Reservoir Geomechanics VISAGE 1726681270

VISAGE
Version 2018.1
Technical Description
VISAGE Technical Description
Table of Contents
Introduction ....................................................................................................................... 1
Applications to reservoir engineering ...................................................................................................... 1
Main features of VISAGE ........................................................................................................................... 1
Essential features ..................................................................................................................................... 1
Materials library ........................................................................................................................................ 2
Special features ........................................................................................................................................ 2
Reservoir engineering applications ......................................................................................................... 2
Related documentation .............................................................................................................................. 3
Scientific background ...................................................................................................... 4
Modeling elasticity ............................................................................................................ 6

3D elasticity theory .................................................................................................................................... 6
Finite element theory ................................................................................................................................. 8
Modeling of plasticity ..................................................................................................... 12

Physical determination ............................................................................................................................ 12
Visco–plastic algorithm ........................................................................................................................... 13
Visco–plasticity ....................................................................................................................................... 14
Convergence criterion ............................................................................................................................. 16
Visco-plastic time step selection ............................................................................................................ 17
Discontinuity model ........................................................................................................ 19

Equivalent material .................................................................................................................................. 19
Constitutive equations of rock mass ..................................................................................................... 19
Elasticity matrix of jointed rock mass .................................................................................................... 20
Nonlinear behavior of jointed rock mass ............................................................................................... 21
Material Constitutive Models ......................................................................................... 23

Elastic material properties ...................................................................................................................... 23
Orthotropic material ................................................................................................................................ 23
Isotropic material .................................................................................................................................... 24
Orientation of the local coordinate system .............................................................................................. 24
Undrained Model ...................................................................................................................................... 26
Invariant yield criteria .............................................................................................................................. 27
Mohr–Coulomb criteria ........................................................................................................................... 28
Drucker–Prager criteria ........................................................................................................................... 30
Von Mises criteria ................................................................................................................................... 31
i
Tresca criteria ......................................................................................................................................... 32

Critical state criteria ................................................................................................................................ 33
Chalk model ............................................................................................................................................ 34
ISAMGEO chalk model ........................................................................................................................... 37
Modified Drucker–Prager model ............................................................................................................. 37
Discontinuity yield criteria ...................................................................................................................... 38
Mohr–Coulomb criteria ........................................................................................................................... 38
Creep Model .............................................................................................................................................. 40
ISAMGEO Chalk Model ................................................................................................... 42

Stress invariants and derivatives ........................................................................................................... 42
Elastic stress strain behavior ................................................................................................................. 44
Elastoplastic stress strain behavior ....................................................................................................... 47
Yield surface ........................................................................................................................................... 47
Flow rule and plastic potential ................................................................................................................ 48
Hardening and softening ......................................................................................................................... 49
The ISAMGEO Chalk Model without rate dependency ......................................................................... 50
Elastic behavior ...................................................................................................................................... 50
Yield surface ........................................................................................................................................... 50
Tensile failure criterion ............................................................................................................................ 52
Yield surface with cohesion .................................................................................................................... 53
Plastic potential ....................................................................................................................................... 54
Special cases θ = -30° and θ = 30° ........................................................................................................ 54
Variation of the cap geometry ................................................................................................................. 55
The ISAMGEO Chalk Model with rate dependency ............................................................................... 55
Basic ideas ............................................................................................................................................. 55
Evaluation of the elastoplastic matrix [Dep] ............................................................................................ 56
Stress correction to the yield surface ...................................................................................................... 58
Rate independent behavior ................................................................................................................. 59
Rate dependent behavior .................................................................................................................... 60
The plastic multiplier λ is evaluated as a negative number ................................................................. 63
Radial return in case of shear failure .................................................................................................. 63
Unloading to elastic behavior .............................................................................................................. 63
In-situ state ............................................................................................................................................... 63

Porosity update ........................................................................................................................................ 64
Water weakening ...................................................................................................................................... 65
References ................................................................................................................................................ 66
Coupled porous media ................................................................................................... 67

Coupled hydro-mechanical equations ................................................................................................... 67
One-way coupled model .......................................................................................................................... 69
Two-way coupled model with permeability updating ........................................................................... 70
ii
Permeability updating methods .............................................................................................................. 71

Two-way coupled model with pore volume updating ........................................................................... 74
Dual Porosities/Permeabilities ...................................................................................... 79

Micro-macro relations of fractured poroelastic medium ...................................................................... 80
Arbitrary planar cracks ............................................................................................................................ 81
Extension to multiple fracture sets ........................................................................................................ 82
Implementation in VISAGE ...................................................................................................................... 82
Equation solvers ............................................................................................................. 84

Linear equation solution techniques ...................................................................................................... 84
Guidelines to finite element modeling .......................................................................... 86

Idealization ................................................................................................................................................ 86
Subdividing the structure ........................................................................................................................ 86
Mesh size .................................................................................................................................................. 87
Element types ........................................................................................................................................... 87
Element shape and size ........................................................................................................................... 87
Linear solver solution .............................................................................................................................. 88
Singularities .............................................................................................................................................. 88
Loading ..................................................................................................................................................... 89
Final notes ................................................................................................................................................ 89
References and bibliography ..............................................................................................

References ................................................................................................................................................ 90
Bibliography ............................................................................................................................................. 91
Further reading ......................................................................................................................................... 91
iii
iv
Introduction
Petrel Reservoir Geomechanics provides a user-interface for the generation, simulation management and results viewing
for the VISAGE finite element geomechanics simulator, with VISAGE performing the numerical calculations of rock
stresses, displacements and failure.
VISAGE can couple to the ECLIPSE industry reference reservoir simulator or INTERSECT, to perform stress- and strain-
dependent reservoir simulations.
This technical manual describes the numerical calculations performed by VISAGE within its Petrel-enabled workflows.
These workflows are self-contained in the Petrel user-interface, which handles all the interaction with VISAGE and
between VISAGE and ECLIPSE/INTERSECT, this removes the need to use VISAGE keywords. The Petrel-enabled
workflow is described in Petrel Help -> Help centre -> Geomechanics.
VISAGE is currently available on several hardware platforms and operating systems, including Microsoft Windows and
Linux. Parallel versions of this software for both CPUs and GPUs are available, enabling effective and efficient solving of
large problems; the parallel option for CPU is based on a distributed memory architecture using MPI (Message Passing
Interface).
Applications to reservoir engineering

Reservoir Geomechanics workflows consist of the VISAGE finite element simulator and Petrel Geomechanics. It offers a
robust, versatile, flexible and user-controlled tool for solving a variety of complex engineering problems encountered in
the oil and gas industry. Fast and efficient, the system can be employed to predict subsidence, compaction and pore
collapse due to high pressure draw-down during production operations. During water injection, micro fracture initiation
and propagation, induced by thermal gradients and dynamic changes in the effective stress state, can be predicted,
monitored and assessed. Post-frac productivity indices for vertical fracture completion can be calculated. An extensive
non-linear material library provides a powerful tool for studying reservoir geomechanics problems. Exciting opportunities
now exist for reservoir engineers to study the effect, on preferred water flood directionality, of thermal gradients, stress
magnitude and orientation, now recognized as key parameters for successful reservoir management. Coupled two and
three phase reservoir simulations can be performed, linking VISAGE to ECLIPSE. The new simulator integrates the
disciplines of rock mechanics and petroleum engineering to assess the effect of porous media deformation on fluid flow
characteristics. Sophisticated 3D reservoir models with complex predefined distributions of faults are readily
accommodated.
During waterflooding, faults and fractures may become conduits of flow or indeed transmissibility barriers if sealing
occurs. Furthermore, the evolution of fractures is constantly traced with hydraulic parameters being updated.
Incorporating experimental data obtained from core samples to update permeabilities and rock fabric characteristics, as
fracturing develops, well orientations and locations can be optimized.
Petrel Geomechanics and VISAGE together provide a unique tool accounting for changes in the effective stress state,
neglected by most commercially available simulators.
Main features of VISAGE

The main features are outlined below.
Essential features
• Modular design.
• Large 3D grids.
• Faults and Fractures
Introduction
1
• Speed.
• Flexibility and efficiency.
Materials library
• Isotropic elasticity.
• Transverse Anisotropic elasticity.
• Mohr–Coulomb.
• Von Mises.
• Tresca.
• Drucker–Prager.
• Modified Drucker–Prager with Cam Clay potential surface.
• Critical State using Modified Cam Clay Model.
• Chalk model.
• ISAMGEO chalk model with and without rate dependency.
• Mohr–Coulomb for fractures and faults.
Special features
• User friendly.
• Materials library.
• Preprocessors.
• Postprocessors.
• Propriety interfaces.
• Parallel versions using distributed memory.
• Hardware independent – VISAGE finite element simulator.
Reservoir engineering applications

Examples of applications for Petrel Reservoir Geomechanics:
• Waterflooding.
• Subsidence.
• Compaction.
• Simulation.
• Thermal fracturing.
• Thermal recovery.
• Reservoir integrity.
• Well planning and optimization.
• Well survivability.
• Gas storage.
Introduction
2
Related documentation
The following manuals are intended to be used in conjunction with this manual:
• ECLIPSE Reference Manual.
• Petrel Help —> Reservoir Geomechanics.
Introduction
3
Scientific background
The science of geomechanics provides an extensive field for research, especially with recent advances in technology.
New, complicated, experimental devices and faster and more powerful computers have assisted in the venture, yielding
improved and elegant solutions to geotechnical problems. Among the numerical techniques that have emerged as an
adequate means for the solution of geomechanics problems, is the finite element method. However, the development of
finite element techniques in geomechanics has remained largely in the academic environment with minimal practical
application. This is due either to uncertainties in the basic finite element techniques in describing soil/rock behavior or the
development of complex numerical algorithms to approximate various soil/rock aspects that were not easily understood.
A finite element system, the VISAGE finite element simulator, which is applicable to most geotechnical problems
encountered in engineering practices, has now increased the engineer's confidence in obtaining complex solutions
efficiently. Thus in determining quantitative results from various types of analyses, a better understanding of the behavior
of structures can now be obtained.
The concept of a numerical approach that will reproduce the stress/strain/volume change behavior of a geomass under
three-dimensional stress states, arbitrary loading paths and drainage conditions, is still a long way off, despite the
extensive research efforts and the development of sophisticated constitutive models. Even if such an approach existed, it
would be unwieldy and too complex. It is therefore necessary to isolate the salient features that govern the soil/rock
behavior in a particular problem.
Generally, soils/rocks behave in a more complicated manner than a simple elastic theory can predict. The main physical
feature of this behavior is the irrecoverability of strains. Consider in the diagram below, a typical stress–strain curve
obtained by a drained triaxial test.
The numerical interpretation of the relationships between stresses and strains are complicated functions depending on
the coefficients selected to represent the soil/rock behavior. In addition soils/rocks are far from uniform and variations in
properties occur from stratum to stratum and from site to site. This variability makes analysis in geomechanics difficult
and somewhat subjective.
4
Figure 2.1. Stress–Strain Behavior of Soil/Rock
It has been shown in the past that simple nonlinear numerical models can take into account stress–strain relationships
and therefore predict soil/rock behavior better than most elastic models.
The simplest nonlinear stress-strain behavior available is the elastic-perfectly plastic model. In this theory, a yield surface
separates stress states, which give rise to both elastic and plastic (irrecoverable) strains. More accurately, this means
that soils/rocks behave elastically until a failure criterion is violated, at which point, plastic behavior occurs.
Numerous techniques are available for the solution of elasto-plastic problems, amongst them ; initial stress, initial strain
and tangential methods. Any of these numerical approaches can be use, to successfully calculate the response to yield
criteria such as Mohr–Coulomb, Drucker Prager and critical state models. These yield criteria take into account plastic
constitutive properties such as the cohesion strength, c ′ , and the friction angle, φ . Critical state criteria can account for
yielding under compression by using cap models.
A representative of initial strain methods is the visco–plastic method. It is the visco–plastic approach that is used
predominately in the VISAGE finite element simulator, since it provides all the necessary attributes required the steady-
state and transient behavior of soil/rock masses.
The tangential method is used for the ISAMGEO chalk model.
5
Modeling elasticity
This chapter gives an overview of elasticity theory and its implementation in the finite element method.
It describes the elasticity constitutive, equilibrium laws and compatibility equations in 3D in terms of small strain solid
elasticity. A later chapter extends this to solving fully coupled porous media flow.
In VISAGE compressive stresses are taken as negative.
Note: However it should be noted that compressive stresses are taken as positive in Petrel Reservoir Geomechanics.
This applies to both the input data and results.
3D elasticity theory
Stress and strain are second order symmetric tensors. Due to their symmetry they can be written in Voigt notation.
σ xx
{}
σ yy
σ zz
σ = {σ } = , or Eq. 3.1
τ xy
τ yz
τ zx
ϵ xx
{}
ϵ yy
ϵ zz
ϵ = {ϵ } = , or ϵ ij Eq. 3.2
γ xy
γ yz
γ zx
where σ xx , and so on are the normal stresses, τ xy etc. are the shear stresses, ϵ xx etc. are the normal strains and γ xy
and so on, are the engineering shear strains (see below). This means that the strain as written in not strictly a tensor
quantity.
Modeling elasticity
6
Figure 3.1. Normal and Shear Stresses
Constitutive Law for Linear Elasticity – Stress Strain relationship is given by Hooke's law and D is the fourth order
stiffness tensor of 81 elastic moduli. However due to symmetry and the existence of a strain-energy density function, the
maximum number of independent constants reduces to 21, which is the maximum degree of anisotropy. Again writing in
Voigt notation we have :
σ xx C 11 C 12 C 13 C 14 C 15 C 16 ϵ xx
{} {}
σ yy C 21 C 22 C 23 C 24 C 25 C 26 ϵ yy
σ zz C 31 C 32 C 33 C 34 C 35 C 36 ϵ zz
= , or σ = Dε Eq. 3.3
τ xy C 41 C 42 C 43 C 44 C 45 C 46 γ xy
τ yz C 51 C 52 C 53 C 54 C 55 C 56 γ yz
τ zx C 61 C 62 C 63 C 64 C 65 C 66 γ zx
Representing a displacement field by the vector:
u (x , y , z )
u=
{ v (x , y , z )
w (x , y , z )
} Eq. 3.4
and compatibility equations relate the strain to displacements by:
1 ∂ ui ∂ uj
ε ij = (
2 ∂ xj
+
∂ xi )
, (i , j = x , y , x ) ; or ε ij =
1
2 ( i ,j
u + u j ,i )
or
Eq. 3.5
∂u ∂v ∂w
εx = , εy = , εz =
∂x ∂y ∂z
∂v ∂u ∂w ∂v ∂u ∂w
γ xy = + , γ yz = + , γ zx = +
∂x ∂y ∂y ∂z ∂z ∂x
Modeling elasticity
7
Where the last set of equations use the engineering shear strains. Note the factor of 1/2 difference between the tensor
shear strains.
Equilibrium equations relate the external forces on a body to the internal forces (stresses)
∂ σx ∂ τ xy ∂ τ xz
+ + + fx = 0
∂x ∂y ∂z
∂ τ yx ∂ σy ∂ τ yz
+ + + fy = 0 , or σ ij ,j +f Eq. 3.6
∂x ∂y ∂z i =0
∂ τ zx ∂ τ zy ∂ σz
+ + + fz = 0
∂x ∂y ∂z
On the boundary Γ can divide into two parts — traction and displacement boundaries, as shown below.
−
u i = u i , on Γ u ( displacement condition )
−
t i = t i , on Γ σ ( traction condition ) Eq. 3.7
(where traction t i = σ ij n j )
Figure 3.2. Boundary Conditions
Solving equations the equilibrium, constitutive and compatibility with the boundary conditions will provide the stress,
strain and displacement fields (15 equations in 15 unknowns for 3D problems). These are difficult to solve analytically so
numerical methods are used.
Finite element theory

The finite element method is a numerical technique for solving problems that are described by partial differential
equations. These can be formulated using functional minimization. A domain of interest is discretized locally by small
regions of arbitrary shapes called finite elements. Hence a continuous physical problem is transformed into a discretized
finite element problem with unknown nodal values. Summing up over the element equations results in a global matrix
which will be solved using numerical linear algebra. Stresses and strains are calculated at integration (gauss) points
inside finite elements using the nodal values and shape functions.
Using the principal of minimum potential energy where:
Total Potential energy (E) = Strain energy (W) - work done by applied loads (L)
a displacement field can be found that minimizes the potential energy.
Modeling elasticity
8
Consider one arbitrary finite element:
δΔE = δΔW − δΔL = 0
ΔW =
1
2 ∫
{Δϵ } T {Δσ } d Vol
vol Eq. 3.8
ΔL = ∫ {Δd } {ΔF } d Vol + ∫ {Δd } {ΔT } d Srf

vol
T
Srf
T
where F are body forces and T are boundary traction forces.

Displacements inside a finite element can be determined with the use of shape functions [N] and the nodal
displacements di
{Δd } = N {Δd } i Eq. 3.9
Strains can be found from the compatibility equations and the nodal displacements
{Δϵ } = B {Δd } i
Eq. 3.10
where B = derivatives of N
{Δσ } = D {Δϵ } Eq. 3.11
Consider the potential energy in one element. Using the constitutive equation to remove stress
ΔE =
1
2 ∫ {Δϵ }
vol
T
D {Δϵ } d Vol − ∫ {Δd } {ΔF }dVol − ∫ {Δd } {ΔT }dSrf
vol
T
Srf
T
Eq. 3.12
and the compatibility equations and the shape functions to replace the variations in displacements by the nodal values
ΔE =
1
2
vol
∫ {Δd } T
n B T
D B {Δd } n d Vol − ∫ {Δd }
vol
T
n N T
{ΔF } dVol
Eq. 3.13
− ∫
Srf
{Δd } Tn N T
{ΔT }dSrf
Minimizing the potential energy gives the equilibrium equation for each finite element
δΔE =
{Δd } Tn
vol
∫ B T
D B {Δd } n
d Vol − ∫
vol
N T
{ΔF }dVol − ∫ Srf
N T
{ΔT }dSrf = 0
Eq. 3.14
−
Srf
∫ N T
{ΔT }dSrf = 0
This can be written as
Modeling elasticity
9
k E {Δd } n = {Δf E }
kE = ∫
vol
B T
D B d Vol (Stiffness matrix )
Eq. 3.15
{Δf E } = ∫
vol
N T
{ΔF }dVol − ∫ Srf
N T
{ΔT }dSrf (Right hand side load )
The individual element equilibrium equations are assembled into the global equilibrium equations.
To solve general geometries, elements must be able to assume general shapes. To accomplish this, elements of simple
shapes are used along with local coordinate systems. Relationships are then established between the local and global
coordinates for each element. The same shape functions used for the interpolation of the displacement field are used to
specify the relationship between global and local coordinate systems, then the elements are known as isoparametric.
{d } = N {d } i
Eq. 3.16
{x } = N {x } i
Shape functions [N] are specified in the local coordinates {ξ , η , ζ }.
Figure 3.3. Gauss Point Local Axes
Not only are the shape functions required but their derivatives in the compatibility matrix [B].
∂ Ni / ∂ x 0 0
0 ∂ Ni / ∂ y 0
0 0 ∂ Ni / ∂ z
B = Eq. 3.17
∂ Ni / ∂ y ∂ Ni / ∂ x 0
0 ∂ Ni / ∂ z ∂ Ni / ∂ y
∂ Ni / ∂ z 0 ∂ Ni / ∂ x
Derivatives can be converted from local to global coordinate systems by means of the partial differentiation rule
Modeling elasticity
10
∂ ∂x ∂y ∂z ∂ ∂
{} {}{}
∂ξ ∂ξ ∂ξ ∂ξ ∂x ∂x
∂ ∂x ∂y ∂z ∂ ∂
= =J Eq. 3.18
∂η ∂η ∂η ∂η ∂y ∂y
∂ ∂x ∂y ∂z ∂ ∂
∂ζ ∂ζ ∂ζ ∂ζ ∂z ∂z
where J is the Jacobian. The inverse of J is used to calculate the derivatives in global coordinate as needed by matrix [B]
∂ ∂
{} {}
∂x ∂ξ
∂ ∂
= J −1 Eq. 3.19
∂y ∂η
∂ ∂
∂z ∂ζ
The determinant of the Jacobian is also used in the transformed integrals from the global to local coordinate system
∫∫∫ d x d y d z = ∫ ∫ ∫
1 1 1
det | J | d ξ d η d ζ Eq. 3.20
−1 −1 −1
Integration of expressions for stiffness matrices and load vectors cannot be performed analytically for a general case of
isoparametric elements. Instead, stiffness matrices and load vectors are typically evaluated numerically using Gauss
quadrature over quadrilateral regions. The Gauss quadrature formula for the volume integral in the three-dimensional
case is of the form:
∫∫∫
1 1 1 n n n
f (ξ , η , ζ ) d ξ d η d ζ ≃ ∑ ∑ ∑ w i w j w k f (ξ i , η j , ζ k ) Eq. 3.21
−1 −1 −1
i =1 j =1 k =1
where wi , wj and wk are weighting coefficients and ξi , ηj and ζ k are coordinate positions (abscissae) within the
element. Typical values for n are 1, 2 and 3. Evaluation of stresses and strains are also calculated at these gauss points.
Modeling elasticity
11
Modeling of plasticity
When an increment of stress is applied to a material, two types of strain may occur: one is elastic and reversible; the
other is plastic and irreversible. Based on St. Venant’s theorem, the principal axes of the increments of plastic strain,
should coincide with the principal axes of stress and not of stress increments. This behavior contrasts sharply with the
behavior of elastic materials. Here the principal axes of the increments of strain coincide with the principal axes of stress
increments and not of stress. Thus, the basic nature of plastic strain is quite different from elastic strain. Soils and rocks
behave like an elastic material at low stress levels and like a plastic material at high stress levels. Hence, when a soil/
rock stress–strain condition is modeled it should take into account the plastic nature of the strain at high stresses.
Various procedures utilizing the finite element method have been used successfully for elasto–plastic problems. The
techniques can be roughly divided into two groups. The first consists of techniques where the element stiffness matrix
remains constant and the loads only are updated during the iterations. In the second group the stiffness are recomputed
at each load step and/or iteration.
Within the first group are the initial stress and initial strain methods. Variable stiffness methods include the tangential and
secant stiffness methods.
The first approach is where the stress–strain relationship in every load increment is adjusted to take plastic deformations
into account. With a properly specified elasto–plastic matrix, this incremental elasticity approach can successfully treat
perfect plasticity, as well as hardening plasticity. In this approach – referred to as the initial stress process – increments
of strain, prescribe the stress system uniquely.
The second approach falls into the initial strain family of processes. In this, during an increment of loading, the increase
of plastic strain is calculated and treated as an initial strain, for which the elastic stress distribution is adjusted. The
visco–plastic approach is the most important of this type of process and is detailed in this chapter.
See also the ISAMGEO Chalk model chapter for details of the solution to elasto-plastic problems using a tangential
method.
Physical determination
In the visco–plastic approach, it is assumed that the only instantaneous strains that can be produced by stresses are
elastic. A time-dependent strain is added and its rate depends on the amount by which the function of the stresses
exceeds a yield value.
To illustrate the situation conceptually, a uniaxial model is introduced. In this model the plastic component can only
become active if σ > ϒ, in which σ is the total actual stress applied and ϒ is the yield value, defined by the yield
criterion. On instantaneous load application, only elastic straining of the spring takes place. The excess ( σ − ϒ), is
taken by the dashpot. This results in some strain, which is time dependent. In a typical model like this, different
modifications can be made, to represent more complex material behavior. For example, the yield value ϒ, can easily
become strain-dependent and also the dashpot and spring characteristics can easily be made nonlinear. This simple
uniaxial conceptual situation can be generalized to a multi-axial case, by placing more than one visco–plastic models in
series and/or parallel.
12
Figure 4.1. Basic one-dimensional elastic–visco–plastic model
This visco–plastic model can easily be degenerated in to an initial stress model, by omitting the dashpot component. In
this manner, a multi-axial situation can be produced by placing a series of springs in series. Thus, there is a fundamental
difference between the two models: the total stress of the visco–plastic model can exceed the yield value
instantaneously, by any desired amount. This has been noticed in many experiments.
In both models, the excess stresses σ — ϒ are maintained by a set of body forces, that are in equilibrium with the
initial system. At this stage of the calculation, the system of body forces can be removed by allowing the structure, which
maintains its elastic properties unchanged, to deform further, thus adding new stresses and strains to the existing set.
Once again, these are likely to exceed the yield value, and the process, therefore, has to be repeated. If the process
converges, nonlinear compatibility and equilibrium conditions will be satisfied.
Visco–plastic algorithm
Having described the visco–plastic approach, the next concern is to implement visco–plasticity, in a way that will
guarantee compatibility. This compatibility will assure the fact, that safe conclusions are drawn for the way in which this
method satisfies both plasticity and equilibrium.
The sequence illustrated below, is independent of the finite element type used in the analysis.
Starting from a converged stress and/or strains at each integration point existing in the finite element mesh:
{ σ o } = σ xo , σ yo , σ zo , σ xyo , σ yzo , σ zxo T Eq. 4.1
{ ϵ o } = ϵ xo , ϵ yo , ϵ zo , ϵ xyo , ϵ yzo , ϵ zxo T
Form the stiffness matrix for each finite element, by integrating over its volume, using all the necessary integration points
to satisfy the incompressibility theorem:
KE = ∫ v B T
D B dV Eq. 4.2
where, B is the strain–displacement matrix, resulting at each integration point and D is the elastic stress-strain
matrix.
13
Assemble all the element stiffness matrices KE , into the global stiffness matrix K .
Apply the external force, either in terms of prescribed nodal displacement changes, or by an increment in the external
force and calculating the nodal displacements, using some form of an elimination process:
−1 Eq. 4.3
{Δδ } = K {ΔP }
At each integration point, calculate the strains corresponding to the nodal displacements of the finite element type on
which it lies:
{Δϵ } = B {Δδ } Eq. 4.4

Assuming elastic behavior, calculate the corresponding stresses at each integration point within the element:
{Δσ } = D {Δε } Eq. 4.5

Add any previous converged stresses to those obtained by the above equation, to evaluate the total stresses at each
integration point:
{σ } = { σ 0} + {Δσ } Eq. 4.6
and from those total stresses, evaluate the stress invariants

Using a preselected failure criterion, evaluate the value of the yield criteria F (function) in terms of the above stress
invariants:
F = F (J 1, J 2, J 3, k ) Eq. 4.7
where, J 1, J 2 and J 3 are the first, second and third stress invariants respectively and k is a function of state parameters
which are related to hardening/softening parameters.
Details on the various forms for F can be found in the Material Constitutive Models chapter.
Consider the sign of equation of F . If F < 0 at all integration points, the material behaves elastically.
If F ≥ 0 at any integration point, the yield criterion used in the analysis has been violated; excess stresses, which must
be redistributed into neighboring integration points in the mesh, exist.
With the yield criterion violated, the visco–plastic approach differs, from other numerical techniques as mentioned earlier,
in the way it iterates to achieve convergence, without violating equilibrium in the end.
Visco–plasticity
The visco–plastic model, shown above, can be generalized to include most behavior patterns encountered in soils. It is
first postulated, that the total strain is a sum of elastic and visco–plastic components. Thus:
{Δε } = {Δε } e + {Δε } v p Eq. 4.8
From figure 4.1, it can be seen that when yield occurs, the rate of movement of the dashpot will be a function of the
magnitude of the yield violation. The general visco–plastic rate law may be written as:
ε̇ vp = γ ⋅ f (F ) ⋅ { ∂∂Qσ } Eq. 4.9
here for real visco–plastic materials:
γ is the fluidity parameter of the dashpot (taken as unity)

f is a function of the yield criteria F,
Q is the visco–plastic potential.
When the method is applied to time independent elasto–plastic materials γ = 1.0 and f (F ) = f .
14
The method can accommodate both associated and non–associated flow rules. Setting Q = F makes the flow rate
associated.
To ensure no visco–plastic flow below the yield limit, it is considered that:
f (F ) = 0 if F < 0 Eq. 4.10

And:
f (F ) = f (F ) if F ≥ 0 Eq. 4.11
if any violation of the yield criterion at any integration point has occurred, which gives a positive value for the function F,
due to an underestimation of the plastic strains that have occurred.
Rather than accumulating stresses back to the yield surface, using the plastic stress–strain matrix used by initial stress
methods, visco–plasticity accumulates the visco–plastic strains at each time step; these are then converted to plastic
stresses and finally to body loads. The time step used here is a “psuedo” time used only to drive the algorithm. The time
step within the visco–plastic approach, as well as its magnitude, will be specified later on in this chapter.
At a specific time step, the visco–plastic strain is given by:
∫
t +Δt
{Δε } vp = ε̇ vp dt Eq. 4.12
t
Bearing in mind that, when the time step is of a small enough value, the strain rate may be assumed to be constant over
it and a linear approximation is used:
{Δε } vp = Δt ⋅ ε̇ vp Eq. 4.13
The excess stresses are then calculated by:
{Δσ } p = D ⋅ {Δε } vp Eq. 4.14
which are used to obtain a set of bodyloads as in equation 4.21 (p.16). These bodyloads are added to the nodal forces
at each time step and the procedure is repeated.
In general, the steps below are followed in the visco–plastic approach between two consecutive time steps, t and
t + Δt .
1. Knowing at time t (iteration number), a nodal redistribution force {ΔP } of the applied force, the nodal
displacements are evaluated, making use of the stiffness matrix of the whole structure, K , thus:
{Δδ } t = K −1
⋅ {ΔP } t Eq. 4.15
2. For each integration point, evaluate the total strain increments, making use of the above nodal displacements:
{Δε } t = B ⋅ {Δδ } t Eq. 4.16
3. At this step, the visco–plastic approach is different to other approaches, because the elastic part of the total strain is
estimated to obtain the stresses. Thus:
{Δσ } t = D ⋅ ({Δε } t − {Δε } tvp ) Eq. 4.17
for the first iteration or for any unyielded integration point, {Δε } vp = 0.
4. Add the above stresses to any converged stress at the previous increment:
{σ } = { σ o } + {Δσ } Eq. 4.18
5. Convert the stresses into invariants and calculate the yield function F.
6. If F= 0, proceed to the next integration point; if not, calculate the visco–plastic strain rate:
ε̇ tvp = γ ⋅ f (F ) ⋅ { ∂∂Qσ } Eq. 4.19
15
7. Update the visco–platic strain increment as follows:
{Δε } tvp+Δt = {Δε } vp

t
+ Δt ⋅ ε̇ ⋅t
vp Eq. 4.20
where Δt is a psuedo time. See the Visco-plastic Time Step Selection section for details of how this time step is
calculated.
8. Evaluate the bodyloads equivalent to the change in visco–plastic strain and then add them to the external nodal
forces:
{ΔP } t +Δt = {ΔP } t + Σ ∫

v
B T
D (Δt ⋅ ε̇ tvp ) dV Eq. 4.21
9. Return to step 1 and repeat the process, until convergence is achieved. If not, if could be assumed that failure of the
material has occurred.
Once convergence has been achieved the displacements, stresses and strains are updated for the next load increment.
This process is shown for the problem of a uni-axially loaded bar below.
Figure 4.2. Application of the visco–plastic algorithm
Convergence criterion
As already mentioned, the visco–plastic approach must satisfy non-linear compatibility and equilibrium conditions,
iterating in the manner described above. It is therefore necessary to establish some rules for these iterative procedures:
this will ensure accuracy for termination. Otherwise, the repetition of the calculations might continue for an indefinite
number of cycles.
There are several possibilities, but an important feature of the whole process, is that it must not allow cumulative errors.
Thus, one or more convergence criteria must be employed within the numerical process. This will take care of the total
unbalanced force at every load increment and at every stage of iteration. The purpose of these convergence criteria will
be to measure the satisfaction of the equilibrium equation. Some of these criteria can be used jointly or separately.
Convergence criteria available are:
• Maximum of displacement changes.
• Value of yield function F.
The displacement criteria:
16
max (Δdt +Δt − Δ d t )

< tol (dimensionless)
max (Δ d t +Δt )
where
Δd t +Δt is the displacement from the current iteration
Δd t is the displacement from the previous iteration

The yield criteria (F) :
F < tol (not dimensionless)

for all gauss points. As F is not dimensionless tol is not dimensionless either. This is the default criteria used.
For each of these convergence criteria tol is a tolerance specified a priori.

However, the influence of any convergence criterion incorporated in any numerical approach, must be closely examined,
so that its influence on the end result will be minimized.
Note: In Petrel Reservoir Geomechanics the yield criteria option is used always.
Visco-plastic time step selection

It can be seen from the above sequence of equations, that the visco–plastic approach depends heavily on the selection
of the appropriate “pseudo” time step t, which must be incorporated in the analysis. This step-by-step integration of a
system of equations in respect to “pseudo” time, leads to a further approximation, and so adds new errors to the usual
space discretization errors. This is due to the fact, that {Δε}vp in each time interval is supposed to remain constant, within
this small period of time. This assumes that any state of stress remains constant during this time increment, while, in
reality, it is varying. Thus, an a priori selection of the time step may cause errors to become excessive and results to be
meaningless.
Time steps, depend on stresses and viscous properties of the material and may vary from one element to another.
Assuming an explicit form of the visco–plastic algorithm this critical time step can be calculated.
The automatic calculation of optimum time steps in visco-plasticity was first addressed by Cormeau (1975) who obtained
analytical solutions for a number of failure criteria including Mohr–Coulomb. Cormeau extracted analytically, the
maximum eigenvalue Ψ from the matrix
∂ ε̇ υρ ∂F ∂Q ∂2 Q
A =
∂σ
=γ { ( ) ( ) ( )}
∂σ
T
∂σ
+F
∂σ2
Eq. 4.22
and computed the visco–plastic time step
Δt c = 2Ξ / Ψ Eq. 4.23
Ξ is an eigenvalue parameter that categorizes the stability of the time-stepping calculations. Cormeau (1975) took Ξ = 1
Stolle and Higgins (1989) , who obtained solutions for the visco–plastic time step for explicit Euler type elastic visco–
plastic time marching schemes, pointed out that Ξ = 1 admits stable oscillatory convergence, whilst Ξ = 0.5 represents an
upper bound for non-oscillatory stable predictions.
To ensure that nonlinear simulations can be performed efficiently, it has been recognized that an automatic calculation of
the visco–plastic time step Δ t is essential.
Some explicit expressions for time step selection are provided from the above equations for simple yield criteria. These
were derived for each yield criterion from the corresponding yield function. Hence, the critical time step for a Von–Mises
type material is given as:
4(1 + v)
Δt VM = Eq. 4.24
3Eγ
17
while for Mohr–Coulomb or Tresca material types it becomes:
4(1 + v)(1 − 2v)

Δt MC = Eq. 4.25
γE(1 − 2v + sin 2φ)
where E is Young's Modulus, v Poisson's Ratio and the φ friction angle.
As can be seen, when the critical time step is derived from the above equations and applied in equation 4.21, the fluidity
parameter γ does not affect the solution, providing it is nonzero.
It can be seen that the time step is a function of Young's modulus. This can have an adverse effect on convergence if
there are large variations in Young's modulus.
When an alternative yield criterion to the three above is used in an analysis, the value computed by equation 4.25 is
used and should be factored. For this purpose a multiplying factor is allowed for.
When the above time steps are used the time step remains constant throughout the analysis.
Alternatively the time step can be calculated automatically by equations 4.22 and 4.23. For the criteria above this will
result in the same critical time steps as given by equations 4.24 and 4.25 for associated flow. However for all other
criteria and non associated flow a varying time step will be calculated as the equations are dependent on the current
state of stress. This will result in a potentially different time step at each integration point. As the visco–plastic algorithm
is being used to solve elasto–plastic problems (that is, no time-dependent plastic behavior), the time step does not have
to be the same for all the integration points during a particular iteration.
Note: Visco-plastic time steps are calculated automatically with VISAGE.
18
Discontinuity model
The mechanical behavior of rock masses is often dominated by the properties and spatial locations of the discontinuities
(joints) within such masses. When interconnected, these discontinuities define individual rock blocks that can become
unstable under certain conditions.
Excavations and other mechanical or thermal fluid pressure loading processes applied to rock masses which may
contain discontinuities, can produce changes in stress and/or geometry that initiate instability.
Discrete zones of weakness, like faults or fractures (DFNs) can be modeled by individual elements and suitable material
properties. However if the number of discontinuities in a rock mass is very large so that modeling becomes impossible, a
different model has to be used. Here we describe constitutive laws which can be used to model such rock masses.
The methodology adopted is that of an equivalent material (See Pande, G. N., Beer, G. and Williams, J. R. : Numerical
Methods in Rock Mechanics., U.K., John Wiley and Sons Ltd, (1990)). In the following sections a brief description of the
constitutive behavior of this material will be given. Here joint refers to either a fault or DFN.
Equivalent material
Analyses assuming an equivalent material behavior rely on the assumptions that the discontinuities occur in regular sets,
and that the spacing between members within each set is much smaller than the dominant dimension describing the
external geometry of the particular problem. Economy in analysis is a compensating factor for the restrictions imposed by
these assumptions.
The equivalent material concept is based on developing a constitutive law for a material that will behave in the same way
as the rock mass, with its discrete sets of discontinuities. This means that the equivalent material is formulated by
smearing out the influence of each discontinuity set throughout the respective volumes that they occupy.
The theory behind this equivalent material is based on the visco–plastic algorithm and is capable in accounting for both
the deformational and failure behavior in the intact rock material and the sets of discontinuities.
The theory is first developed by considering the rheological units that represent each component in the rock mass, that is
intact rock material and the sets of discontinuities. The representation of the rock mass by series connection of these
units leads to the formulation of the general equations describing the behavior of the equivalent material.
Constitutive equations of rock mass

A rock mass is considered to consist of rock material intersected by sets of planar discontinuities The members in each
set of discontinuities are assumed to be parallel, continuous and evenly spaced at dimensions that are small in
comparison with the critical dimension of the rock mass, e.g. size of reservoir.
The definition of local axes with respect to the reference axes is necessary in order to provide links between the global
axes and the local axes of the discontinuities of the rock mass.
Local sets of coordinates are required to describe:
• The elastic anisotropy of the rock material.
• The orientation of each set of discontinuities.
Global coordinates are used to formulate the overall problem of the mechanical response of the rock mass in terms of
the equivalent material. Appropriate transformations are required to transform matrices, such as the stiffness matrix and
vectors, such as stresses and strains, from local to global coordinates and vice versa.
Discontinuity planes are so named because they permit relative displacements to develop. For a set of discontinuity
planes, these relative movements, when divided by the spacing between the planes can be interpreted as strains. The
total global strain vector for the rock mass is given by summing the global strain vector for the rock material with those for
Discontinuity model
19
each of the sets of discontinuities. Each set of discontinuities is considered to occupy a negligible volumetric proportion
of the rock mass and it is therefore assumed that the same global stress vector will be experienced by the rock mass and
each set of discontinuities.
The rock mass is composed of the rock material and sets of discontinuities. In order to enable complex modes of
response to be modeled and for mathematical convenience each component is represented in the analysis by a visco–
plastic unit.
The analysis of the mechanical behavior of the rock mass proceeds by developing the constitutive relations for the
equivalent material whose response considers the interaction of all of the constituent components, i.e. the intact rock, the
discontinuities with associated sets of interfaces. The rheological analogue of the equivalent material is illustrated in
figure 5.1 and can be seen to consist of units connected in series to form a string. Each of the units in a string is
subjected to the same increment of stress, while the increment of strain for the string is given as the sum of the
incremental strains of the units.
It can be seen that the string in figure 5.1 consists of the intact rock and all the sets of discontinuities. The series
connection ensures that the intact rock and the sets of discontinuities will experience an identical increment of stress and
will have increments of strain that are additive.
Elasticity matrix of jointed rock mass

Joints are attributed with a normal and shear stiffness due to poor contact of joint surfaces, resistance of the joints to
closure, friction on the joint walls and compressibility of the in filling gauge material. These stiffnesses (k) can be used to
derive a complete elasticity matrix of a jointed rock mass [DRM] with multiple joint sets. Assuming that the joints are
parallel, with uniform spacing and that the volumetric space is negligible the equivalent material model is used. The
movement of a joint under a stress σ is given by σ/ k.
Consider a joint plane and define a local coordinate axes ( x’,y’,z’) on this plane with z’ normal to the plane and x’ ,y’ lie
within the plane. A compliance matrix s can then be defined which relates stresses on the plane to relative movements:
w S 11 S 12 S 13 σ n
{}
v = S 21 S 22 S 23 τ s 1
u S 31 S 32 S 33 τ s 2
{}
It is unlikely that all components of the compliance matrix will be known. Usually only the normal compliance s11 and a
Eq. 5.1
general shear compliance Ss will be known. It is assumed that S22 and S33 the shear compliances are then equal to
S s / 2 and that all off diagonal terms are zero. It may be noted that under these circumstances the compliances are the
inverses of the stiffnesses.
Assuming a set of parallel joints within a set having a joint frequency f and the same compliance on each joint then the
overall compliance in the global coordinate system is
S = fT sT T Eq. 5.2
where T is a transformation matrix form the local joint coordinate system to the global system and f is the joint frequency
(number of joints per unit length). The frequency is also equal to one over the joint spacing. For faults the spacing is
taken as the perpendicular length across a fault plane between the intersected surfaces of the cell containing the fault.
If there are n sets of joints, then the total compliance (S*) is given by
n
S * = Σ f i T i s i T iT Eq. 5.3
i =1
and the average strain in the joints by
ε J = S *σ Eq. 5.4
The total strains in the rock mass are the sum of the intact rock strains and the joint strains leading to a final form for the
elasticity matrix of the rock mass [DRM] of
Discontinuity model
20
σ = D RM ε
Eq. 5.5
−1 −1
D RM = DI +S*
where [DI] is the intact rock elasticity matrix. This is the complete form of the elasticity matrix of the equivalent material.
These equations take into account the elastic characteristics of the joints (through stiffness), the spacing of the joints,
orientation of the joints and the elastic characteristics of the intact rock. Formation of the rock mass elasticity matrix as
above leads to a general anisotropic elasticity matrix.
Alternatively an anisotropic elasticity matrix could be assumed initially. However to obtain the 21 independent elastic
constants for a general anisotropic material is an extremely difficult task. The above formulation allows for an anisotropic
elastic stiffness matrix to be calculated from an arbitrary joint fabric.
Nonlinear behavior of jointed rock mass

The nonlinear behavior of the jointed rock mass is modeled by the multi-laminate theory. The theory is formulated in
terms of local coordinates and can describe the amount of sliding and opening along a discontinuity plane. The theory is
implemented within the framework of visco–plasticity. As for the elastic model the joint sets are assumed to be parallel
and continuous and to occupy a small volume in relation to the total rock mass. Each set of discontinuity planes is given
strength parameters incorporated within a failure criterion. The model is represented rheologically in 5.1. It assumes that
there are n sets of discontinuities all having the same failure criteria. The stress acting in each component is the same
and the elasticity matrix of the intact rock and joints is all lumped together in the single spring.
Figure 5.1. Rheological model for jointed rock model
The model fits within the framework of visco–plasticity with the following modifications.
Discontinuity model
21
At any integration point the stress within the rock mass is given by
{σ } = σ x , σ y , σ z , σ xy , σ yz , σ zx T Eq. 5.6
which is transformed to normal and shear stresses on a joint plane k
{σ^ } k = σ n , σ s T
k Eq. 5.7
Using the normal and shear stresses within a failure criteria, the visco–plastic strain rate for a joint set in the global
coordinate system is
∂Qk ∂ σ^ k
ε̇ kvp = γ k ⋅ f (F ) k ⋅ { }{ }
∂ σ^ k ∂σ
Eq. 5.8
It is assumed that the intact rock also yields and gives the following visco–plastic strain rate
ε̇ ivp = γ i ⋅ f (F ) i ⋅ { ∂∂Qσ }
i
Eq. 5.9
The total visco–plastic strain rate at an integration point will be the sum of the contributions from each joint set and the
intact rock
n
ε̇ vp = ε̇ ivp + Σ ε̇ kvp Eq. 5.10
k =1
this strain rate is used within the visco–plastic algorithm. (See equation 4.19.)
The total model (elastic and plastic) as described above involves the following :
Intact Rock: elastic properties, strength parameters, failure criterion and flow rule.
Discontinuities: elastic stiffness (Ks ,Kn), orientation, spacing, strength parameters, failure criterion and flow rule.
Discontinuity model
22
Material Constitutive Models

This chapter describes the elastic, plastic (yield) and discontinuity models which can be used in VISAGE.
Elastic models can be either isotropic or transversely anisotropic.
Plasticity models available are Mohr-Coulomb, Drucker-Prager and critical state cap models amongst others.
Also detailed is a discontinuity model for modeling fractures (DFN) and faults.
Note: Not all material models available in VISAGE can be accessed via the Petrel Reservoir Geomechanics workflow.
Elastic material properties

Orthotropic material
The general compliance matrix for orthotropic material requires 9 independent material parameters and is given by:
1 ν yx ν zx
Ex - -
Ey Ez
ν xy 1 ν zy
ϵ xx - Ey - 0 σ xx
Ex Ez
ϵ yy σ yy
ν xz ν yz 1
ϵ zz - - Ez σ zz
Ex Ey
= Eq. 6.1
γ xy 1 τ xy
γ yz G xy τ yz
γ zx 1 τ zx
0
G yz
1
G zx
With x, y and z indicating directions, ε, γ are normal and shear strains respectively; σ is stress; E, ν and G are Young’s
modulus, Poisson’s ratio and shear modulus for the material.
Due to symmetry ν yx , ν xz , ν zy can be calculated from :
ν xy ν yx ν xz ν zx ν yz ν zy
= , = , = Eq. 6.2
Ex Ey Ex Ez Ey Ez
Note: Orthotropic material is not available via Petrel Reservoir Geomechanics.

23
Isotropic material
Two independent material parameters are required to describe an isotropic material.
The compliance matrix for an isotropic material is:
1 ν ν
- -
E E E
ϵ xx ν 1 ν σ xx
- - 0
E E E
ϵ yy σ yy
ν ν 1
ϵ zz - - σ zz
E E E
= Eq. 6.3
γ xy 1 τ xy
γ yz G τ yz
1
γ zx 0 τ zx
G
1
G
With x, y and z indicating directions, ε, γ are normal and shear strains, respectively; σ is stress; the independent
parameters are Young’s Modulus E and Poisson’s Ratio v. The shear modulus G for the material is then given by
E
G = Eq. 6.4
2(1 + v)
Note: 0.0 ≤ v < 0.5
Orientation of the local coordinate system

Transversely isotropic material with a vertical axis of symmetry (TIV) can be either global or local. Global means that the
vertical axis of symmetry is constant. Currently it is assumed that the constant vertical axis of symmetry is aligned with
the vertical axis (z-axis) of the global coordinate system i.e. ß = 0, θ = 0. For an explanation of the angles ß, θ see below.
Using the local coordinate systems is required when the bedding is inclined or folded. The angles ß, θ are then internally
calculated for each cell, from the plane formed at the base of each cell.

24
Figure 6.1. Orientation of local z-axis (ζ-axis) within the global coordinate system
The z-axis of the local coordinate is denoted as ζ-axis. The dashed lines represent the projections of the ζ-axis on the xy-
plane and on the z-axis of the global coordinate system.
The angle ß denotes the angle between the global x-axis and the projection of the local ζ-axis on the xy-plane. θ is the
angle between the global z-axis and the local ζ-axis.
All angles have to be provided in degrees.
Note:
The global z-axis can be transformed into the local ζ-axis by means of two elementary rotations:
First a rotation around the global z-axis by angle ß (mathematical positive, right hand rule).
Followed by a rotation around the global y-axis by angle θ (mathematical positive).
Note:
The angles ß, θ are identical with the first two out of three Euler angles associated with the Euler transformation:
Rotation around the z-axis followed by a rotation around the y-axis followed by a rotation around the z-axis.
The third angle is not needed as long as the material is isotropic perpendicular to the local ζ-axis: Isotropic and TIV
material.

25
Undrained Model
In geomechanical engineering the principle of effective stress underlies much of the analysis model. For the full
implementation of the effective stress principle a time dependent consolidation analysis should be considered Coupled
porous media (p.67). This involves the solution of the fully coupled Biot consolidation equations. Alternatives to the full
consolidation solution are to model the undrained and fully drained limits of the consolidation model or to apply a known
pore pressure change. The behaviour of the non-reservoir regions in a coupled model can be modelled using the fully
drained behaviour or alternatively a known pore pressure change can be applied. This means that there is no pore
pressure change or a given change in pore pressure in these regions. An alternative is to use an undrained model which
will result in pore pressure changes being calculated in those cells from the calculated volume changes.
Using the principle of effective stress
{Δσ } = {Δσ ′} + {m }αΔp Eq. 6.5
where the stress and pore pressure are taken as tension positive. σ is the total stress and σ ′ is the effective stress and
{m } is equal to unity for the normal stresses and zero for the shear components. α is known as Biot’s coefficient and is
defined as
KT
α = 1.0 − Eq. 6.6
Ks
where KT is the bulk modulus of the equivalent material (solid phase plus pores) and Ks is the bulk modulus of the
solid phase.
This definition of effective stress takes into account the volumetric strains caused by the uniform compression of the solid
particles by the pore fluid as well as the overall volumetric strain of the equivalent material. For soils the volume strain of
the solid particles is usually insignificant but for rocks this can become significant. The constitutive equation is then
expressed in terms of effective stresses as
{Δσ ′} = D ′ {Δε } Eq. 6.7
where D ′ is the effective constitutive matrix and inserted into the equilibrium equations
If conditions are fully undrained there will be no movement of the pore fluid relative to the skeleton, consequently the
skeleton and pore fluid elements undergo precisely the same deformation. Their strains {Δε} can therefore be equated.
Total stress changes are related to strain changes by the total stiffness matrix [D]
{Δσ } = D {Δε } Eq. 6.8

f
As the pore pressure changes by Δp the pore fluid undergoes a volumetric strain change Δϵ v then
Δp = K f Δϵ vf Eq. 6.9
where Kf is the pore fluid stiffness. It is not simply the bulk modulus of the pore fluid Kw as the pore fluid occupies only
the voids, but is a composite of the pore fluid and the solid phase Ks . They are related by the porosity n, via
1 n 1−n
= + Eq. 6.10
Kf Kw Ks
When conditions are undrained the solid and pore fluid strains become the same, so that
Δϵ vf = Δϵ v = m T {Δϵ } Eq. 6.11
The change in pore pressure can then be written as
Δp = K f m T {Δϵ } Eq. 6.12
Substitution of the above equations into the effective stress equation leads to the following equation for the stiffness
matrix [D]

26
D = D ′ + α m m T Kf Eq. 6.13
Thus for the undrained analysis the stiffness equations are setup in terms of the total stresses. Effective stresses and
pore pressures are then computed using equations 6.7 (p.26) and 6.12 (p.26) respectively.
Invariant yield criteria

Given a state of stress (total or effective) at any integration (gauss) point
then the stress can be written in terms of invariants, independent of the coordinate system used.
A convenient choice for these invariants is first principal invariant (I1) of σ and the second and third principal invariants
(J2, J3) of the deviatoric stress tensor, defined as
I 1 = Tr (σ ) = ( σ 1 + σ 2 + σ 3)
1 1
J2 = s : s = ( ( σ 1 − σ 2 ) 2 + ( σ 2 − σ 3 ) 2 + ( σ 3 − σ 1) 2 ) Eq. 6.15
2 6
J 3 = det (s ) = s 1 s 2 s 3
I1
Where s =σ − I is the deviatoric stress tensor and I is the identity matrix.
3
Related quantities are (p, J, θ) or (p, q, θ)
Mean stress :
1
p = I Eq. 6.16
3 1
Equivalent stress :
J = J 2 or q = 3 J 2 Eq. 6.17
Lode angle :
( σ 2 − σ 3)
θ = tan-1
1
2 (
3 ( σ 1 − σ 3)
−1 ) − 30 ° ≤ θ ≤ 30 ° Eq. 6.18
where σ 1, σ 2, σ 3 are the principal stresses. These stress invariants can be interpreted as shown below. The value of
p is a measure of the distance along the space diagonal of the current deviatoric plane ( π-plane). J provides a measure
of distance from the space diagonal in the deviatoric plane and θ defines the orientation of the stress state within this
plane.

27
Figure 6.2. Invariants in Principal Stress Space — Space Diagonal
Figure 6.3. Invariants in Principal Stress Space — Deviatoric (π) Plane
The yield function F is defined as a function of stress and state parameters {k } that are related to hardening/softening
parameters:
F = F ({σ }, {k }) Eq. 6.19

and the plastic potential function is of a similar form
Q = Q ({σ }, {m }) Eq. 6.20

where {m} are again state parameters, which are never used, as only derivatives of Q with respect to stress components
are needed.
If Q = F then the flow rule is described as associated. If Q ≠ F then the flow rule is described as non-associated.
Mohr–Coulomb criteria
The yield and potential functions are given by :

28
F =J − ( tanc φ + p )g (θ )
φ
( cosθ + sinsinθsinφ
g (θ ) =
/ 3)
Eq. 6.21
sinθ sin φ
F = p sinφ + J (cosθ − ) − c cos φ
3
where c is the cohesion and φ is the friction angle. The flow rule Q is given by the same equation but with the friction
angle φ replaced by the dilation angle ψ .
Hardening/softening is accounted for in the cohesion which is a function of plastic strain:
c = c ο + HEε dp Eq. 6.22
where co is the initial cohesion, H is the hardening/softening parameter, E is the Young’s modulus and εp d is the
equivalent plastic strain:
2.0 ε x2 + ε y2 + ε z2 + ε xy2 + ε yz2 + ε zx2 1/2

ε pd = Eq. 6.23
3.0
Below is shown the relationship between yield and potential surfaces in p–J space.
Figure 6.4. Mohr–Coulomb Yield and Plastic Potential Functions in p-J Space
F and Q are functions of the Lode angle and the friction and dilation angle respectively.
Below is shown the yield and potential surfaces on the deviatoric plane.

29
Figure 6.5. Mohr–Coulomb Yield and Plastic Potential Functions on Deviatoric Plane
Below is shown the yield surface in principal stress space. Mohr–Coulomb yield surface is an irregular hexagonal
pyramid.
Figure 6.6. Mohr–Coulomb Yield Surface in Principal Stress Space
Drucker–Prager criteria
F =J − ( tanc φ + p )g (θ )
2 3sin φ
g (θ ) = (
3 ± sin φ )
at θ = ± 30 ° Eq. 6.24
6
3(3 ± sin φ ) (
F =J − c cos φ + p sin φ )

30
where c is the cohesion and φ is the friction angle. The flow rule Q is given by the same equation but with the friction
angle φ replaced by the dilation angle ψ. The Drucker–Prager criteria is the same as Mohr–Coulomb in p–J space and a
circle on the deviatoric plane. g(θ) is a constant given by either the inscribed (-) or circumscribed (+) circle of the Mohr
Coulomb hexagon. Only the inscribed circle is formulated.
Hardening/softening is accounted for in the cohesion which is a function of plastic strain:
d
c = c ο + HEε vp Eq. 6.25
where co is the initial cohesion, H is the hardening/softening parameter, E is the Young’s modulus and εvp d is the

ε vpd = Eq. 6.26
3.0
Figure 6.7. Drucker–Prager Yield Surface
Von Mises criteria

F = 3(J − c ) Eq. 6.27

where c is the yield strength. The flow rule Q is given by the same equation.
Hardening/softening is accounted for in the yield stress which is a function of plastic strain:
where co is the initial yield strength, H is the hardening/softening parameter, E is the Young’s modulus and εp d is the

ε pd = Eq. 6.29
3.0

31
Figure 6.8. Von Mises Yield Surface
Tresca criteria
F = J cos(θ ) − c Eq. 6.30

where c is the yield stress. The flow rule Q is given by the same equation.
Hardening/softening is accounted for in the yield stress which is a function of plastic strain:
d
where co is the initial yield strength, H is the hardening/softening parameter, E is the Young’s modulus and ε p is the

ε pd = Eq. 6.32
3.0

32
Figure 6.9. Tresca Yield Surface
Critical state criteria

The Critical State model is modeled on the modified Cam Clay model. This is given by an ellipse in the p–J space and
the formulation assumes a Mohr–Coulomb hexagon for the yield surface in the deviatoric plane given by M.
The equation of the yield surface is then given by
F = 3.0J 2 + M 2 p (p + 2.0p c )
3.0sin φ
where M =
3.0cos θ − sin φ sin θ Eq. 6.33
χϵ (1 + e o )
p c = p co e vp , χ =
(λ − κ)
where φ is the friction angle, χ is the hardening/softening parameter and ε vp is the volumetric plastic strain. Also eo is
the initial voids ratio, λ is the slope of the virgin consolidation line and κ the slope of the swelling line. p co is the initial
radius of the Cam Clay ellipse.
The projection of the yield surface onto the J–p plane and the volumetric behavior are shown below. This shows how the
model can simulate both hardening and softening. The flow potential Q is associated and is also given by equation 6.33
(p.33).

33
Figure 6.10. Projection of Yield Surface onto J–p Plane
Figure 6.11. Volumetric Behavior of Modified Cam Clay Model
Chalk model
This uses the yield function form the JCR V Unified Chalk model from the Norwegian Geotechnical Institute.
The yield function is given by a three surface model that divides the stress space into a pore collapse region, shear
region and tensile region. For each region a yield function is given in terms of the three-dimensional stress invariants
mean stress p, equivalent stress q and Lode angle θ which are functions of the stress invariants.
The pore collapse failure surface is defined by an elliptical cap surface
(p − Rp max ) 2
2
Fc = q − (Yο − a ) 1 − 2
(
(R − 1) p max ) Eq. 6.34
which has an associated flow rule and is assumed to have volumetric strain hardening defined by
p max (1 + e ο ) p
dp max = dε v Eq. 6.35
Γ
The shear failure surface is given by the hyperbolic surface

34
2
F s = q − Y ο + a 2 + G ( θ ) η f ( p − Rp max )
Y ο = q ο + a 2 + G (θ )η f Rp max 2
Eq. 6.36
G (θ ) = G (θ , k )
6 cos ( ϕ)
ηf =
(3 − sin(ϕ))
qo = c * ηf Eq. 6.37
and is assumed fixed and non-hardening but to have a non-associated flow rule. Here c = cohesion and ϕ = friction
angle.
The plastic potential being the same as the failure surface with ηf replaced with η g . G (θ ) accounts for the variation in
the third stress invariant and is given by
c 1cos θ + c 2( c 3 + c 4)
G (θ ) =
c2 + c4
c 1 = 2(1 − k 2)
Eq. 6.38
c 2 = (k − 2)2
c 3 = (5k − 4)k
c 4 = 2c 1cos2 θ
The tensile surface is also hyperbolic and given by
η ο ( p − p t )2
2
(
Ft = q − qο ηο pt + qο ) 1−
pt (ηο pt + qο )
( G (θ )η f )2 Rp max
ηο = − Eq. 6.39
α 2 + G ( θ ) η f R p max 2
η ο σ t2
pt =
2σ t η ο + q ο
which is fixed and non-hardening with an associated flow rule.

35
Figure 6.12. Chalk Model in p–q Plane
Figure 6.13. Chalk Model in Deviatoric Plane

36
ISAMGEO chalk model

See the separate section describing this model ISAMGEO Chalk Model (p.42).
Modified Drucker–Prager model

The failure surface is described by the classic Drucker–Prager criterion for unconsolidated material
F (σ ) = q + Ap = 0 Eq. 6.40
where again :
3
p = σ kk / 3 , q = s s , s ij = σ ij − p δ ij Eq. 6.41
2 ij ij
The parameter A in equation 6.40 (p.37) defines the maximum friction coefficient of the material. This parameter is kept
constant.
The plastic yield function is obtained by introducing a scalar valued hardening function η , so that:
F (σ , η ) = q + η ⋅ Ap = 0, 0 ≤ η ≤ 1 Eq. 6.42
The plastic hardening is described by a function of the generalized plastic strain:
ē p
η = η 0 + (1 − η 0) Eq. 6.43
B + ē p
∫ ( 23 ė ė )
t 1/2
p p
ē p = ij ij dτ Eq. 6.44
0
where:
ε vp
e ijp = ε ijp − δ
3 ij
The effective stress space is divided into two zones as shown below, a plastic compressibility zone and a dilation zone.
The transition from compressibility to dilation is defined by the line:
F (σ ) = q − η c Ap = 0 Eq. 6.45
The following plastic potential is used for the non-associated flow rule:
Q (σ ) = q − η c A p ln ( pp )
0
Eq. 6.46

37
Figure 6.14. Description of Implemented Constitutive Model
Discontinuity yield criteria

Given a state of stress (total or effective) at any integration point
the stress is converted to normal and shear stresses on the joint plane
{σ n } = σ n , τ s T Eq. 6.48
by
{ σ n } T = T {σ } T Eq. 6.49
where T is a transformation matrix based on the direction cosines of the normal to the joint plane. The direction cosines
are calculated automatically from the joint angles (dip and dip direction).
The yield function F is defined as a function of stress and state parameters {k} which are related to hardening/softening
parameters:
F = F ({ σ n }, {k }) Eq. 6.50
and the plastic potential function is of a similar form
Q = Q ({ σ n }, {m }) Eq. 6.51
where {m} are again state parameters, which are never used as only derivatives of Q with respect to stress components
are needed.
If Q = F then the flow rule is said to be associated. If Q ≠ F then the flow rule is said to non–associated.
The Mohr–Coulomb yield criteria is available for modeling joint plasticity.
Mohr–Coulomb criteria
The yield and potential functions are given by:

38
F1 = σ n k =1,n
=0
Eq. 6.52
F2 = | τ s | + σ n tan ∅ k − c k k =1,n
=0
and
Q 1 = σ n − constant k =1,n
=0
Eq. 6.53
Q2 = | τ s | + σ n tan Ψ k − constant k =1,n
=0
where ck is the cohesion, ∅k is the friction angle and Ψ k is the dilation angle of the kth joint set. The yield function is
represented by Mohr–Coulomb failure F2 together with a no tension cut off F1 .
The flow rule Q is given by the same equation but with the friction angle ∅ replaced by the dilation angle Ψ.
Hardening/softening is accounted for in the cohesion which is a function of shear plastic strain varying linearly from the
initial value to a residual cohesion over a given shear plastic strain:
(co − cR )
c = co − τs Eq. 6.54
τ sR
where co is the initial cohesion, cR is the residual cohesion at shear strain.
Figure 6.15. Yield and plastic potential functions
Figure 6.16. Hardening/softening function

39
Creep Model
Creep Theory
Salt rock mechanics is increasingly being seen as important in reservoir engineering. Pre–salt reservoirs sealed by thick
salt rock require drilling though the salt rock layers. Understanding the stability and integrity of well bores requires a
detailed understanding of the behaviour of the salt rock. Modelling of salt rock and similar materials requires a time-
dependent non-linear material behaviour. Creep or visco-elastic/plastic models meet this requirement. A power law creep
model (Norton – Bailey) exists within VISAGE.
In general, the stress-strain relationship in uniaxial creep, can be represented by the following equation
ε c = f (σ , t , T ) Eq. 6.55
where εc is the creep strain, σ is the nominal stress, t is the time and T is the temperature. This equation can be
simplified by separating the effects as follows :
ε c = f 1(σ )f 2(t )f 3(T ) Eq. 6.56
There are a number of alternative expressions. One of the most commonly used is the power stress law attributed to
Norton.
f 1(σ ) = B σ n Eq. 6.57
where B and n are material constants. The Bailey law for time
f 2(t ) = D t m Eq. 6.58
where D and m are material constants. Temperature dependence is of the form
−Q
f 3(T ) = C exp ( ) Eq. 6.59
RT
where Q is activation energy and R is Boltzman’s constant. The creep strain is then
−Q n m
ε c = A exp ( )σ t Eq. 6.60
RT
Under isothermal conditions the creep strain is
ε c = A σ nt m Eq. 6.61
which is known as the Norton-Bailey law and is available within VISAGE.

This equation can describe primary or secondary creep stages. Usually m < 1 is used for primary creep and m = 1 for
secondary creep.
In practical applications stress will vary with time so the creep strain curves with time under constant stress are
converted into variable stress curves. Differentiating gives
ε̇ c = mA σ n t ( m −1) Eq. 6.62
and removing the time variable (by substituting it from the Norton-Bailey equation) gives
( m −1)/m
ε̇ c = mA 1/m σ n /m ( ε c ) Eq. 6.63
These forms are known as the time and strain hardening equations of creep. The strain hardening is considered more
accurate and agrees well with experimental tests. Both versions are available in VISAGE.
The above equations were developed for uniaxial creep. Multiaxial creep laws are similar to the multiaxial plasticity
formulations and are usually based on the von Mises effective stress and the Prandtl-Reuss flow rule. The creep strain
rate is related to the deviatoric stress as follows :

40
dε ijc
ε̇ cij = = λS ij Eq. 6.64
dt
where the factor of proportionality λ, can be determined experimentally from a uniaxial creep test and Sij is the deviatoric
stress. The multiaxial version of the uniaxial Norton-Bailey creep law is therefore
c
ε eff = A (σ eff ) n t m Eq. 6.65
3 1/2
where σ eff = S S is the effective stress. A similar expression exists for the equivalent or effective strain rate
2 ij ij
c
ε eff . Therefore a multiaxial expression for the creep strain rates and the effective (von Mises) stress can be written as :
3
ε̇ cij = mA (σ eff )( n −1) S ij t ( m −1) Eq. 6.66
2
4.1.2 Time procedures

In creep problems the strain is a function of time according to the constitutive equations and the solution needs to be
marched in time by dividing the total time into small time intervals. There are two main methods used in Finite Element
simulations as described below. The strain or displacement rate is assumed to be a function of the current displacement
and time as follows :
du
u̇ = = f (u , t ) Eq. 6.67
dt
The simplest method uses a finite difference approximation for the differential of the displacement with time
du u t +Δt − u t
u̇ t = ( )
dt t
=
Δt
Eq. 6.68
This gives the following “recurrence” relationship
u ( t +Δt ) = u t + Δt f (u t , t ) Eq. 6.69
This is the Forward-Difference (Explicit) Euler method as it uses only the previous point to calculate the new value of
displacement. This method requires small time steps to remain stable and convergent. This is the method currently
implemented in VISAGE.
As stated earlier the algorithm must remain both convergent and stable. The time step length must be chosen to ensure
convergence and stability. A time step which satisfies stability can be found using
2
( ε̇ cij ε̇ cij )Δt = ε̇ cij Δt ≤ τϵ Eq. 6.70
3
′
It is import to note that the effective stress σ = σ + αmp where m = {1 1 1 0 0 0}, α = Biots coefficient, σ is the total
stress and σ' is the effective stress which is the stress used in the creep algorithms above.

41
ISAMGEO Chalk Model

In 2013 Schlumberger acquired IP and sole rights for a specific unified stress-strain model developed for chalk
reservoirs. This model is named the 'ISAMGEO’s Chalk Model', but the model can be applied to other weak and
compressible reservoir rocks.
The model uses the elasto-plastic solution technique, and is characterized by a dependency of pore collapse strength on
volumetric plastic strain rate following the approach proposed by de Waal et al. (1998) (p.66) . It uses a single yield
surface for shear and pore collapse as described by Papamichos et al. (1997) (p.66) . The single yield surface concept
reduces problems in the transition of the numerical solutions from pore collapse to shear failure and vice versa. The
model does not implement the Cosserat approach of micropolar elasticity.
The ISAMGEO Chalk Model is implemented in two forms in Petrel Reservoir Geomechanics and its VISAGE simulator:
• With rate dependent behavior (creep)
• Without rate dependent behavior
The creep parameter 'b' for the ISAMGEO Chalk Model with rate dependent behavior can be determined by lab tests.
Water sensitivity is also included in both implementations of the model, with the variations of pore collapse, strength and
hardening parameter also being determined from lab tests. Water sensitivity in the rock elastic parameters can also
included via the Property Updating Function described in the Petrel Reservoir Geomechanics User Guide.
Currently the ISAMGEO Chalk Model cannot be used in conjunction with any other constitutive material models available
in Petrel Reservoir Geomechanics and its VISAGE simulator, and it and does not compute yield in faults and fractures.
The following sections give a detailed description of the model, but for a better understanding a few details of the general
solution process may be useful:
• Simulations start from an in situ state of stress, which is in equilibrium with the rock strength.
• The simulation is subdivided into simulation steps, which represent time steps of a loading time history, like the
stress changes resulting from the pressure history for a reservoir.
• Each simulation step covers a time step Δt. The time stepping scheme is explicit, i.e. results for the time t + Δt are
calculated based on the variables (like stress) for the time t.
• Depending on the required accuracy, this can require small time steps. On the other hand, computational efforts for a
single time step are reduced, compared with implicit approaches, which require iterations. If the obtained stresses
are not in equilibrium with external loads, out of balance forces are calculated and applied in the next simulation step.
This approach and model have proved themselves in practice for various full field compaction/subsidence studies. The
application of this model and approach to the reservoir chalks of the Valhall field in the North Sea is described in
Kristiansen and Plischke (2010) (p.66) .
Stress invariants and derivatives

Within this chapter all stresses are effective stresses. The prime, which is commonly used to denote effective stresses, is
omitted. Compressive stresses are positive.
The following stress invariants are used in the model:
1st invariant of normal stress I1 :
I1 = σ x + σ y + σ z Eq. 7.1
Alternatively, mean effective stress p m = I 1 / 3 may be used.

2nd invariant of deviatoric stress:
ISAMGEO Chalk Model

42
1
J2 = ⋅ ( σ x − σ y )2 + ( σ y − σ z )2 + ( σ z − σ x )2 + τ xy2 + τ yz2 + τ zx2 Eq. 7.2
6
J2 can also be written as:
1
J2 = ⋅ S x2 + S y2 + S z2 + τ xy2 + τ yz2 + τ zx2 Eq. 7.3
2
with deviatoric stresses Sx = σx - pm, Sy = σy - pm and Sz = σz - pm.
J 2 is linked with the Von-Mises stress q via q = 3 ⋅ J 2.

3rd invariant of deviatoric stress and Lode angle:
J 3 = S x ⋅ S y ⋅ S z + 2 ⋅ τ xy ⋅ τ yz ⋅ τ zx − S x ⋅ τ yz2 − S y ⋅ τ zx2 − S z ⋅ τ xy2 Eq. 7.4
Alernatively the Lode angle θ is used, which is defined via
3⋅ 3 J3
( )
sin 3θ =
2
⋅
J2 ⋅ J2
Eq. 7.5
The range of θ is limited from -30° ( − π / 6) to +30° ( π / 6). A Lode angle of θ = -30° corresponds to triaxial
compression (σ2 = σ3), while θ =30° denotes triaxial extension (σ1 = σ2). It should be noted that the sign of the Lode
angle changes, if compressive stresses are defined as negative.
Derivatives
Derivatives of invariants and related quantities with respect to stress components are needed for the implementation of
elastoplasticity.
For the mean effective stress it is :
δp m δp m δp m 1
= = = Eq. 7.6
δσ x δσ y δσ z 3
For the derivatives of J 2 one obtains:
δ J2 Sx
= Eq. 7.7
δσ x 2 ⋅ J2
δ J2 Sy
= Eq. 7.8
δσ y 2 ⋅ J2
δ J2 Sz
= Eq. 7.9
δσ z 2 ⋅ J2
δ J2 τ xy
= Eq. 7.10
δτ xy J2
δ J2 τ yz
= Eq. 7.11
δτ yz J2
δ J2 τ zx
= Eq. 7.12
δτ zx J2
The derivatives of the third invariant of deviatoric stress can be written as:
ISAMGEO Chalk Model

43
δJ 3 J2
= S y ⋅ S z − τ xy2 + Eq. 7.13
δσ x 3
δJ 3 J2
= S z ⋅ S x − τ zx2 + Eq. 7.14
δσ y 3
δJ 3 J2
= S x ⋅ S y − τ xy2 + Eq. 7.15
δσ z 3
δJ 3
= 2 ⋅ ( τ yz ⋅ τ zx − S z ⋅ τ xy ) Eq. 7.16
δτ xy
δJ 3
= 2 ⋅ ( τ zx ⋅ τ xy − S x ⋅ τ yz ) Eq. 7.17
δτ yz
δJ 3
= 2 ⋅ ( τ xy ⋅ τ yz − S y ⋅ τ zx ) Eq. 7.18
δτ zx
The Lode angle is a function of the second and the third invariants of deviatoric stress. Accordingly, the chain rule must
be used to obtain the derivatives:
δθ δθ δ J2 δθ δJ 3
= ⋅ + ⋅ Eq. 7.19
δσ x δ J2 δσ x δJ 3 δσ x
= ⋅ + ⋅ Eq. 7.20
δσ y δ J2 δσ y δJ 3 δσ y
= ⋅ + ⋅ Eq. 7.21
δσ z δ J2 δσ z δJ 3 δσ z
= ⋅ + ⋅ Eq. 7.22
δτ xy δ J2 δτ xy δJ 3 δτ xy
= ⋅ + ⋅ Eq. 7.23
δτ yz δ J2 δτ yz δJ 3 δτ yz
= ⋅ + ⋅ Eq. 7.24
δτ zx δ J2 δτ zx δJ 3 δτ zx
where
δθ tan(3θ)
= − Eq. 7.25
δ J2 J2
and
δθ 3
= Eq. 7.26
δJ 3 2 ⋅ J 2 ⋅ J 2 ⋅ cos(3θ)
Elastic stress strain behavior

In the context of programming the Finite Element method, stresses and stress increments as well as strains and strain
increments usually are represented as column vectors (Voigt notation), containing the six different components of the
corresponding tensors.
ISAMGEO Chalk Model

44
In the case of linear elastic isotropic stress strain behavior, the material behavior is characterized by two independent
parameters. In the following, the Young’s modulus E and the Poisson’s ratio ν are used to define the so-called D-matrix
which links vectors of effective stress increments and strains increments:
Δσ x Δε x
Δσ y Δε y
Δσ z Δε z
= D el ⋅ Eq. 7.27
Δτ xy Δγ xy
Δτ yz Δγ yz
Δτ zy Δγ zy
with
v v
1 0 0 0
1−v 1−v
v v
1 0 0 0
1−v 1−v
v v
1 0 0 0
E • (1 − v) 1−v 1−v
D el = • Eq. 7.28
(1 + v) • (1 − 2v) 1 − 2v
0 0 0 0 0
2 • (1 − v)
1 − 2v
0 0 0 0 0
2 • (1 − v)
1 − 2v
0 0 0 0 0
2 • (1 − v)
The shear modulus G, the bulk modulus K and the uniaxial compression modulus M can be expressed by E and ν:
E
G = Eq. 7.29
2 ⋅ (1 + v)
E
K = Eq. 7.30
3 ⋅ (1 − 2v)
E ⋅ (1 − v)
M = Eq. 7.31
(1 + v) ⋅ (1 − 2v)
In case of transversally isotropic behavior, one assumes that the rock response is characterized by one symmetry plane.
In this plane, the stress strain behavior is isotropic, whereas the response for loading normal to this plane is different.
Bedding planes are the most common isotropic planes. This often encountered special case of anisotropic behavior can
be characterized by means of five independent parameters: E1, E2, ν 1 , ν 2 and G2, as long as linear elasticity applies.
The definition of these parameters is given in Figure 7.1 (p.46) which also shows the definition of parameters G1 and ν
3, which can be derived from the others.
ISAMGEO Chalk Model

45
Figure 7.1. Definition of elastic constants in case of transversally isotropic, linear elastic stress strain
behavior (according to W.Wittke: “Rock Mechanics”, Springer Verlag, 1990)
From energy considerations, one can derive an upper limit for the parameter G2 (see for example W.Wittke: “Rock
Mechanics”, Springer Verlag, 1990, p.46).
E2
G2 ≤ Eq. 7.32
2 • v 2 ⋅ (1 + v 1) + ( E 2 / E 1 − v 22) • (1 − v 12)
The modulus G2 is difficult to determine in laboratory tests. Its magnitude may be approximated by
E2
G2 = Eq. 7.33
2 • (1 + v 2)
For transversally isotropic behavior and with the y-axis normal to the isotropic plane, the D-matrix reads:
ISAMGEO Chalk Model

46
E 1 • (1 − n • v 22) E1 • v2 E 1 • ( v 1 + n • v 22)
0 0 0
(1 + v 1) • m m (1 + v 1) • m
E1 • v2 E 2 • ( 1 − v 1) E1 • v2
0 0 0
m m m
E 1 • ( v 1 + n • v 22) E1 • v2 E 1 • (1 − n • v 22)
D el = 0 0 0 Eq. 7.34
(1 + v 1) • m m (1 + v 1) • m
0 0 0 G2 0 0
0 0 0 0 G2 0
E1
0 0 0 0 0
2 • (1 + v 1)
with n = E1/E2 and m = (1 − v 1 − 2n ⋅ v 22)
Elastoplastic stress strain behavior

The basic idea of elastoplasticity is the assumption that total strain increments are composed of an elastic part and a
plastic part:
{Δε } = {Δε el } + {Δε pl } Eq. 7.35
Plastic strains are irreversible. Only the elastic part contributes to stress increments:
{Δσ } = D el ⋅ {Δε el } = D el ⋅ ({Δε } − {Δε pl }) Eq. 7.36
The elastic D-matrix D el has been defined in the previous section.
Yield surface
The yield surface F serves as a criterion for the onset of plastic behavior. In the simple case of no hardening, softening or
water weakening, the yield surface is a scalar function of the stress state: F = f ({σ }). If F < 0, the state of stress is
considered as elastic, whereas F = 0 refers to plastic behavior. Stress states with F > 0 are not admissible, although the
numerical implementation of plasticity commonly results in stress states F > 0, which are then corrected back to the yield
surface F = 0. In the stress space, F = 0 describes a surface. In a two dimensional pm - q diagram it is a line. It should be
recalled that in general a stress state cannot be completely described in a p m − q diagram, as at least three invariants
are required. As examples, two yield surfaces are presented, which are quite popular in rock mechanics. The Drucker-
Prager criterion does not depend on the Lode angle and represents a straight line in the p m − q diagram:
F = q − M ⋅ pm − cq Eq. 7.37
M is a parameter characterizing friction and should not be mixed up with the uniaxial compression modulus defined in the
previous section. cq is the intercept on the q-axis and describes the contribution of cohesion to the yield strength. In the
three dimensional stress space defined by the principal stresses, the Drucker-Prager criterion describes a cone. If
parameters M and cq are directly derived from the results of triaxial compression tests (σ2 = σ3), the strength of rock for
different stress states, e.g. close to a bore hole wall, may be severely overestimated, as the impact of the intermediate
principal stress on rock strength is overestimated. As known from the representation by Mohr’s circle, the Mohr-Coulomb
criterion just depends on the maximum and the minimum principal stresses
1 + sin(φ) 2 ⋅ cos(φ)
F = σ1 − ⋅ σ3 − ⋅c Eq. 7.38
1 − sin (φ) 1 − sin (φ)
ISAMGEO Chalk Model

47
where ϕ is the friction angle and c is the cohesion. The correspondence with parameters M and cq of the Drucker-Prager
criterion is as follows (in case of σ2 = σ3):
6 ⋅ sin(φ)
M = Eq. 7.39
3 − sin(φ)
6c ⋅ cos (φ)
cq = Eq. 7.40
3 − sin(φ)
For the implementation of the Mohr-Coulomb criterion into finite element codes, a formulation based on invariants is
preferred:
sin φ ⋅ sin θ
F = (
J 2 ⋅ cos θ +
3 ) − (p m ⋅ sinφ + c ⋅ cos φ ) Eq. 7.41
As the Mohr-Coulomb criterion does not account for the stabilizing effect of the intermediate principal stress in case of σ2
> σ3, it underestimates rock strength for stress states differing from triaxial compression. For a Drucker-Prager criterion
and a Mohr-Coulomb criterion even extremely high mean effective stresses will not induce plastic behavior, as long as
the deviatoric stress is small. The yield surface used for the ISAMEO Chalk Model overcomes this limitation. It is
described in Yield surface (p.50).
Flow rule and plastic potential

The plastic strain increments are evaluated by means of the flow rule from the plastic potential Q:
{Δε } = λ ⋅ { δQδσ }
pl
Eq. 7.42
with
δQ δQ δQ δQ δQ δQ δQ T
{ } {
δσ
= , , , , ,
δσ x δσ y δσ z δτ xy δτ yz δτ zx } Eq. 7.43
λ is a positive multiplier, which is evaluated below. In the simplest case the plastic potential coincides with the yield
surface F. The flow rule is then called “associated”. Usually, associated flow overestimates volumetric plastic strains
associated with plastic shear. Therefore, “non-associated” flow rules are introduced with Q ≠ F. A common approach in
the case of the Mohr-Coulomb criterion is to replace the friction angle in the criterion by the smaller angle of dilatancy ψ.
The parameter λ is determined from the condition that in case of ideal plastic behavior (no hardening or softening),
during plastic flow the state of stress must remain on the yield surface:
{ δFδσ }
T
dF = ⋅ {Δσ } = 0 Eq. 7.44
The combination of Equations 7.36 (p.47), 7.42 (p.48) and 7.44 (p.48) and resolving for λ yields:
δF T
λ =
{ } ⋅ D ⋅ {Δε}
δσ el
Eq. 7.45
{ } ⋅ D ⋅ { δQδσ }
δF T
δσ el
and:
{ δQδσ } ⋅ { δFδσ } ⋅ D
T
D el ⋅
{Δσ } =
( D el −
{ δFδσ } ⋅ D ⋅ { δQδσ }
T
el
el
) ⋅ {Δε } = D ep ⋅ {Δε } Eq. 7.46
For the implementation of elastoplasticity into a finite element code, the elastoplastic matrix D ep can be used instead
of the elastic matrix when calculating the stiffness matrix for the solution of the system of unknown displacements. It
ISAMGEO Chalk Model

48
should be noted that the elastoplastic matrix D ep is a non-symmetric matrix, if a non-associated flow rule is used (Q ≠
F). Accordingly, also the stiffness matrix, which is formed from D ep becomes non-symmetric and a corresponding
solver is required. Alternatively, the matrix may be approximated by a symmetric matrix, e.g. by
approx = ( D ep + D ep T
D ep )/2 Eq. 7.47
Hardening and softening

While it is essential for plastic behavior that a stress state stays on the yield surface, an increase or decrease of the yield
surface may be considered. In the case of isotropic strain hardening, the yield surface increases its size as a function of
plastic strain. Volumetric plastic strain increments are defined as:
Δε v ,pl = Δε xx , pl + Δε yy ,pl + Δε zz ,pl Eq. 7.48
Equivalent plastic strain is a scalar measure of deviatoric plastic strain and is defined via
Δε v ,pl 2 Δε v ,pl 2 Δε v ,pl 2
Δe pl =
2
{ ⋅ Δε xx , pl −
3 (
3
+ Δε yy , pl − ) (3
+ Δε zz , pl −
3 ) ( ) Eq. 7.49
+ Δε 2xy ,pl + Δε 2yz ,pl + Δε zx2 ,pl }

Alternatively, plastic work may be used to define hardening (“work hardening”). The case of kinematic hardening is not
considered here. It represents a shift of the yield surface in the stress space, not an increase. This results in an increase
of rock strength only for a selected stress path. In case of strain hardening/softening, equation 7.44 (p.48) is modified to:
δF T δF
dF = { }
δσ
⋅ {Δσ } +
δκ
⋅ Δκ = 0 Eq. 7.50
where κ is the hardening parameter (e.g. the volumetric plastic strain or the equivalent plastic strain). By introducing
δF 1
H = − ⋅ Δκ ⋅ Eq. 7.51
δκ λ
one obtains:
δF T
λ =
{ }
δσ
⋅ D el ⋅ {Δε}
Eq. 7.52
{ δFδσ } ⋅ δQ
T
D el ⋅ { }
δσ
+ H
and
{ δQδσ } ⋅ { δFδσ } ⋅ D
T
D el ⋅
{Δσ } =
( D el −
{ δFδσ } ⋅ D ⋅ { δQδσ } + H
T
el
el
)
In case of strain hardening with κ equal to volumetric plastic strain, one obtains
⋅ {Δε } = D ep ⋅ {Δε } Eq. 7.53
δF 1
H = − ⋅ Δε v ,pl ⋅ Eq. 7.54
δε v ,pl λ
From equation 7.42 (p.48) follows
δQ
Δε v ,pl = λ ⋅ Eq. 7.55
δp m
Thus λ cancels out in equation 7.44 (p.48):
ISAMGEO Chalk Model

49
δF δQ
H = − ⋅ Eq. 7.56
δε v ,pl δp m
and the elastoplastic [D]-matrix in equation 7.53 (p.49) can be formulated without knowing the magnitude of the
volumetric plastic strain increment.
The ISAMGEO Chalk Model without rate dependency

This section discusses the ISAMGEO Chalk Model without rate dependency. It describes:
• Elastic behavior
• Yield surface
• Tensile failure Criterion
• Plastic potential
• Special cases for Lode angles of θ = -30° and θ = 30°
• Variation of the cap geometry
Elastic behavior
Elastic behavior can either be isotropic or transversally isotropic. In both cases, a dependency on porosity can be used.
For isotropic behavior, the Young’s modulus E can be a linear or an exponential function of porosity Φ:
E = E0 − E n ⋅ Φ Eq. 7.57
or:
E = E 0 ⋅ e −d ⋅Φ Eq. 7.58
The Poisson’s ratio can be defined as a linear function of porosity:
v = v0 − v n ⋅ Φ Eq. 7.59
Note: The two options for porosity dependence are available as input options in the Petrel module. They are called
Linear and Exponential variation. A value should be entered in the required field.
Linear values are entered positive and exponential values entered negative.
Yield surface
The yield surface is the combination of the Mohr-Coulomb criterion with a cap, which describes pore collapse, as it was
proposed by Papamichos et. al. (Papamichos, E., Brignoli, M. and Santarelli, F. J. (1997), An experimental and
theoretical study of a partially saturated collapsible rock. Mech. Cohes.-Frict. Mater., 2: 251–278):
sin(ϕ) ⋅ sin(θ) pc − pm
F = (
J 2 ⋅ cos(θ ) +
3 )− ( p m + p t ) ⋅ sinϕ ⋅
pc + pt
Eq. 7.60
with
- pc = hydrostatic pore collapse strength
- pt = tensile strength
For very large pc -values, the model corresponds to a linear Mohr-Coulomb criterion. For J 2 = 0, the formula is
replaced by:
ISAMGEO Chalk Model

50
pm − pc
F = ⋅ (pm + pt ) Eq. 7.61
pc + pt
In case of p m < (2p c − p t ) / 3, the criterion yields dilatant shear failure behavior, while for p m > (2p c − p t ) / 3,
pore collapse with irreversible pore volume reduction is modelled. Figure 7.2 (p.51) shows the criterion in a p m − q
diagram, assuming σ 2 = σ 3 (triaxial compression). Figure 7.2 (p.51) is limited to positive (compressive) mean
effective stress. It is true that equation 7.61 (p.51) formally represents a tensile failure criterion (tension cut-off for
p m − p t ). However, a tensile failure criterion based on mean effective stress is considered as inadequate. For
example, a stress state with σ1 = 5 MPa, σ2 = 3 MPa and σ3 = -5 MPa yields pm = 1 MPa, and even pt = 0 would not
give tensile failure. Instead, the tensile failure criterion should be based on the minimum principal stress (see next
section). Thus p t in equation 7.60 (p.50) serves primarily to define the cohesion in case of shear failure. Equation 7.60
(p.50) gives for the intercept on the q - axis (i.e. for pm = 0)
6p t ⋅ sin ϕ pc
cq = ⋅ Eq. 7.62
3 − sin ϕ pc + pt
Figure 7.2. Yield surface for the ISAMGEO Reservoir Chalk Model in a pm - q diagram
Comparing equation 7.60 (p.50) with equation 7.41 (p.48) shows for pc >> pt the equivalence
c = p t ⋅ tan ϕ Eq. 7.63
The hydrostatic pore collapse strength pc is not constant, but increases with progressing pore collapse due to volumetric
strain hardening. In the ISAMGEO Reservoir Chalk Model, this hardening is described by
Δp c = h ⋅ max (0, Δε v , pl ) Eq. 7.64
where the hardening parameter can be defined as a function of porosity, either linear:
h = h0 − h n ⋅ Φ Eq. 7.65
or exponential:
−h n ⋅Φ
h = h0e Eq. 7.66
ISAMGEO Chalk Model

51
It should be mentioned that equation 7.64 (p.51) describes a hardening process, which only applies for an increase in
volumetric plastic strain (Δεv,pl > 0). The end cap thus cannot shrink, only expand (this will be modified later, when rate
dependent behavior and water weakening are introduced).
The derivative of F with respect to the stress vector is obtained by means of the chain rule. For J 2 > 0 this yields:
δF
{ }
δσ
=
δF
δp m
⋅ { } + δδFJ ⋅ { δδσJ } + δFδθ ⋅ { δσδθ }
δp m
δσ 2
2
Eq. 7.67
with
δF pc − pm pm + pt
δp m
= sin ϕ ⋅
pc + pt
⋅
1
⋅
2 pc − pm (− 1 ) Eq. 7.68
δF sin ϕ
δ J2 (
= cos θ ⋅ 1 + tan 3θ ⋅ tanθ −
3
⋅ (tan 3θ − tan θ ) ) Eq. 7.69
(here it must be considered that θ is a function of J 2, see equation 7.25 (p.44), and the chain rule must be applied) and
δF cos θ ⋅ sin ϕ
δθ
= (
J 2 ⋅ sin θ −
3 ) Eq. 7.70
For J 2 = 0, one obtains from equation 5.5:

δF 2p m + p t − p c
= Eq. 7.71
δp m pc + pt
where δF / δ J 2 = 0 and δF / δθ = 0.
For Lode angles of θ = 30° and θ = -30°, equation 7.69 (p.52) needs special attention, as tan(3θ) is not defined (see
Special cases θ = -30° and θ = 30° (p.54)).
Here, also the derivative of the yield surface with respect to the hydrostatic pore collapse strength is given, which is
needed for equation 7.54 (p.49). In case of J 2 > 0 one obtains
2
δF 1 (pm + pt ) 1
= − sin ϕ ⋅ ⋅ ⋅
δp c 2 ( p c + p t )2 pc − pm
pc + pt Eq. 7.72
pc − pm 1 ( p m + p t )2
= sin ϕ ⋅ ⋅ ⋅
pc + pt 2 (pc − pm ) ⋅ (pc + pt )
while for J 2 = 0 the derivative from equation 7.61 (p.51) reads:
δF − ( p m + p t )2
= Eq. 7.73
δp c ( p c + p t )2
Tensile failure criterion

σ 3, is considered as more meaningful than the
A tensile failure criterion, which is based on the minimum principal stress
criterion based on the mean effective stress (equation 7.100 (p.57)). With the uniaxial tensile strength σ tens , the
criterion takes the form:
ISAMGEO Chalk Model

52
F = − ( σ 3 + σ tens ) Eq. 7.74
The minimum principal stress can be expressed by means of the invariants:

2 ⋅ J 2 ⋅ sin (θ + 4π / 3)
σ3 = p m + Eq. 7.75
3
It is then straightforward to determine the derivatives with respect to the invariants:
∂F
= −1 Eq. 7.76
∂ pm
∂F 2 ⋅ sin (θ + 4π / 3) ∂ F ∂θ
= − + ⋅ Eq. 7.77
∂ J2 3 ∂ θ ∂ J2
∂F 2 ⋅ J 2 ⋅ cos (θ + 4π / 3)
= − Eq. 7.78
∂θ 3
(for equation 7.77 (p.53) it must again be considered that θ is a function of J 2, see equation 7.25 (p.44), and the chain
rule must be applied). The derivatives with respect to the stress components can be evaluated similarly as in equation
7.67 (p.52). The flow rule is associated (plastic potential identical with yield surface). Again, special attention is needed
for Lode angles around +30° and -30°, as the derivatives
∂θ tan 3θ
=
∂ J2 J2
and
∂θ 3
=
∂ J3 2 ⋅ J 2 J 2 ⋅ cos 3θ
are not defined. For this purpose, the yield criterion is evaluated for Lode angles of 30° and - 30° by replacing
sin (θ + 4π ) explicitly by -0.5 (for θ =-30°) and -1.0 (for θ =30°), respectively. This yields for θ =-30°:
∂F 1
= Eq. 7.79
∂ J2 3
and for θ =30°:
∂F 2
= Eq. 7.80
∂ J2 3
while the derivation with respect to the Lode angle is not considered. For Lode angles in the range of -30°< θ <−29° as
well as 29.5° < θ < 30°, the derivatives are interpolated between the results obtained for ± 30° and for -29° and 29.5°,
respectively. The elastoplastic [D]-matrix is then calculated according to equation 7.46 (p.48). For the correction of the
trial stress state back to the yield surface a radial return algorithm is applied:
F({ σ i })
{σ icorrected } = σ i − ∂F T ∂F Eq. 7.81
{ } { }
∂σ
⋅
∂σ
Yield surface with cohesion

In equation 7.99 (p.57), the parameter pt is used to describe the cohesion (see equation 7.63 (p.51)). At the same
time it defines in equation 7.100 (p.57) a tension cut-off criterion. To uncouple the cohesion and tension cut-off
parameters we introduce the cohesion c as an independent parameter. This would change equation 7.99 (p.57) to
sin φ ⋅ sin θ pc − pm
F = J 2(cos θ + ) − ( p m ⋅ sin φ + c ⋅ cos φ ) ⋅ Eq. 7.82
3 pc + pt
ISAMGEO Chalk Model

53
(see also equation 7.41 (p.48)), while equation 7.63 (p.51) remains unchanged.
In Tensile failure (p.52), a tensile failure criterion is described to replace the tensile failure criterion based on mean
effective stress (equation 7.41 (p.48)).
Plastic potential
The plastic potential Q corresponds to the yield surface F as described above, with the exception that the angle of friction
ϕ is replaced by the dilatancy angle ψ:
sin ψ ⋅ sin θ pc − pm
Q = (
J 2 ⋅ cos θ +
3 ) ( )
− p m + p t ⋅ sin ψ ⋅
pc + pt
Eq. 7.83
With ψ ≠ ϕ, this results in a non-associated flow rule, with the consequence of a non-symmetric stiffness matrix. In the
ISAMGEO model, this is by-passed for that part of the plastic potential, which corresponds to pore collapse: for
p m > 2p c − p t 3, i.e. on the right of the vertex of the curve shown on Yield surface (p.50), the dilatancy angle ψ is
( )/
replaced by the friction angle ϕ. For J 2 = 0, the plastic potential is identical with the yield surface:
pm − pc
Q = F = ⋅ (pm + pt ) Eq. 7.84
pc + pt
The derivative of the plastic potential is:
δQ δQ δp m δQ δ J2 δQ δθ
{ }
δσ
=
δp m
⋅
δσ { } +
δ J2
⋅
δσ { }
+
δθ
⋅
δσ { } Eq. 7.85
which differs from the derivative of the yield surface in case of p m < (2p c − p t ) / 3 because:
δQ pc − pm pm + pt
δp m
= sin ψ ⋅
pc + pt
⋅
1
⋅
2 pc − pm ( − 1 ) Eq. 7.86
δQ sin ψ
δ J2 (
= cos θ ⋅ 1 + tan 3θ ⋅ tan θ −
3
⋅ (tan 3θ − tan θ ) ) Eq. 7.87
and
δQ cos θ ⋅ sin ψ
δθ
− (
J 2 ⋅ sin θ −
3 ) Eq. 7.88
p m = (2p c − p t ) / 3 as discussed above, the derivative

Due to the switch from non-associated to associated flow at
δQ / δp m is continued, while δQ / δ J 2 and δQ / δθ are discontinuous at p m = (2p c − p t ) / 3.
Special cases θ = -30° and θ = 30°

For Lode angles of θ = -30° (triaxial compression, σ2 = σ3) and θ = 30° (triaxial extension, σ2 = σ3) equation 7.69 (p.52) is
not defined because tan(3θ) is not defined. This is overcome by the following method: In the equation for F (equation
7.60 (p.50)) cosθ and sinθ are replaced by their numeric values for θ = -30° and θ = 30°, before F is derivated with
respect to θ. Thus one obtains for θ = -30° with cosθ = √3/2 and sinθ = −1/2 :
sin φ pc − pm
F = J2 ⋅ ( 2
3
−
2 3 ) ( )
− p m + p t ⋅ sin φ ⋅
pc + pt
Eq. 7.89
This yields
ISAMGEO Chalk Model

54
δF 3 sinφ 3 − sin φ
= − = Eq. 7.90
δ J2 2 2 3 2
while δF / δθ = 0. For θ = 30° with cosθ = √3/2 and sinθ = 1/2, F takes the form:
sin φ pc − pm
F = J2 ⋅ ( 2
3
+
2 3 ) ( )
− p m + p t ⋅ sin φ ⋅
pc + pt
Eq. 7.91
and
δF 3 sinφ 3 + sin φ
= + = Eq. 7.92
δ J2 2 2 3 2
Again δF / δθ equals zero. In order to ensure a smooth transition when approaching θ = -30° and θ = 30°, derivatives
obtained as above and derivatives using the correct formulas for θ = -29.5° and θ = 29.5° are linearly interpolated for
-30° < θ < -29.5° and for 29.5° < θ < 30°. The same approach is used for the plastic potential Q.
Variation of the cap geometry

In order to offer an additional option to simulate the pore collapse behavior, the ISAMGEO Reservoir Chalk Model uses a
more general form of the yield surface:
sin φ pc − pm m
F = J2 ⋅ ( 2
3
−
2 3 ) ( )
− p m + p t ⋅ sin φ ⋅ ( pc + pt ) Eq. 7.93
For m = 1/2, this equation is identical with equation 7.60 (p.50). The typical range for m is between 0.4 and 0.6. The
above given derivatives are only slightly modified.
The factor 1/2 in equations 7.68 (p.52) and 7.72 (p.52) is replaced by m and ( p c − p m ) / ( p c + p t ) is replaced by
(pc − pm ) / (pc + pt ) m . The transition from dilatant shear behavior to pore collapse (= apex of the curve shown on
Yield surface (p.50)) is shifted from p m = (2p c − p t ) / 3 to p m = ( p c − m ⋅ p t ) / (1 + m ). The choice of m <

1/2 results in a sharper drop of the q vs. pm curve shown on Yield surface (p.50).
The ISAMGEO Chalk Model with rate dependency

This section discusses the rate dependent model. It describes:
• The basic ideas behind the rate dependent model
• Evaluation of the elastoplastic matrix [Dep]
• Stress correction to the yield surface
Basic ideas
A rate dependent model as suggested by de Waal and Smits (1998) (p.66) is suitable to reproduce the “creep”
behavior of chalk, which is confirmed by the results of lab experiments and field observations.
For the rate dependent model the pore collapse strength p c , which defines the yield surface in equation 7.60 (p.50) is
replaced by the pore collapse strength p cc , which depends on the volumetric plastic strain rate ε̇ v ,pl
ε̇ v .pl = Δε v .pl / t Eq. 7.94
according to:
ISAMGEO Chalk Model

55
ε̇ v ,pl b
p cc = pc 0 ⋅ ( )ε̇ 0
Eq. 7.95
ε̇ 0 is a reference volumetric plastic strain rate, commonly chosen as 0.1 %/h, and p c 0 is the hydrostatic pore collapse
strength at the reference rate. The exponent b is a material parameter.
Strain rate hardening as described for the rate independent model by equation 7.64 (p.51) now applies for the hydrostatic
pore collapse strength at the reference rate:
Δp c 0 = h ⋅ max (0, Δε v ,pl ) Eq. 7.96
The above given rate dependency of pore collapse strength is another hardening/softening mechanism, which affects the
current pore collapse strength concurrently to the conventional strain hardening. It is not independent from strain
hardening, as a finite volumetric plastic strain rate requires strain hardening. The differential form of equation 7.95 (p.56)
can be written as:
p cc ⋅ b
Δp cc = ⋅ Δε̇ v ,pl Eq. 7.97
ε̇ v .pl
If the volumetric plastic strain rate increases, p cc increases (hardening). If the volumetric plastic strain rate decreases,
the yield surface shrinks (softening). In case of a constant volumetric plastic strain rate there is no rate dependent
hardening or softening. If the constant rate equals the reference rate, the rate dependent and the rate independent model
are identical.
For the numerical implementation, an additional assumption is required. If one starts from a state with zero volumetric
plastic strain rate, p cc according to equation 7.95 (p.56) is zero. Correspondingly, zero stresses are the only admissible
starting state. This would require a very complex procedure to simulate a stress path from zero stresses to the in-situ
stresses. Such a tedious process can be by-passed by introducing the so-called “geological rate” ε geo . It can be
interpreted as the current creep rate at in-situ stresses, which results from previous geological processes (e.g.
sedimentation). Usually quite a small rate is assumed. A realistic magnitude for still ongoing natural subsidence in North
Sea chalk fields is 2·10-9 %/hr. For engineering purposes, such small rates are not of relevancy: a rate of 2·10-9 %/hr
yields within a period of 50 years less than 0.001% strain. Thus any volumetric plastic strain rates below such low
geological rates can be neglected. For the finite element implementation of the rate dependent model in the ISAMGEO
model, elastic rock behavior is assumed, as soon as the volumetric plastic strain rate falls below the geological rate
(which is an input parameter). It should be noted that the rate dependent model only considers positive volumetric plastic
strain rates, as they apply for pore collapse. In the course of shear failure, volumetric plastic strains are negative (dilatant
behavior). In this case, as well as in case of elastic behavior, the volumetric plastic strain rate, as it is used in the context
of the rate dependent model, is set to the above introduced geological rate.
The equations derived in the previous sections are to also applicable for the rate dependent model. However major
modifications are necessary in the context of hardening and softening and are presented below.
Evaluation of the elastoplastic matrix [Dep]

For establishing the stiffness matrix, it is necessary to define the elastoplastic matrix D ep , which combines the stress
increment and the strain increment:
{Δσ } = D el ⋅ ({Δε } − {Δε pl }) = D ep ⋅ {Δε } Eq. 7.98
Again, the derivation is based on the flow rule (equation 7.42 (p.48)) and the condition dF = 0 for plastic behavior,
which now includes the strain hardening and the hardening and softening resulting from changes in the volumetric plastic
strain rate. The flow rule is still given by
{Δε pl } = λ ⋅ ∂ σ{ ∂Q }
The yield surface F takes for m = 1/2 the form
ISAMGEO Chalk Model

56
sin ϕ ⋅ sin θ p cc − p m
F = J 2 ⋅ cos θ ⋅ ( 3 )
− ( p m + p t ) ⋅ sin ϕ ⋅
p cc + p t
Eq. 7.99
and for J2 = 0
p m − p cc
F = ⋅ (pm + pt ) Eq. 7.100
p cc + p t
with according to equation 7.95 (p.56). Thus F is also a function of the volumetric plastic strain rate ε v ,pl .
p cc
Accordingly, the equation for dF = 0 (compare with equation 7.50 (p.49)) changes to
{ ∂∂Fσ } ∂F δF
T
dF = ⋅ {Δσ } + ⋅ Δε v ,pl + ⋅ Δε̇ v ,pl = 0 Eq. 7.101
δ∂ ε v ,pl δε̇ v ,pl
In the following, a time increment Δt is considered, during which stress increases by {Δσ } and volumetric plastic strain
by Δε̇ v ,pl . The volumetric plastic strain rate at the beginning of this time increment, which is known, is denoted as
Δε̇ old
v ,pl . Then the volumetric plastic strain increment can be written as
∂Q
Δε v ,pl = λ ⋅ Eq. 7.102
∂ pm
while the change in the volumetric plastic strain rate amounts to
Δε v ,pl λ ∂Q
Δε̇ v ,pl = − ε̇ old = ⋅ − ε̇ old Eq. 7.103
Δt v ,pl Δt ∂ p m v ,pl
With these equations and {Δσ } from equation 7.98 (p.56), dF = 0 can be written as
∂F T ∂Q
dF =
∂σ{ } ⋅ D el ⋅ {Δε } − λ ⋅ (∂σ { }) Eq. 7.104
∂F ∂Q ∂F λ ∂Q
+
∂ ε v ,pl
⋅λ ⋅
∂ pm
+
∂ ε̇ v ,pl
⋅ ⋅
Δt ∂ p m (
− ε̇ vold,pl = 0 )
This equation can be used to determine the plastic multiplier λ. Resolving for λ yields:
∂F T
λ =
{ } ∂σ
⋅ D el ⋅ {Δε}
−
∂F
⋅
ε̇ vold,pl Eq. 7.105
H0 − H C − H R ∂ ε̇ v ,pl H0 − H C − H R
where
{ ∂∂Fσ } { ∂∂Qσ }
T
H0 = ⋅ D el ⋅ Eq. 7.106
while HC corresponds to the rate independent strain hardening term (- H in equation 7.56 (p.50)):
∂F ∂Q ∂F ∂ p cc ∂Q
HC = ⋅ = ⋅ ⋅ Eq. 7.107
∂ ε v ,pl ∂ pm ∂ p cc ∂ ε v ,pl ∂ pm
The hardening/softening term related to the rate dependent behavior is
∂F ∂Q 1
HR = ⋅ ⋅ Eq. 7.108
∂ ε̇ v ,pl ∂ pm Δt
If λ is replaced in equation 7.104 (p.57), you obtain:
ISAMGEO Chalk Model

57
∂Q ∂F T
{Δσ } = D el ⋅ {Δε } −
D el ⋅ { } { }
∂σ
⋅
∂σ
⋅ D el {Δε}
H0 − H C − H R Eq. 7.109
∂ ε̇ vold,pl
+ D el ⋅ { ∂∂Qσ } ⋅ ∂ ε̇∂ F
v ,pl
⋅
H0 − H C − H R
This equation shows that in addition to an elastoplastic matrix D ep given by
∂Q ∂F T
D ep = D el −
D el ⋅ { } { }
∂σ
⋅
∂σ
⋅ D el
Eq. 7.110
H0 − H C − H R
a right hand side contribution must be considered, when the system of equations for unknown displacements is
established. This force must be calculated from the stress-like term
old
∂Q ∂F ε̇ v ,pl
{Δσ r } = D el ⋅ { }
∂ σ ∂ ε̇ v ,pl
⋅
H0 − H C − H R
Eq. 7.111
The above quoted equations contain the term ∂ F / ∂ ε̇ v ,pl , which can be calculated for J 2 > 0 as
∂F ∂F ∂ p cc
= ⋅
∂ ε̇ v ,pl ∂ p cc ∂ ε̇ v ,pl
Eq. 7.112
p cc − p m 1 ( p m +p t )2 p cc ⋅ b
= − sinϕ ⋅ ⋅ ⋅ ⋅
p cc + p t 2 ( p cc − p m ) ⋅ ( p cc − p t ) ε̇ v ,pl
and for J 2 = 0 as:
∂F ∂F ∂ p cc − ( p m + p t )2 p cc ⋅ b
= ⋅ = ⋅ Eq. 7.113
∂ ε̇ v ,pl ∂ p cc ∂ ε̇ v ,pl ( p cc + p t )2 ε̇ v ,pl
There is a problem with these equations, when they are used for the formulation of the elastoplastic matrix D ep , at the
time of establishing the stiffness matrix, the value of ε̇ pl

v for the current time increment is not known. As an
approximation, it is replaced by ε̇ old
v ,pl . It should be noted that the actual implementation of equation 7.111 (p.58)
neglects the term HR in the denominator, as this produced the expected behavior in a better way.
Stress correction to the yield surface

This section discusses stress correction to the yield surface when using the rate dependent model. It describes:
• Rate independent behavior
• Rate dependent behavior
• A special case where the plastic multiplier λ is evaluated as a negative number
• Radial return in case of shear failure
• Unloading to elastic behavior
ISAMGEO Chalk Model

58
Rate independent behavior

Once the displacement increments have been determined by the solution of the system of equations and strain
increments have been evaluated for the individual integration points, the corresponding stress increments must be
determined. In case of elastoplastic behavior, the resulting stress state must be corrected back to the yield surface F, if
the condition F ≤ 0 is not fulfilled. This procedure is rather straightforward in case of rate independent behavior and is
outlined below. This serves as the base for the description of the more complex correction approach in case of rate
dependent behavior.
Starting point is the stress state obtained in the previous simulation step, denoted as { σ n } and the calculated strain
increment for the current state {Δε }.
In a first step, the so-called trial stress state { σ tr } is calculated from {Δε } , assuming elastic behavior:
{ σ tr } = { σ n } + D el ⋅ {Δε } Eq. 7.114
It is then checked, whether this stress state is inside the yield surface, i.e. whetherF ({ σ tr }) < 0. If this is the case,
elastic behavior applies and { σ tr } is an admissible stress state. The new stress state { σ n +1} equals { σ tr }. If
F ({ σ tr }) = 0, the material behaves plastic, but { σ tr } is also an admissible stress state, no further actions are
required.
If F ({ σ tr }) > 0, the response is plastic, and a correction of { σ tr } towards an admissible stress state on the yield
surface F = 0 is required.
Here the sub-increment approach is described, as it is used in the ISAMGEO model. For this approach, the strain
increment {Δε } is subdivided into k sub-increments (k = 45). Using the stress state from the previous simulation step
{ σ n }, the plastic multiplier λ is calculated for the sub-increment according to equation 7.52 (p.49):
∂F T
{Δε}
λ =
{ ∂σ }
⋅ D ⋅
k el
Eq. 7.115
{ ∂∂Fσ } ⋅ D ⋅ { ∂∂Qσ } + H
T
el
with H according to equation 7.56 (p.50). When λ is evaluated, the corresponding plastic strain increment is known and a
stress sub-increment can be calculated from the elastic part of the strain sub-increment:
∂Q
{Δσ i } = D el ⋅ ( {Δε}
k
−λ ⋅{
∂ σ })
Eq. 7.116
An intermediate stress state is obtained by adding this stress increment to { σ n }:

{ σ i } = { σ n } + {Δσ i } Eq. 7.117
λ is also used to calculate a sub-increment of the volumetric plastic strain:
∂Q
Δε v( ,pl
i)
=λ ⋅ Eq. 7.118
∂ pm
(i )
If Δε v ,pl > 0, pore collapse is given and the hydrostatic pore collapse strength is increased by hardening:
Δp c = h ⋅ Δε v( ,pl
i)
Eq. 7.119
Now the next step of the sub-increment loop starts: the plastic multiplier is calculated on the basis of the obtained
intermediate stress state σ i and the updated hydrostatic pore collapse strength. A new stress sub-increment and a
{ }
new intermediate stress state
ISAMGEO Chalk Model

59
{ σ i } = { σ iprevious } + {Δσ i } Eq. 7.120
(i )
are evaluated. The hydrostatic pore collapse strength is updated, if Δε v ,pl > 0. At the end of the loop, when the full
{Δε } has been applied, the intermediate stress state { σ i } represents the final stress state { σ n +1}. When the
increments are sufficiently small, this stress state should be on or close to the yield surface F, i.e. F ({ σ n +1}) ≈ 0. If
the obtained state of stress is not in equilibrium with the applied loads, out-of-balance forces are calculated in the course
of an equilibrium check, which are applied in the next step of the simulation.
Note: The number of sub-increments k can be changed. The user can select pressure steps and the number of
increments within each step. If the change in pressure is large during a pressure step the number of increments should
be increased until a convergent solution is obtained. As an alternative the number of sub-increments could also be
increased. Different numbers of pressure steps (increments) and different sub-increments should be tried to decide if the
solution is in fact convergent. This will depend on each model and its non-linearity, so explicit guidance for all models on
the exact number of increments and sub-increments that might be needed is not possible.
Rate dependent behavior

In principle, the same approach is followed in case of rate dependent behavior. The difference results from the fact that
the yield surface is now also a function of the volumetric plastic strain rate. This must be considered when checking for
F>0. Furthermore, in addition to strain hardening, there is softening/hardening from changes in the volumetric plastic
strain rate. Both mechanisms must be considered during the stress correction procedure, which is complicated by the
fact that strain rate and strains are closely related.
A trial stress state { σ tr } is calculated as described above using the matrix D el . When checking the yield criterion F
for the trial stress state, one must consider the following scenarios:
• the volumetric plastic strain rate from the previous simulation step equals to the geological rate ε geo , which applies
for elastic behavior or shear failure (dilatant): in this case, p cc as used in the yield criterion is set to its minimum
value, defined by the geological rate
ε̇ geo b
p cc ,min = p c 0 ⋅ ( ) ε̇ 0
Eq. 7.121
If the yield criterion check givesF ({ σ tr }, p cc ,min ) < 0, the trial stress state corresponds to the wanted final
stress state { σ n =1}, the material behavior is elastic. If F ({ σ tr }, p cc ,min ) > 0, but ∂ Q / ∂ p m < 0 (shear
failure, dilatant), the behavior is rate independent, and the stress correction approach as outlined above applies. The
volumetric plastic strain rate is kept as ε geo . In case of F σ tr , p cc ,min > 0 and ∂ Q ∂ p m > 0, pore
({ } ) /
collapse is encountered, volumetric plastic strain develops and the volumetric plastic strain rate increases according
to the approach outlined below.
• the volumetric plastic strain rate from the previous simulation step ε̇ old
v ,pl is greater than the geological rate ε̇ geo , i.e.
chalk is in the pore collapse state: in this case, p cc to be used in the yield criterion is given by
ε̇ old b
p cc = pc 0 ⋅ ( )
v ,pl
ε0
Eq. 7.122
If the yield criterion check now gives F ({ σ tr }, p cc ) < 0, it does not necessarily mean that the material behavior
is elastic. The actual volumetric plastic strain rate may decrease and accordingly also the hydrostatic pore collapse
strength p cc . In this case, the rate ε̇ old
v ,pl is arbitrarily decreased by 5% to a test rate:
˙
v ,pl = 0.95 ⋅ ε v ,pl
old
ε̇ test Eq. 7.123
ISAMGEO Chalk Model

60
for which a new (smaller) p cctest is calculated.

ε̇ test b
p cctest = pc 0 ⋅ ( )v ,pl
ε̇ 0
Eq. 7.124
If the yield criterion check still gives F ({ σ tr }, p cctest ) < 0, the test rate is further reduced and the yield criterion is
checked for the reduced pore collapse strength. This is repeated until the test rate either equals the geological rate
or the criterion check gives F ({ σ tr }, p cctest ) > 0. In the first case it is concluded that the trial stress state
corresponds to the wanted final stress state { σ n +1}, the material behavior is elastic. If F ({ σ tr }, p cc ,min ) > 0,
and ∂ Q / ∂ p m > 0, pore collapse will develop within this simulation step, the volumetric plastic strain and the
volumetric plastic strain rate will be evaluated according to the approach outlined below. As a decrease of p cc
affects also the shape of the yield surface on the left of the apex in the p m − q diagram, it may happen that
∂ Q / ∂ p m becomes negative, which corresponds to shear failure and dilatancy. This behavior is rate independent,
and the stress correction approach as outlined above applies. The volumetric plastic strain rate is set to ε̇ geo . If the
yield criterion check using the p cc from the previously obtained rate ε̇ old
v ,pl (equation 7.122 (p.60)) yields
F ({ σ tr }, p cc ) > 0, and ∂ Q / ∂ p m .0, pore collapse will continue in this simulation step, the approach to
calculate plastic strain and plastic strain rate is described below. If ∂ Q / ∂ p m < 0, rate independent shear failure is
encountered.
According to equation 7.86 (p.54), the condition ∂ Q / ∂ p m > 0 is for m = 1/2 equivalent to the condition
p m > (2p cc − p t ) / 3 or in the general case to p m > ( p cc − m ⋅ p t ) / (1 + m ).
As outlined above, λ must be known to evaluate the intermediate stress states in the course of the stress correction
routine:
∂Q
{Δσ i } = D el ⋅ ( {Δε}
k
−λ ⋅{
∂ σ })
Eq. 7.125
For λ, equation 7.105 (p.57) has been derived:
This equation contains, namely in HR (equation 7.108 (p.57)) and in ∂ F / ∂ ε̇ v ,pl (equation 7.113 (p.58)) the term
∂ p cc p cc ⋅ b p cc ⋅ b
= = Eq. 7.126
∂ ε̇ v ,pl ε̇ v ,pl (∂ Q / ∂ p m ⋅ λ) / Δt
If this is inserted into equation 7.125 (p.61), one obtains a quadratic equation for λ:
∂F T C ⋅ Δt ⋅ ε̇ old
λ ⋅ ( H0 − H C ) − λ ⋅
2
({ }
∂σ
⋅ D el ⋅ {Δε } + C + )
∂Q / ∂ pm
v ,pl
=0 Eq. 7.127
with
∂F
C = ⋅ p cc ⋅ b Eq. 7.128
∂ p cc
It can be resolved, resulting in
∂F
{ ∂∂Fσ }
T T 2
{ ∂σ }
⋅ D el ⋅ {Δε} + C ⋅ D el ⋅ {Δε} + C
λ 1,2 =
2 ⋅ ( H0 − H C )
± ( 2 ⋅ ( H0 − H C ) ) −
C ⋅ Δt ⋅ ε̇ old
∂Q / ∂ pm
v ,pl Eq. 7.129
ISAMGEO Chalk Model

61
As λ should be a positive multiplier, only the positive root is meaningful. For the stress correction using the sub-increment
approach, the evaluated strain increment is subdivided as described above. For the first sub-increment, the multiplier λ is
calculated on the basis of the stresses, the p cc and the volumetric plastic strain rate as obtained in the previous
simulation step. The subdivision into k sub-increments also applies to the time increment Δt :
∂F
{ ∂∂Fσ }
T {Δε} T {Δε} 2
{ ∂σ }
⋅ D el ⋅ ⋅ D el ⋅
λ =
k
2 ⋅ ( H0 − H C )
+C
± ( k
2 ⋅ ( H0 − H C )
+C
) −
C ⋅ Δt / k ⋅ ε̇ old
∂Q / ∂ pm
v ,pl Eq. 7.130
The plastic multiplier is used to evaluate a stress increment {Δσ i }, an intermediate stress state { σ i } as well as a sub-
(i )
increment Δε v ,pl of volumetric plastic strain.
Strain hardening is applied for the hydrostatic pore collapse strength at the reference rate:
Δp c 0 = h ⋅ max (0, Δε v( ,pl

i)
) Eq. 7.131
In the first sub-increment, the volumetric plastic strain rate is calculated as
ε v( ,pl
i)
λ ∂Q
Δε̇ (vi,pl
)
= − ε̇ old = ⋅ − ε̇ vold,pl Eq. 7.132
(Δt).k v ,pl Δt ∂ p m
A new p cc according to equation 7.95 (p.56) is calculated, using the updated Δp c 0 and the updated rate. In the
following sub-increments ε̇ old

v ,pl is replaced by the rate obtained in the previous sub-increment, when calculating λ:
Δε v( ,pl
i)
λ ∂Q
Δε̇ (vi,pl
)
= −1)
− ε̇ (vi,pl = ⋅ −1)
− ε̇ (vi,pl Eq. 7.133
Δt / k Δt ∂ p m
At the end of a sub-increment it may turn out that the yield criterion F = 0 is not fulfilled for the calculated volumetric
plastic strain rate and the according p cc . Most probably this results from the fact that rather small strain increments can
change strain rates and also the pore collapse strength for the given strain rate considerably. It is thus assumed that this
inconsistency results from a wrong rate, which should at least partly be corrected. Therefore a correction mechanism has
been implemented. The equation F = 0 (equations 7.100 (p.57) and 7.101 (p.57), respectively) is resolved for the pore
collapse strength. This strength p cn reveals the strength, which is in equilibrium with the current intermediate stress
state. For J 2 it is evaluated as:
pt ⋅ B
p cn = p m + Eq. 7.134
1−B
where for m = 1/2:
sin φ ⋅ sin θ 2
B =
(
J 2 ⋅ cos θ +
3 ) Eq. 7.135
2
(pm + pt )
Now from equation 7.95 (p.56), one can calculate the rate, which corresponds to p cn :
p cn 1/b
Δε v( ,pl
n)
= ε̇ 0 ⋅ ( )
p c 0n
Eq. 7.136
The rate as calculated according to equation 7.132 (p.62) (for the first sub-increment) or 7.133 (p.62), is then arbitrarily
corrected:
Δε̇ (vi,pl
,corrected )
= 0.95 ⋅ Δε̇ (vi,pl
)
+ 0.05 ⋅ Δε v( ,pl
n)
Eq. 7.137
ISAMGEO Chalk Model

62
(n ) (n )
As only 5% of the rate in equilibrium with F = 0 is applied, this correction has a major impact only if Δε v ,pl and Δε v ,pl
differ considerably. It should be kept in mind that this correction approach does not change the pore collapse strength at
the reference rate, as strain hardening is not considered.
The plastic multiplier λ is evaluated as a negative number

If load increments are chosen too big or if plastic strain increments are so small that other inaccuracies may have an
impact, the plastic multiplier λ may become negative. Theoretically, this should not happen, as long as the yield surface
is convex in the stress space. Nevertheless, λ<0 cannot be excluded within a complex simulation. In this case an
arbitrary decrease or increase of the currently evaluated volumetric plastic strain rate is tried to see if this resolves the
problem. If not, λ is set to zero, i.e. in the current sub-increment the elastic stress increment is applied.
Radial return in case of shear failure

An additional radial return algorithm in the case of shear failure ( ( p m < (2p cc − p t ) / 3 or in the general case
p m < ( p cc − m ⋅ p t ) / (1 + m ))), when hardening and rate dependency do not play a role, is no longer
recommended.
Unloading to elastic behavior

In the ISAMGEO model, the question of whether the elastic [D] or elastoplastic [D]- matrix is used, when the stiffness
matrix is formed, is answered on the basis of conditions in the previous simulation step. If this step is elastic, the elastic
matrix is used. If the previous step was elastoplastic, the elastoplastic matrix is used again. This approach is correct. as
long as the loading characteristic does not change. For full field compaction/subsidence studies, it can happen that in the
course of water injection the chalk behavior changes from elastoplastic compaction to elastic behavior. In this case (as
well as in other unloading scenarios) the strain increment {Δε } is calculated with the elastoplastic matrix, which
represents a much softer stiffness, whereas the elastic matrix should have been used. Accordingly, the strain increment
is greater than the correct value. If now the stress increment is calculated using the elastic matrix, the stress increment
may be severely overestimated. In the ISAMGEO model, this situation is therefore handled in such a way that the stress
increment is calculated using the same elastoplastic matrix, as was used for establishing the stiffness matrix, although
F σ tr < 0. If a finite element code uses a more refined implicit approach for establishing the stiffness matrix (e.g. a
({ })
Newton-Raphson iteration), this practice should not be necessary.
In-situ state
For the initial stresses, from which the simulation starts, it must be assured that they are in equilibrium with the rock
strength. In case of pore collapse, F ≤ 0 must be satisfied for the minimum pore collapse strength, which results from the
assumed geological rate defined in equation 7.121 (p.60).
In the context of the rate dependent model, it is often assumed that chalk at all porosities is in or close to the pore
collapse state for the geological rate. Therefore it can be quite a challenge to properly define the effective in-situ stresses
and the hydrostatic pore collapse strength at the reference rate, p c 0, from which according to 7.121 (p.60) the pore
collapse strength at the geological rate is calculated. It must be considered that
• total vertical stress is usually given as a function of depth by the gradient,
• pore pressure is extracted from the flow simulation model (or in the “non active” region also from a gradient),
• porosity is given by the static model,
• creep parameter b depends on porosity,
• pore collapse strength and (see below) creep parameter b may depend on water saturation.
Thus special approaches are necessary to arrive at an in-situ state, in which chalk of all porosities is in or close to the
pore collapse state at the geological rate. In the Petrel module the in-situ stress state should be given by the INITIALISE
or EXPLICIT INITIALIZATION methods and the stress state should satisfy the yield conditions initially.
ISAMGEO Chalk Model

63
Porosity update
In the ISAMGEO model, the following parameters maybe defined as a function of initial porosity :
• the Young’s modulus (equations 7.577.138 (p.50) or 7.587.139 (p.50)) and the Poisson’s ratio (equation 7.59 (p.50)),
• initial hydrostatic pore collapse pc
• the cohesion c
• the friction angle ϕ and the dilatancy angle ψ ,

• the tensile strength pt ,
• the creep parameter b.
and the hardening parameter h on current porosity
The dependency on porosity can be either a linear or an exponential function of porosity φ:
G = G0 − G n ⋅ φ Eq. 7.138
or:
G = G 0 ⋅ e −d ⋅φ Eq. 7.139
where G can be any of the material parameters above.
Note: The two options for porosity dependence are available as input options in the Petrel module. They are called
Linear and Exponential variation. A value should be entered in either of the required fields.
Linear values are entered positive and exponential values entered negative.
Note: If the above equations give incorrect values for any of the material parameters ( i.e. negative values ) then the
value G0 will be taken without any porosity dependence.
Hardening, however, is defined as a function of the updated porosity (compare with 7.65 (p.51) and 7.66 (p.51), which
also apply for the rate dependent model). The porosity change Δφ due to a volumetric strain increment Δε v is
evaluated as
Δφ = − Δε ν ⋅ (1 − φ ) Eq. 7.140
This can differ from other approaches to update porosity, especially in reservoir simulators. As equation 7.140 (p.64) is
not familiar in the reservoir mechanics community, its derivation is given in the following. Starting point is the definition of
porosity φ 0 for an initial state:
Vp 0
φ0 = Eq. 7.141
Vs + Vp 0
where V p 0 is the initial pore volume and V s is the solid volume. The definition of porosity φ 1, which corresponds to a
state, where the pore volume is decreased by ΔV p is given by:
V p 0 − ΔV p
φ1 = Eq. 7.142
V s + V p 0 − ΔV p
Accordingly
ISAMGEO Chalk Model

64
V p 0 ⋅ ( V s + V p 0 − ΔV p ) − ( V s + V p 0) ⋅ ( V p 0 − ΔV p )
Δφ = φ 0 − φ 1 = Eq. 7.143
( V s + V p 0) ⋅ ( V s + V p 0 − ΔV p )
Some arithmetic yields:
ΔV p ⋅ V s
Δφ = Eq. 7.144
( V s + V p 0) ⋅ ( V s + V p 0 − ΔV p )
Using
V s + V p 0 − ΔV p − V p 0 + ΔV p 0 Vs
1 − φ1 = = Eq. 7.145
V s + V p 0 − ΔV p V s + V p 0 − ΔV p
equation 7.144 (p.65) can be written as
ΔV p
Δφ = ⋅ 1 − φ 1)
Vs + Vp 0 (
Eq. 7.146
Neglecting deformation of the grains and with the usual small strain assumption, the volumetric strain increment can be
expressed as
− ΔV p
Δε v = Eq. 7.147
Vs + Vp 0
Here compaction is defined as positive (Δε v > 0). The combination of equations 7.146 (p.65) and 7.147 (p.65) yields
the above given equation 7.140 (p.64).
The integration of equation 7.140 (p.64) yields a relationship between porosity φ for a certain volumetric strain ε v and
initial porosity φ 0( ε v = 0)
φ = φ0 ⋅ e
(εv ) + 1 − e (εv ) Eq. 7.148
Water weakening
It is assumed that water weakening affects mainly the pore collapse strength. The effect was supposed to decrease with
decreasing porosity and should vanish for some “limit porosity”. Water weakening of chalk is based on a variation of rock
parameters with water saturation S w in the general form of:
P ( S w ) = P ( S w = 0) ± ( P ( S w = 0) − P ( S w = 1)) B Eq. 7.149
In order to keep the number of parameters for the chalk model within reasonable limits, we consider such a dependency
only for hydrostatic pore collapse strength at the reference rate and optionally also for hardening.
For the hydrostatic pore collapse strength at the reference rate it is assumed that it includes a portion, which depends on
water saturation. This portion Δp c ( oil −water ) describes the difference between the pore collapse strength for a fully oil
saturated chalk and a fully water saturated chalk. It is assumed that this difference does not change in the course of
hardening.
The decrease of Δp c ( oil −water ) as a function of increasing water saturation Sw is described by the formula:
Δp c = Δp c ( oil −water ) ⋅ (1 − S w ) Bww Eq. 7.150
that is the difference has its maximum at Sw = 0 and reaches zero for Sw = 1. The exponent Bww (the name is chosen
in order to avoid confusion with the Skempton coefficient B) typically has an order of magnitude of 15 - 20. Accordingly,
ISAMGEO Chalk Model

65
most of the pore collapse strength decrease takes place while the water saturation increases to about 25%. Derivation of
equation 7.150 (p.65) with respect to water saturation yields the strength decrease with increasing water saturation:
∂ Δp c
= Δp c ( oil −water ) ⋅ (1 − S w ) Bww −1 ⋅ B ww Eq. 7.151
∂ Sw
As the pore collapse strength p cc at the current volumetric plastic strain rate ε̇ v ,pl ,, which directly governs the plastic
behavior, depends on the pore collapse strength p c 0 at the reference rate via the equation
ε̇ v ,pl b
p cc = p c 0 ⋅ ( ) ε̇ 0
the water sensitivity also affects p cc . From equation 7.99 (p.57) and 7.82 (p.53), respectively, it can be seen that water
weakening can also have an impact on the shear strength. For full field models, in which one material class is used for
chalk with varying porosity, it is recommended to define Δp c ( oil −water ) as a function of (initial) porosity ϕ, for example:
Δp c ( oil −water ) = Δp c ( oil −water ),φ =0 + φ ⋅ Δp c ( oil −water ),n Eq. 7.152
Also hardening may be defined as a function of water saturation. The hardening parameter can be calculated from the
water saturation independent part as given by equation 7.65 (p.51) or 7.66 (p.51) and an additional water saturation
dependent part. The decrease in hardening is :
Δh ( S w ) = Δh ( oil −water ) ⋅ (1 − S w ) Bww Eq. 7.153
In this equation Δh ( oil −water ) denotes the difference in hardening between fully oil saturated chalk and fully water
saturated chalk. Again, a porosity dependency similar as in equation 7.152 (p.66) can be introduced.
It is important to note that only an increase of water saturation has an impact on pore collapse strength and hardening. If
water saturation decreases (for what reason ever), it will not result in an increase of the pore collapse strength nor of the
hardening capability.
Note: Due to differences in the implementation between the Pc and hardening the input values for Pc and hardening
relate to different points on the equations. The input Pc value is the value at initial Sw and the Sw dependent part gives
the decrease in Pc. For hardening the input value is the minimum hardening at Sw = 1 and the Sw dependent part gives
the height of the curve.
References
1. de Waal, J.A. and Smits, R. M. M., “Prediction of Reservoir Compaction and Surface Subsidence: Field Application
of a New Model”, paper SPE 14214, SPE Formation Evaluation (1998), 3, No. 2 347–356.
2. Papamichos, E, Brignoli, M., and Santarelli, F. J., “An experimental and theoretical study of a partially saturated
collapsible rock”, Mechanics of Cohesive-Frictional Materials, (1997) 2, No. 3, 251-278.
3. Kristiansen, T.G. and Plischke B. “History Matched Full Field Geomechanics Model of the Valhall Field Including
Water Weakening and Re-Pressurisation” paper SPE 131505, presented at the SPE EuropeC/EAGE Annual
Conference and Exhibition, Barcelona, Spain, (14–17 June, 2010).
ISAMGEO Chalk Model

66
Coupled porous media

Generally static analyses can be solved which are either drained or undrained. However real rock behavior is often time
related, with pore pressure and the hydraulic boundary conditions playing a part. To account for these effects, it is
necessary to combine the equations governing the flow of pore fluid through the soil/rock skeleton with the equilibrium
and constitutive equations of the rock mass. This theory is called fully coupled and couples the pore fluid flow and stress/
strain behavior. The equations outlined below are generally called Biot’s consolidation equations and make use of the
principle of effective stress. They are detailed below for one phase flow but can be extended for multi-component multi-
phase flow.
Coupled hydro-mechanical equations

The principle of effective stress is written as
{Δσ } = {Δσ ′} + {m }αΔp Eq. 8.1
where the stress and pore pressure are taken as tension positive. σ is the total stress and σ ′ is the effective stress and
{m } is equal to unity for the normal stress and zero for the shear components. α is known as Biot’s coefficient and is
defined as
KT
α = 1.0 − Eq. 8.2
Ks
where KT is the bulk modulus of the equivalent material (solid phase plus pores) and Ks is the bulk modulus of the
solid phase.
This definition of effective stress takes into account the volumetric strains caused by the uniform compression of the solid
particles by the pore fluid as well as the overall volumetric strain of the equivalent material. For soils the volume strain of
the solid particles is usually insignificant but for rocks this can become significant. The constitutive equation is then
expressed in terms of effective stresses as
{Δσ ′} = D ′ {Δε } Eq. 8.3
where D ′ is the effective constitutive matrix and inserted into the equilibrium equations
∇ {σ } + {F } = 0 Eq. 8.4
where {F } are the applied forces.

Flow of pore fluid flow through the porous medium is defined by Darcy’s law
{v } = − k {∇ h }
p Eq. 8.5
h = + i
γf { G }
where h is the hydraulic head, {i G } is a gravity vector and [k] is the permeability matrix. γ f is the unit weight of the pore
fluid.
The equation of continuity and compressibility for the fluid is given as :
∇ T {v } − Q = fluid accumulation terms Eq. 8.6
Terms relating to fluid accumulation are :

67
∂ ϵv ∂ϵ
rate of change of total volumetric strain = {m } T
∂t ∂t
(1 − n) ∂ p
rate of change of the solid volume due to pressure changes
Ks ∂t
n ∂p
rate of change of fluid density
Kw ∂t
′
1 T ∂ {σ }
and change of solid volume due to effective stress changes − {m }
3K s ∂t
∂ { σ ′} 1 ∂ {ε} {m} ∂ p
where substituting for
∂t
from the constitutive equation gives −
3K s (
{m } T D ′
∂t
+
3K s ∂ t )
where n is the porosity and Kw is the bulk modulus of the fluid.
The continuity equation is then given by
{m } T D ′ ∂ ε
(
∇ T {v } − Q = {m } T −
3K s )
∂t
Eq. 8.7
(1 − n) n 1 ∂p
+ + − {m } T D ′{m }
Ks Kw (3K s ) 2 ∂t
where Q represents any fluid flow (sources or sinks) and the terms on the RHS represent any fluid accumulation due to
fluid and matrix compressibility.
In equation 8.7 (p.68) the term is equal to 3KT for an isotropic material and the equation can be written as
∂ε (1 − n) n KT ∂p
∇ T { v } − Q = {m } T a + + − Eq. 8.8
∂t Ks Kw ( K s )2 ∂ t
Introducing Skempton’s coefficient β for an undrained test as
1 1
−
KT Ks
β = Eq. 8.9
1 1 1 1
n − + −
Kw Ks KT KS
it can be shown that the final term in equation 8.8 (p.68) can be rewritten as
(1 − αβ) 1
η =α Eq. 8.10
β KT
which is termed the storage capacity and is equal to 1 over the Biot Modulus.
Using a suitable finite element method such as the Galerkin method, minimum potential energy or principle of virtual
work the equations of equilibrium and continuity defined above become :
K {d {u } / dt } + C {dp / dt } = {dF / dt } Eq. 8.11

where
K = ∫Ω
B T D ′ BdΩ

68
C = ∫B Ω
T
D ′ BdΩ
{F } = ∫ N ∫N
T T ^
dbdΩ + d t dΓ
Ω Γ
and
T Eq. 8.12
KP {p } + C {d {u } / dt } + CP {dp / dt } = {Q } + {G }
where
(∇ N¯ ) T k (∇ N¯ )
KP = ∫ Ω γf
dΩ
C = ∫ N αBdΩ
T ¯ T
Ω
CP = ∫ N η N dΩ ¯ T ¯
Ω
K = ∫ (∇ N ) k ( i ¯ T
G ) dΩ
Ω
Equations 8.11 (p.68) and 8.12 (p.69) can be written as a set of ordinary differential equations:
0 0 {u} K C d {u } {dF / dt }
0 KP { }
p
+
C T CP { } {
dt p
=
{Q } + {G } } Eq. 8.13
To solve the above equation a time marching scheme is required. Assuming a linear variation in time the equation
becomes:
K C
C T − KP θ Δt t − CP { {u}p } k +1
K C
=
C T KP (1 − θ )Δt t − CP { {u}p } k
Eq. 8.14
{ dF }
+ { ({ Q } + {G })Δt t }
where Δt k is the consolidation time step at step (increment) k and θ is a time step parameter ranging form 0.0 to 1.0.
One-way coupled model

The fully coupled equations above can be solved for stress and fluid flow simultaneously. This can be extremely difficult
and time consuming especially for multi-phase flow problems. As a result different coupling methods have been
developed which involve solving the fluid and stress equations within separate simulators.
In a one-way coupled model the fluid flow equations are generally solved using a reservoir simulator (ECLIPSE/
INTERSECT) and pressure, temperature, saturation results passed to the geomechanical simulator (VISAGE) but no
results from the geomechanical simulator are used in the reservoir simulator. In the figure below “n” represents the time
steps.

69
Figure 8.1. One-way coupling
Two-way coupled model with permeability updating

The first extension to one-way coupling, is to dynamically update the permeability using results from the geomechanical
model as shown below. To update permeability either functions or tables are used which define relationships of
permeability against strain or porosity.
Figure 8.2. Two-way coupling

70
Permeability updating methods

Numerous methods are available to define permeability updating. These methods are split between intact material and
fault/fracture options. These options are detailed below.
Intact material permeability update methods

There are two types of method, either using predefined functions or tables defining permeability multipliers KM.
Generally permeability multipliers are calculated which are used with the initial permeability [K0] to get the updated
(current) permeability [K]. Initial permeability is read from the reservoir simulation and porosity is either read from the
simulation files for reservoir cells or input as a material property. The diagonal elements of the permeability tensor are
extracted from reservoir simulation giving the initial permeability tensor :
Kx 0 0
K0 = 0 Ky 0
0 0 Kz
The permeability multiplier then multiplies all entries equally.
Current porosity is used and calculated as (assuming compression positive) :
ϕ = ( ϕ 0 − Δϕ ) / (1 − Δϕ )
Δϕ = Δϵ v
where
ϕ0 = initial porosity
Δϕ = change in porosity
Δϵ v = change in volumetric strain
Change in volumetric strain is calculated from the initialization step of the simulation.
Kozeny Carman function

The Kozeny-Carman equation relates permeability to porosity via
2
K = ϕ 3 c S 2(1 − ϕ )
/
where c is a constant and S the specific surface. Assuming c and S are constant the updated permeability can be
Eq. 8.15
calculated based on the initial and current porosities :
ϕ 3 / (1 − ϕ )2
K = K0
ϕ 03 / (1 − ϕ 0)2
Kozeny-Poiseuille function
Permeability relationship based on the initial and change in porosities :

71
(1 + Δϕ / ϕ )3
K = K0
(1 + Δϕ)
Polynomial law function

User defined variable “ p ” used in the calculation :
K = K 0 (ϕ / ϕ 0) p
Kozeny table
User defined table of permeability multiplier versus change in porosity.
[K] = [K0]* KM
Equivalent plastic strain table

pl
User defined table of permeability multiplier versus equivalent plastic strain ( ϵ eq ) where :
2 dev 1
ϵ eqpl = ϵ : ϵ dev ; ϵ dev = ϵ − tr (ϵ )
3 3
[K] = [K0]* KM
Dual table
pl
Consists of two user defined tables of permeability multipliers versus equivalent plastic strain ( ϵ eq ) and volumetric total
strain (ϵv)
The multiplier is calculated from one of the following options :
if equivalent plastic strain > 0 use equivalent plastic strain to calculate permeability multiplier
else use total volumetric strain to calculate permeability multiplier
[K] = [K0]* KM
Dual table with plasticity differentiation

pl
Consists of three user defined tables of permeability multipliers versus equivalent plastic strain ( ϵ eq ) and volumetric
pl
plastic strain ( ϵ v ) and volumetric total strain (ϵv)
The multiplier is calculated from one of the following options :

Use equivalent plastic strain if failure on cap to calculate permeability multiplier
Use volumetric plastic strain if shear failure to calculate permeability multiplier
Use volumetric total strain if elastic to calculate permeability multiplier
[K] = [K0]* KM
Intact porosity table

Use defined table of permeability versus porosity. Calculate initial Kt0 and current Kt permeabilities from the table using
the initial and current porosities and then calculate permeability multiplier :

72
KM = Kt/Kt0
[K] = [K0]* KM
Note: KM will modify the initial permeability as usual. However the KM parameter is calculated purely from the input table.
For consistency the initial permeability calculated from the table using the initial porosity should match the initial
permeability extracted from the reservoir simulation.
Fault/fracture permeability update methods

Updating of fault/fracture permeability uses the following method :
• First assign a global permeability to the fault/fracture. If a single porosity/permeability model then this is currently set
as half the permeability from the reservoir simulation. The other half is assigned to the intact rock. For dual porosity/
permeability models then the intact rock and fractures are assigned the corresponding permeabilities from the matrix
with the fracture permeability divided equally amongst each discontinuity.
• Calculate the local normal Kn and shear Ks fault/fracture permeability from the global permeability for each fault/
fracture.
• Calculate the change in permeability for each fault/fracture (using either of the methods below)
• Calculate global permeability change by converting local permeability to global and sum for all faults/fractures.
• Set updated permeability = original permeability + permeability change.
• Note: The final updated permeability will usually be a full tensor — only the diagonal terms are passed to the
reservoir simulator.
Standard table
Here permeability is updated from tables of shear permeability multiplier (mult_s) versus normal strain and normal
permeability multiplier (mult_n) versus shear strain.
Normal strain is calculated as the sum of the elastic and plastic normal strains plus the initial gap.
Shear strain is calculated as the sum of the elastic and plastic shear strains.
Here Kn and Ks are the local normal and shear permeabilities of the discontinuity. The updated permeability for each
discontinuity is calculated as :
mult_n * mult_s 0 0
K Mj = 0 mult_s 0
0 0 mult_n * mult_s
Ks 0 0
K 0j = 0 Kn 0
0 0 Ks
ΔK j = K Mj T
* K 0j * K Mj − K 0j
ΔK jis the change of permeability in the local coordinate system, here j refers to each discontinuity. The local
permeabilities are transformed to the global co-ordinate system and summed over the number of discontinuities then
added to the initial global permeability [K0 ].

73
Discontinuity porosity table

Permeability is updated from a table of permeability versus porosity.
Normal strain is calculated as the sum of the elastic and plastic normal strains plus the initial gap and it is assumed that
normal strain is equivalent to the change in porosity.
Using the table original KT0 and current KT permeabilities are calculated from the initial and current porosities.
Here current porosity = initial porosity + normal strain.
K M = KT / KT 0
ΔK j = K 0j * K M − K 0j
Here the enhancement is used equally for all permeability directions. This means that there is a neutral direction
enhancement available for discontinuities.
ΔK jis the change of permeability in the local coordinate system, here j refers to each discontinuity. The local
permeabilities are transformed to the global co-ordinate system and summed over the number of discontinuities then
added to the initial permeability [K0 ].
Reservoir simulation notes

When using permeability updating it is important to ensure that it will not be performed twice. That is once from the
coupling with VISAGE and once within the reservoir simulation.
Another issue is the well connection factors used by the reservoir simulator. Depending on the analysis it maybe required
to update these as the permeability is changing.
ECLIPSE
ECLIPSE specifies transmissibility multiplies under the keyword ROCKTAB or related keywords. When this keyword is
used the transmissibility multiplier should be set to 1.0 prior to running the coupling process to avoid updating the
permeability twice. Another ECLIPSE keyword which may result in incorrect permeability is the MULTIPLY keyword in
the EDIT section.
To force ECLIPSE to recalculate the connection factor after the permeability has changed the COMPDAT or related
keyword should be used after each VISAGE step is reached. Also the data under COMPDAT must be specified correctly
for ECLIPSE to calculate connection factors based on the given permeability. See ECLIPSE Reference Manual and
Technical Description for full details.
INTERSECT
For INTERSECT the following nodes should be considered when preforming permeability updating :
• WellToCellConnections is used to specify connection factors.
• CompactionMultiplierTable is used for transmissibility multipliers.
Two-way coupled model with pore volume updating

The second extension to one-way coupling, is to dynamically update the pore volume using results from the
geomechanical model as shown below.
During the reservoir simulation, the pore volume change is computed using a user-defined pore volume compressibility
factor. The variation of the pore volume depends not only on this pore volume but also on the stress variations. To better

74
characterize the pore volume variations in the reservoir simulation, the pore volume variation computed from the
geomechanical modeling is used to update the pore volume in the reservoir simulation.
The coupled simulation is divided into stress steps. A stress step is delimited by user-defined meeting times between the
reservoir and geomechanical calculations. The coupled pore volume procedure is presented here for one stress step. In
the Figure below “k” represents iterations.
Figure 8.3. Two-way coupling with Pore Volume Updating
ECLIPSE — Pore Volume updating

To update pore volume two methods are available either using the MULTPV keyword or updating the ROCKTAB tables
in ECLIPSE.
Coupled procedure — MULTPV

In this method, the coupling is achieved using step changes in pressure. The rock compressibility is not updated within
each iteration. Instead of correcting the rock compressibility, pressures are corrected directly and the rock compressibility
is kept constant. The step changes in pressures may be seen as corrections to the pressure in the reservoir. This is due
to the corrections to the pore volumes which are needed due to inaccurate rock compressibility values. The corrections
to the pore volumes are made through a factor called MULTPV, which is updated every iteration. MULTPV is placed in
the SCHEDULE section of the ECLIPSE input file which essentially performs an undrained analysis. This causes a step
change in fluid pressure due to change in pore volumes using the fluid compressibility (see ECLIPSE reference manual).
The pore volume variation computed by ECLIPSE between any two stress steps is due to the factor MULTPV along with

75
PORV_MOD, which accounts for the rock compressibility. The pore volume variation from VISAGE is computed from the
volumetric strain occurring as a result of the pore pressure change between the two stress steps.
The equations for the pore volume variation from the two simulators are (assuming small strain and tension positive) :
From VISAGE :
αΔϵ v (Δp, Δσ)

PV geo = PV 0 1 +
φ0
where
PV = pore volume
Δϵ v = change in volume strain (function of change in total stress Δσ and pressure Δp
φ0 = initial porosity
α = Biots coefficient
subscript 0 refers to the initial state and “geo” refers to the pore volume as calculated by the geomechanical simulation.
From ECLIPSE :
PV res = PV 0 * f (p )
or
PV 2res − PV 1res
= (PORV_MOD 2 − PORV_MOD 1)
PV 0
where
f (p ) = user defined data (ROCKTAB)
PORV_MOD = pore volume modifier from the ROCKTAB table.
“res” refers to the pore volume as calculated by the reservoir simulation.
The procedure is described here for one stress step.

The stress step starts with a reservoir simulation. During the reservoir simulation the pore volume variation is ∆PVres
(one value per grid cell) . Then the geomechanical computation is performed and the associated pore volume variation is
∆PVgeo (one value per grid cell). Then the difference between the geomechanical and reservoir pore volumes is
computed. To allow for large differences a relaxation parameter β ϵ [0:1] maybe used. A correction in pore volume is then
computed accounting for the relaxation factor
ΔPV cor = β * (ΔPV geo − ΔPV res )

The aim is then to calculate a corrected pore volume PV res_cor from the current one on the basis of the calculated
correction.
Hence the evolution of the pore volume is defined as
PV res_cor = MULTPV * PV res

with

76
ΔPV cor
MULTPV = 1+
PV res
MULTPV is applied in the next reservoir simulation of the same stress step and then iterations are performed until
convergence is reached.
Note: To use pore volume updating with ECLIPSE the ROCKTAB keyword must be in the model. For models with the
ROCK keyword it will be replaced automatically with an equivalent ROCKTAB keyword. Currently only the ROCKTAB
keyword is supported for pore volume updating.
Note: Use of the MULTIPLY keyword in the ECLIPSE data file with the options PORO or PORV may results in pore
volumes being incorrectly updated.
Coupled procedure — ROCKTAB

The ROCKTAB table is updated for each grid block after each iteration by the use of new rock compressibility values
from the VISAGE calculations. This eliminates the need to update the pore volumes using the MULTPV array because
updating the rock compressibility will automatically attempt to bring the pore volume variation in ECLIPSE into accord
with those from VISAGE. The ROCKTAB table is calculated using the change in pressure from ECLIPSE and the change
in pore volume calculated by VISAGE
Pressure Pore Volume Multiplier

pt 1
p t + Δp Δϵ v (Δp, Δσ)
(1 + α )
φ0
where
pt = pressure at start of time step
Δϵ v = change in volume strain (function of change in total stress Δσ and pressure Δp )
φ0 = initial porosity
For use in ECLIPSE this table will be extended linearly to allow for larger changes in pressure in the ECLIPSE model.
Note: The memory/disk requirements for storing a different ROCKTAB table for each grid block can be very large. For
this reason the current implementation is restricted to ROCKTAB tables containing exactly five lines of data. For
ROCKTAB tables with different numbers of lines the user will have to expand or contract the table to use the ROCKTAB
pore volume method. The workflow will automatically expand any ROCKTAB tables so that there is one table per cell
using the existing distribution of ROCKTAB tables within the model. Again if the model uses the ROCK keyword it will be
replaced by an equivalent ROCKTAB table for each cell.
Pore volume convergence

During the iteration process convergence is achieved when the following condition is meet :
tol = max | PV i − PV i −1 | / max (PV i )

77
where PV is the VISAGE pore volume per cell at successive iterations “i” and “i–1”.
INTERSECT — Pore Volume Updating

Pore volume updating is available with INTERSECT via the MULTPV method of ECLIPSE. Using the method described
above pore volume multipliers are passed to INTERSECT via the following nodes :
• BulkCellSelectionFamily is used to specify the modified cells.
• CellPropertyEdit is used to select the PORE_VOLUME_MULTIPLIER.
and the data is provided via HDF5 files.

78
Dual Porosities/Permeabilities
Fluid transport in dual porosities/ permeabilities media is of prime interest for fractured reservoir. An extensive literature
exists on this subject since the pioneering work of Barenblatt and Zheltov (1960) ] for non-deformable medium. The
importance of the coupling between deformation and fluid flow has been recognized by may authors especially with
regards to the stress dependence of the permeability associated with the fracture network (see Bai and Elsworth (1992) ,
Benveniste (1987) among others). In the case of single phase flow, the mass balance in the matrix and fracture system
can be written as Berryman and Wang (1995) :
∂ ρfϕM
+ ∇ ⋅ ρfqM = − γ (pM − pF )
∂t
Eq. 9.1
∂ ρfϕF
+ ∇ ⋅ ρfqF = γ (pM − pF )
∂t
where ρf denotes the fluid density, ϕ the variation of the matrix and fracture porosities while q M and qF denotes the
fluid discharge vector for the matrix and fracture respectively. These fluxes are related to the corresponding pressure
gradient via Darcy's type laws.
Here we are interested in the coupling between geomechanical and flow simulation for fractured reservoirs modeled as a
dual porosity medium performed via two distinct simulators (flow and geomechanical one). We are especially interested
in the proper up-scaling of the poroelastic properties of such a dual porosity medium. Although, the flow simulation
requires the discretization of both fracture and matrix, the geomechanical simulation models the material as a
homogeneous one for each grid cell. The mesh size of the geomechanical simulations are such that the effect of multiple
fractures within the volume equivalent to one grid cell needs to be properly up-scaled. Moreover, when doing so, the
hydro-mechanical coupling still needs to properly capture changes especially with respect to permeability stress
dependence. The poromechanical behavior of a double porosity / double permeability media can be derived
macroscopically following Biot's theory of porous media (see for example, Berryman and Wang (1995) ):
Note: Contrary to other sections here we use the convention of solid mechanics where stresses are taken positive in
tension and negative in compression. The strains are therefore positive in dilation and negative in contraction.
d Σ = L E − bM d p M − bF d p F
dpM dpF
ϕ M = bM E + +
N MM N MF Eq. 9.2
dpM dpF
ϕ F = bF E + +
N FM N FF
where L denotes the drained elastic stiffness tensor of the fractured media, bM and p M , repectively, b F
pF and
denotes the Biot's coefficient tensor and pore pressure related to the matrix, and the fracture set respectively. Σ and E
denote the macroscopic stress and strain tensors respectively. The macroscopic variation of porosities of the matrix ϕ M
and of the fracture system ϕ F are related to the overall strain and the different pore pressures where the Biot's like
modulus N 's are such that by the reciprocity principle (see Coussy (2004) ) N FM = N MF . Let's recall here that in one
step of a staggered coupling solution to the coupled mechanics/fluid flow problem, the mechanical simulator solves for an
elastic (possibly plastic) problem with internal body forces related to the effect of the pore-pressure field changes in the
matrix and fractures.
79
Note: Pressures and stresses are to be understood as variation from a reference state here, hence the notation d p for
pressure changes.
One has to understand that due to the presence of a finite set of fractures the behavior of the medium is inherently
anisotropic. Isotropic behavior can be recovered only when the fractures are randomly oriented. The constitutive
parameters which appear in such a constitutive law are however not directly linked to the poroelastic parameters of the
intact matrix as well as the orientation and density of the fracture system. They are "macroscopic" parameters which
govern the poromechanical response of the medium at the length scale of one grid cell as depicted in figure 9.1.
Figure 9.1. A representative volume element (RVE) of a fractured reservoir
The goal is to consistently provide a link between intact matrix and fractures parameters and these macroscopic
parameters while keeping in mind the requirements of a coupled flow/geomechanical simulation (especially in regards to
the staggered coupling approach used between ECLIPSE and VISAGE).
The estimation of elastic moduli of media containing micro-cracks has drawn a large interest and a large number of
results are available in the literature (see Cornet (1993) , Berryman and Pride (2002) among others for a review). We will
not detail the derivation of these moduli — merely given the equations as used in VISAGE. We present in the following a
consistent and general way to estimate the macroscopic poroelastic properties of a double porosity medium made of a
poroelastic matrix and an arbitrary number of fractures.
Micro-macro relations of fractured poroelastic medium

Consider the case of an isotropic porous matrix with drained elastic moduli ( k m , g m ) Biot coefficient b m and Biot's
modulus Nm and arbitary planar cracks. The pore pressure within the matrix and the fractures are assumed to be
different and respectively equal to pm and p f . We model the representative volume element of such medium as a
poroelastic composite with different pore pressure. This results in the following equations to obtain a simple estimate of
the macroscopic poroelastic coefficients of such a double porosity medium described by the 9.1 (p.79).
Macroscopic constitutive law
d Σ = L E − b M dp M − b F dp F
d pM d pF
ϕM = bM E + +
N MM N MF
d pM d pF
ϕF = bF E + +
N FM N FF
80
Estimate of the compliance (stiffness) tensor

M = Mm + ℋ
L = M −1
Defect influence compliance tensor

BT
(
ℋ ijkl = P 32 ( B N − B T ) n i n j n k n l + n n δ + n i n k δ jl + n j n l δ ik + n j n k δ il )
4 ( i l jk )
Macroscopic Biot's coefficients
bM = L Mm I bm
b F = L ℋI
Macroscopic Biot's modulus
1 1 b m2
= +
N MM NM N FF
1
= I Mm bF
N FF
1 bm
= −
N MF N FF
In the geomechanical simulator we are interested in the relations for the macroscopic constitutive law only.
Here M m is the matrix compliance tensor. All other parameters as defined previously.
Arbitrary planar cracks

Here we describe in more detail a method for calculating the defects influence compliance tensor ℋ for arbitrary planar
cracks Berryman and Pride (2002) , Eshelby (1957) and Dormieux et. al (2006) .
−
We can introduce the following relation between the mean displacement discontinuity ui across the fracture and the
traction vector applied to the fracture:
−
u i = B ij σ jk n k Eq. 9.3
where B ij denotes the mean fracture compliance (m/Pa) and ni the normal to the fracture plane. Note that we have
used Einstein convention for summation on repeated indices. If one assumes an isotropy of the tangential components of
the fracture compliance (i.e. associated with shearing in the fracture plane), we can express the fracture compliance
tensor as:
B ij = B N n i n j + B T ( δ ij − n i n j ) Eq. 9.4
where BN and BT denotes the mean normal and tangential compliance of the fracture expressed in m/Pa. Assuming
planar fractures, the contribution of the defects to the macroscopic strain field in the RVE:
Δϵ = N ∫ Sc
1
2
( n ⊗ u + u ⊗ n )d S c
under stress applied conditions becomes:
81
1 − −
Δ ϵij = N S c
2
(
ni u j + u i nj )
Eq. 9.5
1
= N S c ( n i B jk σ kl n l + B ik σ kl n l n j )
2
where N is the number of cracks per unit of volume and Sc denotes the crack area. The product P 32 = N S c denotes
the ratio of the total of crack surface per unit of volume. In order to factor out the far-field stress from the previous
1
equation, we can choose the following general form for the stress tensor Σ k lmn = δ δ + δ lm δ kn
2 ( km ln ), such that
Σ k lmn σ mn = σ kl . One finally obtains (see Dvorak and Benveniste (1992) and Berryman and Pride (2002) ):
BT
(
Δ ϵij = P 32 ( B N − B T ) n i n j n k n l + )
n n δ + n i n k δ jl + n j n l δ ik + n j n k δ il ) σ kl = ℋ ijkl σ kl
4 ( i l jk
This allows us to write the macroscopic defect compliance tensor for a number of cracks of orientation ni and normal
and shear compliance BN and BT as:
BT
(
ℋ ijkl = P 32 ( B N − B T ) n i n j n k n l + n n δ + n i n k δ jl + n j n l δ ik + n j n k δ il )
4 ( i l jk ) Eq. 9.6
Note: Note also the relation with the crack density parameter P 32 = ρS x / a 3 where a denotes crack length scale.
Note: Note also that these equations are completely analogous to those described in Discontinuity model (p.19). The
frequency there being analogous to the P32 parameter.
Extension to multiple fracture sets

We have so far assumed that a single set of fractures (similar shape and orientation) is present in the RVE. The results
previously obtained can be extended to the case of a finite number of fracture set in a straightforward manner. In the
case of multiple fracture sets, the overall defect compliance influence tensor ℋ is simply obtained as the sum of the
effect of each fractures set ℋ β defined by equation (9.6 (p.82)) for a given orientation, fracture compliance and
intensity. In the case of N different fracture sets, one obtains:
ℋ= Σ ℋβ Eq. 9.7
β = 1, N
All the results previously derived for the estimate of the macroscopic poroelastic properties remains the same.
Implementation in VISAGE
This section gives information of how dual porosity/permeability models from ECLIPSE/INTERSECT are used in coupled
workflows with VISAGE. Currently if a dual porosity/permeability ECLIPSE/INTERSECT model is used two-way coupled
to VISAGE then only permeability updating is available.
The coupling process follows the workflow given in the Coupled Porous Media (p.67) section. The first difference is that
VISAGE extracts the matrix and fracture pore pressures from ECLIPSE/INTERSECT and these are used as specified
earlier in this section to create two microscopic effective stresses. These effective stresses are calculated with the matrix
and fracture Biots coefficients. These effective stresses are then used in the invariant (matrix) and discontinuity (fracture)
yield criteria as appropriate.
82
The second difference concerns the calculation of updated permeability. Here the matrix and fracture permeabilities
specified in ECLIPSE/INTERSECT are extracted by VISAGE. Then each is updated based on the methods detailed in
Permeability Updating Methods (p.71) section. The updated matrix and fracture permeabilities are then exported for use
in the next ECLIPSE/INTERSECT step. To enable the updating of fracture permeability there must be fractures specified
in the VISAGE model.
83
Equation solvers
During a finite element analysis the solution of the following set of linear equations is required:
K {Δd } = {ΔP } Eq. 10.1

where [K] is the global assembled stiffness matrix and is of size n by n where n is the total number of unknowns in the
problem.
The matrix [K] can contain many zeros and in addition be banded about the diagonal. In general [K] can be asymmetric
but depending on the problem may also be symmetric. Distribution of the zeros and extent of the band are dependent on
the mesh numbering of the problem. Symmetry of the global matrix depends on the material models being used.
Generally two methods are used to solve the set of simultaneous equations, a direct method based on Gaussian
elimination or an iterative technique. Iterative solvers can be run in parallel allowing for greater speed and enhanced
memory distribution.
Linear equation solution techniques

There are two main solution techniques used for the solution of the finite element equations. They are the direct method
and iterative solvers.
Direct solvers
Direct solvers calculate the inverse [K]-1 of [K]. The solution {Δd} can then be solver directly.
For static analyses the direct equation solver both in core and out of core is based on Cholesky elimination and uses a
skyline storage method.
In the skyline solutions the central nonzero diagonal part of the stiffness matrix defining the structure is stored. This is
then solved as a whole or in parts. If it is solved in parts, then an out of core solver is said to be in use, off-loading the
matrix to hard disk. The out of core method is used for larger problems.
It is particularly important for the direct solution method to keep the width of the band as low as possible as the solution
time will be proportional to the square of the bandwidth. The semi-bandwidth will be defined by the maximum number
difference in any element plus one multiplied by the number of degrees of freedom being used.
In some hardware platforms, computer memory limitations usually prevent the solution of large problems and suitable
algorithms have to be developed which will allow the memory bounds to be exceeded by means of reading and writing to
and from disks.
The solution of problems is based on various bandwidth solvers, capable of administrating problem requirements which
exceed such memory bounds. As a general rule it is always good practice to number the structure for optimum
bandwidth and front width. This will enhance the ability to solve problems that demand substantial memory stacks,
although this is not important for the iterative solvers.
Iterative solvers
The iterative solvers in VISAGE are all based on the conjugate gradient (CG) method. Unlike direct solvers they
approach the solution gradually and an error estimate is used to assess when the solution is reached.
Iterative solvers are independent of element and/or nodal ordering sequence. The stiffness matrix must be positive
definite and, for large problems, iterative solvers provide an effective solver from both the CPU and memory
management points of view.
Equation solvers
84
There is a threshold above which the iterative solvers are much faster than the direct solvers. Below this threshold, the
direct solver provides a faster solution as it requires only one inversion of the global stiffness matrix. This threshold value
is dependent on the number of elements, element types and number of dimensions. It also depends on the memory
available on the machine.
However iterative solvers have a weakness in that it can take a large number of iterations to converge when the
underlying physical problem is strongly heterogeneous. This is usually the case in reservoir engineering models.
The slow convergence of the conjugate gradient method can be improved by using preconditioning. Generally this is a
technique for improving the condition number of the matrix. The condition number of the matrix [K] measures how much
the result will change for a small change in the input {Δd}.
−1 −1
K {Δd } = {ΔP } can be solved indirectly by solving M K {Δd } = M {ΔP } where [M] is a symmetric
matrix. The quality of the preconditioner [M] depends on the condition number of [M]-1[K]. The difficulty lies in finding a
preconditioner that will improve the convergence of the conjugate gradient method.
The simplest method is to use the diagonal elements of the matrix [K] as the preconditioner. This is known as diagonal or
Jacobi preconditioning.
An alternative is to use deflation preconditioning, Jonsthovel et. al (2013). Here physical details about the geometry of
the grid are used to construct a superior preconditioner. This method requires more memory but can reduce the number
of CG iterations considerably. In figure 10.1 we present a typical convergence plot for CG with and without deflation
preconditioning.
Figure 10.1. : Convergence of CG and DPCG for single solve
Note: For full field 3D reservoir models created by Petrel Reservoir Geomechanics the iterative solver is used, most
often in parallel. Also the use of the deflation solver is automatically selected.
Equation solvers
85
Guidelines to finite element

modeling
This section gives general details about finite element modeling. Not all aspects may be relevant to the full scale
reservoir geomechanics. In the context of the finite element method, an actual structure or continuum is replaced by an
equivalent idealized structure composed of discrete elements, referred to as finite elements. These are connected
together at a number of specific nodal points. By assuming displacement fields or stress patterns within an element it is
possible to derive a stiffness matrix relating the nodal forces to the nodal displacements within an element. The stiffness
matrix of the assemblage of elements can be formed by considering the nodal contributions in an overall stiffness matrix
of each of the individual elements. If conditions of equilibrium are applied at every node of the idealized structure, a set of
simultaneous algebraic equations can be formed, the solution to which provides all the nodal displacements. These in
turn are used to determine all the internal stresses. The majority of finite element packages assume linear elastic small
strain theory. The VISAGE finite element simulator can offer nonlinear, small strain/displacement capabilities.
Note: Not all the features detailed here are available when creating models with Petrel Reservoir Geomechanics
Idealization
First of all, in any analysis, it is necessary to discretize the continuum, by subdividing the structure into one-, two- or
three-dimensional finite elements by fictitious lines and/or surfaces. Thus, the structure is now represented by an
assemblage of simple geometric shapes rather than having a complex geometric outline. The elements are assumed to
be interconnected at a discrete number of nodal points situated at the element boundaries. The way the structure will be
meshed, will depend on:
• The physical nature of the structure itself, such as material or load boundaries.
• The element types to be used.
• The extent of the area of intersect.
• The type of analysis: static, dynamic, potential, eigenvalue, thermal, or nonlinear.
• The cost of running the analysis and the way the loads are applied.
Subdividing the structure

It is normally trivial to decide how to subdivide the structure into finite elements by the physical specification of the
structure. With frame structures such as oil rigs or buildings there is an obvious correspondence between real and
idealized beams. In more discontinuous structures such as an hydrocarbon reservoirs, the obvious breakpoints are at the
material boundaries.
There are two aspects that should always be looked for initially: symmetry and repetition.
In many structures, symmetry can be used to significantly reduce the size of the problem to be solved. The use of such a
condition reduces not only the number of elements involved, but will normally significantly reduce the bandwidth of the
equations to be solved also. As the cost of the solution is proportional to the square of the bandwidth (for skyline
solvers), this is a very important factor. With symmetry, the symmetric plane is constraint to remain planar. This
technique may also be used with radial symmetry such as in discs. Loading will also need to be symmetric although it is
sometimes possible to use loading combinations to give results for asymmetric loads on symmetric structures, provided
that an elastic analysis is performed.
Guidelines to finite element modeling

86
However, the VISAGE finite element simulator incorporates highly sophisticated equation solvers which assist in
overcoming some of these problems. For more information see Equation solvers (p.84).
Mesh size
Apart from the constraints imposed by the physical nature of the structure, the mesh size will be chosen to provide for
three major factors. The overriding requirement is that the element must be capable of mathematically representing the
physical area it defines to the degree required. Thus, it may be possible to mesh for a large number of fairly simple
elements or a much smaller number of higher order elements. It is for this reason that higher order elements may be
adopted for use within the VISAGE finite element simulator.
The basic mesh may then be refined or degenerated depending on whether it contains an area of interest. With frame
structures, the overall response is normally looked for and no local refinement is called for. In solid body analysis there
will often be a region of main interest, this will need to have an adequately refined mesh. Elsewhere it is only necessary
to ensure that the load will diffuse correctly into the area. Hence mesh refinement may well be necessary in the areas of
restraint or localized load application to ensure load diffusion is correct. In the interface region it will also be necessary to
ensure a gradual change in element size to ensure individual elements can perform correctly.
Element types
The VISAGE finite element simulator uses the displacement method for its basis for finite element analyses. According to
this method, the displacements are chosen as the prime unknowns, with the stresses being determined from the
calculated displacement field. This method is used by most finite element packages. In this technique, the displacement
of the system of nodes is assumed to have unknown values only at the nodal points so that variation within any element
is described in terms of the nodal values by means of interpolation functions, normally referred to as shape functions. It is
important to realize that these functions governing the performance of an element are incorporated between nodes and
hence extreme distortion of any element may give strange results or fail due to a mathematically contorted function. The
strains within any element may be expressed in terms of the element nodal displacements where the strains will be
composed of the derivatives of the shape functions. The element stresses are calculated from the strains in a manner
which ensures the satisfaction of equilibrium and plasticity equations. Provided that the element shape functions have
been chosen so that no singularities exist in the integrands of the functional, the total potential energy of the continuum
will be the sum of the energy contributions of the individual elements. The summation of the contributions, when equated
to zero, results in a system of equilibrium equations for the complete continuum. These equations are then solved by any
standard technique to yield the nodal displacements.
The element formulations will, for the displacement method, ensure displacement continuity across boundaries of
elements. Hence it is essential that all nodes at element boundaries are fully connected to adjacent lines of nodes,
otherwise forces will not be transmitted through the continuum correctly and element edges will be allowed to deform
locally, to pull apart, or overlap.
Element shape and size

A prime requirement in using finite elements is that they are not forced beyond the limits of their governing mathematics.
Hence the use of a higher order element is likely to allow the user much more flexibility and return greater accuracy. It is
probably worthwhile to consider the various elements into two very broad categories of lower and higher order elements.
The lower order elements will probably have linear variation of geometry stress and strain whilst the higher order
elements will have correspondingly higher order variations. Lower order elements will therefore need to be used in
greater numbers to approach the degree of accuracy achieved when using an isoparametric element.
How large an element is and its shape will be governed by attempting to fit a meshing scheme to the structure. The
prime requirement for this will probably be to define a coarse mesh, except at the area of interest, for reasons of
economy. The coarse mesh will therefore need to cope with the load diffusion through the structure. The fine mesh will
need to give enough detail and the necessary stress components output for the engineer to decide how the structure is
performing in that area.
In defining these two areas and the transition region, some elements are likely to be deformed to map the structure
adequately. The simple, lower order elements will not perform well when removed from aspect ratios of 1:1 and simple,
lower order elements should certainly not be used when being pulled more into a spiked shape. Higher order elements

87
can cope with an order of magnitude greater distortion. Thus the eight noded elements can often be given with aspect
ratios of up to 1:7, whilst the equivalent brick element can rise to orders of 1:10 and cause little degradation in
performance.
Whilst working with these general guidelines it must be remembered that between the nodal points defining the element
is the shape function. If these points are positioned so as to cause peculiar edge shapes, they may cause integration
points to move outside the relevant boundary. Such an occurrence causes a singularity of the element and the program
execution will normally cease.
Linear solver solution

There are three main solution techniques used within the finite element method. They are the in-core skyline, the out-of-
core skyline and the iterative solvers. In the skyline solutions the central nonzero diagonal part of the stiffness matrix
defining the structure is stored. This is then solved as a whole or in parts. If it is solved in parts, then an out of core solver
is said to be in use. The out of core method is used for larger problems.
It is particularly important for the banded solution to keep the width of the band as low as possible as the solution time
will be proportional to the square of the bandwidth. The semi-bandwidth will be defined by the maximum number
difference in any element plus one multiplied by the number of degrees of freedom being used.
The iterative solvers are independent of element and/or nodal ordering sequence. The stiffness matrix must be positive
definite and, for large problems, provides an effective solver from both the CPU and memory management points of
view.
There is a threshold above which the iterative solvers are much faster than the skyline solvers. Below this threshold, the
skyline provides a faster solution as it requires only one inversion of the global stiffness matrix. This threshold value is
dependent on the number of elements, element types and number of dimensions. It also depends on the memory
available on the machine.
As a general rule it is always good practice to number the structure for optimum bandwidth and front width. This will
enhance the ability of VISAGE Finite Element Software to solve problems that demand substantial memory stacks,
although this is not important for the iterative solver.
Singularities
The accuracy of the solution will be dependent upon the conditioning of the equations the solver has been asked to
solve. In physical terms this can be regarded as the relative stiffness contributions given by the equations involved. Thus
if one area of the matrix appears to be dividing the structure into two very stiff regions with a very flexible region, the
solution may be highly inaccurate. There would appear to be five main causes commonly associated with this problem:
non–fixed structures, distorted elements, unintentional multi-structures, bad modeling choice of elements and structures
on flexible mountings.
It is a requirement for all these solutions that the structure is not free in space and so restraints must be introduced to
prevent this. It is important however not to overstrain the structure; it should be free to deform in a real physical manner.
If, when using constraint equations, which are forcing freedoms to have their displacements related to others, and
releasing certain freedoms, it is important not to totally release that part of the structure. This may also occur if coincident
nodes or gaps in nodes are used and some elements are not correctly joined.
Distorted elements are a frequent problem, necessitating remodeling. If the elements are used in an unacceptable
fashion it can appear to the solution that a local region of practically zero stiffness occurs hence causing singularity. This
may be caused by the elements physical geometry and the values given for it to build its elasticity matrix such as
Young’s Modulus or Poisson’s Ratio. This could also happen if an element is used in an inappropriate place, for example
a thin element where large stresses needed to be transmitted. Having very large elements next to very small elements
can also cause similar problems.
Occasionally structures need to be modeled that physically do have weak areas supporting stiff structures. Support
conditions for slung bodies often have this problem where the supports used may be varied along the length. Here all
that can be done is take great care in checking the reaction forces and attempting to define the structure so that
numerically the solution progresses from stiff equations to the weaker equations by judicious use of the numbering or
substructuring. This averts sudden jumps in relative stiffness.

88
Loading
All packages provide for a range of loading types. Point loads and distortions are the basis of all loads as applied onto
the final structure. However, more sophisticated patterns of pressure and body forces can be applied using the internal
element shape functions and hence equivalent nodal loads will be calculated to give the correct mathematical distribution
to suit that function. If this was not used, dividing loads on an eight node face equally would tend to cause high distortion
at the corner nodes in comparison with mid-side nodes.
Final notes
In deciding on model selection, appropriate considerations are required to ensure that it can describe the physical
condition with the required accuracy. Make sure that the answers that are obtained in a visual form could be used with
some engineering judgment. Mesh the structure as evenly as possible avoiding any gross distortion of element shape,
but refining to accommodate local high stress and loading as necessary.
Note: Care must be taken in fixing the structure globally, in a manner that will allow it to move as in reality.
Note: Checks must be performed on most input data, so that it conforms to the model to be analysed and ensure that it
conforms to the program input specification.
CAUTION: Inadequate checking is the biggest single cause of error.

89
References and bibliography

References
1. Barenblatt, G.I. and Zheltov, Y.P. , “Fundamentals equations of filtration of homogeneous liquids in fissured rocks”,
Soviet Physics Doklady (1960), 5, 522–525
2. Berryman, J. and Wang F.H., , “The elastic coefficients of double-porosity models for fluid transport in jointed rock”,
Journal of Geophysical Research (1995), 100, No. B12, 24611–24627.
3. Cormeau, I. , “Numerical stability in quasi-static elasto/visco-plasticity”, International Journal for Numerical Methods
in Engineering , (1975) 9, No. 1 109–127
4. Coussy, O. Poromechanics. Chichester, U.K., John Wiley & Sons Ltd, (2004).
5. de Waal, J.A. and Smits, R. M. M., “Prediction of Reservoir Compaction and Surface Subsidence: Field Application
of a New Model”, paper SPE 14214, SPE Formation Evaluation (1998), 3, No. 2 347–356.
6. Elsworth, D and Bai M,. “Flow-Deformation Response of Dual-Porosity media”, ASCE Journal of Geotechnical
Engineering, (1992) 118, 107–124.
7. Jonsthovel T.B., van Gijzen M.B., Vuik C. and Scarpas A., “On the Use of Rigid Body Modes in the Deflated
Preconditioned Conjugate Gradient Method”, SIAM Journal on Scientific Computing (2013), 35, B207-B225.
8. Kachanov, M. and Sevostianov, I., “On quantitative characterization of microstructures and effective properties”,
International Journal of Solids and Structures, (2005), 42, 309–335
9. Kachanov M., Tsukrov I., and Shafiro, B., “Effective moduli of solids with cavities of various shapes”, Applied
Mechanics Review, (1994), 47 (1S), S151 – S174.
10. Kristiansen, T.G. and Plischke B. “History Matched Full Field Geomechanics Model of the Valhall Field Including
Water Weakening and Re-Pressurisation” paper SPE 131505, presented at the SPE EuropeC/EAGE Annual
Conference and Exhibition, Barcelona, Spain, (14–17 June, 2010).
11. Nemat-Nasser, S. and Hori, M,. Micromechanics: Overall properties of heterogeneous materials (2nd ed) North-
Holland, (1999).
12. Pande, G. N.: On an accelerator in elasto–viscoplasticity, Internal Report C/R/301/77, Swansea, U.K., University
College of Swansea, (1977).
13. Papamichos, E, Brignoli, M., and Santarelli, F. J., “An experimental and theoretical study of a partially saturated
collapsible rock”, Mechanics of Cohesive-Frictional Materials, (1997) 2, No. 3, 251-278.
14. Sayers, C. M., “Stress-dependent elastic anisotropy of sandstones,” (2002), Geophysical Prospecting, 50, No. 1, 85–
95.
15. Sayers, C. M., and Kachanov, M., “A simple technique for finding effective elastic constants of cracked solids for
arbitrary crack orientation statistics”, International Journal of Solids and Structure (1991), 27, 671–680.
16. Sayers C. M., and Kachanov M., “Microcrack-induced elastic wave anisotropy of brittle rocks,” Journal of
Geophysical Research, (1995), 100, No. B3, 4149–4156.
17. Smith, I. M. and Griffiths, D. V.: Programming the Finite Element Method, (2nd edition), UK, John Wiley and Sons Ltd,
(1982). 15–17 and 36–40.
18. Stolle, D.F. and Higgins, J.E. “Viscoplasticity and plasticity — numerical stability revisited”, International Conference
on Numerical Methods in Geomechanics, NUMOG III, (1989), 431–438
19. Zhang Y., Sayers C. M., and Adachi J. I., “The use of effective medium theories for seismic wave propagation and
fluid flow in fractured reservoirs under applied stress”, Geophysical Journal International (2009), 177, No. 1, 205–
221.

90
Bibliography
The following bibliography is highly recommended for those who wish to know more about finite element techniques, the
constitutive models used within VISAGE and the classical theories of plasticity and visco-plasticity.
1. Beer, G. and Watson, J. O.: Introduction to Finite and Boundary Element Methods for Engineers, U.K., John Wiley
and Sons Ltd, (1992).
2. Hinton, E. and Owen, D. R. J. : An introduction to finite element computations, Swansea, U.K., Pineridge Press,
(1979).
3. Holman, J. P.: Heat Transfer, (6th edition), McGraw–Hill, (1986).
4. Owen, D. R. J. and Hinton, E.: Finite Elements in Plasticity: Theory and Practice, Swansea, U.K., Pineridge Press,
(1980).
5. Pande, G. N., Beer, G. and Williams, J. R. : Numerical Methods in Rock Mechanics., U.K., Wiley-Blackwell, (1990).
6. Smith, I. M. and Griffiths, D. V. : Programming the Finite Element Method, U.K., John Wiley and Sons Ltd, (1982).
7. Welty James R.: Engineering Heat Transfer SI Version, U.K., John Wiley and Sons Inc, (1978).
Further reading
1. Barton, N. and Choubey, V.: “The Shear Strength of Rock Joints in Theory and Practice”, Rock Mechanics, (1977),
10, No. 1–2, 1–54.
2. Gerrard, C. M. and Pande, G. N. : “Numerical Modelling of Reinforced Jointed Rock Masses I Theory”, Computers
and Geotechnics, (1985), 1, 293–318.
3. Griffiths, D. V. and Koutsabeloulis, N. C. : “Finite Element Analysis of Vertical Excavations”, Computers and
Geomechanics, (1985), 1, 221–235.
4. Heffer, K. J. and Koutsabeloulis, N. C : Fracture Scale Effects in Hydrocarbon Reservoirs, Scale Effects in Rock
Masses 93, Pinto da Cunha (editor), Balkema, Rotterdam, Taylor and Francis Group, (1993)
5. Heffer, K. J., Koutsabeloulis, N. C. and Wong S. K. : Coupled Geomechanical, Thermal and Fluid Flow Modelling as
an Aid to Improving Waterflood Sweep Efficiency, Eurock ’94, Balkema, Rotterdam, (1994).
6. Koutsabeloulis, N. C.: Tunnel Design Using the ’Trapdoor’ Problem., Numerical Methods in Geomechanics
(Innsbruck 1988), Swoboda G. (editor), Rotterdam, The Netherlands, A. A. Balkema, (2004).
7. Koutsabeloulis, N. C.: Numerical Modelling of Soft Reservoir During Fluid Depletion, Geotechnical Engineering of
Hard Soils - Soft Rocks: Proceedings of an International Symposium under the auspices of the ISSMFE, Athens,
Greece, 20-23 September 1993: Vols 1-3. A. Anagnostopoulos (Editor), R. Frank (Editor), N. Kalteziotis (Editor), F.
Schlosser (Editor), Taylor & Francis, (1994).
8. Koutsabeloulis, N. C. and Griffiths, D. V., : “Numerical Modelling of the Trap Door Problem”, Geotechnique, (1989),
39, No. 1, 77–89.
9. Koutsabeloulis, N. C. and Rylance, M. : Acceleration and Deceleration Techniques for ’Viscoplasticity’, Proceedings
of the International Conference on Numerical Methods in Engineering: Theory and Applications, Swansea, U.K. (7–
11 January 1990), Elsevier Applied Science.
10. Koutsabeloulis, N. C. and Griffiths, D. V. : Dynamic Slope Stability Analysis Using the Finite Element Method,
Computer and Physical Modelling in Geomechanical Engineering, Balassubramaniam et al (editors), Rotterdam, The
Netherlands, A. A. Balkema, (1989).
11. Koutsabeloulis, N. C., Heffer, K. J. and Wong S. K.: “ Numerical geomechanics in reservoir engineering ” Computer
Methods and Advances in Geomechanics: Proceedings of the Eighth International Conference, Morgantown, West
Virginia, USA, (May 1994), Siriwardane H. J. and Zaman M.M. (editors) Rotterdam, The Netherlands, A. A. Balkema.
12. Koutsabeloulis, N. C. and Rylance, M. : On the Integration of the Multi–Laminate Model for Soils, Numerical Models
in Geomechanics, Pande and Pietruszczak (editors), Rotterdam, The Netherlands, A. A. Balkema (1992).

91
13. Koutsabeloulis, N. C., Rylance, M. and Eggington, R.: 3D Finite Element Modelling of the Hope Brook Gold Mine,
Computer Methods and Advances in Geomechanics, Beer, Booker and Carter (editors), Rotterdam, The
Netherlands, A. A. Balkema (1991).
14. Naylor, D. J. : “Stresses in Nearly Incompressible Materials by Finite Elements with Application to the Calculation of
Excess Pore Pressures”, International Journal for Numerical Methods in Engineering , (1974), 8, No. 3, 443–460.
15. Pan, X. D. and Hudson, J. A. : A Simplified Three Dimensional Hoek–Brown Yield Criterion., Rock Mechanics and
Power Plants, Romana (editor), Rotterdam, The Netherlands, A. A. Balkema, (1988).
16. Pande, G. N.: “A Constitutive Model of Rock Joints”, Proceedings of the International Symposium on Fundamentals
of Rock Joints ,Bjorkliden, Sweden (15–20 September, 1985).
17. Pande, G. N. , Owen, D. R. J. and Zienkiewicz, O. C.:” Overlay Models in Time–Dependent Non–Linear Materials
Analysis”, Computers and Structures, 77, 435-443, Pergamon Press.

92

Reservoir Geomechanics VISAGE 1726681270

Uploaded by

Copyright:

Available Formats

Reservoir Geomechanics VISAGE 1726681270

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Reservoir Geomechanics VISAGE 1726681270

Uploaded by

Copyright:

Available Formats

VISAGE

Scientific background ...................................................................................................... 4

Modeling elasticity ............................................................................................................ 6

Modeling of plasticity ..................................................................................................... 12

Discontinuity model ........................................................................................................ 19

Material Constitutive Models ......................................................................................... 23

Tresca criteria ......................................................................................................................................... 32

ISAMGEO Chalk Model ................................................................................................... 42

In-situ state ............................................................................................................................................... 63

Coupled porous media ................................................................................................... 67

Permeability updating methods .............................................................................................................. 71

Dual Porosities/Permeabilities ...................................................................................... 79

Equation solvers ............................................................................................................. 84

Guidelines to finite element modeling .......................................................................... 86

References and bibliography ..............................................................................................

Applications to reservoir engineering

Main features of VISAGE

Reservoir engineering applications

Figure 2.1. Stress–Strain Behavior of Soil/Rock

Figure 3.1. Normal and Shear Stresses

and compatibility equations relate the strain to displacements by:

Figure 3.2. Boundary Conditions

Finite element theory

Consider one arbitrary finite element:

δΔE = δΔW − δΔL = 0

ΔL = ∫ {Δd } {ΔF } d Vol + ∫ {Δd } {ΔT } d Srf

where F are body forces and T are boundary traction forces.

{Δd } = N {Δd } i Eq. 3.9

This can be written as

Figure 3.3. Gauss Point Local Axes

Figure 4.1. Basic one-dimensional elastic–visco–plastic model

{ σ o } = σ xo , σ yo , σ zo , σ xyo , σ yzo , σ zxo T Eq. 4.1

{ ϵ o } = ϵ xo , ϵ yo , ϵ zo , ϵ xyo , ϵ yzo , ϵ zxo T

{Δϵ } = B {Δδ } Eq. 4.4

{Δσ } = D {Δε } Eq. 4.5

{σ } = { σ 0} + {Δσ } Eq. 4.6

and from those total stresses, evaluate the stress invariants

{Δε } = {Δε } e + {Δε } v p Eq. 4.8

ε̇ vp = γ ⋅ f (F ) ⋅ { ∂∂Qσ } Eq. 4.9

here for real visco–plastic materials:

γ is the fluidity parameter of the dashpot (taken as unity)

f (F ) = 0 if F < 0 Eq. 4.10

{Δε } vp = Δt ⋅ ε̇ vp Eq. 4.13

The excess stresses are then calculated by:

{Δσ } p = D ⋅ {Δε } vp Eq. 4.14

{Δε } t = B ⋅ {Δδ } t Eq. 4.16

{Δσ } t = D ⋅ ({Δε } t − {Δε } tvp ) Eq. 4.17

{σ } = { σ o } + {Δσ } Eq. 4.18

ε̇ tvp = γ ⋅ f (F ) ⋅ { ∂∂Qσ } Eq. 4.19

7. Update the visco–platic strain increment as follows:

{Δε } tvp+Δt = {Δε } vp

{ΔP } t +Δt = {ΔP } t + Σ ∫

Figure 4.2. Application of the visco–plastic algorithm

max (Δdt +Δt − Δ d t )

Δd t +Δt is the displacement from the current iteration

Δd t is the displacement from the previous iteration

F < tol (not dimensionless)

For each of these convergence criteria tol is a tolerance specified a priori.

Visco-plastic time step selection