International Journal of Solids and Structures 206 (2020) 9–22

Contents lists available at ScienceDirect

International Journal of Solids and Structures

journal homepage: [Link]/locate/ijsolstr

Topology optimization and shape design method for large-span

tensegrity structures with reciprocal struts
Yan Su a,b,c, Jingyao Zhang c, Makoto Ohsaki c, Yue Wu a,b,⇑
a
Key Lab of Structures Dynamic Behavior and Control of the Ministry of Education, Harbin Institute of Technology, Harbin 150090, China
b
Key Lab of Smart Prevention and Mitigation of Civil Engineering Disasters of the Ministry of Industry and Information Technology, Harbin Institute of Technology, Harbin
150090, China
c
Department of Architecture and Architectural Engineering, Kyoto University, Kyoto-Daigaku Katsura, Nishikyo, Kyoto 615-8540, Japan

a r t i c l e i n f o a b s t r a c t

Article history: Tensegrity is a kind of efficient structural system that is composed of continuous cables in tension and
Received 15 December 2019 discontinuous struts in compression. Although the studies on tensegrity have been carried out by many
Received in revised form 20 May 2020 researchers, there are few methods for assembling large-span tensegrity by multiple units. Inspired by
Accepted 3 September 2020
the similarities between reciprocal units and tensegrity units, we present a novel method for designing
Available online 11 September 2020
large-span tensegrity structures with reciprocal struts, which is a kind of new tensegrity structural sys-
tem that can cover a large area, while the rule of discontinuous struts is followed strictly. The proposed
Keywords:
design method includes assembly, topology optimization, and shape design. The prototype; i.e., initial
Tensegrity
Reciprocal structure
topology, of the tensegrity is generated from reciprocal struts. A topology optimization method using
Assembly of tensegrity units ground structure approach is proposed to remove the redundant cables from the initial topology. A shape
Topology optimization design method using fictitious stiffness parameters is proposed to satisfy the architectural shape demand.
Shape design The form-finding problem, which is to find the self-equilibrium shape in the topology optimization as
well as shape design processes, is solved by using the non-rigid-body motion analysis method.
Numerical examples demonstrate the effectiveness and applicability of the proposed method.
Ó 2020 Elsevier Ltd. All rights reserved.

1. Introduction Tensegrity structure transfers loads only by axial force and

maintains its stability with the minimum possible number of
Tensegrity is a kind of efficient structural system that is com- structural members. Therefore, it is considered as one of the ideal
posed of continuous cables in tension and discontinuous struts in structural systems (Sultan, 1999). The cable dome, which is
compression (Motro, 2003). It has applications in many fields, such designed according to the concept of tensegrity but with supports,
as architecture (Kawaguchi and Ohya, 2004), mechanical engineer- has been successfully used in many large-span structures (Yuan
ing (Kan et al., 2019), biomedical engineering (Denton et al., 2003), et al., 2007; Ma et al., 2019). The most famous cable domes are
and mathematics (Donev et al., 2004). As most researchers accept, a the Geiger cable dome and the Levy cable dome. However, these
tensegrity should have the following characteristics: 1) The struc- cable domes are not ‘real’ tensegrity structures in view of the strict
ture is free-standing without any support; 2) There are only two dif- definition, because they are not free-standing. Once the supports
ferent types of members, struts and cables; 3) The struts do not are removed, the cable domes are no longer stable. In fact, few
contact with each other at their ends; 4) Structural self-weight is ‘real’ tensegrity structures have been used to cover a large space.
negligibly small compared with the member forces (Zhang and This is because the design of large-span tensegrity structure is
Ohsaki, 2015). The tensegrity in this paper refers to a cable-strut complex and difficult. Some studies have been conducted to
structure that follows the characteristics above strictly. Skelton assemble large-span tensegrities by multiple units (Gómez-
and de Oliveira (2009) classified the tensegrity structures based Jáuregui et al., 2012; Smaili and Motro, 2005). Most of these assem-
on the number of struts connected to a node; hence, the structure bly methods combine units by connecting their nodes (Zhang et al.,
considered in this paper is classified into Class 1 tensegrity. 2015; Rimoli and Pal, 2017; Feng, 2017). These structures are not
‘real’ tensegrity structures either, because the struts contact with
each other at their ends. Only the tensegrity structures with speci-
⇑ Corresponding author at: Key Lab of Structures Dynamic Behavior and Control
of the Ministry of Education, Harbin Institute of Technology, Harbin 150090, China.
fic rules can remain stable after assembling from multiple units,
E-mail address: wuyue_2000@[Link] (Y. Wu). such as tensegrity tower (Zhang and Ohsaki, 2006a) and tensegrity

[Link]
0020-7683/Ó 2020 Elsevier Ltd. All rights reserved.
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

Fig. 1. A typical reciprocal structure composed of multiple units; (a) reciprocal unit, (b) reciprocal structure.

beam (Ali et al., 2011). However, these structures cannot cover a 3- 2. Configuration of the tensegrity structure with reciprocal
dimensional space. struts
This paper aims to develop a design method of ‘real’ tensegrity
structures covering a large space, which can keep stability without 2.1. Relationship between tensegrity and reciprocal units
supports avoiding the struts from contacting each other at a node.
There are three issues to be resolved for this purpose. First, how to Figs. 2 and 3 show the top views of typical reciprocal and
assemble a ‘real’ large-span tensegrity by multiple units. Second, tensegrity units with four bars (struts), respectively. It can be seen
how to reduce the number of members (cables) composing the that there are some similarities between the two kinds of units.
structure. Third, how to adjust the self-equilibrium shape to satisfy
the geometrical requirements by the architects.
For the first problem, we make use of reciprocal structures, as
shown in Fig. 1, which is a kind of grid structure that consists of
a series of mutually supporting short bars in a regular pattern
(Larsen, 2008). The bars are assembled to a unit first, and then
the units are connected to generate the entire structure as shown
in Fig. 1(b). We utilize the fact that the reciprocal structure can
be assembled by multiple units easily (Su et al., 2019a). There
are some similarities between the reciprocal unit and tensegrity
unit (Snelson, 2012; Widyowijatnoko et al., 2019), which will be
discussed in detail in Section 2.1. Inspired by these similarities,
we propose a large-span tensegrity structure with reciprocal struts,
which is a ‘real’ tensegrity that satisfies the basic requirements
strictly.
For the second and third problems, we define the design process
as topology optimization and shape design problems. To reduce
the number of cables and make the assembled tensegrity simpler
for an architectural engineering purpose, a topology optimization Fig. 2. A reciprocal unit.
method using ground structure approach is proposed. To control
the structural shape close to the desired shape, a shape design
method is proposed. In the topology optimization and shape design
methods, the form finding is a basic process at each step. Non-
rigid-body motion analysis is an effective nonlinear structural
analysis method (Zhang and Ohsaki, 2013) for a free-standing
structure. It will be used as a form finding method in this paper
because the topology optimization and shape design problems
can be solved successfully by adjusting the parameters of the form
finding process.
The rest of this paper is organized as follows. Section 2 intro-
duces the relationship between tensegrity and reciprocal units
and proposes a method for assembling a large-span tensegrity. Sec-
tion 3 introduces the form finding method used in the topology
optimization and shape design process. Sections 4 and 5 formulate
the topology optimization and shape design problems, respec-
tively. Section 6 outlines the flow of designing a large-span tenseg-
rity with reciprocal struts. Section 7 presents four numerical
examples to illustrate the effectiveness and applicability of the
proposed method, and Section 8 concludes the study. Fig. 3. A tensegrity unit.

10
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

Fig. 4. Generation process of a tensegrity unit.

Both units are derived from weaving the struts, while the struts The steps in Fig. 4 can be summarized as follows:
overlap with each other (from the top view). This configuration
brings the concepts of weaving into three dimensions. The struts (a) Place four bars in a cross-shape on a plane.
of the tensegrity unit are separate, while those of the reciprocal (b) Translate each bar to make a square.
unit contact each other. Furthermore, the horizontal positions of (c) Adjust the locations of end nodes so that the bars connect
struts in the tensegrity unit correspond to the bars in the reciprocal with each other reciprocally at upper and lower surfaces to
unit, showing that a tensegrity unit can be generated from the form a reciprocal unit.
reciprocal unit; i.e., by making the bars of reciprocal unit separate, (d) Conduct a form generation process for reciprocal structure
the struts of the reciprocal unit can reach the same position as by self-adaptive form generation method (Su et al., 2019b).
those of a tensegrity unit. Fig. 4 presents the generation process Here, the distance between the bar axes should be set much
of a typical tensegrity unit. larger than the true element radius to ensure that the struts
are separate.
(e) Extend each bar in its axial direction so that there is an
extension of the element from the connection centroid.
(f) Assign enough number of cables to the unit so that it is
stable and statically indeterminate. Three kinds of cables
as shown in Fig. 5 should be assigned in this step; i.e., top
ring, bottom ring, and vertical cables. Each node in the top
ring is connected to the two nearest nodes in the bottom
ring by vertical cables.

2.2. Assembly of tensegrity with reciprocal struts

Besides the similarities mentioned above, both the units can

form complex structures by assembling multiple units. Assembly
Fig. 5. Arrangement of cables in a tensegrity unit with reciprocal struts. of the tensegrity structure is difficult because the topology is

Fig. 6. Assembly of the reciprocal units; (a) Top view, (b) Diagonal view.

11
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

constrained by both geometry and force balance, while assembly of not a minimal tensegrity. Second, the shape of the proposed
the reciprocal structure is simply a geometry problem. Therefore, it tensegrity depends on the configuration of the reciprocal struts,
is practically useful if the assembly of tensegrity with reciprocal which is regular in most cases. A shape design process is needed
struts can be realized in a similar way as the reciprocal structure. to generate a large-span tensegrity close to the target shape satis-
The reciprocal structure realizes the assembly by sharing a strut fying an architectural requirement.
between different units. Fig. 6 shows the assembly process of four
4-bar reciprocal units. The bar marked as ‘shared bar’ belongs to
different reciprocal units. Therefore, there are 12 bars in total com- 3. Form finding method
posing the four units. When the structure is subjected to external
loads, the shared bars can transfer the force between the units Using the proposed method, we design a tensegrity structure
and let the four units work together. with n nodes, mS struts, and mC cables. The model obtained from
Since a reciprocal unit can be transferred into a tensegrity unit the assembly process in Section 2.2 is in the unstressed state;
as described in Section 2.1, the assembled reciprocal structure can therefore, a form finding process is needed for finding the self-
also be used for generating a tensegrity structure with multiple equilibrium shape. Many effective form finding methods have been
units. After assembling the reciprocal units, the cables should be proposed in the existing studies (Zhang and Ohsaki, 2006b; Zhang
attached to the structure. Fig. 7 shows the assembly of four 4-bar et al., 2006; Koohestani and Guest, 2013). Among them, the non-
tensegrity units. Each unit is assigned with cables according to rigid-body motion analysis (Zhang and Ohsaki, 2015) is used here
the arrangement as shown in Fig. 5. The cables should be added because it can generate a self-equilibrium shape close to the initial
to connect an end node of the shared bar with the nodes in differ- shape, even for a self-standing structure without supports, by
ent units so that the units can maintain stability after being assem- effectively eliminating the components of rigid-body motions.
bled. When the structure is subjected to the external loads, the Fig. 8 depicts the input and output parameters in the non-rigid-
shared bar and the cables connected to it can transfer the force body motion analysis. The input parameters include the initial con-
between the four units and let them work together. figuration (the coordinate vectors x, y, z, and the connectivity
The proposed process can assemble a stable large-span tenseg- matrix C), the initial prestress vector s, and the stiffness vector k.
rity structure with reciprocal struts, if the shared bars are closed as The output parameters include the final configuration (xF, yF, zF)
shown in Fig. 7. However, there are still two problems to be and the corresponding prestress sF in the balanced shape.
resolved for using the assembled structural system in real engi- For form finding of the proposed large-span tensegrity with
neering applications. First, some redundant cables should be reciprocal struts, the initial configuration is obtained using the pro-
removed from the assembled tensegrity, because it is generally cess described in Section 2, which is the key step for assembling a

Fig. 7. Assembly of the tensegrity units with reciprocal struts; (a) Top view, (b) Diagonal view.

Fig. 8. Input and output parameters of the non-rigid-body motion analysis.

12
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

large-span tensegrity structure. After assembling the initial config- than a negative bound DS with a moderately large absolute value.
uration in zero-prestress state, the output parameters in Fig. 8 can At the same time, the forces of cables should be larger than a small
be regarded as functions of the initial prestress and stiffness vec- positive bound DC in order to confirm that the cables are in tension.
tors. Therefore, we can modify them to adjust the structural topol- This way, the force constraints can be written as
ogy and shape.
sF;C
k P DC ðk ¼ 1; . . . ; mC Þ ð5Þ
The initial prestress vector s is assigned by the designers. Since
the number of cables in the initial tensegrity is relatively large, we
can adjust the initial prestress vector s to generate the cables with k 6 DS ðk ¼ 1; . . . ; mS Þ
sF;S ð6Þ
small forces in the self-equilibrium shape, and remove them to
where sF;S
is the force in the kth strut at the self-equilibrium
k
obtain the optimal topology in Section 4.
state.
It has been shown by Ohsaki and Zhang (2015) that a fictitious
Finally, the topology optimization problem is defined as
stiffness parameter k can be used to obtain various shapes at self-
follows:
equilibrium state. Therefore, we modify the value of the stiffness
PmC F;C
vector to adjust the structural shape in the shape design process minimize f ðsÞ ¼ k¼1 sk ðsÞ
in Section 5. subject to smin 6 sk 6 smax ðk ¼ 1; . . . ; mC þ mS Þ
sF;C
k ðsÞ P DC ðk ¼ 1; . . . ; mC Þ
4. Topology optimization problem ð7Þ
k ðsÞ 6 DS ðk ¼ 1; . . . ; mS Þ
sF;S
For the topology optimization problem, the ground structure jk6 ðsÞj 6 DE
method (Ehara and Kanno, 2010; Kanno, 2013) is a common and k7 ðsÞ P DE
effective approach. The initial tensegrity structure has enough
The cables with sufficiently interior force are removed after
number of cables; therefore, it is regarded as a ground structure,
optimization to find an optimal topology.
from which the redundant cables are removed to obtain an optimal
topology.
5. Shape design method
The force in a cable is seen as a measure for the contribution of
the cable in the self-equilibrium state. As explained in Section 2, a
The shape of tensegrity structure with optimal topology is fur-
different value of the vector s will lead to a different force distribu-
ther optimized in this section by adjusting the structural shape to a
tion in the self-equilibrium state. Therefore, The topology opti-
target shape. Compared with truss structures (San et al., 2019),
mization process takes the vector s as the variables with the
shape design of tensegrity structures is more complicated because
lower bound s min and upper bound s max as
we have to optimize the self-equilibrium shape under prestresses.
smin 6 sk 6 smax ðk ¼ 1; . . . ; mC þ mS Þ ð1Þ It has been demonstrated by Zhang and Ohsaki (2007) that the
non-rigid-body motion analysis can generate different forms using
Let sF;C
k ðsÞ
denote the force in the kth cable of the self- different fictitious stiffnesses. A shape design method, which com-
equilibrium shape, which is a function of the initial prestress s. bines non-rigid-body motion analysis and optimization algorithm,
The cable can be removed when its force is sufficiently small. is proposed in this section for design of large-span tensegrity. The
Therefore, the sum of forces of all cables, defined as Eq. (2), is cho- optimization algorithm is incorporated to find optimal stiffnesses
sen as the objective function f ðsÞ of the optimization problem. We of members so as to generate the self-equilibrium shape as close
can find the cables with small forces by minimizing this objective as possible to the target shape.
function. The objective function is an error function defined by the differ-
X
mC ence between the current shape and the target shape. When the
f ðsÞ ¼ sF;C
k ð2Þ target shape is described by the heights of some target points,
k¼1 the error function can be formulated as the sum of the height dis-
To avoid the 3-dimensional tensegrity structure degenerating tances between the current and target heights of all target points.
into the 2-dimensional or even 1-dimensional space, constraints Note that a reference point should be chosen to exclude the effect
should be given for the eigenvalues of the tangent stiffness matrix; of the rigid-body displacement. Hence, the objective function is
i.e., the six lowest eigenvalues corresponding to the rigid-body dis- formulated, as follows, as a function of the member stiffness vector
placements of a 3-dimensional structure should be equal to zero. k:
Eigenvalue analysis is conducted for the tangent stiffness matrix X
l  
to obtain the 6th and 7th lowest eigenvalues. The 7th eigenvalue Q ðk Þ ¼ zi ðkÞ  z1 ðkÞ  Dz  ð8Þ
i
should be positive so that the structure is stable. Incorporating i¼2

small numerical errors, the following constraints are given for where Zi is the z-coordinate of the ith target point, Zi is the z-
the 6th and 7th eigenvalues, denoted by k6 and k7 , respectively, coordinate of the reference point, Dzi is the target height of the
using a small positive bound DE : ith target point, and l is the number of target points including
jk6 j 6 DE ð3Þ the reference point.
The shape design problem is formulated as follows:
k7 P DE ð4Þ Pl  
minimize Q ðkÞ ¼ i¼2 zi ðkÞ  z1 ðkÞ  Dz  i
If the structure is statically determinate, the forces of all mem- subject to kmin 6 ki 6 kmax ði ¼ 1; . . . ; mC þ mS Þ
i ðkÞ P DC
sF;C ði ¼ 1; . . . ; mC Þ
bers are zero, which turns out to be the best solution to the opti-
mization problem to minimize f ðsÞ in Eq. (2). However, it is not a ð9Þ
sF;S
i ð k Þ 6 DS ði ¼ 1; . . . ; mS Þ
satisfactory solution for our problem, because we aim at finding
a self-equilibrium form rather than a form without prestress. To jk6 ðkÞj 6 DE
avoid this circumstance, the force of all struts should be smaller k7 ðkÞ P DE

13
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

where k max and k min are the upper and lower bounds of ki . 6. Flow of designing a large-span tensegrity with reciprocal
Comparing the shape design problem (9) with the topology opti- struts
mization problem (7), the form finding methods and the con-
straints are the same, while the objective function and the We have introduced the assembly method in Section 2, form
variables are different. finding method in Section 3, topology optimization method in
Section 4, and shape design method in Section 5 for designing a
large-span tensegrity with reciprocal struts. These methods are
integrated and summarized in Fig. 9. First, we use the assembly
method to generate a prototype of a large-span tensegrity struc-
ture consisting of multiple units without prestress. Based on this
prototype, we can obtain the self-equilibrium shape of the assem-
bled structure using the non-rigid-body motion analysis as the
form finding method. Second, the topology optimization is realized
by adjusting the initial force of the form finding process as variable,
while the stiffness is fixed. Third, shape design is carried out by
adjusting the stiffness of members in the form finding process as
variable, while the initial force is fixed.
It should be noted that the form finding process is included as a
step inside the topology optimization and shape design process
rather than an independent step. The stiffness parameters obtained
through shape design are not the true stiffnesses. The unstressed
length of each cable is to be adjusted using the true stiffness to
generate the same tension force as the result of the form finding
process for the final shape.

7. Numerical examples

In this section, four numerical examples are presented in order

to illustrate the accuracy and effectiveness of the proposed
method. In Example 1, we design a simple 4-bar tensegrity struc-
ture, which has been studied in many papers, to show the accuracy
and effectiveness of the proposed topology optimization and shape
design methods. Example 2 applies the proposed method to a more
complex large-span tensegrity structure to show its applicability
and effectiveness for design of a real-scale roof structure. Example
3 investigates a tensegrity composed of 3-bar units to show that
the proposed method can assemble the units successfully. Example
4 applies the proposed method to a complicated tensegrity com-
posed of different kinds of units to show that the proposed method
can deal with the reciprocal units with different configurations.
The prestresses and nodal coordinates can be scaled proportion-
ally without violating the self-equilibrium relationship of the
structures, where the external loads are zero. Therefore, units are
omitted without loss of generality in the numerical examples.
The tolerance of the norm of unbalanced loads in the form finding
Fig. 9. Flowchart of designing a large-span tensegrity with reciprocal struts. process inside the topology optimization and shape design is set as

Fig. 10. Initial model of Example 1; (a) Top view, (b) Diagonal view.

14
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

10-6. The optimization problems are solved by the interior point

method of the function fmincon in the Optimization Toolbox of
MATLAB R2018a.

Table 1
Nodal coordinates of the initial model of Example 1.

Coordinates
x y z
A 0.34 1.09 0.00
B 0.49 0.35 0.93
C 1.09 0.34 0.00
D 0.35 0.49 0.93
E 0.34 1.09 0.00
F 0.49 0.35 0.93
G 1.09 0.34 0.00
H 0.35 0.49 0.93
Fig. 13. Iteration history of the objective function in shape design process of
Example 1.

Table 3
Nodal coordinates of the shape design result of Example 1.
Table 2
Cable forces of the optimal topology. Coordinates

Cable number Interior force Cable number Interior force x y z

1 1.92 9 1.10 A 0.31 1.10 0.06

2 0.01 10 1.10 B 0.52 0.30 0.86
3 1.92 11 1.10 C 1.10 0.31 0.06
4 0.01 12 1.10 D 0.30 0.53 0.87
5 1.92 13 1.55 E 0.31 1.10 0.06
6 0.01 14 1.55 F 0.52 0.29 0.86
7 1.92 15 1.55 G 1.10 0.30 0.06
8 0.01 16 1.55 H 0.30 0.52 0.86

Fig. 11. Topology optimization result of Example 1; (a) Top view, (b) Diagonal view.

Fig. 12. Shape design result of Example 1; (a) Front view, (b) Diagonal view.

15
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

In the topology optimization process, the member stiffnesses of In the shape design process, the initial prestress is fixed at 1
the struts and cables are set as 100.0 and 10.0, respectively. The for the struts and + 1 for the cables. The member stiffness param-
lower and upper bounds of initial prestresses of members, which eters are taken as the variables, and their lower and upper bounds
are the variables, are set as 3 and + 3, respectively. are 1.0 and 1000.0, respectively.

Fig. 14. Assembled initial model of Example 2; (a) Top view and front view, (b) Diagonal view.

Fig. 15. Assembly process of Example 2.

Fig. 16. Topology optimization result of Example 2; (a) Self-equilibrium forces, (b) Cables that can be removed.

16
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

7.1. Example 1 removed. The tensegrity unit after removing redundant cables is
shown in Fig. 11. The number of prestress modes decreases from
The first example is a 4-strut tensegrity unit composed of 4 3 to 1 after topology optimization. The connectivity of cables and
struts and 16 cables as shown in Fig. 10. It is used for confirming the number of prestress modes are the same as the well-known
effectiveness of the topology optimization and shape design meth- prismatic tensegrity. This result verifies the accuracy and effective-
ods. The initial model is obtained from the process shown in Fig. 3. ness of the proposed topology optimization method.
The node and element numbers are marked in Fig. 10(a), and the Shape design is conducted using the topology optimization
nodal coordinates of the initial model are listed in Table 1. The result as the initial model. The target shape is defined by the
number of prestress modes of the initial model is 5. heights of some target points. Point A is chosen as a reference point
Since the assembly of units is not needed for this tensegrity to exclude the effect of the rigid-body displacement. The target
with a single unit, the topology optimization is conducted directly. height distances between Point A and Points C, E, G are zero and
The sum of the interior forces of cables 1–16 is taken as the objec- the target height distances between Point A and Points B, D, F, H
tive function. The cable forces at self-equilibrium state after topol- are 0.8. Hence, the objective function in the shape design process
ogy optimization are listed in Table 2. It can be seen that the forces of Example 1 can be written as
of cables 2, 4, 6, 8 are less than 1% of those of the other cables, X X
showing that these vertical cables are redundant and can be Q¼ jzi  zA j þ jzi  zA  0:8j ð10Þ
i2fC;E;Gg i2fB;D;F;Hg

The final result after topology optimization and shape design is

shown in Fig. 12. The history of objective function value during the
shape design process is shown in Fig. 13. It can be seen from the
figure that the objective function value decreases gradually and
converges at 0.01. The nodal coordinates after the shape design
are shown in Table 3. It can be seen that the proposed method
can exclude the effect of the rigid-body displacements and success-
fully generate the shape close to the target shape.

7.2. Example 2

Example 2 is a square Archimedes tensegrity as shown in

Fig. 14, which is composed of nine 4-bar units and includes 24
struts and 144 cables. Fig. 15 depicts the assembly process of this
model. The form finding process using non-rigid-body motion
analysis method ensures that the assembled tensegrity is stable;
therefore, the tensegrity consisting of multiple units with recipro-
cal struts can cover a large space.
As shown in Fig. 16, the cables of the tensegrity with multiple
units are relatively complex. Topology optimization is carried
Fig. 17. Topology optimization result. out first to remove some redundant cables. The sum of the

Fig. 18. Shape design result of the target shape 1 of Example 2; (a) Top view, (b) Diagonal view, (c) Front view, (d) Left view.

17
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

Fig. 19. Shape design result of the target shape 2 of Example 2; (a) Top view, (b) Diagonal view, (c) Front view, (d) Left view.

self-equilibrium forces of all cables is taken as the objective func-

tion. The distribution of cable forces of the optimal topology is
shown in Fig. 16(a). The cables shown in Fig. 16(b), whose forces
are less than 1% of those of the other cables, are removed to obtain
the topology in Fig. 17. Compared with the initial ground structure,
24 cables have been removed and the structure becomes lighter
and simpler. The number of prestress modes decreases from 30
to 2 after the topology optimization.
In the shape design process, two different target shapes are con-
sidered in this example. Points 1, 4, 7 and 10 are chosen as target
points that control the structural shape. The same height is
assigned in ‘target shape 10 , while different heights are given in
‘target shape 20 . Shape design is carried out twice for the two target
shapes with the same constraints, while the objective functions are
different. Point 1 is chosen as the reference point. In target shape 1,
the height differences between point 1 and the other three points
are zero; i.e., the four points are on the same plane, and the objec-
tive function is written as

Q ¼ jz7  z1 j þ jz4  z1 j þ jz10  z1 j ð11Þ

The objective function decreases from 35.24 to 0.01 in the shape
design process to obtain the final shape as shown in Fig. 18. It can
be seen that the four target points are on the same plane, showing
the proposed shape design method is effective even for this com-
plex example.
In target shape 2, the height difference is zero between Points 1
and 7, and 10.0 between Point 1 and Points 4 and 10; hence, the
objective function for target shape 2 is written as

Q ¼ jz7  z1 j þ jz4  z1  10j þ jz10  z1  10j ð12Þ

The objective function decreases from 20.35 to 0.01 in the shape
design process of target shape 2 to obtain the final shape as shown
in Fig. 19. There are obvious height differences between the target
points in the shape design result, and it achieves the target shape
as desired. This example confirms that the proposed methods can
realize different target shapes according to different architectural Fig. 20. Assembled initial model of Example 3; (a) Top view, (b) Front view, (c)
requirements. Diagonal view.

18
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

Fig. 21. Assembly process of Example 3.

7.3. Example 3 that the assembled tensegrity units with 3-bar reciprocal struts are
also acceptable for a practical purpose.
Different from the previous two examples, the third example is After assemblage of units, topology optimization is conducted
composed of six 3-bar units as shown in Fig. 20. Fig. 21 shows its first by minimizing the sum of the forces of all cables as the objec-
assembly process by 12 struts and 72 cables. The arrangement of tive function. The cables shown in Fig. 22, whose forces are less
cables in the 3-bar units with reciprocal struts can be seen in than 1% of those of the other cables, can be judged as redundant
Fig. 20(c). It is similar to the arrangement in the 4-bar units as and removed to obtain the optimal topology as shown in Fig. 23.
shown in Fig. 5 except that the number of struts in each unit The number of prestress modes decreases from 19 to 7 after topol-
changes from 4 to 3. The assembled structure is stable, confirming ogy optimization. This way, the topology optimization process
reduces the number of cables and makes the structure lighter
and simpler.
Shape design is conducted from the optimal topology in Fig. 23
as the initial solution. The target shape is controlled by the heights
of the points in the outer and inner ring marked as O1-O6 and I1-I6,
respectively, in Fig. 23. Point O1 is chosen as a reference point to
exclude the effect of the rigid-body displacements. The target
height distance is zero between Points O1 and Points Oi (i–1),
and Dh between Point O1 and Points Ij. Hence, the objective func-
tion in the shape design process of Example 3 can be written as

X
6   X6  
zO  zO  þ  
Q¼ i 1
z I j
 z O1
 D h  ð13Þ
i¼2 j¼1

The shape design is conducted with different target heights Dh

and the results are shown in Fig. 24. All the processes of the four
results are the same except the values of Dh. This result shows that
the proposed shape design method is feasible and easy to adjust
according to different target shapes.
Fig. 22. Cables that can be removed in Example 3.
7.4. Example 4

The structure in Example 4, which is an extension of that in

Example 3, is composed of different reciprocal units. As shown in
Fig. 25, multiple struts and cables are added along the outer ring
of Example 3. The struts of Example 4 are shown separately in
Fig. 26. It can be seen that six 4-bar units and six 3-bar units are
added to the original model of Example 3. Moreover, even 5-bar
unit and 6-bar unit can be found in the model.
In the same manner as previous examples, topology optimiza-
tion is conducted first. The cables shown in Fig. 27 can be judged
as redundant and removed to obtain the optimal topology as
shown in Fig. 28. The number of prestress modes decreases from
66 to 18 after topology optimization. Comparing the cables in
Fig. 27 with those in Fig. 22, they are totally different even though
the inner cables arrangements of the two models are identical. It
shows the cables that can be removed depend on the topology of
the tensegrity, and there is no unified rule for selection of the
redundant cables. Hence, the topology optimization process is
Fig. 23. Topology optimization result of Example 3. essential for the purpose of removing redundant cables.
19
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

Fig. 24. Shape design results of Example 3.

Fig. 27. Cables that can be removed in Example 4.

Fig. 25. Initial model of Example 4.

Fig. 28. Topology optimization result of Example 4.

is controlled by the heights of the points in the outer and inner

rings. The points in the outer ring are marked as A1–A12 in
Fig. 28, and those in the inner ring are the same as Example 3. Point
A1 is chosen as the reference point to exclude the effect of the
Fig. 26. Struts of Example 4.
rigid-body displacements. The target height distance is zero
between Points A1 and Points Ai (i–1), and 1.5 between Point A1
Shape design is conducted by using the optimal topology in and Points Ij. Hence, the objective function in the shape design pro-
Fig. 28 as the initial model. Similar to Example 3, the target shape cess of Example 4 can be written as
20
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

Fig. 29. Shape design result of Example 4; (a) Top view, (b) Diagonal view, (c) Front view, (d) Left view.

X
12   X6  
 zA  zA  þ   Acknowledgments
Q¼ i 1
zIj  zA1  0:4 ð14Þ
i¼2 j¼1
The present work has been conducted with the financial sup-
The shape design result is presented in Fig. 29. It demonstrates port of the Chinese National Natural Science Foundation (Project
that the proposed shape design method is also effective for the designation: 51578186 and 51778182). The first author appreci-
tensegrity composed of units with different configurations. ates the support from the Chinese Scholarship Council (File No.
201806120237) that enables him to conduct cooperative research
8. Conclusions with Prof. Ohsaki and Prof. Zhang at Kyoto University. The second
author appreciates the supports from JSPS KAKENHI (Grant Num-
A new type of large-span tensegrity with reciprocal struts has ber 19H05369).
been proposed in this paper. New methods for assembly, topology
optimization and shape design have been developed, and the con- References
clusions are summarized as follows:
Ali, N.B.H., Rhode-Barbarigos, L., Smith, I.F., 2011. Analysis of clustered tensegrity
1) The proposed configuration of the tensegrity structure with structures using a modified dynamic relaxation algorithm. Int. J. Solids Struct.
48 (5), 637–647.
reciprocal strut is a ‘real’ tensegrity structure that can cover Denton, M.J., Dearden, P.K., Sowerby, S.J., 2003. Physical law not natural selection as
a large space with discontinuous struts. The proposed assem- the major determinant of biological complexity in the subcellular realm: new
bly method can make full use of the relationship between support for the pre-Darwinian conception of evolution by natural law.
Biosystems 71 (3), 297–303.
tensegrity and reciprocal units, and realize the assembly of Donev, A., Torquato, S., Stillinger, F.H., Connelly, R., 2004. A linear programming
tensegrity units successfully. The proposed method is effec- algorithm to test for jamming in hard-sphere packings. J. Comput. Phys. 197 (1),
tive for different types of units and their combination. 139–166.
Ehara, S., Kanno, Y., 2010. Topology design of tensegrity structures via mixed
2) In the topology optimization process, the initial prestresses integer programming. Int. J. Solids Struct. 47 (5), 571–579.
for the non-rigid-body motion analysis are chosen as vari- Feng, X., 2017. The optimal initial self-stress design for tensegrity grid structures.
ables, and the sum of the self-equilibrium forces of cables Comput. Struct. 193, 21–30.
Gómez-Jáuregui, V., Arias, R., Otero, C., Manchado, C., 2012. sNovel technique
is minimized as the objective function. The redundant cables
for obtaining double-layer tensegrity grids. Int. J. Space Struct. 27 (2–3), 155–
to be removed can be successfully found by solving the 166.
topology optimization problem. Kan, Z., Peng, H., Chen, B., Xie, X., Sun, L., 2019. Investigation of strut collision in
tensegrity statics and dynamics. Int. J. Solids Struct. 167, 202–219.
3) In the shape design process, the stiffnesses of members for
Kanno, Y., 2013. Topology optimization of tensegrity structures under compliance
the non-rigid-body motion analysis are chosen as variables, constraint: A mixed integer linear programming approach. Optimiz. Eng. 14 (1),
and the height difference between the current shape and the 61–96.
target shape is minimized as the objective function. This Kawaguchi, K., Ohya, S, 2004. Preliminary report of observation of real scale
tensegrity skeletons under temperature change. In: Proceedings of
way, the structural shape close to the target shape is suc- International Association for Shell and Spatial Structures. Sep. 2004.
cessfully obtained to satisfy architectural requirements. Dif- Koohestani, K., Guest, S.D., 2013. A new approach to the analytical and
ferent shapes can be realized by different assignments of the numerical form-finding of tensegrity structures. Int. J. Solids Struct. 50 (19),
2995–3007.
target shape. Larsen, O.P., 2008. Reciprocal frame architecture. Routledge.
Ma, Q., Ohsaki, M., Chen, Z., Yan, X., 2019. Multi-objective optimization for prestress
design of cable-strut structures. Int. J. Solids Struct. 165, 137–147.
Motro, R., 2003. Tensegrity: Structural Systems for the Future. Elsevier.
Declaration of Competing Interest Ohsaki, M., Zhang, J.Y., 2015. Nonlinear programming approach to form-finding and
folding analysis of tensegrity structures using fictitious material properties. Int.
The authors declare that they have no known competing finan- J. Solids Struct. 69, 1–10.
Rimoli, J.J., Pal, R.K., 2017. Mechanical response of 3-dimensional tensegrity lattices.
cial interests or personal relationships that could have appeared Compos. B. Eng. 115, 30–42.
to influence the work reported in this paper. Skelton, R.E., de Oliveira, M.C., 2009. Tensegrity systems. Springer, New York.

21
Y. Su et al. International Journal of Solids and Structures 206 (2020) 9–22

San, B., Waisman, H., Harari, I., 2019. Analytical and numerical shape optimization Zhang, L.Y., Zhao, H.P., Feng, X.Q., 2015. Constructing large-scale tensegrity
of a class of structures under mass constraints and self-weight. J. Eng. Mech. 146 structures with bar–bar connection using prismatic elementary cells. Arch.
(1), 04019109. Appl. Mech. 85 (3), 383–394.
Smaili, A., Motro, R., 2005. A self-stress maintaining folding tensegrity system by Zhang, J.Y., Ohsaki, M., 2006a. Form-finding of tensegrity structures subjected to
finite mechanism activation. J. Int. Assoc. Shell Spatial Struct. 46 (2), 85–93. geometrical constraints. Int. J. Space Struct. 21 (4), 183–195.
Snelson, K., 2012. The art of tensegrity. Int. J. Space Struct. 27 (2–3), 71–80. Zhang, J.Y., Ohsaki, M., 2006b. Adaptive force density method for form-finding
Su, Y., Ohsaki, M., Wu, Y., Zhang, J., 2019a. A numerical method for form finding and problem of tensegrity structures. Int. J. Solids Struct. 43 (18–19), 5658–5673.
shape optimization of reciprocal structures. Eng. Struct. 198, 109510. Zhang, J.Y., Ohsaki, M., Kanno, Y., 2006. A direct approach to design of geometry and
Su, Y., Wu, Y., Qiao, G., Shen, S., 2019b. Self-adaptive form generation method forces of tensegrity systems. Int. J. Solids Struct. 43 (7–8), 2260–2278.
for reciprocal grid structures. Comput.-Aided Civ. Infrastruct. Eng. 34 (5), 444– Zhang, J.Y., Ohsaki, M., 2007. Optimization methods for force and shape design of
454. tensegrity structures. In: Proc. 7th World Congresses of Structural and
Sultan, C., 1999. Modeling, design, and control of tensegrity structures with Multidisciplinary Optimization, pp. 40–49.
applications. Zhang, J., Ohsaki, M., 2013. Free-form design of tensegrity structures by non-rigid-
Widyowijatnoko, A., Irwanuddin, I., Aditra, R.F., 2019. Rection as a synthesis of body motion analysis. In: Proceedings of International Association for Shell and
reciprocal and tensegrity structure. Nexus Netw. J., 1–14 Spatial Structures, Sep. 2013.
Yuan, X., Chen, L., Dong, S., 2007. Prestress design of cable domes with new forms. Zhang, J.Y., Ohsaki, M., 2015. Tensegrity structures: form, stability, and symmetry.
Int. J. Solids Struct. 44 (9), 2773–2782. In: Mathematics for Industry 6, Springer. 2015

22