Applied Mathematical Modelling 22 (1998) 495515

Analysis of reinforced concrete structures subjected to dynamic

loads with a viscoplastic DruckerPrager model
Juan Jose L
opez Cela

Universidad de Castilla-La Mancha, ETS Ingenieros Industriales de Ciudad Real, Campus Universitario s/n, 13071 Ciudad
Real, Spain
Received 23 July 1997; received in revised form 27 April 1998; accepted 2 June 1998

Abstract
The behavior of reinforced concrete structures subjected to dynamic loads is analyzed. The concrete material is
modelled by an elasto-viscoplastic law, whose inviscid counterpart is the DruckerPrager model. A viscous regularization is introduced in order to avoid the mesh dependency eects that usually appear when strain softening occurs.
The model is implemented in a general nite element computer code for fast transient analysis of uid-structure systems, based on an explicit central dierence scheme. The model is activated to both continuum elements and layered
shell elements. So, realistic numerical analyses of complex 3-D engineering problems are simple and ecient. Three
examples, two of which are modelled with layered shell elements, are presented below. 1998 Elsevier Science Inc.
All rights reserved.
Keywords: Fast transient analysis; Reinforced concrete; Viscoplastic material law; Layered shell elements

Notation
Mathematical model
r
r
c
r0
e
e0
e
F r; a; c
a; c
I
Qr; b; c
b; c
Dt
1

stress tensor
hydrostatic stress
shear stress
deviatoric stress tensor
strain tensor
deviatoric strain tensor
volumetric strain
DruckerPrager yield surface
material parameters dening the yield surface
identity tensor
plastic potential function
material parameters dening the plastic potential function
time increment

Tel.: +34 926 295 300; fax: +34 926 295 361; e-mail: jlopez@ind-cr.uclm.es.

0307-904X/98/$19.00 1998 Elsevier Science Inc. All rights reserved.

PII: S 0 3 0 7 - 9 0 4 X ( 9 8 ) 1 0 0 5 0 - 1

496

M
rf
q
g
v
l
ep
ep
Layered shell formulation
F int
Ve
B
x; y; z
n; g; f
W
L
NL
h
hL
uL
Material parameters
E
v
q
ryield
fc
ft
k
/
W
K
G

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

outward normal to the plastic potential function

viscoplastic stress tensor
vector of internal variables (a,c for the DruckerPerger model)
viscoplastic parameter
fastest elastic wave speed
mesh size parameter
equivalent plastic strain
plastic strain tensor
vector of internal nodal forces for a given element
volume of an element
matrix of derivatives of the shape functions
coordinates
normalized coordinates
weights of numerical integration
layer index
total number of layers
thickness of the shell
thickness of layer L
ratio of the layers thickness to the total thickness
Young's modulus
Poisson's ratio
density
elastic limit
maximum compressive strength
maximum tensile strength
cohesion
friction angle
dilatancy angle
bulk modulus
shear modulus

1. Introduction
The scope of this work is to analyze the behavior of reinforced concrete structures subjected to
dynamic loads such as impacts, explosions, etc. For the concrete material a DruckerPrager
elastoplastic model [1] is assumed. This model is relatively simple (only needs two parameters to
dene the yield surface and one more parameter to dene the plastic potential function and
consequently the ow rule) but can reproduce some features considered typical in concrete:
softening and dierent behavior in tension and compression. The softening phenomenon consists
in a reduction of the load-carrying capacity with increasing strain. This results in mesh sensitivity
problems. Regularization procedures are methods for avoiding this sensitivity [2]. One of these
methods, followed in the present work, is the introduction of viscoplastic terms in the material
constitutive law [3,4]. The numerical implementation is based on the radial return algorithm [5]
generalized for the case of strain hardening or softening [6], as summarized for a von Mises model
in Ref. [7]. A viscoplastic regularization is applied after an inviscid solution has been obtained [8].

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

497

Since the DruckerPrager yield surface in stress space is a cone, the developed algorithm includes
a specic treatment of the stress points that lie close to the cone apex, where the normal to the
yield surface is not dened.
This material behavior law has been activated to 2-D and 3-D continuum elements but emphasis has been done in to set up a simple but eective method for the numerical representation of
thin reinforced concrete structures. The chosen approach is `macroscopic', in that the structures
to be studied are discretized by shell nite elements. A typical element is viewed as a sandwich
composed of several layers, each one having its own, homogeneous material. For simplicity, no
attempt is done to model relative motions of the layers due e.g. to slip or delamination phenomena. This of course reduces the applicability of the method to the cases (i.e., load regimes or
parts of the time transients) where such extreme phenomena do not play a primary role.
The main advantages of the method are its simplicity and its high computational eciency, due
to the fact that a small number of relatively large shell elements are used to model a given
structure in place of myriads of much smaller continuum elements. Another attractive property is
that it is easily implemented as an extension in any general nite element computer code already
containing (homogeneous) shell elements and a library of suitable materials. The formulation is
such that any meaningful combination of layers and materials is easily obtained and the model
can be used, with the limitations pointed out above, not only for the representation of reinforced
concrete structures but also more generally for any composite material that admits a layered
approximation.
A material identication procedure has been developed in order to derive the concrete material
data dening the yield surface from piecewise linearized experimental stressstrain curves. The
procedure aims at reproducing a uniaxial compression curve and the maximum tension strength
value. Two methods have been investigated: in the rst one, the hardening or softening regime is
obtained by expanding the DruckerPrager yield surface without changing the angle of the cone.
The second method considers only variations of the angle. These evolutions of the yield surface
are directly related to variations, during the plastic ow, of the two physical parameters of this
model: cohesion for the rst case and friction angle for the second one. In the present study, these
two eects have been analyzed independently.
The model is implemented in PLEXIS-3C, a general nite element code for fast transient
analysis of uid-structure systems [9]. PLEXIS-3C is jointly developed since 1986 by the French
Commissariat 
a l'Energie Atomique (CEA-CEN Saclay) and by the European Commission (ECJRC Ispra). The code uses a central dierence, explicit time marching algorithm which, combined
with suitable lumping of the mass matrix, leads to a fully explicit implementation. PLEXIS-3C
oers a large library of nite elements, using both continuum and structural representations, and
therefore allows for straightforward implementation of the layer-based technique described in this
work. In addition, the code presents unique capabilities for the simulation of uid-structure interactions. Both the structural and the uid domain can be represented within one single, fully
coupled numerical analysis thanks to sophisticated, yet fully automatic, interfacing algorithms.
These are capable of automatically detecting and properly treating even the most complex uidstructure interfaces of the permanent type, see e.g. Ref. [10].
Three numerical examples are presented in this work. The rst one, consisting of a reinforced
concrete slab subjected to an impact load at the center, has been analyzed by using two-dimensional continuum elements. The second case is an impact on a natural-draught cooling tower,
which is discretized by triangular layered shell elements. The third test case is an example of fully
coupled uid-structure analysis, a gas explosion in a reactor containment. The geometry is axisymmetric and the structures are modelled with layered conical shell elements, while continuum
elements discretize the uids (explosive bubble and air) contained within the reactor building.

498

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

2. Constitutive model for concrete

The constitutive model used in the present study is elasto-viscoplastic. Its inviscid counterpart
is the DruckerPrager model with hardening or softening. A DruckerPrager model is represented by the two-parameter yield surface
F r; a; c ar s c 0;

where

1
r tracer hydrostatic stress;
3
1 p
s p r0 : r0 shear stress;
2

2
3

r0 r rI deviatoric stress;
4
where r is the stress tensor, I is the identity tensor and a; c are material parameters related to the
friction angle and the cohesion, respectively.
The numerical implementation of the model is based on the radial return algorithm but by
taking into due account the particularities of a DruckerPrager model [1]. The main dierence
between the two models is the shape of the yield surface in the stress space: while the von Mises
surface is a cylinder, the DruckerPrager surface is a cone. As a consequence, in a Drucker
Prager model an associative plastic ow produces variations in both hydrostatic and deviatoric
stress components. The radial return algorithm gives the time-discrete evolution equations for the
inviscid solutions rn1 , r0n1 , an1 and cn1 at time tn1 tn Dt.
The plastic ow is dened according to the plastic potential function:
Qr; b; c br s c 0

in the above equation, b is a material parameter related to the dilatancy, representing an inelastic
volume increase. If a b, an associative ow rule is obtained, otherwise the law is said non-associative.
Another particularity of the DruckerPrager model is that is necessary to distinguish two zones
in the yield surface:
A normal zone, where the return to the yield surface from the trial stress points is performed
according to the radial return algorithm, but with corrections in the hydrostatic and deviatoric
stress components, as indicated above.
A singular zone, close to the cone apex, where the normal to the plastic potential surface is not
dened and, therefore, it is not possible to return to it as simply as it is done in the previous
zone. This zone is delimited by a cone whose apex coincides with the apex of the DruckerPrager cone and whose generatrix forms a certain angle (depending on the value of the parameter
b) with the cone axis.
For the singular zone a special algorithm is developed. The normal to the plastic potential
function, that describes the plastic ow, is
oQ 1 r0 1
6

bI:
or 2 s 3
It can be seen that this expression becomes singular at the apex, because s 0. The return to
the yield surface is decomposed into two parts: the rst one is a projection onto the hydrostatic
axis, followed by a translation along this axis until the apex is reached.
In order to obtain the consistency parameter which assures that the nal stress state is in the
yield surface, we impose the continuity of this parameter along the boundary between the normal
M

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

499

region and the singular one. Since when b 0, it is not possible to return to the apex from the
singular zone, the model requires a positive value (possibly very small) of b.
The radial return algorithm is based on a backward Euler scheme. This means that to compute
the nal stress at time n 1 we need to know, at this time, the normal to the plastic potential
function and the consistency parameter. The normal is that corresponding to the trial stress. To
compute the consistency parameter it is necessary to distinguish if the hardening law is linear or
not. In the most general, non-linear case, the equation of the consistency parameter must be
solved by an iteration procedure. When the hardening law is linear, this equation is amenable to
closed form solution. Because in the kind of problems addressed in this work (fast transient
dynamics with explicit time integration) the time step is very small, the error produced by a
linearization of a non-linear hardening law during the time step is not very important. Then, for a
linear hardening law, this implementation and the ones adopted in Refs. [1,3] are the same. But
for a non-linear hardening law, in the implementation in PLEXIS-3C, it is very important to
perform such a linearization in order to maintain the explicit character of the code.
After an inviscid solution has been obtained the viscoplastic correction can be applied. It is
important to note that this correction is applied a posteriori, and that it is independent of the
model used for the inviscid case and of the kind of algorithm used for solving the equations. For
viscoplasticity, the formulation proposed by DuvautLions [11] extended to accommodate
hardening variables [8] is used. By integrating the viscoplastic constitutive model in closed form
[8], the following expressions are obtained:
h
i
1 eDt=g
Dt
rfn1 eDt=g rn 1 e g rn1
7
Dr;
Dt=g
h
i
Dt
Dt
qfn1 e g qn 1 e g qn1 ;
8
where q represents a vector of the internal plastic variables, that for the particular case of the
DruckerPrager model includes the two material parameters a and c; Dt is the step increment of
the central dierence scheme and g is a viscoplastic parameter having a time dimension.
The viscoplastic model is linear and therefore appears to be very poor in representing possible
strain rate eects. However, it should be recalled that this part of the model acts as a regularization of the solution and not as a physical contribution to it. In particular, the g parameter is
chosen according to numerical requirements. Being just a regularization, the model should be as
computationally cheap as possible. It is remarkable that the linear assumption leads to an explicit
evaluation of the stresses. In particular, no (conditionally stable) subincrementation algorithm
will have to be included for this evaluation. Because of the application after the inviscid solution
and the integration in closed form, the use of this regularization has a very low computational
cost indeed.
To select the value of the g parameter the criterion proposed in Ref. [4] is followed
l
6 5;
vg

where v is the fastest elastic wave speed and l is the mesh size parameter, that following common
practice is taken to be the diameter of the largest element in the mesh.
When g ! 0, Dt=g ! 1 and then rfn1 ! rn1 and the inviscid case is recovered. On the other
hand, if g ! 1 Dt=g ! 0 and then rfn1 ! rn Dr and we have the elastic case. For a given Dt
(xed with numerical criteria for the central dierence scheme) it is needed a value of g as low as
possible but suciently high to regularize the solution. Then, when a non-zero value of g is
adopted, dierent results than a pure elasto-plastic model will be obtained. It is important to

500

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

remark that, when mesh dependency occurs, the results with a elasto-plastic model are numerically possible but are not longer valid. On the other hand an elastic model of the material is not a
realistic option. With the current viscoplastic model, the results are not completely accurate, but
the achieved approximation can be considered enough for engineering purposes. In any case, the
simplicity of the model and its computational eciency compensates this inconvenient.
3. Modelling of thin shell reinforced concrete structures
The shell nite element formulations implemented in PLEXIS-3C have in common the basic
assumption that bers (straight lines across the thickness of the undeformed element) remain
straight during deformation. This hypothesis is justied by the fact that, in order to be represented
by shell elements, the structures must be relatively thin.
Fibers may or may not be initially normal to some element mid-plane, usually called the
reference surface. If the orientation of the bers with respect to the reference surface is allowed to
change during deformation, then transverse shear strains and stresses are included in the model
and one has a so-called MindlinReissner formulation. Otherwise, no transverse shear eects are
taken into account, and one has a Kirchho formulation.
In the shell formulation, the through-thickness stress is set to zero. In the two examples,
concerning shell concrete structures, presented in this work one is modelled with three-node
triangular plates and the other one with two-node axisymmetric shell elements. For the plate
element, the MindlinReissner formulation is assumed and then, only the component normal to
the plane of the plate is zero. For the conical shell due to the axial symmetry and because this shell
follows Kirchho theory, the three tangential stresses are zero. Therefore, the only non-zero terms
of the stress tensor are the longitudinal and circumferential components, rx and rh .
For any of the shell theories adopted in PLEXIS-3C, the main step in the element formulation
is the calculation of the internal nodal forces equivalent to the state of stress over the element,
which are then used to solve the equilibrium equations. Use is made of the principle of virtual
work which, at the element level, results in an expression of the type:
Z
int
F BT r dV ;
10
Ve
int

where F is the vector of internal nodal forces for a given element e; V e is the volume of the
element in the current conguration, B is the matrix of derivatives of the shape functions and r is
the Cauchy stress tensor.
Because of the special nature of the shell thickness z direction, the previous integral is usually
in the form (x and y being tangent to the reference surface):
0
1
Z Z Z
@ BT r dzA dx dy
F int
11
xy

and, because of the complexity of the functions involved (non-linearity of material behavior, large
strains, etc.) it is often best evaluated numerically:
"N
#
Nn X
Ng
f
X
X

12
F int
Wn Wg
BT r ngfWf det J :
n1 g1

f1

The W coecients are the weights of the numerical integration process (e.g., Gauss rule) and det J
is the determinant of the Jacobian matrix that describes the transformation of the element volume

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

501

into a normalized domain, characterized by the normalized coordinates n, g, f. In this way

the function is evaluated only at a given number of sampling or integration points within the
element.
Let us now compare a homogeneous element with one composed by several layers, see Fig. 1.
In the homogeneous case, the stress prole across the thickness may assume complex shapes
because of material non-linearities (e.g., plasticity), but is a continuous function. In the multilayer
case, instead, if the materials of the various layers have dierent properties then the stress prole
may become discontinuous across layers. The strain proles are continuous in both cases because
of the assumption of straight bers. In order to compute the internal forces, we apply a procedure
similar to 9, but now to each layer of the element separately
F int

(
"N
#)
Nn X
Ng
NL
f
X
X
X
T

det J :

Wn Wg
uL
B r ngfL WfL
n1 g1

L1

13

f1

Here L is the layer index which varies from 1 to NL , the total number of layers. NfL is the
number of Gauss integration points through the thickness in the layer L and fL the normalized
the ratio of the layers thickness hL to the
thickness coordinate for the layer. Finally, uL represents
P
total thickness h of the element: uL hL =h hence L uL 1.
4. Identication of the material parameters
The objective of this section is to determine the material parameters a and c of the Drucker
Prager constitutive law in order to reproduce, as closely as possible, the behavior of concrete. The
present identication procedure aims at reproducing the experimental uniaxial curves. In other
words, when a numerical uniaxial compression/tension test is performed, the resulting stress
strain curves should be as similar as possible to the experimental (real) ones. As a preliminary
step, it is common practice to approximate the continuous experimental curves with piecewise
linear ones.
The DruckerPrager criterion, characterized by a conical yield surface, can be considered as a
smooth approximation to the MohrCoulomb criterion, which uses an irregular hexagonal

Fig. 1. Idealization of homogeneous and layered shell structures.

502

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

pyramid yield surface. Following Ref. [12], the relations between the material parameters of both
surfaces are calculated as follows:
The starting point is the MohrCoulomb yield surface, where the relation between the material
parameters (cohesion k and friction angle /) and the compressive and tensile strength (fc and
ft ) is given by Mohr's circles:
sin /

fc ft
;
fc ft

14

fc 1 sin / 1 p
15

fc ft :
2 cos /
2
Then, by matching the DruckerPrager surface to the MohrCoulomb surface the desired relations are obtained.
There are several ways to approximate the two surfaces. If matching is imposed at the apex and
either at point A or at point B on the deviatoric plane (Fig. 2), then in the case of point A
matching, the cone circumscribes the hexagonal pyramid along the compressive meridians, while
in the case of point B matching, the DruckerPrager surface is circumscribed by the pyramid
along the tensile meridians.
If the two surfaces are made to agree along both compressive and tensile meridians, the relations of Table 1 are obtained.
These expressions are valid not only for the elastic limit point of the curve, but for all (plastic)
points along the curve. If the expressions of compression tting are used, then the maximum
compression value fc is correctly predicted, but the tensile one ft is overestimated (in this particular case, about a 30%, i.e., from 4 to 5.5 MPa). The same conclusions can be applied to the
other points. Then, due to the fact that it is not possible to t both curves simultaneously, the
examples presented in this work have been performed matching the compression one.
In the case of non-perfect plasticity, the problem arises of writing the material parameters in
terms of the equivalent plastic strain that is the magnitude used by the model implemented in the
k

Fig. 2. Fitting between MohrCoulomb and DruckerPrager criteria in the deviatoric plane.

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

503

Table 1

Fitting between MohrCoulomb and DruckerPrager criteria

Compression tting

Tension tting

2 sin /
a p
33 sin /

2 sin /
a p
33 sin /

3fc 1 sin /
c p
33 sin /

3fc 1 sin /
c p
33 sin /

code. To obtain these relations the yield surface is derived with respect to the material parameters
and it is substituted in the constitutive rate equations, particularized to the case of uniaxial expressions.
In this work, only one of the two possible material parameters k and / has been considered
variable while the other one has been kept constant. This corresponds to two dierent evolutions
of the yield surface, as shown in Fig. 3: expansion of the cone or variation of the angle of the
cone. Both cases represent a form of isotropic hardening.
The procedure is the following. The constitutive rate equation can be splitted into deviatoric
and volumetric parts:
r_ 0 2Ge_0 G

r_ 0 _
k;
s

16

_
r_ K e_ kKb;

17
0

where G and K are the shear and bulk moduli, e_ and e_ the volumetric and deviatoric part of the
rate strain tensor and k_ the consistency parameter.
For an uniaxial compression test along, e.g. direction 1 we have:
1
r r1 ;
3


2
1
1
0
r diag r1 ; r1 ; r1 ;
3
3
3
1 p
r1
s p r0 : r0 p ;
3
2

Fig. 3. Two dierent evolutions of the yield surface.

504

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

e e1 2e2 ;


2
1
1
e0 diag e1 e2 ; e1 e2 ; e1 e2 ;
3
3
3
where r1 and e1 are the stress and strain along direction 1, and e2 the transversal strain. A tension
test should be considered if we decide to match the tension curve. Then, Eqs. (16) and (17) become:
volumetric rate equation
1
_
r_ 1 Ke_1 2e_2 kb;
3

18

deviatoric rate equations (there are three deviatoric equations but because they are coincident,
we only write one of them)
p
2
2
2 3 _
r_ 1 2G e_1 e_2
Gk:
19
3
3
3
Combining Eqs. (18) and (19)
"
#
p
k_ 
20
r_ 1 E e_1 b 3 :
3
On the other hand, the plastic strain rate tensor is dened according to the ow rule as
"
p
p #
 0

_
p

oQ
3
3
k
1
r
1
;b
k_
bI diag b 3; b
e_p k_
2
2
b
2s 3
or
and the equivalent plastic strain is
q
k_
2 p
2b2 3
e_ p
e_p : e_p
3
3
or
3e_ p
k_ q :
2b2 3
Finally, Eq. (20) becomes
2
3
p
b 3 7
6
r_ 1 E4e_1 q e_ p 5:
2b2 3

21

22

23

24

To obtain the two evolutions above explained independently an expression of the DruckerPrager
yield surface in terms of the cohesion k and friction angle /; is needed. This expression, matched
along compressive meridians with a MohrCoulomb surface, is (Ref. [11])


3 sin /
p
k cos / 0
25
F r sin / s
2 3

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

505

and particularizing for an uniaxial compression test, one arrives at

r1

2k cos /
:
1 sin /

26

The next step is to derive Eq. (26) with respect to k and /.

Expansion of the yield surface (k variable and / constant)
r_ 1

dr1 dk _ p
2 cos / dk _ p
e
pe
dk de
1 sin / dep

27

and by eliminating e_ p between Eqs. (24) and (27)

p
dk
b 3 1 sin / r_ 1
0
k E q
:
dep
2 cos / r_ 1 Ee_1
2
2b 3

28

Evolution of the angle of the cone (k constant and / variable)

r_ 1

dr1 d/ _ p
2
d/ _ p
e
e
d/ dep
1 sin / dep

29

and by eliminating e_ p between Eqs. (24) and (29)

p
d/
b 3 1 sin / r_ 1
0
/ E q
:
r_ 1 Ee_1
dep
2k
2b2 3

30

The necessary steps to perform the tting are shown in Table 2 in which the previous equations
are written in discrete form.
With these two tting procedures the compression curve is correctly reproduced for every value
of b. However, the tension curve is overestimated and the post-peak regime exhibits a strong
dependence on b. For b 0 (non-associative ow rule) the results are better than with b a
(associative ow rule). In a similar manner, if it is decided to match the tension curve, the same
kind of problems will appear in the compression curve. There is no special reason to select one of
the two curves. Even the code allows the possibility of using both of them: one part of the
structure modelled with parameters obtained from a compression curve and the other with those
parameters that predict the tension curve. Also, another kind of curve can be suggested: for
example, a triaxial test. In this latter case, none of the two uniaxial curves will be accurately
predicted. This highlights not only the troubles for this simple model to predict the complete
behavior of concrete, but also the big diculties we nd in trying to derive realistic material
parameters.
To obtain the value of b, and because the expression of the potential function (Eq. (5)) is
similar to the expression of the yield one (Eq. (1)), we can use
2 sin W
b p
;
33 sin W

31

where W is the dilatancy angle, which represents an increase of plastic volume under pressure.

506

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

Table 2

Material identication procedure

Cone expansion

Cone angle variation

Step 1. Compute sin / fc ft =fc ft

Step 2. For each value of the piecewise linear
compression curve, compute from Eq. (26)

p
Step 1. Compute k 12 fc ft
Step 2. For each value of the piecewise linear compression curve,
compute from Eq. (26)

ki

ri 1 sin /
2 cos /

Step 3. Compute the DruckerPrager material

parameters a and c
2 sin /

a p
33 sin/
3ri 1 sin /
ci p
33 sin /

/i r2i 4k 2 =r2i 4k 2
Step 3. Compute the DruckerPrager material parameters a and c
2 sin /i
ai p

3 3 sin /i


3ri 1 sin /i
ci p

3 3 sin /i

Step 4. Compute dk=dep i k 0 i for each Dr, De

of the curve:
p
b 3 1 sin /
Dr
ki 0 E q
2 cos / Dr EDe
2
2b 3

Step 4. Compute d/=dep i /0 i for each Dr, De of the curve:

p
b 3 1 sin /
Dr
/i 0 E q
2k
Dr EDe
2
2b 3

Step 5. Compute the value of the increment of

equivalent plastic strain for each increment of the k parameter
Dki
Dep i
ki 0

Step 5. Compute the value of the increment of equivalent plastic

strain for each increment of the / parameter
D/i
Dep i
/i 0

Note. Because in the expression of /i 0 appears sin/ that is variable along the curve some discretization is needed in the linear
branches of the curve.

5. Numerical examples
Several test cases, using 2-D continuum elements, proposed in Refs. [3,4] have been successfully
reproduced with the present model. The main objective of this calculations was to check the
eectiveness of the viscoplastic regularization introduced in the material law to obtain mesh independent results. As a further example, a simply supported reinforced concrete slab subjected to
an impact load at the centre was considered. The structure is also modelled with 2-D continuum
elements in a plane strain situation.
Two additional examples are presented in this section: impact on a cooling tower and gas
explosion in a reactor containment. In these cases, the models of the structures are built with
layered shell elements.
In all computations, the steel is modelled with an elastic-perfectly plastic von Mises material
while to represent the concrete the material law is the described above and in some detail in Ref.
[13].
The material properties are the following. For the steel, elastic-perfectly plastic behavior with
E 210 GPa, q 7800 kg/m3 , m 0:3 and ryield 680 MPa. For the concrete, elastoplastic behavior with softening, E 20 GPa, q 2400 kg/m3 , m 0:2; dilatancy angle W 10 , maximum
compression strength fc 40 MPa, maximum tension strength ft 4 MPa, and softening behavior dened by the compression curve represented in Fig. 4.

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

507

Fig. 4. Concrete slab subjected to impact load: problem denition.

5.1. Concrete slab subjected to impact load

The problem description is shown in Fig. 4. Only one half of the structure was modelled because of the symmetry and two-dimensional continuum elements with four nodes and four integration points were used to discretize the concrete domain. Plane strain conditions are
considered.
The reinforcement consisted of 15 steel bars of 18 mm diameter along a 1 m depth of the slab.
The equivalent steel area was A 3817  106 m2 of steel per square meter of concrete. The
discretization of the steel reinforcement was done with beam/shell elements with the nodes rigidly
attached to the nodes of the continuum elements.
To obtain the model parameters for concrete both yield surface evolutions (variation of the
angle of the cone and cone expansion) were used. Because the dilatancy angle W is 10 a nonassociative ow rule was considered and b 0:2.
Three dierent meshes were used in the study with a value of the viscoplastic parameter
g 1  106 . In these conditions, the factor l=vg was 4.12 for the rst mesh, 3.09 for the second
one and 2.47 for the last and nest one. In all cases it was less than 5, fullling Eq. (9).
A rst set of computations was performed in order to verify again the eectiveness of the
viscoplastic correction: unreinforced slab, steel reinforced slab (bottom-only or bottom-and-top
reinforcement) and concrete characteristics derived with the two yield surface evolutions explained above. In Fig. 5 the deformed meshes of the unreinforced slab are shown for a computation time of 1.05 ms (corresponding to twenty wave reections across the slab). The results for
the inviscid case are mesh dependent (note that just one row of elements fails in each of the
meshes) while those calculated with the viscoplastic regularization avoid this problem. In the rest
of the analyzed cases, also mesh independent results were obtained.
In Fig. 6 are presented some results in terms of equivalent plastic strain distribution for the
same computation time of 1.05 ms. The presented results are those obtained with bottom reinforcement only and with the intermediate mesh, i.e. l=vg 3:09. The concrete characteristics are

Fig. 5. Concrete slab: deformed meshes for three dierent element size: (a) inviscid case; (b) viscoplastic case.

508
J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

509

Fig. 6. Concrete slab: equivalent plastic strain distributions at 1.05 ms for two possible evolutions of the yield surface.

calculated according the two possible yield surface evolutions. The following conclusions can be
drawn from this gure:
Dierent strain distributions appear in the two computations.
Lower strain values result in the case of characteristics obtained by varying the angle of the
DruckerPrager cone. This is due to the fact that when this tting is used, the tension curve
presents, in the softening part, higher values than those obtained with the other tting, i.e.,
the reduction of the load carrying capacity is less pronounced.
The results obtained considering variations in the angle of the cone seem to be closer to reality
than the ones obtained with the other tting, i.e., expansion of the cone.
5.2. Thin shell structures
As an example, consider the frequent case of concrete shell structures having reinforcement
grids placed in two `layers' within the bulk of the concrete material, close to the inner and outer
surfaces (this typical layout will be assumed in both numerical examples presented below). If each
grid has a large number of bars oriented along various directions, it can be approximated by an
isotropic steel layer in the model. The layer thickness must be chosen so as to represent the real
steel fraction present in the structure. The shell is then composed of 5 layers, of which 3 are of
concrete and 2 of steel, see Fig. 7. All layers have 1 integration point along the ber, except the
central one that, because of its thickness, has 2 Gauss points.

Fig. 7. Layered shell element: Location of integration points along the thickness.

510

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

Fig. 8. Impact on a cooling tower: problem denition.

In Fig. 7 a section across the thickness of a 2-D layered element is represented, where the
concrete and steel layers as well as the location of the integration points are shown. For the
presentation of results, all element quantities will be associated with a lamina, understanding for
that the plane containing all the integration points situated at the same height along a ber. So,
the example of Fig. 7 is composed of 5 layers with 6 integration points altogether and consequently 6 laminae.
The viscoplastic parameter g will be determined separately in each example, because it depends
not only on material parameters but also on the element's length.
5.2.1. Impact on a cooling tower
This example is concerned with a typical reinforced concrete structure, namely a naturaldraught cooling tower. This 3-D problem is purely structural since the impact loading is simulated
by an imposed external pressure. The main geometric, material and layer characteristics are
shown in Fig. 8. Because of the symmetry, only one half of the structure has been modelled. The
mesh is composed of 800 three-node triangular plate elements.
To select the g parameter, use is made of Eq. (9). One may obtain a good estimate of the ratio
l=v by letting PLEXIS-3C (automatically) compute the critical time integration step
Dt mine l=v, where the minimum is taken over all elements in the mesh. The value is only
approximated because the code uses a weighted average speed between those of concrete and
steel, but, in any case, it is easier to proceed in this way than manually in a possibly quite
complicated 3-D mesh. For security, and due to this average speed value, we use a factor 1 instead
of a factor 5 in Eq. (9), that is l=vg < 1: For this model, the code automatically computes
Dt 0:684 ms. Since PLEXIS-3C applies a safety factor of 0.5 the g parameter is:

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

511

Dt
32
1:368  103 ) g 2  103 :
0:5
The calculation was performed until a physical time of 100 ms. It required 148 time steps and
120 s of CPU on an HP 9000 712 workstation.
In Fig. 9 the equivalent plastic strain distributions are presented for the 4 concrete laminae at
the nal time. For ease of comparison, the same scale is used in all drawings. The strongest
concentration of strain is located in lamina 6, that is the internal one. The pressure acts toward the
center of the tower, so the inner laminae 4 and 6 are under tension in the center and therefore, the
higher values of plastic strain are located in these zones.
g>

5.2.2. Gas explosion in a reactor containment

This test case is an example of fully coupled uid-structure analysis, a gas explosion within the
reinforced concrete secondary containment building of a nuclear power plant. The geometry of
the problem is given in Fig. 10 and is assumed to be axisymmetric. The lower basement of the
containment is very thick and can be modelled by a rigid boundary. The building walls are relatively thin and can be conveniently represented by two-nodes conical layered shell elements
without a topological thickness. The containment is supposed to be initially lled by air at room
temperature and atmospheric pressure. An explosion is assumed to take place in the lower part of
the building at the initial time of the studied transient.
The explosive is simply represented by a perfect gas at high pressure, which initially occupies
the zone indicated in black on Fig. 11 (t 0). The properties of this gas are: density q 111:5 kg/
m3 , adiabatic exponent c 1:4, specic internal energy i 1:28 MJ/kg. The bulk air is also a

Fig. 9. Impact on a cooling tower: equivalent plastic strain distributions for the 4 concrete laminae at 100 ms.

512

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

Fig. 10. Gas explosion in a reactor containment: problem denition.

perfect gas with q 1:2 kg/m3 , c 4 and i 0:208 MJ/kg. The characteristics of the structural
materials are the same as for the previous example, and those corresponding to the layers are in
the table of Fig. 9.
The mesh is composed of 981 continuum uid elements (mostly 4-node quadrilaterals with a
few 3-node triangles) and 88 conical shells. In this case, proceeding like in the previous example,
one obtains Dt 0:0213 ms, and therefore the viscosity parameter is
g>

Dt
4:26  103 ) g 7  105 :
0:5

33

The calculation required 11750 steps and 4120 s CPU on an HP 9000 712 workstation, for a
physical time of 250 ms.
Fig. 11 shows the evolution of gas pressure within the containment in the rst 120 ms; note the
strong wave propagation and reection eects. Fig. 12 shows the deformation of the structure,
amplied by a factor 10 to highlight the critical spots. The upper drawings correspond to the
present case while the lower ones were obtained by assuming a purely elastic homogeneous
material for the structure. Note that plasticity is reached in the present models's solution both at
the containment top and, even more pronounced, at the inner horizontal oor, which appears to
be seriously damaged at the end of the computer transient.

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

513

Fig. 11. Gas explosion in a reactor containment: uid pressures.

6. Conclusions
A fully explicit and non-iterative approach to the modelling of reinforced concrete structures,
including viscous regularization to avoid mesh sensitivity problems, has been proposed. The
concrete constitutive law is based on a DruckerPrager yield criterion with softening or hardening. The numerical implementation is based on the classical radial return algorithm. The explicit character of the computer code is maintained even in the case of a non-linear hardening law.
This is achieved by performing a linearization in the hardening law during the time step. Due to
the fact that close to the apex of the DruckerPrager cone exists a singular zone where the return
trajectory is not well dened, a special algorithm for the trial stress states that lie in this part of the
stress space has been developed. The main assumption is to impose the continuity of the consistency parameter along the boundary between the normal region and the singular one. Mesh
independent results have been obtained in several test cases modelled with 2-D continuum elements.
For evaluating the concrete material parameters two possible evolutions of the yield surface
were investigated: more realistic plastic strain distributions were obtained by varying the angle of
the DruckerPrager cone rather than by expanding the cone itself. Both yield surface evolutions
represent a form of isotropic hardening or softening.
The model has been extended to layered shell elements in order to analyze in a simple but
ecient manner thin reinforced concrete structures. No modelling of the relative motions between
layers has been attempted in this formulation. The examples have shown that the method is
correctly implemented and may be applied to realistic analysis, with plausible results. However, a

514

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

Fig. 12. Gas explosion in a reactor containment: structural deformations (10).

full validation and calibration campaign remains to be done, by comparison with experiments and
other data available in the literature.

Acknowledgements
This work has been performed during a post-doctoral Human Capital and Mobility fellowship
at the Institute for Systems, Informatics and Safety of the Joint Research Center of the European
Commission, Ispra, Italy. The author thanks Dr. F. Casadei and Dr. P. Pegon for the helpful
discussions while carrying out this work.
References
[1] B. Loret, J.H. Prevost, Accurate numerical solutions for DruckerPrager elastic-plastic methods, Comp. Meth.
Appl. Mech. Eng. 54 (1986) 259277.
[2] R. De Borst et al., Fundamental issues in Finite Element analysis of localization of deformation, Eng. Comput. 10
(1993) 99121.
[3] L.J. Sluys, Wave propagation, localization and dispersion in softening solids, Ph.D. Thesis, Delft University, 1992.
[4] B. Loret, J.H. Prevost, Dynamic strain localization in elasto-(visco)-plastic solids. Part 1. General formulation and
one-dimensional examples. Part 2. Plane strain examples, Comp. Meth. Appl. Mech. Eng. 83 (1990) 247273 and
275294.

J.J. L
opez Cela / Appl. Math. Modelling 22 (1998) 495515

515

[5] M.L. Wilkins, Calculation of elastic-plastic ow, University of California, Lawrence Livermore National
Laboratory, Rep. UCRL-7322 Rev. 1, 1969.
[6] R.D. Krieg, S.W. Key, Implementation of a time-independent plasticity theory into structural computer programs,
in: J.A. Stricklin, K.J. Saczalski (Eds.), Constitutive equations in viscoplasticity: Computational and engineering
aspects (AMD-20), Amer. Soc. Mech. Eng., New York, 1976.
[7] T.J.R. Hughes, Numerical implementation of constitutive models: Rate independent deviatoric plasticity, in:
S. Nemat-Nasser, R.J. Asaro, G.A. Hegemier (Eds.), Theoretical foundation for large-scale computations for
nonlinear material behaviour, Martinus Nujho Publishers, Boston, MA, 1984, pp. 2657.
[8] J.C. Simo, J.G. Kennedy, S. Govindjee, Nonsmooth multisurface plasticity and viscoplasticity. Loading/unloading
conditions and numerical algorithms, Int. J. Num. Meth. Eng. 26 (10) (1985) 21612185.
[9] H. Bung et al., PLEXIS-3C: A Computer code for fast transient problems in structures and uids, 10th
International Conference in Structural Mechanics in Reactor Technology, SMiRT-10, Anaheim, USA, 1419
August, 1989.
[10] F. Casadei, J.P. Halleux, An algorithm for permanent uid-structure interaction in explicit transient dynamic,
Comp. Meth. Appl. Mech. Eng. 128 (1995) 231289.
[11] G. Duvaut, J.L. Lions, Les inequations en Mechanique et Physique, Dunod, Paris, 1972.
[12] W.F. Cheng, Plasticity in reinforced concrete, Mc Graw-Hill, New York, 1982.
[13] J.J. Lopez Cela, P. Pegon, F. Casadei, Fast transient analysis of reinforced concrete structures with DruckerPrager model and viscoplastic regularization, in: D.R.J. Owen, E. Onate, E. Hinton (Eds.), Proceedings of the
Fifth International Conference on Computational Plasticity, Fundamentals and Applications, Barcelona, Spain,
1720 March 1997 (CIMNE).