j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 6 ( 2 0 0 8 ) 7882

journal homepage: www.elsevier.com/locate/jmatprotec

The optimisation of shot peen forming processes

T. Wang a, , M.J. Platts b , J. Wu c
a

School of Engineering and Design, Brunel University, Uxbridge UB8 3PH, UK

Department of Engineering, University of Cambridge, Mill Lane, Cambridge CB2 1RX, UK
c Department of Aircraft Manufacturing Engineering, Northwestern Polytechnical University,
Xian 710072, China
b

a r t i c l e

i n f o

a b s t r a c t

Article history:

An optimisation method based on an effective process model is presented. The calibrated

Received 11 August 2006

and veried process model simulates peening on realistic components and provides a non-

Received in revised form

linear relationship between peening parameters and deections. An optimal solution can be

2 November 2007

found by minimising the deviation of the computed deformed shape from the desired shape

Accepted 2 December 2007

subject to certain constraints. This optimal solution of peening parameters can be directly
used for practical operations. An optimisation example for forming a cylindrical shape on
a 76 mm 76 mm 3 mm sample is provided. The experimental results demonstrate the

Keywords:

applicability of the present optimisation method.

Shot peening

2007 Elsevier B.V. All rights reserved.

(Shot) peen forming

Optimisation
Process model
Finite element analysis (FEA)

1.

Introduction

Bombarding the surface of a metal sheet by a stream of small

hard shot with sufcient kinetic energy can form a specic
shape. This process is called (shot) peen forming, which has
been especially used for contouring integral aircraft skin
panels.
The key problem for peen forming processes is how to
determine an optimal design of peening parameters for the
desired shape. However, in practical applications this has
been through trial and error. Tatton (1986) described shot
peen forming in production and emphasized the importance of experience. Some numerical methods that are
heavily dependent upon tests were also developed. For example, VanLuchene and Cramer (1996) presented a numerical
method for predicting peening intensity patterns by assum-

Corresponding author.
E-mail address: [email protected] (T. Wang).
0924-0136/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmatprotec.2007.12.019

ing the peen forming process as a linear nite element

system responding to the peening induced stresses. Empirical equations determined by physical experiments in Boeing
Commercial Airplane Group were relied upon to set up the
relationship between the induced stresses and the peening intensity. As recognized in their paper, Homer and
VanLuchene (1991) stated that the experimental data are
based on peening of unconstrained specimens but assumed
to be applicable for constrained specimens, which may introduce error because in reality various constraints, such as
geometry, jigs (pre-stresses), and peening sequences, are
possible.
Wang (2003) and Wang et al. (2006) presented a process
model and veried it in various experimental conditions with
different constraints. This model has also been discussed and
further developed by Blackwell et al. (2004) of QinetiQ. Using
the calibrated and veried process model, a non-linear rela-

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 6 ( 2 0 0 8 ) 7882

tionship between the peening parameters and deection of

specied points can be set up. A constrained optimisation procedure can consequently be used to nd an optimal solution
according to specied deection. This optimisation method
will be discussed in this paper.

2.

The process model

Up to a certain intensity of coverage the individual peen

impacts can be regarded as acting independently. A number
of discrete impacts can therefore be assumed to act simultaneously and their effects are distributed in a specic plastic
layer. Based on this assumption, a static loading unit is found
to model the macroscopic effect of this shot peening. The
loading unit can be conveniently calibrated from the peening parameters, such as shot radius, mass ow rate and air
pressure. Realistic peening intensities can then be simulated
by multiple application of this loading unit. The number of
loading cycles is in direct proportion to the peening time,
which makes the model applicable to practical applications.
Each loading cycle includes two steps. In the rst step, a temperature prole is applied to the section points through the
thickness of composite shell elements in ABAQUS when all
nodal degrees of freedom are xed. This application of temperature causes the surface layer to be stretched beyond yield,
thus mimicking the effect of peening. After the application of
this equivalent load to induce a plastic layer, the second step
is realised by simply releasing the extra boundary conditions.
This leaves the material in a deformed conguration. More
details of the model can be found in Wang (2003) and Wang et
al. (2006).

3.

The experimental approach

Three milimetre thick aluminium 5251-H22 sheet was used

in the present study. Its proportional limit is about 103 MPa,
Youngs modulus 69 GPa and Poissons ratio 0.33. Uniaxial
tension tests were conducted to provide the stressstrain relationship for the analysis. Cast steel S660 peens were employed
in an air-blast machine. The mass ow rate was kept as a constant 13.67 g/s. The nozzle was kept vertical to the worktable
at a distance of 300 mm.
Unconstrained 20 psi peening on 76 mm 19 mm 3 mm
samples was used for calibration tests. Only the central area
19 mm 19 mm was peened, with the rest protected by masking, because the distribution of peens within this area can
be approximated as a uniform distribution for the peening
machine being used. Arc height was measured by a digital
Almen gauge like the normal Almen test.
As a particular example for the optimisation, a cylindrical shape on 76 mm 76 mm 3 mm samples with a curvature
radius of 600 mm was desired under the experimental conditions. The cylindrical shape was chosen because it provides a
challenge, as it is known that uniform unconstrained peening
normally produces a double-curved shape. A digital dial gauge
was used to measure the deformed shape on the peened surface. As the sample was not absolutely at before peening, the

79

measurement was also conducted before peening. The deection was nally taken as the subtraction of the initial reading
from the nal one.

4.

The optimisation method

4.1.

The assumption

In a one-dimensional problem (one peening area and one

deection as the calibration test), it is found that a non-linear
relationship
w=

at
b+t

(1)

between the peen-formed arc height w and exposure time t

can t both the experimental results and nite element analyses as shown in Fig. 1.
For a multidimensional problem as shown in Fig. 2, a similar relationship can be assumed as follows

wi =

Aij tj
Bij + tj

(i = 1, 2, . . . , n and j = 1, 2, . . . , m)

(2)

where tj is the exposure time (or a variable which is in direct

proportion to it) on the jth peening element only, wi is the
deection of the ith node, and Aij and Bij are the non-linear tting coefcients. This assumption implies that the deection
at the ith node can be simply added by the components contributed by each peening element. This might involve certain
errors because it is known that the peening sequence affects
the deection. However, as shown in Wang et al. (2006), successively peening on two adjacent areas has a variation of about
5% which is neglectable.
As a convenient means, Eq. (2) can be written in the format
of matrices
Kt = w

(3a)

Fig. 1 The deection development of a one-dimensional

problem.

80

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 6 ( 2 0 0 8 ) 7882

Fig. 3 The conguration for an optimisation example.

Fig. 2 The conguration for a multidimensional problem.

(1) Discretise the 2D blank to peening elements and locate the

specic nodes for assessing the deformed shape.
(2) Determine the shot size and air pressure for the peening
machine being used.
(3) Conduct FE analysis on each of the peening elements to
determine the individual Kij .
(4) Assemble the overall matrix K.
(5) Determine the optimal t subject to Eqs. (4) and (5).

or

w1

t
1
K21 K22 . . . K2m
w2

=
...
... ... ... ...

...
...

... ... ... ...

tm

K11

Kn1

K12

Kn2

...

...

K1m

In Step 5 to determine the optimal t , a sequential quadratic

programming algorithm is used. Gill et al. (1981) pointed out

Knm

(3b)

wn

where Kij = (wi /tj ) = (Aij /(Bij + tj )) (i = 1,2,. . ., n and j = 1,2,. . ., m).
In the physical aspects, it means peening to one unit of tj on
the jth element can produce a deection of Kij at the ith node.

4.2.

The optimisation procedure

the solution to Eq. (3a) is generally not

For a given desired w,
unique because n might not be equal to m and the components
might not be independent. Therefore, an optimisation
of w
procedure is needed to nd an optimal solution t . As a forming
process, the most important concern is the accuracy of the
formed shape. The optimisation problem can be given as to
nd an optimal t which minimises the target function

f (t) = Kt w

(4)

subject to specic constraints, for example,

tl t tu
inf
Kt w

(5)

where tl and tu are lower and upper bounds, respectively, is

the maximum allowed error. The optimisation procedure can
be described as follows,

Fig. 4 The computation result for modelling shot peening

on a central area.

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 6 ( 2 0 0 8 ) 7882

81

that optimisation methods have progressed tremendously in

all problem categories and commented that these developments mean that the typical person who wishes to solve an
optimisation problem would not (and, in our view, should not)
start from scratch to devise his own optimisation method or
write his own implementation. Therefore, from this point
of view, it was suggested that available routines should be
selected from high quality mathematical software to solve
practical problems. In the current optimisation work for peen
forming processes, the optimisation functions in MATLAB are
employed. Particularly, a sequential quadratic programming
algorithm implemented in a function, constr, is used to nd the
minimum of a constrained non-linear multivariate function.
For details of this algorithm, see MathWorks (1997).

5.

Results and discussion

For the purpose of forming a cylindrical shape on

76 mm 76 mm 3 mm samples with a curvature radius
of 600 mm, the 2D shape as shown in Fig. 3 is divided into
16 peening elements and 25 nodal points. Only one peening
element was peened by masking others in each increment.
The whole process was done by incrementally peening all 16
areas. The shot size and air pressure were set as stated above.
Process modelling was correspondingly conducted for a
quarter of the model because the sample is symmetric. The
computation result for modelling peening on a central area is
shown in Fig. 4. The overall matrix K is assembled according
to all FEA results.
Referring to Fig. 3, for a cylindrical shape with a curvature
at point A
radius of 600 mm, the component of the desired w
and E is 0 mm, point B and D is 0.9 mm, and point C is 1.2 mm
as shown in solid lines of Fig. 5(a)(e) for Sections 15. If tl and
tu are set to 0 and 60 s and is 0.47 mm, the optimal solution t is shown in Fig. 3 for each peening element. According
to this optimal solution, the experimental results measured
along Sections 15 at 25 points are shown in Fig. 5. The deviation of the experimental results from the desired shape along
Sections 24 is a few percent, while Sections 1 and 5 have a
relatively larger deviation. This could be attributed to the edge
effects.
It is worth noting that 0.47 mm of is the minimum error
that the optimisation procedure can achieve for this tested
case. If is set to a smaller value, the constraints given by Eq. (5)
cannot be satised. This indicates that a slight double-curved
result cannot be avoided by this unconstrained peening and
pre-stressed peening could be further used to shape a more
single-curved shape.

6.

Fig. 5 The comparison of the desired shape and the

experimental results.

Conclusion

As demonstrated in the present work, the optimisation

method based on the effective process model is able to provide a preliminary optimal design of peening parameters for
the desired shape. The process model serves like virtual tests
on realistic components and can produce a more accurate
simulation than previous models. From the modelling results,
a non-linear relationship between peening parameters spec-

82

j o u r n a l o f m a t e r i a l s p r o c e s s i n g t e c h n o l o g y 2 0 6 ( 2 0 0 8 ) 7882

ied for discrete peening areas and deections specied for

nodal points can be set up. On the basis of this relationship,
a constrained optimisation procedure can be used to nd
an optimal solution which can be directly used for practical
operations.

Acknowledgements
This paper is based on a project initially carried out in
Cambridge University Engineering Department, which was
supported by Airbus UK, Cambridge Overseas Trust, and Universities UK. The authors would like to thank Andy Levers and
David Gardiner for their useful discussion during the project.

references

Blackwell, P.L., Grifths, M.D., Ward, T., Gardiner, S., 2004. A

computer modelling capability for shot peen forming. Metal
Finishing News 5 (3), 1214.

Gill, P.E., Murray, W., Wright, M.H., 1981. Practical Optimization.

Academic Press, London.
Homer, S.E., VanLuchene, R.D., 1991. Aircraft wing skin
contouring by shot peening. Journal of Material Shaping
Technology 9 (2), 89101.
MathWorks, MATLAB Optimisation Toolbox. Users Guide Version
5, 1997.
Tatton, R.J.D., 1986. Shot peen forming. In: Meguid, S.A. (Ed.),
Impact Surface Treatment (The Second International
Conference on Impact Treatment Processes, Craneld
Institute of Technology, Bedford, UK). Elsevier, London, pp.
134143.
VanLuchene, R.D., Cramer, E.J., 1996. Numerical modelling of a
wing skin peen forming process. Journal of Materials
Engineering and Performance 5 (6), 753760.
Wang, T., 2003. Numerical simulation and optimisation of shot
peen forming processes. PhD Thesis. University of
Cambridge.
Wang, T., Platts, M.J., Levers, A., 2006. A process model for shot
peen forming. Journal of Materials Processing Technology 172
(2), 159162.