fast computation of multiphase flow in porous media by implicit discontinuous Galerkin schemes with optimal ordering of elements. Submitted. url: www.math.sintef.no/geoscale. Oddie, G., Shi, H., Durlofsky, L. J., Aziz, K., Pfeffer, B. and Holmes, J. A. 2003. Experimental study of two and three phase flows in large diameter inclined pipes. Int. J. Multiphase Flow 29: 527558.

Ouyang, L. B., Arbabi, S. and Aziz, K. 1998. A single-phase wellbore-flow model for horizontal, vertical, and slanted wells. SPEJ 3: 124-133. Shi, H., Holmes, J. A., Diaz, L. R., Durlofsky, L. J. and Aziz, K. 2005b. Drift-flux parameters for three-phase steady-state flow in wellbores. SPEJ 10 (2): 130-137. Shi, H., Holmes, J. A., Durlofsky, L. J., Aziz, K., Diaz, L. R., Alkaya, B. and Oddie, G. 2005a. Drift-flux modeling of two-phase flow in wellbores. SPEJ 10 (1): 24-33. Wolfsteiner, C., Lee, S. H. and Tchelepi, H. A. 2006. Well modeling in the multiscale finite volume method for subsurface flow simulation. Multi. Mod. Sim. 5: 900-917 Zuber, N. and Findlay, J. A. 1965. Average volumetric concentration in two-phase flow systems. J. Heat Transfer, Trans. ASME 87: 453-468.