See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330420061

Design optimization of the ram structure of friction stir welding robot

Article in Mechanics of Advanced Materials and Structures · January 2019

DOI: 10.1080/15376494.2018.1471758

CITATIONS READS

24 2,013

5 authors, including:

Haitao Luo Peng Wang

Chinese Academy of Sciences Tsinghua University
101 PUBLICATIONS 576 CITATIONS 18 PUBLICATIONS 78 CITATIONS

SEE PROFILE SEE PROFILE

Jinguo Liu
Shenyang Institute of Automation,CAS
251 PUBLICATIONS 4,200 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jinguo Liu on 10 December 2019.

The user has requested enhancement of the downloaded file.

MECHANICS OF ADVANCED MATERIALS AND STRUCTURES
2018, VOL. 0, NO. 0, 1–11
https://doi.org/10.1080/15376494.2018.1471758

ORIGINAL ARTICLE

Design optimization of the ram structure of friction stir welding robot

Haitao Luoa , Jia Fua, Peng Wangb, Jinguo Liua, and Weijia Zhoua
a
Department of Space Automation Technologies & Systems, State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy
of Sciences (CAS), Shenyang, P.R. China; bInstitute of Modern Transmission and Digital Technology, Northeastern University, Shenyang, China

ABSTRACT ARTICLE HISTORY

This paper proposes a method for designing the ram structure of friction stir welding (FSW) robots Received 17 April 2018
using finite element analysis. Under the given working condition, force analysis for the ram struc- Accepted 22 April 2018
ture is performed. By analyzing the boundary and load cases on the ram structure, we optimize
KEYWORDS
the topology and size. Through ram-structure optimization considering the static and dynamic
Designing; ram structure;
characteristics, we can achieve lightweight design of the ram structure and effectively improve the friction stir welding; robots;
FSW robot’s welding precision. The optimization flow and method can be applied to heavy-load dynamics characteristics;
robots and high-stiffness structure. topology; size; optimization

1. Introduction optimization, which is very generic and mathematically

sound [6]. Based on the early work of Prager and Taylor,
Friction stir welding (FSW) is a kind of solid connection
Venkayya et al. proposed a rule-based method for dealing
method, which has many advantages compared with the
with large-scale optimization problems [7]. In 1977, Fleury
traditional fusion welding due to their attractive joint mech- and Sander showed that the optimal criterion method could
anical properties and environmental friendliness [1]. In add- also serve as the dual problem of mathematical program-
ition, neither welding wire nor welding pretreatment is ming method, and one significant contribution of their
required [2]. Therefore, FSW has widely been adopted in work is to combine two methods of structural optimization
various industries worldwide. [8]. The optimal criterion method has been extended to a
To meet the welding requirements in aeronautics and more general system optimization approach, thereby putting
astronautics, FSW robots are usually huge and complex. If forward the so-called composite beam adjustment to ensure
their components or parts are too heavy, the welding preci- the feasibility of the solution [9, 10]. For structural opti-
sion of FSW robots will seriously degrade because the mization, Schmit formulated the section optimization of
weight has a great impact on the static and dynamic charac- component as a mathematical programming problem and
teristics [3]. Therefore, the design optimization of the major proposed to deal with a variety of structural optimization of
parts of FSW robots is of great practical significance for the load cases using mathematical programming methods [11].
research and development of FSW robots. It has shown that it was feasible to solve structural optimiza-
In order to effectively reduce the weight of FSW robots, tion problems with finite element method. And then, math-
researchers have adopted a number of methods to enhance ematical programming methods were further developed to
its structure performance in accordance with the actual integrate the advantage of optimal criterion methods by tak-
working condition by removing redundant material and ing into account the mechanical properties [12], such as
then adding rib-plates. Based on these measures, the final explicit approximation, variable link, effective constraint
design still cannot resist structural deformation and dynamic selection, inverse variable introduction, and dual solution
flutter. To date, the development of advanced design meth- technology, which led to great improvement of the compu-
ods for the simultaneous optimization of topology and size tation efficiency. In recent years, intelligent optimization
is in strong demand. However, in order to put structure algorithms have also become popular in the ram structure
optimization technology into practice, we not only need to optimization [13–20]. In addition, the constructed method
establish a reliable optimization model, but also an effective based on sampling points is a new kind of modeling
and efficient optimization algorithm. There are a variety of method. For the large-scale optimization problems of engin-
existing methods for structure optimization [4, 5]. The opti- eering structural, entity test is not only expensive but also
mal criterion method is one of the early methods for struc- computationally very intensive. With the development of
tural design optimization. In the 1970s, people took the high speed computers and finite element analysis, numerical
Kuhn–Tucker condition as the criteria for structure simulation technologies have attracted more and more

CONTACT Haitao Luo [email protected] State Key Laboratory of Robotics, Shenyang Institute of Automation, Chinese Academy of Sciences (CAS),
Shenyang 110016, P.R. China.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/umcm.
ß 2019 Taylor & Francis Group, LLC
2 H. LUO ET AL.

attention, as they can significantly reduce the computation

time. Such methods mainly include the response surface
(RS) method [21], radial basis function method [22], aug-
mented radial basis function method [23], and Kriging
method [24].
In this paper, we focus on the optimal design of the ram
structure of an FSW robot with the help of finite element
modeling. Force analysis under the given working condition
is also provided. In order to determine the final optimal
structure of the ram, we establish a finite element model
and optimize the topology and size based on the variable
density method and the external penalty function method
assisted by an approximate RS method. Our optimization
results show that the ram mass decreases greatly while the
end deformation just increases a little. Meanwhile, the nat-
ural frequency of the ram has been considerably increased.
The success of the proposed optimization method indicates
that the proposed ideas are promising, providing a new Figure 1. An illustration of a friction stir welding robot.
means for the lightweight design of the relative structures.
The optimization design on the ram structure of FSW missiles, and fight jets. Such aluminum alloys cannot be welded
robots is carried out in two stages, including topological with the traditional fusion welding techniques; otherwise the
optimization and size optimization described as follows. required performance cannot be achieved, such as aluminum
alloy material AL-7075 and AL-2024. To enhance welding per-
1. In the initial structure design, the ram can be taken as a formance, minimize the weight and increase fatigue strength,
cuboid. The topology optimization can help designers to FSW robots have been developed and became popular.
quickly find the ram’s optimal material distribution under Most existing FSW devices were transformed from numeric-
the given objective function and constraint conditions. ally controlled machine tools with a single function, which lack
Here, the objective function is ram’s minimum flexibility or flexibility and are not suited for complex work piece welding [25].
maximum stiffness and the constraint is volume percentage In the aerospace field, most workpieces are of large-scale and have
or mass fraction. However, topology optimization only is a thin-wall surface, where the weld seam is always represented as a
not sufficient, as the ram structure is just a framework. An complex curve in 3D space. Furthermore, a very welding precision
appropriate lattice-rib needs to be filled in its internal struc- is required [26, 27]. Another challenge in using these welding
ture to further enhance its mechanical properties. devices is the complicated path planning and programming [28].
Meanwhile, topology optimization method also takes the The new FSW robots with seven joint axis co-motions have
installation, debugging and process factors into account been developed to address the above difficulties. As Figure 1
during ram’s structure design. Therefore, topology opti- shows, the FSW robot that we designed consists of X
Y
Z
mization is only the initial stage of ram structure design. axis, A
B axis, a stir-welding head and a rotary table. The main
2. Size optimization focuses on different lattice-rib styles and body of the robot includes X
Y
Z axis, A
B axis, and a stir-
its specific geometry sizes located in the internal structure welding head. The X
Y
Z axis contains a transmission system
of the ram. The optimization goal is to minimize the ratio composed of a ball screw and a linear guide, a gravity compensa-
of total mass and its first order natural frequency. In other tion system consisting of balance weights and a gravity compensa-
words, the lighter the weight and the higher the natural tion system. The A
B axis includes worm and gear driving
frequency of the ram structure, the better. The constraint systems and stir-welding head components. Rotary table (T axis)
is the percentage of the natural frequency before and after has a rotary DOF. The workpiece to be welded is fastened by a
the optimization. By means of size optimization, the stiff- special fixture on the rotary table. The rotary table has a scale
ness of the ram will be further enhanced. Although its mark, precise positing and locking function, which can meet the
structure mass increases slightly, the benefit is significant demands of different working conditions. There are seven DOFs
compared with the static deformation of the ram before in the whole system, where, the X
Y
Z axis has three transla-
and after optimization. That is, size optimization is the tional DOFs, A
B axis has two rotational DOFs, and the stir-
more detailed stage of ram structure optimization. welding head has translational and rotational DOF. The rotational
DOF of the stir-welding head is not relevant for the system model.
The body, upright pillar and slip saddle are casted of cast iron.
2. Load cases about the ram structure of FSW robot However, the main load bearing components, such as the ram, is
made up of alloy steel because of its configuration. The mass of
2.1. Brief introduction of the FSW robot
the main body of the robot is about 71 tons, with an envelope size
In recent years, aerospace industry has posed high demands for of 1.8 m  1.8 m  1.6 m.
advanced welding technologies for high-strength and low weight During the welding process, the ram of FSW robot is
aluminum alloys, which are widely used in launching vehicles, always in cantilever and subject to an enormous force and
 MECHANICS OF ADVANCED MATERIALS AND STRUCTURES 3

force in the XZ or XY coordinate system, respectively,

Txn(n ¼ i, j) is the torque of the angular displacement rela-
tive to node i, j, and Myn or Mzn(n ¼ i, j) is the bending
moment in the XZ or XY coordinate system, respectively.
We can then convert the nodal vector into nodal axial dis-
placement, nodal deflection and torsion angle vector as follows.
8
>
> fug ¼ ½ui uj T
>
< fvg ¼ t h t h T
>
i zi j zj
 T (5)
>
> f g
w ¼ w i hyi wj hyj
>
>  
: fhg ¼ h h T
xi xj

Figure 2. Finite element model of 3D beam element. The displacement mode of the axial displacement u can be
expressed by a linear function, while the deflection
and x can
torque as well as considerable amount of variation in FSW be expressed by a cubic polynomial. The displacement mode of
processing. The static and dynamic performances of the ram the torsion angle h is also expressed by a linear function.
will have a strong impact on the geometric accuracy of the According to the displacement vector of an element node, we
weld seam. Therefore, optimization design of the ram struc- can get the displacement mode of 3D beam element as follows.
ture is of great importance for improving the welding per- T
ff g ¼ ½uvwhx  ¼ Nde (6)
formance of the FSW robot.
In the above equation, N denotes the displacement shape
functional matrix of the 3D beam element.
2.2. Modeling of the ram structure Based on the theories of mechanics of materials, the
The ram structure of FSW robots can be analyzed using the 3D strain and displacement components of each point on the
beam element theory. On that basis, we are able to get various flexible body comply with the conditions of geometric equa-
performance indicators for its static and dynamic characteris- tion. Consequently, the linear strain and shear strain equa-
tics. Each node of the 3D beam element has six degrees of free- tions of the 3D beam element are as follows.
8
dom, where the six directions of the nodal displacements >
> ex ¼ du=dx
<
correspond to the six nodal forces, as shown in Figure 2. ey ¼
ydt2 =dy2
(7)
Under the given configuration, a finite element model of >
> e ¼
zdw2 =dz2
: z
the simplified elastic structure can be created by two nodes c ¼ qdhx =dx
of space beam element. For each space beam element, each
node has six DOFs. Therefore, the nodal displacement vec- Rewriting the above in the form of matrix, we obtain:
tor of the beam element can be expressed as feg ¼ Bde (8)
h iT
de ¼ dTi dTj (1) Since the stress and strain components have a linear rela-
tionship and meet the generalized Hooker’s law, the element
In Eq. (1), stress expressed by the nodal displacement is shown as follows.
(  T
di ¼ ui vi wi hxi hyi hzi frg ¼ De ¼ DBde (9)
 T (2)
dj ¼ uj vj wj hxj hyj hzj In the above equation, D denotes the elastic matrix:
2 3
where ln,
n, wn(n ¼ i, j) denotes the linear displacement of E 0 0 0
node i, j relative to XYZ axis in the local coordinate system, 60 E 0 07
D¼6 40
7 (10)
respectively, hxn, hyn, hzn(n ¼ i, j) denotes the angular dis- 0 E 05
placement of the cross section located on node i, j relative 0 0 0 G
to XYZ axis, respectively, hxn is the twist angle of the cross
section around the X axis, and hyn, hzn is the turn angle of According to the virtual displacement principle of an elastic
body, all the work of external force on the elastic body along
cross section in XZ and XY coordinate planes, respectively.
The corresponding nodal force vector can be described as the virtual displacement is equal to its virtual strain energy.
ððð
follows.
h iT d F¼
T
deT rdxdydz (11)
F e ¼ FiT FjT (3) V

In Eq. (3), Then, we obtain the formula of the 3D beam element

(   between equivalent nodal force and equivalent nodal dis-
Fi ¼ Ni Qyi Qzi Txi Myi Mzi T placement.
  (4)
Fj ¼ Nj Qyj Qzj Txj Myj Mzj T Re ¼ K e d e (12)
where Nn(n ¼ i, j) is the axial force of the linear displace- In Eq. (12), Ke means the element stiffness matrix of the
ment relative to node i, j, Qxn or Qyn(n ¼ i, j) is the shear 3D beam element.
4 H. LUO ET AL.

ðð
K ¼
e
BT DBdt (13)

Assume that A is the cross section area of the 3D space

beam element. In the XZ and XY coordinate planes, Iy and
Iz are the section moment-of-inertia. Jk is the element tor-
sion moment-of-inertia. E and G denote the material elastic
modulus and shear modulus, respectively. According to the
classical beam theory, the space beam element stiffness
matrix relative to the local coordinate system can be
expressed as in Eq. (14)
k¼ Figure 3. Force analysis of the ram structure.
2 3
EA
6 l 7
6 7
6
6 0 12EIz 7
7
conditions, we can see that melon-flap longitude welding is
6 l3 7
6
6 12EIy
7
7 the worst working case, which will be taken as an example
6 0 0 7
6 l3 7 to perform force analysis, as shown in Figure 3.
6 7
6 GJk 7
6 0 7 Under the melon-flap longitude welding condition, the
6 0 0 Sym 7
6 l 7
6 7
6 0
6 0
6ELy

2 0
4EIy 7
7
loads mainly include weight G, the plug-in resistance Fn of
6 l l 7
6 7 the FSW head, feeding resistance Ft, transverse fluctuating
6 6EIz 4EIz 7
6 0 0 0 0 7
6 l2 l 7 force Fn(t), rotating torque Tn, and the pulling force F of
6 7
6
EA EA 7
6 0 0 0 0 0 7
6 l
6 l 7
7
steel wire rope. Moreover, e means the distance between
6 12EIz 6ELz 12EIz 7
6 0
6

3
l
0 0 0

l2
0
l3
7
7 shaft axis of the FSW head and inner side of ram, l1 means
6 7
6 12EIy 6ELy 12EIy 7 the distance between the ram centroid and fixed surface, l2
6 0 0
3 0 0 0 0 7
6 l l2 l3 7
6 7 the distance between the ram centroid and the front-end
6 GJk GJk 7
6 0 0 0
0 0 0 0 0 7
6 7
6
6
l l 7
7
hole axis, l3 the distance between steel wire rope and fixed
6 0 6ELy 2EIy 6ELy 4EIy 7
6 0
0 0 0 0 0 7 surface, n the angular between the FSW head plug-in resist-
6 l2 l l2 l 7
4 6ELz 2EIy 6ELz 4EIz 5
0
l2
0 0 0
l
0

l2
0 0 0
l
ance and the vertical direction, and a and b the length and
width of the ram, respectively. According to the above ana-
(14)
lysis, the six-dimensional forces and toques of the ram can
By combining the element stiffness matrix, we get the be given as follows.
global stiffness matrix K of the 3D beam element. And then
we obtain the relationship between the external load F and 1. X axial force:
the nodal displacement X. In the next, we need to obtain
Nx ¼ Fn sin n þ Ft cos n (18)
the inherent structure characteristics, which is the basis of
the subsequent dynamics analysis. The un-damped free 2. Y axial shear force:
vibration equation of the mechanical structure is as follows.
Qy ¼ Fn cos n
Ft sin n þ 2Ft
G (19)
M€a ðtÞ þ KaðtÞ ¼ 0 (15)
3. Z axial shear force:
In the above equation, M means the global mass matrix,
Qz ¼ Fa ðtÞ (20)
which also contains the element mass matrix. The element
mass matrix expression is as follows. 4. X axial toque:
ðð
Tx ¼
Tn sin ðnÞ
ðe þ a=2Þ½Fn cos ðnÞ þ Ft sin ðnÞ
Me ¼ qN T Ndv (16)
(21)
By converting the above equation into solving the gener- 5. Y axial bending moment:
alized eigenvalue and eigenvector problem, we have
My ¼ Tn cos ðnÞ
ðe þ a=2Þ½Fn sin ðnÞ þ Ft cos ðnÞ (22)
Ku
x2 Mu ¼ 0 (17)
6. Z axial bending moment:
In this way, we can obtain the natural frequency x1,
x2, … ,xn and natural model of vibration u1, u2, … ,un. Mz ¼ ½Fn cos ðnÞ
Ft sin ðnÞðl1 þ l2 Þ
Gl1 þ 2Fl l3 (23)
Through the above force analysis, we can see that the
ram is a biaxial bending part mainly subjected to bending
2.3. Force analysis of the ram structure
moment. In order to comprehensively evaluate its mechan-
FSW robots have five kinds of typical working conditions ical property against terminal load, the principal model
under the actual welding process, such as melon-flap longi- needs to be created and its total structure stiffness needs to
tude welding, melon-top ring welding, melon-bottom ring be acquired. Under the melon-valve welding configuration,
welding, cylinder ring welding and cylinder longitude weld- the elastic deformation of the ram is as shown in Figure 4.
ing. But under each typical working condition, the ram’s Assume that the ram is a constant-section cantilever, and
load is different. Comparing these five typical working point A is a fixed end. Three concentrated forces are loaded
 MECHANICS OF ADVANCED MATERIALS AND STRUCTURES 5

8
>
> C2 ¼ ðU
Fl Þl32
2Vl3
>
>
>
< D2 ¼ ðFl
2U þ 3V Þl3
3
>
3
>
> 1
>
> C3 ¼ ðQy
U Þl12 þ ðU
Fl Þl32 þ Vl1
2Vl3 þ Qy l1 l2
>
> 2
: D ¼ ð2U
Q Þl3 þ ð6V
4U þ 2F Þl3
3Vl2
3Q l2 l
3 y 1 l 3 1 y 1 2

(28)
Based on the deflection line equation, the turn angular
equations of AB, BC, and CD segment can be obtained. In
Figure 4. Bending deformation sketch of ram. Figure 4, the maximum displacement and turn angle in the
endpoint D are expressed as follows, respectively.
from point B to point D. According to the basic theory of 8
> 1 1
< jyjmax ¼ yD ¼ Qy l13
Qy ðl1 þ l2 Þl12 þ C3 l1 þ D3
material mechanics, the shear force and bending moment in 6 2 (29)
the fixed end section can be described by: >
: jhjmax ¼ y0 D ¼ 1 Qy l12
Qy ðl1 þ l2 Þl1 þ C3
 2
FRA ¼ 2Fl
G þ Qy
(24)
MA ¼ 2Fl l3
Gl1 þ Qy ðl1 þ l2 Þ Based on Eq. (29), we may limit the maximum deflection
and turn angle, and predefine them according to the actual
As three concentrated forces are applied to different loca-
demand. In this way, the stiffness conditions of the ram can
tions of the ram, the bending moment equation of the ram
be defined as follows.
beam section can be obtained using the piecewise calculation (
method. The bending-moment equations of AB, BC, and CD jyjmax  ½y
(30)
segments can be written as follows. jhjmax  ½h
8  
>
> Mðx1 Þ ¼ ð2Fl
G þ Qy Þx1
2Fl l3
Gl1 þ Qy ðl1 þ l2 Þ ;
>
< where [y] and [h] are allowable deflection and angle.
ð0  x1  l3 Þ
>
> M ð x 2 Þ ¼ Qy ðl1 þ l2
x2 Þ
Gðl1
x2 Þ; ðl3  x2  l1 Þ
>
: Mðx3 Þ ¼ Qy ðl1 þ l2
x3 Þ; ðl1  x3  l1 þ l2 Þ 3. Ram-structure topology of optimization
(25) 3.1. Topology optimization model
Since the moment equations are different, the corre- According to the specified welding condition, the ram’s load
sponding deflection line equations are also different. In add- case is analyzed. For example, under the melon-flap longi-
ition, the elastic deformation belongs to a small tude working condition, the ram (Z axis) is in cantilever
deformation, so its deflection is far less than its span. configuration. Because of the self-weight and overload from
Therefore, the approximate differential equation of deflec- the FSW head end, the static and dynamic characteristics of
tion line and turn angle equation can be described as fol- the ram will degrade sharply, which severely affect the weld-
lows. ing precision.
8   In order to reduce the weight of the ram and maximize
>
> ÐÐ M
>
<y ¼ dx dx þ Cx þ D the structural stiffness, we discretize the structure into finite
Ely
Ð M (26) elements and determine which position of the material can
>
> ¼ þ
>
: h dx C be removed. Finally, we can get the optimal topology of
Ely
the ram.
In the fixed end A, the deflection displacement and turn Topology optimization has three main advantages
angle are equal to zero. Therefore, the deflection equations [29, 30]:
of AB, BC, and CD segments can be shown as follows.
8 1. it is possible to use the material configuration to opti-
>
> ð2Fl þ U Þx13
3ð2Fl l3 þ V Þx12
>
> yAB ¼ mize the structure performance and composition;
>
>
>
>
6EIy 2. it is able to determine the element choices according to
<
Ux23 þ 3Vx22 þ C2 x2 þ D2 the best force transmission path of materials;
yBC ¼ (27)
>
> 6EIy 3. it can automatically form hole-like internal structure.
>
>
>
> Qy x33
3Qy ðl1 þ l2 Þx32 þ C3 x3 þ D3
>
> yCD ¼
: 6EIy Due to the above reasons, topology optimization in
engineering application is of great practical significance.
To clearly formulate the problem, we introduce two con- Methods for the topology optimization mainly include the
stants, U and V, where U ¼ Qy
G and V ¼ Qy(l1 þ l2)
Gl1. homogenization method, the variable density method, the
By virtue of deflection equation’s continuity condition, the four variable thickness method, evolutionary structural optimiza-
integral constants in Eq. (26) can be determined. The expres- tion method, level set method, independent continuous
sions corresponding to C2, D2, C3, and D4 can be expressed as mapping method, isoperimetric method, and bubble method.
follows. Among them, the variable density method has gained much
6 H. LUO ET AL.

SIMP method, the optimization criteria algorithm is applied

for the minimum compliance objective. In addition, these
criteria are also used to update all design variables.
First of all, a Lagrange function is created according to
the objective function and the constraint conditions defined
above.
L ¼ C þ k1 ðV
V  Þ þ k2 ðF
KX Þ
(33)
þk3 ðqmin
xÞ þ l4 ðq
1Þ
In the above equation, k1,k2,k3, and k4 are Lagrange mul-
tipliers. While k1 is a scalar; k2,k3, and k4 are vectors; q is a
column vector whose elements are qi. When qi takes the
extreme value qi , the Lagrange function meets the Kuhn–
Tucker necessary conditions.
8
>
> @L ¼ @C þ k1 @V
k2 @ ðKX Þ
k3 þ k4 ¼ 0
>
>
>
> @qi @qi @qi @qi
>
>
>
< V ¼ V
Figure 5. The influence of the penalty factor on the relationship between elas-
tic modulus and relative density. F ¼ KX (34)
>
>
> k3 ðqmin
q Þ ¼ 0
>
>
>
interest due to its adjustable punishment factor, high com- >
> k4 ðq
1Þ ¼ 0
>
: q  q  1
putation efficiency and effective and global search ability. min
The variable density method introduces a kind of imaginary
where partial derivatives @(KX)/@qi, @V/@qi, and @C/@qi are
variable density material, which is set as variable. An index
sensitivity of displacement, volume, and objective function,
function of cell density can be used to simulate properties of respectively.
materials. 8   k n k
>
> k n k
EðxÞ ¼ E0 qðxÞp (31) < Ci qi ; qmin< Ci qi <1
n
> qmin ; Cik qki  qmin (35)
In Eq. (31), E(x) denotes Young’s modulus, Eo the initial >
: 1;  k n k
Young’s modulus, q the relative density of element, and P C q 1
i i
the penalty factor. P is also referred to as a driving factor, In the above equation, the Cik ¼ pðqiÞðp
1Þ uTi K0 ui =k1 vi ¼ 1
which drives the element relative density to 0 or 1. P is is the design criterion. n is damping factor, which is used to
related to Poisson ratio as shown Figure 5. ensure the stability and convergence of numerical calculation.
The topology optimization of the ram structure using the The main steps for iterative optimization of the ram top-
solid isotropic material with penalization (SIMP) can be for- ology using the SIMP method can be summarized as
mulated as follows, where structure flexibility is set to be the given below.
objective to be minimized:
Find : q ¼ ðq1 ; q2 ; :::; qn ÞT 2 Rn 1. Define the boundary conditions such as design domain,
Min : CðqÞ ¼ F T X ¼ X T KX constraints and loads. The element relative density in
S:T : F ¼ KX design domain can change with the iteration process.
Xn (32) 2. Discretize the structure into finite element meshes, and
V ðqÞ ¼ qi ti  fV0 ¼ Vmax compute the element stiffness matrix before optimization.
i¼1
0<qmin  qi  1; i ¼ 1; 2; :::; n 3. Initialize the element design variables and specify the
initial relative density of elements in the design domain.
In the equations, q denotes the material relative density; 4. Calculate the material characteristic parameters and the
n is the discrete element amount of the ram; C(q) is the stiffness matrix of each discrete element and calculate
objective, namely, the structure flexibility; K is the global the node displacement.
stiffness matrix; X is the overall displacement vector; F is 5. Calculate the compliance and sensitivity value of the
the loads vector;
i is the element volume of the ram; V is overall structure, and solve the Lagrange multiplier.
the volume after optimization; Vo is the initial volume; Vmax 6. Update the design variables using optimization crite-
is the upper limit of volume; f is the rate of the volume, and ria method.
qmin is the minimum relative density. 7. Check if the stop criterion is met. If not, go to step (4).
Because of the large number of design variables in top-
ology optimization, we always solve this problem using the
3.2. Topology optimization results
finite element method. Each analysis needs to reassemble
the structure stiffness matrix and solve multiple equations. When the penalty factor takes the intermediate value five, the
As a result, the analysis incurs a high computation cost. topology optimization result of the ram structure of the FSW
According to the material interpolation theory based on the robot is shown in Figure 6(a). From the figure, it can be found
 MECHANICS OF ADVANCED MATERIALS AND STRUCTURES 7

Figure 6. Topology configuration of the ram structure with (a) material density distribution, P ¼ 5, and topology configuration according to (b) convergence curve.

that the density range distributes in accordance with the layer Because the volume of the ram is huge, the computation
way. In the figure, the red area denotes the region where the time will become prohibitively long using the real simulation
material is kept, and the blue area indicates the region where model. Hence, we adopt an approximation model, also
the material is removed, while the other colors represent tran- known as surrogates, for estimating the objective value.
sition areas. We can see that under the melon-valve configur- Using approximation models for design optimization can be
ation, the final optimized topology of the ram structure is traced back to when the work reported in 1960s, Schmit pro-
similar to an arch bridge. Figure 6(b) shows the convergence posed the approximate model concept in structural optimiza-
profile of the objective function, which indicates that the tion design, which accelerated the optimization process and
objective to be minimized decreases very rapidly in the begin- promoted the application of optimization algorithm in engin-
ning and converges after approximately 10 iterations. eering [11]. Many approximate models have been employed as
The optimization result can provide very instructive insight surrogates, such as the RS methods, Kriging method, and radial
basis function neural network models. In this work, we use the
into the design of ram structures. Note that in order to meet
Kriging method as the approximate model. The size optimiza-
these external constraint conditions and stiffness strength
tion process using the Kriging method is given in Figure 9.
design requirements, the optimized structure needs to be
For constructing Kriging approximation model, training
“repaired” by filling up some of the regions where the materials
samples can be obtained by the Latin hypercube design of
are removed in the topology optimization, such as add rib-plate
experimental (DOE) method. Latin hypercube DOE is based on
and beam column. The final design of the ram structure is pre- Latin square method plus a uniformity criterion, when its value
sented in Figure 7(a). The deformation of the ram along the Z- reached maximum. So this method is also called Uniform Latin
axis is as shown in Figure 7(b) under the melon-valve configur- square DOE. Latin hypercube DOE can generate uniform sam-
ation with the external load acting on the end of the ram and ple points with in the design space, which can greatly reduce
the fixed constraint applied to the other end. Compared with needed sample size for approximating nonlinear problem.
original displacement, the maximum displacement of the opti- In addition to the samples obtained by the Latin hyper-
mized ram has significantly increased by nearly 40%. cube DOE, the accuracy of the Kriging model was further
checked using some random sample points. Once the accur-
acy reaches a predefined threshold, the Kriging model will
4. Size optimization of ram structure no longer be updated and the optimization will be based on
4.1. Ram size optimization model the Kriging only. In this way, the computation for optimiza-
tion can be significantly reduced.
In order to enhance the stiffness and vibration resistance of Many optimization algorithms, such as sequential quad-
the ram structure and to improve its deformation resistance ratic programming, adaptive simulated annealing and the
and external interference, we conduct a further step for gradient based method can be employed for size optimiza-
refining the design obtained by topology optimization. tion assisted by the Kriging model. In this paper, the exter-
Considering the processing and installation demand, we nal penalty function method was adopted. When Kriging
configured the rib-plate type and frame size to achieve the model’s precision meets the predefined accuracy, the finite
optimal performance of the ram in accordance with several element model is no longer need in optimization.
existing design schemes. Different styles of rib-plate cell are Size optimization of the ram structure can be formulated
shown in Figure 8. as follows.
8 H. LUO ET AL.

Figure 7. The results of (a) configuration after optimization, and the comparison between the topology (b) path diagram and original path diagram.

Figure 8. Six kinds of different style rib-plates, (a) “W” style, (b) “*” hole style, (c) round hole style, (d) cross hole style, (e) “*” style, and (f) matts style.

T
Find : X ¼ ½x1 x2 x3 x4 T ¼ ½dtl1 l2  The relationship between the sectional dimensions and
Min : FðX Þ ¼ W=f 0 1 (36)
the output response of the ram is can be described by a
S:T : gðdtl1 l2 Þ ¼ f1
f 0 1  0 Kriging model as follows.
X2R
F ðxÞ ¼ F~ ðxÞ þ e (37)
In the above equation, W denotes the mass of the ram; d
is the bore diameter of the rib-plate cell; t is the thickness of where F(x) is the function describing the actual response
the rib-plate cell; f1/f10 is the natural frequency of the ram ~
value given the design, which is unknown, FðxÞ the approxi-
before and after optimization; and l1 and l2 are the length mate response value, estimated using the RS model, and e is
and width of the rib-plate cell. a random error between the approximation and the actual
 MECHANICS OF ADVANCED MATERIALS AND STRUCTURES 9

Figure 9. Size optimization process using approximate models.

Figure 10. (a) The configuration results of the ram after size optimization, and the size optimization (b) path diagram vs. topology and original path diagram.

value, usually assumed to be a Gaussian noise zero mean model is dependent on the model order and the number of
and standard deviation of r. decision variables. For size optimization of the space beam
Based on the complexity of the actual function F(x), the structure, we can choose a second-order polynomial to meet
order of the polynomial model to approximate the response the accuracy requirement. Then the initialization requires a
can be chosen to be from 1 to 4. Note that the minimum num- minimum of 1/2(N þ 1)(N þ 2) for sample points. In this case,
ber of training samples for constructing the approximate the approximate model can be described as follows.
10 H. LUO ET AL.

Table 1. Comparison between various performances indicators of the ram structure before and after optimization.
Indicator Primary structure After topology optimization After size optimization Contrast between them
Mass (t) 30.15 11.25 14.36 52.4% (reduce)
Displacement (mm) 0.67 0.98 0.76 13.4% (increase)
Natural frequency (Hz) 86 114 133 54.6% (enhance)

Figure 11. (a) The manufactured ram structure and (b) the friction stir robot.

F~ ðxÞ ¼ b0 þ bx x þ b2 x þ ::: þ bN xN Based on topology and size optimization, a ram structure

X has been manufactured, as shown in Figure 11(a). The entire
þ bNþ1 x12 þ bNþ2 x22 þ ::: þ 2
b2Nþ2 x2N þ bij xi xj
i6¼j
friction stir robot is shown in Figure 11(b).
(38)
5. Conclusion
This paper introduces a type of environment-friendly tech-
4.2. Size optimization results nology termed as FSW robots, which can be used for high-
In order to reduce the ram weight and enhance its stiffness, strength lightweight aluminum alloy welding highly
based on topology optimized results, we optimize its internal demanded in the aerospace industry. This kind of welding
structure elements which together form the whole ram frame. facilities can significantly enhance the transportation cap-
We find the lattice gabion is the best pattern element which acity of the space shuttle rockets, and improve the welding
makes up the internal structure of the ram frame. The optimal quality of complex curved surface structures. Thereafter, in
thickness size of each rib-plate from lattice gabion is shown in order to improve the geometric accuracy and mechanical
properties of the seams, design and development of heavy-
Figure 10(a). Finally, we conduct the simulation analysis under
load and high-stiffness FSW robots, especially those are able
the specified condition and got the path diagram as shown in
to be used for welding of large thin-wall aluminum alloy
Figure 10(b). Comparing these results, we can see that the stiff-
structures will be indispensable. Considering the influence of
ness performance of the ram has improved and the mass distri-
the gravity and the endload on the ram due to its cantilever
bution is more reasonable after the size optimization. configuration, we have established a detailed mathematical
In order to reflect the extent of the variations of each model for force analysis, based on which topology and size
basic performance index more intuitively, the mass, displace- optimization using effective and computationally efficient
ment and natural frequency of the ram are listed in Table 1. optimization algorithms have been conducted. By perform-
Fromthetable, we find that the effect of topology optimiza- ing the topology and size optimization, we are able to
tion has considerably reduced the ram’s weight, which also obtained optimized material distribution, structure style and
sheds light on the general design direction. Meanwhile, size key size of ram structure, resulting in much better mechan-
optimization has greatly improved the structural stiffness ical properties.
with only slight increase in the mass of the design. Through Through the dynamic characteristic analysis and struc-
topology and size optimization of the ram structure, the end tural optimization design on the ram structure of FSW
displacement of the ram in the welding process has signifi- robots, we have also gained useful insight into vibration
cantly decreased while ensuring the natural frequency of the resistance, high stiffness and structure lightweight design,
FSW robots to be in the feasible range. leading to improve welding precision of FSW robots.
 MECHANICS OF ADVANCED MATERIALS AND STRUCTURES 11

Acknowledgments [13] L. Xiong, R. Molfino, and M. Zoppi, “Fixture layout optimization

for flexible aerospace parts based on self-reconfigurable swarm
This work was supported by the National Natural Science Foundation intelligent fixture system,” Int. J. Adv. Manuf. Tech., vol. 66, no.
of China under grant number 51505470; Youth Innovation Promotion 9, pp. 1305–1313, 2013. DOI: 10.1007/s00170–012-4408-5.
Association, CAS [grant no. 2018237]. [14] S. Binitha and S. Sathya, “A survey of bio inspired optimization
algorithms,” Int. J. Soft. Comput. Eng., vol. 2, no. 2, pp.
137–151, 2012.
Declaration of interest [15] V. Govindaraj and J. V Ramasmy, “Optimum detailed design of
reinforced concrete continuous beams using genetic algo-
The authors declared no potential conflicts of interest with respect to
the research, authorship, and/or publication of this article. rithms,” Comput. Struct., vol. 84, no. 1–2, pp. 34–48, 2005.
DOI: 10.1016/j.compstruc.2005.09.001.
[16] M. Husken, Y. C. Jin, and B. Sendhoff, “Structure optimization
of neural networks for evolutionary design optimization,” Soft.
Comput., vol. 9, no. 1, pp. 21–28, 2005.
ORCID [17] J. Wu, R. Zhang, and G. Yang, “Design and experiment verifi-
Haitao Luo http://orcid.org/0000-0001-8865-6466 cation of a new heavy friction-stir-weld robot for large-scale
complex surface structures,” Ind. Robot., vol. 42, no. 4, 2015.
[18] Y. K. Liu and Y. M. Zhang, “Fusing machine algorithm with welder
intelligence for adaptive welding robots,” J. Manuf. Process., vol.
27, pp. 18–25, 2017. DOI:10.1016/j.jmapro.2017.03.015.
References
[19] K. Saitou, K. Izui, S. Nishiwaki, and P. Papalambros, “A survey
[1] M. Rahmi and M. Abbasi, “Friction stir vibration welding pro- of structural optimization in mechanical product development,”
cess: Modified version of friction stir welding process,” Int. J. J. Comput. Inf. Sci. Eng., vol. 5, no. 3, pp. 214–226, 2005.
Adv. Manuf. Tech., vol. 90, no. 1, pp. 141–151, 2017. DOI: [20] A. M. Zhou, Y. C. Jin, and Q. F. Zhang, “A population predic-
10.1007/s00170-016-9383-9. tion strategy for evolutionary dynamic multiobjective opti-
[2] B. T. Gibson, D. H. Lammlein and T. J. Prater, “Friction stir mization,” IEEE. Trans. Cybern., vol. 44, no. 1, pp. 40–53,
welding: Process, automation, and control,” J. Manuf. Process., 2014. DOI: 10.1109/TCYB.2013.2245892.
vol. 16, no. 1, pp. 56–73, 2014. DOI: 10.1016/ [21] C. Zhang, S. Yang and Q. Zhang, “Automatic optimization
j.jmapro.2013.04.002. design of a feeder extrusion die with response surface method-
[3] J. D. Backer and G. Bolmsj€ o, “Deflection model for robotic fric- ology and mesh deformation technique,” Int. J. Adv. Manuf.
tion stir welding,” Ind. Robot., vol. 41, no. 4, pp. 365–372, Tech., vol. 91, no. 9, pp. 1–13, 2017.
2014. [22] A. Myslinski, “Radial basis function level set method for struc-
[4] J. Shi, Y. Wang, G. Zhang, and H. Ding, “Optimal design of tural optimization,” Control. Cybern., vol. 39, no. 3, pp.
3-d of pkm module for friction stir welding,” Int. J. Adv. 627–645, 2010.
Manuf. Tech., vol. 66, no. 9, pp. 1879–1889, 2013. DOI: [23] Y. K. Sui, S. P. Li, and Y. Q. Guo, “An efficient global opti-
10.1007/s00170-012-4467-7. mization algorithm based on augmented radial basis function,”
[5] X. M. Liu, X. M. Zhang, J. S. Yao, Y. H. Yin, and P. Y. Xia, Int. J. Simul. Multisci. Des. Optim., vol. 2, no. 1, pp. 49–55,
“Optimization of rotation speed ratio in dual-rotation friction 2008.
stir welding by finite element method,” Int. J. Adv. Manuf. [24] P. Jiang, L. Cao and Q. Zhou,“Optimization of welding process
Tech., vol. 91, no. 5, pp. 2869–2874, 2017. DOI:10.1007/s00170- parameters by combining kriging surrogate with particle swarm
017-0007-9. optimization algorithm,” Int. J. Adv. Manuf. Tech., vol. 86,
[6] R. T. Rockafellar and R. J. B. Wets, “Stochastic convex pro-
no. 9–12, pp. 2473–2483, 2016.
gramming: Kuhn- Tucker conditions,” J. Math. Econ., vol. 2,
[25] P. Wenger and D. Chablat, “Design of a three-axis isotropic
pp. 349–370, 1975. DOI: 10.1016/0304-4068(75)90003-8.
parallel manipulator for machining applications,” Proceedings
[7] V. B. Venkayya, N. S. Khot, and L. Berke, “Application of opti-
of the Workshop on Fundamental Issues and Future Directions
mality criteria approaches to automated design of large practical
for Parallel Mechanisms and Manipulators, Quebec City,
structures,” 2nd Symposium on Structural Optimisation
AGARD-CP-123, Milano, Italy, 1973. Quebec, Canada, 2002.
[8] C. Fleury and G. Sanders, “Relations between optimality criteria [26] S. Mikael and I. Kalaykov, “A robot prototype for friction stir
and mathematical programming in structural optimization,” welding,” Proceedings of the IEEE Conference on Robotics,
Proc. Symp. Appl. Comput. Methods Eng., pp. 507–520, Automation and Mechatronics, Bangkok, Thailand, 2006. DOI:
Los Angeles, 1977. 10.1109/RAMECH.2006.252646.
[9] F. E. Eastep, V. A. Tischler, V. B. Venkayya, and N. S. Khot, [27] Y. Z. Okawa, T. G. Masasi, and H. R. Sugii, “Development of 5-
“Aeroelastic tailoring of composite structures,” J. Aircraft., vol. axis friction stir welding system,” SICE-ICASE International Joint
36, no. 6, pp. 1041–1047, 1999. Conference, Busan, Korea, 2006. DOI: 10.1109/SICE.2006.315435.
[10] R. Ganguli, “Optimal design of composite structures: A histor- [28] W. R. Longhurst, C. D. Cox and B. T. Gibson, “Development
ical review,” J. Indian Inst. Sci., vol. 93, no. 4, pp. 557–570, of friction stir welding technologies for in-space man-
2013. ufacturing,” Int. J. Adv. Manuf. Tech., vol. 90, no. 1–4, pp.
[11] L. A. Schmit, “Structural design by systematic synthesis,” 1–11, 2016. DOI: 10.1007/s00170-016-9362-1.
Proceedings of the Second Conference on Electronic [29] K. Lee, Topology Optimization of Convective Cooling System
Computation, pp. 105–122, New York, 1960. Designs. The University of Michigan, ProQuest LLC, 2012.
[12] N. P. Van, K. Maute, and M. Langelaar, “Level-set methods for [30] Y. Zhang and J. L. Xie, “Testing and analysis of several global
structural topology optimization: A review,” Struct. Multidiscip. optimization algorithms,” Appl. Mech. Mater., vol. 404, pp.
Optim., vol. 48, no. 3, pp. 437–472, 2013. 543–547, 2013. DOI: 10.4028/www.scientific.net/AMM.404.543.

View publication stats