INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN BIOMEDICAL ENGINEERING

Int. J. Numer. Meth. Biomed. Engng. (2012)

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/cnm.2491
Algorithms for a strain-based plasticity criterion for bone
Pankaj Pankaj
*
,
and Finn E. Donaldson
School of Engineering, The University of Edinburgh, Kings Buildings, Edinburgh, EH9 3JL, UK
SUMMARY
A range of stress-based plasticity criteria have been employed in the nite element analysis of the post-
elastic behaviour of bone. There is some recognition now that strain-based criteria are more suitable for this
material because they better represent its behaviour. Moreover, because bone yields at relatively isotropic
strains, a strain-based criterion requires fewer material parameters unlike those required for a stress-based
criterion. Based on a minimum and maximum principal strain criterion, a robust strain-based plasticity
algorithm is developed. As the criterion comprises six piecewise linear surfaces in principal strain space,
it has a number of singular regions. Singularity indicators are developed to direct the algorithm to make
appropriate plastic corrector returns when singularity regions are encountered. The developed algorithms
permit a plastic corrector to be achieved in a single iterative step in all cases. A range of benchmark tests
are developed and conducted after implementing the algorithm in a nite element package. These tests show
that the constitutive behaviour is as expected. Copyright 2012 John Wiley & Sons, Ltd.
Received 21 February 2012; Accepted 26 March 2012
KEY WORDS: stress-based; anisotropy; singularity indicators; benchmark tests
1. INTRODUCTION
Several stress-based criteria have been considered for bone. These range from the extensively used
isotropic von Mises criterion to the anisotropic Tsai-Wu [1] criterion [211]. Although some of the
simpler isotropic criteria, perhaps employed because they were readily available, have been shown
to be unsuitable for bone [3, 8], other anisotropic ones require too many parameters, and their accu-
racy is not sufcient to justify the additional experimental effort required for the determination of
the parameters [3, 12].
There has been some debate as to whether stress-based or strain-based criteria are best for rep-
resenting bone yielding [13]. Some studies have indicated that strain-based criteria offer greater
accuracy [4, 14]. For bone, a strain-based criterion is more suitable than stress-based criteria for
two compelling reasons. Firstly, there is now some evidence to suggest that yielding and failure of
bone is based on strain rather than stress [15] and strain-based criteria offer greater accuracy [4, 14].
Secondly, bone is an anisotropic material, which can be approximated by orthotropy [16, 17]. This
is reected both in its elastic properties and in its strength. As a consequence, the yield criterion for
bone has to be anisotropic in the stress space. However, in the strain space, bone yields at relatively
isotropic strains, and the yield strain is not dependent upon apparent elastic stiffness or density
[4, 14, 1820]. An isotropic, strain-based criterion requires relatively few material properties to be
evaluated and no a priori identication of material orientation. Contrastingly, an anisotropic stress-
based criterion requires numerous material parameters to be determined and must be oriented with
the physical structure of the bone. These reasons suggest that for bone, a strain-based criterion is
*Correspondence to: Pankaj Pankaj, School of Engineering, The University of Edinburgh, Kings Buildings, Edinburgh,
EH9 3JL, UK.

E-mail: pankaj@ed.ac.uk
Copyright 2012 John Wiley & Sons, Ltd.
P. PANKAJ AND F. E. DONALDSON
not only biodelic but also numerically more convenient to employ. Interestingly, although several
studies have found strain-based yield criteria to be more suitable for bone, development of plastic-
ity algorithms has been extremely limited [21]. In fact, some studies have simply used an elastic
analysis to predict regions where strains violate a strain-based criterion [8, 22].
Strain-based plasticity was rst discussed about four decades ago by Naghdi and Trapp [23] and
subsequently developed by others [2428]. Over the decades, it has received very little attention in
comparison with stress-based theories, and generally, nite element codes do not incorporate strain-
based plasticity. It should be noted that for any static stressstrain state, a yield criterion formulated
in stress space can be converted into its equivalent in strain space. However, key differences become
apparent as yielding progresses [23, 29, 30]. First identied by Yoder and Iwan [27], it was illus-
trated by Lan et al. [25] that the sign of the derivative of yield function uniquely identies loading,
neutral and unloading conditions in strain space, but is ambiguous in stress space.
Gupta et al. [21] developed a yield criterion for trabecular bone based upon an isotropic modi-
ed super ellipsoid dened in principal strain space by Bayraktar et al. [18]. The criterion has four
parameters, two of which are essentially the yield strains in compression and tension. This yield
surface is smooth and is therefore free from the issues associated with singularities at sharp corners
in piecewise linear surfaces such as those in MohrCoulomb, Tresca and Rankine criteria; although
algorithms have been developed to effectively handle issues associated with singular yield surfaces
[3133]. It has been demonstrated by previous researchers [3, 8] that reasonable accuracy can be
achieved through the use of a maximum (Saint Venant) and minimum principal strain criterion.
This yield criterion can be visualised as a cube in principal strain space. Indeed, it consists of six
planar yield surfaces and requires only two parameters (yield strains in tension and compression)
to be dened. Furthermore, it has been shown that accurate returns to such piecewise linear sur-
faces (in a predictorcorrector algorithm) can be achieved in a single iterative step [31, 33]. This
greatly reduces the numerical cost of the overall algorithm and also improves its stability and accu-
racy. Such algorithms can be effectively used in biomedical engineering problems [34]. For these
reasons, the minimum and maximum elastic principal strain criterion was selected for this work.
Although it has previously been used as a limit in studies of bone failure [35], it has not previously
been developed as a full plasticity algorithm. The aim of this study was to develop algorithms for
this simple strain-based constitutive model for inclusion in nite element codes.
2. FORMULATION AND DEVELOPMENT OF ALGORITHMS
2.1. Denition of the criterion
The yield criterion is dened by limiting values of the elastic principal strains. The criterion is fully
dened by specication of tensile and compressive yield strains. The yield criterion g, is a function
of the principal elastic strains, c
e
1
, c
e
2
, c
e
3
i.e.
g =g
_
c
e
1
, c
e
2
, c
e
3
_
(1)
As there is little available data on post-yield behaviour of bone, the algorithms developed in this
study are limited to perfect plasticity. The yield criterion is represented in the principal strain space
by the six yield planes
g
1t
=c
e
1
Y
t
: g
2t
=c
e
2
Y
t
: g
3t
=c
e
3
Y
t
g
1c
=c
e
1
Y
c
: g
2c
=c
e
2
Y
c
: g
3c
=c
e
3
Y
c
(2)
where Y
t
and Y
c
are the tensile and compressive yield strains. As usual, tension is taken as positive
and compression as negative. In principal strain space, this yield criterion can be visualised as a
cube with sides Y
t
[Y
c
[ as illustrated in Figure 1. It is important to note that even under conditions
of elastic isotropy, a yield surface in principal strain and principal stress space do not have the
same shape.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
(a)
(b)
Figure 1. The 3D maximum principal strain criterion (a) and its shape in stress space (b) under conditions
of elastic isotropy. Note Y
t
and Y
c
indicate tensile and compressive yield strains, respectively. The line of
coloured locations indicate equivalent points on the two yield surfaces for c
1
=Y
t
, c
2
=0, c
3
=Y
c
Y
t
.
2.2. Predictorcorrector theory
Clausen et al. [36] and Huang and Grifths [37] considered piecewise linear yield functions in
stress-based plasticity. Here, we follow their general arguments. For piecewise linear yield func-
tions, the ow vectors are constant during the corrector phase if it is performed in the principal
coordinate system. This fact can be used to greatly simplify the return mapping process. To realise
these simplications, it is necessary to rotate the predictor state into its principal orientation before
performing the plastic return. The consistent tangent matrix can then be dened in principal coordi-
nates before it and the returned state are rotated back into general coordinates. In the following, the
symbols

and

denote the same term in general and principal coordinate systems. The rotation of
strain and stress vectors in Voigt notation can be expressed as
c

or c

1
c

(3)
o

T
o

or o

T
o

(4)
and elasticity and compliance tensors as

e
=

T
D

1
or D

e
=

(5)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON

e
=

T
or C

e
=

T
(6)
where

is the 6 6 rotation matrix, which can be derived by following the arguments of

Clausen et al. [36] as

=
_

_
c
x
O x
c
x
O x
c
y
O x
c
y
O x
c

O x
c

O x
c
x
O x
c
y
O x
c

O x
c
x
O x
c
y
O x
c

O x
c
x
O y
c
x
O y
c
y
O y
c
y
O y
c

O y
c

O y
c
x
O y
c
y
O y
c

O x
c
x
O y
c
y
O y
c

O y
c
x
O
c
x
O
c
y
O
c
y
O
c

O
c

O
c
x
O
c
y
O
c

O
c
x
O
c
y
O
c

O
2c
x
O x
c
x
O y
2c
y
O x
c
y
O y
2c

O x
c

O y
c
x
O x
c
y
O y
c
x
O y
c
y
O x
c

O x
c
x
O y
c

O y
c
x
O x
c
y
O x
c

O y
c
y
O y
c

O x
2c
x
O
c
x
O x
2c
y
O
c
y
O x
2c

O
c

O x
c
x
O
c
y
O x
c
x
O x
c
y
O
c

O
c
x
O x
c

O x
c
x
O
c
y
O
c

O x
c
y
O x
c

O
2c
x
O y
c
x
O
2c
y
O y
c
y
O
2c

O y
c

O
c
x
O y
c
y
O
c
x
O
c
y
O y
c

O y
c
x
O
c

O
c
x
O y
c
y
O y
c

O
c
y
O
c

O y
_

_
(7)
in which c
i
O
j
=cos
_
[
i
O
j
_
, or the direction cosine of the angle
_
[
i
O
j
_
from axis

to axis i .
Hookes law can be written in terms of the elastic and plastic parts of the total strain as
o

=

D

e
_
c

p
_
=

D

e
c

e
(8)
where

D

e
is the elasticity matrix in principal strain orientations.
The associative plastic ow rule for stress-based criterion , and equivalent strain-based criterion
g, can be expressed in rate form as

p
=dz
d
d o

=dz

e dg
d c

e
(9)
where

C

e
=
_

D

e
_
1
, is the compliance matrix in principal strain orientations and Jz is the plastic
multiplier. The presence of

C

e
in Equation (9) is required to satisfy Iliushins postulate [38]. The
predictorcorrector phase can be written in terms of stress and strain as
o

= o

T
dz

D

e d
d o

(10)
and

D

e
c

e
=

D

e
c

e
T
dz
dg
d c

e
(11)
respectively, where o

T
and c

e
T
are the trial or predictor states of stress and elastic strain. Rearranging
Equation (11) and premultiplying by

C

e
, we have
dz

e dg
d c

e
= c

e
T
c

e
(12)
Premultiplying Equation (12) by
_
dg,d c

e
_
T
and alternatively
_
dg,d c

e
T
_
T
gives
dz
_
dg
d c

e
_
T

e dg
d c

e
=
_
dg
d c

e
_
T
_
c

e
T
c

e
_
(13)
and Jz
_
dg
d c

e
T
_
T

e dg
d c

e
=
_
dg
d c

e
T
_
T
_
c

e
T
c

e
_
(14)
respectively. Following the analysis of convex surfaces by Hiriart-Urruty and Lemarechal [39] and
yield planes by Huang and Grifths [37], we can write
_
d
d o

_
T
_
o

T
o

_
6
_
o

T
_
6
_
d
d o

T
_
T
_
o

T
o

_
(15)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
and
_
dg
d c

e
_
T
_
c

e
T
c

e
_
6 g
_
c

e
T
_
6
_
dg
d c

e
T
_
T
_
c

e
T
c

e
_
(16)
Substituting the left hand sides of Equations (13) and (14) into Equation (16) we get
_
dg
d c

e
_
T

e dg
d c

e
6
g
_
c

e
T
_
dz
6
_
dg
d c

e
T
_
T

e dg
d c

e
(17)
If the yield surface is piecewise linear and the return is to a single surface then dg,d c

e
=dg,d c

e
T
will remain constant throughout the return in principal coordinates. Therefore, Equation (17)
becomes
_
dg
d c

e
_
T

e dg
d c

e
=
g
_
c

e
T
_
dz
=
_
dg
d c

e
T
_
T

e dg
d c

e
giving dz =
g
_
c

e
T
_
_
@g
@O

e
_
T

e
@g
@O

e
=
g
_
c

e
T
_
_
@g
@O

e
T
_
T

e
@g
@O

e
T
(18)
Substituting Equation (18) into Equation (11) we have

e
c

e
=

D

e
c

e
T

_

_
g
_
c

e
T
_
_
@g
@O

e
T
_
T

e
@g
@O

e
T
_

_
dg
d c

e
T
c

e
= c

e
T

_

_
g
_
c

e
T
_
_
@g
@O

e
T
_
T

e
@g
@O

e
T
_

_

C

e dg
d c

e
T
(19)
Equation (19) represents the core return mapping process for returns to a plane. However, in the
present yield criterion, it is also necessary for states of strain that lie in singular regions to be
returned to the following: lines (intersection of two yield planes) and points (intersections of three
yield planes). Singularity indicators are developed later in this section to determine if a return to
these locations is required.
We rst consider the situation in which the singularity indicators reveal that a return to a line is
required. The equation of a line on the yield surface can be expressed as the parametric equation
c

e
=t r

l
c

l
(20)
where t is a scalar parameter, r

l
is a vector parallel to the line and c

l
is an arbitrary point on the
line. An innite number of predictor states may require a return to a given point on a line, therefore,
the direction of J c

p
is unknown. However, the fact that the rst derivatives of the yield surfaces
are orthogonal to the line can be used to derive a closed form expression for the corrected state.
Substitution of Equation (20) into Equation (12) gives
d c

p
=dz

e dg
d c

e
=
_
c

e
T

_
t r

l
c

l
__
(21)
which when premultiplied by

D

e
gives
dz
dg
d c

e
=

D

e
__
c

e
T
c

l
_
t r

l
_
(22)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
Rearranging Equation (22) and premultiplying by
_
r

l
_
T
produces
dz
_
r

l
_
T
dg
d c

e
t
_
r

l
_
T

e
r

l
=
_
r

l
_
T

e
_
c

e
T
c

l
_
(23)
Noting that
_
r

l
_
T
dg
d c

e
=0 (24)
because of the orthogonality condition, the parameter t can be evaluated as
t =
_
r

l
_
T

e
_
c

e
T
c

l
_
_
r

l
_
T

e
r

l
(25)
Substitution of Equation (25) into Equation (20) gives the corrected state as
c

e
=
_

_
_
r

l
_
T

e
_
c

e
T
c

l
_
_
r

l
_
T

e
r

l
_

_ r

l
c

l
(26)
In the case of a return to a point, the corrected state can simply be written down as the coordinates
of the point in the principal space, that is
c

e
=Y
t=c
Y
t=c
Y
t=c
0 0 0|
T
(27)
where Y
t=c
represents the yield strain in tension or compression as appropriate.
2.3. Derivation of the singularity indicators
The equations derived in Section 2.2 can be used to return any predictor state of strain to the yield
surface (Figure 2). It remains to identify which states of strain are required to be returned to a plane,
a line or a point. This problem is confounded by the fact that the return vectors in principal strain
space are not orthogonal to the yield surface (e.g. the predictorcorrector line plotted in Figure 2).
Note that although dg,d c

e
is orthogonal to g, the term

C

e
(dg,d c

)
e
is not. Predictor states of strain
therefore exist, which are in violation of only one yield surface, yet when corrected are in viola-
tion of another. Such strain states are a subset of the singular states. To overcome this obstacle, we
require the derivation of singularity indicators similar to those derived by de Borst et al. [40] and
Pankaj and Bicanic [33] for the MohrCoulomb yield criterion.
Following an argument similar to that of Pankaj and Bicanic [33] for stress space, we continue
in principal strain space, which simplies the algebra required. The process used here effectively
checks whether the elastic principal strains have changed order following a return(s) to the yield
surface(s). By denition, the predictor principal strains, are ranked c
e
T1
> c
e
T2
> c
e
T3
. A singular
predictor state must exist if, following a return(s), the nal elastic principal strains are ranked
other than c
e
1
> c
e
2
> c
e
3
. In principal strain space, the rst derivatives of the yield functions
(Equation (2)) are
dg
1t
d c

e
=
_
_
_
1
0
0
_
_
_
:
dg
2t
d c

e
=
_
_
_
0
1
0
_
_
_
:
dg
3t
d c

e
=
_
_
_
0
0
1
_
_
_
dg
1c
d c

e
=
_
_
_
1
0
0
_
_
_
:
dg
2c
d c

e
=
_
_
_
0
1
0
_
_
_
:
dg
3c
d c

e
=
_
_
_
0
0
1
_
_
_
(28)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
(a)
(b)
Figure 2. The strain-based yield surface (a) in principal strain space and (b) plane representation.
A predictorcorrector step is shown between states of strain c

i
and c

iC1
.
Writing Equation (19) in vector form, we therefore get one of the following
_
_
_
c
e
1
c
e
2
c
e
3
_
_
_
=
_
_
_
c
e
T1
c
e
T2
c
e
T3
_
_
_

g
it

C
e
i i
_

C
e
1i

C
e
2i

C
e
3i
_

_
(29)
_
_
_
c
e
1
c
e
2
c
e
3
_
_
_
=
_
_
_
c
e
T1
c
e
T2
c
e
T3
_
_
_

g
ic

C
e
i i
_

C
e
1i

C
e
2i

C
e
3i
_

_
(30)
where i =1, 2, 3. From these expressions, we can evaluate the corrected principal strains and check
whether the ranking order has changed. A ranking order change denotes a singular state, and this
fact can be utilised to derive singularity indicators. Here, we consider two example derivations; the
rst associated with a line singularity and the second with a point singularity. Other indicators can
be similarly derived.
2.3.1. Example 1 - Singular in tensiontension region. Consider violation of the yield plane g
1t
.
A single return to this yield plane (Equation (29)) results in
c
e
1
=c
e
T1
g
1t
(31)
c
e
2
=c
e
T2
g
1t

C
e
21

C
e
11
(32)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
If this return results in c
e
2
> c
e
1
, then the predictor strain state is in a singular regime, and we can
dene a singularity indicator for the tensiontension region as
j
tt
=c
e
2
c
e
1
> 0
or j
tt
=
_
c
e
T2
g
1t

C
e
21

C
e
11
_

c
e
T1
g
1t
_
(33)
A value of j
t t
> 0 will indicate the need for a return to the line at the intersection of the yield
planes g
1t
and g
2t
.
2.3.2. Example 2 - Singular in tensiontensiontension region. Consider a situation with j
tt
> 0,
indicating that the return needs to be made to the line as discussed earlier. A return to the line
marking the intersection of yield planes g
1t
and g
2t
leads to
c

e
=t r

l
c

l
_
_
_
c
e
1
c
e
2
c
e
3
_
_
_
=t
_
_
_
0
0
1
_
_
_

_
_
_
Y
t
Y
t
0
_
_
_
(34)
from Equation (20). The parameter t is dened as in Equation (25), such that in the present example
t =

D
e
31
(c
T1
Y
t
)

D
e
32
(c
T2
Y
t
)

D
e
33
c
T3

D
e
33
(35)
If this return results in c
e
3
> c
e
1
, then the predictor state is in a singular region, which requires a
return to a point. We can then dene a singularity indicator as
j
ttt
=c
e
3
c
e
1
> 0
or j
ttt
=t Y
t
=
_

D
e
31
(c
T1
Y
t
)

D
e
32
(c
T2
Y
t
)

D
e
33
c
T3

D
e
33
_
Y
t
(36)
2.4. Elasto-plastic tangential modulus matrix
The FE solution of an elasto-plastic problem involves iterative solution typically governed by a
NewtonRaphson scheme, which requires the evaluation of a tangential modulus matrix

D

ep
relat-
ing the incremental change in stress induced by an increment of strain. An increment of stress can
be expressed as
d o

=

D

e
d c

dz
dg
d c

e
(37)
and an elastic strain increment as
d c

e
=d c

dz

e dg
d c

e
(38)
2.4.1. Tangential modulus matrixstrain state on a plane. In the absence of hardening, the
consistency equation can be written as
dg =
_
dg
d c

e
_
T
d c

e
=0 (39)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
substituting from Equation (38) for d c

e
in Equation (39) gives
dg =
_
dg
d c

e
_
T
_
d c

dz

e dg
d c

e
_
=0 (40)
Rearranging Equation (40) gives
_
dg
d c

e
_
T
d c

=dz
_
dg
d c

e
_
T

e dg
d c

e
(41)
and
dz =
_
@g
@O

e
_
T
d c

_
@g
@O

e
_
T

e
@g
@O

e
(42)
which can be substituted into Equation (37) to produce
d o

=
_

_

D

@g
@O

e
_
@g
@O

e
_
T
_
@g
@O

e
_
T

e
@g
@O

e
_

_
d c

J o

=

D

ep
d c

(43)
2.4.2. Tangential modulus matrixstrain state on a line. In the case of a strain state on a line at the
intersection of multiple yield surfaces,

D

ep
can be evaluated as the summation of two 66 matrices
in the form adopted by Clausen et al. [36] for isotropic stress-based yield criteria in principal space

ep
=(

D

)
ep

(44)
where (

D

)
ep
contains elements relating only to normal strain components and is thus non-zero
only in the rst three rows and columns, and

G

contains only the shear stiffness terms rotated into

principal directions (following Equation (5)), that is

T
_
_
0

33
0

33
0

33
D

e
ij
33
_
_

1
, for i , =4, 5, 6 (45)
For a return to a line, the direction of the elastic strain increment, dened in principal strain space,
must be parallel to the line. Therefore,
d c

e
=dj r

l
(46)
where dj is a scalar multiple. The corresponding increment of stress must be dened as
d o

=dj

D

e
r

l
(47)
and

D

ep
must be singular with respect to any vector orthogonal to

D

e
r

l
. Hence, from Equation (37)
and its denition in Equation (43),

D

ep
can be written as

ep
d c

e
=

D

e
d c

e
(48)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
and therefore
(

D

)
ep
r

l
=

D

e
r

l
(

D

)
ep
r

l
_
_
r

l
_
T

e
_
=

D

e
r

l
_
_
r

l
_
T

e
_
(

D

)
ep
=

e
r

l
_
r

l
_
T

e
_
r

l
_
T

e
r

l
(49)
2.4.3. Tangential modulus matrixstrain state at a point. In the present yield criterion, only
singularity points dened by the intersection of three yield surfaces exist. Following the arguments
of [36] at such a point, the tangent modulus matrix is singular with respect to any direction in
principal space, but not with respect to the shear directions. Hence,

ep
=

G

(50)
where

G

is dened in Equation (45).

2.5. Consistent tangent matrix
Use of the tangent modulus matrix

D

ep
derived in Section 2.4 has been shown to impair the
quadratic rate of convergence of the global iteration scheme [41]. Simo and Taylor [42] developed
a consistent tangent matrix to restore the quadratic convergence of the global Newton scheme. This
can be expressed as

epc
=T

ep
(51)
where T

is a modication matrix generally dened as

T

=
_
1

^zD

e
d
_
d ,do

_
do

_
1
(52)
in which the derivative is taken at the corrected state, and 1

is the identity matrix. In the case of

linear yield criteria (such as the present one), it has been established that T

can be evaluated at the

predictor state as
T

=1

^zD

e
d
_
d ,do

_
do

(53)
thus avoiding the inversion required in Equation (52) [43]. Conversion of Equation (53) for a yield
surface dened in strain space gives
T

=1

^zC

e
d
_
dg,dc

e
_
dc

e
=1

d
_
^zC

e
_
dg,dc

e
__
dc

e
=1

d^c

p
dc

e
(54)
where the plastic corrector in general strain space, ^c

p
can be expressed in terms of its
representation in principal strain space, using Equations (3) and (7), as
T

=1

d^c

p
dc

e
=1

d
_

1
^ c

p
_
dc

e
(55)
which can be expanded, noting the matrix identity d

1
,d. =

1
_
d

,d.
_

1
[44, 45], with
the substitution
I

i
=

1
d

dc

e
i

1
^ c

p
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
such that
T

=1

_
I

x
I

y
I

xy
I

x
I

y
_
(56)
The term d

,dc

e
i
represents the rate of change of the principal strain axes, which can be expressed
using the following geometrical arguments, similar to Clausen et al. [36], for principal stress axes.
Expanding d

,dc

e
i
with the chain rule gives
d

dc

e
i
=
d

d[
x
d[
x
dc

e
i

d[
y
d[
y
dc

e
i

d[

d[

dc

e
i
(57)
where J[
x
, J[
y
, and J[

are innitesimal rotation angles about the .-axis, ,-axis and z-axis. In
the case that the principal ( ., ,, z) and general (., ,, z) axes are aligned, the angles between them
can be represented as the tensor
[

0
=
_

_
[
x
O x
[
x
O y
[
x
O
[
y
O x
[
y
O y
[
y
O
[

O x
[

O y
[

O
_

_ =
_
_
0

2

2
0

2

2
0
_
_
(58)
from which the corresponding transformation tensor can be derived as
C

0
=cos [

0
=
_
_
1 0 0
0 1 0
0 0 1
_
_
(59)
Innitesimal rotations of the coordinate system about each of the axes are illustrated in Figure 3 and
produce the transformation tensors
C

x
=cos
_
_
0

2

2
J[
x
2
J[
x

2
J[
x
J[
x
_
_
=
_
_
1 0 0
0 1 J[
x
0 J[
x
1
_
_
(60)
C

y
=cos
_
_
J[
y
2

2
J[
y

2
0

2

2
J[
y
2
J[
y
_
_
=
_
_
1 0 J[
y
0 1 0
J[
y
0 1
_
_
(61)
and
C

=cos
_
_
J[

2
J[

2

2
J[

J[

2

2
0
_
_
=
_
_
1 J[

0
J[

1 0
0 0 1
_
_
(62)
Figure 3. Innitesimal rotation angles about the three coordinate axes; (a) angle J[
x
about the .-axis, (b)
angle J[
y
about the ,-axis, and (c) angle J[

about the z-axis.

Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
respectively. It may be noted that these equations are slightly different from those obtained by
Clausen et al. [36]. From Equations (59)(62) the change in transformation matrix

(Equation (7))
resulting from a rotation around each axis can be expressed as
d

d[
x
J[
x
=

_
C

x
_

_
C

0
_
=
_

_
0 0 0
0

33
0 0 J[
x
0 0 J[
x
0 0 0 0 J[
x
0
0 0 0 J[
x
0 0
0 2J[
x
2J[
x
0 0 0
_

_
(63)
d

d[
y
J[
y
=

_
C

y
_

_
C

0
_
=
_

_
0 J[
y
0
0

33
0 0 0
0 J[
y
0
0 0 0 0 0 J[
y
2J[
y
0 2J[
y
0 0 0
0 0 0 J[
y
0 0
_

_
(64)
and
d

d[

J[

_
C

_
C

0
_
=
_

_
J[

0 0
0

33
J[

0 0
0 0 0
2J[

2J[

0 0 0 0
0 0 0 0 0 J[

0 0 0 0 J[

0
_

_
(65)
where quadratic terms are neglected. To evaluate the terms of Equation (57), it is also necessary
to calculate J[
x
,Jc

e
i
, J[
y
,Jc

e
i
and J[

,Jc

e
i
. Innitesimal changes in the normal strains do not
affect the orientation of the principal strain directions, hence,
J[
i
Jc
e
j
=0

=
d

dc
e
j
=0

=I

j
=0

, for i , =., ,, z (66)

Through consideration of the strain Mohrs circles (Figure 4) the rate of change of the rotation
angles with the shear strains ;
e
xy
, ;
e
x
and ;
e
y
can be found as
Figure 4. Mohrs circles of strain for innitesimal rotation angles J[
x
, J[
y
and J[

.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
sin(2J[
x
) =
1
2
J;
e
y
1
2
_
c
e
2
c
e
3
_ =
J[
x
J;
e
y
=
1
2
_
c
e
2
c
e
3
_ (67)
sin(2J[
y
) =
1
2
J;
e
x
1
2
_
c
e
1
c
e
3
_ =
J[
y
J;
e
x
=
1
2
_
c
e
1
c
e
3
_ (68)
sin(2J[

) =
1
2
J;
e
xy
1
2
_
c
e
1
c
e
2
_ =
J[

J;
e
xy
=
1
2
_
c
e
1
c
e
2
_ (69)
noting that for small angles, sin 0 ~0. Substitution of Equation (67) into Equation (63) gives
d

d;
e
y
=
1
2
_
c
e
2
c
e
3
_
_

_
0 0 0
0

33
0 0 1
0 0 1
0 0 0 0 1 0
0 0 0 1 0 0
0 2 2 0 0 0
_

_
(70)
and similarly from Equation (68) and Equation (64),
d

d;
e
x
=
1
2
_
c
e
1
c
e
3
_
_

_
0 1 0
0

33
0 0 0
0 1 0
0 0 0 0 0 1
2 0 2 0 0 0
0 0 0 1 0 0
_

_
(71)
and Equation (69) into Equation (65)
d

d;
e
xy
=
1
2
_
c
e
1
c
e
2
_
_

_
1 0 0
0

33
1 0 0
0 0 0
2 2 0 0 0 0
0 0 0 0 0 1
0 0 0 0 1 0
_

_
(72)
The consistent modier T

can be calculated by inserting Equations (66) and (70)(72) into

Equation (56) such that
T

=1

_
0

xy
I

x
I

y
_
(73)
2.6. Overview of the algorithm
1.Initialisation:
Pass in, dc

, c

e
, c

p
Set, Y
t
, Y
c
, D

e
, C

e
=
_
D

e
_
1
, c

e
T
=c

e
Jc

2.Transform c

e
T
, D

e
and C

e
into principal coordinates c

e
T
,

D

e
and

e
using Equations (3), (5) and (6)
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
3.Check yield conditions
IF all g( c

e
) 60: no return required,
c

e
= c

e
T
,

D

epc
=

D

e
GO TO 5
ELSE: strain return required,
GO TO 4
4.Check singularity indicators to identify type of return
IF return to a plane:
Calculate J c

p
from Equation (19) and

D

ep
from Equation (43)
c

e
= c

e
T
J c

p
ELSE IF return to a line:
Calculate c

e
from Equation (26) and

D

ep
from Equations (44) and (49)
ELSE IF return to a point:
Calculate c

e
from Equation (27) and

D

ep
from Equation (50)
END IF
Calculate T

from Equation (73)

epc
=T

ep
5.Rotate c

e
and

D

epc
back into general coordinates c

e
and D

epc
using Equations 3 and 5
6.Update stress and state variables
o

=D

e
c

e
, c

p
=c

e
7.Return o

, D

epc
, c

e
and c

p
3. BENCHMARK TESTS
The algorithm was coded as a Fortran user subroutine appended to Abaqus Standard (ABAQUS,
Inc., Providence, RI, USA). The applied conditions were uniaxial tension and compression; hydro-
static tension and compression; triaxial combinations of tension and compression; and planar and
triaxial pure shear. Uniaxial tests (Figure 5(a) and (b)) were applied by xing all nodes on one face
in the z direction and one of these also in the . and , directions; equal displacement was applied to
the remaining nodes in the z direction. In hydrostatic tests (Figure 5(c) and (d)), one node was xed
in all directions, and all nodes with a common face were restrained against translation orthogonally
to that face; equal displacements were applied orthogonally to nodes on the unrestrained faces. In
triaxial tension and compression tests (Figure 5(e) and (f)), nodes were restrained identically to the
hydrostatic case, but the displacements were applied such that the ratio of tensile to compressive
strain was Y
t
,Y
c
. In the planar shear test (Figure 5(g)), one node was fully xed, and other nodes
were displaced to produce a symmetrical diamond shape within the testing plane, without restraint
to displacement out of that plane. In the triaxial shear test (Figure 5(h)), one node was fully xed,
and all others were displaced to produce a symmetrical diamond in 3D. The tests were performed
using tensile and compressive yield strains of 0.5% and 0.7%, respectively. These are typical yield
strains reported for bone [46]. In the benchmark tests, the material was assumed to be isotropic,
to maintain transparency of the solution, with Youngs modulus of 10 GPa and Poissons ration of
0.3. In the following, the results shown are the three principal strain components. Their elastic and
plastic components are plotted against the largest principal component of applied total strain in each
case. The results of the benchmark tests are considered in the following paragraphs. A discussion of
the results is provided in the following section.
Results of uniaxial tests are illustrated in Figure 6(a) (compression) and (b) (tension). Under
uniaxial compression, the elastic strain increased linearly until the yield strain in compression was
reached. Initial plastic returns were made to a single yield surface. Following rst yield, there was no
further increment in the elastic strain component. The minimum elastic principal strain component
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
(a) (b) (c)
(d) (e)
(g)
(h)
(f)
Figure 5. Single element benchmark test boundary conditions for (a) uniaxial compression, (b) uniaxial
tension, (c) hydrostatic compression, (d) hydrostatic tension, (e) compressivecompressivetensile,
(f) tensiletensilecompressive, (g) plane pure shear, and (h) triaxial pure shear. Note D
t
, D
c
and D
s
are
the applied tensile, compressive and shear displacements, respectively.
remained at a constant value of 0.7%. All further applied strain became plastic strain. A similar
pattern of yielding occurred in tension; rst yield occurred when the maximum elastic principal
strain reached the tensile yield strain and remained at a constant value of 0.5% thereafter. In both
cases, it was noted that once rst yield occurred in one direction, the elastic strain in other unyielded
directions did not increase.
The results of hydrostatic tests are illustrated in Figure 6(c) (compression) and (d) (tension).
In Figure 6(c), the plastic strain remained zero until the compressive yield surface was reached.
Plastic returns were made to three yield surfaces. This occurred simultaneously in all principal
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
(a) (b)
(c) (d)
Figure 6. Single element uniaxial and hydrostatic tests of the strain-based yield criterion under (a) uniaxial
compressive, (b) uniaxial tensile, (c) hydrostatic compressive and (d) hydrostatic tensile displacements.
directions. Under further displacement, the elastic strain remained constant at a value of 0.7%,
and all additional strain became plastic. Equivalent behaviour was observed under conditions of
hydrostatic tension (Figure 6(d)). The elastic strain reached the tensile yield strain simultaneously
in all principal directions; further strain became plastic. The limiting strain was 0.5%.
Under compressioncompressiontension conditions, the response was as shown in Figure 7(a).
The ratio of tensile to compressive applied displacements was Y
t
,Y
c
. The principal elastic strains
increased until rst yield occurred; in this case, at two compressive and one tensile yield planes
simultaneously. Further strain induced plastic strain in all principal directions. Once yielded, the
elastic strain components remained constant at their respective yield values. Further applied strain
became plastic. Similar behaviour was observed under tensiontensioncompression conditions,
illustrated in Figure 7(b). In this case, tensile yield occurred for both the maximum and intermedi-
ate principal strain component and compressive yield for the minimum. At post-yield, the principal
elastic strains remained at the yield strain values.
Figure 7(c) shows the response under pure shear displacements in a single plane. Under these con-
ditions, the applied shear strain, ;
app
can be resolved into principal strains of c
1
=;
app
,2, c
2
=0,
and c
3
= ;
app
,2. Therefore, because the yield strain was lower in tension, it was the mode of
rst yield. Following rst yield, a component of plastic strain was induced in all principal directions
caused by the Poisson effect. This reduced the rate of increase of the elastic minimum principal
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
(a) (b)
(c)
(d)
Figure 7. Single element triaxial and shear tests of the strain-based yield criterion. Applied boundary condi-
tions were (a) compressivecompressivetensile, (b) tensiletensilecompressive, (c) plane pure shear and
(d) triaxial pure shear displacements.
strain component (see kinks at points A and B in Figure 7(c)), and induced an equal and opposite
component of elastic strain in the direction of the middle principal strain to satisfy the boundary
condition c
2
= 0. The minimum principal elastic strain then increased until the compressive yield
strain was reached. At this point, no further elastic strain accumulated; further displacement induced
only plastic strain. The principal elastic strains in yielded directions remained at the yield values.
The results of a triaxial state of pure shear are presented in Figure 7(d). The applied shear strains of
;
12
=;
23
=;
31
=;
app
resulted in applied principal strains of c
1
=;
app
, and c
2
=c
3
=;
app
,2.
The elastic strains increased until rst yield occurred in tension. Further straining induced plas-
tic strain in all principal directions. This reduced the rate of accumulation of elastic strain in the
unyielded middle and minimum principal directions (see kinks at points A and B in Figure 7(c)).
Elastic strain accumulated in the unyielded directions until the compressive yield strain was reached.
Following yield, all elastic principal strain components remained at their respective yield strains.
Cyclic loading tests were conducted to examine the behaviour of the developed algorithm under
unloading and reloading. The elastic and plastic strain components, which developed in the single
element subjected to several cycles of uniaxial tensile and compressive displacements are plotted
in Figure 8(a) and (b). Under the initial tensile conditions, the behaviour was linear until it reached
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
(a) (b)
Figure 8. Single element test under cycles of uniaxial tensile and compressive applied strain in the z
direction; (a) elastic and (c) plastic strain components.
the tensile yield strain Y
t
. Prescribed displacement was then progressed until 0.1% plastic strain at
which point the applied displacements were reduced. The elastic strain returned along the pre-yield
gradient with a non-zero total strain component, where the elastic strain was reduced to zero. Addi-
tional compressive displacement was applied until yield occurred on the compressive yield surface.
This was encountered at a total strain value of 0.6%, 0.1% higher than the assigned yield strain as
a result of the preceding tensile plastic strain in the z direction. Compressive displacement was con-
tinued up to 0.3% plastic strain. Two further cycles of tensile followed by compressive loading were
applied and produced similar behaviour. The principal elastic strain components remained within
the assigned yield strains throughout all cycles.
4. DISCUSSION OF BENCHMARK TESTS
Under uniaxial tension and compression, it was demonstrated that the elastic predictorplastic
corrector algorithm worked correctly for non-singular states of strain. The various tests under hydro-
static and combinations of tension and compression involved singular strain states in violation of
both two and three yield surfaces. In every case, the algorithm was seen to achieve a valid plas-
tic return to the yield surface. The planar and triaxial pure shear conditions involved the greatest
possible rotation of the principal strain state from the model coordinate system. This represented
an additional test for the developed algorithm. Under these conditions, the singular strain states
were resolved successfully. Under repeated cyclical uniaxial tensile and compressive strains, the
algorithm correctly resolved the elastic and plastic strain components. These results indicate that
the developed algorithm is able to accurately model the maximum/minimum principal strain yield
criterion in general FE analyses, including loading, unloading and reverse loading conditions.
4.1. Yielding and the Poisson effect
In the results of uniaxial single element tests, it was observed that following yielding in one direc-
tion, the elastic Poisson effect in other directions ceased. More specically, the component of elastic
strain induced by Poissons effect was replaced by an equivalent plastic strain. For example, this
behaviour can be seen in Figure 6(a) with uniaxial compression. Preceding rst yield, the elas-
tic compressive strain increased and was accompanied by associated tensile Poisson strains in the
orthogonal directions. Following rst yield, both the compressive and tensile elastic strain compo-
nents remained constant, although only the former had exceeded the yield surface. This behaviour
was accompanied by the accumulation of plastic strains at an equivalent rate to the pre-yield state.
It is important to note that the Poisson effect is related to elastic strain, and once the elastic strain
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
becomes constant in the principal loading direction, there will be no additional elastic strain because
of the Poisson effect in the orthogonal directions.
In the case of multiaxial and shear tests, similar behaviour was observed. However, following rst
yield in these cases, the rate of increase of elastic strain reduced, but did not cease in the unyielded
orthogonal directions. Although the applied loading at the element boundary continued to increase
at the same rate, both pre-yield and post-yield, the fact that the rate of increase of elastic strain
reduced in the unyielded directions requires explanation.
To explain this, we rst consider the uniaxial case in more detail. Consider an isotropic solid with
elastic compliance tensor

C

e
. If we assume an initial state of strain, which lies on the tensile yield
surface and apply an additional tensile unit stress ^ o

, the elastic predictor, or trial strain increment

in principal space is
^ c

e
T
=

C

e
^ o

=
1
1
_
_
1 v v
v 1 v
v v 1
_
_
_
_
_
1
0
0
_
_
_
=
1
1
_
_
_
1
v
v
_
_
_
(74)
From Equation (12), the plastic strain is
^ c

p
=dz

e dg
d c

e
=dz

e
_
_
_
1
0
0
_
_
_
=
Jz
1
_
_
_
1
v
v
_
_
_
(75)
The increment of elastic principal strain following the plastic return will therefore be
^ c

e
=^ c

e
T
^ c

p
=
1
1
_
_
_
1
v
v
_
_
_

Jz
1
_
_
_
1
v
v
_
_
_
=
1 dz
1
_
_
_
1
v
v
_
_
_
(76)
Although the initial state lay on the tensile yield surface, it follows that the increment of maxi-
mum elastic principal strain must be zero
_
^c
e
1
=0
_
. Therefore, dz = 1 and the plastic increment
exactly matches and eliminates the elastic trial strain. This process effectively converts all compo-
nents of the trial strain state into identical plastic strain components, even if only one of them is in
violation of the yield surface. It should be noted that this behaviour is induced by the presence of
the elastic compliance tensor

C

e
in the ow vector, which is necessary to make a return in strain
space physically valid [38].
Following this discussion, it becomes possible to understand the observed reduction in the rate
of elastic strain for unyielded directions in multiaxial loading conditions. Following rst yield, the
associated Poisson strain in the orthogonal directions becomes plastic. Although the apparent strain
increases at the same rate, to retain equilibrium, the elastic component of strain in the unyielded
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
directions must be reduced by the amount of plastic strain induced by the Poisson effect. To illus-
trate this, consider the case of a planar pure shear boundary condition on an isotropic material
(Figure 7(c)). As previously noted, an applied shear strain of ;
app
resolves into principal strains of
c
1
=;
app
,2, c
2
=0, and c
3
=;
app
,2. The initial state lies upon the tensile yield surface, and an
additional stress ^ o

is applied
^ c

e
T
=

C

e
^ o

=
1
1
_
_
1 v v
v 1 v
v v 1
_
_
_
_
_
1
0
1
_
_
_
=
1
1
_
_
_
1 v
0
1 v
_
_
_
(77)
Using equation 75, the increment of elastic principal strain following the plastic return will
therefore be
^ c

e
=^ c

e
T
^ c

p
=
1
1
_
_
_
1 v
0
1 v
_
_
_

Jz
1
_
_
_
1
v
v
_
_
_
(78)
To achieve a return to the tensile yield surface
_
^c
e
1
=0
_
, this implies Jz =1 v. Therefore,
^ c

e
=
1
1
_
_
_
0
v(1 v)
(v 1)(v 1)
_
_
_
(79)
Here, the discussed plastic Poisson effect changed the value of c
e
2
from zero in the elastic regime,
to proportional to v(1v) following rst yield. Furthermore, the absolute rate of increase in c
e
3
was
reduced by the same effect. This proportional reduction can be calculated as
^c
e
3
^c
e
T3
=
1
E
(1 v)(v 1)

1
E
(1 v)
=1 v (80)
It can similarly be demonstrated that under triaxial shear conditions (Figure 7(d)), the proportional
reduction of the rate of increase in elastic strain following yield in another orthogonal direction is
1 2v. These results demonstrate that the observed behaviour results entirely from the Poisson
effect, which is only present because of the requirements of Iliushinss postulate [38].
5. CONCLUSIONS
A simple but efcient and robust strain-based plasticity algorithm, based on minimum and maxi-
mum principal strain, has been developed for simulating the post-elastic behaviour of bone. Unlike
stress-based criteria, the isotropic nature of the criterion is particularly advantageous. Singularity
indicators are developed to direct the algorithm to make plastic corrector return to the singular
locations of the criterion. The developed algorithms permit a plastic corrector to be achieved in a
single iterative step in all cases. The algorithm was implemented in a 3D nite element package,
and the benchmark tests conducted show that the constitutive behaviour is as expected.
ACKNOWLEDGEMENTS
The authors are grateful to Dr Johan Christian Clausen of Aalborg University, Denmark for his helpful cor-
respondence. Funding from the Carnegie Trust for the Universities of Scotland is gratefully acknowledged.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
ALGORITHMS FOR A STRAIN-BASED PLASTICITY CRITERION FOR BONE
CONFLICT OF INTEREST STATEMENT
The authors have no conict of interest with respect to the presented work.
REFERENCES
1. Tsai SW, Wu EM. A general theory of strength for anisotropic materials. Journal of Composite Materials 1971;
5:5880.
2. Donaldson F, Pankaj P, Law AH, Simpson AH. Virtual trabecular bone models and their mechanical response.
Proceedings of the Institution of Mechanical Engineers, H: The Journal of Engineering in Medicine 2008;
222:11851195.
3. Fenech CM, Keaveny TM. A cellular solid criterion for predicting the axialshear failure properties of bovine
trabecular bone. Journal of Biomechanical Engineering 1999; 121:414422.
4. Ford CM, Keaveny TM, Hayes WC. The effect of impact direction on the structural capacity of the proximal femur
during falls. Journal of Bone and Mineral Research 1996; 11:377383.
5. Keyak JH. Improved prediction of proximal femoral fracture load using nonlinear nite element models. Medical
Engineering & Physics 2001; 23:165173.
6. Keyak JH, Rossi SA, Jones KA, Skinner HB. Prediction of femoral fracture load using automated nite element
modeling. Journal of Biomechanics 1998; 31:125133.
7. Keyak JH, Rossi SA, Jones KA, Les CM, Skinner HB. Prediction of fracture location in the proximal femur using
nite element models. Medical Engineering & Physics 2001; 23:657664.
8. Keyak JH, Rossi SA. Prediction of femoral fracture load using nite element models: an examination of stress- and
strain-based failure theories. Journal of Biomechanics 2000; 33(2):209214.
9. Lotz JC, Cheal EJ, Hayes WC. Fracture prediction for the proximal femur using nite element models: Part 1 linear
analysis. Journal of Biomechanical Engineering 1991; 113:353360.
10. Lotz JC, Cheal EJ, Hayes WC. Fracture prediction for the proximal femur using nite element models: Part
2nonlinear analysis. Journal of Biomechanical Engineering 1991; 113:361365.
11. Voor MJ, Anderson RC, Hart RT. Stress analysis of halo pin insertion by non-linear nite element modeling. Journal
of Biomechanics 1997; 30:903909.
12. Niebur GL. A computational investigation of multiaxial failure in trabecular bone. PhD Thesis, Mechanical
Engineering, Graduate Division, The University of California, Berkley, USA, 2000.
13. Doblare M, Garcia JM, Gomez MJ. Modelling bone tissue fracture and healing: a review. Engineering Fracture
Mechanics 2004; 71:18091840.
14. Keaveny TM, Wachtel EF, Ford CM, Hayes WC. Differences between the tensile and compressive strengths of
bovine tibial trabecular bone depends on modulus. Journal of Biomechanics 1994; 27(9):11371146.
15. Nalla RK, Kinney JH, Ritchie RO. Mechanistic fracture criteria for the failure of human cortical bone. Nature
Materials 2003; 2:164168.
16. Cowin SC, Mehrabadi MM. Identication of the elastic symmetry of bone and other materials. Journal of
Biomechanics 1989; 22(6/7):503515.
17. Donaldson F, Pankaj P, Cooper DML, Thomas CDL, Clement JG, Simpson AHRW. Relating age and micro-
architecture with apparent-level elastic constants: a FE study of female cortical bone from the anterior femoral
midshaft. Proceedings of the Institution of Mechanical Engineers, H: The Journal of Engineering in Medicine 2011;
225:585596.
18. Bayraktar HH, Morgan EF, Niebur GL, Morris GE, Wong EK, Keaveny TM. Comparison of the elastic and yield
properties of human femoral trabecular and cortical bone tissue. Journal of Biomechanics 2004; 37(1):2735.
19. Mercer C, He MY, Wang R, Evans AG. Mechanisms governing the inelastic deformation of cortical bone and
application to trabecular bone. Acta Biomaterialia 2006; 2:5968.
20. Vahey JW, Lewis JL, Vanderby R. Elastic moduli, yield stress, and ultimate stress of cancellous bone in the canine
proximal femur. Journal of Biomechanics 1987; 20:2933.
21. Gupta A, Bayraktar HH, Fox JC, Keaveny TM, Papadopoulos P. Constitutive modeling and algorithmic implemen-
tation of a plasticity-like model for trabecular bone structures. Computational Mechanics 2007; 40:6172.
22. Schileo E, Taddei F, Cristofolini L, Viceconti M. Subject-specic nite element models implementing a maximum
principal strain criterion are able to estimate failure risk and fracture location on human femurs tested in vitro. Journal
of Biomechanics 2008; 41:356367.
23. Naghdi PM, Trapp JA. The signicance of formulating plasticity theory with reference to loading surfaces in
strainspace. International Journal of Engineering Science 1975; 13:785797.
24. Farahat AM, Kawakami M, Ohtsu M. Strainspace plasticity model for the compressive hardeningsoftening
behaviour of concrete. Construction and Building Materials 1995; 9:4559.
25. Lan YM, Sotelino ED, Chen WF. The strainspace consistent tangent operator and return mapping algorithm
for constitutive modelling of conned concrete. International Journal of Applied Science and Engineering 2003;
1:1729.
26. Mizuno E, Hatanaka S. Compressive softening model for concrete. Journal of Engineering Mechanics 1992;
118:15461563.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm
P. PANKAJ AND F. E. DONALDSON
27. Yoder PJ, Iwan WD. On the formulation of strainspace plasticity with multiple loading surfaces. Journal of Applied
Mechanics 1981; 48:773778.
28. Youquan Y. Stress space and strain space formulation of the elasto-plastic constitutive relations for singular yield
surface. Acta Mechanica Sinica 1986; 2:169177.
29. Casey J, Naghdi PMa. Discussion: on the formulation of strainspace plasticity with multiple loading surfaces. ASME
Journal of Applied Mechanics 1982; 49:460462.
30. Casey J, Naghdi PM. On the nonequivalence of the stress space and strain space formulations of plasticity theory.
ASME Journal of Applied Mechanics 1983; 50:350354.
31. de Borst R. Integration of plasticity equations for singular yield functions. Computers and Structures 1987;
26:823829.
32. Knockaert R, Chastel Y, Massoni E. Rate-independent crystalline and polycrystalline plasticity, application to FCC
materials. International Journal of P 2000; 16:17198.
33. Pankaj P, Bicanic N. Detection of multiple active yield conditions for mohr coulomb elasto-plasticity. Computers
and Structures 1997; 62:5161.
34. Donaldson F, Pankaj P, Simpson A. Investigation of factors affecting loosening of Ilizarov ring-wire external xator
systems at the bone-wire interface. Journal of Orthopaedic Research 2012; 30:726732.
35. Oden ZM, Selvitelli DM, Bouxsein ML. Effect of local density changes on the failure load of the proximal femur.
Journal of Orthopaedic Research 1999; 17:661667.
36. Clausen J, Damkilde L, Andersen L. Efcient return algorithms for associated plasticity with multiple yield planes.
International Journal for Numerical Methods in Engineering 2006; 66:10361059.
37. Huang J, Grifths DV. Observations on return mapping algorithms for piecewise linear yield criteria. International
Journal of Geomechanics, ASCE 2008; 8:253265.
38. Iliushin AA. On the postulate of plasticity. Journal of Applied Mathematics and Mechanics 1961; 25:746752.
39. Hiriart-Urruty JB, Lemarechal C. Convex Analysis and Minimization Algorithms, Vol. 1. Springer-Verlag: Berlin,
1993.
40. de Borst R, Pankaj P, Bicanic N. A note on singularity indicators for MohrCoulomb type yield criteria. Computers
and Structures 1991; 39:219220.
41. Nagtegaal JC. On the implementation of inelastic constitutive equations with special reference to large deformation
problems. Computer Methods in Applied Mechanics and Engineering 1982; 33:469484.
42. Simo JC, Taylor RL. Consistent tangent operators for rate-independent elastoplasticity. Computer Methods in Applied
Mechanics and Engineering 1985; 48:101118.
43. Criseld MA. Nonlinear Finite Element Analysis of Solids and Structures: Advanced Topics, Vol. 2. Wiley:
New York, 1997.
44. Petersen KB, Pedersen MS. The matrix cookbook, Oct 2008. http://www2.imm.dtu.dk/pubdb/p.php?3274,
accessed15/11/2010, version 20081110.
45. Selby S. Standard Mathematical Tables. CRC Press: Cleveland, Ohio, 1975.
46. Ebacher V, Tang C, McKay H, Oxland TR, Guy P, Wang R. Strain redistribution and cracking behaviour of human
bone during bending. Bone 2007; 40:12651275.
Copyright 2012 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Biomed. Engng. (2012)
DOI: 10.1002/cnm