You are on page 1of 6

Journal of Materials Processing Technology 80 – 81 (1998) 591 – 596

Plastic and viscoplastic damage models with numerical treatment


for metal forming processes
E. Bellenger *, P. Bussy
LMCAO Uni6ersité de Picardie, Pôle Régional de Picardie, IUT, 48 rue d’Ostende, 02100 Saint-Quentin, France

Abstract

Metal forming processes are problems with a multiplicity of non linearities. There are physical non linearities (plasticity,
viscoplasticity, damage), geometrical non linearities (large displacements, large deformations), contact and friction non linearities.
Therefore, numerical simulations of metal forming processes lead to choosing appropriate constitutive equations and a suitable
algorithm to solve these problems. The work presented here treats these difficulties. Firstly, we proposed a stable elasto-plastic
damage model and a stable elasto-viscoplastic (visco) damage model for large transformations. Secondly, we use the large time
increment method in a version adapted to solve non linear geometrical problems with contact and friction. © 1998 Elsevier Science
S.A. All rights reserved.

Keywords: Plasticity; Viscoplasticity; Damage; Large transformations; Numerical treatment

1. Introduction be taken into account because of its preponderance


over other phenomena such as plasticity or
The presence of finite inelastic deformations in mate- viscoplasticity.
rials gives rise to many difficulties in computational A large number of damage models have been devel-
structural mechanics. This is the reason why the use of oped in order to allow numerical simulation of struc-
finite element codes in metal forming industry, e.g. the tural damage. The implementation of these models in
sheet metal forming or the forging, is still rare. The finite element codes is difficult because it leads to
modeling of metal forming processes involves a large unstable calculations. Convergence becomes impossible
number of non linearities. Firstly, there are geometrical after limiting points. This problem results from bad
non linearities with large displacements and large defor- modelization of non convex free energy. Therefore, we
mations, in addition to conditions of contact and fric- propose a stable elasto-plastic damage model and stable
tion. There are non linearities in material behavior such viscoplastic (visco) damage model for large transforma-
as plasticity, viscoplasticity and damage. Plasticity and tions. These are extensions of the elastic damage model
viscoplasticity have been well modeled in the past, developed by Benkrid and Bussy [1], stable to general-
whereas damage has not. ized standard materials in its formulation for small
However, numerical simulations of metal forming perturbations.
processes, to be realistic, need to consider the phe- Another key problem in the metal forming numerical
nomenon of material damage. One purpose of process simulation, concerns the computational algorithm. This
optimization is to eliminate, or at least minimize, dam- choice is fundamental because it is the principal reason
age development. This is the case for processes such as of the large calculation cost. The finite element codes
extrusion, forging, rolling or deep drawing. Models generally use Newton–Raphson methods for solving
must therefore include damage in order to ensure that it non linear problems. These methods are incremental in
does not appear. On the other hand, in the operations time and iterative on each increment. These are more
of sheet cutting, punching or machining, damage must costly when non linearities accumulate. This is the case
for our problems, which involve several non linearities.
Numerical simulations of metal forming processes with
* Corresponding author. classical software lead to small, and therefore numer-

0924-0136/98/$19.00 © 1998 Elsevier Science S.A. All rights reserved.


PII S0924-0136(98)00166-6
592 E. Bellenger, P. Bussy / Journal of Materials Processing Technology 80–81 (1998) 591–596

ous, loading time increments, leading to a large number convex written with the stress tensor ensures stability.
of global resolutions. We choose the non linear geometric model proposed in
Unlike classical incremental techniques, the large [7]. It introduces a local rotation R defining the co-rota-
time increment method that is used in this paper en- tional configuration. We have to write the damage
ables these problems to be simulated without proceed- model into this co-rotational configuration of the geo-
ing by small increments. This method was first metrical model. In the following, we begin to present
introduced by Ladeveze [2] in the context of structural constitutive laws in small perturbations before we
mechanics. We will use a version proposed in [3], which present their extensions to non linear geometry.
is applicable to non linear problems with geometrical
non linearities, contact and friction. This numerical 2.1. Stable elastic damage model
method does not cut the loading path into small incre-
ments, instead the whole loading path is considered as The damage model is developed within the general
only one increment. This strategy makes it easy to framework of continuum thermodynamics of irre-
integrate the limit points and gives good prediction of versible processes. To formulate the constitutive equa-
the post critical behavior. This algorithm has been used tion, it is necessary to define the free energy and
to simulate sheet metal forming [4] and sheet cutting damage potential as functions of the internal state
[5]. variables. This presentation assumes elasticity coupled
with isotropic damage as an isothermal process also.
The model is defined by the specific free energy c(oe, X)
2. Constitutive equations including damage in the following form:
rc(oe, X)= rc1(oe, X)+ rc2(X) (2)
In order to model changes associated with damage,
some approaches were developed. We can distinguish where r denotes the material density, X is the scalar
two major classes: (i) the phenomenological approach damage variable; it is the internal variable associated
starting with the model of Kachanov, (ii) a microscale– with the variable Y. oe is the elastic strain tensor.
macroscale approach that intends to construct models rc1(oe, X)= 1/2 tr(oe(Ce/1+X)oe) is the elastic damage
at a macroscale using some considerations from the free energy. Ce denotes the linear elastic tensor. c2(X) is
microscale. a convex function.
In this work the phenomenological approach is cho- The associated thermodynamics forces are:
sen by using an internal variable to describe the conse-
quences of damage evolution on the elastic properties,
as well as on the plastic or viscoplastic flow properties, (c 1 Ce
s= r = o

 
of the material. The scalar damage variable D used by (oe 1+ X e
Lemaitre and Chaboche [6] and many others authors, is (c 1 (c 1 Ce
interpreted as the effective area reduction caused by the Y= r + r 2 = − tr oe o + Z(X) (3)
(X (X 2 (1+X)2 e
distributed microscopic cracks and cavities due to mate-
rial damage. In this work we introduce a new scalar
continuous variable, X, related to the usual damage
variable D bounded by 0 and 1. where s denotes the Cauchy stress tensor; tr( · ) is the
trace of a tensor. From Eq. (3), Y takes the following
D 1
X= [ 1−D = (1) form:
1− D 1+X
1
dD must verify the condition D +dD B1, then dX is e s)
Y= Z(X)− tr(sC − 1
(4)
2
not bounded, which avoids numerical problems. X
takes values within the range 0 (for virgin material) to The second principle of thermodynamics invites us to
(corresponding to the complete failure). write and verify the Clausius–Duhem inequality. To
In this paper, constitutive equations are based upon make a standard generalized model [8] we define a
the stable elastic damage model [1] developed in small damage potential of dissipation 8*X (Y). The dissipation
perturbations. This convex model belonging to a set of is positive when we introduce the convex potential of
generalized standard materials allows us to have stable dissipation, which is chosen as the indicator function of
calculations when a point reaches a critical damage the damage convex defined by the yield function:
level. The basic idea in the formulation of this model is 1
e s)− Z(X)− Z0 50
fX (Y)= − Y− Z0 = tr(sC − 1
the following. The damage affects essentially the mate-
2
rial elastic stiffness. It is the coupling between the
(5)
model and geometrical non linearities which enables
strain softening to occur. The choice of a damage Z0 ] 0 is the initial damage threshold.
E. Bellenger, P. Bussy / Journal of Materials Processing Technology 80–81 (1998) 591–596 593

Fig. 3. Mesh and boundary conditions.

where rc3(p) is a convex function.


The associated thermodynamics forces are:
Fig. 1. Co-rotational model. (c1 Ce (c3
s=r = oe, R= r
(oe 1+ X (p
Damage occurs only when the state of stress reaches
the actual yield function. This corresponds to the satis- Y= r
(c2
+r
(c1 1
=Z(X)− tr oe
Ce
o
  (8)
faction of the yield criterion fX =0. Damage continues (X (X 2 (1+X)2 e
to grow if the yield is continuously satisfied, i.e. if
f: X = 0. 2.2.1. Elasto-plastic damage model
The complementary evolution law is: For the elasto-plastic constitutive law (oin =op), we
use the Prandtl–Reuss behavior law and the damage
(fX
X: = − lX = lX (6) elastic model presented above. We formulate the von
(Y Mises flow criterion in terms of Cauchy stress tensor:
The expression of the multiplier lX is deduced from fp = seq − R(p)− R0 5 0, (9)
the consistency condition fX =0 and f: X =0.
where p; = t0
2/3o; p : o; p, p the accumulated plastic
2.2. Stable inelastic damage models strain is the internal variable associated with the
isotropic hardening variable R. seq =
This presentation assumes elasticity coupled with
32 tr(dev s dev s) with dev s the deviatoric part of the
isotropic damage. We consider only isotropic harden- Cauchy stress tensor. R0 denotes the initial plastic
ing. If o, oe and oin denote, respectively, the total, the hardening threshold.
elastic and the inelastic strain tensors we write o =oe + The second principle of thermodynamics invites us to
oin. The constitutive laws of the model are determined write and verify the Clausius–Duhem inequality. From
by introducing the free energy function c(oe, X, p): Eqs. (7) and (8) this inequality takes the following
form:
rc(oe, X, p)= rc1(oe, X) + rc2(X) + c3(p) (7)
tr(so; p)− Rp; − YX: ] 0, (10)
If we assume damage and plastic processes dissipate
independently (Eq. (10)) must be verified in two parts.
To make a standard generalized model we define a
potential of dissipation 8*(s, R, Y) composed of two
terms. 8*X (Y) concerns the damage dissipation and
8*p(s, R) concerns the plastic dissipation. If absolute

Fig. 2. Comparison of the experimentally obtained stress–time with


predicted theoretically for IN100 in temperature of 1000°C. Fig. 4. Isostress.
594 E. Bellenger, P. Bussy / Journal of Materials Processing Technology 80–81 (1998) 591–596

rate independence is sought when modeling the behav- From the computational point of view, the above
ior, Eqs. (5) and (6) are still valid and the plastic constitutive equations (Eqs. (5) and (9)) can be inte-
potential can be chosen as the indicator function of the grated using an elastic predictor/plastic damage correc-
elasticity convex fp. tor algorithm. We assume that at time tn the state
The complementary evolution laws are: variables are known. Given Dun + 1, the displacement
between tn and tn + 1, we can perform an elastic predic-
(fp 3 dev s (fp
o; p = lp = lp , p; = −lp = lp (11) tor, assuming that the plastic and the damage variables
(s 2 seq (R remain frozen. The trial elastic Cauchy stress can be
(fX computed by Eq. (8). If the stress lies outside the elastic
X: = − lX = lX (12) domain fp and the damage domain fX, a damage-plastic
(Y
corrector step is then carried out using a classical radial
where for i= p or X, li ]0 and lifi =0. The plastic and return algorithm [10].
the damage consistency parameters are determined by For the integration of time-dependent behavior laws
the consistency conditions f: i =0 and fi =0. (Eqs. (14) and (16)), the Euler family method is used. It
consists of the following equation:
2.2.2. Elasto-6iscoplastic (6isco) damage model
xtn + 1 = xtn + [(1− u)x; tn + ux; tn + 1]Dt (17)
Unlike plasticity, viscoplasticity (oin =ovp) is a time-
dependent phenomenon. Therefore, a time-dependent Dt is the time increment between tn and tn + 1, u[0, 1],
yield function is introduced, and the stress may depart x {ovp, p, X given by Eq. (16)}.
from the elastic domain. Eqs. (7) and (8) are still valid. Damage and inelastic phenomena evolutions lead to
For viscoplastic dissipation the following particular the definition Cauchy stress tensor at t+ Dt by:
form of 8*vp(s, R) has been proposed in [9]:

8*vp(s, R)=
#
K seq −R(p) − R0 $ n+1
(13)
st + Dt =
Ce
1+ Xt + DX
[Do − Doin + st (1+Xt )] (18)
n +1 K +

K and n are material parameters,  ·  + denotes the 3. Geometric model


positive part.
If we consider for example, that for a constant We consider an initial configuration V0. Loads Fd,
applied load the damage evolution is not directly linked are applied to a part (2V0 on its boundary and displace-
with time as for viscoplasticity, we use Eqs. (5) and (6). ments ud on the part (1V0. The third part (3V0 repre-
Thus, it is the non linear geometrical phenomenon sents the set of points that comes into contact. Body
coupled with viscoplasticity which allows damage to forces density fd is also applied. M0 is a point on V0
grow. In that case the viscoplastic strain rate and the which we find at M at the instant t. M is defined by
damage rate take the following forms: M= M0 + u(M0, t), where u(M0, t) is the displacement
field (Fig. 1).
(*vp(s, R) 3 dev s Generally, the variables used to describe the constitu-
o; vp = = p; , tives laws in large displacements are not objective. To

# $
(s 2 seq
be objective we choose the non linear geometrical
(* (s, R) s −R(p) −R0 n
model proposed in [7]. It introduces in all points of the
p; = vp = eq (14)
(R K + domain a local rotation R defining the co-rotational
(f referential. This rotation locally transforms the configu-
X: = − l X X ration of V into a rotated configuration V (Fig. 1). R is
(Y
associated to the rotation rate W and defined by the
differential equation and the initial condition:
In order to describe a time-dependent damage evolu-
tion, we introduce a time-dependent yield function for
Á
damage. As for viscoplastic flow, we choose the follow- Ã: T 1
R R= W= (L−L T)
ing particular form for 8*X (Y): Í 2 (19)
Ã
8*X (Y)=
1
Y− Z0N++ 1 (15) ÄR t = 0 = I
N+ 1
Eq. (14) is still valid and damage rate is given by: where W is the rotation rate, L= F: F − 1, F=I+du/

X: =
# 1
e s) −Z(X)− Z0
tr(sC − 1 $ N
(16)
dM0 is the gradient tensor. With this rotation we obtain
the strain rate and the stress tensor in the rotated
2 + configuration.
E. Bellenger, P. Bussy / Journal of Materials Processing Technology 80–81 (1998) 591–596 595

!t =RtR T
(20)
Find s= (u, F, 7, P, q) defined on [0, T]× V checking
Öt[0, T]:
D = RDR T
— (F, 7) kinematically admissible:

t and D denote, respectively, the Kirchhoff stress and du


the strain rate in the rotated configuration. D= F(u)= I+ , 7= u ( 3V0, u ( 1V0 = ud (21)
dM0
(1/2)(L + L T). Furthermore, t; , the Kirchhoff stress rate
in the rotated configuration, takes the form: t; =Rt JR T, — (P, q) statically admissible: Öu*U, U= {u ( 1V0 =0}
where t J, denotes the Jaumann derivative of the Kirch-
hoff stress.
We obtain the constitutive laws defined in non linear
& tr(P TF*) dV0 +
& q7* d(3V0

& &
geometry by writing into the co-rotational configura- V0 ( 3V0

tion of the geometrical model between t; and D, the


models developed above in small perturbations. Then, = Fdu* d(2V0 + fdu* dV0 (22)
( 2V0 V0
material behavior laws are objective constitutive laws.
—Constitutive laws:

F: t = A(Pt, t5 t) and R(q, 7; ) (23)


4. Numerical treatment

The development of reliable numerical procedures to where A is a material characteristic function and R
predict materials behavior in metal forming processes, represents the contact and friction relations. P=JsF−T
has encountered many serious obstacles. Together with is the first Piola Kirchhoff stress tensor; J= det F.
the non linearity of the material, other important ef-
fects like the unsteady nature of the process, the large 4.2. The large time increment method
magnitude of the strains and the importance of contact
and friction make the study of metal forming processes This problem formulation displays, on one hand, (i)
so complex that is analysis justifies the use of sophisti- the admissibility relations, Eqs. (21) and (22), which are
cated finite element numerical algorithm, which leads to linear but not locally, and on the other hand (ii) the
large scale computer requirements. Therefore, in order local constitutive relations (Eq. (23)) which involves all
to reduce the calculation cost we use the large time the non linearities (geometrical, material, contact and
increment method [11]. In particular, the version due to friction) of the problem. (i) is called Ad and (ii) is called
Bussy et al. [3], adapted to solve non linear problems G.
with contact and friction. The iterative pattern of this This presentation allows us to find the mathematical
numerical method gives for each iteration a new estima- framework of the large time increment method. This
tion solution for all the loading history. version of the algorithm implemented in our finite
element code OPTIFIA uses the variable rates s=
(u; , F: , 7; , P: , q; ). Each iteration is split into two steps.
4.1. Presentation of the problem Firstly, a non linear and local step that enables us to
calculate a solution ŝG (verifying the behavior laws).
The specificity of a geometrical problem with contact Secondly, a global and linear step verifying the kine-
and friction is the change of the boundary conditions matic and the static admissibilities gives a solution
during the loading. Using a classical incremental al- sn + 1 Ad. The solution of the problem is se  Ad SG.
gorithm, leads to explicitly determine the boundary Therefore, the two steps of the algorithm are
conditions at each increment. To use the large time
increment method it is not possible to conserve in the Local step: From sn = (F: n, 7; n, P: n, q; n )Ad find
global resolution such boundary conditions that are ŝ= (F.: , 7̂; , P.: , q̂; )G verifying on [0, T]:
non linear. To extend the algorithm to contact prob-
lems, the strategy consists of releasing these boundary
conditions and treating these equations as local consti- F:. = F: n in V0 and q̂; −q; n = − k(7̂; − 7; n ) on (3V0 (24)
tutive relations. Two variables are defined: 7 the dis-
placement vector on (3V0 and q the loading vector on
(3V0. With this strategy, the contact and friction are F:. t = A(P. t, t5 t) in V0 R(q̂; , 7̂; ) on (3V0 (25)
treated implicitly. Therefore, a geometrical non linear
problem with contact and friction can be written on the Global step: From ŝG find sn + 1 = (F: n + 1, 7; n + 1,
following form: P: n + 1, q; n + 1)Ad better than sn verifying on [0, T]:
596 E. Bellenger, P. Bussy / Journal of Materials Processing Technology 80–81 (1998) 591–596

du; n + 1 5. Numerical examples


×u; n + 1/F: n + 1 = , u; n + 1 ( 1V0 =u; d
dM0

Öu* U
& tr([P: n + 1 − P: n ]TF*) dV0
The elasto-plastic damage model and the viscoplastic
damage model in large transformations presented
above are implemented in our finite element code OP-
&
V0
TIFIA. The simulations presented in planar calculation
+ (q; n + 1 − q; n )7* d(3V0 =0 (26) use the isoparametric triangle with six nodes and three
( 3V0 Gauss points. The first example is an illustration of the
viscoplastic visco damage model in non linear geometry
P:. −P: n + 1 =Ce(F:. − F: n + 1) in V0 (see Section 2.2.2) with R(p)= Q(1− e − bp) and
(27)
q̂; − q; n + 1 = k(7̂; −7; n + 1) on (3V0 Z(X)= QX (1−e − bX X). This first application concerns
a square sheet subjected to creep condition at high
k is a positive parameter matrix which has the dimen- temperature for IN100 in 1000°C. Fig. 2 shows strain–
sion of stiffness. Search directions defined by Eqs. (24) time behavior and the comparison with experimental
and (27) characterize the large time increment method results. The second example is a test of the simulation
version. of sheet cutting with the elasto-plastic damage model in
In order not to solve the global problem (Eq. (26)) large transformations (see Section 2.2.1) and unilateral
for each time, this problem is decomposed into two contact conditions with R(p)=R0 + kp n and Z(X)=
problems. A first space problem, which depends only ln(1+ X)/A. Fig. 3 describes the mesh and the
on the space variable M0 and a second time problem boundary conditions. Fig. 4 shows the isostress.
which depends only time. Therefore, the solutions are
described by the following forms.
References
p
u; n + 1(t, M0)− u; n (t, M0) = % gi (t)wi (M0) [1] K. Benkrid, P. Bussy, A. Abdali, J. Mater. Proc. Technol. 45
i=1
(1994) 521 – 526.
p [2] P. Ladevèze, Sur une famille d’algorithmes en mécanique des
F: n + 1(t, M0)− F: n (t, M0) = % gi (t)ai (M0) structures, C.R., Acad. des Sci., 300, Serie II, No. 2, 1985.
i=1 [3] P. Bussy, P. Rougée, P. Vauchez, The large time increment method
for numerical simulation of metal forming processes, Proc. NU-
p META, Swansea, 1990.
7; n + 1(t, M0)−7; n (t, M0) = % gi (t)gi (M0) [4] P. Bussy, B. Liu, P. Vauchez, New algorithm for numerical
i=1 simulation of metal forming processes, Proc. NUMIFORM 92,
Sophia-Antipolis, 1992.
p
[5] A. Abdali, K. Benkrid, P. Bussy, J. Mater. Proc. Technol. 60
P: n + 1(t, M0)− P: n (t, M0) = % hi (t)bi (M0) (1996) 255 – 260.
i=1
[6] J. Lemaitre, J.L. Chaboche, J. Mech. Appl. 2 (3) (1978) 317–365.
p [7] P. Ladevèze, Sur la théorie de la plasticité en grandes déforma-
q; n + 1(t, M0)− q; n (t, M0) = % hi (t)di (M0) (28) tions, Rapport Interne No. 9 LMT, Cachan, 1981.
i=1 [8] B. Halphen, Nguyen Quoc Son, Sur les matériaux standards
généralisés, J. Mécanique, 14 (1975).
with ai (M0)=dwi /dM0 and gi (M0) =wi (M0) ( 3V0. gi (t) [9] J. Lemaitre, J.L. Chaboche, Mécanique des matériaux solides,
and hi (t) are the scalar time functions defined on [0, T]. DUNOD, 1985.
[10] J.C. Simo, R.L. Taylor, Comput. Methods Appl. Mech. 48 (1985)
wi (M0), ai (M0), bi (M0) and di (M0) depend only on the 101 – 108.
space variable M0. We find in [3,4,11] all the details to [11] P. Boisse, P. Bussy, P. Ladevèze, Int. J. Numer. Methods Eng. 29
obtain these unknowns. (1990) 647 – 663.

You might also like