You are on page 1of 16

General Approach for Topology-Finding

of Tensegrity Structures
Xian Xu 1; Yafeng Wang 2; and Yaozhi Luo, A.M.ASCE 3
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Abstract: This paper proposes a general approach for topology-finding of tensegrity structures. The ground structure method that is widely
used in topology optimization of truss structures is employed. An optimization model, using the member connections and the internal forces
of members as variables, is developed to find self-stressed tensegrities under given constraints on member distribution, configuration
symmetry, and node stability. Mixed integer linear/quadratic programming is used to solve the optimization problem. Numerical examples
are carried out to verify the proposed approach. Compared to previous approaches, the proposed approach is able to find a wider scope of
tensegrity structures that have a maximum of k (k ≥ 1) struts connecting to a single node and is able to deal with objective functions in
quadratic form. DOI: 10.1061/(ASCE)ST.1943-541X.0001532. © 2016 American Society of Civil Engineers.
Author keywords: Tensegrity; Topology; Optimization; Ground structure method; Structural optimization.

Introduction various numerical methods (Quirant et al. 2003; Quirant 2007; Tran
and Lee 2010). This process is termed force-finding by some re-
Tensegrity structures are stable self-stressed pin-jointed systems searchers (Tran and Lee 2010; Xu and Luo 2010). Whether or
consisting of tensile members and compressive members. There not a solution can be found for a force-finding problem is heavily
are various definitions of tensegrity (Motro 2003). The early def- dependent on the given parameters (i.e., topology, geometry, and
initions given by its inventors require that each node receives one tensile and compressive member distribution) which are predeter-
and only one strut, and the extended definition developed by mined based on the knowledge and experience of the designer. As a
engineering researchers allows more than one strut connected to result, force-finding usually deals with simple and regular tensegr-
each node (Motro 2003). In this paper, the extended definition ity structures whose existence can be qualitatively judged in
is adopted. To systematically classify tensegrity structures, Skelton advance. Most previous works on finding of tensegrity structures
and de Oliveira (2009) proposed the notion of Class k tensegrity are carried out in geometric space, termed form-finding, which in-
structures, where k is the maximum number of struts that touch cludes dynamic relaxation procedure-based methods (Zhang et al.
each other. This classification is also adopted in this paper. Since 2006; Ali et al. 2011), force density formulation-based methods
they were invented in the middle of last century, tensegrity struc- (Vassart and Motro 1999; Masic et al. 2005; Estrada et al. 2006;
tures have attracted considerable attention from a wide range of Zhang and Ohsaki 2006), equilibrium matrix-based methods (Tran
engineering and scientific communities, such as building structures and Lee 2010, 2011), finite-element method (Pagitz and Mirats Tur
(Raducanu 2001; Kono et al. 1999; Pedretti 1998), bridge struc- 2009), and so on. A significant drawback of form-finding is that the
tures (Rhode-Barbarigos et al. 2010, 2012), space structures final nodal locations of the found tensegrity structures are unpre-
(Furuya 1992; Hanaor 1993; Tibert and Pellegrino 2002), cell me- dictable and uncontrollable. If topology, instead of geometry, is
chanics (Ingber 2003; Stamenovic 2005; Xu and Luo 2009), taken as variables in the searching of tensegrity structures, the pre-
robotics (Aldrich et al. 2003; Paul et al. 2006; Rovira and Tur viously mentioned drawback will be eliminated. Recently, Ehara
2009), and so on. and Kanno (2010), and Kanno (2012, 2013a) used mixed-integer
The self-stressibility and stability of a tensegrity structure is de- linear programming (MILP) to find the feasible topology of ten-
pendent on the topology (member connections), geometry (nodal segrity structures. The MILP method showed a promising potential
positions), and distribution of tensile and compressive members. in finding and optimization of Class 1 tensegrity structures (i.e., ten-
Given the topology, geometry, and tensile and compressive member segrity structures with one strut connected to each node) under vari-
distribution of a tensegrity structure, the self-stress mode that ous constraints (Kanno 2013a, b). However, topology-finding/
satisfies the unilateral properties of the members and meanwhile optimization of Class kðk > 1Þ tensegrity structures (i.e., tensegrity
stabilizes the structural system can be determined/designed by structures with at most k struts connected to each node) has not been
addressed yet, and only linear optimization objectives can be applied
1
Associate Professor, Dept. of Civil Engineering, Zhejiang Univ., A- in the MILP method due to the inherent limitation of the algorithm.
823 Anzhong Bldg., 866 Yuhangtang Rd., Hangzhou, Zhejiang 310058, Based on the previous studies, this paper proposes a more-
China (corresponding author). E-mail: xian_xu@zju.edu.cn general approach for topology-finding/optimization of tensegrity
2
Graduate Student, Dept. of Civil Engineering, Zhejiang Univ., A-818 structures. The ground structure method proposed by Dorn et al.
Anzhong Bldg., 866 Yuhangtang Rd., Hangzhou, Zhejiang 310058, China. (1964) and widely used in topology optimization of truss structures
3
Professor, Dept. of Civil Engineering, Zhejiang Univ., A-821 Anzhong
is employed. The member connections, together with the internal
Bldg., 866 Yuhangtang Rd., Hangzhou, Zhejiang 310058, China.
Note. This manuscript was submitted on June 24, 2015; approved on
forces of members, are used as variables. A quadratic expression
February 2, 2016; published online on April 13, 2016. Discussion period that can approximately evaluate the evenness of prestress in cables
open until September 13, 2016; separate discussions must be submitted is established and used as an objective function. An optimization
for individual papers. This paper is part of the Journal of Structural En- model that is able to handle both quadratic objectives and linear
gineering, © ASCE, ISSN 0733-9445. objectives is developed and mixed-integer quadratic programming

© ASCE 04016061-1 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


8 P
(MIQP) or mixed-integer linear programming (MILP) is used to >
> 0 if xi ¼ 0
solve the optimization problem. Numerical examples are carried < i∈Ej
mj ¼ P ð3Þ
out to verify the proposed approach. The main contributions of >
the paper include (1) the proposed approach is able to deal with :1
> if
i∈Ej
xi ≥ 1
Class kðk ≥ 1Þ tensegrities; and (2) the optimization of the even-
ness of cable prestress is realized by using a quadratic objective
and MIQP. Since xi ∈ f0,1g, Eq. (3) is equivalent to
The paper is organized as follows. The section “Problem Def- 1 X X
inition” defines the variables and the constraint conditions for the xi ≤ mj ≤ xi ; ∀j∈B ð4aÞ
jEj j i∈E i∈E
topology-finding problem. The problem is formulated as a MILP/ j j

MIQP model in the section “MILP/MIQP Formulation.” The sec-


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

tion “Numerical Examples” presents four typical examples to mj ∈ 0; 1 ð4bÞ


verify the proposed approach. Finally, the paper is concluded by
the section “Conclusions.”
More details on the definition of mj can be found in the liter-
ature (Kanno 2013a).
Problem Definition

A self-stressed tensegrity structure must satisfy some necessary Class k Condition


conditions, such as self-equilibrium and prestress stability. If For a Class k tensegrity structure, there are at most k struts
the structure is required to be Class 1, it should also satisfy the connecting to each node, which can be expressed as
discontinuity condition of struts, i.e., each node is connected by X
only one strut. xi ≤ k; ∀ j ∈ N ð5Þ
As for a practical tensegrity structure, in addition to the preced- i∈Ej
ing conditions, it should also satisfy some additional conditions.
For instance, the members should not intersect, all the members
connecting to a node should not be coplanar, and a certain degree Equilibrium Condition
of symmetry may be required. Furthermore, the internal forces of Let t ¼ ðt1 ; t2 ; : : : ; tb ÞT , where b is the number of members of the
members are usually required to close to each other so that a small ground structure, denote the internal forces of the members, and it
number of member sections will be needed. In this paper, all these should satisfy the static equilibrium that
conditions except the prestress stability will be integrated into the
optimization model. The prestress stability of the possible tensegr- At ¼ 0 ð6Þ
ity structures will be checked after they are identified by the
where A = equilibrium matrix (Pellegrino and Calladine 1986).
optimization algorithm.
The ground structure is generated based the given nodes in the
three-dimensional space. N and E are used to denote the set of Unilateral Internal Force Constraint
nodes and the set of candidate members, respectively, and Ej is
used to denote the label set of members connecting to node j. The unilateral internal force constraint formulated by Kanno (2012)
is used here. Let M and ε be positive constants, where M is suffi-
ciently large, i.e., 0 < ε ≪ M. Consider the linear inequalities
Integer Variable Definition
−Mxi ≤ ti ≤ Mð1 − xi Þ − ε; ∀i∈E ð7aÞ
Two 0-1 variables, xi and yi , are used to describe the state of
member i as (Ehara and Kanno 2010)
−Myi ≤ −ti ≤ Mð1 − yi Þ − ε; ∀i∈E ð7bÞ
ðxi ; yi Þ ¼ ð1; 0Þ ⇔ member i is a strut ð1aÞ

Since xi ∈ f0,1g, yi ∈ f0,1g, then one can see that Eqs. (7a) and
ðxi ; yi Þ ¼ ð0; 1Þ ⇔ member i is a cable ð1bÞ (7b) is equivalent to
8
< ti < 0 ⇔ xi ¼ 1 ⇔ member i is a strut
ti > 0 ⇔ yi ¼ 1 ⇔ member i is a cable ð8Þ
ðxi ; yi Þ ¼ ð0; 0Þ ⇔ member i is removed ð1cÞ :
ti ¼ 0 ⇔ xi ¼ yi ¼ 0 ⇔ member i is removed

In order to study the internal forces of cables and the internal


In order to consider the symmetry of the structure to some forces of struts respectively, the internal force vector is divided into
extent, a 0–1 variable, mj , is used (Kanno 2013a). Let li denote two parts, i.e.
the length of member i of the ground structure, and let Bj ∈ E be
a set of members that have the same length. Then, mj is defined as t ¼ tcable þ tstrut ð9Þ
 where tcable ¼ jEj × 1 vector in which the components correspond-
0 if there is no strut in Bj
mj ¼ ð2Þ ing to cable members are positive and all the other components are
1 if there is at least one strut in Bj zero; and tstrut is also a jEj × 1 vector in which the components
corresponding to strut members are negative and all the other
By using the variable xi defined in Eq. (1), Eq. (2) can be reex- components are zero. Hence, for a component tcable i of vector
pressed as tcable and a component tstrut
i of vector tstrut, there are

© ASCE 04016061-2 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


−Myi ≤ −tcable
i ≤ −εyi ; ∀i∈E ð10aÞ where Pcross = pairs of intersected ground structure members. More
details about this constraint can be found in Kanno (2012, 2013a).
−Mxi ≤ tstrut
i ≤ −εxi ; ∀i∈E ð10bÞ
Additional Constraints
For the tensegrity structures considers in this paper, it is required
Node Stability Conditions that each node should be connected by both cable(s) and strut(s),
The stability of the whole tensegrity system cannot be formulated i.e., any case that only cables or struts connect to a node is not
as a constraint with the given variables and thus does not considered accepted. To meet this requirement, additional constraints are in-
within the main frame of the proposed optimization model. After- troduced as
wards, a check is needed to determine whether a found tensegrity X
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

structure is a stable or unstable system, as that carried out in the yi ≤ xi 0 ð∀i ∈ Ej Þ; ∀ j ∈ N ð16aÞ
i 0 ∈Ej
examples of this paper. Since node stability is a necessary condition
for the stability of the whole system, two conditions for node X
stability are considered. yi ≤ xi 0 ð∀i ∈ Ej Þ; ∀j∈N ð16bÞ
To ensure stability of a node in space, at least three members are i 0 ∈Ej
required to be connected to it. This condition can be written as
X
ðxi þ yi Þ ≥ 3ā; ∀ j ∈ N ā ∈ f0; 1g ð11Þ
i∈Ej
Valid Inequalities
When a mixed-integer programming is solved with branch-and-cut
where ā = parameter introduced to consider whether all the algorithm, adding some valid inequalities to the constraints can dra-
given nodes must be used in the final tensegrity structure and it matically improve the computational efficiency. In this paper, the
is given by following valid inequality is used:
(
1 if all nodes must be used xi þ yi ≤ 1; ∀i∈E ð17Þ
ā ¼ ð12Þ
0 if not all nodes must be used
which indicates that for each ground structure member i, it is kept
as a strut or a cable or is removed in the final tensegrity structure.
Meanwhile, it is further required that the members connected to
the node should not in a same plane otherwise the node will has
zero stiffness in the normal of the plane (i.e., unstable). This re- Objective Functions and Control Parameters
quirement can be formulated as
P Objective Functions
X i∈Ej ðxi þ yi Þ X
ðxi þ yi Þ þ ≤ ðxi þ yi Þð∀p ∈ Pj Þ; ∀ j ∈ N Two linear objective functions that have been used by Ehara and
i∈cp
jEj j i∈E Kanno (2010) and Kanno (2013a) respectively are adopted in this
j j
paper. The first one f 1 is the number of cables, and the second one
ð13Þ f2 is the total length of cables. These two objective functions can be
expressed as
where Pj ¼ f1,2; : : : ; pj g = label set of coplanar members
connecting to node j, in which pj = number of coplanar member X
n

sets of node j; and cpj = coplanar member set in the pth plane f1 ¼ yi ð18Þ
i¼1
of node j. Eq. (13) is also referred as member coplanarity
constraint.
X
n
f2 ¼ li y i ð19Þ
Geometrical Conditions i¼1

where li = length of member i.


Structure Symmetry Constraint Sometimes the designers expect the internal forces in the
It has been verified that a high order of symmetry in geometry usu- members of the same type (i.e., cable or strut) are close to each
ally implies that the number of different member lengths is small other so that the number of member sections can be reduced for
(Kanno 2013a). To manipulate the symmetry of the found tensegr- the convenience of manufacture or the structural material can be
ity structure, the number of different lengths of struts b̄ can be used efficiently used. To consider this requirement, the variance g of
as a constraint, which is given by the components of vector tcable, which can be seemed as an approxi-
X mate evaluation on the unevenness of the distribution of cable
mj ¼ b̄ ð14Þ forces, is introduced as a new objective. This new objective is
j∈B given by
Pn
ðtcable − t̄Þ2
Member Intersection Constraint g ¼ i¼1 i ð20Þ
n
To ensure that there is no intersection between any two members in P
the final tensegrity, any pairs of intersected ground structure where t̄ ¼ ni¼1 tcable
i =n, and n = number of ground structure
members i and i 0 should not simultaneously adopted by the final members. A similar objective function also can be formulated
tensegrity. By using Eq. (1), this constraint can be expressed as for the internal forces of struts if necessary.
Since tcable consists of nonzero components for final cables and
xi þ xi 0 þ yi þ yi 0 ≤ 1; ∀ ði; i 0 Þ ∈ Pcross ð15Þ zero components for removed members and strut members, t̄ is not

© ASCE 04016061-3 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


P
an accurate expression for the average internal force of cables and g X i∈Ej ðxi þ yi Þ X
is also not an accurate expression for the variance of the internal ðxi þ yi Þ þ ≤ ðxi þ yi Þð∀p ∈ Pj Þ; ∀ j ∈ N
i∈cpj
jEj j i∈Ej
forces of cables. Due to this inherent characteristics, if g is inde-
pendently used as an objective function, trial computations reveal ð22kÞ
that it is prone to yields tensegrity systems with a larger number of
cables. To overcome this shortage, an improved objective function
f 3 , which combines g with f 1 , is proposed, and it is given by 1 X X
xi ≤ mj ≤ xi ; ∀j∈B ð22lÞ
jEj j i∈E i∈E
j j
f 3 ¼ wf 1 þ g ð21Þ
X
where w = combination coefficient which is large enough to ensure yi ≤ xi 0 ð∀i ∈ Ej Þ; ∀j∈N ð22mÞ
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

that f1 is prior to g during the optimization. The purpose of objec- i 0 ∈Ej


tive function f3 is not only to minimize the number of cables but
also to even the cable forces to some extent in the final tensegrity X
structure. xi ≤ yi 0 ð∀i ∈ Ej Þ; ∀j∈N ð22nÞ
i 0 ∈Ej
Since g has a quadratic form, the MILP proposed by previous
studies (Ehara and Kanno 2010; Kanno 2013a) is not applicable for
both g and f 3 . An alternative algorithm, MIQP, is used to handle xi þ yi ≤ 1; ∀i∈E ð22oÞ
these objective functions in quadratic form.
X
Control Parameters mj ¼ b̄ ð22pÞ
j∈B
The first control parameter is k in Eq. (5), which is named here as
Tensegrity Class Parameter. It controls the maximum number of
struts connecting to a single node of the final tensegrity structure. xi ∈ f0; 1g; yi ∈ f0; 1g; ∀i∈E ð22qÞ
The second control parameter is b̄ in Eq. (14), which is here
named as Symmetry Parameter. It controls the number of strut mj ∈ f0; 1g; ∀j∈B ð22rÞ
lengths in the final tensegrity structure. The number of cable
lengths also can be chosen as Symmetry Parameter and can be dealt
with in the same manner. ā ∈ f0; 1g ð22sÞ
The third control parameter is ā in Eq. (11), which is named as
Node Parameter here. It controls whether all the given nodes must
be used in the final tensegrity structure. Numerical Examples

In this section, four typical numerical examples are carried out


MILP/MIQP Formulation to verify the proposed approach. The constant parameters used
in these examples are given as follows: δ ¼ 0.1, M ¼ 100, and
The topology finding/optimization problem mentioned earlier can ε ¼ 0.01. The MILP/MIQP problem is solved by Gurobi Optimizer
be formulated into a MILP/MIQP problem, which gives 6.0.0.
min f ð22aÞ
x;y;m;t
Example 1: Class 1 Tensegrity Structures Using f 1 and
f3
s:t: At ¼ 0 ð22bÞ A ground structure with jNj ¼ 10 and jEj ¼ 45, as shown in Fig. 1,
is considered in this example. This ground structure has been used
t ¼ tcable þ tstrut ð22cÞ by Ehara and Kanno (2010) who found a Class 1 tensegrity with
minimum number of cables using MILP. Here, firstly the same op-
X timization objective as that used by Ehara and Kanno (2010),
xi ≤ k; ∀j∈N ð22dÞ i.e., Eq. (18), is used to verify that the proposed approach is able
i∈Ej

−Mxi ≤ ti ≤ Mð1 − xi Þ − ε; ∀i∈E ð22eÞ

−Myi ≤ −ti ≤ Mð1 − yi Þ − ε; ∀i∈E ð22fÞ

−Mxi ≤ tstrut
i ≤ −εxi ; ∀i∈E ð22gÞ

−Myi ≤ −tcable
i ≤ −εyi ; ∀i∈E ð22hÞ

X
ðxi þ yi Þ ≥ 3ā; ∀j∈N ð22iÞ
i∈Ej

xi þ xi 0 þ yi þ yi 0 ≤ 1; ∀ ði; i 0 Þ ∈ Pcross ð22jÞ Fig. 1. Ground structure of Example 1

© ASCE 04016061-4 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Solutions of Example 1: (a) E1-TS1-1 (f 1 ); (b) E1-TS1-2(f3 )

As shown in Fig. 2(a), when the optimization objective (i.e., f1 )


Table 1. Cable Forces of Tensegrity Structures Found by Using Different same as that used by Ehara and Kanno (2010) is used, the found
Optimization Objectives in Example 1 tensegrity E1-TS1-1 has 5 struts and 16 cables and is same as that
found by Ehara and Kanno (2010). In Fig. 2(a) struts are repre-
Normalized cable Normalized cable
Cable number forces of E1-TS1-1 forces of E1-TS1-2 sented by thick lines and cables are represented by thin lines. This
representation will be used throughout the paper. Fig. 2(b) is the
1 0.4134 0.3514 tensegrity structure found by using f 3 as the optimization objective.
2 0.4134 0.3514
It also has 5 struts and 16 cables and is referred as E1-TS1-2. Both
3 1.0000 0.8498
4 1.0000 0.0556 tensegrity structures have one self-stress mode and four infinitesi-
5 0.0654 0.7071 mal mechanism modes. Further check reveals that for each system,
6 0.8321 0.7071 the infinitesimal mechanism modes can be eliminated by the self-
7 0.8321 0.3333 stress mode, i.e., both tensegrity structures are stable. The normal-
8 0.3922 1.0000 ized cable forces of the two tensegrity structures, together with the
9 0.8321 0.7071 real variances of cable forces, are given in Table 1. It is shown that
10 0.3922 1.0000 the cable force variance of E1-TS1-2 is smaller than that of
11 0.8321 0.7071 E1-TS1-1, i.e., the cable forces of E1-TS1-2 are more evenly dis-
12 1.0000 0.3333
tributed than those of E1-TS1-1.
13 1.0000 0.8498
14 0.0654 0.0556
In order to investigate the effects of the member intersection
15 0.4134 0.3514 constraint Eq. (15) and the member coplanarity constraint Eq. (13),
16 0.4134 0.3514 computations without one or both of them are also conducted. The
Variances 0.1029 0.0890 objection function f 1 is used in these computations.
As shown in Fig. 3(a), when the member intersection constraint
is not considered, the found tensegrity has only 15 cables which is
less than those given in Fig. 2, but there are a strut and a cable
to get the same result as the previous method. Then, the new opti- intersecting with each other. When both the member intersection
mization objective proposed in this paper, i.e., Eq. (21), is used to constraint and the member coplanarity constraint are not consid-
show that the proposed approach is also able to deal with objective ered, there are only 12 cables in the found tensegrity, as shown
functions in quadratic form. k ¼ 1 and ā ¼ 1 are used and no con- in Fig. 3(b), but member coplanarity is observed at a number of
trol on b̄ is imposed. nodes and all the struts intersect with each other. Moreover, further

Fig. 3. Solutions without some constraint(s): (a) without member intersection constraint; (b) without member intersection and member coplanarity
constraint

© ASCE 04016061-5 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Table 2. Summary of Found Tensegrity Structures in Example 2
Number of
Number infinitesimal
Objective Tensegrity Number Variance of self-stress mechanism
function structure of cables of cable forces modes modes
f1 E2-TS2-1 16 0.0494 1 5
f3 E2-TS2-2 16 0.0000 1 6
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Ground structure of Example 2

check finds that both of them are unstable under the prestress given
in the solutions. It seems that both the member intersection con-
straint and the member coplanarity constraint should be considered
to ensure practical applicability and prestress stability of the found
tensegrity systems.

Example 2: Class 2 Tensegrity Structures


Using f 1 and f 3
In this example, a ground structure with jNj ¼ 12 and jEj ¼ 66, as
shown in Fig. 4, is considered. The nodes of this ground structure Fig. 6. Ground structure of Example 3
are located in a way that six of them form a regular hexagon and the
other six of them form two regular triangles. The centers of the
hexagon and triangles are located on the X 3 -axis. The hexagon
and triangles are all parallel to the X 1 X 2 -plane. The two triangles the cable forces of E2-TS2-2 are more evenly distributed than those
have a relative rotation of 180° to each other. The two optimization of E2-TS2-1.
objectives used in Example 1 are also adopted in this example. The Both tensegrity structures have one self-stress mode and more
control parameters used in this example are k ¼ 2, b̄ ¼ 3, than one infinitesimal mechanism mode. Nonetheless, further
and ā ¼ 1. check finds that each of them can be stabilized by the correspond-
Fig. 5(a) shows the tensegrity found by using f 1 as the optimi- ing prestress t obtained in the solution.
zation objective. It has 10 struts and 16 cables and is referred as
E2-TS2-1. There are eight nodes, each of which is connected by
two struts, which indicates that this is a Class 2 tensegrity structure. Example 3: Class 1 Tensegrity Structures Using All
Fig. 5(b) shows the tensegrity structure found by using f3 as the Three Objective Functions
optimization objective. It has 9 struts and 16 cables and is referred In this example, a ground structure modified from the one used in
as E2-TS2-2. It is also a Class 2 tensegrity structure with six nodes, Example 2 is considered. It is generated by removing a pair of
each of which is connected by two struts. A summary of the proper- diagonal vertices of the hexagon from the ground structure of Ex-
ties of the found tensegrity structures is given in Table 2. It shows ample 2 and adding two new vertices on the right above/below the
that the cable force variance of E2-TS2-2 is zero, which indicates centroids of the triangles, as shown in Fig. 6. Both of the new added
that all the cables of E2-TS2-2 have the same internal force. Since vertices are on X 3 -axis and the whole ground structure is symmet-
E2-TS2-1 has a nonzero cable force variance, it is obviously that rical about X 1 X 2 -plane. It also has jNj ¼ 12 and jEj ¼ 66.

Fig. 5. Solutions of Example 2: (a) E2-TS2-1 (f1 ); (b) E2-TS2-2 (f3 )

© ASCE 04016061-6 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 7. Solutions of Example 3 (b̄ ¼ 2): (a) E3-TS1-1-1 (f1 ); (b) E3-TS1-1-2 (f 2 ); (c) E3-TS1-1-3 (f3 )

Fig. 8. Solutions of Example 3 (b̄ ¼ 4): (a) E3-TS1-2-1 (f1 ); (b) E3-TS1-2-2 (f 2 ); (c) E3-TS1-2-3 (f3 )

© ASCE 04016061-7 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


All the three objective functions are used in this example and the Example 4: Class 1 and Class 2 Tensegrity Structures
control parameters k and ā are set to be 1. Meanwhile, the control Using a  0
parameter b̄, ranging from 1 to 6, is used to generate tensegrity The ground structure considered in this example is same as that
structures with different level of symmetry. For b̄ ¼ 2, 4, and 5, used in Example 3 and all the three objective functions are also
stable tensegrity structures are found and they are shown in
used. Here it is allowed that some nodes of the ground structure
Figs. 7–9. They are summarized in Table 3, which shows that
are not used by the final tensegrity structure, i.e., ā ¼ 0. k ¼ 1
the property corresponding to the used objective function is opti-
mized in each group of tensegrity structures found under the same and 2 are investigated respectively. For k ¼ 1, since the number
b̄. All the found tensegrity structures have six discontinuous struts of given nodes is 12, the maximum number of struts, as well as
because k ¼ 1 is used. In addition, a possible tensegrity structure the maximum number of strut lengths, for the possible final ten-
that satisfies all the constraints is also obtained with b̄ ¼ 3, but fur- segrity structures is six. As a result, b̄ ranged from 1 to 6 are con-
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

ther check finds that it is unstable. For b̄ ¼ 1 or b̄ ¼ 6, no structure sidered to find all the possible Class 1 tensegrity structures.
satisfies all the constraints is found. Similarly, for k ¼ 2, the maximum number of strut lengths is

Fig. 9. Solutions of Example 3 (b̄ ¼ 5): (a) E3-TS1-3-1 (f1 ); (b) E3-TS1-3-2 (f 2 ); (c) E3-TS1-3-3 (f3 )

Table 3. Summary of Found Tensegrity Structures in Example 3


Objective Tensegrity Number Total cable Variance Number of Number of infinitesimal
b̄ function structure of cables length of cable forces self-stress modes mechanism modes
2 f1 E3-TS1-1-1 22 49.1795 0.0937 1 3
f2 E3-TS1-1-2 22 47.3814 0.0704 1 3
f3 E3-TS1-1-3 22 49.1795 0.0558 2 4
4 f1 E3-TS1-2-1 23 50.8643 0.0535 1 2
f2 E3-TS1-2-2 24 49.5475 0.0577 1 1
f3 E3-TS1-2-3 23 50.8643 0.0535 1 2
5 f1 E3-TS1-3-1 23 56.0616 0.0621 1 2
f2 E3-TS1-3-2 24 49.5475 0.0649 1 1
f3 E3-TS1-3-3 23 55.1219 0.0539 1 2

© ASCE 04016061-8 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. Solutions of Example 4 (k ¼ 1, b̄ ¼ 2): (a) E4-TS1-1-1 (f1 ); (b) E4-TS1-1-2 (f 2 ); (c) E4-TS1-1-3 (f 3 )

Fig. 11. Solutions of Example 4 (k ¼ 1, b̄ ¼ 3): (a) E4-TS1-2-1 (f1 ); (b) E4-TS1-2-2 (f 2 ); (c) E4-TS1-2-3 (f 3 )

© ASCE 04016061-9 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 12. Solutions of Example 4 (k ¼ 1, b̄ ¼ 4): (a) E4-TS1-3-1 (f1 ); (b) E4-TS1-3-2 (f 2 ); (c) E4-TS1-3-3 (f 3 )

Fig. 13. Solutions of Example 4 (k ¼ 1, b̄ ¼ 5): (a) E4-TS1-4-1 (f1 ); (b) E4-TS1-4-2 (f 2 ); (c) E4-TS1-4-3 (f 3 )

© ASCE 04016061-10 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


12, and b̄ ranges from 1 to 12 are considered to find all the possible found by using f 1 (E4-TS2-3-1) is same to the one found by using
Class 2 tensegrity structures. f2 (E4-TS2-3-2) and in the results with b̄ ¼ 7, the tensegrity found
When k ¼ 1 is used, solutions are found with b̄ ¼ 2–5. The by using f1 (E4-TS2-6-1) is same to the one found by using f3 (E4-
found tensegrity structures together with the nodes have not been TS2-6-3).
used are given in Figs. 10–13. They are summarized in Table 4. It Similar to the Example 3, in each group of tensegrity structures
shows that the found tensegrity structures with b̄ ¼ 2 have identical using the same control parameters, the property corresponding to
properties (Table 4). A closer check on their configurations finds the used objective function has been optimized (Tables 4 and 5). All
that they are identical systems locating in different orientations the tensegrity structures found in this example have one self-stress
(Fig. 10). All the given nodes have been used by the final tensegrity mode. There is no infinitesimal mechanism for most of the tensegr-
structures with b̄ ¼ 5 (Fig. 13) and they are same as those found in ity structures found in this example. For these tensegrity structures
Example 3 with the same b̄ (Fig. 9). without any infinitesimal mechanism, they are kinetically deter-
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

When k ¼ 2 is used, solutions are found with b̄ ¼ 2–9. The mined systems and unconditionally stable. For the other tensegrity
found tensegrity structures together with the nodes that have not structures, further checks find that their infinitesimal mechanism
been used are given in Figs. 14–21. They are summarized in can be eliminated by the corresponding self-stress. Hence, they
Table 5. It is found that in the results with b̄ ¼ 4, the tensegrity are stable too.

Table 4. Summary of Found Class 1 Tensegrity Structures in Example 4


Objective Tensegrity Number Variance Number of Number of infinitesimal
b̄ functions structures of cables Total cable length of cable forces self-stress modes mechanism modes
2 f1 E4-TS1-1-1 9 24.0977 0.0604 1 1
f2 E4-TS1-1-2 9 24.0977 0.0604 1 1
f3 E4-TS1-1-3 9 24.0977 0.0604 1 1
3 f1 E4-TS1-2-1 10 26.5773 0.0986 1 0
f2 E4-TS1-2-2 10 25.7242 0.0617 1 0
f3 E4-TS1-2-3 10 27.0251 0.0472 1 0
4 f1 E4-TS1-3-1 15 35.1069 0.0781 1 0
f2 E4-TS1-3-2 15 33.3087 0.0942 1 0
f3 E4-TS1-3-3 15 34.6808 0.0598 1 0
5 f1 E4-TS1-4-1 23 56.0616 0.0621 1 2
f2 E4-TS1-4-2 24 49.5475 0.0649 1 1
f3 E4-TS1-4-3 23 55.1219 0.0539 1 2

Fig. 14. Solutions of Example 4 (k ¼ 2, b̄ ¼ 2): (a) E4-TS2-1-1 (f1 ); (b) E4-TS2-1-2 (f 2 ); (c) E4-TS2-1-3 (f 3 )

© ASCE 04016061-11 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 15. Solutions of Example 4 (k ¼ 2, b̄ ¼ 3): (a) E4-TS2-2-1 (f1 ); (b) E4-TS2-2-2 (f 2 ); (c) E4-TS2-2-3 (f 3 )

Fig. 16. Solutions of Example 4 (k ¼ 2, b̄ ¼ 4): (a) E4-TS2-3-1 (f1 ); (b) E4-TS2-3-2 (f 2 ); (c) E4-TS2-3-3 (f 3 )

© ASCE 04016061-12 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 17. Solutions of Example 4 (k ¼ 2, b̄ ¼ 5): (a) E4-TS2-4-1 (f1 ); (b) E4-TS2-4-2 (f 2 ); (c) E4-TS2-4-3 (f 3 )

Fig. 18. Solutions of Example 4 (k ¼ 2, b̄ ¼ 6): (a) E4-TS2-5-1 (f1 ); (b) E4-TS2-5-2 (f 2 ); (c) E4-TS2-5-3 (f 3 )

© ASCE 04016061-13 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 19. Solutions of Example 4 (k ¼ 2, b̄ ¼ 7): (a) E4-TS2-6-1 (f1 ); (b) E4-TS2-6-2 (f 2 ); (c) E4-TS2-6-3 (f 3 )

Fig. 20. Solutions of Example 4 (k ¼ 2, b̄ ¼ 8): (a) E4-TS2-7-1 (f1 ); (b) E4-TS2-7-2 (f 2 ); (c) E4-TS2-7-3 (f 3 )

© ASCE 04016061-14 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

Fig. 21. Solutions of Example 4 (k ¼ 2, b̄ ¼ 9): (a) E4-TS2-8-1 (f1 ); (b) E4-TS2-8-2 (f 2 ); (c) E4-TS2-8-3 (f 3 )

Table 5. Summary of Found Class 2 Tensegrity Structures in Example 4


Objective Tensegrity Number Variance Number of Number of infinitesimal
b̄ function structure of cables Total cable length of cable forces self-stress modes mechanism modes
2 f1 E4-TS2-1-1 6 14.0161 0.1013 1 0
f2 E4-TS2-1-2 6 12.7337 0.0937 1 0
f3 E4-TS2-1-3 6 16.8764 0.0134 1 0
3 f1 E4-TS2-2-1 6 16.6546 0.0723 1 0
f2 E4-TS2-2-2 6 12.7337 0.0612 1 0
f3 E4-TS2-2-3 6 14.7596 0.0143 1 0
4 f1 E4-TS2-3-1 6 11.7311 0.1000 1 0
f2 E4-TS2-3-2 6 11.7311 0.1000 1 0
f3 E4-TS2-3-3 6 15.1863 0.0654 1 0
5 f1 E4-TS2-4-1 7 18.0244 0.0732 1 0
f2 E4-TS2-4-2 7 13.8883 0.0767 1 0
f3 E4-TS2-4-3 7 19.0785 0.0652 1 0
6 f1 E4-TS2-5-1 7 16.5792 0.0907 1 0
f2 E4-TS2-5-2 7 14.3144 0.0647 1 0
f3 E4-TS2-5-3 7 16.5582 0.0328 1 0
7 f1 E4-TS2-6-1 8 22.3167 0.0541 1 1
f2 E4-TS2-6-2 8 18.9069 0.0738 1 1
f3 E4-TS2-6-3 8 22.3167 0.0541 1 1
8 f1 E4-TS2-7-1 10 24.4383 0.0635 1 1
f2 E4-TS2-7-2 11 22.7916 0.0591 1 0
f3 E4-TS2-7-3 10 26.6735 0.0422 1 1
9 f1 E4-TS2-8-1 12 29.7151 0.1018 1 1
f2 E4-TS2-8-2 13 26.2161 0.0607 1 0
f3 E4-TS2-8-3 12 32.4165 0.0939 1 1

Conclusions a subassemblage of the corresponding ground structure. An opti-


mization model that uses the member connections and internal
Based on the previous studies (Kanno 2012, 2013a), a more- forces of members as variables is developed for the topology-find-
general approach for topology-finding tensegrity structures is pro- ing of tensegrity structures. Various objective functions can be ap-
posed in this paper. The possible tensegrity system is considered as plied to the optimization model, depending on the designers’

© ASCE 04016061-15 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061


preferences on the tensegrity structures needing to be generated. Motro, R. (2003). Tensegrity: Structural systems for the futures, Kogan
According to the style of objective functions, MILP or MIQP is Page Science, London.
used to solve the optimization model. The versatility of the pro- Pagitz, M., and Tur, J. M. (2009). “Finite element based form-finding
posed approach has been verified by a number of numerical exam- algorithm for tensegrity structures.” Int. J. Solids Struct., 46(17),
3235–3240.
ples. The proposed approach can be deemed as a generalized
Paul, C., Valero-Cuevas, F. J., and Lipson, H. (2006). “Design and control
version of previous methods and it is able to find a more-general of tensegrity robots for locomotion.” IEEE Trans. Rob., 22(5), 944–957.
scope of tensegrity structures. Pedretti, M. (1998). “Smart tensegrity structures for the Swiss Expo 2001.”
Proc., SPIE 3330, Smart Structures and Materials 1998: Sensory Phe-
nomena and Measurement Instrumentation for Smart Structures and
Acknowledgments Materials, International Society for Optics and Photonics, Washington,
DC, 378–386.
Downloaded from ascelibrary.org by HARBIN ENGINEERING UNIVERSITY on 09/03/18. Copyright ASCE. For personal use only; all rights reserved.

This work was supported by National Natural Science Foundation Pellegrino, S., and Calladine, C. R. (1986). “Matrix analysis of statically
of China (Grant Nos. 51378458 and 11402229) and Zhejiang Pro- and kinematically indeterminate frameworks.” Int. J. Solids Struct.,
vincial Natural Science Foundation (Grant Nos. LY13E080002 and 22(4), 409–428.
Quirant, J. (2007). “Self-stressed systems comprising elements with unilat-
LQ14A020003). The authors also would like to thank the anon-
eral rigidity: Selfstress states, mechanisms and tension setting.” Int. J.
ymous reviewers for their suggestions in improving the standard Space Struct., 22(4), 203–214.
of the manuscript. Quirant, J., Kazi-Aoual, M. N., and Laporte, R. (2003). “Tensegrity sys-
tems: The application of linear programmation in search of compatible
selfstress states.” J. Int. Assoc. Shell Spatial Struct., 44(1), 33–50.
References Raducanu, V. (2001). “Architecture et système constructif: Cas des sys-
tèmes de tenségrité.” Ph.D. thesis, Univ. de Montpellier II, Montpellier,
Aldrich, J. B., Skelton, R. E., and Kreutz-Delgado, K. (2003). “Control France (in French).
synthesis for a class of light and agile robotic tensegrity structures.” Rhode-Barbarigos, L., Ali, N. B. H., Motro, R., and Smith, I. F. (2010).
Proc., 2003 American Control Conf., IEEE Press, Piscataway, NJ, “Designing tensegrity modules for pedestrian bridges.” Eng. Struct.,
5245–5251. 32(4), 1158–1167.
Ali, N. B. H., Rhode-Barbarigos, L., and Smith, I. F. (2011). “Analysis of Rhode-Barbarigos, L., Ali, N. B. H., Motro, R., and Smith, I. F. (2012).
clustered tensegrity structures using a modified dynamic relaxation “Design aspects of a deployable tensegrity-hollow-rope footbridge.”
algorithm.” Int. J. Solids Struct., 48(5), 637–647. Int. J. Space Struct., 27(2–3), 81–96.
Dorn, W., Gomory, R., and Greenberg, H. (1964). “Automatic design of Rovira, A. G., and Tur, J. M. M. (2009). “Control and simulation of a
optimal structures.” J. Mech., 3(1), 25–52. tensegrity-based mobile robot.” Rob. Auton. Syst., 57(5), 526–535.
Ehara, S., and Kanno, Y. (2010). “Topology design of tensegrity structures Skelton, R. E., and de Oliveira, M. C. (2009). Tensegrity systems, Vol. 1,
via mixed integer programming.” Int. J. Solids Struct., 47(5), 571–579. Springer, New York.
Estrada, G. G., Bungartz, H. J., and Mohrdieck, C. (2006). “Numerical Stamenović, D. (2005). “Effects of cytoskeletal prestress on cell rheological
form-finding of tensegrity structures.” Int. J. Solids Struct., 43(22), behavior.” Acta Biomater., 1(3), 255–262.
6855–6868. Tibert, A. G., and Pellegrino, S. (2002). “Deployable tensegrity reflectors
Furuya, H. (1992). “Concept of deployable tensegrity structures in space for small satellites.” J. Spacecraft Rockets, 39(5), 701–709.
application.” Int. J. Space Struct., 7(2), 143–151. Tran, H. C., and Lee, J. (2010). “Initial self-stress design of tensegrity grid
Gurobi Optimizer 6.0.0 [Computer software]. Gurobi Optimization, structures.” Comput. Struct., 88(9), 558–566.
Houston. Tran, H. C., and Lee, J. (2011). “Determination of a unique configuration of
Hanaor, A. (1993). “Double-layer tensegrity grids as deployable struc- free-form tensegrity structures.” Acta Mech., 220(1–4), 331–348.
tures.” Int. J. Space Struct., 8(1), 135–143. Vassart, N., and Motro, R. (1999). “Multiparametered formfinding method:
Ingber, D. E. (2003). “Tensegrity. I: Cell structure and hierarchical systems Application to tensegrity systems.” Int. J. Space Struct., 14(2),
biology.” J. Cell Sci., 116(7), 1157–1173. 147–154.
Kanno, Y. (2012). “Topology optimization of tensegrity structures under Xu, X., and Luo, Y. (2009). “Tensegrity structures with buckling members
self-weight loads.” J. Oper. Res. Soc. Jpn., 55(2), 125–145. explain nonlinear stiffening and reversible softening of actin networks.”
Kanno, Y. (2013a). “Exploring new tensegrity structures via mixed integer J. Eng. Mech., 10.1061/(ASCE)EM.1943-7889.0000060, 1368–1374.
programming.” Struct. Multi. Optim., 48(1), 95–114. Xu, X., and Luo, Y. (2010). “Force-finding of tensegrity systems using si-
Kanno, Y. (2013b). “Topology optimization of tensegrity structures under mulated annealing algorithm.” J. Struct. Eng., 10.1061/(ASCE)ST
compliance constraint: A mixed integer linear programming approach.” .1943-541X.0000180, 1027–1031.
Optim. Eng., 14(1), 61–96. Zhang, J. Y., and Ohsaki, M. (2006). “Adaptive force density method for
Kono, Y., Choong, K. K., Shimada, T., and Kunieda, H. (1999). “An form-finding problem of tensegrity structures.” Int. J. Solids Struct.,
experimental investigation of a type of double layer tensegrity grids.” 43(18), 5658–5673.
J. Int. Assoc. Shell Spatial Struct., 40(2), 103–111. Zhang, L., Maurin, B., and Motro, R. (2006). “Form-finding of nonregular
Masic, M., Skelton, R. E., and Gill, P. E. (2005). “Algebraic tensegrity tensegrity systems.” J. Struct. Eng., 10.1061/(ASCE)0733-9445(2006)
form-finding.” Int. J. Solids Struct., 42(16), 4833–4858. 132:9(1435), 1435–1440.

© ASCE 04016061-16 J. Struct. Eng.

J. Struct. Eng., 2016, 142(10): 04016061

You might also like