You are on page 1of 23

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


Published online 9 August 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nag.649

A coupled damageplasticity model for concrete based on


thermodynamic principles: Part II: non-local regularization and
numerical implementation
Giang D. Nguyen1, , , and Guy T. Houlsby2,
1 Department

of Mathematics and Statistics, University of New Mexico, Albuquerque, NM 87131, U.S.A.


of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, U.K.

2 Department

SUMMARY
Non-local regularization is applied to a new coupled damageplasticity model (Int. J. Numer. Anal.
Meth. Geomech. 2007; DOI: 10.1002/nag.627), turning it into a non-local model. This procedure resolves
softening-related problems encountered in local constitutive models when dealing with softening materials.
The parameter identification of the new non-local coupled damageplasticity model is addressed, with
all parameters being shown to be obtainable from the experimental data on concrete. Because of the
appearance of non-local spatial integrals in the constitutive equations, a new implementation scheme is
developed for the integration of the non-local incremental constitutive equations in nonlinear finite element
analysis. The performance of the non-local model is assessed against a range of two-dimensional structural
tests on concrete, illustrating the stability of the stress update procedure and the lack of mesh dependency
of the model. Copyright q 2007 John Wiley & Sons, Ltd.
Received 15 January 2007; Revised 21 April 2007; Accepted 1 June 2007
KEY WORDS:

concrete; damage; plasticity; non-local; numerical implementation

1. INTRODUCTION
The resolving of softening-related problems plays a crucial role in the development of constitutive
models for strain softening materials in general, and for concrete in particular. In the constitutive modelling of concrete, localization due to softening is of great importance, because strain
softening and strength reduction are two of the most important features of the macroscopic material

Correspondence

to: Giang D. Nguyen, Department of Mathematics and Statistics, University of New Mexico,
Albuquerque, NM 87131, U.S.A.

E-mail: giang.nguyen@trinity.oxon.org, gdnguyen@math.unm.edu

Postdoctoral Fellow (formerly Research Student, University of Oxford).


Professor of Civil Engineering.

Copyright q

2007 John Wiley & Sons, Ltd.

392

G. D. NGUYEN AND G. T. HOULSBY

behaviour. The use of damage mechanics, in combination with plasticity theory, enables us to
derive appropriate constitutive models for the material [1]. However, it should be noted here that
in this study a macroscopic approach is used to model complicated underlying micro-fracturing
processes in concrete. Therefore, macroscopic variables such as damage indicator and plastic strain
are definitely not true representatives of those complicated fracturing processes. Nevertheless, to
some extent, they could be considered appropriate in capturing some essential features of the
material behaviour at macroscopic level, such as the stiffness and strength reductions and residual
strain, although the underlying microscopic processes are much more complicated. The damage and
plasticity zones in coupled damageplasticity models therefore could be viewed as macroscopic
representation of the underlying micro-cracking and frictional slips in a non-zero volume fracture
process zone (FPZ).
As the material exhibits significant post-peak softening, appropriate treatments, called regularization techniques, need to be applied to the constitutive modelling as well as the structural
analysis. In the literature, non-local regularization techniques have been found to be appropriate for
the modelling of softening materials [25], and help to avoid pathological problems encountered
in the constitutive modelling of these materials [6]. In this study, these techniques are applied to
a coupled damageplasticity models recently developed [1]. In particular, spatial integrals now
appear in the two damage criteria, with the physical interpretation of redistributing the energies
associated with the damage processes. The yield criterion remains in its local form. The calibration
of parameters for the non-local model is addressed, with the procedures proposed in Nguyen and
Houlsby [7] being used.
Numerical implementation plays an important part in the development of constitutive models for
engineering materials. The implementation here comprises a method for the solution of the partial
differential equations in solid mechanics, and the incorporation of the constitutive models into this
system of governing differential equations. In this study, the finite element method is employed for
solving boundary value problems in continuum mechanics. However, as concrete exhibits highly
nonlinear behaviour after peak stress, the incorporation of the non-local constitutive relationships
into the system of equations is not straightforward. A modified backward integration scheme, based
on the algorithm proposed by Crisfield [8 (Chapter 6, vol. 1)] is employed for the integration of
the non-local incremental constitutive equations. For the solution of systems of nonlinear algebraic
equations in finite element analysis, the arc-length method in combination with local constraint
equations employing dominant displacements [9] is implemented, and proves its reliability in this
study. Numerical examples showing the stability of the integration scheme, and the capability of
the arc-length control procedure to handle highly nonlinear behaviour in structural analysis, will
be presented.
In the next sections, a summary of the model in Nguyen and Houlsby [1] and the incorporation
of non-locality into this model are presented. This is followed by a summary of the procedures
for calibration of parameters for this non-local coupled damageplasticity model. The numerical
implementation of the non-local model is presented in Section 5, followed by numerical examples
simulating real structural problems.

2. THE LOCAL MODEL


The constitutive equations governing the local behaviour of a coupled damageplasticity model [1]
consist of a stressstrain relationship, a plasticity yield criterion yp along with two damage criteria
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

393

ydt and ydc for tensile and compressive damage, respectively. They all are written as follows:
i j =

(1 + )i j kk i j
+ i j
E(1 td )(1 cd )

yp = kk +
ydt =

ydc =

i j i j
2

(1)

k =0

(2)

(1 + pt )i+j i+j pt kk ll 


2E(1 td )2

F1t (td , cd ) = 0

(1 + pc )ij ij pc kk ll 


2E(1 cd )2

F1c (cd ) = 0

(3)

(4)

where   denotes the Macaulay bracket (x = 0, x<0; x = x, x0), i j the Kronecker delta, k
and  two parameters of the yield criterion, pt and pc two parameters of the damage criteria, and
F1t and F1c denote two functions governing the evolutions of damage variables td and cd [10].
The following evolution rules for plastic strain i j and damage variables td and cd are implicitly
embedded in the above constitutive equations:
 i j = p

*yp
= p (r i j + i j ) = p (r i j + i j )
*i j

(5)

 td = td

*ydt
= td
*td

(6)

 cd = cd

*ydc
= cd
*cd

(7)

where p , cd and td are corresponding non-negative scalar multipliers; td and cd two generalized
stresses associated with td and cd , respectively [1]; and
i j i j

k =0
(8)
2
is the yield function written in the generalized stress space i j [1, 11]. It is noted here that yp
(in Equation (8)) is only used to define a non-associated flow rule, through the use of the ratio r
governing the deviation of the plastic strain rate from the normal to the yield surface yp in true
stress space i j (see [10, 12] for details). For computational aspects, the yield surface yp (see
Equation (2)), obtained by substituting i j = i j into yp , is used [1]. The use of both yp and yp
here can be considered similar to the use of a yield criterion and a plastic potential, respectively,
for the definition of a non-associated flow rule in conventional plasticity.
In the damage criteria (3) and (4), the total stress tensor i j is decomposed into two parts
i+j and ij corresponding to tension and compression, respectively (see [13, 14]). The damageinduced softening processes are governed by two functions F1t and F1c , the forms and parameter
yp = [r kk + (1 r )kk ] +

Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

394

G. D. NGUYEN AND G. T. HOULSBY

identification of which have been detailed in Nguyen and Houlsby [1]. They are rewritten here as
follows:
2
2 
E + E pt (1 td )n t
t
c ft
F1 = (1 d )
(9)
2E E(1 td ) + E pt (1 td )n t
and
F1c =

f c2
2E

E + E pc (1 cd )n c [ln(1 + cd )]m c


E(1 cd ) + E pc (1 cd )n c [ln(1 + cd )]m c

2
(10)

Besides material properties such as Young modulus E, Poissons ratio , uniaxial tensile and
compressive strengths f t and f c , the following parameters are used in the above functions: E pt
and n t for tensile damage, and E pc , m c and n c for compressive damage. The physical meaning of
these parameters is explored in Part I of this paper [1], and is summarized in the next section.
The two parameters k and  in yield criterion (2) are defined as
k=

f cy f ty
3

(11)

=

f cy f ty
3

(12)

in which f ty and f cy are yield stresses in uniaxial tension and compression, defined as
f cy = ( f c0 + Hc cp )(1 cd )

(13)

f ty = ( f t0 + Ht tp )(1 td )(1 cd )

(14)

In the above expressions, the two initial yield thresholds are denoted as f c0 = 0.3 f c and f t0 = f t
for uniaxial compression and tension, respectively. Linear hardening rules are assumed here, with
Ht >0 and Hc >0 being the hardening moduli and tp and cp being corresponding accumulated
plastic strains in tension and compression [1]. We rewrite the definitions of tp and cp here:

(15)
tp = cF4t (kk )  i j  i j

cp = cF4c (kk )  i j  i j
(16)
where c =

2/3 and

 
kk + f cy
= min 1,
f cy + f ty

 
f ty kk
= 1 F4t
F4c = min 1,
f cy + f ty
F4t

(17)
(18)

The determination of all parameters of the proposed coupled damageplasticity model, based on
fracture properties of the material and widths of the fracture zones, has been presented in Part I
of this paper [1] in which further details on the model can be found.
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

395

3. NON-LOCAL REGULARIZATION
To deal with softening-related problems, the local constitutive model described in the preceding
section needs to be regularized. The regularization can be based on any of several methods ranging
from simple (e.g. fracture energy regularization by Bazant and Oh [15]) to more complicated
(non-local or gradient approaches). A detailed classification of regularization methods is given by
Jirasek and Patzak [16]. Here full regularization of the constitutive equations based on non-local
theories is used.
Following Pijaudier-Cabot and Bazant [2], it is necessary to apply non-local treatment only
to variables or quantities directly controlling the softening process. Various non-local damage
approaches using different non-local quantities has been analysed at length in Jirasek [17]. Recently,
Jirasek and Marfia [4] developed a non-local averaging scheme using displacement with some
advantages over other averaging schemes (e.g. using strain or stress averaging). In this study, with
damage-induced softening in the constitutive model, the two damage criteria (3) and (4) should
be treated as non-local criteria. Therefore, the regularization in this case is realized through the
non-locality of the two energy-like terms associated with the damage processes (see Equations (3)
and (4)). This is done by placing these energy-like terms in (3) and (4) under two non-local spatial
integrals. A physical interpretation is that the energy redistribution due to micro-crack interactions
[18, 19] is accounted for in the non-local model. The damage energies defined by the first terms
of Equations (3) and (4) and arising from all material points inside the volume Vd govern the
non-local softening processes. The thermomechanical aspects of this non-local regularization have
been discussed by Nguyen [10] and are also subjects of an on-going study [20]. From (3) and (4),
we obtain two non-local damage criteria as follows:
ydt =
ydc

1
G(x)

1
=
G(x)

g(y x)

(1 + pt )i+j i+j pt kk ll 


2E(1 td )2

Vd

g(y x)

dV F1t (td , cd ) = 0

(1 + pc )ij ij pc kk ll 

Vd

2E(1 cd )2

dV F1c (cd ) = 0

(19)

(20)

where all the stress terms in Equations (19) and (20) are evaluated at point y which
lies inside a

sphere (or circle in two dimensions) of volume Vd , centre x and radius R; G(x) = Vd g(y x) dV
is used to normalize the weighting scheme applied to the energy-like terms in (19) and (20); and
g(y x) is a bell-shaped weighting function defined by

0
if r >R



2
g(r ) = g(y x) =
(21)
r2

1
if
r
R
R2
The volume Vd here is defined by the non-local interaction radius R. For an interpretation of the
length parameter R, recent results [21] in bridging micro-scale and macro-scale could be used for
a relationship between the internal length R and the size of the representative volume element.
Further details on non-local formulations and the non-local damage criteria described above can
be found in Bazant and Jirasek [3] and Nguyen [10], respectively.
The set of non-local constitutive equations now comprises Equations (1)(2) and (19)(20). It
should be noted here that the non-locality of damage in (19) and (20) also helps to prevent the
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

396

G. D. NGUYEN AND G. T. HOULSBY

localization of both damage and plastic strain into zones of infinitesimal volumes, as encountered
in local models. Due to this non-locality, damage at a material point, once activated, will also
trigger damage at neighbouring material points, reducing the yield thresholds at those neighbouring
points (Equations (13) and (14)). Plasticity at the neighbouring damaged points can therefore takes
place at the same time as damage (e.g. in uniaxial tension), or can be activated once reduced yield
thresholds are met. With the focus here on the constitutive behaviour of the model at this stage of
the development, an analytical proof that the non-local model achieves these regularization features
is not given, but numerical examples (Section 6) demonstrate its effectiveness.

4. DETERMINATION OF MODEL PARAMETERS


As can be seen in Part I of this paper [1], the determination of parameters controlling the local
stressstrain behaviour of the model strongly depends on the imaginary widths wt and wc of the
FPZs in tension and compression, respectively. Unlike in smeared crack models, where wt and wc
can be directly determined from the finite element size [15, 22], these two widths in the non-local
model are related to the non-local interaction radius R, and also depend on other parameters
controlling the local stressstrain behaviour of the model. It is therefore difficult, even impossible,
to determine analytically the relationship between wt (or wc ) and the non-local interaction radius
R in non-local models, except in some simple cases with a linear softening law (e.g. see [2325]).
For that reason, a condition on the equivalence between the energy dissipated by a non-local
model with imaginary crack bandwidth wt (or wc in compression) and that specified by the fracture
energy G F (or G c ) is used in this study. Non-local numerical analyses of a simple bar under uniaxial
tension (or compression) will be carried out for the determination of a relationship between wt
(or wc ) and the length parameter R of the non-local model. Once this relationship is established,
a value of R to give a numerical crack band width wt close to that proposed in the literature
(e.g. wt 3dmax in [15, 26]), and satisfying the condition of fracture energy balance, is readily
determined.
The determination of the ratios wt /R and wc /R in non-local constitutive models has been
presented at length in Nguyen [10], and by Nguyen and Houlsby [7]. In this study, we use the
procedure proposed in Nguyen and Houlsby [7] to determine wt and wc from a local constitutive
model, experimental material properties, and a specified value of non-local radius R. The calibration
of non-local model parameters using the procedures in Nguyen and Houlsby [7] can only give
consistent responses of the model on the basis of fracture energy. More advanced (and more
computationally expensive) calibration methods (e.g. those based on numerical inverse analysis
and optimization by Carmelliet [27] and Le Bellego et al. [28]) for parameters of non-local
models can be used to obtain a supposedly unique model response corresponding to the material
characteristic length.
A summary of the models and parameters used in the non-local model described above can be
found in Nguyen [10]. Those parameters are not independent, but interact closely. They can be
classified as local parameters and a spatial parameter, which is the non-local radius R in this case.
The local parameters govern the pointwise behaviour of the model and are determined based on
calibration procedures in Nguyen and Houlsby [1, 7]. Following those procedures, the widths wt
and wc of the FPZs are obtained based on a specified value of non-local radius R (see [7]), and
then used to compute these local parameters (see [1]). In those procedures, the mutual effects of
all parameters of the model (spatial and local) on each other are accounted for. Different simplified
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

397

constitutive models can be derived from the full version by dropping out appropriate terms in
the constitutive equations and setting appropriate values for some model parameters, e.g. setting
Ht = Hc = , and f t0 = f c0 = to exclude plasticity from the full model (see [1] for details).
This feature has been incorporated into the finite element code OXFEM used in this study.

5. NUMERICAL IMPLEMENTATION
The finite element implementation of the non-local constitutive model described above is now
presented. A new integration scheme is proposed for the integration of the non-local incremental
constitutive equations derived from Sections 2 and 3. Local arc-length control based on the
algorithm by May and Duan [9] is used to handle snap-through and snap-back responses when
tracing the equilibrium paths for structures made of softening materials. Only the new integration
scheme for non-local incremental constitutive equations is presented here; the implementation of
the local arc-length control is given by Nguyen [10].
Integrating the incremental constitutive equations is a vital part of the numerical implementation
of the model, as it directly affects the stability of the numerical solutions. The integration scheme
must also be able to deal with non-locality in the constitutive equations. An implicit integration
method, based on the integration scheme proposed by several researchers [8 (Chapter 6, vol. 1); 29]
is therefore adopted and modified here.
The responses of every material point in the structure must satisfy entirely the system of constitutive equations (1)(2) and (19)(20), which in general can only be solved using numerical
methods. The analysis is, however, more complex than in models where the evolution laws of
damage are explicitly defined (e.g. in models by Peerlings et al. [30], Peerlings [31], Jirasek and
Patzak [32] and Jirasek et al. [33]). Because of the appearance of the spatial integral in Equations
(19) and (20), two spatial discretization schemes are necessary for solving the boundary value
problem. The first discretization is required for the numerical solution of the partial differential
equations, as is employed in conventional finite element analysis. The second is an inner discretization, which deals with the integration of the rate constitutive equations along a loading path.
Normally, for models based on local theory, only the outer discretization scheme is needed, as
the integration of the constitutive equations can be carried out pointwise. For the implementation
used here, the same discretization scheme is used for the solution of governing partial differential
equations and the integration of non-local constitutive equations. This results in the non-local
evaluation of energy-like quantities in the two damage criteria over the Gauss points used in the
finite element discretization.
For the numerical analysis using finite elements, the integrals in (19) and (20) are transformed
to summations over Gauss points. Denoting by the energy-like quantity to be averaged, we can
express its corresponding non-local counterpart as [10]:
n m e e
n

w g(ye x) det Jie (yie ) 


n i=1
m e i e i e

(x) = e=1
=
w

(22)
e

e=1
i=1 wi g(yi x) det Ji
in which e denotes the element index and n the total number of elements inside the interaction
volume Vd defined by a sphere at centre x and of radius R; i is the ith Gauss point of element
e; m e is the number of Gauss points of this element inside the interaction volume; wie and Jie
are, respectively, the weight and Jacobian matrix at Gauss point i of element e; n
is the total
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

398

G. D. NGUYEN AND G. T. HOULSBY

number of Gauss points inside the interaction volume at point x; w


and
are, respectively, the
weight and energy-like quantity associated with the ith Gauss point of element e, in which w
is
defined by
wie g(yie x) det Jie
m e e
e
e
e=1
i=1 wi g(yi x) det Ji

w
=  n

(23)

An implicit Euler integration scheme is adopted for the integration of the incremental constitutive
equations. However, due to the presence of the spatial integrals in the damage criteria, the stress
update procedure cannot be carried out pointwise as is normal in local models. Non-locality in
this case turns the pointwise-defined stressstrain constitutive equations to a system of non-local
coupled equations, relating the stresses, strains and internal variables at several integration points
in the failure region. This coupling makes the stress update routine more complicated, requiring
considerable effort in the formulation and implementation as well as a time cost in the numerical
computation.
As the constitutive relationships in this case contain coupled equations relating the stress and
strain increments at several integration points, finding the intersection points between the stress
path and the yield/damage surface is almost impossible. To avoid the need for this, an elastic
predictordamageplastic corrector integrating scheme is adopted here for the integration of the
equations. This is based on the algorithm proposed by Crisfield [8 (Chapter 6, vol. 1)] and can
be considered as a form of the backward-Euler integration scheme [8 (Chapter 6, vol. 1)]. This
algorithm uses the normal at the elastic trial point and hence avoids the necessity of computing the
intersection between the elastically assumed incremental stress vector (using secant elasticdamage
behaviour) and the yielddamage surfaces. Furthermore, the method enforces yield (2) and damage
criteria (19), (20) at any stage of the loading process, thus removing the inaccuracies encountered
when the incremental consistency conditions of the yield and damage functions are used [29].
At the starting point, the system of equations are rewritten in the incremental form by taking
the first-order Taylor expansion of the yield and damage functions about the elastic trial point
B (Figure 1, with y in the figure representing either the yield or compressive/tensile damage
surface). In this coupling case, it is assumed here that yielding and both mechanisms of damage
take place at the same instant. Therefore, treatment of corners (see Figure 6 of Nguyen and
Houlsby [1]) on the composite yielddamage surface is automatically taken into account. By
dropping appropriate terms, we can straightforwardly treat simpler situations, in which only one or
two failure mechanisms are activated. From (1), (2), (19), (20), a system of incremental equations
B (Elastic trial point)
y = yB > 0
C
X

y = yC > 0
y = 0 (elastic limit)

Figure 1. Pictorial presentation of the integration scheme (after de Borst [29] and
Crisfield [8 (Chapter 6, vol. 1)]).
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

399

governing the constitutive behaviour of the model can be written:


di j =

(1 + )di j  dkk i j
+ di j
E(1 td )(1 cd )
+

(1 + )i j kk i j t
(1 + )i j kk i j c
d
dd +
t
c
2
E(1 d ) (1 d )
E(1 td )(1 cd )2 d

yp = yp | B +

*yp
*i j

di j +


t
t
yd |C = yd | B + w

ydc | B

*td

dtd +

*yp

*yp

*d

*tp

c
c dd +

dtp +

E(1 td )2
+

(1+ pt )i+j i+j pt kk ll 


E(1 td )3

*yp

*cp

(1+ pt )i+j di j pt kk i j di j

ydc |C

*yp

dtd

(1 + pc )ij di j + pc kk i j di j

(24)
dcp

(25)

*F1t t *F1t c

*t dd *c dd = 0 (26)

d
d

*F1c c

c dd = 0
(1 + pc )ij ij pc kk ll  c *d
+
dd
E(1 cd )3
E(1 cd )2

(27)

The spatial integrals in Equations (26) and (27) obtained from two damage criteria have been
replaced by summations over integration points. The terms yp | B , ydt | B and ydc | B in (25)(27) are
the values of the loading functions at the elastic trial point B. Furthermore, it is implied here
that all the derivatives and terms in Equations (25)(27) are evaluated at this stress point. Due to
the appearance of the summations in (26) and (27), it should also be noted that B for these two
damage functions denotes several elastic trial points at different physical points, from which ydt
and ydc are evaluated, rather than a single point in the original scheme.
As mentioned above, the stresses at the elastic trial point B (Figure 1) are obtained by adding
elastic incremental stresses to the stresses at point X . Our aim is to compute the stress increment
i j | BC , which is needed in going from B to C (Figure 1), from system (24)(27). At first, from
(25), using the flow rule
*yp
di j = p
(28)
*i j
and the definitions of dtp and dcp in (15)(18), the plastic strain rate di j is obtained as

*y
*y
*y
p
p
p

yp | B +
dkl + t dtd + c dcd
*y

*kl
*d
*d
*yp
p

di j = p
= 




*

*i j
c *yp *yp F t ( ) *yp + F c ( ) *yp i j
4 kk
4 kk
*mn *mn
*tp
*cp

(29)

Secondly, back substituting the above plastic strain increment into (24), some manipulation
leads to the relationship between the stress, strain and internal variables, written in incremental
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

400

G. D. NGUYEN AND G. T. HOULSBY

form as follows:




*y
*y

*y
p
p
p

t +
c
| +

d
d
y

B
p
d
d
t
c

*
*
*

i
j
d
d

di j + 



*y
*y
*yp *yp
p
p
t
c
[Distjmn ]1
dmn =
F4 t + F4 c
c

*
*
*
*

kl
kl
p
p

t
c

[(1
+
)
[(1
+
)



]d



]d
ij
kk i j
ij
kk i j

d
d

t
c
t
c
2
2
E(1 d ) (1 d )
E(1 d )(1 d )

(30)

where Distjmn is the constitutive matrix, which is tangent with respect to plasticity and secant with
respect to damage (see [10] for details):
*yp *yp
*i j *mn
Di jmn
Distjmn =



t
c
(1 d )(1 d )
*yp
*yp
*yp *yp
t
c
F4 t + F4 c
c
*kl *kl
*p
*p

(31)

where Di jmn is the elastic compliance matrix.


Finally, substituting (30) into (26) and (27), we obtain a system of linear equations in dtd and
dcd . This system is written for all integration points in the FPZ, each pair (tension and compression)
of which has the following form:



*F1t t
*F1t c
t
t
yd =
w
Q t t dd c dd + yd  = 0
(32)

*d
*d




*F1c c
c
c
yd =
w
Q c c dd + yd  = 0

*d

(33)

where Q t and Q c are defined as follows:

*y
p
yp | B + t dtd + c dcd

*d
*d
*i j
(1 + pt )+

p


t
qq
kl
kl
st 1

[Di jkl ]





E(1 td )2

*y
*y
*yp *yp
p
p

F4t t + F4c c
c

*mn *mn
*p
*p

+
t
(1 + pt )kl pt qq kl st 1 [(1 + )i j qq i j ]dd

Qt =

[D
]
i jkl
t
t
c
2
2

E(1 d )
E(1 d ) (1 d )

+
c

(1 + pt )kl pt qq kl [D st ]1 [(1 + )i j qq i j ]dd

i
jkl
E(1 td )2
E(1 td )(1 cd )2

+ +
(1 + pt )i j i j pt kk ll  t

+
d
d
t
E(1 d )3

Copyright q

2007 John Wiley & Sons, Ltd.

*yp

*yp

(34)

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

and

*yp
+
+

*i j
(1 + pc )
+
p


c
pp
kl
kl
st 1

[Di jkl ]





E(1 cd )2

*y
*y
*y
*y
p
p
p
p

F4t t + F4c c
c

*mn *mn
*p
*p

t
[(1
+
)



]d

(1 + pc )
+
p


i
j
pp
i
j
c
pp
kl
kl
d
st 1
Qc =

[D
]
i jkl
c
t
c
2
2

E(1 d )
E(1 d ) (1 d )

[(1
+
)



]d
ij
pp i j

(1 + pc )kl + pc  pp kl [D st ]1
d

i
jkl
c
t
c
E(1 d )2
E(1 d )(1 d )2

(1 + pc )ij ij pc kk ll  c

+
dd
c
3
E(1 d )

yp | B

*yp

dt
*td d

*yp

401

dc
*cd d

(35)

As can be seen, the strain increments do not appear in (34) and (35) when substituting (30)
into Equations (26) and (27), as they have been applied in moving from point X to the elastic
trial point B [8 (Chapter 6, vol. 1)]. Solving the above system of equations (32)(33) will give
the damage increments at all points in the damaged zone of the structure. Back substituting the
damage increments into (30), noting that the strain increments have been applied in the predictor
step (from X to B) and now must be removed from that expression, we obtain the stress increment
i j | BC in going from B to C. Finally, the stress at point C (see Figure 1) is updated using
[8 (Chapter 6, vol. 1)]:
i j |C = i j | B + i j | BC
in which i j | BC is computed by



*yp t
*yp c *yp

yp | B +

d + c d

*
*
*

ij
d
d





*y
*y
*yp *yp
p
p
t
c
[Distjmn ]1
mn | BC =
F4 t + F4 c
c

*
*
*
*

kl
kl
p
p

t
c

[(1
+
)
[(1
+
)



]



]
i
j
kk
i
j
i
j
kk
i
j

d
d

E(1 td )2 (1 cd )


E(1 td )(1 cd )2

(36)

(37)

In the above expression, all damage increments have been computed by solving systems (32)(33),
and all derivatives are evaluated at the elastic trial point B. Normally, due to the linearization,
the updated stress points do not lie on the yield/damage surfaces (Figure 1) and relevant techniques should be applied to return them to the loading surfaces. Repetition of the same process
but with point B replaced with C, noting that there are no elastic stress increments in the subsequent steps, would be an appropriate and simple way to bring the stresses back to the loading
surfaces. Mathematically, this is an iterative NewtonRaphson process to return the stress at B
to loading surfaces. This simple but rather efficient stress return algorithm is advocated here due
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

402

G. D. NGUYEN AND G. T. HOULSBY

to the complexity of the system of incremental constitutive equations (24)(27). Alternatively,


further enhancement to the exactness of the integration scheme can be obtained through the
combination of sub-incrementation and the repetition of the above predictorcorrector processes,
all of which have been implemented in the finite element code OXFEM. Furthermore, to reduce
the errors in updating the stresses, in this study the integration scheme above was carried out based
on the incremental instead of iterative strains (see [8 (Chapter 6, vol. 1)] for details).

6. NUMERICAL EXAMPLES
This section examines the numerical validation of the non-local coupled damageplasticity model.
Analyses of example structures are carried out to show the features of the model. These tests
range from simple to complex loading cases, and hence require appropriate choice of constitutive
models.
The structural tests used here comprise those exhibiting important features in the behaviour of
quasi-brittle materials in general and concrete materials in particular. They can be classified as
tension tests, bending tests, shear tests and compression-related tests, under monotonic or cyclic
loading. The material models used can be pure tensile damage or tensilecompressive damage
with plastic deformation being accounted for whenever the experimental cyclic loading data are
available. The calibration procedures for parameters of non-local models, proposed in Nguyen and
Houlsby [1, 7] and summarized in Section 4, are used throughout the numerical examples.
All the numerical analyses are carried out using the local arc-length control [9] for the incremental
analysis, and NewtonRaphson method for the iterative technique. Finite element meshes of sixnode triangular elements are used in all examples. The convergence tolerance parameter is 104
for the norm of the out-of-balance force vector in the NewtonRaphson iterative process. The same
tolerance is used in the stress update routine to gauge the errors occurring in returning the stresses to
the loading surfaces. Automatically chosen numbers of increments (see [8 (Chapter 9, vol. 1); 9]),
controlled by the number of iterations required for each load increment, are used throughout the
examples. Due to the complexity of the non-local formulation, a local stiffness matrix based on the
local constitutive matrix in Equation (31) (see [10] for details) is used in all numerical examples.
A loading scheme consisting of three load stages (1st: fully elastic behaviour, 2nd and 3rd: peak and
post-peak stages) is used, in which the controlling minimum and maximum numbers of iterations
for the last two stages are normally 1218 and 1827, respectively.
To achieve convergence in severe cases with snap-back observed in the equilibrium paths, the
constraint equation of the arc-length control is based on relative nodal displacements in the FPZ,
although crack mouth opening displacements (CMODs) or crack mouth sliding displacements
(CMSDs) can also be used without any difficulty (see [9, 10, 29] for details). The local stiffness
matrix, which is tangent with respect to plasticity and secant with respect to damage (Equation (31);
see also [10]), is used throughout the numerical examples in this section. When the dissipation is
totally due to damage, the local secant stiffness matrix is used. For the NewtonRaphson iterative
method, the stiffness matrix is only recomputed at the beginning of the load increment, and kept
unchanged throughout the iterations [10]. Furthermore, the stress update is based on the incremental
instead of iterative strains [8].
Table I summarizes model parameters used in numerical examples in this section. Details on
how to obtain these model parameters can be found in Nguyen [10].
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

403

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

Table I. Model parameters used in numerical examples.


Examples and models

Parameters

6.1

6.2

6.3

6.4

6.5

Pure damage

Pure damage

Tensile damage
and plasticity

Pure damage

Tensile and
compressive damage

2.4 104
0.2
0.059

3.0 104
0.2
0.124

3.8 104
0.2
0.125

2.48 104
0.18
0.1085

2.0 (or 2.86)

3.33

3.0

3.55

12.0
2.02 (1.98)
3090.0 (9081.0)
0.26 (0.36)

16.0
1.96
6899.0
0.32

11.0
6.0
2.32
6381.0
0.33
33 800.0

25.0
1.86
14 129.0
0.37

E (N/mm2 )

G F (N/mm)
G c (N/mm)
f t (N/mm2 )
f c (N/mm2 )
f c0 (N/mm2 )
R (mm)
wt /R
E pt (N/mm2 )
nt
Ht (N/mm2 )
wc /R
E pc (N/mm2 )
mc
nc

3.1 104
0.2
0.072
18.0
4.08
38.0
2.0
2.0
1988.0
0.28
2.2
6000.0
6.0
0.24

60

10

60

60mm
(a)

(b)

(c)

Figure 2. (a) Double edge notched specimen (10-mm thick)geometry; (b) experimental
crack pattern [34]; and (c) FE meshes.

6.1. Double-edge notched specimen under tension


In this example, simulations of a double edge notched specimen under tension [34] are presented
(Figure 2(a)). In the analyses, the specimen is fixed in both directions at the bottom edge, and in
horizontal direction at the top edge. The analyses were carried out using three meshes of six-node
triangular finite elements (Figure 2(c)), with prescribed vertical displacements on the top edge of
the specimen.
The numerical results are depicted in Figure 3, showing the agreement in the loaddisplacement
curves obtained from different finite element meshes, thus proving the lack of mesh dependence
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

404

G. D. NGUYEN AND G. T. HOULSBY

1.3
1.1

1.2

0.9

Load (kN)

0.7

0.8

0.5
0

0.6

0.005

0.01

0.015

0.02

0.025

Experimental
Numerical, mesh 1, ft=2.0MPa
Numerical, mesh 2, ft=2.0MPa
Numerical, mesh 3, ft=2.0MPa
Numerical, mesh 1, ft=2.86MPa

0.4
0.2
0
0

0.05
0.1
0.15
Prescribed displacement (mm)

(a)

0.2

(b)

Figure 3. Double edge notched specimennumerical results: (a) loaddisplacement curves and (b) damage
pattern (mesh 3, prescribed displacement of 0.19 mm).

P/2

P
Crack path

200
100
50

2000

L /2

Figure 4. Geometrical data and half-beam model used in the numerical analysis.

of the model. The experimental peak load (1.13 kN) in the figure can be seen to be bounded by
its two numerical counterparts (0.94 and 1.26 kN) corresponding to two used values of the tensile
strength. In addition, the overall shape of the numerical loaddisplacement curves is consistent
with the experiment. The failure of the specimen can be seen in Figure 3(b). No clear macro-crack
can be observed, as attention here is paid to the structural response of the specimen under loading,
rather than the crack propagation and interaction. A finer mesh and smaller non-local interaction
radius could be used, at higher computational cost, if the observation of crack propagation is
important (see [10]).
6.2. Three-point bending testnotched beam
Analyses are carried out for a notched beam in a three-point bending test (Figure 4), aiming at
investigating Mode I fracture and crack propagation. The geometrical data and material properties
are taken from the experimental test of Petersson [35], with the effect of the beams self-weight
being eliminated from the measured fracture energy (see [35]).
Because of symmetry, only half of the beam was modelled (Figure 5). Results, in the form
of loaddeflection curve and damage pattern, are shown in Figure 6. The damage process zone
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

405

Figure 5. Finite element meshes: (a) full beam and (b) zoomed-in at centre.

800
700
Load (N)

600
500
400
300

Exp., GF=115N/m
Exp., GF=137N/m
Num., mesh a, GF=124N/m
Num., mesh b, GF=124N/m

200
100
0

0.2

0.4
0.6
Deflection (mm)

0.8

Figure 6. Loaddeflection curve and damage pattern at very late stage of the numerical analysis (finer mesh,
zoomed-in at centre part of the half-beam).

can be clearly seen in this figure and the numerical crack path agrees well with the experimental
one in Figure 4. The numerical loaddeflection curves obtained from the two meshes are almost
identical, again demonstrating the lack of mesh dependency of the proposed model. In addition,
they also match the experimental curves quite well.
6.3. Four-point bending testnotched beamcyclic loading
In this example, the four-point bending test experimentally performed by Hordijk [36] is simulated
using the coupled damageplasticity model. The geometry of the specimen and finite element
meshes of a half-beam model are shown in Figure 7.
The numerical loaddeflection curve overestimates the experimental one in the post-peak region
near peak load (Figure 8(a)), while it underestimates the tail behaviour (Figure 8(b)). However, the
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

406

G. D. NGUYEN AND G. T. HOULSBY

150

150

150

P
Crack path
100
30
50

500mm

Figure 7. Four-point bending testgeometry and finite element meshes.

numerical peak load is rather close to the value given by experiment (Figure 8). The permanent
deformations at zero stress state are clearly seen in Figure 8, showing the capability of the model
in capturing residual strains in tension. On the other hand, it can also be seen that there is almost
no difference in the numerical results using different FE meshes, proving the mesh independence
of the obtained numerical results.
In Figure 8(c), we can see the zones of damage and plastic strain along with their increments
at a specific moment of the failure process. Although plasticity is local in the proposed model,
its activation and development in the structure is however induced by the non-local damage
mechanisms. In other words, the reduction of the yield thresholds due to the non-locality of damage
will also result in the activation of plasticity at neighbouring points of the material point under
consideration. Therefore, the accumulated plastic strain defined in (15) also spreads out, realizing
through the sizes of the zones of accumulated plastic strain and its increment in Figure 8(c).
6.4. Four-point shear test
The four-point shear test of Arrea and Ingraffea [37] is selected here to demonstrate the capability
of the model in capturing the structural responses in shear loading. The snap back of the load
displacement curve helps to demonstrate the stability of the proposed numerical integration scheme
and solution methods in dealing with highly nonlinear structural response. The geometrical data
for structure are shown in Figure 9.
To avoid local failure, the load distributors were also modelled in the two finite element meshes,
and assumed to be made of steel (E = 210 GPa,  = 0.3) (Figure 10). The analyses were carried
out with the applied load under indirect control, based on the relative displacements between nodal
points of elements in the fracture zone (local control). The arc-length method was used for the
incremental analysis and the NewtonRaphson method for the iterative equilibrium. This helps to
capture effectively the snap back of the loaddisplacement curve.
Snap-back behaviour can be clearly seen on the loaddisplacement curves in Figure 11(b), with
the vertical displacement taken at point A on the upper side of the beam and under the steel load
distributor (Figure 9). The difference in the load-displacement curves from the analyses using two
meshes comes mainly from the difference in size of the modelled load distributor. This is only
a local effect, and the overall structural responses in Figure 11(a) are almost identical and not
significantly affected by the size of the load distributors.
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

407

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

Numerical, mesh 1
Numerical, mesh 2
Experimental

3
2.5

2
Load (kN)

Load (kN)

2
1.5
1
0.5

1.5

1
0.5

0
0

(a)

Numerical, mesh 1
Numerical, mesh 2
Experimental

2.5

0.05
0.1
0.15
0.2
Mid-span deflection (mm)

0.25

(b)

0.1
0.2
0.3
0.4
Mid-span deflection (mm)

0.5

(c)

Figure 8. Four-point bending testloaddeflection curves: (a) peak and (b) tail behaviour. (c) Contour
lines of internal variables and their increments (mesh 2), at point X (b) on the loaddeflection curve.

There is however a poor match between the experimental and numerical loadCMSD curves;
and this has also been found in other research dealing with this mixed-mode test [38, 39]. It can
be supposed that the observed mismatch comes from the irrelevant use of pure Mode I fracture
energy in a mixed-mode analysis. However, lack of relevant material properties from the real test
means that this mismatch cannot be fully explained.
The numerical crack pattern can be seen in Figures 12 and 13, and compared with the experimentally observed crack path in Figure 9. The numerical crack path seems to be less curved than
its experimental counterpart.
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

408

G. D. NGUYEN AND G. T. HOULSBY

P
A

0.13P

Observed crack path

306

306
82

156
203

61 61

396

396

203

Figure 9. Four-point shear testgeometrical data (dimensions in mm).

Figure 10. Finite element meshes.

Mesh 1
Mesh 2
Experimental, lower
Experimental, upper

160
140

100
80
Load P (kN)

Load P (kN)

120
100
80
60
40

Mesh 1
Mesh 2

60
40
20

20
0

0
0

(a)

0.05
0.1
CMSD (mm)

0.15

(b)

0.05
0.1
0.15
Displacement at load point (mm)

0.2

Figure 11. (a) LoadCMSD responses and (b) loaddisplacement curves.


Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

409

Figure 12. Deformed structure at CMSD of 0.26 mm (mesh 2, magnification of 100).

Figure 13. Four-point shear test: crack pattern at CMSD of 0.26 mm (mesh 2).

6.5. Splitting test on a concrete prism


The splitting test (Brazilian test) here serves as an indirect testing method to measure the tensile
strength of the material, helping to resolve disadvantages in the implementation of direct tensile
test [40]. The splitting strength can then be used to calculate the uniaxial tensile strength of the
material [41, 42]. The testing arrangement was quite simple, as shown in Figure 14, and test on the
prism specimen was adopted for the numerical simulation, with D = 75 mm, B = 50 mm and the
following widths of the load bearing strip: b1 = 0.08D = 6 mm (numerical test 1; named STP75-8
in [40]) and b2 = 0.16D = 12 mm (numerical test 2; STP75-16 in [40]).
In both tests, the results obtained from the two finite element meshes are almost identical, proving
the mesh independence of the proposed non-local model. However, the peak loads obtained in
neither of the numerical tests agree well with the experimental data. In addition, only the numerical
loaddisplacement curve for the first test (b/D = 0.08) shows a close resemblance to the numerical
reference curve [43]. Analyses have also been carried out using a lower value of the compressive
fracture energy (dashed dot curve, Figure 15(a)). However, only the tail response is affected and
the peak load only changes insignificantly. For the second test (b/D = 0.16), oscillation in the
loaddisplacement response can be observed, which has also been found in other numerical results
for this kind of splitting test [43, 44].
The paper by Rocco et al. [40] does not provide any results on the loaddisplacement curves
or the crack pattern of the specimen. In this study, the damage contours obtained from numerical
analysis are presented to illustrate the failure processes in the specimen at different loading
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

410

G. D. NGUYEN AND G. T. HOULSBY

b
D

Figure 14. Splitting test: geometry and finite element model.

30
20

25

15

10
Num., Comi & Perego (2001)
Num., Mesh 1, Gc=18N/mm
Num., Mesh 2, Gc=18N/mm
Num., Mesh 2, Gc=10N/mm
Exp. peak load, b/D=0.08

Load (kN)

Load (kN)

(a)

0.02

0.04

10
Num., Mesh 1
Num., Mesh 2
Num., Comi & Perego( 2001)
Exp. peak load, b/D=0.16

0
0

0.06

Prescribed displacement (mm)

D
15

0
0

20

(b)

0.02

0.04

F
0.06

Prescribed displacement (mm)

Figure 15. Loaddisplacement curves: (a) b/D = 0.08 and (b) b/D = 0.16.

At A (peak)

At B

Figure 16. Tensile damage (left) and compressive damage (right) for b/D = 0.08.

stages. This is useful as it helps to show the complex failure process involving both tensile and
compressive damage mechanisms, which are coupled in this example. The damage zones in both
tests are depicted in Figures 16 and 17, where the splitting effect can be clearly seen in Figure 17
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

411

At E

At C (just after peak)

At D

At F

Figure 17. Tensile damage (left) and compressive damage (right) for b/D = 0.16.

for b/D = 0.16. Tensile damage in this case (b/D = 0.16) occurs at the centre of the specimen
and quickly develops through its height, while failure due to compressive damage just happens at
the corner of the load bearing strip. This is different from the first test (b/D = 0.08) where failure
due to both mechanisms of damage localizes underneath the load bearing strip (Figure 16).

7. CONCLUSIONS
Enhancements using non-local theory to a newly developed constitutive model [1] to cope with
softening-related problems are presented in this paper. The regularized damageplasticity model
employs two independent length parameters in tension and compression. All parameters of this
model are shown to be identifiable and can be calibrated based on the experimental data on
concrete, using the procedures developed by Nguyen and Houlsby [1, 7]. This is an important
feature which makes the model readily applicable in practice.
An integration scheme, based on that by de Borst [29] and Crisfield [8 (Chapter 6, vol. 1)],
was proposed for updating stress during the incremental-iterative solution procedure in nonlinear
finite element analysis. This implicit integration scheme for non-local rate constitutive equations
was shown to be stable through the simulation of real structural tests in Section 6. This stability
is enabled by the use of a local arc-length control [9] in the incremental analysis. The principal
structural responses were captured in numerical examples using the proposed non-local approach,
and the lack of mesh dependency of the non-local constitutive model was established.
The formulation of a non-local coupled damageplasticity in this study (Part I: Nguyen and
Houlsby [1] and Part II: this paper) is considered as an initial step in the development of a
thermomechanical approach for the non-local constitutive modelling of concrete. The computational
aspects of the proposed non-local model were only briefly considered in this study, with the
proposal of an integration scheme for the stress update routine. In addition, use of a local stiffness
matrix, instead of a non-local stiffness matrix consistent with the stress update algorithm, in the
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

412

G. D. NGUYEN AND G. T. HOULSBY

numerical analysis significantly increased the computational costs. Further research will focus on
the thermodynamic, regularization and computational aspects of non-local constitutive modelling,
and the incorporation of anisotropic behaviour to appropriately predict failure mode and orientation
of failure planes, all of which have been discussed by Nguyen [10].
ACKNOWLEDGEMENT

Financial support from the Jenks family (in the U.S.A.) through the Peter Jenks Vietnam scholarship to
the first author, while he was working in Oxford, is gratefully acknowledged.
REFERENCES
1. Nguyen GD, Houlsby GT. A coupled damageplasticity model for concrete based on thermodynamic principles:
Part I: model formulation and parameter identification. International Journal for Numerical and Analytical
Methods in Geomechanics 2007; DOI: 10.1002/nag.627.
2. Pijaudier-Cabot G, Bazant ZP. Non-local damage theory. Journal of Engineering Mechanics (ASCE) 1987;
113(10):15121533.
3. Bazant ZP, Jirasek M. Non-local integral formulation of plasticity and damage: survey of progress. Journal of
Engineering Mechanics (ASCE) 2002; 128(11):11191149.
4. Jirasek M, Marfia S. Non-local damage model based on displacement averaging. International Journal for
Numerical Methods in Engineering 2005; 63(1):77102.
5. Grassl P, Jirasek M. Plastic model with non-local damage applied to concrete. International Journal for Numerical
and Analytical Methods in Geomechanics 2006; 30:7190.
6. Jirasek M, Bazant ZP. Inelastic Analysis of Structures. Wiley: New York, 2002.
7. Nguyen GD, Houlsby GT. Nonlocal damage modelling of concrete: a procedure for the determination of
parameters. International Journal for Numerical and Analytical Methods in Geomechanics 2007; 31:867891.
8. Crisfield MA. Non-linear Finite Element Analysis of Solids and Structures, vols. 1 and 2. Wiley: New York,
1997.
9. May IM, Duan Y. A local arc-length procedure for strain softening. Computers and Structures 1997; 64(14):
297303.
10. Nguyen GD. A thermodynamic approach to constitutive modelling of concrete using damage mechanics and
plasticity theory. D.Phil. Dissertation, Department of Engineering Science, University of Oxford, August 2005.
11. Houlsby GT, Puzrin AM. A thermomechanical framework for constitutive models for rate-independent dissipative
materials. International Journal of Plasticity 2000; 16:10171047.
12. Collins IF, Houlsby GT. Application of thermomechanical principles to the modelling of geotechnical materials.
Proceedings of the Royal Society of London, Series A 1997; 453:19752001.
13. Ladeveze P. Sur une theorie de lendommagement anisotrope. Rapport interne No. 34, LMT Cachan, France,
1983.
14. Ortiz M. A constitutive theory for the inelastic behaviour of concrete. Mechanics of Materials 1985; 4:6793.
15. Bazant ZP, Oh BH. Crack band theory for fracture of concrete. Materials and Structures (RILEM, Paris) 1983;
16:155177.
16. Jirasek M, Patzak B. Models for quasibrittle failure: theoretical and computational aspects. ECCM-2001 European
Conference on Computational Mechanics, Cracow, Poland, 2629 June 2001.
17. Jirasek M. Nonlocal models for damage and fracture: comparison of approaches. International Journal of Solids
and Structures 1998; 35(3132):41334145.
18. Bazant ZP. Why continuum damage is non-local: micromechanics arguments. Journal of Engineering Mechanics
(ASCE) 1991; 117(5):10701087.
19. Bazant ZP. Non-local damage theory based on micromechanics of crack interactions. Journal of Engineering
Mechanics (ASCE) 1994; 120(3):593617.
20. Nguyen GD. A thermodynamic approach to non-local damage modelling of concrete. International Journal of
Solids and Structures 2007, submitted.
21. Gitman IM, Askes H, Aifantis EC. The representative volume size in static and dynamic micromacro transitions.
International Journal of Fracture 2005; 135:L3L9.
Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

A COUPLED DAMAGEPLASTICITY MODEL FOR CONCRETE: PART II

413

22. Rots JG, Nauta P, Kusters GMA, Blaauwendraad J. Smeared crack approach and fracture localization in concrete.
Heron 30(1), Stevin Laboratory, Delft University of Technology, Delft, The Netherlands, 1985.
23. De Borst R, Muhlhaus H-B. Gradient-dependent plasticity: formulation and algorithmic aspects. International
Journal for Numerical Methods in Engineering 1992; 35:521539.
24. Meftah F, Reynouard JM. A multilayered mixed beam element in gradient plasticity for the analysis of localized
failure modes. Mechanics of Cohesive-frictional Materials 1998; 3:305322.
25. Zhao J, Sheng D, Zhou W. Shear banding analysis of geomaterials by strain gradient enhanced damage model.
International Journal of Solids and Structures 2005; 42(20):53355355.
26. Bazant ZP, Pijaudier-Cabot G. Measurement of characteristic length of nonlocal continuum. Journal of Engineering
Mechanics (ASCE) 1989; 115(4):755767.
27. Carmeliet J. Optimal estimation of gradient damage parameters from localization phenomena in quasi-brittle
materials. Mechanics of Cohesive-frictional Materials 1999; 4:116.
28. Le Bellego C, Dube JF, Pijaudier-Cabot G, Gerard B. Calibration of non-local damage model from size effect
tests. European Journal of Mechanics A/Solids 2003; 22:3346.
29. De Borst R. Nonlinear analysis of frictional materials. Ph.D. Dissertation, Delft University of Technology, Delft,
The Netherlands, 1986.
30. Peerlings RHJ, de Borst R, Brekelmans WAM, Geers MGD. Gradient enhanced damage modelling of concrete
fracture. Mechanics of Cohesive-frictional Materials 1998; 3:323342.
31. Peerlings RHJ. Enhanced damage modelling for fracture and fatigue. Ph.D. Dissertation, Eindhoven University
of Technology, Eindhoven, The Netherlands, 1999.
32. Jirasek M, Patzak B. Consistent tangent stiffness for non-local damage models. Computers and Structures 2002;
80:12791293.
33. Jirasek M, Rolshoven S, Grassl P. Size effect on fracture energy induced by non-locality. International Journal
for Numerical and Analytical Methods in Geomechanics 2004; 28:653670.
34. Shi C, van Dam AG, van Mier JGM, Sluys LJ. Crack interaction in concrete. In Materials for Buildings and
Structures EUROMAT, Wittmann FH (ed.), vol. 6. Wiley-VCH: Weinheim, Germany, 2000; 125131.
35. Petersson PE. Crack growth and development of fracture zones in plain concrete and similar materials. Report
TVBM-1006, Division of Building Materials, Lund Institute of Technology, Lund, Sweden, 1981.
36. Hordijk DA. Local approach to fatigue of concrete. Ph.D. Dissertation, Delft University of Technology, Delft,
The Netherlands, 1991.
37. Arrea M, Ingraffea A. Mixed mode crack propagation in mortar and concrete. Report 81-13, Department of
Structural Engineering, Cornell University, Ithaca, NY, U.S.A., 1982.
38. Alfaiate J, Pires EB, Martins JAC. A finite element model for the study of crack propagation in concrete.
Proceedings of 2nd International Conference on Computer Aided Assessment and Control (Localized Damage
92), vol. 1: Fatigue and Fracture Mechanics, Southampton, U.K., 1992; 261280.
39. Jefferson AD. Crafta plastic-damage-contact model for concrete. II. Model implementation with implicit
return-mapping algorithm and consistent tangent matrix. International Journal of Solids and Structures 2003; 40:
60016022.
40. Rocco C, Guinea GV, Planas J, Elices M. The effect of the boundary conditions on the cylinder splitting
strength. In Fracture Mechanics of Concrete Structures, Wittmann FH (ed.); Proceedings of the FRAMCOS-2.
AEDIFICATIO Publishers: Freiburg, 1995.
41. Rocco C, Guinea GV, Planas J, Elices M. Review of splitting-test standards from a fracture mechanics point of
view. Cement and Concrete Research 2001; 31:7382.
42. CEB-FIB Model Code 1990: Design Code/Comite Euro-International du Beton. Thomas Telford Publishing:
London, England, 1993.
43. Comi C, Perego U. Fracture energy based bi-dissipative damage model for concrete. International Journal of
Solids and Structures 2001; 38:64276454.
44. Feenstra PH, de Borst R. A composite plasticity model for concrete. International Journal of Solids and Structures
1996; 33(5):707730.

Copyright q

2007 John Wiley & Sons, Ltd.

Int. J. Numer. Anal. Meth. Geomech. 2008; 32:391413


DOI: 10.1002/nag

You might also like