Journal of Materials Processing Technology 151 (2004) 183191

Blank optimization in sheet metal forming

using an evolutionary algorithm
H. Naceur a , Y.Q. Guo b, , J.L. Batoz c
b

a Lab. Roberval, Universit de Technologie de Compigne, UMR 6066 CNRS, BP 20529-60205, Compigne, France
Lab.GMMS, Faculty of Exact and Natural Sciences, University of Reims, Moulin de la Housse, BP 1039-51687, Reims Cedex, France
c Institut Suprieur dIngnierie de la Conception, 27 Rue dHellieule, 88100 Saint-Di Des Vosges, France

Abstract
In this paper, the authors present a new numerical approach to optimize the shape of the initial blank, which plays an important role on
the quality of the final 3D workpiece obtained by the deep drawing of thin sheets. This new approach is based on the coupling between the
inverse approach used for the forming simulation and an evolutionary algorithm. The preliminary results dealing with the optimization of
the blank contour in the case of a square cup (the Benchmark test of Numisheet93) show the efficiency and the potential interest of the
proposed approach.
2004 Elsevier B.V. All rights reserved.
Keywords: Sheet metal forming; Inverse approach; Topologic optimization; Evolutionary algorithm; Blank design

1. Introduction
The numerical simulation of the stamping process is used
extensively for the analysis and design of industrial parts to
avoid long and expensive experimental try-out procedures.
The proceedings of recent conferences [13,19] contains
pertinent references on this subject. Several incremental
codes have been developed and used for the accurate simulation of the deep drawing of thin sheets. These incremental
implicit or explicit codes are based on shell elements, taking
into account large elasto-plastic strains and the frictional
contact conditions between the tools and the sheet. Some
simplified codes have also been developed for the preliminary design of workpieces and tools [49]. A simplified
one-step method called the inverse approach (IA) has been
proposed and developed by Batoz et al. [4,8,9,11,12] to estimate large elasto-plastic strains using simple shell elements.
Both simplified and incremental approaches allow the
evaluation of the influence of a set of process parameters
on the formability of the stamped part, but if a good combination of these parameters is to be found for numerous
analyses must be done: the manual optimization of parameters appears difficult and tedious. Some studies have been
published on the automatic optimization of deep drawing

Corresponding author. Tel.: +33-3-2691-8441; fax: +33-3-2691-3075.

E-mail address: yq.guo@univ-reims.fr (Y.Q. Guo).

0924-0136/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2004.04.036

process parameters [8,1014] but the subject is still quite

open because of many numerical and technical difficulties
yet to be overcome. The first aspect is to find the best definition of the optimization problem which sometimes can be
multi-objective with severe constraints and a large number
of design variables.
Recently, a simple approach called evolutionary structural optimization (ESO) for the topological optimization of
structures and mechanical parts has been proposed by Xie
and Steven [15,16] and this approach is based on the concept
of progressive removal of inefficient material in a structure,
starting with a simple but large domain. Compared to other
methods of structural optimization such as the method of homogenization [17] and the method of density function [18],
the ESO method is very attractive because of its simplicity
and efficiency. In recent years, the ESO method has been
proved to be very useful in solving several kind of problems
such as section design, shape and topology optimization for
static, dynamic and buckling problems [1618].
In the present work, the authors adapt and evaluate the
ESO method for the shape optimization of the initial blank
in the deep drawing process [20]. In some previous developments, the IA has been successfully combined with a sequential quadratic programming (SQP) algorithm to optimize the
blank contours [8,1012]. However, that procedure requires
the calculation of sensitivities with a very good accuracy;
otherwise, the convergence cannot be assured because of the
highly non-linearities in the deep drawing.

184

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

In this paper, the authors first recall the main assumptions

and the formulation aspects of the AI. Then, they present the
evolutionary optimization procedure: design variables, objective/quality/cost function and functional constraints. Then
this procedure is applied to the shape optimization of the
blank in the deep drawing of a square cup. The data are taken
from the benchmark test of the Numisheet93 Conference.
Some conclusions are drawn from the numerical results.

z,n

uq
z0,n0

xp

xq

up
Z ,W

q0
p0

h0

2. Description of the inverse approach

x0q

In itial
b lank

The IA is based on the a priori knowledge of the final

shape of the workpieces. The final part is meshed with triangular elements and the positions of material points in the
initial flat blank are determined by an iteration procedure
(Fig. 1). The first estimation can be obtained simply by a vertical projection of the nodes onto the horizontal plane, and
then an implicit NewtonRaphson algorithm is used to modify the initial nodal positions in order to satisfy equilibrium
in the final workpiece. The deformation theory of plasticity
based on the assumption of proportional loading is used to
obtain an integrated constitutive law (total Cauchy stresses
as a function of total logarithmic strains). The actions of the
tools are replaced by total (but unknown) nodal forces without taking into account the time dependent contact conditions. The AI (or one-step approach) has been proved very
to be efficient for a fast estimation of the strain distribution
and forming difficulties in the drawn part.

q
p

K n o w n final
3 D w o rkpiece

x0p

k
i

Y ,V

X ,U

Fig. 2. Kinematics of a thin shell in deep drawing.

configurations, respectively.The local Cartesian coordinates

are defined by x = x, y, z. Using Eqs. (1) and (2):



. n 0
.
dxq0 = F01 dxq ,
 p,y ..
F01 = xp,x u
 p,x ..xp,y u
3
(3)



.
.
dxq = Fz dx,
Fz = xp,x + znx ..xp,y + znp,y ..n
(4)
where 3 = z/z0 = h/ h0 is the thickness stretch (assumed
constant through the thickness). The inverse of the deformation gradient tensor in q is then given by
F 1 = F01 Fz1

2.1. Strain definition

dxq0 = F 1 dxq ,

In the IA, only the initial flat blank C0 and the final 3D
shape are considered. The position vectors of a material
point q0 in the initial blank and q in the final part are given
by the following expressions, using a generalized Kirchhoff
assumption (Fig. 2):

It is much simpler to express the tensors F01 and Fz1 in

the local orthogonal basis defined by

xq0 = xp0 + z0 n 0 = xp u

 p + z0 n 0

(1)

xq = xp + zn

(2)

where u
 p is the displacement vector of material point p, n 0
and n are the normals of the middle surface in p0 and p, and
z0 and z are the thickness coordinates in the initial and final
Tool actions

A0

P0

t1 t2

n = 
t1 t2 

(6)

The inverse of the left CauchyGreen tensor can be defined

by
B1 = F T F 1

(7)

The Eigenvalue calculation of B1 can give the two principal

stretches and their directions, so that the logarithmic strains
are then obtained.
(8)

The constant thickness stretch is calculated with the hypothesis of full incompressibility.

U, V

t2 = xp,y ,

[] = [M][ln ][M]T

h, , =? P

Known workpiece

t1 = xp,x ,

(5)

Initial blank h0

Fig. 1. General description of the inverse approach.

2.2. Constitutive equations

B0

Hill criterion of plasticity for anisotropic sheets is expressed in the plane stress state by
=  [P]{} 2 = 0

(9)

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

with


 = x y xy

where is the equivalent stress and [P] the matrix depending

on the average plane anisotropy coefficient r . The hypothesis
of proportional loading (the deformation theory of plasticity)
allows the total plastic strains to be expressed as:
p
{p } = [P]{}
(10)

 
where p =
p [P]{p } is the equivalent plastic strain.
Small elastic strains are given classically by
{e } = [He ]1 {} = [C][]

(11)

Using Eqs. (10) and (11) the total Cauchy stresses are obtained in terms of the total logarithmic strains as
1

1
{}
(12)
[P]
{} = [C] +
Hs
with

2.3. Principle of virtual work and finite element

discretization
The shell element DKT12 described in Fig. 3 is used to
discretize the workpiece [8]. The principle virtual work on
the final configuration C is given by
nelt

e=1

2.3.1. Internal force vector

The internal virtual work for an element is given by the
following expression:


x
 
e
y dv
Wint
x y xy
{} dv =
=
(14)

ve
ve
xy
with

 
   
= e + z ,

1
1
hz h
2
2

(15)

where   contains the virtual membrane and bending

strains, {} is the Cauchy stress vector and the thickness h
is assumed constant by element.
The virtual membrane strains are expressed in terms of
virtual nodal displacements using the strain matrix operator
of the CST element [9]:

  
(16)
e = ux vy uy + vx = [Bm ]{un }
with

   
un = u1 v1 w1 u2 v2 w2 u3 v3 w3

Hs =
p

W=

185

We =

nelt

e
e
Wint
Wext
=0

(13)

e=1

e is the element internal virtual work and W e the

where Wint
ext
element external virtual work due to the actions of the tools.

The virtual bending strains are also expressed in terms of the

virtual nodal displacements using the strain matrix operator
[Bf ] of the plate bending element called DKT6 [9].
The CST membrane element and the DKT6 plate bending
element are combined to define the DKT12 flat shell element (Fig. 3). The element internal force vector in the local
coordinate system x, y, z is thus given by

 e 


[Bm ]T + z[Bf ]T {} dz dA
(17)
Fint =
ve

Since the vertical displacement W and the rotation around

each element side are known, the only unknown global degrees of freedom are the horizontal displacements U and V
at each corner node.
2.3.2. External force vector
The IA is a direct method independent of the deformation
strain paths. Some equivalent forces due to the friction can
be defined for each element under the blank-holder:
f = 2qnt

(18)

where f is the tangential friction force per unit area, qn the

blank-holder pressure, the friction coefficient and t the
unit vector in the sliding direction.
The action of the punch or die on a point is replaced by a
normal force and a tangential friction force. Their resultant
is given by the following expression:

Fig. 3. Shell element DKT12.

P = P n f with n f = 

1
1 + 2

(n t )

(19)

186

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

Here the force intensity P is unknown, but it can be obtained

by the equilibrium at a node k:
k

Fx (U, V)

nx
  

k
k
= 0 (20)
Fyk (U, V)
= nfy
Fint
Fext

f
Fz (U, V) int
nz ext

The projection of the above residual force vector in the n f

direction gives Pk , then the system of Eq. (20) is reduced to
a system with only two unknown U and V at each node.

3. Topologic optimization of initial flat blank

3.1. Introduction
A general method able simultaneously deal with shape
and topology optimization is an open research topic. An important development in this domain is the method of homogenization proposed by Bendse and Kikuchi [17], where the
structure is represented by a model with micro-holes and the
objective is to search for the optimal porosity of the porous
environment while using a criterion of optimality. The use
of a stationary mesh for the design allows to avoid the avoidance of expensive and difficult remeshing. Unfortunately, the
method of homogenization is quite complex mathematically.
In 1993, a simple method for shape and topology optimization, called evolutionary structural optimization
(ESO) was proposed by Xie and Steven [16] and Bendse
and Kikuchi [17]. This method is based on the concept of
a gradual elimination of the material to achieve an optimal
design. The ESO method was used firstly for linear static
problems, then it was generalized to the optimization of
dynamic problems.
The ESO method is simple and intuitive. It uses a model
with a regular mesh to represent the initial design domain.
The optimal shape is one sub-domain of the initial domain.
An iteration of the evolutionary optimization involves a FEM
analysis of the structure and a calculation of indicators to
eliminate some elements in the mesh.
The ESO method can be easily implemented in all FEM
codes. As in the homogenization method, the use of a
fixed mesh in the ESO method leads to an irregular (or
non-smooth) of the blank contour.

Fig. 4. Shape optimization of the addendum surfaces of a drawn piece.

The minimum sheet area under the blank-holder with the

formability constraints is searched for these constraints can
be expressed by the limitations on the thickness variation:
25%

he h 0
+15%
h0

(21)

where h0 and he are the initial and final thickness of an

element.
The plastic work density is naturally chosen as a criterion
for the elimination of elements. This criterion appears quite
appropriate in the deep drawing process since the material
that is subjected to little deformation can be considered useless.
3.2.1. Criterion of the plastic work density
The plastic work density is defined by

p
d=
dp

(22)

Using isotropic hardening with Hollomon stressstrain law

and the assumption of additive deformations:
 1/n

p =

(23)
K
E
with

3.2. Formulation of the optimization problem

= Kn

This section deals with the determination of the optimal shape of the initial blank. The elements under the
blank-holder are considered as design variables. The objective of the optimization is to search for the best geometrical
blanks shape under the blank-holder (Fig. 4) that allows
the obtaining of a drawn part without defects such as rupture or wrinkles. During the iteration process the useful part
of the workpiece remains unaltered: the only surface to be
optimized is the plane surface shown in Fig. 4.

where K is the rigidity coefficient and n is the strain-hardening

exponent. The total equivalent strain is given by

=  [P]1 {}
(25)

From Eq. (23):



 1/(n1)
1

d
dp =
nK K
E

(24)

(26)

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

187

Fig. 7. Thickness variation.

of the following non-linear equation:

 1/n
0
0

=0
K
E

(28)

Integration of Eq. (27) leads to the following plastic work

density:
Fig. 5. Flow chart of the ESO optimization algorithm.

d=
Then the plastic work density can be expressed in terms of
the total equivalent stress:

 1/(n1)


1
1

d=

d
(27)
nK K
E
0
where 0 is the equivalent stress corresponding to p = 0
(before plasticity). This value is determined by the resolution

Fig. 6. Finite element mesh for a quarter of the square cup (5242 triangular
shell elements).

1/n )
(1 + n)( 02 2 ) 2E( 0 ( 0 /K)1/n (
/K)

2E(1 + n)
(29)

The elimination criterion is defined by

d
RR
dmax

(30)

where dmax is the value of the maximal plastic work density

in the workpiece.

Fig. 8. Distribution of the plastic work density.

188

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

Fig. 9. Intermediate configurations obtained during the ESO procedure.

The above equation is applied to each triangular finite

element.
During the optimization procedure, if the criterion (30) is
satisfied, the corresponding elements subjected to little plastic deformation can be removed. As proposed in [15,16,18],
the value of RR can be estimated by

3.2.2. Constraint on the volume

The optimization procedure mentioned above does not
possess a stop test and can therefore continue until the
complete removal of the elements under the blank-holder.
A limitation on the reduction of blank material under the
blank-holder is thus imposed through the test:

RR = a0 a1 SS a2 SS2 a3 SS3



V V0 


 V 0  VR

+ with 0 RR 1

(31)

where SS is the number of design states already found during

the evolution process and a0 = 1, a1 , a2 , a3 , . . . are the evolution coefficients that can be determined by numerical tests.

(32)

where VR is the maximum ratio of the sheet material to be

removed with respect to the volume of the initial blank under
the blank-holder.

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

189

3.3. Algorithm of evolutionary optimization

The ESO algorithm implemented in the REFORM code
(FEM software developed at UTC for the deep drawing simulation by the inverse approach) is presented in Fig. 5.

4. Numerical application
The present numerical application deals with the deep
drawing of a square box with a depth of 25 mm (the Benchmark test of Numisheet93). The geometrical and material
properties are as follows: initial flat blank (before optimization) of 150 mm 150 mm; initial thickness h0 = 0.78 mm;
Youngs modulus E = 206 000 MPa; average anisotropy
coefficient (Lankford) r = 1.77; friction coefficient under the blank-holder = 0.144; total blank-holder force
Q = 19 600 N; Hollomon constitutive law = 567.29
0.2637 MPa.
A quarter of the box is discretized into 5242 DKT12 shell
elements (Fig. 6).
The thickness variation on the part for the initial problem
before optimization (with the contour in Fig. 6) is presented
in Fig. 7. Due to the friction under the blank-holder and
the large initial blank size, a quite strong rate of thinning
(approximately 16%) is observed. The distribution of the
plastic work density is shown in Fig. 8. A strong plastic
work density concentration at the die radius entrance in the
diagonal direction is observed.
The convergence is reached after 66 iterations with an
imposed limit of volume reduction VR = 85% (Eq. (30)).
Fig. 9 shows some intermediate configurations obtained
during the ESO procedure. The final configuration corresponds to a reduction of blank volume of 82% (with respect
to the initial volume of the blank under the blank-holder).
Figs. 10 and 11 represent the curves of the volume reduction and the evolution of the maximal plastic work density,
respectively.

Fig. 10. Volume reduction during optimization.

Fig. 11. Evolution of the maximal plastic work density.

A detailed evaluation of the results shows that the optimal configuration corresponding to the minimal plastic work
density does not coincide with the configuration corresponding to the minimal thickness variation (the last iteration).
These two particular configurations are indicated in Fig. 11.
The optimum contour obtained by the ESO procedure is
then smoothed (Fig. 12) using a B-spline curve in GiD
software. Fig. 13 shows the final workpiece and the corresponding initial blank for the optimized blank using the
Inverse Approach. The thickness distribution is shown in
Fig. 14. A better thickness distribution compared with the
starting case is obtained (Fig. 7) despite (or because of) the
important reduction of blank area.
The distribution of the plastic work density is presented in
Fig. 15. Again the reduction of its maximum value compared
to the initial case (Fig. 8) is observed. Table 1 summarizes
the main results before and after optimization.

Fig. 12. Smoothing of the optimal contour.

190

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

Fig. 13. Final workpiece and initial blank obtained by the IA with the blank optimization.
Table 1
Results before and after optimization
Optimization

++h/h (%)

+h/h (%)

dmax

+V (%)

Before
After

+9.5
+10.2

15.9
13.8

336.6
288.7

0
84.7

5. Concluding remarks

Fig. 14. Distribution of thickness.

Fig. 15. Distribution of plastic work density.

An evolutionary structural optimization (ESO) method

has been developed for the optimization of the contour of the
blank in sheet metal stamping. The method consists simply
in eliminating material elements subjected to a only small
plastic deformation. A criterion based on the plastic work
density was chosen and a limitation on the total eliminated
material was imposed. The present technique needs a fine
mesh.
The ESO method has two important advantages. The first
is that it does not require computations of gradients: only
mesh data (node coordinates, connectivities) and classical
element outputs such as strains, stresses and plastic work
density are needed. The second advantage is that this method
is independent of the forming simulation solver. Although,
the authors used a simplified inverse approach (a one-step
code), it is possible to couple the ESO method with any
industrial simulation code.
Several aspects will be studied in the future, such as the
improvement of the convergence speed (reduction of the
number of iterations), the use of quadrilateral elements to
obtain a smoother contour of the blank and the choice of
other optimization criteria to evaluate their influence on the
final solution.
The authors believe that the ESO method is an attractive
approach for the optimum design of 3D addendum surfaces
in the industrial sheet forming process. As in many other
forming problems, the definition of the best quality functions
with constraints for the optimization of process parameters
remains one of the key issues. This important aspect needs

H. Naceur et al. / Journal of Materials Processing Technology 151 (2004) 183191

close cooperation between the shop floor die engineers and

the software developers to achieve some practical solutions
at the preliminary design stage or at later try-out stages.

References
[1] J.C. Gelin, P. Picart, in: Proceedings of the Fourth International
Conference on Numerical Simulation of 3D Sheet Metal Forming
processes, NUMISHEET99, Besanon, France, 1999.
[2] K. Mori, Simulation of material processing: theory, methods and
applications, in: Proceedings of the Seventh International Conference on Numerical Methods in Industrial Forming Processes,
NUMIFORM2001, Toyohashi, Japan, 2001.
[3] A. Makinouchi, et al., in: Proceedings of the Second International Conference on Numerical Simulation of 3D Sheet
Forming processesVerification of Simulation with Experiment,
NUMISHEET93, 1993.
[4] J.L. Batoz, et al., An efficient algorithm to estimate the large strains
in deep drawing, in: Proceedings of the NUMIFORM89, 1989,
pp. 383388.
[5] J.C. Gerdeen, P. Chen, Geometric mapping method of computer modeling of sheet metal forming, in: Proceedings of the NUMIFORM89,
1989, pp. 437444.
[6] M.P. Sklad, B.A. Yungblud, Analysis of multi-operation sheet
forming processes, in: Proceedings of the NUMIFORM92, 1992,
pp. 543547.
[7] S.D. Liu, A. Assempoor, Development of FAST 3D, A designoriented one-step FEM in sheet metal forming, COMPLAS IV Part
II, 1995, pp. 15151526.
[8] J.L. Batoz, Y.Q. Guo, Analysis and design of sheet forming parts
using a simplified inverse approach, COMPLAS V, Barcelona, Spain,
1997, pp. 178185.
[9] J.L. Batoz, Y.Q. Guo, F. Mercier, The inverse approach with simple
triangular shell elements for large strain predictions of sheet metal
forming parts, Eng. Comput. 15 (6/7) (1998) 864892.

191

[10] H. Naceur, Optimisation de forme de structures minces en presence

de non-linarits gomtriques et matrielles, PhD Thesis, UTC,
1998.
[11] Y.Q. Guo, et al., Modeling and blank optimum design of thin car
panels obtained by sheet metal forming, in: J.L. Batoz, et al. (Eds.),
Integrated Design and Manufacturing in Mechanical Engineering98,
Kluwer Academic Publishers, 1999, pp. 307314.
[12] J.L. Batoz, et al., Design of process parameters in deep drawing
of sheets to improve manufacturing feasibility, in: P. Chedmail,
et al. (Eds.), Integrated Design and Manufacturing in Mechanical
Engineering2000, Kluwer Academic Publishers, 2002, pp. 307314.
[13] O. Ghouati, D. Joannic, J.C. Gelin, Optimization of process parameters for the control of spring-back in deep drawing, in: Proceedings
of the NUMIFORM98, Balkema, 1998, pp. 819824.
[14] T. Ohata, et al., Development of optimum process design system by
numerical simulation, J. Mater. Process. Technol. 60 (1996) 543548.
[15] Y.M. Xie, G.P. Steven, A simple evolutionary procedure for structural
optimization, Comput. Struct. 49 (1993) 885896.
[16] Y.M. Xie, G.P. Steven, Evolutionary Structural Optimization,
Springer-Verlag, Berlin, Germany, 1997.
[17] Y.M. Xie, G.P. Steven, Evolutionary structural optimisation for dynamic problems, Comput. Struct. 58 (1996) 10671073;
M.P. Bendse, N. Kikuchi, Generating optimal topologies in structural design using a homogenisation method, Comput. Meth. Appl.
Mech. Eng. 71 (1998) 197224.
[18] R.J. Yang, C.H. Chuang, Optimal topology design using linear programming, Comput. Struct. 52 (1994) 265275;
O.M. Querin, et al., Comput. Meth. Appl. Mech. Eng. 189 (2000)
559573.
[19] D.Y. Yang, et al., Design innovation through virtual manufacturing,
in: Proceedings of the NUMISHEET 2002, vols. 1 and 2, Jeju Island,
Korea, 2002.
[20] H. Naceur, K. Arar, Blank optimization in sheet metal forming
process using awareness seminar and fenet technology evolutionary
algorithms, Nafems/Fenet Workshops, Session: Product and System
Optimization, Zurich, Switzerland, 2002.