This action might not be possible to undo. Are you sure you want to continue?
A.F. Fossum, R.M. Brannon
Prepared by
Sandia National Laboratories
Albuquerque, New Mexico 87185 and Livermore, California 94550
for the United States Department of Energy
under Contract DEAC0494AL85000
Approved for public release; distribution is unlimited.
SANDIA REPORT
SAND20043226 • UC405
Unlimited Release
Printed August 2004
Issued by Sandia National Laboratories, operated for the United States Department of
Energy by Sandia Corporation.
NOTICE: This report was prepared as an account of work sponsored by an agency of
the United States Government. Neither the United States Government nor any agency
thereof, nor any of their employees, nor any of their contractors, subcontractors, or their
employees, makes any warranty, express or implied, or assumes any legal liability or
responsibility for the accuracy, completeness, or usefulness of any information appara
tus, product, or process disclosed, or represents that its use would not infringe privately
owned rights. Reference herein to any specific commercial product, process, or service
by the trade name, trademark, manufacturer, or otherwise, does not necessarily consti
tute or imply its endorsement, recommendation, or favoring by the United States Gov
ernment, any agency thereof or any of their contractors or subcontractors. The views and
opinions expressed herein do not necessarily state or reflect those of the United States
Government, any agency thereof or any of their contractors.
Printed in the United States of America This report has been reproduced directly from
the best available copy.
Available to DOE and DOE contractors from
Office of Scientific and Technical Information
PO Box 62
Oak Ridge, TN 37831
Prices available from (615) 5768401, FTS 6268401
Available to the public from
National Technical Information Service
US Department of Commerce
5285 Port Royal Rd
Springfield, VA 22161
NTIS price codes
Printed copy: A08
Microfiche copy: A01
SAND20043226
Unlimited Release
Printed August 2004
THE SANDIA GEOMODEL
Theory and User’s Guide
Arlo F. Fossum and Rebecca M. Brannon
Geomechanics, Dept. 6117
Sandia National Laboratories
PO BOX 5800
Albuquerque, NM 87185
Abstract
The mathematical and physical foundations and domain of applicability of
Sandia’s GeoModel are presented along with descriptions of the source code
and user instructions. The model is designed to be used in conventional finite
element architectures, and (to date) it has been installed in five host codes
without requiring customizing the model subroutines for any of these different
installations. Although developed for application to geological materials, the
GeoModel actually applies to a much broader class of materials, including
rocklike engineered materials (such as concretes and ceramics) and even to
metals when simplified parameters are used. Nonlinear elasticity is supported
through an empirically fitted function that has been found to be wellsuited to a
wide variety of materials. Fundamentally, the GeoModel is a generalized plas
ticity model. As such, it includes a yield surface, but the term “yield” is gener
alized to include any form of inelastic material response including microcrack
growth and pore collapse. The geomodel supports deformationinduced anisot
ropy in a limited capacity through kinematic hardening (in which the initially
isotropic yield surface is permitted to translate in deviatoric stress space to
model Bauschinger effects). Aside from kinematic hardening, however, the
governing equations are otherwise isotropic. The GeoModel is a genuine unifi
cation and generalization of simpler models. The GeoModel can employ up to
40 material input and control parameters in the rare case when all features are
used. Simpler idealizations (such as linear elasticity, or Von Mises yield, or
MohrCoulomb failure) can be replicated by simply using fewer parameters.
For highstrainrate applications, the GeoModel supports rate dependence
through an overstress model.
iv
Intentionally Left Blank
v
Acknowledgements
The authors are grateful for the support received over the years from a number of San
dia project managers and colleagues who have provided financial support and encouraged
the development and enhancement of this general geomaterial model for use in Sandia
computer codes. The initial development of the model was begun under a project led by
Mike Stone under the U.S. Department of Energy (DOE) Advanced Computational Tech
nology Initiative (ACTI) program. Additional development was made under an Engineer
ing Sciences Research Foundation (ESRF) project led by Hal Morgan; a Defense
Programs (DP) Penetration project led first by Jim Hickerson and later by Danny Frew;
several DOE Accelerated Strategic Computing Initiative (ASCI) Design and Qualification
(DQ) Materials & Physics Models (M&PM) projects led by Mike McGlaun and Justine
Johannes, a Hard and Deeply Buried Target (HDBT) project led by Paul Yarrington and
Shawn Burns; a Model Accreditation Via Experimental Sciences For Nuclear Weapons
(MAVEN) laboratory testing project led by Moo Lee, two LDRD projects, one led by
Larry Costin and the other by Rich Regueiro; and an Army Research Laboratory (ARL)
project led by Bill Bruchey.
The authors acknowledge individual contributions of Steve Akers, Lupe Arguello,
Paul Booker, Kevin Brown, Ted Carney, Mike Chiesa, Jim Cox, Kent Danielson, Mike
Forrestal, Craig Foster, Joanne Fredrich, Lew Glenn, Ed Hoffman, Dave Holcomb, John
Holland, Joe Jung, Daniel Kletzli, Colby Lavin, Estaban Marin, Yvonne Murray, Bill
Oberkampf, Bill Olsson, John Pott, Tom Pucik, Jonathan Rath, Sal Rodriguez, Bill
Scherzinger, Paul Senseny, Allen Shirley, Len Schwer, Steve Sobolik, Bob Swift, Tom
Voth, Tom Warren, Joel Wirth, and Mike Wong, who have provided code support, user
feedback, bug fixes, theoretical discussions, and constructive criticism.
The first author, AFF, would like to express his sincere appreciation to Rebecca Bran
non for painstakingly going through each line of coding in the material model subroutine
checking theoretical soundness, debugging, documenting, and suggesting improvements.
The quality of the model has vastly improved as a result. The second author, RMB, appre
ciates the numerous program managers and other supporters who have given Arlo Fossum
sufficient time and resources to assemble a relatively comprehensive set of components
that ideally should be discussed in any material model resource document (theory, code,
verification, validation, calibration instructions, and datasets for various materials). RMB
admires Arlo Fossum for his mentoring, enthusiasm, unfailing sense of humor, and his
remarkable ability to discern the salient physics evident in disparate geomechanical data.
Our compatible skills and temperaments have established the foundation for an effective
and enduring materials modeling team. Finally, both authors are grateful to our family
members for their patience and support during the many latenight hours required to reach
this year’s model development and dissemination goals.
Arlo Fossum Rebecca Brannon Sandia National Laboratories
affossu@sandia.gov rmbrann@sandia.gov August 30, 2004.
vi
Contents
Acknowledgements .................................................................................................. v
Introduction.............................................................................................................. 1
GeoModel overview........................................................................................................ 2
Features of the GeoModel ..................................................................................................... 2
Limitations of the GeoModel ................................................................................................. 3
Notation..................................................................................................................... 7
Typesetting conventions.................................................................................................. 7
Vector and Tensor notation............................................................................................. 7
The Stress Tensor..................................................................................................... 9
Stress invariants .............................................................................................................. 11
Derivatives of the stress invariants ................................................................................. 12
Special stress states ......................................................................................................... 13
HYDROSTATIC (HYD) ......................................................................................................... 13
TRIAXIAL (TXC and TXE): ................................................................................................... 14
SIMPLE/PURE SHEAR and PRESSURESHEAR LOADING (SHR) ................................... 18
Spatial and Reference stress (frame indifference) .......................................................... 19
Lode coordinates ............................................................................................................. 20
Octahedral yield profile visualization ................................................................................... 22
Meridional yield profile visualization ................................................................................... 23
Closedform solution for ordered eigenvalues ...................................................................... 24
GeoModel theory...................................................................................................... 25
Elasticity.......................................................................................................................... 26
Nonlinear elasticity................................................................................................................ 27
Assigning values to the elastic constants............................................................................... 27
The elastic limit (yield surface)....................................................................................... 28
The cap function, f
c
................................................................................................................ 36
Elasticplastic coupling ......................................................................................................... 37
Meridional shear limiter function, F
f
.................................................................................... 38
The complete GeoModel yield function................................................................................. 42
The “J3TYPE” Lodeangle function, Γ................................................................................. 43
Advancing the solution (groundwork discussion)........................................................... 47
Evolution equations......................................................................................................... 50
Evolution equation for the porosityrelated internal state variable, κ.................................. 50
Evolution equation for the kinematic hardening backstress tensor....................................... 58
Advancing the solution (final step, consistency parameter) ........................................... 60
Formal equivalence with oblique return algorithms............................................................. 60
Quasistatic inelastic tangent stiffness tensor......................................................................... 61
Stability issues ....................................................................................................................... 61
Rate Dependence...................................................................................................... 63
Viscoplasticity model overview...................................................................................... 63
Viscoplasticity model derivation..................................................................................... 66
Limiting case.......................................................................................................................... 68
Assigning a value to the characteristic material time...................................................... 70
Thermodynamics considerations..................................................................................... 71
GeoModel Numerical Solution scheme .................................................................. 72
Role of the GeoModel within a finiteelement program................................................. 72
Spatial stability (mesh dependence/loss of strong ellipticity)......................................... 74
GeoModel files, subroutines, memory requirements, and model installation requirements. 75
vii
Arguments passed to and from the GeoModel driver routine......................................... 76
PLOTABLE OUTPUT ................................................................................................... 77
GeoModel algorithm....................................................................................................... 78
Rate independent (inviscid) part of the viscoplasticity equations......................................... 78
Viscous part of the viscoplasticity equations ........................................................................ 78
Software “confidence building” activities.............................................................. 79
Preliminary convergence testing........................................................................................... 86
Verification: singleelement problems (regression suite) ............................................... 87
Other singleelement tests ..................................................................................................... 92
Verification: Hendren & Ayier pressurized cylinder...................................................... 93
Elastic freefield wave form (finiteelement verification) .............................................. 95
Parameterization (calibration) ............................................................................... 96
Nonlinear elasticity ......................................................................................................... 97
Elasticplastic coupling................................................................................................... 98
Triaxial Compression...................................................................................................... 99
Parameterization: Rate dependence ................................................................................ 102
Building confidence in the physical theory............................................................ 104
Postcalibration Triaxial loading..................................................................................... 105
Fieldscale penetration.................................................................................................... 106
Free field wave form for spherical shock loading........................................................... 107
Closing Remarks ...................................................................................................... 108
APPENDIX A: Parameterizing the GeoModel ..................................................... A1
APPENDIX B: Nomenclature and Data Sets ........................................................ B1
Model Parameters
(User Input) .............................................................................................................. B2
Internal State Variables
(Plotable Output) ...................................................................................................... B8
Dataset for Salem Limestone .......................................................................................... B14
Dataset for Sidewinder Tuff............................................................................................ B15
Dataset for lab scale intact Climax Granite .................................................................... B16
Dataset for field scale jointed Climax Granite................................................................ B17
Dataset for 23MPa Concrete........................................................................................... B18
Dataset for Conventional Strength Portland Concrete.................................................... B19
Datasets for mimicking classical (simplified) models .................................................... B20
VonMises material ............................................................................................................... B21
Maximum Principal Stress failure......................................................................................... B21
Tresca.................................................................................................................................... B22
MohrCoulomb...................................................................................................................... B22
Subinput set for linear elasticity .......................................................................................... B26
Subinput set for “turning off” all rate dependence ............................................................. B26
Subinput set for disabling kinematic hardening .................................................................. B26
Subinput set for associativity ............................................................................................... B26
Subinput set for disabling cap and crush curve................................................................... B26
Subinput set for disabling tensile cutoff limits.................................................................... B26
REFERENCES
viii
Figures
Figure 1.0. GeoModel continuous yield surface. ........................................................... 1
Figure 1.1. Distinction between a yield surface and the limit surface. .......................... 5
Figure 3.1. Typical hydrostatic (pressure vs. volumetric strain) compression data. ..... 13
Figure 3.2. Triaxial compression (TXC) and triaxial extension (TXE)......................... 15
Figure 3.3. Triaxial and CSD load paths........................................................................ 16
Figure 3.4. An octahedral yield profile. ......................................................................... 22
Figure 3.5. Distortion of the meridional profile when using nonisomorphic stress measures.
23
Figure 4.1. Backstress and shifted stress. ...................................................................... 31
Figure 4.2. Qualitatively meridional profile shapes resulting from (a) porosity alone, (b)
microcracks alone, and (c) porosity and microcracks in combination. ....... 33
Figure 4.3. Distinction between twosurface upperbound models and the GeoModel. 33
Figure 4.4. Basic shapes of the three key functions that characterize the composite shape of the
yield surface................................................................................................. 35
Figure 4.5. GeoModel cap function ............................................................................... 36
Figure 4.6. Kinematic hardening. .................................................................................. 40
Figure 4.7. Yield surface evolution with both microcracking and porosity. ................. 40
Figure 4.8. Shear limiter function (unshifted and shifted)............................................. 41
Figure 4.9. Octahedral yield profiles, plotted at allowable values of the strength ratio. 45
Figure 4.10. Lode angle function (for various ψ strength ratios) plotted vs. the Lode angle
varying from –30 degrees (TXE) to +30 degrees (TXC). ........................... 46
Figure 4.11. Hardening mechanisms. .............................................................................. 50
Figure 4.12. A conventional porosity crush curve (dashed) and state path (solid).......... 52
Figure 4.13. The Gurson theory for porous yield surfaces compared with the GeoModel cap
function at various values of the internal state variable κ. Qualitatively, the theories
are similar. ................................................................................................... 53
Figure 4.14. Relationship between (a) hydrostatic pressure vs. volumetric strain data, (b) the
GeoModel Xfunction, and (c) a traditional porosity vs. pressure crush curve. 54
Figure 4.15. Continuously differentiable GeoModel yield function and some characteristic
dimensions. .................................................................................................. 56
Figure 4.16. Meridional plane in which the magnitude of the stress deviator “r” is plotted against
(which is proportional to the pressure “p”) ................................................. 58
Figure 4.17. The effect of kinematic hardening on the meridional plane that contains the
backstress..................................................................................................... 59
Figure 5.1. Rate dependence. ......................................................................................... 64
Figure 5.2. Highrate weighting factor at various initial states...................................... 68
Figure 5.3. Rate dependence in uniaxial strain loading of a nonhardening pressureinsensitive
material. ....................................................................................................... 69
Figure 6.1. Meaning of the “SHEAR” output variable. ................................................. 77
Figure 7.1. Screen shot of the MS Excel interface for the GeoModel materialmodel driver
(MMD). ....................................................................................................... 83
Figure 7.2. Subcycling test for hydrostatic loading. ...................................................... 86
Figure 7.3. Rate of convergence. ................................................................................... 86
Figure 7.4. Unconfined compression. ............................................................................ 88
Figure 7.5. 20 MPa triaxial compression. ...................................................................... 89
Figure 7.6. Reduced triaxial extension. ......................................................................... 90
Figure 7.7. Plane Strain.................................................................................................. 91
ix
Figure 7.8a. Crush curve and hydrostatic loading verification........................................ 92
Figure 7.8b. Rate dependence TXE/TXC ratio verification for triaxial load/unload. ..... 92
Figure 7.9. Hendron & Ayier verification problem. ...................................................... 94
Figure 7.10. Spherical cavity geometry. .......................................................................... 95
Figure 7.11. Finiteelement vs. analytical/numerical elastic wave velocity at 470 meters from a
velocity spherical cavity source at 204 meters. ........................................... 95
Figure 8.1. Meridional limit curves for some materials already parameterized to the
GeoModel. ................................................................................................... 96
Figure 8.2. Nonlinear elasticity in shear. ....................................................................... 97
Figure 8.3. Nonlinear elasticity in hydrostatic loading.................................................. 97
Figure 8.4. Elasticplastic coupling: deformationinduced changes in elastic moduli (Salem
Limestone)................................................................................................... 98
Figure 8.5. Shear failure limit curve compared with concrete data. .............................. 99
Figure 8.6. Progression of the hardening yield surface (family of blue lines) under a triaxial
compression test (red path), illustrated with correspondence of the meridional
plane to the stressstrain diagram. ............................................................... 100
Figure 8.7. Shearenhanced dilatation under compression. ........................................... 101
Figure 8.8. Kolsky (split Hopkinson) bar apparatus used to obtain data in Fig. 8.9 ..... 102
Figure 8.9. Suitability of the GeoModel overstress ratedependence model to match observed
data. ............................................................................................................. 103
Figure 9.1. (a) GeoModel prediction of uniaxial strain loading. (b) GeoModel predictions of
triaxial stressstrain response at various confining pressures...................... 105
Figure 9.2. After the GeoModel was fitted to laboratoryscale material property tests, it was
used to predict projectile penetration depth using spherical cavity expansion
analysis. ....................................................................................................... 106
Figure 9.3. Spherical cavity geometry. .......................................................................... 107
Figure 9.4. GeoModel + JAS3D finiteelement prediction (blue) vs. measured (red) velocity
and displacement at 470 m. from a spherical cavity velocity source at 204 meters.
107
Figure A.1. Ideal hydrostatic parameterization data. .............................. ............................. A2
Figure A.2. Lesspreferable hydrostatic parameterization data. ............. ............................. A2
Figure A.3. Data representative of a triaxial compression parameterization test. ................ A3
Figure A.4. A family of TXC tests conducted to failure. ....................... ............................. A5
Figure A.5. TXC stressstrain plots and extraction of their peak values to construct the
meridional limit curve (data are for porcelanite). ................ ............................. A6
Figure A.6. The Lode function information originally shown in Fig. 4.10 of the main report,
now displayed with a transformed ordinate. ....................... ........................... A10
Figure A.7. A sketch of how data might be distributed at nontriaxial states. ................... A10
Figure B.1. Other yield surface shapes supported by the GeoModel. .... ........................... B20
Figure B.2. Classical simplified yield surfaces in the octahedral plane. ........................... B20
Figure B.3. A stress at the limit state under MohrCoulomb theory. ..... ........................... B22
x
Intentionally Left Blank
Introduction
1
THE SANDIA GEOMODEL
Theory and User’s Guide
1. Introduction
Simulating deformation and failure of natural geological materials (such as limestone,
granite, and frozen soil) as well as rocklike engineered materials (such as concrete [46]
and ceramics [2]) is at the core of a broad range of applications, including exploration and
production activities for the petroleum industry, structural integrity assessment for civil
engineering problems, and penetration resistance and debris field predictions for the
defense community. For these materials, the common feature is the presence of microscale
flaws such as porosity (which permits inelasticity even in purely hydrostatic loading) and
networks of microcracks (leading to low strength in the absence of confining pressure,
nonnegligible nonlinear elasticity, ratesensitivity, and differences in material behavior
under triaxial extension compared to triaxial compression).
For computational tractability, and to allow relatively straightforward model parame
terization using standard laboratory tests, the Sandia GeoModel [15] strikes a balance
between firstprincipals micromechanics and phenomenological, homogenized, and semi
empirical modeling strategies. The overarching goal is to provide a unified generalpur
pose constitutive model that can be used for any geological or rocklike material, that is
predictive over a wide range of porosities and strain rates. Being a unified theory, the Geo
(a)
z
r
(b)
T
X
C
T
X
E
T
X
E
T
X
C
T
X
E
T
X
C
(c)
Figure 1.0. GeoModel continuous yield surface. (a) threedimensional view in principal stress space with the
high pressure “cap” shown as a wire frame, (b) the meridional “side” view (thick line) with the cap shown on the
more compressive righthand side of the plot using cylindrical coordinates in which points along the compressive
[111] direction, and (c) the octahedral view, which corresponds to looking down the hydrostat (onto planes perpendic
ular to the [111] direction).
z
Introduction
2
Model can simultaneously model multiple failure mechanisms, or (by using only a small
subset of the available parameters) it can duplicate simpler idealized yield models such as
classic Von Mises plasticity and MohrCoulomb failure. Thus, exercising this model can
require as many as 40 parameters for extremely complicated materials to only 2 or 3
parameters for idealized simplistic materials. The model parameters are defined in the
nomenclature table (Appendix B). Appendix A gives stepbystep instructions for using
experimental data to assign values to the GeoModel parameters.
GeoModel overview
The GeoModel shares some features with earlier work by Schwer and Murry [42] in
that a Pelessone function [34] permits dilatation and compaction strains to occur simulta
neously. For stress paths that result in brittle deformation, failure is associated ultimately
with the attainment of a peak stress and worksoftening deformation. Tensile or extensile
microcrack growth dominates the micromechanical processes that result in macroscopi
cally dilatant (volume increasing) strains even when all principal stresses are compressive.
At low pressure, porous brittle materials can fail by shear localization and exhibit strain
softening behavior. At higher pressures, they can undergo strainhardening deformation
associated with macroscopically compactive volumetric strain (i.e. void collapse). Fea
tures and limitations of the GeoModel are summarized below.
Features of the GeoModel
Depending on how the model parameters are set, the GeoModel is capable of any of
the following model features
• Linear and nonlinear, associative or nonassociative DruckerPrager plasticity.
• Linear and nonlinear, associative or nonassociative MohrCoulomb plasticity.
• Linear or nonlinear, associative or nonassociative WillamWarnke plasticity.
• Von Mises perfect plasticity.
• Tresca perfect plasticity.
• SandlerRubin twosurface cap plasticity (approximated).
• Threeinvariant, mixed hardening, continuous surface cap plasticity.
• Linear or nonlinear shear failure with or without kinematic hardening.
• Nonlinear compaction function (pressurevolume) with isotropic hardening.
• Three LodeAngle functions (i.e., noncircular or circular octahedral yield profile).
• Linear or nonlinear elasticity.
• Rateindependent or strainratesensitive yield surface.
• Flexibility that permits reducing the model (and the number of required parameters).
to other more classical failure models.
• Pressure and shear dependent compaction (similar to pα models hydrostatic loading,
but generalized to include shear effects in general loading).
• Ubiquitous jointing (i.e., support for a network of many randomly oriented faults).
Introduction
3
Limitations of the GeoModel.
• The GeoModel version that is described in this report treats the material as initially
isotropic. Kinematic hardening is the only mechanism for deformationinduced
anisotropy. Enhanced versions of the GeoModel that support arbitrary anisotropic
jointing are available but not documented here.
• The elasticity model is hypoelastic rather than hyperelastic.
• While the hydrostatic crush curve is quite general, only an elliptic cap function is
available for modeling shear effects on pore collapse and other mechanisms of plastic
volume reduction. Alternative cap models (such as the Gurson function) can be
incorporated in future revisions if needed.
• The host code is responsible for satisfying frame indifference (by calling the
GeoModel using conjugate reference stress and strain rate measures).
• The GeoModel describes material response up to the onset of softening. The host
code is responsible for handling material postpeak softening and the accompanying
change in type of the governing equations to ensure meshsize independence.
• The GeoModel is not extensively parameterized (or even tested) for tensile loading,
though it is thoroughly validated in compression.
• Compared with simple idealized models (which are well known to give unsatisfactory
results in nontrivial structural applications), the GeoModel is computationally
intensive, though less so than many other models of comparable complexity.
• The GeoModel is limited to relatively small distortional (shape changing) strains,
though large volume changes are permitted. Arbitrarily large rotations are permitted
if the host code manages the reference stress and strain measures properly (see
page 76).
• The triaxial extension/compression strength ratio is presumed constant. It does not
vary with pressure, nor does it evolve in time.
The GeoModel predicts observed material response, without explicitly addressing how
the material behaves as it does. The GeoModel reflects subscale inelastic phenomena en
ensemble by phenomenologically matching observed data to interpolation functions. Con
siderations guiding the structure of the GeoModel’s material response functions are (1)
consistency with microscale theory, (2) computational tractability, (3) suitability to cap
ture trends in characterization data, and (4) physicsbased judgements about how a mate
rial should behave in application domains where controlled experimental data cannot be
obtained.
Fundamentally, the GeoModel is a generalized and unified plasticity model. Here, the
term “plasticity” is defined very broadly to include any mechanism of inelastic deforma
tion. Primarily, the source of inelastic deformation in geological materials (or in rocklike
materials such as concrete and ceramics) is growth and coalescence of microcracks and
pores. Under massive confining pressures, inelasticity could include plasticity in its tradi
tional dislocation sense or, more generally, might result from other microphysical mecha
nisms (internal locking, phase transformation, twinning, etc.).
Introduction
4
The GeoModel is phenomenological and semiempirical because the physical mecha
nisms of inelastic material behavior are handled in an ensemble manner, without explicitly
partitioning and modeling each possible contributor to the inelasticity. The GeoModel
makes no explicit reference to microscale properties such as porosity, grain size, or crack
density. Instead, the overall combined effects of the microstructure are modeled by casting
the macroscale theory in terms of macroscale variables that are realistic to measure in the
laboratory. For example, inelastic compaction followed by shearenhanced dilatation has
long been attributed to an initial phase of void collapse followed later by microcracks
opening in shear. The GeoModel is exceptionally capable of matching this type of
observed compaction/dilatation data, but it does so without demanding that the user sup
ply information about essentially unknowable porosity or microcrack distributions within
the material.
Being a generalized plasticity model, the GeoModel presumes that there exists a con
vex contiguous “elastic domain” of stress states for which the material response can be
construed to be elastic. The boundary of the elastic domain is called the yield surface.
When loading is severe enough that continuing to apply elasticity theory would produce a
stress state lying outside the yield surface, the material response will instead be inelastic
and a different set of equations must then be solved. Aside from supporting kinematic
hardening, the GeoModel is isotropic, which means that the criterion for the onset of plas
ticity depends only on the three principal values of the stress tensor, , but not
on the principal directions. Consequently, as illustrated in Fig. 1.0 (page 1), the yield sur
face may be visualized as a 2D surface embedded in a 3D space where the axes are the
principal stresses. The elastic domain is the interior of this surface. The hydrostat is the
[111] direction, along which all three principal stresses are equal. Any plane that contains
the hydrostat is called a meridional plane. Any “side view” crosssection of the yield
surface on a meridional plane is called a meridional profile. Any plane perpendicular to
the hydrostat is called an octahedral plane, and any crosssection of the yield surface
on an octahedral plane is called an octahedral profile.
Mathematically, the yield surface may be expressed in terms of a yield function
. When hardening is permitted, the yield function additionally depends on
internal state variables that quantify the underlying microstructure (e.g. porosity).
Points on the yield surface satisfy and therefore, because the equation
is phrased in terms of three independent variables, the yield surface
may be visualized in the 3D Cartesian space, called stress space*. When the yield func
tion additionally depends on internal state variables, different values for the internal state
variables result in different yield surfaces in stress space. Points within the elastic domain
satisfy . Brittle materials are very weak in tension, but they can deform elastically
under a much broader range of stress states in compression. Consequently, the elastic
domain (and therefore its boundary, the yield surface) resides primarily in the compressive
* Some people prefer that this be called HaighWestergaard space [31] so that the phrase “stress
space” may be reserved for the higherdimensional space defined by the set of all tensors that com
mute with the stress tensor.
σ
1
σ
2
σ
3
, , ( )
f σ
1
σ
2
σ
3
, , ( )
f 0 =
f σ
1
σ
2
σ
3
, , ( ) 0 =
f 0 <
Introduction
5
part of stress space where all three principal stresses are negative. Thus, the typical rock
yield surface shown in Fig. 1.0(a) is actually being viewed from the compressive [111]
direction. The “cap” part of that yield surface (shown as a wire frame in Fig. 1.0) reflects
the fact that, unlike solid metals, inelasticity can occur in rocks even under purely hydro
static compression as a consequence of void collapse.
A yield surface is the boundary of elastically obtainable stress states, whereas a
limit surface is the boundary of stresses that are quasistatically obtainable by any quasi
static means, elastic or plastic. Points outside the limit surface can be reached only tran
siently in dynamic loading via viscoplastic rate dependence. Points outside a yield surface
might be attainable through a hardening process, but points outside the limit surface are
not attainable via any quasistatic process. Points on the limit surface define the onset of
material softening. Consequently, a state on the limit surface is attainable at least once, but
might not be attainable thereafter. The GeoModel simulates material response only up to
the limit state. The GeoModel does not simulate subsequent softening, if any, because
softening usually induces a change in type of the partial differential equations for momen
tum balance, which therefore requires a response from the host code to alter its solution
algorithm (perhaps by inserting void or by activating special elements that accommodate
displacement discontinuities).
Since the GeoModel does not include postsoftening stress response, the limit surface
may be regarded as fixed — it does not evolve (i.e., move around in stress space) as a
yield surface can. Since the limit surface contains all attainable stress states, it follows that
the set of all possible yield surfaces is contained within the limit surface (see Fig. 1.1).
Plasticity induces microstructural changes that permit the yield surface to evolve through
initial yield surface
isotropically hardened
yield surface
kinematically hardened
yield surface
both isotropically
hardened yield surface
l
i
m
i
t
s
u
r
f
a
c
e
Figure 1.1. Distinction between a yield surface and the limit surface. This sketch shows meridional
profiles of an initial yield surface along with hardened yield surfaces that might evolve from the initial sur
face. All achievable stress states (and therefore all possible yield surfaces) are contained within the limit
surface. Fig. 1.0(b), for example, depicts a family of yield surfaces, all bounded by the limit surface.
r
z
and kinematically
Introduction
6
time, effectively changing the initial material into a mechanically different material. A
material can have an infinite number of yield surfaces generated via various pathdepen
dent hardening processes, but it can have only one limit surface. Limit surface character
ization is accomplished by performing numerous experiments all the way to the point of
material rupture (catastrophic failure). Each such experiment can have only one peak
stress state. Postpeak softening in a material might lead to a stress at rupture that is
smaller than at the peak, but it is the collection of peak — not rupture — stress states that
defines the limit surface. Of course, mapping out the limit surface for a given material
requires using a new sample for every experiment, which itself introduces uncertainty
regarding variability in material composition and microstructure. Presently, the GeoModel
treats the limit surface (and each yield surface) as a sharp threshold boundary. Work is
underway to allow these boundaries in stress space to be “fuzzy” to better account for nat
ural material variability.
The set of all possible yield surfaces is contained within the limit surface. Porous
materials are capable of inelastic deformation even under purely hydrostatic loading. Con
sequently, porous materials tend to have closed convex yield surfaces. Once all pores are
crushed out, however, a material can withstand an unlimited amount of pressure. Thus, as
indicated in Fig. 1.1, the limit surface for any material will always be an open convex set.
Despite being developed primarily for geological applications, the GeoModel is truly a
unification of many classical plasticity models. For example, by using only a small subset
of available parameters, the GeoModel can be instructed to behave precisely like a classi
cal hardening or nonhardening Von Mises model, in which case the yield surface
becomes a cylinder centered about the [111] direction. Other classical models such as
DruckerPrager plasticity, Tresca theory, maximum principal stress theory, and MohrCou
lomb theory are also supported in the GeoModel by using the simplified input sets sum
marized in Appendix B. Replicating analytical results from simplified theories is an
important aspect of verification of the GeoModel. However, full use of nearly all Geo
Model features is often required to adequately validate the model for realistic rocklike
materials.
To describe in greater detail how the GeoModel supports its broad range of micro
mechanisms of failure in a mathematical and computational framework, Chapters 2 and 3
first summarize our notation and outline some important concepts and conventions about
the nature of stress. Chapter 4 describes the GeoModel theory (elasticity, yield surface def
inition and evolution, etc.). The computational algorithm, subroutines, and plotable output
will be discussed in Chapter 6, followed by software quality assurance in Chapter 7. Chap
ters 8 and 9 summarize verification and validation tests that have been completed to date
for a variety of materials. Model parameters (as well as descriptions of internal state vari
ables and other symbols used in this report) are defined in Appendix B, along with sample
input sets for realistic and idealized materials. Instructions for determining appropriate
model parameters from laboratory data are provided in Appendix A.
Notation
7
2. Notation
Typesetting conventions
Throughout this report, blue variables are user input parameters and green vari
ables are internal state variables available for plotting in the numerical implementation.
At the discretion of the host code in which the GeoModel is run, several other field vari
ables (e.g., stress) may be additionally available for plotting.
Vector and tensor equations will be presented using indicial Cartesian notation in
which repeated indices within a term are understood to be summed from 1 to 3 while non
repeated indices are free and take values 1 through 3. Upon occasion, vectors and tensors
will be written in symbolic or “direct” notation in which the number of “tildes” beneath a
symbol equals the tensorial order of that variable. For example, , , and would denote
a scalar, a vector, and a tensor, respectively.
Vector and Tensor notation
For this report, the following standard operations and definitions from vector and ten
sor analysis will be employed:
Dot product between two vectors: . (2.1)
Dot product between a tensor and a vector: means . (2.2)
Dot product between a tensor and a tensor: means . (2.3)
Kronecker delta: . (2.4)
Identity tensor: is the tensor whose components are and whose
component matrix is therefore the identity matrix. (2.5)
Inner product between two tensors: . (2.6)
Magnitude of a vector: . (2.7)
Magnitude of a tensor: . (2.8)
s v
˜
T
˜
˜
u
˜
v
˜
• u
1
v
1
u
2
v
2
u
3
v
3
+ + u
k
v
k
= =
y
˜
A
˜
˜
x
˜
• = y
i
A
ik
x
k
=
C
˜
˜
A
˜
˜
B
˜
˜
• = C
ij
A
ik
B
kj
=
δ
ij
1 if i=j
0 if i j ≠
¹
´
¦
=
I
˜
˜
ij δ
ij
3 3 ×
A
˜
˜
:B
˜
˜
A
ij
B
ij
=
v
˜
v
k
v
k
v
k
2
k 1 =
3
∑
= =
A
˜
˜
A
ij
A
ij
A
ij
2
j 1 =
3
∑
i 1 =
3
∑
= =
Notation
8
Trace of a tensor : . (2.9)
Deviatoric part of a tensor: , or . (2.10)
First invariant (trace) of a tensor : . (2.11)
Second invariant of a tensor : . (2.12)
Third invariant of a tensor : . (2.13)
Throughout this report, invariants of the stress tensor will be written without the super
script identifier. For example, means the same thing as . The GeoModel supports
kinematic hardening in which the shifted stress tensor is defined ,
where is the backstress tensor (defined later). The invariants in the nonhard
ening theory will become when kinematic hardening is used.
In materials modeling, tensors are often regarded as higherdimensional vectors. The
inner product between two tensors, and , is isomorphic to (i.e., geometrically analo
gous to) the dot product between two vectors. This permits the “magnitude” of a tensor,
the “direction” of a tensor, and the “angle” between two tensors to be defined in manners
analogous to ordinary vector definitions. The direction of a tensor plays a role in the Geo
Model by defining the outward normal to the yield surface (which is actually a hypersur
face in higherdimensional tensor space). Likewise, the angle between two tensors is used
to quantify the concept of nonnormality, discussed later.
The derivatives of a scalarvalued function that depends on a secondorder tensor
as well as depending on a scalar are given by
is a secondorder tensor with ij components . (2.14)
is a scalar. (2.15)
Other derivatives are defined similarly. For example, the derivative of a secondorder ten
sor with respect to a another secondorder tensor is a fourthorder tensor with ijkl
components . Fourthorder tensors do not play a significant role in the Geo
Model theory. The only truly important fourthorder tensor is the plastic tangent stiffness
tensor, formally equal to the derivative of the stress rate with respect to the strain rate.
A
˜
˜
trA
˜
˜
A
11
A
22
A
33
+ + A
kk
= =
A
˜
˜
dev
A
˜
˜
1
3

trA
˜
˜
( )I
˜
˜
– = A
ij
dev
A
ij
1
3

A
kk
δ
ij
– =
A
˜
˜
I
1
A
trA
˜
˜
A
kk
= =
A
˜
˜
J
2
A
1
2

tr A
˜
˜
dev
( )
2
  =
A
˜
˜
J
3
A
1
3

tr A
˜
˜
dev
( )
3
  =
σ
˜
˜
J
2
J
2
σ
ξ
˜
˜
ξ
˜
˜
σ
˜
˜
α
˜
˜
– =
α
˜
˜
I
1
J
2
J
3
, ,
I
1
ξ
J
2
ξ
J
3
ξ
, ,
A
˜
˜
B
˜
˜
f σ
˜
˜
κ
∂f
∂σ
˜
˜

∂f
∂σ
ij

∂f
∂κ

A
˜
˜
B
˜
˜
∂A
ij
∂B
kl
⁄
The Stress Tensor
9
3. The Stress Tensor
This section defines the stress tensor, its principal values, its invariants, and its sign
convention. This chapter describes four canonical stress paths used to parameterize the
GeoModel: hydrostatic (HYD), triaxial compression (TXC), triaxial extension (TXE), and
shear (SHR). For transient dynamics, the GeoModel additionally requires Hugoniot and/or
Kolsky (split Hopkinson) bar data to parameterize the viscoplasticity*. This chapter
defines the distinction between the spatial Cauchy stress and the unrotated “reference”
stress. In preparation for a detailed discussion of the GeoModel theory, this chapter closes
with a detailed description of “stress space” and Lode coordinates.
The stress tensor , is defined such that the traction vector (i.e., force per unit
area), acting on any given plane with unit normal , is given by
. (3.1)
Of course, the traction and normal vectors may be described in terms of their Cartesian
components, and with respect to an orthonormal basis. The stress
tensor has a Cartesian component matrix such that the above equation may be writ
ten in matrix form as
, (3.2)
or in indicial form as
, (3.3)
where (recalling the implied summation convention) the repeated index “j” is understood
to be summed from 1 to 3 and the nonrepeated “free” index “i” appearing in each term
takes values from 1 to 3 so that the above equation is actually a compact representation of
three separate equations (one for each value of the free index).
The stress is symmetric, which means that . In continuum mechanics, and in
this report, stress is taken positive in tension. This sign convention can be the source of
considerable confusion, especially when discussing stress invariants. For example, the
trace of the stress, , is positive in tension. However, brittle materials have very low
strength in tension. Consequently, most of the functions defined in this report are nontriv
ial over only a small range of the tensile states where is positive. On the other hand,
* Until rate dependence is discussed separately in Chapter 5, all incremental or rate equations in this
report are understood to apply only under quasistatic loading and may therefore be regarded as
“inviscid” equations. Incorporating viscoplastic rate dependence requires, as a prerequisite, solu
tion of these quasistatic inviscid equations.
σ
˜
˜
t
˜
n
˜
t
˜
σ
˜
˜
n
˜
• =
t
1
t
2
t
3
, , { } n
1
n
2
n
3
, , { }
3 3 ×
t
1
t
2
t
3
σ
11
σ
12
σ
13
σ
21
σ
22
σ
23
σ
31
σ
32
σ
33
n
1
n
2
n
3
=
t
i
σ
ij
n
j
=
σ
ij
σ
ji
=
I
1
I
1
The Stress Tensor
10
most of the GeoModel functions are nontrivial over a relatively large range of compres
sive states where is negative. To help manage the sign convention problem, we will
introduce a new notation that an overbar on a variable denotes the negative. Specifically,
for any variable x,
DEFINITION OF THE “OVERBAR.” (3.4)
In our plots of any variable that varies as a function of , we will usually employ an
abscissa of , which (being the negative of ) is positive in compression. Any variable
typeset with an overbar will be positive more often than negative in most applications.
The principal stresses are the eigenvalues of the stress matrix, posi
tive in tension. Their negatives are positive in compression.* When cast in
terms of the principal basis (i.e., the orthonormal eigenvectors of the stress matrix), the
diagonal components of the stress matrix will equal the principal stresses, and the off
diagonals will be zero.
The stress deviator is the deviatoric part of the stress (see Eq. 2.10):
. (3.5)
Loosely speaking, the stress deviator is a tensor measure of shear stress. An overall scalar
measure of shear will be defined later. The quantity is called the mean stress,
and we will denote it by . The negative of the mean stress, , is called the pres
sure, and is positive in compression. Noting that the mean stress (or pressure) is simply a
multiple of , its value is an invariant, meaning that the sum of diagonal stress compo
nents will have the same value regardless of the orthonormal coordinate system used to
describe the stress components. The principal directions of the stress deviator are the same
as those for the stress itself, and the principal values for the stress deviator are related to
the principal stresses by
(3.6)
(3.7)
. (3.8)
The trace of is zero, which implies that the numerically largest principal value of
any nonzero stress deviator will always be positive and the smallest will always be nega
tive. In model parameterization tests, a sufficiently high confining pressure is typically
superimposed on the stress deviator to make all principal stress components compressive
even though the principal deviatoric stresses always have mixed signs.
* Of course, if principal stresses are ordered such that , then the barred principal stresses
will be ordered .
I
1
x x – ≡
I
1
I
1
I
1
σ
1
σ
2
σ
3
, , ( )
σ
1
σ
2
σ
3
, , ( )
σ
1
σ
2
σ
3
≤ ≤
σ
1
σ
2
σ
3
≥ ≥
S
˜
˜
S
˜
˜
σ
˜
˜
1
3

trσ
˜
˜
( )
I
˜
˜
– =
1
3

trσ
˜
˜
( )
p p p – =
I
1
s
1
σ
1
1
3
 σ
1
σ
2
σ
3
+ + ( ) – =
s
2
σ
2
1
3
 σ
1
σ
2
σ
3
+ + ( ) – =
s
3
σ
3
1
3
 σ
1
σ
2
σ
3
+ + ( ) – =
S
˜
˜
p
The Stress Tensor
11
Stress invariants
The trace operator is an invariant, which means its value will be the same regardless of
which orthonormal basis is used for the stress components. Being symmetric, the stress
tensor has a total of three independent invariants:
= (3.9)
= (3.10)
= . (3.11)
The fact that these invariants are computed from the stress tensor is sometimes empha
sized by typesetting them as , , and . Similarly defined invariants for some other
tensor would be typeset as . Invariants for a tensor would be written
, and so forth. Any invariant written without a clarifying superscript should
be understood to be a stress invariant.
The mean stress is defined to be the average of the principal stresses, whereas pres
sure is just the negative of mean stress:
mean stress: pressure: , where . (3.12)
Superimposing an extra pressure on any stress state causes the pressure to increase
from to , while having no effect on the stress deviator and therefore no effect on
the second and third invariants. Because the stress deviator has a zero trace, it can be
shown that also equals the determinant of the stress deviator so that , and
the second invariant can be written alternatively as .
Equation (3.10) shows that the invariant is never negative, which permits us to
define a supplemental stress invariant, the signed equivalent shear stress as
. (3.13)
The “transfer of sign” operator* is defined
. (3.14)
As defined, the equivalent shear stress will have a numerical sign that is positive in triaxial
extension states (defined below), negative in triaxial compression, and it will be identi
cally equal to the applied shear stress if the stress tensor happens to be in a state of pure
shear (also defined below).
* which is an intrinsic function in most computing languages
I
1
trσ
˜
˜
= σ
1
σ
2
σ
3
+ +
J
2
1
2
 
trS
˜
˜
2
=
1
2

s
1
2
s
2
2
s
3
2
+ + ( )
J
3
1
3
 
trS
˜
˜
3
=
1
3

s
1
3
s
2
3
s
3
3
+ + ( )
σ
˜
˜
I
1
σ
J
2
σ
J
3
σ
ξ
˜
˜
I
1
ξ
J
2
ξ
J
3
ξ
, , { }
α
˜
˜
I
1
α
J
2
α
J
3
α
, , { }
p
p
p
I
1
3
 = p
I
1
3
 = I
1
I
1
– ≡
P
p p P +
S
˜
˜
J
3
J
3
s
1
s
2
s
3
=
J
2
s
1
s
2
s
2
s
3
s
3
s
1
+ + ( ) – =
J
2
τ
τ sign J
2
J
3
,   =
sign x y ,  
x if y 0 ≥
x – if y 0 <
¹
´
¦
=
The Stress Tensor
12
For clarity, the values of the invariants have been shown here in terms of the principal
values of the stress and its deviator. However, because the trace operation gives the same
result regardless of which basis is used, the invariants are computed in practice directly
from fullypopulated component matrices, thereby avoiding the need for an expen
sive eigenvalue analysis.
Derivatives of the stress invariants
For isotropic material modeling, each scalarvalue function of stress, , is pre
sumed to depend only on the principal stress values, not on the principal stress directions.
Equivalently, the function is isotropic if and only if it may be expressed alternatively
as a function of the three stress invariants . In situations where the derivative of
with respect to stress is required, the chain rule can be applied as follows:
. (3.15)
In symbolic tensor notation, this expansion is written as
. (3.16)
Because the three invariants are each proper functions of the stress tensor, their derivatives
may be computed in advance:
= the identity tensor (3.17)
= = the stress deviator (3.18)
= = the “Hill” tensor (3.19)
Thus, Eq. (3.16) may be written
. (3.20)
Of particular interest is the trace of the above expression, . Since both and
are deviatoric, the result is
. (3.21)
The factor of 3 appears simply because . This, by the way, is a good example of
the fallibility of indicial notation. Specifically,
3 3 ×
φ σ
˜
˜
( )
φ σ
˜
˜
( )
I
1
J
2
J
3
, ,
φ
∂φ
∂σ
ij

∂φ
∂I
1

∂I
1
∂σ
ij

∂φ
∂J
2

∂J
2
∂σ
ij

∂φ
∂J
3

∂J
3
∂σ
ij
 + + =
∂φ
∂σ
˜
˜

∂φ
∂I
1

dI
1
dσ
˜
˜

∂φ
∂J
2

dJ
2
dσ
˜
˜

∂φ
∂J
3

dJ
3
dσ
˜
˜
 + + =
dI
1
dσ
˜
˜

I
˜
˜
=
dJ
2
dσ
˜
˜

S
˜
˜
= dev σ
˜
˜
( )
dJ
3
dσ
˜
˜

T
˜
˜
= dev
S
˜
˜
2
( )
∂φ
∂σ
˜
˜

∂φ
∂I
1

\ .
 
I
˜
˜
∂φ
∂J
2

\ .
 
S
˜
˜
∂φ
∂J
3

\ .
 
T
˜
˜
+ + =
tr ∂φ ∂σ
˜
˜
⁄ ( )
S
˜
˜
T
˜
˜
tr
∂φ
∂σ
˜
˜

\ .

 
3
∂φ
∂I
1

\ .
 
=
trI
˜
˜
3 =
The Stress Tensor
13
. (3.22)
Special stress states
This section defines the four main stress states that are used to parameterize the Geo
Model. These are hydrostatic loading (HYD), triaxial compression (TXC), triaxial exten
sion (TXE), and shear (SHR). The purpose of the GeoModel is to interpolate realistically
between known material response at these canonical states to describe material behavior
under general stress states.
HYDROSTATIC (HYD). Loading is “hydrostatic” when components of the stress
tensor are of the form
for hydrostatic stress states. (3.23)
In practice, the pressure is usually compressive (and therefore positive). Hydrostatic
testing is very important to parameterization of the GeoModel because it indirectly char
acterizes the influence of material porosity. When hydrostatically loaded to a high pres
sure and then unloaded, a nonporous material will trace through the same stress states on
both the loading and unloading curves. A porous material, on the other hand, will unload
along a different path. If possible, hydrostatic testing for the GeoModel should be con
ducted to sufficiently high pressures to compress out all pores, as indicated in Fig. 3.1.
*
* Assuming that the matrix material is plastically incompressible, the porosity is ,
where is the logarithmic (Hencky) residual strain after full void collapse. If the residual
strain is small, a Taylor series expansion of this formula gives .
tr
∂φ
∂σ
˜
˜

\ .

 
∂φ
∂σ
kk

k 1 =
3
∑
=
∂φ
∂ σ
kk
k 1 =
3
∑
 ≠
σ
˜
˜
 
p – 0 0
0 p – 0
0 0 p –
=
p
p
ε
v
Partial pore collapse Total pore collapse
p
ε
v
Figure 3.1. Typical hydrostatic (pressure vs. volumetric strain) compression data. Total pore col
lapse is achieved when the unloading curve (here shown as nonlinearly elastic) is tangent to the loading
curve. In this case, the residual volumetric strain approximately equals the initial porosity.
residual strain
initial porosity ≈
1 e
ε
v
residual
–
ε
v
residual
porosity ε
v
residual
≈
The Stress Tensor
14
When a material is loaded under hydrostatic tension instead of compression, inelastic
response is again possible, but the mechanism of failure is catastrophic growth and coales
cence of microcracks, resulting in material softening and, ultimately, complete loss of load
carrying ability.
TRIAXIAL (TXC and TXE): Loading is “triaxial” whenever two principal stresses
(denoted and called the “lateral” stresses) are equal to each other, but distinct from the
third “axial” principal stress (denoted ). Thus, with respect to the principal basis,
for triaxial stress states, (3.24)
and
for triaxial stress states. (3.25)
Also,
for triaxial stress states (3.26a)
for triaxial stress states (3.26b)
for triaxial stress states. (3.26c)
The signed equivalent shear stress for triaxial loading is
(3.27)
The invariants defined here may be written alternatively in terms of compressive stress
measures as
(3.28)
where
. (3.29)
σ
L
σ
A
σ
˜
˜
 
σ
A
0 0
0 σ
L
0
0 0 σ
L
=
S
˜
˜
 
σ
A
σ
L
–
3

2 0 0
0 1 – 0
0 0 1 –
=
I
1
σ
A
2σ
L
+ =
J
2
1
3

σ
A
σ
L
– ( )
2
=
J
3
2
27

σ
A
σ
L
– ( )
3
=
τ
σ
A
σ
L
–
3
 =
I
1
3τ 3σ
L
+ = J
2
τ
2
= J
3
2 3
9

τ
3
=
τ
σ
A
σ
L
–
3
 =
The Stress Tensor
15
The term “triaxial” is a bit of a misnomer because there are not really three indepen
dent loads applied — the lateral stresses are equal. These experiments are normally per
formed on cylindrical test specimens with the lateral load supplied by a pressure bath. For
triaxial compression (TXC) the axial stress is more compressive than the lateral stress.
For triaxial extension (TXE) the axial stress is not necessarily tensile — it is merely
less compressive than the lateral stress. For TXC, the specimen changes shape such that its
lengthtodiameter ratio decreases. For TXE, the lengthtodiameter ratio increases even
though the length and diameter might individually both decrease. Uniaxial stress compres
sion (also called unconfined compression) is a special form of TXC in which the
axial stress is compressive and the lateral stress is zero. Uniaxial stress extension is a spe
cial form of TXE in which the axial stress is tensile and the lateral stress is zero. Uniaxial
strain compression, which is typical in flyerplate impact experiments is a special case
of TXC in which the axial stress is compressive, while the lateral strain is zero (making
the lateral stress also compressive, but less compressive than the axial stress). Biaxial
tension is a special case of TXC in which the lateral stress is tensile and the axial stress
is zero. Biaxial compression is a special case of TXE in which the lateral stress is com
Figure 3.2. Triaxial compression (TXC) and triaxial extension (TXE) Two principal stresses (the
lateral stress) are equal. For TXC, the axial stress is more compressive than the lateral stress. For TXE,
the axial stress is less compressive than the lateral stress. In the labels, ; stress is positive in
tension while stress is positive in compression.
σ σ – = σ
σ
σ
L
σ
L
σ
A
σ
L
<
σ
A
σ
L
>
σ
A
σ
L
>
σ
A
σ
L
<
TXC
TXE
The Stress Tensor
16
pressive and the axial stress is zero. According to Eq. (3.27), the signed shear stress satis
fies for TXC, whereas for TXE. Consequently, Eq. (3.28) shows that
(and hence ) for TXC, while (and hence ) for TXE. Of
course, for both TXC and TXE because it is the square of .
In typical triaxial experiments, the lateral stress is held fixed (via a pressure bath)
while only the axial stress is varied. In this case, Eq. (3.28) implies that
for triaxial stress loading with fixed lateral stress. (3.30)
Being easily achieved in the laboratory, TXC and TXE data are essential to parameter
ize the GeoModel. In a typical triaxial test, the material is first loaded hydrostatically in a
pressure bath until all three principal stresses reach a compressive pressure . There
after, the lateral stresses are held fixed at this value ( ) while the axial stress is
then increased beyond . For some experiments, the axial stress might be increased
only until the stress difference reaches a given value, after which all stresses are again
increased by equal amounts. These are called constant stress difference (CSD)
τ 0 > τ 0 <
J
3
0 > J
3
0 < J
3
0 < J
3
0 >
J
2
0 > τ
dτ
d I
1

1
3
 =
A typical TXC
load path
A typical TXC
CSD load path
τ
I
1
I
1
Figure 3.3. Triaxial and CSD load paths. The material is first compressed hydrostatically to a pre
selected bath pressure ; at this point, the value of the first stress invariant is and
therefore . When the triaxial leg begins, the lateral stress is held constant
while the axial stress is varied. This causes both the first and second invariants to change such that the
path in this stress plot is a straight line with slope . For simple triaxial loading, the stress differ
ence is increased until material failure occurs. For CSD loading, the stress difference is increased to a
preselected value, and then held fixed while all stress components are thereafter varied equally until
failure occurs.
P
bath
I
1
3P
bath
– =
I
1
3P
bath
= σ
L
P
bath
= ( )
1 3 ⁄
3
1
τ
30°
3
1
3P
bath
3P
bath
A typical TXC
τ
I
1
3
1
30°
3P
bath
A typical TXE
CSD load path
I
1
τ –
A typical TXE
load path
τ –
I
1
P
bath
σ
L
P
bath
=
P
bath
The Stress Tensor
17
experiments. Typically, these experiments are run to the point of material failure. Periodic
partial unloading during a test reveals yielding if the unloading stressstrain curve has a
noticeably different slope than the loading curve (without unloading, it would be impossi
ble to definitively distinguish plasticity from nonlinear elasticity).
As illustrated in Fig. 3.3, a series of TXC experiments at various bath pressures and/or
stress differences results in a family of stressatyield points that map out the TXC
meridional profile of the GeoModel yield surface. Similar experiments under TXE map
out the TXC meridional profile. Usually, the TXE failure envelope will be shaped simi
larly to the TXC envelope, but lower in magnitude because, at a given value of , the
value of at failure is generally lower for TXE than for TXC. A plot of the failure
envelope in vs. space is essentially equivalent to the meridional “side” view of the
yield surface (Fig. 1.0b), except with the axes scaled differently. The TXE experiments are
mapping out the crosssection of the yield surface along which pressure varies while stay
ing on the “base” of the triangular octahedral profile in Fig. 1.0c (page 1), whereas TXC
experiments reveal how the apex of the triangle varies with pressure. For metals, there is
little difference between the stress intensity required to initiate failure in TXC compared to
TXE. However, for brittle materials, the difference is quite noticeable and (according to
idealized microphysical theories) can be attributed to internal frictional resistance to shear
crack growth. Because friction increases with pressure, the material strengths in TXC and
TXE tend to increase with pressure but in approximately the same proportions so that the
ratio of TXC strength to TXE strength is approximately pressure independent. Conse
quently, the TXE profiles shown in Fig. 3.3 are shaped identically to the TXC profiles
except smaller in magnitude.
Another form of triaxial loading, commonly used for dynamic material testing, is
uniaxial strain, in which the lateral strain is held constant. If the lateral strain is
held constant while continuing to compress axially, the lateral compressive stress will
increase to prevent lateral motion. For uniaxial strain, , and therefore Hooke’s
law
*
in rate form reduces to
Uniaxial strain: and , (3.31)
where and are, respectively, the tangent elastic bulk and shear moduli.
For uniaxial strain, the rate of the signed equivalent shear stress and the rate of the first
stress invariant are
Uniaxial strain: and , (3.32)
and therefore the path through stress space is a straight line with slope
* The general form of Hooke’s law, applicable to any form of triaxial loading, is given in Eq. (4.11).
Eq. (3.31) is a special case of Eq. (4.11) in which , with Eq. (4.13) used to express Young’s
modulus and Poisson’s ratio in terms of and .
I
1
τ , ( )
I
1
J
2
τ I
1
ε
L
ε
L
σ
L
ε
·
L
0 =
ε
·
L
0 =
E ν K G
σ
·
A
K
4
3
G +
\ .
 
ε
·
A
= σ
·
L
ν
1 ν –
σ
·
A
K
2G
3
 –
\ .
 
ε
·
A
= =
K G
τ
·
2Gε
·
A
3
 = I
·
1 3Kε
·
A
=
The Stress Tensor
18
for triaxial stress with fixed lateral strain, (3.33)
where is Poisson’s ratio. Since Poisson’s ratio typically varies between 0 and ,* this
result shows that the trajectory in the meridional profile will generally have a shallower
slope under uniaxial strain loading than under uniaxial stress loading. This result should
make some intuitive sense. Uniaxial strain experiments are conducted by applying
increasing levels of compression in the axial direction while holding the lateral strain
fixed. As the axial strain is compressed, the material “wants” to expand laterally, but is not
permitted to — a lateral compressive force prevents this outward motion. This constrain
ing lateral compression makes larger than it would be when lateral expansion is
unconstrained. The larger results in a shallower slope in the stress trajectory.
SIMPLE/PURE SHEAR and PRESSURESHEAR LOADING (SHR).
A material is in a state of simple shear with respect to a given coordinate system if
the stress matrix in that system is of the form
for simple shear, (3.34)
where is the shear stress. The eigenvalues of this matrix are . In general, any
stress state that is deviatoric with one eigenvalue being zero is said to be a pure shear
[27, p. 16]. (Thus, simple shear is a special type of pure shear). For an isotropic material
model like the GeoModel, yield depends only on the principal stresses, so there is no prac
tical difference between simple and pure shear (except when the model is anisotropic
because of kinematic hardening).
For conducting material characterization experiments, pure shear of the form
for pure shear (3.35)
is most convenient. For brittle materials, pure shear is difficult to attain because one of the
eigenvalues is always tensile. Frequently, pure shear is superimposed with enough confin
ing hydrostatic pressure to make all principal stresses negative (compressive). Specifi
cally, superimposing the hydrostatic loading of Eq. (3.23) onto the shear stress of
Eq. (3.35), gives a state of combined pressureshear (SHR) loading:
* Strictly speaking, positive definiteness of the elastic stiffness tensor merely requires .
Whereas negative Poisson’s ratio has been observed in manmade materials with reentrant micro
structures, it has not (to our knowledge) been reported for naturally occurring materials. Perfor
mance of the GeoModel has not been verified for materials with negative Poisson’s ratio.
dτ
d I
1

1
3

2G
3K

1
3

1 2ν –
1 ν +

\ .
 
= =
ν 1 2 ⁄
1 – ν 1 2 ⁄ < <
I
1
I
1
σ
˜
˜
  S
˜
˜
 
0 s 0
s 0 0
0 0 0
= =
s s s – 0 , , { }
σ
˜
˜
  S
˜
˜
 
s 0 0
0 s – 0
0 0 0
= =
The Stress Tensor
19
for combined pressureshear loading. (3.36)
The invariants for combined pressureshear loading are
for combined pressureshear loading (3.37a)
for combined pressureshear loading (3.37b)
for combined pressureshear loading (3.37c)
. (3.38)
Spatial and Reference stress (frame indifference)
The spatial Cauchy stress tensor defined in Eq. (3.1) is the “familiar” stress tensor
used in everyday engineering applications. Let us now denote that stress by . The
elasticity component of solids models requires knowledge of both the initial and current
configurations. Moreover, the principle of material frame indifference demands that if a
second problem were considered that had the same initial configuration, but a current con
figuration that is identical to the current configuration of the first problem, except also rig
idly rotated, then the predicted spatial stresses for the second problem should be identical
to those of the first problem, except rigidly rotated by the same amount. This concept is
quite different from a mere basis change because the initial configuration is identical for
both problems.
Satisfying material frame indifference in a spatial context can be computationally
expensive and errorprone because anisotropic internal state variables (such as directions
of material fibers or orientation of the backstress) must be rotated into the spatial frame,
and special “objective” rates must be integrated in constitutive models. A mathematically
equivalent (and numerically more accurate and efficient) strategy instead applies the con
stitutive model within an unrotated reference configuration. With this approach, rotation
of internal variables is not required, and all rates that appear in the constitutive model are
more easily integrated true rates instead of corotational rates.
If is the proper orthogonal tensor (found from a polar decomposition of the defor
mation) that characterizes the material rotation, then the unrotated stress is simply
(3.39)
By working in the unrotated reference configuration, the GeoModel predicts the
stresses for the nonrotating problem. Upon receiving the GeoModel’s update of the unro
tated stress, the host code then rotates the predicted stress back into the spatial frame.
Roughly speaking, this approach will give results identical to a spatial constitutive model
that is cast in terms of polar objective rates.
σ
˜
˜
 
s p – 0 0
0 s – p – 0
0 0 p –
=
I
1
3p – =
J
2
τ
2
=
J
3
0 =
τ s =
σ
˜
˜
spatial
R
˜
˜
σ
˜
˜
unrotated
R
˜
˜
T
σ
˜
˜
spatial
R
˜
˜
• • =
The Stress Tensor
20
For problems involving massive material rotation (e.g., turbine blades, vortices, tum
bling rock fragments, etc.), the “unrotation” strategy can give considerably more accurate
answers because the host code may, optionally, use the rotation tensor at the beginning
of the time step when computing the starting value of , but then the host code
may use at the end of the step when recasting the updated value of (output
of the GeoModel) to the spatial frame. Hence, this approach supports socalled “strong
objectivity” [38] in a very natural way.
Throughout the remainder of this report, the “stress” must be understood to be the
unrotated stress Likewise all other vector or tensor variables (such as the strain
rate) mentioned in this report are understood to be cast in the unrotated configuration
(material frame). Any host code that uses the GeoModel must (1) perform these unrotation
operations, (2) call the GeoModel, and then (3) rerotate the result back to the spatial frame
upon return. For more information, see page 76.
Lode coordinates
Any isotropic yield function may be expressed in terms of the principal stresses
. Therefore, the yield surface may be visualized in a 3D space for which the
Cartesian coordinates are these principal stresses. The value of the yield function
must be independent of the ordering of the eigenvalues. Therefore, as seen
in Fig. 1.0 (page 1), the yield surface must have rotational symmetry about the [111]
(hydrostat) direction and reflective symmetry about the TXE and TXC axes in the octahe
dral plane (i.e., the view looking down the [111] axis).
The principal Cartesian coordinates comprise an adequate choice
for characterizing stress space, but the yield function is often cast in terms of different
independent variables to exploit the yield surface’s natural symmetries optimally. The nat
ural symmetries suggest instead using cylindrical coordinates — called Lode
cylindrical coordinates — for which the zaxis is parallel with the [111] symmetry
axis. We have placed a bar on the symbol for the angular coordinate because we intend
to define it so that it will be positive in TXC and negative in TXE. A constant plane is a
meridional plane, and a plot of vs. . at a given value of is called a meridional
profile. Because most of the yield surface resides in the compressive domain where
, we will usually display meridional profiles as vs. (where ). Any con
stantz plane is an octahedral plane, and any crosssection looking down the [111] axis
(i.e., on a plane of constant ) is in an octahedral profile. Meridional and octahedral
profiles are illustrated in Fig. 1.0 (page 1).
In this report, the Lode angle is defined so that it equals zero in SHR. It varies from
in TXE to in TXC. Superimposing pressure on a stress state changes only the
axial zcoordinate, leaving the octahedral coordinates unchanged, which makes
Lode coordinates a natural choice when decomposing tensors in to their isotropic and
deviatoric parts. The radial coordinate equals the magnitude of the stress deviator. The
coordinate is proportional to the mean stress. The angular coordinate is a measure of the
R
˜
˜
n
σ
˜
˜
unrotated
R
˜
˜
n+1
σ
˜
˜
unrotated
σ
˜
˜
σ
˜
˜
.
unrotated
σ
1
σ
2
σ
3
, , ( )
f σ
1
σ
2
σ
3
, , ( )
120°
σ
1
σ
2
σ
3
, , ( )
r θ z , , ( )
θ
θ
r z θ
z 0 < r z z z – =
z
θ
30° – +30°
r θ , ( )
r z
The Stress Tensor
21
relative proportions of the principal values of the stress deviator. Thus, as the Lode angle
varies from to , the stress transitions through TXE, SHR, and TXC states.
Cylindrical Lode coordinates are especially useful because they may be expressed in
terms of stress invariants, thereby eliminating the need for an eigenvalue decomposition.
The axial zcoordinate is positive on the tensile part of the hydrostat, so is positive
on the compressive hydrostat. We define the Lode angle to be positive in TXC and neg
ative in TXE. The change in variables from principal coordinates to Lode coordinates per
mits the yield function to be alternatively expressed in the form .
When phrased in terms of Lode coordinates, the yield function needs to be defined only
over a sextant on any octahedral plane. The symmetry properties of the yield surface
may be used to reconstruct the octahedral profile over the full range from to .
Performing these necessary but tedious coordinate transformations from principal
stresses to cylindrical Lode coordinates, it can be shown [31] that the cylindrical Lode
coordinates may be determined directly from the , , and scalar stress invariants,
eliminating the need for an eigenvalue analysis. Specifically,
(3.40a)
(3.40b)
(3.40c)
The square root coefficients are merely byproducts of the coordinate transformations. For
example, since the zcoordinate is the projection of the stress onto the [111] axis, the
appears because the magnitude of the [111] vector is . The Lode radius r is zero if
. Also, the Lode angle is undefined when , which should not be too disturb
ing since the angular coordinate for any cylindrical system is undefined when the point in
question lies on the symmetry axis (which, in this case, is the [111] hydrostat).
Later, when we give the mathematical formulation for the GeoModel yield function, it
will be phrased as . Using the above formulas, the yield function is ultimately
implemented in the form . The invariant influences only the Lode angle.
When simpler yield models (DruckerPrager) are independent of the third stress invariant,
they are therefore independent of the Lode angle, which makes their octahedral yield pro
file a circle. The GeoModel must include a noncircular yield profile to reproduce TXE/
TXC strength differences clearly evident in the data for geological materials. Thus, the
GeoModel must necessarily use all three Lode coordinates (equivalently, all three stress
invariants).
30° – +30°
z
θ
f σ
1
σ
2
σ
3
, , ( ) f r θ z , , ( )
60°
0° 360°
I
1
J
2
J
3
r 2J
2
=
3θ sin
J
3
2

3
J
2

\ .
 
3 2 /
– =
z
I
1
3
 =
3
3
J
2
=0 θ J
2
=0
f r θ z , , ( )
f I
1
J
2
J
3
, , ( ) J
3
The Stress Tensor
22
Octahedral yield profile visualization. Given a yield function , yield
profiles may be generated by solving to obtain expressed as a function of .
A meridional profile is generated by plotting vs. at a fixed value of . An octahedral
profile, which corresponds to a yield surface crosssection at a given value of , describes
how the Lode radius at yield varies with the Lode angle. Rather than plotting vs. ,
octahedral profiles are obtained by parametrically plotting Cartesian coordinates
and (3.41)
Here, is an angle that varies over the full range from 0 to . The Lode angle ,
which is permitted to vary only over the range from to , is generated from the
fullrange angle by the sawtooth function
(3.42)
With known, the value of corresponding Lode radius can be found from the yield con
dition, and finally, the family of points on the octahedral yield profile may be
generated parametrically as varies from 0 to , as illustrated in Fig. 3.4.
f r θ z , , ( )
f 0 = r θ z , ( )
r z θ
z
r θ
x
1
r Θ cos = x
2
r Θ sin =
Θ 360° θ
30° – 30°
Θ
θ
1
3
ArcSin 3Θ ( ) sin   =
θ r
x
1
x
2
, ( )
Θ 360°
σ
3 >
σ
2 >
σ
1
σ3
<
σ1
<
σ2
σ
1 >
σ
2 >
σ
3
σ2
<
σ1
<
σ3
SHR ( θ=0 )
S
H
R
(
θ
=
0
)
S
H
R
(
θ
=
0
)
SHR ( θ=0 )
S
H
R
(
θ
=
0
)
S
H
R
(
θ
=
0
)
T
X
C
(
θ
=
3
0
o
)
σ
1
>
σ
3
>
σ
2
σ
1
<
σ
3
<
σ
2
T
X
C
(
θ
=
3
0
o
)
T
X
C
(
θ
=
3
0
o
)
T
X
E
(
θ
=
–
3
0
)
T
X
E
(
θ
=
–
3
0
)
T
X
E
(
θ
=
–
3
0
)
σ
1
σ
3
σ
2
Figure 3.4. An octahedral yield profile. Geological materials tend to be stronger in TXC than in
TXE, which is why the TXC axes are always on an apex of the rounded triangle (i.e., farther from the or
igin, corresponding to higher strength). The Lode angle alternates cyclically from in TXE to
in TXC because the yield threshold must be independent of the eigenvalue ordering.
θ 30° – 30°
The Stress Tensor
23
Meridional yield profile visualization. To draw a geometrically accurate meridi
onal crosssection of the yield surface, the profile should be plotted as vs. for a given
value of . Typically, we will plot the TXC ( ) profile. Using the and Lode
coordinates as the axes in a meridional plot ensures that lengths and angles in the meridi
onal profile will equal corresponding lengths and angles in stress space. Many times, how
ever, we will depict a geometrically distorted view of the meridional profile by instead
plotting vs. , where the signed equivalent shear stress equals , depending on
whether the loading is closer to TXE or TXC. Recalling from Eqs. (3.40a) and (3.40c) that
and , a plot of vs. is equivalent to changing the aspect ratio
of an vs. plot by a factor of , as illustrated in Fig. 3.5. Thus, whenever we
plot the meridional profile as vs. , keep in mind that the actual meridional crosssec
tion in stress space is smaller in width by a factor of 2.45. Meridional profile distortion is
an issue only when ascertaining the direction of the yield surface normal. Figure 3.5
shows that the normal to the yield profile in a distorted plot does not correspond to the
normal in stress space.
r z
θ θ 30° = r z
τ I
1
τ J
2
±
J
2
r 2 ⁄ = I
1
3z = τ I
1
r z 6 2.45 =
τ I
1
r
z
τ J
2
=
I
1
2J
2
p
Figure 3.5. Distortion of the meridional profile when using nonisomorphic stress measures. Only a
plot of vs. will result in a geometrically accurate depiction of a meridional crosssection of stress
space for which angles and lengths are preserved. The middle plot shows the magnitude of the stress devi
ator plotted against the pressure, resulting in a plot eccentricity of . The last plot shows the
equivalent shear stress plotted against the first stress invariant, for a plot eccentricity of .
r z
1 3 ⁄ 0.577 =
6 2.45 =
isomorphic
(geometrically accurate) distorted
distorted
The Stress Tensor
24
Closedform solution for ordered eigenvalues. Recalling that Lode cylindrical
coordinates merely represent a coordinate change from the principal coordinates
to a new set of coordinates , it follows that inverse transformation
formulas should exist for obtaining the principal stresses from Lode coordinates. Each dis
tinct sextant in Fig. 3.4 merely corresponds to a different eigenvalue ordering. Regardless
of the sextant in which the stress resides, falls on the piplane axis correspond
ing to the smallest eigenvalue, whereas falls on the axis of the largest eigen
value. Therefore, transformation formulas that convert cylindrical coordinates back to
Cartesian coordinates only need to be defined over the range from through to
determine the ordered eigenvalues.
Letting the compressive eigenvalues be ordered , the tensile eigenvalues
must be ordered , and the inverse transformation formulas are
low: (3.43a)
middle: (3.43b)
high: . (3.43c)
These formulas constitute a closedform solution for the ordered eigenvalues of any real
symmetric matrix, not just a stress. * Using these formulas, any yield function that is
stated in terms of principal stresses, , can be immediately recast into a form
expressed in terms of stress invariants, , which is more convenient for plastic
ity modeling because it can be differentiated without an eigenvector analysis (see
Eq. 3.16). For example, any material model that seeks to initiate failure when the largest
tensile principal stress reaches a critical value, , can do so by simply substituting
Eq. (3.43c) into the failure criterion, . The above closedform solution for
ordered eigenvalues is applied in Appendix B (page B22) to convert the MohrCoulomb
theory of failure into a formulation expressed in terms of stress invariants, as required in
the GeoModel.
* The solution quoted here is equivalent to the closed form solution derived by Malvern [32] via a
trigonometric substitution. Malvern’s angle is a Lode angle, but defined to be zero in TXE and
in TXC.
σ
1
σ
2
σ
3
, , ( ) r θ z , , ( )
θ 30° =
θ 30° – =
30° – +30°
σ
3
σ
2
σ
1
≤ ≤
σ
3
σ
2
σ
1
≥ ≥
σ
1
z
3

r
2
 θ cos
θ sin
3
 + –
I
1
3
 J
2
θ cos
θ sin
3
 + – = =
σ
2
z
3

2
3
 r θ sin +
I
1
3

2
3
 J
2
θ sin + = =
σ
3
z
3

r
2
 θ cos
θ sin
3
 – +
I
1
3
 J
2
θ cos
θ sin
3
 – + = =
3 3 ×
α
60°
f σ
1
σ
2
σ
3
, , ( )
f I
1
J
2
J
3
, , ( )
σ
3
σ
3
crit
σ
3
σ
3
crit
=
GeoModel theory
25
4. GeoModel theory
Being a generalized plasticity theory, the GeoModel is founded upon an additive
decomposition of the strain rate into separate contributors: from elastic straining and
from inelastic straining:
. (4.1)
The GeoModel permits the host code to employ any definition of the strain so long as its
rate is conjugate to the stress in the sense that the work rate per unit volume is given by
. (4.2)
To satisfy the principle of material frame indifference, the host code must cast the stresses
and strain rates in an unrotated configuration. At present, all implementations of the Geo
Model have approximated the strain rate by the unrotated symmetric part of the velocity
gradient:
, (4.3)
where is the velocity vector, is the current spatial position vector, and the tensor is
the rotation from the polar decomposition of the deformation gradient. The conjugate
stress is the unrotated Cauchy stress defined in Eq. 3.39. Henceforth, all references to the
stress and the strain rate must be understood to be the unrotated stress and strain
rate.
All GeoModel material parameterizations to date have been based on the above
approximation for the strain rate. Using a different choice for the conjugate stress and
strain rate measures would, of course, entail adjusting material parameters appropriately.
The strain rate in Eq. (4.3) is an approximation because, for general deformations, it is not
precisely equal to the rate of any proper function of the deformation
*
. The approximate
strain rate in Eq. (4.3) exactly equals the unrotated logarithmic (Hencky) strain rate for
any deformation having stationary principal stretch directions. It is an excellent approxi
mation to the Hencky strain rate even when principal stretch directions change orientation
as long as the shear strains remain small (volumetric strains may be arbitrarily large). For
geological applications, material rupture generally occurs well before shear strains
become large, so Eq. (4.3) is a prudent choice for the strain rate measure. If, however, the
model is to be subjected to significant cyclical loading (e.g. fatigue), then a proper strain
rate should be used instead of Eq. (4.3) even if the distortional strains are always small.
* Paths can be devised for which the starting and ending configurations are identical, but the time
integral of does not evaluate to zero [11].
ε
˜
·
˜
ε
˜
·
˜
e
ε
˜
·
˜
p
ε
˜
·
˜
ε
˜
·
˜
e
ε
˜
·
˜
p
+ =
σ
˜
˜
σ
˜
˜
:ε
˜
·
˜
ε
·
ij
1
2

∂v
m
∂x
n

∂v
n
∂x
m
 +
\ .
 
R
mi
R
nj
≈
v
˜
x
˜
R
˜
˜
σ
ij
ε
·
ij
ε
·
ij
GeoModel theory
26
Elasticity
The GeoModel supports both linear and nonlinear hypoelasticity*. The GeoModel
presumes the material is elastically isotropic and that the elastic stiffness tensor is
itself isotropic (i.e., deformationinduced elastic anisotropy is not included). Conse
quently, the stress is governed by a rate form of Hooke’s law:
. (4.4)
Because the elastic tangent stiffness tensor, , is presumed to be isotropic, Eq. (4.4)
can be written as two separate and much simpler equations, one for the volumetric
response and the other for the deviatoric response:
(4.5)
and
. (4.6)
Here, and are the tangent shear and bulk elastic moduli; is the pressure (negative
of the mean stress); is the volumetric elastic strain rate computed by the trace opera
tion,
, (4.7)
is the stress deviator; and is the deviatoric part of the elastic strain rate, defined
. (4.8)
We have used the overbar (which, recall, simply denotes the negative of a variable) in our
equation for the pressurevolume response because the mean stress is typically compres
sive (negative) in most applications of the GeoModel and therefore and are typically
positive. Of course, Eq. (4.5) remains valid for volumetric expansion and tensile
mean stresses as well. No overbar is used in Eq. (4.8) because deviatoric tensors
always have eigenvalues of mixed signs.
For linear elasticity, the user merely specifies constant values for the bulk modulus
and the shear modulus . For nonlinear elasticity, the moduli are stressdependent tan
gent moduli (i.e., slopes of the tangents to the stressstrain curves). Three parameters are
available for fitting the nonlinear tangent bulk modulus to laboratory data obtained
from unloading curves in hydrostatic compression. Similarly, three parameters are avail
able for fitting the nonlinear tangent shear modulus indirectly from triaxial test data.
Additional elastic parameters are available for materials whose elastic properties are
affected by inelastic deformation (see Eqs. 4.33 and 4.34). Stepbystep instructions for
determining elastic properties from measured data are provided in Appendix A.
* “Hypoelastic” means the stress can be written as a function of the strain, but is not derivable from
an energy potential. When a potential exists, then the formulation is “hyperelastic.”
C
ijkl
σ
·
ij
C
ijkl
ε
·
kl
e
=
C
ijkl
p
·
Kε
·
v
e
=
S
·
ij
2Gγ
·
ij
e
=
G K p
ε
·
v
e
ε
·
v
e
ε
·
kk
e
=
S
ij
γ
·
ij
e
γ
·
ij
e
ε
·
ij
e 1
3

ε
·
v
e
δ
ij
– ≡
p ε
v
e
ε
v
e
0 < ( )
p 0 < ( )
K
G
K
G
GeoModel theory
27
Nonlinear elasticity. At the user’s option, the GeoModel supports nonlinear elasticity
by permitting the elastic tangent moduli to vary with the stress according to
(4.9)
. (4.10)
In these equations, the and parameters are material constants determined via non
linear regression fitting to the unloading portions of hydrostatic compression and triaxial
compression experiments, as described in Appendix A.
*
Further descriptions of the physi
cal meanings of the parameters in these equations are given Appendix B. The GeoModel’s
functional forms for the nonlinear elastic tangent moduli are phenomenological to permit
tight empirical fits to experimental data for a wide variety of materials. Suitability of these
functions for fitting material data is demonstrated in Chapters 8 and 9, starting on page 96.
Incidentally, it can be shown that an elasticity model of the form in Eq. (4.4) is hyper
elastic (i.e., derivable from an isotropic elastic potential) if and only if the shear modulus
is constant and the bulk modulus depends at most only on . Because Eq. (4.34) permits
the shear modulus to vary, the GeoModel is hypoelastic if .
Assigning values to the elastic constants. The bulk modulus is determined
from the local tangent of the elastic part of a pressure vs. volumetric strain plot obtained
from hydrostatic testing. Rather than determining the shear modulus directly from a shear
loading experiment (where GeoMaterials tend to be weak), the shear modulus is typically
found indirectly from triaxial loading data. For triaxial loading, the stress rates are related
to the strain rates by
†
and , (4.11)
where and are, respectively, Young’s modulus and Poisson’s ratio. If and are
known, then the bulk modulus , the Lame modulus , and the shear modulus may be
determined from the wellknown elasticity equations [18],
. (4.12)
Because the GeoModel casts its elasticity model in terms of the bulk modulus and the
shear modulus , the following formulas are convenient for converting various combina
tions of elastic constants into expressions involving only and (see, for example, Ref.
[18], page 139):
* The generalized nonlinear elasticity formulas on page 37 may be used when elastic properties
appear to be affected by inelastic deformation.
† Here, we are writing Hooke’s Law in rate form to allow for the possibility that the elastic moduli
might be nonlinear. Thus, the elastic properties used here are the tangent moduli (i.e. based on the
local slope of an elastic stress vs. strain curve).
K b
o
b
1
b
2
I
1
 –
\ .
 
exp + =
G g
o
1 g
1
exp g
2
J
2
1 2 /
– ( ) –
1 g
1
–
 =
b
k
g
k
I
1
g
1
0 ≠
K
ε
·
A
σ
·
A
2νσ
·
L
–
E
 = ε
·
L
νσ
·
A
– 1 ν – ( )σ
·
L
+
E
 =
E ν E ν
K λ G
K
E
3 1 2ν – ( )
 = λ
Eν
1 ν + ( ) 1 2ν – ( )
 = G
E
2 1 ν + ( )
 =
K
G
K G
GeoModel theory
28
(4.13)
(4.14)
(4.15)
(uniaxial strain modulus — see Eq. 3.31). (4.16)
Eq. (4.11) implies that triaxial experiments conducted under constant lateral stress
satisfy
. (this applies if lateral stress is constant) (4.17)
A fixed lateral stress implies that . Thus, when a stressstrain curve is
obtained by plotting the stress difference against the axial strain for a triaxial
loading experiment in which the lateral stress is fixed, the local tangent of the curve equals
Young’s modulus . With the bulk modulus having been obtained separately from
hydrostatic test data, Eq. (4.15c) may then be used to determine the tangent shear
modulus .
The elastic limit (yield surface)
Like most plasticity models, the GeoModel begins each solution phase (i.e., each time
step) by tentatively presuming that the loading is elastic. This produces a trial elastic
stress, which is then checked to see if it is inside or on the yield surface. If so, the tentative
assumption of elasticity is validated and the actual updated stress is set equal to the trial
elastic stress. If, on the other hand, the trial elastic stress falls outside the yield surface,
then the tentative assumption of elastic response was wrong, and the solution phase is then
solved anew using the equations governing inelastic deformation. Before discussing these
inelastic governing equations, we must first characterize the yield surface itself. We will
begin by discussing yield surfaces in some generality and then progressively work
towards the precise functional form for the GeoModel yield surface.
Mathematically, the GeoModel is a generalized plasticity model. The term “plasticity”
is broadened to include not only the usual flow of material via dislocations (a phenomenon
that has actually been observed in brittle materials when they are loaded under extraordi
narily high confining pressure), but also any other mechanisms that lead to a marked
departure from elasticity. Examples include crack growth, void collapse, or perhaps even
phase transition. Rather than explicitly tracking each of these microscale failure mecha
nisms explicitly, the “yield” surface itself characterizes them all in an ensemble phenome
nological manner. If the stress state is “not too severe,” then material
response will be elastic and therefore reversible (nondissipative). Once the stress
becomes critically severe, however, the material will undergo irreversible structural
λ K
2
3
G – = ν
3K 2G –
2 3K G + ( )
 = E
9KG
3K G +
 =
E
ν

18KG
3K 2G –
 =
λ
ν

2
3
 3K G + ( ) =
ν
1 ν –

3K 2G –
3K 4G +
 =
E
1 ν –

18KG
3K 4G +
 = 2G λ + K
4
3
G + = G
3KE
9K E –
 =
K
4
3
G +
E 1 ν – ( )
1 ν + ( ) 1 2ν – ( )
 =
σ
·
L
0 = ( )
σ
·
A
Eε
·
A
=
σ
·
A
σ
·
A
σ
·
L
– =
σ
A
σ
L
– ε
A
E K
G
σ
1
σ
2
σ
3
, , ( )
GeoModel theory
29
changes that manifest as inelastic strains (nonrecoverable upon removal of the load). The
material response is elastic whenever the stress is on the inside of the yield surface. If con
tinuing to apply elasticity theory would move the stress into regions outside the yield sur
face, then plasticity equations are applied.
The GeoModel yield criterion and yield function are
GEOMODEL YIELD CRITERION:
. (4.18)
GEOMODEL YIELD FUNCTION:
. (4.19)
The remainder of this chapter is devoted to motivating the functional forms of these equa
tions and defining the numerous variables that appear in them. Briefly, the yield function
is defined such that elastic states satisfy . The yield criterion corresponds to .
The “building block” functions and are used to describe the elastic limit caused by
the presence of microcracks, whereas the function accounts for strength reduction by
porosity. The function represents the ultimate limit on the amount of shear the material
can support in the absence of pores (i.e., represents the softening transition limit thresh
old, sketched in Fig. 1.1, resulting exclusively from microcracks). The material parameter
characterizes the maximum allowed translation of the yield surface when kinematic
hardening is enabled, in which case is the second invariant of the shifted stress tensor
, where is the backstress. When kinematic hardening is disabled (i.e., when
is specified to be zero), the backstress will be zero and therefore would be simply
the invariant of the stress deviator. The function describes the limit strength, whereas
defines the yield threshold associated with cracks, which can evolve toward the
limit surface via kinematic hardening as explained later in the context of Fig. 4.7. The
function , where is the Lode angle of the shifted stress, is used to account for
differences in material strength in triaxial extension and triaxial compression. By appear
ing as a multiple of , the function accommodates material weakening caused
by porosity. The function depends on an internal state variable whose value controls
the hydrostatic elastic limit, as explained later in the context of Figs. 4.4, 4.5, and 4.15.
J
2
ξ
F
f
I
1
( ) N –   F
c
I
1
κ , ( )
Γ θ
ξ
( )
 =
f σ
˜
˜
α
˜
˜
κ , , ( ) J
2
ξ
Γ
2
θ
ξ
( ) F
f
I
1
( ) N –  
2
F
c
I
1
κ , ( ) – =
f
f 0 < f 0 =
F
f
Γ
F
c
F
f
F
f
N
J
2
ξ
ξ
˜
˜
S
˜
˜
α
˜
˜
– = α
˜
˜
N J
2
ξ
F
f
F
f
N –
Γ θ
ξ
( ) θ
ξ
F
f
N –  
2
F
c
F
c
κ
GeoModel theory
30
The principal goal of this section is to describe in great detail the physical motivations
of the GeoModel yield criterion cited in Eq. (4.18). This criterion describes the geometri
cal shape of the yield surface in stress space. For rocks and rocklike materials, the yield
surface will have a shape similar to the one illustrated in Fig. 1.0 (page 1).
*
Figure 1.0(b)
shows a “side” meridional profile of the yield surface in bold, along with a family of
other profiles from which the yield surface might have evolved over time (via continu
ously varying values of the internal state variable); Fig. 4.7 shows a similar plot when
kinematic hardening is allowed. Very little of the yield surface in Fig. 1.0(b) exists in the
tensile domain (left side of the meridional plot), implying that materials of this type are
very weak in tension.
Figure 1.0(c) shows the yield surface profile from a perspective looking down onto a
plane — called an octahedral plane — that is perpendicular to the [111] symmetry axis
and therefore represents a crosssection of the yield surface at a given pressure. Since the
onset of yield must not depend on the ordering of the principal stresses, the yield surface
for any isotropic yield model possesses rotational symmetry about the hydrostat
(i.e., the [111] axis), as well as reflective symmetry about any of the triaxial compression
or triaxial extension axes labeled TXC and TXE in Fig. 1.0(c). As seen in Fig. 1.0(c), the
octahedral profile is somewhat triangular in shape. This periodic asymmetry corresponds
to differences in the failure limit under triaxial compression (TXC) and triaxial extension
(TXE). Because the yield surface is farther from the origin on a TXC axis than on a TXE
axis, this material has higher strength in TXC than in TXE. The function character
izes the shape of the octahedral profile because is proportional to . The size of
the octahedral profile at various pressures is governed by the functions and .
Elastic stress states are “inside” the yield surface ( ). Stress states for which
are said to be “on the yield surface.” Like classical plasticity models, the yield surface in
the GeoModel characterizes the point of departure from elastic to inelastic behavior. When
the stress is on the yield surface, and if applying elasticity theory would result in an
updated stress that falls outside the yield surface, then plasticity equations will be applied.
Stress states outside the yield surface for which are unachievable except through a
hardening evolution of the internal state variables ( and/or ) corresponding to a funda
mental change of the underlying microstructure of the material. Stresses outside the limit
surface are unachievable by any quasistatic means.
The internal state variable controls the location of the yield cap (wire frame in
Fig. 1.0a on page 1). When increases in response to pore collapse, octahedral profiles
that pass through the cap will expand isotropically (i.e., the octahedral profile changes
size, but not shape, and it does so without translating in stress space). The amount of iso
tropic expansion or contraction varies with pressure in such a manner that the family of
yield surfaces corresponding to various values of is bounded by the shear limit surface,
.
* At the user’s option, the GeoModel parameters can be set to alternatively duplicate classical ideal
ized Von Mises or MohrCoulomb theory. Doing this would be inappropriate when modeling real
materials, but it can be useful in benchmark testing.
κ
120°
Γ θ ( )
r 1 Γ θ ( ) ⁄
F
f
F
c
f 0 < f 0 =
f 0 >
κ
α
˜
˜
κ
κ
κ
J
2
F
f
I
1
( ) Γ θ ( ) ⁄ =
GeoModel theory
31
Though not needed in many applications,
the GeoModel supports kinematic hardening for
which the symmetry axis of the yield surface is
permitted to shift in stress space so that the
invariants in the yield function are based on the
shifted stress tensor, defined
. (4.20)
As illustrated in Fig. 4.1, the backstress
is a deviatoric tensorvalued internal state vari
able that defines the origin about which the
yield surface is centered. When the backstress
tensor changes, the yield surface translates in
stress space, thereby supporting deformation
induced anisotropy (Bauschinger effect) in a
limited capacity. The backstress is initially zero,
but then evolves according to an evolution equation described in detail on page 59. The
GeoModel is otherwise fully isotropic, both elastically and plastically. Consequently, the
yield function is isotropic with respect to the shifted stress deviator , implying that it
depends on the invariants of the shifted stress deviator, as well as and an internal state
variable that characterizes isotropic hardening caused by void collapse or softening
caused by porosity increases. Specifically, the GeoModel yield function is of the form
, (4.21)
where
and . (4.22)
In this section, we seek to describe the size and shape of the yield surface at an instant fro
zen in time. Thus, we will focus on how the yield function depends mathematically on the
stress invariants , with the internal state variables ( and ) regarded as con
stants. The means by which the yield surface evolves in response to time variation of
and/or is discussed separately.
Before discussing the physical foundations of the GeoModel yield function in
Eq. (4.19), we will first discuss qualitative features of any yield function of the more gen
eral form in Eq. (4.21). Given that an isotropic yield function possesses alternating
symmetry about the [111] direction in stress space, the yield function in Eq. (4.21) is most
naturally cast in terms of the cylindrical Lode coordinates as
, (4.23)
α
˜
˜
ξ
˜
˜
S
˜
˜
kinematically
hardened
yield
surface
initial
yield
surface
Figure 4.1. Backstress and shifted stress.
ξ
ij
S
ij
α
ij
– ≡
α
ij
ξ
ij
I
1
κ
f f I
1
J
2
ξ
J
3
ξ
κ ; , , ( ) =
J
2
ξ
1
2
tr
ξ
˜
˜
2
= J
3
ξ
1
3
tr
ξ
˜
˜
3
=
I
1
J
2
ξ
J
3
ξ
, , ( ) κ
α
˜
˜
κ
α
˜
˜
30°
f f r θ z κ ; , , ( ) =
GeoModel theory
32
where these Lode coordinates are defined with respect to the kinematically shifted origin
in stress space. For any given values of , , and , there must exist only one radius
that is a solution to (otherwise, the yield surface would not be convex).
Thus, without loss in generality, elastic stress states for any isotropic yield function always
can be characterized in the general functional form
, (4.24)
where is regarded as a material function determined from experimental data and
is introduced here only to discuss the structure of isotropic yield functions in generality
(i.e., the GeoModel has an implied “g” function, but does not construct one explicitly).
The yield function corresponding to Eq. (4.24) may be written
(Any isotropic yield function can be written in this form.) (4.25)
At present, the GeoModel assumes that the shape of the octahedral yield profile is the
same at all pressures — only its size varies with pressure.
*
Moreover, the GeoModel pre
sumes that the shape of the octahedral yield profile is constant in time (i.e., it does not
evolve to any different shape in response to plastic deformation even though it can permis
sibly vary in size and translate in stress space). Consequently, the GeoModel’s yield func
tion is structured such that is separable into the product of two distinct
functions, one depending only on and the other depending only on and , permitting
Eq. (4.24) to be structured in the general form:
(This form results from a separability assumption.) (4.26)
As was the case with the g function, and have been introduced here only to illus
trate the basic structure of the GeoModel’s yield function. The GeoModel’s specific for
mulations will be discussed soon. A degree of ambiguity exists in the definitions of
and because they may be replaced respectively by and for any scalar
without loss in generality. To remove this ambiguity, the function is scaled such that it
merely describes the shape of the octahedral profile (i.e., the view of the yield surface
looking down the [111] direction). The function defines the meridional profile of the
yield function, and therefore this function also defines the size of the octahedral profile.
The GeoModel aims to model rocks and rocklike brittle materials. The mechanical
behavior of such materials is typically driven by two underlying mechanisms: porosity
and microcracks. To date, microphysical research has focused on the effects of only one of
these mechanisms at a time. Figure 4.2(a) shows the qualitative shape of the meridional
profile typically that is predicted when only porosity is considered. In this case, the merid
ional yield profile is a “cap” function that is essentially flat like a Von Mises profile for a
large range of pressures ( is proportional to pressure), and then the profile drops to zero
when pressure becomes large enough to collapse voids. Figure 4.2(b) shows the general
* Some evidence suggests that the octahedral yield profile should in fact vary in shape from strongly
triangular at low pressures to nearly circular at extraordinarily high pressures [23]. Consequently,
the GeoModel’s current assumption of a constant octahedral profile shape might change in future
releases.
θ z κ r
f r θ z κ ; , , ( ) 0 =
r g θ z κ ; , ( ) <
g θ z κ ; , ( )
f r
2
g θ z κ ; , ( )  
2
– =
g θ z κ ; , ( )
θ z κ
r h
1
θ ( )h
2
z κ ; ( ) <
h
1
h
2
h
1
h
2
ηh
1
h
2
η ⁄ η
h
1
h
2
z
GeoModel theory
33
shape of a meridian profile that is typically predicted for theories that consider only the
influence of microcracks without considering porosity. Microcracks lead to low strength in
tension, but strength increases as pressure is increased because pressure generates addi
tional friction at crack faces, thereby reducing the shear load suffered by the matrix mate
rial. The GeoModel, unifies these separate microscale theories to obtain a combined
porosity and microcrack model as sketched qualitatively in Fig. 4.2(c). Loosely speaking,
the GeoModel obtains the combined meridional yield function by multiplying the individ
ual porosity and microcrack profiles (and scaling the ordinate appropriately to match
data).
To date, the combined effect of voids and microcracks remains a poorly developed
branch of materials constitutive modeling. Some early models simply asserted that a mate
rial is elastic (safe from yield) only if it is safe from both crack growth and void collapse,
with each criterion tested separately. However, as illustrated in Fig. 4.3(a), this approach
results in a discontinuous slope in the meridional yield profile and fails to account for
interactions between voids and cracks. The GeoModel [Fig. 4.3(b)], phenomenologically
permits cracks and voids to interact in a way that results in a continuously differentiable
meridional profile, making the GeoModel bettersuited for reproducing observed data.
z z
r
c
z ( ) r
f
z ( )
z
r z ( )
Figure 4.2. Qualitatively meridional profile shapes resulting from (a) porosity alone, (b) microcracks
alone, and (c) porosity and microcracks in combination.
porosity
only
microcracks
only
combined
porosity
and microcracks
(a) (c) (b)
porous only
(without cracks)
z
Figure 4.3. Distinction between twosurface upperbound models and the GeoModel.
c
r
a
c
k
e
d
o
n
l
y
(
w
i
t
h
o
u
t
p
o
r
e
s
)
Simple twosurface
upper bound model
GeoModel
(one continuous surface)
r
z
r
GeoModel theory
34
The GeoModel achieves a combined porous+cracked yield surface by multiplying the
fracture function in Fig. 4.2(b) times the cap function in Fig. 4.2(a) so that
is proportional to . (4.27)
The proportionality factor depends on the Lode angle so that the equivalent shear stress
at yield (which, recall, is simply a constant multiple of ) can be made lower in TXE than
in TXC. Cap functions depend on the porosity level (which controls where the cap curve
intersects the zaxis). The curvature of a cap function controls the degree to which porosity
affects the shear response. Although the GeoModel does not explicitly track porosity, it
does include an internal state variable that equivalently accounts for the presence of
porosity. As explained later (page 50), the value of and one additional material constant,
, determine both the cap curvature and the location where the cap intersects the
hydrostat (the axis). Thus, the cap function implicitly depends on and .
Recalling that the Lode cylindrical radius equals and the Lode axial coordi
nate is proportional to , the GeoModel implements the notion of multiplying fracture
and cap functions by using Eq. (3.40) to express Eq. (4.27) in terms of stress invariants
instead of Lode coordinates, so the GeoModel yield function is of the form
. (4.28)
Comparing with Eq. (4.18), the f and F functions are related by
and . (4.29)
The invariant is computed using the shifted stress tensor , where is the
deviatoric tensorvalued backstress that is nonzero only when kinematic hardening is
enabled. Thus, in addition to depending explicitly on the stress tensor, the yield criterion
depends implicitly on material constants and on two internal state variables, (mentioned
earlier) and .
The fracture function characterizes the crackingrelated portion of the meridional
yield profile. The GeoModel cap function is normalized to have a peak value of 1. The
function characterizes the Lode angle dependence of the meridional profile and is
normalized to equal 1 in triaxial compression . At different Lode angles, usu
ally has values greater than 1, which (because it is a divisor in Eq. 4.28) reduces equiva
lent shear strength. Rather than regarding as a strength reducer, it can be alternatively
interpreted as a stress intensifier. Qualitatively, these functions are typically shaped as
shown in Fig. 4.4.
r
f
z ( ) r
c
z ( )
r z ( ) r
f
z ( )r
c
z ( )
θ
r
κ
κ
R
z r
c
z ( ) κ R
r 2J
2
ξ
z I
1
J
2
ξ
f
f
I
1
( )f
c
I
1
( )
Γ θ ( )
 =
f
f
F
f
N – = f
c
F
c
=
J
2
ξ
ξ
˜
˜
S
˜
˜
α
˜
˜
– ≡
α
˜
˜
κ
α
˜
˜
f
f
f
c
Γ θ ( )
θ 30° = ( ) Γ
Γ
GeoModel theory
35
As explained on page 43, the precise expression for the function is determined by
userspecification of two parameters: the TXE/TXC strength ratio and an integerval
ued option (J3TYPE), which controls the manner in which the octahedral profile radius
varies from the value at TXE to 1 at TXC. As explained below, the porosity (cap)
function is defined by two parameters: the initial intersection on the horizontal axis
and the eccentricity or “shape factor” for the ellipse (i.e., the width to height ratio of the
ellipse). As explained on page 56, the GeoModel internally computes evolution of the cap
function resulting from void collapse. As explained on page 41, the very important
function, which reflects influence of microcracks, is determined by fitting triaxial com
pression data to an exponential spline [up to five parameters ( , , , , )]. Stan
dard experiments (needed to assign values to these parameters) are discussed in
Appendix A. The remainder of this chapter is dedicated to providing further details about
the three key functions , , and used in the GeoModel.
θ
Γ=1
r
a
d
i
u
s
=
ψ
r
a
d
i
u
s
=
1
Γ=1/ψ
I
1
f
f
I
1
( ) F
f
N – =
Figure 4.4. Basic shapes of the three key functions that characterize the composite shape of the yield sur
face. (a) Lode angle dependence, (b) porosity cap curve, (c) limit failure curve. The Lode angle function
is the reciprocal of the radius in the octahedral plane, making it best regarded as a stress intensifier; is
normalized to equal 1 in triaxial compression, which implies that it must equal in triaxial extension (where
is the TXE/TXC strength ratio). Shear influence on void collapse begins at the point where the cap function
branches into an ellipse. Since the cap function is multiplied by the fracture curve, this transition point also
marks where the composite GeoModel failure surface branches away from , beyond which macroscale re
sponse is influenced simultaneously by both cracks and voids. For pure (shearfree) hydrostatic compression,
void collapse begins at the point where the ellipse intersects the horizontal. Only the function has dimensions
of stress (the others are dimensionless).
Γ θ ( )
Γ θ ( )
1 ψ ⁄
ψ
f
f
f
f
I
1
f
c
I
1
( ) F
c
=
1
(a)
(b) (c) κ
Γ
ψ
1 ψ ⁄
f
c
p
0
R
f
f
a
1
a
2
a
3
a
4
N
f
c
f
f
Γ
GeoModel theory
36
The cap function, f
c
. Under compression, the pores in a material can irreversibly col
lapse, thereby resulting in permanent (plastic) volume changes when the load is removed.
Plastic volume changes can occur for porous media even if the matrix material is plasti
cally incompressible. Permanent volume changes can also occur if a material undergoes
an irreversible phase transformation. The GeoModel supports plastic volume changes, but
it does so without explicitly modeling the underlying microphysical mechanisms. None
theless, the GeoModel does reflect the influence of micromechanical theory by phenome
nologically incorporating plastic volume changes observed in hydrostatic loading. To
motivate the GeoModel’s cap theory, we will explain the equations and their qualitative
features in the context of porosity, but keep in mind that any other microphysical compac
tive mechanisms are equally well accommodated by the phenomenological cap model.
The cap function accounts for the presence of pores in a material by controlling
where the yield function will intersect the axis in compression. This intersection point
corresponds to and, because we are considering compressive states, we will
denote the value of at the intersection point by , where is the pressure
(positive in compression) at which inelastic deformation commences in purely hydrostatic
loading for a given level of porosity. As voids compress out, the value of will change,
as explained later when we discuss the evolution equations for the GeoModel’s internal
state variable . Porosity also degrades material shear strength because, recalling
Eq. (4.28), the cap function effectively reduces the nonporous yield strength, defined pre
viously by the fracture function .
The GeoModel employs a cap function
*
defined
(4.30)
* evaluated in the code by a Pelessone [34] function, .
f
c
I
1
J
2
0 =
I
1
I
1
X = X 3 ⁄
X
κ
f
f
I
1
κ
X
f
c
1
Figure 4.5. GeoModel cap function
f
c
2
I
1
κ , ( ) 1
I
1
κ – ( ) I
1
κ – I
1
κ – ( ) + ( )
2 X κ – ( )
2
 – =
f
c
2
I
1
κ , ( )
1 if I
1
κ <
1
I
1
κ –
X κ –

\ .
 
2
– otherwise.
¹
¦
´
¦
¦
=
GeoModel theory
37
Neither nor are usersupplied material parameters. Instead, these variables are com
puted internally within the GeoModel code by enforcing consistency with more intuitive
usersupplied parameters obtained from hydrostatic testing (see page 55).
The equation of the elliptical portion of the cap curve is
. (4.31)
The intersection point will be later related to the value of so that knowledge of the
internal state variable will be sufficient to compute a value for . For now, while
describing the geometry of the yield surface, both and should be regarded as internal
state variables whose values are computed internally in the GeoModel using evolution
equations discussed later. Rather than using directly, recall that the GeoModel uses the
function that is simply the square of given in Eq. (4.30):
. (4.32)
Elasticplastic coupling. The cap model is used when the material being studied
contains enough porosity (or highly compliant second phase inclusions) so that inelastic
volume reduction become possible through irreversible reduction of void space. Intu
itively, one might expect the elastic moduli to stiffen as voids collapse, but the material
might actually become more elastically compliant as shown in Fig. 8.4 (a phenomenon
that might be explained, for example, by rubblization of a ligament network). Regardless
of its microphysical origins, the elastic moduli are permitted to vary with plastic strain by
generalizing the nonlinear elastic moduli expressions in Eqs. (4.9) and (4.10) to
(4.33)
. (4.34)
In the absence of joints, the scale factors and equal 1.0; otherwise, they are com
puted internally within the GeoModel as described in a separate sequel report. In the
above equations, is the equivalent plastic shear strain (which, for proportional load
ing, is conjugate to the equivalent shear stress, ), and is the plastic compaction vol
ume change. Mathematically,
(4.35)
. (4.36)
Though defined mathematically as stated, the GeoModel computes the plastic volume
change, indirectly, as explained later in the context of Eq. (4.67).
κ X
f
c
2
I
1
κ –
X κ –

\ .
 
2
+ 1 =
X κ
κ X
κ X
f
c
F
c
f
c
F
c
f
c
2
=
K f
K
b
o
b
1
b
2
I
1
 –
\ .
 
exp + b
3
Exp
b
4
ε
v
p
 –
\ .

 
–
¹ )
´ `
¦ ¹
=
G f
G
g
o
1 g
1
exp g
2
J
2
1 2 /
– ( ) –
1 g
1
–
 g
3
Exp
g
4
γ
eqiv
p
 –
\ .
 
–
¹ )
´ `
¦ ¹
=
f
G
f
K
γ
eqiv
p
J
2
ε
v
p
γ
eqiv
p
2
γ
˜
·
˜
p
t d
∫
=
ε
v
p
tr
ε
˜
·
˜
p
t d
∫
=
ε
v
p
GeoModel theory
38
Meridional shear limiter function, F
f
. In a loose sense, the previous subsection
described the GeoModel’s cap function by considering a material that contained pores,
but no cracks. In this subsection, we describe the GeoModel’s fracture function by
considering a material that contains microcracks but no porosity. Numerous microphysical
analyses (as well as a preponderance of data) suggest that, for microcracked media, the
onset of “yield” depends on all three stress invariants, which implies that the yield func
tion for microcracked media must depend on all three cylindrical Lode coordinates. The
GeoModel supports this singularly common prediction of microscale damage theory.
Though they differ in specific details, microphysical damage theories and laboratory
observations for brittle materials also tend to share the following qualitative features:
• At a given mean stress, yield in triaxial extension (TXE) occurs at a lower stress than
in triaxial compression (TXC), which implies that octahedral yield profiles are
generally triangular (or distorted hexagon) in shape, with the triangle apex being
located on TXC axes, as sketched in Fig. 1.0(c) on page 1.
• Brittle materials are very weak in tension. This implies that the meridional yield
profile will include few if any tensile stress states. Brittle materials are also vulnerable
to shear cracking at low pressures, but they become able to support increasingly large
shear stresses as pressure is increased because friction at crack faces helps reduce the
shear load that must be suffered by the matrix material itself. Thus, in the absence of
porosity, the meridional profile is expected to monotonically increase with pressure.
• When microscale theories regard brittle crack fracture to be the only failure
mechanism, they predict that the material strength (i.e., the Lode radius at failure) will
increase monotonically with increasing pressure, so that the meridional profile
expands in an everexpanding conelike shape like the limit surface in Fig. 1.1.
Microphysical idealizations such as MohrCoulomb theory predict the meridional
profile is a straight line whose slope is directly related to the friction coefficient.
When microscale theories allow for both crack growth and ductility of the matrix
material, they predict that the increase in strength from friction will continue only
until ductile yield (at extraordinarily high pressures) becomes more likely; such
theories correspond to a meridional yield profile that is conelike at low to moderate
pressures but asymptotes to a zero slope (like a Von Mises cylinder) at high pressures.
Given the wide variety of microscale predictions for the meridional profile, the GeoModel
is equipped with a fourparameter exponential spline that is capable of replicating any of
these microphysical idealized theories, as well as actual observed material yield and rup
ture response at low and moderate pressures (i.e., at pressures well below the cap elastic
limit) so that observed data primarily reflect microcrack damage rather than combined
cracking with pore collapse (covered elsewhere in this report).
f
c
f
f
GeoModel theory
39
In the meridional plane (i.e., at a given value of the Lode angle), the yield surface
characterizes the transition boundary for inelastic flow. Stress states that were, at one time,
outside the yield surface might become realizable through hardening evolution of the yield
surface. However, the allowable amount of hardening is not unbounded. At some point,
the material will fail catastrophically (i.e., rupture). Often, the stress at rupture is smaller
than the peak stress. Stressstrain curves might or might not exhibit postpeak softening,
depending on whether or not the experiment is stresscontrolled or straincontrolled.
The peak stress (not the stress at rupture) defines the stresscarrying limit of the mate
rial. As first mentioned on page 5, the limit surface is the boundary of all stress states
that the material is capable of supporting. Many of these achievable stress states can be
reached only through inelastic processes. Appendix A (step 4 on page A5) describes in
detail how to determine the limit surface from experimental data. Mathematically, the
limit surface is characterized by a limit function that is similar in form to the yield func
tion. Specifically, the limit surface is defined by , where
. (4.37)
The limit function depends only on , not on any internal state variables. A yield
function , on the other hand, depends on the backstress tensor and on the sca
lar internal state variable . Unlike a yield surface, which can evolve over time because it
depends on timevarying internal state variables, the limit surface is fixed in stress space.
The yield function is presumed to share some qualitative features with the shear limit sur
face, but depends additionally on internal state variables as follows:
, where . (4.38)
When examining experimental data, it is generally easier to determine the maximum
limit point than the point at which plasticity first begins. Consequently, the GeoModel
provides an empirical fitting function for the limit surface in the meridional plane, and
the initial meridional yield surface is simply , reduced perhaps by a cap function
if the material initially contains voids. Comparing Eq. (4.38) with (4.37) reveals that the
yield surface inherits its octahedral profile shape (i.e., its Lode angle dependence) from
the limit surface. The size of the yield octahedral profile is generally smaller than the limit
surface profile because of the multiplier (cap) function , which represents the effect of
porosity. The yield surface origin is also offset from the limit surface origin by an amount
governed by the kinematic hardening backstress tensor . The limit surface always has a
[111] symmetry axis passing through the actual (not kinematically shifted) origin in stress
space. When kinematic hardening is enabled, the yield surface has a symmetry axis paral
lel to [111] that is offset so that it does not pass through the origin.
F σ
˜
˜
( ) 0 =
F σ
˜
˜
( ) J
2
σ
˜
˜
( )
F
f
2
I
1
σ
˜
˜
( )  
Γ θ σ
˜
˜
( )  
 – =
F σ
˜
˜
( ) σ
˜
˜
f σ
˜
˜
α
˜
˜
κ , , ( ) α
˜
˜
κ
f σ
˜
˜
α
˜
˜
κ , , ( ) J
2
ξ
˜
˜
( )
F
f
I
1
σ
˜
˜
( )   N – { }
2
F
c
I
1
σ
˜
˜
( ) κ , ( )
Γ θ
ξ
˜
˜
( )  
 – =
ξ
˜
˜
S
˜
˜
α
˜
˜
– ≡
F
f
F
f
N – F
c
F
c
α
˜
˜
GeoModel theory
40
When advancing the solution, the
initial meridional profile can harden
kinematically and/or isotropically. The
equations governing yield surface evo
lution are designed to permit only a
limited amount of hardening. As
sketched in Fig. 4.6, the initial yield
surface is permitted to translate upward
in the meridional plane by no more than
a userspecified limit .
When the yield surface has reached
the limit surface and when the stress
itself lies on the limit surface, the mate
rial will begin to soften. At that point, a
constitutivelevel description of mate
rial response no longer remains possi
ble; the host code must intervene by
inserting void or by invoking special elements capable of supporting displacement discon
tinuities. The limit surface marks the point at which a continuum material model is inade
quate to characterize macroscale material response because softening localization
becomes possible. Before reaching the limit surface, material response is handled entirely
by the GeoModel.
The shifted shear limiter function
defines how the shear stress
at yield varies with pressure for a non
porous, but microcracked, material in
its initial (virgin) state. When this
microcrack yield function is combined
with the cap function the
actual shear stress at yield is further
reduced because porosity makes inelas
ticity possible even for purely hydro
static compression (i.e., loading along
the axis). The GeoModel evolves
this combined porous yield function in
such at way that the yield surface grows
up to the limit surface. The GeoModel
does not handle material response after
reaching the limit surface. Properly, upon receiving a “limitarrival” signal, the host code
must, at this time, initiate scaledependent softening localization through the use of void
I
1
lim
it
s
u
r
f
a
c
e
,
Ff
(
I 1
)
i
n
i
t
i
a
l
J
2
Figure 4.6. Kinematic hardening. The user defines
parameters for the ultimate shear limit
surface. The initial yield surface is identical except
shifted down by a user specified amount . Kinematic
hardening allows the initial yield surface to translate
until reaching the ultimate failure surface (at which
point, the host code must initiate “element death” or
perhaps some other strategy for supporting macroscale
softening.
a
1
a
2
a
3
a
4
, , , ( )
N
N
y
ie
ld
s
u
r
f
a
c
e
,
f
f
=
Ff
(
I 1
)

N
n
o
n
p
o
r
o
u
s
N
I
1
lim
it
s
u
r
f
a
c
e
J
2
Figure 4.7. Yield surface evolution with both micro
cracking and porosity. For real materials, that contain
both cracks and voids, it is difficult to identify a single
envelope for the combined porous/cracked fracture
yield function . However, ultimate failure data can be
more readily mapped out.
f
f
N
F
f
I
1
( ) N –
F
c
I
1
κ , ( )
I
1
GeoModel theory
41
insertion or special elements that support displacement discontinuities. If the host code
fails to initiate a softening algorithm, the GeoModel will treat the limit surface as if it were
a nonhardening yield surface (making its postpeak predictions robust, but undoubtedly
inaccurate).
The GeoModel supports modeling microcracked material by providing flexible fitting
functions that can reproduce octahedral and meridional yield profiles observed for real
materials. In particular, the shear limit function used in the GeoModel is of the form
, where shear limit surface in TXC. (4.39)
where the are userspecified material parameters determined from experimental data as
explained in Appendix A. The initial (nonporous) meridional yield profile is
initial yield surface (nonporous). (4.40)
where is the userspecified shift factor. Therefore, is the zero pressure intercept
of the nonporous meridional yield surface on the axis. Frequently, is taken to be
zero. The shear limiter function, Eq. (4.39), asymptotes to a linear envelope, as indicated
in Fig. 4.8. To force the material to obey a VonMises type yield response at extremely
high pressures, the slope coefficient is merely set to zero.
Let us now explain why the shear limiter function has the general shape depicted in
Fig. 4.8. Then we will list constraints on the model parameters necessary to achieve this
shape. Brittle materials fail at very low shear stresses when the pressure is low, but they
are able to sustain higher levels of shear stress without failing if loaded under higher con
fining pressures. Consequently, the shear limiter function is expected to increase monoton
ically with pressure. Or, since is proportional to the pressure, is expected to
increase monotonically with . Furthermore, a fundamental tenant from plasticity theory
is that the yield function must be semiconvex, which implies that the second partial deriv
ative of must be negative or zero. When we speak of the “shear” stress at failure,
I
1
J
2
F
f
I
1
( ) N – =
a
1
N –
a
3
I
1
J
2
F
f
I
1
( ) =
a
1
a
3
Figure 4.8. Shear limiter function (unshifted and shifted). The shifted function should be regarded
as a nominal shape of the yield surface in the meridional plane, although porosity further lowers and
distorts the meridional yield profile by multiplying the shifted shear limit function by . F
c
shear limit surface
initial yield surface
a
4
a
4
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– a
4
I
1
+ = I
1
I
1
– =
a
k
f
f
I
1
( ) F
f
I
1
( ) N – =
N a
1
N –
J
2
ξ
N
a
4
I
1
F
f
I
1
F
f
I
1
( )
GeoModel theory
42
we are effectively speaking of the value of the Lode radius at failure corresponding to the
Lode angle for the stress state. For any cylindrical coordinate system — including the
Lode system — the radius must always be nonnegative and therefore is defined only
over the domain for which it yields positive values. Finally, in its virgin state, any material
should be unfailed at zero stress, which means that the origin must fall below the meridi
onal yield profile. All of these physical considerations lead to the following constraints on
allowable values for the parameters:
unloaded virgin material must be below yield (4.41)
nonnegative slope at low pressures (4.42)
positive Lode radius (4.43)
convexity condition (4.44)
. nonnegative slope at high pressures (4.45)
Specific values for these model parameters are determined from triaxial test data, as
explained in Fig. 3.3 on page 16 (and in Figs A.4 and A.5 of Appendix A). Sample fits of
the GeoModel’s shear limit function to data can be found in Fig. 8.1 on page 96 and
Fig. 8.5 on page 99.
The complete GeoModel yield function.
Equation (4.28) is the yield criterion. The yield function must be negative for all
elastic states (inside the yield surface), zero for all stress states satisfying the yield crite
rion (on the yield surface), and positive for all stress states (outside the yield surface) that
cannot be reached except through an inelastic process — if at all. For computational rea
sons, the GeoModel’s yield function is based on the square of Eq. (4.28):
. (4.46)
When kinematic hardening is used, the stress invariants, and , are those for the kine
matically shifted stress tensor, . Otherwise, when kinematic hardening is dis
abled, these are simply the stress invariants. Of course, is the first invariant of the
stress tensor . (Since backstress is deviatoric, ).
The building block functions and are implemented in the GeoModel in a slightly
altered form by being expressed in terms of the shear limit function and an alternative
(computationally more efficient) cap function :
(4.47)
. (4.48)
The first of these equations allows the user to specify a maximum amount, , that the
yield function is permitted to translate under kinematic hardening. Thus, the function
can be regarded as a “limit” or “softening” envelope, beyond which stresses can never be
reached quasistatically (not even via hardening). The second equation recasts the func
tion as the square root of a different function for computational reasons. The second
F
f
a
1
a
3
– N – 0 ≥
a
2
a
3
a
4
+ 0 ≥
a
2
0 >
a
3
0 ≥
a
4
0 ≥
f
f Γ
2
θ
ξ
( )J
2
ξ
f
f
2
I
1
( )f
c
2
I
1
κ , ( ) – =
θ
ξ
J
2
ξ
ξ
˜
˜
S
˜
˜
α
˜
˜
– ≡
I
1
I
1
σ
σ
˜
˜
α
˜
˜
I
1
ξ
0 =
f
f
f
c
F
f
F
c
f
f
I
1
( ) F
f
I
1
( ) N – =
f
c
I
1
κ , ( ) F
c
I
1
κ , ( ) =
N
F
f
f
c
F
c
GeoModel theory
43
equation also shows explicitly the presence of the internal state variable related to the
isotropic hardening part of the GeoModel associated with void collapse. The internal state
variable marks the branch point where combined porous/cracked yield surface deviates
from the nonporous yield surface. As explained on page 56, this branch point is deter
mined internally within the GeoModel in a manner that ensures consistency with mea
sured hydrostatic data.
In terms of the new building block functions, the critical Lode radius in triaxial com
pression (TXC), where , may be expressed as a function of the Lode axial coor
dinate as
. (4.49)
Plotting vs. (at a given value of ) will produce a geometrically accurate visual
ization of the meridional yield profile. Often we instead plot vs. to label the axes
with more broadly recognized stress measures, but in doing so we are actually showing a
geometrically distorted view of the yield profile, as explained on page 23.
Substituting Eqs. (4.47) and (4.48) into (4.46) gives the yield criterion cited at the
beginning of this section [Eq. 4.18].
The “J3TYPE” Lodeangle function, Γ. This section describes available func
tional forms for the Lode angle dependence function . This function controls the
shape of the octahedral yield profile. Since this function controls only the shape, not size,
of the octahedral profile, its magnitude is inconsequential. The function is normalized
to equal unity in TXC ( ).
*
At other Lode angles, . Thus, since
appears in the yield function as a multiplier of , it acts as a pseudo stress raiser, causing
yield to occur at smaller values of at Lode angles differing from the fiducial (TXC)
angle where . To ensure convexity of the octahedral yield profile, the Lode
angle function must satisfy
. (4.50)
A hallmark trait of rocks and rocklike materials (concrete, ceramics, etc.) is a higher
strength in triaxial compression than in triaxial tension at any given mean pressure.
Loosely speaking, this characteristic results from friction at crack faces being able to carry
a larger portion of the load under compression, therefore sparing the surrounding matrix
material from having to carry the entire resolved shear stress at crack tips. Classical Mohr
Coulomb theory, which is supported by the GeoModel primarily for comparisons with ide
* With this normalization, the meridional function then quantifies the pressurevarying size of
octahedral profiles
κ
κ
Γ θ ( ) 1 =
z
r
TXC
F
f
3 z ( ) N – ( ) F
c
3 z κ , ( )
2
 =
r
TXC
z κ
J
2
I
1
Γ θ ( )
Γ
θ +30° = Γ θ ( ) 1 > Γ θ ( )
f
f
J
2
ξ
J
2
ξ
Γ θ ( ) 1 =
Γ′′ θ ( ) Γ θ ( ) + 0 ≥
GeoModel theory
44
alized analytical solutions, has an octahedral profile in the shape of a distorted hexagon,
causing considerable computational difficulties when dealing with the vertices. The com
putational attractiveness of removing yield surface corners has motivated numerous pro
posals of smoothed threeinvariant models for frictional materials [29,28,3,47,7], and
Lade [30] was among the first efforts to additionally include curvature in the meridional
plane. According to Borja, et. al. [6], there is evidence that smoothed yield surfaces cap
ture mechanical response more accurately than vertex models, but these authors point to
no data to back up this claim. The GeoModel presently supports three yieldtype options
(specified by a value of 1, 2, or 3 for the user parameter, J3TYPE):
1.Gudehus (an efficient smoothed profile, with restrictions on convexity)
2.WillamWarnke (a relatively inefficient smooth profile with no convexity constraints)
3.MohrCoulomb (distorted hexagon polygon)
The Gudehus and WillamWarnke options both correspond to fully differentiable yield
functions (no vertices). The MohrCoulomb option (which is available principally for
comparisons with analytical solutions) is differentiable everywhere except at triaxial states
where yield surface vertices require special numerical handling.
*
Recognizing logistical constraints of most laboratories, the GeoModel presumes that
experimental data are available at most only for a limited number of canonical loading
paths: perhaps triaxial extension ( ), perhaps simple shear ( ), and almost cer
tainly triaxial compression ( ). Regardless of which yieldtype (Gudehus, Willam
Warnke, or MohrCoulomb) is selected, the shape of the octahedral yield profile is
described, in part, by user specification of a parameter , equal to the triaxial exten
sion/compression (TXE/TXC) strength ratio at a given pressure. The GeoModel
presumes that only the size of the octahedral yield profile — not its shape — varies with
pressure. Consequently, the strength ratio equals its userspecified value at all pressures
and throughout the entirety of the simulation (i.e., is a constant, not a time varying
internal state variable). Appendix A gives instructions for inferring a value of from
experimental data.
* The GeoModel averages directions on either side of the vertex if the strain rate points within the
limiting (Koiter) fan of unit normals. If the strain rate points within a sextant of the octahedral
plane, then the normal in that sextant is used.
θ= π 6 ⁄ – θ=0
θ=π 6 ⁄
ψ
ψ
ψ
ψ
GeoModel theory
45
Precise functional forms of available Lode angle functions are given below:
1. Gudehus: .
To satisfy the convexity requirement of Eq. (4.50), the strength ratio must satisfy .
2. WillamWarnke: ,
where . The WillamWarnke option is convex for .
3. MohrCoulomb: . Here, the internal friction
angle φ is the angle of the failure envelope in the Mohrdiagram ( , where
is the coefficient of friction). Within the GeoModel, is determined from the user
supplied strength ratio by . The MohrCoulomb option is convex for .
These three options are distinguished by how the octahedral yield profile varies in stress space in
the transition from TXE to simple shear to TXC at a fixed pressure. Graphs of the octahedral yield
profile corresponding to any of the above options may be constructed by parametrically plotting
(4.51)
Here, , where . (4.52)
The angle varies from 0 to , and therefore varies between and .
Γ θ ( )
1
2
 1 3θ sin
1
ψ
 1 3θ sin – ( ) + + =
7
9
 ψ
9
7
 < <
Γ θ ( )
4 1 ψ
2
– ( )cos
2
α
∗
2ψ 1 – ( )
2
+
2 1 ψ
2
– ( ) α
∗
cos 2ψ 1 – ( ) 4 1 ψ
2
– ( )cos
2
α
∗
5ψ
2
4ψ – + +
 =
α
∗
π
6
 θ + =
1
2
 ψ 2 ≤ ≤
Γ θ ( )
2 3
3 φ sin –
 θ cos
φ sin θ sin
3
 –
\ .
 
=
φ tan µ = µ
φ
φ sin 3
1 ψ – ( )
1 ψ +
 =
1
2
 ψ 2 ≤ ≤
x r Θ cos = y r Θ sin =
r
2
Γ θ ( )
 = θ
1
3
ArcSin 3Θ sin ( ) =
Θ 360° θ 30° – +30°
Figure 4.9. Octahedral yield profiles, plotted at allowable values of the strength ratio. The compari
son plot corresponds to a strength ratio of ψ=0.8. The function is defined so that all models return a
value of 1 in triaxial compression.
Γ θ ( )
J3TYPE=1
WillamWarnke MohrCoulomb COMPARISON Gudehus
J3TYPE=2 J3TYPE=3
GeoModel theory
46
As illustrated in Fig. 4.9, the MohrCoulomb model (J3TYPE=3) interpolates linearly
in octahedral stress space, resulting in a distorted hexagon if , a perfect hexa
gon (Tresca) if , and a triangle if . For a detailed explanation of the Mohr
Coulomb formulation, see Appendix B (page B22). The MohrCoulomb model has yield
surface vertices at , which results in extra computational effort to determine
plastic strain rates in triaxial states. The Gudehus and WillamWarnke options, on the
other hand, involve no yield surface vertices, which speeds up computations. The Gude
hus option is the default because of its computational simplicity, but it supports only a lim
ited range of TXE/TXC strength ratios, the WillamWarnke option should be used if a
rounded but strongly triangular octahedral yield profile is desired. Appendix A (STEP 6)
provides guidance for selecting the Lode angle option most appropriate for matching
experimental data.
The GeoModel subsumes many simpler (classical) models as special cases. For exam
ple, if failure is hypothesized to occur when the largest principal stress (or strain) reaches a
critical value, then the octahedral yield profile will be a triangle. If, on the other hand, fail
ure is presumed when the equivalent shear stress reaches a critical value, independent of
the Lode angle (like a Von Mises or DruckerPrager criterion), then the octahedral yield
profile is a circle, which can be modeled with the Gudehus option. As a rule, any classical
failure criterion that is expressed directly in terms of the principal stresses will imply an
appropriate J3TYPE option (and an appropriate value for the TXE/TXC ratio), but such
criteria will also imply functional constraints on the meridional failure function as well,
which requires appropriate GeoModel inputs to mimic. Simplified GeoModel input sets,
corresponding to these classical specialcase idealized theories may be found at the end of
Appendix B.
Octahedral profile plots like the ones shown in Fig. 4.9 are most illuminating from a
qualitative perspective. However, for parameterizing the GeoModel to quantitative labora
tory data, simple plots of vs. are more useful. Fig. 4.10 shows the functions for
each of the J3TYPE options. In all cases, the lower bound on is 1.0 at TXC.
1 2 ⁄ ψ 1 < <
ψ=1 ψ 1 2 ⁄ =
θ π 6 ⁄ ± =
f
f
Γ θ Γ θ ( )
Γ
0.4 0.2 0.2 0.4
1.05
1.1
1.15
1.2
1.25
0.4 0.2 0.2 0.4
1.2
1.4
1.6
1.8
2
0.4 0.2 0.2 0.4
1.2
1.4
1.6
1.8
2
TXE SHR TXC
0
π
6
 –
π
6

Γ θ ( ) 1 –
TXE SHR TXC
0
π
6
 –
π
6

Γ θ ( ) 1 –
TXE SHR TXC
0
π
6
 –
π
6

Γ θ ( ) 1 –
Figure 4.10. Lode angle function (for various ψ strength ratios) plotted vs. the Lode angle varying from
–30 degrees (TXE) to +30 degrees (TXC). The larger the value of , the smaller the radial distance to the
octahedral yield profile and therefore the smaller the shear failure strength. The Gudehus and Willam
Warnke options both predict lowest strength (largest value of ) at TXE, whereas MohrCoulomb theory
predicts lowest strength at an intermediate Lode angle somewhere between TXE and TXC. For example,
when , MohrCoulomb theory reduces to Tresca theory and the lowest strength (highest ) occurs
at the zero Lode angle (pure shear).
Γ
Γ
ψ 1 = Γ
GeoModel theory
47
Advancing the solution (groundwork discussion)
So far, we have discussed how some microphysically based, but generally oversimpli
fied theories can be used to predict theoretical shapes of yield or failure surfaces. The
GeoModel implicitly captures microscale phenomena by using macroscale measurable
variables in phenomenological manner. Direct use of idealized theories would require ini
tializing and evolving microscale quantities (such as porosity) that are impractical to mea
sure in the laboratory. Therefore, the algebraic structure of functions used in the
GeoModel is guided by idealized microscale theories, but recast in terms of directly mea
surable macroscopic variables. Simplified failure criteria help guide choices for interpola
tion functions to be fitted to real observed data that likely reflect the specific phenomena
considered in microscale idealizations and possibly some other “unknown” sources of
inelastic flow.
Microphysical theories are also used to guide how the GeoModel treats the partition
ing of inelastic flow, once it begins. For example, most theories of inelastic flow (includ
ing the GeoModel) presume that the total strain rate can be partitioned additively as
, (4.53)
where represents the elastic (or recoverable) part of the strain rate and denotes the
“plastic” part of the strain rate. More correctly, represents the inelastic strain rate,
which reflects contributions from any and all sources of inelastic material response. Many
classical theories presume that the direction of the plastic strain rate is parallel to the nor
mal to the yield surface. In this case, since the normal to the yield surface can be obtained
by the gradient of the yield function , the plastic part of the strain rate is pre
sumed to be of the form
, (4.54)
where is a multiplier (called the consistency parameter) determined by demanding
that the stress must remain on the yield surface during inelastic loading. The subscripts on
the partial derivative merely indicate that the internal state variables are held constant.
When the plastic strain rate direction is determined from the stress gradient of the yield
function, as shown here, the model is said to be “associative” (to indicate that the plastic
strain rate is associated with the yield function*).
* For materials that exhibit elasticplastic coupling, the terms “associativity” and “normality” can
have distinct meanings, depending on whether the portion of the total strain rate attributable to rates
of elastic moduli is absorbed into the elastic strain rate or the inelastic strain rate. When the cou
pling terms (from rates of elastic moduli) are incorporated into the inelastic strain rate, normality
and associativity are not interchangeable terms. If the coupling terms are incorporated into the elas
tic strain rate, then associativity and normality are interchangeable, but at the cost that the elastic
strain rate ceases to be an exact differential with respect to deformation.
ε
˜
·
˜
ε
˜
·
˜
ε
˜
·
˜
e
ε
˜
·
˜
p
+ =
ε
˜
·
˜
e
ε
˜
·
˜
p
ε
˜
·
˜
p
f σ
˜
˜
α
˜
˜
κ , , ( )
ε
·
ij
p
λ
· ∂f
∂σ
ij

\ .
 
α
˜
˜
κ ,
=
λ
·
GeoModel theory
48
While the GeoModel does support associativity at user request, many researchers
report that normality tends to overpredict the amount of volumetric plastic strain [43].
Therefore, nonnormality is supported in the GeoModel as well. For nonnormality the
user specifies a flow function such that
*
. (4.55)
The plastic strain rate includes both deviatoric and isotropic parts. If the flow potential
depends on the first invariant , then applying Eq. (3.21), the volumetric plastic strain
rate is
. (4.56)
The plastic strain rate points normal to the isosurface .
†
If the flow function is
associative, then the plastic potential function is identical to the yield function and the
plastic strain rate will therefore point normal to the yield surface. Flow surface vertices
reside at points where the flow potential is nondifferentiable, in which case the plastic
strain rate points within a “cone of limiting normals” (Koiter fan) at the vertex and is
determined through additionally considering the trial elastic stress rate associated with the
total strain rate.
In the GeoModel, the functional form of is the same as that of , but with different
values for material constants. Specifically, the flow potential can be made to differ from
the yield function by assigning values to , , , and that differ from their
counterpart parameters used to define the yield surface. For associativity,
the potential function parameters should be given values identical to their counterparts in
the yield function.
If continuing to apply elasticity theory would result in a predicted stress lying outside
the yield surface, the governing equations are no longer elastic. At this point, the strain
rate is decomposed into two parts, elastic plus plastic, as mentioned in Eq. (4.53). The
stress rate is determined by applying elasticity. That is,
, (4.57)
* While considerable data does exist to suggest that the inelastic strain rate is not directed normal to
the yield surface for some materials, such behavior is not well understood. The mathematical valid
ity of assuming existence of a nonassociated flow potential function has been called into question
by Sandler and Pucik [40,37], who have demonstrated that such a model is inherently unstable,
inadmissibly generating unbounded energy from quiescent states.
† Because the current stress might not reside on the isosurface , the GeoModel projects the
stress to the nearest point on this isosurface. The need for such revisions is rarely recognized in
plasticity programs (and casts doubt on the very notion flow potentials).
φ σ
˜
˜
α
˜
˜
κ , , ( )
ε
·
ij
p
λ
· ∂φ
∂σ
ij

\ .
 
α
˜
˜
κ ,
=
ε
·
ij
p
I
1
ε
·
v
p
trε
˜
·
˜
p
3λ
· ∂φ
∂I
1
 = =
φ 0 =
φ 0 =
φ f
φ
a
2
PF
a
4
PF
R
PF
ψ
PF
a
2
a
4
R ψ , , , ( )
σ
·
ij
C
ijkl
ε
·
ij
e
C
ijkl
ε
·
ij
ε
·
ij
p
– ( ) = =
GeoModel theory
49
where denotes the isotropic tangent elastic stiffness tensor. In Eq. (4.57), our goal is
to compute the stress rate. The current state is known, and therefore the instantaneous
stress state and elastic moduli are known. In numerical implementations of constitutive
models, the strain rate is known (it is provided by the host code after solution of the
momentum equation). Thus, the only unknown in this equation is the plastic strain rate.
Equation (4.55) allows us to compute the direction of the plastic strain rate from the
known instantaneous stress state. Thus, after substitution of Eq. (4.55) into (4.57), the
stress rate can be written
. (4.58)
Everything on the righthandside of this equation is known except the value of the consis
tency parameter, .
The consistency parameter is obtained by demanding that, not only must the stress be
on the yield surface during plastic loading , it must also remain on the yield sur
face throughout a plastic loading interval. Thus, during plastic loading. The yield
function depends on the stress, but it also depends on the isotropic hardening internal
state variable, and (if applicable) on the kinematic hardening backstress state variable
tensor . Thus, the assertion that can be written via the chain rule as
. (4.59)
The first term may be simplified through application of Eq. (3.20). The last two terms
reflect the fact that the yield surface can evolve in shape and translate in stress space dur
ing inelastic loading. In what follows, we will present “evolution equations” that govern
how the state variables change in response to plastic flow. It will be argued that the evolu
tion of each internal state variable should be proportional to the plastic strain rate. Equiva
lently, these rates must be proportional to our unknown plastic consistency parameter .
By substituting Eq. (4.58) and these soontobederived evolution equations for the inter
nal state variables into Eq. (4.59) we will be able to solve Eq. (4.59) for the consistency
parameter . Once the consistency parameter is known, it can be substituted into
Eq. (4.58) to obtain the stress rate, which may then be integrated numerically to update the
stress. With the consistency parameter known, then rates of internal state variables
(ISVs) become known through their evolution equations, allowing the ISVs themselves to
be updated to the end of the timestep. Thus, the key to advancing the solution is to now
derive in detail the internal state variable evolution equations.
C
ijkl
σ
·
ij
C
ijkl
ε
·
ij
e
C
ijkl
ε
·
ij
λ
· ∂φ
∂σ
ij

\ .
 
α
˜
˜
κ ,
– = =
λ
·
f 0 = ( )
f
·
0 =
f
κ
α
ij
f
·
0 =
f
· ∂f
∂σ
ij
σ
·
ij
∂f
∂κ
κ
· ∂f
∂α
ij
α
·
ij
+ + 0 = =
λ
·
λ
·
λ
·
GeoModel theory
50
Evolution equations
Equation (4.59) may be used to determine the plastic consistency parameter if hard
ening evolution laws can be found for which the rate of each internal state variable ( and
) is proportional to . Once the plastic consistency parameter is known, the evolution
laws may be integrated through time to model the time varying hardening evolution of the
yield surface.
As indicated in Fig. 4.11, isotropic hardening (governed by and related to void col
lapse) causes a change in size of octahedral yield profiles, while kinematic hardening
(governed by the backstress tensor ) produces a translation of all octahedral yield pro
files. In general, both types of hardening can occur simultaneously.
In this section, we will derive explicit expressions for an isotropic hardening modulus
and a kinematic hardening tensor, , such that the evolution of the internal state vari
ables may be written in the forms
(4.60)
and
. (4.61)
Later, substituting these expressions into Eq. (4.59) will lead to an expression for the plas
tic consistency parameter . Once the consistency parameter is known, the above equa
tions can be themselves integrated through time to update and .
Evolution equation for the porosityrelated internal state variable, κ.
We begin this section with some background discussion about the meaning of the internal
state variable , connecting it to some classical microphysical theories for purely porous
(noncracked) materials. The GeoModel’s reinterpretation of for both porous and
cracked materials will lead ultimately to an evolution law of the desired form,
where is called the isotropic hardening modulus.
λ
·
κ
α
ij
λ
·
Figure 4.11. Hardening mechanisms. At a given pressure, isotropic hardening entails an increase in
size, kinematic hardening translates the yield surface, and compound hardening includes both mecha
nisms. Softening corresponds to a yield surface contraction.
Initial yield
surface
isotropic
hardening
kinematic
hardening
compound
hardening
κ
α
˜
˜
h
κ
H
˜
˜
α
κ
·
h
κ
λ
·
=
α
˜
·
˜
H
˜
˜
α
λ
·
=
λ
·
κ
α
˜
˜
κ
κ
κ
·
h
κ
λ
·
=
h
κ
GeoModel theory
51
In the GeoModel, void collapse commences at different pressures depending on the
amount of shear stress present. The effect of shear stress on void collapse is characterized
by the cap function illustrate in Fig. 4.5. Even though void collapse depends on shear
stress, characterizing this effect requires only specification of two numbers and on
the hydrostat [see Fig. 4.5]. In the GeoModel, and are presumed to be interrelated so
that knowledge of is sufficient to compute the value of . We will discuss this relation
ship later. For now, we will focus on how the hydrostat intercept should vary as porosity
is reduced.
If a material is capable of permanent volume change (i.e., if hydrostatic testing exhib
its nonzero residual plastic volumetric strain upon releasing the pressure), then the
material likely contains voids. The hydrostat intercept is proportional to the critical
“elastic limit” pressure required to initiate irreversible void collapse. Therefore, the larger
the porosity, the smaller will be. As porosity is crushed out, the hydrostat intercept will
move to the right so that increasing pressure will be required to continue crushing out the
pores. Recognizing that is an indirect measure of porosity changes, our first goal is to
describe how the relationship between and can be inferred from hydrostatic test
data. Then we will discuss the relationship between and . With these two relationships
in hand, we will ultimately assert that
, (4.62)
from which substitution of Eq. (4.56), during hydrostatic compression or compaction
dominated processes, will give the evolution equation in the desired form,
, where . (4.63)
During dilatationdominated processes, a different form is used for (see Eq. 4.73).
κ X
κ X
X κ
X
ε
v
p
X
X
ε
v
p
ε
v
p
X
κ X
κ
· dκ
dX

\ .
 
dX
dε
v
p

\ .

 
ε
·
v
p
=
κ
·
h
κ
λ
·
= h
k
3
dκ
dX

\ .
 
dX
dε
v
p

\ .

 
∂φ
∂I
1
 =
h
k
GeoModel theory
52
Relationship between and (the hydrostatic crush curve)
If the matrix material for a porous medium is plastically incompressible, then it can be
shown [33] that the unloaded porosity (i.e., the innate porosity at the rest state, not the
slightly different porosity that reflects reversible elastic porosity reduction under loading)
evolves under plastic loading according to
, (4.64)
where is the void volume in a sample divided by the total volume of the sample (both
volumes are those in the unloaded state), and is the trace of the logarithmic plastic
strain rate. To second order accuracy, Eq. (4.64) implies that the change in porosity is
approximately equal to the plastic volumetric strain:
. (4.65)
Considering only hydrostatic loading, early
research on pore collapse focused on deriving and/or
experimentally measuring socalled “crush curves” in
which porosity in a material is plotted as a function of
the applied pressure, as in Fig. 4.12. The GeoModel
uses a similar curve, but inferred directly from hydro
static stressstrain data so that porosity measurements
are not necessary. By using the cap function, the Geo
Model incorporates the results from this specialized
hydrostatic experiment into the general theory in such
a manner that pore collapse will commence at lower
pressures in the presence of shear. Recall that the cap
function, loosely speaking, represents material
response in the absence of microcracks. Porousonly
theories typically predict meridional cap profiles similar to the GeoModel cap function.
For example, Gurson [20] reported the following upperbound yield criterion (expressed
in terms of Lode cylindrical coordinates):
, (4.66)
where is a constant (the yield stress of the matrix material) and is the porosity. Being
independent of the Lode angle, the Gurson yield function is a circle in the octahedral
plane; Gurson’s meridional profile is compared with the GeoModel’s cap function in
Fig. 4.13 for various porosities. As porosity goes to zero, the meridional profile
approaches the pressure insensitive VonMises profile for the matrix material (in other
words, in the absence of microcracks, the yield surface becomes a cylinder in stress space
as the porosity goes to zero). This property holds only in the absence of microcracks.
When microcracks are later included, the common envelope of yield surfaces will be the
shear fracture curve .
X ε
v
p
Π
Π
·
1 Π – ( )ε
·
v
p
=
Π
ε
·
v
p
Π Π
o
– ε
v
p
≈
P
E
Figure 4.12. A conventional porosity
crush curve (dashed) and state path
(solid). Porosity is constant until a crit
ical elastic limit pressure is reached.
Thereafter, porosity is reduced as pres
sure is increased.
P
E
Π
p
r k f
v
2
1 2f
v
z
2k

\ .
 
cosh – + =
k f
v
F
f
GeoModel theory
53
Under Gurson theory, the key material properties are microphysical (the yield stress
for the matrix material and the porosity ), which are difficult to measure directly. To
obtain analytical results, these theories must resort to over simplistic assumptions about
the matrix material and pore morphology (e.g., Gurson theory presumes perfectly spheri
cal voids arranged in a perfectly periodic array). Finally, microphysical theories are typi
cally upper bounds, which are of limited use in applications since the tightness of the
bound is unknown.
Rather than directly using models like Gurson theory, the GeoModel is guided by the
general trends they predict. Except at extremely high porosity, the Gurson model predicts
that the cap surface will be essentially flat for a large range of pressures (zcoordinates).
The loss in shear strength caused by pores is pronounced over only a small range of pres
sures near the hydrostatic limit pressure. This region (beyond which the yield surface
noticeably branches down to zero) is called the cap region. As seen in Figure 4.13, Gur
son theory predicts the yield surface will evolve with porosity in such a manner that the
cap essentially translates along the pressure (z) axis — the curvature of the cap region
does not change significantly. The GeoModel supports these general trends by using a
computationally simpler RubinSandler cap function which is simply constant until a crit
ical branch pressure is reached, after which the cap function drops to zero along an ellipse.
The cap curve evolves by simple translation along the hydrostat without changes in cap
curvature.
In the Gurson model, the current location of the translated cap is a function of the
matrix yield stress and the porosity. Rather than using essentially unknowable matrix
properties like these as internal state variables, the GeoModel recognizes that the appro
priate location for the cap can be determined directly from hydrostatic compression test
10 20
.2
.4
.6
.8
1
r
z
h
i
g
h
p
o
r
o
s
i
t
y
low
porosity
Gurson model
Figure 4.13. The Gurson theory for porous yield surfaces compared with the GeoModel cap function
at various values of the internal state variable κ. Qualitatively, the theories are similar. When microc
rack effects are included, the GeoModel profiles at various porosities form a pressuredependent enve
lope, as in Fig. 1.0(b) on page 1, instead of the horizontal VonMiseslike envelope shown here.
z
r
GeoModel with microcrack
effects suppressed
k
f
v
GeoModel theory
54
data and the branch point at which shear begins to affect pore collapse is presumed in the
GeoModel to translate with the hydrostatic limit point. Thus, characterizing how the entire
cap function evolves in response to plastic loading boils down to characterizing how the
hydrostat intercept point (i.e., where the cap intersects the zaxis) evolves.
The GeoModel presumes that it is experimentally tractable to obtain pressure vs. volu
metric strain data. If possible, the experiment should be run to the point of total pore col
lapse (as in Fig. 3.1 on page 13). The elastic response of the material must be first
determined by fitting the unloading curve to the nonlinear elasticity fitting function in
Eq. (4.9). “Copies” of the elastic unloading function may be superimposed anywhere on
the hydrostatic pressure vs. total strain data. As indicated in Fig. 4.14, the elastic unload
ing curves can be used to determine a shift distance that must be applied at any given pres
sure to remove the elastic part of the strain. After applying these shifts, the pressure vs.
total strain plot is converted to a pressure vs. plastic strain plot, called an Xfunction. The
Xfunction in Fig. 4.14 asymptotes to infinity when the plastic volume strain (i.e., the
change in porosity) has reached its maximum value corresponding to all of the pores hav
ing been crushed out. Rotating the Xcurve and shifting the origin produces a classical
crush curve in which porosity is plotted as a function of pressure. The GeoModel never
explicitly refers to porosity. Instead, the plastic volumetric strain is employed as an indi
rect measure of porosity changes.
Parameterizing the GeoModel so that it will adequately model the changes in the yield
surface resulting from pore collapse requires converting hydrostatic pressure vs. volumet
ric strain data as illustrated in Fig. 4.14 to obtain a classical porosity vs. pressure crush
curve in which the porosity is plotted as a function of the pressure . Specialized parame
s
o
3p
3
p
1
– =
ε
v
p
p
1
3

X =
ε
v
total
p
K p ( )
p
3
p
3
P
E
P
E
P
E
porosity p
3
ε
v
p
– ≈
Figure 4.14. Relationship between (a) hydrostatic pressure vs. volumetric strain data, (b) the GeoModel X
function, and (c) a traditional porosity vs. pressure crush curve. Test data are pressure vs. total volumetric
strain. Once the elastic unloading curves have been parameterized to the GeoModel fitting functions, the elastic
strain at each pressure value may be subtracted from the total strain to generate the Xfunction. This function as
ymptotes to a limit value for the plastic strain when all voids have crushed out (and plastic volume changes there
fore become negligible). The limit strain is approximately the initial porosity in the material. Rotating the Xplot
and moving the origin as shown will produce a traditional porosity vs. pressure crush curve. 90°
Hydrostatic test data
Classical crush curve
GeoModel Xfunction
p
initial slope
shift distance
copies of the
elastic unloading
curve
shift distance
shift distance
shift distance
p
GeoModel theory
55
terization software is available from the model developers to perform this conversion task
and to fit the resulting crush curve to an exponential spline (see Appendix A). The plastic
volume strain of a virgin (predeformation) material is zero. Therefore the userspecified
parameter is approximately equal to the initial porosity in the material. As pressure is
increased from zero, the crush curve [Fig. 4.14(c)] shows that the porosity remains
unchanged for a while until an elastic limit pressure is reached. Continuing to apply
increasing pressure beyond this elastic limit results in irreversible pore collapse and there
fore reduction in porosity. The GeoModel allows fitting the post yielding part of the crush
curve according to the crush curve spline formula,
, where . (4.67)
Here, , , , and are fitting constants. Referring to the porosity vs. pressure crush
curve, these parameters are interpreted physically as follows:
• equals , where is the elastic limit pressure at the initial onset of pore
collapse.
• equals , where is the initial slope of the porosity vs. pressure crush
curve (see Fig. 4.14).
• is an optional fitting parameter that may be used if a measured crush curve has an
inflection point (i.e., initially concave down, transitioning to concave up at high
pressures).
• is the maximum achievable plastic volume strain, which corresponds
approximately to the initial porosity in the material. Complete crushing out of all
pores in the material is recognized in the hydrostatic pressure vs. total strain curve if
the elastic release curve is tangent to the loading curve.
Equation (4.67) coincides with hydrostatic test data only during pore collapse. During the
initial elastic loading, Eq. 4.67 describes the dashed line in Fig. 4.14, which is not required
to coincide with the data. At pressures above the elastic limit ( ), Eq. 4.67 may be
used to compute the plastic volumetric strain according to
, where (4.68)
This relationship may be differentiated to obtain the derivative needed in
Eq. (4.63). Now all we need to compute the isotropic hardening modulus is the rela
tionship between and so that we can substitute the derivative into Eq. (4.63).
p
3
P
E
p
3
ε
v
p
– p
3
e
p
1
p
2
ξ + ( )ξ –
= ξ X p
0
– 3 p P
E
– ( ) = =
p
0
p
1
p
2
p
3
p
0
3P
E
– P
E
p
1
s
o
3p
3
( ) ⁄ s
o
p
2
p
3
ξ 0 >
ε
v
p
p
3
1 e
p
1
p
2
ξ + ( )ξ –
–   = ξ X p
0
– 3 p P
E
– ( ) = =
dX dε
v
p
⁄
h
κ
κ X dκ dX ⁄
GeoModel theory
56
Relationship between and (cap curvature model)
For a purely porous material, Fig. 4.13
suggests that is simply smaller than by a
fixed amount. Therefore, knowing is suffi
cient to determine . For a material that con
tains both pores and microcracks, the
relationship between the branch point and
is similar, but influenced by the pressure
sensitivity of the fracture function. Recall
that the continuously differentiable meridi
onal yield function is constructed by multi
plying a function times a cap porosity
function . Qualitatively, the fracture func
tion marks the onset of shear crack growth,
with significant pressure strengthening being
the result of friction at crack faces. This yield
envelope function might be lower than the
ultimate shear limit envelope if kinematic
hardening is allowed. The cap porosity func
tion intersects the hydrostat ( axis) at
, marking the point at which pressure
under hydrostatic loading would be suffi
cient to induce pore collapse. Variation of the
cap function with shear stress (along the
ellipse) merely reflects an expectation that
pore collapse will commence at a lower pres
sure than the hydrostatic limit when shear
assisted.
Isotropic hardening in the GeoModel is
cast in terms of the branch point located at
(or, equivalently, ) where the
yield function begins to deviate from the
envelope function . The height of the
branch point in Fig. 4.15 is considerably
lower than the peak height .
Between the branch and the peak, mate
rial response begins to be influenced by
porosity, but is still shear crack (dilatation)
dominated, and therefore plastic volume
increases because of crack bulking. Between
the peak and the hydrostatic limit point ,
porosity dominates the material response,
κ X
f
f
I
1
I
1
I
1
f
c
s
h
e
a
r
lim
it
e
n
v
e
lo
p
e
,
Ff
(
I 1
)
cap (porosity) function
yield
J
2
s
h
e
a
r
li
m
i
t
,
Ff
(
I 1
)
BRANCH
POINT
κ
κ
BRANCH
POINT
X
X
profile
1
B
b
a
A
Figure 4.15. Continuously differentiable Geo
Model yield function and some characteristic di
mensions. The ratio is usually larger than 1,
whereas is usually smaller than 1. The bold
blue curve is . The dashed yield curve is .
a b ⁄
A B ⁄
F
f
f
c
f
f
f
c
R
a
b
 =
“
y
ie
l
d
”
e
n
v
e
lo
p
e
,
f
f
=
Ff
(
I 1
)

N
κ X
X
κ
κ
X
f
f
f
f
c
f
f
f
f
F
f
f
c
I
1
X
I
1
κ = I
1
κ =
f
f
f
b
B
X
GeoModel theory
57
resulting in plastic volume compaction (from pore collapse). At the critical zeroslope
point on the yield surface, material response is influenced equally by both cracks and
pores so that so that no net volume change is apparent at the macroscale.
Twosurface models [e.g., Ref. 41] typically construct and evolve the yield function by
making direct reference to the ratio , which (referring to Fig. 4.15) is typically
smaller than unity. The geomodel, however, constructs and evolves its yield function
based on the ratio , which is typically larger than unity. This distinction between
the two ratios is important to emphasize in publications and presentations to avoid confu
sion between the GeoModel and conventional twosurface models.
Guided both by trends in observed data and by microphysical theories (e.g., Fig. 4.13),
the GeoModel presumes that hardening proceeds such that the ratio between the distances
a and b labeled in Fig. 4.15 remains always equal to a userspecified constant . The axis
labels in Fig. 4.15 indicate that and . Therefore the cap eccen
tricity (also called the cap shape parameter) is given by
or, solving for ,
. (4.69)
When written without the overbar denoting the negative, this equation becomes
. (4.70)
This expression is evaluated internally within the GeoModel coding to determine as a
function of the internal state variable . Differentiating both sides of this equation with
respect to gives
. (4.71)
The κ evolution law
Substituting Eq. (4.71) into (4.63) gives
, (4.72)
where the “isotropic hardening parameter” is
(4.73)
A B ⁄
R a b ⁄ ≡
R
a X κ – = b F
f
κ ( ) =
R a b ⁄ =
R X κ – ( ) F
f
κ ( ) ⁄ = X
X κ RF
f
κ ( ) + =
X κ RF
f
κ ( ) – =
X
κ
κ
dX
dκ
 1 RF
f
′ κ ( ) – =
κ
·
h
κ
λ
·
=
h
κ
h
κ
min 3
∂φ
∂I
1

dX dε
v
p
⁄
1 RF
f
′ κ ( ) –

\ .
 
I
1
κ –
R

\ .
 
dX dε
v
p
⁄
1 RF
f
′ κ ( ) –

\ .
 
, =
compactiondominated
hardening modulus
dilatationdominated
hardening modulus
GeoModel theory
58
subject to the constraint, .
*
For
numerical convenience, this constraint is
replaced by , where is
the initial bulk modulus. The derivative
is found from crushcurve data fit
ted to Eq. (4.68).
The first term in the minimum function of
Eq. (4.73) dominates when the stress state
falls on the “compaction dominated” part of
the yield surface, labeled in Fig. 4.16, while
the second term dominates in the dilatation
regime. The second term in the minimum
function is guided by trends in observed data.
Even under monotonic loading, the rela
tive position of the stress state on the yield surface can move from compaction to dilata
tion regimes. Figure 8.6(a) (page 100) shows a triaxial compression load path (angled red
arrow) that falls initially on the porosity (compaction) dominated portion of the yield sur
face, which therefore results in plastic volume reduction and an isotropic expansion of the
yield surface. Compaction from void collapse and dilatation from crack bulking are rela
tively balanced in the vicinity of the critical point. After the stress passes through the crit
ical point, the hardening modulus in Eq. (4.73) transitions from its compaction
dominated value to the dilatationdominated value.
Evolution equation for the kinematic hardening backstress tensor. The
GeoModel supports kinematic hardening, but is otherwise isotropic. Kinematic hardening
entails using a shifted stress tensor in the yield function instead of the actual
stress. The deviatoric tensor internal state variable is called the backstress, and it is
computed using evolution equations described here.
Recall that the yield criterion with kinematic hardening is given by
, (4.74)
where is the second invariant of the shifted stress,
. (4.75)
The backstress tensor is initialized to zero. Upon onset of yielding, the backstress
evolves in proportion to the deviatoric part of the plastic strain rate:
* When using the overbar to denote the negative, Eq. (4.72) may be written subject to
. Thus, since is never negative, this constraint ensures that will never decrease.
Physically, this is equivalent to demanding that porosity must always decrease. Increases in poros
ity (i.e., softening) cannot be accommodated at the material constitutive level — to avoid mesh
dependencies of the solution, softening must be handled at the field scale by the host code.
z
r
Figure 4.16. Meridional plane in which the
magnitude of the stress deviator “r” is plotted
against (which is proportional to the pres
sure “p”) The yield surface (solid) demarks
the onset of inelastic flow. Under continued in
elastic loading, the yield surface hardens (ex
pands and/or translates) toward the shear limit
surface (dashed).
z –
s
h
e
a
r
lim
it s
u
r
fa
c
e
c
o
m
p
a
c
t
i
o
n
ELASTIC DOMAIN
d
ila
ta
tio
n
0 h
κ
∞ – > >
κ
·
h
k
γ
·
=
0 h
k
∞ < < γ
·
κ
0 h
κ
0.01b
0
2
– > > b
0
dX dε
v
p
⁄
h
κ
ξ
˜
˜
S
˜
˜
α
˜
˜
– =
α
˜
˜
J
2
ξ
F
f
I
1
( ) N –  
2
F
c
I
1
( ) =
J
2
ξ
ξ
˜
˜
S
˜
˜
α
˜
˜
– =
α
˜
˜
GeoModel theory
59
, (4.76)
where . (4.77)
Hence, comparing with Eq. (4.61), the kinematic hardening modulus tensor is given by
, (4.78)
where is a material constant and is a scalarvalued decay function designed to
limit the kinematic hardening such that as approaches the shear limit surface,
. Since the yield function itself is defined in terms of , the maximum
kinematic translation that can occur before reaching the limit surface equals the model off
set parameter . The GeoModel uses the function to “slow down” the rate of harden
ing as the limit surface is approached so that will equal zero upon reaching the limit
surface. Specifically, the GeoModel uses the following decay function:
, where . (4.79)
Kinematic hardening causes the octahedral profile to translate so that it no longer remains
centered at the origin (See Fig. 4.11). Consequently, the yield surface will appear to have
translated upward in the meridional plane that contains the backstress (see Fig. 4.17). The
translation distance equals and Eq. (4.79) prevents this distance from ever exceeding
the userspecified offset limit . Of course, on the meridional plane perpendicular to the
backstress, the meridional profile will not appear to have translated.
α
˜
·
˜
HG
α
α
˜
˜
( )
γ
˜
·
˜
p
=
γ
˜
·
˜
p
devε
˜
·
˜
p
dev λ
· ∂φ
∂ξ
˜
˜

\ .

 
= =
H
˜
˜
α
HG
α
α
˜
˜
( )dev
∂φ
∂ξ
˜
˜

\ .

 
=
H G
α
α
˜
˜
( )
G
α
0 → α
˜
˜
F
f
I
1
( ) F
f
I
1
( ) N –
N G
α
H
˜
˜
α
G
α
α
˜
˜
( ) 1
J
2
α
N
 – = J
2
α
1
2
trα
˜
˜
2
=
J
2
α
N
I
1
J
2
F
f
I
1
( ) N – =
I
1
J
2
F
f
I
1
( ) =
Figure 4.17. The effect of kinematic hardening on the meridional plane that contains the back
stress. According to the hardening rule, this meridional profile is permitted to translate in deviatoric
stress space by as much as a shift factor . The meridional plane perpendicular to the back stress
shows no translation at all.
N
J
2
α
GeoModel theory
60
Advancing the solution (final step, consistency parameter)
Recall from Eq. (4.59) that the consistency parameter is determined from the con
sistency condition, , applied during plastic loading intervals. Specifically,
. (4.80)
Recall the key equations governing the rates of the field and internal state variables:
(4.81a)
(4.81b)
(4.81c)
. (4.81d)
With these, the consistency condition in Eq. (4.80) becomes
. (4.82)
From which it follows that
, (4.83)
where
. (4.84)
Formal equivalence with oblique return algorithms. With the plastic parame
ter determined from Eq. (4.83), the stress rate may be written
, (4.85)
where
(trial elastic stress rate) (4.86)
and . (4.87)
In the GeoModel, the stress state is updated through direct integration of the GeoModel
plasticity equations. However, for our upcoming discussion of rate dependence, it is
important to understand that the update formula in Eq. (4.85) implies that the stress may
be alternatively integrated through time by first computing a trial elastic stress at the
end of the timestep which may be projected back to the yield surface (which itself has
been updated to the end of the step) to determine the final stress. If the trial elastic stress
λ
·
f
·
0 =
f
· ∂f
∂σ
ij
σ
·
ij
∂f
∂α
ij
α
·
ij
∂f
∂κ
κ
·
+ + 0 = =
σ
·
ij
C
ijkl
ε
·
ij
e
C
ijkl
ε
·
kl
C
ijkl
ε
·
kl
p
– = =
α
·
ij
H
ij
α
λ
·
=
κ
·
h
κ
λ
·
=
ε
·
ij
p
λ
· ∂φ
∂σ
ij
 =
∂f
∂σ
ij
C
ijkl
ε
·
ij
λ
· ∂φ
∂σ
ij
 –
\ .
 
∂f
∂α
ij
H
ij
α
λ
· ∂f
∂κ
h
κ
λ
·
+ + 0 =
λ
· 1
χ

∂f
∂σ
ij
 C
ijkl
ε
·
kl
=
χ
∂f
∂σ
ij
C
ij kl
∂φ
∂σ
i j

∂f
∂α
ij
H
i j
α
–
∂f
∂κ
h
κ
– =
σ
·
ij
σ
·
ij
E
A
ij
B
kl
σ
·
kl
E
χ
 – =
σ
·
ij
E
C
ijkl
ε
·
kl
=
A
ij
C
ijrs
∂φ
∂σ
rs
 = B
kl
∂f
∂σ
mn
 =
σ
ij
E
GeoModel theory
61
falls outside the yield surface, plastic flow must have occurred during at least part of the
solution interval. Therefore, after evolving the internal state variables appropriately to
update the yield surface to the end of the step, it can be shown that Eq. (4.85) implies that
the stress at the end of the step may be found by obliquely projecting the trial elastic stress
back onto the updated yield surface. Because is not generally proportional to , the
projection is oblique to the yield surface even if plastic normality is used. As explained in
Chapter 5, the trial stress is projected only partly back to the yield surface whenever rate
sensitivity is applied.
Quasistatic inelastic tangent stiffness tensor. For any constitutive model, the
inelastic tangent stiffness is a fourthorder tensor formally equal to the derivative of
the stress rate with respect to the total strain rate. That is,
. (4.88)
Therefore, comparing this equation with Eqs. (4.85) and (4.87), the GeoModel’s tangent
stiffness is given by
quasistatic tangent modulus
where , , and . (4.89)
The tangent stiffness is major symmetric only for associative models
.
*
Stability issues. The last term in Eq. (4.89) is subtracted from the positivedefinite
elastic stiffness , so inelastic flow can potentially make the inelastic tangent stiffness
tensor noninvertible. In other words, the tangent stiffness tensor might eventually
have a zero eigenvalue, marking the onset of softening (yield surface contraction).
Whether or not the occurrence of a zero tangent stiffness results in stressstrain softening
(i.e., a change from a positive to negative slope in a stressstrain plot) depends on the load
ing direction. For example, if the strain rate is orthogonal to the null space of a noninvert
ible tangent stiffness tensor, then no stressstrain softening will be observed and no change
in type of the momentum equation will occur even if the yield surface is contracting.
A standing wave (i.e., a nonmoving discontinuity in displacement or velocity) is
another form of material instability that has been extensively studied in the literature. The
acoustic wave speeds (i.e., the speed at which inelastic perturbations can propagate
through a material in the direction of a given unit vector ) are given by ,
where denotes the eigenvalues of the secondorder acoustic tensor,
(4.90)
* Actually, equivalence of associativity and normality holds only in the absence of elasticplastic
coupling. When elastic moduli can change in response to plastic loading, an associative
model will not exhibit normality and will not have a majorsymmetric tangent stiffness [9].
A
˜
˜
B
˜
˜
T
ijkl
σ
·
ij
T
ijkl
ε
·
kl
=
T
ijkl
C
ijkl
1
χ
P
ij
Q
ij
– =
P
ij
C
i jrs
∂φ
∂σ
rs
 = Q
mn
∂f
∂σ
mn
C
mnkl
= χ
∂f
∂σ
ij
C
ij kl
∂φ
∂σ
i j

∂f
∂α
ij
HG
α
∂φ
∂σ
i j
 –
∂f
∂κ
h
κ
– =
T
ijkl
T
klij
= ( )
f φ = ( )
f φ = ( )
C
ijkl
T
ijkl
n
˜
c x ρ ⁄ =
x
a
jk
n
i
T
ijkl
n
l
=
GeoModel theory
62
Thus, for the GeoModel, substituting Eq. (4.89) into (4.90),
, (4.91)
where
= the elastic acoustic tensor (4.92)
and
and . (4.93)
If the plastic tangent stiffness is not major symmetric, then the acoustic tensor will
not be symmetric. In this case, not only are standing waves possible, but so are
imaginary wave speeds (flutter instability). A complete spectral analysis of acoustic ten
sors of the form in Eq. (4.91) is provided in Ref. [9] where every possible ordering of the
inelastic wave speeds relative to elastic wave speeds is derived and where every possible
acoustic eigenvector is presented. Physically, the eigenvector characterizes the velocity
jump direction. If the eigenvector is parallel to the wave propagation direction , then the
wave is a compression wave. If the eigenvector is perpendicular to , the wave is a shear
wave. For elastic materials, these are the only two possible kinds of waves, but for inelas
tic tangent tensors of the form (4.89), other modes are possible.
For rate dependent materials, the question of material stability must be examined anew
because, as explained in the next section, the dynamic tangent stiffness tensor is generally
stiffer than the quasistatic tangent stiffness.
a
jk
a
jk
e
1
χ
p
j
q
k
– =
a
jk
e
n
i
C
ijkl
n
l
=
p
j
n
i
C
ijrs
∂φ
∂σ
rs
 = q
k
∂f
∂σ
mn
C
mnkl
n
l
=
T
ijkl
x 0 = ( )
n
˜
n
˜
Rate Dependence
63
5. Rate Dependence
The governing equations discussed so far are rate independent, so they only apply for
quasistatic loading. Under high strain rates, the elastic response of a material occurs
almost instantaneously, but the physical mechanisms that give rise to observable inelastic
ity can not proceed instantaneously. Materials have inherent “viscosity” or “internal resis
tance” that retards the rate at which damage can accumulate. For example, cracks grow at
a finite speed — they cannot change instantaneously from one size to another. If a stress
level is high enough to induce crack growth, then the quasistatic solution for material
damage will not be realized unless sufficient time elapses to permit the cracks to change
length. Likewise, void collapse takes finite time. Simple inertia also contributes to rate
dependence. During the time that cracks are growing towards the quasistatic solution, the
stress will drop down toward the quasistatic solution. Until sufficient time has elapsed for
the material to equilibrate, the stress state will lie outside the yield surface. If the applied
strain is released any time during this damage accumulation period, then the total damage
will be ultimately lower than it would have been under quasistatic loading through the
same strain path.
Viscoplasticity model overview
The evolution of the yield function, and the very character of the inelastic deformation
itself, can be dramatically altered by the rate at which loads are applied. In the limit of
extraordinarily high load rates (as near the source of an explosion), material response is
essentially elastic because insufficient time exists for plasticity to fully develop. At high
strain rates, the equation of state (i.e., the pressurevolume part of the elasticity) plays the
predominant role in material response. To allow for rate dependence, an overstress model
is used. The user specifies a “relaxation” parameter governing the characteristic speed at
which the material can respond inelastically. If the loads are applied over a time interval
that is significantly smaller than the characteristic response time, then essentially no
inelasticity will occur during that interval. If, on the other hand, the loads are applied
slowly (as in quasistatic testing), then inelasticity will be evident.
The GeoModel uses a generalized DuvautLions [13] ratesensitive formulation, illus
trated qualitatively in Fig. 5.1. Consider a loading increment during which the strain
increment is prescribed to be . Two limiting solutions for the updated stress can be
readily computed: (1) the lowrate (quasistatic) solution which is found by solving the
rateindependent GeoModel equations described previously, and (2) the highrate solution
corresponding to insufficient time for any plastic damage to develop so that it is sim
ply the trial elastic stress. As explained below and illustrated in Fig. 5.1, the DuvautLions
rate formulation is based on a viscoplastic differential equation, the solution of which
∆t
∆ε
˜
˜
σ
˜
˜
L
σ
˜
˜
H
Rate Dependence
64
shows that the updated stress will be (approximately) a linear interpolation between the
lowrate quasistatic plasticity solution and the highrate purely elastic solution . In
other words, there exists a scalar between 0 and 1 that depends on the strain rate such
that
. (5.1)
σ
˜
˜
L
σ
˜
˜
H
η
σ
˜
˜
σ
˜
˜
L
η
σ
˜
˜
H
σ
˜
˜
L
– ( ) + ≈
~
~
σ
˜
˜
n
H
σ
˜
˜
n 1 –
low
Figure 5.1. Rate dependence. For a given strain increment, two limiting solutions can be readily
found. The “low rate” solution , which lies on the yield surface, is the solution to the rate independent
GeoModel governing equations. The high rate solution is simply the trial elastic stress. The actual up
dated ratedependent viscoplastic stress falls between these two limiting case solutions so that
. The inset graph shows how the scale factor varies with the loading interval. If the
loading interval is long relative to the material’s characteristic response time , then sufficient time exists
to fully develop plastic response and the updated solution therefore coincides with the quasistatic solution
. If the loading interval is considerably shorter than the material’s characteristic response time, then the
solution will be the highrate elastic solution.
σ
˜
˜
L
σ
˜
˜
H
σ
˜
˜ σ
˜
˜
σ
˜
˜
L
η
σ
˜
˜
H σ
˜
˜
L
– ( ) + = η
τ
σ
˜
˜
L
y
i
e
l
d
s
u
r
f
a
c
e
σ
˜
˜
n
low
=
σ
˜
˜
n
L
q
u
a
s
i
s
t
a
t
i
c
p
l
a
s
t
i
c
i
t
y
p
u
r
e
e
l
a
s
t
i
c
i
t
y
σ
˜
˜
n
ηn
(
σn
H –
σn
L )
~
~
r
a
t
e
d
e
p
e
n
d
e
n
t
v
i
s
c
o
p
l
a
s
t
i
c
s
o
l
u
t
i
o
n
σ
˜
˜
n 1 –
σ
˜
˜
n+1
σ
˜
˜
n 1 +
H
σ
˜
˜
n
high
η
n
+
1
(
σ
n
+
1
H
–
σ
n
+
1
L
)
~
~
~
~
σ
˜
˜
n 1 +
low
=
σ
˜
˜
n 1 +
L
y
i
e
l
d
s
u
r
f
a
c
e
p
u
r
e
e
l
a
s
t
i
c
i
t
y
r
a
t
e
d
e
p
e
n
d
e
n
t
v
i
s
c
o
p
l
a
s
t
i
c
s
o
l
u
t
i
o
n
q
u
a
s
i
s
t
a
t
i
c
p
l
a
s
t
i
c
i
t
y
η
∆t
τ

1
Rate Dependence
65
The update for internal state variables is structured similarly, but uses a somewhat differ
ent weighting factor, as explained below. For both the stress and internal state variable
updates, the interpolation factor varies from 1 at high strain rates (when is small) to
0 at low strain rates (when is large), as illustrated in the graph inset of Fig. 5.1, where
the abscissa is normalized by a factor called the material’s “characteristic” response
time. At the end of this chapter, we will describe how the GeoModel assigns a value for
the characteristic material response time . Incidentally, for simplicity, Fig. 5.1 shows a
stationary yield surface. In general, the yield surface will evolve in size or translate
according to the hardening rules described earlier.
A time interval is deemed to be “long” if . A time interval is “short” if
. Soon we solve the viscoplastic equations to prove that, if the initial stress is on the
yield surface, then highrate scale factor internal state variables (ISVs) is
. (5.2)
This is also the rate factor for the stress at the onset of yielding when and coincide.
At the end of a viscoplastic step, the final stress state will not lie on the yield surface. We
will prove that, in this case, the scale factor is smaller than the value cited in Eq. (5.2).
Consequently, the “attraction” that the dynamic stress has for the quasistatic solution
increases somewhat as the stress moves farther from the yield surface.
Referring to Fig. 5.2, the weighting factor is large when the time step is signifi
cantly smaller than the characteristic time required for the material’s plasticity solution to
develop. Effects of plasticity are apparent in the GeoModel only when the time interval is
long or when the characteristic material response time is short so that will be small. In
this case, according to Eq. (5.1), the solution will be near the quasistatic (lowrate) solu
tion .
Figure 5.1 illustrates that the viscoplastic solution will follow a trajectory that is simi
lar to the quasistatic solution except displaced from the yield surface. Consequently,
experimental data for highrate loading scenarios have the appearance of inducing a
higher yield stress in the material. Unlike some plasticity models, the GeoModel does not
alter the material yield stress as a function of strain rate. Instead, the overstress model
accomplishes the same effect in a much more physically justifiable manner.
η ∆t
∆t
τ
τ
∆t ∆t τ »
∆t τ «
η
1 e
∆t τ ⁄ –
–
∆t τ ⁄
 =
σ
˜
˜
σ
˜
˜
low
η
η
σ
˜
˜
L
Rate Dependence
66
Viscoplasticity model derivation
In the context of viscoplasticity, an inviscid (rateindependent) solution for the
stress is presumed to exist. Likewise inviscid (rateindependent) solutions are pre
sumed available for the internal state variables, here denoted collectively by “q”. These
limiting case solutions are merely the solutions of the rateindependent GeoModel equa
tions described in earlier chapters. Viscous effects are incorporated by presuming that the
strain rate is decomposed as the sum of an elastic part plus a viscoplastic part :
. (5.3)
The viscoplastic part of the strain rate includes both the usual plastic strain rate from the
quasistatic (lowrate) solution as well as additional (retarding) contributions resulting
from viscosity. The viscoplastic strain rate is governed by
. (5.4)
The fourthorder tensor is the elastic compliance (inverse of the stiffness), is a
material parameter called the relaxation time, and is the rateindependent stress solu
tion whose value at the beginning of a time increment is tracked as an extra state vari
able (called QSSIGXX, QSSIGYY, etc. in Appendix B).
*
At the end of the time interval,
ultimately has the value , which is found by integrating the rateindependent Geo
Model equations from the earlier chapters. During viscoplastic loading, each internal state
variable is presumed to vary according to
. (5.5)
Here is the value of the internal state variable ( or ) throughout the time interval,
initially being equal to at the beginning of the step, and (through application of the
rateindependent GeoModel) ultimately equalling the low rate solution at the end of
the step. Like the inviscid quasistatic stress, the inviscid quasistatic ISVs (QSEL, QSB
SXX, etc.) must be tracked as distinct extra state variables.
The stress rate is, as usual, given by the elastic stiffness acting on the elastic part of the
strain rate: . Thus, using Eq. (5.3), the stress rate may be written
, where . (5.6)
Here, is the elastic trial stress rate, and therefore is the time varying elastic
trial stress that ultimately equals the highrate solution at the end of the step.
* The “low” or “inviscid” stresses must be tracked independently. They cannot be inferred by pro
jecting the actual stress onto the yield surface. Attempting to do so causes undesirable results in
ratedependent loadunload cycles.
σ
˜
˜
L
q
L
ε
˜
·
˜
e
ε
˜
·
˜
vp
ε
˜
·
˜
ε
˜
·
˜
e
ε
˜
·
˜
vp
+ =
ε
·
ij
vp
1
τ
C
ijkl
1 –
σ
kl
σ
kl
low
–   =
C
ijkl
1 –
τ
σ
˜
˜
low
∆t
σ
˜
˜
low
σ
˜
˜
L
q
q
·
1 –
τ
 q q
low
–   =
q
low
κ α
˜
˜
q
0
q
L
σ
·
ij
C
ijkl
ε
· e
=
σ
·
ij
σ
·
ij
high
1
τ
 σ
ij
σ
ij
low
– ( ) – = σ
·
ij
high
C
ijkl
ε
·
kl
≡
σ
·
ij
high
σ
ij
high
σ
ij
H
Rate Dependence
67
Eq. (5.6) is a set of linear firstorder differential equations which may be integrated
exactly over a time step with the use of integrating factors [19] provided that is
known as a function of time throughout the time step. In principal, we would need to solve
the rate independent equations analytically over the entire time step to integrate Eq. (5.6)
exactly, but this is not tractable in practice. In what follows, we will describe how the time
history of can instead be well approximated over the step. First, let’s introduce a
change of variables by defining
, (5.7)
so that the governing equation for the stress rate may be written
. (5.8)
This equation can be solved exactly if the time variation of is known
throughout the time step. Time variation of is governed by known quasistatic rate
equations. Consequently, the dynamic accuracy can be maximized by presuming that the
rate of is constant over the step so that itself is approximated to vary linearly
over the step.
Recall that the final solution can be presumed known at the end of the step
because is found by integrating the elasticity equations and is found separately by
integrating the inviscid quasistatic plasticity equations. Similarly, the difference
is known at the beginning of the step because is at the beginning of
the step and is retrieved from the saved quasistatic stress extra state variable array.
The highrate stress is simply the elastic trial stress. Thus, it varies linearly through
time from its initial to final value (with small higherorder nonlinearities if the strain rate
and/or elastic moduli are not constant). The quasistatic stress rate is an oblique pro
jection of the trial stress rate onto the yield surface, so also varies approximately lin
early through time (with nonlinear effects from flow potential surface curvature being
higher order). Thus, we may call on the mean value theorem to assert that
. (5.9)
With this approximation, the ODE in Eq. (5.8) may be solved exactly. Since equals
at the beginning of the step, the initial condition is that when . Integrat
ing the ODE, evaluating the result at the end of the step, and applying the definition of
to obtain the updated solution for eventually gives
, (5.10)
where
and . (5.11)
σ
ij
low
σ
ij
low
u
ij
σ
ij
σ
ij
high
– ≡
u
·
ij
1
τ
 u
ij
σ
ij
high
σ
ij
low
– + ( ) – =
σ
˜
˜
high
σ
˜
˜
low
–
σ
˜
˜
low
σ
˜
˜
low
σ
˜
˜
low
σ
˜
˜
H
σ
˜
˜
L
–
σ
˜
˜
H
σ
˜
˜
L
σ
˜
˜
0
high
σ
˜
˜
0
low
–
σ
˜
˜
0
high
σ
˜
˜
σ
˜
˜
0
low
σ
˜
˜
high
σ
˜
˜
·
low
σ
˜
˜
low
σ
˜
˜
high
σ
˜
˜
low
– ( )
σ
˜
˜
H
σ
˜
˜
L
– ( )
t t
0
–
∆t

\ .
 
σ
˜
˜
0
high
σ
˜
˜
0
low
– ( ) 1
t t
0
–
∆t
 –
\ .
 
+ ≈
σ
˜
˜
0
high
σ
˜
˜
0
u
ij
0 = t t
0
=
u
ij
σ
ij
σ
˜
˜
FINAL
σ
˜
˜
L
R
H
σ
˜
˜
H
σ
˜
˜
L
– ( ) r
h
σ
˜
˜
0
high
σ
˜
˜
0
low
– ( ) + + =
R
H
1 e
∆t τ ⁄ –
–
∆t τ ⁄
 = r
h
e
∆t τ ⁄ –
R
H
– =
Rate Dependence
68
With the presence of the last term, the solution in Eq. (5.10) is not precisely of the form
shown in Eq. (5.1) unless . As seen in timestep “n1” in Fig. 5.1, the initial
values for the “high” and “low” rate paths coincide only at the onset of plasticity.
Eq. (5.10) can be put into the form of Eq. (5.1) if we approximate that is par
allel to . With this assumption, Eq. (5.1) becomes
, (5.12)
where
(5.13)
and
. (5.14)
Fig. 5.2 shows how the weighting factor var
ies with the stress difference ratio appearing in
the last term of Eq. (5.13). For plastic loading,
equals 0 only when the initial state is on the yield
surface. Otherwise, beyond this onset of yielding
moment, increases, eventually asymptoting
to 1 under steady strain rates. The lowering of the rate factor caused by nonzero
makes the dynamic stress more strongly attracted to the quasistatic solution as the distance
between them increases.
For the rate dependent update of internal state variables, Eq. (5.5) can be integrated
analytically if the time variation of over the step is approximated by
. The resulting solution for is
, (5.15)
where the highrate weight factor is the same as in Eq. (5.11). The weight factor
for the internal state variables differs from the weight factor for the stress because the
two problems have different initial conditions. Recall that the material responds elastically
at extremely high rates. Hence, because the internal state variables can evolve only when
plasticity occurs, the high rate solution for any internal state variables is simply its
value at the beginning of the time increment.
Limiting case. If a strain rate is held constant for a long enough period then the differ
ence between the dynamic stress and the inviscid stress will sometimes reach a
steadystate value in the laboratory. In this case, the equations outlined in this chapter
imply that this steady state stress difference is given by
(5.16)
σ
˜
˜
0
high
σ
˜
˜
0
low
=
σ
˜
˜
0
high
σ
˜
˜
0
low
–
σ
˜
˜
H
σ
˜
˜
L
–
2 4 6 8 10
0.2
0.4
0.6
0.8
1
η
∆t
τ

2 4 6 8
1.0
0.8
0.6
0.4
0.2
10
ϒ=0
ϒ=1
Figure 5.2. Highrate weighting factor
at various initial states. The largest weight
factor (upper red curve) applies when the
initial stress is on the yield surface. The fac
tor is lower if the initial state is already off
the yield surface at the beginning of the step.
σ
˜
˜
FINAL
σ
˜
˜
L
η σ
˜
˜
H
σ
˜
˜
L
– ( ) + =
η R
H
r
h
ϒ + =
ϒ
σ
˜
˜
0
high σ
˜
˜
0
low
–
σ
˜
˜
H σ
˜
˜
L
–
 ≡
η
ϒ
ϒ
ϒ
η ϒ
q
q
low
q
H
q
L
q
H
– ( ) t t
o
– ( ) ∆t ⁄ + = q
q
FINAL
q
L
R
H
q
H
q
L
– ( ) + =
R
H
R
H
η
q
H
σ
ij
σ
ij
L
σ
ij
σ
ij
L
–
τ
χ
P
ij
Q
mn
ε
·
mn
=
Rate Dependence
69
where , , and are defined in Eq. (4.89).
In the very simplified context of nonhardening VonMises plasticity, this equation
becomes
(5.17)
where is the shear modulus and is a unit tensor in the direction of the stress deviator.
For example,
for simple shear (5.18)
for uniaxial (axisymmetric) loading (5.19)
Therefore, in the case of simple shear for nonhardening Von Mises plasticity, Eq. (5.17)
implies that
for simple shear (5.20)
for uniaxial (axisymmetric) loading (5.21)
These simple analytical results can be used to trendtest the numerical implementation of
rate sensitivity, as described in Fig. 5.3.
χ P
ij
Q
ij
σ
ij
σ
ij
L
– 2Gτ ( )S
ˆ
ij
S
ˆ
mn
ε
·
mn
=
G S
ˆ
ij
S
ˆ
 
1
2

0 1 0
1 0 0
0 0 0
=
S
ˆ
 
1
6

2 0 0
0 1 – 0
0 0 1 –
=
σ
12
σ
12
L
– 2Gτε
·
12
=
σ
11
σ
11
L
–
4G
3
τε
·
11
=
σ
11
σ
HEL
⁄
ε
11
σ
11
σ
11
L
– ( )
σ
11
σ
11
L
time
Figure 5.3. Rate dependence in uniaxial strain loading of a nonhardening pressureinsensitive
material. In the stressstrain plot, the normalizer is the quasistatic uniaxial yield stress. The
dynamic (black) stressstrain plot exhibits an apparent increase in strength relative to the quasis
tatic (dashed blue) solution. This rateinduced stress difference is plotted in the second figure
(normalized by the peak value), where the dashed red line shows the analytical asymptote enve
lope from Eq. (5.21). This calculation included a lower strength in tension (interestingly, the dif
ference between dynamic and quasistatic strengths is higher in tension).
σ
HEL
1
–1
Rate Dependence
70
Assigning a value to the characteristic material time
The GeoModel permits the user to control the value of the characteristic time through the
use of up to seven positivevalued parameters, through which are employed in the
code to assign a value of the characteristic time according to the following formulas:
, (5.22)
where
, and , (5.23)
and (McCauley brackets). (5.24)
If a constant characteristic time is desired, then set , and all other T’s to zero.
Fig. 5.3 used a constant , as did the simpleshear ratedependent simulation shown in
Fig. 7.8(b) on page 92.
Suitability of the GeoModel’s overstress ratedependence theory for predicting labora
tory data is illustrated in Fig. 8.9 on page 103.
T
1
T
7
τ
f ε
· equiv
( ) if ε
v
p
0 ≤
T
6
f ε
·
equiv
( ) 1 <T
7
I
1
T
5
+ ( )> ( )
2
+ ( ) if ε
v
p
0 > and I
1
T
5
– ≥ ¹
´
¦
=
f ε
· equiv
( )=T
1
1
ε
· equiv

\ .
 
T
2
ε
· equiv
ε
˜
˜
·
=
<x>
0 if x 0 ≤
x if x 0 >
¹
´
¦
=
τ T
1
τ =
τ
Rate Dependence
71
Thermodynamics considerations
At present, the GeoModel’s equation of state is incorporated within the purely
mechanical model — it contains no thermal terms (i.e., terms involving material proper
ties such as the specific heat and Grüneisen parameter). Nonetheless, the GeoModel’s
nonlinear elasticity model has been successfully fitted to Hugoniot shock data. How can
this be? In thermodynamics, pressure is typically expressed as a function of two variables:
the density and a thermal variable (usually temperature or entropy). For example,
. (5.25)
In thermodynamics, you can always use a purely mechanical equation of state if you
restrict the class of allowable problems so that one or two of the thermal variables are
interrelated in some known way. If, for example, you restrict attention to isothermal load
ing, then the pressure will be expressible in the form , where the “material con
stants” in the equation (such as the bulk modulus) must be set to their isothermal values.
Likewise, if you can consider only adiabatic loading, then the pressure is again expressible
as a mechanical function if the parameters such as the bulk modulus are set to their adia
batic values.
The GeoModel is parameterized at low pressure and low strain rates under isothermal
conditions (room temperature), but at high pressures and high strain rates under adiabatic
conditions. Thus, the nonlinear fit for the bulk modulus may be regarded to transition
from isothermal to adiabatic properties as pressure is increased. This implies a vague
“domain of applicability” for the mechanical GeoModel. Specifically, the model may be
used for problems where low pressure regions are also isothermal (and near room temper
ature) and high pressure regions are adiabatic (high rate). Without information on the ther
mal properties of the material, the validity of the mechanical GeoModel in other domains
cannot be ascertained. In particular, using the model in highrate/lowpressure applica
tions (acoustics) or lowrate/high pressure problems (e.g., creep of underground salt bod
ies) will possibly require reparameterization.
A good method for gauging the degree to which the mechanical GeoModel can be
applied in broad thermodynamic domains would be to compare isothermally measured
elastic moduli (inferred by the slope of a stress strain curve) with acoustically measured
moduli, which are the lowpressure isentropic elastic constants. The difference between
the isothermal and isentropic bulk modulus (at a given pressure) is proportional to the
square of the thermal expansion coefficient divided by the specific heat. Thus, if no differ
ences are observed between the isothermal and isentropic moduli at a given pressure, the
domain of applicability of the purely mechanical model is likely broad.
ρ
p f ρ T , ( ) =
p F ρ ( ) =
GeoModel Numerical Solution scheme
72
6. GeoModel Numerical Solution scheme
Aside from kinematic hardening, the GeoModel is isotropic and therefore the yield
function depends only on principal stresses. An eigenvalue analysis is avoided by casting
the yield function in terms of stress invariants. The principal stress directions (eigenvec
tors) are not needed to evaluate the yield function. For any isotropic elasticity model, how
ever, evolution of plastic response must allow for rotation of principal stress directions
caused by the elastic portion of the loading. Thus, the governing equations must be cast in
incremental tensorial form, requiring all six independent components of symmetric ten
sors to be passed to the model. Careful numerical integration schemes [6] are required to
ensure accuracy and convergence.
This chapter begins with a description of how the GeoModel is to be used within a
host (finiteelement) code, followed by a discussion of the influence of material softening
on fieldscale stability (i.e., stability of the spatial finiteelement solution, not stability of
the GeoModel’s internal time integration algorithm). Next, GeoModel installation instruc
tions are provided that describe the public
*
subroutines and memory requirements. Fol
lowing a summary of plotable GeoModel output, the GeoModel’s time integration
algorithm is briefly summarized.
Role of the GeoModel within a finiteelement program
The GeoModel is designed for use in host codes (typically finiteelement programs)
that solve the momentum balance PDE,
, (6.1)
where is the spatial Cauchy stress tensor (denoted in Eq. 3.39), is the spa
tial position vector, is the body force per unit volume, is the mass density, and is
the material acceleration that is related to the spatial velocity field by material time
derivative
(6.2)
* i.e., called directly from the host code.
∂σ
ik
∂x
k
 f
i
+ ρa
i
=
σ
ik
σ
˜
˜
spatial
x
k
f
i
ρ a
i
v
i
a
i
∂v
i
∂t

∂v
i
∂x
k
v
k
+
Dv
i
Dt

∂v
i
∂t

\ .
 
X
˜
= = =
GeoModel Numerical Solution scheme
73
Here, is the timezero reference position vector, which serves to identify Lagrangean
material points. The spatial position vector is related to the reference position
through the deformation mapping function, such that the deformation gradient tensor is
(6.3)
The stress and velocity fields (as well as displacement or velocity boundary conditions)
are known at the beginning of each time step, so that application of Eq. (6.1) permits eval
uation of the acceleration field. In most host codes, the updated position of a material par
ticle is computed to secondorder accuracy with respect to the time step through
application of
(6.4)
or
. (6.5)
Equivalently,
, (6.6)
where . (6.7)
To date, all installations of the GeoModel have approximated the unrotated strain rate by
the unrotated symmetric part of the velocity gradient defined in Eq. (4.3), evaluated at the
halfstep by using the velocity field in Eq. (6.7). The GeoModel integrates the unrotated
strain rate to predict the unrotated stress at the end of the step, which must then be rotated
into the spatial configuration by the host code. As mentioned on page 20 (and clarified
later in this chapter), optimizing the accuracy of the spatial solution for problems involv
ing massive material rotation requires the host code to apply its un/rotation operations
using polar rotation tensors that are consistent with the part of the time step (beginning,
half, or end) at which un/rotation operations are required.
Of course, once the GeoModel and any other constitutive models in the problem have
been applied to determine stresses at the end of a time step, Eq. (6.1) may be integrated
again to update the acceleration field, thus launching a new timestep cycle.
X
˜
x
˜
X
˜
F
ij
∂x
i
∂X
j

\ .
 
t
=
∆t
x
˜
n 1 +
x
˜
n
∂x
˜
∂t

\ .
 
X
˜
n
∆t
1
2
 
∂
2
x
˜
∂t
2

\ .
 
X
˜
n
∆t ( )
2
+ + =
x
˜
n 1 +
x
˜
n
v
˜
n
∆t
1
2
a
˜
n
∆t ( )
2
+ + =
x
˜
n 1 +
x
˜
n
v
˜
n+1 2 ⁄
∆t + =
v
˜
n+1 2 ⁄
v
˜
n
1
2
 
a
˜
n
∆t + ≡
GeoModel Numerical Solution scheme
74
Spatial stability (mesh dependence/loss of strong ellipticity)
As rock deforms inelastically, the initial yield surface (lower curve in Fig. 4.16)
evolves toward — and is not permitted to evolve beyond — the limit surface (upper curve
in Fig. 4.16). If the stress reaches the limit surface, the rock has, in a loose sense, failed
catastrophically. More correctly, the GeoModel has reached the limit of its applicability
because large scale cracking and subsequent loss in strength cannot be modeled locally at
the constitutive level. Material softening generally produces a change in type of the
momentum equation, requiring intervention from the host code to change its solution
scheme appropriately (for further details, see page 61). In the GeoModel, the state variable
“CRACK” is a flag equaling 1.00 whenever a principal stress (or ) cutoff has been
applied or 2.00 when the stress has reached the limit surface and can harden no further;
otherwise “CRACK” equals 0.0. In either case, a positive value of “CRACK” marks the
onset of softening.
Because the GeoModel comes equipped with its own flags for failure, this model
should not be used with other fracture models such as a maximum principal stress crite
rion. Instead, the “CRACK” flag should be queried by the host code to determine when it is
appropriate to add void (when pressure is tensile) or to apply discontinuous shear dis
placement element shape functions (when pressure is compressive) or to apply any other
appropriate response
*
to material softening that will ensure localization response that con
verges as the spatial mesh size is reduced.
If the host code fails to activate any special response when the “CRACK” flag becomes
nonzero, the GeoModel will continue to run, but its predictions are suspect. Without a
meaningful hostcode response to failure, the geomodel will handle the inelastic response
at the limit state in a manner similar to nonhardening plasticity (i.e., rather than properly
softening down away from the limit state, the GeoModel will force the stress to dwell at
the limit state). To summarize, the GeoModel is intended to model only the portion of
material response that is appropriate to compute at the local constitutive level. The Geo
Model sends flags back to the host code at the onset of softening (a nonlocal phenome
non). The host code is responsible for responding appropriately to these flags by initiating
material softening.
* We are investigating a fieldscale softening strategy that introduces a length scale based on Weibull
perturbations of the material strength field, which is especially appealing because (unlike “element
death” and cohesive zone models) it can be easily justified physically. Specifically, softening
results from subgridscale flaw clustering, which can be shown to have a Weibull (or nearly
Weibull) distribution and can be parameterized via standard laboratoryscale experiments. Prelimi
nary investigations [4] have shown that Weibulllike softening strategies lead to very realistic frag
ment patterns and are meshindependent.
I
1
GeoModel Numerical Solution scheme
75
GeoModel files, subroutines, memory requirements, and
model installation requirements.
This section is a software requirements specification that must be followed by anyone
who installs the GeoModel into a host code.
*
The GeoModel is designed to be imple
mented into multiple host codes without any revision of the source code. As described
below, the model has three public
†
subroutines (GEOCHK, GEORXV, and ISOTROPIC_
GEOMATERIAL_CALC). To support portability, the GeoModel conforms to Sandia’s
Model Interface Guidelines (MIG) [10]. Therefore, the model presumes that calculations
entail three distinct phases, the first two of which are performed at startup while the last
one is applied for every element at every timestep:
1. User input. The GeoModel requires the host code to acquire user input values and save
them into a single array using the keywords and ordering listed in Appendix B. This property
array must be passed to the subroutine GEOCHK for “domain certification” (i.e., verifying that
input values fall within allowable ranges, as explained on page 81). Additionally, the routine
GEOCHK sets defaults for unspecified user inputs.
2. Storage. To be portable, the GeoModel does not actually allocate storage for internal state
variables (ISVs) — this is the responsibility of the host because data layout varies from code
to code. The GeoModel provides a list of storage requirements by requiring the host code to
call subroutine GEORXV. This routine returns physical dimensions, initial values, plot
keywords, and advection requirements for each internal state variable. To use this routine, the
host code loops over the lists returned by GEORXV to then allocate the storage, define plot
options, and initialize the ISV fields.
3. Execution. Every cycle, ISOTROPIC_GEOMATERIAL_CALC must be called to update
the stress to the end of the step. Detailed descriptions of the inputoutput arguments are
provided below.
The GeoModel must, upon occasion, relay messages to the user or terminate the calcula
tion. Log message protocols and bombing procedures vary among host codes. For porta
bility, the GeoModel follows MIG guidelines by calling subroutines LOGMES, FATERR,
or BOMBED whenever it needs to relay messages to the user, log fatal errors, or terminate
calculations, respectively. These routines are not part of the GeoModel source code.
Because these actions require hostcode responses that vary from code to code, these rou
tines must be written and maintained by the hostcode architects. Likewise, a routine
called TOKENS (used only in the extra variable request routine) is expected to preexist in
the host code. Any host code that already supports MIG models will already have these
MIGutilities in their repository. Host code architects of nonMIGcompliant codes may
request sample MIGutilities that they may customize to suit their own code’s protocols
for information passing and code termination.
* Any deviation from these model installation instructions (as well as any modification of the Geo
Model source code itself) may result in loss of technical support. Model installers who believe that a
deviation from these instructions is warranted are encouraged to contact the GeoModel developers.
† i.e., routines that are called directly from the host code. All other routines in the GeoModel are “pri
vate” and should not be called by the host. To serve codes that remesh, one additional public rou
tine, isotropic_geomaterial_state, is available that will repair advection errors.
GeoModel Numerical Solution scheme
76
The GeoModel source code is available in both FORTRAN 77 and 90. To date, large
scale production codes have used the F77 version to best ensure portability, so those rou
tines will be described here. The F90 routines, which are not significantly different, are
generated from the F77 master files to ensure consistency.
Arguments passed to and from the GeoModel driver routine
The following list describes variables passed between the host code and the Geo
Model’s driver routine (isotropic_geomaterial_calc):
INPUT:
• NBLK: The number of cells or finite elements to be processed. Parallel codes send
only one cell at a time (NBLK=1).
• NINSV: The number of internal state variables for the GeoModel.
• DT: The time step
• PROP: the userinput array, filled with real numbers, as summarized at the top of the
nomenclature table in Appendix B and also summarized within the source code
prolog itself.
• SIG: The unrotated Cauchy stress tensor at time n. The six independent components
of the stress must be passed in the ordering . Within the
FORTRAN, this array is dimensioned “SIG(6, NBLK)” so that the stress components
for any given finite element are in six contiguous memory locations.
• D: The unrotated strain rate tensor, preferably evaluated at time n+1/2 because the
GeoModel treats the strain rate tensor as constant over the entire interval. Most codes
approximate the strain rate tensor as the unrotated symmetric part of the velocity
gradient (see Eq. 4.3). Component ordering and contiguous storage are the same as
for stress.
• SV: the internal state variable array containing reals, as described in the nomenclature
table in Appendix B.
OUTPUT:
• SIG: The unrotated stress tensor at time n+1. The component ordering is the same as
described above.
• SV: The internal state variable array (updated to time n+1)
• USM: Uniaxial strain (constrained) elastic modulus equal to . The host
code may use the USM output to compute an upper bound on the wave speed ( ,
where is mass density) when setting the timestep.
These arguments require unrotation of spatial stress at time n, unrotation of the strain rate
at time n+1/2, and rotation of the updated stress back to the spatial frame at time n+1. For
problems involving significant material rotation, this requires using three different polar
rotation tensors [14]. For moderate rotation problems, it might be acceptable to use the
rotation tensor evaluated at time n+1/2 for all three operations, but the accuracy conse
quences of this simplified approach have not been carefully examined in the computa
tional mechanics literature.
σ
11
σ
22
σ
33
σ
12
σ
23
σ
31
, , , , , { }
H K
4
3

G + =
H ρ ⁄
ρ
GeoModel Numerical Solution scheme
77
PLOTABLE OUTPUT
In addition to the stress, any variable in the SV state variable array is available for plot
ting. The plot keywords (and ordering of variables in the SV array) are listed in Nomencla
ture Appendix B.
As mentioned earlier, the “CRACK” flag may be plotted to visualize softening regions.
The “INDEX” flag may be plotted to locate regions that are now or have ever deformed
inelastically (even if elastic at the given instant). To visualize regions that are currently
deforming inelastically, the “SHEAR” variable should be plotted.
The “SHEAR” variable is an informational output, equal to zero during elastic cycles
and equaling a measure of normality of the trial elastic stress rate during plastic intervals.
Specifically, as illustrated in Fig. 6.1, “SHEAR” ranges from zero when the trial stress rate
is tangent to the yield surface to unity when it is normal to the yield surface; an intermedi
ate value of “SHEAR” indicates oblique plastic loading relative to the yield surface. While
“SHEAR” quantifies the plastic loading direction relative to the yield surface, the internal
state variable “DCSP”, which is the plastic consistency parameter , may be plotted as a
measure of the magnitude or intensity of plastic loading.
The “EQDOT” variable may be plotted to gain an overall sense of intensity of the cur
rent strain rate. Small values of “EQDOT” correspond to relatively quiescent regions. Plot
ting “EQPS” will show equivalent plastic shear strain, while “EQPV” gives plastic volume
strain (and is roughly equal to the porosity change from inelastic void collapse).
The stress invariant “I1” is three times the negative of pressure. “ROOTJ2” may be
regarded as a scalar measure of effective shear stress and is proportional to the radial coor
dinate of the stress in the octahedral plane. The Lode angle, “LODE”, quantifies the angu
lar location of the stress in the octahedral plane, and it varies from 30 for triaxial
compression to –30 for triaxial extension (0 for simple or pure shear). If kinematic harden
ing is enabled, “BACKRN” quantifies the distance that the origin of the octahedral profile
has shifted in stress space.
The complete list of other (less useful) plotable output is in Appendix B.
λ
·
500 1000 1500 2000
30
20
10
10
L
o
d
e
a
n
g
l
e
time
elastic plastic
–30
(a)
500 1000 1500 2000
0.2
0.4
0.6
0.8
1
time
elastic plastic
S
H
E
A
R
(c)
Figure 6.1. Meaning of the “SHEAR” output variable. This calculation uses a Von Mises yield sur
face. The strain rate remains in triaxial extension for half of the calculation, which is why the Lode angle
(a) is initially constant at . During this interval, the stress reaches the yield surface and continues to
push directly against it, which is why SHEAR (c) jumps to and holds at 1.0. Halfway through the problem,
the strain rate direction is changed in stress space [as indicated by arrows in (b)] to move the stress toward
triaxial compression. At the beginning of this transition, SHEAR first jumps to 0.5 where the normal and
tangential components are equal, and moves back toward 1.0 as the tangential component decays.
30° –
(b)
GeoModel Numerical Solution scheme
78
GeoModel algorithm
The GeoModel presumes that the strain rate is constant throughout the entire step, and
the stress is integrated as follows:
Rate independent (inviscid) part of the viscoplasticity equations.
STEP 1. To guard against unpredictable hostcode advection errors (or similar corruption of the
updated state from the last time step), apply a return algorithm to ensure the initial
stress is on or inside the yield surface.
STEP 2. Compute the nonlinear elastic tangent moduli appropriate to the stress at time n.
STEP 3. Apply Hooke’s law in rate form to obtain the elastic stress rate at time n.
STEP 4. Integrate the elastic stress rate using firstorder differencing to obtain an estimate for
the trial elastic stress at the end of the step.
STEP 5. Evaluate the yield function at the trial elastic stress. If the yield function evaluates to a
negative number, the trial elastic stress is accepted as the final updated stress, and the
inviscid algorithm returns (i.e., go to STEP 16). Otherwise, continue.
STEP 6. To reach this step, the trial elastic stress state was found to lie outside the yield surface.
At this point, the time step is divided into an internally determined number of
subcycles. All subsequent steps described below this point apply to the smaller time
steps associated with subcycles.
STEP 7. Evaluate the gradients of the yield function for eventual use in Eq. (4.83).
STEP 8. Evaluate the flow potential gradients for eventual use in Eqs. (4.87) and (4.85)
STEP 9. Evaluate the isotropic hardening coefficient in Eq. (4.72).
STEP 10. Evaluate the function in Eq. (4.76).
STEP 11. Apply Eq. (4.83) to obtain the consistency parameter.
STEP 12. Use forward differencing (within the subcycle) to integrate Eqs. (4.72) and (4.76),
thereby updating the internal state variables, and . Similarly integrate Eq. (4.85)
to advance the stress to the end of the subcycle.
STEP 13. The above steps will have directly integrated the governing equations through the end
of the subcycle, so the updated stress will be in principle already on the yield surface.
However, guard against slight roundoff and integration errors by applying an iterative
return correction to place the stress exactly on the yield surface.
STEP 14. Increment the subcycle counter, and save the partially updated inviscid internal state
variables.
STEP 15. If subcycles remain to be evaluated, go to STEP 7. Otherwise, continue to STEP 16.
Viscous part of the viscoplasticity equations.
STEP 16. The previous set of steps govern computation of the equilibrium state. Now apply
Eq. (5.22) to compute the characteristic material response time.
STEP 17. Using the trial elastic stress corresponding to an update to the end of the time step,
apply Eq. (5.10) to compute the dynamic stress. Apply Eq. (5.15) to similarly compute
the dynamic values of internal state variables to account for rate sensitivity.
STEP 18. Save the values of the internal state variables into the state variable array.
STEP 19. STOP.
h
κ
G
α
κ α
ij
Software “confidence building” activities
79
7. Software “confidence building” activities
This chapter describes progress towards Software Quality Assurance (SQA), which
encompasses a broad range of activities including code maintenance, documentation, and
(most importantly) code verification. “Code verification” is defined by the IEEE [22] as
“formal proof of program correctness” in the sense that the governing equations are
numerically solved correctly within a tolerable degree of accuracy. Model validation
*
will
not be discussed until Chapter 9.
We make no claims at this point that the GeoModel software has been exhaustively
verified. In other words, we cannot state with absolute certainty that the governing equa
tions presented in this report are in fact solved correctly. One might challenge the Verifi
cation and Validation (V&V) community to prove that “formal proof of program
correctness” is even possible. Realistically, the confidence one can place in the veracity of
any model prediction can be based only on the extent to which documented evidence sug
gests that the equations are solved correctly. It seems acceptable, therefore, to speak of
varying degrees of progress towards verification
†
, or, more generally, varying degrees of
SQA. In this sense, the GeoModel has undergone a higher level of SQA than is normally
applied to modern material constitutive models of comparable complexity.
‡
Even though
we claim that the GeoModel’s verification and SQA status is above average, we do not
assert that such activities have progressed to the point where we consider the job “fin
ished.” Here in this chapter, we aim only to build confidence in the GeoModel by summa
rizing some of the SQA activities that have been applied to the GeoModel to date (a
comprehensive detailed discourse would fall well outside the scope of this report).
Once a constitutive model is installed within a host code, it becomes only a single
component of a much larger and different model (the finiteelement code). Constitutive
SQA should include ensuring that the model can be installed and run in a variety of host
codes, but verification of the installation is primarily a host code (not constitutive) respon
sibility. Constitutive SQA in the context of largerscale model integration is limited to pro
* Whereas “Verification” seeks to confirm that the equations are solved correctly without questioning
their appropriateness, “validation” compares model predictions against experimental data to deter
mine whether or not the equations themselves are indeed suitable for the application.
† Suppose, for example, that one constitutive model has been verified for both uniaxial strain and
simple shear, whereas another one has been tested only in uniaxial strain. If these are the only tests,
neither model is “well verified,” but the first one is certainly better verified.
‡ From a practical (rather than philosophical) standpoint, the complexity of a model must be consid
ered when speaking about how well a model has been validated. With a given level of financial and
computational resources, solids models cannot be tested to the same level of certainty as fluids
models. Conclusively demonstrating only firstorder accuracy of an anisotropic solid constitutive
model would require more than twenty times the effort needed to verify firstorder accuracy of a
simple fluids model (this follows because a general anisotropic stiffness tensor has 21 independent
components, and therefore 21 independent strain paths would be required to conclusively verify
accuracy; moreover comprehensive testing for solids models requires coordinate invariance tests
that are not needed for scalar fluids models).
Software “confidence building” activities
80
viding adequate model installation instructions (including operational constraints),
delivering correct solutions to the constitutive governing equations, and ensuring that the
model will, wherever feasible, “trap” invalid calling arguments (much as a compiler
“traps” IEEE errors such as division by zero).
Numerical schemes for solving field PDEs (such as NavierStokes equation or Max
well’s electromagnetism equations) have received considerable attention in the V&V liter
ature, while verification of constitutive models has been relatively ignored. Constitutive
models are subcomponents within fieldscale calculations. As such, a partitioning of
responsibility for SQA is needed. (As an analogy, note that the quality assurance responsi
bilities for a turnsignal manufacturer must be different from those of an automobile man
ufacturer.) Constitutive verification aims to build confidence that the model will return
correct solutions to the governing equations, presuming that the host code sends inputs
falling within the admissible domain for the model. SQA may additionally include some
checking of the inputs themselves, as long as doing so does not compromise efficiency.
When a host code sends inadmissible inputs (such as corrupted strain rates caused by
mesh entanglement, advection, or artificial viscosity errors), then correcting such errors is
not the responsibility of a constitutive modeler unless it can be proved by the host code
developers that such errors originated from constitutive model output errors.
Similarly, because the GeoModel is a local constitutive model (i.e., because it does not
solve spacetime PDEs), demonstrating convergence with respect to the spatial mesh is
not a GeoModel verification responsibility. It is well known that mesh dependence can
occur when local constitutive models permit material softening. Therefore, we regard sup
pression of softening as an implicit software requirement specification because our cus
tomers (finiteelement code teams) do not have the code infrastructure that is needed to
properly handle the change in type of their governing PDEs that occurs upon softening.
Until such enhancements are made at the host code (not constitutive) level, the GeoModel
predicts only the onset of catastrophic failure, not its subsequent evolution into macroscale
fragments and fractures.
Software “confidence building” activities
81
In the lexicon of Ref. [25], “static” SQA testing is a prerequisite to the verification
process that encompasses tests that are performed without running the code. To date, the
GeoModel has undergone the following static SQA:
• Independent
*
linebyline review of the source code. The source code was
aggressively inspected to locate and remove possibilities for IEEE errors (e.g. trying to
take the ArcSin of a number larger than unity†, dividing by zero, etc.). Another goal was
to confirm that the equations being solved in the code were indeed the equations
documented in this report.
• Model domain certification (preventing “Garbage In ⇒ Garbage Out”). No function
or set of equations is well posed without a domain of applicability. The domain for the
GeoModel includes constraints on runtime subroutine arguments (e.g., the backstress
must be deviatoric) as well as constraints on the input parameters (e.g., the bulk modulus
must be positive). Except where computational efficiency would be degraded, SQA
includes appraising quality of both user input and runtime arguments sent from the host
code. A new routine (geochk) was recently added to the GeoModel that terminates
calculations if the userinput falls outside allowable ranges. Runtime testing of this
routine is discussed in the next section. Direct runtime testing of time varying
subroutine arguments is computationally inefficient, so the GeoModel’s domain
certification for variable calling arguments relies primarily on our model installation
instructions (page 75), which serve as software requirement specifications that must be
obeyed by the host code developers.
• Portability and version control. The GeoModel has been designed such that it can be
implemented in multiple host codes without altering the source code, thus allowing the
GeoModel developers to maintain a single master version. A host code owner who
faithfully obeys the model installation instructions on page 75 may update the
GeoModel by simply replacing three FORTRAN files (posted on WebFileShare under
keyword “GeoModel”) with the latest GeoModel release and then recompiling. Each
GeoModel release is identified by a six digit code
‡
that prints to the screen at run time.
The GeoModel has been compiled on multiple platforms (Sun, Dell, and HP
workstations or clusters running Linux and/or Windows) in multiple host codes (Alegra,
Presto, Pronto, JAS3D, and two independent drivers) using multiple commercial
compilers (gnu, pgf77, pgf90, and Compaq visual FORTRAN). The source code compiles
without warnings when using stringent SQA options (such as Wall O3 in the gnu
compiler).
• Model documentation. This report is the first publication that describes the numerical
algorithm, provides input definitions, and gives model installation instructions that
describe how SQA responsibilities are partitioned between the model and the host code.
• Technical support. Two standalone singlecell codes that exercise the GeoModel
under prescribed strain and stress paths are available to assist host (finiteelement) code
owners verify their GeoModel installations. The GeoModel includes a “problem
resolution” feature that generates a debugging file (geo.barf) and terminates calculations
whenever unacceptable solution quality is detected. The debugging file may be emailed
* Here, “independent” (which is not synonymous with “unbiased”) means that the source code has
been inspected by an individual who has not written the code. GeoModel code inspection was first
performed by Brannon (the second author of this report) upon joining this modeling effort at the
beginning of fiscal year 200304. Subsequent new code revisions and enhancements have been
reviewed by the member of the FossumBrannon team who did not write the new code.
† This can happen because of slight roundoff errors, as in ASIN(1.0000001).
‡ The code is simply the release date in the form “yymmdd”.
For example, “GeoModel version 040526” was released in 2004 on May 26.
Software “confidence building” activities
82
to the GeoModel team, who can import it into their driver to resolve the problem quickly.
Socalled “dynamic” SQA refers to tests performed by actually running GeoModel. Some
of the dynamic SQA activities to date are summarized here. A detailed documentation of
all GeoModel SQA would require a second report itself, so the following list should be
regarded as simply an overview. A small number specific examples will be given later.
• Model domain certification: The “geochk” feature for checking quality of user inputs
was tested by confirming that the GeoModel would abort when intentionally sent invalid
inputs. As explained below, the “geobarf” problemresolution feature can often
indirectly trap invalid forcing functions (caused, for example, by mesh entanglement).
However, the responsibility for sending valid forcing functions to the GeoModel
remains the onus of the host code, not the GeoModel.
• Runtime monitoring of the solution quality: The problemresolution (geo.barf)
feature has been verified to (1) detect “Garbage Out” predictions such as negative plastic
work, (2) terminate calculations, and (3) write a debugging file that can be emailed to
model developers to determine whether the problem was caused by bad user input, bad
arguments passed from the host code, or a bug in the internal coding.
• Driver regression testing: Two standalone drivers are available for exercising the
GeoModel in a homogenous deformation field.*
Considerable dynamic testing was
performed using our research (nonproduction) model driver that allows visualizing (and
algebraically processing) the output within Mathematica [48]. The other driver [36],
which runs either from a command line or from an Excel frontend (see Fig.7.1), is now
also deployed in the WISDM materials information database [21], allowing the
predictions of the GeoModel to be compared directly against experimental calibration
data. These drivers have been used to assemble a suite of regression tests (hydrostatic
loading, two types of shear loading, uniaxial strain, uniaxial stress, biaxial plane stress,
and numerous mixed loadunload problems using a variety of input parameters), several
of which are simple enough to admit analytical solutions for verification purposes. The
driver regression problems (15 problems to date) are all rerun and inspected for
undesirable changes whenever any change is made to the GeoModel.
• Trend testing: Engineering judgement was used to ensure that solutions vary as
expected when parameters change. For example, Fig. 6.1 depicted a simple trend test in
which analytical (exact) arguments could be used to prove that the “SHEAR” should
equal 1.0 at the onset of yielding and should dwell at 1.0 until the loading direction
changes, after which it should drop instantaneously to 0.5 and then asymptote to 1.0.
Similarly, Figs. 5.3 and Figure 7.8b confirmed that increasing the GeoModel’s
characteristic response time (see Eq. 5.22) would indeed produce the analytically
predicted increase in the apparent strength of the material. These tests were quite
valuable because they allowed correction of a serious bug in an earlier version of the
GeoModel where the response trend upon load reversal was clearly flawed. Trend and
robustness testing also demands that the GeoModel must predict qualitatively
reasonable trends when subjected to deformations that exceed what is expected in
applications (e.g., massively large elongations or pressures). For example, one problem
in our driver regression suite verifies that load/unload curves do indeed asymptote
toward each other under hydrostatic compression as sketched in Fig. 3.1 and verified in
Fig. 7.1. Many other trend tests such as these have been conducted, but (for lack of
space) will not be described in detail in this report.
* Standalone testing obviates many constraints and sources of nonconstitutive errors in production
finiteelement simulations (artificial viscosity, hourglassing, timestep control, boundarycondi
tion errors, code compilation and runtime overhead, etc.).
Software “confidence building” activities
83
Figure 7.1. Screen shot of the MS Excel interface for the GeoModel materialmodel driver (MMD).
This tool serves as a reliable platform for exercising the GeoModel under homogeneous deformations in
a simplified host code architecture that is free from solution corruption caused by unneeded finiteelement
code features such as artificial viscosity, wave motion, contact algorithms, etc.
stress or strain
(or mixed)
loading
Problem description:
Fossum Geomodel
Salem Limestone
Hydrostatic Compression (StressDriven)
**********************************************************************
Path to Fortran MMD executable:
C:\home\data\mdrive\rb\mmd0402c\Debug\mmd0402c.exe
**********************************************************************
Material model:
Index Name (optional)
11 Fossum Geomodel
**********************************************************************
Material model input parameters:
B0 B1 B2 B3 B4 G0 G1 G2 G3 G4
(Pa) (Pa) (Pa) (Pa) (ndim) (Pa) (ndim) (1/Pa) (Pa) (ndim)
1.30E+10 4.25E+10 4.11E+08 1.20E+10 0.021 9.86E+09 0 0 0 0
RJS RKS RKN A1 A2 A3 A4 P0 P1 P2
(m) (Pa/m) (Pa/m) (Pa) (1/Pa) (Pa) (rad) (Pa) (1/Pa) (1/Pa^2)
0.00E+00 2.00E+13 1.00E+12 8.43E+08 2.73E10 8.22E+08 1.00E10 3.14E+08 1.22E10 1.28E18
P3 CR RK RN HC CUTI1 CUTPS T1 T2 T3
(1) (ndim) (ndim) (Pa) (Pa) (Pa) (Pa) (s) (1/s) (ndim)
8.40E02 6 0.72 5.00E+06 9.00E+10 1.00E+11 3.00E+06 0.00E+00 0 0
T4 T5 T6 T7 J3TYPE A2PF A4PF CRPF RKPF SUBX
(1/s) (Pa) (s) (Pa) (ndim) (1/Pa) (rad) (ndim) (ndim) (ndim)
0.00E+00 0.00E+00 0.00E+00 0.00E+00 3 2.73E10 1.00E10 6 0.72 0
**********************************************************************
Load path specifications:
leg time npts nprt ltype* c11 c22 c33 c12 c13 c23
(ndim) (s) (ndim) (ndim) (ndim)
0 0 0 0 111111 0 0 0 0 0 0
1 1 100 2 444222 1.00E+09 1.00E+09 1.00E+09 0 0 0
2 2 100 2 444222 0.00E+00 0.00E+00 0.00E+00 0 0 0
*Load type parameter for each component, cij:
1 = strain rate ( 1/s )
2 = strain ( 1 )
3 = stress rate ( Pa/s)
4 = stress ( Pa )
**********************************************************************
1.1E+09
1.0E+09
9.0E+08
8.0E+08
7.0E+08
6.0E+08
5.0E+08
4.0E+08
3.0E+08
2.0E+08
1.0E+08
0.0E+00
5.0% 4.0% 3.0% 2.0% 1.0% 0.0%
Axial Strain (%)
A
x
i
a
l
S
t
r
e
s
s
(
P
a
)
Software “confidence building” activities
84
• Symmetry testing: Consistent predictions were verified for identical loading applied
in different directions (e.g. uniaxial strain in the 1direction compared with the 2
direction). Consistency has also been tested for stress paths that intersect the yield
surface at symmetrically equivalent points and for trial stress rates normal to the yield
surface (to verify that the stress rate had no tangential component).
• Extensive comparison testing:
*
To date, the GeoModel has been implemented in
five finiteelement codes: (ALEGRA [8,49], PRESTO [26], JAS3D [5], EPIC [24],
legacy PRONTO3D [45]). A disturbing number of discrepancies (e.g., wave arrival
times differing by as much as 10% or peak stresses differing by almost an order of
magnitude) have been identified by comparing predictions for the same problem
simulated by different finiteelement codes. In all but a few cases, these discrepancies
have been traced to hostcodelevel (not constitutivelevel) errors (e.g., hourglassing,
handling of boundary conditions, artificial viscosity, timestep control, etc.). Resolving
such discrepancies is the model verification responsibility of the host code owners, not
the GeoModel developers.
†
The ability of the GeoModel to reduce to simpler models
(e.g., nonhardening elastoplasticity) has often proved invaluable in determining if a
simulation problem originates from the GeoModel or elsewhere in the host code. When,
for example, an undesired feature in a calculation persists even when the GeoModel is
run using simplified parameters, a comparison test can be performed using the existing
(presumably better verified) independent version of that model within the finiteelement
code. We have, for example, often compared a host code’s standard elasticity model
with the GeoModel run in an elastic mode. In one instance where a discrepancy was
traced to the GeoModel, it was attributed to failure of the code owner to follow the
GeoModel installation instructions (page 75). In a few cases, comparison testing did
indeed reveal GeoModel bugs that have since been corrected.
• Comparison with exact analytical solutions: Analytical solutions are not
available to verify all of the GeoModel’s features acting simultaneously. However, each
GeoModel feature has been individually verified to ensure that the promised
quantitative material response is delivered (e.g., accurate tracking of a specified porosity
crush curve, exact apparent strengthening in high rate loading, etc.). Depending on how
the model parameters are set, the GeoModel can be idealized to a form for which some
problems admit exact solutions. One example is that of a linear MohrCoulomb material
with an associative or a nonassociative flow rule in homogeneous loading. This class of
problems was recently studied in a verification and benchmarking activity sponsored by
the Defense Threat Reduction Agency (DTRA) through its Advanced Concepts
Technology Demonstration (ACTD) Project. These exercises were part of a larger
Verification and Validation (V&V) effort to increase confidence in prediction of low
yield nuclear damage of underground (tunnel) facilities in jointed (in situ) rock mass.
The problems are designed to increase in complexity, by invoking additional physics in
the material models, until a level is reached that is deemed sufficient to model precision
field tests. This work will be described in further detail later in this chapter.
* Testing a numerical model in multiple host codes has been vital to our SQA process. If two codes
agree, then no conclusion may be drawn. If, however, two codes predict different answers, then at
least one of them is not solving the equations properly (or is not solving the same equations).
Resolving such discrepancies has time and again expedited bug identification and resolution.
† Incidentally, the disquieting frequency of bugs originating within the host code (not the constitutive
model) reiterates the importance of specialized constitutive model drivers in constitutive model
verification. Constitutive models should not use finiteelement codes as their primary verification
platform.
Software “confidence building” activities
85
• Consideration of the method of manufactured solutions (MMS): The MMS
method of SQA entails first solving an inverse problem in which simple (e.g.,
quadratically varying) analytical time histories for the response functions are substituted
into the governing differential operators to obtain (probably with the assistance of a
symbolic mathematics program such as Mathematica) an expression for the input
functions that would produce the prespecified response function. The numerical model
is then sent this analytically determined input function to verify that the original (pre
selected) response function is recovered. For the GeoModel, using MMS would require
preselecting an analytically simple stress history to determine a strain history to use as
input in the numerical simulation. To date, this technique has not been used in the
GeoModel verification process. Solving the inverse problem is quite difficult because
the GeoModel’s differential operators are “branched” (one set is used during elastic
deformation, while another is used during plastic loading, and the internal state variable
evolution equations themselves are coupled to the location of the stress on the current
yield surface). However, by using simplified model input parameters, solving the
inverse problem might be tractable, so MMS might indeed prove useful in the ongoing
(still incomplete) GeoModel verification process.
• orderofconvergence:
*
Time steps for complicated plasticity models often must
be much smaller than the time step used by the host (finiteelement) code. The
GeoModel’s governing equations change upon reaching yield, and this transition
typically occurs somewhere in the middle of the host code’s time step, which implies
that the constitutive model must break the step into elastic and plastic parts. Moreover,
further subcycling within the constitutive model is required to avoid “drift” of internal
state variables into nonphysical or inconsistent domains.
†
Subcycling complicates the
meaning of a convergence study performed at the host code level. Preliminary tests (see
page 86) indicate that each subcycle within the GeoModel is firstorder convergent, but
(recall the last footnote on page 79) we have not yet performed sufficient tests to
consider this claim fully verified for a broad variety of load paths. Subcycling in the
GeoModel has been massively improved over earlier versions. Problems that formerly
took 2000 code steps to converge to an acceptable accuracy
‡
can now be run to the same
pointwise accuracy in only 10 apparent hostcode steps (internally, the GeoModel still
runs ~2000 subcycle steps, but this improvement in allowable hostcode step size is
essential in fieldscale finiteelement simulations). An independent research effort [16]
is nearing completion in which an implicit integration scheme has been developed for a
simplified version of the GeoModel, but verification of its orderofconvergence is not
yet completed.
* The term “orderofconvergence” is preferred over “orderofaccuracy” because a converged solu
tion is never necessarily a correct solution. For example, if the return direction is incorrect in a clas
sical predictorcorrector plasticity scheme, then the algorithm will converge, but to the wrong
result. The GeoModel, by the way, does not use a return method — it explicitly integrates the equa
tions, using subcycling to assist with the change of governing equations upon yielding.
† Of course, higherorder integration is also an option, but the total computational overhead some
times exceed that of a wellwritten subcycling algorithm. Moreover, higherorder integration algo
rithms are notoriously difficult to maintain when the governing equations themselves are being
revised during parallel development of the physical theory.
‡ At this stage in our ongoing verification process, we are using socalled “viewgraph” assessment of
accuracy in which solutions are compared visually by plotting them together. This easy assessment
method simply bounds the discrepancy between two seemingly overlaying plots to be less than dif
ferences perceptible to the human eye relative to the size of the graph.
Software “confidence building” activities
86
Preliminary convergence testing. Let
denote a fineresolution solution curve (e.g. stress as
a function of strain) corresponding to evenly
spaced time steps. Let denote a coarser solution
corresponding to time steps. In both cases, the
continuous curves and are here regarded
as piecewise linear interpolations between discrete
function values at the timesteps. Both curves are
normalized by the peak value of .
The “integrated discrepancy” is defined
(7.1)
The “pointwise discrepancy” is defined
(7.2)
Figure 7.3 shows that the basic algorithm without
subcycling is firstorder accurate. Figure 7.2 shows
that subcycling makes the solution nearly pointwise
converged ( ) even for inordinately
coarse calculations of 1 to 3 time steps. This highly
desirable behavior can be of paramount importance
when the model is run in finiteelement codes that
take large steps.
2 time steps
2
2
time steps
2
3
time steps
2
4
time steps
2
5
time steps
Figure 7.2. Subcycling test for hydro
static loading. The ordinate is the first
stress invariant and the abscissa is volu
metric strain . Red is without subcycling.
Green is with subcycling. Black, , cor
responds to time steps.
I
1
ε
v
F x ( )
2
15
F x ( )
N
f x ( )
n N <
F x ( ) f x ( )
F x ( )
F x
K
( ) f x
K
( ) –  
2
K 1 =
N
∑
\ .


 
1 2 /
F x
k
( ) f x
k
( ) –  
2
k 1 =
n
∑
\ .


 
1 2 /
discrepancy 2
6 –
≈
log
2
number of cycles ( )
log
2
discrepancy ( )
Figure 7.3. Rate of convergence. The dashed blue line
(shown for reference) has a slope exactly equal to –1.
integrated without subcycling
integrated with subcycling
pointwise without subcycling
pointwise with subcycling
Software “confidence building” activities
87
Verification: singleelement problems (regression suite)
A singleelement problem is one for which the stress and strain fields do not vary in
space. As summarized below, this section presents four different singleelement load
paths, each separately solved using two different GeoModel parameter sets (associative
and nonassociative linear MohrCoulomb — see Appendix B, page B22). To determine
corresponding sets of material parameters for the GeoModel, the linear MohrCoulomb
parameters were used to create simulated data pairs for the limit state and plastic potential
functions. Following the instructions in Appendix A, GeoModel was then fitted to these
data pairs to determine the limitstate parameters and nonassociative
material parameters . The compaction parameters
were selected such that no compaction occurred over the stress
range specified for the load paths. Likewise, hardening and rate dependence were dis
abled. Results are summarized in Figs. 7.4 through 7.7.
Table 7.1: MohrCoulomb Parameters for singleelement verification testing
Parameter
(Material Properties)
Symbol Associative Nonassociative
Young’s Modulus E 31.0 GPa 31.0 GPa
Poisson’s Ratio ν 0.26 0.26
Friction angle φ
Dilation angle
Cohesion 15.7 MPa 15.7 MPa
Dry bulk density 2.34 MPa/m
3
2.34 MPa/m
3
ψ a
1
a
2
a
3
a
4
, , , ,
ψ
PF
a
2
PF
a
1
PF
, , { }
R R
PF
p
0
p
1
p
2
p
3
, , , , , { }
sigy=sigx
sigx
epsz
epsy=0
sigx=0
epsz
sigy=sigx
sigx
epsz
sigy=sigx
sigx
epsz
epsy=0
sigx=0
epsz
epsy=0
sigx=0
epsz
• Simple load paths clearly
demonstrate basic model response
• Run for both full and partial
associativity
• Triaxial Compression
— Activates compression meridian
— Same load path as a real lab test
• Reduced Triaxial Extension
— Activates extension meridian
— Same load path as a real lab test
• Plane Strain
— Activates meridians at Lode
angle(s) (triaxiality) between
compression and extension
Point sigx=sigy epsz
1 0 0.000000
2 0 0.005000
3 0 0.003274
TXC0
Point sigx=sigy epsz
1 0 0.000000
2 20 0.000310
3 20 0.005000
4 20 0.002060
TXC20
Point sigx=sigy epsz
1 0 0.000000
2 100 0.001548
3 100 0.005000
4 100 0.002295
RTX100
Point epsz
1 0.000000
2 0.010000
3 0.008391
PSTRN
Point sigx=sigy epsz
1 0 0.000000
2 0 0.005000
3 0 0.003274
TXC0
Point sigx=sigy epsz
1 0 0.000000
2 20 0.000310
3 20 0.005000
4 20 0.002060
TXC20
Point sigx=sigy epsz
1 0 0.000000
2 100 0.001548
3 100 0.005000
4 100 0.002295
RTX100
Point epsz
1 0.000000
2 0.010000
3 0.008391
PSTRN
29° 26°
ω 29° 14°
S
o
ρ
Software “confidence building” activities
88
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 0.1 0.2 0.3 0.4 0.5 0.
ε
z, %
ε
y
,
%
ε
z
(%)
ε
y
(
%
)
–0.3
0.0
0.0 0.5
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.
ε
z, %
ε
y
,
%
ε
z
(%)
ε
y
(
%
)
–0.5
0.0
0.0 0.5
10
0
10
20
30
40
50
60
0 0.1 0.2 0.3 0.4 0.5 0.
εz, %
σ
z
,
M
P
a
σ
z
(
M
P
a
)
ε
z
(%)
0.0 0.5
0
60
Associative Nonassociative
10
0
10
20
30
40
50
60
0 0.1 0.2 0.3 0.4 0.5 0.
εz, %
σ
z
,
M
P
a
σ
z
(
M
P
a
)
ε
z
(%)
0.0 0.5
0
60
Figure 7.4. Unconfined compression. Exact (black) solutions compared with ALEGRA GeoModel simulation
(pink). (a) lateral strain vs. axial strain (b) Stress vs. strain.
(b)
(a)
SINGLE ELEMENT VERIFICATION RESULTS
UNCONFINED COMPRESSION
Associative Nonassociative
Software “confidence building” activities
89
0.2
0.15
0.1
0.05
0
0.05
0 0.1 0.2 0.3 0.4 0.5 0.
ε
z, %
ε
y
,
%
ε
z
(%)
0.0
–0.2
0.0
0.5
ε
y
(
%
)
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0.05
0 0.1 0.2 0.3 0.4 0.5 0.
εz, %
ε
y
,
%
ε
z
(%)
ε
y
(
%
)
0.0
0.5
0.0
–0.3
0
20
40
60
80
100
120
0 0.1 0.2 0.3 0.4 0.5 0.
ε
z, %
σ
z
,
M
P
a
σ
z
(
M
P
a
)
ε
z
(%)
0
120
0.0
0.5
Associative Nonassociative
0
20
40
60
80
100
120
0 0.1 0.2 0.3 0.4 0.5 0.
εz, %
σ
z
,
M
P
a
σ
z
(
M
P
a
)
ε
z
(%)
0.0
0.5
0
120
Figure 7.5. 20 MPa triaxial compression. Exact (black) solutions compared with ALEGRA GeoModel simulation
(pink). (a) lateral strain vs. axial strain (b) Stress vs. strain.
(b)
(a)
SINGLE ELEMENT VERIFICATION RESULTS
20 MPa TRIAXIAL COMPRESSION
Associative Nonassociative
Software “confidence building” activities
90
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.
ε
z, %
ε
y
,
%
ε
z
(%)
ε
y
(
%
)
0.0
–0.5 0.0 0.2
0.4
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.
εz, %
ε
y
,
%
ε
z
(%)
0.0
0.3
–0.5 0.0 0.2
ε
y
(
%
)
0
20
40
60
80
100
120
0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.
εz, %
σ
z
,
M
P
a
σ
z
(
M
P
a
)
ε
z
(%)
–0.5 0.0 0.2
120
0
Associative Nonassociative
0
20
40
60
80
100
120
0.6 0.5 0.4 0.3 0.2 0.1 0 0.1 0.
εz, %
σ
z
,
M
P
a
σ
z
(
M
P
a
)
ε
z
(%)
–0.5 0.0 0.2
120
0
Figure 7.6. Reduced triaxial extension. Exact (black) solutions compared with ALEGRA GeoModel simulation
(pink). (a) lateral strain vs. axial strain (b) Stress vs. strain.
(b)
(a)
SINGLE ELEMENT VERIFICATION RESULTS
REDUCED TRIAXIAL EXTENSION
Associative Nonassociative
Software “confidence building” activities
91
10
0
10
20
30
40
50
60
0 0.2 0.4 0.6 0.8 1
εz, %
S
t
r
e
s
s
,
M
P
a
sigz, analytic
sigz, numerical
sigy. analytic
sigy, numerical
ε
z
(%)
s
t
r
e
s
s
(
M
P
a
)
σ
z
σ
y
0
0.8
0
60
3
2.5
2
1.5
1
0.5
0
0 0.2 0.4 0.6 0.8 1 1.
ε
z, %
ε
x
,
%
ε
z
(%)
ε
x
(
%
)
0
1
0
–3 1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.
εz, %
ε
x
,
%
ε
x
(
%
)
ε
z
(%)
0
1
0
–1.6
Associative Nonassociative
10
0
10
20
30
40
50
60
0 0.2 0.4 0.6 0.8 1
εz, %
S
t
r
e
s
s
,
M
P
a
sigz, analytic
sigz, numerical
sigy. analytic
sigy, numerical
ε
z
(%)
0
0.8
0
60
s
t
r
e
s
s
(
M
P
a
)
σ
z
σ
y
Figure 7.7. Plane Strain. Exact (black) solutions compared with ALEGRA GeoModel simulation (pink).
(a) lateral strain vs. axial strain (b) Stress vs. strain.
(b)
(a)
SINGLE ELEMENT VERIFICATION RESULTS
PLANE STRAIN
Associative Nonassociative
Software “confidence building” activities
92
Discussion.
In Fig. 7.7, predictions of the GeoModel overlay the analytical results.
*
Though not
yet proved conclusively, we contend that the slight discrepancy apparent in Figs 7.4 and
7.5 arises not from constitutive model errors, but instead from host code errors in the han
dling of stress boundary conditions. The fully straincontrolled problem (Fig. 7.7) exhibits
no significant solution discrepancy. However, moderate error is apparent in Figs 7.4 and
7.5, which involve two stress boundary conditions (the lateral stresses). Like most consti
tutive models, the GeoModel takes strain rate as input and returns updated stress as output.
If the host finiteelement code handles stress boundary conditions improperly, then it will
have slight errors in the strain rate that it sends to the GeoModel, thereby causing predic
tions to deviate slightly from analytical solutions. Such errors are not uncommon when
dynamic finiteelement codes are used to attempt to simulate homogeneous deforma
tions.
†
To reiterate, we believe boundary condition errors in the host code (not the Geo
Model) are responsible for the solution errors, and we anticipate rerunning these
simulations in a true constitutive model driver to verify this claim.
Other singleelement tests. Whenever the source code is changed, we perform
approximately 20 singlefeature singleelement verification checks for each of our regres
sion tests performed under loading that is simple enough to admit analytical solutions.
Fig. 7.8a depicts results from a hydrostatic loading simulation in which plastic volu
metric strain (EQPV) is plotted against pressure (–I1/3). The thick yellow curve in
Fig. 7.8a is the crush curve (Eq. 4.67) determined independently by user specified values
of the GeoModel parameters . The predicted volumetric strain (black line in
Fig 7.8a) is verified to remain zero until the pressure reaches the crush curve, after which
* Precisely quantifying verification error lags far behind other more important constitutive SQA pri
orities, so we will be satisfied here and throughout this chapter with assessing agreement between
computed and analytical results via a socalled “viewgraph metric,” where the error is nebulously
bounded by what can be perceived visually, given the plot size.
† The only way to ensure a precisely homogeneous deformation in a dynamics finiteelement code is
to bypass solution of the momentum equation. More correctly, homogeneous loading requires a
body force identically equal to particle acceleration , making the momentum equation trivial. b
˜
a
˜
1 2 3 4 5 6 7
0.02
0.04
0.06
0.08
0.02 0.04 0.06 0.08 0.1
1·10
8
2·10
8
3·10
8
4·10
8
5·10
8
6·10
8
7·10
8
p
r
e
s
s
u
r
e
p
1
p
3
total volumetric strain
pressure/p
1
p
l
a
s
t
i
c
v
o
l
u
m
e
t
r
i
c
s
t
r
a
i
n
p
l
a
s
t
i
c
v
o
l
u
m
e
t
r
i
c
s
t
r
a
i
n
1
0
–p
3
3 2 1 1 2 3
1
0.5
0.5
quasistatic
dynamic
σ
11
ε
11
Figure 7.8a. Crush curve and hydrostatic loading verifi
cation.
Figure 7.8b. Rate dependence TXE/TXC
ratio verification for triaxial load/unload.
p
1
p
2
p
3
, , ε
v
p
Software “confidence building” activities
93
it drops along the crush curve as it should. The stressstrain curve (inset in 7.8a) unloads
correctly to the userspecified peak strain parameter . For nonhydrostatic loading, we
have verified (trendtest) that shearenhanced pore collapse causes inelasticity to com
mence prior to the pressure reaching the crush curve.
Fig. 7.8b simultaneously verifies the GeoModel’s ability to predict an apparent
increase in strength under dynamic loading and its ability to predict different strengths in
triaxial extension vs. compression. In that problem, pressure dependence of yield was sup
pressed and the TXE/TXC strength ratio was set to 1/2, resulting in a tensile strength half
as large as the compressive strength.
*
Verification: Hendren & Ayier pressurized cylinder
The problem depicted below, solved for subcases of associative and nonassociative
flow, involves a circular tunnel in a MohrCoulomb material loaded in a plane strain con
figuration. A DTRA contractor provided the analytical solutions. The material parameters
are the same as those used in the previous verification problems.
The results in Fig. 7.9 show that the GeoModel solutions agree with analytical results.
Though not confirmed conclusively, the very slight discrepancies are hypothesized to
result from different strain definitions used in the GeoModel and analytical solution (or
possibly host code traction boundary condition issues similar to those discussed earlier).
* The TXE/TXC strength ratio applies to TXE and TXC states at the same pressure. When pressure
dependence of yield is allowed, the TXE peak in a verification test like this will not and should not
be 1/2 the magnitude of the TXC peak.
p
3
+
a
r
p(t)
+
a
r
p(t)
150 MPa
p t ( )
time, t
Geometry
Tunnel
radius
a 1 m 1 m
Far field
radius
b 10 m 10 m
Loading
Internal
pressure
0 0
Far field
pressure
0150 MPa 0150 MPa
p
a
p
b
Software “confidence building” activities
94
4
3
2
1
0
1
2
3
1.0 1.5 2.0 2.5
r/a
S
t
r
a
in
(
%
)
r/a
S
t
r
a
i
n
(
%
)
0.0
0.5
1.0
1.5
2.0
2.5
0.0000E+00 2.5000E+01 5.0000E+01 7.5000E+01 1.0000E+02 1.2500E+02 1.5000E+02
Pressure (MPa)
C
lo
s
u
r
e
(
%
)
Pressure (MPa)
C
l
o
s
u
r
e
(
%
)
0
1
2
3
4
0.0000E+00 2.5000E+01 5.0000E+01 7.5000E+01 1.0000E+02 1.2500E+02 1.5000E+0
Pressure (MPa)
C
lo
s
u
r
e
(
%
)
Pressure (MPa)
C
l
o
s
u
r
e
(
%
)
0
50
100
150
200
250
300
1 2 3 4 5
r/a
S
t
r
e
s
s
(
M
p
a
)
r/a
S
t
r
e
s
s
(
M
P
a
)
12
9
6
3
0
3
6
1.0 1.5 2.0 2.5
r/a
S
t
r
a
in
(
%
)
r/a
S
t
r
a
i
n
(
%
)
Associative Nonassociative
0
50
100
150
200
250
300
1 2 3 4 5
r/a
S
t
r
e
s
s
(
M
p
a
)
r/a
S
t
r
e
s
s
(
M
P
a
)
Figure 7.9. Hendron & Ayier verification problem. Exact (dash) solutions compared with ALEGRA GeoModel
simulation (solid). (a) Tunnel closure vs. farfield pressure (b) Stress components vs. normalized range, (c) strain
components vs. normalized range.
(c)
(b)
(a)
HENDRON & AYIER VERIFICATION RESULTS
Software “confidence building” activities
95
Elastic freefield wave form (finiteelement verification)
This section describes a fieldscale test for verifying installation of the GeoModel in host
finiteelement codes. As emphasized earlier, verifying a constitutive model is distinct from verify
ing its implementation within a finiteelement code. After a constitutive model becomes one of
many components within a much larger finiteelement model, the potential sources of solution
error expand to now include boundary conditions, artificial viscosity, and other aspects of the host
code’s method of solving the partial differential equation that governs momentum balance.
As indicated in Fig. 7.10, a time varying velocity
(identical to the one later discussed on page 107) was
applied at the boundary of a spherical cavity (radius
204m). The GeoModel’s yield features were disabled
to allow predictions for the velocity at the outer
radius (470m) to be compared with an analytical
elasticity solution.
*
Implementations of the Geo
Model in two finiteelement codes were tested. One
code was unable to reproduce the correct response
because of bugs in roller boundary conditions.
Fig. 7.11 demonstrates that the second code (JAS3D)
was capable of reproducing the analytical solution well enough to suggest that the GeoModel is
performing correctly. While the moderate solution error might be attributable to the analytical
solution’s presumption of small strains, further study (by the code owners, not the GeoModel
developers) is warranted to determine if the solution errors result from underintegration, or some
other aspect of the finiteelement model such as artificial viscosity (both codes’ solutions were
strongly affected by artificial viscosity — default settings for artificial viscosity were insufficient
to reproduce the analytical results). This elasticity verification problem is revisited and generalized
in the plasticity validation test on page 107, where code predictions are compared with data.
* Aldridge’s analytical solution [1] is expressed in terms of integrals that are evaluated numerically
in the frequency domain using independent software provided by Aldridge.
Figure 7.10. Spherical cavity geometry.
roller
roller
JAS3DExplicit Dynamic FEM + GeoModel elastic
AldridgeAnalytical/Numerical*
Figure 7.11. Finiteelement vs. analytical/numerical elastic wave velocity
at 470 meters from a velocity spherical cavity source at 204 meters.
Parameterization (calibration)
96
8. Parameterization (calibration)
Appendix A describes how to characterize a material for the GeoModel. Using these
procedures, several materials have already been fit, as summarized in Fig. 8.1 (compare
these with Figs. 1.1 and 3.3). GeoModel parameters for these materials are in Appendix B.
Model PredictionVersus LimitState DatafromTriaxial CompressionTestsConducted
onAERADiatomite
0
5
10
15
20
25
30
120 100 80 60 40 20 0 20 40
I1 (MPa)
J
2
1
/2
(
M
P
a
)
Data
Model
2.1MPa
3.1MPa
4.1MPa
13.8MPa
13.8MPa
Opal CT Diatomite Climax Stock Granodiorite
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2000 0 2000 4000 6000 8000 10000 12000
I1 (MPa)
J
2
^
1
/2
(
M
P
a
)
Model
Data
2216 MPa
1527 MPa
1161 MPa
439 MPa
Climax Stock Granodorite
Antelope Tuff
0
5
10
15
20
25
30
35
40
200 150 100 50 0 50
I1 (MPa)
J
2
1
/2
(
M
P
a
)
Data
Model
0 MPa
5 MPa
10 MPa
25 MPa
70 MPa
Antelope Tuff
Salem Limestone
0
50
100
150
200
250
300
350
400
2500 2000 1500 1000 500 0 500
I1 (MPa)
J
2
1
/2
(
M
P
a
)
Data
Model
0 MPa
20 MPa
50 MPa
100 MPa
200 MPa
400 MPa
Salem Limestone
23 MPa Concrete
0
50
100
150
200
250
300
350
400
450
2500 2000 1500 1000 500 0 500
I1 (MPa)
J
2
1
/2
(
M
P
a
)
Data
Model
0 MPa
10 MPa
20 MPa
40 MPa
50 MPa
75 MPa
100 MPa
200 MPa
300 MPa
400 MPa
23 MPa concrete
Conventional Strength Concrete
0
50
100
150
200
250
300
350
2500 2000 1500 1000 500 0 500
I1 (MPa)
J
2
1
/2
(
M
P
a
)
Model
Data
0 MPa
20 MPa
50 MPa
100 MPa
200 MPa
400 MPa
Conventional Strength Concrete
Model VersusDatafromTriaxial CompressionTestsConductedonSidewinder
Tuff
0
50
100
150
200
250
300
200 0 200 400 600 800 1000 1200 1400
I1 (MPa)
J
2
1
/2
(
M
P
a
)
Peak Stress
Model
200MPa
100MPa
50MPa
25MPa
10MPa
5MPa
0MPa
A1=496.83MPa
A2=6.293E41/MPa
A3=481.08MPa
A4=0.0Radians
Sidewinder Tuff
Frozen Soil
d
a
t
a
d
a
t
a
d
a
t
a
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
d
a
t
a
d
a
t
a
d
a
t
a
G
e
o
M
o
d
e
l
d
a
t
a
d
a
t
a
d
a
t
a
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
r
f
a
c
e
G
e
o
M
o
d
e
l
f
i
t
t
o
l
i
m
i
t
s
u
rfa
c
e
I
1
(MPa)
400
0
2000 0
J
2
(
M
P
a
)
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
d
a
t
a
400
0
J
2
(
M
P
a
)
I
1
(MPa)
2000 0
300
0
J
2
(
M
P
a
)
I
1
(MPa)
2000 0
I
1
(MPa)
150
0
I
1
(MPa) 10000
0
I
1
(MPa)
0
80
I
1
(MPa)
100 0
I
1
(MPa)
1200
0
30
0
J
2
(
M
P
a
)
40
0
J
2
(
M
P
a
)
300
0
J
2
(
M
P
a
)
4
0
J
2
(
M
P
a
)
2000
0
J
2
(
M
P
a
)
d
a
t
a
Figure 8.1. Meridional limit curves for some materials already parameterized to the GeoModel.
Parameterization (calibration)
97
Nonlinear elasticity
Figures 8.2 and 8.3 demonstrate the ability of the GeoModel’s nonlinear elasticity fit
ting functions (Eqs. 4.10 and 4.9) to reproduce nonlinear elasticity data. Parameters were
assigned using the leastsquares model calibration tools described in Appendix A.
Figure 8.2. Nonlinear elasticity in shear. This figure shows the GeoModel fit to concrete data [46]
from the unload portion of a triaxial compression test conducted at a confining pressure of 200 MPa. Here,
the principal stress difference is plotted against the principal strain difference, thereby making the slope
equal to twice the shear modulus.
0
50
100
150
200
250
300
350
400
0.19 0.195 0.2 0.205 0.21 0.215
Principal Strain Difference
P
r
i
n
c
i
p
a
l
S
t
r
e
s
s
D
i
f
f
e
r
e
n
c
e
(
M
P
a
)
Data
Model
0
50
100
150
200
250
0.000 0.020 0.040 0.060 0.080 0.100
Volume Strain
H
y
d
r
o
s
t
a
t
i
c
S
t
r
e
s
s
(
M
P
a
)
Data
Model
Figure 8.3. Nonlinear elasticity in hydrostatic loading. This figure shows the GeoModel fit to con
crete data [46] from the unload portion of a hydrostatic compression test from 200 MPa. Here, the slope
equals the bulk modulus.
Parameterization (calibration)
98
Elasticplastic coupling
For many materials, elastic moduli are unaffected by inelastic deformation. However,
Fig. 8.4 shows data for a material whose elastic properties are affected by inelasticity.
Modeling this effect requires using the enhanced moduli fitting functions, Eqs. (4.33) and
(4.34), which permit the shear and bulk moduli to vary with equivalent plastic strain
(determined from data by the residual strain upon unloading to a zero stress, as explained
in Appendix A). For the GeoModel to be considered a good fit to data, the simulated
unloading curves merely need to be parallel to experimental unloading curves (not neces
sarily overlapping unless the data and simulation unload from the same strain).
Figure 8.4. Elasticplastic coupling: deformationinduced changes in elastic moduli (Salem Lime
stone). (a) the tangent bulk modulus can change in response to changes in porosity (i.e., volumetric
plastic strain). (b) Likewise, the tangent shear modulus, especially at low shear stresses, can change in re
sponse to plastic deformation.
600
500
400
300
200
100
0
0.1 0.08 0.06 0.04 0.02 0 0.02
Volume Strain
M
e
a
n
N
o
r
m
a
l
S
t
r
e
s
s
(
M
P
a
)
Data
GeoModel
HYDROSTATIC LOADING
TRIAXIAL LOAD DATA CONVERTED TO SHEAR DATA
(a)
(b)
Parameterization (calibration)
99
Triaxial Compression
Triaxial compression (TXC) and triaxial extension (TXE) testing is integral to param
eterization of the GeoModel. By performing a series of tests, as described in Fig. 3.3, the
limit state (onset of softening) can be mapped out. For example, Fig. A.5(a) on page A6
shows raw data from a a suite of TXC tests, indicating how the peak values in each test
provide one data point on the GeoModel limit surface Fig. A.5(b). Data from a similar set
of experiments for concrete, along with the GeoModel leastsquares fit of Eq. (4.8) are
shown in Fig. 8.5. Figure 8.1 shows similar plots for other materials. Appendix A
describes the leastsquares fitting procedures in more detail.
Recall that triaxial testing normally begins with a hydrostatic “loadup” phase, indicated
by the horizontal red arrow Fig. 8.6a (where ). During the hydrostatic leg, defor
mation is initially nonlinearly elastic until the virgin yield surface is reached, at which
time microscale stress concentrations caused by the presence of pores become too large to
resist elastically. Continuing to push the hydrostatic stress to higher levels results in
inelastic pore collapse with associated hardening (expansion) of the yield surface. In
Fig. 8.6, the target hydrostatic stress state for a given experiment (which marks the transi
tion from the hydrostatic leg to the triaxial leg) was MPa, giving an value of 
1200 MPa. Pressurevolume data taken during the hydrostatic leg may be used to deter
Figure 8.5. Shear failure limit curve compared with concrete data. Ref. [46].
0
50
100
150
200
250
300
350
400
450
0 500 1000 1500 2000
Data
Model
A = 1255.68 MPa
B = 1.93E04 1/MPa
C = 1248.20 MPa
θ
= 0.0 Rad
MPa J
2
MPa I
1
−
J
2
0 =
P 400 = I
1
Parameterization (calibration)
100
mine the GeoModel parameters by following instructions in Appendix A.
Because multiple triaxial experiments must be performed to fully characterize a geologi
cal material, variations observed during hydrostatic loading from different tests can be
used to quantify the material property variability.
After the hydrostatic leg, the triaxial leg (angled red arrow in Fig. 8.6a) commences by
increasing the axial load on the specimen while holding the lateral stress constant. As
explained on page 16, the stress path follows a straight trajectory in the meridional plane
with a slope given by
(8.1)
p
0
p
1
p
2
p
3
, , , { }
Figure 8.6. Progression of the hardening yield surface (family of blue lines) under a triaxial compres
sion test (red path), illustrated with correspondence of the meridional plane to the stressstrain diagram.
The straight red line segments shown show this twostage stress trajectory (hydrostatic loading followed
by triaxial loading) in the meridional plot of (which is proportional to the effective shear) versus .
Shearenhanced dilatation corresponds to reaching a zero local yield slope.
J
2
I
1
(a)
(b)
d J
2
( )
d I
1
( )

1
3
 =
Parameterization (calibration)
101
The transition from hydrostatic to triaxial loading is reflected by a pronounced change
in slope in the stressstrain plot of Fig. 8.6. As the axial stress is increased during the triax
ial leg, the yield surfaces continues to harden outward even more, now further assisted by
the presence of a nonzero stress deviator. In Fig. 8.6 the slope of the yield surface is ini
tially negative at the stress state (i.e., where the straight red load path and curved blue
yield surface lines intersect). Consequently, the outward normal to the yield surface during
this early part of the triaxial phase has both a deviatoric component and a compressive iso
tropic component. When the normal to the yield surface is compressive, the inelastic volu
metric strain will be compressive as well. However, the isotropic component of the yield
surface normal changes direction towards the end of the triaxial leg (i.e., the local slope of
the yield surface changes sign), which means that the inelastic volumetric strain is dilata
tional (expanding) even though all stress components are compressive. The “critical state”
at which the yield surface has a zero local slope on the load path marks the onset of shear
enhanced dilatation. Thus, as illustrated qualitatively in Fig. 8.7 and explained in detail in
Appendix A, triaxial loading is used to determine parameters in the GeoModel that govern
yield surface evolution and cap curvature.
Figure 8.7. Shearenhanced dilatation under compression. The exaggerated schematic shows that
crack faces must overcome surface incompatibilities — they cannot slip over each other without opening
even if they are in compression. Moreover, fragments of broken material can become lodged in the crack
face and cause crack opening by their rotation. Crack kinking (in the direction of Mode I loading) further
contributes to the dilatation associated with crack opening. The dark regions in the micrograph [12] are
pores (which collapse under sufficient pressure). The cracks in inclusions produce “microrubble” that ulti
mately generates, through rotation, inelastic volume increase under shear loading even if all principal
stresses are compressive.
DiGiovanni et al, 2000 DiGiovanni et al, 2000
DiGiovanni & Fredrich [12]
Parameterization (calibration)
102
Physically, an increase in inelastic volume during compression (which is quite commonly
observed for brittle materials) is typically attributed to the growth of microcracks under
shear. As the surfaces of these cracks move relative to one another, the crack must open up
(dilatate) both to overcome geometric incompatibilities in their surface roughness and to
permit crack kinking. This interpretation of shearenhanced dilatation is illustrated sche
matically and through SEM imaging in Fig. 8.7.
Parameterization: Rate dependence
The GeoModel’s relaxation parameters may be determined through a
series of labscale laterallyconfined Kolsky bar tests [Fig. 8.8] in which a sample is sub
jected to uniaxial compression at various strain rates.
Figure 8.9 shows results for a series of Kolsky (split Hopkinson) bar strain rate tests
conducted on unconfined compression specimens of Salem Limestone [17]. The peak
stress in these experiments is used to assign values of the GeoModel relaxation parameters
to properly correlate apparent strength with strain rate. As seen, the unconfined compres
sive strength increases with strain rate and is wellaccommodated by GeoModel theory.
For a discussion of how the data in Fig. 8.9b are used to assign values to ,
see Appendix A.
T
1
… T
7
, , ( )
Axial Confining
Vessel
Radial Confining
Vessel
TieRods
Air Gun
Bar Supports
Axial Confining
Vessel
Radial Confining
Vessel
TieRods
Air Gun
Bar Supports
Figure 8.8. Kolsky (split Hopkinson) bar apparatus used to obtain data in Fig. 8.9
T
1
… T
7
, , { }
Parameterization (calibration)
103
160
140
120
100
80
60
40
20
0
0.012 0.010 0.008 0.006 0.004 0.002 0.000 0.002 0.004 0.006 0.008 0.010
Strain
S
t
r
e
s
s
(
M
P
a
)
100/s
10/s
1/s
0.1/s
1.E05/s
Volumetric
Axial Radial
0
20
40
60
80
100
120
140
160
1.0E05 1.0E04 1.0E03 1.0E02 1.0E01 1.0E+00 1.0E+01 1.0E+02 1.0E+03
Strain Rate (s
1
)
U
c
o
n
f
i
n
e
d
C
o
m
p
r
e
s
s
i
v
e
S
t
e
n
g
t
h
(
M
P
a
)
Data
Model
Figure 8.9. Suitability of the GeoModel overstress ratedependence model to match observed da
ta. (a) Model predictions of stressstrain behavior at various strain rates. (b) Corresponding Kolsky
(Hopkinson) bar unconfined compressive strength as a function of strain rate
Building confidence in the physical theory
104
9. Building confidence in the physical theory
In a weak sense, demonstrating (as we did in the previous chapter) that the GeoModel
is capable of being parameterized from controlled laboratory data lends some credibility
to the physical foundations of the model. However, truly validating a model after its
parameters have been determined from standard laboratory data requires showing that the
model can, without any change in precalibrated parameters, predict material response
under different (noncalibration) loading scenarios. Ideally, to validate a constitutive
model (not its implementation into a host finiteelement code), one would prefer to com
pare the model predictions against test data from homogeneous loading experiments that
were different from the homogeneous loading tests conducted for calibration.
Thorough testing of any material constitutive model along with its implementation in a
host code must, of course, include simulation of applications for which model predictions
can be compared with structural response measurements. The goal is to assess the degree
to which the integrated model (i.e., its installation into a host code) is capable of predicting
material system response to nontrivial loading scenarios.
Parameterization entails fitting to a subset of discrete points in parameterization data
tables (e.g., as described on page 99, the limit function is parameterized by using only the
peak stress values, not all values measured in the test). Model validation therefore
includes assessing the fitted model’s ability to interpolate well between other points in
these stressstrain response curves (i.e., points that were not used in calibration). Similarly,
the GeoModel’s rate dependence parameters are determined by using only the peak stress
values in Kolsky bar experiments. Therefore, the model’s ability to match the other data
points in those experiments is a validation test.
In addition to merely ensuring that all data in parameterization tests are well modeled,
a better validation test should exercise the model in application domains in which multiple
physical mechanisms are acting simultaneously. The goal is to assess whether or not the
GeoModel parameterization instructions in Appendix A can lead to a highquality set of
material model properties that are predictive in general loading scenarios. This chapter
describes some validation problems that have been studied to date. In all cases, these prob
lems were run using only the single GeoModel parameter set obtained from calibrating to
other data for the material — no parameter adjustments were made to improve model
agreement for these tests.
Building confidence in the physical theory
105
Postcalibration Triaxial loading
The nonlinear elasticity Eqs. (4.9) and (4.10) are parameterized from shear and hydro
static unloading data. The shear limit function Eq. (4.39) parameterized through peak
states in triaxial testing. The crush curve Eq. (4.67) parameterized through purely hydro
static testing. Once parameterized in this way, the GeoModel may be applied to predict the
irreversible plasticallyhardening stressstrain response at a variety of other stress paths.
For example, Fig. 9.1 compares GeoModel predictions with simple triaxial data at various
confining pressures (these simulations all use a single set of GeoModel parameter values).
Figure 9.1. (a) GeoModel prediction of uniaxial strain loading. (b) GeoModel predictions of triaxial
stressstrain response at various confining pressures. From Ref. [46]. The ability of a single parameter
set to agree so remarkably well with this suite of data is a validation.
0
200
400
600
800
1000
1200
0.05 0 0.05 0.1 0.15 0.2
Strain
A
x
i
a
l
S
t
r
e
s
s
(
M
P
a
)
Data
Model
400MPa
200MPa
100MPa
50MPa
Axial Radial
Strain
A
x
i
a
l
S
t
r
e
s
s
(
M
P
a
)
(b)
0
200
400
600
800
1000
1200
0.00 0.05 0.10 0.15 0.20 0.25
Uniaxial Strain
A
x
i
a
l
S
t
r
e
s
s
(
M
P
a
)
Data
Model
Uniaxial strain
A
x
i
a
l
S
t
r
e
s
s
(
M
P
a
)
(a)
Building confidence in the physical theory
106
Fieldscale penetration
Figure 9.2 shows results from a pretest prediction of depth of penetration and dis
placement histories. In tests like these, material constitutive models are often considered
“above average” if they are predictive within 20%. As seen, the GeoModel performed
exceptionally well by this metric.
0.0
0.5
1.0
1.5
2.0
100 150 200 250 300 350 400
Projectile Striking Velocity (m/s)
D
e
p
t
h
o
f
P
e
n
e
t
r
a
t
i
o
n
(
m
)
Measured
Predicted
Projectile Striking Velocity (m/s)
D
e
p
t
h
o
f
P
e
n
e
t
r
a
t
i
o
n
(
m
)
X
Y
X
Y
2000.0
0.0
2000.0
4000.0
6000.0
8000.0
10000.0
0.0 1.0 2.0 3.0 4.0 5.0
Time (ms)
D
e
c
e
l
e
r
a
t
i
o
n
(
G
's
)
SCE/PRONTO 3D (2500Hz)
Data
2000.0
0.0
2000.0
4000.0
6000.0
8000.0
10000.0
0.0 1.0 2.0 3.0 4.0 5.0
Time (ms)
D
e
c
e
l
e
r
a
t
i
o
n
(
G
'
s
)
SCE/PRONTO 3D (2500Hz)
Data
2000.0
0.0
2000.0
4000.0
6000.0
8000.0
10000.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Time (ms)
D
e
c
e
l
e
r
a
t
i
o
n
(
G
's
)
SCE/PRONTO 3D (2500Hz)
Data
2000.0
0.0
2000.0
4000.0
6000.0
8000.0
10000.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Time (ms)
D
e
c
e
l
e
r
a
t
i
o
n
(
G
's
)
SCE/PRONTO 3D (2500Hz)
Data
(a) (b)
(c) (d)
data
model
d
e
c
e
l
e
r
a
t
i
o
n
(
k
G
’
s
)
10
time (ms)
0
0
5
data
model
d
e
c
e
l
e
r
a
t
i
o
n
(
k
G
’
s
)
10
0
time (ms)
0 5
data
model
d
e
c
e
l
e
r
a
t
i
o
n
(
k
G
’
s
)
10
0
data
model
d
e
c
e
l
e
r
a
t
i
o
n
(
k
G
’
s
)
10
0
time (ms)
0 6
time (ms)
0 6
Figure 9.2. After the GeoModel was fitted to laboratoryscale material property tests, it was used to
predict projectile penetration depth using spherical cavity expansion analysis.
Building confidence in the physical theory
107
Free field wave form for spherical shock loading
Figure 9.4 shows displacement and velocity his
tories for an underground test in which a wave propa
gates from an explosive point source. A measured
velocity history at a point 204 meters from the source
was used as the velocity boundary condition of the
simulation. The goal was to predict a second mea
sured velocity history at a point 470 meters from the
source (see Fig. 9.3). Unlike the similar elasticity
verification test described on page 95, this validation
test used elasticplastic fieldscale parameters that
were determined by applying the GeoModel’s cali
bration procedures described in Appendix A (backfitting alteration of these independently deter
mined calibration parameters was disallowed, as should be the case for any validation test).
The GeoModel (along with its implementation within the host finiteelement code, JAS3D)
comes far closer to matching data than simpler models such as nonhardening Von Mises plasticity,
indicating that the GeoModel’s advanced physical features (especially pressure sensitivity) are
important. Because the GeoModel falls short of a compelling agreement with data, further study is
warranted. Unlike the verification study presented on page 95, this validation simulation is over
predicting peak velocity. As was the case in the verification study, code predictions were strongly
affected by artificial viscosity, suggesting that disagreement with data might be rooted in host code
problems as much as shortcomings of the GeoModel. Of course, natural spatial variability of in
situ rock (which is neglected in the simulation) may play a role, as might the response time of
gauges used to acquire the data, or myriad other possible error sources.
This concludes our overview of preliminary validation testing. More extensive model val
idation activities (now underway) require publication of this theory and user’s guide to pro
vide a resource document that allows analysts to apply the GeoModel properly.
Figure 9.3. Spherical cavity geometry.
roller
roller
Figure 9.4. GeoModel + JAS3D finiteelement prediction (blue) vs.
measured (red) velocity and displacement at 470 m. from a spherical cavity
velocity source at 204 meters.
Closing Remarks
108
10. Closing Remarks
The preceding chapters have attempted to elucidate the physical foundations and
domain of applicability of Sandia’s GeoModel. This model was developed in response to
the need for a predictive model that could be used for a wide range of applications while
maintaining numerical tractability in the context of Sandia’s solid mechanics finite
element software. Three key applications for this work are in projectile penetration
research, analysis of hard and deeply buried targets, and reservoirscale modeling of
formation compaction caused by pore pressure drawdown during oil or gas production.
With the emergence of the capability to simulate the largescale mechanical behavior of
complex geosystems by virtue of recent advances in software and hardware, Sandia
recognized the need to enhance the material modeling capabilities for geomaterials. A
largescale, longterm effort was begun that brings together activities in laboratory testing,
basic research, software development, verification, validation, documentation, and quality
assurance. The goal is to provide a rock mechanics predictive capability that fully
accounts for the complex nature of in situ rock masses. While this is an ongoing program,
the GeoModel has reached a stage of maturity that warrants documentation of the effort to
date. The GeoModel is a genuine unification and generalization of simpler models, and as
such it is capable of satisfying the needs of almost any structural application involving
geomaterials. While “firstprinciples” microscale theories have influenced the general
model framework, physically motivated phenomenological judgements about relations
between stress and strain have been given ultimate priority to more accurately match
observed laboratory behavior.
In the laboratory, most rocks exhibit nonlinear elastic deformation upon unloading
and reloading, hysteresis loops, different behavior in extension than in compression,
strainrate sensitivity, pressure dependence, and postpeak softening. Moreover, high
porosity rocks under compressive mean stresses and nonzero deviatoric stresses, involve
a complex interplay of deformations from competing mechanisms including pore collapse
and microcrackmicrovoid development, which occur simultaneously allowing
macroscopic prefailure dilatation to occur even as pores continue to collapse. In addition,
the strainrate sensitivity of some porous rocks depends on the predominant deformation
mechanism, e.g., the strainrate sensitivity of shear and extensional failure is different
from that of pore collapse, and the strainrate sensitivity itself may be pressure dependent.
This report summarizes the progress toward achieving a realistic rock constitutive theory
that can be calibrated via standard laboratory experiments and is numerically tractable for
massively parallel calculations using tens of millions of threedimensional finite elements
and loading conditions that involve ground shock and other nonquiescent processes.
While the GeoModel has achieved many of the stated objectives, there is still much to be
accomplished. Current modelenhancement activities (to be documented in future reports)
include pressure dependence of the extension/compression strength ratio, softening and
failure (fracture), anisotropic jointed rock behavior (as observed in situ), and natural
spatial variability resulting in localization.
A1
APPENDIX A. Parameterizing the GeoModel
APPENDIX A.
Parameterizing the GeoModel
The following steps describe how to determine values for parameters used in the Geo
Model. In many cases, specialized parameterization software (e.g., HYDROFIT,
SHEARFIT, etc.) was used to perform nonlinear regression to determine optimized param
eter values. This supplemental software, which applies nonlinear least squares fitting is
available from Arlo Fossum. Alternatively, nonlinear optimization is available via the
“Solver” addin of MS Excel.
STEP 1.Use hydrostatic pressure vs. total volumetric strain data to obtain
the nonlinear elastic bulk modulus parameters ( , , and ) as well
as the crush curve parameters , , , . Hydrostatic data are used to
obtain the nonlinear bulk modulus parameters in Eq. (4.9),
(A.1)
as well as the parameters in the crush curve, Eq. (4.68),
, where (A.2)
First express data as a column or space delimited table with the first column being
total volume strain and the next column , where is the pressure
(positive in compression). Determine the parameters in Eq. (A.1) using Fossum’s
HYDROFIT program* (which assumes linear elasticity when ). The HYDROFIT
program employs standard nonlinear regression fitting procedures [35] and includes
parameter sensitivities [i.e., derivatives of Eq. (4.68) with respect to each
parameter]. The input file for the HYDROFIT program contains one line with five
values (initial guesses for the parameters , , , , in that order) and a
second line that contains five integers, each having a value 1 or 0 to indicate
whether that parameter is to be optimized (if not, its value is fixed). The peak
inelastic volume strain, , roughly equals the initial porosity. Refer to Fig. 4.14 for
guidance on how to set initial guesses for the other crush parameters. The HYDROFIT
program outputs optimized values for , , , , , , and .
* HYDROFIT is available upon request. At present, this software presumes linear elasticity when
. If the data show nonlinear elasticity, as in Fig. A.1, HYDROFIT will provide only approxi
mate parameter values. At present, fitting nonlinear parameters is not automated.
b
0
b
1
b
2
p
0
p
3
p
1
p
2
K b
o
b
1
b
2
I
1
 –
\ .
 
exp + =
p
3
ε
v
p
– p
3
e
p
1
p
2
ξ + ( )ξ –
= ξ X p
0
– 3 p P
E
– ( ) = =
ε
v
tot
X 3p = p
X p
0
<
X p
0
<
p
3
p
1
p
2
p
0
b
0
p
3
p
0
p
3
p
1
p
2
b
0
b
1
b
2
A2
APPENDIX A. Parameterizing the GeoModel
volume strain
p
r
e
s
s
u
r
e
(
M
P
a
)
elastic
unloading
p
3
inelastic
loading
initial elastic
loading
(not visible)
Figure A.1. Ideal hydrostatic parameterization data. The unloading portion is used to obtain the
elastic bulk modulus parameters , , and . With these parameters, the HYDROFIT program con
verts the loading portion of the data to a crush curve (see Fig. 4.14), and employs nonlinear regression
to obtain the crush parameters , , , and . For this material (a ceramic powder, zircoa), the ini
tial elastic loading curve is so small in comparison to the scale that it is not visible.
b
0
b
1
b
2
p
0
p
1
p
1
p
2
Figure A.2. Lesspreferable hydrostatic parameterization data. Like the data for the materi
al shown in Fig. A.1, this hydrostatic compression test (for frozen soil) was conducted nearly to
full pore collapse, but the loading curve shows signs of material creep, which is not included in
the GeoModel.
A3
APPENDIX A. Parameterizing the GeoModel
STEP 2.Use triaxial compression data to obtain the shear modulus
parameters , , and . This step determines the parameters in Eq. (4.10),
(A.3)
Arrange the triaxial data as a twocolumn table, the first column being the axial
strain and the second column being the stress difference . Then use
nonlinear least squares regression analysis to obtain the shear modulus parameters
, , and .
As illustrated in Fig. 3.3 in the main text, triaxial testing is typically performed as a
twostage process in which the material is first compressed hydrostatically to a
given pressure. Then, during the second (triaxial) leg, the lateral stress is held fixed
while the axial stress is varied. Only the elastic unloading data should be used for
determining the nonlinear elastic shear modulus parameters. As explained on
page 28, the plot of stress difference vs. axial strain will have a slope
equal to Young’s modulus . Rather than directly using Young’s modulus as a
userspecified material parameter, the GeoModel requires the shear modulus .
Recall that the nonlinear elastic bulk modulus was found previously in step 1.
The SHEARFIT program computes the shear modulus from and by using the
standard linear elasticity formula, , cited in Eq. (4.15). If fitting to rock
g
0
g
1
g
2
G g
o
1 g
1
exp g
2
J
2
1 2 /
– ( ) –
1 g
1
–
 =
ε
A
σ
A
σ
L
– ( )
g
0
g
1
g
2
Figure A.3. Data representative of a triaxial compression parameterization test. The slope of the
unloading curve is the nonlinear tangent Young’s modulus. Through standard moduli conversion formu
las from linear elasticity, the previously determined bulk modulus parameters are used to obtain the
elastic shear modulus parameters , , and . g
0
g
1
g
2
250
200
150
100
50
0
0.1 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0
23 MPa Concrete
Triaxial Compresion Test
75 MPa Confining Pressure
inelastic
loading
elastic
unloading
initial elastic loading
can be difficult to distinguish
from onset of inelasticity.
σ
A
σ
L
– ( ) ε
A
E E
G
K
G K E
G
3KE
9K E –
 =
A4
APPENDIX A. Parameterizing the GeoModel
data, Poisson’s ratio, , is typically in the neighborhood of 0.2 (this
is a useful “sanity” check).
STEP 3.(optional) maintain a record of all peak stress states ever measured
for every available quasistatic loadtofailure experiment ever
performed for the material of interest. The softening threshold (peak limit)
envelope is the boundary of any and all stress states quasistatically achievable for
the material, including both elastically obtainable stress states and stress states that
can be reached only through inelastic deformation. Unlike a yield surface, which is
the boundary of elastically obtainable stresses (and which therefore will, in general,
evolve as the microstructure is altered in response to inelastic deformation), the
limit envelope is fixed in time (see Fig. 1.1 on page 5). All achievable yield surfaces
(an infinite set) are contained within the single limit envelope. Characterizing the
limit envelope requires numerous different experiments. Typically, each individual
experiment has precisely one stress state at which the second stress invariant
achieves a peak value. If the material softens before rupturing, the value of at
failure might be lower than . It is the peak that is of interest, not the post
softening value at failure.
For every available quasistatic loadtofailure experiment, find the stress state at
which is larger than for any of the other stress states in that experiment.
Construct a table of data triplets , where and
are the values of and at the stress state for which is at its peak
value. The number of entries in your peak stress table will equal the number of
experiments run to failure. The goal here is to gather sufficient data to parameterize
both the Lode angle function and the shear limit envelope in Eq. (4.39),
, where (shear limit function). (A.4)
ν 3K E – ( ) 6K ( ) ⁄ =
J
2
J
2
J
2
peak
J
2
I
1
@peak
J
2
peak
θ
@peak
, , ( ) I
1
@peak
θ
@peak
I
1
θ J
2
Γ θ ( )
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– ( ) a
4
I
1
+ = I
1
I
1
– =
A5
APPENDIX A. Parameterizing the GeoModel
STEP 4.Use peak stresses from a family of triaxial compression tests to
determine the shear failure envelope parameters , , , and .
Recall that the function represents a peak shear limit envelope in the meridional
plane for which is plotted against for triaxial compression (TXC) stress
states. Thus, the set of all stress states ever observed in TXC (Lode angle
) must fall below the curve , modulo experimental scatter.
Stated differently, this curve defines boundary of all stress states that ever have been
(or ever can be) observed in quasistatic TXC loading. A sufficient number of TXC
experiments must be conducted under various confining pressures so that the
bounding surface begins to take form. On other meridional planes (i.e., at other
Lode angles), the GeoModel theory presumes the bounding curve is adequately
described by , which simply means that the GeoModel
presumes that the limit function at nonTXC Lode angles is simply a scalar multiple
of the TXC function. Once the function has been determined in this
parameterization step, the peak stress data at other Lode angles will be used later to
determine the proportionality function.
Using only the TXC ( ) data create a scatter plot of all everachieved TXC
values of and . A scatter plot of all TXC stress data measured at various
confining pressures might look somewhat as sketched below:
In Fig. A.4, the peak data pairs (darkened dots) correspond to the
values from your table of peakstress invariants collected in STEP 3. A suite of actual
TXC experiments for porcelanite is shown in Fig. A.5, along with further
a
1
a
2
a
3
a
4
F
f
J
2
I
1
θ +30° = J
2
F
f
I
1
( ) =
J
2
F
f
I
1
( ) Γ θ ( ) ⁄ =
F
f
Γ
θ 30° =
J
2
I
1
J
2
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– ( ) a
4
I
1
+ = =
I
1
Figure A.4. A family of TXC tests conducted to failure. The boundary of data
points defines the function. All other (subpeak) data points fall below this line.
Plots like the are shown for various materials in Fig. 8.1.
F
f
J
2
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– ( ) a
4
I
1
+ = =
θ
@peak
30° =
A6
APPENDIX A. Parameterizing the GeoModel
illustration of how discrete peak points from these experiments are transferred to a
limit surface meridional plot for fitting to Eq. (A.4)
Once enough TXC experiments have been conducted for a welldefined shear limit
boundary to emerge, the next step is to determine values of , , , and that
best fit the function to this boundary.
Given triaxial stress difference vs. axial strain data, Eqs. (3.28) and
(3.29) show that value of at peak is given by and the value
of (i.e., the trace of the stress) at peak equals , where
is the (constant) lateral confining pressure. The data pairs from TXC
experiments (i.e., those for which ) may be fed into a nonlinear
regression parameterization program, SHEARFIT (available upon request) to
Figure A.5. TXC stressstrain plots and extraction of their peak values to construct the meridional
limit curve (data are for porcelanite). (a) Each shear stress vs. axial strain plot has exactly one
peak value, as labeled. The value of at this peak is found by applying Eq. (3.28),
with and equal to the lateral confining pressure for the test. (b) The peak states are
transferred to a plot of vs. for fitting to the meridional limit function.
J
2
I
1
I
1
3τ 3σ
L
+ =
τ J
2
peak
= σ
L
J
2
I
1
F
f
(a)
(b)
a
1
a
2
a
3
a
4
F
f
σ
A
σ
L
– ε
L
J
2
J
2
peak
σ
A
σ
L
–
peak
3 ⁄ =
I
1
I
1
@peak
3σ
L
σ
A
σ
L
– ( )
peak
+ = σ
L
I
1
@peak
J
2
peak
, ( )
θ
@peak
+30° =
A7
APPENDIX A. Parameterizing the GeoModel
optimize the parameters in Eq. (A.4). To use this program, create an input file
containing, in the first line, your initial guesses for and, in the second
line, four logical integers (0 or 1) to indicate whether or not to optimize on the
corresponding parameter. Refer to Fig. 4.1(a) to help decide appropriate initial
guesses. If the data suggest a linear envelope at high pressures, then your first guess
for should be an approximation of the slope of this envelope. If in doubt, take
(i.e., assume the data asymptote to a constant value as pressure goes to
infinity). Eyeball the data to set your first guess for equal to the zeropressure
value of the linear asymptote line (extrapolate visually if necessary), set to equal
minus your best estimate for the actual ordinate intercept of the lowpressure
data (again, extrapolate if necessary). Finally, set to equal an estimate for the
initial (lowpressure) slope of the data divided by . Using these initial guesses,
the SHEARFIT program applies nonlinear regression and outputs values for the ,
, , and parameters.
BEWARE: In typical TXC experiments, all principal stresses are compressive,
making it is possible that SHEARFIT will return a meridional fit to the available data
that corresponds to a shear limit envelope that does not corral the origin (implying
nonphysically that zero stress is “unachievable”). If this occurs, you might want to
append your table of observed data with an entry , where
is your best estimate for the theoretical hydrostatic tensile strength of the
material (therefore is a negative number). Include this entry multiple times
if necessary to force it to have greater weight in the nonlinear regression (or,
preferably, perform more experiments to obtain a larger number of real data points
at low values of ).
STEP 5.Use peak stresses from a family of triaxial extension tests to
determine the extension to compression ratio . As was done above for
TXC tests, construct a table of data pairs corresponding to peak
attained stresses in triaxial extension (TXE). These are the data pairs in your peak
data table from STEP 3 that correspond to . For each of these TXE
data pairs, compute
(A.5)
where is the TXC shear limit function parameterized in the previous step. Of
course, each TXE experiment is likely to result in slightly different values for .
At present, the GeoModel presumes that the TXE/TXC ratio is constant.
a
1
a
2
a
3
a
4
, , ,
a
4
a
4
0.0 =
a
1
a
3
a
1
a
2
a
3
a
1
a
2
a
3
a
4
I
1
@peak
J
2
peak
, ( ) I
1
limit
– 0 , ( )
I
1
limit
I
1
limit
–
I
1
ψ
I
1
peak
J
2
peak
, ( )
θ
@peak
30° – =
ψ
J
2
peak in TXE
F
f
I
1
@ peak in TXE
( )
 =
F
f
ψ
ψ
A8
APPENDIX A. Parameterizing the GeoModel
Therefore, set the GeoModel parameter equal to the average of each computed
using Eq. (A.5) for each available TXE experiment.
It is possible that the data might imply that the strength ratio must vary with
pressure, but the GeoModel presently assumes that is the same at all pressures.
Thus, until the GeoModel is enhanced to support pressurevarying strength ratios,
modeling errors must be managed by measuring the parameter at a confining
pressures in the neighborhood where the GeoModel is likely to be applied. Similar
statements can be made regarding any and all parameters used to define the
GeoModel yield function.
Sometimes, it might be impractical — or overly expensive — to obtain TXE data.
In this case, an engineering approach for estimating presumes that it obeys the
same coupling to the meridional profile slope as predicted in classical Mohr
Coulomb theory. Using Eq. (A.4) to set the TXC slope in Eq. B.19 in Appendix B,
an estimate for the pressurevarying strength ratio is
(A.6)
In the future, we hope to thoroughly explore the merits of this formula, which
correlates the strength ratio with the TXC meridional slope. At present, the
GeoModel presumes the strength ratio is constant. If, however, the above
correlation formula can be substantiated, then we will likely incorporate it into
future releases, thereby possibly eliminating the need for a userspecified
altogether. Until then, you may evaluate the above formula at a value of in the
neighborhood where you plan to apply the GeoModel to obtain a reasonable
estimate for .
ψ ψ
ψ
ψ
ψ
ψ
ψ
1
1 3 a
4
a
2
a
3
e
a
2
I
1
–
+   +
 =
ψ
I
1
ψ
A9
APPENDIX A. Parameterizing the GeoModel
STEP 6.Determine the appropriate Lode function option J3TYPE. Recall that the
GeoModel’s function defines the shape of the octahedral profile shape for
Lode angles spanning the range from TXE ( ) to TXC ( ). The
GeoModel parameter J3TYPE (see page 45) dictates the functional form to be used
for the function. Nontriaxial data are difficult to acquire. If no such data are
available, you will need to use engineering judgement as to an appropriate choice
for J3TYPE. This parameterization step aims to guide the choice in the happy
circumstance that nontriaxial data are available.
As illustrated in Fig. 4.10, the GeoModel’s function is defined to equal 1 in TXC
and in TXE. Large values of correspond to small shear strengths. In this
model parameterization step, all available peakstate data obtained in nontriaxial
loading paths are considered to help decide an appropriate choice for the J3TYPE
option. Looping over your table (collected in STEP 3) of “allobserved” peak stress
invariant triplets , especially those at nontriaxial states,
create a new twocolumn table of data pairs, where
(A.7)
By comparing a scatter plot of these data to the graphs in Fig. 4.10, an
appropriate choice for J3TYPE should be more clear. To assist in the decision, it
might be easier to instead scale the ordinate as shown in Fig. A.6. By overlaying
data with the family of plots in Fig. A.6 an appropriate choice for J3TYPE should be
apparent, as illustrated in Fig.A.7. If nontriaxial data are unavailable (a common
problem), select J3TYPE=1 if the material is judged to be moderately ductile;
otherwise, select J3TYPE=3.
Γ θ ( )
θ 30° – = θ +30° =
Γ
Γ
1 ψ ⁄ Γ
I
1
peak
J
2
peak
θ
peak
, , ( )
θ Γ , ( )
Γ
F
f
I
1
@peak
( )
J
2
peak
 =
θ Γ , ( )
A10
APPENDIX A. Parameterizing the GeoModel
0.4 0.2 0.2 0.4
0.2
0.4
0.6
0.8
1
0.4 0.2 0.2 0.4
0.2
0.4
0.6
0.8
1
0.4 0.2 0.2 0.4
0.2
0.4
0.6
0.8
1
1.2
1.4
TXE SHR TXC
0
π
6
 –
π
6

Γ θ ( ) 1 –
1 ψ ⁄ 1 –

TXE SHR TXC
0
π
6
 –
π
6

Γ θ ( ) 1 –
1 ψ ⁄ 1 –

TXE SHR TXC
0
π
6
 –
π
6

Γ θ ( ) 1 –
1 ψ ⁄ 1 –

Figure A.6. The Lode function information originally shown in Fig. 4.10 of the main report, now dis
played with a transformed ordinate. Knowing that in TXE and in TXC, it makes sense
to scale the ordinate as shown so that, regardless of the value of , the scaled ordinate equals 1 in TXE
and 0 in TXC. With this scaling, all Gudehus Lode functions overlap; the others vary with .
Γ 1 ψ ⁄ = Γ 1 =
ψ
ψ
0.4 0.2 0.2 0.4
0.2
0.4
0.6
0.8
1
1.2
1.4
Γ 1 –
1 ψ ⁄ 1 –

θ
TXE
(a moderate collection
of data is likely to be
available)
Nontriaxial
(very little data likely
to be available)
TXC
(a “large” collection
of data is likely to be
available)
Figure A.7. A sketch of how data might be distributed at nontriaxial states. In this contrived
example, most of the data at nontriaxial states falls in the red WillamWarnke region and therefore
J3TYPE=2 would be appropriate for this material. In practice, data at nontriaxial Lode angles are
rarely available. In this case, the user must resort to engineering judgement to decide which J3TYPE
option to select.
A11
APPENDIX A. Parameterizing the GeoModel
STEP 7.Hardening parameters. In the previous steps, we determined the GeoModel
parameters that define the outer limit surface. No stress state outside this fixed (non
evolving) limit surface can be achieved through any load path. Consequently, the
infinite set of all possible yield surfaces must be contained within this limit
envelope. Unlike the limit surface, which bounds all possible stress states, the yield
surface merely bounds the set of elastically obtainable stress states. Unlike the limit
surface, a yield surface evolves (hardens) through time as a result of microstructural
changes induced in the material under inelastic loading. The initial yield surface is
typically much smaller than the limit surface (see Fig. 1.1 in the main text).
Isotropic hardening permits the initial yield surface to expand on octahedral planes
(by amounts that vary with pressure) up until the limit surface is reached. Kinematic
hardening permits the yield surface to translate in stress space until the limit surface
is reached. Both types of hardening may occur simultaneously.
As a rule, the amount of kinematic hardening relative to total stress is high at lower
pressures. Therefore, kinematic hardening data are best inferred from unconfined
compression tests. In the previous parameterization steps, we determined crush
parameter values , elastic parameters , the limit
surface parameters , and J3TYPE. Now these values should be
used in a finiteelement (or single cell driver) implementation of the GeoModel, to
obtain a simulated table of axial stress vs. volume strain to compare with available
experimental data. The goal is to determine values for the asyetunknown
GeoModel parameters through a systematic simulation sequence. As a first guess,
set the “yettobedetermined” GeoModel parameters as follows:
• offset RN=N=0 (i.e. suppress hardening)
• kinematic hardening parameter HC=1e5
• shape parameter CR=R=10
• plastic potential function parameters for nonassociativity A2PF==A2,
A4PF==A4, RKPF=RK, CRPF=CR (i.e., tentatively assume associativity)
• joint spacing RJS=s=0.0
• rate sensitivity parameters T1 through T7 =0.0 (no rate sensitivity).
Run the finiteelement code for unconfined
compression and output axial stress vs. volumetric
strain (EVOL). It is unlikely that this result will
replicate observed unconfined compression data on
the first try. We ultimately hope to assign values to
the above parameters so that the volume strain will
“turn around” as Fig. 8.6(b) (even though axial strain
increases monotonically, the volume strain turns
around because of the lateral bulking strains).
p
0
p
1
p
2
, , ( ) b
0
b
1
b
2
g
0
g
1
g
2
, , , , , ( )
a
1
a
2
a
3
a
4
ψ , , , , ( )
a
x
i
a
l
s
t
r
e
s
s
volume strain
A12
APPENDIX A. Parameterizing the GeoModel
The phenomenological fitting functions employed in the GeoModel are designed to
extrapolate reasonably well into regions where data are not available, but these
functions are also selected in part for computational tractability and they therefore
serve only as approximations. Consequently, there will certainly be modeling error.
Ideally, one should use regression fitting procedures to select GeoModel hardening
parameters that minimize modeling error relative to available data. We have already
described the programs “HYDROFIT” for regression optimization of the elastic
parameters and SHEARFIT for finding the limit surface parameters, but no similar
fitting software has yet been developed to determine the hardening parameters.
Consequently, for these parameters, a more traditional exploratory manual search
method must be used as described below.
In the following, we will be exploring adjustments of CR, RN, and HC to try to
achieve a strain “turn around” at the correct (observed) strain and stress.
If no turn around is apparent in your simulation, try decreasing CR. Continue to
decrease CR until turn around occurs. If you get no turn around, bring down the
initial yield surface. The intercept on the ordinate on the meridional profile is
located at . Try increasing N to a value no larger than (our initial
guess of presumed that the initial yield surface coincided with the shear
limit surface. By setting N to a nonzero value, we are now permitting kinematic
hardening). You will likely see “turn around” start to occur.
Next try changing HC. Increasing (say, doubling) HC will increase the strain value
at which turnaround occurs. Lowering HC lowers the turnaround strain. If you
continue to have trouble getting turn around, double check that you have correctly
entered the previously determined (known) parameter values.
If the turn around stress is too high, try lowering HC.

Once shearinduced dilatation (turnaround) has been
adequately modeled for unconfined TXC, go then to
highconfinement data. Try changing CR (e.g., from
7 to 10 if computed peak strain is too large).

Go back to unconfined, and work on HC and RN.
Continue to go back and forth until you are satisfied
with both confined and unconfined results.
This concludes our guidance for parameterizing new materials to the GeoModel. Further
revisions of this manual will likely include additional tips and parameterization experi
ment suggestions.
J
2
a
1
a
3
– = a
1
a
3
–
N 0 =
a
x
i
a
l
s
t
r
e
s
s
volume strain
B1
APPENDIX B. Nomenclature and Data Sets
APPENDIX B.
Nomenclature and Data Sets
This appendix contains three tables defining (1) model parameters, (2) plotable vari
ables, and (3) other symbols or acronyms used internally within this manual. In each table,
the first column shows the typeset symbol for the variable. The next column contains the
ASCII string used for the variable in code input files and/or within the source code. In the
SI units column, a “1” indicates that the variable is dimensionless. A “–” indicates that
dimensions vary, while N/A means dimensions are not applicable. The defining equation
(or page number) in the last column of the tables cites the location in this report where the
quantity is defined or discussed.
Reminder: in mechanics, stress and strain are typically taken positive in tension. How
ever, in applications, they are taken positive in compression. To manage this potentially
confusing conflict of conventions, recall
DEFINITION OF THE “OVERBAR” (B.1)
For example, denotes the trace of stress (positive in compression). Therefore,
is positive in compression.
x x – ≡
I
1
I
1
I
1
– ≡
B2
APPENDIX B. Nomenclature and Data Sets
Model Parameters
(User Input)
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
PROP(1)
(B0)
Initial elastic bulk modulus (for intact material if model
ing joints). The tangent bulk modulus is given by
Pa Eq. 4.9
Eq. A.1
PROP(2)
(B1)
High pressure coefficient in nonlinear elastic (intact) bulk
modulus function (see above formula). For linear elastic
ity, set . For nonlinear elasticity, set so that the
Bulk modulus will asymptote to a value at high
pressures.
Pa Eq. 4.9
Eq. A.1
PROP(3)
(B2)
Curvature parameter in nonlinear elastic (intact) bulk
modulus function (see above formula). For linear elastic
ity, set . For nonlinear elasticity, set to a small
value to transition rapidly from the low pressure bulk
modulus to the high pressure modulus. Larger values of
will result in a broader transition range.
Pa Eq. 4.9
Eq. A.1
PROP(4)
(B3)
Coefficient in nonlinear elastic bulk modulus to allow for
plasticityinduced changes in the elastic properties. To
neglect this effect, set . When this parameter is
nonzero, the elastic tangent bulk modulus is computed by
Pa Eq. 4.33
PROP(5)
(B4)
Power in nonlinear elastic bulk modulus to allow for plas
ticityinduced changes in the elastic properties. To neglect
this effect, set . Otherwise, see above formula.
1 Eq. 4.33
PROP(6)
(G0)
Initial elastic shear modulus (for intact material if model
ing joints). The tangent shear modulus is computed by
Pa Eq. 4.10
Eq. A.3
PROP(7)
(G1)
Parameter used to define the elastic (intact) shear modulus
at large shears (see above formula). Specifically, the shear
modulus will asymptote to a value as shear
stress increases. Must be less than 1.0. For linear elastic
ity, set . For the shear modulus to decrease with
shearing, set . For the shear modulus to increase
with shearing, set .
1 Eq. 4.10
Eq. A.3
b
0
K b
o
b
1
b
2
I
1
 –
\ .
 
exp + =
b
1
b
1
=0 b
1
K b
0
b
1
+
b
2
b
2
0 = b
2
b
2
b
3
b
3
0 =
K f
K
b
o
b
1
b
2
I
1
 –
\ .
 
exp + b
3
Exp
b
4
ε
v
p
 –
\ .

 
–
¹ )
´ `
¦ ¹
=
b
4
b
4
0 =
g
0
G g
o
1 g
1
exp g
2
J
2
1 2 /
– ( ) –
1 g
1
–
 =
g
1
g
0
1 g
1
– ( ) ⁄
g
1
0 =
0 g
1
1 < <
g
1
0 <
B3
APPENDIX B. Nomenclature and Data Sets
PROP(8)
(G2)
Curvature parameter in nonlinear elastic (intact) shear
modulus function (see above formula). For linear elastic
ity, set . For nonlinear elasticity small values of
cause the shear modulus to transition rapidly from its ini
tial value to its high shear value . Larger
values of make this transition more gradual.
1/Pa Eq. 4.10
Eq. A.3
PROP(9)
(G3)
Coefficient in nonlinear elastic shear modulus to allow for
plasticityinduced changes in the elastic properties. To
neglect this effect, set . When nonzero, the tangent
shear modulus is computed by
Pa Eq. 4.34
PROP(10)
(G4)
Power in nonlinear elastic shear modulus to allow for
plasticityinduced changes in the elastic properties. To
neglect this effect, set . Otherwise, see above for
mula.
1 Eq. 4.34
PROP(11)
(RJS)
Joint spacing. Set this parameter to zero if the material has
no geological (or rocklike) faults.
meter
PROP(12)
(RKS)
Joint shear stiffness. Set this parameter to zero if the mate
rial has no geological (or rocklike) faults.
Pa
PROP(13)
(RKN)
Joint normal stiffness. Set this parameter to zero if the
material has no geological (or rocklike) faults.
Pa
PROP(14)
(A1)
Constant term in the fitting function for the meridional
profile, , for the ultimate
shear limit surface. Here, and therefore is
three times the pressure. At zero pressure,
while at high pressure (large ),
. Thus, is the vertical intercept of the
linear asymptote, whereas is the vertical intercept
of the limit function itself. These are parameters define the
ultimate limit curve, at which the maximum possible
hardening has occurred and softening is imminent. The
initial onset of yield is described by .
Thus if fitting to data for yield onset, recognize that the
constant term will be lower than by an amount .
Pa Eq. 4.39
Fig. A.4
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
g
2
g
2
0 = g
2
g
0
g
o
1 g
1
– ( ) ⁄
g
2
g
3
g
3
0 =
G f
G
g
o
1 g
1
exp g
2
J
2
1 2 /
– ( ) –
1 g
1
–
 g
3
Exp
g
4
γ
eqiv
p
 –
\ .

 
–
¹ )
¦ ¦
´ `
¦ ¦
¦ ¹
=
g
4
g
4
0 =
a
1
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– a
4
I
1
+ =
I
1
trσ
˜
˜
– = I
1
F
f
I
1
( ) a
1
a
3
– = I
1
F
f
I
1
( ) a
1
a
4
I
1
+ ∼ a
1
a
1
a
3
–
f
f
I
1
( ) F
f
I
1
( ) N – =
a
1
N
B4
APPENDIX B. Nomenclature and Data Sets
PROP(15)
(A2)
Curvature decay parameter in the fitting function for the
meridional profile, . Keep
in mind that , where is pressure. Set
for a linear meridional profile as in MohrCoulomb the
ory. Assign a large value to quickly asymptote to the
highpressure profile slope.
1/Pa Eq. 4.39
Fig. A.4
PROP(16)
(A3)
Parameter in the shear limit meridional fit function,
.
Eq. 4.39
Fig. A.4
PROP(17)
(A4)
Highpressure meridional slope parameter in the fit func
tion, .
1 Eq. 4.39
Fig. A.4
PROP(18)
(P0)
Value of at the onset of pore collapse for hydrostatic
compression of virgin material. This parameter will be
negative because is negative in compression. In the
lexicon of traditional  crush models, this variable
would equal , where is the elastic limit pressure
in hydrostatic compression. In many other publications
about the GeoModel, this parameter is denoted .
Pa Eq. 4.67
Fig. 4.1
4
Eq. A.2
PROP(19)
(P1)
One third of the slope of a porosity vs. pressure crush
curve at the elastic limit. In many other publications about
the GeoModel, this parameter is denoted .
1/Pa Eq. 4.67
Fig.4.14
Eq. A.2
PROP(20)
(P2)
Extra fitting parameter for hydrostatic crush curve data,
used when the crush curve has an inflection point. In
many other publications about the GeoModel, this param
eter is denoted .
1/Pa
2
Eq. 4.67
Fig. 4.14
Eq. A.2
PROP(21)
(P3)
Asymptote (limit) value of the absolute value of the plas
tic volume strain. This parameter is approximately equal
to the initial porosity in the material and may be inferred
from hydrostatic crush data. In many other publications
about the GeoModel, this parameter is denoted .
1 Eq. 4.67
Fig. 4.14
Eq. A.2
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
a
2
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– ( ) a
4
I
1
+ =
I
1
3p – = p a
2
0 =
a
2
a
3
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– ( ) a
4
I
1
+ =
Pa
2
a
4
F
f
I
1
( ) a
1
a
3
e
a
2
I
1
–
– ( ) a
4
I
1
+ =
p
0
I
1
I
1
p α
3P
E
– P
E
X
o
p
1
D
1
p
2
D
2
p
3
W
B5
APPENDIX B. Nomenclature and Data Sets
PROP(22)
(CR)
Shape parameter that allows porosity to affect shear
strength. equals the eccentricity (width divided by
height) of the elliptical cap function, so it is the ratio
(not ) in Fig. 4.15. This parameter affects the stress
level at which dilatation will occur in triaxial compres
sion. If dilatation is occurring too soon, increase the value
of . Decreasing will decrease the influence of poros
ity on shear strength and therefore enhance the effect of
void space creation associated with crack growth. To rep
licate older classical pore collapse models (which initiate
pore collapse only at a critical pressure, regardless of the
level of shear stress), set to a very small number.
1 Fig. 4.15
Eq. 4.69
pg. A11
PROP(23)
(RK)
TXE/TXC (triaxial extension to compression) strength
ratio. Convexity of the yield surface requires that
(or if using J3TYPE=1). Real
materials generally satisfy . Future releases of
the GeoModel will likely allow to be pressuredepen
dent.
1 page 45
Fig. 4.4
Eq. A.5
PROP(24)
(RN)
Offset parameter. Must be nonnegative. Set to
suppress kinematic hardening. For problems with kine
matic hardening, the backstress invariant will not be
permitted to grow any larger than . The initial yield sur
face is defined by , where
describes the shear limit surface (softening threshold).
Roughly speaking, the shear strength can increase by an
amount before softening will commence.
1 Fig. 4.6
Eq. 4.40
page 59
pg. A11
PROP(25)
(HC)
Kinematic hardening parameter. Set and to
suppress kinematic hardening. Otherwise, this parameter
affects how “quickly” the yield surface evolves toward the
ultimate shear failure surface.
Pa Eq. 4.76
pg. A11
PROP(26)
(CTI1)
Tensile cutoff in allowable value of the first stress invari
ant . Value must be positive. If the first invariant (which
is proportional to the negative of pressure) reaches this
cutoff value, the isotropic part of the stress is replaced
with this value. This parameter is available only tempo
rarily to permit reasonable strength predictions in tension.
Current enhancements of this model are focused on more
physically rigorous tensile failure modeling. In principle,
the yield function should limit achievable values of .
This option is available if the yield function is not cutting
off strength at a tensile pressure low enough for the appli
cation at hand.
Pa
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
R
R
a b ⁄
A B ⁄
R R
R
ψ
1 2 ⁄ ψ 2 ≤ ≤ 7 9 ⁄ ψ 9 7 ⁄ ≤ ≤
1 2 ⁄ ψ 1 ≤ ≤
ψ
N
N 0 =
J
2
α
N
F
f
initial
I
1
N , ( ) F
f
I
1
( ) N – = F
f
N
H
H 0 = N 0 =
I
1
cut
I
1
I
1
B6
APPENDIX B. Nomenclature and Data Sets
—
PROP(27)
(CTPS)
Principal stress tensile cutoff. If a predicted principal
stress is found to exceed this value, then it is replaced with
this value. See discussion regarding the state variable
“CRACK” in the next table.
Pa
PROP(28)
(T1)
Primary rate dependence parameter. To specify a constant
intrinsic material response time, set this user input equal
to the characteristic response time. Use the other “T”
parameters to enable dependence on strain rate and pres
sure.
sec Eq. 5.22
PROP(29)
(T2)
Rate dependence parameter. See main text. 1/s Eq. 5.22
PROP(30)
(T3)
No longer used. Set to zero. 1 Eq. 5.22
PROP(31)
(T4)
No longer used. Set to zero. 1/s Eq. 5.22
PROP(32)
(T5)
Rate dependence parameter. See main text. Pa Eq. 5.22
PROP(33)
(T6)
Rate dependence parameter. See main text. sec Eq. 5.22
PROP(34)
(T7)
Rate dependence parameter. See main text. 1/Pa Eq. 5.22
—
PROP(35)
(J3TYPE)
Integervalued control parameter for specifying the
desired type of invariant yield surface:
1  Gudehaus
2  WillamWarnke
3  MohrCoulomb
1 page 45
Fig. A.7
PROP(36)
(A2PF)
Potential function parameter (=A2 for associative). Assign
this parameter in the same way you would assign a value
to , except that this parameter is used to generate the
flow potential surface (i.e., the plastic strain rate will be
normal to the flow potential surface). Be sure to set
equal to if plastic normality is desired.
1/Pa page 48
PROP(37)
(A4PF)
Potential function parameter (=A4 for associative). Assign
this parameter in a manner similar to except that this
parameter will be used to generate the flow potential. Be
sure to set equal to if plastic normality is desired.
1 page 48
PROP(38)
(CRPF)
Potential function parameter (=CR for associative). Flow
potential analog of the yield surface parameter . Be sure
to set equal to if plastic normality is desired.
1 page 48
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
T
1
T
2
T
3
T
4
T
5
T
6
T
7
3
rd
a
2
PF
a
2
a
2
PF
a
2
a
4
PF
a
4
a
4
PF
a
4
R
PF
R
R
PF
R
B7
APPENDIX B. Nomenclature and Data Sets
PROP(39)
(RKPF)
Potential function parameter (=RK for associative). Flow
potential analog of the yield surface parameter .Be sure
to set equal to if plastic normality is desired.
1 page 48
—
PROP(40)
(SUBX)
Subcycle control parameter. If zero, the GeoModel will
select an appropriate subcycle step size. If SUBX is non
zero, then the GeoModel’s default subcycle increment will
be multiplied by 10**SUBX (ten raised to the power). If,
for example, you want the code to decrease its subcycle
size by a factor of 10, then set SUBX=–1. If you want the
subcycle size altered by a factor “x”, then set SUBX to the
base ten log of “x” [ ].
1 —
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
ψ
PF
ψ
ψ
PF
ψ
log
10
x x ln ( ) 10 ln ( ) ⁄ =
B8
APPENDIX B. Nomenclature and Data Sets
Internal State Variables
(Plotable Output)
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
(or page)
SV(1)
[KAPPA]
ELN
EL
The value of at which the meridional yield profile first
branches away from the crack failure surface and begins
to morph into the cap function associated with porosity.
Recalling that , the internal state variable typi
cally will be negative. As a rule of thumb: increasing the
userinput parameter will increase (and therefore
decrease ). Recalling Fig. 4.15, is not the point at
which the meridional profile has a zero slope — it is the
branching location. The zero slope point is reached at a
higher pressure. Isotropic hardening is controlled by the
evolution of . In the GeoModel physics source code,
is denoted by ELN or by EL in the subcycling.
Pa Fig. 4.5
Eq. 4.30
Fig. 4.15
Eq. 4.70
—
SV(2)
[INDEX]
Indicator for isotropic plastic hardening. This flag will
equal zero up until the first time isotropic hardening (i.e.,
evolution of the state variable) occurs. Thereafter, this
flag will equal 1.0 even if the stress later becomes elastic.
1 page 77
SV(3)
[EQDOT]
(Frobenius) norm of input strain rate tensor.
.
1/s page 77
SV(4)
[I1]
First stress invariant (positive in tension). Pa Eq. 3.9
page 77
SV(5)
[ROOTJ2]
Square root of the second stress invariant (always
positive). This the equivalent shear stress in the material.
Pa Eq. 3.10
Eq. 3.13
page 77
SV(6)
[ALXX]
11 component of the backstress Pa page 58
SV(7)
[ALYY]
22 component of the backstress Pa page 58
SV(8)
[ALZZ]
33 component of the backstress Pa page 58
SV(9)
[ALXY]
12 component of the backstress (= 21 component) Pa page 58
SV(10)
[AXYZ]
23 component of the backstress (= 32 component) Pa page 58
SV(11)
[ALXZ]
31 component of the backstress (= 13 component) Pa page 58
κ
I
1
κ κ – = κ
R κ
κ κ
κ κ
κ
ε
·
equiv
L
2
ε
· equiv
ε
·
ij
ε
·
ij
=
I
1
J
2
J
2
α
11
α
22
α
33
α
12
α
23
α
31
B9
APPENDIX B. Nomenclature and Data Sets
SV(12)
[GFUN]
Kinematic hardening decay function, equal to 1.0 initially
and then decays down to 0.0 as the max allowable kine
matic hardening (determined by the shift parameter ) is
approached.
1 Eq. 4.79
page 58
SV(13)
EQP
[EQPS]
Equivalent uniaxial plastic shear strain (conjugate to
). Specifically, , where is
the deviatoric part of the plastic strain rate .
1 Eq. 4.35
SV(14)
[EQPV]
Equivalent plastic volume strain: .
1 page 77
SV(15)
[EL0]
Calculated initial value for (the cap branch value of
). This is not really an internal state variable. Its value
will remain constant throughout the calculation. Keep in
mind: is typically positive and therefore is
typically negative.
Pa Fig. 4.15
SV(16)
[HK]
(Isotropic hardening parameter) Proportionality factor
appearing in the relationship .
Pa
2
Eq. 4.73
SV(17)
[EVOL] Total volume strain, .
1 page 77
SV(18)
[BACKRN]
Square root of the second backstress invariant. This is like
the equivalent shear stress, except applied to the back
stress. The value of BACKRN is not permitted to exceed
the user specified limit value of
Pa Eq. 4.79
—
SV(19)
[CRACK]
Flag that equals 1.0 at the onset of softening whenever the
maximum tensile cutoff has been applied (or when the
limit surface is reached). The geomodel simply replaces
the stress with the cutoff stress (user input CUTPS). Phys
ically, complete loss in loadcarrying ability in the appro
priate direction is actually desired. However, simply
replacing the principal stress with zero at the constitutive
level would result in meshdependencies in host codes that
lack macroscale fracture services. More advanced codes
should examine the internal state variable flag “CRACK”
to determine when this cutoff is being applied (and it is
therefore appropriate to initiate void insertion, element
death or, preferably, more advanced fracture response).
1 page 77
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
(or page)
G
α
N
γ
eqiv
p
J
2
γ
eqiv
p
2
γ
˜
·
˜
p
t d
∫
=
γ
˜
·
˜
p
ε
˜
·
˜
p
ε
v
p
ε
v
p
trε
˜
·
˜
p
t d
∫
=
κ
0 κ
I
1
κ κ κ – ≡
h
κ
κ
·
λ
·
h
κ
=
ε
v
tot
trε
˜
˜
·
t d
∫
J
2
α
N
B10
APPENDIX B. Nomenclature and Data Sets
—
SV(20)
[SHEAR]
Flag equal to 0.0 if the material response is elastic. Other
wise, if positive, the response is plastic. The value ranges
from 0.0 if the trial stress rate is tangent to the yield sur
face to 1.0 if the trial stress rate is normal to the yield sur
face (i.e., pushing directly against it).
1 Fig. 6.1
page 77
SV(21)
[YIELD]
F
Value of the yield function Pa
2
Eq. 4.46
SV(22)
[LODE]
Lode angle in degrees ranging from 30 in triaxial exten
sion to +30 in triaxial compression. The Lode angle is fre
quently denoted β in other publications about the
GeoModel.
deg Eq. 3.40
page 77
SV(23)
[QSSIGXX]
Internal GeoModel variable (quasistatic “low” stress) Pa Eq. 5.10
SV(24)
[QSSIGYY]
Internal GeoModel variable (quasistatic “low” stress) Pa Eq. 5.10
SV(25)
[QSSIGZZ]
Internal GeoModel variable (quasistatic “low” stress) Pa Eq. 5.10
SV(26)
[QSSIGXY]
Internal GeoModel variable (quasistatic “low” stress) Pa Eq. 5.10
SV(27)
[QSSIGYZ]
Internal GeoModel variable (quasistatic “low” stress) Pa Eq. 5.10
SV(28)
[QSSIGZX]
Internal GeoModel variable (quasistatic “low” stress) Pa Eq. 5.10
SV(29)
[DCSP]
Plastic consistency parameter. Eq. 4.55
Eq. 4.83
SV(30)
[QSEL]
Internal GeoModel variable (quasistatic “low” value of
). In Eq. 5.15, stands for any internal state variable.
This is the particular instance for which , which is
the quasistatic value of the isotropic hardening ISV.
Pa Eq. 5.15
SV(31)
[QSBSXX]
Quasistatic backstress. In Eq. 5.15, stands for .
Pa Eq. 5.15
SV(32)
[QSBSYY]
Quasistatic backstress. In Eq. 5.15, stands for .
Pa Eq. 5.15
SV(33)
[QSBSZZ]
Quasistatic backstress. In Eq. 5.15, stands for .
Pa Eq. 5.15
SV(34)
[QSBSXY]
Quasistatic backstress. In Eq. 5.15, stands for .
Pa Eq. 5.15
SV(35)
[QSBSYZ]
Quasistatic backstress. In Eq. 5.15, stands for .
Pa Eq. 5.15
SV(36)
[QSBSZX]
Quasistatic backstress. In Eq. 5.15, stands for .
Pa Eq. 5.15
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
(or page)
f
θ
σ
11
L
σ
22
L
σ
33
L
σ
12
L
σ
23
L
σ
31
L
λ
·
1
Pa s ⋅

κ
L
κ q
L
q
L
κ
L
=
α
11
L
q
L
α
ij
L
α
22
L
q
L
α
ij
L
α
33
L
q
L
α
ij
L
α
12
L
q
L
α
ij
L
α
23
L
q
L
α
ij
L
α
31
L
q
L
α
ij
L
B11
APPENDIX B. Nomenclature and Data Sets
Other symbols
(used only in this report)
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
(or page)
Fourthorder elastic stiffness tensor Pa Eq. 4.4
DELTA
Kronecker delta 1 Eq. 2.4
DEIJPL
Deviatoric plastic strain increment =
1 —
VOLPLAS
Plastic volume strain increment =
1 —
DFDSIG
Derivative of the plastic potential with respect to stress — —
time s —
DT
time increment s —
Young’s modulus Pa Eq. 4.12
The part of the meridional yield profile function associ
ated with microcracks;
Pa Fig. 4.6
Eq. 4.29
FF
Meridional shear limiter function Pa Eq. 4.39
The part of the meridional yield profile function associ
ated with porosity;
1 Fig. 4.5
Eq. 4.29
FC
Meridional nominal yield function (Pelessone function) 1 Eq. 4.32
Eq. 4.30
PHI
Friction angle (for MohrCoulomb theory) 1 Eq. B.3
G
Flow potential function Pa
2
Eq. 4.55
GAMMAP
Octahedral yield shape function (depends on J3TYPE) 1 page 45
, Kinematic hardening modulus tensor Pa Eq. 4.78
HYD Acronym: Hydrostatic loading. The stress is diagonal (no
shears).
page 13
RI1
First stress invariant, . This is positive in ten
sion.
Pa Eq. 3.9
Negative of . This is positive in compression
Pa Eq. 3.9
C
ij kl
δ
ij
∆γ
ij
p
ε
·
ij
p
1
3
ε
·
kk
δ
i j
–
\ .
 
t ∆
∆ε
v
p
ε
·
kk
p
t ∆
t
t ∆
E
f
f
f
f
F
f
N – =
F
f
f
c
f
c
F
c
=
F
c
φ
φ
Γ
H
ij
α
H
˜
˜
I
1
I
1
trσ
˜
˜
=
I
1 I
1
B12
APPENDIX B. Nomenclature and Data Sets
RJ2
Second stress invariant, (never negative).
Geometrically, equals the magnitude of the stress
deviator, and it therefore equals the “length” of the projec
tion of the pseudo stress vector in 3D HaighWestergaard
(stress) space onto the octahedralplane (i.e., is the
Lode radius).
Pa
2
Eq. 3.10
RELJ2ZP
Second invariant of the shifted stress Pa
2
page 8
Eq. 4.22
RJ3
Third stress invariant, . This is positive when
the stress deviator is closer to TXE than to TXC.
Pa
3
Eq. 3.11
Negative of . This is positive when the stress deviator is
closer to TXC than to TXE.
Pa
3
Eq. 3.11
RELJ3ZP
Third invariant of the shifted stress Pa
3
page 8
Eq. 4.22
Poisson’s ratio 1 Eq. 4.12
Mean stress, . The mean stress is positive
in compression. The pressure , which is positive in com
pression is the negative of the mean stress: .
Pa page 11
Pressure = negative of mean stress,
Pa page 11
Eq. 3.12
Porosity (unloaded). 1 Eq. 4.65
PONE
Code parameter equal to 1.0 1 —
PTWO
Code parameter equal to 2.0 1 —
PFOUR
Code parameter equal to 4.0 1 —
Polar rotation tensor 1 Eq. 3.39
Eq. 4.3
mass density kg/m
3
Stress deviator,
Pa Eq. 3.5
The stress tensor. Pa Eq. 3.1
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
(or page)
J
2
J
2
1
2
trS
˜
˜
2
=
2J
2
2J
2
J
2
ξ
J
3
J
3
1
3
trS
˜
˜
3
=
J
3 J
3
J
3
ξ
ν
p
p
1
3
trσ
˜
˜
I
1
3
 = =
p
p p – =
p
p
1
3
trσ
˜
˜
–
I
1
3
 = =
Π
R
˜
˜
ρ
S
˜
˜
S
˜
˜
σ
˜
˜
1
3
 trσ
˜
˜
( )
I
˜
˜
– =
σ
˜
˜
B13
APPENDIX B. Nomenclature and Data Sets
SHR Acronym: Shear loading (one principal value of the stress
deviator is zero and the others are therefore negatives of
each other). SHR applies even when all principal stresses
are compressive — all that matters is the nature of the
stress deviator.
N/A page 18
TXE Acronym: Triaxial extension: Two “lateral” principal
stresses are equal and the distinct eigenvalue is more ten
sile than the lateral stresses. A stress state can be in TXE
even when all principal stresses are compressive — the
axial stress merely needs to be less compressive than the
lateral stresses.
N/A page 15
TXC Acronym: Triaxial compression (two “lateral” principal
stresses are equal and the distinct eigenvalue is more com
pressive than the lateral stresses)
N/A page 15
The Hill tensor, defined to be the deviatoric part of and
therefore given by
Pa
Plastic tangent stiffness tensor Pa Eq. 4.89
Signed equivalent shear stress, . This
equals . It is positive when , and negative
when .
Pa Eq. 3.13
Negative of . This is positive when the stress is closer to
TXC, and negative when closer to TXE
Pa Eq. 3.4
TAU
Characteristic material response time second Eq. 5.4
RLODE
Lode angle associated with the shifted stress tensor
( when is TXC and
when is TXE).
radian
(or deg
in plot
output)
Eq. 3.40
SIN3BTAP
Sine of three times the Lode angle
1 Eq. 3.40
ZETA
Shifted stress tensor, . When kinematic harden
ing is activated, the yield surface origin will be at
instead of at the zero stress origin. The backstress tensor
represents the amount by which the origin has trans
lated.
Pa page 58
Symbol
or
Acronym
ASCII
Name
Name and meaning
SI
units
defining
equation
(or page)
T
˜
˜
S
˜
˜
2
T
˜
˜
S
˜
˜
2
2
3
J
2
I
˜
˜
– =
T
ijkl
τ
τ sign J
2
J
3
,   =
J
2
± J
3
0 ≥
J
3
0 <
τ
τ
τ
θ
ξ
ξ
˜
˜
σ
˜
˜
α
˜
˜
– = θ
ξ
30° =
ξ
˜
˜
θ
ξ
+30° =
ξ
˜
˜
3θ
ξ
sin
θ
ξ
ξ
˜
˜
ξ
˜
˜
σ
˜
˜
α
˜
˜
– =
ξ
˜
˜
0
˜
˜
=
α
˜˜
B14
APPENDIX B. Nomenclature and Data Sets
Datasets
This section cites parameters for some materials that have already been fit to the Geo
Model. Datasets undergo continual refinement as additional material data become avail
able, so consult the model developers to obtain the latest values. This section concludes
with simplified datasets for mimicking simpler classical theories (Von Mises plasticity,
MohrCoulomb theory, etc.)
NOTE TO ALEGRA USERS: The GeoModel must be run using the “Generic EOS”.
Appropriate EOS parameters are quoted at the bottom of each dataset.
Dataset for Salem Limestone
$
$ GeoModel parameters for Salem Limestone
$
B0 = 13.0e9 $Pa
B1 = 42.47e9 $Pa
B2 = 0.4107e9 $Pa
B3 = 12.0e9 $Pa
B4 = 0.021 $Dimensionless
G0 = 9.86e9 $Pa
G1 = 0.0 $Dimensionless
G2 = 0.0 $1/Pa
G3 = 0.0 $Pa
G4 = 0.0 $Dimensionless
RJS = 0.0 $Meters
RKS = 0.0 $Pa/Meter
RKN = 0.0 $Pa/Meter
A1 = 843.02e6 $Pa
A2 = 2.731e10 $1/Pa
A3 = 821.92e6 $Pa
A4 = 1.e10 $Radians
P0 = 314.4e6 $Pa
P1 = 1.22e10 $1/Pa
P2 = 1.28e18 $1/Pa^2
P3 = 0.084 $strain
CR = 6.0 $Dimensionless
RK = .72 $Dimensionless
RN = 12.e6 $Pa
HC = 100000.e6 $Pa
CTI1 = 3.e6 $Pa
CTPS = 1.e6 $Pa
T1 = 4.e4 $sec
T2 = 0.835 $1/sec
T3 = 0.0 $Dimensionless
T4 = 0.0 $1/sec
T5 = 0.0 $Pa
T6 = 3.0 $sec
T7 = 0.0 $1/Pa
J3TYPE = 3 $Dimensionless
A2PF = 0.0 $$$$$ = zero means A2PF defaults to A2 for normality
A4PF = 0.0 $$$$$ = zero means A4PF defaults to A4 for normality
CRPF = 0.0 $$$$$ = zero means CRPF defaults to CR for normality
RKPF = 0.0 $$$$$ = zero means RKPF defaults to RK for normality
SUBX = 0. $Dimensionless
$model 2 generic eos
$ rho ref = 2300. $ kg/m^3
$ tref = 298. $ K
$ ref sound speed = 5400.0 $ m/s
$end
B15
APPENDIX B. Nomenclature and Data Sets
Dataset for Sidewinder Tuff
$ GeoModel parameters for Sidewinder Tuff
$ Units are SI
B0 = 4.0e9 $Pa
B1 = 6.5e9 $Pa
B2 = 0.1e9 $Pa
B3 = 0.0 $Pa
B4 = 0.0 $Dimensionless
G0 = 3.69e9 $Pa
G1 = 0.0 $Dimensionless
G2 = 0.0 $1/Pa
G3 = 0.0 $Pa
G4 = 0.0 $Dimensionless
RJS = 0.0 $Meters
RKS = 0.0 $Pa/Meter
RKN = 0.0 $Pa/Meter
A1 = 496.83e6 $Pa
A2 = 6.293e10 $1/Pa
A3 = 481.08e6 $Pa
A4 = 1.e10 $Radians
P0 = 70.e6 $Pa
P1 = 1.8e11 $1/Pa
P2 = 2.15e19 $1/Pa^2
P3 = 0.08 $strain
CR = 15.0 $Dimensionless
RK = 0.7 $Dimensionless
RN = 0.0 $Pa
HC = 0.0 $Pa
CTI1 = 3.e6 $Pa
CTPS = 1.e6 $Pa
T1 = 0.0 $sec
T2 = 0.0 $1/sec
T3 = 0.0 $Dimensionless
T4 = 0.0 $1/sec
T5 = 0.0 $Pa
T6 = 0.0 $sec
T7 = 0.0 $1/Pa
J3TYPE = 3 $Dimensionless
A2PF = 0.0 $$$$$ default = A2 for normality
A4PF = 0.0 $$$$$ default = A4 for normality
CRPF = 0.0 $$$$$ default = CR for normality
RKPF = 0.0 $$$$$ default = RK for normality
$model 2 generic eos
$ rho ref = 1870. $ kg/m^3
$ tref = 298. $ K
$ ref sound speed = 2800.0 $m/s
$end
B16
APPENDIX B. Nomenclature and Data Sets
Dataset for lab scale intact Climax Granite
$
$ GeoModel parameters for Labscale Intact Climax Granite
$
B0 = 43.00e9 $Pa
B1 = 750.0e9 $Pa
B2 = 100.0e9 $Pa
B3 = 0.0 $Pa
B4 = 0.0 $Dimensionless
G0 = 34.73e9 $Pa
G1 = 0.0 $Dimensionless
G2 = 0.0 $Pa
G3 = 0.0 $Pa
G4 = 0.0 $Dimensionless
RJS = 0.0 $Meters (Joint spacing)
RKS = 0.0 $Pa/Meter (Joint shear stiffness)
RKN = 0.0 $Pa/Meter
A1 = 1355.e6 $Pa
A2 = 3.43e10 $1/Pa
A3 = 1328.e6 $Pa
A4 = 3.82e2 $Dimensionless
P0 = 556.e6 $Pa
P1 = 9.e14 $1/Pa
P2 = 0. $1/Pa^2
P3 = 0.05 $strain
CR = 227.5 $Dimensionless
RK = 0.72 $Dimensionless
RN = 17.0e6 $Pa
HC = 150000.e6 $Pa
CTI1 = 30.e6 $Pa
CTPS = 10.e6 $Pa
T1 = 0.0 $sec
T2 = 0.0 $1/sec
T3 = 0.0 $Dimensionless
T4 = 0.0 $1/sec
T5 = 0.0 $Pa
T6 = 0.0 $sec
T7 = 0.0 $1/Pa
J3TYPE = 3 $dimensionless
$ Flow potential parameters (for associativity, equate with yield parameters)
A2PF = 0.0 $zero means will default to equal A2 (associative)
A4PF = 0.0 $zero means will default to equal A4 (associative)
CRPF = 0.0 $zero means will default to equal CR (associative)
RKPF = 0.0 $zero means will default to equal RK (associative)
$model 2 generic eos
$ rho ref = 2635 $kg/m^3
$ tref = 298 $ K
$ ref sound speed = 17847 $m/s
$ end
B17
APPENDIX B. Nomenclature and Data Sets
Dataset for field scale jointed Climax Granite
$
$ GeoModel parameters for fieldscale JOINTED Climax Granite
$
B0 = 43.00e9 $Pa
B1 = 750.0e9 $Pa
B2 = 100.0e9 $Pa
B3 = 0.0 $Pa
B4 = 0.0 $Dimensionless
G0 = 34.73e9 $Pa
G1 = 0.0 $Dimensionless
G2 = 0.0 $Pa
G3 = 0.0 $Pa
G4 = 0.0 $Dimensionless
RJS = 0.06 $Meters (Joint spacing)
RKS = 8.0e10 $Pa/Meter (Joint shear stiffness)
RKN = 1.6e11 $Pa/Meter
A1 = 1379.e6 $Pa
A2 = 6.51e11 $1/Pa
A3 = 1328.e6 $Pa
A4 = 0.0 $Dimensionless
P0 = 556.e6 $Pa
P1 = 9.e14 $1/Pa
P2 = 0. $1/Pa^2
P3 = 0.05 $strain
CR = 227.5 $Dimensionless
RK = 0.80 $Dimensionless
RN = 17.0e6 $Pa
HC = 150000.e6 $Pa
CTI1 = 30.e6 $Pa
CTPS = 10.e6 $Pa
T1 = 0.0 $sec
T2 = 0.0 $1/sec
T3 = 0.0 $Dimensionless
T4 = 0.0 $1/sec
T5 = 0.0 $Pa
T6 = 0.0 $sec
T7 = 0.0 $1/Pa
J3TYPE = 3 $dimensionless
$ Flow potential parameters (for associativity, equate with yield parameters)
A2PF = 0.0 $zero means will default to equal A2 (associative)
A4PF = 0.0 $zero means will default to equal A4 (associative)
CRPF = 0.0 $zero means will default to equal CR (associative)
RKPF = 0.0 $zero means will default to equal RK (associative)
$model 2 generic eos
$ rho ref = 2635 $kg/m^3
$ tref = 298 $ K
$ ref sound speed = 17847 $m/s
$ end
B18
APPENDIX B. Nomenclature and Data Sets
Dataset for 23MPa Concrete
B0 = 5.5e9 $Pa
B1 = 28.78e9 $Pa
B2 = 0.623e9 $Pa
B3 = 0.0 $Pa
B4 = 0.0 $Dimensionless
G0 = 1.9026e9 $Pa
G1 = 0.890513 $Dimensionless
G2 = 3.55e9 $1/Pa
G3 = 0.0 $Pa
G4 = 0.0 $Dimensionless
RJS = 0.0 $Meters
RKS = 0.0 $Pa/m
RKN = 0.0 $Pa/m
A1 = 1255.7e6 $Pa
A2 = 1.93e10 $1/Pa
A3 = 1248.2e6 $Pa
A4 = 0.0 $Radians
P0 = 1.067e8 $Pa
P1 = 7.66e10 $1/Pa
P2 = 3.88e20 $1/Pa^2
P3 = 0.1538 $Dimensionless(strain)
CR = 10.0 $Dimensionless
RK = 1. $Dimensionless
RN = 3.0e6 $Pa
HC = 1.0e11 $Pa
CTI1 = 3e6 $Pa
CTPS = 1.e6 $Pa
T1 = 0.0 $Sec
T2 = 0.0 $1/Sec
T3 = 0.0 $Dimensionless
T4 = 0.0 $1/Sec
T5 = 0.0 $Pa
T6 = 0.0 $Sec
T7 = 0.0 $1/Pa
J3TYPE=3
A2PF = 0.0 $zero defaults to A2 (associative)
A4PF = 0.0 $zero defaults to A4 (associative)
CRPF = 0.0 $zero defaults to CR (associative)
RKPF = 0.0 $zero defaults to RK (associative)
SUBX = 0.0
$model 2 generic eos
$ rho ref = 2030. $ kg/m^3
$ tref = 298. $ K
$ ref sound speed = 4500.0 $ m/s
$end
B19
APPENDIX B. Nomenclature and Data Sets
Dataset for Conventional Strength Portland Concrete
B0 = 1.0954e10 $Pa
B1 = 0.0 $Pa
B2 = 0.0 $Pa
B3 = 0.0 $Pa
B4 = 0.0 $Dimensionless
G0 = 7.5434e9 $Pa
G1 = 0.0 $Dimensionless
G2 = 0.0 $1/Pa
G3 = 0.0 $Pa
G4 = 0.0 $Dimensionless
RJS = 0.0 $Meters
RKS = 0.0 $Pa/m
RKN = 0.0 $Pa/m
A1 = 4.26455e8 $Pa
A2 = 7.51e10 $1/Pa
A3 = 4.19116e8 $Pa
A4 = 1.0e10 $Radians
P0 = 1.95520e8 $Pa
P1 = 1.2354e9 $1/Pa
P2 = 0.0 $1/Pa^2
P3 = 0.065714 $Dimensionless(strain)
CR = 12.0 $Dimensionless
RK = 1. $Dimensionless
RN = 0.0 $Pa
HC = 0.0 $Pa
CTI1 = 3.0e6 $Pa
CTPS = 1.0e6 $Pa
T1 = 0.0 $Sec
T2 = 0.0 $1/Sec
T3 = 0.0 $Dimensionless
T4 = 0.0 $1/Sec
T5 = 0.0 $Pa
T6 = 0.0 $Sec
T7 = 0.0 $1/Pa
J3TYPE=3
A2PF = 0.0 $zero defaults to A2 (associative)
A4PF = 0.0 $zero defaults to A4 (associative)
CRPF = 0.0 $zero defaults to CR (associative)
RKPF = 0.0 $zero defaults to RK (associative)
SUBX = 0.0
$model 2 generic eos
$ rho ref = 2250. $(kg/cubic meter)
$ tref = 298.
$ ref sound speed = 3056.0
$end
B20
APPENDIX B. Nomenclature and Data Sets
Datasets for mimicking classical (simplified) models
The GeoModel is truly a unification of many simpler theories. By appropriately setting
parameters, the GeoModel can be made to exactly replicate results from simpler idealized
theories, as illustrated below.
In most finiteelement codes, you can modify an existing data set (e.g., one for a real
material) by simply redefining a material parameter in a separate input line, leaving the
original value unchanged. By deviating from a correct input set to a “toy” input set in this
way (rather than overwriting preferred values), you can retain a record of what the mate
rial parameters should be, thereby mitigating unintentional dissemination of physically
bad input sets.
The following specialized input sets use an “aprepro” syntax to show where you need
to provide values. Specifically, all required or computed values appear in braces {}. Spec
ify numerical values wherever “VALUE” appears, and then ensure all other values in
braces are computed as shown (they can be computed by hand or piped into aprepro).
0
10
20
30
0
10
20
30
0
10
20
30
0
10
20
30
0
10
20
0
Von Mises Max stress Max shear (Tresca) MohrCoulomb
Figure B.1. Other yield surface shapes supported by the GeoModel. The grid lines shown on these
sketches correspond to lines of constant and constant . z θ
Max
stress
Mohr
Coulomb
Tresca
Von Mises
Figure B.2. Classical simplified yield surfaces in the octahedral plane. None of these models ade
quately describes rock failure surfaces, but the failure surfaces for real rocks sometimes share some qual
itative features with these models, depending on the level of confining pressure. In these figures, the axes
represent projections of the compressive principal stress axes onto the octahedral plane, taking stress to
be positive in tension. For the max stress and MohrCoulomb models, the size of the octahedral profile
increases with pressure. For all of these models, the meridional profile is a straight line.
B21
APPENDIX B. Nomenclature and Data Sets
VonMises material. Von Mises theory can be duplicated by using the following sim
plified set of GeoModel input values:
$
$ GeoModel parameters for replicating nonhardening VonMises plasticity
$ All parameters not listed are defaulted to zero.
$ Items in braces must be replaced with numbers
$
A1 = {yield_stress_in_shear = VALUE}
B0 = {linear_elastic_bulk_modulus = VALUE}
G0 = {linear_elastic_shear_modulus = VALUE}
J3TYPE = 1 $Use Gudehaus, which is capable of a circular octhedral profile
RK = 1 $Set TXE/TXC ratio = 1.0 to make a circlular octahedral profile
P0 = 1.e99 $make yield in hydrostatic compression impossible
CTI1 = 1.e99 $ set pressure cutoff to “infinity”
CTPS = 1.e99 $ set shear cutoff to “infinity”
CR = 0.001 $ minimize the size of the curved part of the cap
A2 = 0.0
A4 = 0.0
P3 = 0.0
HC=0.0
T1 = 0. $sec
T2 = 0. $1/sec
T3 = 0. $Dimensionless
T4 = 0. $1/sec
T5 = 0. $Pa
T6 = 0. $sec
T7 = 0. $1/Pa
RN=0.
P0=1.e11
A2PF = 0.0
A4PF = 0.0
CRPF = 0.0
RKPF = 0.0
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
Maximum Principal Stress failure. The very simplistic fracture criterion that ini
tiates failure when the largest principal stress reaches a critical value can be modeled in
the GeoModel by using the following parameter set:
$ GeoModel parameters for duplicating a maximum principal stress criterion
$
${max_allowed_principal_stress = VALUE}
B0 = {VALUE} $bulk modulus
G0 = {VALUE} $shear_modulus
P0 = 1.e99 $ turn off the cap function
J3TYPE=3
A1={SQRT(3)*max_allowed_principal_stress}
A4={1/SQRT(3.0)}
RK = 0.5
A4PF = {1/SQRT(3.0)}
RKPF = 0.5
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
B22
APPENDIX B. Nomenclature and Data Sets
Tresca. The simplistic criterion that a material fails when its largest shear stress reaches
a critical value can be modeled by using the following GeoModel parameters:
$ User must specify values in braces
B0 = {bulk_modulus}
G0 = {shear_modulus}
P0 = 1.e99 $ turn off the cap function
J3TYPE=3
A1={2.0**yield_stress_in_shear/sqrt(3.0)}
RK = 1.0
RKPF = 1.0
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
MohrCoulomb.
Classical MohrCoulomb theory for brit
tle failure can be derived from an idealization
that the material contains a large population
of equal sized cracks. Being all the same size,
any given crack loaded in pure shear will fail
(grow) if the resolved shear stress on the
crack face exceeds a critical threshold value
. If a crack face is additionally subjected to
a normal compressive stress , then the
applied shear needed to induce crack
growth must be larger than by an amount
, where is the coefficient of
friction. Stated differently, a given crack is
safe from failure if
, (B.2)
or, recalling that ,
, where (B.3)
This criterion must be satisfied by all cracks in the material. Since MohrCoulomb theory
arises from an idealization that the material contains a large population of cracks (uni
formly random in orientation), a material is safe from failure under general stress states
only if all points on the Mohr’s diagram for the stress fall below the “failure line” defined
by . Failure is therefore deemed to occur when the outer Mohr’s circle
first “kisses” the failure line. Working out the geometry of Fig. B.3, a circle of radius ,
centered at , will be tangent to the failure line if and only if
(B.4)
For the outer Mohr’s circle,
τ
σ
N
f
a
ilu
r
e
lin
e
S
0
φ
Figure B.3. A stress at the limit state
under MohrCoulomb theory.
σ
3
σ
2
σ
1
τ
S
0
σ
N
τ
S
0
τ
fric
µσ
N
= µ
τ τ
fric
– S
0
<
τ
fric
µσ
N
=
τ φ tan σ
N
– S
0
< φ tan µ ≡
τ S
0
φ tan σ
N
+ =
R
σ
N
C =
R S
0
φ cos C φ sin + =
B23
APPENDIX B. Nomenclature and Data Sets
and (B.5)
or, removing the overbars (defined such that ),
and (B.6)
Substituting these into Eq. (B.4) gives
(B.7)
This is the MohrCoulomb failure criterion cast in terms of principal stresses. Eqs. (3.43a)
and (3.43c) on page 24 of the main report imply
(B.8a)
. (B.9b)
Therefore, the stress invariant version of Eq. (B.7) is
, (B.10)
or, solving for ,
. (B.11)
Recall from Eq. (4.39) that the GeoModel’s limit function fits the triaxial compression
(TXC) meridional profile to the following functional form
in TXC ( ) (B.12)
In triaxial compression (TXC), the Lode angle is so that
and (B.13)
Therefore, Eq. (B.11) specializes to TXC loading as
(MohrCoulomb in TXC) (B.14)
Being careful to note that , comparing this result with Eq. (B.12) implies that
the GeoModel limit surface parameters for MohrCoulomb theory should be set as
R
σ
1
σ
3
–
2
 = C
σ
1
σ
3
+
2
 =
σ
k
σ
k
– =
R
σ
3
σ
1
–
2
 = C
σ
3
σ
1
+ ( ) –
2
 =
σ
3
σ
1
–
2
 S
0
φ cos
σ
3
σ
1
+
2
 φ sin – =
σ
3
σ
1
–
2
 J
2
θ cos =
σ
3
σ
1
+
2

I
1
3

J
2
3
 θ sin – =
J
2
θ cos S
0
φ cos
I
1
3
 φ sin –
J
2
3
 θ sin φ sin + =
J
2
J
2
S
0
φ cos
I
1
3
 φ sin –
θ cos
1
3
 θ sin φ sin –
 =
F
f
J
2
a
1
a
3
e
a
2
I
1
–
– a
4
I
1
+ = θ 30° =
θ 30° =
θ cos 30° cos
3
2
 = = θ sin 30° sin
1
2
 = =
J
2
2 3
3 φ sin –
 S
0
φ cos
I
1
3
 φ sin –
\ .
 
=
I
1
I
1
– =
B24
APPENDIX B. Nomenclature and Data Sets
(B.15a)
(B.15b)
(B.15c)
(B.15d)
Therefore, the general MohrCoulomb criterion in Eq. (B.11) may be written in the form
required for the GeoModel as
,
where ,
and (B.16)
Naturally, J3TYPE=3 is appropriate. Moreover, to make the cap function equal to 1,
the cap and crush curve features should be disabled as described on page B26. Classical
MohrCoulomb theory is meant to apply to brittle rupture, so kinematic hardening should
be disabled as described on page B26. An appropriate value for the TXE/TXC strength
ratio must be determined by evaluating Eq. (B.11) in TXE where :
(MohrCoulomb in TXE) (B.17)
The TXE/TXC strength ratio is then given by the ratio of the righthand sides of
Eqs. (B.17) and (B.14):
(B.18)
For this classical MohrCoulomb theory, the slope of the TXC meridional profile,
, equals . Therefore, may be eliminated from Eqs. (B.15d) and (B.18) to
reveal that the TXE/TXC strength ratio is coupled to the slope of the TXC meridional pro
file according to
, where (B.19)
This relationship will be explored in future releases of the GeoModel for allowing pres
sure dependence of the octahedral profile shape when (unlike MohrCoulomb theory) the
meridional profile has a nonconstant slope.
a
1
2 3
3 φ sin –

\ .
 
S
0
φ cos =
a
2
0 =
a
3
0 =
a
4
2 3
3 φ sin –

\ .
 
φ sin
3
 =
J
2
Γ θ ( ) f
f
I
1
( )f
c
I
1
( ) =
f
f
I
1
( ) a
1
a
4
I
1
+ = f
c
I
1
( ) 1 =
Γ θ ( )
2 3
3 φ sin –
 θ cos
φ sin θ sin
3
 –
\ .
 
=
f
c
ψ θ 30° – =
J
2
2 3
3 φ sin +
 S
0
φ cos
I
1
3
 φ sin –
\ .
 
=
ψ
3 φ sin –
3 φ sin +
 =
d J
2
d I
1
⁄ a
4
φ sin
ψ
1
1 3A +
 = A
∂ J
2
∂ I
1

\ .

 
θ=30°
≡
B25
APPENDIX B. Nomenclature and Data Sets
Below, we show the skeleton required for setting the GeoModel inputs to run classical
MohrCoulomb theory. This input set also shows how to set parameters if you wish to run
with nonassociativity (where the flow potential function differs from the yield function
only by having a different value for ). Values for and for are
set by using Eq. (B.15). Values for and are set by using Eq. (B.18).
$ GeoModel parameters for duplicating a classical MohrCoulomb material
$ in BOTH the octahedral and meridional profiles.
$
$ Replace every occurance of “VALUE” in this file with the appropriate
$ MohrCoulomb parameter. You may generate GeoModel input values for the
$ by using the command ‘aprepro this_file output_file’. Preferably, if
$ your finite element code supports embedded aprepro directives, then
$ drop this file DIRECTLY into your finite element GeoModel input block.
$
$ {friction_angle = VALUE} $angle “phi” in GeoModel Appendix B
$ {dilation_angle = VALUE} $for associativity, this equals “phi”
$
$ {cohesion = VALUE} $parameter “S_sub0” in GeoModel Appendix B
$ $This equals yield stress in simple shear.
$ Let Aprepro compute some helper quantities
$ {scalef = 2.0*sqrt(3)/(3.0sin(friction_angle))}
$ {scaleg = 2.0*sqrt(3)/(3.0sin(dilation_angle))}
$
B0 = {VALUE} $bulk modulus
G0 = {VALUE} $shear modulus
J3TYPE=3
A1 = {scalef*cohesion*cos(friction_angle)}
A2 = 0.0
A3 = 0.0
A4 = {scalef*sin(friction_angle)/3.0}
RK = {(3.0sin(friction_angle))/(3.0+sin(friction_angle))}
A1PF = {scaleg*cohesion*cos(dilation_angle)}
A4PF = {scaleg*sin(dilation_angle)/3.0}
RKPF = {(3.0sin(dilation_angle))/(3.0+sin(dilation_angle))}
P0 = 1.e99 $ turn off the cap and crushcurve features
P1 = 0.0 $ no cap
P2 = 0.0 $ no cap
P3 = 0.0 $ zero porosity
CR = 0.001 $ prevent cap influence on shear response
CRPF = 0.001 $ prevent cap influence on shear response
HC = 0.0 $ disable kinematic hardening
RN = 0.0 $ disable kinematic hardening
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
φ a
1
a
2
a
3
a
4
, , , ( ) a
2
PF
a
4
PF
, ( )
ψ ψ
PF
B26
APPENDIX B. Nomenclature and Data Sets
To conclude this appendix, subinput sets are summarized for controlling various fea
tures of the GeoModel.
Subinput set for linear elasticity.
B1 = 0.0
B2 = 0.0
B3 = 0.0
B4 = 0.0
G1 = 0.0
G2 = 0.0
G3 = 0.0
G4 = 0.0
Set B0 to the constant bulk modulus and set G0 to the constant shear modulus. To additionally disable any form of plas
ticity, set A1 to a very large number and disable cutoffs as well as the cap and crush curve as described below.
Subinput set for “turning off” all rate dependence.
T1 = 0.0
T2 = 0.0
T3 = 0.0
T4 = 0.0
T5 = 0.0
T6 = 0.0
T7 = 0.0
For LINEAR rate dependence, set T1 to the material’s characteristic response time and all other T parameters to zero.
Subinput set for disabling kinematic hardening.
HC=0.0
RN=0.0
Subinput set for associativity.
CRPF=0.0
RKPF=0.0
A2PF=0.0
A4PF=0.0
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters. If not, associativity
requires that all parameters that end in “PF” be set equal to their yield parameter counterparts.
Subinput set for disabling cap and crush curve.
P0 = 1.e99 $put the cap at infinity
P1 = 0.0
P2 = 0.0
P3 = 0.0 $set porosity to zero
CR = 0.001 $ minimize the size of the curved part of the cap
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters. The GeoModel user
input processor and initializer will recognize these inputs as “cap disablers” and adjust the model appropriately.
Subinput set for disabling tensile cutoff limits.
CTI1 = 1.e99 $ set pressure (I1) cutoff to “infinity”
CTPS = 1.e99 $ set principal stress cutoff to “infinity”
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
REFERENCES
1
ALDRIDGE, D.F. (2002) Elastic wave radiation from a pressurized spherical cavity, Sandia Na
tional Laboratories Report SAND20021882, Albuquerque, NM.
2
ARGUELLO, J.G., A.F. FOSSUM, D.H. ZEUCH and K.G. EWSUK. (2001) Continuumbased FEM
modeling of alumina powder compaction. KONA, 19: 166177.
3
ARGYRIS,J.H., FAUST,G., SZIMMAT,J., WARNKE,E.P., and WILLAM,K.J. (1974) Recent develop
ments in the finite element analysis of prestressed concrete reactor vessel, Nuclear Engrg. Des.
28, p. 4275.
4
BECKER, R. (2003) Lawrence Livermore National Laboratories, unpublished presentation materials.
5
BLANFORD, M.L. (1998) Jas3d: A multistrategy iterative code for solid mechanics analysis. Un
published. Sandia National Laboratories.
6
BORJA, RONALDO I., SAMA, KOSSI M., and SANZ, PABLO F. (2003) On the numerical integra
tion of three invariant elastoplasticity constitutive models, Comput. Methods Appl. Mech. Engr.
192, p. 12271258.
7
BOSWELL, L.F. and CHEN,Z. (1987) A general failure criterion for plain concrete, Int. J. Solids
Struct. 23, p. 621630
8
BOUCHERON, E.A., K.H. BROWN, K.G. BUDGE, S.P. BURNS, D.E. CARROLL, S.K. CARROLL, M.A.
CHRISTON, R.R. DRAKE, C.G. GARASI, T.A. HAILL, J.S. PEERY, S.V. PETNEY, J. ROBBINS, A.C. ROBIN
SON, R.M. SUMMERS, T.E. VOTH, and M.K. WONG (2002) ALEGRA: User Input and Physics De
scriptions Version 4.2, Sandia National Laboratories Report SAND20022775, Albuquerque, NM.
9
BRANNON, R.M. and W.J. DRUGAN (1993) Influence of nonclassical elasticplastic constitutive
features on shock wave existence and spectral solutions, J. Mech. Phys. Solids 41:2, p. 297330.
10
BRANNON, R.M. and M. K. WONG (1996) MIG Version 0.0 Model Interface Guidelines: Rules to
Accelerate Installation of Numerical Models Into Any Compliant Parent Code,” Sandia National
Laboratories Report SAND962000, Albuquerque, NM.
11
BRANNON, R.M. (2004) Rotation and Reflection: a review of useful theorems associated with
orthogonal tensors. Sandia National Laboratories Report SAND2003XXXX, Albuquerque, NM, in
process.
12
DIGIOVANNI, A.A. and J.T. FREDRICH (2003) private communication, Sandia National Laborato
ries, Albuquerque, NM 87185, USA.
13
DUVAUT, G. and J.L. LIONS (1972) Les Inequations en Mecaniquie et en Physique, Dunod, Paris.
14
FARNSWORTH, G.V. and A.C. ROBINSON (2003) Improved Kinematic Options in ALEGRA, San
dia National Laboratories Report SAND20034510, Albuquerque, NM.
15
FOSSUM, A.F. and R.M. BRANNON (2004) Unified compaction/dilation, strainrate sensitive,
constitutive model for rock mechanics structural analysis applications, ARMA/NARMS 04546,
in press.
16
FOSTER, C.D., R.A. REGUEIRO, A.F. FOSSUM, and R.I. BORJA (2004) Implicit numerical integra
tion of a threeinvariant, isotropic/kinematic hardening cap plasticity model for geomaterials,
submitted to Elsevier Science.
17
FREW, D.J., M.J. FORRESTAL, and W. CHEN (2001) A split Hopkinson Pressure Bar Technique to
determine compressive stressstrain data for rock materials, Experimental Mechanics, 41: 4046.
18
FUNG, Y.C. (1965) Foundations of Solid Mechanics, PrenticeHall, Englewood Cliffs, NJ.
19
GREENBERG, M.D. (1978) Foundations of Applied Mathematics, PrenticeHall, Englewood Cliffs,
NJ.
20
GURSON, A.L. (1977) J. Engng Mater. Technol., 99, p. 2.
21
HOFFMAN, E.L., C.A. LAVIN (2004) WISDM: Weapons Information System and Data Manage
ment, Version 2.2.0., Apr. 2002. Sandia National Laboratories. Electronic. Sandia Network.
22
JAY, F. (1984), Ed., IEEE Standard Dictionary of Electrical and Electronic Terms, Institute of
Electrical and Electronic Engineers, ANSI/IEEE Std. 1001984.
23
JIANG, J., PIETRUSZCZAK, S. (1988) Convexity of yield loci for pressure sensitive materials,
Comput. Geotech. 5, p. 5163.
24
JOHNSON, G.R. et al. (1994) User Instructions for the 1995 Version of the EPIC Code, Alliant
Techsystems, Inc., technical report, Hopkins, Minnesota.
25
KNUPP, P. and K. SALARI (2000) Code verification by the method of manufactured solutions, San
dia National Laboratories Report SAND20001444, Albuquerque, NM.
26
KOTERAS, J.R. and A.S. GULLERUD (2002) Verification of Computer Codes in Computational
Science and Engineering, Chapman & Hall/CRC.
27
LUBLINER, J. (1990) Plasticity Theory, Macmillan, New York, NY.
28
MATSUOKA, H., and NAKAI, T. (1974) Stressdeformation and strength characteristics of soil un
der three different principal stresses, Proc. JSCE 232, p. 5970.
29
LADE, P.V., and DUNCAN, J.M. (1975) Elastoplastic stressstrain theory for cohesionless soil, J.
Geotech. Engrg. Div., ASCE 101, p. 10371053.
30
LADE, P.V. (1977) Elastoplastic stressstrain theory for cohesionless soil with curved yield sur
faces, Int. J. Solids Struct. 13, p. 10191035.
31
LUBLINER, JACOB (1990) Plasticity Theory, MacMillian, New York.
32
MALVERN, L.E. (1969) Introduction to the Mechanics of a Continuous Medium, PrenticeHall,
Englewood Cliffs, NJ
33
NEEDLEMAN, A. and J.R. RICE (1978) Mech. Mater. 7, p. 163.
34
PELESSONE, D. (1989) A modified formulation of the cap model Gulf Atomics Report GA
C19579, prepared for the Defense Nuclear Agency under Contract DNA001086C0277, January.
35
PRESS, W.H., FLANNERY, B.P., TEUKOLSKY, S.A., VETTERLING, W.T. (1992) Numerical Recipes
in FORTRAN, Cambridge University Press, Melbourne, Australia.
36
PUCIK, T.A. (2004) Memorandum of Record, Material Model Driver, 11 April 2004, Pucik Con
sulting Services.
37
PUCIK, T.A. (1993) “A case study of instability and nonuniqueness induced by nonassociated
plastic flow”, Unpublished Manuscript.
38
RASHID, M (1993) Incremental kinematics for finiteelement applications. Int. J. Numerical Meth. in
Engr. 36, no.23, pp. 39373956.
39
ROACHE, P.J. (1998) Verification and Validation in Computational Science and Engineering,
Hermosa publishers, Albuquerque, NM.
40
SANDLER, I.S. and T.A. PUCIK (1994) Nonuniqueness in dynamic rateindependent nonassociated
plasticity, in G.Z. Voyiadjis, et al. (eds.) Mechanics of Materials and Structures, Elsevier Science
B.V., New York, pp. 221240.
41
SANDLER, I.S. and D. RUBIN (1979) An algorithm and a modular subroutine for the cap model,
Int. J. Numer. & Anal. Meth. Geomech. 3, p. 173.
42
SCHWER, L.E., Y.D. MURRY (1994) A threeinvariant smooth cap model with mixed hardening,
Int. J. for Num. And Anal. Mech. In Geomech. 18: 657688.
43
SENSEY, P.E., A.F. FOSSUM, & T.W. PFEIFLE (1983) Nonassociative constitutive laws for low
porosity rocks, Int. J. for Num. and Analytical Meth. in Geomech. 7, pp. 101115.
44
SIMONS, D. (2003) SingleElement Load Paths to Complement ACTD V&V Problems 1&2,
Northrop Grumman Information Technology unpublished presentation slides for ACTD V&V Work
ing Group Meeting, Albuquerque, NM 17 April 2003.
45
TAYLOR, L.M. and D.P. FLANNAGAN (1989) PRONTO 3D a threedimensional transient solid
dynamics program, Sandia National Laboratories Report SAND871912, Albuquerque, NM.
46
WARREN, T.L., A.F. FOSSUM, and D.J. FREW (2004) Penetration into low strength (23 MPa) con
crete: target characterization and simulations. Int J. Impact Engr. (accepted).
47
WILLAM,K.J. and WARNKE,E.P. (1975) Constitutive model for the triaxial behavior of concrete,
ISMES Seminar on Concrete Structures Subjected to Triaxial Stresses, Bergamo, Italy, p. 130.
48
WOLFRAM, STEPHEN, The Mathematica Book, 4th Ed, Cambridge Univ. Press, 1999.
49
WONG, M.K., J.R. Weatherby, C.D. Turner, A.C. Robinson, T.A. Haill, D.E. Carroll (2001) Physics
Applications in the ALEGRA Framework, First MIT Conference on Computational Fluid and
Solid Mechanics, June, 2001.
Issued by Sandia National Laboratories, operated for the United States Department of Energy by Sandia Corporation.
NOTICE: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, nor any of their contractors, subcontractors, or their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by the trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government, any agency thereof or any of their contractors or subcontractors. The views and opinions expressed herein do not necessarily state or reflect those of the United States Government, any agency thereof or any of their contractors.
Printed in the United States of America This report has been reproduced directly from the best available copy. Available to DOE and DOE contractors from Office of Scientific and Technical Information PO Box 62 Oak Ridge, TN 37831 Prices available from (615) 5768401, FTS 6268401 Available to the public from National Technical Information Service US Department of Commerce 5285 Port Royal Rd Springfield, VA 22161 NTIS price codes Printed copy: A08 Microfiche copy: A01
SAND20043226 Unlimited Release Printed August 2004
THE SANDIA GEOMODEL
Theory and User’s Guide
Arlo F. Fossum and Rebecca M. Brannon
Geomechanics, Dept. 6117 Sandia National Laboratories PO BOX 5800 Albuquerque, NM 87185
Abstract
The mathematical and physical foundations and domain of applicability of Sandia’s GeoModel are presented along with descriptions of the source code and user instructions. The model is designed to be used in conventional finite element architectures, and (to date) it has been installed in five host codes without requiring customizing the model subroutines for any of these different installations. Although developed for application to geological materials, the GeoModel actually applies to a much broader class of materials, including rocklike engineered materials (such as concretes and ceramics) and even to metals when simplified parameters are used. Nonlinear elasticity is supported through an empirically fitted function that has been found to be wellsuited to a wide variety of materials. Fundamentally, the GeoModel is a generalized plasticity model. As such, it includes a yield surface, but the term “yield” is generalized to include any form of inelastic material response including microcrack growth and pore collapse. The geomodel supports deformationinduced anisotropy in a limited capacity through kinematic hardening (in which the initially isotropic yield surface is permitted to translate in deviatoric stress space to model Bauschinger effects). Aside from kinematic hardening, however, the governing equations are otherwise isotropic. The GeoModel is a genuine unification and generalization of simpler models. The GeoModel can employ up to 40 material input and control parameters in the rare case when all features are used. Simpler idealizations (such as linear elasticity, or Von Mises yield, or MohrCoulomb failure) can be replicated by simply using fewer parameters. For highstrainrate applications, the GeoModel supports rate dependence through an overstress model.
Intentionally Left Blank iv .
a Defense Programs (DP) Penetration project led first by Jim Hickerson and later by Danny Frew.S. a Hard and Deeply Buried Target (HDBT) project led by Paul Yarrington and Shawn Burns. and datasets for various materials). 2004. Joe Jung. user feedback. appreciates the numerous program managers and other supporters who have given Arlo Fossum sufficient time and resources to assemble a relatively comprehensive set of components that ideally should be discussed in any material model resource document (theory. RMB. John Pott. Len Schwer. Tom Voth. The authors acknowledge individual contributions of Steve Akers. who have provided code support. Steve Sobolik. Finally. verification. RMB admires Arlo Fossum for his mentoring. and suggesting improvements. John Holland. Lupe Arguello. several DOE Accelerated Strategic Computing Initiative (ASCI) Design and Qualification (DQ) Materials & Physics Models (M&PM) projects led by Mike McGlaun and Justine Johannes. and Mike Wong. Dave Holcomb. two LDRD projects. The initial development of the model was begun under a project led by Mike Stone under the U. and his remarkable ability to discern the salient physics evident in disparate geomechanical data. Arlo Fossum affossu@sandia. one led by Larry Costin and the other by Rich Regueiro. Craig Foster. unfailing sense of humor. Sal Rodriguez. Tom Pucik. and an Army Research Laboratory (ARL) project led by Bill Bruchey. documenting. both authors are grateful to our family members for their patience and support during the many latenight hours required to reach this year’s model development and dissemination goals. Mike Forrestal. Jonathan Rath. Lew Glenn. theoretical discussions.Acknowledgements The authors are grateful for the support received over the years from a number of Sandia project managers and colleagues who have provided financial support and encouraged the development and enhancement of this general geomaterial model for use in Sandia computer codes. Bob Swift. Mike Chiesa. Estaban Marin.gov Sandia National Laboratories August 30. a Model Accreditation Via Experimental Sciences For Nuclear Weapons (MAVEN) laboratory testing project led by Moo Lee. would like to express his sincere appreciation to Rebecca Brannon for painstakingly going through each line of coding in the material model subroutine checking theoretical soundness. AFF. Colby Lavin. Department of Energy (DOE) Advanced Computational Technology Initiative (ACTI) program. The first author. The second author. Tom Warren. Bill Oberkampf. Joanne Fredrich. debugging. and constructive criticism. Kent Danielson.gov Rebecca Brannon rmbrann@sandia. bug fixes. Our compatible skills and temperaments have established the foundation for an effective and enduring materials modeling team. Ed Hoffman. Daniel Kletzli. The quality of the model has vastly improved as a result. Bill Scherzinger. enthusiasm. Paul Senseny. Ted Carney. Paul Booker. Jim Cox. Yvonne Murray. Joel Wirth. Additional development was made under an Engineering Sciences Research Foundation (ESRF) project led by Hal Morgan. Kevin Brown. validation. Allen Shirley. v . code. calibration instructions. Bill Olsson.
................................................................................................................................................................... Derivatives of the stress invariants .................................................................................................................... 72 Role of the GeoModel within a finiteelement program............................................................................. consistency parameter) .................................................................................. 7 7 7 The Stress Tensor.................................................................................................... Formal equivalence with oblique return algorithms ..............................................................Contents Acknowledgements .................................................................... Limitations of the GeoModel ............ 70 71 GeoModel Numerical Solution scheme ............................................................................................................................................ Meridional shear limiter function.................................... Octahedral yield profile visualization ............................................................................................................................................................ The cap function............................ GeoModel overview...................... Vector and Tensor notation................. Evolution equations................ Elasticplastic coupling ............ 74 GeoModel files........................................................................................................................................................ Viscoplasticity model derivation............................................................................................................................................... TRIAXIAL (TXC and TXE): ......................... Evolution equation for the porosityrelated internal state variable........ v 1 2 2 3 Notation.......................................................................................................................................................................................................... Ff ...... 75 vi ........... and model installation requirements............ Introduction.......................... Closedform solution for ordered eigenvalues ............................................................................................................................................................................................................................ subroutines..................................................................................................................................................................................................... Assigning values to the elastic constants........................................... 63 63 66 68 Assigning a value to the characteristic material time..... Special stress states ..................................... 19 20 22 23 24 GeoModel theory.................................................. SIMPLE/PURE SHEAR and PRESSURESHEAR LOADING (SHR) .......................................................... κ ............................................................................................................................................................................... 60 60 61 61 Rate Dependence............................... Features of the GeoModel .......................................................................................... 47 50 50 58 Advancing the solution (final step............................................. 25 26 27 27 The elastic limit (yield surface)........................................................................................................................................................................................................ Limiting case............................................... Stability issues ......................... Lode coordinates ...... Evolution equation for the kinematic hardening backstress tensor...................................................... Typesetting conventions............ Thermodynamics considerations...................................... HYDROSTATIC (HYD) .................. Nonlinear elasticity............................................................................................................................... Viscoplasticity model overview............................................................................................................................................................................................................................................................................... 9 11 12 13 13 14 18 Spatial and Reference stress (frame indifference) ......................................... memory requirements......... 28 36 37 38 42 43 Advancing the solution (groundwork discussion)............................................................ Meridional yield profile visualization ................................... Elasticity............................................ Γ .................................................................................................... The “J3TYPE” Lodeangle function........................................................................ 72 Spatial stability (mesh dependence/loss of strong ellipticity)........................................................ The complete GeoModel yield function .................................... Stress invariants .................................................. Quasistatic inelastic tangent stiffness tensor........................... fc ....................................................................................
......................................................................................................................................... 108 A1 B1 B2 B8 B14 B15 B16 B17 B18 B19 B20 B21 B21 B22 B22 B26 B26 B26 B26 B26 B26 REFERENCES vii ............................................................................................................................................... Dataset for lab scale intact Climax Granite ...................................................................................................................... Preliminary convergence testing................... Nonlinear elasticity ............................................................. Dataset for Sidewinder Tuff............................................... Tresca .................... Elasticplastic coupling ......................... Rate independent (inviscid) part of the viscoplasticity equations......................... Subinput set for linear elasticity ..................................................................................... Triaxial Compression................................... Postcalibration Triaxial loading............................................................................................................. Free field wave form for spherical shock loading................................ Subinput set for “turning off” all rate dependence ................................. Model Parameters (User Input) ....................................................................................... Other singleelement tests .................................................................................................................................................. APPENDIX A: Parameterizing the GeoModel ..................................................................... Fieldscale penetration ... Viscous part of the viscoplasticity equations ......................................................... APPENDIX B: Nomenclature and Data Sets ................................ Internal State Variables (Plotable Output) ............................................................... 79 86 Verification: singleelement problems (regression suite) ............ Subinput set for disabling cap and crush curve................................................................................................. 96 97 98 99 102 Building confidence in the physical theory.......................................................................................... Dataset for Conventional Strength Portland Concrete....................................... GeoModel algorithm ................................................................... Maximum Principal Stress failure....................................................................................................................................................................................................................................................................................................................................................................... VonMises material ............................................. Dataset for Salem Limestone .................................................................................................................................................................... Subinput set for disabling kinematic hardening ............................. Elastic freefield wave form (finiteelement verification) .................. MohrCoulomb ....................................................................................................................................................................................... Dataset for 23MPa Concrete ................................................. Subinput set for disabling tensile cutoff limits...... 76 77 78 78 78 Software “confidence building” activities........................................................................................................................................................ 93 95 Parameterization (calibration) ...... Datasets for mimicking classical (simplified) models ....... Dataset for field scale jointed Climax Granite........................................................... PLOTABLE OUTPUT .............................................................................................................................. 104 105 106 107 Closing Remarks ....Arguments passed to and from the GeoModel driver routine.......................................................................................................................................................................................................................................... Subinput set for associativity .......................................................................................................................................................................................................... 87 92 Verification: Hendren & Ayier pressurized cylinder........................................... Parameterization: Rate dependence ...................
................................................... Figure 3....... Figure 4................ volumetric strain) compression data.....8.. .9.................... volumetric strain data.... .................3..............................3............................................................ Figure 5....1.......... Figure 4.................. ..............................2..7.......17.. Figure 7............. Figure 7.... 50 A conventional porosity crush curve (dashed) and state path (solid).... and (c) a traditional porosity vs...................................................4........... 91 viii ................. 68 Rate dependence in uniaxial strain loading of a nonhardening pressureinsensitive material..... Figure 4...................... Figure 4......................... 86 Unconfined compression... 22 Distortion of the meridional profile when using nonisomorphic stress measures.. .... 69 Meaning of the “SHEAR” output variable.............. Figure 7. 58 The effect of kinematic hardening on the meridional plane that contains the backstress................... 88 20 MPa triaxial compression..................... Figure 4......... Figure 7...5................................ Figure 4............... .................. plotted at allowable values of the strength ratio................. 33 Basic shapes of the three key functions that characterize the composite shape of the yield surface....... ...................................................... Figure 3......... ........ Figure 4.. Figure 3...Figures Figure 1..... 45 Lode angle function (for various ψ strength ratios) plotted vs... Figure 5............. 52 The Gurson theory for porous yield surfaces compared with the GeoModel cap function at various values of the internal state variable κ................................................... Qualitatively................... Figure 4...........................1........ ... Figure 5.. Figure 4.......................................................... and (c) porosity and microcracks in combination....................................16....... (b) the GeoModel Xfunction.............6...........1.............................................2..... Figure 7.. 13 Triaxial compression (TXC) and triaxial extension (TXE).. 64 Highrate weighting factor at various initial states.... (b) microcracks alone...... 36 Kinematic hardening.... 33 Distinction between twosurface upperbound models and the GeoModel............ Figure 1.12............................ 59 Rate dependence..........15......................... 86 Rate of convergence....... .................................. 31 Qualitatively meridional profile shapes resulting from (a) porosity alone. Figure 3....2......... the theories are similar...................5............................................ 15 Triaxial and CSD load paths.................... Figure 3............4.......... ................... 41 Octahedral yield profiles................................... 35 GeoModel cap function ................1.......... Figure 4............ Figure 6.......1... the Lode angle varying from –30 degrees (TXE) to +30 degrees (TXC).................................... 16 An octahedral yield profile.... 53 Relationship between (a) hydrostatic pressure vs.......5.. pressure crush curve........7...... 40 Yield surface evolution with both microcracking and porosity......11................... 1 Distinction between a yield surface and the limit surface.....13....10............................. 23 Backstress and shifted stress................... 54 Continuously differentiable GeoModel yield function and some characteristic dimensions.................................. Figure 4............................................ Figure 4........ Figure 4.. ...............3..............................................1............................. 56 Meridional plane in which the magnitude of the stress deviator “r” is plotted against (which is proportional to the pressure “p”) ................... 5 Typical hydrostatic (pressure vs............................. ...........3.. Figure 7.... Figure 7............ Figure 4....... Figure 4............................. 89 Reduced triaxial extension.14......... ............................6....... 77 Screen shot of the MS Excel interface for the GeoModel materialmodel driver (MMD)......... 40 Shear limiter function (unshifted and shifted)..4........0....... Figure 4.................... GeoModel continuous yield surface.............................. 46 Hardening mechanisms............ Figure 4..... 83 Subcycling test for hydrostatic loading................ 90 Plane Strain..............................................................................................2....... ............... ...............
....... measured (red) velocity and displacement at 470 m.........10 of the main report........................9............................................... Figure 7....................8a.... B20 Classical simplified yield surfaces in the octahedral plane.. Figure A..... Hendron & Ayier verification problem... Kolsky (split Hopkinson) bar apparatus used to obtain data in Fig............................ ................................................................................... .. .......4................. analytical/numerical elastic wave velocity at 470 meters from a velocity spherical cavity source at 204 meters. Figure 7...... Figure A.... ...................................................... .................... ......7.............. ....... ............7.....Figure 7. 96 97 97 Elasticplastic coupling: deformationinduced changes in elastic moduli (Salem Limestone)..... Figure A.... Figure B.2......... Figure 8....1............. .......... . 101 102 Figure 8...........4...... Figure B..........A6 The Lode function information originally shown in Fig.....A2 Data representative of a triaxial compression parameterization test..... . Nonlinear elasticity in shear.....................................1.................8b.............................6.. Figure 8....... .3.. Figure 9............ Figure 8.............................. Crush curve and hydrostatic loading verification.. 99 Progression of the hardening yield surface (family of blue lines) under a triaxial compression test (red path).............2............... Figure A................6.......3.................. ...........................4................. 107 GeoModel + JAS3D finiteelement prediction (blue) vs......... Figure 8.....1......... ........................ ix ..3.. 106 Spherical cavity geometry...........5....................... Figure A............................... Spherical cavity geometry..8.......................... B22 Figure 9............... illustrated with correspondence of the meridional plane to the stressstrain diagram... 8................11...... Figure A...........A5 TXC stressstrain plots and extraction of their peak values to construct the meridional limit curve (data are for porcelanite).. (b) GeoModel predictions of triaxial stressstrain response at various confining pressures..9 ............................... Figure A........... 107 Ideal hydrostatic parameterization data................................. Figure 8..........................5.................... Figure 8.................. Figure 7. ... ...A3 A family of TXC tests conducted to failure........................... 105 After the GeoModel was fitted to laboratoryscale material property tests....9....... 100 Shearenhanced dilatation under compression....... from a spherical cavity velocity source at 204 meters....... 103 (a) GeoModel prediction of uniaxial strain loading........ 95 Meridional limit curves for some materials already parameterized to the GeoModel. now displayed with a transformed ordinate..... Figure B............................. Figure 8....... Rate dependence TXE/TXC ratio verification for triaxial load/unload............. 98 Shear failure limit curve compared with concrete data.... ... . ...................... ................. Nonlinear elasticity in hydrostatic loading..A2 Lesspreferable hydrostatic parameterization data..2........ Figure 9........................ ....................A10 A sketch of how data might be distributed at nontriaxial states................. Figure 8........... it was used to predict projectile penetration depth using spherical cavity expansion analysis... 4................... ...........3. 92 92 94 95 Finiteelement vs....2.................................. .................. .10.. Suitability of the GeoModel overstress ratedependence model to match observed data...... ........ ... Figure 7........ B20 A stress at the limit state under MohrCoulomb theory....A10 Other yield surface shapes supported by the GeoModel. Figure 9...1......
Intentionally Left Blank x .
Being a unified theory. structural integrity assessment for civil engineering problems. and differences in material behavior under triaxial extension compared to triaxial compression). nonnegligible nonlinear elasticity. (a) threedimensional view in principal stress space with the high pressure “cap” shown as a wire frame. and frozen soil) as well as rocklike engineered materials (such as concrete [46] and ceramics [2]) is at the core of a broad range of applications.Introduction THE SANDIA GEOMODEL Theory and User’s Guide TXC TX E r E TX z C TX TX C (a) (b) (c) Figure 1. For these materials. homogenized. GeoModel continuous yield surface. and (c) the octahedral view. ratesensitivity. which corresponds to looking down the hydrostat (onto planes perpendicular to the [111] direction). and to allow relatively straightforward model parameterization using standard laboratory tests. the common feature is the presence of microscale flaws such as porosity (which permits inelasticity even in purely hydrostatic loading) and networks of microcracks (leading to low strength in the absence of confining pressure. including exploration and production activities for the petroleum industry. that is predictive over a wide range of porosities and strain rates.0. granite. 1. the Geo1 TXE . The overarching goal is to provide a unified generalpurpose constitutive model that can be used for any geological or rocklike material. and penetration resistance and debris field predictions for the defense community. For computational tractability. the Sandia GeoModel [15] strikes a balance between firstprincipals micromechanics and phenomenological. and semiempirical modeling strategies. (b) the meridional “side” view (thick line) with the cap shown on the more compressive righthand side of the plot using cylindrical coordinates in which z points along the compressive [111] direction. Introduction Simulating deformation and failure of natural geological materials (such as limestone.
exercising this model can require as many as 40 parameters for extremely complicated materials to only 2 or 3 parameters for idealized simplistic materials. Linear or nonlinear shear failure with or without kinematic hardening. Flexibility that permits reducing the model (and the number of required parameters). support for a network of many randomly oriented faults). At low pressure. mixed hardening. The model parameters are defined in the nomenclature table (Appendix B).e. to other more classical failure models.e. Tresca perfect plasticity. At higher pressures. or (by using only a small subset of the available parameters) it can duplicate simpler idealized yield models such as classic Von Mises plasticity and MohrCoulomb failure. noncircular or circular octahedral yield profile). Features and limitations of the GeoModel are summarized below. Features of the GeoModel Depending on how the model parameters are set. SandlerRubin twosurface cap plasticity (approximated). • Ubiquitous jointing (i. associative or nonassociative MohrCoulomb plasticity. continuous surface cap plasticity. Threeinvariant. Three LodeAngle functions (i. but generalized to include shear effects in general loading). Linear or nonlinear elasticity. Von Mises perfect plasticity. porous brittle materials can fail by shear localization and exhibit strainsoftening behavior. • Pressure and shear dependent compaction (similar to pα models hydrostatic loading. associative or nonassociative DruckerPrager plasticity. associative or nonassociative WillamWarnke plasticity. GeoModel overview The GeoModel shares some features with earlier work by Schwer and Murry [42] in that a Pelessone function [34] permits dilatation and compaction strains to occur simultaneously. Tensile or extensile microcrack growth dominates the micromechanical processes that result in macroscopically dilatant (volume increasing) strains even when all principal stresses are compressive. 2 . For stress paths that result in brittle deformation. the GeoModel is capable of any of the following model features • • • • • • • • • • • • • Linear and nonlinear. Linear or nonlinear. Linear and nonlinear. failure is associated ultimately with the attainment of a peak stress and worksoftening deformation. Appendix A gives stepbystep instructions for using experimental data to assign values to the GeoModel parameters. Rateindependent or strainratesensitive yield surface.e.Introduction Model can simultaneously model multiple failure mechanisms.. Thus. Nonlinear compaction function (pressurevolume) with isotropic hardening. void collapse). they can undergo strainhardening deformation associated with macroscopically compactive volumetric strain (i..
• The GeoModel is limited to relatively small distortional (shape changing) strains. The GeoModel reflects subscale inelastic phenomena en ensemble by phenomenologically matching observed data to interpolation functions. though large volume changes are permitted. phase transformation. Alternative cap models (such as the Gurson function) can be incorporated in future revisions if needed. though it is thoroughly validated in compression. only an elliptic cap function is available for modeling shear effects on pore collapse and other mechanisms of plastic volume reduction. the GeoModel is computationally intensive. nor does it evolve in time. though less so than many other models of comparable complexity. (3) suitability to capture trends in characterization data. more generally. the source of inelastic deformation in geological materials (or in rocklike materials such as concrete and ceramics) is growth and coalescence of microcracks and pores.Introduction Limitations of the GeoModel. and (4) physicsbased judgements about how a material should behave in application domains where controlled experimental data cannot be obtained. • The GeoModel is not extensively parameterized (or even tested) for tensile loading. (2) computational tractability.). inelasticity could include plasticity in its traditional dislocation sense or. Primarily. might result from other microphysical mechanisms (internal locking. twinning. Arbitrarily large rotations are permitted if the host code manages the reference stress and strain measures properly (see page 76). • Compared with simple idealized models (which are well known to give unsatisfactory results in nontrivial structural applications). • The GeoModel version that is described in this report treats the material as initially isotropic. Here. 3 . the term “plasticity” is defined very broadly to include any mechanism of inelastic deformation. Under massive confining pressures. without explicitly addressing how the material behaves as it does. • While the hydrostatic crush curve is quite general. • The GeoModel describes material response up to the onset of softening. the GeoModel is a generalized and unified plasticity model. Considerations guiding the structure of the GeoModel’s material response functions are (1) consistency with microscale theory. Kinematic hardening is the only mechanism for deformationinduced anisotropy. Enhanced versions of the GeoModel that support arbitrary anisotropic jointing are available but not documented here. Fundamentally. The GeoModel predicts observed material response. It does not vary with pressure. • The elasticity model is hypoelastic rather than hyperelastic. The host code is responsible for handling material postpeak softening and the accompanying change in type of the governing equations to ensure meshsize independence. • The host code is responsible for satisfying frame indifference (by calling the GeoModel using conjugate reference stress and strain rate measures). • The triaxial extension/compression strength ratio is presumed constant. etc.
because the equation f ( σ 1. but not on the principal directions. σ 2. porosity). and any crosssection of the yield surface on an octahedral plane is called an octahedral profile. σ 2. When hardening is permitted. Any “side view” crosssection of the yield surface on a meridional plane is called a meridional profile. which means that the criterion for the onset of plasticity depends only on the three principal values of the stress tensor. σ 3 ) = 0 is phrased in terms of three independent variables. the yield surface may be visualized as a 2D surface embedded in a 3D space where the axes are the principal stresses. σ 3 ) . The GeoModel makes no explicit reference to microscale properties such as porosity. Any plane perpendicular to the hydrostat is called an octahedral plane. Aside from supporting kinematic hardening. different values for the internal state variables result in different yield surfaces in stress space. Consequently. Any plane that contains the hydrostat is called a meridional plane. σ 3 ) . Points within the elastic domain satisfy f < 0 . inelastic compaction followed by shearenhanced dilatation has long been attributed to an initial phase of void collapse followed later by microcracks opening in shear. For example. Consequently. grain size. When loading is severe enough that continuing to apply elasticity theory would produce a stress state lying outside the yield surface. σ 2. Mathematically.0 (page 1). but they can deform elastically under a much broader range of stress states in compression. as illustrated in Fig. the yield surface) resides primarily in the compressive * Some people prefer that this be called HaighWestergaard space [31] so that the phrase “stressspace” may be reserved for the higherdimensional space defined by the set of all tensors that commute with the stress tensor. the yield surface may be visualized in the 3D Cartesian space. or crack density. the GeoModel presumes that there exists a convex contiguous “elastic domain” of stress states for which the material response can be construed to be elastic.g. The elastic domain is the interior of this surface. the overall combined effects of the microstructure are modeled by casting the macroscale theory in terms of macroscale variables that are realistic to measure in the laboratory. Instead. the yield surface may be expressed in terms of a yield function f ( σ 1. the elastic domain (and therefore its boundary. the material response will instead be inelastic and a different set of equations must then be solved. The GeoModel is exceptionally capable of matching this type of observed compaction/dilatation data. Points on the yield surface satisfy f = 0 and therefore. The boundary of the elastic domain is called the yield surface. When the yield function additionally depends on internal state variables. 4 .Introduction The GeoModel is phenomenological and semiempirical because the physical mechanisms of inelastic material behavior are handled in an ensemble manner. the yield function additionally depends on internal state variables that quantify the underlying microstructure (e. 1. Being a generalized plasticity model. The hydrostat is the [111] direction. ( σ 1. but it does so without demanding that the user supply information about essentially unknowable porosity or microcrack distributions within the material. the GeoModel is isotropic. without explicitly partitioning and modeling each possible contributor to the inelasticity. Brittle materials are very weak in tension. called stress space*. along which all three principal stresses are equal.
depicts a family of yield surfaces.e. The GeoModel does not simulate subsequent softening.0(b). the limit surface may be regarded as fixed — it does not evolve (i. Points outside the limit surface can be reached only transiently in dynamic loading via viscoplastic rate dependence. The “cap” part of that yield surface (shown as a wire frame in Fig. which therefore requires a response from the host code to alter its solution algorithm (perhaps by inserting void or by activating special elements that accommodate displacement discontinuities). 1. Distinction between a yield surface and the limit surface. elastic or plastic. for example. the typical rock yield surface shown in Fig. r it lim s a ur f ce both isotropically and kinematically hardened yield surface kinematically hardened yield surface z initial yield surface isotropically hardened yield surface Figure 1. unlike solid metals. if any. it follows that the set of all possible yield surfaces is contained within the limit surface (see Fig. whereas a limit surface is the boundary of stresses that are quasistatically obtainable by any quasistatic means.0(a) is actually being viewed from the compressive [111] direction. but might not be attainable thereafter. all bounded by the limit surface. but points outside the limit surface are not attainable via any quasistatic process. The GeoModel simulates material response only up to the limit state. a state on the limit surface is attainable at least once. Points outside a yield surface might be attainable through a hardening process. Thus. move around in stress space) as a yield surface can. Since the limit surface contains all attainable stress states. Since the GeoModel does not include postsoftening stress response. 1. All achievable stress states (and therefore all possible yield surfaces) are contained within the limit surface. Points on the limit surface define the onset of material softening. Fig. This sketch shows meridional profiles of an initial yield surface along with hardened yield surfaces that might evolve from the initial surface.. 1. Consequently.Introduction part of stress space where all three principal stresses are negative.1). A yield surface is the boundary of elastically obtainable stress states. Plasticity induces microstructural changes that permit the yield surface to evolve through 5 . because softening usually induces a change in type of the partial differential equations for momentum balance. 1. inelasticity can occur in rocks even under purely hydrostatic compression as a consequence of void collapse.1.0) reflects the fact that.
Chapters 2 and 3 first summarize our notation and outline some important concepts and conventions about the nature of stress. Once all pores are crushed out. Presently. Model parameters (as well as descriptions of internal state variables and other symbols used in this report) are defined in Appendix B. Work is underway to allow these boundaries in stress space to be “fuzzy” to better account for natural material variability. The computational algorithm. For example. Tresca theory. porous materials tend to have closed convex yield surfaces. Postpeak softening in a material might lead to a stress at rupture that is smaller than at the peak. Of course. Chapters 8 and 9 summarize verification and validation tests that have been completed to date for a variety of materials. Thus. Despite being developed primarily for geological applications.1. maximum principal stress theory. in which case the yield surface becomes a cylinder centered about the [111] direction.). as indicated in Fig. along with sample input sets for realistic and idealized materials. Chapter 4 describes the GeoModel theory (elasticity. 1. The set of all possible yield surfaces is contained within the limit surface. and MohrCoulomb theory are also supported in the GeoModel by using the simplified input sets summarized in Appendix B. and plotable output will be discussed in Chapter 6. the GeoModel treats the limit surface (and each yield surface) as a sharp threshold boundary. full use of nearly all GeoModel features is often required to adequately validate the model for realistic rocklike materials. However. by using only a small subset of available parameters. A material can have an infinite number of yield surfaces generated via various pathdependent hardening processes. a material can withstand an unlimited amount of pressure. mapping out the limit surface for a given material requires using a new sample for every experiment. but it can have only one limit surface. Other classical models such as DruckerPrager plasticity. subroutines. but it is the collection of peak — not rupture — stress states that defines the limit surface. which itself introduces uncertainty regarding variability in material composition and microstructure. however. Consequently. Each such experiment can have only one peak stress state. Replicating analytical results from simplified theories is an important aspect of verification of the GeoModel. the GeoModel is truly a unification of many classical plasticity models. Porous materials are capable of inelastic deformation even under purely hydrostatic loading. 6 . To describe in greater detail how the GeoModel supports its broad range of micromechanisms of failure in a mathematical and computational framework. the limit surface for any material will always be an open convex set. Instructions for determining appropriate model parameters from laboratory data are provided in Appendix A. followed by software quality assurance in Chapter 7. Limit surface characterization is accomplished by performing numerous experiments all the way to the point of material rupture (catastrophic failure).Introduction time. the GeoModel can be instructed to behave precisely like a classical hardening or nonhardening Von Mises model. yield surface definition and evolution. etc. effectively changing the initial material into a mechanically different material.
˜ ˜ Magnitude of a vector: v = ˜ Magnitude of a tensor: A = ˜ vk vk = (2. and a tensor. At the discretion of the host code in which the GeoModel is run. k=1 2 ∑ ∑ Aij . Vector and tensor equations will be presented using indicial Cartesian notation in which repeated indices within a term are understood to be summed from 1 to 3 while nonrepeated indices are free and take values 1 through 3.8) 7 . the following standard operations and definitions from vector and tensor analysis will be employed: Dot product between two vectors: u • v = u 1 v 1 + u 2 v 2 + u 3 v 3 ˜ ˜ = uk vk . (2. and T would denote ˜ ˜ a scalar. i = 1j = 1 3 3 3 (2.g. Inner product between two tensors: A :B = A ij B ij .6) ∑ vk2 . vectors and tensors will be written in symbolic or “direct” notation in which the number of “tildes” beneath a symbol equals the tensorial order of that variable. ˜ ˜ ˜ Dot product between a tensor and a tensor: C = A • B means C ij = A ik B kj . a vector. For example. v .7) A ij A ij = (2. (2. Vector and Tensor notation For this report. several other field variables (e. stress) may be additionally available for plotting. blue variables are user input parameters and green variables are internal state variables available for plotting in the numerical implementation.4) Identity tensor: I is the tensor whose ij components are δ ij and whose ˜ component matrix is therefore the 3 × 3 identity matrix.2) Dot product between a tensor and a vector: y = A • x means y i = A ik x k . Upon occasion. respectively. if i ≠ j (2.Notation 2. s .1) (2.5) (2. Notation Typesetting conventions Throughout this report..3) ˜ ˜ ˜ 1 Kronecker delta: δ ij = 0 if i=j .
Notation
Trace of a tensor A : ˜
trA = A 11 + A 22 + A 33 = A kk . ˜
(2.9) (2.10) (2.11) (2.12) (2.13)
dev Deviatoric part of a tensor: A dev = A – 1 ( trA )I , or A ij = A ij – 1 A kk δ ij . 3 ˜ ˜ 3 ˜ ˜ A First invariant (trace) of a tensor A : I 1 = trA = A kk . ˜ ˜
Second invariant of a tensor A : ˜ Third invariant of a tensor A : ˜
A J 2 = 1 tr [ ( A dev ) 2 ] . 2 ˜ A J 3 = 1 tr [ ( A dev ) 3 ] . 3 ˜
Throughout this report, invariants of the stress tensor σ will be written without the super˜ σ script identifier. For example, J 2 means the same thing as J 2 . The GeoModel supports kinematic hardening in which the shifted stress tensor ξ is defined ξ = σ – α , ˜ ˜ ˜ ˜ where α is the backstress tensor (defined later). The invariants I 1, J 2, J 3 in the nonhard˜ ξ ξ ξ ening theory will become I 1 , J 2 , J 3 when kinematic hardening is used. In materials modeling, tensors are often regarded as higherdimensional vectors. The inner product between two tensors, A and B , is isomorphic to (i.e., geometrically analo˜ ˜ gous to) the dot product between two vectors. This permits the “magnitude” of a tensor, the “direction” of a tensor, and the “angle” between two tensors to be defined in manners analogous to ordinary vector definitions. The direction of a tensor plays a role in the GeoModel by defining the outward normal to the yield surface (which is actually a hypersurface in higherdimensional tensor space). Likewise, the angle between two tensors is used to quantify the concept of nonnormality, discussed later. The derivatives of a scalarvalued function f that depends on a secondorder tensor σ ˜ as well as depending on a scalar κ are given by ∂f ∂f  is a secondorder tensor with ij components  . ∂σ ∂σ ij ˜ ∂f is a scalar. ∂κ
(2.14)
(2.15)
Other derivatives are defined similarly. For example, the derivative of a secondorder tensor A with respect to a another secondorder tensor B is a fourthorder tensor with ijkl ˜ ˜ components ∂A ij ⁄ ∂B kl . Fourthorder tensors do not play a significant role in the GeoModel theory. The only truly important fourthorder tensor is the plastic tangent stiffness tensor, formally equal to the derivative of the stress rate with respect to the strain rate. 8
The Stress Tensor
3. The Stress Tensor
This section defines the stress tensor, its principal values, its invariants, and its sign convention. This chapter describes four canonical stress paths used to parameterize the GeoModel: hydrostatic (HYD), triaxial compression (TXC), triaxial extension (TXE), and shear (SHR). For transient dynamics, the GeoModel additionally requires Hugoniot and/or Kolsky (split Hopkinson) bar data to parameterize the viscoplasticity*. This chapter defines the distinction between the spatial Cauchy stress and the unrotated “reference” stress. In preparation for a detailed discussion of the GeoModel theory, this chapter closes with a detailed description of “stress space” and Lode coordinates. The stress tensor σ , is defined such that the traction vector t (i.e., force per unit ˜ ˜ area), acting on any given plane with unit normal n , is given by ˜ t = σ • n. (3.1) ˜ ˜ ˜ Of course, the traction and normal vectors may be described in terms of their Cartesian components, { t 1, t 2, t 3 } and { n 1, n 2, n 3 } with respect to an orthonormal basis. The stress tensor has a 3 × 3 Cartesian component matrix such that the above equation may be written in matrix form as
t1 t3 σ 11 σ 12 σ 13 n 1 σ 31 σ 32 σ 33 n 3
t 2 = σ 21 σ 22 σ 23 n 2
,
(3.2)
or in indicial form as t i = σ ij n j ,
(3.3)
where (recalling the implied summation convention) the repeated index “j” is understood to be summed from 1 to 3 and the nonrepeated “free” index “i” appearing in each term takes values from 1 to 3 so that the above equation is actually a compact representation of three separate equations (one for each value of the free index). The stress is symmetric, which means that σ ij = σ ji . In continuum mechanics, and in this report, stress is taken positive in tension. This sign convention can be the source of considerable confusion, especially when discussing stress invariants. For example, the trace of the stress, I 1 , is positive in tension. However, brittle materials have very low strength in tension. Consequently, most of the functions defined in this report are nontrivial over only a small range of the tensile states where I 1 is positive. On the other hand,
* Until rate dependence is discussed separately in Chapter 5, all incremental or rate equations in this report are understood to apply only under quasistatic loading and may therefore be regarded as “inviscid” equations. Incorporating viscoplastic rate dependence requires, as a prerequisite, solution of these quasistatic inviscid equations.
9
The Stress Tensor
most of the GeoModel functions are nontrivial over a relatively large range of compressive states where I 1 is negative. To help manage the sign convention problem, we will introduce a new notation that an overbar on a variable denotes the negative. Specifically, for any variable x,
x ≡ –x
DEFINITION OF THE “OVERBAR.”
(3.4)
In our plots of any variable that varies as a function of I 1 , we will usually employ an abscissa of I 1 , which (being the negative of I 1 ) is positive in compression. Any variable typeset with an overbar will be positive more often than negative in most applications. The principal stresses are the eigenvalues ( σ 1, σ 2, σ 3 ) of the stress matrix, positive in tension. Their negatives ( σ 1, σ 2, σ 3 ) are positive in compression.* When cast in terms of the principal basis (i.e., the orthonormal eigenvectors of the stress matrix), the diagonal components of the stress matrix will equal the principal stresses, and the offdiagonals will be zero. The stress deviator S is the deviatoric part of the stress (see Eq. 2.10): ˜ S = σ – 1 ( trσ ) I . ˜ ˜ 3 ˜ ˜
(3.5)
Loosely speaking, the stress deviator is a tensor measure of shear stress. An overall scalar measure of shear will be defined later. The quantity 1 ( trσ ) is called the mean stress, 3 ˜ and we will denote it by p . The negative of the mean stress, p = – p , is called the pressure, and is positive in compression. Noting that the mean stress (or pressure) is simply a multiple of I 1 , its value is an invariant, meaning that the sum of diagonal stress components will have the same value regardless of the orthonormal coordinate system used to describe the stress components. The principal directions of the stress deviator are the same as those for the stress itself, and the principal values for the stress deviator are related to the principal stresses by
1 s 1 = σ 1 –  ( σ 1 + σ 2 + σ 3 ) 3 1 s 2 = σ 2 –  ( σ 1 + σ 2 + σ 3 ) 3 1 s 3 = σ 3 –  ( σ 1 + σ 2 + σ 3 ) . 3
(3.6) (3.7) (3.8)
The trace of S is zero, which implies that the numerically largest principal value of ˜ any nonzero stress deviator will always be positive and the smallest will always be negative. In model parameterization tests, a sufficiently high confining pressure p is typically superimposed on the stress deviator to make all principal stress components compressive even though the principal deviatoric stresses always have mixed signs.
* Of course, if principal stresses are ordered such that σ 1 ≤ σ 2 ≤ σ 3 , then the barred principal stresses will be ordered σ 1 ≥ σ 2 ≥ σ 3 .
10
Being symmetric. Because the stress deviator S has a zero trace. whereas pressure p is just the negative of mean stress: I1 mean stress: p = 3 I1 pressure: p = . J 2 . The “transfer of sign” operator* is defined x sign [ x.9) (3. and so forth. 3 (3. J 2 . and the second invariant can be written alternatively as J 2 = – ( s 1 s 2 + s 2 s 3 + s 3 s 1 ) . J 3 } .11) 3 3 ˜ The fact that these invariants are computed from the stress tensor σ is sometimes empha˜ σ σ σ sized by typesetting them as I 1 . the signed equivalent shear stress τ as τ = sign [ J 2. negative in triaxial compression. which means its value will be the same regardless of which orthonormal basis is used for the stress components. * which is an intrinsic function in most computing languages 11 . and it will be identically equal to the applied shear stress if the stress tensor happens to be in a state of pure shear (also defined below). J 3 ] . it can be ˜ shown that J 3 also equals the determinant of the stress deviator so that J 3 = s 1 s 2 s 3 . where I 1 ≡ – I 1 .10) shows that the invariant J 2 is never negative. Invariants for a tensor α would be written ˜ α.The Stress Tensor Stress invariants The trace operator is an invariant. J α˜ J α } .14) (3. the stress tensor has a total of three independent invariants: I 1 = trσ ˜ J 2 = 1 trS 2 2 ˜ = σ1 + σ2 + σ3 2 2 . (3. y ] = – x if if y≥0 .12) Superimposing an extra pressure P on any stress state causes the pressure to increase from p to p + P . while having no effect on the stress deviator and therefore no effect on the second and third invariants. y<0 (3. Any invariant written without a clarifying superscript should { I1 2 . which permits us to define a supplemental stress invariant.10) 3 3 .3 J 3 = 1 trS 3 = 1 ( s1 + s2 + s3 ) .13) As defined. Similarly defined invariants for some other ξ ξ ξ tensor ξ would be typeset as { I 1 .. Equation (3. The mean stress p is defined to be the average of the principal stresses. 3 be understood to be a stress invariant. the equivalent shear stress will have a numerical sign that is positive in triaxial extension states (defined below).2 = 1 ( s1 + s2 + s3 ) 2 (3. and J 3 .
= S = dev ( σ ) = the stress deviator ˜ dσ ˜ ˜ dJ 3 . not on the principal stress ˜directions. Equivalently.+ ..∂J 2 ∂φ. Specifically. In situations where the derivative of φ with respect to stress is required. their derivatives may be computed in advance: dI 1 . ∂I 1 ∂σ ij ∂J 2 ∂σ ij ∂J 3 ∂σ ij ∂σ ij In symbolic tensor notation. the invariants are computed in practice directly from fullypopulated 3 × 3 component matrices. T .= ..dJ 2 ∂φ.The Stress Tensor For clarity. (3. Eq. is a good example of ˜ the fallibility of indicial notation.18) (3. This. However. Derivatives of the stress invariants For isotropic material modeling. the result is ∂φ ∂φ tr  = 3  . J 3 . this expansion is written as ∂φ dI 1 ∂φ. Since both S and T ˜ ˜ ˜ are deviatoric.17) (3.19) ∂φ ∂φ∂φ∂φ . S + .15) (3.21) ∂I 1 ∂σ ˜ The factor of 3 appears simply because trI = 3 . tr ( ∂φ ⁄ ∂σ ) . 12 . thereby avoiding the need for an expensive eigenvalue analysis.20) ∂I 1 ˜ ∂J 2 ˜ ∂J 3 ˜ ∂σ ˜ Of particular interest is the trace of the above expression.16) may be written (3. each scalarvalue function of stress. is presumed to depend only on the principal stress values.+ .= I = the identity tensor ˜ dσ ˜ dJ 2 .. (3.=  I + .+ . φ ( σ ) . the function φ ( σ ) is isotropic if and only if it may be expressed alternatively ˜ as a function of the three stress invariants I 1.∂J 3 ∂φ . by the way. J 2.. the chain rule can be applied as follows: ∂φ ∂I 1 ∂φ. the values of the invariants have been shown here in terms of the principal values of the stress and its deviator.16) ∂I 1 dσ ∂J 2 dσ ∂J 3 dσ ∂σ ˜ ˜ ˜ ˜ Because the three invariants are each proper functions of the stress tensor.dJ 3 ∂φ .. because the trace operation gives the same result regardless of which basis is used.. (3.. (3.+ .= T = dev ( S 2 ) = the “Hill” tensor ˜ ˜ dσ ˜ Thus.= ..
If possible. residual is the logarithmic (Hencky) residual strain after full void collapse.22) Special stress states This section defines the four main stress states that are used to parameterize the GeoModel. the residual volumetric strain approximately equals the initial porosity. * Assuming that the matrix material is plastically incompressible. Loading is “hydrostatic” when components of the stress tensor are of the form [σ] = ˜ –p 0 0 0 –p 0 0 0 –p for hydrostatic stress states. triaxial extension (TXE). A porous material. Typical hydrostatic (pressure vs. Hydrostatic testing is very important to parameterization of the GeoModel because it indirectly characterizes the influence of material porosity. 3. These are hydrostatic loading (HYD).23) In practice. The purpose of the GeoModel is to interpolate realistically between known material response at these canonical states to describe material behavior under general stress states. If the residual where ε v residual . will unload along a different path. triaxial compression (TXC). a nonporous material will trace through the same stress states on both the loading and unloading curves.* p Partial pore collapse p Total pore collapse εv residual strain ≈ initial porosity εv Figure 3. the porosity is 1 – e v . 3 ∂ ∑ σ kk k=1 (3. volumetric strain) compression data. the pressure p is usually compressive (and therefore positive).The Stress Tensor ∂φ tr  = ∂σ ˜ k=1 ∑ ∂σ kk 3 ∂φ ≠ ∂φ . a Taylor series expansion of this formula gives porosity ≈ ε v ε residual 13 .1. (3. on the other hand.1.. In this case. as indicated in Fig. Total pore collapse is achieved when the unloading curve (here shown as nonlinearly elastic) is tangent to the loading curve. strain is small. and shear (SHR). hydrostatic testing for the GeoModel should be conducted to sufficiently high pressures to compress out all pores. When hydrostatically loaded to a high pressure and then unloaded. HYDROSTATIC (HYD).
(3.τ 9 (3. but distinct from the third “axial” principal stress (denoted σ A ).26a) (3.26b) (3.( σ A – σ L ) 3 27 The signed equivalent shear stress for triaxial loading is A L τ = 3 σ –σ (3.27) The invariants defined here may be written alternatively in terms of compressive stress measures as I1 = where A L τ = . with respect to the principal basis. (3. ultimately. but the mechanism of failure is catastrophic growth and coalescence of microcracks. resulting in material softening and. (3..0 – 1 0 3 ˜ 0 0 –1 for triaxial stress states. Thus.26c) 2.29) 14 . complete loss of load carrying ability.25) Also. inelastic response is again possible.24) and σA – σL 2 0 0 [ S ] = . σA 0 0 [σ] = ˜ 0 σL 0 0 0 σL for triaxial stress states.The Stress Tensor When a material is loaded under hydrostatic tension instead of compression. TRIAXIAL (TXC and TXE): Loading is “triaxial” whenever two principal stresses (denoted σ L and called the “lateral” stresses) are equal to each other.28) σ –σ (3. 3 3τ + 3 σL J2 = τ2 J3 = 2 3 3 . I1 = J2 = J3 = σ A + 2σ L 1 .( σ – σ ) 2 L 3 A for triaxial stress states for triaxial stress states for triaxial stress states.
σ = – σ .The Stress Tensor σA > σL σA < σL σA < σL σA > σL TXC TXE σL σL Figure 3. Uniaxial stress compression (also called unconfined compression) is a special form of TXC in which the axial stress is compressive and the lateral stress is zero. In the labels. the specimen changes shape such that its lengthtodiameter ratio decreases. These experiments are normally performed on cylindrical test specimens with the lateral load supplied by a pressure bath. For TXC. The term “triaxial” is a bit of a misnomer because there are not really three independent loads applied — the lateral stresses are equal.2. Uniaxial strain compression. Uniaxial stress extension is a special form of TXE in which the axial stress is tensile and the lateral stress is zero. the lengthtodiameter ratio increases even though the length and diameter might individually both decrease. For triaxial compression (TXC) the axial stress is more compressive than the lateral stress. For triaxial extension (TXE) the axial stress is not necessarily tensile — it is merely less compressive than the lateral stress. For TXE. Biaxial compression is a special case of TXE in which the lateral stress is com 15 . but less compressive than the axial stress). Biaxial tension is a special case of TXC in which the lateral stress is tensile and the axial stress is zero. stress σ is positive in tension while stress σ is positive in compression. which is typical in flyerplate impact experiments is a special case of TXC in which the axial stress is compressive. the axial stress is less compressive than the lateral stress. For TXE. Triaxial compression (TXC) and triaxial extension (TXE) Two principal stresses (the lateral stress) are equal. the axial stress is more compressive than the lateral stress. while the lateral strain is zero (making the lateral stress also compressive. For TXC.
Being easily achieved in the laboratory. and then held fixed while all stress components are thereafter varied equally until failure occurs.28) implies that dτ1. the stress difference is increased to a preselected value. The material is first compressed hydrostatically to a preselected bath pressure P bath . For CSD loading. the lateral stresses are held fixed at this value ( σ L = P bath ) while the axial stress is then increased beyond P bath . after which all stresses are again increased by equal amounts. Consequently.30) τ A typical TXC load path τ A typical TXC CSD load path 1 30° 3P bath 3 1 3 I1 3P bath I1 I1 I1 A typical TXE load path A typical TXE CSD load path –τ –τ Figure 3. the stress difference is increased until material failure occurs. at this point. the value of the first stress invariant is I 1 = – 3P bath and therefore I 1 = 3P bath . the lateral stress is held constant ( σ L = P bath ) while the axial stress is varied. Eq. According to Eq.27). In this case. the lateral stress is held fixed (via a pressure bath) while only the axial stress is varied. J 2 > 0 for both TXC and TXE because it is the square of τ . the axial stress might be increased only until the stress difference reaches a given value.The Stress Tensor pressive and the axial stress is zero. the material is first loaded hydrostatically in a pressure bath until all three principal stresses reach a compressive pressure P bath . (3. In a typical triaxial test. Thereafter. TXC and TXE data are essential to parameterize the GeoModel. (3. Of course. When the triaxial leg begins. This causes both the first and second invariants to change such that the path in this stress plot is a straight line with slope 1 ⁄ 3 . For simple triaxial loading. dI1 3 (3. Eq.for triaxial stress loading with fixed lateral stress. In typical triaxial experiments.3. whereas τ < 0 for TXE. the signed shear stress satisfies τ > 0 for TXC. For some experiments. while J 3 < 0 (and hence J 3 > 0 ) for TXE. (3. Triaxial and CSD load paths. These are called constant stress difference (CSD) 16 .= .28) shows that J 3 > 0 (and hence J 3 < 0 ) for TXC.
1. For metals.0b). (3. However.0c (page 1). is uniaxial strain. 17 . except with the axes scaled differently. in which the lateral strain ε L is held constant. (4. τ ) stressatyield points that map out the TXC meridional profile of the GeoModel yield surface. respectively. (4. (3. 3. it would be impossible to definitively distinguish plasticity from nonlinear elasticity). the material strengths in TXC and TXE tend to increase with pressure but in approximately the same proportions so that the ratio of TXC strength to TXE strength is approximately pressure independent.σ A = K – . 1–ν 3 (3. Another form of triaxial loading. for brittle materials. Periodic partial unloading during a test reveals yielding if the unloading stressstrain curve has a noticeably different slope than the loading curve (without unloading.G ε A 3 and ν · · 2G · σ L = . these experiments are run to the point of material failure. applicable to any form of triaxial loading. ε A .11) in which ε L = 0 . ε L = 0 .3 are shaped identically to the TXC profiles except smaller in magnitude. commonly used for dynamic material testing.13) used to express Young’s modulus E and Poisson’s ratio ν in terms of K and G . the TXE failure envelope will be shaped similarly to the TXC envelope. Typically. 1. For uniaxial strain. a series of TXC experiments at various bath pressures and/or stress differences results in a family of ( I 1. the rate of the signed equivalent shear stress and the rate of the first stress invariant are · · 2Gε A · · Uniaxial strain: τ = and I 1 = 3Kε A .31) where K and G are. Consequently. is given in Eq. · Eq. the lateral compressive stress σ L will · increase to prevent lateral motion. at a given value of I 1 . If the lateral strain ε L is held constant while continuing to compress axially. whereas TXC experiments reveal how the apex of the triangle varies with pressure.3. the value of J 2 at failure is generally lower for TXE than for TXC. there is little difference between the stress intensity required to initiate failure in TXC compared to TXE.11).The Stress Tensor experiments. the tangent elastic bulk and shear moduli.32) 3 and therefore the path through stress space is a straight line with slope * The general form of Hooke’s law. and therefore Hooke’s law* in rate form reduces to Uniaxial strain: 4 · · σ A = K + . but lower in magnitude because. For uniaxial strain. (4. As illustrated in Fig. the difference is quite noticeable and (according to idealized microphysical theories) can be attributed to internal frictional resistance to shear crack growth. Usually.31) is a special case of Eq. 3. Because friction increases with pressure. Similar experiments under TXE map out the TXC meridional profile. A plot of the failure envelope in τ vs. the TXE profiles shown in Fig. with Eq. The TXE experiments are mapping out the crosssection of the yield surface along which pressure varies while staying on the “base” of the triangular octahedral profile in Fig. I 1 space is essentially equivalent to the meridional “side” view of the yield surface (Fig.
superimposing the hydrostatic loading of Eq.35). simple shear is a special type of pure shear). In general.The Stress Tensor dτ . This constraining lateral compression makes I 1 larger than it would be when lateral expansion is unconstrained.1 – 2ν = . The larger I 1 results in a shallower slope in the stress trajectory. it has not (to our knowledge) been reported for naturally occurring materials. SIMPLE/PURE SHEAR and PRESSURESHEAR LOADING (SHR). Since Poisson’s ratio typically varies between 0 and 1 ⁄ 2 . positive definiteness of the elastic stiffness tensor merely requires – 1 < ν < 1 ⁄ 2 . The eigenvalues of this matrix are { s.35) is most convenient.23) onto the shear stress of Eq. Uniaxial strain experiments are conducted by applying increasing levels of compression in the axial direction while holding the lateral strain fixed. for triaxial stress with fixed lateral strain. For brittle materials.33) 3 1 + ν where ν is Poisson’s ratio.= dI1 1 2G . pure shear of the form s 0 0 [ σ ] = [ S ] = 0 –s 0 ˜ ˜ 0 0 0 for pure shear (3. pure shear is difficult to attain because one of the eigenvalues is always tensile. pure shear is superimposed with enough confining hydrostatic pressure to make all principal stresses negative (compressive). 16]. but is not permitted to — a lateral compressive force prevents this outward motion.34) where s is the shear stress. (Thus. (3. (3. 0 } . This result should make some intuitive sense. Whereas negative Poisson’s ratio has been observed in manmade materials with reentrant microstructures. For an isotropic material model like the GeoModel. p. gives a state of combined pressureshear (SHR) loading: * Strictly speaking. A material is in a state of simple shear with respect to a given coordinate system if the stress matrix in that system is of the form [σ] = [S] = ˜ ˜ 0 s 0 s 00 000 for simple shear. 18 . – s. yield depends only on the principal stresses. For conducting material characterization experiments. As the axial strain is compressed.3 3K 1. Performance of the GeoModel has not been verified for materials with negative Poisson’s ratio. Frequently. any stress state that is deviatoric with one eigenvalue being zero is said to be a pure shear [27. the material “wants” to expand laterally. (3. so there is no practical difference between simple and pure shear (except when the model is anisotropic because of kinematic hardening). .* this result shows that the trajectory in the meridional profile will generally have a shallower slope under uniaxial strain loading than under uniaxial stress loading. (3. Specifically.
except rigidly rotated by the same amount. and special “objective” rates must be integrated in constitutive models. but a current configuration that is identical to the current configuration of the first problem.37c) (3. the principle of material frame indifference demands that if a second problem were considered that had the same initial configuration.1) is the “familiar” stress tensor used in everyday engineering applications.39) By working in the unrotated reference configuration. then the predicted spatial stresses for the second problem should be identical to those of the first problem. this approach will give results identical to a spatial constitutive model that is cast in terms of polar objective rates. Upon receiving the GeoModel’s update of the unrotated stress.36) The invariants for combined pressureshear loading are I 1 = – 3p J2 = τ 2 J3 = 0 τ = s . except also rigidly rotated. and all rates that appear in the constitutive model are more easily integrated true rates instead of corotational rates. This concept is quite different from a mere basis change because the initial configuration is identical for both problems. (3. Roughly speaking.37b) (3. the GeoModel predicts the stresses for the nonrotating problem. (3. for combined pressureshear loading for combined pressureshear loading for combined pressureshear loading (3. then the unrotated stress is simply σ unrotated = R T • σ spatial • R ˜ ˜ ˜ ˜ (3.The Stress Tensor [σ] = ˜ s–p 0 0 0 –s–p 0 0 0 –p for combined pressureshear loading. If R is the proper orthogonal tensor (found from a polar decomposition of the defor˜ mation) that characterizes the material rotation.38) Spatial and Reference stress (frame indifference) The spatial Cauchy stress tensor defined in Eq. Let us now denote that stress by σ spatial .37a) (3. Moreover. the host code then rotates the predicted stress back into the spatial frame. A mathematically equivalent (and numerically more accurate and efficient) strategy instead applies the constitutive model within an unrotated reference configuration. With this approach. rotation of internal variables is not required. The ˜ elasticity component of solids models requires knowledge of both the initial and current configurations. Satisfying material frame indifference in a spatial context can be computationally expensive and errorprone because anisotropic internal state variables (such as directions of material fibers or orientation of the backstress) must be rotated into the spatial frame. 19 .
(2) call the GeoModel. turbine blades. see page 76. ˜ rate) mentioned in this report are understood to be cast in the unrotated configuration (material frame). tumbling rock fragments.0 (page 1). the “stress” σ must be understood to be the ˜ unrotated stress σ unrotated Likewise all other vector or tensor variables (such as the strain . The z coordinate is proportional to the mean stress. the “unrotation” strategy can give considerably more accurate answers because the host code may. Therefore.. Any host code that uses the GeoModel must (1) perform these unrotation operations.g. The natural symmetries suggest instead using cylindrical ( r. σ 2. which makes Lode coordinates a natural choice when decomposing tensors in to their isotropic and deviatoric parts. we will usually display meridional profiles as r vs. Therefore. Superimposing pressure on a stress state changes only the axial zcoordinate. vortices.. The radial r coordinate equals the magnitude of the stress deviator. σ 3 ) . but the yield function is often cast in terms of different independent variables to exploit the yield surface’s natural symmetries optimally. optionally. the yield surface must have 120° rotational symmetry about the [111] (hydrostat) direction and reflective symmetry about the TXE and TXC axes in the octahedral plane (i.e. Meridional and octahedral profiles are illustrated in Fig. 1. at a given value of θ is called a meridional profile. θ ) coordinates unchanged.The Stress Tensor For problems involving massive material rotation (e. the view looking down the [111] axis). z (where z = – z ). etc. the Lode angle θ is defined so that it equals zero in SHR. In this report. and a plot of r vs. A constant θ plane is a meridional plane. z . The principal Cartesian coordinates ( σ 1.e. leaving the octahedral ( r. For more information. It varies from – 30° in TXE to +30° in TXC. σ 3 ) comprise an adequate choice for characterizing stress space.. σ 3 ) must be independent of the ordering of the eigenvalues. Hence. 1. We have placed a bar on the symbol θ for the angular coordinate because we intend to define it so that it will be positive in TXC and negative in TXE. Throughout the remainder of this report.0 (page 1). The value of the yield function f ( σ 1. use the rotation tensor R n at the beginning of the time step when computing the starting value of σ unrotated . as seen in Fig. The angular coordinate is a measure of the 20 .). on a plane of constant z ) is in an octahedral profile. σ 2. Any constantz plane is an octahedral plane. this approach supports socalled “strong objectivity” [38] in a very natural way. Because most of the yield surface resides in the compressive domain where z < 0 . and any crosssection looking down the [111] axis (i. but ˜then the host code ˜ may use R n+1 at the end of the step when recasting the updated value of σ unrotated (output ˜ ˜ of the GeoModel) to the spatial frame. the yield surface may be visualized in a 3D space for which the Cartesian coordinates are these principal stresses. z ) coordinates — called Lode cylindrical coordinates — for which the zaxis is parallel with the [111] symmetry axis. σ 2. θ. and then (3) rerotate the result back to the spatial frame upon return. Lode coordinates Any isotropic yield function may be expressed in terms of the principal stresses ( σ 1.
40a) J3 3 3 / 2 . when we give the mathematical formulation for the GeoModel yield function.The Stress Tensor relative proportions of the principal values of the stress deviator. in this case. Thus. is the [111] hydrostat). eliminating the need for an eigenvalue analysis. θ. and TXC states. For example. since the zcoordinate is the projection of the stress onto the [111] axis. Also. all three stress invariants).40b) (3. r = 2J 2 (3. and J 3 scalar stress invariants. When simpler yield models (DruckerPrager) are independent of the third stress invariant. 21 . Specifically. When phrased in terms of Lode coordinates. they are therefore independent of the Lode angle. so z is positive on the compressive hydrostat. it can be shown [31] that the cylindrical Lode coordinates may be determined directly from the I 1 . Later. Performing these necessary but tedious coordinate transformations from principal stresses to cylindrical Lode coordinates. J 2. SHR.40c) The square root coefficients are merely byproducts of the coordinate transformations. as the Lode angle varies from – 30° to +30° . the 3 appears because the magnitude of the [111] vector is 3 . The axial zcoordinate is positive on the tensile part of the hydrostat.  2 J 2 I1 z = 3 (3.sin 3θ = – . The GeoModel must include a noncircular yield profile to reproduce TXE/ TXC strength differences clearly evident in the data for geological materials. Using the above formulas. which makes their octahedral yield profile a circle. σ 2. the GeoModel must necessarily use all three Lode coordinates (equivalently. Thus. thereby eliminating the need for an eigenvalue decomposition. the yield function is ultimately implemented in the form f ( I 1. θ. The Lode radius r is zero if J 2 =0 . z ) . J 2 . it will be phrased as f ( r. σ 3 ) to be alternatively expressed in the form f ( r. z ) . The change in variables from principal coordinates to Lode coordinates permits the yield function f ( σ 1. Cylindrical Lode coordinates are especially useful because they may be expressed in terms of stress invariants. which should not be too disturbing since the angular coordinate for any cylindrical system is undefined when the point in question lies on the symmetry axis (which. the stress transitions through TXE. the Lode angle θ is undefined when J 2 =0 . J 3 ) . The invariant J 3 influences only the Lode angle. We define the Lode angle θ to be positive in TXC and negative in TXE. The symmetry properties of the yield surface may be used to reconstruct the octahedral profile over the full range from 0° to 360° . the yield function needs to be defined only over a 60° sextant on any octahedral plane.
An octahedral profile. z at a fixed value of θ . A meridional profile is generated by plotting r vs. Θ is an angle that varies over the full range from 0 to 360° . which corresponds to a yield surface crosssection at a given value of z . SH R 22 .ArcSin [ sin ( 3Θ ) ] 3 (3.4. which is why the TXC axes are always on an apex of the rounded triangle (i. Given a yield function f ( r. corresponding to higher strength). octahedral profiles are obtained by parametrically plotting Cartesian coordinates x 1 = r cos Θ and x 2 = r sin Θ (3. as illustrated in Fig. Rather than plotting r vs. x 2 ) points on the octahedral yield profile may be generated parametrically as Θ varies from 0 to 360° . θ . θ.41) Here. and finally.4. TXE ( θ= –30 ) SH R TX C (θ =0 σ3 ) R SH =0 (θ (θ =3 o 0 ) <σ 1 σ2 3 >σ 2 <σ 3 σ 3 >σ 2 >σ 1 C TX 0 =3 (θ o ) ) σ 1< σ1 >σ SHR ( θ=0 ) SHR ( θ=0 ) σ 3< σ2 TXC ( θ=30o ) E TX (θ =0 ) ) 30 =– (θ σ1 σ 1 >σ 2 >σ 3 <σ 2 <σ 1 σ3 σ2 (θ ) =0 TX E (θ =– R SH 30 ) Figure 3. yield profiles may be generated by solving f = 0 to obtain r expressed as a function of ( θ. An octahedral yield profile. which is permitted to vary only over the range from – 30° to 30° . z ) . is generated from the fullrange angle Θ by the sawtooth function 1 θ = . the value of corresponding Lode radius r can be found from the yield condition. The Lode angle θ alternates cyclically from – 30° in TXE to 30° in TXC because the yield threshold must be independent of the eigenvalue ordering. 3.. The Lode angle θ . farther from the origin. the family of ( x 1. z ) . describes how the Lode radius at yield varies with the Lode angle.42) With θ known.The Stress Tensor Octahedral yield profile visualization. Geological materials tend to be stronger in TXC than in TXE.e.
I 1 .5 shows that the normal to the yield profile in a distorted plot does not correspond to the normal in stress space. whenever we plot the meridional profile as τ vs.The Stress Tensor Meridional yield profile visualization.40a) and (3. The middle plot shows the magnitude of the stress deviator plotted against the pressure. keep in mind that the actual meridional crosssection in stress space is smaller in width by a factor of 2. however. 3.577 . we will plot the TXC ( θ = 30° ) profile. depending on whether the loading is closer to TXE or TXC. we will depict a geometrically distorted view of the meridional profile by instead plotting τ vs. Thus. resulting in a plot eccentricity of 1 ⁄ 3 = 0. Recalling from Eqs. The last plot shows the equivalent shear stress plotted against the first stress invariant.5. 23 . z for a given value of θ . Typically. To draw a geometrically accurate meridional crosssection of the yield surface. z plot by a factor of 6 = 2. (3.5. the profile should be plotted as r vs.40c) that J 2 = r ⁄ 2 and I 1 = 3 z . Meridional profile distortion is an issue only when ascertaining the direction of the yield surface normal. as illustrated in Fig. r isomorphic (geometrically accurate) 2J 2 distorted τ = J2 distorted z p I1 Figure 3.45 . for a plot eccentricity of 6 = 2. I 1 is equivalent to changing the aspect ratio of an r vs.45 . Figure 3. where the signed equivalent shear stress τ equals ± J 2 . Only a plot of r vs. Using the r and z Lode coordinates as the axes in a meridional plot ensures that lengths and angles in the meridional profile will equal corresponding lengths and angles in stress space.45. Many times. z will result in a geometrically accurate depiction of a meridional crosssection of stress space for which angles and lengths are preserved. I 1 . a plot of τ vs. Distortion of the meridional profile when using nonisomorphic stress measures.
Therefore.43a) middle: σ2 = (3.43c) into the failure criterion. 3. z ) . f ( σ 1. 24 . Each distinct sextant in Fig.43c) These formulas constitute a closedform solution for the ordered eigenvalues of any real * symmetric 3 × 3 matrix. J 3 ) . σ 3 = σ 3 . θ.+ J 2 cos θ – 3 3 (3. Using these formulas.cos θ – 2 3 3 I1 sin θ = . as required in the GeoModel.The Stress Tensor Closedform solution for ordered eigenvalues. σ 2.+ . θ = 30° falls on the piplane axis corresponding to the smallest eigenvalue.cos θ + 2 3 3 z.43b) high: σ3 = .+ . the tensile eigenvalues must be ordered σ 3 ≥ σ 2 ≥ σ 1 . can do so by simply substituting crit Eq. whereas θ = – 30° falls on the axis of the largest eigenvalue. Recalling that Lode cylindrical coordinates merely represent a coordinate change from the principal coordinates ( σ 1. σ 2. The above closedform solution for ordered eigenvalues is applied in Appendix B (page B22) to convert the MohrCoulomb theory of failure into a formulation expressed in terms of stress invariants. but defined to be zero in TXE and 60° in TXC. σ 3 ) to a new set of coordinates ( r. which is more convenient for plasticity modeling because it can be differentiated without an eigenvector analysis (see Eq.16). Letting the compressive eigenvalues be ordered σ 3 ≤ σ 2 ≤ σ 1 .4 merely corresponds to a different eigenvalue ordering. (3.J 2 sin θ 3 3 I1 sin θ = . f ( I 1. Malvern’s angle α is a Lode angle. (3. * The solution quoted here is equivalent to the closed form solution derived by Malvern [32] via a trigonometric substitution. it follows that inverse transformation formulas should exist for obtaining the principal stresses from Lode coordinates.– J 2 cos θ + 3 3 I1 2 = . J 2.rsin θ . any material model that seeks to initiate failure when the largest crit tensile principal stress σ 3 reaches a critical value. Regardless of the sextant in which the stress resides. For example. can be immediately recast into a form expressed in terms of stress invariants. σ 3 . and the inverse transformation formulas are low: σ1 = z. not just a stress. σ 3 ) .+ 2 r sin θ 3 3 rzsin θ . any yield function that is stated in terms of principal stresses.– . transformation formulas that convert cylindrical coordinates back to Cartesian coordinates only need to be defined over the range from – 30° through +30° to determine the ordered eigenvalues. 3.
The conjugate stress is the unrotated Cauchy stress defined in Eq. all implementations of the GeoModel have approximated the strain rate by the unrotated symmetric part of the velocity gradient: 1 ∂v m ∂v n · ε ij ≈ . the GeoModel is founded upon an additive · · decomposition of the strain rate ε into separate contributors: ε e from elastic straining and ˜ ˜ · ε p from inelastic straining: ˜ · · · ε = εe + εp . entail adjusting material parameters appropriately.1) ˜ ˜ ˜ The GeoModel permits the host code to employ any definition of the strain so long as its rate is conjugate to the stress σ in the sense that the work rate per unit volume is given by ˜ · σ :ε . 3.3) exactly equals the unrotated logarithmic (Hencky) strain rate for any deformation having stationary principal stretch directions. If. x is the current spatial position vector.39. Using a different choice for the conjugate stress and strain rate measures would. At present. (4. All GeoModel material parameterizations to date have been based on the above approximation for the strain rate.3) even if the distortional strains are always small. so Eq. (4. It is an excellent approximation to the Hencky strain rate even when principal stretch directions change orientation as long as the shear strains remain small (volumetric strains may be arbitrarily large). The approximate strain rate in Eq. but the time · integral of ε ij does not evaluate to zero [11].GeoModel theory 4. and the tensor R is ˜ ˜ ˜ the rotation from the polar decomposition of the deformation gradient.+ . fatigue). material rupture generally occurs well before shear strains become large.2) ˜ ˜ To satisfy the principle of material frame indifference. (4.3) is a prudent choice for the strain rate measure. R mi R nj . (4. (4. it is not precisely equal to the rate of any proper function of the deformation*. GeoModel theory Being a generalized plasticity theory. 25 . For geological applications. The strain rate in Eq. * Paths can be devised for which the starting and ending configurations are identical. for general deformations. then a proper strain rate should be used instead of Eq. however. (4. the model is to be subjected to significant cyclical loading (e. the host code must cast the stresses and strain rates in an unrotated configuration.3) where v is the velocity vector. Henceforth. of course.3) is an approximation because. 2 ∂x n ∂x m (4. all references to the · stress σ ij and the strain rate ε ij must be understood to be the unrotated stress and strain rate. .g.
(4. Stepbystep instructions for determining elastic properties from measured data are provided in Appendix A.e. defined (4. Additional elastic parameters are available for materials whose elastic properties are affected by inelastic deformation (see Eqs. the moduli are stressdependent tangent moduli (i.6) (4. 3 (4. Similarly. ε v is the volumetric elastic strain rate computed by the trace operation. the user merely specifies constant values for the bulk modulus K and the shear modulus G .5) Here. Eq.4) can be written as two separate and much simpler equations. · e · e . and γ ij is the deviatoric part of the elastic strain rate.5) remains valid for volumetric expansion ( ε v < 0 ) and tensile mean stresses ( p < 0 ) as well.e. For nonlinear elasticity.GeoModel theory Elasticity The GeoModel supports both linear and nonlinear hypoelasticity*..4) Because the elastic tangent stiffness tensor. For linear elasticity. is presumed to be isotropic. (4.8) because deviatoric tensors always have eigenvalues of mixed signs. one for the volumetric response and the other for the deviatoric response: ·e · p = Kε v and ·e · S ij = 2Gγ ij .33 and 4. G and K are the tangent shear and bulk elastic moduli. When a potential exists.7) ·e S ij is the stress deviator. The GeoModel presumes the material is elastically isotropic and that the elastic stiffness tensor C ijkl is itself isotropic (i. recall. Eq. then the formulation is “hyperelastic.34). (4. deformationinduced elastic anisotropy is not included). ·e ·e ε v = ε kk . (4.8) We have used the overbar (which. Three parameters are available for fitting the nonlinear tangent bulk modulus K to laboratory data obtained from unloading curves in hydrostatic compression. three parameters are available for fitting the nonlinear tangent shear modulus G indirectly from triaxial test data. p is the pressure (negative ·e of the mean stress). simply denotes the negative of a variable) in our equation for the pressurevolume response because the mean stress is typically comprese sive (negative) in most applications of the GeoModel and therefore p and ε v are typically e positive. (4. * “Hypoelastic” means the stress can be written as a function of the strain.” 26 . Of course. Consequently.. but is not derivable from an energy potential.· e γ ij ≡ ε ij – 1 ε v δ ij . the stress is governed by a rate form of Hooke’s law: ·e · σ ij = C ijkl ε kl . slopes of the tangents to the stressstrain curves). 4. No overbar is used in Eq. C ijkl .
† Here.4) is hyperelastic (i. The GeoModel’s functional forms for the nonlinear elastic tangent moduli are phenomenological to permit tight empirical fits to experimental data for a wide variety of materials. page 139): * The generalized nonlinear elasticity formulas on page 37 may be used when elastic properties appear to be affected by inelastic deformation. the Lame modulus λ . the GeoModel is hypoelastic if g 1 ≠ 0 . 27 . G = go 1 – g1 (4. the stress rates are related to the strain rates by† · · σ A – 2νσ L · ε A = E and · · – νσ A + ( 1 – ν )σ L · ε L = . the shear modulus is typically found indirectly from triaxial loading data. strain curve).e. and the shear modulus G may be determined from the wellknown elasticity equations [18]. I 1 1 1 – g 1 exp ( – g 2 J 2 / 2 ) .10) In these equations. Because Eq. volumetric strain plot obtained from hydrostatic testing. then the bulk modulus K .* Further descriptions of the physical meanings of the parameters in these equations are given Appendix B. Assigning values to the elastic constants. 2(1 + ν) (4. At the user’s option. we are writing Hooke’s Law in rate form to allow for the possibility that the elastic moduli might be nonlinear. Suitability of these functions for fitting material data is demonstrated in Chapters 8 and 9. respectively. (4. Incidentally.. derivable from an isotropic elastic potential) if and only if the shear modulus is constant and the bulk modulus depends at most only on I 1 . Thus. E K = 3 ( 1 – 2ν ) Eν λ = ( 1 + ν ) ( 1 – 2ν ) E G = . For triaxial loading.9) (4. starting on page 96.. The bulk modulus K is determined from the local tangent of the elastic part of a pressure vs. for example.12) Because the GeoModel casts its elasticity model in terms of the bulk modulus K and the shear modulus G . the b k and g k parameters are material constants determined via nonlinear regression fitting to the unloading portions of hydrostatic compression and triaxial compression experiments.GeoModel theory Nonlinear elasticity.11) where E and ν are. (4.34) permits the shear modulus to vary. as described in Appendix A. If E and ν are known. the following formulas are convenient for converting various combinations of elastic constants into expressions involving only K and G (see. it can be shown that an elasticity model of the form in Eq. Rather than determining the shear modulus directly from a shear loading experiment (where GeoMaterials tend to be weak). Young’s modulus and Poisson’s ratio. the GeoModel supports nonlinear elasticity by permitting the elastic tangent moduli to vary with the stress according to b2 K = b o + b 1 exp – .. E (4.. the elastic properties used here are the tangent moduli (i. Ref. based on the local slope of an elastic stress vs.e. [18].
G = 3 ( 1 + ν ) ( 1 – 2ν ) 3K – 2G ν = 2 ( 3K + G ) λ = 2 ( 3K + G ) 3 ν 4 2G + λ = K + . Eq. void collapse.15c) may then be used to determine the tangent shear modulus G .” then material response will be elastic and therefore reversible (nondissipative).. each time step) by tentatively presuming that the loading is elastic. If. but also any other mechanisms that lead to a marked departure from elasticity.e. the GeoModel is a generalized plasticity model. the tentative assumption of elasticity is validated and the actual updated stress is set equal to the trial elastic stress. however. σ 2. We will begin by discussing yield surfaces in some generality and then progressively work towards the precise functional form for the GeoModel yield surface. The term “plasticity” is broadened to include not only the usual flow of material via dislocations (a phenomenon that has actually been observed in brittle materials when they are loaded under extraordinarily high confining pressure). on the other hand. the trial elastic stress falls outside the yield surface. and the solution phase is then solved anew using the equations governing inelastic deformation. Before discussing these inelastic governing equations.17) · · · A fixed lateral stress implies that σ A = σ A – σ L . the material will undergo irreversible structural 28 .13) (4. If the stress state ( σ 1. the GeoModel begins each solution phase (i. With the bulk modulus K having been obtained separately from hydrostatic test data. Once the stress becomes critically severe. σ 3 ) is “not too severe. If so. (4.= 1–ν 3K + 4G 3KE G = 9K – E (4. Thus.16) (uniaxial strain modulus — see Eq.11) implies that triaxial experiments conducted under constant lateral stress · ( σ L = 0 ) satisfy · · σ A = Eε A . the local tangent of the curve equals Young’s modulus E .14) (4. when a stressstrain curve is obtained by plotting the stress difference σ A – σ L against the axial strain ε A for a triaxial loading experiment in which the lateral stress is fixed. (4. 3.G 3 9KG E = 3K + G ν 3K – 2G . the “yield” surface itself characterizes them all in an ensemble phenomenological manner. we must first characterize the yield surface itself. or perhaps even phase transition. This produces a trial elastic stress. Eq. Mathematically. (this applies if lateral stress is constant) (4. then the tentative assumption of elastic response was wrong. The elastic limit (yield surface) Like most plasticity models.= 1–ν 3K + 4G 4 E( 1 – ν ) K + . Rather than explicitly tracking each of these microscale failure mechanisms explicitly. Examples include crack growth.G 3 E = 18KG ν 3K – 2G E18KG .15) (4.GeoModel theory 2 λ = K – .31). which is then checked to see if it is inside or on the yield surface.
When kinematic hardening is disabled (i.1. in which case J 2 is the second invariant of the shifted stress tensor ξ = S – α . The function F c depends on an internal state variable κ whose value controls the hydrostatic elastic limit. sketched in Fig. If continuing to apply elasticity theory would move the stress into regions outside the yield surface. is used to account for differences in material strength in triaxial extension and triaxial compression. The GeoModel yield criterion and yield function are GEOMODEL YIELD CRITERION: [ F f ( I 1 ) – N ] F c ( I 1. The function F f represents the ultimate limit on the amount of shear the material can support in the absence of pores (i. The function Γ ( θ ξ ) . The material parameter N characterizes the maximum allowed translation of the yield surface when kinematic ξ hardening is enabled. κ ) .5. 4.4. κ ) ξ J 2 = . The material response is elastic whenever the stress is on the inside of the yield surface. which can evolve toward the limit surface via kinematic hardening as explained later in the context of Fig. the function F c accommodates material weakening caused by porosity.7. The “building block” functions F f and Γ are used to describe the elastic limit caused by the presence of microcracks. when ˜ ˜ ˜ ξ ˜ N is specified to be zero).15.e.GeoModel theory changes that manifest as inelastic strains (nonrecoverable upon removal of the load). where α is the backstress. whereas the function F c accounts for strength reduction by porosity. κ ) = J 2 Γ 2 ( θ ξ ) – [ F f ( I 1 ) – N ] 2 F c ( I 1. as explained later in the context of Figs. the backstress will be zero and therefore J 2 would be simply the invariant of the stress deviator. Briefly. 29 . The yield criterion corresponds to f = 0 . whereas F f – N defines the yield threshold associated with cracks. F f represents the softening transition limit threshold.. α.e.19) The remainder of this chapter is devoted to motivating the functional forms of these equations and defining the numerous variables that appear in them. where θ ξ is the Lode angle of the shifted stress. resulting exclusively from microcracks). ˜ ˜ (4. and 4. the yield function f is defined such that elastic states satisfy f < 0 . The function F f describes the limit strength. By appearing as a multiple of [ F f – N ] 2 .. then plasticity equations are applied..18) (4. 1. 4. 4. Γ( θξ ) GEOMODEL YIELD FUNCTION: ξ f ( σ.
Stress states outside the yield surface for which f > 0 are unachievable except through a hardening evolution of the internal state variables ( κ and/or α ) corresponding to a funda˜ mental change of the underlying microstructure of the material. the GeoModel parameters can be set to alternatively duplicate classical idealized Von Mises or MohrCoulomb theory. When the stress is on the yield surface.. 1.0(c) shows the yield surface profile from a perspective looking down onto a plane — called an octahedral plane — that is perpendicular to the [111] symmetry axis and therefore represents a crosssection of the yield surface at a given pressure. the yield surface for any isotropic yield model possesses 120° rotational symmetry about the hydrostat (i.18). Stresses outside the limit surface are unachievable by any quasistatic means. along with a family of other profiles from which the yield surface might have evolved over time (via continuously varying values of the κ internal state variable).* Figure 1.e.0a on page 1).0(c). Since the onset of yield must not depend on the ordering of the principal stresses. 30 .0(b) shows a “side” meridional profile of the yield surface in bold.. The size of the octahedral profile at various pressures is governed by the functions F f and F c . 1. the octahedral profile changes size. the octahedral profile is somewhat triangular in shape.0 (page 1). (4. Because the yield surface is farther from the origin on a TXC axis than on a TXE axis. the yield surface in the GeoModel characterizes the point of departure from elastic to inelastic behavior. and it does so without translating in stress space). 1. octahedral profiles that pass through the cap will expand isotropically (i. implying that materials of this type are very weak in tension. then plasticity equations will be applied. When κ increases in response to pore collapse.” Like classical plasticity models. As seen in Fig. Stress states for which f = 0 are said to be “on the yield surface. but not shape. the yield surface will have a shape similar to the one illustrated in Fig. Doing this would be inappropriate when modeling real materials.e. This criterion describes the geometrical shape of the yield surface in stress space. Fig. but it can be useful in benchmark testing. This periodic asymmetry corresponds to differences in the failure limit under triaxial compression (TXC) and triaxial extension (TXE). Very little of the yield surface in Fig. The internal state variable κ controls the location of the yield cap (wire frame in Fig. The Γ ( θ ) function characterizes the shape of the octahedral profile because r is proportional to 1 ⁄ Γ ( θ ) .GeoModel theory The principal goal of this section is to describe in great detail the physical motivations of the GeoModel yield criterion cited in Eq. this material has higher strength in TXC than in TXE. * At the user’s option.7 shows a similar plot when kinematic hardening is allowed. For rocks and rocklike materials. 1. as well as reflective symmetry about any of the triaxial compression or triaxial extension axes labeled TXC and TXE in Fig. Figure 1. and if applying elasticity theory would result in an updated stress that falls outside the yield surface. The amount of isotropic expansion or contraction varies with pressure in such a manner that the family of yield surfaces corresponding to various values of κ is bounded by the shear limit surface.0(b) exists in the tensile domain (left side of the meridional plot). Elastic stress states are “inside” the yield surface ( f < 0 ). J2 = Ff ( I1 ) ⁄ Γ ( θ ) . the [111] axis). 4. 1.0(c).
22) In this section. The means by which the yield surface evolves in response to time variation of κ and/or α is discussed separately. the yield function in Eq. induced anisotropy (Bauschinger effect) in a Figure 4.21) where 1 ξ J 2 = .21) is most naturally cast in terms of the cylindrical Lode coordinates as f = f ( r.1. defined ξ ij ≡ S ij – α ij . with the internal state variables ( κ and α ) regarded as con˜ stants. as well as I 1 and an internal state variable κ that characterizes isotropic hardening caused by void collapse or softening caused by porosity increases. θ. Given that an isotropic yield function possesses alternating 30° symmetry about the [111] direction in stress space.tr ξ 3 .21).tr ξ 2 2 ˜ and 1 ξ J 3 = . J 2 . 3 ˜ (4. ˜ Before discussing the physical foundations of the GeoModel yield function in Eq. J 3 . (4. the yield surface translates in stress space. Consequently. Specifically.23) 31 . Thus. z . the yield function is isotropic with respect to the shifted stress deviator ξ ij .κ ) .GeoModel theory Though not needed in many applications. both elastically and plastically. The GeoModel is otherwise fully isotropic. the GeoModel yield function is of the form ξ ξ f = f ( I 1. (4. the backstress α ij ˜ is a deviatoric tensorvalued internal state variable that defines the origin about which the initial yield surface is centered. 4. we will focus on how the yield function depends mathematically on the ξ ξ stress invariants ( I 1. we will first discuss qualitative features of any yield function of the more general form in Eq. α ˜ ξ ˜ (4. thereby supporting deformationBackstress and shifted stress.20) kinematically hardened yield surface S As illustrated in Fig. the GeoModel supports kinematic hardening for which the symmetry axis of the yield surface is permitted to shift in stress space so that the invariants in the yield function are based on the shifted stress tensor.1. The backstress is initially zero.19).κ ) . (4. we seek to describe the size and shape of the yield surface at an instant frozen in time. but then evolves according to an evolution equation described in detail on page 59. When the backstress yield surface tensor changes. (4. J 2 . J 3 ) . implying that it depends on the invariants of the shifted stress deviator. (4. limited capacity.
26) As was the case with the g function. h 1 and h 2 have been introduced here only to illustrate the basic structure of the GeoModel’s yield function. Figure 4.GeoModel theory where these Lode coordinates are defined with respect to the kinematically shifted origin in stress space. (4. but does not construct one explicitly). Thus. the GeoModel’s yield function is structured such that g ( θ. the view of the yield surface looking down the [111] direction). and then the profile drops to zero when pressure becomes large enough to collapse voids. z . 32 .κ ) ] 2 (Any isotropic yield function can be written in this form.) (4.κ ) is separable into the product of two distinct functions. In this case. The yield function corresponding to Eq. one depending only on θ and the other depending only on z and κ . the GeoModel assumes that the shape of the octahedral yield profile is the same at all pressures — only its size varies with pressure.25) At present. microphysical research has focused on the effects of only one of these mechanisms at a time. (4..24) may be written f = r 2 – [ g ( θ.κ ) = 0 (otherwise.κ ) (This form results from a separability assumption. z .κ ) is regarded as a material function determined from experimental data and is introduced here only to discuss the structure of isotropic yield functions in generality (i.24) to be structured in the general form: r < h 1 ( θ )h 2 ( z .2(b) shows the general * Some evidence suggests that the octahedral yield profile should in fact vary in shape from strongly triangular at low pressures to nearly circular at extraordinarily high pressures [23]. The GeoModel’s specific formulations will be discussed soon. For any given values of θ .κ ) .e. Consequently. without loss in generality. Consequently. z . the GeoModel has an implied “g” function.. elastic stress states for any isotropic yield function always can be characterized in the general functional form r < g ( θ .2(a) shows the qualitative shape of the meridional profile typically that is predicted when only porosity is considered. z . it does not evolve to any different shape in response to plastic deformation even though it can permissibly vary in size and translate in stress space). and κ . the GeoModel presumes that the shape of the octahedral yield profile is constant in time (i. θ.* Moreover. Figure 4. The function h 2 defines the meridional profile of the yield function. z . there must exist only one radius r that is a solution to f ( r. and therefore this function also defines the size of the octahedral profile. The mechanical behavior of such materials is typically driven by two underlying mechanisms: porosity and microcracks. The GeoModel aims to model rocks and rocklike brittle materials. A degree of ambiguity exists in the definitions of h 1 and h 2 because they may be replaced respectively by ηh 1 and h 2 ⁄ η for any scalar η without loss in generality. z . permitting Eq. the meridional yield profile is a “cap” function that is essentially flat like a Von Mises profile for a large range of pressures ( z is proportional to pressure). To date. the GeoModel’s current assumption of a constant octahedral profile shape might change in future releases. To remove this ambiguity. the yield surface would not be convex).e. (4.24) where g ( θ.e.) (4. the function h 1 is scaled such that it merely describes the shape of the octahedral profile (i..
However. To date. as illustrated in Fig. 33 . the GeoModel obtains the combined meridional yield function by multiplying the individual porosity and microcrack profiles (and scaling the ordinate appropriately to match data). making the GeoModel bettersuited for reproducing observed data. rc ( z ) rf ( z ) microcracks only r(z) combined porosity and microcracks porosity only (a) z (b) z (c) z Figure 4.2. The GeoModel.3. Simple twosurface upper bound model r porous only (without cracks) ly on es) ed t por ck u cra itho (w GeoModel (one continuous surface) r z Figure 4. Loosely speaking. unifies these separate microscale theories to obtain a combined porosity and microcrack model as sketched qualitatively in Fig. but strength increases as pressure is increased because pressure generates additional friction at crack faces. z Distinction between twosurface upperbound models and the GeoModel. The GeoModel [Fig. 4.3(a). the combined effect of voids and microcracks remains a poorly developed branch of materials constitutive modeling.2(c). (b) microcracks alone. and (c) porosity and microcracks in combination. 4. thereby reducing the shear load suffered by the matrix material.GeoModel theory shape of a meridian profile that is typically predicted for theories that consider only the influence of microcracks without considering porosity. Microcracks lead to low strength in tension. this approach results in a discontinuous slope in the meridional yield profile and fails to account for interactions between voids and cracks.3(b)]. 4. phenomenologically permits cracks and voids to interact in a way that results in a continuously differentiable meridional profile. Some early models simply asserted that a material is elastic (safe from yield) only if it is safe from both crack growth and void collapse. Qualitatively meridional profile shapes resulting from (a) porosity alone. with each criterion tested separately.
2(b) times the cap function r c ( z ) in Fig. so the GeoModel yield function is of the form f f ( I 1 )f c ( I 1 ) ξ J 2 = . determine both the cap curvature and the location where the cap intersects the hydrostat (the z axis).28) and fc = Fc . the cap function r c ( z ) implicitly depends on κ and R .GeoModel theory The GeoModel achieves a combined porous+cracked yield surface by multiplying the fracture function r f ( z ) in Fig.27) in terms of stress invariants instead of Lode coordinates. The GeoModel cap function f c is normalized to have a peak value of 1. (3. (4. it can be alternatively interpreted as a stress intensifier. Although the GeoModel does not explicitly track porosity. where α is the ˜ ˜ ˜ ˜ deviatoric tensorvalued backstress that is nonzero only when kinematic hardening is enabled.2(a) so that r ( z ) is proportional to r f ( z )r c ( z ) .27) The proportionality factor depends on the Lode angle θ so that the equivalent shear stress at yield (which.40) to express Eq. which (because it is a divisor in Eq. recall. in addition to depending explicitly on the stress tensor. Γ(θ) Comparing with Eq. As explained later (page 50). 4. R . 4. The function Γ ( θ ) characterizes the Lode angle dependence of the meridional profile and is normalized to equal 1 in triaxial compression ( θ = 30° ) . κ (mentioned earlier) and α . Thus. these functions are typically shaped as shown in Fig. the f and F functions are related by ff = Ff – N (4. ξ Recalling that the Lode cylindrical radius r equals 2J 2 and the Lode axial coordinate z is proportional to I 1 . 34 . (4. (4. ˜ The fracture function f f characterizes the crackingrelated portion of the meridional yield profile.18). it does include an internal state variable κ that equivalently accounts for the presence of porosity. the value of κ and one additional material constant. The curvature of a cap function controls the degree to which porosity affects the shear response. Cap functions depend on the porosity level (which controls where the cap curve intersects the zaxis).4. At different Lode angles. Qualitatively. Rather than regarding Γ as a strength reducer.. Γ usually has values greater than 1.28) reduces equivalent shear strength. 4. Thus. 4. (4. the GeoModel implements the notion of multiplying fracture and cap functions by using Eq. is simply a constant multiple of r ) can be made lower in TXE than in TXC.29) ξ The invariant J 2 is computed using the shifted stress tensor ξ ≡ S – α . the yield criterion depends implicitly on material constants and on two internal state variables.
GeoModel theory
fc ( I 1 ) =
s= diu 1
Fc
ff ( I 1 ) = Ff – N
Γ=1 θ Γ=1/ψ I1 I1 1
ra
rad iu
s= ψ
(a)
(b) (c) κ Figure 4.4. Basic shapes of the three key functions that characterize the composite shape of the yield surface. (a) Lode angle dependence, (b) porosity cap curve, (c) limit failure curve. The Lode angle function Γ ( θ ) is the reciprocal of the radius in the octahedral plane, making it best regarded as a stress intensifier; Γ ( θ ) is normalized to equal 1 in triaxial compression, which implies that it must equal 1 ⁄ ψ in triaxial extension (where ψ is the TXE/TXC strength ratio). Shear influence on void collapse begins at the point where the cap function branches into an ellipse. Since the cap function is multiplied by the fracture curve, this transition point also marks where the composite GeoModel failure surface branches away from f f , beyond which macroscale response is influenced simultaneously by both cracks and voids. For pure (shearfree) hydrostatic compression, void collapse begins at the point where the ellipse intersects the horizontal. Only the function f f has dimensions of stress (the others are dimensionless).
As explained on page 43, the precise expression for the Γ function is determined by userspecification of two parameters: the TXE/TXC strength ratio ψ and an integervalued option (J3TYPE), which controls the manner in which the octahedral profile radius varies from the value 1 ⁄ ψ at TXE to 1 at TXC. As explained below, the porosity (cap) function f c is defined by two parameters: the initial intersection p 0 on the horizontal axis and the eccentricity or “shape factor” R for the ellipse (i.e., the width to height ratio of the ellipse). As explained on page 56, the GeoModel internally computes evolution of the cap function resulting from void collapse. As explained on page 41, the very important f f function, which reflects influence of microcracks, is determined by fitting triaxial compression data to an exponential spline [up to five parameters ( a 1 , a 2 , a 3 , a 4 , N )]. Standard experiments (needed to assign values to these parameters) are discussed in Appendix A. The remainder of this chapter is dedicated to providing further details about the three key functions f c , f f , and Γ used in the GeoModel.
35
GeoModel theory
The cap function, fc. Under compression, the pores in a material can irreversibly collapse, thereby resulting in permanent (plastic) volume changes when the load is removed. Plastic volume changes can occur for porous media even if the matrix material is plastically incompressible. Permanent volume changes can also occur if a material undergoes an irreversible phase transformation. The GeoModel supports plastic volume changes, but it does so without explicitly modeling the underlying microphysical mechanisms. Nonetheless, the GeoModel does reflect the influence of micromechanical theory by phenomenologically incorporating plastic volume changes observed in hydrostatic loading. To motivate the GeoModel’s cap theory, we will explain the equations and their qualitative features in the context of porosity, but keep in mind that any other microphysical compactive mechanisms are equally well accommodated by the phenomenological cap model. The cap function f c accounts for the presence of pores in a material by controlling where the yield function will intersect the I 1 axis in compression. This intersection point corresponds to J 2 = 0 and, because we are considering compressive states, we will denote the value of I 1 at the intersection point by I 1 = X , where X ⁄ 3 is the pressure (positive in compression) at which inelastic deformation commences in purely hydrostatic loading for a given level of porosity. As voids compress out, the value of X will change, as explained later when we discuss the evolution equations for the GeoModel’s internal state variable κ . Porosity also degrades material shear strength because, recalling Eq. (4.28), the cap function effectively reduces the nonporous yield strength, defined previously by the fracture function f f . fc 1
κ
Figure 4.5.
X
I1
GeoModel cap function
The GeoModel employs a cap function* defined 1 if I 1 < κ κ) = I1 – κ 2 1 –  otherwise. X–κ
( I – κ)( I – κ + ( I – κ)) 2( X – κ )2
2 f c ( I 1,
(4.30)
1 1 1 2 * evaluated in the code by a Pelessone [34] function, f c ( I 1, κ ) = 1 –  . 
36
GeoModel theory
Neither κ nor X are usersupplied material parameters. Instead, these variables are computed internally within the GeoModel code by enforcing consistency with more intuitive usersupplied parameters obtained from hydrostatic testing (see page 55). The equation of the elliptical portion of the cap curve is I1 – κ 2 f c2 +  = 1 . X–κ
(4.31)
The intersection point X will be later related to the value of κ so that knowledge of the internal state variable κ will be sufficient to compute a value for X . For now, while describing the geometry of the yield surface, both κ and X should be regarded as internal state variables whose values are computed internally in the GeoModel using evolution equations discussed later. Rather than using f c directly, recall that the GeoModel uses the function F c that is simply the square of f c given in Eq. (4.30):
2 Fc = fc .
(4.32)
Elasticplastic coupling. The cap model is used when the material being studied contains enough porosity (or highly compliant second phase inclusions) so that inelastic volume reduction become possible through irreversible reduction of void space. Intuitively, one might expect the elastic moduli to stiffen as voids collapse, but the material might actually become more elastically compliant as shown in Fig. 8.4 (a phenomenon that might be explained, for example, by rubblization of a ligament network). Regardless of its microphysical origins, the elastic moduli are permitted to vary with plastic strain by generalizing the nonlinear elastic moduli expressions in Eqs. (4.9) and (4.10) to b4 b2 K = f K b o + b 1 exp –  – b 3 Exp –  I p εv 1
1 1 – g 1 exp ( – g 2 J 2 / 2 ) g4 G = f G g o  – g 3 Exp –  . γp 1 – g1 eqiv
(4.33)
(4.34)
In the absence of joints, the scale factors f G and f K equal 1.0; otherwise, they are computed internally within the GeoModel as described in a separate sequel report. In the p above equations, γ eqiv is the equivalent plastic shear strain (which, for proportional loadp ing, is conjugate to the equivalent shear stress, J 2 ), and ε v is the plastic compaction volume change. Mathematically,
p γ eqiv = p εv =
∫
·
·p 2 γ dt ˜
(4.35) (4.36)
∫ tr ε p dt . ˜
Though defined mathematically as stated, the GeoModel computes the plastic volume p change, ε v indirectly, as explained later in the context of Eq. (4.67). 37
Microphysical idealizations such as MohrCoulomb theory predict the meridional profile is a straight line whose slope is directly related to the friction coefficient.e.0(c) on page 1. as well as actual observed material yield and rupture response at low and moderate pressures (i. Ff . with the triangle apex being located on TXC axes. which implies that octahedral yield profiles are generally triangular (or distorted hexagon) in shape. at pressures well below the cap elastic limit) so that observed data primarily reflect microcrack damage rather than combined cracking with pore collapse (covered elsewhere in this report). the GeoModel is equipped with a fourparameter exponential spline that is capable of replicating any of these microphysical idealized theories. they predict that the increase in strength from friction will continue only until ductile yield (at extraordinarily high pressures) becomes more likely. Thus.e. The GeoModel supports this singularly common prediction of microscale damage theory. which implies that the yield function for microcracked media must depend on all three cylindrical Lode coordinates. the onset of “yield” depends on all three stress invariants. the previous subsection described the GeoModel’s cap function f c by considering a material that contained pores. so that the meridional profile expands in an everexpanding conelike shape like the limit surface in Fig.GeoModel theory Meridional shear limiter function.1. we describe the GeoModel’s fracture function f f by considering a material that contains microcracks but no porosity. 1. 1. the meridional profile is expected to monotonically increase with pressure. 38 . In this subsection.. • When microscale theories regard brittle crack fracture to be the only failure mechanism. Given the wide variety of microscale predictions for the meridional profile. Brittle materials are also vulnerable to shear cracking at low pressures. such theories correspond to a meridional yield profile that is conelike at low to moderate pressures but asymptotes to a zero slope (like a Von Mises cylinder) at high pressures. In a loose sense. they predict that the material strength (i. When microscale theories allow for both crack growth and ductility of the matrix material. but no cracks. as sketched in Fig.. microphysical damage theories and laboratory observations for brittle materials also tend to share the following qualitative features: • At a given mean stress. for microcracked media. the Lode radius at failure) will increase monotonically with increasing pressure. This implies that the meridional yield profile will include few if any tensile stress states. in the absence of porosity. • Brittle materials are very weak in tension. but they become able to support increasingly large shear stresses as pressure is increased because friction at crack faces helps reduce the shear load that must be suffered by the matrix material itself. yield in triaxial extension (TXE) occurs at a lower stress than in triaxial compression (TXC). Numerous microphysical analyses (as well as a preponderance of data) suggest that. Though they differ in specific details.
A yield ˜ ˜ function f ( σ.e.. its Lode angle dependence) from the limit surface.37) reveals that the yield surface inherits its octahedral profile shape (i. the limit surface is fixed in stress space..38) with (4. the yield surface has a symmetry axis parallel to [111] that is offset so that it does not pass through the origin.. the stress at rupture is smaller than the peak stress. As first mentioned on page 5. α. The yield surface origin is also offset from the limit surface origin by an amount governed by the kinematic hardening backstress tensor α . which can evolve over time because it ˜ lar internal state depends on timevarying internal state variables. 39 . Specifically. the material will fail catastrophically (i. the limit surface is characterized by a limit function that is similar in form to the yield function. κ ) = J 2 ( ξ ) – . ˜ ˜ ˜ Γ[ θ( ξ ) ] ˜ where ξ ≡ S – α . the limit surface is defined by F ( σ ) = 0 . the GeoModel provides an empirical fitting function F f for the limit surface in the meridional plane.GeoModel theory In the meridional plane (i. reduced perhaps by a cap function F c if the material initially contains voids. Consequently. depending on whether or not the experiment is stresscontrolled or straincontrolled. on the other hand. Many of these achievable stress states can be reached only through inelastic processes. which represents the effect of porosity. Comparing Eq. but depends additionally on internal state variables as follows: { F f [ I 1 ( σ ) ] – N } 2 F c ( I 1 ( σ ).38) ˜ ˜ ˜ When examining experimental data.e. α. depends on the backstress tensor α and on the sca˜ ˜ variable κ . rupture). The peak stress (not the stress at rupture) defines the stresscarrying limit of the material. Stress states that were. The yield function is presumed to share some qualitative features with the shear limit surface. The size of the yield octahedral profile is generally smaller than the limit surface profile because of the multiplier (cap) function F c .37) ˜ ˜ σ)] Γ[θ( ˜ The limit function F ( σ ) depends only on σ .e. Stressstrain curves might or might not exhibit postpeak softening. κ ) ˜ ˜ f ( σ. However. F( (4. at a given value of the Lode angle).. At some point. and the initial meridional yield surface is simply F f – N . the limit surface is the boundary of all stress states that the material is capable of supporting. κ ) . the allowable amount of hardening is not unbounded. When kinematic hardening is enabled. (4. it is generally easier to determine the maximum limit point than the point at which plasticity first begins. at one time. Often.. The limit surface always has a ˜ [111] symmetry axis passing through the actual (not kinematically shifted) origin in stress space. (4. Unlike a yield surface. not on any internal state variables. the yield surface characterizes the transition boundary for inelastic flow. Appendix A (step 4 on page A5) describes in detail how to determine the limit surface from experimental data. Mathematically. where ˜ F f2 [ I 1 ( σ ) ] ˜ σ ) = J 2 ( σ ) – . outside the yield surface might become realizable through hardening evolution of the yield surface.
the initial yield l n itia in surface is permitted to translate upward in the meridional plane by no more than I1 a userspecified limit N . Kinematic hardening. The user defines the limit surface and when the stress parameters ( a 1.7. When the yield surface has reached Figure 4.. e initial meridional profile can harden fac N ur (I 1) it s kinematically and/or isotropically. that contain this combined porous yield function in both cracks and voids.point. material response is handled entirely by the GeoModel.GeoModel theory When advancing the solution. the host code must. a 4 ) for the ultimate shear limit surface. ultimate failure data can be up to the limit surface. At that point. The lim = Ff ff ce. the I) N F f( 1 J2 . The GeoModel evolves cracking and porosity. does not handle material response after reaching the limit surface.6. The initial yield surface is identical except itself lies on the limit surface. The limit surface marks the point at which a continuum material model is inadequate to characterize macroscale material response because softening localization becomes possible. ble. the host code must initiate “element death” or rial response no longer remains possi. loading along Figure 4. The GeoModel more readily mapped out. For real materials. However. Before reaching the limit surface. a hardening allows the initial yield surface to translate until reaching the ultimate failure surface (at which constitutivelevel description of mate. a 2.6. material in its initial (virgin) state. As po on sketched in Fig.e. Yield surface evolution with both microthe I 1 axis). at this time. the host code must intervene by inserting void or by invoking special elements capable of supporting displacement discontinuities.perhaps some other strategy for supporting macroscale softening. a3. equations governing yield surface evorfa su lution are designed to permit only a ield sy rou limited amount of hardening. 4.shifted down by a user specified amount N . it is difficult to identify a single envelope for the combined porous/cracked fracture such at way that the yield surface grows yield function ff . The shifted shear limiter function ce N J2 rf a su F f ( I 1 ) – N defines how the shear stress it lim at yield varies with pressure for a nonporous. When this microcrack yield function is combined with the cap function F c ( I 1. Kinematic rial will begin to soften. but microcracked. κ ) the actual shear stress at yield is further reduced because porosity makes inelasI1 ticity possible even for purely hydrostatic compression (i. upon receiving a “limitarrival” signal. initiate scaledependent softening localization through the use of void 40 . Properly. the mate.
In particular. a 1 – N is the zero pressure intercept ξ of the nonporous meridional yield surface on the J2 axis. Let us now explain why the shear limiter function has the general shape depicted in Fig. Brittle materials fail at very low shear stresses when the pressure is low. When we speak of the “shear” stress at failure. Furthermore. although porosity further lowers and distorts the meridional yield profile by multiplying the shifted shear limit function by F c . a fundamental tenant from plasticity theory is that the yield function must be semiconvex. the shear limit function used in the GeoModel is of the form F f ( I 1 ) = a 1 – a 3 e –a2 I 1 + a 4 I 1 . The shifted function should be regarded as a nominal shape of the yield surface in the meridional plane. (4. as indicated in Fig. The shear limiter function. Then we will list constraints on the model parameters necessary to achieve this shape. The initial (nonporous) meridional yield profile is ff ( I1 ) = Ff ( I1 ) – N initial yield surface (nonporous).8. 4. where I 1 = – I 1 shear limit surface in TXC. N is taken to be zero. the shear limiter function is expected to increase monotonically with pressure. J2 = Ff ( I1 ) J2 = Ff ( I1 ) – N a4 a1 a3 a1 – N a3 a4 shear limit surface I1 initial yield surface I1 Figure 4. the GeoModel will treat the limit surface as if it were a nonhardening yield surface (making its postpeak predictions robust. Shear limiter function (unshifted and shifted).39) where the a k are userspecified material parameters determined from experimental data as explained in Appendix A. The GeoModel supports modeling microcracked material by providing flexible fitting functions that can reproduce octahedral and meridional yield profiles observed for real materials. Therefore. 41 . asymptotes to a linear envelope. To force the material to obey a VonMises type yield response at extremely high pressures.8.GeoModel theory insertion or special elements that support displacement discontinuities. Or. since I 1 is proportional to the pressure. Consequently. Frequently. which implies that the second partial derivative of F f ( I 1 ) must be negative or zero. Eq. the slope coefficient a 4 is merely set to zero. If the host code fails to initiate a softening algorithm.39). but they are able to sustain higher levels of shear stress without failing if loaded under higher confining pressures. F f is expected to increase monotonically with I 1 . (4. 4. (4.8.40) where N is the userspecified shift factor. but undoubtedly inaccurate).
42) (4. κ ) . in its virgin state. I 1 = 0 ).1 on page 96 and Fig. The complete GeoModel yield function. Equation (4.44) (4. (Since backstress α is deviatoric. N . the function F f can be regarded as a “limit” or “softening” envelope. (4. κ ) = F c ( I 1.41) (4. Thus. The second equation recasts the f c function as the square root of a different function F c for computational reasons. For computational reasons. as explained in Fig. The second 42 . are those for the kinematically shifted stress tensor. Finally. zero for all stress states satisfying the yield criterion (on the yield surface). The yield function f must be negative for all elastic states (inside the yield surface). Otherwise. All of these physical considerations lead to the following constraints on allowable values for the parameters: a1 – a3 – N ≥ 0 a2 a3 + a4 ≥ 0 a2 > 0 a3 ≥ 0 a4 ≥ 0 . (4.5 of Appendix A). Sample fits of the GeoModel’s shear limit function to data can be found in Fig. 8. when kinematic hardening is dis˜ ˜ ˜ σ abled. I 1 is the first invariant I 1 of the ξ stress tensor σ . For any cylindrical coordinate system — including the Lode system — the radius must always be nonnegative and therefore F f is defined only over the domain for which it yields positive values. the GeoModel’s yield function is based on the square of Eq.GeoModel theory we are effectively speaking of the value of the Lode radius at failure corresponding to the Lode angle for the stress state. 3.48) The first of these equations allows the user to specify a maximum amount.5 on page 99.43) (4. θ ξ and J 2 . ˜ ˜ The building block functions f f and f c are implemented in the GeoModel in a slightly altered form by being expressed in terms of the shear limit function F f and an alternative (computationally more efficient) cap function F c : ff ( I1 ) = Ff ( I1 ) – N f c ( I 1. the stress invariants. any material should be unfailed at zero stress.28) is the yield criterion.4 and A.3 on page 16 (and in Figs A. that the yield function is permitted to translate under kinematic hardening. beyond which stresses can never be reached quasistatically (not even via hardening). which means that the origin must fall below the meridional yield profile. 8. and positive for all stress states (outside the yield surface) that cannot be reached except through an inelastic process — if at all. unloaded virgin material must be below yield nonnegative slope at low pressures positive Lode radius convexity condition nonnegative slope at high pressures (4. (4. ξ ≡ S – α .45) Specific values for these model parameters are determined from triaxial test data.28): ξ 2 f = Γ 2 ( θ ξ )J 2 – f f2 ( I 1 )f c ( I 1. these are simply the stress invariants.46) ξ When kinematic hardening is used. Of course. κ ) .47) (4.
The internal state variable κ marks the branch point where combined porous/cracked yield surface deviates from the nonporous yield surface. Classical MohrCoulomb theory. (4.50) A hallmark trait of rocks and rocklike materials (concrete. Γ ( θ ) > 1 . therefore sparing the surrounding matrix material from having to carry the entire resolved shear stress at crack tips. ceramics. z (at a given value of κ ) will produce a geometrically accurate visualization of the meridional yield profile. This function controls the shape of the octahedral yield profile. 4. Often we instead plot J 2 vs.47) and (4. The Γ function is normalized to equal unity in TXC ( θ = +30° ). Since this function controls only the shape.. of the octahedral profile. which is supported by the GeoModel primarily for comparisons with ide* With this normalization. the f f meridional function then quantifies the pressurevarying size of octahedral profiles 43 .49) Plotting r TXC vs.* At other Lode angles. as explained on page 23. Γ. 2 (4.48) into (4. Substituting Eqs. As explained on page 56. it acts as a pseudo stress raiser. κ ) r TXC = . etc. its magnitude is inconsequential. not size.) is a higher strength in triaxial compression than in triaxial tension at any given mean pressure. To ensure convexity of the octahedral yield profile.46) gives the yield criterion cited at the beginning of this section [Eq. (4. causing ξ yield to occur at smaller values of J 2 at Lode angles differing from the fiducial (TXC) angle where Γ ( θ ) = 1 . where Γ ( θ ) = 1 . but in doing so we are actually showing a geometrically distorted view of the yield profile. The “J3TYPE” Lodeangle function. the critical Lode radius in triaxial compression (TXC). since Γ ( θ ) ξ appears in the yield function as a multiplier of J 2 . this characteristic results from friction at crack faces being able to carry a larger portion of the load under compression. Thus.GeoModel theory equation also shows explicitly the presence of the internal state variable κ related to the isotropic hardening part of the GeoModel associated with void collapse. I 1 to label the axes with more broadly recognized stress measures. the Lode angle function must satisfy Γ ′′ ( θ ) + Γ ( θ ) ≥ 0 . this branch point is determined internally within the GeoModel in a manner that ensures consistency with measured hydrostatic data. This section describes available functional forms for the Lode angle dependence function Γ ( θ ) . In terms of the new building block functions. may be expressed as a function of the Lode axial coordinate z as ( F f ( 3 z ) – N ) F c ( 3 z. Loosely speaking.18].
The GeoModel presumes that only the size of the octahedral yield profile — not its shape — varies with pressure.* Recognizing logistical constraints of most laboratories. Appendix A gives instructions for inferring a value of ψ from experimental data. by user specification of a parameter ψ . or 3 for the user parameter. or MohrCoulomb) is selected. and almost certainly triaxial compression ( θ=π ⁄ 6 ). The GeoModel presently supports three yieldtype options (specified by a value of 1. 2.Gudehus (an efficient smoothed profile. with restrictions on convexity) 2. the GeoModel presumes that experimental data are available at most only for a limited number of canonical loading paths: perhaps triaxial extension ( θ= – π ⁄ 6 ).7]. The MohrCoulomb option (which is available principally for comparisons with analytical solutions) is differentiable everywhere except at triaxial states where yield surface vertices require special numerical handling.WillamWarnke (a relatively inefficient smooth profile with no convexity constraints) 3. there is evidence that smoothed yield surfaces capture mechanical response more accurately than vertex models. ψ is a constant. the shape of the octahedral yield profile is described. then the normal in that sextant is used. et. has an octahedral profile in the shape of a distorted hexagon. * The GeoModel averages directions on either side of the vertex if the strain rate points within the limiting (Koiter) fan of unit normals. the strength ratio ψ equals its userspecified value at all pressures and throughout the entirety of the simulation (i.. causing considerable computational difficulties when dealing with the vertices.GeoModel theory alized analytical solutions. Consequently. but these authors point to no data to back up this claim. in part. Regardless of which yieldtype (Gudehus.MohrCoulomb (distorted hexagon polygon) The Gudehus and WillamWarnke options both correspond to fully differentiable yield functions (no vertices). J3TYPE): 1. 44 . al.28. [6].3. The computational attractiveness of removing yield surface corners has motivated numerous proposals of smoothed threeinvariant models for frictional materials [29.e. not a time varying internal state variable). According to Borja. WillamWarnke. perhaps simple shear ( θ=0 ). equal to the triaxial extension/compression (TXE/TXC) strength ratio at a given pressure. and Lade [30] was among the first efforts to additionally include curvature in the meridional plane. If the strain rate points within a sextant of the octahedral plane.47.
the strength ratio must satisfy 7 . 2 where α∗ = π + θ . 45 . y = r sin Θ where 1 θ = . Gudehus: Γ ( θ ) = .1 + sin 3θ + . and therefore θ varies between – 30° and +30° . 1+ψ 1 The MohrCoulomb option is convex for . 9 7 2.( 1 – sin 3θ ) .. (4.. The Γ ( θ ) function is defined so that all models return a value of 1 in triaxial compression.8.52) 2 r = . where µ is the coefficient of friction). plotted at allowable values of the strength ratio.9.51) (4.< ψ < 9 . 6 2 3 sin φ sin θ 3. the internal friction 3 – sin φ 3 angle φ is the angle of the failure envelope in the Mohrdiagram ( tan φ = µ . 2 ψ To satisfy the convexity requirement of Eq. Here.50). φ is determined from the usersupplied strength ratio by (1 – ψ) sin φ = 3 . WillamWarnke: 4 ( 1 – ψ 2 )cos 2 α∗ + ( 2ψ – 1 ) 2 Γ ( θ ) = . Graphs of the octahedral yield profile corresponding to any of the above options may be constructed by parametrically plotting x = r cos Θ Here.ArcSin ( sin 3Θ ) ..≤ ψ ≤ 2 . 3 (4. Within the GeoModel. The comparison plot corresponds to a strength ratio of ψ=0. cos θ – . Octahedral yield profiles.GeoModel theory Precise functional forms of available Lode angle functions are given below: 1 1 1.≤ ψ ≤ 2 . 2 ( 1 – ψ 2 ) cos α∗ + ( 2ψ – 1 ) 4 ( 1 – ψ 2 )cos 2 α∗ + 5ψ 2 – 4ψ 1 The WillamWarnke option is convex for . 2 These three options are distinguished by how the octahedral yield profile varies in stress space in the transition from TXE to simple shear to TXC at a fixed pressure. MohrCoulomb: Γ ( θ ) = . . Γ(θ) The angle Θ varies from 0 to 360° . J3TYPE=1 Gudehus J3TYPE=2 WillamWarnke J3TYPE=3 MohrCoulomb COMPARISON Figure 4.
MohrCoulomb theory reduces to Tresca theory and the lowest strength (highest Γ ) occurs at the zero Lode angle (pure shear).2 0. Fig.2 0. Appendix A (STEP 6) provides guidance for selecting the Lode angle option most appropriate for matching experimental data.2 0. involve no yield surface vertices. but such criteria will also imply functional constraints on the meridional failure function f f as well. corresponding to these classical specialcase idealized theories may be found at the end of Appendix B.2 1. However.4 0 TXC π 6 TXE π – 6 SHR 0. In all cases. 4. any classical failure criterion that is expressed directly in terms of the principal stresses will imply an appropriate J3TYPE option (and an appropriate value for the TXE/TXC ratio).2 Γ(θ) – 1 2 Γ(θ) – 1 2 1.4 0 TXC π 6 Figure 4.4 0 TXC π 6 TXE π – 6 SHR 0. The Gudehus and WillamWarnke options both predict lowest strength (largest value of Γ ) at TXE. 4. for parameterizing the GeoModel to quantitative laboratory data.05 1.8 1. resulting in a distorted hexagon if 1 ⁄ 2 < ψ < 1 . a perfect hexagon (Tresca) if ψ=1 . failure is presumed when the equivalent shear stress reaches a critical value. whereas MohrCoulomb theory predicts lowest strength at an intermediate Lode angle somewhere between TXE and TXC. For example. the lower bound on Γ is 1.10 shows the Γ ( θ ) functions for each of the J3TYPE options. then the octahedral yield profile will be a triangle. independent of the Lode angle (like a Von Mises or DruckerPrager criterion).9. Γ(θ) – 1 1. which results in extra computational effort to determine plastic strain rates in triaxial states.1 1. The Gudehus option is the default because of its computational simplicity.25 1. The larger the value of Γ . As a rule. which requires appropriate GeoModel inputs to mimic. Octahedral profile plots like the ones shown in Fig. and a triangle if ψ = 1 ⁄ 2 .2 0. 46 .4 0. which can be modeled with the Gudehus option. then the octahedral yield profile is a circle.4 0. but it supports only a limited range of TXE/TXC strength ratios. on the other hand.6 1.15 1.2 0.10.8 1. simple plots of Γ vs. Simplified GeoModel input sets. 4.4 1. on the other hand. the Lode angle varying from –30 degrees (TXE) to +30 degrees (TXC). the smaller the radial distance to the octahedral yield profile and therefore the smaller the shear failure strength.6 1. θ are more useful. if failure is hypothesized to occur when the largest principal stress (or strain) reaches a critical value. The GeoModel subsumes many simpler (classical) models as special cases.9 are most illuminating from a qualitative perspective.4 0.2 0.0 at TXC.2 TXE π – 6 SHR 0.4 1. If. the WillamWarnke option should be used if a rounded but strongly triangular octahedral yield profile is desired. the MohrCoulomb model (J3TYPE=3) interpolates linearly in octahedral stress space. see Appendix B (page B22). For example. when ψ = 1 . For a detailed explanation of the MohrCoulomb formulation. which speeds up computations. The MohrCoulomb model has yield surface vertices at θ = ± π ⁄ 6 . Lode angle function (for various ψ strength ratios) plotted vs. The Gudehus and WillamWarnke options.GeoModel theory As illustrated in Fig.
most theories of inelastic flow (includ· ing the GeoModel) presume that the total strain rate ε can be partitioned additively as ˜ · = εe + εp . ε p represents the inelastic strain rate. Simplified failure criteria help guide choices for interpolation functions to be fitted to real observed data that likely reflect the specific phenomena considered in microscale idealizations and possibly some other “unknown” sources of inelastic flow. normality and associativity are not interchangeable terms. we have discussed how some microphysically based. but generally oversimplified theories can be used to predict theoretical shapes of yield or failure surfaces. κ ) . Direct use of idealized theories would require initializing and evolving microscale quantities (such as porosity) that are impractical to measure in the laboratory. the plastic part of the strain rate is pre˜ ˜ sumed to be of the form ·p · ∂f ε ij = λ  .53) ˜ ˜ ˜ · · where ε e represents the elastic (or recoverable) part of the strain rate and ε p denotes the ˜ ˜ · “plastic” part of the strain rate. Many classical theories presume that the direction of the plastic strain rate is parallel to the normal to the yield surface. ˜ which reflects contributions from any and all sources of inelastic material response. In this case. ∂σ ij α. Microphysical theories are also used to guide how the GeoModel treats the partitioning of inelastic flow. but recast in terms of directly measurable macroscopic variables. The GeoModel implicitly captures microscale phenomena by using macroscale measurable variables in phenomenological manner.GeoModel theory Advancing the solution (groundwork discussion) So far. κ ˜ (4. as shown here. since the normal to the yield surface can be obtained by the gradient of the yield function f ( σ. the terms “associativity” and “normality” can have distinct meanings.54) · where λ is a multiplier (called the consistency parameter) determined by demanding that the stress must remain on the yield surface during inelastic loading. When the coupling terms (from rates of elastic moduli) are incorporated into the inelastic strain rate. For example. · · ε (4. * For materials that exhibit elasticplastic coupling. then associativity and normality are interchangeable. but at the cost that the elastic strain rate ceases to be an exact differential with respect to deformation. the algebraic structure of functions used in the GeoModel is guided by idealized microscale theories. More correctly. If the coupling terms are incorporated into the elastic strain rate. depending on whether the portion of the total strain rate attributable to rates of elastic moduli is absorbed into the elastic strain rate or the inelastic strain rate. The subscripts on the partial derivative merely indicate that the internal state variables are held constant. Therefore. once it begins. 47 . the model is said to be “associative” (to indicate that the plastic strain rate is associated with the yield function*). α. When the plastic strain rate direction is determined from the stress gradient of the yield function.
The stress rate is determined by applying elasticity.57) * While considerable data does exist to suggest that the inelastic strain rate is not directed normal to the yield surface for some materials. If the flow potential depends on the first invariant I 1 . In the GeoModel. The need for such revisions is rarely recognized in plasticity programs (and casts doubt on the very notion flow potentials).GeoModel theory While the GeoModel does support associativity at user request. That is. a 4 . For nonnormality the user specifies a flow function φ ( σ. κ . 48 . as mentioned in Eq.53). † Because the current stress might not reside on the isosurface φ = 0 . · ·e · ·p σ ij = C ijkl ε ij = C ijkl ( ε ij – ε ij ) .55) ∂σ ij α.37]. ∂I 1 ˜ (4. ·p The plastic strain rate ε ij includes both deviatoric and isotropic parts.. the strain rate is decomposed into two parts. then the plastic potential function is identical to the yield function and the plastic strain rate will therefore point normal to the yield surface. For associativity. such behavior is not well understood.56) ˜ The plastic strain rate points normal to the isosurface φ = 0 . who have demonstrated that such a model is inherently unstable. the volumetric plastic strain rate is ·p · ∂φ · ε v = trε p = 3λ . the flow potential φ can be made to differ from PF PF the yield function by assigning values to a 2 . the functional form of φ is the same as that of f . a 4. ψ ) used to define the yield surface. The mathematical validity of assuming existence of a nonassociated flow potential function has been called into question by Sandler and Pucik [40. R PF . nonnormality is supported in the GeoModel as well. and ψ PF that differ from their counterpart parameters ( a 2. Therefore. but with different values for material constants. many researchers report that normality tends to overpredict the amount of volumetric plastic strain [43]. R.† If the flow function is associative. the potential function parameters should be given values identical to their counterparts in the yield function. At this point. α. inadmissibly generating unbounded energy from quiescent states. the governing equations are no longer elastic. κ ) such that* ˜ ˜ ·p · ∂φ ε ij = λ  (4. (4. in which case the plastic strain rate points within a “cone of limiting normals” (Koiter fan) at the vertex and is determined through additionally considering the trial elastic stress rate associated with the total strain rate. If continuing to apply elasticity theory would result in a predicted stress lying outside the yield surface. (4.21). the GeoModel projects the stress to the nearest point on this isosurface. Specifically. (3. then applying Eq. Flow surface vertices reside at points where the flow potential is nondifferentiable. elastic plus plastic.
GeoModel theory where C ijkl denotes the isotropic tangent elastic stiffness tensor.59) The first term may be simplified through application of Eq.59) we will be able to solve Eq. (4.55) into (4. It will be argued that the evolution of each internal state variable should be proportional to the plastic strain rate.59) for the consistency · parameter λ . κ and (if applicable) on the kinematic hardening backstress state variable · tensor α ij .57).20).55) allows us to compute the direction of the plastic strain rate from the known instantaneous stress state. (4.· ∂f · · f = . κ ˜ (4. after substitution of Eq.57). these rates must be proportional to our unknown plastic consistency parameter λ .α ij = 0 . the strain rate is known (it is provided by the host code after solution of the momentum equation). not only must the stress be on the yield surface during plastic loading ( f = 0 ) . 49 . In Eq. The yield function f depends on the stress. the key to advancing the solution is to now derive in detail the internal state variable evolution equations. The last two terms reflect the fact that the yield surface can evolve in shape and translate in stress space during inelastic loading. Thus. Equiva· lently. The consistency parameter is obtained by demanding that. the only unknown in this equation is the plastic strain rate. In numerical implementations of constitutive models.· ∂f . (4. it must also remain on the yield sur· face throughout a plastic loading interval. ∂σ ij α. we will present “evolution equations” that govern how the state variables change in response to plastic flow. By substituting Eq.κ + .58) Everything on the righthandside of this equation is known except the value of the consis· tency parameter. allowing the ISVs themselves to be updated to the end of the timestep. Thus. it can be substituted into Eq. Equation (4. λ . the stress rate can be written · ·e · · ∂φ σ ij = C ijkl ε ij = C ijkl ε ij – λ  . Thus. and therefore the instantaneous stress state and elastic moduli are known. then rates of internal state variables (ISVs) become known through their evolution equations. Once the consistency parameter is known. (3. The current state is known. ∂κ ∂α ij ∂σ ij (4. but it also depends on the isotropic hardening internal state variable. With the consistency parameter λ known. the assertion that f = 0 can be written via the chain rule as ∂f. (4. Thus. Thus.58) and these soontobederived evolution equations for the internal state variables into Eq.σ ij + .58) to obtain the stress rate. f = 0 during plastic loading. In what follows. which may then be integrated numerically to update the · stress. (4. our goal is to compute the stress rate. (4.
Hardening mechanisms. H α . both types of hardening can occur simultaneously. isotropic hardening entails an increase in size. while kinematic hardening (governed by the backstress tensor α ) produces a translation of all octahedral yield pro˜ files. we will derive explicit expressions for an isotropic hardening modulus h κ and a kinematic hardening tensor. substituting these expressions into Eq. 50 . In general. such that the evolution of the internal state vari˜ ables may be written in the forms · · κ = hκ λ (4. Once the plastic consistency parameter is known. the above equations can be themselves integrated through time to update κ and α . ˜ Evolution equation for the porosityrelated internal state variable. Once the consistency parameter is known. We begin this section with some background discussion about the meaning of the internal state variable κ .11. connecting it to some classical microphysical theories for purely porous (noncracked) materials. As indicated in Fig. κ . The GeoModel’s reinterpretation of κ for both porous and · · cracked materials will lead ultimately to an evolution law of the desired form. Initial yield surface isotropic hardening kinematic hardening compound hardening Figure 4. kinematic hardening translates the yield surface. (4. 4.11.60) and · · α = H λ. the evolution laws may be integrated through time to model the time varying hardening evolution of the yield surface. In this section.59) may be used to determine the plastic consistency parameter λ if hardening evolution laws can be found for which the rate of each internal state variable ( κ and · α ij ) is proportional to λ . (4. κ = h κ λ where h κ is called the isotropic hardening modulus. isotropic hardening (governed by κ and related to void collapse) causes a change in size of octahedral yield profiles.GeoModel theory Evolution equations · Equation (4. At a given pressure. Softening corresponds to a yield surface contraction. and compound hardening includes both mechanisms.59) will lead to an expression for the plas· tic consistency parameter λ .61) ˜ ˜α Later.
the smaller X will be. If a material is capable of permanent volume change (i. void collapse commences at different pressures depending on the amount of shear stress present. dX dε p ∂ I v 1 (4. where dκ dX ∂φ h k = 3  . Then we will discuss the relationship between κ and X . then the material likely contains voids. during hydrostatic compression or compaction dominated processes. 4. 4. 4. a different form is used for h k (see Eq..GeoModel theory In the GeoModel. Even though void collapse depends on shear stress. (4. With these two relationships in hand. Therefore.5]. we will focus on how the hydrostat intercept X should vary as porosity is reduced.56). The hydrostat intercept X is proportional to the critical “elastic limit” pressure required to initiate irreversible void collapse. will give the evolution equation in the desired form. The effect of shear stress on void collapse is characterized by the cap function illustrate in Fig.e. · · κ = h κ λ . Recognizing that ε v is an indirect measure of porosity changes. ε v .73). dX dε p v (4. 51 . As porosity is crushed out. we will ultimately assert that dκ dX · p · κ =  . the larger the porosity. characterizing this effect requires only specification of two numbers κ and X on the hydrostat [see Fig.62) from which substitution of Eq. .5. if hydrostatic testing exhibp its nonzero residual plastic volumetric strain ε v upon releasing the pressure). the hydrostat intercept will move to the right so that increasing pressure will be required to continue crushing out the p pores. In the GeoModel. κ and X are presumed to be interrelated so that knowledge of X is sufficient to compute the value of κ ..63) During dilatationdominated processes. We will discuss this relationship later. For now. our first goal is to p describe how the relationship between ε v and X can be inferred from hydrostatic test data.
65) Considering only hydrostatic loading. The GeoModel uses a similar curve. To second order accuracy. Gurson [20] reported the following upperbound yield criterion (expressed in terms of Lode cylindrical coordinates): z2 r = k f v + 1 – 2f v cosh  . Being independent of the Lode angle.13 for various porosities.64) implies that the change in porosity is approximately equal to the plastic volumetric strain: p Π – Πo ≈ εv .12. When microcracks are later included. the GeoPE Model incorporates the results from this specialized Figure 4. in the absence of microcracks. 4. the innate porosity at the rest state. 2k (4. loosely speaking. Recall that the cap Thereafter. (4. porosity is reduced as presfunction. By using the cap function. then it can be shown [33] that the unloaded porosity Π (i. the yield surface becomes a cylinder in stress space as the porosity goes to zero). (4. As porosity goes to zero. not the slightly different porosity that reflects reversible elastic porosity reduction under loading) evolves under plastic loading according to · ·p Π = ( 1 – Π )ε v . E pressures in the presence of shear. the meridional profile approaches the pressure insensitive VonMises profile for the matrix material (in other words. represents material sure is increased. response in the absence of microcracks. and ε v is the trace of the logarithmic plastic strain rate. early Π research on pore collapse focused on deriving and/or experimentally measuring socalled “crush curves” in which porosity in a material is plotted as a function of the applied pressure.. the common envelope of yield surfaces will be the shear fracture curve F f .GeoModel theory p Relationship between X and ε v (the hydrostatic crush curve) If the matrix material for a porous medium is plastically incompressible. Gurson’s meridional profile is compared with the GeoModel’s cap function in Fig. Porosity is constant until a crita manner that pore collapse will commence at lower ical elastic limit pressure P is reached. as in Fig.12. This property holds only in the absence of microcracks. Eq. but inferred directly from hydrostatic stressstrain data so that porosity measurements p are not necessary. Porousonly theories typically predict meridional cap profiles similar to the GeoModel cap function. A conventional porosity hydrostatic experiment into the general theory in such crush curve (dashed) and state path (solid). 4. the Gurson yield function is a circle in the octahedral plane.66) where k is a constant (the yield stress of the matrix material) and f v is the porosity. 52 .e. (4. For example.64) where Π is the void volume in a sample divided by the total volume of the sample (both ·p volumes are those in the unloaded state).
the GeoModel is guided by the general trends they predict.0(b) on page 1. the theories are similar. microphysical theories are typically upper bounds. When microcrack effects are included. The loss in shear strength caused by pores is pronounced over only a small range of pressures near the hydrostatic limit pressure. after which the cap function drops to zero along an ellipse. these theories must resort to over simplistic assumptions about the matrix material and pore morphology (e. Qualitatively. Rather than using essentially unknowable matrix properties like these as internal state variables. the Gurson model predicts that the cap surface will be essentially flat for a large range of pressures (zcoordinates). The cap curve evolves by simple translation along the hydrostat without changes in cap curvature.13. The Gurson theory for porous yield surfaces compared with the GeoModel cap function at various values of the internal state variable κ.6 porosity 10 20 . 1. Gurson theory presumes perfectly spherical voids arranged in a perfectly periodic array). Except at extremely high porosity. In the Gurson model.4 . the current location of the translated cap is a function of the matrix yield stress and the porosity.2 z z Figure 4. which are of limited use in applications since the tightness of the bound is unknown.g. which are difficult to measure directly.. the key material properties are microphysical (the yield stress k for the matrix material and the porosity f v ). The GeoModel supports these general trends by using a computationally simpler RubinSandler cap function which is simply constant until a critical branch pressure is reached. To obtain analytical results.GeoModel theory r 1 .8 Gurson model low porosity r GeoModel with microcrack effects suppressed h igh . As seen in Figure 4. Rather than directly using models like Gurson theory.13. the GeoModel recognizes that the appropriate location for the cap can be determined directly from hydrostatic compression test 53 . Under Gurson theory. instead of the horizontal VonMiseslike envelope shown here. as in Fig. the GeoModel profiles at various porosities form a pressuredependent envelope. This region (beyond which the yield surface noticeably branches down to zero) is called the cap region. Gurson theory predicts the yield surface will evolve with porosity in such a manner that the cap essentially translates along the pressure (z) axis — the curvature of the cap region does not change significantly. Finally.
pressure crush curve. the pressure vs. This function asymptotes to a limit value for the plastic strain when all voids have crushed out (and plastic volume changes therefore become negligible). (b) the GeoModel Xfunction. the elastic strain at each pressure value may be subtracted from the total strain to generate the Xfunction. the experiment should be run to the point of total pore collapse (as in Fig. 4.e. pressure crush curve. total strain data. Thus. where the cap intersects the zaxis) evolves.1 on page 13). total strain plot is converted to a pressure vs. The limit strain is approximately the initial porosity in the material. “Copies” of the elastic unloading function may be superimposed anywhere on the hydrostatic pressure vs. As indicated in Fig.14 to obtain a classical porosity vs.. After applying these shifts. plastic strain plot. characterizing how the entire cap function evolves in response to plastic loading boils down to characterizing how the hydrostat intercept point (i. called an Xfunction.14 asymptotes to infinity when the plastic volume strain (i. Specialized parame54 . 3. The Xfunction in Fig. The GeoModel never explicitly refers to porosity. volumetric strain data as illustrated in Fig. Hydrostatic test data elastic unloading copies of the curve p shift distance shift distance shift distance shift distance GeoModel Xfunction p = 1X 3 Classical crush curve p porosity ≈ p 3 – ε v initial slope s o = – 3p 3 p 1 K(p) PE total εv PE p3 p3 p εv PE p Figure 4. the plastic volumetric strain is employed as an indirect measure of porosity changes. 4. Once the elastic unloading curves have been parameterized to the GeoModel fitting functions.14. If possible. The elastic response of the material must be first determined by fitting the unloading curve to the nonlinear elasticity fitting function in Eq. 4. volumetric strain data.. and (c) a traditional porosity vs. volumetric strain data.GeoModel theory data and the branch point at which shear begins to affect pore collapse is presumed in the GeoModel to translate with the hydrostatic limit point. Test data are pressure vs. (4. Parameterizing the GeoModel so that it will adequately model the changes in the yield surface resulting from pore collapse requires converting hydrostatic pressure vs. pressure crush curve in which the porosity is plotted as a function of the pressure p . Rotating the Xcurve and shifting the origin produces a classical crush curve in which porosity is plotted as a function of pressure. Relationship between (a) hydrostatic pressure vs. The GeoModel presumes that it is experimentally tractable to obtain pressure vs. the change in porosity) has reached its maximum value corresponding to all of the pores having been crushed out. total volumetric strain. Rotating the Xplot 90° and moving the origin as shown will produce a traditional porosity vs. Instead.14.9). the elastic unloading curves can be used to determine a shift distance that must be applied at any given pressure to remove the elastic part of the strain.e.
Eq.67 describes the dashed line in Fig. As pressure is increased from zero. pressure crush curve (see Fig. Complete crushing out of all pores in the material is recognized in the hydrostatic pressure vs. 4. the crush curve [Fig. total strain curve if the elastic release curve is tangent to the loading curve. which corresponds approximately to the initial porosity in the material. • p 1 equals s o ⁄ ( 3p 3 ) . • p 3 is the maximum achievable plastic volume strain.14.GeoModel theory terization software is available from the model developers to perform this conversion task and to fit the resulting crush curve to an exponential spline (see Appendix A). where P E is the elastic limit pressure at the initial onset of pore collapse.14). During the initial elastic loading. At pressures above the elastic limit ( ξ > 0 ). 4.67) coincides with hydrostatic test data only during pore collapse.14(c)] shows that the porosity remains unchanged for a while until an elastic limit pressure P E is reached.e. these parameters are interpreted physically as follows: • p 0 equals – 3P E . Eq.68) p This relationship may be differentiated to obtain the derivative dX ⁄ dε v needed in Eq. transitioning to concave up at high pressures).67 may be used to compute the plastic volumetric strain according to p ε v = p 3 [ 1 – e –( p 1 + p2 ξ )ξ ] . where ξ = X – p0 = 3 ( p – PE ) . The plastic volume strain of a virgin (predeformation) material is zero. (4. Therefore the userspecified parameter p 3 is approximately equal to the initial porosity in the material. pressure crush curve. p p 3 – ε v = p 3 e –( p1 + p2 ξ )ξ . which is not required to coincide with the data. 55 .. Now all we need to compute the isotropic hardening modulus h κ is the relationship between κ and X so that we can substitute the derivative dκ ⁄ dX into Eq. p 0 .63). where ξ = X – p0 = 3 ( p – PE ) (4. (4. initially concave down. 4. Referring to the porosity vs. p 2 . and p 3 are fitting constants. Continuing to apply increasing pressure beyond this elastic limit results in irreversible pore collapse and therefore reduction in porosity.67) Here. (4. 4. • p 2 is an optional fitting parameter that may be used if a measured crush curve has an inflection point (i. 4. Equation (4. The GeoModel allows fitting the post yielding part of the crush curve according to the crush curve spline formula. p 1 .63). where s o is the initial slope of the porosity vs.
material response begins to be influenced by porosity. Between the peak and the hydrostatic limit point X . whereas A ⁄ B is usually smaller than 1. the fracture function f f marks the onset of shear crack growth. (cap curvature model) ff te mi r li a ) (I 1 . Fig. The bold blue curve is F f f c . f f= ope el env ld” “yie I1 fc 1 cap (porosity) function BRANCH POINT I1 κ X ) (I 1 f . I 1 = κ ) where the yield function f begins to deviate from the envelope function f f .15. 4. The dashed yield curve is f f f c .15 is considerably lower than the peak height B . The cap porosity function f c intersects the hydrostat ( I 1 axis) at X .13 suggests that κ is simply smaller than X by a fixed amount. 4. equivalently. Variation of the cap function with shear stress (along the ellipse) merely reflects an expectation that pore collapse will commence at a lower pressure than the hydrostatic limit when shear assisted.GeoModel theory Relationship between κ and X For a purely porous material. porosity dominates the material response. The height of the branch point b in Fig. F t mi r li ea yield sh profile J2 b BRANCH POINT B κ R = a b A a I1 X Figure 4. the relationship between the branch point κ and X is similar. Isotropic hardening in the GeoModel is cast in terms of the branch point located at I 1 = κ (or. Continuously differentiable GeoModel yield function and some characteristic dimensions. marking the point at which pressure under hydrostatic loading would be sufficient to induce pore collapse. with significant pressure strengthening being the result of friction at crack faces. Qualitatively. For a material that contains both pores and microcracks. Ff pe N elo nv (I 1) e sh Ff . and therefore plastic volume increases because of crack bulking. Therefore. The ratio a ⁄ b is usually larger than 1. Recall that the continuously differentiable meridional yield function f is constructed by multiplying a function f f times a cap porosity function f c . but is still shear crack (dilatation) dominated. 56 . but influenced by the pressure sensitivity of the fracture function. knowing X is sufficient to determine κ . Between the branch and the peak. This yield envelope function f f might be lower than the ultimate shear limit envelope F f if kinematic hardening is allowed.
GeoModel theory resulting in plastic volume compaction (from pore collapse). . however. .g. 4. Twosurface models [e. At the critical zeroslope point on the yield surface. solving for X .15 indicate that a = X – κ and b = F f ( κ ) . This distinction between the two ratios is important to emphasize in publications and presentations to avoid confusion between the GeoModel and conventional twosurface models.63) gives · · κ = hκ λ . constructs and evolves its yield function based on the ratio R ≡ a ⁄ b .15) is typically smaller than unity.71) into (4. X = κ + RF f ( κ ) .13). this equation becomes X = κ – RF f ( κ ) . Therefore the cap eccentricity (also called the cap shape parameter) R = a ⁄ b is given by R = ( X – κ ) ⁄ F f ( κ ) or. Ref. 41] typically construct and evolve the yield function by making direct reference to the ratio A ⁄ B ..72) where the “isotropic hardening parameter” h κ is compactiondominated hardening modulus dilatationdominated hardening modulus h κ = min p p I1 – κ dX ⁄ dε v ∂φ dX ⁄ dε v 3 .69) This expression is evaluated internally within the GeoModel coding to determine X as a function of the internal state variable κ . . The geomodel. which (referring to Fig.15 remains always equal to a userspecified constant R . which is typically larger than unity.70) (4. 4. Differentiating both sides of this equation with respect to κ gives dX = 1 – RF ′ ( κ ) . 4.71) The κ evolution law Substituting Eq. the GeoModel presumes that hardening proceeds such that the ratio between the distances a and b labeled in Fig. Fig. (4. material response is influenced equally by both cracks and pores so that so that no net volume change is apparent at the macroscale.. The axis labels in Fig. R 1 – RF f ′ ( κ ) ∂I 1 1 – RF f ′ ( κ ) (4. Guided both by trends in observed data and by microphysical theories (e. f dκ (4.g. (4. (4.73) 57 . . When written without the overbar denoting the negative. 4.
while against – z (which is proportional to the pres“p”) the second term dominates in the dilatation sureonset of The yield surface (solid) demarks the inelastic flow. Eq. The GeoModel supports kinematic hardening.68). this constraint ensures that κ will never decrease. (4. (4. The derivative p dX ⁄ dε v is found from crushcurve data fitted to Eq. ELASTIC DOMAIN Evolution equation for the kinematic hardening backstress tensor. The second term in the minimum elastic loading.75) ˜ ˜ ˜ The backstress tensor α is initialized to zero. softening must be handled at the field scale by the host code. the hardening modulus h κ in Eq. After the stress passes through the critical point. Thus. Recall that the yield criterion with kinematic hardening is given by ξ J2 = [ Ff ( I1 ) – N ] 2 Fc ( I1 ) .73) dominates when the stress state Meridional plane in falls on the “compaction dominated” part of Figure 4. Even under monotonic loading. Kinematic hardening entails using a shifted stress tensor ξ = S – α in the yield function instead of the actual ˜ ˜ ˜ stress.GeoModel theory subject to the constraint. 58 . labeled in Fig. the relative position of the stress state on the yield surface can move from compaction to dilatation regimes. the backstress ˜ evolves in proportion to the deviatoric part of the plastic strain rate: · · * When using the overbar to denote the negative. which therefore results in plastic volume reduction and an isotropic expansion of the yield surface.74) ξ = S – α.01b 0 . Figure 8. Compaction from void collapse and dilatation from crack bulking are relatively balanced in the vicinity of the critical point. softening) cannot be accommodated at the material constitutive level — to avoid mesh dependencies of the solution. pands and/or translates) toward the shear limit surface (dashed). since γ is never negative. Physically. Increases in porosity (i. r e rfac it su im ar l she tion com dilata pa cti on The first term in the minimum function of z Eq. Upon onset of yielding.6(a) (page 100) shows a triaxial compression load path (angled red arrow) that falls initially on the porosity (compaction) dominated portion of the yield surface. 0 > h κ > – ∞ . the yield surface hardens (exfunction is guided by trends in observed data. but is otherwise isotropic.16.16. (4. (4. 4. The deviatoric tensor internal state variable α is called the backstress. (4.* For numerical convenience.72) may be written κ = h k γ subject to · 0 < h k < ∞ . where b 0 is the initial bulk modulus. this constraint is 2 replaced by 0 > h κ > – 0. this is equivalent to demanding that porosity must always decrease.73) transitions from its compactiondominated value to the dilatationdominated value.e.. the stress deviator “r” which the magnitude of is plotted the yield surface. (4. ξ where J 2 is the second invariant of the shifted stress. Under continued inregime. and it is ˜ computed using evolution equations described here.
The meridional plane perpendicular to the back stress shows no translation at all. ˜ ˜ ∂ ξ ˜ (4. the GeoModel uses the following decay function: α J2 G α ( α ) = 1 – .76) (4. Consequently.61). ˜ ˜ ∂ ξ ˜ ∂φ H α = HG α ( α )dev . Since the yield function itself is defined in terms of F f ( I 1 ) – N . the kinematic hardening modulus tensor is given by (4.GeoModel theory · · α = HG α ( α ) γ p . The effect of kinematic hardening on the meridional plane that contains the backstress.11).17). (4. 4. The α translation distance equals J 2 and Eq.. J2 = Ff ( I1 ) – N J2 = Ff ( I1 ) I1 α J2 I1 Figure 4. 59 . 2 ˜ (4.tr α 2 . The GeoModel uses the G α function to “slow down” the rate of hardening as the limit surface is approached so that H α will equal zero upon reaching the limit ˜ surface. Specifically. on the meridional plane perpendicular to the backstress. .79) prevents this distance from ever exceeding the userspecified offset limit N . comparing with Eq. ˜ F f ( I 1 ) .17. 4. According to the hardening rule. the yield surface will appear to have translated upward in the meridional plane that contains the backstress (see Fig. the maximum kinematic translation that can occur before reaching the limit surface equals the model offset parameter N .78) where H is a material constant and G α ( α ) is a scalarvalued decay function designed to ˜ limit the kinematic hardening such that G α → 0 as α approaches the shear limit surface. ˜ ˜ ˜ · ∂φ · · where γ p = devε p = dev λ .77) Hence. the meridional profile will not appear to have translated. ˜ N where 1 α J 2 = . (4. .79) Kinematic hardening causes the octahedral profile to translate so that it no longer remains centered at the origin (See Fig. this meridional profile is permitted to translate in deviatoric stress space by as much as a shift factor N . Of course.
.α · ∂f.80) (4. ∂σ ij ∂κ ∂σ ij ∂α ij (4.81a) (4. (4.H ij λ + . consistency parameter) · Recall from Eq.81b) (4.85) implies that the stress may E be alternatively integrated through time by first computing a trial elastic stress σ ij at the end of the timestep which may be projected back to the yield surface (which itself has been updated to the end of the step) to determine the final stress. ∂α ij ∂σ ij ∂σ ij ∂κ From which it follows that 1 ∂f · · λ = .80) becomes ∂f ∂f .κ = 0 ..85) In the GeoModel. (4. the consistency condition in Eq.· · · ∂φ .83). (4.81d) (4. the stress state is updated through direct integration of the GeoModel plasticity equations.C ijkl ε ij – λ  + .C ijkl ε kl χ ∂σ ij where ∂f ∂φ.GeoModel theory Advancing the solution (final step. applied during plastic loading intervals.σ ij + .α ij + . ∂σ ij ∂α ij ∂κ Recall the key equations governing the rates of the field and internal state variables: · ·e · ·p σ ij = C ijkl ε ij = C ijkl ε kl – C ijkl ε kl · α· α ij = H ij λ · · κ = hκ λ ·p · ∂φ ε ij = λ . If the trial elastic stress 60 .∂f . (4.84) Formal equivalence with oblique return algorithms.59) that the consistency parameter λ is determined from the con· sistency condition. ∂f ..h κ .α ∂fχ = .87) (4.– . ∂σ ij With these. Specifically. χ where ·E · σ ij = C ijkl ε kl ∂φ ∂f A ij = C ijrs .· ∂f.83) (4. f = 0 .H ij – .82) .h κ λ = 0 . With the plastic parameter determined from Eq.C ijkl .81c) (4.· ∂f · · f = .86) (4. (4.. it is important to understand that the update formula in Eq. for our upcoming discussion of rate dependence. ∂σ rs ∂σ mn (trial elastic stress rate) (4.and B kl = . the stress rate may be written ·E · · E A ij B kl σ kl σ ij = σ ij – . However.
P ij Q ij χ ∂φ ∂f where P ij = C ijrs .. after evolving the internal state variables appropriately to update the yield surface to the end of the step. a nonmoving discontinuity in displacement or velocity) is another form of material instability that has been extensively studied in the literature. The last term in Eq.C ijkl .85) and (4.e. · · σ ij = T ijkl ε kl . comparing this equation with Eqs. marking the onset of softening (yield surface contraction).90) 61 . the GeoModel’s tangent stiffness is given by 1 T ijkl = C ijkl – .* Stability issues.C mnkl .88) Therefore. plastic flow must have occurred during at least part of the solution interval. In other words. Quasistatic inelastic tangent stiffness tensor. (4. the tangent stiffness tensor might eventually have a zero eigenvalue. a change from a positive to negative slope in a stressstrain plot) depends on the loading direction. equivalence of associativity and normality holds only in the absence of elasticplastic coupling. For example.e.87). then no stressstrain softening will be observed and no change in type of the momentum equation will occur even if the yield surface is contracting. a jk = n i T ijkl n l * Actually. (4. ˜ where x denotes the eigenvalues of the secondorder acoustic tensor. it can be shown that Eq. For any constitutive model.85) implies that the stress at the end of the step may be found by obliquely projecting the trial elastic stress back onto the updated yield surface. so inelastic flow can potentially make the inelastic tangent stiffness tensor T ijkl noninvertible. When elastic moduli can change in response to plastic loading. an associative ( f = φ ) model will not exhibit normality and will not have a majorsymmetric tangent stiffness [9]. As explained in Chapter 5.89) is subtracted from the positivedefinite elastic stiffness C ijkl .∂f ∂φ.89) The tangent stiffness is major symmetric ( T ijkl = T klij ) only for associative models ( f = φ ) .. Because A is not generally proportional to B . The acoustic wave speeds (i. the trial stress is projected only partly back to the yield surface whenever rate sensitivity is applied.. Whether or not the occurrence of a zero tangent stiffness results in stressstrain softening (i.e. the speed at which inelastic perturbations can propagate through a material in the direction of a given unit vector n ) are given by c = x ⁄ ρ . (4. the inelastic tangent stiffness T ijkl is a fourthorder tensor formally equal to the derivative of the stress rate with respect to the total strain rate.h κ . ∂σ rs ∂σ mn quasistatic tangent modulus and ∂φ. A standing wave (i.– . (4.HG α . if the strain rate is orthogonal to the null space of a noninvertible tangent stiffness tensor. That is.∂f∂f χ = .– .GeoModel theory falls outside the yield surface. Qmn = .. (4. Therefore. the ˜ projection is oblique to the yield surface even if˜plastic normality is used. ∂σ ij ∂α ij ∂σ ij ∂κ ∂σ ij (4.
the wave is a shear ˜ wave. 62 . Physically. substituting Eq. not only are standing waves ( x = 0 ) possible.91) (4.91) is provided in Ref.C mnkl n l .92) and ∂φ p j = n i C ijrs ∂σ rs and ∂f q k = . the dynamic tangent stiffness tensor is generally stiffer than the quasistatic tangent stiffness. as explained in the next section. then the acoustic tensor will not be symmetric. (4. then the ˜ wave is a compression wave.93) If the plastic tangent stiffness T ijkl is not major symmetric. for the GeoModel.89). other modes are possible. but for inelastic tangent tensors of the form (4. the question of material stability must be examined anew because. In this case. the eigenvector characterizes the velocity jump direction. [9] where every possible ordering of the inelastic wave speeds relative to elastic wave speeds is derived and where every possible acoustic eigenvector is presented. A complete spectral analysis of acoustic tensors of the form in Eq. If the eigenvector is perpendicular to n . 1 e a jk = a jk – . For elastic materials. (4.p j q k . If the eigenvector is parallel to the wave propagation direction n . these are the only two possible kinds of waves. ∂σ mn (4. but so are imaginary wave speeds (flutter instability).90).GeoModel theory Thus. χ where e a jk = n i C ijkl n l = the elastic acoustic tensor (4. For rate dependent materials.89) into (4.
illustrated qualitatively in Fig. At high strain rates.1. If. In the limit of extraordinarily high load rates (as near the source of an explosion). material response is essentially elastic because insufficient time exists for plasticity to fully develop. cracks grow at a finite speed — they cannot change instantaneously from one size to another. If the loads are applied over a time interval that is significantly smaller than the characteristic response time. Viscoplasticity model overview The evolution of the yield function. the pressurevolume part of the elasticity) plays the predominant role in material response. Consider a loading increment ∆t during which the strain increment is prescribed to be ∆ ε . Rate Dependence The governing equations discussed so far are rate independent. so they only apply for quasistatic loading. As explained below and illustrated in Fig. 5. then essentially no inelasticity will occur during that interval.. If the applied strain is released any time during this damage accumulation period. Two limiting solutions for the updated stress can be ˜ readily computed: (1) the lowrate (quasistatic) solution σ L which is found by solving the ˜ rateindependent GeoModel equations described previously. the loads are applied slowly (as in quasistatic testing). void collapse takes finite time. the elastic response of a material occurs almost instantaneously. Materials have inherent “viscosity” or “internal resistance” that retards the rate at which damage can accumulate. Under high strain rates. the solution of which 63 . Likewise. the DuvautLions rate formulation is based on a viscoplastic differential equation. on the other hand. Simple inertia also contributes to rate dependence. then the quasistatic solution for material damage will not be realized unless sufficient time elapses to permit the cracks to change length. To allow for rate dependence.Rate Dependence 5. can be dramatically altered by the rate at which loads are applied. the stress will drop down toward the quasistatic solution.e.1. the equation of state (i. the stress state will lie outside the yield surface. The GeoModel uses a generalized DuvautLions [13] ratesensitive formulation. 5. but the physical mechanisms that give rise to observable inelasticity can not proceed instantaneously. then the total damage will be ultimately lower than it would have been under quasistatic loading through the same strain path. The user specifies a “relaxation” parameter governing the characteristic speed at which the material can respond inelastically. and the very character of the inelastic deformation itself. If a stress level is high enough to induce crack growth. For example. During the time that cracks are growing towards the quasistatic solution. an overstress model is used. then inelasticity will be evident. Until sufficient time has elapsed for the material to equilibrate. and (2) the highrate solution σ H corresponding to insufficient time for any plastic damage to develop so that it is sim˜ ply the trial elastic stress.
Rate Dependence shows that the updated stress will be (approximately) a linear interpolation between the lowrate quasistatic plasticity solution σ L and the highrate purely elastic solution σ H . The inset graph shows how the scale factor η varies with the loading interval. If the loading interval is considerably shorter than the material’s characteristic response time. For a given strain increment. there exists a scalar η between 0 and 1 that depends on the strain rate such that σ ≈ σL + η ( σH – σL ) . then the ˜ solution will be the highrate elastic solution. two limiting solutions can be readily found. ˜ ˜ ˜ ˜ σH ˜n+1 pure elasticity (5.1. which lies on the yield surface. In ˜ ˜ other words. Rate dependence.1) σ ˜ n+1 ˜ yie ld nt ion de olut en ep ic s e d ast rat copl vis su rfa ce σ low = σ L ˜n+1 ˜n+1 ˜ ˜ η n+ qu as H+1 ( σn ~ 1 + ~n –σ L1) is ta tic ic i ty η σ low = σ L ˜n ˜n ˜ ˜ σn η n( ~ i as qu H –σ n ~ L) 1 ∆t τ σ low ˜n–1 pure elastici ty pl as t σ high ˜n σ ˜n σH ˜n ˜ dent tion epen lu rate d lastic so p visco Figure 5. then sufficient time exists to fully develop plastic response and the updated solution therefore coincides with the quasistatic solution σ L . If the ˜ ˜ ˜ ˜ loading interval is long relative to the material’s characteristic response time τ . The actual up˜ dated ratedependent viscoplastic stress σ falls between these two limiting case solutions so that ˜ σ = σ L + η ( σ H – σ L ) . The “low rate” solution σ L . ic at st tic as pl ity σ ˜n–1 ld yie ce rfa su 64 . The high rate solution σ H is simply the trial elastic stress. is the solution to the rate independent ˜ GeoModel governing equations.
2). Effects of plasticity are apparent in the GeoModel only when the time interval is long or when the characteristic material response time is short so that η will be small. experimental data for highrate loading scenarios have the appearance of inducing a higher yield stress in the material. but uses a somewhat different weighting factor. where the abscissa is normalized by a factor τ called the material’s “characteristic” response time. the final stress state will not lie on the yield surface. Consequently. the overstress model accomplishes the same effect in a much more physically justifiable manner. We will prove that. in this case. the “attraction” that the dynamic stress has for the quasistatic solution increases somewhat as the stress moves farther from the yield surface. the scale factor is smaller than the value cited in Eq.1 illustrates that the viscoplastic solution will follow a trajectory that is similar to the quasistatic solution except displaced from the yield surface.Rate Dependence The update for internal state variables is structured similarly. ˜ Figure 5. according to Eq. we will describe how the GeoModel assigns a value for the characteristic material response time τ .2) This is also the rate factor for the stress at the onset of yielding when σ and σ low coincide.1 shows a stationary yield surface.1. for simplicity. Fig. Soon we solve the viscoplastic equations to prove that. the solution will be near the quasistatic (lowrate) solution σ L . the GeoModel does not alter the material yield stress as a function of strain rate. At the end of this chapter. the interpolation factor η varies from 1 at high strain rates (when ∆t is small) to 0 at low strain rates (when ∆t is large). A time interval is “short” if ∆t « τ . Consequently. 65 . A time interval ∆t is deemed to be “long” if ∆t » τ .1). (5. In general. 5. ˜ ˜ At the end of a viscoplastic step. as illustrated in the graph inset of Fig.. Incidentally. (5. Referring to Fig. the η weighting factor is large when the time step is significantly smaller than the characteristic time required for the material’s plasticity solution to develop.2. the yield surface will evolve in size or translate according to the hardening rules described earlier. if the initial stress is on the yield surface. Instead. In this case. then highrate scale factor internal state variables (ISVs) is 1 – e –∆t ⁄ τ η = . Unlike some plasticity models. For both the stress and internal state variable updates. as explained below. 5. ∆t ⁄ τ (5. 5.
using Eq.( σ ij – σ ij ) . τ is a material parameter called the relaxation time. The stress rate is. an inviscid (rateindependent) solution σ L for the ˜ stress is presumed to exist. * The “low” or “inviscid” stresses must be tracked independently.6) · high high Here. (5. These limiting case solutions are merely the solutions of the rateindependent GeoModel equations described in earlier chapters.3).4) –1 The fourthorder tensor C ijkl is the elastic compliance (inverse of the stiffness). ˜ initially being equal to q 0 at the beginning of the step.3) ˜ ˜ ˜ The viscoplastic part of the strain rate includes both the usual plastic strain rate from the quasistatic (lowrate) solution as well as additional (retarding) contributions resulting from viscosity. given by the elastic stiffness acting on the elastic part of the · · strain rate: σ ij = C ijkl ε e . τ (5.* At the end of the time interval. in Appendix B). the inviscid quasistatic ISVs (QSEL. which is found by integrating the rateindependent Geo˜ ˜ Model equations from the earlier chapters. (5. etc. Like the inviscid quasistatic stress. σ ij is the elastic trial stress rate. QSSIGYY. Viscous effects are incorporated by presuming that the · · strain rate is decomposed as the sum of an elastic part ε e plus a viscoplastic part ε vp : ˜ ˜ · · e ε vp · ε = ε + .[ q – q low ] .C ijkl [ σ kl – σ kl ] .) must be tracked as distinct extra state variables. (5. τ where · high · σ ij ≡ C ijkl ε kl . as usual. The viscoplastic strain rate is governed by 1 –1 · vp low ε ij = . σ low ultimately has the value σ L . and therefore σ ij is the time varying elastic H trial stress that ultimately equals the highrate solution σ ij at the end of the step. Thus. They cannot be inferred by projecting the actual stress onto the yield surface. each internal state variable q is presumed to vary according to –1 · q = . 66 . and (through application of the rateindependent GeoModel) ultimately equalling the low rate solution q L at the end of the step. Attempting to do so causes undesirable results in ratedependent loadunload cycles. here denoted collectively by “q”. Likewise inviscid (rateindependent) solutions q L are presumed available for the internal state variables.Rate Dependence Viscoplasticity model derivation In the context of viscoplasticity. the stress rate may be written · · high 1 low σ ij = σ ij – . etc. QSBSXX. During viscoplastic loading.5) Here q low is the value of the internal state variable ( κ or α ) throughout the time interval. τ (5. and σ low is the rateindependent stress solu˜ tion whose value at the beginning of a time increment ∆t is tracked as an extra state variable (called QSSIGXX.
Since σ 0 equals σ at the beginning of the step. τ (5. Recall that the final solution σ H – σ L can be presumed known at the end of the step ˜ ˜ because σ H is found by integrating the elasticity equations and σ L is found separately by ˜ ˜ integrating the inviscid quasistatic plasticity equations.8) This equation can be solved exactly if the time variation of σ high – σ low is known ˜ ˜ throughout the time step. we may call on the mean value theorem to assert that t – t0 t – t0 high low ( σ high – σ low ) ≈ ( σ H – σ L )  + ( σ 0 – σ 0 ) 1 –  . so σ low also varies approximately lin˜ early through time (with nonlinear effects from flow potential surface curvature being higher order). In what follows. we would need to solve the rate independent equations analytically over the entire time step to integrate Eq. (5. the dynamic accuracy can be maximized by presuming that the rate of σ low is constant over the step so that σ low itself is approximated to vary linearly ˜ ˜ over the step. let’s introduce a change of variables by defining high u ij ≡ σ ij – σ ij .( u ij + σ ij – σ ij ) . (5. it varies linearly through ˜ time from its initial to final value (with small higherorder nonlinearities if the strain rate · and/or elastic moduli are not constant). Thus. ˜ The highrate stress σ high is simply the elastic trial stress. Integratt0 ij ˜0 ing the ODE.11) 67 . ˜ ˜ ˜ ˜ ˜0 ˜0 where – ∆t ⁄ τ RH = 1 – e ∆t ⁄ τ (5.6) is a set of linear firstorder differential equations which may be integrated low exactly over a time step with the use of integrating factors [19] provided that σ ij is known as a function of time throughout the time step. we will describe how the time low history of σ ij can instead be well approximated over the step.7) so that the governing equation for the stress rate may be written 1 high low · u ij = – .10) and r h = e –∆t ⁄ τ – R H . Similarly. and applying the definition of u ij to obtain the updated solution for σ ij eventually gives σ FINAL = σ L + R H ( σ H – σ L ) + r h ( σ high – σ low ) . evaluating the result at the end of the step. the difference σ high – σ low is known at the beginning of the step because σ high is σ at the beginning of ˜0 ˜0 ˜0 ˜ low the step and σ 0 is retrieved from the saved quasistatic stress extra state variable array. Time variation of σ low is governed by known quasistatic rate ˜ equations. (5. First. the initial condition is that u = 0 when t = ˜ . (5.8) may be solved exactly. Consequently. but this is not tractable in practice. In principal.Rate Dependence Eq.6) exactly. The quasistatic stress rate σ low is an oblique pro˜ jection of the trial stress rate onto the yield surface. (5. (5. the ODE in Eq. Thus.9) ˜ ˜ ˜ ˜ ∆t ˜ ˜ ∆t high With this approximation.
1) becomes ˜ ˜ η FINAL = σ L + η ( σ H – σ L ) .2 2 2 4 4 6 6 8 8 ∆t 10 τ 10 Fig. (5. (5. (5. Hence.13) 0. Recall that the material responds elastically at extremely high rates. The largest weight the last term of Eq.1) if we approximate that σ 0 – σ 0 is par˜ ˜ allel to σH – σ L .8 0. ϒ increases. As seen in timestep “n1” in Fig. moment. The lowering of the rate factor η caused by nonzero ϒ makes the dynamic stress more strongly attracted to the quasistatic solution as the distance between them increases. Highrate weighting factor ies with the stress difference ratio ϒ appearing in at various initial states. 5. Otherwise. For the rate dependent update of internal state variables. Limiting case. σH – σL ˜ ˜ (5.16) 68 . the equations outlined in this chapter imply that this steady state stress difference is given by τ · L σ ij – σ ij = . (5. Eq.10) can be put into the form of Eq. ϒ factor (upper red curve) applies when the initial stress is on the yield surface.1. (5. (5. the high rate solution q H for any internal state variables is simply its value at the beginning of the time increment. high low Eq. In this case. With this assumption.1) unless σ 0 ˜ ˜0 values for the “high” and “low” rate paths coincide only at the onset of plasticity.2 ϒ=1 0.4 ϒ=0 (5.11).0 where 1 η = RH + rh ϒ and σ high – σ low ˜0 ˜0 ϒ ≡ .12) σ ˜ ˜ ˜ ˜ 1.4 0.13). (5. If a strain rate is held constant for a long enough period then the differL ence between the dynamic stress σ ij and the inviscid stress σ ij will sometimes reach a steadystate value in the laboratory. For plastic loading. beyond this onset of yielding the yield surface at the beginning of the step.10) is not precisely of the form high = σ low .2 shows how the weighting factor η varFigure 5..5) can be integrated analytically if the time variation of q over the step is approximated by q low = q H + ( q L – q H ) ( t – t o ) ⁄ ∆t . the initial shown in Eq.2. (5. (5. (5.6 0. The facequals 0 only when the initial state is on the yield tor is lower if the initial state is already off surface. Eq. eventually asymptoting to 1 under steady strain rates. The resulting solution for q is q FINAL = q L + R H ( q H – q L ) .P ij Q mn ε mn χ (5.Rate Dependence With the presence of the last term. 5. because the internal state variables can evolve only when plasticity occurs. the solution in Eq. The weight factor R H for the internal state variables differs from the weight factor η for the stress because the two problems have different initial conditions.8 0.14) 0.15) where the highrate weight factor R H is the same as in Eq.6 0.
17) implies that · L σ 12 – σ 12 = 2Gτε 12 4G · L σ 11 – σ 11 = .Rate Dependence where χ . Eq. 010 1ˆ [ S ] = . 5.τε 11 3 for simple shear for uniaxial (axisymmetric) loading (5. (4.1 0 0 2 000 for simple shear (5. and Q ij are defined in Eq. in the case of simple shear for nonhardening Von Mises plasticity. This calculation included a lower strength in tension (interestingly.21) These simple analytical results can be used to trendtest the numerical implementation of rate sensitivity.89).18) 2 0 0 1ˆ [ S ] = . the normalizer σ HEL is the quasistatic uniaxial yield stress. as described in Fig. σ 11 ⁄ σ HEL σ 11 L σ 11 L ( σ 11 – σ 11 ) 1 time ε 11 –1 Figure 5. this equation becomes ˆ ˆ · L σ ij – σ ij = ( 2Gτ )S ij S mn ε mn (5.3.20) (5. This rateinduced stress difference is plotted in the second figure (normalized by the peak value). In the very simplified context of nonhardening VonMises plasticity. For example.17) ˆ where G is the shear modulus and S ij is a unit tensor in the direction of the stress deviator.19) Therefore. (5. (5. where the dashed red line shows the analytical asymptote envelope from Eq.3. In the stressstrain plot. P ij . the difference between dynamic and quasistatic strengths is higher in tension). The dynamic (black) stressstrain plot exhibits an apparent increase in strength relative to the quasistatic (dashed blue) solution. Rate dependence in uniaxial strain loading of a nonhardening pressureinsensitive material. 69 .0 – 1 0 6 0 0 –1 for uniaxial (axisymmetric) loading (5.21).
Fig. T 1 through T 7 which are employed in the code to assign a value of the characteristic time according to the following formulas: · p f ( ε equiv ) if ε v ≤ 0 τ = . and all other T’s to zero. 8. (5. ˜ (5. If a constant characteristic time τ is desired. · equiv ε and 0 <x> = x if x ≤ 0 if x > 0 and · · ε equiv = ε . 5.3 used a constant τ . 70 .8(b) on page 92.24) (McCauley brackets).22) · p T 6 f ( ε equiv ) ( 1 + ( <T 7 ( I 1 + T 5 )> ) 2 ) if ε v > 0 and I 1 ≥ – T 5 where 1 T2 · f ( ε equiv )=T 1  . Suitability of the GeoModel’s overstress ratedependence theory for predicting laboratory data is illustrated in Fig. as did the simpleshear ratedependent simulation shown in Fig. 7. then set T 1 = τ .23) (5.9 on page 103.Rate Dependence Assigning a value to the characteristic material time The GeoModel permits the user to control the value of the characteristic time through the use of up to seven positivevalued parameters.
Thus. but at high pressures and high strain rates under adiabatic conditions.. How can this be? In thermodynamics. the model may be used for problems where low pressure regions are also isothermal (and near room temperature) and high pressure regions are adiabatic (high rate). for example. A good method for gauging the degree to which the mechanical GeoModel can be applied in broad thermodynamic domains would be to compare isothermally measured elastic moduli (inferred by the slope of a stress strain curve) with acoustically measured moduli. you restrict attention to isothermal loading.. terms involving material properties such as the specific heat and Grüneisen parameter). If. Nonetheless. p = f ( ρ. using the model in highrate/lowpressure applications (acoustics) or lowrate/high pressure problems (e.e. Specifically. The GeoModel is parameterized at low pressure and low strain rates under isothermal conditions (room temperature). where the “material constants” in the equation (such as the bulk modulus) must be set to their isothermal values. creep of underground salt bodies) will possibly require reparameterization. the GeoModel’s nonlinear elasticity model has been successfully fitted to Hugoniot shock data. Without information on the thermal properties of the material. the GeoModel’s equation of state is incorporated within the purely mechanical model — it contains no thermal terms (i. pressure is typically expressed as a function of two variables: the density ρ and a thermal variable (usually temperature or entropy). which are the lowpressure isentropic elastic constants. T ) . then the pressure is again expressible as a mechanical function if the parameters such as the bulk modulus are set to their adiabatic values.25) In thermodynamics. 71 . Likewise. then the pressure will be expressible in the form p = F ( ρ ) . The difference between the isothermal and isentropic bulk modulus (at a given pressure) is proportional to the square of the thermal expansion coefficient divided by the specific heat.g. In particular. (5. you can always use a purely mechanical equation of state if you restrict the class of allowable problems so that one or two of the thermal variables are interrelated in some known way. For example. the nonlinear fit for the bulk modulus may be regarded to transition from isothermal to adiabatic properties as pressure is increased. if you can consider only adiabatic loading. Thus. the domain of applicability of the purely mechanical model is likely broad. This implies a vague “domain of applicability” for the mechanical GeoModel.Rate Dependence Thermodynamics considerations At present. if no differences are observed between the isothermal and isentropic moduli at a given pressure. the validity of the mechanical GeoModel in other domains cannot be ascertained.
called directly from the host code. GeoModel Numerical Solution scheme Aside from kinematic hardening.. ∂x k (6. This chapter begins with a description of how the GeoModel is to be used within a host (finiteelement) code. f i is the body force per unit volume.39).1) where σ ik is the spatial Cauchy stress tensor (denoted σ spatial in Eq. GeoModel installation instructions are provided that describe the public* subroutines and memory requirements. requiring all six independent components of symmetric tensors to be passed to the model. ∂σ ik . Careful numerical integration schemes [6] are required to ensure accuracy and convergence. 72 . Role of the GeoModel within a finiteelement program The GeoModel is designed for use in host codes (typically finiteelement programs) that solve the momentum balance PDE. ρ is the mass density. Following a summary of plotable GeoModel output.e.e. 3. and a i is the material acceleration that is related to the spatial velocity field v i by material time derivative ∂v i Dv i ∂v ∂v i a i = i + .+ f i = ρa i .GeoModel Numerical Solution scheme 6. For any isotropic elasticity model. x k is the spa˜ tial position vector. An eigenvalue analysis is avoided by casting the yield function in terms of stress invariants.2) * i.v k = . The principal stress directions (eigenvectors) are not needed to evaluate the yield function. the governing equations must be cast in incremental tensorial form. stability of the spatial finiteelement solution. however. ∂t X ∂t ∂x k Dt ˜ (6. not stability of the GeoModel’s internal time integration algorithm). Thus. the GeoModel’s time integration algorithm is briefly summarized.= . Next. evolution of plastic response must allow for rotation of principal stress directions caused by the elastic portion of the loading.. the GeoModel is isotropic and therefore the yield function depends only on principal stresses. followed by a discussion of the influence of material softening on fieldscale stability (i.
3). all installations of the GeoModel have approximated the unrotated strain rate by the unrotated symmetric part of the velocity gradient defined in Eq.a n ( ∆t ) 2 . (6. x n + 1 = x n + v n+ 1 ⁄ 2 ∆t . once the GeoModel and any other constitutive models in the problem have been applied to determine stresses at the end of a time step.GeoModel Numerical Solution scheme Here.  ( ∆t ) 2 ˜ ˜ ∂t X 2 ∂t 2 X ˜ ˜ ˜ ˜ (6. evaluated at the halfstep by using the velocity field in Eq. ˜ ˜ ˜ where v n+ 1 ⁄ 2 ≡ v n + 1 a n ∆t . (6. half. The GeoModel integrates the unrotated strain rate to predict the unrotated stress at the end of the step.7). 2˜ ˜ ˜ ˜ Equivalently. ∆t + .7) To date.1) may be integrated again to update the acceleration field. (4. thus launching a new timestep cycle. ∂X j t (6. 2˜ ˜ ˜ (6. the updated position of a material particle is computed to secondorder accuracy with respect to the time step ∆t through application of ∂x n 1 ∂2 x n x n + 1 = x n + . or end) at which un/rotation operations are required.6) (6. The spatial position vector x is related to the reference position X ˜ ˜ through the deformation mapping function. which must then be rotated into the spatial configuration by the host code. so that application of Eq. In most host codes. Of course. Eq. X is the timezero reference position vector. 73 .5) (6. (6.1) permits evaluation of the acceleration field. As mentioned on page 20 (and clarified later in this chapter). optimizing the accuracy of the spatial solution for problems involving massive material rotation requires the host code to apply its un/rotation operations using polar rotation tensors that are consistent with the part of the time step (beginning. such that the deformation gradient tensor is ∂x i F ij = .3) The stress and velocity fields (as well as displacement or velocity boundary conditions) are known at the beginning of each time step. which serves to identify Lagrangean ˜ material points.4) or 1 x n + 1 = x n + v n ∆t + .
In either case. failed catastrophically. the GeoModel has reached the limit of its applicability because large scale cracking and subsequent loss in strength cannot be modeled locally at the constitutive level. Instead. rather than properly softening down away from the limit state. 4. otherwise “CRACK” equals 0.GeoModel Numerical Solution scheme Spatial stability (mesh dependence/loss of strong ellipticity) As rock deforms inelastically. More correctly.00 whenever a principal stress (or I 1 ) cutoff has been applied or 2. If the stress reaches the limit surface.e.00 when the stress has reached the limit surface and can harden no further. Specifically. The host code is responsible for responding appropriately to these flags by initiating material softening. The GeoModel sends flags back to the host code at the onset of softening (a nonlocal phenomenon). softening results from subgridscale flaw clustering. requiring intervention from the host code to change its solution scheme appropriately (for further details. 74 . which can be shown to have a Weibull (or nearly Weibull) distribution and can be parameterized via standard laboratoryscale experiments.0. In the GeoModel. the GeoModel will force the stress to dwell at the limit state). which is especially appealing because (unlike “element death” and cohesive zone models) it can be easily justified physically. see page 61). Because the GeoModel comes equipped with its own flags for failure. the state variable “CRACK” is a flag equaling 1. Material softening generally produces a change in type of the momentum equation. To summarize. If the host code fails to activate any special response when the “CRACK” flag becomes nonzero.16) evolves toward — and is not permitted to evolve beyond — the limit surface (upper curve in Fig. * We are investigating a fieldscale softening strategy that introduces a length scale based on Weibull perturbations of the material strength field. this model should not be used with other fracture models such as a maximum principal stress criterion.. 4. the GeoModel will continue to run. the geomodel will handle the inelastic response at the limit state in a manner similar to nonhardening plasticity (i. the “CRACK” flag should be queried by the host code to determine when it is appropriate to add void (when pressure is tensile) or to apply discontinuous shear displacement element shape functions (when pressure is compressive) or to apply any other appropriate response* to material softening that will ensure localization response that converges as the spatial mesh size is reduced. the GeoModel is intended to model only the portion of material response that is appropriate to compute at the local constitutive level.16). Preliminary investigations [4] have shown that Weibulllike softening strategies lead to very realistic fragment patterns and are meshindependent. the rock has. a positive value of “CRACK” marks the onset of softening. Without a meaningful hostcode response to failure. but its predictions are suspect. in a loose sense. the initial yield surface (lower curve in Fig.
75 . This property array must be passed to the subroutine GEOCHK for “domain certification” (i. as explained on page 81). Likewise.. the GeoModel conforms to Sandia’s Model Interface Guidelines (MIG) [10].e. these routines must be written and maintained by the hostcode architects. To support portability. the model has three public† subroutines (GEOCHK. a routine called TOKENS (used only in the extra variable request routine) is expected to preexist in the host code. Log message protocols and bombing procedures vary among host codes. As described below. Additionally. subroutines. routines that are called directly from the host code. Host code architects of nonMIGcompliant codes may request sample MIGutilities that they may customize to suit their own code’s protocols for information passing and code termination.* The GeoModel is designed to be implemented into multiple host codes without any revision of the source code. the model presumes that calculations entail three distinct phases. ISOTROPIC_GEOMATERIAL_CALC must be called to update the stress to the end of the step. For portability. 3. define plot options. the GeoModel follows MIG guidelines by calling subroutines LOGMES. and ISOTROPIC_ GEOMATERIAL_CALC). Detailed descriptions of the inputoutput arguments are provided below. the host code loops over the lists returned by GEORXV to then allocate the storage. Therefore. Every cycle. verifying that input values fall within allowable ranges. or BOMBED whenever it needs to relay messages to the user. This section is a software requirements specification that must be followed by anyone who installs the GeoModel into a host code. 2. This routine returns physical dimensions. relay messages to the user or terminate the calculation. initial values. is available that will repair advection errors. Because these actions require hostcode responses that vary from code to code. Storage. Any host code that already supports MIG models will already have these MIGutilities in their repository. plot keywords. upon occasion. * Any deviation from these model installation instructions (as well as any modification of the GeoModel source code itself) may result in loss of technical support.e. All other routines in the GeoModel are “private” and should not be called by the host. These routines are not part of the GeoModel source code.. isotropic_geomaterial_state. one additional public routine. the first two of which are performed at startup while the last one is applied for every element at every timestep: 1. the GeoModel does not actually allocate storage for internal state variables (ISVs) — this is the responsibility of the host because data layout varies from code to code. GEORXV. The GeoModel must. † i. memory requirements. respectively. and initialize the ISV fields. and model installation requirements. User input. the routine GEOCHK sets defaults for unspecified user inputs. The GeoModel provides a list of storage requirements by requiring the host code to call subroutine GEORXV. To use this routine. To be portable. Execution. or terminate calculations.GeoModel Numerical Solution scheme GeoModel files. and advection requirements for each internal state variable. To serve codes that remesh. log fatal errors. Model installers who believe that a deviation from these instructions is warranted are encouraged to contact the GeoModel developers. The GeoModel requires the host code to acquire user input values and save them into a single array using the keywords and ordering listed in Appendix B. FATERR.
as described in the nomenclature table in Appendix B. Arguments passed to and from the GeoModel driver routine The following list describes variables passed between the host code and the GeoModel’s driver routine (isotropic_geomaterial_calc): INPUT: • NBLK: The number of cells or finite elements to be processed. unrotation of the strain rate at time n+1/2.GeoModel Numerical Solution scheme The GeoModel source code is available in both FORTRAN 77 and 90. The F90 routines. largescale production codes have used the F77 version to best ensure portability. are generated from the F77 master files to ensure consistency. σ 23. OUTPUT: • SIG: The unrotated stress tensor at time n+1. this array is dimensioned “SIG(6. σ 22. • D: The unrotated strain rate tensor. NBLK)” so that the stress components for any given finite element are in six contiguous memory locations. preferably evaluated at time n+1/2 because the GeoModel treats the strain rate tensor as constant over the entire interval. The six independent components of the stress must be passed in the ordering { σ 11. Parallel codes send only one cell at a time (NBLK=1). 4. • SV: the internal state variable array containing reals. σ 12. and rotation of the updated stress back to the spatial frame at time n+1. where ρ is mass density) when setting the timestep. this requires using three different polar rotation tensors [14]. The host 3 code may use the USM output to compute an upper bound on the wave speed ( H ⁄ ρ . These arguments require unrotation of spatial stress at time n. which are not significantly different. Component ordering and contiguous storage are the same as for stress.G . • NINSV: The number of internal state variables for the GeoModel. For moderate rotation problems. Within the FORTRAN. For problems involving significant material rotation. The component ordering is the same as described above. To date. σ 33. as summarized at the top of the nomenclature table in Appendix B and also summarized within the source code prolog itself. • DT: The time step • PROP: the userinput array. but the accuracy consequences of this simplified approach have not been carefully examined in the computational mechanics literature. it might be acceptable to use the rotation tensor evaluated at time n+1/2 for all three operations. 76 . so those routines will be described here. filled with real numbers. • SV: The internal state variable array (updated to time n+1) • USM: Uniaxial strain (constrained) elastic modulus equal to H = K + 4. • SIG: The unrotated Cauchy stress tensor at time n. σ 31 } .3). Most codes approximate the strain rate tensor as the unrotated symmetric part of the velocity gradient (see Eq.
GeoModel Numerical Solution scheme PLOTABLE OUTPUT In addition to the stress. quantifies the angular location of the stress in the octahedral plane. while “EQPV” gives plastic volume strain (and is roughly equal to the porosity change from inelastic void collapse).1. “ROOTJ2” may be regarded as a scalar measure of effective shear stress and is proportional to the radial coordinate of the stress in the octahedral plane. If kinematic hardening is enabled. While “SHEAR” quantifies the plastic loading direction relative to the yield surface. This calculation uses a Von Mises yield surface. Halfway through the problem. and it varies from 30 for triaxial compression to –30 for triaxial extension (0 for simple or pure shear). The “EQDOT” variable may be plotted to gain an overall sense of intensity of the current strain rate. an intermediate value of “SHEAR” indicates oblique plastic loading relative to the yield surface. the internal · state variable “DCSP”. The “SHEAR” variable is an informational output. may be plotted as a measure of the magnitude or intensity of plastic loading. Meaning of the “SHEAR” output variable. which is why SHEAR (c) jumps to and holds at 1. During this interval. the “SHEAR” variable should be plotted. 77 . The complete list of other (less useful) plotable output is in Appendix B. The “INDEX” flag may be plotted to locate regions that are now or have ever deformed inelastically (even if elastic at the given instant).0 as the tangential component decays.8 elastic plastic 10 20 0. 6.2 –30 30 (a) (b) SHEAR time 2000 0. The Lode angle. the “CRACK” flag may be plotted to visualize softening regions. As mentioned earlier.6 0. SHEAR first jumps to 0. 1 Lode angle 10 elastic 500 plastic 1000 1500 0. and moves back toward 1. To visualize regions that are currently deforming inelastically.4 (c) 500 1000 1500 time 2000 Figure 6. “LODE”. The plot keywords (and ordering of variables in the SV array) are listed in Nomenclature Appendix B. any variable in the SV state variable array is available for plotting. “SHEAR” ranges from zero when the trial stress rate is tangent to the yield surface to unity when it is normal to the yield surface. equal to zero during elastic cycles and equaling a measure of normality of the trial elastic stress rate during plastic intervals. Specifically. The stress invariant “I1” is three times the negative of pressure. Small values of “EQDOT” correspond to relatively quiescent regions.5 where the normal and tangential components are equal. the strain rate direction is changed in stress space [as indicated by arrows in (b)] to move the stress toward triaxial compression. At the beginning of this transition. as illustrated in Fig.1.0. The strain rate remains in triaxial extension for half of the calculation. which is the plastic consistency parameter λ . “BACKRN” quantifies the distance that the origin of the octahedral profile has shifted in stress space. the stress reaches the yield surface and continues to push directly against it. which is why the Lode angle (a) is initially constant at – 30° . Plotting “EQPS” will show equivalent plastic shear strain.
STEP 8. thereby updating the internal state variables. Apply Hooke’s law in rate form to obtain the elastic stress rate at time n. Compute the nonlinear elastic tangent moduli appropriate to the stress at time n. and the inviscid algorithm returns (i. STEP 15. (4. (4.GeoModel Numerical Solution scheme GeoModel algorithm The GeoModel presumes that the strain rate is constant throughout the entire step. STEP 6. 16. STEP STEP 5. All subsequent steps described below this point apply to the smaller time steps associated with subcycles. (4. 2. (4. Save the values of the internal state variables into the state variable array. (4.85) Evaluate the isotropic hardening coefficient h κ in Eq. Otherwise. apply a return algorithm to ensure the initial stress is on or inside the yield surface. Viscous part of the viscoplasticity equations. Use forward differencing (within the subcycle) to integrate Eqs.. 11. STEP 1. To reach this step.87) and (4. Evaluate the G α function in Eq. guard against slight roundoff and integration errors by applying an iterative return correction to place the stress exactly on the yield surface. the trial elastic stress is accepted as the final updated stress. go to STEP 7.10) to compute the dynamic stress. STEP 12. Apply Eq. 7.22) to compute the characteristic material response time. At this point. Now apply Eq. If subcycles remain to be evaluated. apply Eq.85) to advance the stress to the end of the subcycle. so the updated stress will be in principle already on the yield surface.72) and (4. continue to STEP 16. Using the trial elastic stress corresponding to an update to the end of the time step. the time step is divided into an internally determined number of subcycles.76).83) to obtain the consistency parameter. (4. go to STEP 16). The above steps will have directly integrated the governing equations through the end of the subcycle. STEP 19. STEP 14.72). The previous set of steps govern computation of the equilibrium state. 9.76). (5. (5. Evaluate the gradients of the yield function for eventual use in Eq. and save the partially updated inviscid internal state variables. STEP 13. However. (4. STEP 78 . Otherwise. STEP 4. (5. Evaluate the yield function at the trial elastic stress.83). and the stress is integrated as follows: Rate independent (inviscid) part of the viscoplasticity equations. Increment the subcycle counter. 10. Similarly integrate Eq. STEP 18. continue. Apply Eq.e. STEP 17. If the yield function evaluates to a negative number. STEP STEP STEP STEP To guard against unpredictable hostcode advection errors (or similar corruption of the updated state from the last time step). Evaluate the flow potential gradients for eventual use in Eqs. Integrate the elastic stress rate using firstorder differencing to obtain an estimate for the trial elastic stress at the end of the step. the trial elastic stress state was found to lie outside the yield surface. κ and α ij . STOP. STEP 3.15) to similarly compute the dynamic values of internal state variables to account for rate sensitivity.
neither model is “well verified.” but the first one is certainly better verified. varying degrees of SQA. Once a constitutive model is installed within a host code. to speak of varying degrees of progress towards verification†. Constitutive SQA in the context of largerscale model integration is limited to pro* Whereas “Verification” seeks to confirm that the equations are solved correctly without questioning their appropriateness. it becomes only a single component of a much larger and different model (the finiteelement code). In this sense. “Code verification” is defined by the IEEE [22] as “formal proof of program correctness” in the sense that the governing equations are numerically solved correctly within a tolerable degree of accuracy. It seems acceptable. documentation. Conclusively demonstrating only firstorder accuracy of an anisotropic solid constitutive model would require more than twenty times the effort needed to verify firstorder accuracy of a simple fluids model (this follows because a general anisotropic stiffness tensor has 21 independent components. which encompasses a broad range of activities including code maintenance. Model validation* will not be discussed until Chapter 9. that one constitutive model has been verified for both uniaxial strain and simple shear. but verification of the installation is primarily a host code (not constitutive) responsibility. the GeoModel has undergone a higher level of SQA than is normally applied to modern material constitutive models of comparable complexity. Constitutive SQA should include ensuring that the model can be installed and run in a variety of host codes. for example.‡ Even though we claim that the GeoModel’s verification and SQA status is above average. or. ‡ From a practical (rather than philosophical) standpoint. 79 . more generally. In other words. and therefore 21 independent strain paths would be required to conclusively verify accuracy. we cannot state with absolute certainty that the governing equations presented in this report are in fact solved correctly. † Suppose. Realistically. With a given level of financial and computational resources. whereas another one has been tested only in uniaxial strain. If these are the only tests.Software “confidence building” activities 7. therefore. and (most importantly) code verification. the complexity of a model must be considered when speaking about how well a model has been validated. One might challenge the Verification and Validation (V&V) community to prove that “formal proof of program correctness” is even possible. moreover comprehensive testing for solids models requires coordinate invariance tests that are not needed for scalar fluids models). we aim only to build confidence in the GeoModel by summarizing some of the SQA activities that have been applied to the GeoModel to date (a comprehensive detailed discourse would fall well outside the scope of this report). solids models cannot be tested to the same level of certainty as fluids models. “validation” compares model predictions against experimental data to determine whether or not the equations themselves are indeed suitable for the application. we do not assert that such activities have progressed to the point where we consider the job “finished. the confidence one can place in the veracity of any model prediction can be based only on the extent to which documented evidence suggests that the equations are solved correctly.” Here in this chapter. We make no claims at this point that the GeoModel software has been exhaustively verified. Software “confidence building” activities This chapter describes progress towards Software Quality Assurance (SQA).
a partitioning of responsibility for SQA is needed. As such. Constitutive models are subcomponents within fieldscale calculations. 80 . Similarly. Numerical schemes for solving field PDEs (such as NavierStokes equation or Maxwell’s electromagnetism equations) have received considerable attention in the V&V literature. When a host code sends inadmissible inputs (such as corrupted strain rates caused by mesh entanglement. while verification of constitutive models has been relatively ignored. wherever feasible..Software “confidence building” activities viding adequate model installation instructions (including operational constraints). we regard suppression of softening as an implicit software requirement specification because our customers (finiteelement code teams) do not have the code infrastructure that is needed to properly handle the change in type of their governing PDEs that occurs upon softening. because the GeoModel is a local constitutive model (i. Therefore. SQA may additionally include some checking of the inputs themselves. advection. It is well known that mesh dependence can occur when local constitutive models permit material softening.e. the GeoModel predicts only the onset of catastrophic failure. Until such enhancements are made at the host code (not constitutive) level. delivering correct solutions to the constitutive governing equations. and ensuring that the model will. as long as doing so does not compromise efficiency. demonstrating convergence with respect to the spatial mesh is not a GeoModel verification responsibility.) Constitutive verification aims to build confidence that the model will return correct solutions to the governing equations. not its subsequent evolution into macroscale fragments and fractures. presuming that the host code sends inputs falling within the admissible domain for the model. because it does not solve spacetime PDEs). (As an analogy. or artificial viscosity errors). then correcting such errors is not the responsibility of a constitutive modeler unless it can be proved by the host code developers that such errors originated from constitutive model output errors. “trap” invalid calling arguments (much as a compiler “traps” IEEE errors such as division by zero). note that the quality assurance responsibilities for a turnsignal manufacturer must be different from those of an automobile manufacturer.
.g. Direct runtime testing of time varying subroutine arguments is computationally inefficient. To date. dividing by zero. • Technical support. so the GeoModel’s domain certification for variable calling arguments relies primarily on our model installation instructions (page 75).g. provides input definitions. the GeoModel has undergone the following static SQA: • Independent* linebyline review of the source code. JAS3D. The source code was aggressively inspected to locate and remove possibilities for IEEE errors (e. Dell. [25]. The domain for the GeoModel includes constraints on runtime subroutine arguments (e.g. Runtime testing of this routine is discussed in the next section. the bulk modulus must be positive). pgf90. No function or set of equations is well posed without a domain of applicability. trying to take the ArcSin of a number larger than unity†. A host code owner who faithfully obeys the model installation instructions on page 75 may update the GeoModel by simply replacing three FORTRAN files (posted on WebFileShare under keyword “GeoModel”) with the latest GeoModel release and then recompiling. This report is the first publication that describes the numerical algorithm. For example. as in ASIN(1. • Model domain certification (preventing “Garbage In ⇒ Garbage Out”). etc. • Model documentation. Two standalone singlecell codes that exercise the GeoModel under prescribed strain and stress paths are available to assist host (finiteelement) code owners verify their GeoModel installations. Except where computational efficiency would be degraded. ‡ The code is simply the release date in the form “yymmdd”. GeoModel code inspection was first performed by Brannon (the second author of this report) upon joining this modeling effort at the beginning of fiscal year 200304.). The source code compiles without warnings when using stringent SQA options (such as Wall O3 in the gnu compiler). “independent” (which is not synonymous with “unbiased”) means that the source code has been inspected by an individual who has not written the code. • Portability and version control. Presto. “static” SQA testing is a prerequisite to the verification process that encompasses tests that are performed without running the code. Subsequent new code revisions and enhancements have been reviewed by the member of the FossumBrannon team who did not write the new code. and HP workstations or clusters running Linux and/or Windows) in multiple host codes (Alegra. “GeoModel version 040526” was released in 2004 on May 26. The GeoModel has been compiled on multiple platforms (Sun. Another goal was to confirm that the equations being solved in the code were indeed the equations documented in this report. and two independent drivers) using multiple commercial compilers (gnu.Software “confidence building” activities In the lexicon of Ref. A new routine (geochk) was recently added to the GeoModel that terminates calculations if the userinput falls outside allowable ranges. 81 .barf) and terminates calculations whenever unacceptable solution quality is detected. and gives model installation instructions that describe how SQA responsibilities are partitioned between the model and the host code.0000001). Each GeoModel release is identified by a six digit code‡ that prints to the screen at run time. thus allowing the GeoModel developers to maintain a single master version.. The debugging file may be emailed * Here. and Compaq visual FORTRAN). The GeoModel has been designed such that it can be implemented in multiple host codes without altering the source code. SQA includes appraising quality of both user input and runtime arguments sent from the host code. which serve as software requirement specifications that must be obeyed by the host code developers. Pronto. pgf77. The GeoModel includes a “problemresolution” feature that generates a debugging file (geo. † This can happen because of slight roundoff errors. the backstress must be deviatoric) as well as constraints on the input parameters (e.
3. which runs either from a command line or from an Excel frontend (see Fig. biaxial plane stress. Fig.0 at the onset of yielding and should dwell at 1.8b confirmed that increasing the GeoModel’s characteristic response time (see Eq.1.7. A small number specific examples will be given later. who can import it into their driver to resolve the problem quickly.). • Model domain certification: The “geochk” feature for checking quality of user inputs was tested by confirming that the GeoModel would abort when intentionally sent invalid inputs. two types of shear loading. 5. However. and (3) write a debugging file that can be emailed to model developers to determine whether the problem was caused by bad user input.5 and then asymptote to 1. timestep control. The driver regression problems (15 problems to date) are all rerun and inspected for undesirable changes whenever any change is made to the GeoModel. uniaxial strain.3 and Figure 7..1 and verified in Fig. A detailed documentation of all GeoModel SQA would require a second report itself.1). For example. • Trend testing: Engineering judgement was used to ensure that solutions vary as expected when parameters change.22) would indeed produce the analytically predicted increase in the apparent strength of the material. 7. These tests were quite valuable because they allowed correction of a serious bug in an earlier version of the GeoModel where the response trend upon load reversal was clearly flawed. These drivers have been used to assemble a suite of regression tests (hydrostatic loading. not the GeoModel. several of which are simple enough to admit analytical solutions for verification purposes. 82 . or a bug in the internal coding. Socalled “dynamic” SQA refers to tests performed by actually running GeoModel. code compilation and runtime overhead.g. For example. (2) terminate calculations. after which it should drop instantaneously to 0. so the following list should be regarded as simply an overview. * Standalone testing obviates many constraints and sources of nonconstitutive errors in production finiteelement simulations (artificial viscosity. the “geobarf” problemresolution feature can often indirectly trap invalid forcing functions (caused. Some of the dynamic SQA activities to date are summarized here. massively large elongations or pressures). is now also deployed in the WISDM materials information database [21]. by mesh entanglement). and numerous mixed loadunload problems using a variety of input parameters).1 depicted a simple trend test in which analytical (exact) arguments could be used to prove that the “SHEAR” should equal 1.barf) feature has been verified to (1) detect “Garbage Out” predictions such as negative plastic work. uniaxial stress. bad arguments passed from the host code. the responsibility for sending valid forcing functions to the GeoModel remains the onus of the host code. Trend and robustness testing also demands that the GeoModel must predict qualitatively reasonable trends when subjected to deformations that exceed what is expected in applications (e. one problem in our driver regression suite verifies that load/unload curves do indeed asymptote toward each other under hydrostatic compression as sketched in Fig. Figs.* Considerable dynamic testing was performed using our research (nonproduction) model driver that allows visualizing (and algebraically processing) the output within Mathematica [48].0 until the loading direction changes. Similarly. Many other trend tests such as these have been conducted. but (for lack of space) will not be described in detail in this report. • Runtime monitoring of the solution quality: The problemresolution (geo. allowing the predictions of the GeoModel to be compared directly against experimental calibration data. 6. 5. boundarycondition errors.0. • Driver regression testing: Two standalone drivers are available for exercising the GeoModel in a homogenous deformation field. The other driver [36]. hourglassing. for example. etc.Software “confidence building” activities to the GeoModel team. As explained below.
28E18 T3 (ndim) 0 SUBX (ndim) 0 G2 (1/Pa) 0 P0 (Pa) 3.0E+08 1.22E+08 CUTI1 (Pa) 1.0E+08 7.0% 4.00E+00 0 1.86E+09 A3 (Pa) 8.0% 2.00E+00 B2 (Pa) 4.1E+09 1.73E10 HC (Pa) 9.1.00E+00 0 1.00E+00 B4 (ndim) 0.22E10 T2 (1/s) 0 RKPF (ndim) 0. Screen shot of the MS Excel interface for the GeoModel materialmodel driver (MMD).0E+08 6.00E+10 J3TYPE (ndim) 3 G0 (Pa) 9.0E+08 4. etc.00E+09 0.00E10 CUTPS (Pa) 3.72 T6 (s) 0.25E+10 RKS (Pa/m) 2.00E+06 A4PF (rad) 1.exe ********************************************************************** Material model: Index 11 Name (optional) Fossum Geomodel ********************************************************************** Material model input parameters: B0 (Pa) 1.00E10 Axial Stress (Pa) Hydrostatic Compression (StressDriven) 1.00E+00 0 0 0 0 0 0 0 0 0 stress or strain (or mixed) loading *Load type parameter for each component.021 A2 (1/Pa) 2.0E+08 2.00E+13 CR (ndim) 6 T5 (Pa) 0.00E+11 A2PF (1/Pa) 2.0% 3.0% 1.30E+10 RJS (m) 0.40E02 T4 (1/s) 0.11E+08 RKN (Pa/m) 1.0E+08 8.43E+08 RN (Pa) 5. contact algorithms.20E+10 A1 (Pa) 8.0E+00 0. This tool serves as a reliable platform for exercising the GeoModel under homogeneous deformations in a simplified host code architecture that is free from solution corruption caused by unneeded finiteelement code features such as artificial viscosity.00E+09 0.72 G4 (ndim) 0 P2 (1/Pa^2) 1.00E+00 CRPF (ndim) 6 Axial Strain (%) ********************************************************************** Load path specifications: leg (ndim) 0 1 2 time (s) 0 1 2 npts (ndim) 0 100 100 nprt (ndim) 0 2 2 ltype* (ndim) 111111 444222 444222 c11 c22 c33 c12 c13 c23 0 1.0E+08 5. cij: 1 = strain rate 2 = strain 3 = stress rate 4 = stress ( 1/s ) ( 1 ) ( Pa/s) ( Pa ) ********************************************************************** Figure 7. 83 .0% 5.73E10 G1 (ndim) 0 A4 (rad) 1.14E+08 T1 (s) 0.0E+08 3.0E+08 0. wave motion.00E+00 P3 (1) 8.00E+00 B1 (Pa) 4.Software “confidence building” activities Problem description: Fossum Geomodel Salem Limestone ********************************************************************** Path to Fortran MMD executable: C:\home\data\mdrive\rb\mmd0402c\Debug\mmd0402c.00E+09 0.00E+06 T7 (Pa) 0.0E+09 9.00E+00 B3 (Pa) 1.0% G3 (Pa) 0 P1 (1/Pa) 1.00E+12 RK (ndim) 0.
In a few cases. However.g.). the disquieting frequency of bugs originating within the host code (not the constitutive model) reiterates the importance of specialized constitutive model drivers in constitutive model verification. These exercises were part of a larger Verification and Validation (V&V) effort to increase confidence in prediction of lowyield nuclear damage of underground (tunnel) facilities in jointed (in situ) rock mass. however. * Testing a numerical model in multiple host codes has been vital to our SQA process.g. an undesired feature in a calculation persists even when the GeoModel is run using simplified parameters. JAS3D [5].g. then at least one of them is not solving the equations properly (or is not solving the same equations). the GeoModel has been implemented in five finiteelement codes: (ALEGRA [8. timestep control. often compared a host code’s standard elasticity model with the GeoModel run in an elastic mode. A disturbing number of discrepancies (e. wave arrival times differing by as much as 10% or peak stresses differing by almost an order of magnitude) have been identified by comparing predictions for the same problem simulated by different finiteelement codes. Depending on how the model parameters are set. If. for example.Software “confidence building” activities • Symmetry testing: Consistent predictions were verified for identical loading applied in different directions (e. legacy PRONTO3D [45]). In one instance where a discrepancy was traced to the GeoModel. exact apparent strengthening in high rate loading. until a level is reached that is deemed sufficient to model precision field tests.. • Extensive comparison testing:* To date. The ability of the GeoModel to reduce to simpler models (e. The problems are designed to increase in complexity.. EPIC [24].). This work will be described in further detail later in this chapter. In all but a few cases.49]. † Incidentally. PRESTO [26]. uniaxial strain in the 1direction compared with the 2direction). the GeoModel can be idealized to a form for which some problems admit exact solutions. accurate tracking of a specified porosity crush curve... 84 . two codes predict different answers. • Comparison with exact analytical solutions: Analytical solutions are not available to verify all of the GeoModel’s features acting simultaneously. This class of problems was recently studied in a verification and benchmarking activity sponsored by the Defense Threat Reduction Agency (DTRA) through its Advanced Concepts Technology Demonstration (ACTD) Project. it was attributed to failure of the code owner to follow the GeoModel installation instructions (page 75). Constitutive models should not use finiteelement codes as their primary verification platform. If two codes agree. not † the GeoModel developers. comparison testing did indeed reveal GeoModel bugs that have since been corrected. each GeoModel feature has been individually verified to ensure that the promised quantitative material response is delivered (e. hourglassing. nonhardening elastoplasticity) has often proved invaluable in determining if a simulation problem originates from the GeoModel or elsewhere in the host code. Resolving such discrepancies has time and again expedited bug identification and resolution. handling of boundary conditions. etc.g. for example. etc. We have. a comparison test can be performed using the existing (presumably better verified) independent version of that model within the finiteelement code. Resolving such discrepancies is the model verification responsibility of the host code owners.g. artificial viscosity. Consistency has also been tested for stress paths that intersect the yield surface at symmetrically equivalent points and for trial stress rates normal to the yield surface (to verify that the stress rate had no tangential component). When. by invoking additional physics in the material models. these discrepancies have been traced to hostcodelevel (not constitutivelevel) errors (e. then no conclusion may be drawn. One example is that of a linear MohrCoulomb material with an associative or a nonassociative flow rule in homogeneous loading.
The GeoModel. ‡ At this stage in our ongoing verification process. An independent research effort [16] is nearing completion in which an implicit integration scheme has been developed for a simplified version of the GeoModel. Subcycling in the GeoModel has been massively improved over earlier versions. Preliminary tests (see page 86) indicate that each subcycle within the GeoModel is firstorder convergent. so MMS might indeed prove useful in the ongoing (still incomplete) GeoModel verification process. but to the wrong result. solving the inverse problem might be tractable. The GeoModel’s governing equations change upon reaching yield. Solving the inverse problem is quite difficult because the GeoModel’s differential operators are “branched” (one set is used during elastic deformation. we are using socalled “viewgraph” assessment of accuracy in which solutions are compared visually by plotting them together.g. To date. does not use a return method — it explicitly integrates the equations. • orderofconvergence:* Time steps for complicated plasticity models often must be much smaller than the time step used by the host (finiteelement) code. this technique has not been used in the GeoModel verification process. if the return direction is incorrect in a classical predictorcorrector plasticity scheme. Moreover. but the total computational overhead sometimes exceed that of a wellwritten subcycling algorithm. by using simplified model input parameters. * The term “orderofconvergence” is preferred over “orderofaccuracy” because a converged solution is never necessarily a correct solution. For example. but this improvement in allowable hostcode step size is essential in fieldscale finiteelement simulations). higherorder integration algorithms are notoriously difficult to maintain when the governing equations themselves are being revised during parallel development of the physical theory. by the way. which implies that the constitutive model must break the step into elastic and plastic parts. † Of course. but (recall the last footnote on page 79) we have not yet performed sufficient tests to consider this claim fully verified for a broad variety of load paths. then the algorithm will converge. The numerical model is then sent this analytically determined input function to verify that the original (preselected) response function is recovered. This easy assessment method simply bounds the discrepancy between two seemingly overlaying plots to be less than differences perceptible to the human eye relative to the size of the graph. and the internal state variable evolution equations themselves are coupled to the location of the stress on the current yield surface).Software “confidence building” activities • Consideration of the method of manufactured solutions (MMS): The MMS method of SQA entails first solving an inverse problem in which simple (e. For the GeoModel. using subcycling to assist with the change of governing equations upon yielding. while another is used during plastic loading. However.† Subcycling complicates the meaning of a convergence study performed at the host code level. further subcycling within the constitutive model is required to avoid “drift” of internal state variables into nonphysical or inconsistent domains. the GeoModel still runs ~2000 subcycle steps. quadratically varying) analytical time histories for the response functions are substituted into the governing differential operators to obtain (probably with the assistance of a symbolic mathematics program such as Mathematica) an expression for the input functions that would produce the prespecified response function.. but verification of its orderofconvergence is not yet completed. 85 . higherorder integration is also an option. Moreover. using MMS would require preselecting an analytically simple stress history to determine a strain history to use as input in the numerical simulation. and this transition typically occurs somewhere in the middle of the host code’s time step. Problems that formerly took 2000 code steps to converge to an acceptable accuracy‡ can now be run to the same pointwise accuracy in only 10 apparent hostcode steps (internally.
The dashed blue line (shown for reference) has a slope exactly equal to –1.Software “confidence building” activities Preliminary convergence testing.1) The “pointwise discrepancy” is defined n ∑ [ F ( xk ) – f ( xk ) ] 2 k = 1 1/2 (7. Black. Rate of convergence.3 shows that the basic algorithm without subcycling is firstorder accurate. This highly desirable behavior can be of paramount importance when the model is run in finiteelement codes that take large steps. stress as a function of strain) corresponding to N evenly spaced time steps.2 shows that subcycling makes the solution nearly pointwise converged ( discrepancy ≈ 2 –6 ) even for inordinately coarse calculations of 1 to 3 time steps. Figure 7. 86 . Figure 7. F ( x ) .2. 24 time steps log 2 ( discrepancy ) integrated without subcycling integrated with subcycling 25 time steps log 2 ( number of cycles ) pointwise without subcycling pointwise with subcycling Figure 7.g. In both cases. Let f ( x ) denote a coarser solution corresponding to n < N time steps. Green is with subcycling. Subcycling test for hydrostatic loading. corresponds to 2 15 time steps. the continuous curves F ( x ) and f ( x ) are here regarded as piecewise linear interpolations between discrete function values at the timesteps.3. The “integrated discrepancy” is defined N [ F ( x K ) – f ( x K ) ] 2 ∑ K = 1 1/2 2 time steps 22 time steps (7. Both curves are normalized by the peak value of F ( x ) . The ordinate is the first stress invariant I 1 and the abscissa is volumetric strain ε v .2) 23 time steps Figure 7. Red is without subcycling. Let F ( x ) denote a fineresolution solution curve (e.
p } were selected such that no compaction occurred over the stress { R.26 26° 14° 15.000000 0. GeoModel was then fitted to these data pairs to determine the limitstate parameters ψ. R 0 1 2 3 range specified for the load paths. As summarized below.7.005000 0.Software “confidence building” activities Verification: singleelement problems (regression suite) A singleelement problem is one for which the stress and strain fields do not vary in space. Likewise. the linear MohrCoulomb parameters were used to create simulated data pairs for the limit state and plastic potential functions. Following the instructions in Appendix A. a 1 } . Results are summarized in Figs. a 1. p . each separately solved using two different GeoModel parameter sets (associative and nonassociative linear MohrCoulomb — see Appendix B.002060 epsz epsy=0 sigx=0 Point 1 2 3 4 epsz 0.1: MohrCoulomb Parameters for singleelement verification testing Parameter (Material Properties) Young’s Modulus Poisson’s Ratio Friction angle Dilation angle Cohesion Dry bulk density Symbol E ν φ ω So ρ Associative 31. Point 1 2 3 epsz TXC0 sigx=sigy epsz 0 0. a 3. a 2. p .26 29° 29° Nonassociative 31.34 MPa/m3 15. The compaction parameters PF.000000 0.005000 0. page B22).7 MPa 2.0 GPa 0.000310 0.000000 2 0. a 4 and nonassociative PF PF material parameters { ψ PF. To determine corresponding sets of material parameters for the GeoModel. this section presents four different singleelement load paths. hardening and rate dependence were disabled.003274 TXC20 sigx=sigy 0 20 20 20 RTX100 sigx=sigy 0 100 100 100 • Simple load paths clearly demonstrate basic model response • Run for both full and partial associativity • Triaxial Compression — Activates compression meridian — Same load path as a real lab test • Reduced Triaxial Extension — Activates extension meridian — Same load path as a real lab test • Plane Strain — Activates meridians at Lode angle(s) (triaxiality) between compression and extension sigy=sigx sigx Point 1 2 3 4 epsz 0.34 MPa/m3 87 . a 2 .000000 0 0.002295 PSTRN Point epsz 1 0.010000 3 0.005000 0 0.001548 0. 7.7 MPa 2.4 through 7.0 GPa 0. p .008391 Table 7.
0 0 0.5 εy%(%) ε.1 0.4 0.4 0.1 Nonassociative 0. (a) 0 0.1 0.5 Associative 60 60 50 (b) 60 60 Nonassociative 50 σσ .Software “confidence building” activities SINGLE ELEMENT VERIFICATION RESULTS UNCONFINED COMPRESSION Associative 0 0.5 0.0 0. strain.% εzε (%) z 0.2 0.3 0. 88 .0 εz (%) ε z. Exact (black) solutions compared with ALEGRA GeoModel simulation (pink).4.0 0.1 0.2 0. Figure 7.5 –0.15 0. y 0.3 0. 0.% y 0. axial strain (b) Stress vs. 0 0. % 0.1 0. MPa 30 z 40 40 z 20 10 00 10 0.3 0.35 0.4 0.6 0.3 0. % 0. % 0.2 εz (%) ε z.4 0.00 0.4 0. (a) lateral strain vs. Unconfined compression.1 εyε(%) .3 0.(MPa) z MPa 30 20 10 00 10 0 0.3 –0.2 0.5 0.5 σ (MPa) σz.2 εz (%) ε z.0 .5 0.05 0.5 0.5 0.2 0.5 0.25 0.3 0.
0 εz (%) ε z.2 . % 0. axial strain (b) Stress vs.3 z 0.5 0.1 0. (MPa) 80 z z 60 40 20 00 0 0.Software “confidence building” activities SINGLE ELEMENT VERIFICATION RESULTS 20 MPa TRIAXIAL COMPRESSION Associative 0 0.5 ε .0 0.3 0. % 0. (a) lateral strain vs.2 0. strain. % 0.0 0.5 0.5 Associative 120 120 (b) Nonassociative 120 120 100 100 σzMPa σ .% 0.4 0.1 0.35 0.5 0. 0.(MPa) 80 60 40 20 0.2 0.5.05 0 0.0 0. 89 .2 0.4 0.05 0 0.2 –0. 20 MPa triaxial compression. 00 0 0.05 (a) 0 0. 0.% εzε (%) 0.25 0.3 –0. Figure 7.3 0.3 0.% εy (%) y 0.1 εy ε(%) .15 0.1 0.1 Nonassociative 0.0 εz (%) ε z.0 σ MPa σz .1 0.5 0.4 0.3 0.5 0.5 0.2 0. Exact (black) solutions compared with ALEGRA GeoModel simulation (pink).2 εz (%) ε z.05 y 0.4 0.15 0.
axial strain (b) Stress vs. 90 .1 0. % 0.5 0.2 0. MPa σzσ(MPa) 80 60 40 20 0 0 0. (%) y 0.4 0.3 0. 0.2 0.0 0.3 εyε .1 0 0.6.1 0.35 0.0 0.2 –0.25 y 0.3 0. Exact (black) solutions compared with ALEGRA GeoModel simulation (pink).6 0.4 0.25 0.2 σz.Software “confidence building” activities SINGLE ELEMENT VERIFICATION RESULTS REDUCED TRIAXIAL EXTENSION Associative 0.4 0.0 –0.0 0.1 0 0.6 0.5 –0.1 0 0.3 εz (%) ε z.35 0.4 0.2 0. (a) lateral strain vs. (a) 0. 0.4 0.1 0.15 0.2 0. Reduced triaxial extension.5 0.0 0.05 0 0.3 εz (%) ε z.2 0.3 0. MPa σ (MPa) z 80 60 z 40 20 00 0.(%) % 0.1 0 0. 0.1 0. % 0.5 ε z.6 0.4 Nonassociative 0.1 0.5 –0.05 0 0. strain.15 0.2 Figure 7.3 0.5 εz (%) ε z.5 0.1 0.2 0. % ε % εy .6 0.2 Associative 120 120 (b) 120 120 Nonassociative 100 100 .0 0.5 0. % εz (%) 0.
Software “confidence building” activities SINGLE ELEMENT VERIFICATION RESULTS PLANE STRAIN Associative 00 0 0.4 3 –3 0 1 –1.4 0.8 10 1 0 0 0. 0.8 1 1.6 0.% εz (%) z 0.2 0.8 1 Figure 7.2 0. 91 . stressMPa (MPa) 40 30 sigz.2 2.8 0. (a) 0 0. analytic sigy. Exact (black) solutions compared with ALEGRA GeoModel simulation (pink).4 . % 1 0.5 % εx.2 Nonassociative 00 0.4 ε.6 1.6 0.6 0. numerical sigy.4 10 ε.8 0.8 1 1. analytic sigy.2 0.4 0.6 0. stress MPa (MPa) 40 30 sigz. (a) lateral strain vs.2 0.5 0. axial strain (b) Stress vs.5 1.6 0.7.8 2 1 1.% εxε (%) x 1. analytic sigz. Plane Strain. numerical sigy. numerical 20 σy 20 σy 10 10 0 0 00 0 0 0. strain. analytic sigz.6 0 εz (%) ε z.x ε (%) εz (%) ε z. εz%(%) z 0. % 1 Associative 60 60 (b) Nonassociative 60 60 σz σz 50 50 Stress. numerical Stress.
† The only way to ensure a precisely homogeneous deformation in a dynamics finiteelement code is to bypass solution of the momentum equation. 7·10 8 pressure 0 0.8a depicts results from a hydrostatic loading simulation in which plastic volumetric strain (EQPV) is plotted against pressure (–I1/3). given the plot size.5.4 and 7. ˜ ˜ 92 .8a) is verified to remain zero until the pressure reaches the crush curve.02 0.” where the error is nebulously bounded by what can be perceived visually. we believe boundary condition errors in the host code (not the GeoModel) are responsible for the solution errors. thereby causing predictions to deviate slightly from analytical solutions. 4. 7. 7. and we anticipate rerunning these simulations in a true constitutive model driver to verify this claim. * In Fig. However. after which * Precisely quantifying verification error lags far behind other more important constitutive SQA priorities.67) determined independently by user specified values p of the GeoModel parameters p 1. making the momentum equation trivial. Fig. Though not yet proved conclusively. The thick yellow curve in Fig. Other singleelement tests. moderate error is apparent in Figs 7. then it will have slight errors in the strain rate that it sends to the GeoModel. cation.08 0. Rate dependence TXE/TXC ratio verification for triaxial load/unload.5 arises not from constitutive model errors. homogeneous loading requires a body force b identically equal to particle acceleration a . the GeoModel takes strain rate as input and returns updated stress as output. which involve two stress boundary conditions (the lateral stresses).5 dynamic 6·10 8 5·10 8 plastic volumetric strain 4·10 8 quasistatic 1 2 3 3·10 8 0.8b.06 p3 0. Such errors are not uncommon when dynamic finiteelement codes are used to attempt to simulate homogeneous deforma† tions.8a is the crush curve (Eq.08 σ 11 0. The fully straincontrolled problem (Fig. so we will be satisfied here and throughout this chapter with assessing agreement between computed and analytical results via a socalled “viewgraph metric.04 0.02 –p3 1 1 2 pressure/p1 3 4 5 6 7 Figure 7. predictions of the GeoModel overlay the analytical results. More correctly. The predicted volumetric strain ε v (black line in Fig 7.Software “confidence building” activities Discussion.1 0. 7. To reiterate. If the host finiteelement code handles stress boundary conditions improperly.4 and 7.7) exhibits no significant solution discrepancy.5 total volumetric strain 1 0.8a. but instead from host code errors in the handling of stress boundary conditions. we contend that the slight discrepancy apparent in Figs 7. 7. Like most constitutive models. p 3 . Whenever the source code is changed.04 0. Crush curve and hydrostatic loading verifi Figure 7. we perform approximately 20 singlefeature singleelement verification checks for each of our regression tests performed under loading that is simple enough to admit analytical solutions.7.06 p1 2·10 8 3 1·10 8 2 1 ε 11 0. p 2.
t The results in Fig. the very slight discrepancies are hypothesized to result from different strain definitions used in the GeoModel and analytical solution (or possibly host code traction boundary condition issues similar to those discussed earlier). p(t) +a Tunnel radius Far field radius a b Geometry 1m 10 m 1m 10 m r Loading p(t) 150 MPa Internal pressure Far field pressure pa pb 0 0150 MPa 0 0150 MPa time. The material parameters are the same as those used in the previous verification problems. A DTRA contractor provided the analytical solutions. involves a circular tunnel in a MohrCoulomb material loaded in a plane strain configuration. 7. compression. 7. In that problem.* Verification: Hendren & Ayier pressurized cylinder The problem depicted below. resulting in a tensile strength half as large as the compressive strength. Fig. * The TXE/TXC strength ratio applies to TXE and TXC states at the same pressure. For nonhydrostatic loading.8b simultaneously verifies the GeoModel’s ability to predict an apparent increase in strength under dynamic loading and its ability to predict different strengths in triaxial extension vs. pressure dependence of yield was suppressed and the TXE/TXC strength ratio was set to 1/2. When pressure dependence of yield is allowed.9 show that the GeoModel solutions agree with analytical results. we have verified (trendtest) that shearenhanced pore collapse causes inelasticity to commence prior to the pressure reaching the crush curve.8a) unloads correctly to the userspecified peak strain parameter p 3 . solved for subcases of associative and nonassociative flow. the TXE peak in a verification test like this will not and should not be 1/2 the magnitude of the TXC peak. The stressstrain curve (inset in 7. Though not confirmed conclusively.Software “confidence building” activities it drops along the crush curve as it should. 93 .
5 2.0 0.5000E+02 Pressure (MPa) Pressure (MPa) (b) 300 300 250 250 Stress (MPa) 200 Stress (MPa) Stress (Mpa) 1 2 200 Stress (Mpa) 150 150 100 100 50 50 0 r/a 3 r/a 4 5 0 1 2 r/a r/a 3 4 5 (c) 6 3 3 2 Strain (%) Strain (%) Strain (%) 1. normalized range.0 1.2500E+02 1.2500E+02 1.0 1.Software “confidence building” activities HENDRON & AYIER VERIFICATION RESULTS Associative (a) 4 Nonassociative 2. Hendron & Ayier verification problem.5 2.5 Closure (%) 1.5000E+0 0.5000E+01 5.5 0 1 Strain (%) 0 3 1 6 2 9 3 12 r/a r/a 2.9. normalized range.5 r/a r/a 2. (c) strain components vs.5 4 1.5000E+01 5. farfield pressure (b) Stress components vs.0000E+01 7.0000E+00 2. (a) Tunnel closure vs.0000E+02 1. Exact (dash) solutions compared with ALEGRA GeoModel simulation (solid).0000E+01 7.0 2.5000E+01 Pressure (MPa) 1.0000E+00 Closure (%) Closure (%) 1.0 2.5000E+01 Pressure (MPa) 1.5 Figure 7. 94 .0000E+02 1.0 0.0 3 Closure (%) 2 1 0 0.
or some other aspect of the finiteelement model such as artificial viscosity (both codes’ solutions were strongly affected by artificial viscosity — default settings for artificial viscosity were insufficient to reproduce the analytical results). Figure 7. While the moderate solution error might be attributable to the analytical solution’s presumption of small strains.10. Finiteelement vs.11 demonstrates that the second code (JAS3D) was capable of reproducing the analytical solution well enough to suggest that the GeoModel is performing correctly. artificial viscosity.11. JAS3DExplicit Dynamic FEM + GeoModel elastic AldridgeAnalytical/Numerical* Figure 7. As emphasized earlier. further study (by the code owners. 7. Fig.Software “confidence building” activities Elastic freefield wave form (finiteelement verification) This section describes a fieldscale test for verifying installation of the GeoModel in host finiteelement codes. roller As indicated in Fig. and other aspects of the host code’s method of solving the partial differential equation that governs momentum balance.10. The GeoModel’s yield features were disabled to allow predictions for the velocity at the outer radius (470m) to be compared with an analytical elasticity solution. Spherical cavity geometry. where code predictions are compared with data. One code was unable to reproduce the correct response roller because of bugs in roller boundary conditions. a time varying velocity (identical to the one later discussed on page 107) was applied at the boundary of a spherical cavity (radius 204m). not the GeoModel developers) is warranted to determine if the solution errors result from underintegration. 95 . verifying a constitutive model is distinct from verifying its implementation within a finiteelement code. This elasticity verification problem is revisited and generalized in the plasticity validation test on page 107. * Aldridge’s analytical solution [1] is expressed in terms of integrals that are evaluated numerically in the frequency domain using independent software provided by Aldridge. analytical/numerical elastic wave velocity at 470 meters from a velocity spherical cavity source at 204 meters. 7. After a constitutive model becomes one of many components within a much larger finiteelement model.* Implementations of the GeoModel in two finiteelement codes were tested. the potential sources of solution error expand to now include boundary conditions.
0 Radians ce J 2 (MPa) Peak Stress Model 200 MPa 100 MPa 50 MPa 25 MPa 10 MPa 5 MPa 0 MPa 4 Frozen Soil to J 2 (MPa) 200 J21/2 (MPa) M od el ta 150 100 0 0 200 G da ta 50 e e od oM 200 da 0 0 400 I 1 (MPa) I (MPa) 1 600 800 1000 1200 1200 1400 0 Ge o 0 data l ta data da fi t data 80 to it lim urfac it s lim e fi t Figure 8. Parameterization (calibration) Appendix A describes how to characterize a material for the GeoModel.83 MPa A2 = 6.1 MPa 4. 1.1 MPa 3.3).293E4 1/MPa A3 = 481. Meridional limit curves for some materials already parameterized to the GeoModel. as summarized in Fig. 450 400 400 350 J J 2 (MPa) (MPa) 300 250 200 150 100 50 0 0 dat a 100 50 0 0 500 G 500 0 0 500 I 1 I (MPa) (MPa) 1 1000 1500 2000 2000 2500 da t a G M eo 0 0 o da t a M eo da t a 23 MPa concrete ce rfa su it m li to t fi el od da t a 23 MPa Concrete 400 400 350 Salem Limestone Salem Limestone Data Model 0 MPa 10 MPa 20 MPa 40 MPa 50 MPa 75 MPa 100 MPa 200 MPa 300 MPa 400 MPa 300 JJ2 (MPa) (MPa) 250 200 150 l de f it dat 1000 1 1/2 2 to 2 1/2 a 1500 lim it s fa ur ce Data Model 0 MPa 20 MPa 50 MPa 100 MPa 200 MPa 400 MPa 500 I 1 I (MPa) (MPa) 2000 2000 2500 350 300 300 J J 2 (MPa) (MPa) 250 J 2 (MPa) J21/2 (MPa) 200 1/2 od el 150 Conventional Strength Concrete Conventional Strength Concrete e f ac ur it s lim Model to Data it 0 MPa f 40 40 35 30 25 20 15 10 5 Antelope Tuff Antelope Tuff da t a 100 d a ta oM Ge 50 0 0 500 0 0 500 I 1 I (MPa) (MPa) 1 1000 1500 2000 2000 2500 0 0 50 G 0 0 da 20 MPa 50 MPa 100 MPa 200 MPa 400 MPa d ata M eo ta o da l de ta fit to lim u it s rfa ce Data Model 0 MPa 5 MPa 10 MPa 25 MPa 70 MPa 2 50 100 I 1 (MPa) I1 (MPa) 150 150 200 30 30 Model Prediction Versus LimitState Data from Triaxial Compression Tests Conducted on AERA Diatomite Opal CT Diatomite ts fa ur 25 ce 2000 2000 1800 1600 Climax Stock Granodiorite Climax Stock Granodorite J 2 (MPa) 20 J2 1/2 (MPa) ta 15 ta 200 0 0 40 20 0 0 20 300 300 Model Versus Data from Triaxial Compression Tests Conducted on Sidewinder Tuff I 1I (MPa) (MPa) 40 1 60 80 100 100 120 0 0 2000 0 0 dat a 5 da Ge o oM Ge 600 400 da t a 10 Mo el od da de l J2^ 800 dat fit to lim J 2 (MPa) (MPa) i 1400 1200 1000 Data Model 2. I 1 (MPa) 96 . GeoModel parameters for these materials are in Appendix B.1 MPa 13.Parameterization (calibration) 8.1.08 MPa A4 = 0.1 and 3. several materials have already been fit. 8.8 MPa a 1/2 f it to l t su im i e rfac Model Data 2216 MPa 1527 MPa 1161 MPa 439 MPa 2000 I 1 I (MPa) (MPa) 4000 6000 1 8000 10000 10000 12000 Sidewinder Tuff rfa su 250 A1 = 496.1 (compare these with Figs. Using these procedures.8 MPa 13.
4.080 0.100 Volume Strain Figure 8.040 0.2. This figure shows the GeoModel fit to concrete data [46] from the unload portion of a hydrostatic compression test from 200 MPa.Parameterization (calibration) Nonlinear elasticity Figures 8. Nonlinear elasticity in hydrostatic loading. 400 Principal Stress Difference (MPa) 350 300 250 200 150 100 50 0 0.10 and 4.195 0.21 0. 250 Hydrostatic Stress (MPa) 200 150 100 50 0 0.000 Data Model 0. 97 .205 0. Parameters were assigned using the leastsquares model calibration tools described in Appendix A.19 0. thereby making the slope equal to twice the shear modulus.9) to reproduce nonlinear elasticity data. This figure shows the GeoModel fit to concrete data [46] from the unload portion of a triaxial compression test conducted at a confining pressure of 200 MPa.215 Data Model Principal Strain Difference Figure 8.2 and 8.2 0.3. the principal stress difference is plotted against the principal strain difference.3 demonstrate the ability of the GeoModel’s nonlinear elasticity fitting functions (Eqs. Nonlinear elasticity in shear.060 0. the slope equals the bulk modulus.020 0. Here. Here.
98 . However. Elasticplastic coupling: deformationinduced changes in elastic moduli (Salem Limestone). which permit the shear and bulk moduli to vary with equivalent plastic strain (determined from data by the residual strain upon unloading to a zero stress. the tangent shear modulus. (4. 600 500 HYDROSTATIC LOADING Mean Normal Stress (MPa) Data GeoModel 400 300 200 100 0 0. Eqs. volumetric plastic strain). Fig..06 0. (a) the tangent bulk modulus can change in response to changes in porosity (i. 8. especially at low shear stresses.34).02 (a) 0 0. can change in response to plastic deformation.4. elastic moduli are unaffected by inelastic deformation.Parameterization (calibration) Elasticplastic coupling For many materials.1 Volume Strain TRIAXIAL LOAD DATA CONVERTED TO SHEAR DATA (b) Figure 8. For the GeoModel to be considered a good fit to data. Modeling this effect requires using the enhanced moduli fitting functions.08 0. the simulated unloading curves merely need to be parallel to experimental unloading curves (not necessarily overlapping unless the data and simulation unload from the same strain).33) and (4.02 0.4 shows data for a material whose elastic properties are affected by inelasticity.04 0. as explained in Appendix A). (b) Likewise.e.
A.Parameterization (calibration) Triaxial Compression Triaxial compression (TXC) and triaxial extension (TXE) testing is integral to parameterization of the GeoModel. Recall that triaxial testing normally begins with a hydrostatic “loadup” phase. giving an I 1 value of 1200 MPa. as described in Fig. J 2 MPa 450 400 350 300 250 200 150 100 50 0 0 500 1000 1500 2000 A = 1255.5. For example.5(a) on page A6 shows raw data from a a suite of TXC tests. [46]. 8.3. deformation is initially nonlinearly elastic until the virgin yield surface is reached. Shear failure limit curve compared with concrete data. the target hydrostatic stress state for a given experiment (which marks the transition from the hydrostatic leg to the triaxial leg) was P = 400 MPa.0 Rad Data Model − I1 MPa Figure 8.20 MPa θ = 0. During the hydrostatic leg. (4. Appendix A describes the leastsquares fitting procedures in more detail. Continuing to push the hydrostatic stress to higher levels results in inelastic pore collapse with associated hardening (expansion) of the yield surface. Data from a similar set of experiments for concrete. 3. the limit state (onset of softening) can be mapped out.6a (where J2 = 0 ). along with the GeoModel leastsquares fit of Eq. indicating how the peak values in each test provide one data point on the GeoModel limit surface Fig. Pressurevolume data taken during the hydrostatic leg may be used to deter99 .68 MPa B = 1. Figure 8.93E04 1/MPa C = 1248. at which time microscale stress concentrations caused by the presence of pores become too large to resist elastically. Fig.6.1 shows similar plots for other materials. By performing a series of tests. indicated by the horizontal red arrow Fig.5(b). Ref.5. In Fig. 8. 8. A.8) are shown in Fig.
illustrated with correspondence of the meridional plane to the stressstrain diagram. variations observed during hydrostatic loading from different tests can be used to quantify the material property variability. Shearenhanced dilatation corresponds to reaching a zero local yield slope. 8. the stress path follows a straight trajectory in the meridional plane with a slope given by d ( J2 ) 1. (a) (b) Figure 8. p 3 } by following instructions in Appendix A. the triaxial leg (angled red arrow in Fig. p 1. p 2. After the hydrostatic leg. Progression of the hardening yield surface (family of blue lines) under a triaxial compression test (red path). Because multiple triaxial experiments must be performed to fully characterize a geological material.= d( I 1) 3 (8.6. The straight red line segments shown show this twostage stress trajectory (hydrostatic loading followed by triaxial loading) in the meridional plot of J 2 (which is proportional to the effective shear) versus I 1 .1) 100 .Parameterization (calibration) mine the GeoModel parameters { p 0.6a) commences by increasing the axial load on the specimen while holding the lateral stress constant. As explained on page 16.
. triaxial loading is used to determine parameters in the GeoModel that govern yield surface evolution and cap curvature. Crack kinking (in the direction of Mode I loading) further contributes to the dilatation associated with crack opening. However. Consequently. the outward normal to the yield surface during this early part of the triaxial phase has both a deviatoric component and a compressive isotropic component. the yield surfaces continues to harden outward even more. now further assisted by the presence of a nonzero stress deviator.6. the local slope of the yield surface changes sign). As the axial stress is increased during the triaxial leg. Moreover. 8. The dark regions in the micrograph [12] are pores (which collapse under sufficient pressure). DiGiovanni & Fredrich [12] DiGiovanni et al. the inelastic volumetric strain will be compressive as well. inelastic volume increase under shear loading even if all principal stresses are compressive. Thus.. 2000 Figure 8. When the normal to the yield surface is compressive. fragments of broken material can become lodged in the crack face and cause crack opening by their rotation. 101 .7 and explained in detail in Appendix A. The cracks in inclusions produce “microrubble” that ultimately generates. the isotropic component of the yield surface normal changes direction towards the end of the triaxial leg (i. where the straight red load path and curved blue yield surface lines intersect).6 the slope of the yield surface is initially negative at the stress state (i. Shearenhanced dilatation under compression. as illustrated qualitatively in Fig.e.Parameterization (calibration) The transition from hydrostatic to triaxial loading is reflected by a pronounced change in slope in the stressstrain plot of Fig.e.7. In Fig. 8. The “critical state” at which the yield surface has a zero local slope on the load path marks the onset of shearenhanced dilatation. which means that the inelastic volumetric strain is dilatational (expanding) even though all stress components are compressive. The exaggerated schematic shows that crack faces must overcome surface incompatibilities — they cannot slip over each other without opening even if they are in compression. through rotation. 8.
This interpretation of shearenhanced dilatation is illustrated schematically and through SEM imaging in Fig. As seen. TieRods Kolsky (split Hopkinson) bar apparatus used to obtain data in Fig. the crack must open up (dilatate) both to overcome geometric incompatibilities in their surface roughness and to permit crack kinking. Bar Supports Air Gun Radial Confining Vessel Axial Confining Vessel Figure 8.9b are used to assign values to { T 1. 8. As the surfaces of these cracks move relative to one another.Parameterization (calibration) Physically. 8.8] in which a sample is subjected to uniaxial compression at various strain rates. see Appendix A. 8.8.9 Figure 8. Parameterization: Rate dependence The GeoModel’s relaxation parameters ( T 1. an increase in inelastic volume during compression (which is quite commonly observed for brittle materials) is typically attributed to the growth of microcracks under shear. …. 102 .7. For a discussion of how the data in Fig. The peak stress in these experiments is used to assign values of the GeoModel relaxation parameters to properly correlate apparent strength with strain rate. T 7 } . …. the unconfined compressive strength increases with strain rate and is wellaccommodated by GeoModel theory. 8. T 7 ) may be determined through a series of labscale laterallyconfined Kolsky bar tests [Fig.9 shows results for a series of Kolsky (split Hopkinson) bar strain rate tests conducted on unconfined compression specimens of Salem Limestone [17].
006 0.1/s 1.004 0.000 0.0E+03 Strain Rate (s ) Figure 8. (a) Model predictions of stressstrain behavior at various strain rates.0E04 1.010 0. (b) Corresponding Kolsky (Hopkinson) bar unconfined compressive strength as a function of strain rate 103 .008 0.004 0. Suitability of the GeoModel overstress ratedependence model to match observed data.0E+01 1.0E01 1.0E05 1.0E03 1.Parameterization (calibration) 160 140 Radial 120 Volumetric Axial 100/s 10/s 100 Stress (MPa) 1/s 80 0.0E+00 1 1.010 0.008 0.002 0.0E+02 1.9.012 Strain 160 140 Uconfined Compressive Stength (MPa) 120 100 80 60 Data Model 40 20 0 1.002 0.006 0.E05/s 60 40 20 0 0.0E02 1.
the GeoModel’s rate dependence parameters are determined by using only the peak stress values in Kolsky bar experiments.e. as described on page 99.Building confidence in the physical theory 9. However. The goal is to assess the degree to which the integrated model (i. demonstrating (as we did in the previous chapter) that the GeoModel is capable of being parameterized from controlled laboratory data lends some credibility to the physical foundations of the model. truly validating a model after its parameters have been determined from standard laboratory data requires showing that the model can. points that were not used in calibration). Model validation therefore includes assessing the fitted model’s ability to interpolate well between other points in these stressstrain response curves (i. these problems were run using only the single GeoModel parameter set obtained from calibrating to other data for the material — no parameter adjustments were made to improve model agreement for these tests. include simulation of applications for which model predictions can be compared with structural response measurements. of course. In all cases. Therefore.g. 104 . the limit function is parameterized by using only the peak stress values. a better validation test should exercise the model in application domains in which multiple physical mechanisms are acting simultaneously. In addition to merely ensuring that all data in parameterization tests are well modeled. Thorough testing of any material constitutive model along with its implementation in a host code must. the model’s ability to match the other data points in those experiments is a validation test.. Building confidence in the physical theory In a weak sense. predict material response under different (noncalibration) loading scenarios. one would prefer to compare the model predictions against test data from homogeneous loading experiments that were different from the homogeneous loading tests conducted for calibration. Ideally. without any change in precalibrated parameters. Similarly.. its installation into a host code) is capable of predicting material system response to nontrivial loading scenarios. Parameterization entails fitting to a subset of discrete points in parameterization data tables (e.e.. The goal is to assess whether or not the GeoModel parameterization instructions in Appendix A can lead to a highquality set of material model properties that are predictive in general loading scenarios. not all values measured in the test). This chapter describes some validation problems that have been studied to date. to validate a constitutive model (not its implementation into a host finiteelement code).
05 Axial Axial Stress (MPa) Stress (MPa) Radial Data Model Axial 400MPa (b) 200MPa 100MPa 50MPa 0 0. From Ref. The ability of a single parameter set to agree so remarkably well with this suite of data is a validation. [46]. (a) GeoModel prediction of uniaxial strain loading. (4.1 Strain 0.20 0. Fig. For example.2 Strain Figure 9.15 0. 105 .10) are parameterized from shear and hydrostatic unloading data. 9.67) parameterized through purely hydrostatic testing.1.25 Uniaxial Strain Uniaxial strain 1200 1000 800 600 400 200 0 0.00 Data Model 0. The shear limit function Eq.9) and (4.05 0.15 0. Once parameterized in this way.05 0. 1200 Axial Stress (MPa) 1000 Axial Stress (MPa) (a) 800 600 400 200 0 0. (4. the GeoModel may be applied to predict the irreversible plasticallyhardening stressstrain response at a variety of other stress paths. (b) GeoModel predictions of triaxial stressstrain response at various confining pressures. The crush curve Eq.1 compares GeoModel predictions with simple triaxial data at various confining pressures (these simulations all use a single set of GeoModel parameter values).10 0.39) parameterized through peak states in triaxial testing.Building confidence in the physical theory Postcalibration Triaxial loading The nonlinear elasticity Eqs. (4.
0 2000.0 5 (a) 10000.0 data 6000. 106 .0 0 SCE/PRONTO 3D (2500Hz) Data model 2000.0 data Deceleration (G's) Deceleration (G's) 6000.0 2.0 0 1.0 10 deceleration (kG’s) 10 8000.0 4000.0 6. it was used to predict projectile penetration depth using spherical cavity expansion analysis.0 0 1.0 time (ms) 3.0 2000.0 Deceleration (G's) data 6000.0 6 (c) Figure 9. After the GeoModel was fitted to laboratoryscale material property tests.0 6 2000.0 time (ms) 3. material constitutive models are often considered “above average” if they are predictive within 20%.0 3.0(ms) 4.0 2000.0 time (ms) 4.0 0 0.0 10 10 data deceleration (kG’s) 8000.0 0.5 0.0 0 1.0 Time (ms) 4.0 Time (ms) 4.0 deceleration (kG’s) 8000.2 shows results from a pretest prediction of depth of penetration and displacement histories.0 0.0 SCE/PRONTO 3D (2500Hz) Data model 2.0 time3.0 0.0 6.0 0.0 0 SCE/PRONTO 3D (2500Hz) Data model 2.0 0.0 100 150 Projectile Striking Velocity (m/s) Projectile Striking Velocity (m/s) 10000. In tests like these.0 Deceleration (G's) 6000.0 2000.0 5.0 Time (ms) (d) 5.0 0 1.0 200 250 300 350 400 10000.0 2.2.0 (b) 10000.0 4000.0 SCE/PRONTO 3D (2500Hz) Data model deceleration (kG’s) 8000.0 5 2000. As seen.5 Measured Predicted 1.0 4000.0 0.0 4000.0 2000.0 Time (ms) 5.0 Depth of Penetration (m) Depth of Penetration (m) 1.0 0.0 0 0.0 5. the GeoModel performed exceptionally well by this metric.Building confidence in the physical theory Fieldscale penetration Figure 9. Y X 2.
This concludes our overview of preliminary validation testing.3. The goal was to predict a second measured velocity history at a point 470 meters from the source (see Fig. natural spatial variability of insitu rock (which is neglected in the simulation) may play a role. Unlike the similar elasticity verification test described on page 95. GeoModel + JAS3D finiteelement prediction (blue) vs. Spherical cavity geometry. as should be the case for any validation test). measured (red) velocity and displacement at 470 m. 9. from a spherical cavity velocity source at 204 meters. were determined by applying the GeoModel’s calibration procedures described in Appendix A (backfitting alteration of these independently determined calibration parameters was disallowed. or myriad other possible error sources. As was the case in the verification study.Building confidence in the physical theory Free field wave form for spherical shock loading roller Figure 9.4 shows displacement and velocity histories for an underground test in which a wave propagates from an explosive point source. this validation roller test used elasticplastic fieldscale parameters that Figure 9. further study is warranted. Of course. Unlike the verification study presented on page 95. A measured velocity history at a point 204 meters from the source was used as the velocity boundary condition of the simulation.4. Because the GeoModel falls short of a compelling agreement with data. More extensive model validation activities (now underway) require publication of this theory and user’s guide to provide a resource document that allows analysts to apply the GeoModel properly. 107 . this validation simulation is overpredicting peak velocity. suggesting that disagreement with data might be rooted in host code problems as much as shortcomings of the GeoModel. The GeoModel (along with its implementation within the host finiteelement code. as might the response time of gauges used to acquire the data.3). code predictions were strongly affected by artificial viscosity. indicating that the GeoModel’s advanced physical features (especially pressure sensitivity) are important. JAS3D) comes far closer to matching data than simpler models such as nonhardening Von Mises plasticity. Figure 9.
Closing Remarks 10. validation. and as such it is capable of satisfying the needs of almost any structural application involving geomaterials. most rocks exhibit nonlinear elastic deformation upon unloading and reloading. The GeoModel is a genuine unification and generalization of simpler models. and quality assurance. pressure dependence.g. and natural spatial variability resulting in localization. In the laboratory. documentation. there is still much to be accomplished. strainrate sensitivity. longterm effort was begun that brings together activities in laboratory testing. e. 108 .. software development. Sandia recognized the need to enhance the material modeling capabilities for geomaterials. In addition. and reservoirscale modeling of formation compaction caused by pore pressure drawdown during oil or gas production. anisotropic jointed rock behavior (as observed in situ). hysteresis loops. Three key applications for this work are in projectile penetration research. While this is an ongoing program. and postpeak softening. the strainrate sensitivity of some porous rocks depends on the predominant deformation mechanism. analysis of hard and deeply buried targets. which occur simultaneously allowing macroscopic prefailure dilatation to occur even as pores continue to collapse. Moreover. high porosity rocks under compressive mean stresses and nonzero deviatoric stresses. the GeoModel has reached a stage of maturity that warrants documentation of the effort to date. With the emergence of the capability to simulate the largescale mechanical behavior of complex geosystems by virtue of recent advances in software and hardware. Closing Remarks The preceding chapters have attempted to elucidate the physical foundations and domain of applicability of Sandia’s GeoModel. physically motivated phenomenological judgements about relations between stress and strain have been given ultimate priority to more accurately match observed laboratory behavior. verification. Current modelenhancement activities (to be documented in future reports) include pressure dependence of the extension/compression strength ratio. the strainrate sensitivity of shear and extensional failure is different from that of pore collapse. This model was developed in response to the need for a predictive model that could be used for a wide range of applications while maintaining numerical tractability in the context of Sandia’s solid mechanics finiteelement software. softening and failure (fracture). While “firstprinciples” microscale theories have influenced the general model framework. basic research. different behavior in extension than in compression. and the strainrate sensitivity itself may be pressure dependent. involve a complex interplay of deformations from competing mechanisms including pore collapse and microcrackmicrovoid development. The goal is to provide a rock mechanics predictive capability that fully accounts for the complex nature of in situ rock masses. While the GeoModel has achieved many of the stated objectives. A largescale. This report summarizes the progress toward achieving a realistic rock constitutive theory that can be calibrated via standard laboratory experiments and is numerically tractable for massively parallel calculations using tens of millions of threedimensional finite elements and loading conditions that involve ground shock and other nonquiescent processes.
derivatives of Eq. total volumetric strain data to obtain the nonlinear elastic bulk modulus parameters ( b 0 . its value is fixed). b 1 . where p is the pressure (positive in compression). specialized parameterization software (e. b2 K = b o + b 1 exp – . This supplemental software. nonlinear optimization is available via the “Solver” addin of MS Excel. p 2 . each having a value 1 or 0 to indicate whether that parameter is to be optimized (if not. STEP 1. (4. b 0 in that order) and a second line that contains five integers. Alternatively. Hydrostatic data are used to obtain the nonlinear bulk modulus parameters in Eq. roughly equals the initial porosity. * HYDROFIT is available upon request. 4. The HYDROFIT program outputs optimized values for p 0 . b 1 . p 3 .68) with respect to each parameter]. A.) was used to perform nonlinear regression to determine optimized parameter values. In many cases.2) First express data as a column or space delimited table with the first column being tot total volume strain ε v and the next column X = 3p . where ξ = X – p 0 = 3 ( p – P E ) (A. SHEARFIT. The peak inelastic volume strain. and b 2 ) as well as the crush curve parameters p 0 . Eq. (4.1.Use hydrostatic pressure vs. The input file for the HYDROFIT program contains one line with five values (initial guesses for the parameters p 3 . as in Fig.68). Determine the parameters in Eq. p 1 . p 0 . b 0 . this software presumes linear elasticity when X < p 0 . p 3 . and b 2 .e. Parameterizing the GeoModel The following steps describe how to determine values for parameters used in the GeoModel. The HYDROFIT program employs standard nonlinear regression fitting procedures [35] and includes parameter sensitivities [i. p 1 .g. A1 . p 2 .1) using Fossum’s HYDROFIT program* (which assumes linear elasticity when X < p 0 ). p 2 . etc.APPENDIX A.1) (A. (4. HYDROFIT will provide only approximate parameter values. (A. fitting nonlinear parameters is not automated.. HYDROFIT.. p 1 .14 for guidance on how to set initial guesses for the other crush parameters. I 1 as well as the parameters in the crush curve. which applies nonlinear least squares fitting is available from Arlo Fossum. If the data show nonlinear elasticity. Refer to Fig. p 3 . p p 3 – ε v = p 3 e –( p1 + p2 ξ )ξ . At present.9). Parameterizing the GeoModel APPENDIX A. At present.
the initial elastic loading curve is so small in comparison to the scale that it is not visible.1. 4. and b 2 . which is not included in the GeoModel.2. this hydrostatic compression test (for frozen soil) was conducted nearly to full pore collapse. Parameterizing the GeoModel pressure (MPa) p3 initial elastic loading (not visible) inelastic loading elastic unloading volume strain Figure A. and p 2 . A2 . zircoa). Ideal hydrostatic parameterization data. the HYDROFIT program converts the loading portion of the data to a crush curve (see Fig. but the loading curve shows signs of material creep. p 1 . Like the data for the material shown in Fig. A.14). p 1 . With these parameters. For this material (a ceramic powder. Figure A.1. Lesspreferable hydrostatic parameterization data. b 1 . The unloading portion is used to obtain the elastic bulk modulus parameters b 0 .APPENDIX A. and employs nonlinear regression to obtain the crush parameters p 0 .
250 23 MPa Concrete Triaxial Compresion Test 75 MPa Confining Pressure 200 150 inelastic loading 100 50 initial elastic loading can be difficult to distinguish from onset of inelasticity.08 0.1 0. triaxial testing is typically performed as a twostage process in which the material is first compressed hydrostatically to a given pressure.. As explained on page 28. the previously determined bulk modulus parameters are used to obtain the elastic shear modulus parameters g 0 . Then use nonlinear least squares regression analysis to obtain the shear modulus parameters g 0 . 0 0. Parameterizing the GeoModel STEP 2. 3. As illustrated in Fig.04 0. axial strain ε A will have a slope equal to Young’s modulus E . g 1 . 1 1 – g 1 exp ( – g 2 J 2 / 2 ) G = g o 1 – g1 (A. and g 2 .16 0.15). Then. the first column being the axial strain ε A and the second column being the stress difference ( σ A – σ L ) . Only the elastic unloading data should be used for determining the nonlinear elastic shear modulus parameters. g 1 . and g 2 . the lateral stress is held fixed while the axial stress is varied.3) Arrange the triaxial data as a twocolumn table.12 elastic unloading 0 0. during the second (triaxial) leg.3.APPENDIX A. (4. Recall that the nonlinear elastic bulk modulus K was found previously in step 1. The SHEARFIT program computes the shear modulus G from K and E by using the 3KE standard linear elasticity formula. and g 2 . This step determines the parameters in Eq. Through standard moduli conversion formulas from linear elasticity. (4. The slope of the unloading curve is the nonlinear tangent Young’s modulus.1 Figure A.14 0. Rather than directly using Young’s modulus E as a userspecified material parameter.3 in the main text. cited in Eq. the GeoModel requires the shear modulus G .Use triaxial compression data to obtain the shear modulus parameters g 0 . g 1 .10).02 0. the plot of stress difference ( σ A – σ L ) vs. If fitting to rock 9K – E A3 . G = . Data representative of a triaxial compression parameterization test.06 0.
2 (this is a useful “sanity” check). Characterizing the limit envelope requires numerous different experiments.(optional) maintain a record of all peak stress states ever measured for every available quasistatic loadtofailure experiment ever performed for the material of interest. Parameterizing the GeoModel data. Typically.1 on page 5). F f ( I 1 ) = ( a 1 – a 3 e – a2 I 1 ) + a 4 I 1 . The softening threshold (peak limit) envelope is the boundary of any and all stress states quasistatically achievable for the material. is typically in the neighborhood of 0.APPENDIX A. (4.39). the limit envelope is fixed in time (see Fig. (A. where I 1 = –I1 (shear limit function). The goal here is to gather sufficient data to parameterize both the Lode angle function Γ ( θ ) and the shear limit envelope in Eq. the value of J 2 at peak failure might be lower than J 2 . STEP 3. The number of entries in your peak stress table will equal the number of experiments run to failure.4) A4 . find the stress state at which J 2 is larger than for any of the other stress states in that experiment. in general. It is the peak that is of interest. If the material softens before rupturing. 1. each individual experiment has precisely one stress state at which the second stress invariant J 2 achieves a peak value. not the postsoftening value at failure. @peak @peak peak and Construct a table of data triplets ( I 1 . Unlike a yield surface. Poisson’s ratio. For every available quasistatic loadtofailure experiment. J 2 . evolve as the microstructure is altered in response to inelastic deformation). θ @peak ) . including both elastically obtainable stress states and stress states that can be reached only through inelastic deformation. where I 1 @peak are the values of I 1 and θ at the stress state for which J 2 is at its peak θ value. which is the boundary of elastically obtainable stresses (and which therefore will. ν = ( 3K – E ) ⁄ ( 6K ) . All achievable yield surfaces (an infinite set) are contained within the single limit envelope.
Once the F f function has been determined in this parameterization step. the GeoModel theory presumes the bounding curve is adequately described by J 2 = F f ( I 1 ) ⁄ Γ ( θ ) . The boundary of data points defines the F f function. a 3 . A. A suite of actual TXC experiments for porcelanite is shown in Fig. which simply means that the GeoModel presumes that the limit function at nonTXC Lode angles is simply a scalar multiple of the TXC function. 8. along with further A5 . A scatter plot of all TXC stress data measured at various confining pressures might look somewhat as sketched below: J2 = F f ( I1 ) = ( a 1 – a 3 e –a2 I 1 ) + a 4 I 1 I1 Figure A. and a 4 . at other Lode angles).5. A sufficient number of TXC experiments must be conducted under various confining pressures so that the bounding surface begins to take form. On other meridional planes (i. In Fig.e.4. the set of all stress states ever observed in TXC (Lode angle θ = +30° ) must fall below the curve J 2 = F f ( I 1 ) . this curve defines boundary of all stress states that ever have been (or ever can be) observed in quasistatic TXC loading.Use peak stresses from a family of triaxial compression tests to determine the shear failure envelope parameters a 1 . Stated differently. Thus.. a 2 . the peak data pairs (darkened dots) correspond to the θ @peak = 30° values from your table of peakstress invariants collected in STEP 3. Plots like the are shown for various materials in Fig.4.1. A. Recall that the F f function represents a peak shear limit envelope in the meridional plane for which J 2 is plotted against I 1 for triaxial compression (TXC) stress states. A family of TXC tests conducted to failure. All other (subpeak) data points fall below this line. Using only the TXC ( θ = 30° ) data create a scatter plot of all everachieved TXC values of J 2 and I 1 . Parameterizing the GeoModel STEP 4. modulo experimental scatter.APPENDIX A. the peak stress data at other Lode angles will be used later to determine the Γ proportionality function.
APPENDIX A. Parameterizing the GeoModel
illustration of how discrete peak points from these experiments are transferred to a limit surface meridional plot for fitting to Eq. (A.4)
(a)
(b)
Figure A.5. TXC stressstrain plots and extraction of their peak values to construct the meridional limit curve (data are for porcelanite). (a) Each shear stress J 2 vs. axial strain plot has exactly one peak value, as labeled. The value of I 1 at this peak is found by applying Eq. (3.28), I 1 = 3τ + 3σ L peak and σ equal to the lateral confining pressure for the test. (b) The peak states are with τ = J 2 L transferred to a plot of J 2 vs. I 1 for fitting to the F f meridional limit function.
Once enough TXC experiments have been conducted for a welldefined shear limit boundary to emerge, the next step is to determine values of a 1 , a 2 , a 3 , and a 4 that best fit the F f function to this boundary. Given triaxial stress difference σ A – σ L vs. axial strain ε L data, Eqs. (3.28) and peak (3.29) show that value of J 2 at peak is given by J2 = σA – σL peak ⁄ 3 and the value @peak = 3σ + ( σ – σ ) peak , where σ of I 1 (i.e., the trace of the stress) at peak equals I1 L A L L @peak peak is the (constant) lateral confining pressure. The ( I 1 , J2 ) data pairs from TXC experiments (i.e., those for which θ @peak = +30° ) may be fed into a nonlinear regression parameterization program, SHEARFIT (available upon request) to A6
APPENDIX A. Parameterizing the GeoModel
optimize the parameters in Eq. (A.4). To use this program, create an input file containing, in the first line, your initial guesses for a 1, a 2, a 3, a 4 and, in the second line, four logical integers (0 or 1) to indicate whether or not to optimize on the corresponding parameter. Refer to Fig. 4.1(a) to help decide appropriate initial guesses. If the data suggest a linear envelope at high pressures, then your first guess for a 4 should be an approximation of the slope of this envelope. If in doubt, take a 4 = 0.0 (i.e., assume the data asymptote to a constant value as pressure goes to infinity). Eyeball the data to set your first guess for a 1 equal to the zeropressure value of the linear asymptote line (extrapolate visually if necessary), set a 3 to equal a 1 minus your best estimate for the actual ordinate intercept of the lowpressure data (again, extrapolate if necessary). Finally, set a 2 to equal an estimate for the initial (lowpressure) slope of the data divided by a 3 . Using these initial guesses, the SHEARFIT program applies nonlinear regression and outputs values for the a 1 , a 2 , a 3 , and a 4 parameters. BEWARE: In typical TXC experiments, all principal stresses are compressive, making it is possible that SHEARFIT will return a meridional fit to the available data that corresponds to a shear limit envelope that does not corral the origin (implying nonphysically that zero stress is “unachievable”). If this occurs, you might want to @peak peak append your table of observed ( I 1 , J2 ) data with an entry ( – I 1limit, 0 ) , where I 1limit is your best estimate for the theoretical hydrostatic tensile strength of the material (therefore – I 1limit is a negative number). Include this entry multiple times if necessary to force it to have greater weight in the nonlinear regression (or, preferably, perform more experiments to obtain a larger number of real data points at low values of I 1 ).
STEP
5.Use peak stresses from a family of triaxial extension tests to determine the extension to compression ratio ψ . As was done above for peak peak TXC tests, construct a table of ( I 1 , J 2 ) data pairs corresponding to peak attained stresses in triaxial extension (TXE). These are the data pairs in your peak data table from STEP 3 that correspond to θ @peak = – 30° . For each of these TXE data pairs, compute
peak J 2 in TXE ψ = @ F f ( I 1 peak in TXE )
(A.5)
where F f is the TXC shear limit function parameterized in the previous step. Of course, each TXE experiment is likely to result in slightly different values for ψ . At present, the GeoModel presumes that the TXE/TXC ratio ψ is constant. A7
APPENDIX A. Parameterizing the GeoModel
Therefore, set the GeoModel parameter ψ equal to the average of each ψ computed using Eq. (A.5) for each available TXE experiment. It is possible that the data might imply that the strength ratio ψ must vary with pressure, but the GeoModel presently assumes that ψ is the same at all pressures. Thus, until the GeoModel is enhanced to support pressurevarying strength ratios, modeling errors must be managed by measuring the parameter ψ at a confining pressures in the neighborhood where the GeoModel is likely to be applied. Similar statements can be made regarding any and all parameters used to define the GeoModel yield function.
Sometimes, it might be impractical — or overly expensive — to obtain TXE data. In this case, an engineering approach for estimating ψ presumes that it obeys the same coupling to the meridional profile slope as predicted in classical MohrCoulomb theory. Using Eq. (A.4) to set the TXC slope in Eq. B.19 in Appendix B, an estimate for the pressurevarying strength ratio is 1 ψ = 1 + 3 [ a 4 + a 2 a 3 e –a2 I 1 ]
(A.6)
In the future, we hope to thoroughly explore the merits of this formula, which correlates the strength ratio with the TXC meridional slope. At present, the GeoModel presumes the strength ratio is constant. If, however, the above correlation formula can be substantiated, then we will likely incorporate it into future releases, thereby possibly eliminating the need for a userspecified ψ altogether. Until then, you may evaluate the above formula at a value of I 1 in the neighborhood where you plan to apply the GeoModel to obtain a reasonable estimate for ψ .
A8
A.7. Large values of Γ correspond to small shear strengths.6 an appropriate choice for J3TYPE should be apparent. Parameterizing the GeoModel STEP 6.10. Recall that the GeoModel’s function Γ ( θ ) defines the shape of the octahedral profile shape for Lode angles spanning the range from TXE ( θ = – 30° ) to TXC ( θ = +30° ). as illustrated in Fig. select J3TYPE=1 if the material is judged to be moderately ductile.Determine the appropriate Lode function option J3TYPE. Γ ) data to the graphs in Fig. This parameterization step aims to guide the choice in the happy circumstance that nontriaxial data are available. 4. A. A9 . Looping over your table (collected in STEP 3) of “allobserved” peak stress peak peak invariant triplets ( I 1 . create a new twocolumn table of ( θ. it might be easier to instead scale the ordinate as shown in Fig. 4. θ peak ) .7) By comparing a scatter plot of these ( θ. especially those at nontriaxial states.10. By overlaying data with the family of plots in Fig. If no such data are available. As illustrated in Fig. an appropriate choice for J3TYPE should be more clear.6. you will need to use engineering judgement as to an appropriate choice for J3TYPE. The GeoModel parameter J3TYPE (see page 45) dictates the functional form to be used for the Γ function. If nontriaxial data are unavailable (a common problem). select J3TYPE=3. Γ ) data pairs. Nontriaxial data are difficult to acquire. where @peak Ff ( I 1 ) Γ = peak J2 (A.APPENDIX A. J 2 . To assist in the decision. all available peakstate data obtained in nontriaxial loading paths are considered to help decide an appropriate choice for the J3TYPE option. In this model parameterization step.A. the GeoModel’s Γ function is defined to equal 1 in TXC and 1 ⁄ ψ in TXE. otherwise.
4 Γ(θ) – 1 1⁄ψ–1 1 0.2 0.4 0.8 0. With this scaling.2 1 0.4 θ Figure A.4 0 TXC π 6 Figure A. the user must resort to engineering judgement to decide which J3TYPE option to select.4 0.7.2 0. A10 .2 TXE π – 6 SHR 0.2 0. Knowing that Γ = 1 ⁄ ψ in TXE and Γ = 1 in TXC.2 0. In this contrived example.4 0. most of the data at nontriaxial states falls in the red WillamWarnke region and therefore J3TYPE=2 would be appropriate for this material. regardless of the value of ψ .2 0. In this case.2 0. data at nontriaxial Lode angles are rarely available.8 0. the scaled ordinate equals 1 in TXE and 0 in TXC.4 0.2 0. the others vary with ψ .8 0.6. it makes sense to scale the ordinate as shown so that. Γ–11⁄ψ–1 1. now displayed with a transformed ordinate. The Lode function information originally shown in Fig.10 of the main report.4 0 TXC π 6 TXE π – 6 SHR 0.2 Nontriaxial (very little data likely to be available) TXC (a “large” collection of data is likely to be available) 1 0.2 0.6 0.6 0.4 1.6 Γ(θ) – 1 1⁄ψ–1 1 0.6 1. A sketch of how data might be distributed at nontriaxial states.4 0.2 0.4 0. 4.2 0. Parameterizing the GeoModel Γ(θ) – 1 1⁄ψ–1 1.APPENDIX A. all Gudehus Lode functions overlap.4 0.4 0.4 0 TXC π 6 TXE π – 6 SHR 0.8 TXE (a moderate collection of data is likely to be available) 0. In practice.2 0.
Both types of hardening may occur simultaneously. Unlike the limit surface. which bounds all possible stress states.6(b) (even though axial strain increases monotonically. 1. the infinite set of all possible yield surfaces must be contained within this limit envelope. the amount of kinematic hardening relative to total stress is high at lower pressures. a 2.. the volume strain turns around because of the lateral bulking strains). kinematic hardening data are best inferred from unconfined compression tests. We ultimately hope to assign values to the above parameters so that the volume strain will “turn around” as Fig. volume strain A11 .0 (no rate sensitivity). b 1. Therefore. we determined the GeoModel parameters that define the outer limit surface. set the “yettobedetermined” GeoModel parameters as follows: • • • • • • offset RN=N=0 (i. b 2. Unlike the limit surface. The goal is to determine values for the asyetunknown GeoModel parameters through a systematic simulation sequence. and J3TYPE.Hardening parameters. to obtain a simulated table of axial stress vs.e. p 2 ) . tentatively assume associativity) joint spacing RJS=s=0. It is unlikely that this result will replicate observed unconfined compression data on the first try.APPENDIX A. In the previous parameterization steps.1 in the main text). a yield surface evolves (hardens) through time as a result of microstructural changes induced in the material under inelastic loading. a 4. axial stress Run the finiteelement code for unconfined compression and output axial stress vs. As a rule. In the previous steps. g 1. As a first guess. Kinematic hardening permits the yield surface to translate in stress space until the limit surface is reached. we determined crush parameter values ( p 0. the yield surface merely bounds the set of elastically obtainable stress states.e. g 0. a 3. Parameterizing the GeoModel STEP 7. 8.0 rate sensitivity parameters T1 through T7 =0. CRPF=CR (i. ψ ) . p 1. No stress state outside this fixed (nonevolving) limit surface can be achieved through any load path. elastic parameters ( b 0. The initial yield surface is typically much smaller than the limit surface (see Fig. g 2 ) . volumetric strain (EVOL). volume strain to compare with available experimental data. A4PF==A4. Now these values should be used in a finiteelement (or single cell driver) implementation of the GeoModel. Consequently. suppress hardening) kinematic hardening parameter HC=1e5 shape parameter CR=R=10 plastic potential function parameters for nonassociativity A2PF==A2. Isotropic hardening permits the initial yield surface to expand on octahedral planes (by amounts that vary with pressure) up until the limit surface is reached. the limit surface parameters ( a 1. RKPF=RK.
Next try changing HC. Parameterizing the GeoModel The phenomenological fitting functions employed in the GeoModel are designed to extrapolate reasonably well into regions where data are not available. Once shearinduced dilatation (turnaround) has been adequately modeled for unconfined TXC. a more traditional exploratory manual search method must be used as described below. Consequently. The intercept on the ordinate on the meridional profile is located at J2 = a 1 – a3 . try lowering HC. If the turn around stress is too high. Continue to go back and forth until you are satisfied with both confined and unconfined results. bring down the initial yield surface. If you continue to have trouble getting turn around.APPENDIX A. Consequently. Lowering HC lowers the turnaround strain. Try increasing N to a value no larger than a 1 – a 3 (our initial guess of N = 0 presumed that the initial yield surface coincided with the shear limit surface. If you get no turn around. try decreasing CR. but no similar fitting software has yet been developed to determine the hardening parameters. we will be exploring adjustments of CR. Increasing (say. By setting N to a nonzero value. for these parameters. there will certainly be modeling error. from 7 to 10 if computed peak strain is too large). In the following. but these functions are also selected in part for computational tractability and they therefore serve only as approximations. RN. axial stress volume strain This concludes our guidance for parameterizing new materials to the GeoModel. and HC to try to achieve a strain “turn around” at the correct (observed) strain and stress. and work on HC and RN.. Try changing CR (e. Continue to decrease CR until turn around occurs. Ideally.g. double check that you have correctly entered the previously determined (known) parameter values. doubling) HC will increase the strain value at which turnaround occurs. A12 . If no turn around is apparent in your simulation. We have already described the programs “HYDROFIT” for regression optimization of the elastic parameters and SHEARFIT for finding the limit surface parameters. we are now permitting kinematic hardening). go then to highconfinement data. one should use regression fitting procedures to select GeoModel hardening parameters that minimize modeling error relative to available data. You will likely see “turn around” start to occur. Go back to unconfined. Further revisions of this manual will likely include additional tips and parameterization experiment suggestions.
(2) plotable variables. a “1” indicates that the variable is dimensionless. To manage this potentially confusing conflict of conventions. In each table. recall x ≡ –x DEFINITION OF THE “OVERBAR” (B. Nomenclature and Data Sets APPENDIX B. and (3) other symbols or acronyms used internally within this manual. the first column shows the typeset symbol for the variable. while N/A means dimensions are not applicable. B1 . stress and strain are typically taken positive in tension. in applications. A “–” indicates that dimensions vary. The next column contains the ASCII string used for the variable in code input files and/or within the source code.1) For example. The defining equation (or page number) in the last column of the tables cites the location in this report where the quantity is defined or discussed.APPENDIX B. I 1 denotes the trace of stress (positive in compression). Therefore. I 1 ≡ – I 1 is positive in compression. they are taken positive in compression. Reminder: in mechanics. However. Nomenclature and Data Sets This appendix contains three tables defining (1) model parameters. In the SI units column.
Initial elastic shear modulus (for intact material if modeling joints). 4. To neglect this effect. the elastic tangent bulk modulus is computed by b4 b2 K = fK b o + b 1 exp – .33 g0 PROP(6) (G0) Pa Eq. the shear modulus will asymptote to a value g 0 ⁄ ( 1 – g 1 ) as shear stress increases. I 1 SI units defining equation b0 Pa Eq. 4. For linear elasticity.10 Eq.10 Eq. For the shear modulus to increase with shearing. Larger values of b 2 will result in a broader transition range. For nonlinear elasticity.1 b3 PROP(4) (B3) Coefficient in nonlinear elastic bulk modulus to allow for plasticityinduced changes in the elastic properties.9 Eq. 4. For linear elasticity. 4. A. For the shear modulus to decrease with shearing. Nomenclature and Data Sets Model Parameters (User Input) Symbol or Acronym ASCII Name PROP(1) (B0) Name and meaning Initial elastic bulk modulus (for intact material if modeling joints). set 0 < g 1 < 1 . set b 2 to a small value to transition rapidly from the low pressure bulk modulus to the high pressure modulus. set g 1 < 0 . Otherwise. Pa Eq. 4. set b 3 = 0 .9 Eq. The tangent shear modulus is computed by 1 1 – g 1 exp ( – g 2 J 2 / 2 ) G = g o 1 – g1 1 Eq. set b 4 = 0 . A.3 B2 . I p εv 1 Pa Eq. When this parameter is nonzero. 1 Eq. The tangent bulk modulus is given by b2 K = b o + b 1 exp – . Must be less than 1. Pa Eq. For nonlinear elasticity.3 g1 PROP(7) (G1) Parameter used to define the elastic (intact) shear modulus at large shears (see above formula).33 b4 PROP(5) (B4) Power in nonlinear elastic bulk modulus to allow for plasticityinduced changes in the elastic properties. set b 1 =0 . set g 1 = 0 . A. see above formula.0. set b 2 = 0 .1 b2 PROP(3) (B2) Curvature parameter in nonlinear elastic (intact) bulk modulus function (see above formula).APPENDIX B. 4.1 b1 PROP(2) (B1) High pressure coefficient in nonlinear elastic (intact) bulk modulus function (see above formula). To neglect this effect. For linear elasticity. Specifically. A. – b 3 Exp – . 4. A. set b 1 so that the Bulk modulus K will asymptote to a value b 0 + b 1 at high pressures.9 Eq.
whereas a 1 – a 3 is the vertical intercept of the limit function itself. I 1 = – trσ and therefore I 1 is three times the pressure.39 Fig. Set this parameter to zero if the material has no geological (or rocklike) faults. Thus. A. Here. 4. Set this parameter to zero if the material has no geological (or rocklike) faults. 4. At ˜ 1 Eq. The initial onset of yield is described by f f ( I 1 ) = F f ( I 1 ) – N . F f ( I 1 ) = a 1 – a 3 e – a 2 I 1 + a 4 I 1 . These are parameters define the ultimate limit curve.APPENDIX B. recognize that the constant term will be lower than a 1 by an amount N . A. Nomenclature and Data Sets Symbol or Acronym ASCII Name PROP(8) (G2) Name and meaning Curvature parameter in nonlinear elastic (intact) shear modulus function (see above formula). Joint spacing. set g 4 = 0 . at which the maximum possible hardening has occurred and softening is imminent. see above formula.10 Eq. B3 . the tangent shear modulus is computed by 1 1 – g 1 exp ( – g 2 J 2 / 2 ) g4 G = f G g o . To neglect this effect. 4.4 a1 PROP(14) (A1) zero pressure. 4. for the ultimate shear limit surface. To neglect this effect. Ff ( I1 ) = a1 – a3 while at high pressure (large F f ( I 1 ) ∼ a 1 + a 4 I 1 . SI units defining equation g2 1/Pa Eq. Joint shear stiffness. For nonlinear elasticity small values of g 2 cause the shear modulus to transition rapidly from its initial value g 0 to its high shear value g o ⁄ ( 1 – g 1 ) . Set this parameter to zero if the material has no geological (or rocklike) faults. set g 3 = 0 . Larger values of g 2 make this transition more gradual.34 PROP(11) (RJS) PROP(12) (RKS) PROP(13) (RKN) meter Pa Pa Pa Eq. a 1 is the vertical intercept of the linear asymptote. Constant term in the fitting function for the meridional profile. I 1 ).– g 3 Exp –  p 1 – g1 γ eqiv Pa Eq. For linear elasticity.34 g4 PROP(10) (G4) Power in nonlinear elastic shear modulus to allow for plasticityinduced changes in the elastic properties. set g 2 = 0 .3 g3 PROP(9) (G3) Coefficient in nonlinear elastic shear modulus to allow for plasticityinduced changes in the elastic properties. Otherwise. When nonzero. Joint normal stiffness. Thus if fitting to data for yield onset.
A. 1/Pa Eq. A. this parameter is denoted D2 . this parameter is denoted W.67 Fig.67 Fig.4 Eq. where p is pressure. Pa 2 Eq. SI units defining equation a2 1/Pa Eq. used when the crush curve has an inflection point. In the lexicon of traditional p . A. 1/Pa2 p2 PROP(20) (P2) Extra fitting parameter for hydrostatic crush curve data. 4.67 Fig. In many other publications about the GeoModel. this parameter is denoted D1 . Keep in mind that I 1 = – 3p . This parameter will be negative because I 1 is negative in compression. A.2 p3 PROP(21) (P3) Asymptote (limit) value of the absolute value of the plastic volume strain.4.2 p1 PROP(19) (P1) One third of the slope of a porosity vs. 4.14 Eq.39 Fig.2 Eq. this parameter is denoted 1 PROP(18) (P0) Pa Eq.α crush models.39 Fig. 1 Eq. A. Assign a 2 a large value to quickly asymptote to the highpressure profile slope.2 Xo .4 a3 a4 p0 PROP(16) (A3) Parameter in the shear limit meridional fit function.1 4 PROP(17) (A4) Highpressure meridional slope parameter in the fit function. 4. A. Value of I 1 at the onset of pore collapse for hydrostatic compression of virgin material. A.14 Eq. B4 . F f ( I 1 ) = ( a 1 – a 3 e – a2 I 1 ) + a 4 I 1 .14 Eq. pressure crush curve at the elastic limit. 4. This parameter is approximately equal to the initial porosity in the material and may be inferred from hydrostatic crush data. F f ( I 1 ) = ( a 1 – a 3 e –a 2 I 1 ) + a 4 I 1 .4 Eq. where P E is the elastic limit pressure in hydrostatic compression. 4. F f ( I1 ) = ( a1 – a 3 e –a2 I 1 ) + a 4 I 1 . In many other publications about the GeoModel. 4. Set a 2 = 0 for a linear meridional profile as in MohrCoulomb theory.67 Fig. In many other publications about the GeoModel. Nomenclature and Data Sets Symbol or Acronym ASCII Name PROP(15) (A2) Name and meaning Curvature decay parameter in the fitting function for the meridional profile. 4. 4.APPENDIX B. In many other publications about the GeoModel. 4.39 Fig. this variable would equal – 3P E . 4.
Otherwise. the shear strength can increase by an amount N before softening will commence. Roughly speaking. Offset parameter. Value must be positive.4 Eq.APPENDIX B. 4. Future releases of the GeoModel will likely allow ψ to be pressuredependent. A11 cut I1 PROP(26) (CTI1) Pa B5 . Must be nonnegative. Real materials generally satisfy 1 ⁄ 2 ≤ ψ ≤ 1 . this parameter affects how “quickly” the yield surface evolves toward the ultimate shear failure surface. 4. Pa Eq. regardless of the level of shear stress). For problems with kinematic hardening. A11 ψ PROP(23) (RK) 1 page 45 Fig. R equals the eccentricity (width divided by height) of the elliptical cap function.6 Eq. This parameter affects the stress level at which dilatation will occur in triaxial compression. This option is available if the yield function is not cutting off strength at a tensile pressure low enough for the application at hand. 4. A. Convexity of the yield surface requires that 1 ⁄ 2 ≤ ψ ≤ 2 (or 7 ⁄ 9 ≤ ψ ≤ 9 ⁄ 7 if using J3TYPE=1). N ) = F f ( I 1 ) – N . Nomenclature and Data Sets Symbol or Acronym ASCII Name PROP(22) (CR) Name and meaning Shape parameter that allows porosity to affect shear strength.15. TXE/TXC (triaxial extension to compression) strength ratio. Fig. To replicate older classical pore collapse models (which initiate pore collapse only at a critical pressure. 4. If dilatation is occurring too soon. A11 H PROP(25) (HC) Kinematic hardening parameter. the isotropic part of the stress is replaced with this value. In principle. Set H = 0 and N = 0 to suppress kinematic hardening. Set N = 0 to suppress kinematic hardening. so it is the ratio a ⁄ b (not A ⁄ B ) in Fig. increase the value of R . Tensile cutoff in allowable value of the first stress invariant I 1 .69 pg. where F f describes the shear limit surface (softening threshold). set R to a very small number. This parameter is available only temporarily to permit reasonable strength predictions in tension. the backstress invariant α J 2 will not be SI units defining equation R 1 Fig. If the first invariant (which is proportional to the negative of pressure) reaches this cutoff value.76 pg. Current enhancements of this model are focused on more physically rigorous tensile failure modeling.15 Eq. Decreasing R will decrease the influence of porosity on shear strength and therefore enhance the effect of void space creation associated with crack growth.5 N PROP(24) (RN) 1 permitted to grow any larger than N . the yield function should limit achievable values of I 1 . 4. The initial yield surface is defined by F finitial ( I 1.40 page 59 pg. 4. 4.
Be sure to set R PF equal to R if plastic normality is desired.22 T2 T3 T4 T5 T6 T7 — PROP(29) (T2) PROP(30) (T3) PROP(31) (T4) PROP(32) (T5) PROP(33) (T6) PROP(34) (T7) PROP(35) (J3TYPE) 1/s 1 1/s Pa sec 1/Pa 1 Eq. If a predicted principal stress is found to exceed this value.WillamWarnke 3 .7 PF a2 PROP(36) (A2PF) 1/Pa page 48 PF a4 PROP(37) (A4PF) Potential function parameter (=A4 for associative).22 Eq.Gudehaus 2 . See main text. See main text. See discussion regarding the state variable “CRACK” in the next table.22 Eq. To specify a constant intrinsic material response time. Set to zero. 5.22 Eq.22 page 45 Fig.. 5. SI units defining equation — Pa T1 PROP(28) (T1) sec Eq. Integervalued control parameter for specifying the desired type of 3 rd invariant yield surface: 1 . A. No longer used.22 Eq. Flow potential analog of the yield surface parameter R . Rate dependence parameter. Be sure to set a 2 equal to a 2 if plastic normality is desired. No longer used.22 Eq. Assign this parameter in the same way you would assign a value to a 2 . Set to zero.APPENDIX B. then it is replaced with this value. 5. set this user input equal to the characteristic response time. 5. 5. Use the other “T” parameters to enable dependence on strain rate and pressure. Be PF sure to set a 4 equal to a 4 if plastic normality is desired. the plastic strain rate will be PF normal to the flow potential surface). 1 page 48 R PF PROP(38) (CRPF) Potential function parameter (=CR for associative). Primary rate dependence parameter. Nomenclature and Data Sets Symbol or Acronym ASCII Name PROP(27) (CTPS) Name and meaning Principal stress tensile cutoff. Assign this parameter in a manner similar to a 4 except that this parameter will be used to generate the flow potential. 5.MohrCoulomb Potential function parameter (=A2 for associative). 1 page 48 B6 . Rate dependence parameter. 5. See main text. Rate dependence parameter. See main text. except that this parameter is used to generate the flow potential surface (i.e. Rate dependence parameter.
Be sure to set ψ PF equal to ψ if plastic normality is desired. you want the code to decrease its subcycle size by a factor of 10. for example. then set SUBX=–1. If you want the subcycle size altered by a factor “x”. Nomenclature and Data Sets Symbol or Acronym ASCII Name PROP(39) (RKPF) Name and meaning Potential function parameter (=RK for associative). Flow potential analog of the yield surface parameter ψ .APPENDIX B. If zero. the GeoModel will select an appropriate subcycle step size. then the GeoModel’s default subcycle increment will be multiplied by 10**SUBX (ten raised to the power). SI units defining equation ψ PF 1 page 48 — PROP(40) (SUBX) Subcycle control parameter. If SUBX is nonzero. If. 1 — B7 . then set SUBX to the base ten log of “x” [ log10 x = ( ln x ) ⁄ ( ln 10 ) ].
1/s page 77 I1 J2 α 11 α 22 α 33 α 12 α 23 α 31 SV(4) [I1] SV(5) [ROOTJ2] First stress invariant (positive in tension).e. In the GeoModel physics source code. evolution of the κ state variable) occurs. As a rule of thumb: increasing the userinput parameter R will increase κ (and therefore decrease κ ).15. this flag will equal 1.30 Fig. 3. This the equivalent shear stress in the material. Thereafter. 4.70 κ 1 page 77 — SV(2) [INDEX] Indicator for isotropic plastic hardening.13 page 77 page 58 page 58 page 58 page 58 page 58 page 58 Square root of the second stress invariant J 2 (always positive). Pa Pa Eq. κ is not the point at which the meridional profile has a zero slope — it is the branching location. 4. Recalling that κ = – κ . · ε equiv = · · ε ij ε ij . 4.0 even if the stress later becomes elastic. Isotropic hardening is controlled by the evolution of κ . the internal state variable κ typically will be negative. The zero slope point is reached at a higher pressure. This flag will equal zero up until the first time isotropic hardening (i. · ε equiv SV(3) [EQDOT] L 2 (Frobenius) norm of input strain rate tensor.APPENDIX B.10 Eq. 3. 3. 4.5 κ The value of I 1 at which the meridional yield profile first branches away from the crack failure surface and begins to morph into the cap function associated with porosity. Pa Eq.9 page 77 Eq. Recalling Fig. 4. Nomenclature and Data Sets Internal State Variables (Plotable Output) Name and meaning Symbol or Acronym ASCII SI units Name SV(1) [KAPPA] ELN EL defining equation (or page) Fig. is denoted by ELN or by EL in the subcycling. SV(6) [ALXX] SV(7) [ALYY] SV(8) [ALZZ] SV(9) [ALXY] SV(10) [AXYZ] SV(11) [ALXZ] 11 component of the backstress 22 component of the backstress 33 component of the backstress 12 component of the backstress (= 21 component) 23 component of the backstress (= 32 component) 31 component of the backstress (= 13 component) Pa Pa Pa Pa Pa Pa B8 ..15 Eq.
˜ · 1 page 77 SV(18) [BACKRN] Square root of the second backstress invariant. more advanced fracture response). Physically. 4. 4. This is like the equivalent shear stress. 4.79 — SV(19) [CRACK] 1 page 77 B9 . However.0 initially and then decays down to 0. preferably. SV(17) [EVOL] Total volume strain. except applied to the backstress. Pa Eq. element death or. κ ≡ – κ is Pa2 Eq. Equivalent uniaxial plastic shear strain (conjugate to J 2 ). Nomenclature and Data Sets Symbol or Acronym ASCII Name SV(12) [GFUN] Name and meaning Kinematic hardening decay function. p γ eqiv = SI units defining equation (or page) Gα 1 Eq. 4. where ˜ · γ p is ˜ · p Equivalent plastic volume strain: ε v = ∫ trε p dt . ˜ p εv SV(14) [EQPV] SV(15) [EL0] ·p 2 γ dt .APPENDIX B.79 page 58 p γ eqiv SV(13) EQP [EQPS] 1 Eq. Keep in mind: κ is typically positive and therefore typically negative.35 ∫ · the deviatoric part of the plastic strain rate ε p . equal to 1. This is not really an internal state variable.0 at the onset of softening whenever the maximum tensile cutoff has been applied (or when the limit surface is reached). Specifically. simply replacing the principal stress with zero at the constitutive level would result in meshdependencies in host codes that lack macroscale fracture services. The value of BACKRN is not permitted to exceed the user specified limit value of N Flag that equals 1. The geomodel simply replaces the stress with the cutoff stress (user input CUTPS).73 hκ tot εv α J2 SV(16) [HK] (Isotropic hardening parameter) Proportionality factor · · appearing in the relationship κ = λ h κ . ˜ 1 Pa page 77 Fig. complete loss in loadcarrying ability in the appropriate direction is actually desired. Its value will remain constant throughout the calculation.0 as the max allowable kinematic hardening (determined by the shift parameter N ) is approached. 4. ∫ tr ε dt .15 κ0 Calculated initial value for κ (the cap branch value of I 1 ). More advanced codes should examine the internal state variable flag “CRACK” to determine when this cutoff is being applied (and it is therefore appropriate to initiate void insertion.
46 Lode angle in degrees ranging from 30 in triaxial extension to +30 in triaxial compression. the response is plastic.10 Eq. L Quasistatic backstress. q L stands for any internal state variable. 3.0 if the material response is elastic.e. 5. The value ranges from 0. In Eq. Nomenclature and Data Sets Symbol or Acronym ASCII Name SV(20) [SHEAR] Name and meaning Flag equal to 0. Value of the yield function SI units defining equation (or page) — 1 Fig. 5. Internal GeoModel variable (quasistatic “low” value of κ ). 5. 6. 5. In Eq. 5. 5.1 page 77 f θ SV(21) [YIELD] F SV(22) [LODE] Pa2 Eq. 5.15. L Quasistatic backstress. q L stands for α ij . 5. 5. 5. pushing directly against it).40 page 77 L σ 11 L σ 22 L σ 33 L σ 12 L σ 23 L σ 31 SV(23) [QSSIGXX] Pa Pa Pa Pa Pa Pa 1 Pa ⋅ s Eq.15 Eq. q L stands for α ij .15 Eq. 5.15. Otherwise. In Eq. In Eq.15 Eq.0 if the trial stress rate is normal to the yield surface (i. 5.10 Eq.15.0 if the trial stress rate is tangent to the yield surface to 1.15. In Eq. L Quasistatic backstress. 5. In Eq. The Lode angle is frequently denoted β in other publications about the GeoModel. 5. q L stands for α ij . 5. 4. 5. 5.15 Eq.15. 5.10 Eq.15.55 Eq. q L stands for α ij . q L stands for α ij . L Quasistatic backstress. 5. Pa Pa Pa Pa Pa Pa Eq. This is the particular instance for which q L = κ L .. L Quasistatic backstress. 5.10 Eq.15 Eq. 4. if positive. q L stands for α ij .10 Eq. In Eq. deg Eq. which is the quasistatic value of the isotropic hardening ISV.APPENDIX B.10 Eq.15.15 SV(24) [QSSIGYY] SV(25) [QSSIGZZ] SV(26) [QSSIGXY] SV(27) [QSSIGYZ] SV(28) [QSSIGZX] SV(29) [DCSP] SV(30) [QSEL] · λ κL Pa L α 11 L α 22 L α 33 L α 12 L α 23 L α 31 SV(31) [QSBSXX] SV(32) [QSBSYY] SV(33) [QSBSZZ] SV(34) [QSBSXY] SV(35) [QSBSYZ] SV(36) [QSBSZX] L Quasistatic backstress. 4. Internal GeoModel variable (quasistatic “low” stress) Internal GeoModel variable (quasistatic “low” stress) Internal GeoModel variable (quasistatic “low” stress) Internal GeoModel variable (quasistatic “low” stress) Internal GeoModel variable (quasistatic “low” stress) Internal GeoModel variable (quasistatic “low” stress) Plastic consistency parameter.83 Eq.15 B10 .
55 page 45 Eq.3 Eq.29 DFDSIG Derivative of the plastic potential with respect to stress time t ∆t E ff Ff fc Fc φ φ FC FF DT time increment Young’s modulus The part of the meridional yield profile function associated with microcracks.4 — δ ij p ∆γ ij p ∆ε v Kronecker delta · p . f c = Fc Pa 1 Eq.39 Fig. B.9 B11 .9 I1 Negative of I 1 . This is positive in compression Pa Eq.30 Eq. = trσ . This is positive in ten˜ Pa Eq. 4.32 Eq.78 page 13 PHI G GAMMAP Γ α H ij . 4. I 1 sion. 3. 4.APPENDIX B. 4. 4. The stress is diagonal (no shears).29 Meridional nominal yield function (Pelessone function) Friction angle (for MohrCoulomb theory) Flow potential function Octahedral yield shape function (depends on J3TYPE) Kinematic hardening modulus tensor Acronym: Hydrostatic loading. f f = F f – N Meridional shear limiter function The part of the meridional yield profile function associated with porosity.5 Eq. 4. 4.· Deviatoric plastic strain increment = ε ij – 1 ε kk δ ij ∆t ·p Plastic volume strain increment = ε kk ∆t VOLPLAS 1 — s s Pa Pa — — — — Eq. 4. 1 1 Pa2 1 Pa Eq.6 Eq.4 Eq. 4. 4. 4. 2.12 Fig. 3. H ˜ HYD I1 RI1 First stress invariant. Nomenclature and Data Sets (used only in this report) Symbol or Acronym C ijkl ASCII Other symbols Name and meaning SI units Name defining equation (or page) Fourthorder elastic stiffness tensor DELTA DEIJPL Pa 1 1 3 Eq.
3.12 Eq. 4. J 3 = .trS 3 .22 Eq. 4.12 page 11 ν p Pa in compression. The pressure p .11 J3 J3 ξ J3 RJ3 the stress deviator is closer to TXE than to TXC.5 Eq.1 The stress tensor. ξ J2 RELJ2ZP 2J 2 is the Second invariant of the shifted stress 1 Third stress invariant.39 Eq.10 Geometrically. J 2 = .22 Eq. The mean stress is positive 3 ˜ I 3 page 8 Eq. 3. B12 . 3.3 Porosity (unloaded). 4.e. 3. p = . Lode radius). 3. Nomenclature and Data Sets Symbol or Acronym ASCII Name RJ2 Name and meaning 1 Second stress invariant. 4.0 Code parameter equal to 4. 3. p Π PONE PTWO PFOUR 1 Pressure = negative of mean stress. which is positive in compression is the negative of the mean stress: p = – p . p = – .APPENDIX B.( trσ ) I ˜ ˜ ˜ ˜ 1 3 1 1 1 1 1 kg/m3 Pa Pa R ˜ ρ S ˜ σ ˜ Eq. 2J 2 equals the magnitude of the stress deviator. 2 ˜ SI units defining equation (or page) J2 Pa2 Eq. 3. and it therefore equals the “length” of the projection of the pseudo stress vector in 3D HaighWestergaard (stress) space onto the octahedralplane (i..trσ =  1 3 ˜ I 3 Pa page 11 Eq.11 Third invariant of the shifted stress Poisson’s ratio 1 1 Mean stress. RELJ3ZP 3 ˜ Pa3 Pa3 1 Eq.0 Code parameter equal to 2. S = σ – .trS 2 (never negative). Negative of J 3 . 4.65 — — — Eq. This is positive when the stress deviator is closer to TXC than to TXE. This is positive when Pa2 Pa3 page 8 Eq.0 Polar rotation tensor mass density Stress deviator.. Code parameter equal to 1.trσ = .
40 sin 3θ ξ ξ ˜ SIN3BTAP Sine of three times the Lode angle θξ Eq. Nomenclature and Data Sets Symbol or Acronym ASCII Name Name and meaning Acronym: Shear loading (one principal value of the stress deviator is zero and the others are therefore negatives of each other). This is positive when the stress is closer to TXC. 5. 3. SHR applies even when all principal stresses are compressive — all that matters is the nature of the stress deviator. Pa Pa Eq. Acronym: Triaxial compression (two “lateral” principal stresses are equal and the distinct eigenvalue is more compressive than the lateral stresses) The Hill tensor. τ = sign [ J 2. The backstress tensor α represents the amount by which the origin has trans˜ lated. 4.4 Eq. This equals ± J 2 . and negative when closer to TXE Characteristic material response time Lode angle associated with the shifted stress tensor ξ = σ – α ( θ ξ = 30° when ξ is TXC and θ ξ = +30° ˜ ˜ ˜ ˜ when ξ is TXE).J 2 I ˜ ˜ ˜ 2 3 ˜ SI units defining equation (or page) SHR N/A page 18 TXE N/A page 15 TXC N/A page 15 T ˜ T ijkl Pa Plastic tangent stiffness tensor Signed equivalent shear stress. the yield surface origin will be at ξ = 0 ˜ ˜ instead of at the zero stress origin. and negative when J 3 < 0 . B13 .4 Eq. J 3 ] . ˜ Pa second radian (or deg in plot output) 1 Pa Eq.APPENDIX B. 3. It is positive when J 3 ≥ 0 .40 page 58 ZETA Shifted stress tensor. 3. When kinematic harden˜ ˜ ˜ ing is activated. A stress state can be in TXE even when all principal stresses are compressive — the axial stress merely needs to be less compressive than the lateral stresses. 3. Acronym: Triaxial extension: Two “lateral” principal stresses are equal and the distinct eigenvalue is more tensile than the lateral stresses. ξ = σ – α .89 Eq. defined to be the deviatoric part of S 2 and therefore given by T = S 2 – .13 τ τ τ θξ TAU RLODE Negative of τ .
021 $Dimensionless G0 = 9.0 $Dimensionless G2 = 0.47e9 $Pa B2 = 0.e4 $sec T2 = 0.0 $Pa/Meter A1 = 843. Datasets undergo continual refinement as additional material data become available.0 $1/sec T5 = 0. Dataset for Salem Limestone $ $ GeoModel parameters for Salem Limestone $ B0 = 13.e6 $Pa T1 = 4.e10 $Radians P0 = 314.0 $Pa G4 = 0.0 $1/Pa G3 = 0.02e6 $Pa A2 = 2.0 $Meters RKS = 0.e6 $Pa CTI1 = 3.0 $$$$$ = zero means A2PF defaults A4PF = 0.0e9 $Pa B1 = 42.835 $1/sec T3 = 0.0 $Pa T6 = 3.e6 $Pa CTPS = 1.0 $$$$$ = zero means A4PF defaults CRPF = 0. $Dimensionless $model 2 generic eos $ rho ref = 2300. $ K $ ref sound speed = 5400.0 $$$$$ = zero means RKPF defaults SUBX = 0.72 $Dimensionless RN = 12. so consult the model developers to obtain the latest values.APPENDIX B.0e9 $Pa B4 = 0.) NOTE TO ALEGRA USERS: The GeoModel must be run using the “Generic EOS”.22e10 $1/Pa P2 = 1.731e10 $1/Pa A3 = 821.0 $$$$$ = zero means CRPF defaults RKPF = 0.0 $Dimensionless T4 = 0.0 $Pa/Meter RKN = 0.28e18 $1/Pa^2 P3 = 0. Nomenclature and Data Sets Datasets This section cites parameters for some materials that have already been fit to the GeoModel. etc.0 $ m/s $end to to to to A2 A4 CR RK for for for for normality normality normality normality B14 .084 $strain CR = 6.4e6 $Pa P1 = 1.86e9 $Pa G1 = 0.0 $sec T7 = 0.0 $Dimensionless RJS = 0.e6 $Pa HC = 100000.92e6 $Pa A4 = 1. This section concludes with simplified datasets for mimicking simpler classical theories (Von Mises plasticity. $ kg/m^3 $ tref = 298. MohrCoulomb theory.4107e9 $Pa B3 = 12. Appropriate EOS parameters are quoted at the bottom of each dataset.0 $1/Pa J3TYPE = 3 $Dimensionless A2PF = 0.0 $Dimensionless RK = .
5e9 $Pa B2 = 0.15e19 $1/Pa^2 P3 = 0.0 $Pa T6 = 0.0 $Pa B4 = 0.69e9 $Pa G1 = 0.0 $$$$$ default = RK for $model 2 generic eos $ rho ref = 1870.0 $Pa/Meter A1 = 496.0 $end normality normality normality normality $ kg/m^3 $ K $m/s B15 .e6 $Pa T1 = 0.0 $1/Pa G3 = 0.0 $Dimensionless RJS = 0.0 $sec T2 = 0.0 $$$$$ default = A2 for A4PF = 0.e6 $Pa P1 = 1.0 $Dimensionless RK = 0. $ tref = 298.0 $Dimensionless G2 = 0.8e11 $1/Pa P2 = 2. Nomenclature and Data Sets Dataset for Sidewinder Tuff $ GeoModel parameters for Sidewinder Tuff $ Units are SI B0 = 4.0 $Pa G4 = 0.0 $$$$$ default = CR for RKPF = 0.0e9 $Pa B1 = 6.0 $Dimensionless T4 = 0.0 $1/sec T5 = 0.08 $strain CR = 15.0 $1/Pa J3TYPE = 3 $Dimensionless A2PF = 0. $ ref sound speed = 2800.83e6 $Pa A2 = 6.0 $1/sec T3 = 0.e6 $Pa CTPS = 1.0 $sec T7 = 0.e10 $Radians P0 = 70.0 $Pa CTI1 = 3.7 $Dimensionless RN = 0.0 $Dimensionless G0 = 3.0 $Pa HC = 0.1e9 $Pa B3 = 0.0 $Pa/Meter RKN = 0.0 $Meters RKS = 0.08e6 $Pa A4 = 1.APPENDIX B.0 $$$$$ default = A4 for CRPF = 0.293e10 $1/Pa A3 = 481.
0 $Dimensionless G2 = 0.0e9 $Pa B3 = 0.e6 $Pa A4 = 3.e6 $Pa P1 = 9.e14 $1/Pa P2 = 0.82e2 $Dimensionless P0 = 556. equate with yield parameters) A2PF = 0.05 $strain CR = 227. $1/Pa^2 P3 = 0.0e9 $Pa B2 = 100.0 $zero means will default to equal A4 (associative) CRPF = 0.e6 $Pa A2 = 3.0 $Pa G3 = 0.0 $Pa G4 = 0.00e9 $Pa B1 = 750.43e10 $1/Pa A3 = 1328.0 $zero means will default to equal CR (associative) RKPF = 0.0 $sec T7 = 0.0 $1/sec T3 = 0.0 $Pa T6 = 0.72 $Dimensionless RN = 17.0 $Dimensionless T4 = 0.e6 $Pa CTI1 = 30.5 $Dimensionless RK = 0.0e6 $Pa HC = 150000.0 $Dimensionless G0 = 34.e6 $Pa CTPS = 10. Nomenclature and Data Sets Dataset for lab scale intact Climax Granite $ $ GeoModel parameters for Labscale Intact Climax Granite $ B0 = 43.73e9 $Pa G1 = 0.APPENDIX B.0 $Pa B4 = 0.0 $zero means will default to equal A2 (associative) A4PF = 0.0 $Pa/Meter (Joint shear stiffness) RKN = 0.0 $1/Pa J3TYPE = 3 $dimensionless $ Flow potential parameters (for associativity.e6 $Pa T1 = 0.0 $Dimensionless RJS = 0.0 $sec T2 = 0.0 $1/sec T5 = 0.0 $Pa/Meter A1 = 1355.0 $zero means will default to equal RK (associative) $model 2 generic eos $ rho ref = 2635 $kg/m^3 $ tref = 298 $ K $ ref sound speed = 17847 $m/s $ end B16 .0 $Meters (Joint spacing) RKS = 0.
73e9 $Pa G1 = 0.51e11 $1/Pa A3 = 1328.0 $Pa B4 = 0.0 $Dimensionless T4 = 0.0 $1/Pa J3TYPE = 3 $dimensionless $ Flow potential parameters (for associativity.0 $zero means will default to equal A4 (associative) CRPF = 0.0 $Pa G4 = 0.0 $sec T7 = 0.0 $Dimensionless G2 = 0.e6 $Pa P1 = 9.00e9 $Pa B1 = 750.05 $strain CR = 227.0 $Pa G3 = 0.0 $zero means will default to equal RK (associative) $model 2 generic eos $ rho ref = 2635 $kg/m^3 $ tref = 298 $ K $ ref sound speed = 17847 $m/s $ end B17 .0 $sec T2 = 0.80 $Dimensionless RN = 17.e6 $Pa CTI1 = 30.e6 $Pa A2 = 6.0 $Dimensionless P0 = 556.e6 $Pa A4 = 0.0 $Dimensionless G0 = 34.0 $Dimensionless RJS = 0.0 $zero means will default to equal A2 (associative) A4PF = 0.5 $Dimensionless RK = 0.0 $1/sec T3 = 0.e6 $Pa T1 = 0.0e9 $Pa B2 = 100.0e6 $Pa HC = 150000.e14 $1/Pa P2 = 0.0e9 $Pa B3 = 0.0 $Pa T6 = 0. $1/Pa^2 P3 = 0.0 $zero means will default to equal CR (associative) RKPF = 0.06 $Meters (Joint spacing) RKS = 8.e6 $Pa CTPS = 10.6e11 $Pa/Meter A1 = 1379.0e10 $Pa/Meter (Joint shear stiffness) RKN = 1.APPENDIX B. equate with yield parameters) A2PF = 0.0 $1/sec T5 = 0. Nomenclature and Data Sets Dataset for field scale jointed Climax Granite $ $ GeoModel parameters for fieldscale JOINTED Climax Granite $ B0 = 43.
0 T3 = 0.1538 CR = 10.e6 T1 = 0. Nomenclature and Data Sets Dataset for 23MPa Concrete B0 = 5.0 T7 = 0.2e6 A4 = 0. RN = 3.88e20 P3 = 0.78e9 B2 = 0.0 $zero CRPF = 0.0 RJS = 0.0 G0 = 1.0e6 HC = 1. $ tref = 298.0 G4 = 0.623e9 B3 = 0.0 $end $ kg/m^3 $ K $ m/s B18 .0 $zero A4PF = 0.0 T2 = 0.0 B4 = 0. $ ref sound speed = 4500.0 $zero SUBX = 0.0 T4 = 0.0 T6 = 0.55e9 G3 = 0.9026e9 G1 = 0.0 T5 = 0.0 A1 = 1255.0 $zero RKPF = 0.0e11 CTI1 = 3e6 CTPS = 1.067e8 P1 = 7.890513 G2 = 3.0 J3TYPE=3 A2PF = 0.5e9 B1 = 28.APPENDIX B.66e10 P2 = 3.93e10 A3 = 1248.0 $Pa $Pa $Pa $Pa $Dimensionless $Pa $Dimensionless $1/Pa $Pa $Dimensionless $Meters $Pa/m $Pa/m $Pa $1/Pa $Pa $Radians $Pa $1/Pa $1/Pa^2 $Dimensionless(strain) $Dimensionless $Dimensionless $Pa $Pa $Pa $Pa $Sec $1/Sec $Dimensionless $1/Sec $Pa $Sec $1/Pa defaults defaults defaults defaults to to to to A2 A4 CR RK (associative) (associative) (associative) (associative) $model 2 generic eos $ rho ref = 2030.0 RKS = 0.0 P0 = 1.0 RK = 1.0 RKN = 0.7e6 A2 = 1.
0e10 P0 = 1.0 G4 = 0.0 G3 = 0.0 T5 = 0.0e6 T1 = 0.0 T4 = 0.0 RKS = 0.5434e9 G1 = 0.0 P3 = 0. RN = 0.0 T7 = 0.0 $zero RKPF = 0.0 $end B19 .0 T2 = 0.95520e8 P1 = 1.065714 CR = 12.0 RJS = 0.0 RKN = 0.0 $zero SUBX = 0. $(kg/cubic meter) $ tref = 298.0 HC = 0.0 B3 = 0.0954e10 B1 = 0.0e6 CTPS = 1.0 $zero A4PF = 0.0 G2 = 0.2354e9 P2 = 0.0 G0 = 7. $ ref sound speed = 3056.26455e8 A2 = 7.0 T6 = 0.0 T3 = 0.0 RK = 1. Nomenclature and Data Sets Dataset for Conventional Strength Portland Concrete B0 = 1.APPENDIX B.0 J3TYPE=3 A2PF = 0.0 $Pa $Pa $Pa $Pa $Dimensionless $Pa $Dimensionless $1/Pa $Pa $Dimensionless $Meters $Pa/m $Pa/m $Pa $1/Pa $Pa $Radians $Pa $1/Pa $1/Pa^2 $Dimensionless(strain) $Dimensionless $Dimensionless $Pa $Pa $Pa $Pa $Sec $1/Sec $Dimensionless $1/Sec $Pa $Sec $1/Pa defaults defaults defaults defaults to to to to A2 A4 CR RK (associative) (associative) (associative) (associative) $model 2 generic eos $ rho ref = 2250.0 $zero CRPF = 0.0 B4 = 0.0 B2 = 0.0 CTI1 = 3.19116e8 A4 = 1.51e10 A3 = 4.0 A1 = 4.
APPENDIX B. Nomenclature and Data Sets
Datasets for mimicking classical (simplified) models
The GeoModel is truly a unification of many simpler theories. By appropriately setting parameters, the GeoModel can be made to exactly replicate results from simpler idealized theories, as illustrated below.
Von Mises Max stress Max shear (Tresca) MohrCoulomb
30 0
20
10 0
10 0 20
0 10 20 30
30
Figure B.1. Other yield surface shapes supported by the GeoModel. sketches correspond to lines of constant z and constant θ .
The grid lines shown on these
Von Mises
Max stress
Tresca
Mohr Coulomb
Figure B.2. Classical simplified yield surfaces in the octahedral plane. None of these models adequately describes rock failure surfaces, but the failure surfaces for real rocks sometimes share some qualitative features with these models, depending on the level of confining pressure. In these figures, the axes represent projections of the compressive principal stress axes onto the octahedral plane, taking stress to be positive in tension. For the max stress and MohrCoulomb models, the size of the octahedral profile increases with pressure. For all of these models, the meridional profile is a straight line.
In most finiteelement codes, you can modify an existing data set (e.g., one for a real material) by simply redefining a material parameter in a separate input line, leaving the original value unchanged. By deviating from a correct input set to a “toy” input set in this way (rather than overwriting preferred values), you can retain a record of what the material parameters should be, thereby mitigating unintentional dissemination of physically bad input sets. The following specialized input sets use an “aprepro” syntax to show where you need to provide values. Specifically, all required or computed values appear in braces {}. Specify numerical values wherever “VALUE” appears, and then ensure all other values in braces are computed as shown (they can be computed by hand or piped into aprepro). B20
APPENDIX B. Nomenclature and Data Sets
VonMises material. Von Mises theory can be duplicated by using the following simplified set of GeoModel input values:
$ $ GeoModel parameters for replicating nonhardening VonMises plasticity $ All parameters not listed are defaulted to zero. $ Items in braces must be replaced with numbers $ A1 = {yield_stress_in_shear = VALUE} B0 = {linear_elastic_bulk_modulus = VALUE} G0 = {linear_elastic_shear_modulus = VALUE} J3TYPE = 1 $Use Gudehaus, which is capable of a circular octhedral profile RK = 1 $Set TXE/TXC ratio = 1.0 to make a circlular octahedral profile P0 = 1.e99 $make yield in hydrostatic compression impossible CTI1 = 1.e99 $ set pressure cutoff to “infinity” CTPS = 1.e99 $ set shear cutoff to “infinity” CR = 0.001 $ minimize the size of the curved part of the cap A2 = 0.0 A4 = 0.0 P3 = 0.0 HC=0.0 T1 = 0. $sec T2 = 0. $1/sec T3 = 0. $Dimensionless T4 = 0. $1/sec T5 = 0. $Pa T6 = 0. $sec T7 = 0. $1/Pa RN=0. P0=1.e11 A2PF = 0.0 A4PF = 0.0 CRPF = 0.0 RKPF = 0.0 This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
Maximum Principal Stress failure. The very simplistic fracture criterion that initiates failure when the largest principal stress reaches a critical value can be modeled in the GeoModel by using the following parameter set:
$ GeoModel parameters for duplicating a maximum principal stress criterion $ ${max_allowed_principal_stress = VALUE} B0 = {VALUE} $bulk modulus G0 = {VALUE} $shear_modulus P0 = 1.e99 $ turn off the cap function J3TYPE=3 A1={SQRT(3)*max_allowed_principal_stress} A4={1/SQRT(3.0)} RK = 0.5 A4PF = {1/SQRT(3.0)} RKPF = 0.5
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
B21
APPENDIX B. Nomenclature and Data Sets
Tresca. The simplistic criterion that a material fails when its largest shear stress reaches a critical value can be modeled by using the following GeoModel parameters:
$ User must specify values in braces B0 = {bulk_modulus} G0 = {shear_modulus} P0 = 1.e99 $ turn off the cap function J3TYPE=3 A1={2.0**yield_stress_in_shear/sqrt(3.0)} RK = 1.0 RKPF = 1.0
This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.
MohrCoulomb. Classical MohrCoulomb theory for brittle failure can be derived from an idealization that the material contains a large population of equal sized cracks. Being all the same size, any given crack loaded in pure shear will fail (grow) if the resolved shear stress τ on the crack face exceeds a critical threshold value S 0 . If a crack face is additionally subjected to a normal compressive stress σ N , then the applied shear τ needed to induce crack growth must be larger than S 0 by an amount τ fric = µσ N , where µ is the coefficient of friction. Stated differently, a given crack is safe from failure if τ – τ fric < S 0 , or, recalling that τ fric = µσ N , τ – tan φ σ N < S 0 , where tan φ ≡ µ
(B.2)
τ
e lur fai
line
S0
φ σ3 σ2 σ1
σN
Figure B.3. A stress at the limit state under MohrCoulomb theory.
(B.3)
This criterion must be satisfied by all cracks in the material. Since MohrCoulomb theory arises from an idealization that the material contains a large population of cracks (uniformly random in orientation), a material is safe from failure under general stress states only if all points on the Mohr’s diagram for the stress fall below the “failure line” defined by τ = S 0 + tan φ σ N . Failure is therefore deemed to occur when the outer Mohr’s circle first “kisses” the failure line. Working out the geometry of Fig. B.3, a circle of radius R , centered at σ N = C , will be tangent to the failure line if and only if R = S 0 cos φ + C sin φ For the outer Mohr’s circle,
(B.4)
B22
. 1cos θ – .4) gives σ3 + σ1 σ3 – σ1 .7) is I1 J2 J 2 cos θ = S 0 cos φ – .7) This is the MohrCoulomb failure criterion cast in terms of principal stresses.APPENDIX B.sin θ . 3 3 or. Nomenclature and Data Sets σ1 – σ3 R = 2 σ3 – σ1 R = 2 and σ1 + σ3 C = 2 –( σ3 + σ1 ) C = 2 (B. the Lode angle is θ = 30° so that and sin θ = sin 30° = 1 2 (B.6) Substituting these into Eq. (B.sin φ 3 3 – sin φ (MohrCoulomb in TXC) (B.sin θ sin φ . I1 S 0 cos φ – .sin φ + . S 0 cos φ – . solving for J2 .12) In triaxial compression (TXC).– . (B.43a) and (3.13) Therefore.11) specializes to TXC loading as I1 2 3 J 2 = . (3.10) (B.39) that the GeoModel’s limit function F f fits the triaxial compression (TXC) meridional profile to the following functional form J2 = a1 – a3 e –a2 I 1 + a4 I 1 3 cos θ = cos 30° = 2 in TXC ( θ = 30° ) (B. and (B.14) Being careful to note that I 1 = – I 1 . Eqs. the stress invariant version of Eq.= . (B. (B.8a) J2 σ3 + σ1 I1 .sin φ 3 J 2 = .sin φ 2 2 (B. (4.11) Recall from Eq.= 2 J 2 cos θ (B. Eq.9b) (B. comparing this result with Eq.43c) on page 24 of the main report imply σ3 – σ1 .5) or.12) implies that the GeoModel limit surface parameters for MohrCoulomb theory should be set as B23 .= S 0 cos φ – . removing the overbars (defined such that σ k = – σ k ). 2 3 3 Therefore.sin θ sin φ 3 (B.
19) This relationship will be explored in future releases of the GeoModel for allowing pressure dependence of the octahedral profile shape when (unlike MohrCoulomb theory) the meridional profile has a nonconstant slope.15d) and (B.16) 2 3 sin φ sin θ and Γ ( θ ) = . the general MohrCoulomb criterion in Eq. S 0 cos φ – . 3 – sin φ 3 Naturally.11) in TXE where θ = – 30° : I1 2 3 J 2 = . Nomenclature and Data Sets 2 3 a 1 =  S 0 cos φ 3 – sin φ a2 = 0 a3 = 0 2 3 . Therefore. J3TYPE=3 is appropriate. (B..APPENDIX B.15b) (B. cos θ – . (B. An appropriate value for the TXE/TXC strength ratio ψ must be determined by evaluating Eq.sin φ 3 + sin φ 3 (MohrCoulomb in TXE) (B.15d) Therefore.17) and (B.18) to reveal that the TXE/TXC strength ratio is coupled to the slope of the TXC meridional profile according to 1 ψ = . so kinematic hardening should be disabled as described on page B26. sin φ may be eliminated from Eqs. the slope of the TXC meridional profile. ∂ I 1 θ=30° (B. Moreover.11) may be written in the form required for the GeoModel as J 2 Γ ( θ ) = f f ( I 1 )f c ( I 1 ) .sin φ a 4 =   3 – sin φ 3 (B. (B. B24 .15c) (B. Classical MohrCoulomb theory is meant to apply to brittle rupture. to make the cap function f c equal to 1. the cap and crush curve features should be disabled as described on page B26. (B.17) The TXE/TXC strength ratio is then given by the ratio of the righthand sides of Eqs. fc ( I 1 ) = 1 (B. equals a 4 .15a) (B.18) For this classical MohrCoulomb theory. where f f ( I 1 ) = a 1 + a 4 I 1 . d J2 ⁄ d I 1 .14): 3 – sin φ ψ = 3 + sin φ (B. 1 + 3A where ∂ J 2 A ≡ .
$angle “phi” in GeoModel Appendix B $for associativity. then $ drop this file DIRECTLY into your finite element GeoModel input block. Nomenclature and Data Sets Below.0 A3 = 0.001 CRPF = 0. $ $ {friction_angle = VALUE} $ {dilation_angle = VALUE} $ $ {cohesion = VALUE} $ $parameter “S_sub0” in GeoModel Appendix B $This equals yield stress in simple shear.0} RK = {(3. This input set also shows how to set parameters if you wish to run with nonassociativity (where the flow potential function differs from the yield function PF PF only by having a different value for φ ). a 2.0sin(friction_angle))/(3.0 RN = 0.0*sqrt(3)/(3. (B. Values for ( a 1.0sin(dilation_angle))} $ B0 = {VALUE} $bulk modulus G0 = {VALUE} $shear modulus J3TYPE=3 A1 = {scalef*cohesion*cos(friction_angle)} A2 = 0. a 4 ) are set by using Eq.0 A4 = {scalef*sin(friction_angle)/3.0 P2 = 0. Values for ψ and ψ PF are set by using Eq. $ GeoModel parameters for duplicating a classical MohrCoulomb material $ in BOTH the octahedral and meridional profiles. a 3. Preferably.0+sin(dilation_angle))} P0 = 1. B25 .15). $ $ Replace every occurance of “VALUE” in this file with the appropriate $ MohrCoulomb parameter.e99 $ turn off the cap and crushcurve features P1 = 0. this equals “phi” $ Let Aprepro compute some helper quantities $ {scalef = 2.18). if $ your finite element code supports embedded aprepro directives.0 $ no cap $ no cap $ zero porosity $ prevent cap influence on shear response $ prevent cap influence on shear response $ disable kinematic hardening $ disable kinematic hardening This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters. You may generate GeoModel input values for the $ by using the command ‘aprepro this_file output_file’. we show the skeleton required for setting the GeoModel inputs to run classical MohrCoulomb theory.APPENDIX B. (B.0 CR = 0.0sin(friction_angle))} $ {scaleg = 2.0} RKPF = {(3.0+sin(friction_angle))} A1PF = {scaleg*cohesion*cos(dilation_angle)} A4PF = {scaleg*sin(dilation_angle)/3. a 4 ) and for ( a 2 .0sin(dilation_angle))/(3.0 P3 = 0.0*sqrt(3)/(3.001 HC = 0.
0 0. The GeoModel userinput processor and initializer will recognize these inputs as “cap disablers” and adjust the model appropriately. Subinput set for “turning off” all rate dependence.0 0.0 0.e99 = 1.0 0.e99 $ set pressure (I1) cutoff to “infinity” $ set principal stress cutoff to “infinity” This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.0 0.0 0.e99 0.0 RKPF=0.0 A4PF=0. P0 P1 P2 P3 CR = = = = = 1.0 0. associativity requires that all parameters that end in “PF” be set equal to their yield parameter counterparts. HC=0. CTI1 CTPS = 1.0 For LINEAR rate dependence. Nomenclature and Data Sets To conclude this appendix. B26 . Subinput set for disabling tensile cutoff limits.0 This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters. set A1 to a very large number and disable cutoffs as well as the cap and crush curve as described below.0 RN=0.APPENDIX B. B1 B2 B3 B4 G1 G2 G3 G4 = = = = = = = = 0.0 A2PF=0.0 Subinput set for associativity.0 0.0 0.0 0.0 0. set T1 to the material’s characteristic response time and all other T parameters to zero. Subinput set for disabling cap and crush curve. subinput sets are summarized for controlling various features of the GeoModel.0 0.0 0. Subinput set for disabling kinematic hardening. To additionally disable any form of plasticity. CRPF=0. If not. T1 T2 T3 T4 T5 T6 T7 = = = = = = = 0. Subinput set for linear elasticity.001 $put the cap at infinity $set porosity to zero $ minimize the size of the curved part of the cap This list presumes that your implementation of the GeoModel sets defaults for unlisted parameters.0 0.0 0.0 Set B0 to the constant bulk modulus and set G0 to the constant shear modulus.0 0.
K.F. (2001) Continuumbased FEM modeling of alumina powder compaction. M. and A. technical report. Sandia National Laboratories Report SAND20021882. constitutive model for rock mechanics structural analysis applications. Englewood Cliffs. NM. VOTH. 19 GREENBERG. Sandia National Laboratories. M. T. CHRISTON. BLANFORD. Sandia National Laboratories. . 621630 8 BOUCHERON. ARGYRIS. submitted to Elsevier Science. (1988) Convexity of yield loci for pressure sensitive materials.M. (1987) A general failure criterion for plain concrete. Dunod. Institute of Electrical and Electronic Engineers.E. R. Y. unpublished presentation materials.G.G. F. R. p.D. and M.. Geotech.R. and R.A. SAMA. Sandia National Laboratories Report SAND20022775. Methods Appl.H. Nuclear Engrg. Engng Mater. 7 11 BECKER. (1977) J. HAILL. and CHEN. FOSSUM. J. J. S. Inc. in press. R.. PETNEY.M. p. p. Sandia National Laborato13 ries. (1998) Jas3d: A multistrategy iterative code for solid mechanics analysis. Ed. L. C.. A.C.. NJ.0. K. ROBBINS.” Sandia National Laboratories Report SAND962000. DRAKE. Albuquerque.T. 9 BRANNON. D.P.L.H. WARNKE. J. KONA. GARASI. isotropic/kinematic hardening cap plasticity model for geomaterials.A. p.G. (1974) Recent developments in the finite element analysis of prestressed concrete reactor vessel.K.C. LAVIN (2004) WISDM: Weapons Information System and Data Management. 41: 4046. D. BORJA. (1994) User Instructions for the 1995 Version of the EPIC Code. ANSI/IEEE Std. KOSSI M. Solids 41:2. G. and J. Alliant Techsystems.2. Sandia Network. ROBINSON. Engr. 23 JIANG. PrenticeHall. EWSUK.R.M. FAUST. Comput.G.A.J. G.V.. S. R. SZIMMAT. 2002. BRANNON (2004) Unified compaction/dilation.V. A. K. R. NM. ARGUELLO. 10 BRANNON.D.P.. FORRESTAL. 23.E. Paris. NJ. CARROLL. ROBINSON (2003) Improved Kinematic Options in ALEGRA.J. (1978) Foundations of Applied Mathematics. ZEUCH and K. and R. SUMMERS.H. 5163. Mech. Phys. E. PIETRUSZCZAK. 12271258.. 19: 166177. strainrate sensitive. p. Sandia National Laboratories Report SAND20034510. DRUGAN (1993) Influence of nonclassical elasticplastic constitutive features on shock wave existence and spectral solutions. BUDGE. Englewood Cliffs.. NM.Z.J. PrenticeHall. NM. USA.F. 16 FOSTER. Minnesota. A. REGUEIRO. 22 JAY. and M. WONG (1996) MIG Version 0. 21 HOFFMAN. IEEE Standard Dictionary of Electrical and Electronic Terms. and W. PEERY. 4 5 6 192. R. Solids Struct.M. Sandia National Laboratories Report SAND2003XXXX.K. NM. WONG (2002) ALEGRA: User Input and Physics Descriptions Version 4. J. Technol.S..2. p. S.. Des. Albuquerque. BORJA (2004) Implicit numerical integration of a threeinvariant.M. (2004) Rotation and Reflection: a review of useful theorems associated with orthogonal tensors. Experimental Mechanics.J.F. 5. (2002) Elastic wave radiation from a pressurized spherical cavity. and W. (2003) Lawrence Livermore National Laboratories. BOSWELL. RONALDO I...A.C. Apr. 18 FUNG. Comput. (1984). LIONS (1972) Les Inequations en Mecaniquie et en Physique. and SANZ. BROWN.L. 4275. T. FOSSUM.. Hopkins. DUVAUT. M. A. 12 DIGIOVANNI. 15 FOSSUM. Albuquerque.J. R. CARROLL. 14 FARNSWORTH. 1001984. and WILLAM. NM 87185. 297330. Albuquerque. et al. Version 2. Int.A.. (2003) On the numerical integration of three invariant elastoplasticity constitutive models. Mech.I. S. K.0 Model Interface Guidelines: Rules to Accelerate Installation of Numerical Models Into Any Compliant Parent Code. C. 99. 24 JOHNSON.L. 28. Albuquerque. 20 GURSON. (1965) Foundations of Solid Mechanics. C.E. BRANNON. 17 FREW.L. 2. G. and J. D. BURNS. J. E. D.F. Albuquerque. A. PABLO F.J. A.F.A. CHEN (2001) A split Hopkinson Pressure Bar Technique to determine compressive stressstrain data for rock materials. FREDRICH (2003) private communication.REFERENCES 1 2 3 ALDRIDGE.G. ARMA/NARMS 04546.. Electronic. in process.. J. M. Unpublished.
p. 40 SANDLER. MURRY (1994) A threeinvariant smooth cap model with mixed hardening.23. (1969) Introduction to the Mechanics of a Continuous Medium. and DUNCAN. A. no. 163. Elsevier Science B. RICE (1978) Mech.C.K. A. Solids Struct. 29 LADE. J. 42 SCHWER. GULLERUD (2002) Verification of Computer Codes in Computational Science and Engineering. J. (eds.J. 49 WONG. Voyiadjis. T. (1990) Plasticity Theory. 47 WILLAM. A. Chapman & Hall/CRC. NY. June. Haill. P. 27 LUBLINER. & Anal. and D. p. L. 3. T.A.J. in Geomech. And Anal.E. L. M. In Geomech. Meth. 36 PUCIK.E. A. RUBIN (1979) An algorithm and a modular subroutine for the cap model. January.D. NM 17 April 2003.. ASCE 101. 39373956. and NAKAI. Int. Int J. PFEIFLE (1983) Nonassociative constitutive laws for low porosity rocks. Mater. 11 April 2004. D. (2004) Memorandum of Record. Mech.. Numer. 30 LADE. 43 SENSEY.P.F.R. 221240. PUCIK (1994) Nonuniqueness in dynamic rateindependent nonassociated plasticity. STEPHEN. B.L. P.A. Carroll (2001) Physics Applications in the ALEGRA Framework. W.. 28 MATSUOKA. TEUKOLSKY. P. Albuquerque.. D. Int. Material Model Driver. and Analytical Meth. ISMES Seminar on Concrete Structures Subjected to Triaxial Stresses. and A.R. Albuquerque. 48 WOLFRAM. J. 7. MacMillian.) Mechanics of Materials and Structures. et al. Robinson. (1975) Constitutive model for the triaxial behavior of concrete. 34 PELESSONE. 26 KOTERAS. Sandia National Laboratories Report SAND20001444. Macmillan. J. 35 PRESS. Geomech. P. (1992) Numerical Recipes in FORTRAN. 5970. FLANNERY. Turner. in G. Pucik Consulting Services. New York.S. NM. p.. T. T. I. NM. 101115. M (1993) Incremental kinematics for finiteelement applications. FOSSUM. 38 RASHID. (2003) SingleElement Load Paths to Complement ACTD V&V Problems 1&2. Int. Proc. J. P.V.W.A.E. p..Z. Albuquerque. Bergamo. S. and D. D. Engrg. FOSSUM. 130. FLANNAGAN (1989) PRONTO 3D a threedimensional transient solid dynamics program. (1998) Verification and Validation in Computational Science and Engineering. prepared for the Defense Nuclear Agency under Contract DNA001086C0277. . and WARNKE. The Mathematica Book.M. 18: 657688. Weatherby. 39 ROACHE.P.H. 2001. 32 MALVERN. J. C. Sandia National Laboratories Report SAND871912. (1989) A modified formulation of the cap model Gulf Atomics Report GAC19579. Impact Engr. L.S. for Num. pp. in Engr. J.V. Melbourne.25 KNUPP. and K. JACOB (1990) Plasticity Theory. Div. 10191035. 33 Englewood Cliffs. and J. New York. (1993) “A case study of instability and nonuniqueness induced by nonassociated plastic flow”. Cambridge Univ.K. PrenticeHall. I.E. 173. SALARI (2000) Code verification by the method of manufactured solutions.J.. p.P. Italy. 1999. 7. J. and T. & T. J.R. 36.. for Num.. Press. H. NM. pp. p.D. W. Northrop Grumman Information Technology unpublished presentation slides for ACTD V&V Working Group Meeting. VETTERLING. NJ NEEDLEMAN. Geotech. 31 LUBLINER.S.E. Y. Int. (1975) Elastoplastic stressstrain theory for cohesionless soil. 10371053.T. JSCE 232. FREW (2004) Penetration into low strength (23 MPa) concrete: target characterization and simulations. Albuquerque. 45 TAYLOR. Hermosa publishers.F. 46 WARREN. pp. T. 13.. (accepted)..V. Australia. 41 SANDLER. (1974) Stressdeformation and strength characteristics of soil under three different principal stresses. 37 PUCIK. First MIT Conference on Computational Fluid and Solid Mechanics. Cambridge University Press. 4th Ed.A. 44 SIMONS. Int. New York. (1977) Elastoplastic stressstrain theory for cohesionless soil with curved yield surfaces.A. and D. J.M. Numerical Meth. Unpublished Manuscript.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.