SPE 147110

Numerical Modeling of Massive Sand Production

Jianlin Wang, David Yale, Ganesh Dasari, ExxonMobil Upstream Research Company

Copyright 2011, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Annual Technical Conference and Exhibition held in Denver, Colorado, USA, 30 October–2 November 2011.

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 have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper 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 SPE copyright.

Abstract

Primary production of heavy oils where significant sand is produced (generally referred to CHOPS for Cold Heavy Oil
Production with Sand) is one of the heavy oil recovery processes that have been used to recover bitumen from weakly
consolidated reservoirs. There has been significant debate over the exact mechanisms that allow the sand to be produced and
whether the process creates a radial zone of high permeability around the wellbore or more linear high permeability channels
(wormholes) out from the reservoir. Numerical modeling of the sand production process presents significant numerical
challenges mainly because it is a highly-coupled nonlinear process and requires constitutive models that are able to capture the
material instability. In this paper, we present a numerical approach to explicitly model the sand production process using the
Finite Element Method with Arbitrary Lagrangian Eulerian (ALE) formulation. A number of novel features have been
developed and integrated into the ALE formulation to tackle the challenges associated with sand production simulation, such
as Eulerian boundary, automatic adaptive remeshing, and advanced constitutive models. By combining these features, our
model enables stable coupling between fluid flow and geomechanics, and is capable of modeling large deformation and highly
nonlinear geomechanical behaviors. Some examples are provided to demonstrate the soundness, accuracy, and effectiveness of
the numerical model. For simulation of the CHOPS process, the model shows that a key aspect of the process is the shifting of
overburden stress from parts of the reservoir sand that fail and are produced to those where the sand has not failed. It is this
mechanism that allows wormholes to form in the numerical models. The models have also shown conditions which can be
induced in the reservoir to enhance formation of wormholes.

Introduction

Extracting bitumen from oil sand reservoirs generally leads to production of sand in methods such as Cold Heavy Oil Production
with Sand (CHOPS), Cyclic Steam Stimulation (CSS), Steam Assisted Gravity Drainage (SAGD), and other cold flow or
primary production processes. The amount of sand and water produced may vary from very small to large and it depends on
the type of method, stress-state within the reservoir, pressure drawdown and depletion. In cases of CSS and SAGD, sand
production is not desirable, while sand production is encouraged in cases of cold production processes such as CHOPS and
other primary production processes. In the CHOPS, production of some sand with the oil enhances the permeability of the
subsurface allowing higher oil production rates than would be obtainable from Darcy flow through the original reservoir sand.
Because the amounts of sand and water produced are very large in processes such as CHOPS, safety and optimization of such
complex large scale production require realistic simulation models. Modeling the sand production presents significant
numerical challenges because it is a highly coupled geomechanical-fluid flow process and also because the productions are
from weakly consolidated reservoirs where the material behaviors are very complex.

The very early methods (e.g. Salama and Venkatesh 1983) for predicting sand production are based on empirical relations,
which depend on fluid velocity, strength of the formation, grain size, etc. These methods are case specific and can not be
considered as general predictive methods. Many other researchers have also investigated the mechanisms for prediction of sand
production (Bratli and Risnes, 1981; Risnes and Bratli, 1981; Perkins and Weingarten, 1988; Veeken et al. 1991; Ramos 1994;
Van den Hoek et al, 1996, 2000). They identified compressive shear failure, tensile failure and erosion as the three main
mechanisms of sand production. Vadoulakis et al. (1996) presented some mathematical aspects of the piping and surface erosion
as related to sand production. These methods generally do not account for the stress redistribution inside the reservoir or the
interaction between the reservoir and the surrounding formation.
2 SPE 147110

During the last two decades, finite element methods have also been used to predict limited sand production (e.g. Morita et
al., 1989; Papamichos and Stavropoulou, 1996; Yi et al., 2005). Ong et al. (2007) presented a method for predicting the onset
of sand production in terms of critical drawdown pressure in high flow rate gas wells. The fluid flow principles are coupled
with a Mohr-Coulomb material model. Sand production was assumed to initiate when the drawdown pressure condition induces
tensile stresses. In all these simulations, rock volume around the wellbore experiencing at least one of the failures (compression,
shear and erosion) is computed as a function of time. The sand production rates are calculated by assuming that the failed sand
will be produced. However, none of these models simulate actual sand production. Instead, the failure of sand is considered as
an indication of sand production. Since these models do not consider removal of material, they cannot account for interaction
of failed/removed sand and the rest of the formation. Wan and Wang (2004) and Wang et al. (2005) presented a different
method to predict sand production based on mixture theory with erosion mechanics. This method assumes that mobilized sand
is one of the internal variables in the governing equations. This model also fails to simulate the removal of material from the
reservoir, but the sand production rate is calculated as an internal variable.

Existing methods as described above, from the very early empirical correlations the finite element models, are suitable
where the amount of sand produced is small. When sand production rates are large, it is critical to simulate the sand removal
process to account for interactions between produced sand, remaining reservoir sand, reservoir fluid, and the surrounding
(overburden/underburden) formations. Currently, no commercial software or published numerical methods can be readily used
to simulation this process. In this paper, we describe a numerical approach developed recently by the authors to explicitly model
the sand production process using the Finite Element Method. The approach is based on the Arbitrary Lagrangian Eulerian
(ALE) formulation (Longatte et al., 2003; Anderson et al., 2004), but the numerical model presented is beyond the standard
fluid flow in porous media flow model and is well beyond the boundary conditions and material models normally used for
reservoir simulation. With tight coupling between fluid flow and geomechanics, the model is capable of modeling large
deformation and highly nonlinear geomechanical behaviors. A number of novel features have been integrated into the finite
element formulation to tackle the challenges associated with sand production simulation, such as the Eulerian boundary
condition, automatic adaptive remeshing, and also advanced constitutive models to represent the material behavior in weakly
consolidated reservoirs (Dasari et al., 2009). By combining these features, the model we developed allows for explicit modeling
of sand failure, sand displacement, and removal of sand during production. Our model also enables the interactions among
reservoir sand and fluid, failed sand, and the surrounding formations, and thus can account for the pressure evolution and stress
redistribution during the sand production.

Although the model developed in this study is general enough to be used for a wide range of reservoir and near wellbore
studies involving sand production, it is especially useful for studying the complex behavior of wormhole production during
primary production of heavy oil or bitumen from unconsolidated reservoirs (or the CHOPS process). During CHOPS, a
sufficiently low pressure environment is created immediately around the wellbore such that sand is dragged into the wellbore
and produced with the bitumen. Sand fractions can be between 10% and 40% of the total volume of material produced for
periods of time. There has been significant debate over the exact mechanisms that allow the sand to be produced and whether
the process creates a radial zone of high permeability around the wellbore or more linear high permeability channels out from
the reservoir. Modeling and time-lapse seismic imaging of reservoirs strongly suggests that the sand often does not just come
from the near wellbore region but from significant distances from the wellbore, thus leading many to suppose the existence of
a channel or high porosity zone away from the wellbore referred to as a “wormhole”. The enhanced porosity and permeability
of the wormhole allows significantly higher oil production rates into the wellbore than would be possible considering the
original permeability of the reservoir and the high viscosity of the oil.

A number of authors have discussed potential ways that “wormholes” might develop during the CHOPS process (Geilikman
et al, 1994a, 1994b; Wang et al, 2001; Tremblay et al, 1996, 1998a, 1998b). Although these authors postulate that failure and
erosion of the sand from the inside of the wormhole due to the high viscosity flow of the oil through the wormhole is the
primary mechanism for wormhole development, they do not fully consider the geomechanical coupling and in-situ stress
implications of wormhole development and growth. There is a significant body of literature (Willson et al, 2002; Varziri et al,
2002; Papamichos et al, 2001; Risnes and Bratli, 1981) on how sand production in the near wellbore region may initiate due to
mechanical failure of the sand due to drawdown and fluid flow. However, most near wellbore sand production models focus
on failure of the sand immediately around the wellbore due to stress concentrations. Only a few papers (Risnes and Bratli,
1982; Vaziri, 1986) account for fluid drag forces in the equilibrium of momentum. But there has been little extension of those
theories into the much more sustained sand production from apparently far away from the wellbore that occurs in CHOPS.
There are conceptual models of a growing yielded zone around the wellbore due to sand production (Geilikman et. al., 1994)
but the yielded zone is definition is often limited to a similar form as in near wellbore failure which depends only on stresses
and pore pressures, not on how the drag forces of the fluid flow destabilize the sand matrix and lead to localized stress arching
and load shedding which we believe to be a central tenant to the limitation of most wormhole growth to be in one direction
from the wellbore.
SPE 147110 3

The numerical model we developed for reservoir processes like CHOPS shows the key elements of why and how sand is
produced during primary production of heavy oil and why it impacts that process as it does. In the rest of the paper, we will
describe the governing equations and solution algorithm of our numerical model, the novel features developed to simulation
sand production, as well as material model calibration using the laboratory measurements. Some examples are provided to
demonstrate the capabilities and effectiveness of the model. For simulation of the CHOPS process, the model shows how the
sand fails under fluid drag forces, and how overburden stress is shifted from parts of the reservoir sand that fail and are produced
to those where the sand has not failed. It is this key mechanism that allows wormholes to form in the numerical models. The
models have also shown conditions which can be induced in the reservoir to enhance formation of wormholes.

Governing Equations and the Solution Algorithms

The effective permeability to water of heavy oil formations is relatively low and the drawdown gradient is correspondingly
relatively high. Coupled analysis is required to account for loading induced by pore pressure gradient. According to Biot’s
poroelasticity theory, two basic equations need to be considered to simulate flow in porous media – the momentum balance
and the mass balance equations. For simplicity, these equations are written below for a saturated medium containing a single
fluid phase:
σ  ρg  0
κ         p u
(1)

  p  g   

 K K
    
 t t (2)


f
   f s 
Eq. (1) is the momentum balance equation, where σ is the total Cauchy stress tensor,  is the saturated mass density, and g
is the gravitational vector. The total stress σ in Eq. (1) can be defined as
σ  σ' p I (3)

where σ' is the effective stress on the rock skeleton, is the Biot’s constant, p is the pore pressure, and I is the identity
matrix. The effective stress σ' can be written in terms of the strain ε and the elasticity tensor D as
σ'  D ε (4)

where D is also called the rock’s constitutive matrix. For sand production problem, the material behavior is highly nonlinear
so that plasticity has to be introduced to fully define the constitutive relation of the rock, which will be described in detail later
in the paper. The mass density  in Eq. (1) is defined as
  (1) s    f (5)

where s and f are the solid and fluid densities respectively. Eq. (2) is the fluid mass balance equation, where  is the
permeability tensor,  is the fluid viscosity, Kf and Ks are the bulk modulus of the liquid and the solid particle, respectively,
and u is the displacement of the solid matrix. The divergence of the displacement field in Eq. (2) representing the pore volume
change can also be written in terms of the volumetric strain v as
 v   u (6)

It can be seen that the fluid flow and geomechanics are coupled through Eqs. (1) and (2). The effect of pore pressure on the
stress change is captured in Eqs. (1) and (3). The pore pressure is coupled with rock deformation induced pore volume change
in Eqs. (2) and (6). The rock deformation also affects the rock porosity and the permeability. Eq. (2) also assumes that Darcy’s
law is valid for fluid flow

q κ
    
p   g   v  v s  (7)
 f 
f
f

where q is the fluid flux, and vf and vs are the fluid and solid matrix velocities, respectively. It is worth noting that Darcy’s law
is written in terms of relative flux in Eq. (7)
4 when the solid is not moving, it is the standard Darcy’s law, both the sand
SPEthe
147110
fluid
are moving, the Darcy’s law is written in terms of relative fluid flux, or the slip velocity (vf - vs ) between the fluid and solid
matrix. When the solid is not moving, the slip velocity is equal to the fluid velocity, which represents the standard Darcy’s flow
condition. When the viscous drag force exceeds the solid material strength, the solid will fail and move with the fluid. In this
case, it is the slip velocity between the fluid and solid that creates the drag force.
SPE 147110 5

Finite element discretization of Eqs. (1) and (2) leads to a linear system of equations with solid phase displacement u and
the pore pressure p as the primary unknown variables to be solved in the finite element system, which can be written in the
following form (e.g. Zheng et al., 2004):
K u  Qp  f u
u p (8)
Hp  QT  S fp

t t 
T T T κ T     

where K  B DB d , Q   B  m N f d , H   B f B f d and S   N f 
  N f d,
K



  K s 
    f 
B  Nu and B f  N f ,

Nu and N f are the shape functions for the displacement and the pore pressure fields, respectively, and
f u and f p are the right hand side load vectors.
Please note that the geomechanical field in Eq. (8) is written in static form where only the stiffness matrix of the solid phase K
is involved. There is an alternative way of solving the geomechanical field using dynamic relaxation. Neglecting damping, the
first set of equations in (8) in dynamic equilibrium may be written as

M u&  K u  Q p  f u (9)

where M is the mass matrix and u& is acceleration. In this paper, we use the dynamic relaxation for the geomechanical field
because of its advantage in solving problems with unstable material responses.

The standard approach to solve the above finite element system is to use an implicit fully-coupled algorithm where the
mechanical and fluid flow fields are solved simultaneously and the global equilibrium at a given load level is achieved via
iteration. While the fully-coupled implicit algorithm is effective for many reservoirs depletion problems where the drawdown
gradient is relatively low and material behavior is rather stable, it is less suitable for non-stable material response such as sand
problem problems. In addition, the fully-implicit algorithm requires matrix inversion in each step, which leads to potentially
high CPU cost for large systems. For these considerations, we use explicit dynamic relaxation scheme for the geomechanical
field, where the solution is advanced in time based on locally evaluated quantities. In this case the time steps must be smaller
than the implicit scheme as the integration algorithm is only conditionally stable.

There are two solution schemes to couple the explicit geomechanical field with the flow field:
(1) Staggered solution with dynamic relaxation for geomechanical field and implicit integration of the flow field, or
Explicit/Implicit coupling
(2) Tightly coupled staggered solution with dynamic relaxation for geomechanical field and explicit integration of the
flow field, or Explicit/Explicit coupling.
The staggered coupling scheme is generally more efficient, however, it may require small coupling steps for certain application
with high strain rates where mechanical deformation cause large or rapid changes in pore pressure. Therefore, we use the tightly
coupled staggered solution scheme to solve sand production problems. With explicit dynamic relaxation for geomechanical
field with explicit integration of the flow field, it is suitable for both stable and non-stable material response and has potential
advantages for large systems through high efficiency parallelization. The Explicit/Explicit coupling procedures are described
in the following table.

Geomechanical Field Flow Field

Step 1. From time t to t+t, evaluate the mid-point
velocity (v t+t/2)
Step 2: Solve for the flow equations for pressure (pt+t)
t+t
Step 3: Compute the displacement (u ) and stres
(t+t)

Description of the Numerical Model

6 SPE 147110
Although explicit dynamic relaxation for geomechanical field can help the solution of unstable post-failure response, it is not
sufficient for problems with massive sand production, which necessitating some new features such as automatic adaptive
remeshing, Eulerian boundary, and advanced constitutive models. Rockfield’s finite element software ELFEN (Rockfield,
SPE 147110 7

2007) has some fundamental procedures necessary for simulation of the sand production, for example, the fracture energy based
straining softening regularization (Crook et al. 2003), tightly staggered coupling algorithm as described above, and the Arbitrary
Lagrangian-Eulerian (ALE) formation with adaptive remeshing. We choose ELFEN as the numerical platform to develop our
sand production model and have been working with Rockfield to enhance and optimize ELFEN’s capabilities on automatic
adaptive remeshing and results remapping, Explicit/Explicit coupling, and constitutive modeling. We have also developed
Eulerian Boundary condition and automatic flow control algorithms as the key parts of our numerical model and have integrated
these components into ELFEN to combine with the other features mentioned above for simulation of sand production. A brief
review of some key features in our numerical model is provided below.

Automatic Adaptive Remeshing

Production of large volumes of sand leads to excessive mesh distortion when a Lagrangian discretisation is used for the solid
matrix. While conceptually this could be overcome using an Eulerian formulation where the variables are expressed in terms
of Euclidian coordinates, this class of formulation is not ideally suited to resolving the evolution of localised damaged zones
with high relative movements, due to the difficulty in initially resolving and subsequently convecting the motion of the localised
damaged zones (Crook et al., 2003). To address this issue, an automatic adaptive remeshing technique is used in the current
model. In this technique, when mesh quality (e.g. measured by distortion) drops below a threshold, a new mesh is created
automatically from the deformed configuration. All the results from old mesh are mapped onto new mesh and subsequent
calculations are carried out on the new mesh. The automatic adaptive remeshing technique used in the current model allows
simulation of large movements of sand and fluid by eliminating the distortion of the elements. The following figure (Fig. 1)
provides a simple demonstration of using the adaptive remeshing procedure to simulate the Thick Wall Cylinder (TWC)
experiment, which can be utilized to understand the mechanisms of wellbore breakout.

(a) Initial unstructured mesh (b) Adaptive refinement in the zones (c) Simulation subsequent to gross
of high shear strain localization collapse through adaptive remeshing

Figure 1: Automatic adaptive remeshing for TWC simulation (Rockfield, 2007)

Eulerian Boundary

In sand production problem, there is considerable volume of solid materials transported into the wellbore. Although adaptive
remeshing can somehow remedy the solution breakdown, a numerical procedure is required to move the materials from the
wellbore; otherwise, the wellbore will be clogged with produced material and may reach an incorrect post-failure state. Eulerian
boundary condition is another important feature we introduced to simulate sand and water movements at the boundaries of the
wellbore.

The Eulerian boundaries are defined by a series of geometric lines in 2-D or geometric surfaces in 3-D. Material is then
allowed to freely flow through these geometric entities. After each analysis step the displacement of the boundary nodes
corresponding to these geometric entities is assessed, and if this exceeds a pre-defined value, all the nodes on the Eulerian
boundary are returned to their original position (as illustrated in Fig. 2). As repositioning of the nodes does not generate strain
in the elements adjacent to the Eulerian boundary there is a volume loss on an exit surface. Repositioning of the nodes adds to
the distortion of the elements adjacent to the Eulerian boundary and, in particular, elements adjacent to an exit boundary will
gradually decrease in volume. Consequently, the automatic adaptive remeshing procedure is used in conjunction with the
Eulerian boundaries to periodically trigger a re-mesh of the domain.
8 SPE 147110

Wellbore boundary

t=0

Figure 2: Demonstration of an exit Eulerian Boundary

The use of Eulerian boundary conditions at producer is described using Fig. 3. Typical initial mesh around wellbore is shown
in Fig. 3a. Parts of finite elements may enter into the producer due to various forces acting on them (Fig. 3b). The Eulerian
boundary condition at the producer then “absorbs” the parts of the elements that enter into the producer. The area/volume of
elements entered into producer is the sand produced at that time. The automatic adaptive remeshing then makes a new mesh
such that no sand is within the producer (Fig. 3c). With the Eulerian boundary condition, the model can automatically remove
sand that is produced into wellbore thereby decreasing computational effort to deal with failed sand and can also calculate
volumes and rates of produced while maintaining constant wellbore geometry.

(c) New mesh after removing the

(a) Initial mesh (b) Material enters into wellbore material that entered into wellbore
Figure 3: Finite element meshes around producer

Advanced Constitutive Models

Understanding of the geomechanical properties of oil sands is important for modeling and implementation of most heavy oil
and bitumen recovery processes. For example, the production of significant sand in CHOPS arises from material instability in
the wellbore region due to pore pressure gradient induced loading exceeding material strength, which causes unusual
deformation patterns in the subsurface. Simulating this process is complex necessitating constitutive models for weakly
cemented rock formation that are able to reproduce a physically realistic evolution of the material from the intact state to the
fully damaged state. In this study, we use a critical state based constitutive model, called Soft Rock Model, that was
originally proposed by Rockfield (Crook et al., 2003). This model was similar to the superior sand model developed by
Drescher and Mroz (1997) for simulation of sand liquefaction. We worked with Rockfield to further develop the Soft Rock
Model and calibrate the model for simulation of sand production.

As a critical state model, the Soft Rock model unifies the treatment of the shearing and consolidation properties to
provide a rational hardening or softening mode. However, the Soft Rock model extends the original Cam Clay critical state
model (Wood, 1990) in a number of ways in order to represent the experimentally observed response of rock formulation in
hydrocarbon applications, including the elastic law, the plastic yield and potential surfaces, as well as the hardening and
softening law. For the elastic law, empirical relations are developed for the Young’s modulus and Poisson’s Ratio based on
experimental data, which are written as
SPE 147110 9

  '
n

E  Eref 1 3  (10)

 A 
where Eref is the reference Young’s Modulus,  3 ' is the minimum principal stress (or the most tensile stress) , A and n are
material constants; and

1 e m ' 
 
 min  max  3
(11)
 min

where  min and  max are the values of the Poisson’s Ratio at  3 ' = 0 and at very high  3 ' , and m is the material constant.
As an illustration, the empirical relations expressed in Eqs. (10) and (11) are plotted in Fig. 4 and are compared with some
experimental measurements.

(a) Young’s Modulus (b) Poisson’s Ratio

Figure 4: Empirical relations for the elastic law

The yield function, or the failure criterion, of the Soft Rock model is a smooth three-invariant surface defined as:
 p' p 1/ n

F  p', q  q  g( , p')   p' pt tan 0 c  (12)
 pt  pc 

where p’ is the effective mean stress, q is the deviatoric stress,  is the Lode angle, pt is the tensile intercept of the yield surface
with the hydrostatic axis, pc is the pre-consolidation pressure or compressive tensile intercept of the yield surface with the
hydrostatic axis, 0 and n are material constants which define the shape of the yield surface in the p’-q plane and g( ,p’) is a
function that controls the shape of the yield function in the deviatoric plane (or the -plane), defined as

 1  sin(3 )   p 
g( , p')    and   0 exp1 p' c  (13)

 1    pc0 

where , 0 , 1 are material constants, and pc0 and pc are the initial and current pre-consolidation pressure, respectively.
The shape of the yield surface in the p’-q stress plan and in the -plane are plotted in Fig. 5. It can be seen from the figure
that the yield surface in the -plane is more circular at higher effective mean stress and more triangular at lower mean
effective stress. In the p’-q plane, the yield surfaces at different Lode angles are between the red and green curves in Fig. 5a,
which are cases corresponding to  = 30o and  = -30o , respectively.

The plastic potential surface, which defines the direction of plastic flow, is of the similar form as the yield surface defined
in Eq. (12), but has a different shape to take into account the non-associativity. One such function to define the potential surface
10
is Eq. (12) with the friction parameter 0 replaced by a dilation parameter 0. Detailed expressions are omittedSPE 147110
here. The
hardening/softening law of the Soft Rock model is a more general relationship between the pre-consolidation pressure pc (or
the yield cap size) and the plastic volumetric strain  p , which overcomes the limitation of the original hardening law in
the Cam Clay model. As an example, one such relation is plotted in Fig. 6 to show the variation of pc with the plastic
volumetric strain, with the detailed expressions omitted.
SPE 147110 11

 = 30o

pc  = -30o

(a) p’-q stress plane (the green and red lines denote  = 30o and -30o
respectively; the blue line is the critical state line at  = 30o ) (b) the -plane

Figure 5: Soft Rock model yield surface in the p’-q plane and the -plane

Figure 6: Hardening/Softening relation of the Soft Rock model

Another important component of the material model is permeability coupling with the volumetric strain. We have developed
relation for the permeability to water as a function of dilation. Using a model of water saturation increase as a function of
volumetric strain increase due to dilation coupled with relative permeability equations (i.e. permeability to water as function of
water saturation), we obtained the water permeability as a function of volumetric strain that reproduce the experimentally
observed behavior. The permeability model has described in detail along with experimental validations in another paper
published elsewhere (Yale at al., 2010).

Demonstration of Numerical Model Capabilities

Modeling of Sand Behavior under Injection

Many heavy oil recovery processes involve the injection of fluids or reduction in effective stresses to allow fluids to flow.
The sand behavior under injection has been studied experimentally (Yale et al., 2010), but very few models can fully capture
the behavior, particularly the stress paths and the volumetric responses. As a numerical example to validate the constitutive
models described in the paper, we simulate some of the injection tests of oil sand core samples conducted in our lab. In the
tests, core samples oriented with vertical direction from the field perpendicular to one of the radial directions of the core. The
core samples were initially loaded to the reservoir condition, followed by water injection under constant total confining stress
in the radial direction and no-strain boundary condition in the axial direction. These tests were described in (Yale et al. 2010).
12 SPE 147110

Fig. 7 shows both the numerical and experimental results for one of the Celtic sand core samples during injection. It is first
observed that the numerical results agree well with the experimental measurements. The stress paths in Fig. 7a reflect how the
differential stress changes with the reduction of effective mean stress. The no-strain boundary conditions in the axial direction
lead to the increase in differential stress because the vertical total stress increases lead to the vertical effective stress decreasing
more slowly than the horizontal effective stress. After the stresses hit the yield surface, the sand will start to fail with both
effective mean stress and deviatoric stress decreasing as the yield surface shrinks. The material continues to soften along with
volumetric dilation until the stresses reduce to nearly zero. During this process, dilation appears to increase roughly linearly
until failure, accelerating to nearly exponential dilation post failure (Fig. 7b). The good agreement between the numerical
results and experimental observations confirms that the constitutive models employed in this study are able to represent the
evolution of sand formation from an initially intact to a completely damaged state.

(a) Stress paths (b) Volumetric response

Figure 7: Numerical and experimental modeling of sand behavior under injection

Massive Sand Production Simulation

The example below is provided to demonstrate that the combined use of automatic adaptive remeshing, Eulerian Boundary
Condition, advanced constitutive models, and large-strain coupled geomechanics-fluid flow formulations with the dynamic
relaxation scheme allows for simulation of large-volume sand production. A hypothetical reservoir of 60m wide, 60m long and
10m thick with a producer in the center is considered. A quarter symmetrical model shown in Fig. 8, with appropriate boundary
conditions, is used to represent the reservoir, including the overburden and the underburden above and below the reservoir. Fig.
8a shows the initial finite element mesh, and Fig. 8b shows the initial material grid (in red and grey stripes) of in the horizontal
plane at the middle of the reservoir and two vertical planes along the symmetry boundaries. The analyses reported here were
carried out using the ELFEN (Rockfield, 2007) finite element software.

Overburden

Reservoir
z

Underburden x
y
(b) Initial material grids in the horizontal (xy)
(a) Initial finite element mesh plane and two vertical (xz and yz) planes

Figure 8: A quarter symmetrical finite element reservoir model

SPE 147110 13

As this example is only to demonstrate the applicability of the proposed numerical model, the details of the reservoir
condition are not very relevant and thus are omitted hear. Assume that the initial reservoir pressure is 4000 kPa and initial
permeability to water is 10 mD. The initial equilibrium was disturbed by decreasing pore pressure at the producer from 4000
kPa to 850 kPa. When the drag force due to pressure gradient exceeds the frictional resistance of the sand in place, the sand
will move toward the producer. The production of sand at the producer increases the porosity and thus the permeability of the
sand in the reservoir. Some of the simulation results at the early stage of the sand production (time = 20 days) are plotted in
Fig. 9. It is observed from the results that the high pressure gradient in the near wellbore region (Fig. 9a) induced localized
damage (Fig. 9b), which leads to progressive failure and mobilization of the sand. The pattern of the sand movement depends
on the pressure gradient, the reservoir stress state, and the mechanical properties of the sand (as described in the constitutive
models).

(a) (b)

Figure 9: Numerical results of (a) pore pressure and (b) the effective plastic strain

During the simulation, the initial finite mesh in Fig. 8a is continuously updated with new meshes generated in the adaptive
remeshing process. Consequently, the finite element meshes at any stages do not represent the actual deformation. The material
grids depicted in Fig. 8b are used to track the sand movement and to represent the accumulated displacement.

(a) (b)

(c) (d)

Figure 10: Material grids of the reservoir at different stages of sand production
14 SPE 147110

Fig. 10(a-d) shows a series of material grids of three cross sections of the reservoir at different stages of sand production. The
volume of the sand produced at any stage can be obtained by calculating the volume of the elements that have passed through
Eulerian boundary at the producer.

Modeling of Wormhole Development and Impact of Reservoir Conditioning

We have also attempted to deduce the potential mechanisms of wormhole grow during CHOPS using the full first principal
physics of fluid flow forces and grain-grain forces on the reservoir sand system as described earlier in this paper. Analysis of
the model shows that a given element of sand will move if the drag forces due to fluid flow exceed the frictional forces holding
the sand in place. In addition, a given element of sand can only move if the element of sand next to it in the direction of potential
movement is also moving. However, the calculation of the drag forces required to move the sand under most conditions of in-
situ stress would be very high unless an element of sand is very close to the wellbore. Thus there is the general observation that
most sand production problems are confined to the near wellbore region where the 1/r relationship between pressure gradient
and flow rate can lead to high fluid drag forces and the near wellbore stresses can either aid in the failure of the sand or at least
reduce one of the principal stresses due to the free surface boundary condition of an open wellbore. However, our model also
shows quite clearly, that defects or inhomogeneities that might lead to the start of sand production in the near wellbore region,
could also lead to a redistribution of stresses in the area which may lead to a stress shadowing that would sufficiently relieve
local stresses to allow a wormhole to propagate further away from the wellbore in a given direction. In addition, this stress
arching or stress shadowing would tend to allow only one or two wormholes to propagate rather than the dendritic pattern that
has been postulated by others (e.g. Wang et. al., 2001)

(a) (b)

(c) (d)

(e) (f)

Figure 11: Vertical stress redistribution in the reservoir during wormhole formation (Dark orange and blue
colors denote low and high effective stresses, respectively)
SPE 147110 15

The results from one of our earlier VISAGE models are shown in Fig. 11 (a-f). Since the purpose for this particular study
was to understand the mechanisms and patterns of wormhole growth rather than to explicitly simulate sand production, we did
not use the more sophisticated ELFEN model demonstrated in the earlier example for massive sand production simulation. In
this model there is a layer of reservoir sand 5 m thick with two shale layers 10m thick on the top and bottom of the sand. Fig.
11 shows just the reservoir layer. A single producer well is in the lower right corner and the horizontal dimensions of the model
are 30m by 30m. The properties of all the reservoir sand elements are the same, except that some elements along the right hand
boundary, lower boundary, and those along a 45 degree angle from the producer have friction angles of 20 deg, which are 10
degrees less than the friction angle of the remainder of the sand. The reservoir elements have a stress state that is representative
for a 450m deep oil sand reservoir (i.e. vertical stress 8000 kPa, horizontal stress 9000 kPa, and pore pressure 3600 kPa). The
production well is drawdown by 2000 kPa to start production. The sequence of Fig. 11 (a-f) shows the progression of sand
failure once drawdown is initiated. The observed changes in vertical effective stress of the elements along the wormhole path
show the load shedding from the mechanically failed to neighboring elements. The higher vertical stress on these neighboring
elements keeps them from failing and producing sand even though fluid flow rates through those elements are dramatically
increased (due to the increased permeability of the wormhole channel) and the low stress along one principal direction due to
the failed sand in the wormhole. It is interesting to note that the elements along the 45 degree angle from the producer do not
produce sand as the stress shadowing from the failed elements creates too much load on them.

As part of an investigation into processes that may help increase sand production and thus wormhole growth, a concept was
developed called “reservoir conditioning” (Yale et al., 2006). In this concept, reservoir pressure is first increased via cold water
injection to partially relieve the overburden stress on the sand. Subsequent production (of water, bitumen, gas, and some sand)
occurs in an environment where lower pressure gradients are needed to initiate sand production and therefore wormhole
development may be enhanced. Fig. 12 (a-e) shows a similar model to Fig. 11 with 2000 kPa drawdown pressure except that
the reservoir pressure in the sand layer has been increased to within 600 kPa of the overburden stress (i.e. 9000 kPa vertical
and horizontal stresses and 8400 kPa pore pressure).

(a) (b)

(d)
(c)

(e)

Figure 12: Vertical stress redistribution in the conditioned

reservoir during wormhole formation (Dark orange and blue
colors denote low and high effective stresses, respectively)
16 SPE 147110

The sequence of Fig. 12 (a-e) shows a more complex and extensive development of wormholes than in the previous case.
All three of the directions seeded with lower friction angle sands show wormhole growth and wormholes develop in orthogonal
directions to some of the seeded directions. Both these models are very coarse and were done to test the capabilities of the
numerical model to show sand motion deep into the reservoir, but they show an interesting potential method to increase
wormhole growth and thus cold production. What has not been fully investigated is whether the injection of water aids or
detracts from the initial sand production and wormhole development. The current model did take into account that the
permeability of the system to a water phase would be increased due to the conditioning. And since the process of sand
production depends on pressure gradient and not what fluid is flowing, we believe that water production could initiate sand
production. Once sufficient water production had occurred, oil production along the now established wormholes would likely
exceed the oil production along more limited wormholes from standard CHOPS.

Further work with more advanced ELFEN model described earlier is necessary to fully understand how to predict sand
production and oil production from a CHOPS or conditioning enhanced CHOPS process. However, the above model does show
aspects of wormhole development that have not been fully explored in the literature. Some authors have suggested that
wormholes grow especially well in thin sands that have stiffer shale layers above them giving credence to this concept that
wormhole development is at least aided by if not requiring load shedding to adjacent sand to propagate. But these preliminary
results suggest that there could be enhancements to the standard CHOPS process that could dramatically increase recoveries of
bitumen above the generally accepted 10% of OOIP that most cold flow wells achieve.

Conclusions

To overcome the numerical challenges associated with simulation of sand production, a finite element based numerical method
is developed through combined use of the automatic adaptive remeshing, Eulerian Boundary, and advanced constitutive models
within the coupled geomechanics-fluid flow formulation. An explicit dynamic relaxation solution scheme is used to handle the
unstable post-failure behavior involved in sand production. The numerical examples provided in the paper demonstrate that the
proposed solution procedure is able to capture the responses both prior to and subsequent to massive sand production in the
reservoir under any stress states. The model not only predicts the pore pressure and the stress evolution or redistribution within
the reservoir, the conditions of the onset of sand production, but also calculates the sand production volume as a function of
time. The model developed in this study is general enough to be used for a wide range of reservoir and near wellbore studies
involving sand production. It is especially useful for studying the complex behavior of wormhole production during CHOPS.

Acknowledgement

The authors would like to thank Rockfield for their support on ELFEN and thank ExxonMobil Upstream Research Company
management for the permission to publish this paper.

References

Anderson R.W., Elliott N.S., and Pember R.B. (2004). “An arbitrary Lagrangian-Eulerian method with adaptive mesh refinement for the
solution of the Euler equations,” Journal of Computational Physics, Vol. 199, 598–617.
Bratli R.K. and Risnes R. (1981). “Stability and failure of sand arches,” SPE Journal, 883-898.
Crook T., Willson S.M., Yu J.G., and Owen D.R.J. (2003). “Computational modeling of the localized deformation associated with borehole
breakout in quasi-brittle materials,” J. Petr. Soc. Eng., Vol. 38, 177-186.
Dasari G.R., Yale D.P., and Wang J. (2009). “Sand and fluid production and injection modeling methods,” International Patent No.
WO2010/059288A1, PCT/US2009/057720.
Drescher A. and Mroz Z. (1997). “A refined superior sand model,” In ‘Numerical Methods in Geomechanics NUMOG VI’, eds
S.Pietruszcak, G.N. Pande, Rotterdam, Balkema, 21-26.
Geilikman M.B., Dusseault M.B., and Dullien F.A. (1994) “Fluid Production Enhancement by Exploiting Sand Production,” SPE/DOE
27797, presented at SPE/DOE Ninth Symposium on Improved Oil Recovery, Tulsa, Oklahoma, April 1994.
Geilikman M.B. and Dusseault M.B. (1997). “Fluid rate enhancement from massive sand production in heavy-oil reservoirs,” J. Petr. Sci.
Eng., Vol. 17, 5-18.
Longatte E., Bendjeddou Z., and Souli M. (2003). “Application of arbitrary Lagrange Euler formulations to flow-induced vibration
problems,” Journal of Pressure Vessel Technology, Vol. 125, 411–417.
Morita N., Whitfill D.L., Fedde O.P., and Lovik T.H. (1989). “Realistic Sand-Production Prediction: Numerical Approach,” SPE
Production Engineering, February 1989, 25-33.
Ong S.H., Ramos G.G., and Zheng Z. (2007). “Method of predicting the on-set of formation of solid production in high-rate perforated and
open hole gas wells,” US Patent No. 7200539B2.
Papamichos E. and Stavropoulou M. (1996). “An erosion-mechanical model for sand production rate prediction,” Int. J. of Rock Mechanics
and Mining Sciences, Vol. 35, 531-532.
Perkins T.K. and Weingarten J.S. (1988). “Stability andf failure of spherical cavities in unconsolidated sand and weakly consolidated
rock,” SPE 18244, 63rd Annual Tech. Conf., Houston.
SPE 147110 17

Ramos G.G. (1994). “Sand production in vertical and horizontal wells in a friable sandstone formation, North Sea,” SPE28065.
Risnes R. and Bratli R.K. (1981) “Sand stresses around a wellbore,” SPE 9650, presented at Middle East Oil Technical Conference,
Manama, Bahrain, March 1981.
Rockfield (2007). “User manuals for ELFEN finite element software,” Rockfield Software Ltd, Swansea, UK.
Salama M.M. and Venkatesh E.S. (1983). “Evaluation of Erosional Velocity Limitations in Offshore Gas Wells,” OTC 4485, Proc. the 15th
Annual Offshore Technology Conference, Houston, Texas.
Tremblay B., Sedgwick G., and Forshner K. (1996) “Imaging of Sand Production in a Horizontal Sand Pack by X-Ray Computed
Tomography,” SPE Formation Evaluation, Vol. June 1996, 94-98, SPE 30248.
Tremblay B., Sedgwick G., and Forshner K. (1998) “Modelling of Sand Production from Wells on Primary Recovery,” Journal of
Canadian Petroleum Technology, Vol. 37, 41-50.
Van den Hoek P.J., Hertogh G.M.M., Kooijman A.P., de Bree P., Kenter C.J., and Papamichos E. (1996). “A new concept of sand
production prediction: Theory and laboratory experiments,” SPE36418. SPE Annual Tech. Conf., Denver, Colorado.
Van den Hoek P.J., Kooijman A.P., De Bree P.H., Kenter C.J., Sellmeyer C.J., and Willson S.M. (2000). “Mechanisms of downhole sand
cavity re-stabilisation in weakly consolidated sandstones,” SPE 65183.
Vardoulakis I., Stavropoulou M., and Papanastasiou P, (1996). “Hydro-mechanical aspects of the sand production problem,” Jounral of
Transport in Porous Media, Vol. 22, 225-244.
Vaziri H.H. (1986) “Mechanics of Fluid and Sand Production from Oil Sand Reservoirs,” Petroleum Society Paper 86-37-75, presented at
37th Annual Technical Meeting of the Petroleum Society of CIM, Calgary, Canada, June 1986.
Vaziri H., Xiao Y., and Palmer I. (2002) “Assessment of several sand prediction models with particular reference to HPHT wells,”
SPE/ISRM 78235, presented at SPE/ISRM Rock Mechanics Conference, Irving, Texas, October 2002.
Veeken C.A.M., Davies D.R., Kenter C.J., and Kooijman A.P. (1991). “Sand Production Prediction Review: Developing an Integrated
Approach,” SPE 22792, Proc. of 66th SPE Annual Technical Conference, Dallas, Texas.
Wang J., Settari A., Wan R., and Walters D. (2005), “Prediction of Volumetric Sand Production and Wellbore Stability Analysis of a Well
at Different Completion Schemes,” Alaska Rocks 2005, The 40th U.S. Symposium on Rock Mechanics (USRMS): Rock Mechanics for
Energy, Mineral and Infrastructure Development in the Northern Regions, held in Anchorage, Alaska, June 25-29, 2005
Wan R.G. and Wang J. (2004). “Analysis of sand production in unconsolidated oil sand using a coupled erosional-stress-deformation
model,” Journal of Canadian Petroleum Technology, Vol. 43, 47–53.
Wang Y., Chen C.C., and Dusseault M.B. (2001) “An Integrated Reservoir Model for Sand Production and Foamy Oil Flow During Cold
Heavy Oil Production,” SPE 69714, presented at the 2001 SPE International Thermal Operations and Heavy Oil Symposium, Margarita,
Venezuela, March 2001.
Willson S.M., Moschovidis Z.A., Cameron J.R., and Palmer I.D. (2002) “New Model for Predicting the Rate of Sand Production,”
presented at SPE/ISRM Rock Mechanics Conference, Irving, Texas, October 2002.
Wood D.M. (1990). Soil Behavior and Critical State Soil Mechanics. Cambridge University Press.
Yale D.P., Boone T.J., and Smith R.J. (2006). “Application of Reservoir Conditioning in Petroleum Reservoirs,” International Patent No.
WO2009/042333A1, PCT/US2008/074342.
Yale D.P., Mayer T., and Wang J. (2010). “Geomechanics of oil sands under injection,” In Proc. of the 44 th U.S. Rock Mechanics
Symposium, Salt Lake City, Utah, June 2010.
Yi X., Valko P.P., and Russell J.E. (2005). ”Effect of rock strength criterion on the predicted onset of sand production,” Int. J. of
Geomechanics, Vol. 5, 66-73.
Zheng Y., Burridge R., and Burns D. (2004). “Reservoir simulation with the finite element method using Biot poroelastic approach,” MIT
Research Report.