this document downloaded from Terms and Conditions of Use:

vulcanhammer.net
All of the information, data and computer software
(“information”) presented on this web site is for general
information only. While every effort will be made to insure
Since 1997, your complete its accuracy, this information should not be used or relied on
for any specific application without independent, competent
online resource for professional examination and verification of its accuracy,
suitability and applicability by a licensed professional. Anyone
information geotecnical making use of this information does so at his or her own risk
and assumes any and all liability resulting from such use.
engineering and deep The entire risk as to quality or usability of the information

foundations: contained within is with the reader. In no event will this web
page or webmaster be held liable, nor does this web page
or its webmaster provide insurance against liability, for
The Wave Equation Page for any damages including lost profits, lost savings or any
other incidental or consequential damages arising from
Piling the use or inability to use the information contained
within.
Online books on all aspects of
This site is not an official site of Prentice-Hall,
soil mechanics, foundations and Pile Buck, the University of Tennessee at
marine construction Chattanooga, or Vulcan Foundation
Equipment. All references to sources of
Free general engineering and software, equipment, parts, service
or repairs do not constitute an
geotechnical software endorsement.

And much more...

Visit our
companion site
http://www.vulcanhammer.org
 double precision
do precision,
on, intent(inout) :: ep
doub
do
double
uble precision
ub precision,
on, intent(out) :: f
doub
double
uble
le precision :: c, H
intege
integer,
ger
r, intent(out) :: regio
region,
ion, error

if (this
(this%s_type==1)
is%s_typ
is ype=
e==1) then
Elasto-Plastic Strain Hardening Mohr-Coulomb Model
erro
error=1
ror=1
ro
Derivation andreturn
retu
return
Implementation into the Finite Element Model using Principal Stress Space
else if
el f (this%s_type==2
(this
is%s_t
_typ
ype==2=2 ..or. ththis%s_t
this%s_type==3)
_typ
ype==3) thenen
allo
al
allocate(A(4,4))
locate(A
(A(4
(4,4))))
else if (this%s_t
(this%s_type==4)
_typ
ype==4)
ype=
yp e==4)
e==4
e= =4) then
=4) th
allocate(A(6,6
allocate(A(6,6))
(6,6))
(6,6
end
d if

this%getcH(ep,
call this%ge
s%ge H(ep
s%getcH(
s% ep, c, H)
call toprincipal(sigb,
topri
opri
oprinc
ncip
ipal(sig
ip igb,
b, sigbp, A, s_t
s_typ
_typei
ypein=
ein=th
this
is%s
is%s_typ
ype)
ype)
s_typein=this%s_type)

region=0
re
regi
gion
gion=0
f=this
f= is%k
%k*s
*sig
igbp
igbp(1
bp(1,1
(1,1)-
)-si
sigb
gbp(
p(3,
3,1)
3,1)-2*c*sqrt(thi
1) (thi
(this%
s%k)
k)
f=this%k*sigbp(1,1)-sigbp(3,1)-2*c*sqrt(this%k)

if (f
(f > 0) th
then
en
call t
thi
his%getregio
ion(
n(si
sigbp, ep,
this%getregion(sigbp, p, regio
ion)
n)
region)
end if

sele
lect
ct cas
select ase(reg
egio
ion)
n)
case(region)
case(0)
case
case(0)
sigc
sigcp=sigbp
gcp=
p=sigb
gbp
case(1)
case(1
(1))
call this%returntoMCP(sigbp,
this%re
return
rnto
toMCP(si
to sigb
gbp, eep, sigcp,
p, sig
igcp,
igcp
ig cp, er
error)
r)
call tthi
this%DepcMCP(sigbp,
his%De
DepcMC
MCP(
P(sigbp, sigcp
P( sigcp,
cp,
cp, ep, Depc
Depcp)
pcp)
p)
case
case(2)
se(2)
call
ca ll this%returntoMCL1(sigbp,
this%
s%returnto
toMC
MCL1(sig
MC igbp,
igbp
bp, ep
bp ep, si
sigc
sigcp,
gcp,
gcp, eerror)
or)
)
call
ll this%De
this%DepcMCL1(sigbp,
DepcMCL1(s
(sig
(sigbp,
bp, si
bp sigcp,
p, ep, DDep
Depcp)
epcp))
case(3)
call this%returntoMCL6(sigbp,
this%retu
turntoMC
tu toMC
MCL6
L6(sigbp
bp, ep, sigcp,
p, error
error)
or)
or)
call this%Depc
this%DepcMCL6(sigbp,
pcMC
MCL6(s
MC (sig
igbp, si
sigc
sigcp,
gcp, e
ep,
p, Dep
Depcp)
epcp)
ep
case
case(4)
se(4)
call this%returntoMCS(sigbp,
thi
his%return
his% rnto
rntoMC
MCS(
S(si
sigb
sigbp,
p, e
ep, sig
sigcp,
igcp, er
erro
error)
ror)
ror)
call t
ca thi
this%DepcMCS(sigbp,
his%DepcMC
hi MCS(sigbp,
MC p, sigcp
sigcp,
cp,
cp, ep
ep,
, De
Depcp)
p)
end select
ct

call
ll f
frompri
rinc
ncip
ipal
ipal(sigcp, sigc, A, D
al
fromprincipal(sigcp, Dep
epcp, Depc, s_
Depcp, s_ty
type
typein
pein=t
in=this%
=t s%s_ty
s%s_type)
s_ty
s_typein=this%s_type)

end subroutine update

subroutine
outine
tine g
get
etre
regi
gion(thi
his, sigb, e
getregion(this, ep,
p, r
reg
egion)
n)
region)
implicit none
one
class(MohrCoulomb) :: this
double pre
double precision
precision,
reci
cision, dimension(3,1
dimens
di
dimension(3,1),
nsion(3
(3,1),
,1), intent(in)
,1 intent(i
in (in) ::
: sigb
gb
double
do le pre
precision,
reci
cision, in
intent
intent(in) Emil Smed Sørensen, Aalborg University
nt(in) :: ep
double
le pre
precision,
reci
cision, di
dimens
dimension(4,1)
nsion(4
(4,1) :: BPeva
BPeval
val
double
le pre
precision
reci
cision :: c, H,
, apex
ex M.Sc. 4th Semester, 8 June 2012
intege
integer,
ger, in
intent(o
intent(out)
(out) :: r
regio
region
ion

call this%getcH(ep, c, H)
apex=2*c*sqrt(this%k)/(this%k-1)

BPeval(1,1)=dot_product(this%nL1(:,1),sigb(:,1)-apex)
BPeval(2,1)=dot_product(this%nL6(:,1),sigb(:,1)-apex)
 School of Engineering
and Science
Sohngårdsholmsvej 57
Telefon 96 35 97 31
Fax 98 13 63 93
http://ses.aau.dk

Summary:

The purpose of this report is to derive and implement

Title: Elasto-plastic Hardening Mohr- a strain hardening Mohr-Coulomb model based on re-
Coulomb Model - Derivation and Implemen- turn mapping in principal stress space by the use of
tation into the Finite Element Method Using boundary planes. The report aims at modeling strain
hardening rock material through a Mohr-Coulomb ap-
Principal Stress Space
proximation of the generalized Hoek-Brown criterion.
Theme: Master Thesis
Firstly, the classification of rock materials as well
Project period: M.Sc. 4th semester, spring as the generalized Hoek-Brown criterion are pre-
2012 sented. Afterwards follows an introduction to the
Project group: B124C Mohr-Coulomb criterion and the approximations used
Participants: for the generalized Hoek-Brown criterion.
Next, the fundamentals of plasticity and hardening is
presented along with the theory behind return map-
ping in general stress space, including the derivation
Emil Smed Sørensen
of the consistent constitutive matrix used in the global
FEM equilibrium iterations. Then the advantages of
return mapping in principal stress space is outlined.
Supervisor: Johan Clausen Following is the derivation of a non-associated
isotropic strain hardening Mohr-Coulomb model
Circulation: 3 based on the introduced theory.
Number of pages: 70 Finally, the derived model is implemented in two ex-
Submitted: 8th of June 2012 amples. The first example tries to model a strip foot-
ing while the second example models a tunnel exca-
vation. The obtained results are compared with per-
fectly plastic solutions utilizing the peak and residual
strength of the rock material.
 Contents

1 Introduction 1
1.1 Statement of Intent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Classification of Rock Materials and the Generalized Hoek-Brown Criterion 5

3 The Mohr-Coulomb Criterion 9

3.1 Mohr-Coulomb Approximation of Hoek-Brown criterion . . . . . . . . . . . . . . . . . 10

4 Plasticity Fudamentals 13
4.1 The Yield Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2 Plastic Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.3 Hardening and Softening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.4 State Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.5 Time-Independency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.6 Infinitesimal Constitutive Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.7 Multiple Yield Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Return Mapping in General Stress Space 21

5.1 Non-linear Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.2 Return Mapping Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.3 Return to One Active Yield Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.4 Return to Two Active Yield Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.5 Return to Three Active Yield Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.6 Determination of Correct Return Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

6 Return Mapping in Principal Stress Space 29

6.1 Modificaton Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.2 Boundary Planes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

7 Implementation of Strain Hardening Mohr-Coulomb Model 33

7.1 Basic Premises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

v
vi Contents

7.2 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
7.3 Yield Criterion Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
7.4 Return Regions and Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
7.5 Return Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
7.6 Consistent Constitutive Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

8 Computational Example: Strip Footing 43

8.1 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
8.2 Material Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
8.3 Mesh Coarseness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
8.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

9 Computational Example: Tunnel Excavation 51

9.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
9.2 Material Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
9.3 Mesh Coarseness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
9.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

10 Conclusion 59

Bibliography 61
 Chapter 1

Introduction

A large part of the earth’s crust consists of material which can be classified as rock. With advances
within the field of civil engineering and the ever growing need for real estate and infrastructure, more
and more structures are build in or on rock material. For some civil engineering structures, this is a
major advantage, since rock material is often very strong and stiff. Properties which are beneficial for
a foundation. However, rock material also tends to be quite brittle and posses inferior tensile strength.
Properties, which are dangerous to tunnel excavations.
Civil engineering problems involving rock material, as well as many other problems, are often hand-
led by the use of finite element modeling, where the generally non-linear governing equations of the
model are discretized into a finite number of elements, for which the solution to the governing equations
can be approximated with polynomials. Afterwards the system of equations is solved in an incremental
iterative manner until equilibrium is reached. A crucial part in the finite element method is the choice
of constitutive model, which gives the relationship between the strains and the stresses in a given point.
Part of the constitutive model is to predict when plastic straining of the material occurs, which is
dictated by the yield criterion. For rock materials, two often used yield criteria are the old-fashioned
and thoroughly tested Mohr-Coulomb criterion and the fairly new generalized Hoek-Brown criterion.
The Mohr-Coulomb criterion describes a linear relationship between the shear stress in the material
and the corresponding normal stress, which when satisfied, results in plastic straining of the material.
The Hoek-Brown criterion is an empirical non-linear refinement of the Mohr-Coulomb criterion and
is specifically designed for rock-like materials. However, due to the simplicity of the Mohr-Coulomb
criterion, many calculations regarding rock-like material is still carried out using this simpler criterion.
Another part of the constitutive model is to predict how the material behaves under plastic straining.
Generally, materials respond in three different ways, see Figure 1.1. One possibility is, that the material
strengthens during plastic loading until some ultimate strength is reached, in which case the material is
said to harden. Another possibility is, that the material maintains a constant strength, and the material
is said to be perfectly plastic. The third possibility is, that the material has a peak strength, and weakens
until a residual strength is reached, a phenomenon known as softening. The phenomenon of gaining or
losing strength during plastic loading is sometimes referred to simply as hardening, regardless that the
material is softening.
The rate of change of the strength of the material is also a significant factor, which has to be consid-
ered when modeling materials. The behavior of rock materials generally evolve in three possible ways.
Hard, good quality rock material tends to show an elastic-brittle behavior, in which the strength drops

1
2 Chapter 1. Introduction

a) Hardening b) Perfectly plastic c) Softening

Ultimate strength
Stress

Stress

Stress
Yield strength Peak strength
Initial yield strength
Residual strength

Strain Strain Strain

Figure 1.1: Material behaviour under plastic loading.

rapidly, once the material is introduced to plastic straining, see Figure 1.2. Average quality rock mate-
rial tends to show a strain softening behavior similar to the one shown in Figure 1.1c, while very poor
quality rock material shows an elastic-perfectly plastic behavior, see Figure 1.1b, [Hoek and Brown,
1997]. Many finite element models rely on the material to behave in a perfectly plastic manner with a

Elastic-Brittle
Peak strength
Stress

Residual strength

Strain

Figure 1.2: Elastic-brittle behavior in hard rock material

yield strength equivalent to the residual strength. This greatly reduces the problem and thus makes it
easier to solve. However, it could also possibly lead to over sized structures if the material is softening,
because local zones of high plastic straining dictate the strength of the entire material. For example if a
perfectly plastic approach is used with a strength above the residual strength, material located in plastic
zones is stronger than it should be, and hence the model is on the unsafe side. On the other hand, if the
residual strength is used for the entire material, the model might be too conservative.
Constitutive models based on the Mohr-Coulomb criterion which include hardening/softening are
already available, see e.g. de Souza Neto et al. [2008]. However, a model based on the principal stress
space framework presented by Clausen [2007] has yet to be developed.
1.1. Statement of Intent 3

1.1 Statement of Intent

The aim of this project is to derive a strain hardening/softening constitutive model for use in finite
element calculations based on the Mohr-Coulomb criterion which utilize derivations in principal stress
space, both regarding the updated stress state and the consistent constitutive matrix needed for the
global equilibrium iterations.
To test and demonstrate the usefulness of the model, it is used to estimate the influence of the
hardening/softening properties on the bearing capacity of a strip footing as well as the risk of failure
during a tunnel excavation.

1.2 Prerequisites
Strains and stresses are tensors of the 2nd order and the constitutive relation between them is a 4th
order tensor. However, symmetric properties of the strain and stress tensors allow for a formulation in
which they can be expressed equally accurate as vectors, and the constitutive relation can be expressed
as a matrix. In this report, the latter formulation will be used due to its simplicity and ease of use when
writing computer code. Throughout the report, a number of variables, vectors and matrices are used.
To keep track of these, a number of guidelines will be presented in the following.
A scalar is presented in ordinary text as σ1 , whereas a vector or a matrix is symbolized in bold as
e.g. σ or D . By default, vectors are 6 × 1 and matrices are 6 × 6. Vectors and matrices with an overline,
e.g. σ and D are related to the principal stress components and have dimensions of 3 × 1 and 3 × 3
respectively. Vectors and matrices with a tilde, e.g. σ̃σ and T̃
T are related to the shear stress components
and have dimensions of 3 × 1 and 3 × 3 respectively. Vectors and matrices with a hat, e.g. σ̂ σ and D̂
D are
full 6 × 1 vectors and 6 × 6 matrices, where the axes are aligned with those of the principal stresses.
The ordering of the strain vector, ε , and the stress vector, σ , is given as

ε = [ε x εy εz 2ε xy 2ε xz 2εyz ]
T
(1.1)

σ = [σ x σy σz σ xy σ xz σyz ]
T
(1.2)
Stresses are taken as positive in tension unless otherwise stated. D is the elastic constitutive matrix
relating elastic strains to stresses and is given by
⎡ ⎤
⎢D 0 ⎥
⎢ 3x3⎥
D=⎢ ⎥
⎢ 0 G̃ ⎥
(1.3)
⎢3x3 G ⎥
⎣ ⎦
G are given by
where D and G̃
⎡1 − ν ν ⎤
⎢ ⎥
⎢ ⎥
ν
D=
E ⎢ 1−ν ν ⎥
⎢
(1 + ν)(1 − 2ν) ⎢ ⎥
⎥
ν (1.4)
⎢ ν 1 − ν⎥
⎣ ν ⎦
⎡1 0 0⎤
⎢ ⎥
⎢ ⎥
G=
E ⎢0 1 0⎥
⎢
2(1 + ν) ⎢ ⎥
⎥
G̃ (1.5)
⎢0 0 1⎥
⎣ ⎦
E is Young’s modulus and ν is Poisson’s ratio.
 Chapter 2

Classification of Rock Materials and the

Generalized Hoek-Brown Criterion

Rock material is a wide expression used to describe a solid made up of minerals. Depending on the size
of a rock sample, see Figure 2.1, rock material typically ranges from isotropic intact rock mass without
any discontinuities, through very anisotropic rock mass with a few dominating discontinuities, to an
isotropic jointed rock mass with an indistinct number of randomly oriented evenly space discontinuities
with the same characteristics [Merifield et al., 2006].

Structure

Intact rock

Single discontinuity

Two discontinuities

Several
discontinuities

Jointed rock mass

Figure 2.1: The material which is modeled, should be isotropic compared to the size of the
structure.

If a representative sample of the rock material, which is sought modeled, can be regarded as either
intact rock or as jointed rock mass, then the material can be regarded as isotropic, provided that the
sample size is small compared to the structure at hand. If the discontinuities of the sample are oriented
in a non-random order, it might be necessary to model the rock material as an anisotropic continuum. If
large fractures(faults) dominate the construction site of the structure, it may also be necessary to include

5
6 Chapter 2. Classification of Rock Materials and the Generalized Hoek-Brown Criterion

such fractures in the model mesh. In the following, it is assumed, that the rock material can be modeled
as an isotropic continuum.
In order to be able to include rock material in finite element models, the properties of the rock
material need to be known and somehow quantified. Extensive empirical research has lead to the
formulation of the generalized Hoek-Brown criterion, equation (2.1), which predict the stress states
that cause failure in rock materials [Hoek and Brown, 1997].
a
σ′3
σ′1 = σ′3 + σci (mb + s) (2.1)
σci

σ′1 and σ′3 are the major and minor effective principal stresses respectively, where compression is taken
as positive. As the criterion suggests, four parameters are needed in order to asses the strength of
the rock material, namely the uniaxial compressive strength of the intact rock material, σci , and the
constants mb , s and a. The constants can be estimated based on the Geological Strength Index(GSI),
the disturbance factor, D, and the intact rock material constant, mi , by using the following expressions
[Hoek et al., 2002]
GSI − 100
mb = mi exp( )
28 − 14D
(2.2)
GSI − 100
s = exp( )
9 − 3D
(2.3)
−GSI −20
a = + (exp( ) − exp( ))
1 1
(2.4)
2 6 15 3
The Geological Strength Index is a measure of the rock material’s quality based on field observations,
which takes into account the composition and structure of the in-situ rock material as well as the surface
conditions, see Figure 2.3 on page 8. Based on this, the GSI is assigned on a scale ranging from 0 to
100, where 100 indicates a very good quality [Hoek, 2007].
The disturbance factor, D, is used to take into account the blast damage, that part of the rock mate-
rial might suffer from. It ranges from 0 to 1, where 0 indicates undisturbed rock material. The material
constant mi and the uniaxial compressive strength of the intact rock material, σci , is found using labo-
ratory tests on the intact rock material. The elastic modulus of the rock material can be estimated by
[Hoek and Diederichs, 2006]

1 − D/2
Erm = 100,000 MPa( )
1 + exp((75 + 25D −GS I)/11)
(2.5)

Once the rock material has reached a stress state which causes failure, it loses some of its strength,
as mentioned in chapter 1. The manner in which the strength drops is not entirely determined, but
three possible characteristics are mentioned in Hoek and Brown [1997]. One possibility is to assume an
elastic-brittle behavior, where the strength of the rock material rapidly drops to some residual strength
once the failure criteria is reached, see Figure 1.2. Another possibility is to assume a strain soften-
ing relationship between the strength of the material and the plastic straining which it undergoes, see
Figure 1.1c. The third options is to assume that the rock material exhibits in a elastic-perfectly plastic
way, see Figure 1.1b. In this report, it is assumed that the rock material behaves in a strain-softening
manner. For an implementation of an elastic-perfectly plastic generalized Hoek-Brown criterion see
Clausen [2007] and Sørensen [2012].
Chapter 2. Classification of Rock Materials and the Generalized Hoek-Brown Criterion 7

In order to conform with most finite element codes, where tension is taken as positive, the gener-
alised Hoek-Brown criterion can be expressed as
σ1 a
σ3 = σ1 − σci (s − mb ) (2.6)
σci
where the apostrophes signifying effective stresses have been omitted for simplicity. In order to express
the above as a yield function, resulting in a negative number for elastic states and a positive number for
non-allowable states, it can further be rewritten to the following

f (σ
σ,σci , s,mb ,a) = σ1 − σ3 − σci (s − mb
σ1 a
) =0 (2.7)
σci
The stress states which are solutions to the above equation form a six sided pyramid along the hydro-
static axis with curved sides as can be seen in Figure 2.2. Any stress state inside the pyramid is elastic,
whereas any stress state located outside is unobtainable.

σ3

Hydrostatic axis, σ1 = σ2 = σ3

σ2

σ1

Figure 2.2: The generalized Hoek-Brown criterion visualized in principal stress space.
8 Chapter 2. Classification of Rock Materials and the Generalized Hoek-Brown Criterion

GEOLOGICAL STRENGTH INDEX FOR

JOINTED ROCKS (Hoek and Marinos, 2000)

Slickensided, highly weathered surfaces with compact

Slickensided highly weathered surfaces with soft clay

From the lithology, structure and surface condi-

Smooth, moderately weathered and altered surfaces

tions of the discontinuities, estimate the

Rough, slightly weathered, iron stained surfaces

average value of GSI. Do not try to be too
precise. Quoting a range from 33 to 37 is more
realistic than stating that GSI = 35. Note that

coatings or fillings or angular fragments

Very rough, fresh unweathered surfaces
the table does not apply to structurally
controlled failures. Where weak planar
structural planes are present in an unfavourable
orientation with respect to the excavation face,

SURFACE CONDITIONS
these will dominate the rock mass behaviour.
The shear strength of surfaces in rocks that are
prone to deterioration as a result of changes in

coatings or fillings
moisture content will be reduced if water is

VERY GOOD
present. When working with rocks in fair to

VERY POOR
very poor categories, a shift to the right may be
made for wet conditions. Water pressure is

GOOD

POOR
FAIR
dealt with by effective stress analysis.

STRUCTURE DECREASING SURFACE QUALITY

INTACT OR MASSIVE - intact rock

specimens or massive in situ rock with
90 N/A N/A
few widely space discontinuities
DECREASING INTERLOCKING OF ROCK PIECES

80
BLOCKY - well interlocked undis-
turbed rock mass consisting of cubical 70
blocks formed by three intersecting
discontinuity sets
60
VERY BLOCKY - interlocked,
partially disturbed mass with multi-
faceted angular blocks formed by 4 or
50
more joint sets

BLOCKY/DISTURBED/SEAMY - 40
folded with angular blocks formed by
many intersecting discontinuity sets.
Persistence of bedding planes or
30
schistosity
DISINTEGRATED - poorly
interlocked, heavily broken rock mass
20
with mixture of angular and rounded
rock pieces

LAMINATED/SHEARED - lack of 10
blockiness due to close spacing of weak N/A N/A
schistosity or shear planes

Figure 2.3: Geological strength index for jointed rock masses [Marinos and Hoek, 2000].
 Chapter 3

The Mohr-Coulomb Criterion

Even though the Hoek-Brown criterion is specifically developed with rock materials in mind, a lot of the
finite element models made today still utilize the much older Mohr-Coulomb failure criterion, which
states, that once the shear stress, τ, and the normal stress, σn , is reached in a plane in a continuum,
which satisfies
τ = c − σn tan(ϕ) (3.1)
yielding occurs. c is a measure of the cohesion present in the material, while tan(ϕ) is a measure of
the friction coefficient between the grains, thus ϕ is denoted the friction angle. In the above expression,
tension is taken as positive. This linear relationship can be visualized as shown in Figure 3.1.

τ = c − σn tan(ϕ) τ

σ3 σ2 σ1 c ϕ σn

Figure 3.1: The Mohr-Coulomb criterion in σn -τ space.

Any Mohr circle situated below the Mohr-Coulomb line denotes an elastic state, whereas a Mohr
circle, which touches the line denotes a state of yielding. From Figure 3.1, the shear stress causing

σ1 − σ3
failure can be shown to be
τ= cos(ϕ) (3.2)
2
where σ1 and σ3 are respectively the largest and smallest principal stresses. Similarly, the normal stress

σ1 + σ3 σ1 − σ3
on the failure plane is given by
σn = + sin(ϕ) (3.3)
2 2

9
10 Chapter 3. The Mohr-Coulomb Criterion

Substitution back into (3.1) and rewriting results in

σ1 − σ3 + (σ1 + σ3 )sin(ϕ) = 2ccos(ϕ) (3.4)

If the implicit function above is plotted in principal stress space, it is forming a six-sided pyramid along
the hydrostatic axis as shown on Figure 3.2. This shape represents the yield surface, and it is evident
from the figure, that the criterion is pressure dependent. Any stress state situated inside the stress space
bounded by the six planes is elastic, while stress states outside are non-allowable.

σ3

Hydrostatic axis, σ1 = σ2 = σ3

σ2

σ1

Figure 3.2: The Mohr-Coulomb criterion plotted in principal stress space.

Written as a yield function, the Mohr-Coulomb criterion takes the following formulation

f (σ
σ,c,ϕ) = σ1 − σ3 + (σ1 + σ3 )sin(ϕ) − 2ccos(ϕ) = 0 (3.5)

which can be refined to give

f (σ
σ,σc ,k) = kσ1 − σ3 − σc = 0 (3.6)
where k is given by
1 + sin(ϕ)
k=
1 − sin(ϕ)
(3.7)

and the uniaxial compressive strength, σc , is given by

√
σc = 2c k (3.8)

3.1 Mohr-Coulomb Approximation of Hoek-Brown criterion

In order to use the Mohr-Coulomb criterion to model rock materials, a method of relating the Mohr-
Coulomb parameters to the rock properties obtained using the Hoek-Brown criterion needs to be iden-
tified. According to Hoek et al. [2002] the friction angle can be calculated using

6amb (s + mb σ3n )a−1

ϕ = sin (
−1
) (3.9)
2(1 + a)(2 + a) + 6amb (s + mb σ3n )a−1
3.1. Mohr-Coulomb Approximation of Hoek-Brown criterion 11

and the corresponding cohesion is given by

σci ((1 + 2a) s + (1 − a)mb σ3n )(s + mb σ3n )a−1

c= √ (3.10)
(1 + a)(2 + a) 1 + (6amb (s + mb σ3n )a−1 )/((1 + a)(2 + a))

where σ3n is given by

σ3n =
σ3,max
(3.11)
σci
and σ3,max is the upper bound over which the Mohr-Coulomb criterion is sought fitted, cf. Figure 3.3.
Notice that compression is taken as positive in these derivations.

Approximated Mohr-Coulomb criterion Hoek-Brown Criterion σt

σ1

σ3,max
σ3

Figure 3.3: Mohr-Coulomb approximation of Hoek-Brown criterion. Compression is taken as

positive.

The upper boundary of the stress range should be chosen based on the problem at hand, such that it
covers the stress range of the model. For deep tunnels, the following relationship gives a good estimate
[Hoek et al., 2002]
−0.94
= 0.47( )
σ3,max σcm
(3.12)
σcm γH
where
(mb + 4s − a(mb − 8s))( m4b + s)
a−1
σcm = σci
2(1 + a)(2 + a)
(3.13)

and γ is the unit weight of the rock mass, and H is the depth of the tunnel below the surface. For slope
stability, the following gives a good estimate of σ3,max
−0.91
= 0.72( )
σ3,max σcm
(3.14)
σcm γH

A more general approach is to define σ3,max as [Rocscience Inc., 2007]

σ3,max =
σci
(3.15)
4
 Chapter 4

Plasticity Fudamentals

In this chapter, some of the basics of material plasticity is outlined. However, a detailed description
is beyond the scope of this report. For a more thorough exposition, see de Souza Neto et al. [2008],
Ottosen and Ristinmaa [2005] and Crisfield [2000].

4.1 The Yield Function

The strains that develop within a material when exposed to a load can basically be divided into two
separate parts. Part of the strains are what is known as elastic strains. Theses strains are characterized
by the fact that once the external load disappears, so does the elastic strains that developed during
loading. The part of the strains which are not elastic are known as plastic. These strains remain even
after the material has been unloaded. See Figure 4.1.
Stresses, σ

E
1 Strains, ε
εp εe

Figure 4.1: Elastic and plastic strains.

So the total strains, ε , are made up of elastic strains and plastic strains, which can be written as

ε = ε e +εε p (4.1)

where ε e is the elastic strain vector and ε p is the plastic strain vector. Plastic strains start to develop
once the material reaches its yielding limit, which is defined by some yield function f = 0. This could,
for example, be the Hoek-Brown criterion or the Mohr-Coulomb criterion discussed earlier. The yield

13
14 Chapter 4. Plasticity Fudamentals

function, f , is a function of the stresses as well as some hardening parameters, K , which describe the
strength of the material, i.e.
f = f (σ K)
σ,K (4.2)
Sometimes, a material might require more than one yield function in order to be modeled sufficiently
accurate, this is discussed in section 4.7. The hardening parameters are usually determined by some
state parameters, κ , that determine the internal state of the material
K = K (κκ ) (4.3)
The yield function is a scalar valued function, which gives a negative value for all stress states that are
elastic. Once the yield function reaches a value of zero, plastic strains start to develop. The stress states
which fulfill this criterion form a surface in stress space known as the yield surface, see e.g. Figure 2.2
and 3.2. Further, the yield function remains zero during plastic loading, which implies that the time
derivative of f during plastic loading is zero, which can be written as
∂ f T dσ ∂ f T ∂K
= +( ) +( ) =0
d f ∂ f dt σ K dκκ
(4.4)
dt ∂t dt ∂σσ dt K
∂K ∂κκ dt
which is known as the consistency relation. Since the yield function is time-independent, it simplifies
to
∂ f T ∂K
= aT +( ) =0
df dσσ K dκκ
(4.5)
dt dt ∂KK ∂κκ dt
where a is given by
a=
∂f
(4.6)
∂σσ
The time-dependency is discussed further in section 4.5. A stress state which returns a positive value
of the yield function is inadmissible. The stress state within the material is determined by the elastic
strains through the constitutive matrix, D , as
σ = Dε e = D (εε −εε p ) (4.7)
where (4.1) has been used. If no plastic straining has occurred in the material, the relationship between
stresses and strains is one-to-one. I.e. it is possible to determine the stress state based on the total
strains, which merely consist of elastic strains. However, if plastic straining has developed within the
material, the one-to-one relationship is lost, see Figure 4.2. The stress state is said to be path-dependent.

σa
Stresses, σ

σb

Strains, ǫ
ǫx

Figure 4.2: The one-to-one relationship between strains and stresses are lost once plastic
strains have developed.
4.2. Plastic Potential 15

Because of this path-dependence, it is necessary to adopt an incremental approach in order to find

matching strain-stress relations. This is done by taking the time derivative of (4.7), which can be
written as
dεε dεε p
= D = D( − )
dσσ dεε
(4.8)
dt dt dt dt

4.2 Plastic Potential

Once the yield function reaches zero and plastic strains start to develop, it is crucial to know in which
direction they develop. However there is no conclusive way to determine this. A way to get around
this, is to define a plastic potential function, g. The plastic potential is a scalar valued function, which
usually depends upon the stress state and some hardening parameters

g = g(σ K)
σ,K (4.9)

The partial derivative of this plastic potential with respect to the stresses define the direction of the
plastic strains. A common choice for the plastic potential is to use the yield function. If this is the case,
it is referred to as associated plasticity. If another function is chosen, it is referred to as non-associated
plasticity. The length of the incremental plastic strain is controlled by a so called plastic multiplier, dλ,
which is a non-negative scalar. Thus the plastic strain increment is given by
dεε p dλ ∂g dλ
= = b (4.10)
dt σ dt
dt ∂σ
where the abbreviation b has been introduced to improve readability. This relation is known as the flow
rule.

4.3 Hardening and Softening

As mentioned earlier, rock material tend to lose some of its strength once plastic straining occurs,
which is known as softening. However many metals tend to show an increase in strength during plastic
straining, see Figure 1.1a, which is known as hardening. Usually both phenomena are simply referred
to as hardening. If the material is considered perfectly plastic, the yield criterion is independent of the
hardening parameters K , and simply reduces to

f (σ K ) = f (σ
σ,K σ) = F (σ
σ) = 0 (4.11)

where F has been introduced for readability and designates a perfectly plastic yield function. If harden-
ing is employed in the model, two distinct methods are normally chosen, namely the isotropic hardening
model and the kinematic hardening model, see Figure 4.3. The isotropic hardening model expands or
contracts the yield surface, which can be achieved by adding or subtracting an appropriate amount from
the yield criterion based upon the state variables κ . However, the position and shape of the yield surface
in stress space is unaltered. This can be expressed as

f (σ K ) = F (σ
σ,K σ) − K = 0 (4.12)

On the other hand, kinematic hardening shifts the yield surface from one location in stress space to
another. The size and shape of the yield surface remains unaltered. This can be achieved by shifting the
16 Chapter 4. Plasticity Fudamentals

stresses by some amount defined by the state variables as

f (σ K ) = F (σ
σ,K σ −K) = 0 (4.13)

The two different hardening models can be used simultaneously, in which case it is referred to as mixed
hardening. Mixed hardening alters the size and position of the yield surface and leaves the shape
unaltered. This can be written as

f (σ K ) = F (σ
σ,K σ − K kin ) − K iso = 0 (4.14)

where K kin and K iso are the hardening parameters associated with kinematic hardening and isotropic
hardening respectively.

F (σ
σ − K kin ) − K iso

F (σ
σ −K)
F (σ
σ) − K

F (σ
σ) F (σ
σ) F (σ
σ)

Figure 4.3: Isotropic, kinematic and mixed hardening.

4.4 State Parameters

The state parameters which control the hardening of the material need to be identified and their time rate
of change has to be established, the so-called evolution law. The two most common state parameters
are the accumulated plastic strain, denoted, ε p , and the dissipated plastic work, W p , defined by
εp
Wp = ∫ σ T dεε p (4.15)
0

The accumulated plastic strain can be defined in different manners, in which the most common is the
Von Mises accumulated plastic strain defined by
√
2 dεε p T dεε p
ε =∫ ( )
t
p
dt (4.16)
0 3 dt dt
Alternatively, the state parameters can also be defined by some potential function, j, which is a function
of the stress state and the hardening variables

j = j(σ K)
σ,K (4.17)
4.5. Time-Independency 17

and a plastic multiplier, using the following expression

=−
dκκ dλ ∂ j
(4.18)
dt K
dt ∂K
For instance, if the state parameter is the accumulated plastic strain, ε p , and the hardening parameter is
the cohesion, c, the increment of the accumulated plastic strain is given as

dε p
=−
dλ ∂ j
(4.19)
dt dt ∂c
If j is assumed equal to f , the evolution law is said to be associated, and if j is different from f , the
evolution law is said to be non-associated.

4.5 Time-Independency
As can be seen from the above equations, there are a lot of first order time derivatives, which represent
the load rate of the problem. If a solution is sought, which is independent of the load rate, these time
rate of changes can simply be thought of as changes in the variables which are being differentiated. For
example, the time rate of change of the plastic strains

dεε p
(4.20)
dt
can be replaced with
dεε p (4.21)
and thought of as a nothing more than an infinitesimal change in the plastic strains, regardless of time.
By adopting this independency, the consistency relation, equation (4.5), can be written as

∂ f T ∂K
d f = a T dσ
σ+( ) dκκ = 0
K
(4.22)
K
∂K ∂κκ
The stress increment, equation (4.8), can be written as

σ = D (dεε − dεε p )
dσ (4.23)

The flow rule, equation (4.10), can be written as

dεε p = dλbb (4.24)

and finally, the evolution law defined by a potential function, equation (4.18), can be written as

dκκ = −dλ
∂j
(4.25)
K
∂K

4.6 Infinitesimal Constitutive Matrix

The infinitesimal constitutive matrix, D ep , relates infinitesimal strain increments with infinitesimal
stress increments as follows
σ = D ep dεε
dσ (4.26)
18 Chapter 4. Plasticity Fudamentals

Combining the consistency relation, equation (4.22), the infinitesimal stress increment, equation (4.23),
the plastic flow rule, equation (4.24), and the evolution law, a solution for the infinitesimal increment
of the plastic multiplier, dλ, can be found. If the hardening law is assumed to be defined by a potential
function, j, as in equation (4.25), dλ is found to be

a T D dεε
dλ = (4.27)
a T Db + ( ∂K
K)
∂f T
K ∂j
∂K
∂κκ ∂K
K

If this solution is substituted back into equation (4.23), the infinitesimal constitutive matrix can be found
to be
Dba T D
D ep = D − (4.28)
a T Db + ( ∂K
K )
∂ f T ∂K
K ∂j
∂κκ ∂K
K

4.7 Multiple Yield Functions

Some yield criteria might consist of multiple yield functions

f1 (σ K ), f2 (σ
σ,K K ), ..., fn (σ
σ,K K)
σ,K (4.29)

Each yield function defines a surface in stress space. In this case, the elastic stress states are bounded
by the stress states which return a negative value of all the yield functions. See Figure 4.4

f1 = 0

f1 < 0
Discontinuity

f1 < 0 ∩ f2 < 0

f2 = 0
Discontinuity f2 < 0

Figure 4.4: The elastic stress states (blue) of a yield criterion with multiple yield functions
(green).

In these cases, the combined yield surface contains intersections between the individual yield func-
tions, which require special attention. These intersections can be visualized as curves and points in
principal stress space and is known as yield curves and yield points, see e.g. Figure 2.2 and Figure 3.2.
These intersections usually result in discontinuities where the surfaces of the yield functions intersect,
see Figure 4.4. If multiple yield functions are utilized, each yield function typically have a unique plas-
tic potential, gi , and hardening potential, ji . In such cases, the strain direction at a discontinuous part
of a yield criterion, see Figure 4.5, is a linear combination of the different strain directions involved
4.7. Multiple Yield Functions 19

[Koiter, 1953]
k
dεε p = ∑ dλib i (4.30)
i=1

where k is the number of plastic potentials, that is part of the intersection at hand. Similarly, the
evolution law is given by
k
dκκ = − ∑ dλi
∂ ji
(4.31)
i=1 K
∂K

b2
dεε
f2

b1

f1 < 0 ∩ f2 < 0 f1

Figure 4.5: The plastic strain direction at a discontinues part of the yield criteria.
 Chapter 5

Return Mapping in General Stress

Space

In this chapter, the theory behind return mapping is introduced. However, a short introduction to the
non-linear finite element method is given first, in order show the need and applicability of return map-
ping. For a more detailed description of the theory behind return mapping and finite element methods,
see de Souza Neto et al. [2008], Cook et al. [2002] and Crisfield [2000]. The derivations of this chapter
rely on a evolution law of the form given by equation (4.25) and (4.31).

5.1 Non-linear Finite Element Method

Problems involving the displacement and stress distribution throughout a model can be formulated
as partial differential equations made up of the governing equations behind the problem and some
boundary conditions, which make the model unique. However, for complex models, an analytical
solution to these boundary value problems is very hard or simply impossible to establish. Because
of this, the problem is sought solved through numerical integration, which is where the finite element
method comes into play.
As the name suggests, the model is discretized into a finite number of elements, for which the
solution to the governing equations can be approximated with polynomials. A large range of different
elements exist, each with advantages and disadvantages, however this is beyond the scope of this report.
Based on this discretization, the stiffness of the entire model can be calculated. Because the stiffness
of the model is non-linear and path dependent, the boundary conditions are applied incrementally in
what is known as load steps. The system of equations is solved iteratively in each load step, to make
sure that equilibrium is fulfilled. Usually by the use of a Newton-Raphson scheme. This process can be
schematized as shown in Table 5.1. The highlighted points of the procedure are material dependent and
is the main focus of this report. The updated stress state should ideally be found through and integration
of the infinitesimal elasto-plastic constitutive matrix along the path of the strain increment as

σ k = σ k−1 + ∫ σ = σ k−1 + ∫
ε k−1 +∆εε ε k−1 +∆εε
dσ D ep dεε (5.1)
ε k−1 ε k−1

where equation (4.26) has been used. However, the integration of equation (5.1) is no easy task, since
the strain path is unknown and D ep is stress dependent. Several methods exist, which try to circumvent
this problem. Return mapping is one of these methods, and is the method used throughout this report.

21
22 Chapter 5. Return Mapping in General Stress Space

ÿ Load step k = 1,2,...

○ p k = p k−1 + ∆ppk Initiation of the kth load vector
○ ∆uu1 = 0 Initiation of the displacement increment
○ σk = σk−1 Initiation of start guess of σk
○ ÿ Global equilibrium iterations j = 1,2,...
○ ○ r = p k −qq (σ σk ) Force residual, r , from p k and internal forces, q
○ ○ K (D D )
epc
Form the global tangent stiffness matrix, K
○ ○ δuu = K −1r Solve the FEM equations
○ ○ ∆uu j+1 = ∆uu j + δuu Update displacement increment
○ ○ ∆εε = B ∆uu j+1 Calculate strain increment
○ ○ σ k (σ σk−1 ,∆εε) Update stresses
○ ○ D (σ k )
epc σ
Update consistent constitutive matrix
○ Stop iterations when ∥rr ∥ < ǫ∥ppk ∥ ǫ is a prescribed tolerance
○ u k = u k−1 + ∆uu j+1 Update displacement vector
○ ε k = Bu k Update strain vector
End of load step

Table 5.1: Schematic of the incremental nature of the non-linear finite element method and
the Newton-Raphson procedure used in the global equilibrium iterations. Based on [Clausen,
2007]

Further, if the infinitesimal constitutive matrix, D ep , relating infinitesimal strain increments with
infinitesimal stress increments, is used in the global equilibrium iteration scheme of the finite ele-
ment code, where finite increments are used, the quadratic convergence of the problem will be lost
[Nagtegaal, 1982]. Because of this, a consistent constitutive matrix, D epc , is developed, which main-
tains the quadratic convergence by relating infinitesimal changes of the finite strain increments with
infinitesimal changes of the finite stress increments

σ = D epc d∆εε
d∆σ (5.2)

5.2 Return Mapping Basics

The fundamental idea of return mapping is to try out, whether the entire strain increment, ∆εε, is elastic
by introducing the elastic predictor stress increment

σe = D ∆εε
∆σ (5.3)

Adding this, to the initial stress state, see Figure 5.1,

σ A = σ k−1 (5.4)

the predicted elastic stress state, σ B , becomes

σ B = σ A + ∆σ
σe (5.5)
5.2. Return Mapping Basics 23

σB

σe
∆σ σp
-∆σ
σC f (σ KC ) = 0
σC ,K
σ
∆σ
σA
σ
f (σ K A) = 0
σ A ,K

Figure 5.1: Sketch of the updating scheme.

If σ B returns a negative value of the yield function, i.e. f (σ

σ B ) ≤ 0, the entire strain increment is purely
elastic, and no further calculations are needed. However, if the predicted stress state falls outside the
yield surface, a part of the strain increment must consist of plastic strains. According to equation (4.23),
the stress increment must be given by

σ=∫ D (dεε − dεε p ) = D ∆εε − D ∆εε p

ε k−1 +∆εε
∆σ (5.6)
ε k−1

where use has been made of the fact, that D is independent of ε , because linear elasticity is assumed.
Introducing the plastic corrector stress increment, see Figure 5.1

σ p = D ∆εε p
∆σ (5.7)

together with equation (5.3), the stress increment is given by

σ = ∆σ
∆σ σe − ∆σ
σp (5.8)

and the final updated stress state, σC , becomes

σC = σ A + ∆σ
σ (5.9)

which can also be written as

σC = σ B − ∆σ
σp (5.10)
σ p,
Using equation (4.24), the plastic strain increment, ∆εε p , used in calculating the plastic corrector, ∆σ
is given by
λ+∆λ
∆εε p = ∫ bdλ (5.11)
λ

The integration of equation (5.11) is just as complicated as equation (5.1), however, in the return map-
ping framework, the plastic strain increment is approximated with

∆εε p ≈ ∆λ b ∣C (5.12)
24 Chapter 5. Return Mapping in General Stress Space

σ p , can be written as
which results in the plastic corrector increment, ∆σ

σ p ≈ ∆λD
∆σ D b ∣C (5.13)

and thus the problem boils down to finding the updated stress state, σC , which fulfills equation (5.10)
and lies on the yield surface. If the updated stress state belongs to a single active yield function, cf.
equation (4.24), the plastic corrector increment is given as shown above. However, if the updated stress
state belongs to an intersection of two or more yield functions, the plastic strain direction is given by
equation (4.30). Because of this, slightly different return mapping procedures have to be deployed,
depending on the number of active yield functions that the updated stress state, σC , belongs to.

5.3 Return to One Active Yield Function

The updated stress state, σC , belongs to the yield surface defined by the yield function hence

f (σ KC ) = 0
σC ,K (5.14)

where K C are the updated hardening variables

K C = K (κκC ) (5.15)

and κ C are the updated state parameters. In case of a hardening law based upon a potential function,
this could be written as follows
κ C = κ A − ∆λ ∣
∂j
(5.16)
K C
∂K
In order to find the correct updated stress state and the plastic multiplier, equation (5.10) and (5.14) are
solved using an iterative procedure, for instance a Newton-Raphson procedure, which is used in this
text.

5.3.1 Consistent Constitutive Matrix

D epc is derived by taking the total derivative of (5.8) with respect to ∆εε, using (5.3) and (5.13) as follows

= − ⋅ − ⋅
d∆σσ dD D∆εε ∂∆λD Db d∆λ ∂∆λD σ
Db d∆σ
(5.17)
d∆εε d∆εε ∂∆λ d∆εε σ d∆εε
∂∆σ
Multiplying with d∆εε on both sides yields

σ = D d∆εε − Db d∆λ − ∆λD

∂bb
d∆σ D σ
d∆σ (5.18)
σ
∂σ
Rearranging leads to
∂bb −1
σ = (II + ∆λD
d∆σ D ) D (d∆εε − d∆λbb) (5.19)
σ
∂σ
which can be written on the form
σ = D c d∆εε − d∆λD
d∆σ D cb (5.20)
where

Dc = T D (5.21)
−1
T = (II + ∆λD )
∂bb
D (5.22)
σ
∂σ
5.4. Return to Two Active Yield Functions 25

T is known as the modification matrix. Using the consistency condition, (4.22), an expression for d∆λ
can be found in much the same way as it was found in (4.27), and substituted back into (5.20), which
gives the consistent constitutive matrix as

D cba T D c
D epc = D c − (5.23)
a T D cb + ( ∂K
K)
∂f T
K ∂j
∂K
∂κκ ∂K
K

If a is equal to b , it is seen, that D epc is symmetric.

5.4 Return to Two Active Yield Functions

If the yield criterion consists of two yield functions, f1 and f2 , with the appertaining plastic potentials
g1 and g2 and the hardening potentials j1 and j2 , it is possible, that the updated stress state belongs to
the intersection of these two yield surfaces, see Figure 4.4. If this is the case, the direction of the plastic
strains is given by equation (4.30), and thus the corrector stress is also a linear combination of the stress
directions involved giving
∆σσ p = ∆λ1 D b 1 ∣C + ∆λ2 D b 2 ∣C (5.24)
where

b1 =
∂g1
(5.25)
σ
∂σ
b2 =
∂g2
(5.26)
σ
∂σ
(5.27)

Similarly, the hardening law, equation (5.16) expands to

κ C = κ A − ∆λ1 ∣ − ∆λ2 ∣
∂ j1 ∂ j2
(5.28)
K
∂K C K C
∂K
when using the hardening potential method. The updated stress state belonging to the yield curve still
needs to fulfill (5.10) as well as f1 (σ K C ) = 0 and f2 (σ
σC ,K K C ) = 0. This results in eight equations with
σC ,K
eight unknowns, namely σC , ∆λ1 and ∆λ2 . To find the updated stress state, σC , an iterative procedure
is implemented in which the residual, r , of equation (5.10) is defined by

r (σ
σC ,∆λ1 ,∆λ2 ) = σC − (σ
σ B − ∆λ1 D b 1 ∣C − ∆λ2 D b 2 ∣C ) = 0 (5.29)

Expanding r in a first order Taylor series leads to

r (σ
σCi+1 ,∆λ1,i+1 ,∆λ2,i+1 ) =rr (σ σ,∆λ1,i + d∆λ1 ,∆λ2,i + d∆λ2 )
σCi + dσ

=rr (σ
σCi ,∆λ1,i ,∆λ2,i ) +
(5.30)
σ+ d∆λ1 +
∂rr ∂rr ∂rr
dσ d∆λ2
σ
∂σ ∂∆λ1 ∂∆λ2
where
σC ∂σ σ B ∂∆λ1 Db 1 ∂∆λ2 Db 2
= − + +
∂rr ∂σ
(5.31)
σ ∂σ
∂σ σ ∂σσ ∂σσ σ
∂σ
= I + ∆λ1 D + ∆λ2 D
∂bb1 ∂bb2
(5.32)
∂σσ ∂σσ
26 Chapter 5. Return Mapping in General Stress Space

and

= Db 1
∂rr
(5.33)
∂∆λ1
= Db 2
∂rr
(5.34)
∂∆λ2
substituting back into (5.30) yields

r (σ
σCi+1 ,∆λ1,i+1 ,∆λ2,i+1 ) =rr (σ
σCi ,∆λ1,i ,∆λ2,i ) + (II + ∆λ1 D + ∆λ2 D )dσ
σ+
∂bb1 ∂bb2
∂σσ ∂σσ (5.35)
+ Db1 ⋅ d∆λ1 + Db2 ⋅ d∆λ2
and solving for r (σ
σCi+1 ,∆λ1,i+1 ,∆λ2,i+1 ) = 0 gives
∂bb2 −1
σ = (II + ∆λ1 D + ∆λ2 D ) (−rr (σ σCi ,∆λ1,i ,∆λ2,i ) − Db 1 d∆λ2 − Db 2 d∆λ2 )
∂bb1
dσ
∂σσ ∂σσ (5.36)
T r (σ
= −T σCi ,∆λ1,i ,∆λ2,i ) − D cb 1 d∆λ2 − D cb 2 d∆λ2
where
∂bb2 −1
T = (II + ∆λ1 D + ∆λ2 D )
∂bb1
(5.37)
σ
∂σ ∂σσ
Having an initial guess of σCi , ∆λ1,i and ∆λ2,i , a Taylor expansion of the two yield criteria results in
f1 (σ K Ci+1 ) = f1 (σ
σCi+1 ,K σCi + dσ K)
K Ci + dK
σ,K
∂ f1 T (5.38)
= f1 (σ K Ci ) +aaT1 dσ
σCi ,K σ+( ) dK
K
K
∂K
where

K= d∆λ1 +
K ∂κκ
∂K ∂KK ∂κκ
dK d∆λ2
∂κκ ∂∆λ1 ∂κκ ∂∆λ2
=− d∆λ1 −
∂KK ∂ j1 ∂KK ∂ j2
d∆λ2 (5.39)
∂κκ ∂K
K ∂κκ ∂K
K
which gives
f1 (σ K Ci ) = f1 (σ
σCi+1 ,K K Ci+1 ) −aaT1 T r (σ
σCi ,K σCi ,∆λ1,i ,∆λ2,i ) −aaT1 D cb 1 d∆λ2 −aa1 D cb 2 d∆λ2 −
∂ f1 T ∂K (5.40)
( ) ( d∆λ1 + d∆λ2 )
K ∂κκ K ∂κκ
∂K
K
∂K ∂κκ ∂∆λ1 ∂κκ ∂∆λ2
and similarly for f2
f2 (σ K Ci ) = f2 (σ
σCi+1 ,K K Ci+1 ) −aaT2 T r (σ
σCi ,K σCi ,∆λ1,i ,∆λ2,i ) −aaT2 D cb 1 d∆λ2 −aa2 D cb 2 d∆λ2 −
∂ f2 T ∂K (5.41)
( ) ( d∆λ1 + d∆λ2 )
K ∂κκ ∂KK ∂κκ
∂KK ∂κκ ∂∆λ1 ∂κκ ∂∆λ2
Equating (5.40) and (5.41) with 0, leads to two equations with two unknowns, namely d∆λ1 and d∆λ2
which can be found. Once d∆λ1 and d∆λ2 are obtained, dσ σ can be found using (5.36), which leads to
C
a new σ i+1 . Further ∆λ1 and ∆λ2 are updated by
∆λ1,i+1 = ∆λ1,i + d∆λ (5.42)
∆λ2,i+1 = ∆λ2,i + d∆λ (5.43)
And new values of d∆λ1 and d∆λ2 can again be found. The above-mentioned steps are repeated until
satisfactory precision is reached.
5.4.1. Consistent constitutive matrix 27

5.4.1 Consistent constitutive matrix

The consistent constitutive matrix of a point belonging to two active yield functions is found in much the
same was as it was found for the point belonging to one yield function, namely by taking the derivative
of (5.8) and utilizing (5.24)

= − ⋅ − ⋅ − ⋅ − ⋅
σ dD
d∆σ D∆εε ∂∆λ1 Db 1 d∆λ1 ∂∆λ1 Db 1 d∆σσ ∂∆λ2 Db 2 d∆λ2 ∂∆λ2 Db 2 d∆σσ
(5.44)
d∆εε d∆εε ∂∆λ1 d∆εε σ
∂∆σ d∆εε ∂∆λ2 d∆εε σ
∂∆σ d∆εε
which can be rewritten to
σ = T D (d∆εε − d∆λ1b 1 − d∆λ2b 2 )
d∆σ (5.45)
Using the consistency condition of both yield criteria together with equation (5.36) and (5.39), results
in
∂ f1 T
σ+(
aT1 d∆σ ) dK
K =aaT1 T D (d∆εε − d∆λ1b1 − d∆λ2b2 )−
K
∂K
(5.46)
∂ f1 T ∂K
( ) ( d∆λ1 + d∆λ2 ) = 0
K ∂ j1 K ∂ j2
∂K
∂KK ∂κκ ∂K
K ∂κκ ∂K
K
∂f T
σ + ( 1 ) dK
a T2 d∆σ K =aaT2 T D (d∆εε − d∆λ1b 1 − d∆λ2b 2 )−
K
∂K
(5.47)
∂ f2 T ∂K
( ) ( d∆λ1 + d∆λ2 ) = 0
K ∂ j1 K ∂ j2
∂K
∂KK ∂κκ ∂K
K ∂κκ ∂K
K

⎡ T c ⎤ ⎡ ⎤ ⎡ ⎤
This can also be written as
⎢a 1 D ∆εε⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ − A ⎢dλ1 ⎥ = ⎢0⎥
⎢ T c ⎥ ⎢ ⎥ ⎢ ⎥
⎢a 2 D ∆εε⎥ ⎢dλ2 ⎥ ⎢0⎥
(5.48)
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
⎡ ⎤
where
⎢A11 A12 ⎥
A=⎢
⎢
⎥
⎥
⎢A21 A22 ⎥
(5.49)
⎣ ⎦
∂ fi T ∂K
Aik = a Ti D cb k + ( )
K ∂ jk
(5.50)
K
∂K ∂κκ ∂K
K
Thus, dλ1 and dλ2 can be found to be
⎡ ⎤ ⎡ ⎤
⎢dλ1 ⎥ ⎢ ⎥
⎢ ⎥ = B ⎢a 1 D d∆εε⎥
c
⎢ ⎥ ⎢ ⎥
⎢dλ2 ⎥ ⎢a 2 D d∆εε⎥
(5.51)
⎣ ⎦ ⎣ ⎦
c

⎡ ⎤
where
⎢ B11 B12 ⎥
B = A −1 = ⎢
⎢
⎥
⎥
⎢ B21 B22 ⎥
(5.52)
⎣ ⎦
Substituting back into (5.45) gives

σ = D c (d∆εε − B11a T1 D c d∆εεb 1 − B12a T2 D c d∆εεb 1

d∆σ
−B21a T1 D c d∆εεb 2 − B22a T2 D c d∆εεb 2 )
(5.53)

Using (5.53), D epc can be derived to

D epc =D
Dc − B11b 1a T1 D c − B12b 1a T2 D c − B21b 2a T1 D c − B22b 2a T2 D c (5.54)
28 Chapter 5. Return Mapping in General Stress Space

which can also be written as

2 2
D epc = D c − ∑ ∑ Bi jb ia Tj D c (5.55)
i=1 j=1

If needed, the infinitesimal constitutive matrix is found by replacing D c in (5.55) with D .

5.5 Return to Three Active Yield Functions

An updated stress state might also be returned to the intersection of three yield surfaces, f1 , f2 and f3 ,
with the plastic potentials g1 , g2 and g3 and the hardening potentials j1 , j2 and j3 . This scenario is very
similar to the scenario with two active yield surfaces, which was discussed in the previous section, and
will only be touched upon briefly. The plastic corrector is given by

σ p = ∆λ1 D b 1 ∣C + ∆λ2 D b 2 ∣C + ∆λ3 D b 3 ∣C

∆σ (5.56)

and the evolution law is assumed to be given by

κ C = κ A − ∆λ1 ∣ − ∆λ2 ∣ − ∆λ3 ∣

∂ j1 ∂ j2 ∂ j3
(5.57)
K C
∂K K C
∂K K C
∂K
The return algorithm is almost identical to the one mentioned in the previous section, except that an
extra unknown, ∆λ3 needs to be found, which is possible because of the extra equation introduced
by the consistency condition of the third yield criterion. The derivation of this procedure is omitted,
however the modification matrix, T , is given by

∂bb3 −1
T = (II + ∆λ1 D + ∆λ2 D + ∆λ3 D )
∂bb1 ∂bb2
(5.58)
∂σσ ∂σσ ∂σσ
Similarly, the consistent constitutive matrix can be found to be given by
3 3
D epc = D c − ∑ ∑ Bi jb ia Tj D c (5.59)
i=1 j=1

where B is the 3 × 3 equivalent matrix to the one in the previous section.

5.6 Determination of Correct Return Type

In the general six-dimensional stress space, there is no easy way of determining, which of the above
mentioned return algorithms, that should be applied to a certain predictor stress. Because of this, a
commonly used strategy is to start out with returning to a single yield surface. The updated stress state
is then evaluated based upon some specific requirements. In case these requirements are not met, the
predictor stress is returned using a return to two yield surfaces and so on. In general stress space, it is
theoretically possible, that an updated stress state has to be returned using as much as six active yield
surfaces. However, in the three dimensional principal stress space, geometric arguments can be applied,
to establish which method is to be applied.
 Chapter 6

Return Mapping in Principal Stress

Space

If the material in question is isotropic, the stress states within the material can be expressed in principal
stresses through a coordinate transformation, see Figure 6.1. Thus the updated stress state can be

Coordinate transformation

σ = [σ x σy σz σ xy σ xz σyz ]
T

σ = [σ1 σ2 σ3 ]
T

Figure 6.1: Coordinate transformation of general stress space into principal stress space

found in principal stress space, and only the three principal stresses need to be found. And since
only three values are involved, the problem can be visualized in 3D space, where advantages can be
made of geometrical arguments, which can be used to simplify the expressions used in chapter 5. The
drawback is, that coordinate transformation calculations have to be performed. The predictor stress,
σ B , is transformed into principal stresses using ordinary coordinate transformation. Afterwards, the
updated principal stress state, σC , and the consistent constitutive matrix aligned with the principal axes,
epc
D , is found and then transformed back into the original coordinate system of the model, cf. Table 6.1.
D̂
The derivations of chapter 5 should of course still hold in principal stress space for an isotropic
material. However, σ is reduced from the six components of (1.2) to only three components, namely

σ = [σ1 σ2 σ3 ]
T
(6.1)

where the overbar, , is used to indicate, that we are dealing with principal stresses.

29
30 Chapter 6. Return Mapping in Principal Stress Space

○ σB → σB Transform predicted stress state into principal stresses

○ σC (σ σB) Find the updated principal stress state
○ D (σ σC )
epc
D̂ Find consistent constitutive matrix aligned with principal axes
○ σ → σC
C
Transform updated principal stress state to general stresses
aligned with model axes
○ → D epc
epc
D
D̂ Transform consistent constitutive matrix into general stresses
aligned with model axes

Table 6.1: Schematic of return mapping in principal stress space.

6.1 Modificaton Matrix

The modification matrix used in finding the consistent constitutive matrix, T , is still created as a full
6×6 matrix, however, it is aligned with the principal stress axes, meaning that σ x is in the same direction
as σ1 , σy is in the same direction as σ2 and σz is in the direction of σ3 . This is denoted by a hat, ˆ . T̂
T

⎡ ⎤
is divided into two parts
⎢T 0 ⎥
T =⎢⎢
⎥
⎥
⎢ 0 T̃ ⎥
T̂ (6.2)
⎣ T ⎦
T relates to the shear stresses. For a return to a yield surface
where T relates to the principal stresses and T̃
using (5.22) results in
−1
T = (II + ∆λD
D )
∂bb
(6.3)
∂σσ
and for a return to two yield surfaces using (5.37) results in
−1
T = (II + ∆λ1 D + ∆λ2 D )
∂bb1 ∂bb2
(6.4)
∂σσ ∂σσ

T is given by [Clausen et al., 2006]

and so on. Further, T̃

⎡ σC1 −σC2 ⎤
⎢ σB −σB ⎥
⎢ 1 2 ⎥
⎢ ⎥
T =⎢
⎢
⎥
⎥
σC1 −σC3
⎢ ⎥
T̃ (6.5)
⎢ σC2 −σC3 ⎥
σ1B −σ3B
⎢ ⎥
⎣ σ2 −σ3 ⎦
B B

epc
T is found, the consistent constitutive matrix aligned with the principal axes, D̂
Once T̂ D , is calculated
c
analogous to equation (5.23), (5.55), or (5.59), except that a , b and D is replaced by

âa = [a T 0 0 0]
T
(6.6)

b̂b = [b 0 0 0]
T T
(6.7)
D = T̂
c
D̂ TD (6.8)
6.2. Boundary Planes 31

6.2 Boundary Planes

As mentioned earlier, the principal stress space is a three dimensional space in which the yield criterion
can be visualized. Further, from (5.13), it is known, that if a predicted stress state should be returned to
σ p , is given by
a surface, the direction, s , see Figure 6.2, of the plastic corrector, ∆σ

s = Db (6.9)

b
f (σ K) = 0
s
σC ,K

σC
K) = 0
σC ,K
g(σ

Figure 6.2: Direction of the plastic corrector of a return with one active yield function

By evaluating this expression along the boundaries of the yield surface, the predictor stress states,
which can be returned to the yield surface is outlined. The principle is shown for a Mohr-Coulomb
criterion on Figure 6.3. The stress space within these boundaries is known as a return region, belonging
to the specific yield surface.

σ3

s
σ2
σ1

s
s

Figure 6.3: Outlining of a return region by the use of boundary planes.

32 Chapter 6. Return Mapping in Principal Stress Space

Similarly, the direction of the plastic corrector for a predictor stress which is to be returned to two active
yield surfaces, see Figure 6.4, is given by

s = ηss1 + ρss2 (6.10)

b2 s 2
g1 (σ K) = 0
σC ,K b1
s1
σC

f1 (σ K) = 0
σC ,K
g2 (σ K) = 0
σC ,K

f2 (σ K) = 0
σC ,K

Figure 6.4: Direction of the plastic corrector of a return with two active yield functions is a
linear combination of the two plastic corrector directions involved

where η and ρ are some arbitrary positive scalars. Similar arguments apply for a predictor stress with
three active yield surfaces and so on. With this knowledge, it is possible to determine which part of
the yield criterion, a specific predictor stress state should be returned to. Further, if the plastic potential
is linear, the different return regions are made up of planes, since s is independent of the position in
yield space. With this knowledge, it is possible to determine the correct return algorithm, without the
trial-and-error approach of the general stress space updating scheme.
 Chapter 7

Implementation of Strain Hardening

Mohr-Coulomb Model

In this chapter, the theory of the previous chapters will be applied to a Mohr-Coulomb model using
linear elasticity, non-associated plasticity and isotropic strain hardening, along with the evolution laws
of equation (4.25) and (4.31).

7.1 Basic Premises

As mentioned in chapter 3, the Mohr-Coulomb yield criterion can take the form of
f (σ
σ,σc ,k) = kσ1 − σ3 − σc = 0 (7.1)
which will be used in the current implementation due to its simplicity compared to (3.5). The yield
criterion is a function of both the friction angle, ϕ, and the cohesion c. In this implementation, it is
assumed, that the friction angle remains constant. Thus, the hardening parameters vector, K , simplifies
to a scalar, namely the cohesion, c. Further the state parameters vector, κ , of the material, is chosen to
be the scalar accumulated plastic strain, ε̄P . Thus
K (κκ ) = c(ε̄P ) (7.2)
The plastic potential, g, is chosen as
g = (σ
σ,c,ψ) = σ1 − σ3 + (σ1 + σ3 )sin(ψ) − 2ccos(ψ) (7.3)
where ψ represents the angle of dilation. The evolution of the accumulated plastic strain is given by the
hardening potential function, j, cf. equation (4.25), and is chosen as
j = (σ
σ,c,ϕ) = σ1 − σ3 + (σ1 + σ3 )sin(ϕ) − 2ccos(ϕ) (7.4)

7.2 Derivatives
The derivative of f with respect to σ is given by
⎡k⎤
⎢ ⎥
∂f ⎢ ⎥
a= =⎢ 0⎥
σ ⎢ ⎥
⎢
⎥
(7.5)
⎢−1⎥
∂σ
⎣ ⎦

33
34 Chapter 7. Implementation of Strain Hardening Mohr-Coulomb Model

The derivative of f with respect to the harding variable, c, is given by

√
= = = −2 k
∂f ∂f ∂ f dσc
(7.6)
K ∂c ∂σc dc
∂K
The derivative of g with respect to σ is given by
⎡ 1 + sin(ψ) ⎤
⎢ ⎥
∂g ⎢ ⎥
b= =⎢
⎢
⎥
⎥
σ ⎢ ⎥
0 (7.7)
⎢−1 + sin(ψ)⎥
∂σ
⎣ ⎦
The derivative of j with respect to the hardening variables is given by

= = −2cos(ϕ)
∂j ∂j
(7.8)
K ∂c
∂K
Finally, the derivative of the hardening parameters, K , with respect to the state parameters, κ , is given
by
= P =H
K
∂K ∂c
(7.9)
κ
∂κ ∂ε̄
where H is the gradient of the chosen ε̄P − c-curve at the current point of accumulated plastic strain.
This curve could be modeled as a function, however, in the current implementation, it is defined by
a number of predefined (ε p ,c) points, which makes it possible to choose an arbitrary curve, without
changing the computer code . The specific value of c and H is interpolated between these points, see
Figure 7.1. Using this approach, the model is able to handle all of the hardening models shown in
Figure 1.1 as well as the elastic brittle behavior shown in Figure 1.2.

1
c

0 εp

Figure 7.1: Example of arbitrary ε̄P − c-curve defined by a set of (ε p ,c) points.

7.3 Yield Criterion Regions

As shown in Figure 3.2, the Mohr-Coulomb yield criterion consists of six yield surfaces in principal
stress space. However, if by definition, σ1 is the major principal stress, and σ3 is the minor principal
7.3. Yield Criterion Regions 35

stress when transforming the general stress state into principal stress space, it is only necessary to
operate with the yield surface, that satisfies this condition, named f1 , and the two neighboring yield
surfaces, f2 and f6 . See Figure 7.2.

σ3
σ3 ≥ σ1 ≥ σ2 σ3 ≥ σ2 ≥ σ1

f5 f4

σ1 ≥ σ3 ≥ σ2 f6 f3 σ2 ≥ σ3 ≥ σ1

σ1 f1 f2 σ2

σ1 ≥ σ2 ≥ σ3 σ2 ≥ σ1 ≥ σ3

Figure 7.2: The Mohr-Coulomb criterion seen from the hydrostatic axis. The primary yield
surface, f1 , is shown in blue, while the remaining yield surfaces are faded out.

f1 , f2 and f6 is given by

f1 (σ
σ,σc ,k) = kσ1 − σ3 − σc = 0 (7.10)
f2 (σ
σ,σc ,k) = kσ2 − σ3 − σc = 0 (7.11)
f6 (σ
σ,σc ,k) = kσ1 − σ2 − σc = 0 (7.12)

Ie. σ1 is switched with σ2 for yield surface f2 , and σ3 is switched with σ2 for yield surface f6 . This
leads to the following derivatives
⎡k⎤ ⎡0⎤ ⎡k⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
a1 = ⎢
⎢0⎥
⎥ a2 = ⎢
⎢k⎥
⎥ a6 = ⎢ ⎥
⎢−1⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
(7.13)
⎢−1⎥ ⎢−1⎥ ⎢0⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
The index swapping is also valid for the plastic and hardening potentials belonging to these yield sur-
faces. Thus, the plastic strain direction is given by
⎡ 1 + sin(ψ) ⎤ ⎡ ⎤ ⎡ 1 + sin(ψ) ⎤
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
0
b1 = ⎢
⎢
⎥
⎥ b2 = ⎢ ⎥
⎢ 1 + sin(ψ) ⎥ b6 = ⎢
⎢−1 + sin(ψ)⎥
⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
0 (7.14)
⎢−1 + sin(ψ)⎥ ⎢−1 + sin(ψ)⎥ ⎢ ⎥
⎣ ⎦ ⎣ ⎦ ⎣ 0 ⎦
The intersection between yield surface f1 and f2 is a line in principal stress space denoted l1 , see
Figure 7.3. Along this line σ1 = σ2 and is thus given by
⎡ σ ⎤
⎢ ⎥
⎢ ⎥
σ l1 = ⎢ σ1 ⎥
⎢
1

⎥
⎢ ⎥
(7.15)
⎢kσ − σ ⎥
⎣ 1 c⎦
36 Chapter 7. Implementation of Strain Hardening Mohr-Coulomb Model

σ3
a

r l6
l6 l1
r l1

σ1 σ2

f1

Figure 7.3: Naming convention for the Mohr-Coulomb criterion.

Similarly, the intersection between f1 and f6 is denoted l6 where σ2 = σ3 , and is given by

⎡ σ ⎤
⎢ ⎥
⎢ ⎥
σ l6 = ⎢kσ1 − σc ⎥
⎢
1

⎥
⎢ ⎥
(7.16)
⎢kσ − σ ⎥
⎣ 1 c⎦

The intersection of f1 , f2 and f6 is a point in principal stress space, and is denoted a, given by

⎡1⎤
⎢ ⎥
σc ⎢ ⎥
⎢1⎥
σa =
k−1 ⎢ ⎥
⎢
⎥
(7.17)
⎢1⎥
⎣ ⎦

Any predicted stress state falling outside the yield criterion is to be returned to either the primary yield
surface, f1 , the yield lines, l1 or l6 , or the apex, a.

7.4 Return Regions and Boundaries

The region of principal stress space, that returns to f1 is denominated R f1 . Stress points which are to
be returned to l1 is denominated Rl1 and similarly for Rl6 and Ra . Due to the linearity of the Mohr
Coulomb criterion, the different return regions are bounded by planes. Based on the predictor stress
state’s location relative to these planes, the correct return region can be established and hence the
correct return algorithm can be applied. A plane can be defined by it’s normal vector, n , and a point
belonging to the plane, x 0 , as
n T (xx − x 0 ) = 0 (7.18)
Any vector, x , for which the above is satisfied is situated on the plane. A point lying below the plane
gives a negative number, and a point lying above the plane results in a positive number. Thus, it is
necessary to identify the normals of each plane, which can be calculated based on the directions, in
which the planes span.
7.4. Return Regions and Boundaries 37

The plastic corrector direction belonging to a surface return to yield surface f1 is given by
⎡1 + sin(ϕ) − 2ν⎤
⎢ ⎥
⎢ ⎥
s 1 = Db 1 = −
E ⎢ 2ν sin(ϕ) ⎥
⎢
(1 + ν)(2ν − 1) ⎢ ⎥
⎥
(7.19)
⎢2ν − 1 + sin(ϕ)⎥
⎣ ⎦
and the intersection between the f1 -yield surface and the f2 -yield surface is fully determined by σl1 ,
(7.15), which by differentiation gives the direction of the intersection line, r l1 , see Figure 7.3,
⎡1⎤
⎢ ⎥
σl1 ⎢ ⎥
r l1 = =⎢1⎥
∂σ
⎢
∂σ1 ⎢ ⎥ ⎥
(7.20)
⎢k ⎥
⎣ ⎦
By taking the cross product between r l1 and s 1 , the normal of the plane separating the return region
belonging to yield surface f1 , and those belonging to line l1 can be established as

n R f1 →Rl1 = s 1 ×rr l1 (7.21)

where the arrow designates, that the normal of the plane is pointing from the region belonging to f1 , to
the region belonging to l1 . Similarly, the direction of l6 is given by
⎡1⎤
⎢ ⎥
σl6 ⎢ ⎥
r l6 = =⎢ k⎥
∂σ
∂σ1 ⎢ ⎥
⎢
⎥
(7.22)
⎢k ⎥
⎣ ⎦
and thus the normal of the plane which creates the boundary between the region of predictor stresses
belonging to f1 and those belonging to l6 can be found to give

nRl6 →R f1 = s1 ×rr l6 (7.23)

The boundary plane separating Rl1 from Ra is spanned by the direction of s 1 and s 2 , which is the plastic
corrector direction belonging to f2 . Thus

nRl1 →Ra = s1 × s2 (7.24)

and similarly for the boundary plane which separates Rl6 from Ra

n Rl6 →Ra = s 6 × s 1 (7.25)

In order to completely define the boundary planes, a point on each plane is also needed. Since all the
planes go through the apex of the criterion, this point is simply chosen to represent all four boundary
planes. Based upon this, four boundary planes, see Figure 7.4, can be defined by

pR f1 →Rl1 (σ
σ B ) = n TR f →Rl1 (σ σa ) = 0
σ B −σ (7.26)
1

pRl6 →R f1 (σ
σ B ) = n TRl →R f1 (σ σa ) = 0
σ B −σ (7.27)
6

pRl1 →Ra (σ
σ B ) = n TRl →Ra (σ
σ B
σa ) = 0
−σ (7.28)
1

pRl6 →Ra (σ
σ B ) = n TRl →Ra (σ
σ B
σa ) = 0
−σ (7.29)
6

Using these boundary planes, a rule set can be set up, which determines the correct return algorithm
based upon the evaluation of these planes, which has been done in Table 7.1.
38 Chapter 7. Implementation of Strain Hardening Mohr-Coulomb Model

Rule # Conditions Return to

1 pR f1 →Rl1 (σ
σ ) ≤ 0 ∧ pRl6 →R f1 (σ
σB σ )≥0
σB f1
2 pR f1 →Rl1 (σ
σ B ) ≥ 0 ∧ pRl1 →Ra (σ
σB) ≤ 0 l1
3 pRl6 →R f1 (σ
σ ) ≤ 0 ∧ pRl6 →Ra (σ
σB σ )≤0
σB l6
4 pRl1 →Ra (σσ ) ≥ 0 ∧ pRl6 →Ra (σ
B σB) ≥ 0 a

Table 7.1: Rule set for retun algorithms. See Figure 7.4 for further details of boundary plane
location and naming.

σ3
pRl6 →Ra
pRl1 →Ra

pRl6 →R f1 pR f1 →Rl1

σ1 σ2

Figure 7.4: Boundary planes of the Mohr-Coulomb criterion. The visible side of each bound-
ary plane in the figure is defined as the positive direction of the normal.

7.5 Return Algorithms

The return algorithms used for the Mohr-Coulomb criterion are particularly simple. Because of the
linearity of the plastic potential, which has been utilized, the plastic strain direction is independent of
the stresses. This means, that the plastic strain direction is the same, whether it is evaluated at the
predictor stress point, σ B , or at the updated stress point, σC . Thus, equation (5.13) can be simplified to

σ p ≈ ∆λD
∆σ D b ∣B (7.30)

σ p , is the plastic multiplier ∆λ.

This means, that the only unknown in calculating the plastic corrector, ∆σ

7.5.1 Return to yield surface f1

If the predictor point qualifies for a return to f1 , only one yield surface is active, and hence only one
plastic multiplier needs to be found. The updated stress state, σC , is given as

σC = σ B − ∆σ
σ p = σ B − ∆λD
Db 1 (7.31)

which have to satisfy the condition

f1 (σ
σC ,σCc ,k) = kσC1 − σC3 − σCc = 0 (7.32)
7.5.2. Return to yield lines l1 and l6 39

With the use of equation (7.31), σC1 and σC3 can be expressed as

σC1 = σ1B − ∆λs1,1 (7.33)

σC3 = σ3B − ∆λs1,3 (7.34)

where s1,1 and s1,3 is the first and third component of s 1 respectively. Further, with the use of equation
(5.16), the uniaxial compressive strength of the updated stress state, σCc , depends on the accumulated
plastic strain at the updated stress state, ε p,C , which is given as

ε p,C = ε p,A − ∆λ = ε p,A + ∆λ2cos(ϕ)

∂j
(7.35)
∂c
where equation (7.8) has been used. Thus, the compressive uniaxial strength of the updated stress state,
σCc , is given by √
σCc = 2c(ε p,C ) k (7.36)
Substituting back into equation (7.32) gives

f1 (∆λ) = k (σ1B − ∆λs1,1 ) − (σ3B − ∆λs1,3 ) − σCc = 0 (7.37)

which is solved using an ordinary Newton-Raphson iteration procedure with respect to ∆λ. The gradient
of the equation with respect to ∆λ is
dσC dcC dε p,C
= −ks1,1 + s1,3 − Cc = −ks1,1 + s1,3 + dσc
d f1
(7.38)
d∆λ dc dε p,C d∆λ

√
where
dσCc dcC dε p,C
dσc = − p,C
= 4H cos(ϕ) k (7.39)
dcC dε d∆λ
using equation (7.9), (7.35) and (7.36). An initial guess of ∆λ is made and then updated via
d f1 −1
∆λi+1 = ∆λi − ( ∣ ) f1 (∆λi ) (7.40)
d∆λ i
until the required precision is reached.

7.5.2 Return to yield lines l1 and l6

Returning to one of the yield lines, l1 and l6 is a simple expansion of the procedure used for the f1
return. However, in this case, the plastic corrector is given by

σC = σ B − ∆σ
σ p = σ B − ∆λ1 Db 1 − ∆λ1 Db 2 (7.41)

where ∆λ1 and ∆λ2 are unknown. σC has to fulfill both yield criteria. For the l1 return, this results in

f1 (σ
σC ,σCc ,k) = kσC1 − σC3 − σCc = 0 (7.42)
f2 (σ
σC ,σCc ,k) = kσC2 − σC3 − σCc = 0 (7.43)

where

σC1 = σ1B − ∆λ1 s1,1 − ∆λ2 s2,1 (7.44)

σC2 = σ2B − ∆λ1 s1,2 − ∆λ2 s2,2 (7.45)
σC3 = σ3B − ∆λ1 s1,3 − ∆λ2 s2,3 (7.46)
40 Chapter 7. Implementation of Strain Hardening Mohr-Coulomb Model

The accumulated plastic strain of the updated stress state is given by

2
ε p,C = ε p,A − ∑ ∆λi = ε p,A − 2cos(ϕ)(∆λ1 + ∆λ2 )
∂ ji
(7.47)
i=1 ∂c

The two yield criteria are embedded in the residual vector F as

⎡ ⎤
⎢ f1 (∆λ1 ,∆λ2 )⎥
λ) = ⎢
F (∆λ ⎢
⎥
⎥
⎢ f2 (∆λ1 ,∆λ2 )⎥
⎣ ⎦
⎡ ⎤ ⎡ ⎤ (7.48)
⎢k (σ1 − ∆λ1 s1,1 − ∆λ2 s2,1 ) − (σ3B − ∆λ1 s1,3 − ∆λ2 s2,3 ) − σCc ⎥ ⎢0⎥
⎢ ⎥=⎢ ⎥
B
=⎢ ⎥ ⎢ ⎥
⎢k (σ2B − ∆λ1 s1,2 − ∆λ2 s2,2 ) − (σ3B − ∆λ1 s1,3 − ∆λ2 s2,3 ) − σCc ⎥ ⎢0⎥
⎣ ⎦ ⎣ ⎦

⎡ ⎤
where
⎢∆λ1 ⎥
⎢
λ=⎢ ⎥
⎥
⎢∆λ2 ⎥
∆λ (7.49)
⎣ ⎦
The gradient is found to be
⎡ ⎤
⎢−ks1,1 + s1,3 + dσc −ks2,1 + s2,3 + dσc ⎥
F
∂F
= ⎢ ⎥
λT ⎢ ⎥
⎢−ks1,2 + s1,3 + dσc −ks2,2 + s2,3 + dσc ⎥
(7.50)
⎣ ⎦
∂∆λ

where dσc is given by (7.39). The system of equations is solved using a Newton-Raphson iteration,
λ is made, an afterwards updated as
where an initial guess of ∆λ
−1
λi+1 = ∆λ
λi − ( ∣ ) F (∆λ
λi )
F
∂F
∆λ (7.51)
λT i
∂∆λ
until the required precision is reached. The return algorithm for the l6 return is analogous to the above,
except that the f2 yield surface is replaced by the f6 yield surface. This gives
⎡ ⎤
⎢ f1 (∆λ1 ,∆λ2 )⎥
λ) = ⎢
F (∆λ ⎢
⎥
⎥
⎢ f6 (∆λ1 ,∆λ2 )⎥
⎣ ⎦
⎡ ⎤ ⎡ ⎤ (7.52)
⎢k (σ1 − ∆λ1 s1,1 − ∆λ2 s2,1 ) − (σ3B − ∆λ1 s1,3 − ∆λ2 s2,3 ) − σCc ⎥ ⎢0⎥
⎢ ⎥=⎢ ⎥
B
=⎢ ⎥ ⎢ ⎥
⎢k (σ1B − ∆λ1 s1,1 − ∆λ2 s2,1 ) − (σ2B − ∆λ1 s1,2 − ∆λ2 s2,2 ) − σCc ⎥ ⎢0⎥
⎣ ⎦ ⎣ ⎦
with the corresponding gradient
⎡ ⎤
⎢−ks1,1 + s1,3 + dσc −ks2,1 + s2,3 + dσc ⎥
F
∂F
= ⎢ ⎥
λT ⎢ ⎥
⎢−ks1,1 + s1,2 + dσc −ks2,1 + s2,2 + dσc ⎥
(7.53)
⎣ ⎦
∂∆λ

7.5.3 Return to apex point a

The return algorithm to the apex of the Mohr-Coulomb criterion is a further expansion of the l1 and l6
algorithms, in which case f1 , f2 and f3 needs to be fulfilled. The approach is similar to the above and
7.6. Consistent Constitutive Matrix 41

only the main results will be given here. The residual vector is found to be given by

⎡ f (∆λ ,∆λ ,∆λ )⎤

⎢ 1 3 ⎥
⎢ ⎥
λ) = ⎢ f2 (∆λ1 ,∆λ2 ,∆λ3 )⎥
⎢
1 2
F (∆λ ⎥
⎢ ⎥
⎢ f (∆λ ,∆λ ,∆λ )⎥
⎣ 6 1 2 3 ⎦
⎡k (σB − ∆λ s − ∆λ s − ∆λ s ) − (σB − ∆λ s − ∆λ s − ∆λ s ) − σC ⎤
⎢ c⎥
⎢ ⎥
=⎢ C⎥
1 1 1,1 2 2,1 3 3,1 3 1 1,3 2 2,3 3 3,3

⎢k (σ2 − ∆λ1 s1,2 − ∆λ2 s2,2 − ∆λ3 s3,2 ) − (σ3 − ∆λ1 s1,3 − ∆λ2 s2,3 − ∆λ3 s3,3 ) − σc ⎥
B B
⎢ ⎥
(7.54)
⎢k (σB − ∆λ s − ∆λ s − ∆λ s ) − (σB − ∆λ s − ∆λ s − ∆λ s ) − σC ⎥
⎣ 1 1 1,1 2 2,1 3 3,1 2 1 1,2 2 2,2 3 3,2 c⎦
⎡0⎤
⎢ ⎥
⎢ ⎥
=⎢ ⎥
⎢0⎥
⎢ ⎥
⎢0⎥
⎣ ⎦

⎡∆λ ⎤
where
⎢ 1⎥
⎢ ⎥
λ=⎢ ⎥
⎢∆λ2 ⎥
⎢ ⎥
∆λ (7.55)
⎢∆λ ⎥
⎣ 3⎦
and the gradient matrix is given by

⎡−ks + s + dσ −ks + s + dσ −ks + s + dσ ⎤

⎢ 1,1 1,3 c⎥
⎢ ⎥
= ⎢−ks1,2 + s1,3 + dσc −ks2,2 + s2,3 + dσc −ks3,2 + s3,3 + dσc ⎥
⎢
c 2,1 2,3 c 3,1 3,3
F
∂F
⎥
⎢ ⎥
(7.56)
⎢−ks + s + dσ −ks + s + dσ −ks + s + dσ ⎥
λ
∂∆λ T
⎣ 1,1 1,2 c 2,1 2,2 c 3,1 3,2 c⎦

7.6 Consistent Constitutive Matrix

In the evaluation of the consistent constitutive matrix, the modification matrix aligned with the principal
T is needed. The part related to the principal stresses, T , see equation (6.3), simplifies to the
stresses, T̂

⎡1 0 0⎤
unit matrix
⎢ ⎥
⎢ ⎥
T =⎢
⎢0 1 0⎥
⎥
⎢ ⎥
(7.57)
⎢0 0 1⎥
⎣ ⎦
because the derivative of b with respect to σ results in the zero-matrix

⎡0 0 0⎤
⎢ ⎥
∂bb ⎢ ⎥
= ⎢0 0 0⎥
⎢
⎥
σ ⎢ ⎥
(7.58)
⎢0 0 0⎥
∂σ
⎣ ⎦

T , that relates to the shear stresses, T̃

The part of T̂ T , is simply evaluated by equation (6.5). Thus it is
c
D . Afterwards, the consistent constitutive matrix aligned with the principal axes,
possible to evaluate D̂
epc
D
D̂ , is calculated using either equation (5.23), (5.55) or (5.59), depending on the return algorithm used
in finding the updated stress state.
 Chapter 8

Computational Example: Strip Footing

To test the hardening Mohr-Coulomb model, a simple bearing capacity calculation of a strip footing is
carried out in plane strain. This has been done by implementing the strain hardening Mohr-Coulomb
model in FORTRAN and then utilizing it in a FEM-code written in MatLab, which rely on a Newton-
Raphson procedure in the global equilibrium iterations, as shown in Table 5.1.

8.1 The Model

The model consists of a rigid rough foundation resting on top of a strain hardening Mohr-Coulomb
material. Since a strip footing in plain strain is examined, advantage is made of the symmetry line of
the problem, see Figure 8.1. The domain size is governed by L and H, which has been set to 30 m and
20 m respectively. The total width of the foundation is 2 m. The domain is meshed using 2-dimensional
6-node triangular linear strain elements utilizing a gaussorder of 6. The load of the foundation is
modeled using a prescribed displacement of 1000 mm in the negative y-direction of the nodes situated
directly under the foundation.

y Rigid foundation
x
Strain softening
Mohr-Coulomb
Fixed in x-direction
Symmetry line

material

Fixed in y-direction

Figure 8.1: Sketch of the model of the foundation. Not to scale.

43
44 Chapter 8. Computational Example: Strip Footing

8.2 Material Parameters

The parameters of the Mohr-Coulomb material are found based on an approximation to a material with
the Hoek-Brown parameters listed in Table 8.1.

State GSI mi σci D a mb s E ν γ

Peak 50 12 80 MPa 0 0.51 2.01 0.0039 9 GPa 0.25 20 /m3
kN

Residual 25 12 80 MPa 0.5 0.53 0.34 0 9 GPa 0.25 20 kN/m2

Table 8.1: Hoek-Brown material parameters of the peak and residual strength of the rock
material at hand.

As can be seen, for the peak strength of the rock material, intact rock is assumed, D = 0, with a uniaxial
compressive strength of 80 MPa, a GSI value of 50 and a mi constant of 12. For the residual strength,
the GSI value is lowered to 25, and the disturbance factor, D, is set to 0.5. The Hoek-Brown constants
a, mb and s are calculated based on equation (2.2)-(2.4). Young’s modulus is assumed to be 9 GPa.
The parameters orignate from Sharan [2008] and the material associated with it is described as average
quality rock mass, which according to Hoek and Brown [1997] should behave in a strain softening way.
Further, the rock mass is assumed to have a specific weight, γ, of 20 kN/m3 . Using the equations in
chapter 3, the corresponding Mohr-Coulomb parameters can be seen in Table 8.2, where the general
approach of equation (3.15) has been applied.

Parameter ϕ c
○
Peak 32.07 4.21 MPa
○
Residual 17.93 1.91 MPa

Table 8.2: Mohr-Coulomb approximation of the Hoek-Brown parameters listed in Table 8.1.
The approximation utilizes equation (3.15).

Due to the restrictions of the current model, where only the cohesion changes during plastic loading, the
model is unable to account for the change in friction angle. Because of this, the peak friction angle is
used for the residual strength as well. In order to fully implement the strain softening behavior, it is also
necessary to know how fast the strength drops. I.e. a relationship between the accumulated plastic strain
and the cohesion. In the current example, this is modeled using three curve segments, as illustrated on
Figure 8.2. The first segment consists of a linear softening curve with a constant slope of H, which is
defined between the two points (0,cPeak ) and the intermediate point (εInt p
,cInt ). The second segment,
defined between the intermediate point (εInt ,cInt ) and the residual point (εRes
p p
,cRes ), is modeled as a
Bézier curve, with an initial slope of H, and an end slope of 0. The last segment defines a perfectly
plastic behavior, once the residual strength is reached. This is done by modeling a line segment with a
slope of 0, which extents to infinity. The intermediate cohesion, cInt , is taken as

cInt = cRes + 0.3 ⋅ (cPeak − cRes ) (8.1)

p p
The intermediate and residual accumulated plastic strains, εInt and εRes , are varied, to study the influ-
ence on the results. For simplicity and numerical stability, associated plasticity is assumed. The final
8.3. Mesh Coarseness 45

cPeak
1
H

Cohesion, c
cInt

cRes

εRes
εInt
p

p
0 Accumulated plastic strain, ε p

Figure 8.2: Sketch of the ε p − c-curve used in the example.

parameters used in the model are listed in Table 8.3. Further, two perfectly plastic cases are computed.
Namely one where the strength is equal to the peak strength, and one where the strength is equal to the
residual strength.

Parameter ϕ cPeak cInt cRes ψ

○
Value 32.07 4.21 MPa 2.60 MPa 1.91 MPa 32.07○

Table 8.3: Material parameters used in the model.

8.3 Mesh Coarseness

In order to estimate the needed coarseness of the mesh, a convergence analysis has been performed on
p p
a model, where εInt and εRes have been set to 1 and 2 respectively. The mesh has then been generated
with increasingly more degrees of freedom in order to estimate the influence on the peak and residual
bearing capacity of the model. Based on this, a mesh coarseness is chosen, where further refinement
only results in minor changes of the bearing capacity. As can be seen from Figure 8.3, the peak and
residual bearing capacity is dependent upon the coarseness of the mesh. However, at around 10000
degrees of freedom, the bearing capacities start to stabilize, and only minor differences in the bearing
capacities can be observed. Based on this, the model is meshed with 3825 elements, giving a total of
15692 degrees of freedom and 22950 gauss points. The mesh is shown in Figure 8.4.

8.4 Results
The load displacement curve for five different scenarios along with their ε p − c-curves are shown in
Figure 8.5 and Figure 8.6 respectively. The curves named “Perfectly Plastic Peak Strength” and “Per-
46 Chapter 8. Computational Example: Strip Footing

400
Peak bearing capacity
Residual bearing capacity
Bearing capacity, [MN/m ]

350

300

250

200

150
102 103 104 105
Number of degrees of freedom, ndo f
Figure 8.3: Convergence analysis of the peak and residual bearing capacity of a model with
p
εInt = 1 and εInt
p
= 2.

-5
y [m]

-10

-15

-20
0 5 10 15 20 25 30
x [m]
Figure 8.4: Mesh used in the current model, consisting of 3825 elements, 15692 degrees of
freedom and 22950 gauss points.

fectly Plastic Residual Strength” are the perfectly plastic models using the peak and residual strength
respectively, which give a load carrying capacity of 309.2 MN/m and 141.6 MN/m .
To verify these results, they are compared to the analytical solution given by Terzaghi’s bearing
capacity formula
R = γb2 Nγ + qbNq + cbNc
1
(8.2)
2
8.4. Results 47

where R is the bearing capacity, b is the foundation width, q is the overburden pressure and Nγ , Nq and
Nc are the dimensionless bearing capacity factors given by [Ovesen et al., 2007]

1 + sin(ϕ)
Nq = exp(πtan(ϕ))
1 − sin(ϕ)
(8.3)

Nγ = ((Nq − 1)cos(ϕ))
1 3/2
(8.4)
4
Nq − 1
Nc = (8.5)
tan(ϕ)

Using the values of Table 8.3 together with equation (8.2), results in a peak and residual load carrying
capacity of 296.4 MN/m and 137.2 MN/m . However, equation (8.2) is known to give a conservative bearing
capacity, and thus, the perfectly plastic models seems to be in tune with the analytical solutions. The
load carrying capacity of these models should mark the upper and lower bound of the expected load
carrying capacity of the strain softening materials.
p p
The three other curves of Figure 8.5 utilize strain softening, where the numbers indicate εInt and εRes
respectively. The figure shows, that the strain softening materials never reach the load carrying capacity

350

300

250
Line load, p [MN/m ]

200

150

100 Strain softening 0.10-0.20

Strain softening 0.15-0.30
Strain softening 1.00-2.00
50 Perfectly Plastic Peak Strength
Perfectly Plastic Residual Strength
0
0 0.2 0.4 0.6 0.8 1.0
Displacement, u [m]
Figure 8.5: Load displacement curve of strip footing. u is the vertical displacement of the
p p
foundation. The numbers following the strain softening models are εInt and εRes respectively.

of the perfectly plastic model utilizing the peak strength. On the other hand, they do not approach the
load carrying capacity of the perfectly plastic model utilizing the residual strength either. This seems
realistic, since some gauss points reach plasticity before others and weakens, and thus should not be
able to carry the peak load. Similarly, not all gauss points reach the residual strength in the model, and
should thus be able to carry more than the residual load.
48 Chapter 8. Computational Example: Strip Footing

4.5

4.0
Strain softening 0.10-0.20
Cohesion, c [MPa]

Strain softening 0.15-0.30

3.5
Strain softening 1.00-2.00
Perfectly Plastic Peak Strength
3.0
Perfectly Plastic Residual Strength
2.5

2.0

1.5
0 0.5 1.0 1.5 2.0 2.5 3.0
Accumulated plastic strain, ε p , [−]

Figure 8.6: The strain softening behavior of the models shown in Figure 8.5.

p
Figure 8.7 shows the relationship between the residual accumulated plastic strain, εRes , and the
peak and residual bearing capacity of 12 different strain softening models, as well as the peak and
residual strength of the perfectly plastic models. Together with Figure 8.5, it can be seen, that smaller

300
Bearing capacity, [MN/m ]

250 Peak bearing capacity

Residual bearing capacity
Perfectly Plastic Peak Strength
200 Perfectly Plastic Residual Strength

150

0.0 0.5 1.0 1.5 2.0

p
εRes
Figure 8.7: The peak and residual bearing capacity of 12 different strain softening models as
a function of ε p . The bearing capacity of the perfectly plastic models is also shown.

p p
εInt and εRes result in a lower peak bearing capacity, as well as a more rapid decrease in the load
carrying capacity towards a residual bearing capacity. The maximum load bearing capacity of the strain
p p
softening models is influenced by the choice of εInt and εRes . The 1.00−2.00 model predicts a maximum
8.4. Results 49

bearing capacity of 298.2 MN/m , while the 0.15 − 0.30 model predicts a maximum bearing capacity of
268.1 MN/m , which is significantly less than that of the bearing capacity of the perfectly plastic model
utilizing the peak strength. However, it is also almost twice the size of the bearing capacity of the
perfectly plastic model using the residual strength. The 1.00 − 2.00 model predicts a residual bearing
capacity of 157.2 MN/m , while the 0.15 − 0.30 model predicts a residual bearing capacity of 150.8 MN/m .
An improvement compared to the bearing capacity of the perfectly plastic residual model of about
5 − 10 %. Thus, using the perfectly plastic model with the peak strength results in an unsafe model,
while use of the residual strength results in a very conservative estimate of the bearing capacity.
Further, it is noticed, that the 0.10 − 0.20 model failed to converge, at some point just after the peak
bearing capacity was reached, see Figure 8.5. The 0.15−0.30 model was the steepest model, which was
found to converge. This is troublesome if even steeper drops in the load carrying capacity is needed.
However, this might be solved by utilizing a more advanced global equilibrium iteration procedure than
the standard Newton-Raphson. For example an arc-length method.
Figure 8.8 shows a plot of the cohesion throughout the soil of the 0.15 − 0.30 model, which shows,
that the material located along the slip lines have reached the residual strength. This is consistent with
the fact, that these zones experience the most plastic straining. The zones outside these slip lines are
almost unaffected by the softening behavior.

c[MPa]
0
4.0

-5
3.5
y[m]

-10 3.0

-15 2.5

2.0
-20
0 5 10 15 20 25 30
x [m]

Figure 8.8: Plot of the cohesion throughout the soil at the end of the 0.15-0.30 model.

Figure 8.9 show a comparison of the accumulated plastic strains around the rupture zone of the
model. The topmost graph shows the perfectly plastic model utilizing the peak strength, and the bot-
tommost graph shows the perfectly plastic model utilizing the residual strength. The model in the
middle is the strain softening 0.15 − 0.30 model. From the figure, it is seen, that the accumulated plastic
strains of the softening model are more concentrated around the slip lines compared to the perfectly
plastic models. This is probably due to the fact, that once plastic strains has developed in a point, the
point weakens and thus further plasticity is more likely in this point, than in the neighboring material
which surrounds it.
50 Chapter 8. Computational Example: Strip Footing

Perfectly Plastic Peak Strength ε p [−]

1
102
0

-1
100
y[m]

-2
10−2
-3

-4
10−4
-5
0 5 10 15 20
x [m]
Strain Softening 0.15-0.3 ε p [−]
1
102
0

-1
100
y[m]

-2
10−2
-3

-4
10−4
-5
0 5 10 15 20
x [m]
Perfectly Plastic Residual Strength ε p [−]
1
102
0

-1
100
y[m]

-2
10−2
-3

-4
10−4
-5
0 5 10 15 20
x [m]
Figure 8.9: Plot of the accumulated plastic strain throughout the soil of the perfectly plastic
peak strength model, strain softening 0.15-0.30 model and perfectly plastic residual strength
model.
 Chapter 9

Computational Example: Tunnel

Excavation

To further test the approach, a tunnel excavation is carried out using an axisymmetric model, where the
purpose is to estimate the tunnel wall displacement as well as the plastic zone.

9.1 The model

The idea behind the model is to simulate a tunnel excavation in an infinite rock mass. This is done by
reducing the pressure, p, on the tunnel wall, from the in situ stress state, pinf , to zero, see Figure 9.1,
and record the ensuing tunnel wall displacement. The radius of the tunnel is given by rT , which has
been set to 2,5 m. The tunnel axis is aligned in the z-direction.

Strain softening Mohr-Coulomb material

Tunnel wall

θ
r
rT z

Figure 9.1: Concept of the tunnel model.

The axisymmetric properties of the problem is utilized, and a 1 m section of the tunnel in the z-
direction is modeled, see Figure 9.2. Further, the infinite rock mass is bounded by a domain with a
radius of rBC , which has been set to 50 m in the example at hand. As with the example in chapter 8, the

51
 yy

yy
yy

yy
yy

yy

yy
yy

yy

yy
yy

yy

yy

yy
yy

yy

yy

yy
yy

yy

yy

yy

yy
yy

yy

yy

yy
yy

yy
yy

yy
y

yy
yy

yy
y

yy
yy

yy
yy

yy
yy

yy
yy
y

yy
yy

yy
y

yy
yy
52 Chapter 9. Computational Example: Tunnel Excavation

y
yy
yy

yy
yy

yy

yy
yy
y
y
yy

yy
y

yy

yy
y

yy
yy
yy
yy

yy
yy

yy
yy

yy
y
yy
y
yy

yy

yy
y

yy
y

yy

yy
yy

y
yy

yy
yy

yy
yy

yy
yy

yy
y
yy
yy

y
y

yy

yy
yy

yy
rT

yy

yy
y
yy

yy
yy

yy
yy

yy

y
y

yy
yy

yy
yy
yy

yy
yy

yy
y

y
yy

yy
y
yy

yy

yy

yy
yy

yy
yy

yy
y

yy
yy
y
yy

yy

yy
yy

yy
y

yy

yy
yy

yy
yy

yy
yy

yy

y
y

y
yy
Fixed in the z-direction

yy

y
y

y
yy

yy
yy
yy

y
yy
y

yy

y
yy
yy

y
y

y
yy
y

y
yy

yy
yy

yy
y

yy
yy
yy

yy
yy
yy

yy
yy
yy

y
yy

y
y

yy
yy

yy r-direction
y

yy
yy

y
yy

yy
yy

yyyy

yy
y

yy
y

y
y
yy

yy
yy

yy

yy

y
y

yy
y

y
yy yyyyyy yyyyyy yyyyyy

y
y

y
yy
yy

yyy
yy

yy

y
yy
yy

yy
yy
y

yy
yy
Symmetry line

y
y
y

yyyy

y
yy

yy
y

yyy

yy
y
y

yy
y

y
yyy

y
y

yy

yyyy

yy
y

y
y
yy

y
y

yy

y
yy

yy
yy
y

yy
y

y
yyyy
yyy

y
y
yy

yy
yyy

yy
y

yy
yyy

y
yyy

yy
yy

y
y

yy
y
yy

yy

yy
y
y

yy
y
y

y
yyyy
y

yyy
y

y
y
p

yyyy
1m

y
yyy

yy
y
yyy

yy
y
y

y
yy

y
y
yy y yyy yyy

y
yy

the

yy
y

yyy

yy
y

yy
y

y
y
y

y
y

yyyy

yy
yyy

yy
y

yy
yyy
y

y
y

y
y
yyyy

yy

y
yy
yy

y
y
y

y
yyyy

yy

yy
y

yy
y

yyy

yy
y

y
y

yy
y

yin
y

yy

y
yyyy

yy

yy
yyyy

y
y
y

yy
y
y

y
y
yy

yy
y

y
y

yyy
y

y
yy

yy
y

yy

y
y

Fixed
yy

y
y
yy

y
yyyy

yy
y

y
y

y
y

y
yy

yy
y
yy
y
yy

yy
y

yy
yy
yy

yy
y

y
y

yy
y
yyy

yy
y

y
yy
y

y
yy
yy

yy
y

y
yy
z

yy

y
y
y

yy
y

yy
y

yy
y

yy

yy
yyy
yy

yy
y

yy
y

y
y

y
y
yy

yy

yy
yy

yy
yy

y
y

yy
yy

y
yy

yy
y

yy
Fixed in the z-direction

yy
yy

yy

yy
y

yy

yy
y

yy

yy
y
r y

y
yy

yy
y

y
yy

y
yy

yy

y
y

yy
yy

y
yy

yy
yy
yy

yy
yy
yy
y

y
yy

yy
y

y
yy

yy

yy
yy

yy

y
yy

yy
yy

y yy
y
y

yy

yy
yy

yy
y

yy

yy
yy

rBC yy

yy
yy

yy
y

y
yy

yy

y
yy

yy
yy

yy
yy

yy
y

y
yy
y

y
y

yy

y
y

y
yy

yy

yy

yy
y

yy
yy

yy
yy

yy
yy

yy

y
y

yy
y

yy
yy
y

yy
yy

yy
yy

y
y

y
y

yy
Figure 9.2: Sketch of the tunnel model.
yy

yy

yy
y

y
yy

y
y

yy
yy

yy
yy

yy
yy
yy

yy
y

yy
y

yy

y
y
yy

yy
y

yy

y
yy

yy

yy
yy

yy
y

yy
y

y
yy

y
yy

yy

yy
y
yy

y
yy

yy
y

y
yy
yy

yy
yy

y
mesh is generated with 2-dimensional 6-node triangular linear strain elements.
yy
y

yy
y
yy

yy

y
yy
yy

yy
y
y

yy
yy

y
y yy

yy
yy

yy

y
yy

9.2 Material Parameters

yy

yy
yy
yy

The model tries to simulate an excavation in rock material with the Hoek-Brown parameters listed in
Table 9.1.

State GSI mi σci D a mb s E ν γ

kN
Peak 30 8 20 MPa 0 0.52 0.66 0.0004 1.4 GPa 0.3 26 /m3
Residual 15 8 20 MPa 0.5 0.56 0.14 0 1.4 GPa 0.3 26 kN/m2

Table 9.1: Hoek-Brown material parameters of the peak and residual strength of the rock
material at hand.

The parameters of the peak strength are taken from Sharan [2008], which describe the rock material
as very poor. According to Hoek and Brown [1997], very poor rock material tends to behave perfectly
plastic. However, in the current example, it is assumed that the material softens during plastic straining.
Thus the residual strength is found in much the same way as it was done in chapter 8. Namely by
reducing the GSI-value to half of the original value and setting the disturbance factor to 0.5.
Assuming the excavation takes place 100 m below the surface and using the approximations of
chapter 3 along with the estimate of σ3,max based on deep tunnels, equation (3.12), the Mohr-Coulomb
parameters listed in Table 9.2 are obtained.
9.3. Mesh Coarseness 53

Parameter ϕ c
○
Peak 33.74 256 kPa
Residual 19.28○ 103 kPa

Table 9.2: Mohr-Coulomb approximation of the Hoek-Brown parameters listed in Table 8.1.
The approximation utilizes equation (3.12).

Similarly to the example of the strip footing, the friction angle of the model is taken to be given
by the peak friction angle, and the softening behavior is modeled according to Figure 8.2 and equation
(8.1). Thus, the final material parameters are given in Table 9.3.

Parameter ϕ cPeak cInt cRes ψ

○
Value 33.74 256 kPa 149 kPa 103 kPa 33.74○

Table 9.3: Material parameters used in the model.

The in situ stress state of the rock mass, pinf , is assumed to be a hydrostatic pressure, given by the
depth, and the unit weight of the rock

pin f = 26 kN/m3 ⋅ 100 m = 2.6 MPa (9.1)

The effects of gravity are neglected in the model. Similarly to the example of the strip footing, the
p
model is tried with different values of the intermediate and residual accumulated plastic strain, εInt and
p
εRes , as well as two perfectly plastic cases with the peak and residual strength respectively.

9.3 Mesh Coarseness

Similar to the example of chapter 8, a convergence analysis has been made in order to find an appropri-
ate mesh coarseness. This has been done with a model where εInt p
= 0.025 and εRes
p
= 0.050. The tunnel
wall displacement as a function of the number of degrees of freedom is shown in Figure 9.3. From this
figure it is seen, that only minor changes in the wall displacement is found once the number of degrees
of freedom is above 10000. Based on this, the model is meshed using 4800 elements, resulting in 28800
gauss points and 19602 degrees of freedom. The mesh can be seen in Figure 9.4.

9.4 Results
The load displacement curve of the tunnel wall of eight models is shown in Figure 9.5 together with
a close up in Figure 9.6. The strain softening behavior of the models can be seen on Figure 9.7.
From the figures, it is seen that the perfectly plastic model using the peak strength gives a tunnel wall
displacement of 30 mm, while the model using the residual strength gives a displacement of 126 mm.
Using the analytical solution for the perfectly plastic case presented in Carranza-Torres [2003], the
tunnel wall displacement is found to be 30.5 mm and 130 mm respectively, which suggests that the
finite element model is sound. All of the strain softening models predict wall displacements in between
the two perfectly plastic models, which is what was expected. Further, the extent of the plastic zone,
54 Chapter 9. Computational Example: Tunnel Excavation

0.07
Wall displacement [m]

0.06

0.05

0.04

0.03
102 103 104 105
Number of degrees of freedom, ndo f
Figure 9.3: Displacement of the tunnel wall as a function of the number of degrees of freedom
p
of a model with εInt = 0.025 and εRes
p
= 0.050.

1.0

0.8

0.6
z [m]

0.4

0.2

0.0
0 10 20 30 40 50
r [m]
Figure 9.4: Mesh used in the current model, consisting of 4800 elements, 19602 degrees of
freedom and 28800 gauss points.

sketched in Figure 9.8, ranges from 4.12 m for the peak strength model to 5.8 m for the residual strength
model as shown in Figure 9.9. The analytical solution of Carranza-Torres [2003] results in plastic zones
of 4.11 m and 5.72 m for the perfectly plastic peak and residual models.
9.4. Results 55

3.0
Strain softening 0.005-0.010
Pressure on tunnel wall, p [MPa]

2.5 Strain softening 0.010-0.020

Strain softening 0.025-0.050
2.0 Strain softening 0.050-0.100
Strain softening 0.100-0.200
1.5 Strain softening 1.000-2.000
Perfectly Plastic Peak Strength
Perfectly Plastic Residual Strength
1.0

0.5

0.0
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14
Wall displacement in r-direction [m]
Figure 9.5: Displacement of the tunnel wall. The numbers following the strain softening
p p
models are εInt and εRes respectively.

0.10
Strain softening 0.005-0.010
Pressure on tunnel wall, p [MPa]

Strain softening 0.010-0.020

0.08
Strain softening 0.025-0.050
Strain softening 0.050-0.100
0.06 Strain softening 0.100-0.200
Strain softening 1.000-2.000
Perfectly Plastic Peak Strength
0.04 Perfectly Plastic Residual Strength

0.02

0.00
0.02 0.04 0.06 0.08 0.10 0.12 0.14
Wall displacement in r-direction [m]

Figure 9.6: Close up of Figure 9.5.

56 Chapter 9. Computational Example: Tunnel Excavation

0.250
Strain softening 0.005-0.010
Strain softening 0.010-0.020
Strain softening 0.025-0.050
Cohesion, c [MPa]

0.200 Strain softening 0.050-0.100

Strain softening 0.100-0.200
Strain softening 1.000-2.000
Perfectly Plastic Peak Strength
0.150 Perfectly Plastic Residual Strength

0.100
0.00 0.50 1.00 1.50 2.00
Accumulated plastic strain, ε p , [−]

Figure 9.7: The strain softening behavior of the models.

Plastic zone

Tunnel wall

Figure 9.8: Illustration of the extent of the plastic zone.

9.4. Results 57

0.200
Accumulated plastic strian, ε p , [−]

Strain softening 0.005-0.010

Strain softening 0.010-0.020
0.150 Strain softening 0.025-0.050
Strain softening 0.050-0.100
Strain softening 0.100-0.200
0.100 Strain softening 1.000-2.000
Perfectly Plastic Peak Strength
Perfectly Plastic Residual Strength
0.050

0.000
2.50 3.00 3.50 4.00 4.50 5.00 5.50 6.00
r, [m]

Figure 9.9: The extent of the plastic zone of the models.

 Chapter 10

Conclusion

The successful derivation and implementation of the strain hardening Mohr-Coulomb criterion de-
scribed in chapter 7 shows, that it is possible to make use of return mapping in principal stress space
for materials, which exhibit hardening/softening properties.
The proof-of-concept examples of the strip footing and tunnel excavation show the expected results,
where the strain softening materials result in solutions somewhere in between the two extremities of the
perfectly plastic models using the peak and residuals strengths. Thus the hardening properties allow for
a more detailed description of the problem at hand, which should result in a more safe solution, than a
model utilizing perfect plasticity along with the peak strength, as well as less conservative solution than
a model utilizing perfect plasticity along with the residual strength.
Since the implemented model can handle any arbitrary development of the cohesion during plastic
straining, it should be applicable to a great deal of problems. However, as is evident from the examples,
approximations to the Hoek-Brown parameters with a Mohr-Coulomb failure criterion suggest, that the
friction and dilation angles change during plastic straining as well. Something that the current model is
unable to account for.
To further test the use of hardening properties along with principal stress updating schemes, a more
advanced constitutive model, utilizing several hardening and state parameters could be developed. For
example a Mohr-Coulomb model, where the friction and dilation angles are dependent of the state
parameters of the material. Moreover, the expressions for the consistent constitutive matrix of two and
three active yield surfaces, equation (5.55) and (5.59), together with return mapping in principal stress
space, allows for a fairly straight forward way of developing a hardening Hoek-Brown model, which
eliminate the need for the Mohr-Coulomb approximation of hardening rock material.
Next step would be to implement these models into commercial finite element programs, in order to
be truly useful for the professional engineering community, which require ease of use and accessibility.

59
 Bibliography

Carlos Carranza-Torres. Dimensionless graphical representation of the exact elasto-plastic solution of

a circular tunnel in a mohr-coulomb material subject to uniform far-field stresses. Rock Mechanics
and Rock Engineering, 36:237–253, 2003.

Johan Clausen. Efficient Non-Linear Finite Element Implementation of Elasto-Plasticity for Geotech-
nical Problems. PhD thesis, Aalborg University, 2007.

Johan Clausen, Lars Damkilde, and Lars Andersen. Efficient return algorithms for associated plasticity
with multiple yield planes. International Journal For Numerical Methods In Engineering, 66:1036–
1059, 2006.

Robert D. Cook, David S. Malkus, Michael E. Plesha, and Robert J. Witt. Concepts and Applications
of Finite Element Analysis. John Wiley & Sons, Ltd., 2002.

Michael Anthony Crisfield. Non-Linear Finite Element Analysis. John Wiley & Sons, Ltd., 2000.

Eduardo de Souza Neto, Djordje Peric, and David Owens. Computational Methods for Plasticity -
Theory and Applications. John Wiley & Sons, Ltd., 2008.

Evert Hoek. Practical Rock Engineering. 2007.

Evert Hoek and Edwin T. Brown. Practical estimates of rock mass strength. International Journal of
Rock Mechanics & Mining Sciences, 34, No. 8:1165–1186, 1997.

Evert Hoek and Mark S. Diederichs. Empirical estimation of rock mass modulus. International Journal
of Rock Mechanics & Mining Sciences, 43:203–215, 2006.

Evert Hoek, Carlos Carranza-Torres, and Brent Corkum. Hoek-Brown failure criterion - 2002 edition,
2002.

Warner Tjardus Koiter. Stress-strain relations, uniqueness and variational theorems for elasto-plastic
materials with a singular yield surface. Quarterly of applied mathematics., 11:350–354, 1953.

Paul Marinos and Evert Hoek. GSI: A geologically friendly tool for rock mass strength estimation.
In Proceedings of the International Conference on Geotechnical and Geological Engineering, Geo-
Eng2000, 2000.

61
62 Bibliography

Richard S. Merifield, Andrei V. Lyamin, and Scott W. Sloan. Limit analysis solutions for the bearing
capacity of rock masses using the generalised Hoek-Brown criterion. International Journal of Rock
Mechanics & Mining Sciences, 43:920–937, 2006.

Joop C. Nagtegaal. On the implementation of inelastic constitutive equations with special reference to
large deformation problems. Computer Methods in Applied Mechanics and Engineering, 33:469–
484, 1982.

Niels Saabye Ottosen and Matti Ristinmaa. The Mechanics of Constitutive Modeling. Elsevier Ltd,
2005.

Niels Krebs Ovesen, Leif Fuglsang, Gunnar Bagge, Anette Krogsbøll, Carsten S. Sørensen, Bent
Hansen, Klaus Bødker, Lotte Thøgersen, Jens Galsgaard, and Anders H. Augustesen. Lærebog i
Geoteknik. Polyteknisk Forlag, 2007.

Rocscience Inc. Roclab - user’s guide, 2007.

Shailendra K. Sharan. Analytical solutions for stresses and displacements around a circular opening in
a generalized Hoek-Brown rock. International Journal of Rock Mechanics & Mining Sciences, 45:
78–85, 2008.

Emil Smed Sørensen. Modified generalised Hoek-Brown model, 2012. Aalborg University, M.Sc. 3rd
semester report.