Topological Synthesis of

Compliant Mechanisms Using

Hong Zhou
Research Assistant Spanning Tree Theory
Kwun-Lon Ting This paper introduces the spanning tree theory to the topological synthesis of compliant
ASME Fellow
Professor mechanisms, in which spanning trees connect all the vertices together using a minimum
e-mail: [email protected] number of edges. A valid topology is regarded as a network connecting input, output,
support, and intermediate nodes, which contains at least one spanning tree among the
Center for Manufacturing Research, introduced nodes. Invalid disconnected topologies can be weeded out if no spanning tree
Tennessee Technological University, is included. No further deformation analysis and performance evaluation is needed to
Cookeville, TN 38505 invalidate disconnected topologies. Problem-dependent objectives are optimized for to-
pological synthesis of compliant mechanisms. Constraints about maximum input dis-
placement and force, maximum stress and overlapping connections are directly imposed
during optimization process. The discrete optimization problem is solved by genetic al-
gorithm with penalty function handling constraints. Two examples are given to verify the
effectiveness of the proposed synthesis procedure. 关DOI: 10.1115/1.1900148兴

1 Introduction ent formulations to satisfy different requirements for CMs includ-

ing the spring model and the multicriteria model. The spring
A compliant mechanism 共CM兲 is a mechanical device used to
model captures the resistance of the workpiece against the mecha-
transmit motion, force or energy. Unlike traditional rigid-link
nism by incorporating a spring at the output deflection location.
mechanisms, CMs gain all or part of their mobility from elastic
The multicriteria model meets the dual requirements of flexibility
deformation of flexible members rather than joints only 关1兴. With
before contact with the workpiece and stiffness after contact.
the reduction or elimination of joints, backlash, friction, wear and
Larsen and Sigmund 关15兴 defined the design of CMs as distrib-
lubrication need are eliminated or significantly reduced and so are
uting a given amount of material within the design domain. The
the vibration and noise from joints 关2,3兴. Manufacturing and as-
error in obtaining the prescribed values of mechanical and geo-
sembly cost and time are reduced considerably since many CMs
metric advantages was minimized. Sigmund 关16兴 presented an ap-
may be manufactured from an injection-moldable material and
proach by maximizing the mechanical advantage of CMs subject
constructed of one piece. More and more useful applications are
to a limited material volume and input displacement. Constraining
now being found for CMs 关4,5兴.
input displacement was also used to indirectly control the maxi-
A CM is often made of one piece of material and its configu-
mum stress in CMs. Homogenization approaches make adequate
ration is described by topology, shape and size. Topological syn-
optimization on topology and enhance computational efficiency
thesis of CMs is to determine material connectivity among differ-
by changing the discrete optimization problems into continuous
ent ports such as input, output and support ports. These three
optimization problems. Since the discretized elements may have
special ports and other material intersection ports are called nodes.
any intermediate material densities from void to solid, postpro-
Topology defines the skeleton of a CM and connection of nodes.
cessing or designer’s interpretation to the final optimal solution is
The synthesis of CMs with lower pairs was investigated by
necessary to get an appropriate topology.
Shoup and McLarnan 关6兴. They developed the concepts of simple
Designs generated by linear elastic models are not accurate for
and compound flexible links. Midha et al. 关1,7兴 identified kine-
large deformations and, furthermore, design specifications involv-
matic properties of CMs, introduced the concept of link compli-
ing nonlinear force-deflection characteristics and curved path out-
ance, and classified CMs according to link and segment charac-
put cannot be realized. Geometrically nonlinear finite element
teristics. They derived different CMs by replacing rigid links with
models were employed 关17–19兴 to synthesize CMs for different
flexible ones and imposing discrete compliances at the joints of a uses.
given rigid-link kinematic chain. Howell and Midha 关8–10兴 devel- Genetic algorithms were used by Jakiela et al. 关20–22兴 for the
oped a pseudo rigid-body model to approximate the behavior of optimal topology search of continuum structures. The design
cantilever-like flexible segments. The tip path of an end-loaded space was discretized into small elements with all of the finite
cantilever was approximated as a circular arc centered at a certain elements forming a binary-coded bit-string chromosome, 0 and 1
ratio of span from the fixed end. The pseudo rigid-body model for absence and presence of an element in the structure, respec-
allowed several well known rigid-body mechanism design meth- tively. Parsons and Canfield 关23兴 developed a genetic program-
ods amenable for CMs. Murphy et al. 关11兴 applied graph theory to ming procedure for the design of CMs. To facilitate the transmis-
topological synthesis of CMs, and permutated nonisomorphic to- sion of topology and shape characteristics across generations, Tai
pologies by adding flexure pivots and compliant segments to a et al. 关24,25兴 utilized arrangements of skeleton and flesh surround-
rigid-link mechanism. ing the bones to represent structural geometry, which is based on
Ananthasuresh, Kota, and Kikuchi 关12兴 extended homogeniza- the structure of living organisms and simulates the anatomical
tion methods 关13,14兴 to the design of CMs and posed the synthesis description of vertebrates.
problem as a design for desired deflection, namely, accomplishing In this paper, the spanning tree theory is incorporated into to-
a specified displacement at a certain point. They proposed differ- pological synthesis procedure of CMs. With a set of input, output,
support and intermediate nodes introduced into the design do-
Contributed by the Mechanisms and Robotics Committee for publication in the
main, the valid topologies connect all the nodes together and
JOURNAL OF MECHANICAL DESIGN. Manuscript received: March 4, 2004; revised: July should contain at least one spanning tree. If no spanning tree is
23, 2004. Associate Editor: G. K. Ananthasuresh. included in a topology, it is an invalid disconnected topology and

Journal of Mechanical Design Copyright © 2005 by ASME JULY 2005, Vol. 127 / 753

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 no further performance evaluation is needed for it. The following
sections describe the spanning tree theory, the optimization prob-
lem formulation, the genetic algorithm and two synthesis
examples.

2 Spanning Tree Theory

In topological synthesis of CMs, the optimal topology is typi-
cally searched within a structural universal formed by nodes in the
design domain and possible connections among the nodes. Many
invalid disconnected topologies may be included in the search Fig. 1 A structural universal example
space. How to efficiently weed out such kinds of invalid topolo-
gies is an important issue during the topological optimization pro-
cess. ning trees will be huge. It is not convenient to check if a topology
Three kinds of nodes are required: input, output and support contains spanning tree based on tree polynomial. An effective
nodes. These nodes can be determined according to design re- checking approach is described below.
quirements and must be included in the topology. Selecting a to- A walk in a graph G is an alternating sequence of vertices and
pology for a CM may be regarded as finding a connecting network edges, beginning and ending with vertices, in which each edge is
among nodes. These nodes may be connected with or without incident with two vertices immediately preceding and following
other intermediate nodes. With nodes in the design domain, an it. Let the number of vertices in G is p and the adjacency matrix
optimal connecting network is then searched, which forms a con- is A. Then the 共i , j兲 entry of Ak is the number of different walks of
nected skeleton of the CM. length k from vertex vi to vertex v j. For any two different vertices
In graph theory, a tree is a connected graph without any circuit in a connected graph, there is walk connecting them. If G is dis-
关26兴. The minimum number of edges is used in a tree to connect connected, there exist vertices i and j such that the entry 共i , j兲 of
the vertices. There are 共n − 1兲 edges in a tree with n vertices. A tree Ak is zero for all k ⬍ p. A matrix to check if a spanning tree is
is said to be a spanning tree of a connected graph G if the tree contained in a topology can be defined as follows.
contains all vertices of G. Since the vertices of G are barely hang- p−1

兺A
ing together in a spanning tree, it is often referred to as a skeleton
or scaffold of G. Mc = k
共1兲
k=1
A structural universal may be regarded as a labeled connected
graph 关26兴 when nodes and connections are taken as vertices and If zero entry exists in M c, the spanning tree is not contained in
edges in a graph. Nodes and connections in a structural universal the topology. The product of all entries in any line in M c can be
have their status, locations or sizes in the design domain and are used to check if the spanning tree is contained or not. If the
not interchangeable among them. All nodes and connections can product is nonzero, the spanning tree is contained. Otherwise, the
be labeled to distinguish each other, so the structural universal can topology is disconnected.
be regarded as a labeled connected graph. A valid topology has the A structural universal example is shown in Fig. 1. Nodes n1 and
nodes joined together by connections and includes at least one n3 are input and output nodes, n2 , n4 , n5 , n6 are intermediate
spanning tree. Hence all valid topologies can be regarded as add- nodes, n7 and n8 are support nodes. There are 8 vertices and 15
ing different possible edges to the spanning trees. The discon- edges in the graph, which are labeled as n1 , … , n8 and e1 , … , e15.
nected topologies can be weeded out if no spanning tree is in- Figure 2 shows a topology derived from the structural universal in
cluded. The question now is how to know if a topology contains a Fig. 1.
spanning tree or not. Absent connections in the topology are denoted by dashed
Each edge in a graph joins two vertices and is said to be inci- lines. Matrices A and M c can be found to be

冤 冥
dent with each of them. The degree of a vertex is the number of
edges incident with it. A graph can be considered as a symmetric 0 0 0 1 1 0 0 0
irreflexive relation on its vertices. The relation is called adjacency 0 0 1 0 0 0 0 0
and can be represented by a binary matrix A = 关aij兴. In A, the ith 0 1 0 0 0 1 0 0
row and column correspond to the ith vertex, aij = 1 if vertices i
and j are joined by an edge, and aij = 0 otherwise. Matrix A is 1 0 0 0 1 0 1 0
A= 共2兲
called the adjacency matrix of the graph. There is one-to-one cor- 1 0 0 1 0 0 1 1
respondence between a labeled graph with p vertices and its p 0 0 1 0 0 0 0 0
⫻ p symmetric adjacency matrix. The degree matrix ⌬ = 关dij兴 of a
0 0 0 1 1 0 0 0
graph with p vertices is defined as the p ⫻ p diagonal matrix with
dii being the degree of vertex i. With the adjacency and degree 0 0 0 0 1 0 0 0
matrices, the total number of spanning trees in a labeled con-
nected graph can be obtained as the value of any co-factor of the
matrix ⌬ − A. All the spanning trees can also be listed symboli-
cally. Assume that a labeled connected graph G has p vertices and
q edges which are labeled as n1 , n2 , … , n p and e1 , e2 , … , eq, re-
spectively. Let M = 关mij兴 be the p ⫻ p matrix in which mij = −ek if
edge ek joins vertices ni and n j, mii = ek1 + ek2 + ¯ + ekd if these are
the d edges incident with ni, and mij = 0 if no edge joins vertices ni
and n j. Let each spanning tree of G associate with the product of
its edges. The product is called the term of that tree and the tree
polynomial of G is the sum of the terms for all spanning trees.
Then the value of any cofactor of matrix M is the tree polynomial
of G 关27,28兴. Fig. 2 A disconnected topology from the structural universal
When the structural universal is complex, the number of span- in Fig. 1

754 / Vol. 127, JULY 2005 Transactions of the ASME

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 3 Optimization Model for Topological Synthesis of
Compliant Mechanisms
The purpose of optimal topological synthesis of CMs is to make
them best perform the desired mechanical functions. To this goal,
the general optimization process can be stated as

Optimize f共X兲 共6兲

Subject to 兵g j共X兲 艋 0, j = 1,2,…,m其 共7兲

Fig. 3 A connected topology from the structural universal in
Fig. 1 Here X is the design variable vector including cross-sectional
sizes of all possible connections and positions of all movable
nodes. The term f共X兲 is the problem-dependent objective describ-
ing the desired function of the synthesized CM; g j共X兲 are inequal-
ity constraints such as maximum stress and overlapping elements

冤 冥
limitations, which are also defined based on the CM application. If
265 0 0 349 394 0 265 138
equality constraints are involved during the synthesis, they can be
0 7 15 0 0 7 0 0 replaced by equivalent inequality constraints. Hence only inequal-
0 15 14 0 0 15 0 0 ity constraints are considered during the optimization process.
With structural universal established in the design domain, to-
349 0 0 437 487 0 349 183 pological synthesis of CMs entails determining the presence or
Mc = 共3兲
394 0 0 487 530 0 394 211 absence status of every possible connection, cross-sectional sizes
0 7 15 0 0 7 0 0 of connections present in the topology, and positions of all mov-
able nodes. After topological synthesis, shape and size optimiza-
265 0 0 349 394 0 265 138 tion is usually performed to synthesize the detailed dimensions of
138 0 0 183 211 0 138 71 CMs and further improve their performances. All connections can
There are some zero entries in every line of M c. The topology is thus be simplified as straight uniform connections at topological
disconnected and no spanning tree is contained. synthesis stage. Discrete sets of values can be used to define
Figure 3 shows another topology from the structural universal cross-sectional sizes of connections and positions of movable
nodes. Besides, presence or absence status of a possible connec-
in Fig. 1, which is actually a spanning tree. Matrices A and M c are
tion can be combined with its cross-sectional size variable, ab-
sence corresponding to zero size. Cross-sectional size of a con-

冤 冥
0 1 0 1 0 0 0 0 nection can thus be mapped as a binary variable. The larger the
value of the binary variable, the larger the cross-sectional size,
1 0 1 0 0 0 0 0 and zero corresponds to absence of the connection.
0 1 0 0 0 1 0 0 A structural universal can be represented as a labeled connected
1 0 0 0 1 0 1 0 graph in which different nodes and connections correspond to
A= 共4兲 different symbols. A spanning tree can be checked to know if it is
0 0 0 1 0 0 0 0 included in a topology by the product of all entries in any line of
0 0 1 0 0 0 0 1 the corresponding matrix M c of the topology. If the product is
0 0 0 1 0 0 0 0 nonzero, the spanning tree is included and the topology is a valid
connected topology. If the product is zero, the spanning tree is
0 0 0 0 0 1 0 0 excluded and topology is an invalid disconnected topology.
The discrete positions of movable nodes can also be mapped as
binary variables like cross-sectional sizes of connections. Hence

冤 冥
35 57 21 76 20 27 20 6 all design variables can be encoded and joined together as a bi-
57 29 49 27 27 19 27 19 nary string. Genetic algorithm 共GA兲 关29,30兴 is used in this work to
handle the binary bit-string representations of topologies.
21 49 27 35 7 41 7 13 GA searches the solution space through the strategy of the sur-
76 27 35 48 49 8 49 8 vival of the fittest. The fittest individuals of a population have a
Mc = 共5兲 higher probability to reproduce and survive to the next generation.
20 27 7 49 14 8 14 1
GA explores the search space from many different points simul-
27 19 41 8 8 21 8 22 taneously to find an optimum with a more global search strategy
20 27 7 49 14 8 14 1 rather than a localized gradient search approach. GA is not af-
fected by the characteristics of objective and constraint functions
6 19 13 8 1 22 1 8 such as discontinuity, undifferentiability and multimodal.
No zero entry exists in matrix M c. The topology is connected To increase the probability of survival of better designs and
and contains a spanning tree. prevent the overly fit designs from dominating the selection pro-
Different numbers of intermediate nodes and possible connec- cess, the mating pool is generated in this paper by tournament
tions lead to different feasible topologies. The more the interme- selection 关31,32兴. The elitist strategy is incorporated into the se-
diate nodes and possible connections, the more feasible topolo- lection process to make the evolution more reliable and efficient
gies. More possible connections mean more design variables and 关33兴. Multipoint crossover and bit-flip mutation are employed in
more feasible topologies, which of course increase complexity the evolutionary process. Constraints are handled by penalty func-
and computational cost of the optimization problem. Designers tions 关34,35兴. If no spanning tree is included in a topology, a
can reasonably select the number of intermediate nodes and pos- heavy penalty is imposed on it and no computation on objective
sible connections in the design domain for different applications, and constraint functions is needed. If spanning tree is included in
which results in satisfactory topological solutions with affordable a topology and one or several inequality constraints are violated,
computational costs. the fitness of the topology can be penalized as follows

Journal of Mechanical Design JULY 2005, Vol. 127 / 755

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Table 1 Design parameters in example 1 synthesized; n1 , n2 and n4 are input, output and support nodes.
Design domain ⫽ 50 mm⫻ 50 mm Two intermediate nodes n3 , n5 and seven possible connections are
Material: Nylon introduced in the structural universal.
Elastic modulus: E = 2600 N / mm2 The function of this CM is to invert and amplify the input
Yield strength: ␴y = 99 N / mm2 displacement. The objective function to be maximized can be de-
Thickness of beam ⫽ 1.0 mm fined as
Maximum width of beam ⫽ 2.5 mm
f共X兲 = sign共Dout兲兩Dout兩 共9兲
Here sign共Dout兲 specifies a certain direction of Dout as positive,

␾共X兲 = ␺共X兲 冒冉 1+
m

兺 K Max„0,g 共X兲…
j=1
j j 冊 共8兲
its opposite direction as negative. The design variables are widths
of seven beams 共w1 , ¯ , w7兲 and positions of two movable inter-
mediate nodes 共n3 , n5兲. All design variables can be encoded and
joined together as a binary string, in which two bits are assigned
Here ␺共X兲 is the fitness value of design X, which is based only to each design variable. There are four discrete values evenly
on the value of objective function f共X兲. The term ␾共X兲 is the distributed in the range of a design variable. Zero is included in
modified fitness value after considering all the constraints. K j is a beam width, and nonzero beam width is not close to zero.
positive parameter that is proportional to the penalty for violation To directly restrict the required values of input displacement
of constraint j. and input force, maximum input displacement and input force are
taken as constraints. Maximum stress is also considered as a con-
4 Topological Synthesis Examples straint to prevent CM failure
To demonstrate the topological synthesis procedure, two ex- 兩Din兩
amples are presented. For any valid topology, performance is g1共X兲 = −1艋0 共10兲
evaluated by nonlinear co-rotational finite element procedure Da
关36,37兴. All connections are regarded as beams, and node degrees
of freedom are considered for deformation analysis at topological 兩Fin兩
g2共X兲 = −1艋0 共11兲
synthesis stage. Since all nodes in valid topologies are joined Fa
together by connections whose cross-sectional sizes are not close
to zero, the singularity condition of stiffness matrix is good and 兩 ␴ m兩
the deformation analysis process is stable. g3共X兲 = −1艋0 共12兲
␴a
4.1 Example 1: Displacement Inverter and Amplifier. In Da , Fa and ␴a are the allowable maximum input displacement,
this topological synthesis example, the output displacement direc- input force and stress. Din , Fin and ␴m are the actual maximum
tion is required to be opposite to the input displacement and the input displacement, input force and stress.
output displacement magnitude needs to be amplified. The design ␴a is defined as
domain size, material and connection beam properties are given in
Table 1. Figure 4 shows the specified input, output and support ␴y
ports in the design domain. ␴a = 共13兲
SF
The beam structural universal is constructed and shown in Fig.
5. The input, output and support nodes are denoted as circle with ␴y is the yield strength of the material, SF is the safety factor for
dot, circle with cross and solid circles, respectively. Regular the CM.
circles represent intermediate nodes. The upper and lower parts Da , Fa and SF are specified as 1.5 mm, 20 N and 1.5 in this
are symmetric, so only the upper half is labeled, analyzed and example. The input is the displacement at node n1, which is in-
creased until its maximum value is reached or the force needed to
exert the input displacement reaches its maximum value. So con-
straints g1 and g2 are always satisfied through deformation analy-
sis process.
Every design variable needs to specify its lower and upper
bounds, i.e., the range, in GA. The range for the seven width
variables is 关0, 2.5兴. The lower and upper bounds for x and y
coordinates of movable nodes n3 and n5 are selected as
关20 10 35 20 兴 艋 关xn3 y n3 xn5 y n5 兴 艋 关30 15 50 25 兴
共14兲
Node n1 is taken as the origin of the coordinate system. Cross-
Fig. 4 The input, output and support ports in example 1 over and mutation probabilities are specified as 0.85 and 0.03.
Population size and maximum generation number are 50 and 70.
Fitness is the output displacement at node n2 in the desired direc-
tion. Penalty parameter K for the violation of maximum stress
constraint is selected as 10. The optimal discrete values of the
seven width variables and four position variables are
关w1 w2 w3 w4 w5 w6 w7 兴 = 关0 3 0 3 2 3 1 兴
共15兲

关xn3 y n3 xn5 y n5 兴 = 关3 0 1 3 兴 共16兲

The undeformed and deformed configurations of the optimal
topology are denoted in Fig. 6 by solid and dotted lines, respec-
Fig. 5 The structural universal of example 1 tively. For the deformed configuration of the optimal topology,

756 / Vol. 127, JULY 2005 Transactions of the ASME

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 Table 2 Design parameters in example 2
Design domain ⫽ 150 mm⫻ 75 mm
Material: nylon
Elastic modulus: E = 2600 N / mm2
Yield strength: ␴y = 99 N / mm2
Thickness of beam ⫽ 4.0 mm
Maximum width of beam ⫽ 4.0 mm

clamping force needs to be amplified. The design domain, mate-

rial and connection beam properties are given in Table 2. Figure 8
shows the specified input, output and support ports in the design
domain.
The structural universal is shown in Fig. 9, which is constructed
by discretizing the design domain into a mesh of quadrilaterals
and adding a pair of crossing lines inside every quadrilateral. Like
example 1, only the upper half is labeled, analyzed and synthe-
sized because of symmetry; n10 and n1 are input and output nodes;
n3 and n7 are support nodes. Six intermediate nodes and 21 pos-
Fig. 6 The topological synthesis result of example 1 sible connections are introduced in the design domain.
For compliant clamping mechanism design, a spring is usually
placed at the output port to represent the reaction force from
only node displacements are actually concerned and all deformed workpiece and simplify the problem. Compared to the compliant
beams are approximately represented as straight. The real curved clamping mechanism, workpiece material is usually much stiffer
deformation shape of the topology will be considered at dimen- in many cases. There is often a gap, Dg, between the workpiece
sional synthesis or postprocessing stage of the CM. The input and output port before the input force is applied. With the gap, the
displacement and required input force are 1.49 mm and 20.0 N. CM is required to be both flexible and stiff in order to close the
The output displacement and geometric advantage are 8.34 mm gap and transmit the force 关12,16兴. The workpiece in this work is
and 5.60. The maximum stress constraint is satisfied by all beams regarded as rigid and a gap is considered when topology of the
in the topology. Figure 7 shows the maximum and average fitness clamping mechanism is synthesized.
values of every generation through the whole evolutionary The function for this CM is to generate a large clamping force
process. at the output port, so the objective function to be maximized can
In this example, we can see a simple and effective structural be defined as
universal has been employed. Such a kind of structural universal
is usually established based on experience and an idea from de- f共X兲 = sign共Fout兲兩Fout兩 共17兲
signers. If the designers have no preconceived ideas about the Design variables are the widths of 21 beams 共w1 , … , w21兲. Po-
structural universal, they can discretize the design domain into a sitions of intermediate nodes are regarded as fixed in this example
mesh of quadrilaterals and regard all quadrilateral vertices in the and can be adjusted during dimensional synthesis of the selected
mesh as nodes. A pair of crossing lines can be added into every
quadrilateral. The edges of all quadrilaterals and their inside
crossing lines consist of connections. A structural universal is then
constructed by the introduced nodes and connections. This kind of
structural universal will be demonstrated in the next example.
4.2 Example 2: Compliant Clamping Mechanism. The to-
pology of a hand tool used for clamping workpiece is to be syn-
thesized in this example. To securely clamp a workpiece, the

Fig. 8 The input, output and support ports in example 2

Fig. 7 The evolutionary convergence history of example 1 Fig. 9 The structural universal of example 2

Journal of Mechanical Design JULY 2005, Vol. 127 / 757

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 extra constraints in this example which are about the overlapping
of connections. There are four sets of overlapping connections in
the structural universal: 共e3 , e4兲 , 共e10 , e11兲 , 共e13 , e14兲 , 共e16 , e17兲. To
avoid overlapping, at least one of the two connections in the four
sets should be missing. The overlapping constraints can be defined
as follows
g4共X兲 = Min共w3,w4兲 − 0.000 01 艋 0 共18兲

g5共X兲 = Min共w10,w11兲 − 0.000 01 艋 0 共19兲

g6共X兲 = Min共w13,w14兲 − 0.000 01 艋 0 共20兲

g7共X兲 = Min共w16,w17兲 − 0.000 01 艋 0 共21兲

Fig. 10 The topological synthesis result of example 2 The range for the 21 width variables is 关0, 4.0兴. Da , Fa , SF and
Dg are specified as 7 mm, 50 N, 1.5 and 2.0 mm. Crossover and
mutation probabilities and penalty for stress constraint violation
topology if necessary. Similar to example 1, constraints about are the same as example 1. Penalty parameter K for the existence
maximum input displacement, input force and stress in the CM of overlapping connections is selected as 5. Population size and
are imposed. Input displacement and input force constraints are maximum generation number are 90 and 110. The optimal discrete
satisfied through deformation analysis process. There are some values of the 21 width variables are

关w1 w2 w3 w4 w5 w6 w7 w8 w9 w10 ¯ w11 w12 w13 w14 w15 w16 w17 w18 w19 w20 w21 兴
= 关3 1 3 0 0 0 0 3 1 3 0 1 0 0 3 0 3 3 3 3 3 兴 共22兲

The synthesis result is shown in Fig. 10. The input displace- be removed from the solution. Connection e9 connects two fixed
ment and required input force are 7.00 mm and 38.0 N. The output nodes together and is trivial in the solution. But e9 is important in
force and mechanical advantage are 98.3 N and 2.59. The maxi- the structural universal since we do not know one or two support
mum stress constraint is satisfied by all beams in the topology. nodes are needed in the final solution and e9 can be used to re-
Figure 11 shows the maximum and average fitness values of every move a redundant support node.
generation through the whole evolutionary process. The clamping Table 3 shows clamping forces of the optimal topology for
force will be 196.6 N when upper and lower parts are considered different gaps between workpiece and output port. When the gap
together. is within 2.0 mm, the clamping mechanism works well. Clamping
Since all the introduced nodes will remain and be connected in force decreases dramatically when the gap is above 2.0 mm and
the final solution, there might be some redundant nodes and con- there is no clamping force for 3.0 mm gap. It is appropriate for the
nections in it. Such kinds of nodes and connections are not diffi- compliant clamping mechanism to be designed based on required
cult to detect and remove from the final solution since they have maximum gap.
no contributions to motion and force transmissions. They usually
exist in the form of dangling nodes and connections. There is one 5 Conclusions
redundant node and connection in Fig. 10. To keep node n2 in the
final solution, connection e1 exists in dangling form and needs to A spanning tree based topological synthesis method of compli-
ant mechanisms is developed in this paper. Valid topologies are
easy to describe and invalid topologies easy to weed out in the
presented method. Only valid topologies are analyzed for perfor-
mance evaluation. The nodes in valid topologies are joined to-
gether by connections with cross-sectional sizes away from zero.
This leads to a good singularity condition of stiffness matrix and a
stable deformation analysis process. The presence or absence sta-
tus of every possible connection in a topology is combined with
its discrete cross-sectional size, zero size corresponding to its ab-
sence. The discrete values of cross-sectional sizes are employed
for the existence checking of spanning trees and overlapping con-

Table 3 Clamping forces of the optimal topology for different

gaps

Gap Input displacement Input force Clamping force

共mm兲 共mm兲 共N兲 共N兲

0.0 2.11 50.0 266.0

1.0 4.80 50.0 264.0
2.0 7.00 38.0 196.6
3.0 7.00 2.6 0.0
Fig. 11 The evolutionary convergence history of example 2

758 / Vol. 127, JULY 2005 Transactions of the ASME

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
 nections. This makes such kinds of checking straightforward and ogy Optimization,” Mech. Struct. Mach., 25共4兲, pp. 493–524.
关17兴 Buhl, T., Pedersen, C. B. W., and Sigmund, O., 2000, “Stiffness Design of
also improves the computational efficiency. The direct imposition Geometrically Nonlinear Structures Using Topology Optimization,” Struct.
of constraints about input displacement, input force, maximum Multidiscip. Optim., 19共2兲, pp. 93–104.
stress and overlapping connections makes optimization solutions 关18兴 Bruns, T. E., and Tortorelli, D. A., 2001, “Topology Optimization of Non-
reasonable. The topological synthesis results from this method are Linear Elastic Structures and Compliant Mechanisms,” Comput. Methods
well defined. No jagged or fuzzy connections exist in the solu- Appl. Mech. Eng., 190共26–27兲, pp. 3443–3458.
关19兴 Saxena, A., and Ananthasuresh, G. K., 2001, “Topology Synthesis of Compli-
tions, and little designers’ interpretation is needed for the final ant Mechanisms for Nonlinear Force Deflection and Curved Path Specifica-
solutions. The synthesis results are practical for further shape and tions,” ASME J. Mech. Des., 123共1兲, pp. 33–43.
size optimization or for direct fabrication and use. 关20兴 Chapman, C. D., Saitou, K., and Jakiela, M. J., 1994, “Genetic Algorithms as
an Approach to Configuration and Topology Design,” ASME J. Mech. Des.,
116共4兲, pp. 1005–1012.
References 关21兴 Chapman, C. D., and Jakiela, M. J., 1996, “Genetic Algorithm-Based Struc-
关1兴 Midha, A., Norton, T. W., and Howell, L. L., 1994, “On the Nomenclature, tural Topology Design with Compliance and Topology Simplification Consid-
Classification, and Abstractions of Compliant Mechanisms,” ASME J. Mech. erations,” ASME J. Mech. Des., 118共1兲, pp. 89–98.
Des., 116共1兲, pp. 270–279. 关22兴 Jakiela, M. J., Chapman, C. D., Duda, J. W., Adewuya, A., and Saitou, K.,
关2兴 Ananthasuresh, G. K., and Kota, S., 1995, “Designing Compliant Mecha- 2000, “Continuum Structural Topology Design with Genetic Algorithms,”
nisms,” Mech. Eng. 共Am. Soc. Mech. Eng.兲 117共11兲, pp. 93–96. Comput. Methods Appl. Mech. Eng., 186共2兲, pp. 339–356.
关3兴 Howell, L. L., 2001, Compliant Mechanisms, Wiley, New York. 关23兴 Parsons, R., and Canfield, S. L., 2002, “Developing Genetic Programming
关4兴 Ananthasuresh, G. K., 2003, Optimal Synthesis Methods for MEMS, Kluwer, Techniques for the Design of Compliant Mechanisms,” Struct. Multidiscip.
Boston. Optim., 24共1兲, pp. 78–86.
关5兴 Bendse, M. P., and Sigmund, O., 2002, Topology Optimization, Springer, Ber- 关24兴 Tai, K., and Chee, T. H., 2000, “Design of Structures and Compliant Mecha-
lin. nisms by Evolutionary Optimization of Morphological Representations of To-
关6兴 Shoup, T. E., and McLarnan, C. W., 1971, “A Survey of Flexible Link Mecha- pology,” ASME J. Mech. Des., 122共4兲, pp. 560–566.
nisms Having Lower Pairs,” J. Mech. 6共1兲, pp. 97–105. 关25兴 Tai, K., Cui, G. Y., and Ray, T., 2002, “Design Synthesis of Path Generating
关7兴 Her, I., and Midha, A., 1987, “A Compliance Number Concept for Compliant Compliant Mechanisms by Evolutionary Optimization of Topology and
Mechanisms, and Type Synthesis,” ASME J. Mech., Transm., Autom. Des., Shape,” ASME J. Mech. Des., 124共3兲, pp. 492–500.
109共3兲, pp. 348–355. 关26兴 Narsingh, D., 1990, Graph Theory With Applications to Engineering and Com-
关8兴 Howell, L. L., and Midha, A., 1994, “A Method for the Design of Compliant puter Science, Prentice–Hall, Englewood Cliffs, NJ.
Mechanisms with Small-Length Flexural Pivots,” ASME J. Mech. Des., 关27兴 Yellen, J., and Gross, J. L., 1998, Graph Theory and Its Applications, CRC,
116共1兲, pp. 280–290. Boca Raton, FL.
关9兴 Howell, L. L., and Midha, A., 1995, “Parametric Deflection Approximations 关28兴 Chen, W. K., 1997, Graph Theory and Its Engineering Applications, World
for End-Loaded, Large-Deflection Beams in Compliant Mechanisms,” ASME Scientific, Singapore.
J. Mech. Des., 117共1兲, pp. 156–165. 关29兴 Goldberg, D. E., 1989, Genetic Algorithms in Search, Optimization, and Ma-
关10兴 Howell, L. L., and Midha, A., 1996, “A Loop-Closure Theory for the Analysis chine Learning, Addison–Wesley, Reading, MA.
and Synthesis of Compliant Mechanisms,” ASME J. Mech. Des., 118共1兲, pp. 关30兴 Michalewicz, Z., 1996, Genetic Algorithms+ Data Structures= Evolutionary
121–125. Programs, Springer, Berlin.
关11兴 Murphy, M. D., Midha, A., and Howell, L. L., 1996, “The Topological Syn- 关31兴 Chen, S. Y., and Rajan, S. D., 2000, “A Robust Genetic Algorithm for Struc-
thesis of Compliant Mechanisms,” Mech. Mach. Theory, 31共2兲, pp. 185–199. tural Optimization,” Struct. Eng. Mech., 10共4兲, pp. 313–336.
关12兴 Ananthasuresh, G. K., Kota, S., and Kikuchi, N., 1994, “Strategies for Sys- 关32兴 Yang, J., and Soh, C. K., 1997, “Structural Optimization by Genetic Algo-
tematic Synthesis of Compliant MEMS,” ASME J. Dyn. Syst., Meas., Control, rithms with Tournament Selection,” J. Comput. Civ. Eng., 11共3兲, pp. 195–200.
DSC-Vol. 55-2, pp. 677–686. 关33兴 Chen, S. Y., and Rajan, S. D., 1998, “Improving the Efficiency of Generatic
关13兴 Bendse, M. P., and Kikuchi, N., 1988, “Generating Optimal Topologies in Algorithms for Frame Designs, J. Eng. Optimization,” 30共3/4兲, pp. 281–307.
Structural Design Using A Homogenization Method,” Comput. Methods Appl. 关34兴 Krishnamoorthy, C. S., 2001, “Structural Optimization in Practice: Potential
Mech. Eng., 71共2兲, pp. 197–224. Applications of Genetic Algorithms,” Struct. Eng. Mech., 11共2兲, pp. 151–170.
关14兴 Suzuki, K., and Kikuchi, N., 1991, “A Homogenization Method for Shape and 关35兴 Deb, K., and Gulati, S., 2001, “Design of Truss-Structures for Minimum
Topology Optimization,” Comput. Methods Appl. Mech. Eng., 93, pp. 291– Weight Using Genetic Algorithms,” Finite Elem. Anal. Design, 37共5兲, pp.
318. 447–465.
关15兴 Larsen, V. D., and Sigmund, O., 1997, “Bouwstra S. Design and Fabrication of 关36兴 Crisfield, M. A., 1991, Non-Linear Finite Element Analysis of Solids and
Compliant Micromechanisms and Structures with Negative Poisson’s Ratio,” Structures, Wiley, New York.
J. Microelectromech. Syst., 6共2兲, pp. 99–106. 关37兴 Belytschko, T., Liu, W. K., and Moran, B., 2000, Nonlinear Finite Elements
关16兴 Sigmund, O., 1997, “On the Design of Compliant Mechanisms Using Topol- for Continua and Structures, Wiley, New York.

Journal of Mechanical Design JULY 2005, Vol. 127 / 759

Downloaded 27 Apr 2011 to 147.91.1.45. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm