Computers and Structures 88 (2010) 237246

Contents lists available at ScienceDirect

Computers and Structures

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

Advanced form-nding of tensegrity structures

Hoang Chi Tran, Jaehong Lee *
Department of Architectural Engineering, Sejong University, 98 Kunja Dong, Kwangjin Ku, Seoul 143-747, Republic of Korea

a r t i c l e

i n f o

Article history:
Received 22 May 2009
Accepted 28 October 2009
Available online 30 November 2009
Keywords:
Tensegrity structure
Single value decomposition
Form-nding
Eigenvalue problem

a b s t r a c t
A numerical method is presented for form-nding of tensegrity structures. The topology and the types of
members are the only information that requires in this form-nding process. The eigenvalue decomposition of the force density matrix and the single value decomposition of the equilibrium matrix are performed iteratively to nd the feasible sets of nodal coordinates and force densities which satisfy the
minimum required rank deciencies of the force density and equilibrium matrices, respectively. Based
on numerical examples it is found that the proposed method is very efcient and robust in searching
self-equilibrium congurations of tensegrity structures.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
The tensegrity structures rst proposed by Fuller [1] have been
developed in recent years due to their innovative forms, lightweight and deployability. They belong to a class of free-standing
pre-stressed pin-jointed cablestrut system where contacts are allowed among the struts [2]. Their classication is presented as
Class 1 (where bars do not touch) and Class 2 (where bars do connect to each other at a pivot) [3]. As a pioneering work of formnding, so-called force density method was proposed by Schek
[4] for form-nding of tensile structures. Motro et al. [5] presented
the dynamic relaxation which has been reliably applied to tensile
structures [6] and many other non-linear problems. Vassart and
Motro [7] employed the force density method in symbolic form
for searching new congurations. Recently, Masic et al. [8], Zhang
and Ohsaki [9] and Estrada et al. [10] developed new numerical
methods using a force density formulation. Micheletti et al. [11]
used a marching procedure for nding stable placements of a given
tensegrity, and Zhang et al. [12] employed a rened dynamic relaxation procedure for form-nding of nonregular tensegrity systems.
Most recently, Rieffel et al. [13] introduced an evolutionary algorithm for producing large irregular tensegrity structures. Tibert
and Pellegrino [14] presented a review paper for the existing methods for form-nding problem of tensegrity structures. The most recent review for this problem can be found in Juan and Tur [15].
In most available form-nding methods, the symmetry of structure and/or some of member lengths must be assumed known in
advance. Moreover, element force density coefcients are solved
in symbolic form. For example, (i) symmetric properties are
* Corresponding author. Tel.: +82 2 3408 3287; fax: +82 2 3408 3331.
E-mail address: [email protected] (J. Lee).
0045-7949/$ - see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2009.10.006

employed in a group theory so as to simplify form-nding problem

as presented in [8]; (ii) a number of member lengths are specied
at the start in a dynamic relaxation procedure and non-linear programming, as presented in [5,6,14]; however, these information
may not always be available or easy to dene at the beginning;
and (iii) force density coefcients are considered as symbolic variables which cannot be applied for structures with large number of
members [14].
In this paper, a numerical method is presented for form-nding
of tensegrity structures. Both Classes 1 and 2 of tensegrity structures are considered. Tensegrities satisfying either stability (i.e.,
the tangent stiffness matrix is positive denite) or super stability
(i.e., the geometrical stiffness matrix is positive denite) can be obtained by present form-nding procedure in a few of remarkable
iterations, which is more efcient and versatile than other available methods only dealing with super stable tensigrities (e.g.
[9,10]) that are more restrictive in the real mechanical structures.
The topology and the types of members, i.e. either compression
or tension, are the only information that requires in this form-nding process. In other words, any initial nodal coordinate, symmetric
properties, and member lengths are not necessary for the present
form-nding. The force density matrix is derived from an incidence
matrix and an initial set of force densities assigned from prototypes, while the equilibrium matrix is dened by the incidence matrix and nodal coordinates. The eigenvalue decomposition of the
force density matrix and the single value decomposition of the
equilibrium matrix are performed iteratively to nd the feasible
sets of nodal coordinates and force densities which satisfy the minimum required rank deciencies of the force density and equilibrium matrices, respectively. The evaluation of the eigenvalues of
tangent stiffness is also included for checking the stability of the
tensegrity structures.

238

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

Table 1
The incidence matrix of the 2-D two-strut tensegrity structure.

2. Formulation of self-equilibrium equations

2.1. Fundamental assumptions

In this study, the following assumptions are made in tensegrity

structures:
 The topology of the structure in terms of nodal connectivity is
known.
 Members are connected by pin joints.
 No external load is applied and the self-weight of the structure is
neglected during the form-nding procedure.
 There are no dissipative forces acting on the system.

2.2. Force density method for the self-equilibrium and free-standing

tensegrity structures
For a d-dimensional tensegrity structure with b members, n free
nodes and nf xed nodes (supports), its topology can be expressed
by a connectivity matrix Cs 2 Rbnnf as discussed in [4,16,9].
Suppose member k connects nodes i and ji < j), then the ith and
jth elements of the kth row of Cs are set to 1 and 1, respectively,
as follows:

Csk;p

8
for p i;
>
<1
1 for p j;
>
:
0
otherwise:

Member/node

(1)
(2)
(3)
(4)
(5)
(6)

1
0
0
1
1
0

1
1
0
0
0
1

0
1
1
0
1
0

0
0
1
1
0
1

Q diagq;

where q2 R suggested in [4] is the force density vector, dened

by

q fq1 ; q2 ; . . . ; qb gT

in which each component of this vector is the force fi to length li ratio qi fi =li known as force density or self-stressed coefcient in [7].
Without external loading, Eq. (3) can be rewritten neglecting the
self-weight of the structure as

Dx Df xf ;

6a

Dy Df yf ;

6b

Dz Df zf ;

6c
nn

where matrices D2 R

and Df 2 R

nnf

are, respectively, given by

D C QC;

7a

Df C QCf

If the free nodes are numbered rst, then to the xed nodes, Cs can
be divided into two parts as

or by

Cs C Cf ;

D CT diagqC;

7b

8a

where C2 Rbn and Cf 2 Rbnf describe the connectivities of the

members to the free and xed nodes, respectively. Let x; y; z2 Rn
and xf ; yf ; zf 2 Rnf denote the nodal coordinate vectors of the free
and xed nodes, respectively, in x, y and z directions. For a simple
two-dimensional two-strut tensegrity structure as shown in Fig. 1,
which consists of six members (b 6, four cables and two struts)
and four nodes, the connectivity matrix C2 R64 is given in Table
1.
The equilibrium equations in each direction of a general pinjointed structure given by [4] can be stated as

CT QCx CT QCf xf px ;

3a

CT QCy CT QCf yf py ;

3b

CT QCz CT QCf zf pz ;

3c

Df C diagqCf :

It is noted that when external load and self-weight are ignored, a

tensegrity system does not require any xed node, and the tensegrity geometry is dened by the relative position of the nodes, and
the system can be considered as free-standing, forming a rigid-body
free in space [7,16]. In this context, Eq. (7b) and (8b) vanish, and Eq.
(6) becomes:

Dx 0;

9a

Dy 0;

9b

Dz 0;

9c

where D known as force density matrix [14,10] or stress matrix

[1719] can be written directly by [7,18,19] without using Eq. (7)
or (8) as follows:

where px ; py and pz 2 R are the vectors of external loads applied

at the free nodes in x; y and z directions, respectively. The symbol,
T , denotes the transpose of a matrix or vector. And Q 2 Rbb is
diagonal square matrix, calculated by

8b

Di;j

8
q
>
>
< Pk
qk

>
> k2X
:
0

if nodes i and j are connected by member k;

for i j;
otherwise
10

in which X denotes the set of members connected to node i. For

example, for the two-dimensional two-strut tensegrity structure
in Fig. 1, D can be written explicitly from Eq. (10) as

4
(3)

(4)

(5)

1
(1)

3
(6)

(2)

2
Fig. 1. A two-dimensional two-strut tensegrity structure. The thick and thin lines
represent the struts and cables, respectively.

q1 q4 q5
6
q1
6
D6
4
q5
q4

q1
q1 q2 q6

q5
q2

q4
q6

q2

q2 q3 q5

q3

q6

q3

q3 q4 q6

3
7
7
7:
5
11

From Eq. (10), it is obvious that D is always square, symmetric and

singular with a nullity of at least 1 since the sum of all components

239

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

in any row or column is zero for any tensegrity structure. Different

from the matrix D of the cable net, which is always positive denite
[4], D of the tensegrity structure is semi-denite due to the existence of struts as compression members, with qk < 0. Consequently,
it cannot be invertible. For simplicity, Eq. (9) can be reorganized as

Dx y z 0 0 0:

12

However, by substituting Eq. (8) into Eq. (9) the equilibrium equations of the tensegrity structure can be expressed as

CT diagqCx 0;

13a

C diagqCy 0;

13b

CT diagqCz 0:

13c

Eq. (13) can be reorganized as

Aq 0;

20

where s is known as the number of independent states of self-stress,

while the number of innitesimal mechanisms is computed by
m dn  r A , as presented in [21,22]. It is clear that Eq. (20) allows
Eq. (14) to create at least one useful particular solution [20].
In short, based on these two rank deciency conditions, Eqs.
(17) and (20), the proposed form-nding procedure searches for
self-equilibrium congurations that permit the existence of at least
one state of self-stress in the structure. It should be noted that
these two are necessary but not sufcient conditions which have
to be satised for any d-dimensional tensegrity structure to be in
a self-equilibrium state [17,14,16]. The sufcient conditions for
tensegrity or self-stressed pin-jointed structures can be found in
[24,23].

14

where A2 Rdnb is known as the equilibrium matrix in [16], dened by

s nA P 1;

CT diagCx

B
C
A @ CT diagCy A:

15

C diagCz
Eq. (12) presents the relation between force densities and nodal
coordinates, while Eq. (14) shows the relation between projected
lengths in x; y and z directions, respectively and force densities. Both
Eqs. (12) and (14) are linear homogeneous systems of self-equilibrium equations with respect to nodal coordinates and force densities, respectively.

3. Requirement on rank deciency conditions of force density

and equilibrium matrices
Let q be the vector of force density and C be the incidence matrix of a d-dimensional tensegrity structure in self-equilibrium. It is
well known that the set of all solutions to the homogeneous system of Eq. (12) is the null space of D. The dimension of this null
space or rank deciency of D is dened as

nD n  rD ;

16

where rD rankD. It is obvious that vector I1 f1; 1; . . . ; 1gT

2 Rn1 , is a solution of Eq. (12) since the sum of the elements of
a row or a column of D is always equal to zero. The most important
rank deciency condition related to semi-denite matrix D of
Eq. (12) is dened by

nD P d 1:

17

This condition forces Eq. (12) to yield at least d useful particular

solutions [20] which exclude the above vector I1 due to degenerating geometry of tensegrity structure [14,9]. These d particular
solutions form a vector space basis for generating a d-dimensional tensegrity structure. Therefore, the minimum rank deciency or nullity of D must be d 1 for conguration of any
tensegrity structure embedding into Rd , which is equivalent to
the maximum rank condition of D proposed by [1719,16] as
follows:

4. Form-nding process
The proposed form-nding procedure only needs to know the
topology of structure in terms of the incidence matrix C, and type
of each member, i.e. either cable or strut which is under tension or
compression, respectively. Based on element type, the initial force
density coefcients of cables (tension) are automatically assigned
as +1 while those of struts (compression) as 1, respectively, as
follows:

8
<
q 1
1 1
:|{z}
0

cables

9T
=

1
{z
1     1} :
|
;

21

struts

Subsequently, the force density matrix D is calculated from q0 by

Eq. (8). After that, the nodal coordinates are selected from the
eigenvalue decomposition of the matrix D which is discussed in
the next section. These nodal coordinates are substituted into Eq.
(14) to dene force density vector q by the single value decomposition of the equilibrium matrix A which is also presented in the next
section. The force density matrix D is then updated by Eq. (8). The
process is iteratively calculated for searching a set of nodal coordinates x y z and force density vector q until the rank deciencies of
Eqs. (17) and (20) are satised, which forces Eqs. (12) and (14) become true. In this context, at least one state of self-stress can be created, s P 1. In this study, based on required rank deciencies from
Eqs. (17) and (20) the form-nding process is stopped as

nD d 1;
nA 1;

22a
22b

where nD and nA are minimum required rank deciencies of the
force density and equilibrium matrices, respectively.
4.1. Eigenvalue decomposition of force density matrix
The square symmetric force density matrix D can be factorized
as follows by using the eigenvalue decomposition [20]:

D UKUT ;

23

Similarly, the set of all solutions to the homogeneous system of Eq.

(14) lies in the null space of A. Let nA denote dimension of null space
of the equilibrium matrix A which is computed by

where U2 Rnn is the orthogonal matrix (UUT In , in which

In 2 Rnn is the unit matrix) whose ith column is the eigenvector basis /i 2 Rn of D. K2 Rnn is the diagonal matrix whose diagonal
elements are the corresponding eigenvalues, i.e., Kii ki . The eigenvector /i of U corresponds to eigenvalue ki of K. The eigenvalues are
in increasing order as

nA b  r A ;

k1 6 k2 6    6 kn :

maxrD n  d 1:

18

19

where r A rankA. The second rank deciency condition which ensures the existence of at least one state of self-stress can be stated as

24

It is clear that the number of zero eigenvalues of D is equal to

the dimension of its null space. Let p be the number of zero and

240

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

negative eigenvalues of D. There are two cases need to be considered. The rst one is p 6 nD , and the other is p > nD .
nD

Case 1. The rst

orthonormal eigenvectors of U are directly
taken as potential nodal coordinates

x y

 /
z 2 U
1

/2



/n :
D

25

The algorithm then iteratively modies the force density vector

q as small as possible to make the rst nD eigenvalues of D become null as

ki 0 i 1; 2; . . . ; nD :

26
bnD

All the projected lengths L2 R

b members are computed by

 C/ C/
L CU
1
2



 along n directions for

of U
D

C/n 

to remove one vector /i among

C/i 0

nD

27

 if
eigenvector bases of U

28

or which /i causes

detjLd Ld T j 0;

q
C/i 2 C/j 2 C/k 2 :

30

Case 2. Where p > nD , the rank deciency may be forced to be

larger than requirement or enough but D may not be positive
semi-denite during iteration. Additionally, the proposed
form-nding procedure will evaluate the tangent stiffness
matrix of the tensegrity structure which is given in
[24,23,26,9] as follows:

KT KE KG ;

31

where

ea

KG I3  D

AT ;

32a

k1 k2    kr 0;
|
{z}
rigid-body motions

34
where r dd 1=2 is the number of independent rigid-body
motions. Using this criterion, stability of any pre-stressed or
tensegrity structure can be controlled by checking eigenvalues
of tangent stiffness matrix of the structure [24,23]. In short,
the best scenario of conguration in 3-dimensional space is
formed by three best candidate eigenvectors selected from the
rst fourth eigenvector bases which corresponding to the rst
fourth smallest eigenvalues, respectively. These eigenvalues
will be gradually modied to be zero by the proposed iterative
form-nding algorithm. In other words, the proposed formnding procedure has repeatedly approximated equilibrium
conguration such that

Dx y

z 0 0 0:

35

The equilibrium matrix A is computed by substituting the set of

approximated nodal coordinates [x y z] from Eq. (35) into Eq.
(15). In order to solve linear homogeneous system (Eq. (14)) the
single value decomposition [20] is carried out on the equilibrium
matrix A:

A UVWT ;

36

where U2 Rdndn u1 u2    udn  and W2 Rbb w1 w2

   wb  are the orthogonal matrices. V2 Rdnb is a diagonal matrix with non-negative single values of A in decreasing order as

r1 P r2 P    P rb P 0:

37

As indicated in Eq. (22b), the iterative form-nding algorithm is

successful in case of nA 1. Accordingly, there are also two cases
for s during the iterative form-nding procedure:
Case 1. s 0, there exists no null space of A. In other words,
the right single value rb of A in V is not equal to zero. It
denotes that Eq. (14) has no nonzero force density vector q
as a solution. In this case, if the right single vector basis wb
in W corresponding to smallest singular value rb in V is used
as the approximated q, the sign of q may not match with that
of q0 . Thus, all columns of W employed to compute a vector q
that best matches q0 are scanned by form-nding procedure.
The procedure stops sign-nding until the sign of all components of wj j b; b  1; . . . ; 1 is identical to that of q0 , i.e.
signwj
signq0 . That vector wj is directly taken as the
approximated q. In doing so, the form-nding procedure denes
the approximated q that matches in signs with q0 , such that

Aq 0:

38

32b

in which KE is the linear stiffness matrix, KG is the geometrical

stiffness matrix induced by pre-stressed or self-stressed state;
e; a and l 2 Rb denote the vectors of Youngs moduli, cross-sectional areas and pre-stressed lengths of b members of the
tensegrity structure, respectively; I3 2 R33 and  are the unit
matrix and tensor product, respectively. If the tangent stiffness
matrix is positive denite, then the structure is stable when its
rigid-body motions are constrained; i.e., the quadratic form of
KT is positive with respect to any non-trivial motion d as

d KT d > 0

positive stiffness

4.2. Single value decomposition of the equilibrium matrix

Eq. (28) shows /i is linearly dependent with the vector I1 while
Eq. (29) proves at least one of b members in d-dimensional
structure has a zero length. If there is no /i which satises
 are chosen
Eqs. (28) and (29), the rst three eigenvectors of U
as nodal coordinates x y z for 3-dimensional tensegrity
structure.
Accordingly, D will nally have the required rank deciency nD
without any negative eigenvalue. It implies D is positive semidenite, and any tensegrity structure falling into this case is
super stable regardless of material properties and level of
self-stress coefcients [1719,25].

KE A diag

eigKT k1 P k2 P    P kdnr > 0

|{z}

29

where Ld 2 Rb which indicates the vector of lengths of b members from any combination of d eigenvectors among nD above
eigenvector bases in d-dimensional space (assuming d 3) is
given by

Ld

or

33

Case 2. s 1, it is known [27] that the bases of vector spaces of

force densities and mechanisms of any tensegrity structure are
calculated from the null spaces of the equilibrium matrix. In
this case, the matrices U and W from Eq. (36) can be expressed,
respectively, as

U u1
W w1

u2
w2




urA jm1



wb1 jq1 

mdnrA 

39a
39b

where the vectors m2 Rdn denote the m dn  rA innitesimal mechanisms; and the vector q1 2 Rb matching in signs

241

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

with q0 is indeed the single state of self-stress which satises

the homogeneous Eq. (14).

(5)

(6)

In summary, the eigenvalue decomposition of force density matrix D and the single value decomposition of the equilibrium matrix A are performed iteratively to nd the feasible sets of nodal
coordinates [x y z] and force density vector q which satisfy
the minimum required rank deciencies of the force density and
equilibrium matrices as presented in Eq. (22), respectively.
Let the coordinates of node i be denoted as pi xi yi zi  2 R3 .
It should be noted that since the tensegrity structure without xed
nodes is a free body in the space, the force density vector q does
not change under afne transformation [28,29,8] which transforms
 i as follows:
pi to p

 i pi T t;
p

40

where T2 R33 and t 2 R3 represent afne transformation which

is a composition of rotations, translations, dilations, and shears. It
preserves collinearity and ratios of distances; i.e., all points lying
on a line are transformed to points on a line, and ratios of the distances between any pairs of the points on the line are preserved
[30]. Hence, numerous geometries in self-equilibrium are derived
by Eq. (40) with the same force density vector q. For simplicity, in
this paper, it is assumed that t = 0 and T is considered as

1 0 0

(7)

(8)

(9)

(3)

(2)

Fig. 2. A two-dimensional hexagonal tensegrity structure.

Table 2
The force density coefcients of the 2-D hexagonal tensegrity structure.
Force density
coefcients

Tibert and Pellegrino

[14]

Estrada et al.
[10]

Present

q1 q6
q7 q9

1.0
0.5

1.0
0.5

1.0
-0.5

0.4

B
C
T @ 0 1 0 A:
0 0 1

41

0.2
0.0

Since the tensegrity structure should satisfy the self-equilibrium

conditions, the vector of unbalanced forces ef 2 Rdn dened as follows can be used for evaluating the accuracy of the results:

-0.2

ef Aq:

-0.4

q
 ef ef T :

(1)

The Euclidean norm of

(4)

42

-0.6

ef is used to dene the design error  as

43

Two set of parameters which are nodal coordinates and force density vector can be simultaneously dened by proposed form-nding
procedure through following algorithm.

-0.4 -0.2 0.0

0.2

0.4

Fig. 3. The obtained geometry of the two-dimensional hexagonal tensegrity

structure.

Algorithm
 Step 1: Dene C by Eq. (1) for the given topology of tensegrity
structure.
 Step 2: Specify the type of each member to generate initial force
density vector q0 by Eq. (21). Set i 0.
 Step 3: Calculate Di using Eq. (8).
 Step 4: Carry out Eq. (23) to dene x y zi through Eq. (35).
 Step 5: Determine Ai by Eq. (15).
 Step 6: Perform Eq. (36) to dene qi1 through Eq. (38).
 Step 7: Dene Di1 with qi1 by Eq. (8). If Eq. (22) is satised, the
solutions exist. Otherwise, set i i 1 and return to Step 4.
 Step 8: If D is positive semi-denite, go to Step 10. Otherwise,
specify material property, force density coefcient for each
member based on force density vector found.
 Step 9: If Eq. (34) is fullled, go to Step 10. Otherwise, set i i 1
and return to Step 4.
 Step 10: The process is terminated until Eq. (43) has been
checked. The nal coordinates and force density vector are the
solutions. Otherwise, set i i 1 and return to Step 4.

(4)
(5)

(8)

(3)

(7)
(1)

(6)

(2)

2
Fig. 4. A two-dimensional four-strut tensegrity structure.

0.4
0.2
0.0
-0.2
-0.4

5. Numerical examples
Numerical examples are presented for several tensegrity structures using Matlab Version 7.4 (R2007a) [31]. Based on algorithm

-0.4 -0.2 0.0

0.2

0.4

Fig. 5. The obtained geometry of the two-dimensional four-strut tensegrity

structure.

242

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

Table 3
Eigenvalues of the tangent stiffness matrix of the 2-D four-strut tensegrity structure.
Axial stiffness

k1

k2

k3

k4

k5

k6

k7

ei ai 10

11.2930

11.2930

14.1421

20.0000

34.1421

43.1334

43.1334

developed, both nodal coordinates and force density v ector are

simultaneously dened with limited information of nodal connectivity and the type of the each member. A hexagon, expandable
octahedron and truncated icosahedron belong to Class 1 tensegrity
structure, while the remainings are of Class 2.
5.1. Two-dimensional tensegrity structures
5.1.1. Hexagon
The initial topology of the hexagonal tensegrity structure comprising three struts and six cables (Fig. 2) which was studied by Tibert and Pellegrino [14] and Estrada et al. [10] is herein used for
verication purpose. Initial nodal coordinates, member lengths
and force density coefcients are unknown in advance. The only

(21)
(13)
(14)
9

11

(15)

(9)
(1)

4
(22) (24)
12
(30)

q0 fq1 q6 1; q7 q9 1gT :

44

The obtained force density vector normalized with respect to the

force density coefcient of the cable 1, as presented in Table 2,
agrees well with those of Tibert and Pellegrino [14] and Estrada
et al. [10]. The associated stable conguration of the structure is
plotted in Fig. 3. The form-nding procedure converges in only
one iteration with the design error  8:028  1016 . The structure
obtained has only one self-stress state s 1 and one innitesimal
mechanism m 1 when its three rigid-body motions are constrained indicating it is statically and kinematically indeterminate
[22]. The force density matrix D is positive semi-denite, and the
structure is certainly super stable regardless of materials and

(23)

0.7

(18) (20)

(16)
(26)
(17)
(29)

(28)
(25) (11)
(10)
5
(2)
(27)
(4)
(3)
(5)
1

0.6

(19)
10

8
(12)

(7)

(6)

0.5
Design error ()

information is the incidence matrix C and the type of each member

which is employed to automatically assign the initial force density
vector by proposed form-nding procedure as

(8)

0.4
0.3
0.2

Fig. 6. An expandable octahedron tensegrity structure.

0.1
Table 4
The force density coefcients of the expandable octahedron tensegrity structure.
Force density
coefcients

Tibert and Pellegrino

[14]

Estrada et al.
[10]

Present

q1 q24
q25 q30

1.0
1.5

1.0
1.5

1.0
1.5

b 0.4

0.3

0.4

12

0.2
0.3

0.1
5

0.2

0
0.1
0

9
7

10
8

4
11

0.3

12

0.2

11

0.3

0.5

0.1
0.2

0.1

0.4

12

10
5

0.4
0.5

0.4 0.3 0.2 0.1

0.1

0
0.2

0.3

14

Fig. 8. The convergence of the proposed iterative algorithm for the expandable
octahedron.

a
0.5

10

Number of iterations toward selfequilibrium

0.4

0.5

0.4

0.2

Fig. 7. The obtained geometry of the expandable octahedron tensegrity structure. (a) Top view, (b) perspective view.

0.2

0.4

243

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

prestress levels [1719,25]. Consequently, the proposed form-nding procedure with limited information about the incidence matrix
and element prototype is indeed capable of nding a self-equilibrium stable tensegrity structure by imposing the two necessary
rank deciency conditions.
5.1.2. Four-strut tensegrity
As a next example, consider a two-dimensional four-strut
tensegrity structure comprising four struts and four cables
(Fig. 4). Similarly, the only information is the incidence matrix C
and the type of each member which is employed to automatically
assign the initial force density vector by proposed form-nding
procedure as

q0 fq1 q4 1; q5 q8 1gT :

45

The obtained force density vector normalized with respect to the

force density coefcient of the cable 1 is as follows:

q fq1 q4 1; q5 q8 2gT :

procedure converges in only one iteration with the design error

 2:220  1015 . The structure achieved has only one self-stress
state s 1 and no innitesimal mechanism m 0 after constraining its three rigid-body motions indicating it is statically indeterminate and kinematically determinate [22].
Contrary to all the other approaches, the proposed form-nding
does not require the force density matrix positive semi-denite.
Particularly in this problem, the force density matrix D is negative
semi-denite (possessing one negative eigenvalue) implying that
the structure is not super stable. Accordingly, tangent stiffness of
the structures has been investigated using Eq. (34). For simplicity,
all members are assumed to have the same axial stiffness ei ai 10.
The force density coefcients of the members are particularly set
as Eq. (46). The nonzero eigenvalues of KT after neglecting rst
three zero ones corresponding to three rigid-body motions are
listed in Table 3 which shows the smallest is 11.2930; i.e., the
structure is denitely stable.

46

The associated conguration of the structure whose nodal

coordinates are obtained and presented in Fig. 5. The form-nding
Table 5
Main specication in form-nding procedure of the expandable octahedron tensegrity structure at every iteration.
Iteration

q1 q24

q25 q30

ls =lc

1
2
3
4
5
6
7
8
9
10
11
12
13

0.9073
0.8856
0.8811
0.8802
0.8800
0.8800
0.8800
0.8800
0.8800
0.8800
0.8800
0.8800
0.8800

1.2658
1.3095
1.3179
1.3196
1.3199
1.3200
1.3200
1.3200
1.3200
1.3200
1.3200
1.3200
1.3200

1.6180
1.6325
1.6330
1.6330
1.6330
1.6330
1.6330
1.6330
1.6330
1.6330
1.6330
1.6330
1.6330

6.321
1.181
2.330
4.700
9.000
2.000
3.722
7.443
1.489
2.977
5.955
1.191
2.382

5.2. Three-dimensional tensegrity structures

5.2.1. Expandable octahedron
The tensegrity structure consisting of 6 struts and 24 cables
(Fig. 6) was investigated rst by Tibert and Pellegrino [14] and then

0.7
 101
 101
 102
 103
 104
 104
 105
 106
 106
 107
 108
 108
 109

0.6

Design error ()

0.5
0.4
0.3
0.2
0.1

Table 6
The force density coefcients of the truncated icosahedron tensegrity structure.
Force density coefcients

Estrada et al. [10]

q1 q60
q61 q90
q91 q120

1.0000
0.6775
0.3285

a
0.15
56

0.1
0.05
0

47
9

42

26
27

39
40

0.05
25

60
34

1
21 5

38
39
37

13
41

42

22
2
12

40
36

15
19
18

0.2
0.2 0.15 0.1 0.05

0.1

35

50

21
28
56

53
5
54

3
33
1

14

5227 29

17

2630
51
55

322

3511

78

13 1231

16

69
10

0.2
0.15

22
57

16
60

0.2

17

0.05

25
49

15

0.15

11
20

30

46

58

34

3
14

0.15

47
24
48

23

20
59

0.1

23

4341

19

36
37

0.05

45

35
43

25

18
38

24
6

10

31
44

20

45
44

0.1

33
58

0.05
0.1

846

48

59

15

0.2

40

Fig. 10. The convergence of the proposed iterative algorithm for the truncated
icosahedron.

0.15

30
28

32

10

Number of iterations toward selfequilibrium

29

50

49
55

57

0.2

51

53
54

1.0000
0.6775
0.3285

52

0.2

Present

0.2

0.2

0.1

Fig. 9. The obtained geometry of the truncated icosahedron tensegrity structure. (a) Top view, (b) perspective view.

0.1

0.2

244

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

tions (a little bit more effective in convergence in comparison

with 45 iterations presented by [10]) is 7:572  109 . The obtained
structure possesses one state of self-stress s 1 and 55 innitesimal mechanisms m 55 excluding its six rigid-body motions.
The force density matrix D is positive semi-denite which leads
structure to super stable regardless of materials and prestress levels [1719,25]. It is clear that the introduction of single prestress
stiffens all innitesimal mechanisms to make the structure stable.

by Estrada et al. [10] using symbolic analysis and numerical method, respectively. The calculated force density vector after normalizing with respect to the force density coefcient of the cable 1, as
presented in Table 4, agrees well with those of Tibert and Pellegrino [14] and Estrada et al. [10]. The associated stable conguration
of the structure is presented in Fig. 7. All of struts and cables have
the same length, respectively. The length ratio between struts and
cables is ls =lc 1:633 coinciding with the one ls =lc 1:63299 stated by Estrada et al. [10]. The form-nding procedure converges in
13 iterations. The design error convergence is described in Fig. 8
with  2:382  109 .
In order to illustrate the proposed form-nding procedure, main
specication at every step of iteration is presented in Table 5. As
can be seen from this table, for the practical purpose it seems
the proposed algorithm sufciently converges at iteration 6, but
for the demonstration of the efciency all iterations are displayed.
The structure obtained has only one self-stress state s 1 and
one innitesimal mechanism m 1 except for its six rigid-body
motions. Accordingly, it belongs to statically and kinematically
indeterminate structure [22]. The force density matrices D is positive semi-denite which leads structure to super stable regardless
of materials and prestress levels [1719,25].

5.2.3. Octahedral cell

The system shown in Fig. 11 has ve struts and eight cables.
Four of ve struts form a quadrangular polygon. The calculated
force density vector after normalizing with respect to the force
density coefcient of the cable 1 is:

q fq1 q8 1; q9 q12 1; q13 2gT :

The associated conguration of the structure whose nodal coordinates are obtained and displayed in Fig. 12. The form-nding procedure converges in only one iteration with the design error
 3:279  1015 . The structure obtained has only one state of
self-stress s 1 and no innitesimal mechanism m 0 after
constraining its six rigid-body motions indicating it is statically
indeterminate and kinematically determinate [22].
Similar to the force density matrix D of the four-strut tensegrity,
that of the octahedral cell is also negative semi-denite (possessing one negative eigenvalue) implying that the structure is not
super stable. Accordingly, tangent stiffness of the structures has
been investigated using Eq. (34). For simplicity, all members are
assumed to have the same axial stiffness ei ai 2. The force density
coefcient of each member is directly set as Eq. (47). The nonzero
eigenvalues of KT after neglecting rst six zero ones corresponding
to six rigid-body motions are listed in Table 7 which shows the
smallest is 0.1002; i.e., the structure is mechanically stable.

5.2.2. Truncated icosahedron

More complicated example is the truncated icosahedron
tensegrity structure with 30 struts and 90 cables which was analyzed by Estrada et al. [10]. The calculated force density vector
after normalizing with respect to the force density coefcient of
the cable 1, as presented in Table 6, agrees well with that of Estrada
et al. [10]. The associated stable conguration of the structure is
plotted in Fig. 9. The design error  shown in Fig. 10 which exposes the convergence of the proposed form-nding in 39 itera-

6
(5)
(6)

(12)

(9)

(11)
(10)

(13)
2

(1)

5.2.4. Tensegrity dome

A three-dimensional tensegrity dome displayed in Fig. 13 has
six struts and 12 cables. The calculated force density vector after
normalizing with respect to the force density coefcient of the
cable 1 is:

(7)

(8)

(4)

(2)

q fq1 q6 1:0000; q7 q12 1:4574;

(3)

q13 q15 0:5000; q16 q18 0:5426gT :

48

Fig. 14 shows the obtained self-equilibrium conguration. The

form-nding procedure converges in only one iteration with the design error  1:935  1015 . The structure obtained has only one

5
Fig. 11. An octahedral cell tensegrity structure.

47

2
1

0.5

b
5

0.6

0.4

0.4

0.3
0.2
0.2

0.1

0.2

0.4

0.1

0.6

0.2

0.3
0.4

6
5

0.5
3

0.5
0.6

0.4

0.2

0.2

0.4

0.5

0.6

0.4

0.2

Fig. 12. The obtained geometry of the octahedral cell tensegrity structure. (a) Top view, (b) perspective view.

0.2

0.4

245

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

Table 7
Eigenvalues of the tangent stiffness matrix of the octahedral cell tensegrity structure.
Axial stiffness

k1

k2

k3

k4

k5

k6

k7

k8

k9

k10

k11

k12

ei ai 2

0.1002

1.9605

2.4870

2.8353

3.6434

3.9516

4.1584

4.1802

9.2446

9.4750

9.5693

12.3043

3
(3)

(17)
8

(6)

(8)
(9)

(15)

(7)

(18)
9

(5)

(14)

(2)

(11)
(12)

(16)

(13) 7
(10)

(1)
(4) 5
1
Fig. 13. A three-dimensional six-strut tensegrity dome structure.

0.6

0.4

5
7

0.4

0.2

0.2

9
8

0.2

5
4
7
6
1

0.2

82

0.4

4
6

0.6
0.4
0.2

0.4

0
0.2
0.4

0.4

0.2

0.2

0.4

0.4

0.2

0.2

0.4

0.6

0.6

Fig. 14. The obtained geometry of the three-dimensional six-strut tensegrity dome structure (a) Top view, (b) Perspective view.

state of self-stress s 1 and four innitesimal mechanisms

m 4 after constraining its six rigid-body motions indicating it
is statically and kinematically indeterminate [22]. The force density
matrix D is positive semi-denite which leads structure to super
stable regardless of materials and prestress levels [1719,25].
6. Concluding remarks
The advanced form-nding procedure for both Classes 1 and 2
tensegrity structures has been proposed. The force density matrix
is derived from the incidence matrix and initial set of force densities formed by the vector of type of member forces. The elements
of this vector consist of unitary entries +1 and 1 for members
in tension and compression, respectively. The equilibrium matrix
is dened by the incidence matrix and nodal coordinates. The
eigenvalue decomposition of the force density matrix and the single value decomposition of the equilibrium matrix are performed
iteratively to nd the range of feasible sets of nodal coordinates
and force densities which satisfy the required rank deciencies of
the force density and equilibrium matrices, respectively.

Any assumption about initial nodal coordinates, symmetric

properties, member lengths and the positive semi-denite condition of the force density matrix is not necessary in the proposed
form-nding procedure. A rigorous denition is given for the required rank deciencies of the force density and equilibrium matrices that lead to a stable non-degenerate d-dimensional tensegrity
structure. The eigenvalues of the tangent stiffness are also evaluated for checking the stability of the tensegrity structures. In the
numerical examples, a very good convergence of the proposed
method has been shown for two-dimensional and three-dimensional tensegrity structures. The proposed algorithm is strongly
capable of searching novel congurations with limited information
of topology and the members type. As a natural extension of this
research, form-nding with more complicated constraints awaits
further attention.
Acknowledgments
This research was supported by a grant (code#06-R&D-B03)
from Cutting-edge Urban Development Program funded by the

246

H.C. Tran, J. Lee / Computers and Structures 88 (2010) 237246

Ministry of Land, Transport and Maritime Affairs of Korean government. The authors also would like to thank the anonymous reviewers for their suggestions in improving the standard of the
manuscript.
References
[1] Fuller RB. Synergetics explorations in the geometry of thinking. London,
UK: Macmillan; 1975.
[2] Wang BB. Free-standing tension structures. From tensegrity systems to cable
strut systems. London and New York: Taylor and Francis; 2004.
[3] Pinaud JP, Solari S, Skelton RE. Deployment of a class 2 tensegrity boom. Proc
SPIE Smart Struct Mater 2004;5390:15562.
[4] Schek HJ. The force density method for form nding and computation of
general networks. Comput Methods Appl Mech Eng 1974;3:11534.
[5] Motro R, Najari S, Jouanna P. Static and dynamic analysis of tensegrity systems.
In: Proceedings of the international symposium on shell and spatial structures,
computational aspects. Springer; 1986. p. 2709.
[6] Barnes MR. Form nding and analysis of tension structures by dynamic
relaxation. Int J Space Struct 1999;14(2):89104.
[7] Vassart N, Motro R. Multiparametered form nding method: application to
tensegrity systems. Int J Space Struct 1999;14(2):14754.
[8] Masic M, Skelton R, Gill P. Algebraic tensegrity form-nding. Int J Solids Struct
2005;42(1617):483358.
[9] Zhang JY, Ohsaki M. Adaptive force density method for form-nding problem
of tensegrity structures. Int J Solids Struct 2006;43(18-19):565873.
[10] Estrada G, Bungartz H, Mohrdieck C. Numerical form-nding of tensegrity
structures. Int J Solids Struct 2006;43(22-23):685568.
[11] Micheletti A, Williams WO. A marching procedure for form-nding for
tensegrity structures. J Mech Mater Struct 2007;2(5):10126.
[12] Zhang L, Maurin B, Motro R. Form-nding of nonregular tensegrity systems. J
Struct Eng ASCE 2006;132(9):143540.
[13] Rieffel J, Valero-Cuevas F, Lipson H. Automated discovery and optimization of
large irregular tensegrity structures. Comput Struct 2009;87(56):36879.

[14] Tibert AG, Pellegrino S. Review of form-nding methods for tensegrity

structures. Int J Space Struct 2003;18(4):20923.
[15] Juan SH, Tur JMM. Tensegrity frameworks: static analysis review. Mech Mach
Theory 2008;43(7):85981.
[16] Motro R. Tensegrity: Structural Systems for the Future. London: Kogan Page;
2003.
[17] Connelly R. Rigidity and energy. Invent Math 1982;66(1):1133.
[18] Connelly R, Terrell M. Globally rigid symmetric tensegrities. Topol Struct
1995;21:5978.
[19] Connelly R. Tensegrity structures why are they stable. In: Thorpe MF, Duxbury
PM, editors. Rigidity Theory and Applications. Kluwer Academic Publishers;
1999. p. 4754.
[20] Meyer CD. Matrix analysis and applied linear algebra. SIAM; 2000.
[21] Calladine CR. Buckminster Fullers tensegrity structures and Clerk Maxwells
rules for the construction of stiff frames. Int J Solids Struct 1978;14(2):16172.
[22] Pellegrino S, Calladine CR. Matrix analysis of statically and kinematically
indeterminate frameworks. Int J Solids Struct 1986;22(4):40928.
[23] Ohsaki M, Zhang JY. Stability conditions of prestressed pin-jointed structures.
Int J Nonlinear Mech 2006;41(10):110917.
[24] Murakami H. Static and dynamic analyses of tensegrity structures. Int J Solids
Struct 2001;38(20):361529.
[25] Connelly R. Generic global rigidity. Disc Comput Geom 2005;33(4):54963.
[26] Guest S. The stiffness of prestressed frameworks: a unifying approach. Int J
Solids Struct 2006;43(34):84254.
[27] Pellegrino S. Structural computations with the singular value decomposition of
the equilibrium matrix. Int J Solids Struct 1993;30(21):302535.
[28] Tarnai T. Duality between plane trusses and grillages. Int J Solids Struct
1989;25(12):1395409.
[29] Connelly R, Whiteley W. Second-order rigidity and prestress stability for
tensegrity frameworks. SIAM J Disc Math 1996;9(3):45391.
[30] Weisstein EW. Afne transformation; MathWorld, <http://www.mathworld.
wolfram.com/AfneTransformation.html>.
[31] Yang WY, Cao W, Chung TS. Applied numerical methods using Matlabs. United
States of America: Wiley; 2005.