Sie sind auf Seite 1von 13

Arch Appl Mech (2010) 80: 543–555

DOI 10.1007/s00419-009-0382-2

SPECIAL ISSUE

D. Garcia · Philippe K. Zysset · M. Charlebois · A. Curnier

A 1D elastic plastic damage constitutive law for bone tissue

Received: 13 November 2008 / Accepted: 21 March 2009 / Published online: 22 October 2009
© Springer-Verlag 2009

Abstract Motivated by applications in orthopaedic and maxillo-facial surgery, the mechanical behaviour of
cortical bone in cyclic overloads at physiological strain rates is investigated. To this end, a new one-dimensional
rate-independent constitutive model for compact bone is proposed to simulate the damage accumulation occur-
ring during tensile or compressive overloading. We adopted a macroscopic and phenomenological description
of the mechanics of cortical bone. The mathematical formulation of the model is established within the frame-
work of generalized standard materials and is based on the definition of three internal state variables: a tensile
and a compressive damage variable that represent the microcrack density and a residual strain variable that
represents the permanent strain associated with the sliding behaviour of these microcracks. Distinct damage
threshold stresses are used in tension and compression. As the macroscopic mechanical behaviour of bovine
cortical bone is very similar to that of human cortical bone and of much easier access, we first achieved the
validation and identification of the material constants of the constitutive laws in tension using new uniaxial
experimental results of bovine compact bone of our own. With adequate original hardening rules, the consti-
tutive model is able to reproduce the main features of cortical bone behaviour under arbitrary cyclic tensile
and compressive loadings. The proposed algorithm applies for the first time three distinct projections based
on the relationship between the three internal variables and criteria. The predicted stress–strain curves exhibit
a damaged reloading which is collinear with the origin as many cyclic overloading experiments on cortical
bone have shown. Note that as our model was identified for physiological strain rates, it hardly can be applied
in high strain rate situations like the ones involved in impact studies.
Keywords Cortical bone · Generalized standard materials · Damage · Tension · Compression

1 Introduction

In view of optimizing the reliability of a bone-implant structure or predicting bone fracture, a realistic model
of the mechanical behaviour of bone is required. Hence, skeletal research has an important need for realistic
constitutive laws for bone tissue including heterogeneities, anisotropic elasticity and damage behaviour, i.e.
progressive degradation of its elastic properties. In addition to elastic stiffness reduction, quasistatic cyclic
overloading (i.e. damaging) experiments have shown that such models should account for permanent strains

D. Garcia · A. Curnier
Laboratory of Applied Mechanics and Reliability, Ecole Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland

P. K. Zysset (B) · M. Charlebois


Institute of Lightweight Design and Structural Biomechanics (ILSB), Vienna University of Technology (TU-Wien),
Gußhausstrasse 27-29, 1040 Vienna, Austria
E-mail: philippe.zysset@ilsb.tuwien.ac.at
Tel.: +43-1-5880131723
Fax: +43-1-5880131799
544 D. Garcia et al.

[10,11]. Much of the experimental work to date regarding the damage behaviour of cortical bone tissue has
been limited to simple geometries and proportional loadings.
Furthermore, little has been accomplished in describing the mechanical behaviour of cortical bone tissue
in cyclic uniaxial overloading situation. A comprehensive mathematical one-dimensional model is therefore
needed in order to interpret and quantify the damaging process. Previous damage models did not succeed in
describing the reloading phase properly and/or did not include compressive loading. On the one hand, the
general damage model proposed by [20] did not predict a linear damaged reloading curve collinear with the
origin as many tensile and compressive experiments on compact and trabecular bone have shown [7,10,11]
(i.e. after having been overloaded, cortical bone reloads with an elastic modulus equal to its undamaged mod-
ulus, but then develops a reduced modulus characteristic of a perfectly damaging material, cf. Figs. 2, 5a for
an illustration). On the other hand, the damaging visco-elastoplastic model proposed by [5] did not include
compressive loading histories.
Therefore, the objective of this study was to develop a new constitutive model for cortical bone tissue that
predicts the experimental damage stress–strain curves in cyclic tensile and compressive loading. The approach
chosen in this work is that of a macroscopic description and formulation of cortical bone tissue. Indeed, our
model does not take into consideration the hierarchical structure of the tissue, in particular the micro- and
nano-structural levels. The proposed model uses the internal state variable approach common in continuum
damage mechanics and allows a straightforward interpretation of the constitutive behaviour of cortical bone.
Although cortical bone tissue has been shown to behave as a viscoelastic inelastic damageable material
over a large strain rate range [13], rate-dependent effects have a moderate impact at physiological strain rates
[7]. Hence, in this work we chose first to include the main features of the mechanical behaviour of cortical
bone in a rate-independent model (Sect. 2).
Even if human and bovine cortical bone differ at each hierarchical level of their structure (e.g. [2]), it has
been shown that at the macroscopic level adopted in this work, their damage behaviour was observed to be
highly similar [6]. Indeed, those tissues differ at the lamellar level but not at the ultrastructural level where
damage is believed to occur [20]. As bovine cortical bone is of easy access, novel cyclic uniaxial experiments
carried out on bovine cortical bone are then presented in Sect. 3 and serve as a basis for the identification
of the material constants characterizing the model. In order to be fully applicable to human compact bone,
the values of these parameters should be adjusted, based on novel tests with human cortical bone specimens.
The relevance, accuracy and validity of the model will be fully appreciable from the comparison between
experimental and simulation results displayed in Sect. 3 and discussed in Sect. 4.

2 Rate-independent model

2.1 Rheological model

Upon analysing previous experimental results [5,6,11], three crucial regimes of compact bone behaviour under
cyclic tensile and/or compressive loading were identified. The first regime, abbreviated by (E), is pure elasticity
with modulus 0 . The second regime, a damage mode (D), which corresponds to the generation and opening
of microcracks leading to a reduction in modulus and an accumulation of permanent strain. The third regime,
an inelastic mode (I), is due to friction of closing (or sliding) of microcracks leading to residual strains but no
extension of damage. Furthermore, damaged reloading is collinear with the origin in that mode.
To account for this behaviour, a rheological model was designed using a primary elastic spring connected
in series with a damage element composed of a secondary elastic spring undergoing rate-independent damage
mounted in parallel with a rate-independent inelastic pad (Fig. 1).
The mathematical formulation of the rate-independent model is based on the definition of three internal
state variables: a tensile (D+ ∈ [0, 1[) and a compressive (D− ∈ [0, 1[) damage variable that represent the
microcrack density that reduces tissue stiffness and an inelastic strain variable E i ∈ R that represents the defor-
mation associated with these microcracks. The fourth (external) state variable is the total strain E. In order to
clarify the presentation, we present now separately the properties of each rheological element composing the
model.
The elastic spring with constant stiffness 0 mounted in series accounts for the unaltered elastic response
of the calcified extracellular bone matrix.
The inelastic pad accounts for permanent strain due to microcracks. It is characterized by two yield
stresses σ+i (D− ) ≥ 0 and −σ−i (D+ ) ≤ 0 depending on the sign of its inelastic strain rate Ė i (tension or
A 1D elastic plastic damage constitutive law for bone tissue 545

S SD
D
0
σ+
D
1−(D++D− )
E -E i (b) D := D+ +D− 0 Ei
-σ−D

(c)
(a) Si
i
σ+
-σ−i Ė i
Ei
E
Fig. 1 One-dimensional rheological setup describing cortical bone damage. It was designed using a primary elastic spring (a)
connected in series with a damage element composed of a secondary elastic spring (b) undergoing rate-independent damage
mounted in parallel with a rate-independent inelastic pad (c)

compression). The functions σ+i and σ−i depend monotonically on the damage variables D− and D+ , respec-
tively, to express a non-isotropic hardening of the inelastic yield stress. Initially equal to zero, the thresholds
increase with the accumulated damage. Motivated by experimental identification of Sect. 3, we assume the
following exponential hardening functions for the inelastic thresholds:

σ+i (D− ) = χ i (1 − exp (−l D− )) (1)


σ−i (D+ ) = χ i (1 − exp (−l D+ )) (2)

where χ i > 0 and l > 0 are inelastic hardening coefficients.


The damageable spring accounts for bone elastic damage due to microcracks. The spring stiffness

1 − (D+ + D− )
 D := 0 (3)
D+ + D−

decreases from ∞ to 0 as the total damage D+ + D− increases from 0 to 1. As long as D+ + D− is equal to


zero, the stiffness of the damageable spring is infinite due to the lack of initial thickness of the cracks and is
thus equivalent to a rigid rod. Indeed, if the inelastic pad is sliding, initially carrying no stress, the equivalent
stiffness of the two springs mounted in series becomes (1 − (D+ + D− ))0 which is the well-known result for
a damaged elastic material. The total damage D+ + D− is therefore limited to values between 0 and 1 (zero
damage to full damage) and has a simple interpretation (e.g. [12]).
Moreover, two distinct damage threshold stresses beyond which damage occurs characterize the damage-
able spring (i.e. σ+D (D+ , D− ) > 0 for the tensile and −σ−D (D+ , D− ) < 0 for the compressive case, respectively).
We assume that these thresholds obey similar hardening rules than the ones defined in (1) and (2):
 
σ+D (D+ , D− ) = σ0+
D
1 + χ D (1 − exp (−k(D+ + D− ))) (4)
 
σ−D (D+ , D− ) = σ0−
D
1 + χ D (1 − exp (−k(D+ + D− ))) (5)

where σ0+ D
> 0, σ0− D
> 0 are the initial tensile, compressive damage threshold stresses, respectively, and
χ > 0 and k > 0 are damage hardening parameters.
D

In summary, the model is characterized by four variables: the total strain E, the inelastic strain E i and the
tensile and compressive damage variables D+ and D− , respectively. Their conjugate variables are, respec-
tively, the total stress S, the inelastic stress S i and the damage energies W D+ and W D− (conjugate in the sense
that their scalar or duality product S Ė, S i Ė i , W D+ D˙+ and W D− D˙− represents their power). Seven material
constants allow to adjust the desired response: 0 , χ i , l, σ0+D
, σ0−
D
, χ D and k.
546 D. Garcia et al.

2.2 Generalized standard materials model

The nonlinear and rate-independent damage behaviour of the model is entirely defined by two nonsmooth
convex functions: a free energy potential that represents the stored elastic energy density:

(E, E i , D+ , D− )
1 1 1−(D+ +D− ) i 2
2 0 (E − E ) + 2 0 D+ +D− E + I (D+ , D− ) if D+ + D− > 0
i 2
= (6)
2 0 (E) + I{0} (E ) if D+ + D− = 0
1 2 i

with  = ([0, 1[×[0, 1[), I{.} the indicator function of {.} (defined in Appendix A) and a dissipation potential
that implicitly contains the evolution rules for the internal variables:

( Ė i , Ḋ+ , Ḋ− ; E, E i , D+ , D− ) := i ( Ė i ; D+ , D− ) +  D ( Ḋ+ , Ḋ− ; E, E i , D+ , D− ) (7)


with
i ( Ė i ; D+ , D− ) := σ−i (D+ )DR+ ( Ė i ) + σ+i (D− )DR− ( Ė i ) (8)
where D{.} is the distance function to {.} (defined in Appendix A) and

 D ( Ḋ+ , Ḋ− ; E, E i , D+ , D− ) := φ( Ḋ+ , Ḋ− ; E, E i , D+ , D− ) + IR+ ×R+ ( Ḋ+ , Ḋ− ) (9)


where

h + (D+ , D− ) Ḋ+ if E i > 0 or E i = 0 and E ≥ 0
φ( Ḋ+ , Ḋ− ; E, E , D+ , D− ) :=
i
h − (D+ , D− ) Ḋ− if E i < 0 or E i = 0 and E < 0

σ±D2 (D+ , D− )
and h ± (D+ , D− ) := .
20 (1 − (D+ + D− ))2
We note that  is not affected by a permutation of D+ and D− . The distinct tensile and compressive behav-
iours of cortical bone tissue is included in the definition of an asymmetric dissipation potential. In contrast
with the variables Ė i , Ḋ+ and Ḋ− , the arguments appearing in  after the semicolon symbol (;) (Eq. 7) are
only parameters. The definition (9) implies that the function φ, and thus , is not continuous along E i = 0.
However, this discontinuity has no incidence on the generalized standard material formalism because E i is
only a parameter (e.g. [19]).
The state laws which derive from the free energy potential are

0 (E − E i ) if D+ + D− ∈ ]0, 1[
S ∈ ∂ E  = (10)
0 E if D+ + D− = 0
 i
0 E − 0 D+E+D− if D+ + D− ∈ ]0, 1[
S ∈ −∂ E i  =
i
(11)
R if D+ + D− = 0
 i2
D+
1
0 (D E+D )2 if D+ + D− ∈ ]0, 1[
W ∈ −∂ D+  = 2 + − (12)
[0, ∞[ if D+ + D− = 0
 i2
D−
1
0 (D E+D )2 if D+ + D− ∈ ]0, 1[
W ∈ −∂ D−  = 2 + − (13)
[0, ∞[ if D+ + D− = 0
where the scalar quantities appearing in the right-hand side of (10)–(13) are to be considered as singletons.
The same remark holds for further similar expressions.
The dual dissipation potential ∗ is obtained via the Legendre–Fenchel transform of . As the dissipa-
tion potential is separate in Ė i and { Ḋ+ , Ḋ− } and as these variables are independent (uncoupled, mounted in
parallel), the conjugate of the sum is simply the sum of the conjugates. Thus, we find

∗ (S i , W D+ , W D− ; E, E i , D+ , D− ) = i∗ (S i ; D+ , D− ) +  D∗ (W D+ , W D− ; E, E i , D+ , D− ) (14)
A 1D elastic plastic damage constitutive law for bone tissue 547

with

i∗ (S i ; D+ , D− ) := I[−σ i (D+ ),σ i (D− )] (S i ) (15)


− +

and

 D∗ (W D+ , W D− ; E, E i , D+ , D− )

I (W D+ ) if E i > 0 or E i = 0 and E ≥ 0
:= [−∞,h + (D+ ,D− )] D− (16)
I[−∞,h − (D+ ,D− )] (W ) if E i < 0 or E i = 0 and E < 0

The associated flow rules (inverse complementary laws) are




⎪ ∅ if S i < −σ−i (D+ )



⎨ ] − ∞, 0] if S = −σ− (D+ )
i i

Ė i ∈ ∂ Si ∗ = 0 if − σ−i (D+ ) < S i < σ+i (D− ) (17)





⎪ [0, +∞[ if S i = σ+i (D− )


∅ if S i > σ+i (D− )

where {∅} denotes the empty set and, if E i > 0 or E i = 0 and E ≥ 0, then Ḋ− = 0 and


⎨ 0 if W D+ ∈ ] − ∞, h + (D+ , D− )[
Ḋ+ ∈ ∂W D+ ∗ = [0, +∞[ if W D+ = h + (D+ , D− ) (18)

⎩ ∅ if W D+ > h + (D+ , D− ),

whereas if E i < 0 or E i = 0 and E < 0, then Ḋ+ = 0 and




⎨ 0 if W D− ∈ ] − ∞, h − (D+ , D− )[
Ḋ− ∈ ∂W D− ∗ = [0, +∞[ if W D− = h − (D+ , D− ) (19)

⎩ ∅ if W D− > h − (D+ , D− )

Let us note that the kinematic complementarity relation Ḋ+ Ḋ− = 0 ( Ḋ+ ≥ 0, Ḋ− ≥ 0) follows from the defi-
nition of the dissipation potential and its parametric dependence on E i (distinguishing tensile and compressive
states).

2.3 Numerical algorithm

In order to implement the model in a finite element mechanical analysis program, the elastic inelastic damage
evolution law needs to be discretized in time. Its algorithmic implementation is presented in detail in the fol-
lowing as it applies three non-trivial projections based on the relationship between the three internal variables
and their corresponding criteria.
First of all, let us define the inelastic yield and damage threshold functions which result from the dissipation
i ∈ ∂ , W D+ ∈ ∂ D−
potential via the complementary laws (i.e. S Ė i  Ḋ+  and W ∈ ∂ Ḋ− ):
 
 σ i (D− ) − σ−i (D+ )  σ+i (D− ) + σ−i (D+ )
Y i (S i ; D+ , D− ) :=  S i − + − (20)
2 2
and

Y D (W D+ , W D− ; E, E i , D+ , D− )

W D+ − h + (D+ , D− ) if E i > 0 or E i = 0 and E ≥ 0
:= (21)
W D− − h − (D+ , D− ) if E i < 0 or E i = 0 and E < 0
548 D. Garcia et al.

2.3.1 Algorithm formulation

Let us start with an initial strain state E 0 , E 0i , D+,0 , D−,0 for any given discretized time increment (D+,0 +
D−,0  = 0). To this state corresponds the unique initial stress state S0 , S0i , W0 + , W0 − defined by the state laws
D D

(10)–(13). This state is supposed to be inelastically and damageably admissible in the sense that it satisfies
both inelastic and damage criteria, i.e.

Y i (S0i ; D+,0 , D−,0 ) ≤ 0 and Y D (W0 + , W0 − ; E 0 , E 0i , D+,0 , D−,0 ) ≤ 0


D D
(22)

Given a new total strain E, consider the trial strain state E, E 0i , D+,0 , D−,0 (to which corresponds the
trial stress state ST = S(E, E 0i ), STi = S i (E, E 0i , D+,0 , D−,0 ), WT + = W D+ (E 0i , D+,0 , D−,0 ), WT − =
D D

W D− (E 0i , D+,0 , D−,0 )) which may violate the yield and threshold conditions (22).
If D+,0 + D−,0 = 0, then STi , WT + and WT − are undefined (Eqs. (11)–(13)). Thus, the yield and damage
D D

criteria cannot be tested. The indetermination can be lifted by defining a global equivalent criterion which
depends on the trial total stress

ST − σ0+
D
if ST ≥ 0
Y (ST ) := (23)
ST + σ0− if ST < 0
D

In the case of damage process, i.e. Y (ST ) > 0, the final values of S i , W D+ and W D− are well defined at the
projection point Y i (S i , D+ , D− ) = 0 and Y D (W D+ , W D−; E, E i , D+ , D− ) = 0 and normality of the flow rule
is guaranteed through the incremental process.
The purpose of the algorithm is to find the final strain state E, E i , D+ , D− (and corresponding final
stress state S, S i , W D+ , W D− ) which is admissible. The details of computation of the final states obtained after
projection on their respective criteria are given in Appendix B for each mode of the model.

2.3.2 Incremental linearization algorithm

For a uniform deformation test with homogeneous stress state, a situation occurring in a single finite element,
the problem is equivalent to the following local problem: find E such that the equation of force equilibrium is
satisfied, i.e.

S(E, E i , D+ , D− ) − S̄ = 0

where S̄ denotes the imposed stress. We can use the generalized Newton [1] method to solve this problem:

E j+1 = E j + E j with j = 0, 1, 2, . . . (24)

where
dS −1 j i  
E i = − (E , E , D+ , D− ) S(E j , E i , D+ , D− ) − S̄ (25)
dE
dS
where dE is the total stress derivative with respect to the strain (given in Appendix B for each deformation
mode of the model).

2.3.3 Algorithm

The elastic inelastic damage algorithm of the model is described in BOXES 1, 2, 3a, 3b and 4 of Table 1.
A range of mechanical tests were successfully simulated covering both strain- and stress-driven constant step
(relaxation and creep), monotonic ramp, saw tooth like cycles, cyclic sine both in tension and/or compression.
The Newton integration scheme used to overcome the nonlinearity of the stress-strain law under stress control
shows asymptotically quadratic convergence (in the absence of a damage state change).
The three modes of deformation (elastic (E), inelastic (I) and damage (D)) defined in Subsect. 2.1 and
explicited in the algorithm of Table 1 are highlightened in Fig. 2. It shows the response of the numerical model
A 1D elastic plastic damage constitutive law for bone tissue 549

(a) to five tensile strain-driven cycles of increasing amplitude and (b) to five stress-driven cycles of same
amplitude.

3 Identification and illustration

3.1 Uniaxial tests

We carried out a series of mechanical tests on bovine cortical bone specimens in order to identify the constants
characterizing our constitutive law. We will not describe the complete specimen preparation and experimental
procedure here, but the reader is referred to [8] for the details. In summary, let us mention that after hav-
ing isolated bovine femur diaphyses, we milled down the cortical bone along its longitudinal direction into
cylindrical traction specimens with dimensions 44 × 5 mm (Fig. 3). Uniaxial tests were carried out on an
electromechanical testing system (MicroTester, Instron, High Wycombe) at room temperature and under wet
conditions. Prior to each test, we carried out a series of preconditioning cycles in the elastic range in order to
measure the longitudinal initial elastic modulus. The tests included tensile or compressive saw tooth cycles
and saw tooth cycles mixing tension and compression, all being conducted along the longitudinal direction of
the diaphysis (Fig. 4).

Table 1 One-dimensional elastic inelastic damage algorithm (BOXES 1, 2, 3a, 3b and 4)


550 D. Garcia et al.

Table 1 continued

3.2 Identification and validation

All experiments delivered qualitatively the expected constitutive behaviour. In view of the large quantitative
scatter between each test, probably due to specimen density, porosity and mineralization scatter, identification
was done in the sense of a good qualitative agreement retaining the main features of the material behaviour.
Thus, only the most representative experimental stress–strain curves are shown in Fig. 5. They are expressed
in terms of strain H := (L t − L 0 )/L 0 and nominal stress P := p̄/A0 , where L 0 and L t are the initial and
actual gauge lengths, p̄ the load and A0 the initial specimen area.
We recall that the elastic mode of our one-dimensional model is described by the initial Young’s modulus
0 . Damage-dependent hardening functions account for the inelastic and damage evolution modes. If expo-
nential hardening is assumed as we did for both inelastic and damage criteria, the hardening functions are
characterized by six parameters: σ0+ D
, σ0−
D
, χ i , χ D , k and l. A summary of the identified material constants is
given in Table 2.
Although the geometry of the specimens ensured reliable measurements in tension, the small aspect ratio
of the specimens added to heterogeneities facilitated their buckling and rupture in compression. These insta-
bilities lead to near values for the tensile and compressive ultimate strains which is not usual for cortical bone
A 1D elastic plastic damage constitutive law for bone tissue 551

Table 1 continued

(a) (b)
(D) (D)
(E) (E)
Stress P [MPa]
Stress P [MPa]

(I) (I)

3
Strain H [-] Strain H [-]
Fig. 2 Response of the numerical model (a) to five tensile strain-driven cycles of increasing amplitude and (b) to five stress-driven
cycles of same amplitude. The elastic (E), inelastic (I) and damage (D) mode of deformation are represented by solid lines (green),
dotted lines (blue) and dashed lines (red), respectively. The model guarantees a damaged reloading which is collinear with the
origin (thin black solid lines), an important feature of cortical bone behaviour

(a) (b) R 2.5


φ 3mm
φ 5mm

10mm 20mm 12mm

44mm
Fig. 3 Bovine cortical bone traction specimen geometry. The specimens were extracted from bovine femur diaphyses (a) and
milled down into cylindrical specimens with dimensions shown above (b)

(e.g. [3]). Therefore, the tensile part of the stress-strain curves has been privileged in order to identify the
material constants of the models. Successful compressive experiments like the ones presented by [16] which
show a similar damage behaviour in compression and tension are needed in order to further validate or adjust
the material constants of our models.
The comparison of the experimental curves with the numerical model is shown in Fig. 6.
552 D. Garcia et al.

(a) (b)
Strain H [-] Strain H [-] 0.002
0.0025 0.002
0.0025
0.0025 t 0.002
t
-0.002
0.002
Fig. 4 Uniaxial strain schedules: (a) tensile cycles of increasing amplitude and (b) cycles mixing tension and compression (strain
rate: 10−2 s−1 ). We recall that H is the strain H := (L t − L 0 )/L 0 , where L 0 and L t are the initial and actual gauge lengths of
the extensometer used to record the deformation of the specimen

(a) Tensile cycles (b) Tensile and compressive cycles

Stress P [M P a]
Stress P [M Pa]

Strain H [-] Strain H [-]


Fig. 5 Selected stress–strain curves of bovine cortical bone in uniaxial testing: (a) tensile cycles of increasing amplitude showing
the intact reloading phase, a stiffness reduction (with damaged reloading collinear with the origin, dashed lines) and an increasing
energy dissipation and (b) cycles of increasing amplitude mixing tension and compression highlighting the stiffening occurring
when crossing from the tensile to the compressive part of the diagram due to the closing of cracks

Table 2 Overview of the identified material constants of the two rate-independent and rate-dependent one-dimensional damage
models for bovine cortical bone

Material constant 0 σ0+


D
σ0−
D
χi χD k l
Unit [G Pa] [M Pa] [M Pa] [M Pa] [−] [−] [−]
24.4 32 53 50 85.6 14 14
Stress P [M P a]

Predicted
Experimental

Strain H [-]
Fig. 6 Experimental versus predicted stress-strain curves of the model. The model successfully reproduces the experimental
curve in tension

4 Discussion

We sought to investigate the effects of cyclic overloads at physiological strain rates on the mechanical behaviour
of cortical bone tissue. Our own uniaxial cyclic mechanical tests carried out on bovine cortical bone confirm the
presence of three main modes of deformation of cortical bone: the elastic, inelastic and damage deformation
A 1D elastic plastic damage constitutive law for bone tissue 553

regimes are clearly identified by comparing Fig. 5a with Fig. 2a. The comparison between experimental and
simulation results is displayed in Fig. 6. It reveals the relevance, accuracy and validity of our model thus
showing that inelastic and damage mechanisms are closely coupled. Thus, the aforementioned (Subsect. 2.1)
interpretation of the three modes is reinforced: an instantaneous linear elastic regime due to compact bone
cohesion, a rate-dependent damage accumulation mode where microcracks are generated and a microcrack
gliding mode with growing friction.
The elastic mode suggests that the same longitudinal elastic modulus holds in tension and compression. The
very low values found for the initial tensile and compressive threshold stresses, associated with an exponential
hardening and a power law damage evolution rule implies that the elastic domain of cortical bone is extremely
small in the range of a few MPa.
The identified damage mode shows a disymmetry between the compressive and the tensile damage thresh-
old stresses. We find that in magnitude, the longitudinal compressive threshold stress is approximately 1.7
times higher than the tensile one, a result which agrees well with previous findings (e.g. [3]).
Our model is more general than the one proposed by [5] as it also includes the constitutive behaviour of
cortical bone in compression. Furthermore, it was designed using classical rheological elements which are char-
acterized by standard constitutive behaviours and do not need any restriction in order to be thermodynamically
admissible.
The experimental curve mixing tension and compression (Fig. 5b) reveals one of the limitations of our
model. Indeed, it does not predict the observed stiffening of the inelastic mode occurring when crossing from
the tensile to the compressive part of the stress–strain diagram (and the softening going from the compressive
to the tensile part, respectively). It is emphasized that it is not an experimental artefact as this behaviour may be
explained by the closing of cracks and has also been observed in concrete [15]. A model with different elastic
damage in tension and compression would be required to take into account this observed behaviour. More
precisely, instead of taking (1 − D+ − D− ) as the stiffness reduction factor, one could define a tensile damage
function f + (D+ , D− ) and a compressive one f − (D+ , D− ) such that (1− f + (D+ , D− )) and (1− f − (D+ , D− )
would denote the stiffness reduction in tension and compression, respectively.
The potential of the numerical model is shown in Figs. 2a and b. The proposed constitutive law success-
fully reproduces the main features of cortical bone’s behaviour: the intact elastic reloading phase, a stiffness
reduction and an increasing energy dissipation (accumulation of inelastic strains). Furthermore, it guarantees
a damaged reloading collinear with the origin which is an unique aspect of the model.
In this work, a one-dimensional thermodynamically consistent constitutive law for cortical bone is pre-
sented. Based on the results of this study, the model has been generalized in three dimensions in order to
simulate the realistic damage behaviour of cortical bone with the finite element method [9]. Note that we could
have directly presented the three-dimensional model, but the one-dimensional models appear to be very useful
to present the basic ideas, computations and experiments.

Acknowledgments This work was supported by the Swiss National Science Foundation, Grant no. 32-6194400. We thank
Romain Balet for carrying out some of the experimental tests and Peter Zioupos for providing us a tensile loading curve for
cortical bone.

Appendix A: Convex analysis functions

Let U be an arbitrary set and V a subset of U . We define the indicator function of V as:

IV : U → {0, +∞}

0 if x ∈ V
x → (A.1)
+∞ if x ∈/V

Let U be a convex subset of R. We define the distance function to U as

DU : R → R+


⎨ 0 if x ∈ U
x → x − sup U if x ≥ sup U (A.2)

⎩ | x − inf U | if x ≤ inf U
554 D. Garcia et al.

Appendix B: Internal variables evolution

In the following, the three projections based on the relationship between the three internal variables and
their corresponding criteria are presented. We assume that the initial total damage is not equal to zero, i.e.
D+,0 + D−,0  = 0.
(E) Elastic mode,
i.e. Y i (STi ; D+,0 , D−,0 ) ≤ 0 and Y D (WT + , WT − ; E, E 0i , D+,0 , D−,0 ) ≤ 0
D D

Final strain state: E i = E 0i , D+ = D+,0 , D− = D−,0


S = 0 (E − E ), dE
i dS = 
0
dS
where dE is the total stress derivative with respect to the strain.
(I) Pure inelastic mode,
i.e. Y i (STi ; D+,0 , D−,0 ) > 0 and Y D (WT + , WT − ; E, E 0i , D+,0 , D−,0 ) ≤ 0
D D

The radial return algorithm [14,18], or projection of the final inelastic stress on the convex elastic set (e.g.
[17] or [4]) leads in the tensile case ( Ė i > 0):
Final strain state:

σ+i (D− )
E = (D+ + D− ) E −
i
0 , D+ = D+,0 , D− = D−,0

S = 0 (E − E i ), dS
dE = (1 − (D+ + D− ))0
In the compressive case ( Ė i < 0), σ+i (D− ) has to be replaced by −σ−i (D+ ).
(D) Damage and inelastic mode
i.e. Y i (STi ; D+,0 , D−,0 ) > 0 and Y D (WT + , WT − ; E, E 0i , D+,0 , D−,0 ) > 0
D D

The radial return algorithm combined with the projection of the final damage energy on the convex undam-
aged set leads in the tensile case ( Ė i > 0):

Final strain state :



σ+i (D− )
E = (D+ + D− ) E −
i
, D+ | f (D+ ) = 0, D− = D−,0
0 (B.1)
dS
S = 0 (E − E i ), = Equation (B.9)
dE
with f : [0, 1[ → R defined by

σ i (D− )
f (D+ ) := 0 (1 − (D+ + D− )) E − + − σ+D (D+ , D− ) (B.2)
0

In the compressive case ( Ė i < 0), σ+i (D− ) has to be replaced by −σ−i (D+ ), D+ | f (D+ ) = 0, D− = D−,0
by D− | f (D− ) = 0, D+ = D+,0 and σ+D (D+ , D− ) by −σ−D (D+ , D− ), respectively.
We can use the generalized Newton method to solve the C 0 nonlinear equation f (D+ ) = 0:
j
j+1 j f (D+ )
D+ = D+ − j
with j = 0, 1, 2, . . . (B.3)
f
(D+ )

In the tensile case, the derivative of f reads

∂σ+D (D+ , D− )
f
(D+ ) = −0 E + σ+i (D− ) − (B.4)
∂ D+

The tangent operator required by the algorithm can be computed as follows. As Ḋ− = 0 in the tensile case,
we have
dS ∂S ∂ S dD+
= + (B.5)
dE ∂E ∂ D+ dE
A 1D elastic plastic damage constitutive law for bone tissue 555

Let us rewrite the function f defined in (B.2) as a function of the two variables E and D+ . As
∂f ∂f
f˙(E, D+ ) = Ė + Ḋ+ = 0 (B.6)
∂E ∂ D+
we get
dD+ ∂ f /∂ E
=− (B.7)
dE ∂ f /∂ D+
with ∂ f /∂ D+ given by (B.4).
Inserting the expression of the final inelastic strain E i given by (B.1) into S = 0 (E − E i ) and computing
∂ S/∂ E, ∂ S/∂ D+ , Equation (B.5) leads to
dS dD+
= (1 − (D+ + D− ))0 − (0 E − σ+i (D− )) (B.8)
dE dE
With (B.7), we finally get
dS   
= (1 − (D+ + D− ))0 1 − −0 E + σ+i (D− ) / f
(D+ ) (B.9)
dE
In the compressive case ( Ė i < 0), σ+i (D− ) has to be replaced by −σ−i (D+ ) and σ+D (D+ , D− ) by
−σ−D (D+ , D− ), respectively.

References

1. Alart, P., Curnier, A.: A mixed formulation for frictional contact problems prone to Newton like solution methods. Comput.
Methods Appl. Mech. Eng. 92, 353–375 (1991)
2. Ascenzi, A., Bonucci, E., Ripamonti, A., Roveri, N.: X-ray diffraction and electron microscope study of osteons during
calcification. Calcif. Tiss. Res. 25, 133–143 (1978)
3. Cowin, S.C.: Bone Mechanics. CRC Press, Boca Raton (1989)
4. Curnier, A.: Méthodes numériques en mécanique des solides. Presses Polytechniques et Universitaires Romandes (PPUR),
Lausanne (1993)
5. Fondrk, M.T., Bahniuk, E.H., Davy, D.T.: A damage model for nonlinear tensile behavior of cortical bone. J. Biomed.
Eng. 121, 533–541 (1999)
6. Fondrk, M.T., Bahniuk, E.H., Davy, D.T.: Inelastic strain accumulation in cortical bone during rapid transient tensile load-
ing. J. Biomech. Eng. 121, 616–621 (1999)
7. Fondrk, M.T., Bahniuk, E.H., Davy, D.T., Michaels, C.: Some viscoplastic characteristics of bovine and human cortical
bone. J. Biomech. 21(8), 623–630 (1988)
8. Garcia, D.: Elastic plastic damage constitutive laws for cortical bone. Ph.D. Thesis. http://library.epfl.ch/theses/?nr=3435
(2006)
9. Garcia, D., Zysset, P.K., Charlebois, M., Curnier, A.: A three-dimensional elastic plastic damage constitutive law for bone
tissue. Biomech. Model. Mechanobiol. 8(4), 149–165 (2009)
10. Keaveny, T.M., Wachtel, E.F., Kopperdahl, D.L.: Mechanical behavior of human trabecular bone after overloading. J. Orthop.
Res. 17, 346–353 (1999)
11. Kotha, S.P., Guzelsu, N.: Tensile damage and its effects on cortical bone. J. Biomech. 36(11), 1683–1689 (2003)
12. Lemaitre, J.: A Course on Damage Mechanics, 2nd edn. Springer, New York (1996)
13. McElhaney, J.H.: Dynamic response of bone and muscle tissue. J. Appl. Physiol. 21, 1231–1236 (1966)
14. Moreau, J.J.: Application of convex analysis to some problems of dry friction. In: Zorski, H. (ed.) Trends in Applications
of Pure Mathematics to Mechanics II, pp. 263–280. Pitman, London (1979)
15. Ortiz, M.: A constitutive theory for the inelastic behavior of concrete. Mech. Mater. 4(1), 67–93 (1985)
16. Reilly, D.T., Burstein, A.H.: The elastic and ultimate properties of compact bone tissue. J. Biomech. 8(6), 393–396 (1975)
17. Simo, J.C., Hughes, T.J.R.: Computational Inelasticity. Springer, Berlin (1999)
18. Wilkins, M.L.: Calculation of elastic-plastic flow. Methods Comput. Phys. 3, 211–263 (1964)
19. Ziegler, H.: An Introduction to Thermomechanics, 2nd edn. North-Holland Publishing Company, Amsterdam (1983)
20. Zysset, P.: A constitutive law for trabecular bone. Ph.D. Thesis, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne
(1994)

Das könnte Ihnen auch gefallen