Journal of Materials Processing Technology 173 (2006) 101110

Optimisation of springback predicted by experimental and numerical

approach by using response surface methodology
R. Bahloul a, , S. Ben-Elechi b , A. Potiron a
a LPMI-ERTGI, ENSAM, 2 Boulevard du Ronceray, 49035 ANGERS, France
b Laboratoire de Mecanique Roberval, Universite de Technologie de Compiegne, 60205 Compiegne, France
Received 7 March 2005; accepted 3 November 2005

Abstract
Bending has significant importance in the sheet metal product industry. Moreover, the springback of sheet metal should be taken into consideration
in order to produce bent sheet metal parts within acceptable tolerance limits and to solve geometrical variation for the control of the manufacturing
process. Nowadays, the importance of this problem increases because of the use of sheet-metal parts with high mechanical characteristics (high
strength low alloy steel). This work describes robust methods of predicting springback of parts in 3D modelling subjected to bending and unbending
deformations. Also the effects of tool geometry in the final shape after springback are discussed. The first part of this paper presents the laboratory
experiments in wiping die bending, in which the influence of process variables, such as die shoulder radius, punch-die clearance, punch nose radius
and materials properties are discussed. The second part summarises the finite element analysis by using ABAQUS software and compares these
results with some experimental data. The final results of the FEM simulation are in good agreement with the experimental ones. An optimisation
methodology based on the use of experimental design method and response surface technique is proposed in the third part of this paper. This makes
it possible to obtain the optimum values of clearance between the punch and the die and the optimum die radius, which can reduce the springback
without cracking and damage of product.
2005 Elsevier B.V. All rights reserved.

Keywords: Sheet metals; Wiping die bending; Springback; FEM simulation; Optimisation; Experimental design; Response surface technique; Damage

1. Introduction eters in each method of springback reduction by applying a

In the bending process, when the forming forces have been
removed, the metal tries to return to its original shape and results
in phenomenon called springback. Springback is a function of
both the material properties and die configuration; the greater
the strength lower the elastic modulus of material, and the
larger the bend radius and die gap, the greater the springback
will be [1]. Chou and Hung [2] established the relationship in
U-channel bending processes between the springback and the
forming parameters through finite element simulation for the
following techniques of bending: arc bottoming, pinching die
and spanking. An optimisation analysis was coupled with the
finite element analysis to find the optimum forming parame-
ters for each springback reduction technique. And the contour
surface plots are used to observe the sensitivity of each param-
Corresponding author. Tel.: +33 2 41 20 73 24; fax: +33 2 41 20 73 20.
E-mail address: riadh.bahloul@angers.ensam.fr (R. Bahloul).
response surface methodology. Li et al. [3] studied the effect
of the material-hardening mode on the springback simulation
accuracy of V-free bending. The springback is analysed by
using a self-developed 2D elasto-plastic finite element program.
A linear-hardening model and an elasto-plasic power-exponent
hardening model are adopted in their study. The results show
that the material-hardening mode directly affects the springback
simulation accuracy, and greater the veracity of the hardening
mode, greater the springback accuracy. Samuel [4] described a
robust method of predicting sprigback and side wall curls in 2D
operations under plane strain stretching, bending and unbending
deformations. A finite element (FE) program has been used to
analyse the sheet metal axisymmetric U-bending process. Also
the effects of tool geometry and blank holder force on the final
shape after springback are taken into account. The accuracy of
the model is verified by comparison with FE of MARC package
and experimental results. A process optimisation in stamping
was developed by Wu and Altan [5] in a case of flanging a
clutch hub from steel plate. A number of design modifications

0924-0136/$ see front matter 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2005.11.009
102 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110

material type as the main variables. The experimental investiga-

Nomenclature tions presented on the effect of process variables on springback
in straight flanging process. In the second part of Livatyalis
C clearance
work [9], numerical models by FEM was developed by using
Cel Hooke elastic operator
ABAQUS and DEFORM softwares. The principal objective of
D damage variable
this analysis consist of predicting the springback for flanging
Dc critical damage at fracture
and compares these results with some experimental data. Fur-
E Youngs modulus
thermore, flanging with coining to reduce springback has been
f yield function coupled with damage
investigated and evaluated by FEM for better dimensional con-
Rd die radius
trol in flanging and hemming processes. Finite element analysis
Rp punch radius
with 2D can be used to predict springback, bendability and
Rh blank holder radius
tools loads accurately, but the analysis can take several hours,
Sij components of the deviatoric stress tensor S
and pre- and post-processing are even more time consuming.
t thickness of sheet
For this objective, Livatyali et al. [10] proposed a method to
n strain hardening exponent
solve this problem using advanced bending theory, implement-
Greek letters ing the geometric details of the straight flanging process to a
bending angle mathematical model and computerised numerical analysis. The
ij Kronecker delta proposed method predicts springback and tool loads, particularly
D threshold logarithmic strain at which Lemaitre the minimum pad force necessary. Esat et al. [11], used commer-
damage initiates cially available FEA software to analyse bending and springback
el elastic part of the strain tensor of different aluminium materials with different thickness. The
eq equivalent plastic strain amount of springback in terms of springback factors and ver-
ij strain tensor components tical displacement, total equivalent von Mises are presented.
pl plastic part of the strain tensor The FEA results are compared to empirical data. Several studies
R logarithmic strain value at fracture have been carried out on different aspects of finite element anal-
tot total strain tensor ysis of springback. Karafillis and Boyce [12], studied tooling
plastic multiplier factor design in sheet metal-forming compensating for springback and
e , e Lame constants developed a so-called force descriptor algorithm based on calcu-
friction coefficient lating the traction distributions on the deformed sheet in order to
Poissons ratio optimise the tool design. They also examined [13] the manufac-
eq equivalent von Mises stress turing problems in sheet metal-forming processes and proposed
el yield strength a method for tooling and binder design to obtain the desired part
0 isotropic non-linear hardening law shape. Shu and Hung [14] studied finite element analysis and
H hydrostatic pressure the optimisation of springback reduction using the double-bend
ij stress tensor components technique. Pourboghrat and Chu [15] presented a method for
n stress tensor at increment n predicting springback and side-wall curls in two-dimensional
n normal stress plane strain draw bending operations, using moment-curvature
n+1 Gauchy stress tensor at increment n + 1 relationships derived for sheets undergoing plane-strain stretch-
nT elastic prediction of the stress tensor components ing, bending and unbending deformations using finite element
at increment n analysis software.
f friction shear stress This paper describes robust methods of predicting springback
of parts in 3D modelling subjected to bending and unbending
deformations. The effects of process variables (die radius value
and punch-blank clearance) upon the final shape after unloading
have been evaluated to determine the optimum parameters for are studied in details. For convenience of presentation, a series
the selected process conditions. The results have been validated of experimental work was conducted and the results from simu-
with observations of the actual process. lation model are compared with those obtained by experimental
Frequently, the bending process such as wiping-die bending studies. The combination of process parameters are analysed by
(Fig. 1) induces manufacturing problems primarily due to the adopting optimisation methods based on the use of experimental
effect of process variables on springback. The influence of the design method [16] and response surface methodology [1719].
tools, particularly the bending die design has been detailed in That makes it possible to obtain the optimum values of clear-
several works [6,7], mentioning that it affects highly the bending ance between the punch and the die and the optimum die radius
angle value. Various other approaches were developed in order which can reduce the springback without cracking and dam-
to predict and to eliminate the springback in wiping die bending. age of product. The contour plots of response surface are very
In the first work of Livatyali in 2001 [8], a certain number of useful and consequently represented to observe the sensitivity
bending experiments were carried out. They included die corner of each design variable on the springback evolution in order to
radius, punch-die clearance, punch nose radius, pad force and reduce it.
 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110 103

Fig. 1. The process sequence photographed during the wiping die bending experiments of high strength low alloyed steel (S500MC). (a) Initial position, (b)
intermediate phase, (c) Springback.

2. Objectives and approach ratio . D is the threshold at which damage is initiated,

R is the logarithmic strain value at fracture and eq the log-
Based on numerical simulations and experimental observa- arithmic plastic strain. Dc is the critical damage value at
tions, develop a process design for the wiping die bending of fracture.
sheet metal. The objective of this study is the characterisation of Identification of this model consists of the quantitative
influence of certain geometrical bending parameters on the final evaluation of the five coefficients Dc , R , D , and n char-
geometry of security parts used in the automotive sector and acteristics by mean tensile tests [20,21].
their service behaviour. Experimental investigations are used (d) The integration of the von Mises yield function coupled with
to determine the influences of die corner radius, punch (nose) damage given by:
radius and punch-sheet clearance on springback in wiping die
bending. Finite element analysis of these experiments is carried f = eq (1 D) (el + 0 ) (4)
out by using ABAQUS/Standard software. In order to optimise where el is the yield strength obtained by a tensile test
the process, a full factorial design is established which takes and 0 is the non-linear hardening law. eq is the equivalent
into account all the possible combinations of these geometrical stress given by
parameters. The effect of the most unfavourable conditions of 
bending such as negative clearance and small die radii leading 3
to cracking of specimens are studied by numerical and experi- eq = Sij Sij , (5)
2
mental approaches.
where Sij are the components of the deviatoric stress tensor
S expressed by
3. Lemaitres damage model
Sij = ij 13 kk ij . (6)
From a numerical point of view the problem consists of
resolving the constitutive equations describing the behaviour of (e) The plastic flow deduced from the Prandtl-Reuss rule as
the sheet. The set of equations defining the problem are derived f
pl
from: ij = , (7)
ij
(a) The decomposition of the deformation into an elastic part where is the plastic multiplier factor, which can be
and a plastic part: obtained by satisfying the consistency and normality condi-
tions. This requires that f = 0 and the stress state remains on
tot = el + pl (1)
the yield surface during plastic deformation. Thus
(b) The elastic law is coupled with Lemaitre damage law:
f = eq (1 D)(el + 0 ) = 0. (8)
ij = (1 D)(e ij el
kk + 2el
ij ), (2)
The plasticity model is completed by describing the evolu-
where e and e are Lames constants and ij is the Kro- tion of the state variables with continuing plastic straining,
necker delta. ij is the Gauchy true stress components. we can write these equations as
(c) The integration of the damage law:

dH = h (dij , ij , H , D), (9)
  
Dc 2 H 2
D = (1 + ) + 3(1 2) where H , = 1, 2, . . ., n is a set of scalar state variables and
R D 3 eq h is the hardening law for H .
(eq )2/n deq . (3) The integration methods of the non-linear constitutive
equations are based on the use of algorithms which solve
This model depend upon material constants for damage the equations in incremental form. For this purpose, dur-
properties, the hardening exponent n and the Poissons ing a small time interval [tn , tn+1 ], it is assumed that the
104 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110
whole increment is purely elastic, then an elastic prediction
is defined as
T Material
n+1 = n + = (1 Dn )Cel [(tot )n+1 (pl )n ]. (10)
The superscript ()T refers to a trial test.
If this elastic predictor satisfies the yield condition: f < 0,
the prediction is true and the local procedure is completed.
Then it can be stated that:
T
n+1 = n+1 . (11)
Otherwise, this state must be corrected by means of a plastic
correction defined as developed as follow:
The variables at tn+1 must satisfy the yield condition, i.e.
the evolution laws written in the incremental form and elas-
ticity law must satisfy the system:
f = 0, (12)
n+1 (1 Dn+1 )Cel [n + pl
n ] = 0,
pl
(13)
H = h ( ij , (ij )n+1 , Dn+1 ). (14)
The Newton method is used to solve the non-linear global
equilibrium equations as well as the non-linear local equa-
tions obtained by the fully implicit integration of the con-
stitutive equations. The consistent local tangent modulus
is obtained by the exact linearisation of the algorithm. In
this way, an implicit algorithm has been developed which
allows for the integration of the constitutive equations. The
scheme obtained seems well suited to the non-linearity of
the behaviour law [22].
4. Experimental procedure
In the literature, the springback is one of phenomena most treated in sheet
metal forming process and particularly in sheets bending. It is primarily due to
the redistribution of stresses in material when the tools are removed. It can be
distinguished by three classes of parameters which act on geometrical quality
and the final shape of manufactured parts:
Geometry of the bent part (ratio Rd /t).
Adjustment parameters of the press (punch-die clearance, punch stroke, tools
geometry. . .).
Material properties (Youngs modulus, the formability of sheet. . .).
This study concerns the wiping die bending of specimens on mechanical
press Grimar-13675 among a bending line and by using an instrumented punch-
die system. The tool is connected to a signal acquisition and processing system
which produces a curve for each bending operation, displaying the force as a
function of the real punch travel along the sheet. Each specimen was bent to
approximately 90 . The bending experiments are illustrated in Fig. 1. In this
work, similar experiments are performed for a type of specimens with oblong
hole. They are cut out in transverse direction (90 -direction). They make it
possible to characterise the influence of the bend orientation on springback after
punch removal. Lubrication of the part and the tools before bending operation
was used with a full-strength oil [8]. By scaling these photographs, phenomena,
such as increase in punch-die clearance, punch radius, sheet thinning could be
visualised and measured. The following relations were investigated:
Influence of clearance (C) on springback;
Influence of die radius (Rd ) on springback;
Influence of punch radius (Rp ) on springback;
Table 1
Major mechanical properties of S500MC steel
Lematres model
E (MPa) y (MPa) K (MPa) n D R Dc DR
2.10E5 0.28 560 800 0.745 0 0.35 0.21 0.99
Fig. 2. Initial (a) and final (b) steps of bending tests of considered specimens.
Mechanical properties of the tested material are provided in Table 1. During
the preliminary phase of the experiments, the machine tool programs, and tool
set-up and measurement methods were established. The actual experiments were
conducted with a range of punch die-clearance C varying from 0.6 mm to
+0.6 mm and die radius having as value 1, 2, 4, 6 mm. The punch radius and the
stroke are maintained constant equal to 4 and 29 mm, respectively.
After having bent all the specimens, visual observations and measurements
of the geometry were made to evaluate springback thanks to projector of profile.
This operation enabled the real angle of the part to be measured which represents
a manifestation of the existence of residual stresses in material. The cut specimen
used to examine the effect of process parameters on springback phenomenon is
given in Fig. 2.
5. Numerical simulation and springback prediction
The influence of the springback prediction on the behaviour
of sheets and the final geometrical quality of parts is a signifi-
cant aspect of modelling. It will be clarified by a comparison
between the experimental results and those determined by a
numerical simulation. The problem studied in this work consists
of three-dimensional numerical simulation of specimens by wip-
ing die bending process allowing a representation of reality of
springback prediction. The numerical simulation of the damage
evolution has been modelled by means of continuum damage
mechanics approach to describe the behaviour of the sheet in
production bending. Specimens designed with oblong holes are
retained for a better modelling. The successive geometrical con-
figurations of the part during the bending operation are presented
in Fig. 3.
The sheet material is considered as elastic-plastic object
while the punch, die and blank holder are defined as rigid bod-
ies. In these simulations the meshing of the model has been
performed using linear cubic elements (C3D8) showed to be suf-
ficient to describe the bending operation. The effects of punch
velocity are neglected. The friction contact between sheet and
tools follows Coulombs law:
f = n (15)
where f is friction shear stress, n is normal stress at interface,
and is the friction coefficient.
 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110
Fig. 3. Numerical approach: various simulation sequences of wiping die bending. Appearance of springback phenomenon.
The value of coefficient of friction between the punch and the
part surface is equal to 0.09 and the contact at the interfaces of
die, sheet and blank holder is modelled by a value of 0.15. The
goal of this application is to show the aptitude of this model to
predict the springback and to study the influence of certain geo-
metrical parameters related to the process on this phenomenon.
This makes it possible to know on which technological param-
eters is necessary it to act in order to increase or reduce the
springback.
5.1. Inuence of die radius on springback
Fig. 4 shows the effect die radius Rd on the evolution of
springback after bending. Many experiments were carried out on
specimens cut in transversal direction (90 -direction), by main-
taining punch-die clearance equal to the thickness. The treatment
of experimental and numerical results leads to a better com-
prehension of mechanical behaviour and geometrical quality of
specimens after bending. It can be noted on Fig. 4 that a large die
radius favours a significant increase of the final bending angle
compared to the desired one. The loading which the sheet under-
goes on the bent zone strongly affects its geometrical properties
because the transformations of the microstructure during bend-
ing operation depends on the conditions retained. For vertical
bending force applied by the punch, a smaller die corner radius
rquires more deformation force. This can be explained by the
Fig. 4. Effect of die radius Rd on springback angle in case of constant sheet-die
clearance.
105
larger deformation energy needed to bend the sheet to a smaller
radius.
The results obtained by two approaches make it possible to
give the relative variation of springback compared to the exper-
iment and expressed by: (%) = [( exp num )/ exp ] 100.
They are 9% and 10.2%, respectively for die radii of 1 and 6 mm.
Consequently, the springback evolution curve determined by the
numerical approach is thus in good agreement with the experi-
mental one.
In case of small die radius equal to 1 mm, the springback is
not significant any more because there is an almost total rupture
of specimens by cracking as shown in Fig. 5. The section of
resistant material is very decreased, which explains the reduction
of springback. So it can be concluded that smaller the die radius,
the less will be the springback.
5.2. Inuence of punch-blank clearance on springback
For the same test conditions, experiments have been con-
ducted on specimens by always keeping the same cutting direc-
tion, but by varying sheet-punch clearance C in the interval
[0.6, 0.6] mm with uniform increment of 0.2 mm. A total
35 experiments was carried out which include five tests for
each clearances. Fig. 6 shows the evolution of springback
predicted by numerical and experimental approach at vari-
ous clearances. More the distance between punch and blank,
more the springback after bending which is confirmed and val-
idated by these experiments. A smaller clearance produces a
larger maximum strain in sheet metal part, which decreases the
Fig. 5. Cracking of bent specimen for bending radius Rd of 1 mm.
106 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110

Table 2
Springback values in (degrees) determined experimentally for various die radii
and clearances
Rd (mm) Clearance (mm)

0.6 0.4 0.2 0 0.2 0.4 0.6

1 2.5 2.666 2.783 3 3.633 4.016 4.466

2 2.966 2.95 3 3.25 3.866 4.366 4.833
4 3.033 3.13 3.2 3.566 4.416 5.116 6.05
6 3.2 3.216 3.216 3.916 4.8 5.8 6.65

6. Optimisation problem

6.1. Design of experiments

Fig. 6. Effect of punch-blank clearance C on springback angle in case of constant
die radius.
relative influence of elastic deformation. The curves deduced
from the two approaches have the same trend. They evolve sim-
ilarly in a non-linear way.
As shown in Fig. 6, little variation of springback occurs
for negative clearances varying between 0.6 and 0.2 mm in
which the conditions of friction between the punch and sheet are
very severe. For the same die radius, it seems that the variation
of the bending angle is marked when the ratio t/(t + C) is lower
than 1. t and C represent the sheet thickness and the clearance,
respectively. A relative variation of springback compared to the
experiment is equal to 22% can be noted for highest clearance
value. For negative or zero clearances the friction will be one
of the causes of increase of loads allowing to a marked stress
state especially at the interface between punch and blank. Fig. 7
shows the state of bent specimens subjected to friction gener-
ated by punch. For die radius of 1 mm, Fig. 7a corresponds to
positive clearance C of 0.6 mm and Fig. 7b represents a bending
test with negative clearance of 0.6 mm.
Especially, for specimens with oblong hole, localisation
zones of damage and cracks are detected at the rounded-offs
of oblong hole. The same phenomenon was observed for other
die radii. On the marked zone there will be significant reduction
of the thickness of specimen and the appearance of cracks, as
result of punch die clearance lower than the thickness.
A set of data points where the response y is calculated or
measured is required to obtain the response surface. It is desir-
able to select a set of data points that maximises accuracy of the
approximation for given number of data points. The process of
selecting a set of the data points is known as design of experi-
ments (DOE) [16]. The main objective of this investigation was
to study the effects of interaction of tools geometrical param-
eters in sheet metal forming process by wiping die bending in
order to optimise the process. Experiments and simulations were
conducted for an extensive combination of die radius values Rd
and punch-blank clearance values C. A full factorial design of
tow factors is carried out by varying the design parameters and
analysing the response related to the springback. The experi-
mental and numerical procedure had needed 28 cases of tests
in press tools and numerical simulation, respectively: four die
radius [1, 2, 4, 6] mm seven clearance values [0.6, 0.4,
0.2, 0, 0.2, 0.4, 0.6] mm. The experimental and numerical
response of springback according to these process parameters
are reported in Tables 2 and 3 in the form of experiment matrix.
6.2. Response surface methodology
6.2.1. Curve tting
Response surface methodology is applied to obtain an
approximation to a response function in terms of predictor vari-
ables. The response model is generally written as:
y = F (x1 , x2 , . . . , xn ) + , (16)

Fig. 7. Crack formation in the bending zone and marks of friction for Rd = 1 mm. (a) C = 0.6 mm. (b) C = 0.6 mm. (c) Damage contour at the bent zone.
 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110 107

Table 3 6.2.2. Analytical formulation of objective function

Springback values in (degrees) determined numerically for various die radii and The aim was to study the influence and the interaction
clearances
between punch-sheet clearance C and the die radius Rd on
Rd (mm) Clearance (mm) the variation of springback at the end of bending operation.
0.6 0.4 0.2 0 0.2 0.4 0.6 The response surface methodology is used to construct global
approximation of the response at various points of design space
1 2.3 2.45 2.504 2.73 3.142 3.446 3.732
2 2.539 2.655 2.731 3.138 3.061 3.789 4.249
and to compare between experimental and numerical approach.
4 2.786 2.892 2.965 3.119 3.886 4.453 5.149 In the experiment matrices, the physical variables are stan-
6 2.721 2.954 2.843 3.517 4.203 5.046 5.799 dardised and reported to adimensional values which makes it
possible to generalise the method with all types of variables
and to compare their respective influence on the responses stud-
where y is the response, xi (i = 1,2, . . ., n) are predictor variables, ied. For that springback angle is reported to the desired angle
and is the error term. If F is a model that exactly describes the ( = 90 ).
physical process being modelled, may be considered to rep-

resent random error resulting from numerical and experimental (%) = 100 (23)
noise. The function F is normally selected to be a polynomial. 90
For a quadratic polynomial, F is written as: The design domain is the space framed by the considered vari-
n n
ables J and Rd retained for this study. These two parameters are
  normalised to the thickness t such that the reduced variables are
y = 0 + i xi + ij xi xj , (17)
i=1 i=1,j>i
defined by:
C Rd
where represents unknown coefficients. Let us consider a J = and Rd = (24)
case employing tow variables and a quadratic polynomial. The t t
response surface is expressed as: The approximate function is generally written as follows:

y = 0 + 1 x1 + 2 x2 + 3 x12 + 4 x22 + 5 x1 x2 . (18) y = F (x1 , x2 , x3 , . . . , xn ) + (25)

For ease of notation, let x3 = x12 , x4 = x22 , x5 = x1 x2 . Thereby, In the previous expression y represents the response according
the equation becomes: to the variables xi (i = 1, . . ., n)(parameters of the operation) and
is a term representing the error compared to the exact solution.
y = 0 + 1 x1 + 2 x2 + 3 x3 + 4 x4 + 5 x5 . (19) Here it has been found that the third order approximation can
simply present the tendency of observation surface. Contrary it
The unknown coefficients i (i = 0, 1, . . ., 5) in Eq. (18) are
has been noted that the fourth order polynomial approximation
estimated by a linear multiple regression. The linear multiple
was shown to be reliable and sufficient to fit the experimental and
regression model is rewritten in matrix form as:
numerical data and to represent the evolution of the springback.
Y = X + (20) It can be written in the following form:
n
 n
 n
 n

where y = 0 + i x i + ii xi2 + ij xi xj + ii xi3

y1 1 x11 x12 x1k i=1 i=1 i<j i=1



y2 1 x x2k n n n
21 x22   
Y= , X= ..

.. , + ij xi2 xj + ii xi4 + ij xi3 xj
...
.
.. .. ..
. .

. . i=1,i=j i=1 i=1,i=j

yn 1 xn1 xn2 xnh n 2
+ x + (26)
i=1 i
1 1






2
2 Optimisation is carried out in order to determine the optimal
= . , and = . (21) values of the geometrical parameters of the process, which allow

. ..
.






a minimisation of springback . The multi-objective problem can
be formulated as follows:
n n

where is the error vector. The unbiased estimator b of the Minimise Exp = f (Rp , Rd , J) (27a)
coefficient vector is obtained using the last square error method Minimise Num = f (Rp , Rd , J) (27b)
as
1 with the geometrical constraints:
b = (XT X) XT Y (22)
Rdmin Rd Rdmax (28a)
By obtaining coefficients vector b from Eq. (22), the response
surface is prepared. Jmin J Jmax (28b)
108 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110

{, , , } are the regression coefficients representing the 7. Results and discussion

approximate functions of the polynomials Exp and Num . The
theoretical development allows to minimise the square error err It is worth noting from the following figures that springback
between the real design points yi and the estimated values com- predicted by simulation gives lower values than those mea-
puted by using the fourth order function y. One defines err as sured by experiments for all considered combinations. Globally,
follows: the mean deviation between experimental and numerical val-
 ues varies in the range [8%, 16%] for Rd = 1 mm and in the
 nexp
 range [8%, 14%] for Rd = 6 mm. The maximum error for the
err =  (y yi ) (29) proposed model is still acceptable to obtain results close to real-
i=1 ity. Experimentally, maximum springback is found for highest
The yi represent the nexp experimental values that the function die radius values and highest clearance. The same results have
y will have to approach as well as possible. In this case, the been deduced with numerical simulations: the maximum spring-
method allows to solve a linear system in , , , and (Eq. back is about 6.44% of desired angle of folding reached for
(26)) to evaluate the functions Exp and Num . The proposed [Rd , J] = [1.5, 0.15].
function offers the possibility of the prediction of optimal solu- Minimum springback is obtained for the most severe con-
tion for any values of die radius and clearance included in the ditions designed by less die radius and negative clearances. In
considered ranges without necessity to conduct additional coast this case, important shearing phenomenon becomes inevitable
tests. which leads to decrease in strength of material and then its reac-
The regression coefficients of polynomials of objective func- tion designed here by small relative springback angle.
tions are reported in Tables 4 and 5 below. In the negative domain of clearance, variation of springback
does not seem to be very sensitive to the variation of die radius
but it is very pronounced to relative clearance. Whereas, angular
Table 4
error increases more rapidly in the positive domain of relative
Coefficients of objective function of springback deduced from experimental tests
clearance C/t.
Coefficients Values Fig. 8a2 and b2 represent contour surfaces of relative spring-
0 3.136 back measured and predicted by experimental and numerical
1 1.095 approach. It can be observed that both figures have almost the
2 11.1 same trend.
11 0.035
The slope of springback variation with die radius increases
22 32.396
12 17.603 with the increasing of relative clearance. On other hand, it can
11 0.302 be evoked that the dependency of springback is more marked
22 37.69 when clearance varies than die radius variation. This point is
12 31.221 noted from the non-linear regression of springback designed by
21 52.659
the both response surfaces.
11 0.125
22 1080.80 Globally, map obtained by numerical prediction is coherent
12 11.506 with that of experiment one in the considered ranges of bending
21 71.512 parameters.
15.543 Comparing the response surfaces 8a2 and 8b2 , respectively
to Fig. 8a1 and 8b1 , it can be said that the interpolation function
order is satisfactory to represent the evolution of springback
Table 5
Coefficients of objective function of springback deduced from numerical according to process parameters.
simulation From practical point of view, it should be noted that net-shape
parts can not be obtained easily with good geometrical properties
Coefficients Values
in wiping die bending process for positive relative punch-sheet
0 2.847 clearances. However, it is very advisable to produce bent parts
1 0.769
as soon as possible by adopting low die radius.
2 5.239
11 0.078 Curves representing contours plots of relative springback
22 22.522 were given in Fig. 8a3 and 8b3 . It was shown that numerical
12 1.571 results are in good agreement with experiments. The regions
12 0.092 with less springback are found for small values of clearance
22 25.097
and die radius. They were limited by amplitudes of 3.13% and
12 8.605
21 52.328 2.81%, respectively for measurements and predictions, respec-
11 0.004 tively.
22 944.61 The error between experiments and simulations can be
12 2.740 explained by the contact effect, which is based in the default
21 65.001
Coulomb model. Particularly, when clearance is negative, the
17.417
formulation of contact conditions becomes complicated and
 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110
Fig. 8. Springback results given in form of parametric curves, response surfaces and contours plots. (a1 , a2 and a3 ): experimental approach. (b1 , b2 and b3 ): results
of finite element analysis.
theoretically its not easy to make up friction model that takes into
account all process and environmental parameters such tempera-
ture at interfaces, velocity, lubrication characteristics, contacting
area and surface roughness.
Even if error between experimental and numerical results
exists, it is can be judiciously considered acceptable. Coher-
ence of springback results predicted by Finite Element analysis
and experimental data seems to be very well and reliability of
numerical model is achieved.
8. Conclusions
In this work, examination and comprehension of final geo-
metrical quality of parts after wiping die bending operation are
conducted by part design with oblong hole for a better investiga-
tion of study of safety parts. The springback prediction has been
validated with some laboratory experiments and was performed
on specimens of 4 mm-thickness on high strength low alloy steel
(HSLA). This method was applied on a specific type of flang-
ing in which the edge is coined at the end of the operation. The
parameters related to the process, the die radius Rd and sheet-
punch clearance C were retained in adimensional form. A set
109
of finite element simulations of the process has been completed
using commercial programs ABAQUS.
The influence of die radius and clearance on springback vari-
ation and the zone where cracks appear are compared between
experimental and numerical approaches and sources of error
were discussed.
As a set of discussions given above, the following conclusions
have been made:
Springback after wiping die bending operation is heavily
dependent on die corner radius. It means that large die radius
favours a significant increase of the final bending angle
compared to the desired one. In case of small die radius
equal to 1 mm, the springback is not significant any more
because there is an almost total rupture of specimens by
cracking.
More the distance between punch and blank, more the
evolution of springback after bending operation and con-
firmed and validated by experiments. A smaller clearance
defines a larger maximum strain at the edge of sheet
metal part, which decreases the relative influence of elastic
deformation.
110 R. Bahloul et al. / Journal of Materials Processing Technology 173 (2006) 101110
Numerical simulation coupled to ductile damage is showed
to be satisfactory to predict correctly the failure regions, par-
ticularly, when oblong hole design is retained.
In order to optimise the process, a full factorial design is estab-
lished which takes into account all the possible combinations
of these geometrical parameters. Response surface technology
algorithms are specifically developed for optimisation problems.
That makes it possible to obtain the optimum values of clear-
ance between the punch and the die and the optimum die radius,
which can reduce the springback without cracking, and damage
of product. The results of springback optimisation are given in
form of parametric curves, response surfaces and contours plots
which are very useful and consequently represented to observe
the sensitivity of each design variable on the springback evolu-
tion in order to reduce it. Results from the set-up agree very well
with those of numerical model.
References
[1] N.M. Wang, Efficiency in sheet metal forming, in: Proc. of the 13th
Biennial Congress IDDRG, Melbourne, Australia, 1984.
[2] I.N. Chou, C. Hung, Finite element analysis and optimization of spring-
back reduction, Int. J. Mach. Tools Manufact. 39 (1999) 517536.
[3] X. Li, Y. Yang, Y. Wang, J. Bao, Li. Shunping, Effect of the material-
hardening mode on the springback simulation accuracy of V-free bend-
ing, J. Mater. Process. Technol. 123 (2002) 209211.
[4] M. Samuel, Experimental and numerical prediction of springback and
side wall curl in U-bendings of anisotropic sheet metals, J. Mater. Pro-
cess. Technol. 105 (2000) 382393.
[5] H-C. Wu, T. Altan, Process optimisation in stampinga case study for
flanging a clutch hub from steel plate, J. Mater. Process. Technol. 146
(2004) 819.
[6] Z. Kampus, The influence of the shape of die inlet opening on bending,
Proc. Int. Conf. on Sheet Metal. no. 8 (2000) 497504.
[7] Z. Tekiner, An experimental study on the examination of springback of
sheet metals with several thicknesses and properties in bending dies, J.
Mater. Process. Technol. 145 (2004) 109117.
[8] H. Livatyali, T. Altan, Prediction and elimination of springback in
straight flanging using computer aided design methods. Part 1: exper-
imental investigations, J. Mater. Process. Technol. 117 (2001) 262
268.
[9] H. Livatyali, H.C. Wu, T. Altan, Prediction and elimination of spring-
back in straight flanging using computer-aided design methods. Part 2:
FEM predictions and tool design, J. Mater. Process. Technol. 120 (2002)
348354.
[10] H. Livatyali, G.L. Kinzel, T. Altan, Computer aided die design of straight
flanging using approximate numerical analysis, J. Mater. Process. Tech-
nol. 142 (2003) 532543.
[11] V. Esat, H. Darendeliler, M. Ilhan Gokler, Finite element analysis of
springback in bending of aluminium sheets, Mater. Desig. 23 (2002)
223229.
[12] A.P. Karafillis, M.C. Boyce, Tooling design for sheet metal forming
using springback calculations, Int. J. Mech. Sci. 34 (2) (1992) 113131.
[13] A.P. Karafillis, M.C. Boyce, Tooling and binder design for sheet metal
forming processes compensating springback error, Int. J. Mach. Tools.
Manufact. 36 (4) (1996) 503526.
[14] J-S. Shu, C. Hung, Finite element analysis and optimisation of spring-
back reduction: the double bend technique, Int. J. Mach. Tools. Manu-
fact. 36 (4) (1996) 423434.
[15] F. Pourboghrat, E. Chu, Prediction of spring-back and side-wall curl in
2-D draw bending, J. Mater. Process. Technol. 50 (1995) 361374.
[16] A. Todoroki, T. Ishikawa, Design of experiments for stacking sequence
optimisations with genetic algorithm using response surface approxima-
tion, Compos. Struct. 64 (2004) 349357.
[17] R. Bahloul, Ph. Dal Santo, A. Potiron, Shape Optimisation of Auto-
motive Security-parts Regarding the Influence of Residual Stresses and
Material Damage, in: Proceedings of the Fourth European Congress on
Computational Methods in Applied Sciences and Engineering, Finland,
2004, p. 490.
[18] T. Ohata, Y. Nakamura, T. Katayama, E. Nakamachi, Developement of
optimum process design system for sheet fabrication using response
surface method, J. Mater. Process. Technol. 143144 (2003) 667
672.
[19] S.G. Gilmour, R. Mead, A Bayesian design criterion for locating the
optimum point on a response surface, Statis. Proba. Lett. 64 (2003)
235242.
[20] R. Hambli, Etude experimentale, numerique et theorique du decoupage
des toles en vue de loptimisation du procede. These de doctorat,
ENSAM dAngers, October (1996) (in French).
[21] J. Lemaitre, A continuous damage mechanics model for ductile fracture,
J. Eng. Mater. Technol. 107 (1985) 8389.
[22] Hibbit, Karlsson and Sorensen, Inc, Mechanical constitutive theories,
ABAQUS theory manual (1995). Version 5.5 4.1.1.14.3.4.3.