Anda di halaman 1dari 88

Int J Adv Manuf Technol (2011) 56:631647

DOI 10.1007/s00170-011-3211-z

ORIGINAL ARTICLE

Modeling the powder compaction process using the finite


element method and inverse optimization
Meftah Hrairi & Hedi Chtourou & Augustin Gakwaya &

Michel Guillot

Received: 19 May 2009 /Accepted: 31 January 2011 /Published online: 26 February 2011

# Springer-Verlag London Limited 2011

Abstract This paper focuses on studying and e-mail: meftah@iium.edu.my


adapting modeling techniques using the finite
element method to simulate the rigid die H. Chtourou
compaction of metal powders. First, it presents the
1
implementation of the cap constitutive model into Department of Technology, Second order identity
ABAQUS FE software using the closest point IPEIS, Sfax University,
tensor
projection algorithm. Then, an inverse modeling
procedure was proposed to alleviate the problems Sfax, Tunisia
raised by the interpretation of the experimental tests
and to more accurately determine the material :
A. Gakwaya M. Guillot
parameters. The objective function is formed, based
on the discrepancy in density data between the Mechanical Engineering
Department, Laval University,
numerical model prediction and the experi-ment. C
Minimization of the objective function with respect Fourth-order elasticity
to the material parameters was performed using an Quebec City, Canada tensor
in-house optimization software shell built on a
modified Levenberg Marquardt method. Thus, an Keywords Powder
integrated simulation module consisting of an .
metallurgy Compaction
inverse optimization method and a finite element
method was developed for modeling the powder . .
process Finite element
compaction process as a whole. The simulation and
identification module developed was applied to .
Density Inverse
simulate the compaction of some industrial parts. cj
. . Constraint functions
The results reveal that the maximum absolute error identification Cap model
between densities is 2.3%. It corresponds to the .
precision of the experimental method. Abaqus Material
.
parameters Iron powder

M. Hrairi (*)
Nomenclature
Mechanical Engineering Department, International E()
Islamic University Malaysia, Youngs modulus as a
function of compact
Gombak, Malaysia
First order derivatives of
f the weighted penalty
density 1 functions with respect to
the parameters
q
Second order derivatives
of the weighted penalty
functions with respect to
Yield function of the the
Fs tension surface
parameters
Function Fs I 1 a qI1 g exp bI1 ep
f
2
H Elastoplastic
Yield function of the tangent moduli
shear surface I Fourth-order
in the shear failure surface equation
identity tensor
I1 First invariant of
Fc
stress tensor
Function Fc I 1; k s k; k k s k
2 J2 Second invariant
f3
of deviatoric stress tensor
I 1L k &
2 Yield function of the cap
surface J Jacobian matrix of
K() Bulk modulus as a
function of compact density

[K] Stiffness matrix


G
R2
Shear modulus Number of measured data

Number of parameters to
in the cap surface equation
be optimized

q Number of
constraints
cgljm
n, n+1 s
v
632 Int J Adv Manuf Technol (2011) 56:631647

Logarithmic strain
R volumetric
Relative density
Aspect ratio of the cap surface
0 Initial loose
Superscripts
state relative density
R Kirchhoff stress
Unbalanced load vector Cauchy stress tensor cal
calculated using finite
r z Weighted element
Residual vector penalty function e
elastic
r r r exp
1, 2, 3 x Tolerance for experimental
Material parameters describing the aspect convergence (specified k
positive number) iteration
s p
r
ratio of the cap surface R r1 r r2 3 w Non-negative weight pk plastic
Parameter correction
= Helmholtz
Deviatoric stress tensor free energy function
Cost function
* Objective
p function
tk s k s:s
Hardening parameter
Norm of the deviatoric stress tensor
li Plastic
consistency parameters T tangent
Levenberg-Marquardt parameter
Time t
() Trial trial or elastic
T Increment of a quantity prediction
Material parameter (tension cutoff) @
Partial derivative of a
u quantity 1 Introduction
Nodal displacement vector
Subscripts Powder metallurgy
W,D (P/M) has become
Material parameters for hardening rule 0 index for initial widely recognized as a
value robust process for
v
p
producing high-quality
1, 2, 3 indices for parts for a variety of
tension, shear and cap important applications.
failure surfaces This success is due to
the advantages the
" process offers over
respectively cap surface
geometric linearization other metal forming
technologies such as
forging and metal
index for the number of
casting; advantages in
constraints j=1,q
W 1 eDX k material utilization,
material
shape complexity, near-
net shape dimensional
Greek symbols indices for successive
control, among others [
,,, Material parameters for the shear failure Fs I 1 configurations shear
1]. The P/M process can
bI failure surface
a g e 1 qI1 be broken down into
three main manufacturing steps: mixing of the Since these models are a are naturally sensitive to
powder, compaction, and sintering. More than any simple extension of hydrostatic stress and
other operation, it is the compaction, which is classic elastoplasticity, consider the cohesion
generally performed using a set of punches and dies, they are consequently and friction among the
mounted onto a hydraulic or mechanical press that suitable to model the particles of granular
controls the practical application of powder behavior of powders in matter. As a result of the
metallurgy. The semi-finished product resulting from high-density zones but suggestion that the low
the compaction operation is a compact having a unfortunately cannot be strength of powder
used to model the first compacts in shear and
nearly final form but a very weak mechanical
tensile stress field must
resistance sufficient only to ensure its transfer to the stages of compression [
5]. Also, this type of be taken into account,
next step in the process: sintering.X models, such as the
model would require a
plasticity limit that is DruckerPrager cap
The modeling of the compaction operation is used to equivalent during model [ 6], the Cam
optimize not only this step, but ulterior steps as well. In compres-sion and Clay model [ 7], or the
recent years, several types of constitutive models have tension, which is not
DiMaggioSandler
been developed and applied in the analysis of the adequate for granular
compaction of metal powders. Most of the proposed model [ 8, 9] have been
mediums. Granular
constitutive models belong to one of the following two material models, adapted from soil
classes: porous material models and granular material originally developed for mechanics literature.
models. Porous material models are based on the These models are
rocks, soils, and other
classical von Mises model in which some modifications geological materials, are characterized by a yield
are introduced in order to account for the influence of based on a multiple yield criterion, a hardening
function and a flow rule
the hydrostatic pressures on the yield criterion [ 2 4]. criterion approach and
and theyX
Int J Adv Manuf Technol (2011) 56:631647
633

respectively.X ABAQUS POST for the


differ by the functional form of the yield surface. In post-processing of the
the DruckerPrager cap model, the yield surface results. In addition, the
In this work, an enhanced
consists of an elliptical cap and a straight failure line. Cap model for simulating sizeable amount of cap
model parameters and
In the CamClay model, both the failure surface and the behavior of powders
experimental efforts
the cap are characterized by elliptic arcs with during rigid die
required that these
different eccentricity. In the DiMaggio Sandler compaction has been
parameters to be
model the failure surface is given by an exponential developed. This model determined, constitute a
function approaching the yield stress of the fully allows for using variable
handicap for this family
dense material at high pressures. elasticity as well as a
of models. From this,
density dependent cap
comes the necessity to
aspect ratio. It was
Because it captures important aspects of the physics of create an alternate
implemented into the
compaction, the DiMaggioSandler Cap model, on calibration procedure for
finite element (FE) code
which the present work is based, is a frequently used the model, based on
ABAQUS as a user
continuum model for modeling the mechanical inverse problems, in
defined subroutine using
behavior of pharmaceutical powders during compaction order to determine the
a return mapping
[ 10 15] as well as in the simulation of metal powder algorithm. The material parameters
directly from production
compaction processes [ 16 20]. Indeed, this model simulations were run by
components. Thus, this
reflects the yielding, frictional, and densification means of an integrated
work creates a complete
characteristics of powder along with strain and computer module that
module capable of using
geometrical hardening which occur during the compac- uses I-DEAS Master
finite elements and
tion process. However, the counterpart of this flexibility Series software for the
inverse optimization to
is the significant amount of experimental work required geometrical definition of
evaluate the density
to properly calibrate the model parameters for a the problem, an in-house
distribution resulting
particular blend of powders [ 21 24]. Moreover, in developed interface
from the compaction
these studies, the elastic moduli, as well as the cap program for the
process.
aspect ratio, were assumed to be constant which was determination of
not suitable to describe the observed nonlinear boundary conditions,
unloading behavior of metal powders and to better ABAQUS software for
represent ductile powder compaction behavior, the FE solution, and 2 Cap plasticity model
4
To formulate the elastic-plastic constitutive
equations in a large deformation framework, one and where y is
should first choose an Helmholtz free energy Shearfailure

function, C is the
appropriate work conjugate pair of stress and strain fourth-order elasticity
measurements [ 25, 26]. The logarithmic strains tensor, I and 1 are the
fourth and second order surface
were chosen because they are the only measures that
identity tensors
permit an additive decomposition of the strain in the
e
respectively, G and K
large deforma-tion range into elastic ( ) and plastic are the shear and bulk
p
( ) deformation tensors [ 25 27].X moduli expressed as a f2=0
function of the powders
e p relative density .
"" " 1

Another main
For the chosen strain measure, the conjugate stress
component of the Capsurface
measure is the Kirchhoff stress which can be constitutive model is the
approxi-mated to the true Cauchy stress tensor in yield functions that
the case where elastic deformations are small separate the purely
compared to unity. The adopted stressstrain elastic behavior from f3=0

relationship corresponds to that of an isotropic the elastic plastic one.


hyperelastic solid where the elasticity parameters are The yield surface of this Tensionfailure
Elasticdomain
density dependent: elasto-plastic model has
a moving cap,
intersecting the
e p
s C" C " " hydrostatic loading line,
whose position is a
2 function of plastic
volumetric strain, as
with shown in Fig. 1. The surface
main features of the cap
model include a failure
surface and an elliptical
yield cap which closes
f1=0
the open space between
C the failure surface and
2 the hydrostatic axis. The
@y
2Gr I K r 2 = 3G r 1 yield cap expands in the
1 stress space according T
L()
3 to a specified hardening X()
rule. The functional
forms for these surfaces
areX
2
@"
Fig. 1 Multi-surfaced Cap
f1 I1 T 0 elastoplasticity model
634 Int J Adv Manuf Technol (2011) 56:631647
k
f2
q
s
k
Fs
I1

0 p p

where I1 is the first stress


where Fs
f3 Fc I 1; k s k; k Fs k invariant, k s k s : s 2J2 is
I1 0 the norm of the deviatoric stress
2
tensor, is the hardening
a
parameter, , , , , R and T are

material parameters, and


2
q I1 I 1 Lk &

bI
ge 1
6 L k
k if k > 0
7
k where FcI1
;
s

; k ksk 0 if k 0

X k k R Fs k
8

The tension and shear


failure surfaces, f1 and f2,
respectively, represent the
stress states that may cause
fracture of the powder
compact during its ejection
or its compaction with very
large local distortion. This is
often caused by the different
5
compaction ratios of powder
columns in multilevel parts.
The cap yield surface f3 is
2
R

an elliptical function, with R


denoting the ratio of two
elliptical caps diameters.
Traditionally, R was
considered to be constant,
which may be appropriate
for non-cohesive and non-
ductile materials such as
geological materials.
Howev-er, it is not realistic
to model metal powder
behavior using a cap having
a constant axes ratio R. In
this work, R has been
k implemented as an
increasing function of the density. This allows for the proper otherwise desired quadratic rate of
modeling of the behavior similar to that of soils in the low convergence. The scheme
pressure zone. But, when the pressures become important used thus required the
and the density becomes elevated, the plastic yielding updating of the consistent
becomes more susceptible to occur by a shear load than by a tangent stiffness at each
The volumetric plastic strain
hydrostatic load. A flattened cap would be more p
iteration and the use of the
representative of the compacted powder behavior at high " v is the trace of the plastic algorithmic tangent material
strain tensor . It is used to
densities. p
operator [ 30]. Hence, at the
update the actual aggregate end of each time step, the
For strain hardening in the compressive regime, an density using the initial loose current stress state satisfies
exponential function for the effective volumetric plastic the equilibrium of the
p state density 0 as follows: mechanical system with the
strain " v is used to govern the expansive and contractive
behavior of the elliptical cap through the evolution of the external forces in a weak
state variable k. This evolution is given by the hardening rule " sense and complies with the
p r r0 e vp full consistency of the
relating k to the effective volumetric plastic strain " v in a 11 elastoplastic material and
form suggested by the results of the classical hydrostatic frictional contact
pressing test
formulation.X
i 3 Computational algorithm
of cap model In this nonlinear context
p
" v X k W 1 e
DX k
9
where the principal
3.1 Finite element problem unknown is the incremental
p
nodal displacements vector f
where, W and D are two material parameters and " v is the un1g between
The compaction process is
effective plastic strain expressed in terms of the stress first assumed to be a quasi-static configurations n and n+1,
transformation with the final the discrete linearized FE
spatial distribution of density approximation to be solved
(x) as the driving unknown at each iteration k can be
invariant X(k) and defined in a manner that prevents the cap written as follows:
strain from softening: [ 28, 29]. Every material
point undergoes finite strain K
T
deformation and
elastoplastic trans-
formations take place under
( the interaction of the powder k

medium and the tooling


p components. The u
p corresponding nonlinear
structural FE analysis k

involves integration of the


differential elastoplastic
equations in time and space.
p
The time integration in the
present context is due to the
"v
if "v > 0 or if k > 0 and k > I1
discretization of the loading
10 history, while the spatial
integration is normally
performed via Gauss
quadratures at the FE level
and then assembled in terms
of the mechanical degrees of
freedom.X

A nonlinear solver based on


the NewtonRaphson
method has been used. This
"v choice is motivated by the
presence of strong
0 nonlinearities and the
T
k

u
n1g
k

Km

12
Kg

R
n1g
Kl

f f
T
g
h

i k

Int J Adv Manuf Technol (2011) 56:631647


635

h i 2
Solve for
k

T
where K m
3.2 Integration of the behavior law
displacement vector
The numerical integration procedure of the behavior model u
was completed according to the closest point projection n 1

algorithm. Calculation of the stresses is performed follow-ing 0


an implicit scheme of predictioncorrection and by using an to
operator split method [ 31]. First, the plastic evolution is
assumed to be momentarily frozen and an elastic prediction their values at the
is performed. If the plasticity criterion is violated by the
prediction, then a plastic correction must be performed to
take the yield that occurs into account. This algorithm is
1f un1g k the linearizedk problem:
coded in Fortran 77 and implemented into the ABAQUS configuration n
software using the User Material option UMAT. At each f
iteration, performed for a single load step corresponding to
g
the passing of an equilibrium configura-tion Cn to following
configuration Cn+1, the ABAQUS software calls upon the T

UMAT subroutine to perform an integration of the behavior k

law. This access to theX

Table 1
Algorithm for the finite element solver

0 f un1g
Rn1g
f Rn1g

Step
Set the iteration counter k to 0
Calculate the residual vector f

Update the solution:

k
Step
1 Set k=k+1
Initialize the global Jacobian
K
T
0 k 1
and the nodal k

fu n1g
f un1g f un1g
;k
and
r

in addition

n1ep
@
ep
n 1 3

to the elastoplastic tangent moduli @fi s ; h


Hn1
Step
Check for convergence on both the displacement and
s

3
the residual @
"


l
Store all the material point i
variables

Compute and assemble the


residual vector and the material
13
stiffness matrix

Go back to step 2 for another


equilibrium iteration
If convergence, then set n=n+1 and go to next load
subroutine is done for each
of the numerical integration
point elements. The
@s

increment
schematic is based on the
algorithms proposed by
Simo et al. [ 32] for
implementation of the Cap
model. Since this is a multi- i1

surface elastoplasticity
model, there should be an
intermediate algorithmic
step between the elastic
prediction and the plastic
correction: that of the active
plasticity mode detection.X
If no convergence, then:
The numerical integration of
*For each element:
the behavior law is based on
the integration of the flow
*For each material integration point X
rule according to an implicit
Compute the strain increment
Euler scheme, with time
increment
Retrieve the material point variables sn; kn and rn t between the instants

Perform constitutive model integration using the closest point projection


algorithm by calling the UMAT subroutine tn and tn+1:
and obtain the values of s
n1
16 assumption, designated as the
elastic predictor or the trial
where, li represents the plastic consistency parameter increment
Furthermore, the tensor stress, is given by:
P
and designates the tensor of the plastic strain increment. This
chosen to define the total
tensor is expressed as a function of
the plastic deviatoric strain increment strain increment can sn
Trial p
C : " n1 "n C
1
p
e and the plastic validly be decomposed e
additively into an elastic :"n
p
volumetric strain increment tensor e "n1
and a plastic part:
:
v
e p
e e e
p p 1 p 17
e e 3 "v 1
14 sn C : "n1
3.2.1 Elastic prediction
18
In a similar fashion, the total strain increment tensor is
From a physical point of
expressed as a function of the deviatoric strain increment and view, the elastic prediction
the volumetric strain increment. The same applies to the consists of freezing the
stress tensor that can be broken onto a deviatoric part plastic yield and applying an
elastic increment. This is
s and a spherical part I1:
basically to consider that, at
1
instant tn+1, the plastic strain
" e 3 "v 1 is fixed and equal to the value
15 at instant tn. The stress
1
s s 3 I1 1 obtained following this
636 Int J Adv Manuf Technol (2011) 56:631647
s material tangent moduli
CompressiveCornerMode
corresponds to the stress
Using the stress tensor decoupling principle (Eq. 16), the
variation caused by an
spherical and deviatoric components of the stress tensor are infinitesimal strain
then obtained from the following equation:X ShearFailureMode variationX

Trial
I1 ;n1 I1;n 3K ep
H
"v;n1 f2=0 ds
19 n1
21
Trial
sn 1 sn 2G CapMode

e
n1 TensionCorner d"
f3=0

This last equation is used to predict if a plasticity state has


Trial
been reached. Indeed, the yield function fi; n1 test values Mode
n1
are obtained with this prediction. If all the test values for the
Trial
fi; n1 functions are negative, then the load state was
purely elastic. On the other hand, if some of the test
functions become positive, the real stress state will be
obtained after a plastic correction [ 33].X It is therefore determined by
ElasticMode
TensionFailureMode a simple mathematical
f1=0 derivation of the relation
3.2.2 Plastic correction
between the actualized
stresses and the strains:
The state of the material undergoing elastoplastic deforma-tions T L(K) X(K) p
is generally characterized by loading and unloading conditions. dsn1 C : d"n1 d"
Thus, the plastic loading or the elastic loading/ unloading is 22
formulated according to the Kuhn-Tucker relations that govern Fig. 2 Various elastoplastic modes n 1
within the Cap model
the evolution of the plastic consistency parameters li;n1. These
conditions, that must always be verified, stipulate that at each This formulation requires
instant there must be: the determination of the
plastic strain variation. The
3.2.3 Elastoplastic derivation of this module is
li;n1 0; fi;n1 0 and fi;n1 li;n1 0 ; i 1; 2; 3:
tangent moduli performed differently if the
active mode presents perfect
20
The last step in the numerical plasticity or plasticity with
integration of the constitutive hardening [ 33].X
This equation is used to characterize each of the Cap models evolution equations is
modes: elastic mode, tension mode, shear mode, cap mode, concerned with the
singular tension mode at I1 =T, and singular compression determination of the powder
mediums algorithmic 4 Inverse identification of
mode at I1 = (Fig. 2).X
material contribution to the the material parameters
FE tangent stiffness matrix.
Each mode will be identified in terms of limits to the elastically This contribution is computed Judicious use of the finite
predicted stresses. Then, the plastic consistency parameters that at the integration point level element simulation method
allow calculation of the plastic strain are determined. The stress and is called the elastoplastic rests on the precision and
conditions of each one of the plastic modes is presented in Table 2, tangent moduli. In order to exactness of the behavior
together with the corresponding values of the plastic consistency preserve a quadratic rate of model used. To this end, it is
parameters increments. These values are then used to determine the convergence for the global FE not sufficient for the chosen
plastic strain increment (Eq. 13) and thus, the real stress state (Eq. problem, this moduli should model to be well adapted to
19).X be derived from the the type of behavior being
algorithmic and not from the modeled but additionally, it is
continuum mechanics important to ensure that the
formulation [ 33]. In fact, this model parameters represent
the behavior of the material in question. Traditionally, the Cap However, the practical powders is fairly limited, it is
model parameters are determined from a rheological analysis of application of this hypothesis mostly due to the difficulty in
experimental tests chosen for their suitability to the behavior is a very difficult task for its parametrical and
being studied: resonant frequency, hydraulic compression, mechanical tests that are often experimental characterization.
hydro-static compression, triaxial compression, simple uniaxial not uniform and were aimed Often, because of the
compression, and special tests such as the Brazilian disk test [ at characterizing the material. hypotheses required by the
Furthermore, if the use of the experimental
33 36]. In the identification process, the interpretation of the
Cap model for modeling the characterizationX
experimental trials usually requires the use of the homogenous
compression of ductile
deformation hypothesis in order to thresh out the results.
Int J Adv Manuf Technol (2011) 56:631647
637

Table 2 Boundaries and plastic consistency parameters of the plastic modes


Failure mode
Stress state boundaries

Incremental consistency parameters

Tension
(
Trial
T

l
1;n1
Trial
T I1 ;n1

I
1;n1
9K

sTrial

Fs
T
l 0
2;n1

n1

l 0
3;n1
1

k
Trial
sn 1

k
Fs T

Tension corner
ITrial
1

T
l
1;n

8
1;n

2G

Trial
2G

Trial
T I1 ;n1

l
2
n
1

Trial
T I1 ;n1

l
1n

1
dFs T
>
F
s
T

<

s
n 1

< Fs
T

9K
;

dI1
<

9K dFs T =dI1

l 0
3;n1

>

Shear
:
I1Trial;n1
kn
l
1;n
1
0

T
Trial

Trial

I
1 ;n1

Fs T

I
<
1;n 1

Trial

k
l2 ;n 1

s
n1
k

Fs

>

s
n 1


2G

>

dFs T =dI1

l
3;n 1

>
>

>

>

k
n

Trial
<

Trial

I1;n 1
>

s
n1

< F
skn dFs kn =dI1

>

>

l
1;n

>

Compressive corner
>

Trial

>
Trial

:
I <k
1;n1 n

8
Trial
I1 ;n1

l
2;n 1

kI
n 1:n1

Trial
2G

kn

9KdFs
kn

=dI1

>

s
n

> Fs

kn

Trial

<

9K dFs k n=dI1

l
3;n1

s
n1

Fs k n

l
2;n1

Cap
>

Trial

kn

2G
:
Lkn

< I1
n

Trial

Trial

or

>X

kn

l
1;n

Trial

I
1;n 1

l
2;n1 p

<

s
n1

> Fc

I
1;n1

0
l
3;n 1

R2 F k
"v;n1 s n1

3 I1;n1 kn1

(r1, r2, r3) parameters for the surface and the hardening
approach, the parameter sets available do not reproduce real plasticity Cap model that law of the constitutive
practical situations. Thus, the identification of parameters are represented by the model are estimated by
using the inverse method directly from industrial produc-tion vector p=(E, G, T, a; b; g; means of inverse modeling,
components represents an attractive alternative to the the elastic parameters as
q, W, D, r1, r2, r3).
classical procedure since it emanates from practical sit- well as the shear failure
However, a sensitivity study parameters are
uations and avoids the hypothesis of homogenous defor- of the model showed that,
mation altogether [ 37, 38].X predetermined from an
when dealing with powder experimental material
compaction with no
4.1 Nonlinear optimization process characteriza-X
excessive distortion, only
the harden-ing law and cap
The optimization task is based on the LevenbergMar-quardt surface parameters exert a tion of the cap model [ 41].
algorithm [ 39, 40] in order to determine the parameter preponderant influence on Consequently, the parameter
corrections while the finite element solver handles the the final density prediction vector to be identified is
solution of the direct problem. The principal goal of this in the compact [ 41]. Thus, reduced to p=(W, D, r1, r2,
optimization process is to determine elasticity (E, G), failure only the parameters r3).X
surface (a; b; g; q), hardening law (W, D), and cap surface describing the cap failure
Estimating these parameters involves adjusting the is subject to constraints on
parameters in the finite element models until the calculated the parameters based on the
density vector
cal
coincides as the least square with the
2
i1

physics of the problem. The


constraints are written in the
measured density vector
exp
2

. This is accomplished by form:X


minimizing the cost function with respect to the vector of
parameters p: cj p 0 j 1; q
25
6 p where cj are the constraint
1 X functions and q is the number
m
2 of constraints. The set of p
r ip & which satisfies the constraints
1 is called the feasible region.
T In many problems, it is
r r
23 important to maintain the
feasibility of the parameters
where m is the number of throughout the solution
measurements. The residual process in order to carry out
vector, r, is defined by the finite element analyses. In
the current estimation,
bounds are required to
prevent the parameters from
r rexp rcal becoming negative.
24

The minimization of Eq. 23


638 Int J Adv Manuf Technol (2011) 56:631647

The parameter constraints, enforced by the interior penalty


function method, are incorporated directly in the objective
function, , as follows:
*

q q
26 @zj
the optimization process will
6 p 6p z jp stop and the final
optimization is achieved. The
k
parameter correction, p , at
X iteration k is calculated from g

the following system of


s j 1

j1
equations:
where the weighted penalty functions, zj, are the inverse
barrier functions: T
s 1; n
31

Jk Jk mkI hk
wj T
j 1; q dpk Jk rk gk @p s

27 29

z p where is Levenberg
j cj p
Marquardt damping
parameter (a non-negative
scalar), J is the Jacobian

matrix of , and g and h


with j being the non-negative weights. contain the first and the
second order derivatives,
respectively, of the weighted
The nonlinear least squares problem is solved using a modified
penalty functions with respect
LevenbergMarquardt method to account for the weighted
to the parameters. The
penalty functions included in the objective function . Starting elements of J, g, and h are
*

from an initial feasible parameter guess, the modified given by


LevenbergMarquardt method deter-mines a sequence of
corrections to the parameters until convergence is achieved
according to specified criteria. The convergence criterion for the Jis
@ri
optimization study is based on the variation of the objective
function value. If differences in the objective function value
between two subsequent iterations is less than a specified P
positive number ,
cal
@ri

6
i 1; m ; s 1; n
p k1 30

6 pk
<x @ps

28

@ps

2
@ zj
34

The scaled matrix A and the


h
st j 1
scaled vector b are then
defined as:
A
st written as:

s; t 1; n A
32 st

s; t

@p s @p t

1; n

35

A m I dp b

p
Ass A tt

37

And p is calculated from dp using

dp
dps

p s
s 1; n 38
A
P
ss

The fact that Ass 1; s


1; n, allows to be
controlled in a manner that
is problem independent and
consistent from iteration to
iteration.
bs
Recall that m is the number of measured data and n is the b
Because the parameters are
s
number of parameters.
p A ss contained implicitly in the
s 1; n finite element formulation, a
finite difference approxima-
To facilitate the choice of the LevenbergMarquardt tion of the Jacobian matrix J
36
parameter and improve the conditioning of the in Eq. 29 is made. The
The Jacobian matrix is
recalculated at every
scaled form of Eq. 29 to be solved for p is then

iteration by perturbing the


parameters one at a time and
system, Eq. 29 is scaled before it is solved. Defining A and b solving a direct finite
as:X element problem for each
perturbation.X
T
AJ Jh
33 The first derivatives of the
T
bJ rg penalty functions required
to calculate g are given by:
@zj @pt
The second derivatives of
the penalty functions
wj @cj
required to calculate h are
j 1; q ; s 1; n 40
approximated by
39
2 s; t 1; n
@ zj
2 The modified Levenberg
wj Marquardt method and
finite element analysis are
@ cj integrated in the solution
algorithm as shown in Table
@ cj 3.X

j 1; q ; 4.2 Application to the


@ps identification of the cap
model parameters of iron
2 powder
cj @ps@pt
The method described
3
@ps cj above is applied to
determine the cap material
parameters of MP37HD iron
@ps
powder
Int J Adv Manuf Technol (2011) 56:631647
639

Table 3 Inverse finite element optimization algorithm

and of the Levenberg-Marquardt damping parameter


(0) (0)
Initialize the parameters p

and evaluate the cost function


(0) (0)
Solve the direct finite element problem at p
0
Initialize the penalty functions weights j. Evaluate the weighted penalty functions z j and the objective function
*(0)

For iteration k=0,1,2,K

(k)
Calculate the Jacobian matrix J using the finite differences method by solving n direct problems.

(k) (k)
Calculate the penalty function derivatives to form h and g

k k 1 k k
Solve Eq. 29 to determine the correction step to be applied to the parameters, dp and update the parameters p p dp X
k 1
and
(k+1) *(k+1)
Solve the direct finite element problem at p . Evaluate z j
*(k+1) *(k)
Check if <
k k
If false, increase and return to step 4c m 10 m

If true, reduce wj and


k 1 k
wj 0:1 wj mk 1 0:1 mk
Verification that the convergence criteria is met

If false, k=k+1 and return to step 4a

If true, stop
CAX4 shape function is
supplemented by placing
4.2.1 Finite element fine elements at the critical
regions of the part and by
manufactured by Domfer. The MP37HD powder grade used calculation of the answer the fact that the initial mesh
in this study was premixed with 0.375% Acrawax and was oriented in the direction
0.375% ST Lithium as lubricant and has a chemical Considering both loading
composition by weight of Cu 1.80%, graphite 0.75%, and the and geometric symmetries, of compaction (Fig. 4).X
balance is Fe. The industrial part, selected to apply the an axisymmetric element
inverse identification approach to, is one whose geometry needs to be used. Our The FE model consists of
and filling positions for the different tools are given in Fig. preliminary analyses only about 456 elements
3. It is an axisymmetrical part with two levels. The tooling, revealed that the eight-node and 591 nodes. We also
used in the manufacturing of this part comprises a die, a core CAX8 element has a place contact surfaces [ 42]
rod, an upper punch, and two lower punches.X problem with discontinuous at both the powder and tool
equivalent plastic strain at surfaces of Fig. 4. In
its mid-node. Thus, we use addition, since our main
The compaction sequence consists of four steps. The first the four node axisymmetric
two consist of the descent of upper punch 1 and lower punch element and two degrees of objective was the
2. Lower punch 1 stays fixed throughout the entire process. freedom per node CAX4 determination of the
During the third compaction phase, the die and core rod join which provides better compact"s density
the descent of the two punches. In the last step, only the die numerical stability distribution, the simplest
and the upper punch continue to descend. The compaction especially in the presence of and numerically cheapest
kinetics, described by the movement of the different tools way to achieve this was to
high nonlinearities [ 42]. impose nodal displacements
(position in mm), is given in Table 4.X
The lower degree of the atX
Corerod R C
24.5mm L
Upperpunch1 =
Lowerpunch1 14.98mm
Lowerpunch2 A
Fig. 3 Picture, geometry and tooling of the R=32.5mm Die
S 55.52mm
industrial part #1 e
12.7mm

640

Int J Adv Manuf Technol (2011) 56:631647

Table 4 Compression kinetics


Filling position
First step
Second step
Third step
Fourth step

for industrial part #1

Core rod
48
48
48
29.2
29.2

Die
48
48
48
29.2
22.6

Lower punch 1
25.1
25.1
25.1
25.1
25.1

Lower punch 2
30.42
29.62
20.22
0
0

Upper punch 1
45.4
41.6
30.3
9.6
4.9

To this numerical
distribution, the
experimental distribution
the powder boundary node sets. Also, as it can be seen in Fig. 4,each one of the steps as
, determined using the
exp
large displacement contact with Coulomb friction was adopted indicated in Table 5. This
for the interface regions between the powder cavity and tooling table summarizes the number Vick-ers hardness method [
component sides. Moreover, since very small powder sliding of global equilibrium and 45], is correlated. It is
occurs on punch faces, tied (sticking) contact was adopted for contact iterations required by crucial that the initial guess
these interface regions. Both contact conditions were modeled the solver for each of the four of parameters is reasonable,
using the master-slave contact pair formulation in ABAQUS [ load increments. It also shows e.g., parameters for a
42]. This formula-tion requires the association of the master the average local iterations
different powder mix. The
surface with the stiffer interacting body (tooling component) needed at the integration
optimization procedure has
whereas the slave surface is associated with the second point level by the
to struggle with many local
deformable body (powder cavity). The friction coefficient was implemented algorithm. It is minima which may appear
clear that despite of the
assumed to be equal to 0.2 as in previous studies [ 43, 44] and far from physically
relatively large deformation
its reliability was verified by comparing simulated results with acceptable values. It is
increments, convergence of
experi-mental measurements as discussed in Section 5.X the closest point algorithm normally necessary to start
has required a relatively small the optimization from
Moreover, the initial relative density of the powder is number of iterations.X different starting points to
assumed to be uniformly spread and corresponds to a value check the behavior in the
of 36.5%. The applied load is translated into imposed motion 4.2.2 Parameters neighborhood of the
on the surface of the tools according to Table 4. The tool computed optimum. The
optimization
motion is imposed in increments. For each increment of initial parameter values
motion and for each element, the relative density of the were set close to the values
The optimization strategy obtained through the
powder at the integration points is calculated. Furthermore, consists of solving the direct
in the finite element calculation, the multi-criteria Cap type problem using an initial set traditional experimental pa-
constitutive model and the numerical integration schematic of the parameters, thus rameter estimation
procedure for another type
described above are used.X generating a calculated of powder, namely Distaloy
distribution of the final
AE [ 41].X
Furthermore, the four-step FE problem was solved by density of the compact, .
cal
ABAQUS with a relatively small number of increments for
deformed mesh surface)
Restrictedr
displacement Stickingcontact(slavesurface)
Imposedzdisplacement
Fig. 4 Boundary conditions and initial and
Stickingcontact(master Frictionalcontact(mastersurface)
Frictionalcontact(slave surface)
Int J Adv Manuf Technol (2011) 56:631647

641

Table 5 Summary of the nu-

Statistic
Step 1
Step2
Step 3
Step 4

merical performance

Needed increments
17
18
18
18

Average number of global iterations per increment


9
9
13
13

Average number of local (UMAT) iterations per increment


12
10
13
14

In order to make the


The correspondence between the calculated density and the simulation of powder
experimental density is hardly automatic. Indeed, the Vickers compaction an attractive
hardness measurement points are spread in a random manner design tool, an integrated
on the parts surface and the meshing nodes, where the simulation module has been
density is evaluated, do not coincide very often. developed. This module is related tasks. It is composed
Consequently, a smoothing technique had to be used in order mainly intended to facilitate of the I-Deas Master Series
to make the two measurement types positions coincide. and to automate some of the CAD software [ 46], the
Three smoothing techniques were investigated: neural tedious modeling- ABAQUS nonlinear FE
networks, krigeage, and interpolation functions. The two first solver [ 47], the ABAQUS-
methods required processing that was external to the Table 6 Values of identified Post post-processing [ 48]
optimization module, thus offsetting the automatic aspect of parameters for the iron powder software, an in-house
the identification procedure. Furthermore, their application to developed pre-processing
this case seemed more complicated and required a higher Parameter
and interfacing program
calculation time. Thus, the interpolation functions that could Initial value IDEQUS [ 49], and finally
Identified value
be used on all the elements of the meshing, since they all an inverse identification
possess the same reference element (quadrilateral element material parameters module
with 4 nodes), was deemed more suitable. INVMOD. The integrated
software layout, as well as
In the case under consideration here, the number of W
0.8 the main functions of its
parameters to be identified is n =5. The number of 0.864 components, is described in
measurements m, must therefore be greater than or equal to D Fig. 8.X
n=5. This condition is guaranteed to always be respected 0.1E-04
since the number of Vickers hardness measurements is 0.4091E-04
r 5.1 Pre-processing
generally quite high (in this case, m=78). Table 6 contains 1
the identified parameter values for the iron powder whereas 1.0 The pre-processing is first
Figs. 5, 6, and 7 represent the evolution of the objective 0.621
performed through I-Deas,
function and of different material parameters throughout the r2
in which the component
18.0
identification process.X geometries are defined and
16.7
r3 then meshed using a semi-
0.5 automatic mapped pattern.
5 An integrated software environment 0.383 User-defined macro
functions, implemented into I-Deas, are then used for the
selection and the identification of the special boundary
regions of the powder cavity and tooling components.
IDEQUS automatically identifies, generates, and pairs off
contact surfaces using the element sets selected by the macro
functions. All the information is written to a universal FE file
which is then completed with additional information and then
translated into the ABA-QUS format by way of the IDEQUS
program.

5.2 Processing
of cost
0.030 0.015
ABAQUS is an FE solver capable of handling both 4
geometric and material nonlinearities, as well as the 8
frictional contact nonlinearities. One of its main advantages 12
is the open facility it offers for the definition of user material 16
models. In fact, the cap material model was

0
0.045 0.025

function Norm
0.040

Iteration
MP37HD

0.035
0.020 Fig. 5 Cost function evolution

642

0.93 MP37HD

4.5E-05
4.0E-05

0.91
3.5E-05
D

W
W

3.0E-05
Parameter

Parameter
0.89
D
2.5E-05

0.87
0.85

2.0E-05
1.5E-05
1
5
9

13

17

Iteration

Fig. 6 Convergence of the hardening parameters


Int J Adv Manuf Technol (2011) 56:631647 the physical system under study, e.g., by comparing
density distributions. The experimental densities are
5.4 Inverse identification module determined by means of the Vickers hardness technique.
The numerical density distributions are computed with the
The main task of this module is to identify a selected set of FE model. The a priori unknown material parameters in the
unknown parameters of the numerical cap model. The FE model are iteratively updated in such a way that the
unknown parameters are determined iteratively by mini- computed densities match the measured densities as
mizing a cost function which expresses the discrepancy closely as possible. Figure 9 shows a flowchart of the
between the experimental and the computed response of applied inverse method.X

implemented into
ABAQUS via the UMAT
facility that allows the user r2
to define a constitutive
model and implement its
numerical integration
algorithm as a FOR-TRAN
subroutine. Thus, UMAT is
called by the main program r1
at each element integration MP37HD
point within every
equilibrium iteration of
each load increment in the r3
deformation process. The
global FE problem was
solved using the classical
NewtonRaphson method
with a line-search algorithm
whereas the local material 0.900
integration was handled by
the closest point projection
algorithm.
0.800
5.3 Post-processing

The ABAQUS-Post
software is used mainly for
the visualization of the
predicted density maps
within the compacts. The
stress distribution, as well
as the deformed shape of
the tooling components,
could also be post-
processed by the same
software.

24

22
2
20

Parameter
18

0.700
,r

0.600
Parameters
16

0.500
0.400

12

14

1
5

13
17

0.300

Iteration

Fig. 7 Convergence of the


coefficients with respect to the
cap surface axes ratio
triaxial load paths as well as find the volumetric plastic hardening parameters W
data points generated by the strain "pv as a function of I1, =0.847 and D=0.4418E-04.
hydrostatic tests were used the hydrostatic component W being a measure of the
to draw the iso-density of stress. For every maximum achievable
6 Validation curves representing the cap isostatically pressed volumetric plastic strain.X
surfaces at different relative p
6.1 Validation of the inverse densities. This fit shows an specimen we determined " v A sensitivity study of the
results with classical tests increasing aspect ratio R as using the initial loose model, based on simulations
a function of density that density and the final planned and analyzed using
The Cap plasticity material follows the empirical measured post compaction
the Taguchi method,
parameters of the iron relationX density (Eq. 11). The
revealed that, when dealing
powder has been identified obtained values were plotted
with powder compaction
by Hrairi [ 41] based on the r against the first stress
Rr r1r r2 3 with no excessive distortion,
classical homogenous tests. invariant corresponding to
0:208 the most influential
The aspect ratio and the 1:148r 38 the hydrostatic pressures
parameters on the final
hardening parameters are 41 that served to produce the
density distribution are
determined by triaxial tests specimen. This fitting
those related to the
and isostatic compaction To determine the premix process (Eq. 9) permitted hardening of the material.
respectively. Data from the hardening law, we had to the determination of the The identified values of the
Int J Adv Manuf Technol (2011) 56:631647
643

Meshing
Fig. 8 Integrated software framework for powder compaction simulation u material
Definepowd n
erand
tooling UMAT
parametersby

A
Graphical B
Identificatio inverse
nof A
macros materials B CapMaterialmodel
andfriction
coefficients
Boundaryregionselectionandidentification

interface N modeling
and o
boundary implementedinto

USER R
Transformth
eIdeas e
universal
file
I
IDEAS N
p
limits AbaqusUser
r
IDEQUS intoan
Abaqusinpu s
tfile oE
s MaterialRoutine
Definitionoftooling
Prescribethedisplacement

Generation
ofaFE
andpowdergeometries
sequenceofeachtool
minimum?

different parameters are summarized in


Table 7 as well as their respective
relative influence on the final density
distribution in a compact, as determined
by the sensitivity analysis.X

It can be noted from Table 7 that the


results of the inverse method and the Difference
classic, homogeneous tests are quite
similar when it comes to the hardening
parameters but differ considerably for
the aspect ratio parameters. This
discrepancy can probably be explained
by the fact of the almost no contribution
of these parameters to the final density
distribution in a compact. As one can see
fromX parameters
No

Boundary

Initialparameter

Yes

Experimentalsetup
calculation

Costfunction
estimates

conditions

Sensitivity

Material
Ta sure-ment of final core rod and die,
FEmodel bl density, is rather two upper
e insensitive to the punches and three
7, values of R (). lower punches. W
th Lower punch 1 0.864
As a result it is
e does not move, 0.847
hard to determine
inf lower punch 2 is 65.6
r1, r2, and r3 as mounted on a D
lu
en accurately as W spring, and there 0.4091E-4
and D. Therefore, is a level in upper 0.4418E-4
ce 8.0
of in addition to the punch 2.X
r
R( final density in 1
the compact, 0.621
) more measurable The compaction 1.148
su responses should sequence is made 1.45
parameters ms be taken into up of three steps. r2
to account in future During the first 16.7
research in order step, lower
4. 38
punches 2 and 3, 1.07
4 to accurately
r3
Parameterestimationstrategy
% determine all the as well as the two 0.383
co parameters of the upper punches, 0.208
m Cap plasticity begin to descend. 1.88
pa Lower punch 1
model.X
re remains fixed
d throughout the
6.2 Application of process. During
to
the results on an the second
a
industrial part
w
ho Table 7 Values and
pp The pressing of a relative importance
gear, as shown in of the iron powder
in parameters
g Fig. 10, is
65 simulated to
Parameter
.6 evaluate the Inverse method
Fig. 9 Flow-chart of the inverse method for
material parameters identification % estimated Classical tests
an material Contribution (%)
d parameters.
8 Manufacture of
% this part requires,
other than the
644

Int J Adv Manuf Technol (2011) 56:631647

Corerod

Upperpunch1
Upperpunch2

Lowerpunch1

Lowerpunch2

Lowerpunch3

L
2.97

Die

11.45

8.09

2.01

.5

7.3

4.41
24.7
.6

Fig. 10 Picture, geometry and tooling of industrial part #2


measurements was typically
2
green compacts and the under 3 kg/mm . This
compaction step, the die and the core rod join the descent of logarithm of Vickers results in an inaccuracy of
the punches. At the end of this step, the two upper punches hardness can be correlated approximately 2% in the
continue to descend, thus ensuring the compaction of the part with a linear model: worst cases, i.e., at low
from the top. The compaction kinetics, describing the motion density ranges.
of the various tools (position in mm), is given in Table 8.X r 19:306 log HV
39:831 The computed density
42 distribution, based on the
Results from the simulations are depicted in Fig. 11 together
esti-mated parameters,
with an experimentally measured density distribu-tion using the
Vickers hardness method. This method was applied on Error limits introduced by shows a reasonable
this model, by sample agreement with the
compacted parts according to the prescribed kinematics (Table
preparation factors and by experimental result along
8) leading to the density distributions shown in Fig. 11b. This the hardness measurements the outer boundary and in
Vickers hardness measurement technique was used to obtain can the center. However, along
apparent hardness also called macrohardness, i.e., the average p

the core rod, the computed


hardness of a porous material. Its procedure is well documented be established to 0 :05 results differ more
7:8=n with a confidence
in ASTM E92-82 standard [ 50]. This standard offers various compared to the experiment.
level of 68%. Curiously,
loading from 1120 kgf applied to an indenter of pyramidal As illustrated at some
geometry with face angle of 136. In this work, a 5 kgf load (HVmost of the error is caused critical points in the
by the Vickers
5) was used as recommended in MPIF standard 43 [ 51]7. compact, the maximum
measurement, and thus, can
Measurements were performed on five billets of different absolute error between
significantly be reduced by
densities. Each billet had two faces on which n=15 hardness densities was found to be
measurements were realized in a 35 pattern. Using a linear
taking and averaging n
measurements in a small 2.3% corresponding to the
regression technique, the relative density ofX precision of the
area. The standard variation
on Vickers hardness experimental method.
Lower 0
Filling position Core rod punch 1 Lower punch 3
12.2 8.819
Table 8 Compression kinetics for industrial part First step 43.821
12.2 7.877
#2 Second step 43.821
Third step 31.765 12.2 2.989
27.255 12.2 2.989
Die Lower Upper punch 1
43.821 punch 2 43.595
43.821 14.68 36.385
31.765 11.716 24.355
20.005 0 12.5
Upper punch 2 5.373
24.613 4.411
17.403
Int J Adv Manuf Technol (2011) 56:631647
645

Fig. 11 Final density distribu-tion within the compact:

a numerical, b experimental

A
A

C
C

Relativedensity(%)
Relativedensity(%)

63.5
D

65.1

80.5
66.8

81.0

68.5

|85.1 -82.8|=2.3%

81.6

B
70.1

82.2

71.8

|83.8 - 86.0| =2.2%

82.8

73.5
75.1

83.4

76.8

|82.7 -80.7|=2.0%
84.0

78.5

84.6

80.1

85.2
81.8

|81.8 - 80.5| =1.3%

83.5

85.7
85.1

86.3
PM part has been performed
material, geometric as well successfully thus
as contact nonlinearities. demonstrating the practical
While the last two industrial applications of the
However, one can notice a big discrepancy between the computational approach.
range of the relative density of simulation results which nonlinearities were handled
Indeed, a comparison
varies from 63.5% to 85.1% and that of the experimental dataautomatically by the between the experimental
which shows a variation between 80.5% and 86.38%. This ABAQUS FE solver, the results and those obtained
difference in range is seen only in the circled zone where first involves a material
model unavailable in by simulation, using
there is a sharp change of section (Fig. 11). The numerical parameters identi-fied
ABAQUS and hence the
simulation rightly predicts this important density gradient elastoplastic cap model had inversely, revealed that the
(63.580.1%). However, no local density measure-ments to be formulated and maximum absolute error
could be achieved in this region of the part. In fact, this integrated into the software between densities was found
region was fractured during its metallographic prepa-ration. in order to be able to to be 2.3% corresponding to
This fracture was predictable from the simulation results that simulate the behavior of the the precision of the
showed an important density gradient near this corner of the metal powder medium. The experimental method. Also,
part. Consequently, a global density measur-ing technique closest point projection the results from the current
was applied to the fractured region. The obtained value algorithm was used for the work show that inverse
(72%) agrees well with simulated results, stating that in that numerical integration of the modeling of the powder-
region, density varies between 63.5% and 80.1%X multisurface plasticity pressing experiment is a
model. Due to its flexibility viable method for deter-
Since the results obtained showed the desired degree of and capacity to represent all mining material parameters.
precision, the meshing initially proposed was deemed adequate the compaction stages, the It serves as an alternative to
and did not require further adjustment. The simulation of this cap material model was the already established
compaction sequence without the necessity for an intermediate shown to yield very good methods and is generally
meshing would not have been possible if the initial meshing had results as far as the final less expensive to carry out.
not been conceived as a function of the kinetics used in the density was the main The concept proposed in
manufacturing of this part. concern and despite the fact this work is a flexible
that the model was not approach for identification
completely characterized. of cap material model and
7 Conclusion This is however expectable can be useful to apply the
since the final density is results to metalworking
This work has been focused on a unified approach for mainly sensitive to the cap industry. The information
parameters identification of P/M specimens in powder hardening parameters which derived can be used for the
compaction process. Thus, it has created an integrated seem to have been correctly subsequent quantitative
simulation environment capable of using finite elements and identified. design as well as
inverse optimization to evaluate the density distribution optimization of the powder
resulting from the compaction process. The problem of rigid die Using this integrated metallurgy processes.
compaction of ductile metal powders involves simulation module, the
simulation of the
compaction of an industrial
646 Int J Adv Manuf Technol (2011) 56:631647

In the current work, the parameters describing the cap failure Wu CY, Ruddy OM, Bentham AC, Szantoa M, Bierb W, Fragec N,
surface and the hardening law of the constitutive model have Hancock BC, Best SM, Elliott JA Hartmannb S, Yosibash Z (2008)
(2005) Modeling the mechanical Experimental based finite
been determined. To be able to determine the elastic moduli behavior of pharmaceutical element simulation of cold
as a function of density and compacting pressure by means ofpowders during compaction. isostatic pressing of metal
inverse modeling, several unloading steps should be Powder Technol 152:107117 powders. Int J Mech Sci 50:405
incorporated in the experiment. This was not done in the 421
experiments carried out here, thus calls for further Wu CY, Hancock BC, Mills A,
investigations. It may be desirable also to have a separate Bentham AC, Best SM, Elliott JA Pizette P, Martin CL, Delette G,
(2008) Numerical and Sornay P, Sans F (2010)
experiment for the shear failure part of the model. Another experimental investigation of
research orientation will be the development of a 3D Compaction of aggregated
capping mechanisms during ceramic powders: from contact
simulation module, capable of modeling general non- pharmaceutical tablet laws to fracture and yield
axisymmetric applications. compaction. Powder Technol surfaces. Powder Technol
181:121129 198:240250
Acknowledgements The authors would like to acknowledge the
assistance of Lynn Mason in editing this manuscript. Frenning G (2007) Analysis of Andersson DC, Larsson PL,
pharmaceutical powder compac- Cadario A, Lindskog P (2010)
tion using multiplicative On the influence from punch
hyperelasto-plastic theory. geometry on the stress
References Powder Technol 172:102112 distribution at powder
compaction. Powder Tech
German RM (1994) Powder metallurgy science. Metal Powder Gurson AL, Posteraro RA (1992) 202(13):7888
Industries Federation, Princeton, NJ Yield functions for metal
powders for use in the numerical Peric D (1992) On consistent
simulation of powder compac- stress rates in solid mechanics:
Gurson AL (1977) Continuum theory of ductile rupture by void nucleation and
growth. Part I. Yield criteria and flow rules for porous ductile media. Transactions tion. TMS Conference, San computational implications. Int J
of the ASME. J Eng Mat Technol 99:215 Diego, CA Numer Methods Eng 33:799
817
Shima S, Oyane M (1976) Plasticity theory for porous materials. Int J
Mech Sci 18:285291 Peric D, Owen DRJ, Honnor ME
(1992) A model for finite strain
Chtourou H, Gakwaya A, Guillot
Gurson AL, McCabe J (1992) Experimental determination of yield elasto-plasticity based on
M (2002) Modeling of the metal
functions for compaction of blended metal powder. Adv Powder Metall powder compaction process logarithmic strains:
Part Mater 2:133147 computational issues. Comput
using the cap model. Part II:
Meth Appl Mech Eng 94:3561
numerical implementation and
Abou-Chedid G, Brown S (1992) On the mechanical behavior of metal practical applications. Int J
powder compaction. Adv Powder Metall Part Mater 2:110 Solids Struct 39:10771096 Chtourou H, Gakwaya A, Guillot
M, Hrairi M (1995) Implement-
ing a cap material model for the
Drucker D, Prager W (1952) Soil mechanics and plastic analysis on Erhart T, Wall WA, Ramm E
simulation of metal powder
limit design. Quart J Applied Math 10:157175 (2005) A robust computational
compaction. Net Shape
approach for dry powders under
Processing of Powder Materials,
quasi-static and transient impact
Schofield AN, Wroth CP (1968) Critical state solid mechanics. McGraw AMD-vol. 216, ASME, San
loadings. Comput Meth Appl
Hill, New York Fransisco, CA, pp. 1927
Mech Eng 194:41154134

DiMaggio, FL, Sandler, IS (1971) Material model for granular soils. J Alm O (1983) Mechanical testing
Rossi R, Alves MK, Al-Qureshi
Eng Mech Div 97:935950 of powders and powder
HA (2007) A model for the
compacts. Scand J Metall
simulation of powder compaction
Sandler IS, Rubin D (1979) An algorithm and a modular subroutine for processes. J Mater Process 12:302311
the Cap model. Int J Numer Anal Methods Geomech 3:173186 Technol 182:286296
Weber GG, Brown SB (1989)
Falgon D, Vidal-Salle E, Boyer JC, Peczalski R, Andrieu J (2005) Simulation of the compaction of
Khoei AR, Azami AR, Azizi S
Identification procedure of a hardening law for powder compaction. powder components. Adv
(2007) Computational modeling
Powder Technol 157:183190 Powder Metall Part Mater 1:105
of 3D powder compaction
118
processes. J Mater Process
Michrafy A, Ringenbacher D, Tchoreloff P (2002) Modeling the Technol 185:166172
Simo JC, Ortiz M (1985) A unified
compaction behavior of powders: application to pharmaceutical
approach to finite deformation
powders. Powder Technol 127:257266 Lee SC, Kim KT (2007) A study elastoplasticity based on the use of
on the Cap model for metal and hyperelastic constitutive equations.
Cunningham JC, Sinka IC, Zavaliangos A (2004) Analysis of tablet ceramic powder under cold Comput Meth Appl Mech Eng
compaction. I. Characterization of mechanical behavior of powder and compaction. Mater Sci Eng, A 49:222245
powder/tooling friction. J Pharm Sci 93:20222039 445 446:163169
Simo JC, Taylor RL (1986) A return mapping algorithm for plane stress thesis, University Joseph Fourier-
elastoplasticity. Int J Numer Methods Eng 22:649670 Grenoble I, Grenoble, France
Pavier E, Doremus P (1996)
Mechanical behaviour of a
Simo JC, Ju JW, Pister KS, Taylor RL (1988) Assessment of Cap model: lubricated iron powder. In: Cadle Gelin JC, Ghouati O (1995) An
consistent return algorithms and rate-dependent extension. J Eng Mech TM, Narasimhan KS (eds) inverse method for material
114:191218 Advances in powder metallurgy parameters estimation in the
and particulate materials1996. inelastic range. Comput Mech 16
Hofstetter G, Simo JC, Taylor RL (1993) A modified Cap model: closest Metal Pow-der Industries (3):143150
point solution algorithms. Comput Struct 46(2):203214 Federation, USA, pp 640
Mahnken R, Stein E (1996) A
Mosbah P (1995) Modeling and unified approach for parameter
Chtourou H, Gakwaya A, Guillot M (2002) Modeling of the metal
experimental study of metal identification of inelastic material
powder compaction process using the cap model. Part I: experimental
powders behaviour during models in the frame of the finite
material characterization and validation. Int J Solids Struct 39:1059 element method. Comput Meth
1075 comaction in closed die. Ph.D.
Appl Mech Eng 136:225258
Int J Adv Manuf Technol (2011) 56:631647
647

Levenberg K (1944) A method for the solution of certain nonlinear MPIF Standard 43 (1991).
problems in least squares. Quart Appl Math 2:164 168 Method for determination of
hardness of powder metallurgy
Marquardt DW (1963) An algorithm for least squares estimation of products, Princeton, NJ
nonlinear parameters. SIAM J Appl Math 11:431441

Hrairi M (1999) Modlisation Numrique et Identification Directe et Inverse


du Comportement des Poudres Mtalliques lors du Procd du Compactage.
Ph.D. Thesis, Laval University, Qubec

Abaqus (1995) Theory Manual Version 5.5. Hibbit, Karlsson and


Sorenson, Inc., Rhode Island

Samuelson P, Bolin B (1983) Experimental studies of frictional


behavior of hard metal powders sliding on cemented carbide walls.
Scand J Metall 12:315322

Khoei AR, Biabanaki SOR, Vafa AR, Taheri-Mousavi SM (2009) A new


computational algorithm for 3D contact modeling of large plastic
deformation in powder forming processes. Comput Mater Sci 46:203
220

Guillot M, Chtourou H (1996) Generalization of the Vickers hardness


local density measurement technique to different powder materials.
Proceedings of the World Congress on Powder Metallurgy and
Particulate Materials, vol. 1, part 4, pp.3140

SDRC Corporation (1994) Ideas Master Series Student Guide

Hibbitt K (1997) ABAQUS theory manual, Version 5.7, Sorensen, Inc.,


Boston, USA

Hibbitt K (1997) ABAQUS-post user guide, Version 5.7, Sorensen, Inc.,


Boston, USA

Chtourou H, Gakwaya A, Guillot M, Hrairi M (1996) Finite element


simulation of the rigid die compaction of metal powder components.
Proceedings of the Canadian Societyof Mechanical Engineers Forum.
McMaster University, Hamilton, Canada

ASTM E92-82 (1987) Standard test method for Vickers hardness of


metallic materials