Abaqus

A plasticity model for metal
powder forming processes

Roland W. Lewis
a,
*, Amir R. Khoei
b
a
Mechanical Engineering Department, University of Wales Swansea, Singleton Park,
Swansea SA2 8PP, UK
b
Civil Engineering Department, Sharif University of technology, PO Box 11365-9313, Tehran, Iran
Received in nal revised form 14 June 2000
Abstract
In this paper, a double-surface plasticity model, based on a combination of a convex yield
surface consisting of a failure envelope, such as a MohrCoulomb yield surface and, a hard-
ening cap model, is developed for the nonlinear behaviour of powder materials in the concept
of a generalized plasticity formulation for the description of cyclic loading. This model reects
the yielding, frictional and densication characteristics of powder along with strain and geo-
metrical hardening which occur during the compaction process. The solution yields details on
the powder displacement from which it is possible to establish the stress state in the powder
and the densication is derived from consideration of the elemental volumetric strain. A
hardening rule is used to dene the dependence of the yield surface on the degree of plastic
straining. Finally, an adaptive nite element model (FEM) analysis is employed by the upda-
ted Lagrangian formulation to simulate the compaction of a set of complex powder forming
processes. # 2001 Elsevier Science Ltd. All rights reserved.
Keywords: A. Powder compaction; B. Elasticplastic material; B. Constitutive behaviour; C. Numerical
algorithms; C. Finite elements
1. Introduction
The cold pressing process is of great importance in the manufacturing of hard
metal components from powders. If, in this process, inhomogeneities are introduced
or if fracture occurs, the component will be rejected or it will perform badly in its
intended use. Thus, it is of great interest to be able to accurately predict the
mechanical behaviour of the powder during a compaction process.
International Journal of Plasticity 17 (2001) 16591692
www.elsevier.com/locate/ijplas
0749-6419/01/$ - see front matter # 2001 Elsevier Science Ltd. All rights reserved.
PI I : S0749- 6419( 00) 00096- 6
* Corresponding author.
The constitutive modelling of powder is clearly a keystone of successful quantita-
tive solution possibilities. Without a reasonable constitutive model, which can
reproduce powder behaviour under loading conditions, the computations are
worthless. However, a good constitutive model, without a computational framework
in which to use it, is only an academic exercise. In order to adequately reproduce
powder material behaviour under cyclic loading, as a sequence of loading/unload-
ing/reloading, modications have to be introduced to the classical constitutive
models. The success for an elasto-plastic solution to a problem is dependent on:
. a reasonable elasto-plastic constitutive model which can reproduce powder
material behaviour under complicated loading conditions;
. an accurate and stable integration algorithm for the elasto-plastic constitutive
relation.
In recent years, two approaches, based on `porous material' and `granular mate-
rial' models respectively, have been developed which describe the eect of stress
state on the response of the powder material. The porous material model, generally
known as a modied von Mises, has been used for the simulation of metal forming
and powder forming processes (Corapcioglu and Uz, 1978; Sluzalec, 1989; Hisat-
sune et al., 1991). This model includes the inuence of the hydrostatic stress com-
ponent and satises the symmetry and convexity conditions required for the
development of a plasticity theory. The granular material model which has been used
for the modelling of frictional materials such as soil or rock, is adopted to describe
the behaviour of metal powder (Crawford and Lindskog, 1983; Haggblad, 1991;
Riero and Prado, 1994). This model reects the yielding, frictional and densication
characteristics of powder along with strain and geometrical hardening which occur
during the compaction process.
The yielding of porous materials is more complicated than that of fully dense
materials because the onset of yielding is inuenced not only by the deviatoric stress
component but also by the hydrostatic stress. It is for this reason that a von Mises
yield function cannot be used for the development of a plasticity theory for porous
materials. Therefore, a yield function for porous materials, which can be considered
as an extension of von Mises's concept of yielding of fully dense materials, has been
considered by many researchers as (Kuhn and Downey, 1971; Green, 1972; Shima
and Oyane, 1976; Doraivelu et al., 1984)
AJ
/
2
BJ
2
1
= oY
2
0
= Y
2
,
(1)
where J
1
is the rst invariant of the stress tensor, J
/
2
is the second invariant of the
stress deviator and Y
0
and Y
,
are the yield stress of the solid and aggregate, or
partially dense, material having relative density ,, respectively. The parameters A,
B and o are functions of relative density. A typical yield surface for a porous
material with a relative density ,, expressed by the above equation, has the form of
an ellipsoid whose major axis coincides with the o
m
axis, and is shown in Fig. 1.
Eq. (1) represents a prolate spheroid in principal stress space which is a smooth,
convex, bounded surface of a very simple form. The analogy with the comparable
1660 R.W. Lewis, A.R. Khoei / International Journal of Plasticity 17 (2001) 16591692
arguments for the use of the von Mises yield criterion for non-porous materials is
very compelling.
It has been shown that the yield functions proposed by the various researchers
satisfy the required conditions and reduce, as expected, to the von Mises yield
function for fully dense materials (,=1). However, these functions do not predict
the dependence of compressive yield stress on relative density, as described by the
large discrepancies between experiment and theory given in Doraivelu et al. (1984).
Most of the case studies reporting compaction start at a relative density of about
0.7 however, for the powder compaction process, the relative density of the loose
powder ll is about 0.250.40. Furthermore, this type of material model neglects the
hardening factor associated with the densication process where this is one of the
main features in the process. Thus, the use of a yield criterion for a porous material
is not suitable for loose metal powder and the adoption of a model which represents
a frictional granular material is more applicable.
The purpose of the present paper is to develop a plasticity model for metal powder
forming processes, based on a granular material model. A double-surface plasticity
model is presented, based on a combination of a convex yield surface consisting of a
failure envelope, such as a MohrCoulomb yield surface and a hardening elliptical
cap for the nonlinear behaviour of powder materials. The model comprise two sur-
faces, one to reect shear failure and the second to capture densication. The solu-
tion yields details on the powder displacement from which it is possible to establish
the stress state in the powder and the densication is derived from consideration of
the elemental volumetric strain. A hardening rule is used to dene the dependence of
the yield surface on the degree of plastic straining. A generalized plasticity for-
mulation is introduced to describe powder material behaviour under cyclic loading.
A numerical integration algorithm is also presented for stressstrain relation.
For a successful modelling of the highly non-linear behaviour of the powder
materials, an adaptive analysis of metal powder forming process is simulated by the
updated Lagrangian nite element formulation. Finally, the powder behaviour during
the compaction of a set of complex powder forming processes are analysed numerically.
The predictive compaction forces at dierent displacements are compared with the
available experimental details reported in the literature. The simulation of the
Fig. 1. A typical yield surface for porous material.
R.W. Lewis, A.R. Khoei / International Journal of Plasticity 17 (2001) 16591692 1661
deformation is shown as well as the distribution of relative density and stress con-
tours at dierent time stages. It is shown that the proposed plasticity model along
with the adaptive nite element model (FEM) approach is capable of simulating
metal powder compaction processes in an ecient and accurate manner.
2. Constitutive models for powder materials
In the analysis of powder compaction problems, the MohrCoulomb and
DruckerPrager criteria suer from two deciencies. Firstly, the assumption of an
associated ow rule leads to excessive dilatency and secondly it has been seen that
both models imply that the material can support an unlimited hydrostatic compres-
sion. These deciencies can be removed by use of the close-ended conical yield sur-
faces which assume that the yield surface comprises two distinct parts.
Apparently, Drucker et al. (1957) were the rst to suggest an elasto-plastic work-
hardening material model. They proposed that successive yield surfaces might
resemble extended DruckerPrager cones with convex end spherical caps. As the
material strain hardens, both the cone and the end cap expand, the position and size
of which depend upon the hydrostatic strain. In this model, there are two important
innovations. The rst is the introduction of the idea of a spherical cap tted to the
cone. The second is the use of current material density (specic volume or void ratio)
as the strain-hardening parameter to determine the successive loading surfaces. This
model demonstrates such phenomena as an increase in volume due to a decrease in
hydrostatic stress and hence a decrease in yield strength and an increase or decrease
in plastic volume due to yielding, the rate of which is a function of the state of
yielding.
A number of yield criteria have been developed and tested to describe the beha-
viour of granular material with consideration of yielding and densication (Roscoe
et al., 1963; Baltov and Sawczuk, 1965; Roscoe and Burland, 1968; DiMaggio and
Sandler, 1971; Desai, 1980; Desai and Siriwardane, 1980; Crawford and Lindskog,
1983; Desai and Hashmi, 1989; Trassoras et al., 1989; Haggblad, 1991; Haggblad
and Oldenburg, 1994; Sun and Kim, 1997). The most practical yield functions which
have been employed for powder materials are the critical state model, Cam-clay
model, modied Cam-clay model and cap model (Fig. 2).
There are certain dierences between the cap and Cam-clay models. In the cap
models, the portion of the cap above the xed yield, or failure surface (Fig. 2e) is
omitted and only the yield surfaces f
1
()and f
2
(, k) are considered. In the Cam-clay
models, the moving cap plays the main role in dening yielding, and the xed yield
surface is used essentially to dene the critical state. On the other hand, in the cap
model, both the xed and moving surfaces are used to dene the yielding process.
Furthermore, it should be noted that the most successful applications of the Cam-
clay, or modied Cam-clay model, have been to problems involving the loading of
geotechnical constructions on clay which exhibit high cohesion stresses and there-
fore the model is not appropriate for loose powder materials. For this reason, the
cap model is referred to as a Cam-clay model and has been used in the analysis of
nonlinear behaviour of powder forming processes (Crawford and Lindskog, 1983;
Trassoras et al., 1989; Haggblad and Oldenburg, 1994).
2.1. MCEC cap plasticity model
A generalized cap model, based on classical plasticity theory, originally proposed
by DiMaggio and Sandler (1971) and Sandler et al. (1976), allows the control of
dilatancy by means of a hardening cap that intersects a xed failure envelope in a
nonsmooth fashion. A commonly accepted interpretation of this model, given by
Fig. 2. Plasticity models proposed by researchers for granular materials.
Desai and Siriwardane (1984), Chen and Baladi (1985), Simo et al. (1988) and Hof-
stetter et al. (1993), assumes a one-to-one correspondence between the hardening of
the cap and the plastic volume change. However, if a one-to-one hardening law is pos-
tulated, softening response may occur when the stress point is located at the compres-
sive corner region. Motivated by this softening behaviour, Sandler and Rubin (1979)
proposed a modied hardening lawwhich prevents softening response. The generalized
cap models have also been expanded by Nelson and Baladi (1977), Nelson (1978)
and Sandler and Baron (1979) to include rate eects, and anisotropic behaviour within
the yield surface and visco-plastic behaviour during yielding. In the present study, a
constitutive model based on a combination of the MohrCoulomb and elliptical cap
(MCEC) model is developed to describe the nonlinear behaviour of powder materials.
The cap model is a plasticity model dened by a non-softening convex yield sur-
face and a plastic strain rate vector that is normal to the yield surface in stress space.
The yield surface of this elasto-plastic model has a moving cap, intersecting the
hydrostatic loading line, whose position is a function of plastic volumetric strain
(Fig. 3). The main features of the cap model include a failure surface and an ellip-
tical yield cap which closes the open space between the failure surface and the
hydrostatic axis. The yield cap expands in the stress space according to a specied
hardening rule. The functional forms for both the perfectly-plastic and the strain-
hardening portions may be quite general and would allow for the tting of a wide
range of material properties. The movement of the cap is controlled by the increase,
or decrease, of the plastic volumetric strain. Thus, strain-hardening can be reversed
and it is this mechanism that leads to an eective control of dilatancy, which can be
kept quite small (eectively zero) as required for many powders.
In Fig. 3, the two-invariant-rate-independent cap model is dened by a convex
yield surface, which consists of a failure envelope, such as MohrCoulomb yield
surface, F
1
() and an elliptical yield cap surface F
2
(,o
c
), where o
c
is the hardening
parameter being a function of plastic volumetric strain. The functional form for the
failure envelope is
F
1
( ) =
1
3
J
1
sin
J
/
2
_
cos
J
/
2
3
_
sinsin ccos = 0
loi o
m
> L o
c
( )
(2)
Fig. 3. A combination of the MohrCoulomb and elliptical cap models (MCEC).
where J
1
and J
/
2
are the rst invariants of the stress tensor and the second invar-
iant of the deviatoric stress tensor, respectively; o
m
is the mean stress, dened as
o
m
=
1
3
J
1
and is the dilatancy angle. Note that o
m
is chosen as being positive in
tension. The parameters c and are the cohesion and the internal friction angle of
the metal powder. These parameters can be easily determined by a series of standard
tests such as a direct shear or triaxial test.
It is implicitly understood that the yield cap function satises the relation
04F
2
. o
c
( )4F
1
( ) loi L o
c
( )5o
m
5 o
c
( ) (3)
where L(o
c
) and (o
c
) dene the rst stress invariant range of the cap (Fig. 3). Cer-
tain limitations must be placed on the functional form that dene the model. In
order to avoid work-softening in the model, the functions L(o
c
) and (o
c
) must be
continuous and monotonically increasing functions and
oF
2
oo
m
50 ano
oF
2
oo
c
- 0 (4)
The cap must extend from the o
m
-axis to a point on the failure envelope. The
functional form for the cap, which is elliptically shaped, is given by
F
2
. o
c
( ) = J
/
2
M
2
o
m
o
c
( )
2
o
c

c
lan
_ _
2
_ _
= 0
loi L o
c
( )5o
m
5 o
c
( )
(5)
where M is the slope of the MohrCoulomb failure surface in the space
J
/
2
.
_
o
m
_ _
and o
c
is the critical stress governed by the function of plastic volumetric strain,
which is based on a `hardening rule'. A hardening rule is used to dene the depen-
dence of the yield surface on the degree of plastic straining. The evolution of the
yield surface must respond to either `hardening' or `softening'. The `hardening' is
caused by the elimination of voids with an increase in particle resistance to plastic
ow and consequently the powder material itself becomes harder. Conversely,
`softening' is caused due to void expansion, the compacted powder may crack espe-
cially at corners or at other points of high stress concentration, which will cause a
reduction of the ultimate strength of the nished component.
The hardening behaviour of the material is reected in a movement of the densi-
cation line and its intersection with the shear envelop (Naylor et al., 1981) and was
applied by Gethin et al. (1995) for powder material. The hardening parameter L(o
c
)
is chosen so that
L o
c
( ) =- o
c
>=
o
c
il o
c
- 0
0 il o
c
50
_
(6)
and the point of intersection of the cap with the o
m
-axis is dened as
o
c
( ) = 2o
c

c
lan
(7)
where o
c
is dened in term of c
p
v
as
o
c
= f c
p
v
_ _
= o
c0
ex
c
p
v
_ _
(8)
where o
c0
is the initial hydrostatic stress of compaction which is derived from a
consideration of die lling and assumed to be uniform, and is the plastic hard-
ening coecient which is derived from an uniaxial compression/relaxation test. c
p
v
denotes the plastic volumetric strain which, in the present case of an associative ow
rule, takes the form
c
.
p
v
= l
.
oF
oo
ij
o
ij
(9)
where l denotes a proportionality constant termed the plastic multiplier. The eec-
tive plastic strain c
p
may be dened in a rate form as
c
.
p
v
2
3
c
.
p
ij
c
.
p
ij
_
(10)
where e
ij
are the deviatoric strains, e.g. e
ij
=c
ij
1
3
o
ij
.
c
kk
. In Eq. (8), c
p
v
is the eective
volumetric plastic strain, given by the dierential relation
c
.
p
v
=
c
.
p
v
il c
.
p
v
40. oi o
c
- o
m
ano o
c
- 0
0 olleiwise
_
(11)
The mathematical condition o
c
- o
m
when c
.
p
v
> 0 in the rst line of Eq. (11) is a
consequence of the stability requirement that the cap be prevented from acting as a
softening yield surface and the model behaves as in perfect plasticity. If dilatancy
occurs when o
m
> 0, Eq. (11) limits the shrinking of the cap to o
c
=L(o
c
)=0 and
this condition ensures that the cap remains nite.
The models can handle a wider range of powder material characteristics from
loose powders, with ,=0.2, to dense material, with ,=1.0, where , denotes the
relative density. It should be mentioned that the parameters c and change as the
powder is compressed, approaching the dense (solid) material where =0 and the
MohrCoulomb function approaches the von Mises form.
In this model, the diculty of yielding at stresses below the failure envelope can be
avoided if it is recognized that yielding may occur before the material reaches a
failure envelope. Fig. 4 shows a family of yield criteria forming `caps' to the open-
ended MohrCoulomb envelope. As the stress exceeds the yield stress value at point
A and densication occurs, the plastic strain vector, normal to the original ellipse,
produces a plastic volumetric decrease which causes the material to harden and the
ellipse expands until eventually the point B is reached. The elliptical yield surface is
strain dependent and expands when the material hardens, which is associated with a
decrease in volume. Conversely, this contracts when the material softens and is
associated with an increase in volume. As the loading increases to point C, a crack,
or failure, occurs since softening takes place due to the increase in volumetric strain.
2.2. Generalized plasticity
The object of the mathematical theory of plasticity is to provide a theoretical
description of the relationship between increments of stress d and increments of
strain d as
o = Do (12)
where D is the elasto-plastic constitutive matrix depends on the current state of
stress and strain, its history, and importantly, the direction of the stress increment.
To account for this dependence in a simple way, two dierent matrices D
L
and D
U
may be postulated for stress increments corresponding to loading and unloading
situations (Nayak and Zienkiewicz, 1972; Zienkiewicz and Mroz, 1984; Pastor et al.,
1990; Auricchio and Taylor, 1995). A loading direction vector n is now introduced to
discriminate between loading and unloading by the sign of the projection d
T
n, i.e.
o
T
n > 0 loi loaoing
o
T
n - 0 loi unloaoing
and the situation of d
T
n=0 is dened as neutral loading. However, the above
denition fails if strain softening occurs. Therefore, a more convenient denition
was suggested by Zienkiewicz and Mroz (1984) to check the loading or unloading
direction by the elastic stress increment d
e
as
o
T
e
n > 0 loi loaoing
o
T
e
n - 0 loi unloaoing
o
T
e
n = 0 loi neulial loaoing
Fig. 4. Stress paths in strain hardening capped yield criterion.
As continuity of material behaviour under neutral loading should be achieved
with loading and unloading strain increments being identical, certain restrictions on
the loading and unloading stiness matrices must be enforced. Therefore, the most
general forms of D
1
I
and D
1
\
can be written as (Zienkiewicz and Mroz, 1984)
D
1
I
= D
1
e
n
gI
n
T
,H
I
(13)
D
1
\
= D
1
e
n
g\
n
T
,H
\
(14)
where H
L
and H
U
are the plastic hardening/softening modulus and n
gL
and n
gU
are
arbitrary unit stress vectors for loading and unloading directions. The last terms, in
Eqs. (13) and (14), denote the plastic matrix for loading and unloading where the
plastic characteristics of the powder are incorporated. It is easily veried that con-
tinuity is satised for neutral loading.
The strain increment d can be conveniently decomposed into elastic and plastic
parts as
o = o
e
o
(15)
with
o
e
= D
-1
e
o
o
= n
gI
n
T
o,H
I
loi loaoing
o
= n
g\
n
T
o,H
\
loi unloaoing (16)
and the elasto-plastic relations can be written as
o = D
I
o loi loaoing
o = D
\
o loi unloaoing (17)
in which the value of the matrices D
L
and D
U
can be obtained by direct inversion if
H
L/U
,= 0, but more generally can be written as
D
I
= D
e

D
e
n
gI
n
T
D
e
H
I
n
T
gI
D
e
n
(18)
D
\
= D
e

D
e
n
g\
n
T
D
e
H
\
n
T
g\
D
e
n
(19)
Obviously, a complete elasto-plastic behaviour can thus be specied by giving at
each point of the stress space; loading direction vector n, plastic ow direction vectors
n
gL
and n
gU
, plastic hardening/softening modulus H
L
and H
U
, and elastic matrix D
e
.
It is noted that if n
gL
and n
gU
are dierent fromn, which is the case commonly referred
to as non-associative plasticity, the above matrices will be non-symmetric.
In the theory of plasticity the behaviour of a material is specied by surfaces in the
space which enclose the elastic regions. The yield surface F(, k)=0 denes the
loading and unloading directions, giving the unit normal vector as
n =
oF,o
oF,o ( )
T
oF,o ( )
_ _
1,2
(20)
Similarly, the ow direction n
g
is dened by the plastic potential surface Q(,
k)=0 with
n
g
=
oQ,o
oQ,o ( )
T
oQ,o ( )
_ _
1,2
(21)
The hardening moduli H
L
and H
U
follow from the above denitions if we postu-
late that the stresses must be conned to the yield surface during `loading', i.e. that
oF =
oF
o
_ _
T
o
oF
ok
ok
o
_ _
T
o
(22)
Substituting d
p
=n
g
n
T
d/H in Eq. (22), we can obtain, after some manipulation,
(Zienkiewicz and Mroz, 1984)
H =
oF,ok ( ) ok,o ( )
T
n
g
oF,o ( )
T
oF,o ( )
_ _
1,2
(23)
It is clear now that classical plasticity is indeed a special case of generalized plas-
ticity. This simple and general description of generalized plasticity was introduced
by Zienkiewicz and Mroz (1984). It has been proved that the generalized plasticity
forms are extremely useful in describing the complex behaviour of powder material
during compaction process as the surfaces need not be dened.
In the metal powder forming process, especially in the case of loose powder, the
volume of powder reduces as compaction proceeds, so the volumetric strain is
negative and under this circumstance the associated ow rule is most appropriate
since the non-associated ow rule is used when the plastic strain is not in the same
direction as the yield function. Thus, for metals, it has always been assumed that the
yield function is identical to the potential function, F Q, i.e. an associated ow
rule applies since it is close to experimental observation (Bleich, 1972). Since the
ow rule is associated, the gradient oF,o is the same as oQ,o, i.e. it is normal to
the yield surface at any point in its stress space.
2.3. Integration of the constitutive relation
For elasto-plastic problems, the accuracy and stability of the global solution is
strongly aected by the accuracy and stability of the computation of the stress
increments, i.e. the integration of the stressstrain relation and the convergence of
nonlinear iteration. In this section, we shall discuss the main features of integration
algorithms and also the derivation of the tangent elasto-plastic matrix. A simple and
practical integration algorithm for the incremental constitutive relation is commonly
based on an explicit integration scheme within each increment. The simplest scheme
for such an explicit solution is obtained essentially by employing the elasto-plastic
matrix evaluated at the beginning of the incremental step, i.e.
n1
=
n
L
n
(24)
L
n
= D
n
. k
n
( )L
n
(25)
in which D(
n
, k
n
) is the elasto-plastic constitutive matrix evaluated at the beginning
of step n with stress
n
and parameter k
n
, which is the internal parameter for history
dependency. It should be noted that in each time step, several iterations are gen-
erally needed to obtain a converged solution for the nonlinear equation system.
Starting from the last accepted stress-strain state
n
and
n
at time station t
n
, the
incremental displacement ou
i
n
in ith iteration is rst found. Then, by using the
straindisplacement relation, the strain increment o
i
n
in this iteration is determined.
The total incremental strain in the time duration (t
n
, t
n+1
) can be written as
L
i
n
=
i
k=1
o
i
n
(26)
However, in a large deformation analysis the relationship between strain and dis-
placement is no longer linear and therefore Eq. (26) is invalid. In this case, the total
incremental strain in one time step has to been evaluated by using total increment
displacements. Corresponding to the strain increment D
i
n
, the stress increment D
i
n
is determined by an integration of the form
L
i
n
=
_
L
i
n
0
D
n
. k
n
( )o (27)
and the predicted stress for the next iteration is thus given by
i1
n1
=
n
L
i
n
(28)
In order to obtain an accurate stress integration of Eq. (27), a modied Euler
scheme with strain subincrementation is performed for a large strain increment. The
strain increment L
i
n
is rstly divided into M subincrementations of equal size and
then the stress is calculated as
i1
n1
=
n

M
j=1
L
nj
(29)
where
L
nj
= D
nj
. k
nj
_ _
L
i
n
,M (30)
in which
nj
=
n

j1
r=1
L
nr
(31)
and k
nj
is the updated internal parameter at stress
nj
.
An error estimator may be used as a guide for selecting the number of subdivi-
sions M (Khoei, 1998). There are even more sophisticated schemes for the stress
integration, though the computational cost needed may increase considerably. For
instance, instead of the uniform strain subdivisions, we may control the error in
every iteration by predicting the size of each strain subdivision, or instead of using
the Euler scheme, we may use other schemes of higher order such as a RungeKutta
scheme (Sloan, 1987).
3. Computational strategy
Metal powder forming processes often cause nonlinear stress behaviour in mate-
rials and also geometrically nonlinear elastic behaviour. Therefore, both material
and geometric nonlinear eects should be considered. The nonlinear stress beha-
viour of powder along with the constitutive equations and its numerical integration
have been discussed in Section 2. The aim of this section is to describe the treatment
of geometric nonlinear behaviour via a large displacement nite element formula-
tion. As the forming process in powder compaction problems is a large deformation
process, an elasto plastic analysis often utilises a large strain model. However, in
many forming processes where displacement are large, this is not sucient and must
be coupled with an updating of the calculation region mapped by the elements.
In the present study, the transient dynamic analysis of metal powder is simulated
by the nite element method based on a total and updated Lagrangian formulation.
A generalized Newmark scheme is used for the time domain discretization and then
the nal nonlinear equations are solved by a NewtonRaphson procedure. In order
to predict the behaviour of the powdertool friction, an interface element formula-
tion is developed to simulate sliding resistance at the powdertool interface. The
constitutive relations for friction are derived from a Coulomb friction law. Finally,
an adaptive FEM strategy is applied into the large displacement nite element for-
mulation by employing an error estimator and applying an adaptive mesh rene-
ment procedure which consist in dening a new appropriate nite element mesh
within the updated, deformed geometry and interpolating (mapping) the pertinent
variables from one mesh to another in order to continue the simulation.
3.1. Finite element formulation
In the motion of the body, three successive congurations can be considered; the
initial (unstressed, undeformed) conguration ,
0
, the current conguration , and the
incremented conguration ,. The Lagrangian description requires a known reference
conguration to formulate the equilibriumequations; the total Lagrangian description
(TLD) uses the initial unstressed conguration of the body ,
0
and the updated
Lagrangian description (ULD) uses the current deformed conguration , (incidentally,
it may be mentioned that an Eulerian approach would use , as its reference congura-
tion). However, any intermediate conguration may be used as a Lagrangian refer-
ence. Cescotto et al. (1979) presented a generalized Lagrangian description (GLD)
where the equilibrium equations are referred to an arbitrary known deformed cong-
uration ,
R
. In the two limit cases where ,
R
coincides with ,
0
and ,, the TLD or ULD,
respectively, will be recovered by the generalized Lagrangian description (GLD).
Whether the displacements, or strains, are large or small it is imperative that the
equilibrium conditions between the internal and external forces have to be satised.
Thus, the dynamic equilibrium equation of a body in a deformed conguration ,
can be written in a standard form as
o
ij.j
,b
i
,u
i
= 0 (32)
where u
i
is the current (deformed) displacement, b
i
is the body force accelaration, ,
is the current density of powder and o
ij
is the total (Cauchy) stress. A general de-
nition of strains, which is valid whether the displacements or strains are large or
small, was introduced by Green and St. Venant. Based on Green's strain tensor, the
non-linear strain displacement relationship can be dened in terms of the innitesi-
mal and large displacement components as
=
I

I
=
I

1
2
A
(33)
where
L
and
NL
are the linear and nonlinear strains, is the displacement gradiant
and A
is a suitably dened matrix operator which contains displacement derivatives

and can be dened from the denition of total strain (Lewis and Khoei, 1998).
Applying the virtual work principle and then a standard nite element Galerkin
discretization process to Eq. (32) with the independent approximations of u dened
as u = N
u
u, we will arrive at
_
O
B
T
oOMu
f
u
= 0 (34)
where the mass matrix, M, and the load vector, f
u
, are dened as
M =
_
O
N
T
u
,N
u
oO ano f
u
=
_
O
N
T
u
,boO
_
I
t
N
T
u
t
oT (35)
and B
is the well known strain matrix relating the increments of strain and dis-
placement (i.e. d=B
du). The bar sux is added as, if displacements are large, the
strains depend non-linearly on displacements and the matrix B
is dependent on u.
Thus, we can write B
in terms of the linear and nonlinear strain-displacement

matrix, from Eq. (33), as
B
= B
I
B
I
u ( )
= B
I
A
G
(36)
where B
L
is the same matrix as in a linear innitesimal strain analysis and only B
NL
depends on the displacement. In Eq. (36), G contains the Cartesian shape function
derivatives = Gu ( ).
The denition of the total tangential stiness matrix K
T
, used in the rst term of
Eq. (34), is
K
T
= K
I
K
I
K
o
=
_
O
B
T
D
T
B
oOK
o
(37)
where the rst term is the tangential stiness matrix evaluated in the current con-
guration and the second term K
o
is the geometric matrix (Khoei, 1998). In Eq. (37),
K
L
represents the usual, small displacement stiness matrix and the matrix K
NL
is
due to the large displacements. The consistent tangential stiness matrix D
T
is a
function of the displacement and can be obtained by performing a full dierentia-
tion on the internal force term.
In order to complete the numerical solution, the dierential Eq. (34) is integrated
in time by the generalized Newmark scheme (GNpj method), an unconditionally
stable direct solution procedure. Applying the GN22 scheme for displacement and
then the NewtonRaphson procedure for nonlinear iteration, the matrix of the
solution can be given by the following expression (Khoei, 1998)
M
n1
K
Tn1
[Lt
2
_ _
i
o u
n1
_ _ _ _
i
= G
u
{ ]
i
n1
(38)
where G
u
stands for the vector evaluated from the known initial values. The para-
meter [ is chosen in the range of 01 (Zienkiewicz and Taylor, 1989).
3.2. Interface element formulation
In the powder forming processes, the required shape changes are obtained by
pressing the material between tools which are much stier than the shaped material.
Since there is a large dierence between the deformability of the tool and the mate-
rial, relative movement occurs in the contact area, the magnitude of which will
depend on the geometry and frictional conditions at the interface. The implementa-
tion of these frictional conditions into the compaction nite element model plays a
crucial role in the estimation of the density distribution. A non-homogeneous density
distribution induces cracks and residual stresses during compaction and sintering
which is detrimental to the strength of the component. As the shape of a component
gets more complex the consequences of friction get more dicult to foresee. Thus, it
is important to predict the behaviour of the powdertool friction.
There are important links between the nite element contact problem and mathe-
matical programming techniques. Indeed, because the contact problems usually
involve inequality constraints (varying contact areas), the mathematics can be rela-
ted to the method of variational inequalities. Basically, two main constraint methods
of solution have been employed in the nite element solution of contact problems;
the method of `Lagrangian multipliers' and the `penalty approach'.
The method of Lagrangian multipliers has been applied to friction often with the
aid of a plasticity technique which follows from the observation of the close links
between friction and plasticity made by Michalowski and Mroz (1978). In this
method, contact forces are taken as primary unknowns and the non-penetration
condition is enforced. The algorithm takes a non-associative ow rule so that the
tangent stiness matrix is non-symmetric. The Lagrange multiplier method does
expand the original global stiness matrix but does not aect the original global
tangent stiness matrix if small displacement are assumed, or if a quasi-Newton
method is used under large displacements and large rotation.
The penalty approach is closely related to techniques which attempt to introduce a
genuine (although high and possibly non-linear) stiness for the contact region
(Curnier, 1984; Wriggers et al., 1985; Curnier and Alart, 1988; Peric and Owen,
1992). In this method, penetration between two contacting boundaries is introduced
and the normal contact force is related to the penetration by a penalty parameter.
One alternative procedure involves the elimination of degrees of freedom and can be
considered as a form of non-linear master slave approach (Jelenic and Saje, 1995).
Although the penalty method does not introduce any new unknown it is not neces-
sarily more ecient than the Lagrange multiplier method, which does introduce new
unknowns. This is because the penalty method will change the prole and skyline of
the global tangent stiness matrix which will greatly inuence the solution time.
3.2.1. Continuum model of friction
A numerical model of the frictional contact based on penalty approach is used to
simulate sliding resistance at the tool-workpiece interface (Rodic and Owen, 1989;
Peric and Owen, 1992). Consider two bodies, a master (target) and a slave body,
with initial congurations denoted by O
S
and O
T
. The relative displacement from
the point x on the contact surface of the slave body to the point y on the contact
surface of the target may be dened as
g
x ( ) = x ( ) y ( )
_ _
.N on o O
c
( ) = o O
S
_ _
o O
T
_ _
(39)
where g
N
(x) is the gap between the two bodies, (x) and (y) are the conguration
mapping of the slave and target bodies, o O
S
_ _
and o O
T
_ _
are the slave and target
body surfaces respectively, o O
c
( ) denotes a surface where contact between the two
bodies occurs and N is the unit outward normal vector on the target surface. During
the contact and sliding of the bodies, we dene p
N
and p
T
as the normal and tan-
gential load acting on the point x, respectively. The contact conditions may be
expressed in the standard KuhnTucker form as
g
50 p
40 p
.g
= 0 (40)
which is best suited for a variational formulation. Since there is no gap between the
two bodies in the sliding contact problems of powder forming processes, the normal
displacement u
N
is assumed to be zero and a tangential displacement u
T
is only con-
sidered, which consists of stick (or adhesion) and slip decompositions, u
A
T
and u
S
T
, and
is in principle the same as the decomposition of elastic and plastic behaviour, i.e.
u
T
= u
A
T
u
S
T
= u
e
T
u
T
(41)
The normal load p
N
can be obtained from multiplication of the penalty factor k
N
and the displacement in the normal direction u
N
. Similarly, the stick (or elastic)
component of the tangential load may be obtained by multiplying the penalty factor
k
T
and elastic part of the displacement in the tangential direction u
A
T
. The penalty
factors k
N
and k
T
can be considered as being the normal stiness constant and shear
stiness constant, respectively. Constitutive laws for the contact loads can now be
summarized as
p
= D
u
e
(42)
p
T
= D
T
u
e
(43)
where D
N
and D
T
are the normal and tangential parts of the elastic modulus tensor
for friction. The plastic part of the tangential displacement u
T
, in Eq. (41), can be
dened by the direction of an appropriate slip rule which is derived from the gra-
dient of a convex potential Z, as
u
.
T
= l
.
oZ
op
T
= l
.
T (44)
where T =
p
T
p
T
is the unit tangential vector.

3.2.2. Numerical modelling of friction
Evaluation of the stiness matrices for the interface elements requires linearization
of the governing equations for the frictional contact problem. Strict mathematical
linearization results in non-symmetric constitutive matrices D
f
, due to the non-
associated slip rules employed. To preserve the symmetry of the numerical for-
mulation the o-diagonal term in D
f(2 2)
, which represents the coupling between
the normal and tangential stresses at the interface, is neglected, so that its eect is
brought into the formulation via residual `pseudo loads'. In this way the problem is
articially decomposed into a pure contact in the normal direction and a frictional
resistance in the tangential direction, which are linearized separately. Thus, the stress
vector at the interface element is assumed to have only two components; normal stress
o
n
and shear stress t and the constitutive material matrix D
f
can be simply dened as
D
l
=
G
l
0
0 E
l
_ _
(45)
where E
f
is chosen as an arbitrary large number for numerical convenience and the
frictional non-linearity is modelled by an appropriate variation of G
f
which can be
obtained from the shear stressstrain (t ,) relationship for the stick and slip
region (Gethin et al., 1995). It must be noted that the shear modulus G
f
is divided
into two components, the rst where the compact has not moved but there is a rapid
buildup in load. Under this condition, a stick shear modulus G
f
=G
A
is derived
directly from the slope of the t, curve. On commencement of movement, the slip
shear modulus condition G
f
=G
B
is now appropriate and this is derived similarly. A
friction criterion is the indicator which determines whether tangential sticking or
slipping occurs. A type of slip criterion, which is based on Coulomb's law of friction
with a tension cut-o model, is employed for the constitutive modelling of the fric-
tional behaviour (Khoei and Lewis, 1998a).
Finally, the stiness matrix of an interface element is dened in the standard nite
element manner as
K
l
=
_
O
B
T
l
D
f
B
l
oO (46)
where the B
f
matrix is equal to
1
w
N with w denoting the width of the interface ele-
ment considered to be very small, i.e. w 0. It should be noted that t and o
n
are
decoupled in the stiness matrix K
f
. However, in the stress calculation o
n
is eval-
uated rst so that t can then be obtained from the (t,) relationship. Also, if a
positive value of o
n
is registered, then the tension cut-o condition is applied.
3.3. Adaptive FEM analysis
As metal powder forming analyses involve large deformations where distortion of
the mesh is inevitable if an updated Lagrangian coordinate scheme is used (Brekel-
mans et al., 1991; Lewis and Khoei, 1998), it is important to make use of the error
estimate and adaptive remeshing scheme to control the mesh and therefore to improve
the solution quality. This adaptive procedure can be applied as follows; an error and
mesh control component, remeshing of the deformed body, mapping of history
dependent variables from the old to the new mesh and a restart of the simulation.
The aim of error estimation in any analysis is rstly, to determine the error for the
chosen, perhaps distorted, mesh and secondly, to reduce this error to a permissible
value by automatic adaptive remeshing. Various methods for the estimation of spa-
tial discretization errors and adaptive renement have been proposed and used in
practice (Babuska et al., 1986). Basically, there are two types of discretization error
estimates; à-priori' estimates and à-posteriori' estimates. A-priori estimates are
based on a knowledge of the characteristics of the solution and provide qualitative
information about the asymptotic rate of convergence as the number of degrees of
freedom goes to innity. A-posteriori estimates employ the solution obtained by the
numerical analysis, in addition to some a-priori assumptions about the solution.
This method can provide quantitatively accurate measures of the discretization
error, while a-priori estimate method cannot.
An adaptive remeshing procedure for nite element solutions infers an improve-
ment in the quality of the solutions by enriching the approximation in some manner
so as to achieve the best solution for a given computational eort. There are three
methods used for renement of the nite element solutions; the `h-renement' which
is based on the knowledge that the smaller the size of the elements the more accurate
the solutions, the `p-renement' which increases the order of the polynomials used in
the shape functions in the approximation while keeping element size constant in
order to get more exact solutions, and the `hp renement' which combines the
previous two methods. As mentioned by Babuska et al. (1986) the rate of con-
vergence of the p-renement is twice that of h-renement in the presence of singu-
larities. However, for a large class of engineering problems, the hp renement has
an exponential rate of convergence, while the h- and p-renements have only a
polynomial rate of convergence. In forming analyses, the h-renement is often pre-
ferable because incompressibility consideration, changes in boundary location as
well as the occurrence of distorted elements require certain types of elements and
frequent remeshing.
It is clear that all adaptive processes aiming at a specied accuracy solution
involve three stages which have to be applied cyclically:
. obtaining a solution with a specied renement and estimating the error;
. predicting the renement required for achieving a specied accuracy (specied
error) most economically;
. implementing this renement and obtaining a new solution and estimating the
new error, followed by a return to the second stage if this does not meet the
requirements.
In this study, a simple yet reliable a-posteriori error estimator and an adaptive
procedure developed by Zienkiewicz and Zhu (1987, 1991), which is not only rea-
sonably accurate but whose evaluation is computationally simple and can be readily
implemented in existing nite element codes, are applied to the analysis of static and
dynamic behaviour of metal powder forming processes. For this purpose an error
estimation using an L
2
norm of strain by a recovery procedure in which the improved
value of nite element solution is the basis of an adaptive procedure, is proposed and
an h-renement adaptive procedure using unstructured meshes is also employed for
the computation of the new element size (Khoei and Lewis, 1998b, 1999).
3.3.1. Error estimation
In the present nite element Galerkin discretization formulation, C
0
continuity is
assumed for the displacement interpolation function N
u
. For variables such as the
total strain , dened as o = Bou, then the nite element approximation is dis-
continuous and diers from the exact value . The discretization error represents the
dierence between the exact and the nite element solutions for the mathematical
model. Since exact solutions for practical, nonlinear problems are usually not
available, an estimated error, instead of the real error, is adopted to guide the nite
element analysis. Thus, the error in strain may be dened as
e
c
-
+
(47)
where * represent an improved, or recovered, solution. A pointwise denition of
errors, as given in Eq. (47) is generally not convenient and occasionally misleading.
For instance, under a point load the errors in strains will be locally innite but the
overall solution may well be acceptable. Similar situations will exist near re-entrant
corners where, as is well known, stress singularities exist in elastic analyses and
gradient singularities develop in eld problems. For this reason various `norms'
representing some integral scalar quantity are often introduced to measure the error,
or indeed the function itself.
The error estimates which we shall use here will be presented in the L
2
norm of
strain. For instance, the L
2
norm of the strain error e
c
can be written as
e
c
| | =
+

_
_
_
_
=
_
O
+
( )
T
+
( )oO
_ _1
2
(48)
in which the improved solution * can be obtained by a global smoothing using the
interpolation function N
u
and nodal parameter * as
+
= N
u
+
(49)
where
+
=
_
O
N
T
u
N
u
oO
_ _
1
_
O
N
T
u
ôO (50)
and we can obtain the square of each L
2
norm dened in Eq. (48) by summing ele-
ment contributions as
e
c
| |
2
=
m
i=1
e
c
| |
2
i
(51)
where i represents an element contribution and m is the total number of elements.
Indeed for an optimal mesh, we generally try to make the contributions to this
square of the norm equal for all elements.
3.3.2. Adaptive mesh renement
The adaptive mesh renement strategy depends on the nature of the criteria on
accuracy which we wish to satisfy. A very common requirement is to specify the
achievement of a certain minimum percentage error in the L
2
norm. Thus, we
require that after remeshing each element will have the same error and the overall
percentage error is equal to some target percentage error, i.e.
j =
e
c
| |
_
_
_
_
4
e
c
| |
aim
_
_
_
_
= j
aim
(52)
where j
aim
is the prescribed percentage error. If we assume that the error is equally
distributed between elements the above requirements can be translated into our
placing a limit on the error in each element. Thus, for each element we require that
e
c
| |
i
( )
aim
=
1
m
_
_
_
_
_
j
aim
(53)
and the new element size can be predicted by the following equation
h
i
( )
new
h
i
( )
olo
=
e
c
| |
i
( )
aim
e
c
| |
i
( )
olo
_ _1
l
(54)
In this study, l is assumed to be 2 when quadratic triangular elements are used
and 1 when a linear triangular element is employed.
The complete mesh regeneration procedure, according to the rule of Eq. (54), is
probably the most ecient method of allowing mesh `derenement' where originally
too many elements were used. It must be coupled with an ecient mesh generator
which allows the new mesh to be constructed according to a predetermined size
distribution. Such a generator was developed for triangular elements by Peraire et
al. (1987) and is widely used in this research.
Once a new mesh is generated, state variables and history-dependent variables
need to be mapped from the old nite element mesh to the new one. The state vari-
ables consist of the nodal displacements and the variables of the contact algorithm.
The history-dependent variables consist of the Cauchy stress tensor, the strain ten-
sor, the plastic strain tensor and a vector of internal variables. A superconvergent
patch recovery (SPR) method developed by Zienkiewicz and Zhu (1992a,b) is
employed to project the values from the old Gauss points to the old nodal points
(Khoei and Lewis, 1999). In this case, we rst obtain the smoothed nodal point
values for each of the history-dependent variables, then use the inverse mapping
technique to transfer these variables to the nodal points of the new nite element
mesh and nally the corresponding values at the new Gauss points can be calculated
using interpolation functions in the new nite element mesh.
4. Numerical simulation results
In order to illustrate the applicability of the present formulation, the powder
behaviour during the compaction of a set of complex forming processes are analysed
numerically. The problems have been solved with displacement control by increasing
the punch movement and predicting the compaction forces at dierent displace-
ments. An incremental manner is also used to simulate the dynamic analysis of a
compaction process subjected to a surface step loading. The iron powder properties
chosen in the analysis are typical for industrial application (Gethin et al., 1995) and
for all examples are E=40.0 N/mm
2
, n=0.35, c=5.0 N/mm
2
and =30
. The die
wall friction is simulated using a set of interface elements with the Coulomb friction
coecient j=0.08. The initial relative density is ,
0
= 0.4. The plastic hardening
coecient and the initial hydrostatic stress of compaction o
c0
are assumed to have
values of 0.16 and 0.2 N/mm
2
, respectively.
4.1. A plain bush component
The rst example is chosen to show that the updated Lagrangian description (ULD),
using a combination of the MohrCoulomb and elliptical cap (MCEC) model, can give
good answers in simulating the metal powder during compaction. The nite element
modelling of a plain bush component is performed employing an axisymmetric repre-
sentation which is illustrated in Fig. 5(a). Since a displacement based formulation is
used, the implementation of loading in the nite element code is achieved by the use
of prescribed nodal displacements. Fixed nodal values are employed to represent the
xed bottom punch. These infer an assumption that there is no relative movement
between the punch and the powder. Where there is a relative movement, the inter-
face element is used which in this case is along the sleeve (die) and core rod surface.
Fig. 5. An axisymmetric plain bush component: (a) nite element modelling, geometry and boundary
conditions; (b) predicted compaction forces at dierent displacements.
In order to compare the performance of the total and updated Lagrangian for-
mulation for both the MohrCoulomb (MC) and combination of MohrCoulomb
and elliptical cap (MCEC) model, the predictive compaction forces at dierent dis-
placements are shown in Fig. 5(b) using the triangular quadratic continuous dis-
placement elements (T6C) with the six-noded quadratic interface elements (I6C).
Also plotted in this gure are the experimental results and numerical solutions cor-
responding to a plain bush component given by Tran et al. (1993) and Gethin et al.
(1994). Obviously the MohrCoulomb constitutive model fails to predict the com-
paction forces however, the results can be improved signicantly when a combi-
nation of the MohrCoulomb and elliptical cap model is adopted.
4.2. A rotational anged component
The second compaction example is of a multi-level component which is modelled
by an axisymmetric representation as illustrated in Fig. 6. The geometry and
boundary conditions of this rotational anged component are also shown in this
gure. Since a single top, or bottom, punch motion results in severe distortion of
nite element mesh which can cause large approximation errors or make the Jaco-
bian determinant negative (Lewis and Khoei, 1998), a double-ended compaction
process has been employed using a xed mesh analysis. In this case, when simulta-
neous top and bottom punch motions are applied, mesh distortion does not occur,
due to the powder being compacted by both moving punches. The loading char-
acteristic is achieved by the use of prescribed nodal displacements for the top and
bottom punches.
A comparison between the measured and predicted density prole, taken along
section identical to those in the experimental measurements (Lewis and Tran, 1994),
is presented in Fig. 7. The density prole is obtained at the nal stage of compaction
where the top and bottom punch movements are d
t
=6.06 mm and d
b
=7.70 mm.
The density prole shows good agreement with the experimental data.
Fig. 6. Finite element modelling, geometry and boundary conditions of a rotational anged component.
As mentioned before, the crucial numerical problem in this more practical geo-
metry is associated with the distorted mesh. For this reason, an adaptive FEM
analysis has been carried out during the complex movements of the top and bottom
punches. The whole compaction process is simulated by a top punch movement of 3.44
mm to compact the upper part of the section at the rst stage, then a bottom punch
movement of 7.70 mmto compact the lower part of the section at the second stage, and
nally a further top punch movement of 2.62 mm in the third stage of compaction.
The analysis commences with an uniform distribution of elements. The non-linear
solution is carried out for various loading steps (or prescribed displacements) until
the estimated error exceeds a prescribed value, or the sign of the Jacobian determi-
nant becomes negative. At this time a new mesh is generated using an appropriate
renement criterion. The rst adaptive mesh renement is carried out on the
deformed mesh obtained by the rst top punch motion for a specied target error
j
aim
= " 4.0 and a new mesh is generated. After changing the mesh, all state vari-
ables are transferred from the old mesh to the integration points of the new mesh by
using nodal points as the reference points for storing the information. For this pur-
pose, the history dependent values in the old mesh are rst transfered to the nodes
and then, the values at the new nodal points in the new mesh are computed by a
simple interpolation of the old nodal values using shape functions. The computation
is continued on the new mesh starting from the beginning of the step. A second
adaptive analysis is employed after the bottom punch motion takes place for a speci-
ed target error j
aim
= " 2.0. The analysis is carried out by transferring the infor-
mation from the old integration points to the new ones in the new mesh. Finally, a
further top punch movement is applied to obtain a fairly uniform density eld.
The variation of the top and bottom punch forces at dierent displacements are
plotted in Fig. 8. These results can be compared with the predicted compaction
forces for the double-ended compaction process given by authors (Lewis and Khoei,
Fig. 7. Comparison of the relative density prole by the present formulation and experimental results for
a anged component at level of y=21.4 mm.
1998). In Fig. 9, the density distribution of this anged component is shown at the
nal stage of bottom punch movement as 3D plots where the third direction repre-
sents the relative density. The results clearly indicate that the algorithm makes it
possible to simulate the powder forming problems eciently and automatically.
Fig. 8. Predicted top and bottom punches compaction forces at dierent displacements using an adaptive
analysis for a anged component.
Fig. 9. Density distribution at the nal stage of bottom punch movement for a anged component.
4.3. An axisymmetric shaped tip
The third compaction example is of an engineering tip component in hard metal
powder which includes several complications not present in the plain bush and
anged compaction. Some problems are caused by discontinuities in geometry, for
example ow around corners and multipunch set up, i.e. an upper and lower punch.
In order to illustrate a comparison between the density distribution obtained by the
present formulation and those of given by Haggblad and Oldenburg (1994), the
simulation has been performed using the remaining pressing distance of 3.0 mm
from above and an under-pressing of 1.45 mm.
The nite element modelling of a shaped tip component using the eight-noded
quadrilateral elements (Q8C) and six-noded quadratic interface elements (I6C) is
performed employing an axisymmetric representation which is illustrated in Fig. 10.
Also plotted in this gure are the geometry and boundary conditions of this com-
ponent. Since the crucial numerical problem in this complicated die geometry is
associated with the distorted mesh, a double-ended compaction process has been
used employing a xed mesh analysis. The numerical results of the dynamic analysis
of a shaped tip component, simultaneously subjected to a surface step loading at
both the top and bottompunches are shown in Fig. 11(a). In this gure, the variations
Fig. 10. Finite element modelling, geometry and boundary conditions of an axisymmetric shaped tip
component.
with time of the relative densities are plotted at dierent nodes of the shaped tip
component. The distribution of relative density on this tip component is presented in
Fig. 11(b) at the nal stage of compaction as 3D plots. It shows a good agreement
between the density distributions obtained by the present formulation and given by
Haggblad and Oldenburg (1994).
Fig. 11. A shaped tip component: (a) variation with time of the relative densities at dierent nodes; (b)
density distribution at the nal stage of compaction.
4.4. A shaped tablet
The next example is chosen to demonstrate the performance of the present for-
mulation in the tablet pressing processes. A shaped tablet component, with top and
bottom punch motion is numerically analysed using eight-noded quadrilateral ele-
ments (Q8C) and six-noded quadratic interface elements (I6C). The nite element
mesh, geometry and boundary conditions for this tablet are shown in Fig. 12. The
deformed mesh at the nal stage of compaction is also plotted in this gure, where
the simultaneous top and bottom punch motions are d
t
=d
b
=2.0 mm. The relative
density and normal stress, o
y
, contours at the nal stage of compaction are plotted
in Fig. 13. From the contours, it can be seen that the density in the right hand region
Fig. 12. An axisymmetric shaped tablet: (a) nite element modelling, geometry and boundary condition;
(b) deformed meshes at d
t
=d
b
=2.0 mm.
Fig. 13. A shaped tablet: (a) relative density contour, and (b) normal stress o
y
contour at the nal stage of
compaction.
of the axisymmetric tablet are greater than in other regions. However, the low den-
sity region, which appears between high density regions, can cause a crack or a
reduction of the ultimate strength of the nished compacted tablet.
4.5. A modelled multi-level component
The last example chosen demonstrates the performance of the present formulation
for the complicated die geometry of a multi-level component as shown in Fig. 14.
The simulation of an adaptive FEM analysis for this complex shape has been done
by a top punch movement of 5.0 mm, then a bottom punch movement of 8.0 mm,
and a second top punch movement of 3.0 mm as illustrated in Fig. 15.
Fig. 14. Shape, geometry and boundary conditions of the modelled multi-level component.
Fig. 15. Adaptive analysis for a modelled multi-level component: (a) initial mesh (163 elements); (b) rst
adaptive mesh after top punch motion, j
aim
= " 4.0 (385 elements); (c) second adaptive mesh after bot-
tom punch motion, j
aim
= " 2.0 (1001 elements).
The computations are started with an initial uniform mesh as given in Fig. 15(a).
The rst adaptive remeshing is carried out on the deformed mesh obtained for the
rst top punch motion for a specied target error of j
aim
= " 4.0 and followed by
transferring the information from the old integration points to the new ones in the
new mesh in Fig. 15(b). The analysis is continued with a second adaptive mesh
renement after the bottom punch motion takes place for a specied target error of
j
aim
= " 2.0 (Fig. 15c). After transferring all the state variables from the old mesh
to the integration points of the new mesh in Fig. 15(c), a second top punch move-
ment is applied to obtain the nal stages of compaction. In Fig. 16, the relative
density contour at nal stage of compaction are plotted. It can be seen that the
proposed adaptive nite element approach is capable of simulating metal powder
compaction processes in an ecient and accurate manner.
Fig. 16. Relative density contour at the nal stage of compaction for a multi-level component.
Fig. 17. Finite element modelling of a non-symmetric anged component.
5. Conclusion
In the present paper, a plasticity model for metal powder forming processes was
developed based on a granular material model. Adouble-surface plasticity model was
presented, based on a combination of a convex yield surface consisting of a failure
envelope, such as a MohrCoulomb yield surface and a hardening elliptical cap for
the nonlinear behaviour of powder materials. The model comprise two surfaces, one
to reect shear failure and the second to capture densication. The solution yields
details on the powder displacement from which it is possible to establish the stress
state in the powder and the densication was derived from consideration of the ele-
mental volumetric strain. A hardening rule was used to dene the dependence of the
yield surface on the degree of plastic straining. A generalized plasticity formulation
was introduced to describe powder material behaviour under cyclic loading and a
numerical integration algorithm was also presented for the stressstrain relation.
For a successful modelling of highly non-linear behaviour of powder materials, an
adaptive analysis of metal powder forming process was simulated by an updated
Lagrangian nite element formulation. The adaptive strategy employs an error esti-
mator and applies an adaptive mesh renement procedure which involves dening a
new appropriate nite element mesh within the updated, deformed geometry and
interpolating (mapping) the pertinent variables from one mesh to another in order
to continue the simulation. In order to predict the behaviour of the powdertool
friction, an interface element formulation was developed to simulate sliding resis-
tance at the powdertool interface. The constitutive relations for friction are derived
from a Coulomb friction law. Finally, the powder behaviour during the compaction
of a set of complex powder forming processes was analysed numerically. The pre-
dictive compaction forces at dierent displacements are compared with the available
experimental details reported in the literature. The simulation of the deformation is
shown as well as the distribution of relative density and stress contours at dierent
time stages. It is shown that the proposed plasticity model, along with the adaptive
FEM approach, was capable of simulating powder compaction processes in an e-
cient and accurate manner.
Having completed the 2D analysis of powder forming processes, it is worthwhile
developing an appropriate modelling technique for establishing the forming of 3D
parts and the subsequent modelling of complex 3D multi-level parts. Fig. 17 shows
the nite element modelling of a non-symmetric anged component as a typical
example illustrating everyday requirements from industry. In later work, a mod-
ication and extension of this 2D model will be presented as a 3D analysis of the
powder compaction process.
Apart from the issues mentioned in this paper, several important aspects of the
powder forming analysis which are related to nonlinear industrial applications need
further attention:
1. In any metal powder forming process, especially when forming from loose
powder, the volume of powder reduces as the compaction proceeds. In this cir-
cumstance, an associated ow rule is most appropriate since the non-associated
ow rule is used when the plastic strain is not in the same direction as the yield
function. The non-associated ow rule may be used in the case of an elastic
ideally-plastic material model when plastic deformation occurs with no further
volumetric strain. It appears that the non-associated ow rule should be con-
sidered, especially in the early stages of deformation.
2. Forming of non-metal powder material requires an understanding of the
deformation behaviour of hard particles, which is a subject of current interest
in industry. These materials exhibit rate-dependent characteristics for which an
elasto-viscoplastic material model would be more appropriate. The numerical
approach will require further development in order to capture this behaviour.
In summary, this plasticity model will provide adequate powder forming analyses
which are needed as crucial links within a decision support system for an important
class of problems of current industrial interest.
References
Auricchio, F., Taylor, R.L., 1995. Two material models for cyclic plasticity: nonlinear kinematic hard-
ening and generalized plasticity. Int. J. Plasticity 11, 6598.
Babuska, I., Zienkiewicz, O.C., Gago, J., Oliveira, E.R. de A., 1986. Accuracy Estimates and Adaptive
Renement in Finite Element Computations. Wiley.
Baltov, A., Sawczuk, A., 1965. A rule for anisotropic hardening. Acta Mech. 1 (2), 8192.
Bleich, H.H., 1972. On uniqueness in ideally elasto-plastic problems in case of non-associated ow rules.
J. Appl. Mech. Trans., ASME, 983987.
Brekelmans, W.A.M., Janssen, J.D., Van de Ven, A.A.F., With, G.de, 1991. An Eulerian approach for die
compaction processes. Int. J. Num. Meth. Eng. 31, 509524.
Cescotto, S., Frey, F., Fonder, G., 1979. Total and updated Lagrangian descriptions in nonlinear struc-
tural analysis: a unied approach. In: Glowinski, R., et al. (Eds.) Energy Methods in Finite Element
Analysis. Wiley, pp. 283296 (Chapter 15).
Chen, W.F., Baladi, G.Y., 1985. Soil Plasticity, Theory and Implementation. Elsevier, Amsterdam.
Corapcioglu, Y., Uz, T., 1978. Constitutive equation for plastic deformation of porous materials. Powder
Tech. 21, 269274.
Crawford, J.E., Lindskog, P., 1983. Constitutive equations and their role in the modelling of the cold
pressing process. Scand. J. Metall. 12, 277281.
Curnier, A., 1984. A theory of friction. Int. J. Solids Structures 20, 637647.
Curnier, A., Alart, P., 1988. Generalisation of Newton type methods to contact problems with friction. J.
de Mecanique Theorique et Appliquee 7, 6782.
Desai, C.S., 1980. A general basis for yield, failure and potential functions in plasticity. Int. J. Numer.
Anal. Methods Geomech. 4, 361375.
Desai, C.S., Hashmi, Q.S.E., 1989. Analysis, evaluation, and implementation of a nonassociative model
for geologic materials. Int. J. Plasticity 5, 397420.
Desai, C.S., Siriwardane, H.J., 1980. A concept of correction functions to account for non-associative
characteristics of geologic media. Int. J. Numer. Anal. Methods Geomech. 4, 377387.
Desai, C.S., Siriwardane, H.J., 1984. Constitutive Laws for Engineering Materials, with Emphasis on
Geologic Materials. Prentice Hall, NJ.
DiMaggio, F.L., Sandler, I.S., 1971. Material models for granular soils. J. Eng. Mech. Div., ASCE, 97, 935
950.
Doraivelu, S.M., Gegel, H.L., Gunasekera, J.S., Malas, J.C., Morgan, J.T., Thomas, J.F., 1984. A new
yield function for compressible P/M materials. Int. J. Mech. Sci. 26, 527535.
Drucker, D.C., Gibson, R.E., Henkel, D.J., 1957. Soil mechanics and work hardening theories of plasti-
city. Trans. ASCE 122, 338346.
Gethin, D.T., Tran, D.V., Lewis, R.W., Arin, A.K., 1994. An investigation of powder compaction
processes. Int. J. Powder Metall. 30, 385398.
Gethin, D.T., Lewis, R.W., Arin, A.K., 1995. Modelling compaction and ejection processes in the gen-
eration of green powder compacts. ASME Int. Mech. Eng. Congress and Exposition, San Francisco
ADM-216, 2745.
Green, R.J., 1972. A plasticity theory for porous solids. Int. J. Mech. Sci. 14, 215224.
Haggblad, H.-A., 1991. Constitutive models for powder material. Powder Technol. 76, 127136.
Haggblad, H.-A., Oldenburg, M., 1994. Modelling and simulation of metal powder die pressing with use
of explicit time integration. Model. Simul. Mater. Sci. Eng. 2, 893911.
Hisatsune, T., Tabata, T., Masaki, S., 1991. A yield criterion of porous material with anisotropy caused
by geometry or distribution of pores. J. Eng. Mater. Technol. Trans., ASME 113, 425429.
Hofstetter, G., Simo, J.C., Taylor, R.L., 1993. A modied cap model: closest point solution algorithms.
Int. J. Comput. Struct. 46, 203214.
Jelenic, G., Saje, M., 1995. A kinematically exact space nite strain beam modelnite element formula-
tion by generalized virtual work principle. Comp. Meth. Appl. Mech. Eng. 120, 131161.
Khoei, A.R., 1998. Finite Element Modelling and Adaptive Remeshing in Static and Dynamic Analysis of
Metal Powder Forming Processes (C/PH/217/98). University of Wales, Swansea.
Khoei, A.R., Lewis, R.W., 1998a. Finite element simulation for dynamic large elasto-plastic deformation
in metal powder forming. Finite Elements in Analysis and Design 30, 335352.
Khoei, A.R., Lewis, R.W., 1998b. Modelling of powder compaction processes via the displacement
formulation and adaptive analysis. In: Huetink, J., Baaijens, F.P.T. (Eds.), Simulation of Materials
Processing: Theory, Methods and Applications (NUMIFORM 98). A.A Balkema, Rotterdam, pp. 385
390.
Khoei, A.R., Lewis, R.W., 1999. Adaptive nite element remeshing in a large deformation analysis of
metal powder forming. Int. J. Num. Meth. Eng. 45 (7), 801820.
Kuhn, H.A., Downey, C.L., 1971. Deformation characteristics and plasticity theory of sintered powder
materials. Int. J. Powder Metall. 7, 1525.
Lewis, R.W., Khoei, A.R., 1998. Numerical modelling of large deformation in metal powder forming.
Comput. Meth. Appl. Mech. Eng. 159, 291328.
Lewis, R.W., Tran, D.V., 1994. Finite element approach to problems in particulate media with special
reference to powder metal forming. Bull. Tech. Univ. Istanbul 47, 295310.
Michalowski, R., Mroz, Z., 1978. Associated and non-associated sliding rules in contact friction
problems. Arch. Mech. 39, 259276.
Nayak, G.C., Zienkiewicz, O.C., 1972. Elasto-plastic stress analysis. Generalization for various con-
stitutive relations including strain softening. Int. J. Num. Meth. Eng. 5, 113135.
Naylor, D.J., Pande, G.N., Simpson, B., Tabb, R., 1981. Finite Elements in Geotechnical Engineering.
Pineridge Press, Swansea.
Nelson, I., 1978. Constitutive models for use in numerical computations. In: Proc. Int. Symp. on Dynamic
Methods in Soil and Rock Mechanics, Balkema, Rotterdam, pp. 45-97.
Nelson, I., Baladi, G.Y., 1977. Outrunning ground shock computed with dierent models. J. Eng. Mech.
Div., ASCE 103, 377393.
Pastor, M., Zienkiewicz, O.C., Chan, A.H.C., 1990. Generalized plasticity and the modelling of soil
behaviour. Int. J. Num. Anal. Meth. Geomech. 14, 151190.
Peraire, J., Vahdati, M., Morgan, K., Zienkiewicz, O.C., 1987. Adaptive remeshing for compressible ow
computations. J. Comp. Phys. 72, 449466.
Peric, D., Owen, D.R.J., 1992. Computational model for 3-D contact problems with friction based on the
penalty method. Int. J. Num. Meth. Eng. 35, 12891309.
Riero, M.D., Prado, J.M., 1994. Modelling of cold compaction process of metal powders. In: Powder
Metallurgy World Congress, Vol. 1, Paris, pp. 685-688.
Rodic, T., Owen, D.R.J., 1989. A plasticity theory of friction and joint elements. In: Owen, D.R.J. et al.
(Eds.), Computational Plasticity II: Models, Software and Applications. Pineridge Press, Swansea, pp.
10431062.
Roscoe, K.H., Schoeld, A.N., Thurairjah, A., 1963. Yielding of clays in state wetter than critical. Geo-
technique 13 (3), 211240.
Roscoe, K.H., Burland, J.B., 1968. On the generalized stressstrain behaviour of wet clay. In: Heyman, J.,
Leckie, F.A. (Eds.), Engineering Plasticity. Cambridge University Press, UK, pp. 535609.
Sandler, I.S., Baron, M.L., 1979. Recent development in the constitutive modelling of geological materials.
In: Proc. 3rd Int. Conf. Numerical Methods in Geomechanics, Aachen, Germany, pp. 363376.
Sandler, I.S., Rubin, D., 1979. An algorithm and a modular subroutine for the cap model. Int. J. Numer.
Anal. Methods Geomech. 3, 173186.
Sandler, I.S., DiMaggio, F.L., Baladi, G.Y., 1976. Generalized cap model for geological materials. J.
Geotech. Eng. Div., ASCE 102, 683699.
Shima, S., Oyane, M., 1976. Plasticity theory for porous metals. Int. J. Mech. Sci. 18, 285291.
Simo, J.C., Ju, J.W., Pister, K.S., Taylor, R.L., 1988. An assessment of the cap model: consistent return
algorithms and rate-dependent extensions. J. Eng. Mech. Div., ASCE 114, 191218.
Sloan, S.W., 1987. Substepping schemes for the numerical integration of elastoplastic stressstrain rela-
tions. Int. J. Num. Meth. Eng. 24, 893911.
Sluzalec, A., 1989. Yield function in simulation of the powder metal forming. In: Owen, D.R.J. et al.
(Eds.), Computational Plasticity II: Models, Software and Applications. Pineridge Press, Swansea, pp.
9951005.
Sun, X.-K., Kim, K.-T., 1997. Simulation of cold die compaction densication behaviour of iron and
copper powders by cam-clay model. Powder Metall. 40 (3), 193195.
Tran, D.V., Lewis, R.W., Gethin, D.T., Arin, A.K., 1993. Numerical modelling of powder compaction
processes: displacement based nite element method. Powder Metall. 36, 257266.
Trassoras, K.J., Tony, M., Ferguson, B.L., 1989. Modelling of powder compaction using the nite
element method. In: Proc. Powder Metallurgy Conf. Exhibition VI, pp. 85104.
Wriggers, P., Simo, J.C., Taylor, R.L., 1985. Penalty and augmented Lagrangian formulations for contact
problems. In: Middleton, J., Pande, G.N. (Eds.), Numerical Methods in Engineering: Theory and
Applications, NUMETA 85. A.A. Balkema, pp. 97106.
Zienkiewicz, O.C., Mroz, Z., 1984. Generalised plasticity formulation and applications to geomechanics.
In: Desai, C.S., Gallagher, R.H. (Eds.), Mechanics of Materials. Wiley, New York, pp. 665679.
Zienkiewicz, O.C., Taylor, R.L., 1989. The Finite Element Method, Vol. 2. McGraw-Hill, London.
Zienkiewicz, O.C., Zhu, J.Z., 1987. A simple error estimator and adaptive procedure for practical engi-
neering analysis. Int. J. Num. Meth. Eng. 24, 337357.
Zienkiewicz, O.C., Zhu, J.Z., 1991. Adaptivity and mesh generation. Int. J. Num. Meth. Eng. 32, 783810.
Zienkiewicz, O.C., Zhu, J.Z., 1992a. The superconvergent patch recovery and a posteriori error estimates,
part 1: the recovery technique. Int. J. Num. Meth. Eng. 33, 13311364.
Zienkiewicz, O.C., Zhu, J.Z., 1992b. The superconvergent patch recovery and a posteriori error estimates,
part 2: error estimates and adaptivity. Int. J. Num. Meth. Eng. 33, 13651382.

Abaqus

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Abaqus

Diunggah oleh

Hak Cipta:

Format Tersedia

A plasticity model for metal

powder forming processes

is a suitably dened matrix operator which contains displacement derivatives

in terms of the linear and nonlinear strain-displacement

is the unit tangential vector.

Anda mungkin juga menyukai