Professional Documents
Culture Documents
RAMAKANTA PANIGRAHI
JUNE, 2007
DEVELOPMENT, ANALYSIS AND MONITORING
OF DISMOUNTABLE TENSEGRITY STRUCTURE
By
Ramakanta Panigrahi
Submitted
in fulfillment of the requirements of the degree of
Doctor of Philosophy
to the
by Mr. Ramakanta Panigrahi to Indian Institute of Technology, Delhi, for the award of
the degree of the Doctor of philosophy is a record of bonafide research work carried out
by him. He worked under our supervision for the submission of this thesis, which to the
The research reports and the results presented in this thesis have not been submitted in
parts or in full to any other University or Institute for the award of any degree or diploma.
INDIA INDIA
i
ACKNOWLEDGEMENTS
I owe my sincere heartfelt thanks and deep sense of gratitude to my supervisors Professor
Ashok Gupta and Dr. Suresh Bhalla. It was my privilege that I was particularly blessed in
having two professors providing me with the assistance and the invaluable guidance
study. They were easily accessible whenever I needed. I am eternally grateful for their
kindness, helpful guidance, discussions and timely help. They truly exemplify the role of
advisors.
I am grateful to Professor K.G. Sharma, Professor T.K. Dutta and Professor Y. Nath, my
thought process.
I would also like to thank the technical staff of structural laboratory Mr. Sukanta Sahu
and Mr. Nitin Chaurasia, who helped me in conducting the experiments and in extending
all the facilities required. I also express my sincere thanks to Mr. K.K. Bali and Mr. Ram
Kumar for their help in carrying out my fabrication work in the workshop of Civil
Engineering Department.
ii
I am thankful to my dear Oriya friends Bulu, Sushanta, Deba and Punam now doing
research at IIT Delhi, without all of whom, things would have been very difficult. They
were ready to help me at the time of need. My sincere thanks to all the people who made
I want to thank my wife, Rosy, for her endless love and support for helping me to keep
my focus on my research work. She shared all the difficulties and disappointments and
without her help, I wou1d not be where I am today. Last but not the least, to my son,
Sibun, a thank you, for constantly reminding me of what is truly important in my life!
Government of Orissa and many people who contribute to this work both directly and
indirectly.
iii
ABSTRACT
Thousands of people are rendered homeless and displaced during natural calamities in
different parts of the globe every now and then. A low-cost dismountable shelter can be a
boon to the homeless during these situations. Tensegrity structures offer the most
attractive solution for the purpose of temporary shelter due to their flexibility,
This thesis primarily focuses on the development, analysis and monitoring the behaviour
of a new type of tensegrity based dismountable roof structure. First, a single module,
Instrumentation is carried out right at the fabrication stage. The structure is subjected to
destructive load test during which continuous monitoring of the prestress levels,
deflections and strains in the members is carried out. The monitored structure is also
analyzed using finite element method (FEM) and the numerical results compared with the
parametric studies. The investigations are then extended to a tensegrity grid, 2mx2m in
size, built by cohesive integration of four single tensegrity modules. In order to make sure
cables. The behaviour of the tensegrity structure under different loads (dead loads, live
loads, lack of fit in cable, thermal loads and wind loads) and their combinations is studied
in detail.
iv
Apart from the fabrication and testing, a new methodology has been developed for design
of tensegrity structures using artificial neural network (ANN) based on multilayered feed
new low cost version of EMI is successfully applied on tensegrity structures. Both strain
gauges and piezo-transducers are found to complement each other by assisting in load
history retrieval (LHR) and structural health monitoring (SHM) respectively. Piezo
transducers is also used for global dynamic characterization of tensegrity structures and
found suitable for obtaining information concerning the modes of vibration and damage
Reasonable agreements are found between experimental and numerical studies. The
research concludes with the findings obtained through experiments and numerical
studies. It is hoped that this research will make significant contributions in the field of
tensegrity structures.
v
CONTENTS
Page No.
CERTIFICATE i
ACKNOWLEDGEMENTS ii
ABSTRACT iv
CONTENTS vi
NOTATIONS xxii
ABBREVIATIONS xxvi
1.1 INTRODUCTION 1
1.1.3 Properties 7
1.1.4 Advantages 11
1.1.5 Applications 12
1.1.6 Disadvantages 13
vi
1.2.3 Dynamic Analysis of Tensegrity Structures 24
Structures 31
2.1 INTRODUCTION 40
RESULTS 64
vii
CHAPTER 3 TESTING AND ANALYSIS OF DISMOUNTABLE
3.1 INTRODUCTION 87
RESULTS 92
STRUCTURES 98
LOADS
TEMPERATURE 119
viii
4.2.4 Thermal Loads 120
ix
5.3 NEURAL NETWORK ARCHITECTURE 155
x
SENSORS 188
Technique 189
REFERENCES 220
PUBLICATIONS 239
APPENDIX 243
xi
LIST OF FIGURES
Figure 1.1 Snelson’s needle tower of 30m high made of X shaped basic
strut mode 51
xii
mode 52
arrangement 69
strut mode 73
Figure 2.28 Comparison of strut force for different stages of model updating in
strut mode 74
xiii
Figure 2.29 Comparison of deflection for different stages of model updating in
cable mode 75
Figure 2.30 Comparison of strut force for different stages of model updating in
cable mode 75
Figure 2.32 Comparison of strut force between experiment and updated numerical
Figure 2.33 Comparison of bottom cable force between experiment and updated
Figure 2.34 Comparison of top cable force between experiment and updated
Figure 2.36 Comparison of strut force between experiment and updated numerical
Figure 2.39 Percentage increase in strut force between rigidity ratio 50 & 20 84
and 1500 N 85
xiv
Figure 3.1 Detail of central bottom joint 89
Figure 3.14 Deflection of 4m×4m tensegrity grid structure at bottom nodes 101
Figure 3.15 Deflection of 4m×4m tensegrity grid structure at top nodes 101
xv
Figure 3.17 Variation of forces in struts in 4m×4m grid structure 103
Figure 3.19 Variation of top cable force in 4m×4m grid structure 104
Figure 3.20 Variation of leg cable force in 4m×4m grid structure 104
Figure 3.21 Variation of inner bottom cable force in 4m×4m grid structure 105
Figure 3.24 Variation of maximum strut force in 4m×4m grid structure for
Figure 3.27 Maximum deflection for 8×8m grid structure for different heights 110
Figure 3.28 Maximum deflection for 8×8m grid structure for different rigidity
ratio 110
Figure 3.29 Load carrying capacity of 8×8m grid structure for different
Figure 3.30 Load carrying capacity of 8×8m grid structure for different heights 111
Figure 3.31 Maximum strut force of 8×8m grid structure for different heights 112
Figure 3.32 Maximum strut force of 8×8m grid structure for different rigidity
ratios 112
xvi
Figure 3.33 Comparison of maximum deflection in 8×8m grid structure for
Figure 3.34 Comparison of maximum strut force in 8×8m grid structure for
Figure 3.35 Comparison of maximum strut force in 8×8m grid structure for
Figure 3.36 Comparison of load carrying capacity of 8×8m grid structure for
Figure 3.37 Comparison of load carrying capacity of 8×8m grid structure for
Figure 3.38 Comparison of load carrying capacity different grid structure 117
Figure 4.2 Variation of deflection at node 3 for different load cases 129
Figure 4.3 Variation of top cable force for different load cases
Figure 4.5 Variation of leg cable force for different load cases
Figure 4.6 Variation of bottom cable force for different load cases
xvii
Figure 4.8 a) 8m grid with node numbers b) Loading for case 1 143
Figure 5.2 Artificial Neuron with input Function and Output 157
Figure 5.7 Comparison between target value and ANN output for
Figure 5.8 Comparison between target value and ANN output for
Figure 5.9 Comparison between target value and ANN output for
Figure 5.10 Comparison between target value and ANN output for
Figure 5.11 Comparison between target value and ANN output for
Figure 5.12 Comparison between target value and ANN output for
Figure 5.13 Comparison of between target value and ANN output for
Figure 5.14 Comparison between target value and ANN output for
xviii
Figure 6.1 Modelling PZT-structure interaction
Figure 6.5 a) Circuit for measuring PZT admittance by Peairs et al. (2004) 190
Figure 6.9 PZT patch bonded on a strut close to the joint 196
hammering 197
Figure 6.13 Change in signature with increase in damage in the cable 200
Figure 6.15 Change in signature with increase in damage in the strut 202
xix
(b) Second damage 203
Figure 6.18 Signature of the grid structure obtained before and after damage 205
xx
LIST OF TABLES
Table 3.1 Strut number and their connectivity in 4m×4m grid structure 102
Table 4.4 Maximum and minimum in elements due to combined load 127
Table 4.7 Maximum deflection due to combined live load and wind load
Table 4.8 Member force range for case 1 & 2 due to combined live load
Table 4.9 Member force range for case 3 & 4 due to combined live load
Table 5.1 Validation of trained and tested data for tensegrity grid structure 164
xxi
NOTATIONS
C = Capacitance
E = Young’s modulus
K = Stiffness matrix
P = Load vector
Q = Charge
V = Voltage
Z = Mechanical impedance
i = Mode number
m = Number of links
n = Number of struts
p = Element number
r = Rank
xxii
As = Cross sectional area of strut
D3 = Electric displacement
Fs = Force in strut
{F } app
= Vector of externally applied nodal point loads at time t+Dt
{F } i
nr
= Vector of nodal point loads at time t
I (t ) = Input vector
[K ] i = Tangent matrix
xxiii
KNL = Nonlinear component of stiffness matrix
Ls = Length of strut
Oo (t ) = Network output
Vi = Input voltage
Vo = Output voltage
T1 = Stress
To (t ) = Desired output
Yi = Electro-mechanical admittance
connection
Whi = Adaptive rule of weights for input layer to hidden layer connection
ΔP = Incremental load
xxiv
for a particular mode
η = Learning rate
δ = Structural deformation
λ = Slenderness ratio
xxv
ABBREVIATIONS
CN = Cyanoacrylate
DR = Dynamic Relaxation
ER = Electro-rheological
GI = Galvanized Iron
IS = Indian Standards
xxvi
NDE = Nondestructive Evaluation
xxvii
CHAPTER 1
1.1 INTRODUCTION
Natural disasters like floods, earthquakes and hurricanes affected more than 2.5 billion
people across the globe between 1994 and 2003. The figure represents a 60% increase
January 2005. However, this figure does not include millions others distressed by the
Tsunami of December 2004, which killed an estimated 180,000 people and the
earthquakes which struck Pakistan and India in 2005. It is observed that larger number of
people are affected by natural calamities now a days, due to increase in population in the
coastal and the earthquake prone areas. Thousands of people become homeless and
displaced after such natural calamities in different parts of the globe. Due to the rise in
the sea level, expanding deserts and catastrophic weather induced flooding; the number
Nations experts. Leaving aside the natural disasters, it has been found that large number
of people use footpaths as shelter during nights. During disasters, shelters are required for
displaced people. Permanent shelters are not only expensive but also take months
altogether to be built. A low cost temporary shelter can be a boon to the homeless during
these situations. Dismountable temerity structures are the most attractive solution for
temporary shelters due to their lightweight, easy assembling and dismantling for
constructing large grid roofs in the field, low cost and reuse.
1
‘Tensegrity’ is a relatively new concept using which one can create amazing lightweight
and adaptable structures, giving the impression of a cluster of struts floating in the air.
Tensegrity structures can be defined as a pattern that results when the ‘push’ provided by
struts and the pull provided by tendons achieve a win-win relationship with each other.
While pull is continuous, push is discontinuous, and the two balance each other,
compression (bars) and tension (cables) members, where the cables surround bars. In
1921, Johanssen presented a sculpture made with three struts and eight cables during an
exhibition held in Moscow (Gough, 1998). This system had no rigidity and mechanism
could be activated with one of the cables. In 1948, Kenneth Snelson worked on similar
structures with Richard Buckminster Fuller, in Black Mountain College. Fuller expressed
his idea of “islands of compression inside a sea of tension”. Snelson, who followed Fuller
in the field of tensegrity, developed three distinct models. The first was a balancing
shape, the second a linking system with tensile elements and the third an X- shape, which
was used as the basic tensegrity module in the structure shown in Fig. 1.1. “Continuous
tension, discontinuous compression structures” was the title of the patent awarded to him
(Snelson 1965).
Fuller (1962) introduced the term “Tensegrity”, which has been patented in U.S.A., as a
contraction of the two words “tension” and “integrity”. The tension elements provide the
himself the inventor of this new structural system. He patented a system called “Pearl
2
Frameworks” at INPL (Institut National de la Propriete Industrielle), France. These
pioneers i.e. Fuller, Snelson and Emmerich patented their systems, all of which indicate a
similarity.
(a)
(b)
Figure 1.1 Snelson’s needle tower 30 m high made of X shaped basic modules
(a) Needle tower
(b) Basic X module ( Snelson 2004)
3
In contrast to the cable structures, tensile forces in the tensegrity structures are controlled
by inner self-stress states. They are self-supporting and do not require expensive
anchorages. The resilience, which is due to flexible linking of the tensile components, is
are rigidly interconnected, they will lose their ability to transmit forces throughout the
that shear forces are not a major factor in the gross behaviour of the structure. Based on
the tensegrity principle, Burkhardt (2004) described the function of structural framework
of the non-woody plant. A young plant is completely composed of cells of water that
behave much like the balloon. The skin of the cell is a flexible inter-linkage of molecules
held in tension by the force of the water in a contained cell. As the plant is stretched and
bent by wind, rain and other natural forces, the forces are distributed throughout the
plant, without disturbing to its overall structural integrity. It springs back to its usual
shape even in the course of the natural upheavals and finds itself not distorted from the
original shape.
4
functioning in the same manner that is accomplished by pneumatic structures, at the sub
According to Emmerich (1963) “Self stressing structures consist of bars and cables
assembled in such a way that the bars remain isolated in a continuum cables. All these
elements must be spaced rigidly and at the same time interlocked by the prestressing
resulting from the internal stress of the cables without the need for external bearings and
anchorage. The whole is maintained firmly like a self supporting structure, whence the
term self-stressing”.
stable volume in space”. Schodeck and Daniel (1993) labelled tensegrities as rigid
which each component has one degree of member redundancy. Hanaor (1994) defined
which the cables or tendons are tensioned against a system of bars and struts”. Miura and
Pellegrino (1999) described, “A tensegrity structure is any structure realised from cables
and struts, to which a state of prestress is imposed that imparts tension to all cables”.
They further added to the above definition that “as well as imparting tension to all cables,
the state of prestress serves the purpose of stabilising the structure, thus providing first
5
In addition to the above major definitions, following other definitions can be found in the
literature. “Tensegrity systems are spatial reticulated systems in a state of self stress. All
their components are rectilinear and of equivalent size. Tensioned elements (cables) have
have no rigidity in traction and constitute a discontinuous set” as defined by Motro and
structure if at most “k” compressive members are connected to any node. In general, any
traditional tensegrity structure is a class one structure because only one compression
structure.
The most recent definition of tensegrity structure was given by Motro (2003) “A
tension tendons, the combined parts mutually support in such a way that the struts do not
touch one another, but press outwardly against nodal points in the tension network to
form a firm, triangulated and prestressed systems of tension and compression unit.
6
1.1.2 Configurations of Tensegrity Structures
Tensegrity structures are divided into two broad structural classes i.e. prestressed and
into different patterns i.e. diamond, circuit and zigzag. Depending upon the joining
structures into grids, mast and domes. Spherical system is further classified into rhombic
A tensegrity prism is obtained by introducing a relative rotation between the upper and
lower polygons. Kenner (1976) determined the shape of the symmetric tensegrity
π π
structures by using the formula α = − where α is the minimum rotation angle and n
2 n
is the number of struts. Fig. 1.2 shows the perspective and the top view of a half-
cuboctahedron based tensegrity structure and Fig 1.3 shows a simplex type tensegrity
structure. A family of tensegrity prisms is presented in Fig 1.4. Snelson’s needle tower of
30m height shown in Fig 1.1, was constructed with the X shaped basic modules in 1967
1.1.3 Properties
with similar resistance. In other words, they have a high resistance in comparison
7
(a) (b)
Top ties
Leg ties
Struts
Bottom ties
8
(a)
(b) (c) (d)
(h)
(e) (f) (g)
9
• They do not depend on gravity due to their self-stability, so they do not need to
be anchored or leaned on any surface. The systems are stable in any position. The
higher too. Using the analogy of the balloon, if a balloon is more inflated, the
tension forces in the skin are greater and it is harder to deform it.
The forces of compression are located along specific and short lines of action, so
• Due to the discontinuity in compression, they are not acted by torque at all.
• They exhibit the property of synergy, hence the overall behaviour of the system
separately.
employed, and the method of assembly. They can be very flexible or very rigid
• They have the ability of respond as a whole, so local stresses are transmitted
10
• The response to the loads is non linear. They are more flexible under light loads,
but their stiffness increases rapidly as the load increases, just like in a suspension
bridge.
1.1.4 Advantages
etc. employing tensegrity principles will make them highly resilient and, at the
• Tensile forces naturally transmit themselves over the shortest distance between
two points; hence the members are precisely positioned to best withstand stress.
• The fact that these structures vibrate readily means that they transfer loads very
quickly, so the loads cannot stress the structure locally. This is very useful in
together. This conception implies the option of the endless extension of the
assembled piece.
11
• The kinematic indeterminacy of these structures is sometimes an advantage. In
configuration because the shape changes with the equilibrium of the structure.
1.1.5 Applications
The concept of tensegrity was initiated on sculptures but soon found place in architecture
and mathematics. Presently, the concept is used in different fields like civil engineering,
and biomechanics for research activities. Ingbar introduced the concept in biology first
circumstances:
• Portable and foldable structures: such as for disaster struck areas, nomadic
rugged terrains).
• Frames over large areas for environmental control, energy transformation and
food production.
12
• Earthquake-resistant buildings, bridges, shelters, etc. As these structures are
extremely resilient and they could withstand large structural shocks like
earthquakes.
• Tensegrity structures can be used as tent like structures and shadow roofs,
construction of tensegrity arches, foldable reflector antennas and masts for large
ornaments, as footbridges.
The example of application of tensegrity dome is shown in Fig. 1.5 and a foot bridge
assembling several simplex based tensegrity modules along their main axis is shown in
Fig. 1.6.
1.1.6 Disadvantages
13
• The absence of adequate commercial design tools has also been a limitation until
now. There was a lack of design and analysis techniques for these structures.
14
1.2 LITERATURE REVIEW
The basic issues related to tensegrity structures are their form finding, static and dynamic
analysis, design and deployment in the field. The literature reported regarding these
mechanics but no solution can dispense with either of these two aspects. As reported by
Motro (2003), Snelson used a “form controlled” method for form finding. Since stability
was obtained by trial and error process, it is not an impressive solution. The second
method is known as force controlled method, which was developed to ensure the
mechanical requirements using a theoretical model. Although this method may generate
Emmerich (1959) mainly used a geometric approach, using polyhedra, to achieve two
main conditions i.e. the nodes are the apices of polyhedra and the system has at least one
stable self stress condition. For very simple systems, like the three strut tensegrity
module, a static approach was used. The equilibrium conditions for one node determine
the resulting shape of the structure. For regular tensegrity systems, the shape can be
defined by the single parameter i.e. r =s/c, ‘s’ being the length of the strut and ‘c’ that of
the cables. In this case, the form finding is a single parameter process. Motto (2003) used
15
these concepts and analysed a system with six struts and twenty-four cables i.e. an
“expanded octahedron”.
Vassart (1997) developed a multiparameter form finding process for irregular modules
and assemblies, based on the force density approach (covered in later sections). He
defined irregular shapes, either by choice of the force density coefficients or by analysis
and studied the resulting graphic developments. He also designed irregular shapes and
mapped them on double curvature surfaces with positive or negative Gaussian curvature
or both.
Vassart and Motro (1999) derived numerical procedures for the form finding, analysis
and fabrication of wide-span cable nets and grid shells, uniform or variably prestressed
fabric membranes and battened membrane roofs, based on the method of dynamic
relaxation with kinetic damping. Two sets of form-finding parameters were identified i.e.
the prestress (or selfstress) coefficients of the members and the coordinates or the
redundant nodes. This method was applied to tensegrity systems and further extended to
Two basic approaches for the overall form finding of tensile structures have been
developed and applied in practice- the matrix method and the vector method (Wakefield
1999). The matrix method is typically an application of the standard non-linear structural
matrix is solved incrementally until convergence is obtained. Special controls limiting the
16
maximum incremental deflections and the nodal residual forces may be required. The
stress/strain relations for the individual components are coupled with equilibrium and the
compatibility requirements for the complete structure. Matrix methods can be further
classified as incremental and iterative methods (Barnes, 1977). The approach consists of
solving a system of equations which links the stiffness matrix K with the load vector P to
P=K.δ (1.1)
Because of high flexibility and large displacements associated with tensegrity structures,
be taken into account by adding the nonlinear component K NL to the linear stiffness
matrix K. The incremental Euler method solves this system of equations by applying the
ΔP = (K L + K NL ) ⋅ Δδ (1.2)
Iterative methods, such as the Newton-Raphson method, can also be employed to the
above equation. Only difference is that instead of applying the load stepwise, the residual
The force density method and dynamic relaxation method are most popular among all the
available methods for form finding of the tensegrity structures. These are described in
detail.
17
1.2.1.1 Force density method
Sheck (1974) introduced the force density method to determine the possible shapes of
equilibrium of a pin-jointed network consisting of cables and bars. In this method, any
equations. The ratio between the element forces and the element lengths of the network is
called the force density. For given loading, support conditions and force densities, the
As noted Maurin and Motro (1997), the main drawback of the force density method is
that the final distribution of stresses is difficult to control. However, this can be overcome
by iterating with the updated force densities until the desired smooth stress distribution is
achieved, but this would seem to negate the advantage of a linearised solution. Once a
form has been found, a vector or matrix method must be used to analyse its response
under load. Masic et al. (2005) found the force density method as the best method for
form finding of large-scale tensegrity structures. They also observed that this method
additional expense as it does not require unknown knowledge about the structure a priori.
For form finding of tensegrity structures, Zhang and Ohsaki (2006) explained adaptive
force density method based on eigenvalue analysis and spectral decomposition of the
equilibrium matrix with respect to the nodal coordinates. However, they opined that this
method has no direct and exact control over the geometrical and mechanical properties of
the structure. Estrada et al. (2006) proposed a numerical form finding procedure for
18
tensegrity structures without assuming any initial condition to be imposed as a priori,
except the type of each member and the connectivity of the nodes. To search a state of
self stress with minimal elastic potential energy, they included maximal rank of force
density matrix and minimal member length in the form finding procedure.
The dynamic relaxation method is an energy minimization technique that searches the
static equilibrium state by simple vector iteration method. However, such structural
analysis of tensegrity structures must account for geometrical nonlinearity. The dynamic
relaxation method correctly models the static behaviour of tensegrity structures in most
situations. In this method, the static solution of both linear and non-linear structures
vibrations excited by that load. The physical basis of the DR method was initially
perceived as the step by step solution, for small time increments Δt of the Newton’s
damping (Day, 1962). He used dynamic relaxation method to analyze concrete structure.
Barnes (1994) used the DR method for form finding and analyzed prestressed membrane
DR method with or without kinetic damping can also be used for single parameter
systems as done by Kebiche et al. (1999). They proposed a method of computation based
linearities. This method was applied to study the behaviour of a single as well as a four
19
bar tensegrity system for various self-stress levels under external actions i.e. traction,
compression, flexure and torsion. They found an increase in systems rigidity with
increase in the applied load and self-stress level except for compression loading.
Spline beam element is used for modelling spline beams or grid shells employing
requires only three translational degrees of freedom per node and no rotational degrees of
system, Adriaenssens and Barnes (2001) developed a numerical analysis based on the DR
method specifically for form finding and load analysis of this type of structure. The
bending components were treated in a finite difference form with three degrees of
freedom per node rather than six. The method was found to be particularly useful for grid
Fest et al. (2003) fabricated and tested a full-scale prototype of an adjustable tensegrity
structure at Swiss Federal Institute of Technology (EPFL). The test results indicated the
linear behaviour of the structure subjected to vertical loads on single joints and nonlinear
behaviour for loads applied to several joints. DR method was found to be reliable for
predicting the response of the prototype structure under symmetric and asymmetric
was obtained by increasing the number of adjustable struts from two to five per module
20
1.2.2 Static Analysis of Tensegrity Structures
Although the origin of the tensegrity structures can be pinpointed to 1921, the main
investigations were carried out nearby fifty years later. Fuller (1975) and Pugh (1976)
using mechanics were later developed aimed at establishing the theoretical framework for
analysis and design. As per Maxwell’s rule, a truss having b bars and j joints will be stiff
if b =3j-r where r is number of reactions. Without satisfying Maxwell’s rule, some of the
Buckminster’s tensegrity structures are stiff structures having less number of bars. For
these special cases, the stiffness will be of lower order and also permit at least one state
of “self-stress” in the frame, as stated by Maxwell. Using linear algebra, Calladine (1978)
obtained the number of "incipient" modes of low-order stiffness depending upon the
number of the bars, the joints and the independent states of self-stress. Further, he found
the effects of self-stress in the frame imparting first-order stiffness to the frame and
Maxwell’s rule cannot be applied to tensegrity systems to determine whether the system
is stable (Calladine, 1978). The composition and analysis of an equilibrium matrix (H),
which links the nodal loads to the member forces, provides deeper insights into the
Hanaor (1988) classified prerstressable structures into two classes - class I and class II.
Class I structures are geometrically rigid and statically indeterminate structures. Class II
structures, on the other hand are statically and kinematically indeterminate with
21
occurs in class I structures, prestress by means of imposed lack of fit can substantially
enhance the strength. Structures of class II mainly depend on prestress for their geometric
integrity. Assuming small displacements and using flexibility method, Hanaor developed
a unified algorithm for the analysis and design of prestressed of pin-jointed structures of
Hanaor and Liao (1991) analyzed double layer tensegrity grids (DLTG) consisting of
triangular prisms subjected to static loads using a first order linear analysis (small
found sharp increase in member forces with increase in span and that of decrease with
structural depth.
span DLTG consisting of seven triangular prismatic units. The actual response i.e. the
deflections and the bar forces was nonlinear and higher than that predicted by the linear
model. He found a high degree of structural redundancy in the DLTGs constructed from
individual prismatic units. Load bearing capacity was practically not changed by loss of a
member.
By using the principle of virtual work, Sultan et al. (2001) formulated the general
22
certain tensegrity structures and found the state of stress of the structure to depend only
each stage is 2m−5 and is independent of the number of stages. Murkami and Nishimura
(2001 a) presented analytical expressions for initial geometry and the associated pre-
stress modes for cyclic frustum tensegrity modules for an arbitrary number of stages with
infinitesimal mechanism modes in proportion to the square root of the amplitude of the
pre-stress mode. Murkami and Nishimura (2001 b) derived the initial configurations and
the prestress modes for regular truncated icosahedral and dodecahedral tensegrity
modules. The infinitesimal mechanism modes were classified into subspaces based on the
Cesar (2001) and Crane et al. (2005) determined the equilibrium position of a tensegrity
structure subjected to external forces and external moments by static analysis using the
principle of virtual work together with concepts related to geometry of lines. Considering
anti-prism tensegrity structures and assuming struts as massless, of same length, with
only one external force applied per strut and no dissipative force acting on the system,
they obtained a MATLAB based solution using Newton Raphson method and verified it
23
Tran (2002) analysed tensegrity structures considering three of the side ties composed of
a compliant and a noncompliant segment in series. In order to control the shape and the
studied how the lengths of the noncompliant segments are affected due to an external
wrench applied to the structure. He also analysed the new tensegrity structure by
replacing one or two side ties by struts and found the tensegrity behaviour remained
unchanged by connecting the struts to each other with ball and socket joints.
Motro et al. (1986) initiated structural dynamics related research on tensegrity structures.
Connelly and Whitely (1996) defined two concepts of rigidity for tensegrity frameworks
i.e. prestress stability and second-order rigidity. They also proved that in a plane
tensegrity framework, the vertices and bars form a strictly convex polygon with
additional cables across the interior and the overall structure is rigid if and only if it is
Using equations of motion for spatial trusses developed within the framework of three-
dimensional theory of elasticity for large deformation, Murakami (2001 a) derived a set
of equations for static and dynamic analysis of tensegrity structures. He studied the modal
analysis of a three-bar tensegrity module and a six-stage tensegrity beam by utilizing the
24
infinitesimal mechanism modes were found to increase in proportion to the square root of
deformation modes with nonzero elastic force were observed with increasing pre-stress
amplitudes.
Further, Murakami (2001 b) presented static analysis of cyclic tensegrity modules using
shape and stiffness in prestressed condition and sensitivity of the initial geometrical
a class of tensegrity structures with Maxwell number less than zero. For a pre-stressed
mechanisms are isotropically stiffened at each node by a single pre-stress mode. Both
mechanism modes. For analysis of static and dynamic responses, he emphasized the need
damping in elastic cables and a much slower rate of decay of amplitude of vibration due
suggested that to control effective damping, augmenting the natural damping would be
inefficient. By ensuring damping associated with angular motion between the structural
25
elements, they found a natural mode of damping leading to linearly damped equations
analytical form and examined the nonlinear vibrations of a three bar tensegrity structure.
They examined the vibration and the nonlinear relation between torque and rotation in the
direction of the infinitesimal flex. They observed the presence of additional frictional
effects at the joints in real systems ensuring the system to return to the stable equilibrium
Murkami and Nishimura (2001 c) presented a set of procedures for characterizing static
and dynamic response of a six-bar tensegrity module and a two-stage tensegrity module
with three bars at each stage. The procedure was used to compute the Maxwell number,
to find the initial shape, to determine an admissible prestress mode, to carry out modal
space. The singular value decomposition of the initial equilibrium matrix revealed
prestress and infinitesimal mechanism modes. The prestress stiffening effect of the
Skeleton et al. (2001) developed an analytical model of the nonlinear dynamics of a large
elastic cables. The kinematics was described by positions and velocities of the ends of the
26
rigid rods to avoid angular velocities of each rod. The model was intended for shape
control and design of deployable structures and explicit analytical expressions were
derived to study stable equilibrium and controllability. They classified the tensegrity
structures as Class I and Class II. In Class I structures, there is no contact between the
rods and a stable equilibrium is achieved by pretension in the tendons. However, in class
II structures, rods are in contact at nodal points through a ball joint, although transmitting
no torques. Further, they developed the exact nonlinear equations for a Class 1 tensegrity
shell, having nm rigid rods and n (10m-2) tendons, subject to the assumption that the
tendons are linearly elastic and rigid rods of constant length. Finally, they found that the
columns, beams, plates, and shells) is that the elements of the tensegrity structure are
subjected only to axial loads. All members were found to be uni-directionally loaded due
to prestressing and no reversal in the direction of the load carried by the member. Hence,
a host of nonlinear problems known to create difficulties in control (i.e. hysterics, dead
Sultan et al. (2002 a) derived nonlinear equations of motion for tensegrity structures
using the Lagrange methodology under very general modelling assumptions. This
resulted in a finite set of second order ordinary differential equations. For some tensegrity
structures, under symmetrical motions, these equations reduced to a smaller set. They
27
Sultan et al. (2002 b) also derived linear models describing the approximate dynamics of
configurations. At these configurations, the stiffness matrix is linear and the control
matrix quadratic in the level of pretension. The mass and stiffness matrices are positive
definite as the damping matrix is positive semidefinite and hence these reference
solutions stable. They also found an increase in the modal dynamic range with the
pretension and also an increase in the modal damping range with damping.
Williams et al. (2003) derived the necessary and sufficient conditions for equilibrium of
terms of the member force coefficients and the string and the bar connectivity
functions, vectors were used to describe each element. By enlarging the vector space, the
linear algebra problems. Their results characterize the equilibrium conditions of the
tensegrity structures in terms of a very small number of variables and could be used for
Defossez (2003) reported the shape memory effect of symmetric tensegrity structures and
their use as actuators. It was found that change in the original length of elastic elements
causes the variation of the ratio of potential energy between two equilibrium states. As
tensegrity models show the tendency to go back to the more stable state when perturbed
28
by external forces, they exhibit a shape memory effect and thus can be used as actuators.
One of the main advantages of tensegrity structures is that their elastic tensile
components provide excellent opportunities for sensing functions and also act as
structural elements.
Lazzari et al. (2003) analyzed the tensegrity based structure at La Plata Stadium in
Argentina under the effects of wind by using the geometrically non-linear finite element
procedure named “Loki”, developed according to the total Lagrangian formulation. They
considered wind action as follower loading i.e. the case of body-attached pressure load,
where only the direction depends on the deformation within each time step. The structure
was analyzed under static loads; under free vibration and under a general load history i.e.
Domer et al. (2003b) studied two methods i.e. simulated annealing and a new algorithm,
actuator effects. For high and low required accuracies, PGSL provided good solutions
and the simulated annealing offered better results for the intermediate cases.
Sultan and Skeleton (2004) designed a smart sensor consisting of a tensegrity structure
composed of 18 elastic tendons, six rigid bars, and a rigid top to measure three
orthogonal forces and three orthogonal torques. They used the pretension coefficient as a
tuning factor for the static characteristics and found their shape change from a very
29
nonlinear to an approximately linear one due to increase in the pretension. Dynamic
characteristics were tuned through friction and pretension coefficient arrangements. They
observed decrease in the structural natural frequency with increase in friction and
transformation of the low frequency oscillatory mode into pure exponential decaying
ones. Further, they also designed an optimal estimator based on the linearized
of 30 rods and 90 cables and found two non-linear effects when the structure was close to
its integrity limit before collapse. The first one was the use of tensegrity structure as
the magnitude of the applied force. The second one was an increase in the resilience of a
slightly prestressed tensegrity structure due to an applied force more than the prestressed
equivalent structure. This paradoxical stiffening effect due to increase in prestress may
Tensegrity structures are good from serviceability criteria point of view due to their
serviceability criteria includes design and construction of an active control system for a
of a control strategy for nonlinear, coupled and under actuated systems that have no
closed form solutions for direct calculation of commands. Fest et al. (2004) described the
active prototype tensegrity structure including the actuators and sensors embedded in a
30
closed control loop. Based on stochastic search, they proposed and evaluated a quasistatic
Active tensegrity structures are good candidates for implementing active structural
control. With the number of actuators simulated for all possible control commands and
tested against all constraints, computational time increases exponentially for objective
function. Domer and Smith (2005) studied a unique tensegrity structure containing
They reduced the search time by stochastic search methods and found increase in
reasoning system.
Masic and Skeleton (2006) investigated the optimal distribution of the prestress in a
controlled tensegrity structure by jointly designing the structure and its control. They
used gradient based algorithm to select prestress and observed a monotonic decrease of
the objective function inside a feasible region in terms of the extreme direction of the
prestress cone.
The process through which the structure changes from one equilibrium configuration into
apparent that these structures are capable of large displacements and can easily change
31
their shape. They can also be built without, or with very few, complicated bar to bar
joints. These structures are very promising deployable structures due to packaging
Furuya (1992) examined deployment of tensegrity structures, but only at the conceptual
levels. Hanaor (1993) obtained deployability and prestress by extending bars in case of
DLTGs. He fabricated a seven unit domical model consisting of pyramidal units where
copper tubes were used as the outer struts, wooden dowels as the inner struts and a
moulded silicon rubber head as the seal between the two. He also fabricated a three unit
flat model consisting of T prisms with telescopic steel tubes with O-ring seals as struts
and deployment was achieved by an air compressor. He also found that the influence of
prestress level was quite large on the stiffness of geometrically flexible grids and small in
case of geometrically rigid configurations. Sultan and Skelton (1998) analysed the
deployment of tensegrity structures using tendon control and proposed a procedure for
Archimedean antiprism with the top platform rotated 90 degrees and developed generic
design equations for self-deployable systems. Assuming the top and the bottom platforms
parallel to each other and equal lengths of the ties composing a platform, the static
analysis was performed on 3, 4, 5 and 6-strut tensegrity systems. Each tensegrity system
was studied at the minimum limit position and a coordinate system was established at the
32
centre of the bottom platform. He assumed symmetry of forces within the structure
(based on the symmetry of the geometry) and used the theory of elasticity to develop
generic design equations to calculate lengths of the struts and the cables needed to obtain
Hanaor and Levy (2001) reviewed various structural systems available for deployable
space enclosures and evaluated them according to their structural efficiency, technical
dismountable structures. Deployable structures are assembled in the stowed state and
involve no component assembly during the erection process, whereas the dismountable
structures with large aperture, so that they can be easily packaged in a small envelope.
Tibert and Pellegrino (2002) proposed a deployable tensegrity prism forming a ring
structure using two identical cable nets (front and rear) interconnected by tension ties,
and a reflecting mesh was attached to the front net. The geometric configuration of the
structure was optimized by rotating the cable nets about the axis of the reflector to reduce
Tibert and Pellegrino (2003) presented a complete design of a deployable tensegrity mast,
including initial form finding, structural analysis, manufacturing and deployment. The
33
deployment of the mast was achieved using self-locking hinges and the constructing
cables, forming the outer envelope of mast, by two-dimensional weaving. Further, they
reported the weak bending stiffness of the mast in comparison to an articulated truss.
Quirant et al. (2003) studied the various parameters i.e. level of selfstress, distribution of
selfstress and rigidity ratios between strut and cable elements for controlling the
behaviour of the tensegrity systems and their design. Considering different parameters,
they formulated a design procedure according to Euro codes and more particularly, used
the partial safety factors to verify the ultimate limit states stability. The partial safety
factors were calculated considering the sensitivity of the stress of the elements to
manufacturing and the assembly accuracy. Further, they applied the procedure to a
Sultan and Skeleton (2003) developed a new deployment strategy for tensegrity
structures based on the assumption that the structure yields an initial equilibrium
configuration with all tendons in tension. They found that the control variable takes value
only in the set of the equilibrium manifold. The control variables were considered as the
length of the struts in case of telescopic struts, the lengths of the tendons in the case of
tendon control, or a combination of both. Further, they considered two examples, one
showing a time optimal deployment of a relatively simple structure, and the other the
34
For designers having a non-mathematical background in the creative form finding of
tensegrities, Sakantamis and Larsen (2004) developed Cocoon method i.e. a practical
physical modelling tool for designing tensegrity systems. This modelling involves
devising a membrane to be able to enclose the model to be built, inserting struts inside
the membrane one by one, until the required topology is obtained, superimposing linear
ties between the nodes of the struts and removing the membrane safely and transition
Using tubular based joint design with proper tolerance, Vu et al. (2005 & 2006) explored
the deployability of deployable strut tensioned structures explaining their design and
structural efficiency. Fu (2005) studied the structural behaviour and structural types of
tensegrity domes and proposed a design methodology for them by summarizing the
results obtained through non-linear software using the numerical method. Zhang and
Ohsaki (2006) derived the stability conditions for tensegrity structures based on positive
definiteness of the tangent stiffness matrix (i.e. sum of the linear and geometrical
stiffness matrices). For the force density matrix positive semi definite having minimum
rank deficiency d+1 and rank of geometry matrix being d(d+1)/2, they found a stable
Masic et al. (2006) optimized the design of tensegrity structures through a systematic
buckling constraints of bars and shape constraints. The static response was obtained by
35
using the non linear large displacement model through a software package for sparse
1. 3 RESEARCH OBJECTIVES
The qualities of tensegrity structures, which make the technology attractive for human
use, are their resilience and ability to use materials in a very economical way. The
ethereal tensile members have predominant role to play, while the more material
found in the literature on theoretical aspects of tensegrity structures, such as form finding
deployable tower structures. The research on tensegrty grid structure for roof applications
is very limited. The design of tensegrity structure is complex and iterative process.
artificial neural networks and their monitoring using advance sensors. The main
cuboctahedron configuration
36
3. Detailed experimental investigations on dismountable tensegrity structure i.e. to
findings
This thesis comprises seven chapters including introduction and literature review in
Chapter 1. Static and dynamic analysis, design methods and deployment procedure of
Chapter 2 presents the development of single dismountable tensegrity module in strut and
cable mode of arrangement. The structure is instrumented using sensors to obtain the
prestress level and forces in members. The deflection in top nodes is also measured using
LVDTs. The experimental results are compared with the numerical results as obtained by
finite element modelling using ANSYS. To minimize the variation of the results, the
numerical model of the single module tensegrity structure is updated to match the
experimental values with the numerical values. The allowable strengths of the members
are compared using Indian code of practice. Further, the updated model is used for
37
parametric study i.e. effect of prestress level, rigidity ratio on maximum deflection and
tensegrity grid structure are described. This grid structure is different from other
tensegrity grids available in literature in fabrication as the entire grid is being constructed
as a single cohesive unit instead of joining the individual modules. From experiment, the
prestress level, the deflection of the bottom nodes and the forces in various members are
obtained. The experimental results are compared with those of the updated numerical
model. Parametric study is carried out on different tensegrity grid sizes i.e.4m×4m,
6m×6m and 8m×8m. The effect of prestress level, height of the structure and effect of
supports on maximum deflection, maximum strut force, cable forces and load carrying
Chapter 4 describes the behaviour of tensegrity grid structure under thermal loads, lack of
fit and the combination of different loads. The grid structure is also analysed for wind
forces for Delhi region and a suitable grid size is recommended for practical use for the
explored. The data are generated by numerical analysis for different grid sizes with
different heights, rigidity ratios, loads and support conditions. Using feed forward
multilayerd back propagation, the network is trained for different grid sizes to obtain the
38
cross sectional area of the tension members i.e. cables and also tested for a grid size
In Chapter 6, different types of sensors i.e. piezo sensors and strain gauges for monitoring
of structures have been critically reviewed. Piezo sensors have been applied to tensegrity
Chapter 7 interprets the results briefly and highlights their importance for tensegrity
structures. Finally, the conclusions and future scope of this research are presented.
The next chapter describes the development, analysis and testing of a single module
tensegrity prototype.
39
CHAPTER 2
2.1 INTRODUCTION
structures, both in strut and cable mode, followed by their instrumentation and destructive
testing. In the strut mode, the closing and opening mechanism is devised in one of the
struts of the structure for easy assembling and dismantling by closing/opening the strut.
Similarly, in cable mode, the closing and opening mechanism is devised in one of the top
cable through a turnbuckle by loosening which the structure can be dismantled. A single
configuration using galvanised iron (GI) pipes as struts and high tensile stranded cables
as tension elements. Detailed instrumentation is carried out right at the fabrication stage,
consisting of strain gauges and linear variable differential transducers (LVDTs). The
structure is subjected to destructive load test during which continuous monitoring of the
prestress levels, deflections and strains in the struts and cables are carried out. The
monitored structure is also analysed using finite element method (FEM) and the
numerical results compared with the experimental observations. The numerical model is
updated to match the experimental results and the updated model is considered for further
parametric study.
40
2.2 MATERIAL CHARACTERIZATION
It is essential to determine the material properties i.e. the Young’s modulus (E), yield
strength and the ultimate strength of both the tension members (cables) and the
compression members (struts) before they can be employed for fabrication. In this study,
galvanized iron (GI) pipes of medium type, commercially available in India conforming
to the Indian Standard IS 1239-I (1990), were used as compression members. Table 2.1
presents the key specifications of the pipes as per IS 1239-I (1990). 4 mm nominal
diameter mild steel stranded wires of 6 x 19, commercially available in the market
confirming to IS 3459 (1977), were used as tensile members. Table 2.2 presents key
PARAMETER VALUE
Thickness 2.6mm
41
Table 2. 2 Properties of stranded wire as per IS 3459
PARAMETER VALUE
Nominal diameter 4 mm
Type Round
The GI pipes were tested in the universal testing machine (UTM) of 1000 kN capacity as
shown in Fig. 2.1. Tension test was carried out rather than compression test for ease of
performance. Four strain gauges of 5mm gauge length, manufactured by Tokyo Sokki
Kenkyujo Company Limited (TML, 2005), were surface bonded in the middle portion
along the circumference of the GI pipe using Cyanoacrylate (CN) adhesive. The strain
gauges were connected to the data logger card fixed on a personal computer for
automatically recording the strains using strain smart data system, version 3.1. Average
strain was considered for determining the Young’s modulus of the pipe. The internal and
external diameters of four pipes were measured as 15.9mm and 21.375 mm respectively
on an average, resulting in a cross sectional area of 160.284mm2. Three pipes were tested
and the stress strain curves are shown in Fig. 2.2 The average value of Young’s modulus
was computed as 2.05 × 105 N/mm2. The average breaking load was found to be 65.727
42
kN and the corresponding ultimate stress was 410 N/mm2. Hence, the GI pipes conform
to 240 grade as per IS 1161 (1998) and the yield stress is 240 N/mm2. Typical failure
The stranded wires used for the reticulated cable network were made of 0.25 mm
galvanized high carbon steel wires, conforming to Indian Standard IS 1835 (1976). A
6 × 19 stranded wire consisted six strands with nineteen wires in each strand surrounding
a steel core. The net sectional area was found as 6.53 mm2. The wires were tested in a
UTM of 250 KN capacity as shown in Fig. 2.4. Three samples of stranded wires were
tested and the stress strain plot is shown in Fig. 2.5.The Young’s modulus of the stranded
wire was found to be 0.954 × 105 N/mm2. The average failure load was found to be 9660
N. The average yield stress and 0.2% proof stress were found to be 1421.335 N/mm2 and
1119.575 N/mm2 respectively. The failure pattern of the wires is shown in Fig. 2. 6.
43
400
350
300 Sample 1
250
Stress (N/mm2)
200 Sample 3
150
100 Sample 2
50
0
0 0.0005 0.001 0.0015 0.002 0.0025
Strain
Figure 2.2 Stress strain curve of GI pipes
44
250 kN universal testing machine Wedge grip with specimen
45
1600
1400
Sample1
1200
Stress (N/mm2)
1000
800
600
Sample 2
400
200 Sample3
0
0 0.5 1 1.5 2 2.5
Strain in %
Failure location
46
2.3 FABRICATION AND TESTING OF TENSEGRITY MODULE
Fig. 2.7 shows the perspective view and the top view of the tensegrity structure - a half-
tensegrity modules were fabricated using the GI pipes and the 6 × 19 stranded wires tested
previously. The length of the bottom cables was 1m and that of top and side cables was
0.707m. The struts were 1.224m in length. All the lengths were measured from centre to
centre of joints. At each joint, 12 mm eyebolts were used for connecting the cables and
the strut member. The cables were tightened using ferrule A × 4.5 (IS 10942) by means
of hydraulic press in such a manner that there was no slip in the cables. The ends of the
pipes were plugged up to 50mm length and then 12 mm hole was drilled so as to make
easy for pipes to be put in tight position at the joints. Two structures were fabricated: one
in the strut mode and the other in the cable mode, as described below
4 7
7(0.5,1 ,0.5)
6(0,1,0) 8(1,1,0)
2 5
5(1,0.5,0.
4(0,0.5,0.5)
1 8 3(1,0,0)
1(0,0,0)
2(0.5,0,0.
5)
3
(a (b)
)
Figure 2.7 Schematic view of tensegrity structure module developed
(a) Perspective (b) Top view (Spatial coordinates are shown in brackets)
47
2.3.1 Strut Mode of Arrangement
A special mechanism was made in one strut for assembling and dismantling of the
structure in the strut mode. This pipe was made in three parts. Two end parts were of 15.9
inner diameter mm G.I pipes, having lengths of 47 cm and 27.5 cm respectively and the
middle third part was of 25 mm inner diameter GI pipe having 30 cm length. A solid rod
of 20 cm length was welded to the pipe of 27.5 cm length with 15 cm projecting outside
the pipe, which was threaded. At one end of the 25 mm middle pipe, 25 mm socket was
fitted and a bush of 38mm was attached so that the threaded portion of the rod could
move inside and outside for length adjustment. The other end of the 25 mm pipe was
welded with a solid iron rod of 35 mm diameter and 50 mm length, and a 16 mm hole
was made so that 15.9 mm diameter pipe can be easily fitted to it. Since length of the
strut is adjusted to achieve prestress, this arrangement can be called as the ‘strut mode’.
The different stages of the constructed half-cuboctahedron prototype in the strut mode are
shown in Fig. 2.8. Fig. 2.9 shows the half-cuboctahedron in self stressed position after
For the cable mode of arrangement, all the struts were chosen 1.224 m in length. The
opening and closing mechanism was devised in one top cable by means of a turnbuckle.
This cable had three parts i.e. two small cables connected by a turnbuckle. From one end,
cable length of 17 cm and other end of 23 cm were connected to the turnbuckle in the
shown in Fig. 2.10. (b), was made on one of the top cables, so that the three parts can be
48
connected easily after placing the struts in the required position. After keeping the struts
in required position, the length of that top cable was 10 cm longer than the required
length. The length was reduced by tightening the turnbuckle. Once the length of
adjustable top cable became equal to 0.707m, the structure attained self-stressed
equilibrium. This arrangement can be called as ‘cable mode’. The stepwise erection is
In order to do analysis using finite element method (FEM), it is essential to obtain the
prestress level in each cable and strut in the equilibrium configuration. To find the forces
in the GI pipes and the cables, either strain gauge can be bonded or extensometers can be
attached. Since extensometers are expensive, in the present case, 5mm strain gauges
manufactured by TML and confirming to product FLA-5-11 (TML, 2005) were surface
bonded using CN adhesive. For accurate results, four strain gauges were bonded on the
strain in stranded cables is a difficult task due to small diameter. After several trials, this
appropriate for this type of stranded wire. Two each of these 2mm strain gauges were
bonded on the top and the bottom cables and one on the leg cable of the strut mode
tensegrity module. In the cable mode, four 5mm strain gauges were bonded in the GI
pipes in similar manner. The total numbers of strain gauges used were 21 and 40 in the
strut and the cable mode of arrangement respectively. In both the cases, two linear
variable displacement transducers (LVDT) were fitted below two nodes to obtain the
49
deflection in vertical direction. All the sensors were connected to the data logger
connected to the computer having strain smart data system, version 3.1 for monitoring the
(c) (d)
50
Strut
Cable
Eyebolt
i d
51
Turule
(a) (b)
(c)
52
In both the cases, the measurements were recorded at the prestress levels i.e. self-stressed
equilibrium position. The average prestress forces in the top cable, bottom cable, and the
strut were found to be 1.33 kN, 0.81 kN and 2.49 kN respectively in strut mode. The
average prestress force in the strut was found to be 2.58 kN in cable mode of deployment.
The load was gradually applied till failure. A typical loading condition is shown in Fig.
2.12. In the strut mode, the failure of the structure occurred at a load of 1.86 kN, the
failure mode is shown in Fig. 2.13. It is observed that failure occurred due to buckling of
the middle threaded portion of the special strut, which controlled erection. In the cable
mode, the failure occurred at a load of 3.25 kN and the failure pattern is shown in Fig.
2.14. This resulted from the buckling of a strut and occurred at much higher loads as
53
Concrete cubes Iron plate
54
Failure of welded rod connected to pipe
55
Failure of GI pipe by buckling
Figure 2.14 Failure of test structure in cable mode of arrangement
The detailed procedure for large deflection analysis of prestressed cable networks using
matrix displacement approach has been described by Argyris and Scharpf (1972). The
behaviour of tensegrity prototypes tested (as described in the previous sections) was
modelled using FEM using ANSYS 9. A characteristic feature of the tensegrity structures
under the applied loads. That is, the stiffness matrix [K] is a function of the
56
displacement {u}. The stiffness matrix changes because the shape changes and/ or the
members rotate. In general, there are four types of geometric nonlinearities: large strains,
large rotations, stress stiffening and spin softening (Cook et al. 2003).
In tensegrity structures, stress stiffening is more prominent than the other types of non-
linearitis. In this type of nonlinearity, both strains and rotations are small. Stress
stiffening, also called geometric stiffening; incremental stiffening, initial stress stiffening
due to its stress state. This stiffening effect needs to be considered for thin structures
such as cables, thin beams and shells that have the bending stiffness vary small compared
to the axial stiffness. In such structures, the in-plane and transverse displacements are
coupled. This effect also augments the regular nonlinear stiffness matrix produced by
large strain or large deflection effects. Generating and then using additional stiffness
matrix called stress stiffness matrix accounts for the effect of stress stiffening. It may be
The stress stiffness matrix is computed, based on the stress state of the previous
equilibrium iteration. Thus, to generate valid stress stiffened problem, at least two
iterations are normally required, where the first iteration is used to determine the stress
stiffness matrix of the second iteration. If this additional stiffness affects the stresses,
57
The strain-displacement equations for the general motion of a differential length fibre are
derived below. Two different results have been obtained and both discussed. Consider the
motion of a differential fibre, originally dS, and then ds after deformation, as shown in
Fig. 2.15(a). One end moves {U }, and the other end moves {U + dU } . The motion of one
end, with the rigid body translation removed, is {U + dU } − {U } = {dU }, which may be
expanded as
⎧du⎫
{dU} = ⎪⎨dv⎪⎬ (2.1)
⎪dw⎪
⎩ ⎭
where u is the displacement parallel to the original orientation of the fibre, as shown in
Fig 2.15(b). Note that X, Y, Z represent the global Cartesian axes, and x, y, z represent
axes based on the original orientation of the fibre. By the Pythagorean theorem,
ds = ( dS + du ) + (dv) 2 + (dw) 2
2
(2.2)
2 2 2
ds ⎛ du ⎞ ⎛ dv ⎞ ⎛ dw ⎞
A= = ⎜1 + ⎟ +⎜ ⎟ +⎜ ⎟ (2.3)
dS ⎝ dS ⎠ ⎝ dS ⎠ ⎝ dS ⎠
2 2 2
⎛ du ⎞ ⎛ dv ⎞ ⎛ dw ⎞
A = ⎜1 + ⎟ + ⎜ ⎟ + ⎜ ⎟ (2.4)
⎝ dx ⎠ ⎝ dx ⎠ ⎝ dx ⎠
∂u ⎛ ∂u ⎞ ⎛ ∂v ⎞ ⎛ ∂w ⎞
2 2 2
A = 1+ 2 +⎜ ⎟ +⎜ ⎟ +⎜ ⎟ (2.5)
∂x ⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎝ ∂x ⎠
58
A A 2 A3
1+ A = 1+ − + − ... (2.6)
2 8 16
U+dU
dS
ds
U
(a)
X
Z
dv
Y
dw X
du
x
dU y
dS
ds
z
(b)
59
where A2 < 1, one should be aware that using a limited number of terms of this series may
restrict its applicability to small rotation and small strains. If the first two terms of the
∂u 1 ⎡⎛ ∂u ⎞ ⎛ ∂v ⎞ ⎛ ∂w ⎞ ⎤
2 2 2
A = 1+ + ⎢⎜ ⎟ + ⎜ ⎟ + ⎜ ⎟ ⎥ (2.7)
∂x 2 ⎢⎣⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎥⎦
The resultant strain (same as extension, since strains are assumed to be small) is then
∂u 1 ⎡⎛ ∂u ⎞ ⎛ ∂v ⎞ ⎛ ∂w ⎞ ⎤
2 2 2
ε x = A −1 = + ⎢⎜ ⎟ + ⎜ ⎟ + ⎜ ⎟ ⎥ (2.8)
∂x 2 ⎢⎣⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎦⎥
If, more accurately, the first three terms of Equation (2.6) are used and displacement
derivatives of third order and above are dropped, Equation (2.7) reduces to
∂u 1 ⎡⎛ ∂v ⎞ ⎛ ∂w ⎞ ⎤
2 2
A = 1+ + ⎢⎜ ⎟ + ⎜ ⎟ ⎥ (2.9)
∂x 2 ⎢⎣⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎥⎦
∂u 1 ⎡⎛ ∂v ⎞ ⎛ ∂w ⎞ ⎤
2 2
ε x = A −1 = + ⎢⎜ ⎟ + ⎜ ⎟ ⎥ (2.10)
∂x 2 ⎢⎣⎝ ∂x ⎠ ⎝ ∂x ⎠ ⎥⎦
For 1-D structures, Equation (2.10) is used for its greater accuracy and causes no
60
For a spar element such as link 8 (ANSYS, 2004), the stress stiffness matrix is derived, as
⎡0 0 0 0 0 0⎤
⎢0 1 0 0 − 1 0 ⎥⎥
⎢
⎢0 0 1 0 0 − 1⎥
[S l ] = F ⎢ ⎥ (2.11)
L ⎢0 0 0 0 0 0⎥
⎢0 − 1 0 0 1 0⎥
⎢ ⎥
⎣⎢0 0 − 1 0 0 1 ⎦⎥
The element matrices and load vectors are derived using an updated Lagrangian
⎡ − ⎤
⎢⎣ K i ⎥⎦ ΔU i = { F } − { Fi }
app nr
(2.12)
where {F app } = Vector of externally applied nodal point loads at time t+Δt
⎢⎣ ⎥⎦
⎡ − ⎤
⎢⎣ K i ⎥⎦ = [K i ] + [S i ] (2.13)
[K i ] = ∫ [B i ] [D i ][B i ]d (vol )
T
(2.14)
[Bi ] is the strain displacement matrix in terms of the current geometry {x n } and [Di ] is
the current stress strain matrix. [S i ] is the stress stiffness or geometric stiffness
61
[S i ] = ∫ [G i ] [τ i ][G i ]d (vol )
T
(2.15)
where [Gi ] is a matrix of shape function derivatives and [τ i ] is a matrix of the current
In the present case, the finite element analysis of the tensegrity structure was performed
using ANSYS 9 (ANSYS, 2004). All the cable and strut elements were considered as 3D
spar elements, having three degrees of freedom in translation at each node, which were
considered at the ends of the elements. The material was assumed as linear, elastic and
isotropic. Fig. 2.16 shows the model generated using the preprocessor of ANSYS 9. The
degrees of freedom were locked in vertical direction only for bottom nodes i.e. 1, 3, 6 and
8. The Young’s modulus values of 2.05×105 N/mm2 and 0.952 ×105 N/mm2 were used
for struts and cable element respectively, as obtained experimentally. The prestress force
determine the initial strain in the elements. As experimental data for all the elements were
not available, the equations developed from static analysis of n-strut tensegrity system by
Stern (1999) were used to obtain the prestress force in other elements.
F F
t
= s
(2.16)
L t L s
62
where Fa is the force in the top cable, Fb in the bottom cable, Ft in the leg ties, Fs in the
strut. a is the length of top cable, b the bottom cable, Lt the leg tie and L s the strut.
Further,
2 Ls ⎛π ⎞
Fs = Fa sin⎜ ⎟ (2.18)
b ⎝n⎠
2 Lt ⎛π ⎞
Ft = Fa sin ⎜ ⎟ (2.19)
b ⎝n⎠
Using above equations, the relationships for the half cuboctahedron configuration can be
found considering a=0.707m, b=1.0 m, Lt =0. 707m and L s =1.224m. Hence, all forces
Fa = 0.578 Fs (2.20)
F b = 0 . 707 F a (2.21)
F a = F t (2.22)
As per above relationships, the equilibrium prestress forces in top and bottom cable were
calculated as 1.44 kN and 1.02 kN respectively in strut mode, considering strut force i.e.
2.49 kN as reference. Similarly, the average strut force of 2.58 kN was considered as the
reference for cable mode of deployment. The model was simulated with external loads as
applied on the structure during the experiment. The load was distributed equally among
all top nodes as concentrated loads. The large deformation effect for static analysis was
considered.
63
Top cable
Loads
7
4 Struts
Leg 5
cable 2
6 Leg
cable
8
Support
1
3
Bottom cable
Fig. 2.17 shows the deflection at nodes 5 and 7 obtained from the experiment, in the strut
mode of arrangement. The deflection curve indicates nonlinear variation and the stiffness
increases with increase in the loading. The variation of the deflection between two nodes
lies within 4 %. The experimental (average) and numerical deflections are compared in
Fig. 2.18. The trend remains same in both the cases but the numerical deflections are
somewhat higher than the experimental values. The experimental average values are
64
nearly 60% and 25 % less than numerical values at 60N and 1540N respectively, which
In the strut mode of arrangement, the forces obtained in the struts by measuring strains
using strain gauges are compared with the numerical forces in Fig. 2.19. The pattern of
variation of the strut force with increase in load remains same. The individual forces vary
between 50 to 70% with each other. Fig. 2.20 compares the average strut force with that
of the numerically obtained force. At smaller values of the applied load, the experimental
value is only marginally higher than the numerical value. However, with increase in the
2000
Node 5
1500 Node 7
Load (N)
Average
1000
500
0
0 10 20 30 40 50 60
Deflection (mm)
65
2000
Experimental
1500 average
Load (N) Numerical
1000
500
0
0 10 20 30 40 50 60 70
Deflection (mm)
Applied load in N
-2000 Strut 3
Strut 2
-3000
-4000
Average
-5000
Strut 1
-6000
-7000
Applied load in N
66
Applied load in N
-2000
0 500 1000 1500 2000
-3000
Strut force (N)
-4000
experimental
-5000 average
-6000 Numerical
-7000
Fig. 2.21 shows the variation of the average force in the top cables in the strut mode. The
average force in the top cables obtained experimentally is lower than the numerical value
and the difference increases with increase in the load. The difference is nearly 10 % at the
prestress level and 53% at 1800N. In addition, the slope of the curve is much steeper in
case of the numerical results. The force in the bottom cable in the strut is compared in
Fig. 2.22. The figure indicates that the force in the bottom cable obtained experimentally
is marginally higher than the numerical value at the initial loading. However, the
numerical value becomes higher with increase in the applied load and the deviation is
67
3500
3000 Experimental
3000
Bottom cable force (N )
2500 Experimental
2000 Numerical
1500
1000
500
0
0 500 1000 1500 2000
Applied load (N)
68
In the cable mode of arrangement, the experimental deflections at nodes 4 and 7 and their
average values are plotted in Fig. 2.23. The variation in the individual values is very
small, less than 1%. The comparison of the average experimental and the numerical
values is shown in Fig. 2.24. The numerical plot follows the trend of the experimental
plot. However, the experimental values remain less than the numerical value for all the
loads. At initial loading, the experimental deflection is nearly 22% less than the
numerical value, but with increase in the load, the difference becomes less than 10%. Fig.
2.25 shows the force variation in the struts measured experimentally in the cable mode.
The force variation between the individual struts varies in the range of 1% to 70%.
However, the trend remains more or less the same. Comparison of the average
experimental strut force and the corresponding numerical values is shown in Fig. 2.26.
The numerical strut force is marginally less than the experimental forces.
2000
1800
node 4
1600
node 7
1400
average
Load in N
1200
1000
800
600
400
200
0
0 10 20 30 40 50 60 70
Deflection in mm
Figure 2.23 Experimental deflections at nodes 4 & 7 in the
cable mode arrangement
69
2000
Experimental
Load (N) 1500 average
Numerical
1000
500
0
0 20 40 60 80
Deflection (mm)
-4000
-6000
-8000 Strut 1
-10000 Strut 4
Average
-12000 Strut 3
-14000
70
Applied load in N
-1000
-6000
-7000
The structure started to fail at a load of 2.45 kN and at this load, the force in the strut was
measured as 8.24 kN. The structure finally collapsed at a load of 3.25 kN. Considering
the actual slenderness ratio of the struts (with pin-pin connection at ends); the
compression capacity of a strut was found to be 5.145 kN, much less than that actually
observed. This indicates the effective length has been much less and partial fixity at the
ends developed with increase in load. The end conditions tend to be close to fixed
boundary near failure. This causes the compression members to carry more loads than
expected.
Brownjohn et al. (2003) upgraded a highway bridge by dynamic testing and finite model
updating. The governing criterion for design of tensegrity structure is the maximum
deflection of the structure due to external loading. Although the behaviour of a tensegrity
71
structure can be reasonably well modeled using FEM, as demonstrated in the previous
section, small discrepancies still exist. For example, the numerical model tends to
overestimate the deflections, as evident from Fig. 2.18 and 2.24. Hence, an updating of
the numerical model is necessary so that it produces results as close as possible to the
actual experiment. An attempt has been made to update the model in both the strut and
the cable mode by trial and four cases are considered as discussed below.
In the first case, the numerical model was analysed considering all tension members of
the structures as two noded 3D spar elements with three translational degrees of freedom
at each node. The compression members were assumed as two noded 3D elastic beam
elements with six degrees of freedom (i.e. three translations and three rotations) at each
node. Joints were assumed to be pin jointed i.e. friction at the joints is negligible. The
degrees of freedom at bottom support nodes were fixed in the vertical direction only. For
the second case, all conditions of first case remained same except the degrees of freedom
at the bottom support nodes, all the translational degrees of freedom are locked.
The third case was the model considered before in section 2.6. In the fourth case, all the
translational degrees of freedom were restrained keeping other conditions same as the
third case. FEM analyses of all the above cases were carried out using ANSYS 9 for a
total load of 1.86 kN. The numerical values of deflection and strut force with
experimental values were compared for both the strut and the cable mode of arrangement.
72
Fig. 2.27 shows comparison of the deflection for different stages of model updating in the
strut mode of arrangement. It is clear from the graph that the numerical value is closely
matched with first case i.e. tension members as spar elements and compression members
as beam elements, with the degrees of freedoms locked in vertical direction only at the
supports. In other cases, the difference is much larger. Comparison of the strut force for
different stages of model updating in strut mode is shown in Fig. 2.28. The strut force for
the first and the second case is nearly same as observed from the comparison. Up to 1 kN
load, the experimental values are marginally higher than the numerical values and less
than the numerical values beyond 1 kN. However, the first case is more suitable as the
updated model in strut mode of arrangement since the numerical deflection in first case is
1800
1600 case 1
1400 case 2
1200 case 3
Load (N)
1000
case 4
800
Expt.
600
400
200
0
0 10 20 30 40 50 60 70
Deflection (mm)
73
Applied load in N
-1000
0 500 1000 1500 2000
-2000
case 1
Strut force (N)
-3000
case 2
-4000 case 3
case 4
-5000
Expt.
-6000
-7000
The model updating fr0m the point of view of deflections in the cable mode of
arrangement is shown in Fig. 2.29. In this comparison, it is found that numerical values
of fourth case are closest to experimental results. Fig. 2.30 shows the comparison of the
strut forces for different stages of model updating. The experimental strut forces are
comparable with the numerical values obtained for fourth case. Hence, the fourth case
where both struts and cables were spar elements and all the degrees of translations were
74
2000
1800 case 1
1600 case 2
1400 case 3
Load (N) 1200 case 4
1000 expt.
800
600
400
200
0
0 20 40 60 80
Deflection (mm)
Applied load in N
-2000
0 500 1000 1500 2000
-3000
Strut force (N)
-4000 case 1
case 2
-5000
case 3
-6000 case 4
expt.
-7000
Fig. 2.31 shows the variation of the deflection of the updated model with the
experimental deflection in strut mode of arrangement. The results are closely matched
75
and the maximum difference is within 10%. The experimental strut force is compared
with the updated model in strut mode in Fig. 2.32. At 1.86 kN, the maximum deviation of
the numerical force is 15% as compared to the experimental strut force. The numerical
values of bottom and top cable forces are compared in Fig. 2.33 and Fig. 2.34
respectively. In both the cases, the trends are same, though there are minor variations in
the magnitudes. However, all the results obtained by model updating show an
improvement over the initial analysis done in Section 2.6. Similarly, the updated model
results for deflection and strut force in the cable mode of arrangement are shown in Fig.
2.35 and Fig. 2.36 respectively. Again, the numerical values obtained for updated model
In comparison to the strut mode, the fabrication, assembling and dismantling is simpler in
cable mode. The load carrying capacity is also higher in the cable mode of arrangement.
Hence, further studies were carried out only for cable mode of arrangement.
2000
Experimental
1500 Updated model
Load (N)
1000
500
0
0 10 20 30 40 50 60
Deflection (mm)
76
Applied load in N
-2000
0 500 1000 1500 2000
-2500
Experimental
Strut force (N)
-4000
-4500
-5000
2500
Experimental
Bottom cable force (N)
2000
Updated model
1500
1000
500
0
0 500 1000 1500 2000
Applied load (N)
77
3000
Experimental
2500
2000
Experimental
1500
Model updated
Load (N)
1000
500
0
0 10 20 30 40 50 60 70
Deflection (mm)
78
Applied load in N
0
-1000 0 500 1000 1500 2000
Experimental
-2000
Strut froce (N)
Model updated
-3000
-4000
-5000
-6000
-7000
The experiment conducted for both the strut and the cable mode of arrangement shows
that the failure of structure is due to the failure of struts, caused by buckling. So, it is
l
λ= (2.23)
rmin
Where rmin is the radius of gyration and l the effective length of the strut. Taking the
effective length coefficient as 0.7, the effective length was computed as 0.857m (actual
79
The radius of gyration was determined from
I
rmin = (2.24)
A
where I is the moment of inertia of the strut and A the cross sectional area. From
Equations (2.23) and (2.24), the slenderness ratio of the strut was found to be 128.65.
Using Table 5.1 of IS: 800-1984, the permissible stress of the strut comes out to be
56N/mm2, which gives the compression capacity as 8.97 kN. From the numerical analysis
in cable mode of arrangement (using updated model), a strut force 5.91 kN was obtained
at failure.
Since the resultant force obtained numerically is less than the allowable value found from
codal provision, the struts are safe in buckling. However, the strut force corresponding to
failure was found as 12.77 kN from experiment in cable mode. It indicates that the actual
failure load is much higher than the permissible value of 8.97 kN. It shows that the
conditions at the ends of the strut are not pin-pin (i.e. effective length coefficient 0.7) as
assumed in the analysis and somewhat effective length coefficient is between 0.5 and 0.7.
From the experiments, no failure of any cable was observed during the test conducted on
the structure. However, the numerical cable force of the updated model is compared with
the permissible value. The permissible tensile stress in the cable is found experimentally
as 1119.575 N/mm2 (section 3.2). Thus, the maximum load that the cable can carry is
7.31 kN. Using the updated model, a force of 3.7 kN was observed at an total applied
80
load of 1.86 kN, which is much less than permissible value i.e. 7.31 kN. Hence, the cable
After updating the numerical model, parametric study was carried out for the cable mode
of arrangement. The controlling parameter for failure of the structure is the vertical
deflection and the strut force. The updated model is considered to study the effect of
prestress level and rigidity ratio on deflection and strut force. Rigidity ratio (r) depends
on cross sectional area of members and the Young’s modulus of the material.
Mathematically it is expressed as
As E s
r= (2.25)
Ac E c
In the parametric study, the prestres levels are considered as 1 kN, 1.5 kN and 2 kN. The
rigidity ratios are considered as 20, 30, 40 and 50. Keeping the strut cross sectional area
i.e. 160.284 mm2 constant, the cable cross sectional areas are found to be 17.257 mm2,
11.505 mm2, 8.629 mm2 and 6.903 mm2 for the rigidity ratio 20, 30, 40 and 50
respectively. The structure was analysed numerically for the different combinations of the
81
prestress levels and the rigidity ratios and the vertical deflection and strut forces were
The effect of the rigidity ratio on the deflection is shown in Fig. 2.37. Keeping the
prestress value as1.5 kN in the top cable, the rigidity ratio was varied from 50 to 20. It is
observed that with decrease in the rigidity ratio, the deflection decreases substantially at
higher loads. For example, the decrease in values between rigidity ratio 50 and 20 is
1.96%, 15.65% and 24.3% at load of 100N, 600N and 2500 N respectively. This is due to
increase in cable size keeping the strut size constant. Fig 2.38 shows the effect of rigidity
ratio on maximum strut force at a prestress level of 1.5 kN. With decrease in rigidity
ratio, the strut forces increases at higher loads. The percentage increase at different loads
between rigidity ratio 50 and 20 is shown in Fig. 2.39. After a load of 1500 N, the
increase is very marginal and the curve tends to be asymptotic to about 20%.
The effect of prestress on deflection is shown in Fig. 2.40. Keeping the rigidity ratio same
i.e. 50, the level of prestress is considered as 1000 N, 1500 N and 2000 N. With increase
in the prestress level, the deflection values tend to decrease. This decrease varies
nonlinearly, as shown in Fig. 2.41. The effect of the prestress level on the strut force is
shown in Fig. 2.42. The comparison shows that the strut force increases with increase in
the prestress level. However, the percentage increase becomes less as the applied load
increases. At lower loads, the difference is nearly 100% and at higher loads, it reduces to
8 %.
82
3000
2500
2000 r 20
Load (N)
r 30
1500 r 40
r 50
1000
500
0
0 20 40 60 80
Deflection (mm)
0
0 500 1000 1500 2000 2500 3000
-2000
Strut force (N)
-4000
-6000 r 20
r 30
-8000 r 40
r 50
-10000
Applied load (N)
Figure 2.38 Effect of rigidity ratio on strut force (Prestress = 1.5 kN)
83
25
% increase 20
15
10
5
0
0 500 1000 1500 2000 2500 3000
Applied load (N)
3000
2500
2000
Load (N)
prestress 1000
1500 prestress 1500
prestress 2000
1000
500
0
0 20 40 60 80
Deflection (mm)
84
60
50
% decrease
40
30
20
10
0
0 500 1000 1500 2000 2500 3000
Applied load (N)
0
-1000 0 500 1000 1500 2000 2500 3000
-2000
Strut force (N)
-3000
-4000
-5000
-6000
Prestress 1000
-7000 prestress 1500
-8000 prestress 2000
-9000
85
2.10 CONCLUDING REMARKS
based on halfcuboctahedron, in strut as well as cable mode has been presented. The struts
and cables were tested in the laboratory to obtain material properties to be used during
the numerical study. The average prestress levels obtained experimentally were used as
initial strains in the numerical analysis. The experimental results of both strut and cable
mode was compared with the numerical results considering the structure restrained in the
vertical direction only. Since large variation was found between the experimental and
numerical results, the numerical model was updated to match the experimental results.
The results obtained by the updated model for both the strut and the cable mode show
good agreement with the experimental values. During the experiment, the failure of
structure in both the strut and the cable mode was due to failure of the strut due to
buckling and no sign of failure was noticed either in the joints or the cables. The load
carrying capacity of the structure is higher in cable mode than strut mode. Further, the
arrangement. Hence, the structure in cable mode was considered for parametric study. As
it is not possible to conduct the experiments for different prestress levels, different
dimensions and rigidity ratio of the structure, a parametric study was done numerically to
The next chapter presents the development and analysis of a dismountable grid structure
86
CHAPTER 3
GRID STRUCTURE
3.1 INTRODUCTION
In this chapter, the fabrication and testing of a tensegrity grid, 2mx2m in size, by
each unit though turnbuckle. The structure is subjected to destructive load test during
which continuous monitoring of the prestress levels, key deflections and strains in the
struts and the cables, is carried out. The behaviour of the grid is also compared with
results obtained analysing the same grid structure using FEM. In addition, detailed
A dismountable tensegrity grid of size 2m×2m was fabricated keeping the dimensions of
the cables and the struts same as described in the case of the single module in Chapter 2.
This grid was different in fabrication as compared to other similar grids reported in the
literature (Hanaor 1993, Quirant et al. 2003, Fest et al. 2004), in which the grid was
constructed by joining the individual units at the joints. In the present case, however, the
grid was fabricated as a single cohesive unit. The different sub units were connected to
the common nodes in the top layer and the common cables in the bottom layer. To keep
the cross sectional area same, the interior bottom cables i.e. 4mm stranded wires were
87
replaced by 6mm stranded wires. As a result, the numbers of cables reduced by four as
compared to the conventional approach. For connecting the top and the bottom cables,
holes were drilled in a mild steel plate of 10 cm diameter and 5 mm thickness. For struts,
16 mm diameter mild steel rods of 80mm length were cut at a vertical angle of 24
degrees, and welded at the required positions. Mild steel hooks were also welded for
connecting the leg cables. A typical bottom joint is shown in Fig. 3.1.
Fig. 3.2 shows the tensegrity grid fabricated in the workshop of the Civil Engineering
Department, IIT Delhi. For testing the grid, the structure was supported on four supports
made of composite columns of 75 cm diameter and 1m height. One LVDT was fixed
under the central bottom node and another under a side bottom node. All members of one
of the four units were instrumented with electrical strain gauges. Four 5mm long strain
gauges were bonded on the surface of each pipe and two 2mm long strain gauges (TML,
2005) on each cable, as in case of the single module. A total of forty sensors were
instrumented for monitoring the structure. The detailed instrumentation is shown in Fig.
3.3. The load was applied gradually on the structure by electrically operated crane as
uniformly distributed load, as shown in Fig. 3.4. The structure failed at a total load of
6.19 kN due to failure of the U hook of one of top joint as shown in Fig. 3.5. There was
no sign of failure of any member at this load. Two joints of the structure underwent
88
Figure 3.1 Detail of central bottom joint
Supports
89
Strain
gauges
connected
to data
logger
90
Failure of top joint
The behaviour of the tensegrity grid structure tested (as described in the previous
sections) was modelled using FEM. The analysis was performed using ANSYS 9.0. All
the cables and the struts were considered as 3D spar elements, having three degrees of
freedom at each node, which were considered at the end of the elements. The material
was assumed as linear, elastic and isotropic as in the case of single tensegrity module.
Fig. 3.6 shows the model of the 2m×2m tensegrity grid generated using the preprocessor
in ANSYS. All three degrees of freedom were locked for corner bottom nodes i.e. 1, 5,
17 and 21. The Young’s modulus values of 2.05×105 N/mm2 and 0.952 ×105 N/mm2, as
obtained experimentally, were used for the strut and the cable element respectively. The
were used to determine the initial strain in the elements. The average prestress in the
91
struts was found to be 2.13 kN. The analysis procedure is same as described in Chapter 2
All the numerical values were obtained by using the updated model for the cable mode as
discussed in Chapter 2. Fig. 3.7 shows the deflection at the bottom node number 3 i.e. a
side bottom joint of the tensegrity grid (see Fig. 3.6). A maximum deviation of 6% is
observed between the experimental and the numerical plot. A very good agreement can
be observed between the two. Fig. 3.8 (a) and (b) show the force variation in the strut
between node 3 and 10 and strut between 6 and 11 respectively. Fig. 3.9 (a) shows the
force variation in the bottom inner cable connecting nodes 3 and 11. Here, the maximum
92
difference between the experimental and the numerical values is about 13%. Fig. 3.9 (b)
similarly shows the force variation in the bottom inner cable connecting nodes 1 and 9.
The numerical values lie within ±30% of the experimental values, which indicates
somewhat greater difference. Fig. 3.10 and Fig. 3.11 show the force variation in the leg
and top cables respectively. In general, most deviations are small and within reasonable
limits. The observed deviation between the experimental and numerical results is largely
7000
6000 Numerical
5000 Experimental
Load(N)
4000
3000
2000
1000
0
0 2 4 6 8 10
Deflection(mm)
93
Applied load (N)
-2000
0 1000 2000 3000 4000 5000 6000 7000
-3000
Experimental
Strut force(N)
Numerical
-4000
-5000
-6000
(a)
-1100
Experimental
-1300
-1500
-1700
-1900
-2100
-2300
(b)
94
2900
2700 Numerical
Experimental
Cable force(N)
2500
2300
2100
1900
1700
1500
0 1000 2000 3000 4000 5000 6000 7000
Applied load(N)
(a)
1000
Numerical
800
Cable force (N)
Experimental
600
400
200
0
0 1000 2000 3000 4000 5000 6000 7000
Applied load (N)
(b)
95
1400
Cable force(N) 1200 Numerical
1000 Experimental
800
600
400
200
0
0 1000 2000 3000 4000 5000 6000 7000
Applied load(N)
(a)
4500
4000
Numerical
3500
Cable force(N)
3000 Experimental
2500
2000
1500
1000
500
0
0 1000 2000 3000 4000 5000 6000 7000
Applied load(N)
(b)
96
1800
1600
Numerical
Cable force (N) 1400
Experimental
1200
1000
800
600
400
200
0
0 1000 2000 3000 4000 5000 6000 7000
Applied load(N)
(a)
1800
Numerical
1500
Experimental
Cable force(N)
1200
900
600
300
0
0 1000 2000 3000 4000 5000 6000 7000
Applied load(N)
(b)
97
3.5 PARAMETRIC STUDY ON TENSEGRITY GRID STRUCTURES
The comparison of the experimental and the numerical results indicates reasonable
agreement between the two for 2m×2m tensegrity grid. In real field, large tensegrity grid
difficult to perform. Hence, the numerical model is used for further parametric study on
tensegrity grid structures of different sizes. In this section, numerical models of grid of
sizes 4m×4m and 8m×8m are used to study variation of deflection, maximum and
minimum forces in members, effect of rigidity ratio, effect of prestress level, height of
In order to observe the variation of member forces and deflection at joints, a 4m×4m grid
structure consisting of 16 basic modules was analysed using FEM. The prestress level on
top cable was assumed 1.5 kN as the reference value. Fig. 3.12 and Fig. 3.13 show the
top and the perspective view of the 4m×4m grid’s model. The top and the bottom nodes
are indicated in the figure. All the three degrees of translation of bottom corner nodes i.e.
1, 9, 7 and 65 were locked. The structure was analysed for a live load of 400 N/m2, which
correspond to a total load of 6.4 kN. For the rigidity ratio of 50, the bottom and the top
deflections are shown in Fig. 3.14 and Fig. 3.15 respectively. The maximum bottom
deflection is found to be on node 33 i.e. the bottom central node and the same for the top
i.e. on node 6. The deflections at the nodes vary in a wide range. Among all the nodes,
including top and bottom, the maximum vertical deflection is observed on central bottom
node i.e. node 33 shown in Fig. 3.16. The deflection is expected to vary nonlinearly as in
98
case of single module. Although geometrical nonlinearities have been considered, the
observed curve is largely linear due to relatively small deflection (typically less than
span/250).
The numerical analysis shows the forces in 64 struts in the 4m×4m grid can be grouped
into four types due to symmetry. The strut numbers and the nodes connected by them are
listed in Table 3.1, corresponding to Fig. 3.12. Fig. 3.17 shows the forces in 16 struts of
the 4m grid structure. The maximum and minimum force is observed in strut 1 and strut 4
Corresponding to a deflection of span/250, the maximum observed force is less than the
As discussed, the maximum and minimum force occurs in particular members. The 4m
grid structure was analysed for higher loads, and the variation of the maximum and the
minimum force in the struts, the top cables, the leg cables and the bottom cables is shown
in Fig. 3.18, 3.19, 3.20 and 3.21 respectively. With increase in the load, it is observed
from the figures that maximum force increases and the minimum force decreases and
stress reversal also occurs at higher loads. However, the load at which stress reversal
starts is higher than the load corresponding to a vertical deflection of span/250. For
example, the stress reversal occurs in the strut, the top cable, the bottom cable and the leg
cable occurs at loads 12 kN, 5 kN, 7.2 kN and 9.6 kN respectively. The limiting
deflection is 16mm for 4m grid which is observed at a total load of 4 kN i.e. 250 N/m2.
99
As there is redundancy in the structure, stress reversal will not affect the overall stability
of the grid, which is also verified during experiment of the 2m grid described before.
100
7000
node 3
6000
5000 node 5
7000
node 2
6000
node 4
5000 node 6
node 8
Load (N)
4000
node 11
3000 node 12
node 13
2000
node 18
1000 node 20
node 26
0
0 5 10 15 20 25 30
Deflection (mm)
101
30000
25000
20000
Load(N)
15000
10000
5000
0
0 20 40 60 80 100 120 140
Deflection(mm)
Table 3.1 Strut number and their connectivity in 4m×4m grid structure
Strut Bottom Top Strut Bottom Top Strut Bottom Top Strut Bottom Top
number node node number node node number node node number node node
1 1 11 5 3 12 9 5 13 13 17 26
2 3 16 6 5 18 10 7 20 14 19 32
3 17 10 7 19 11 11 21 12 15 33 25
4 15 2 8 17 4 12 19 6 16 31 18
102
0
strut 1
0 1000 2000 3000 4000 5000 6000 7000
-500 strut 2
strut 3
-1000
strut 4
-1500 strut 5
strut 6
Strut force (N)
-2000
strut 7
-2500 strut 8
-3000 strut 9
strut 10
-3500 strut 11
-4000 strut 12
strut 13
-4500
strut 14
-5000 strut 15
6000
4000
Maximum
2000
0
Strut force(N)
-4000
-6000
-8000
-10000 Minimum
-12000
-14000
Applied load(N)
103
15000
10000 Maximum
Cable force(N)
5000
Minimum
0
0 5000 10000 15000 20000 25000 30000
-5000
-10000
Applied load(N)
10000
8000 Maximum
Cable force(N)
6000
4000
2000 Minimum
0
-2000 0 5000 10000 15000 20000 25000 30000
-4000
Applied force(N)
104
10000
Maximum
8000
Cable force(N)
6000
4000 Minimum
2000
0
0 5000 10000 15000 20000 25000 30000
Applied load(N)
Figure 3.21 Variation inner bottom cable force in 4×4m grid structure
(Prestress level 1.5 kN, Rigidity ratio 50)
In this section, the effect of prestress level and the rigidity ratio in tensegrity grid is
studied. For this, 4m×4m grid structure of 0.5m height was modelled. For analysis,
prestress level of 1 kN, 1.5 kN and 2 kN and rigidity ratio of 10, 25 and 50 were
considered. Fig. 3.22 shows the maximum deflection of the structure for different
prestress levels at a rigidity ratio 10. The variation of deflection is only marginal i.e. less
than 0.5%. This indicates that the level of prestress has very little effect on the maximum
deflection. The maximum deflection for different rigidity ratios having prestress level of
1 kN is shown in Fig. 3.23. The rigidity ratios were enforced by changing the cable areas,
keeping the strut areas constant. At a particular load, the deflection increases with
increase in the rigidity ratio. This is because of the reduced cable area with increasing
rigidity ratios. The deflection varies 360% at lower loads and 440% at higher loads.
Hence, rigidity ratio has large influence on deflection. Fig. 3.24 shows variation of the
105
maximum strut force for different prestress levels for a rigidity ratio of 10. The strut force
increases with increase in the prestress level. Comparing the forces for 2 kN and 1 kN
prestresses, it is found the force variation is nearly 90% at lower loads and the difference
decreases to 15% at higher load i.e. 25600N (or 1600 N/m2). Fig. 3.25 shows the
maximum force variation in the struts for different rigidity ratios for prestress level of 1
kN. It is observed the rigidity ratio has negligible effect up to load 10 kN and has little
30000
25000
20000
Load(N)
presterss 1000
15000 prestress 1500
10000 prestress 2000
5000
0
0 50 100 150
Deflection(mm)
106
30000
25000
20000
Load(N)
r10
15000 r25
10000 r50
5000
0
0 50 100 150
Deflection(mm)
14000
12000
Strut force (N)
10000
pre1000
8000
pre1500
6000
pre2000
4000
2000
0
0 5000 10000 15000 20000 25000 30000
Applied load (N)
107
14000
12000
Strut force(N)
10000
r10
8000
r25
6000
r50
4000
2000
0
0 5000 10000 15000 20000 25000 30000
Applied load (N)
In this section, the effect of the height on the deflection, the strut force and the load
carrying capacity is studied. For this study, grid size of 8m×8m was simulated using
FEM. All the degrees of translation at corner nodes were locked. The structure was
analyzed for five rigidity ratios i.e. 10, 20, 30, 40 and 50, at a prestress level of 1.5 kN
and same load intensity i.e. 400 N/m2. Fig. 3.26 shows the perspective view of the
8m×8m tensegrity grid. The variation of the maximum deflection for different heights for
a rigidity ratio of 10 is shown in Fig. 3.27. With increase in height, the maximum
deflection decreases considerably for same load intensity. Comparing the results of 0.8m
with respect to 0.5m height, it is found that the deflection deceases by 33% at smaller
loading intensity and to 60% at higher loads. Fig. 3.28 shows the maximum deflection for
8×8m grid structure for different rigidity ratios for 0.8m height at 1.5 kN prestress. The
increase in deflection is very high with increase in the rigidity ratio. Load carrying
108
capacity of grid structure increases nonlinearly with a decrease in rigidity ratio, as shown
in Fig. 3.29. The load carrying capacity increases by 5.4 times by decreasing rigidity ratio
five times. Fig. 3.30 shows the load carrying capacity of the grid structure for different
heights having a rigidity ratio of 10. By increasing the height from 0.5m to 0.8m i.e. 1.6
times, the load carrying capacity increases by 2.3 times. The variation of the maximum
strut force for different heights at a rigidity ratio 10 is presented in Fig. 3.31. The strut
force decreases with increase in height. The decrease is 7% at smaller load intensity and
about 26% at 400 N/m2. Fig. 3.32 shows the variation of strut force for different rigidity
ratios. With increase in the rigidity ratio, the strut force increases. This is expected, since
struts have to bear greater fraction of the applied loads due to reduction in cable area.
However, the increase is very negligible at lower load intensities and less than 5% at
higher loads. In general, the rigidity ratio has only small influence on variation of strut
force.
109
450
400
350
Load (N/m2)
300
250
height 0.5m
200
height 0.6m
150
height 0.7m
100 height 0.8m
50
0
0 20 40 60 80 100
Deflection (mm)
Figure 3.27 Maximum deflection for 8×8m grid structure for different
heights (Prestress 1.5 kN, Rigidity ratio 10)
450
400
350
300
Load (N/m2)
250 r10
200 r20
150 r30
100 r40
50 r50
0
0 50 100 150 200
Deflection(mm)
Figure 3.28 Maximum deflection for 8×8m grid structure for different
rigidity ratio (Prestress 1.5 kN, Height 0.8m)
110
400
350
300
250
Load (N/m2)
200
150
100
50
0
0 10 20 30 40 50 60
Rigidity ratio
Figure 3.29 Load carrying capacity of 8×8m grid structure for different
rigidity ratio (Prestress 1.5 kN, Height 0.8m)
400
350
300
Load (N/m2)
250
200
150
0.5 0.6 0.7 0.8
Height (m)
Figure 3.30 Load carrying capacity of 8×8m grid structure for different
heights (Prestress 1.5kN, Rigidity ratio 10)
111
16000
height 0.5m
14000
12000 height 0.6m
Strut force(N)
10000 height 0.7m
8000 height 0.8m
6000
4000
2000
0
0 5000 10000 15000 20000 25000 30000
Appled load(N)
Figure 3.31 Maximum strut force of 8×8m grid structure for different
height (Prestress 1.5 kN, Rigidity ratio 10)
12000
10000
Strut force(N)
8000 r10
6000 r20
4000 r30
r40
2000
r50
0
0 5000 10000 15000 20000 25000 30000
Applied load(N)
Figure 3.32 Maximum strut force of 8×8m grid structure for different
rigidity ratios (Prestress 1.5 kN, Height 0.8m)
112
3.5.4 Effect of Number of Supports
Supports have significant effect on the deflection, the load carrying capacity and the
maximum strut force. 8m×8m grid structure was considered to study the effect of
support. In the first case, the structure was supported on four corners with all three
degrees freedom locked. In the second case, the structure was supported with
intermediate supports on the periphery only locked in vertical direction in addition to first
case. Fig. 3.33 shows the comparison of the maximum deflection for the grid structure
with and without the intermediate support. The maximum deflection is observed at the
central bottom node in both the cases. It is also observed that the deflection reduces
considerably by providing one intermediate support. For example, the deflection reduces
from 0.98 mm to 0.32 mm for 5 N/m2 and 86.5 mm to 20.6 mm for 400 N/m2. Hence,
greater reduction of deflection occurs at higher load intensity. This is due to increased
apparent stiffness as the structure deforms under load. Fig. 3.34 shows the variation of
the maximum strut force for different heights with intermediate supports. The decrease in
strut force is very negligible for small load intensity, for example the force reduces from
2.65 kN to 2.62 kN at 5 N/m2. However, the decrease is nearly 15% at 400 N/m2 loading
intensity.
Keeping the prestress level, the height and the rigidity ratio same, the variation of the
maximum strut force with and without intermediate support is shown in Fig. 3.35. When
the grid structure is supported only at the corners, the maximum strut force is observed in
the strut connected to the support. By adding intermediate supports, the maximum value
is observed in a strut connected to the intermediate support and the force in the strut
113
connected to the corner support reduces considerably. With intermediate supports, the
maximum strut force reduces from 3.26 kN to 2.8 kN at an applied load of 1.6 kN and
13.88 kN to 5.92 kN at applied load of 25.6 kN. The load carrying capacity of the 8m
grid structure with and without intermediate supports for different rigidity ratios and
0.5m height is shown in Fig. 3.36. In both cases, the load carrying capacity increases
nonlinearly with increase in the rigidity ratio, keeping other parameters constant. With
intermediate supports, the load carrying capacity is much higher than that without
intermediate supports at same rigidity ratio. Fig. 3.37 shows the load carrying capacity of
the 8m grid structure with and without intermediate supports for different heights at a
constant rigidity ratio of 50. Comparing the values, it is found that the load carrying
capacity is nearly 3.8 times with intermediate supports for 0.5 m height and this value
increases with increase in the height i.e. 4.3 times for 0.8m height. For prestress level of
1.5 kN, height 0.5 m and rigidity ratio of 50, the load carrying capacity of the grid
structure for various spans with and without intermediate support is shown in Fig. 3.38.
In both the case, the load carrying capacity decreases with increase in the grid size. The
difference in the load carrying capacity is higher for 4m grid size (nearly 357%) and
3.6 CONCLUSION
This chapter has presented the fabrication, testing and the finite element analysis of
2m×2m tensegrity grid structure. The comparison of the deflection and the various
between the two. Further, the numerical model was considered for parametric study of
114
500
400 Case1
Load (N/m2) Case 2
300
200
100
0
0 20 40 60 80 100
Deflection (mm)
7000
6000
Strut force(N)
5000
4000
height 0.5m
3000
heigh 0.6m
2000
height 0.7m
1000
height 0.8m
0
0 5000 10000 15000 20000 25000 30000
Applied load(N)
115
16000
14000 Case 2
12000 Case 1
Strut force(N)
10000
8000
6000
4000
2000
0
0 5000 10000 15000 20000 25000 30000
Applied load (N)
Figure 3.35 Comparison of maximum strut force in 8×8m grid structure for
different support condition (Prestress 1.5 kN, Height 0.5m, Rigidity ratio 10)
Case 1: Intermediate support in periphery
Case 2: Without intermediate support in periphery
700
600
500
Load (N/m2)
400 Case 2
300 Case 1
200
100
0
0 10 20 30 40 50 60
Rigidity ratio
116
350
300 Case 1
250 Case 2
Load (N/m2)
200
150
100
50
0
0.5 0.55 0.6 0.65 0.7 0.75 0.8
Height (m)
1200
1000 Case 1
800 Case 2
Load (N/m2)
600
400
200
0
4 5 6 7 8
Grid size (m)
117
larger grid size structures. The effect of height, rigidity ratio, prestress level and support
The proposed dismountable tensegrity grid is easy to fabricate and easy to assemble and
The structure requires less space for storage and is easy to transport as the components
can be dismantled. The tensegrity grids reported so far in the literature require
proposed method reduces the number of cables. There was no indication of failure of any
experimental values of deflection and member forces showed slight nonlinear variation
where as the numerical values showed more or less a linear variation though large
The next chapter describes the behaviour of the tensegrity grids under various possible
118
CHAPTER 4
4.1 INTRODUCTION
This chapter investigates the behaviour of the tensegrity structures under various loads
such as dead loads, live loads, lack of fit in members, thermal loads and wind loads. The
effects of combination of all these loads on the maximum deflection and the maximum
and the minimum forces in different elements are also studied separately. A 2m×2m grid
structure is considered for analysis of all loads and their combinations. The prestress
force 1.5 kN on top cable and rigidity ratio of 50 is considered for analysis. Finally, an
The grid structure of 2m×2m size was analysed for live loads, lack of fit and thermal
loads and their combination. The FEM analysis was done using ANSYS as described in
Chapter 2. The perspective and top view of the 2m grid with the node and the elements
number generated in ANSYS is shown in Fig. 4.1. For analysis, three degrees of freedom
in translation were locked in all load cases at the bottom corner joints.
The maximum live load considered on the roof structure was 750 N/m2 as per IS 875
(1987) and distributed among the interior top nodes in proportion to their tributary area
i.e. a load of 187.5 N on the top corner nodes and 375 N on the top interior nodes.
119
Numerical analysis was carried out for these loads to determine the deflections and
member forces.
Analysis for lack of fit in the cable was carried out particularly to study the effect of any
defect at the time of fabrication of the structure. In this case, length of one of the cables
i.e. cable joining the nodes 1 and 3, was considered 1mm shorter than its ideal length.
This increases the initial strain of the shortened cable by 0.001. Initial strain of all other
In this case, length of one of the struts i.e. the strut connecting the nodes 1 and 7, was
considered 1mm shorter than its original length. This increased the initial compressive
strain by 0.001247, keeping the initial strain of the other elements unaltered. Numerical
analysis was carried out to determine the member force in the elements and deflection in
nodes.
In case of roof structures, a temperature difference generally exists between the exposed
outside surface and the inside surface. Due to this difference, there may be the
the effect of the differential temperature on the maximum deflection and the member
forces. In this study, the reference temperature was taken as 25O C and temperature of all
120
the top cables was increased to 40O C i.e. a temperature difference of 15O C existed
between the top and the bottom surfaces. The coefficient of thermal expansion for the
(a)
(b)
121
4.2.5 Combined Action of Lack of Fit and Thermal loads
For this analysis, the thermal load as described in 4.2.4 and the lack of fit in the cable as
discussed in 4.2.3 were combined to predict the effect of the combined load on the
Table 4.1 shows the deflection at different nodes for live load, lack of fit in the cable and
the strut, thermal load and the combination of the thermal load and the lack of fit in the
cable. The maximum deflection due to live load was found in the top nodes i.e. 4, 6, 16
and 18. The deflection of the central bottom node was less than the bottom side nodes. It
is observed that due to lack of fit in the bottom cable connecting node 1 and 3, the nodes
having direct connection with these two nodes have been lifted by small amount upwards
where as the remaining nodes have no or very negligible deflection. However, all the
nodes deflect by substantial amount due to the thermal load. Further, the combination of
the thermal load and the lack of fit in the cable show that the deflection due to thermal
effect is predominant.
The axial force in various members due to live load, lack of fit in the cable and the strut,
thermal load and combination of thermal load and lack of fit in the cable is shown in
Table A in appendix. The prestress values for the strut, the top, the leg and the bottom
cable were considered as 2.59 kN, 1.5 kN, 1.5 kN and 1.06 kN respectively. It is
observed that due to live load, the maximum increase of force with respect to prestress
equilibrium value in the strut, the top, the leg and the bottom cable is 49%, 22%, 48%
and 36% respectively. Similarly, the maximum decrease of force with respect to prestress
122
equilibrium value in the strut, the top, the leg and the bottom cable is 21%, 29%, 38%
and 35 % respectively. Due to lack of fit in cable 33 (connecting node 1 and 3), the
change in force in struts and cables is within 2.5% as compared to the prestress values.
However, there was an increase 33.3% in the cable 33 and 28.5% in cable 34 (connecting
node 3 and 5). Lack of fit by 1mm in strut i.e. element 45 changes the force in the
elements of other modules within 2%. But, the force in the elements of the module (in
which there is lack of fit) reduced between 14% and 23%. By increasing temperature of
the top cables by 15 OC with respect to the reference value, the force in all struts, top as
well as bottom cables reduced by nearly 11% where as the forces in all the bottom cables
increased by 18% as compared to the prestress equilibrium values. The combined effect
of the lack of fit in the cable and thermal load reduced the force in all the elements except
the bottom cables, in which an increase can be observed. The maximum increase of 52%
The live load, thermal loads and lack of fit in cable were applied simultaneously on the
structure and the analysis was carried out to study behaviour of the structure under the
In this analysis, initial strain of one of the cable i.e. the element connecting nodes 1 and
3, was increased by 0.001. Live load of 187.5 N was applied on the exterior nodes and
123
Table 4.1 Deflection in mm on nodes
124
4.3.2 Analysis for Combined Live Load and Thermal Load
In this case, a live load of 187.5 N was applied on the exterior nodes and 375 N on the
interior nodes. A temperature difference of 15OC was applied between the top and the
interior elements.
4.3.3 Analysis for combined live Load, Lack of Fit and Thermal Load
For this analysis, temperature difference of 15OC was considered between the top
elements and the interior elements. Reference temperature was taken as 25OC and the
increased temperature on the exposed elements as 40OC. Initial strain of one of the cable
Tables 4.3 and Table B in appendix show the maximum deflection at various nods and
the forces in the different elements due to the combined load. The maximum and
minimum forces in the strut, the top, the leg and the bottom cable is highlighted in Table
B in appendix and summarised in Table 4.4. The comparison of force is shown in (Figs.
4.3, 4.4, 4.5 and 4.6) corresponding to the four cases. The different load combinations
125
Table 4.3 Deflection in mm due to load combinations
126
Table 4.4 Maximum & minimum force in elements due to combined load
46 3 10 Strut maximum
49 3 8 Strut minimum
Fig. 4.2 shows a comparison of the deflection at the bottom node 3 for different load
combinations. It can be observed that lack of fit in the cable acting together with live load
reduces the deflection as compared to that live load only. Up to 15 N/m2 load, the
difference is more than 100%. But, this difference decreases with increase in load and it
is only 8% at 750 N/m2. The deflection is maximum for the combination of live and
thermal load. The difference is more than 100% up to 35 N/m2 load. However, this
difference decreases to nearly 15.5% at a load of 750 N/m2. Due to the combination of
live load, lack of fit in the cable and thermal load, the deflection value is more than the
value obtained for live load alone. The difference is 88.6% and 7.3% corresponding to
127
load intensity of 3 N/m2 and 750 N/m2 respectively. In all the cases, the trend of variation
Figs. 4.3, 4.4 and 4.5 show the comparison of the maximum and the minimum forces in
the top cable, the strut and the leg cable respectively. In all the case, variation trend for
the maximum and the minimum force is more or less the same. The maximum force due
to live loads is less than the force due to the combination of the lack of fit in the cable and
the live loads. The maximum force due to thermal and live load is less than the
corresponding values due to live load alone. The difference is very less at smaller loads
and increases nonlinearly with increase in the applied load. However, the minimum force
in the elements is more in case of live loads alone as compared to the combined load. The
minimum force in the elements is observed in case of lack of fit in cable combined with
live and thermal load. The variation of the maximum and the minimum forces in the
bottom cable is shown in Fig. 4.6. The maximum force in the bottom cable occurs due to
the combination of live loads, thermal loads and the lack of fit in the cable. A large
variation is observed from the figure. The difference in the minimum force in the bottom
cable due to live loads alone and the combination of lack of fit and the live load is very
negligible. Similarly, the minimum force in the bottom cable due to live loads, thermal
loads and the lack of fit is the same as the force due to live loads and thermal load. This
analysis indicates that the effect of thermal load is much more predominant in tensegrity
grid structure as compared to lack of fit in the elements, provided lack of fit is within the
tolerances considered.
128
3500
3000
A p p lie d lo a d (N )
2500 case 1
2000 case 2
1500 case 3
1000 case 4
500
0
0 1 2 3 4 5
Deflection (mm)
129
2000
Top cable force (N)
1900
case 1
1800
case 2
1700
case 3
1600 case 4
1500
1400
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(a)
1600
Top cable force(N)
1400
case 1
case 2
1200
case 3
case 4
1000
800
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(b)
Figure 4.3 Variation of top cable force for different load cases
(a) Maximum value (b) Minimum value
130
4000
2000
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(a)
2800
Strut force (N)
2400 case 1
case 2
2000 case 3
case 4
1600
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(b)
131
2400
1200
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(a)
1600
Leg cable force (N)
1200 case 1
case 2
case 3
800 case 4
400
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(b)
Figure 4.5 Variation of leg cable force for different load cases
(a) Maximum value (b) Minimum value
132
2000
1000
0 400 800 1200 1600 2000 2400 2800 3200
Applied load (N)
(a)
1400
1200
Bottom cable force(N)
1000
800 case 1
600 case 2
400 case 3
200 case 4
0
0 1000 2000 3000 4000
Applied load (N)
(b)
Figure 4.6 Variation of bottom cable force for different load cases
(a) Maximum value (b) Minimum value
133
4.4 ANALYSIS FOR WIND LOADS
In field applications, the tensegrity grid structures will be subjected to wind load. Hence,
the structure should have sufficient strength to withstand wind load. In this study, a
2m×2m grid structure was analysed for wind load for Delhi region assuming all corner
nodes locked in the three translational degrees of freedom. The rigidity ratio, height and
prestress values were considered as 50, 0.5m and 1.5 kN respectively. The structure was
assumed to be covered with sheets having dead load of 0.2 kN/m2. Wind load was
considered according to the provisions laid down in IS: 875 (Part-3), 1987 in both
horizontal and vertical directions. In the analysis, positive wind load indicates the force
acting towards the structural element and negative load indicates a load away from it.
The wind load, ‘F’, acting in a direction normal to the individual structural element or
F = (C pe − C pi ) ⋅ A ⋅ Pz (4.1)
P z = 0.6 ⋅ V z2 (4.2)
V z = Vb ⋅ K 1 ⋅ K 2 ⋅ K 3 (4.3)
134
where, K 1 = Risk coefficient, for general buildings and structures (taken as 1, considering
K 2 = Factor to obtain the design wind speed variation with height in different terrains for
different classes of buildings or structures (In this study, considered a value of 0.82, for
terrain category 3 and class ‘C’ structures with height of the building of 3m).
The tributary areas considered for different nodes in the windward and leeward sides.
These values lead to a design wind speed of 38.54 m/sec from Equation 4.3 and design
wind pressure of 891.2 N/m2 from Equation 4.2. The tributary areas of the various nodes
are shown in Fig. 4.7, along with the exact values of the areas in Table 4.5.
4 5
4 5
2
1 3
1
2 3
135
Table 4.5 Tributary areas of nodes
(m2)
1 0.09375
2 0.25
3 0.09375
4 0.15625
5 0.15625
In this analysis, a building width of 2m and a total height of 3.5m was considered,
assuming 3m clear height. For this case, IS: 875 recommends following coefficients
Case 1
Windward side
136
Leeward side
Case 2
Windward side
C pi = -0.7
Leeward side
Numerical analysis was carried out using ANSYS 9.0 for the above two cases of
horizontal wind load taking dead load (weight of the sheeting) on the top nodes to be
200N/m2 in both cases. Dead load was distributed among the top nodes according to their
137
4.4.1.2 Vertical wind force analysis
For roof angle zero degree, wind angle zero degree and h/w = 1.5, C pe = -0.8 (IS 875,
Case 1
C pi = +0.7, C pe = -0.8
Where, (2*2) is the area of the roof perpendicular to the applied load. This is the total
wind load acting on the roof. This force is distributed among the top nodes in proportion
with their tributary area as explained earlier. After distribution, force at the top exterior
nodes was found to be –334.2N and the same at top interior nodes are obtained as –668.4
N.
Distributing the dead weight of the structure i.e. 200N/m2 proportionately among the top
nodes according to their tributary areas, force at the top interior nodes comes out to be
+100N and at the exterior nodes as +50 N. Hence, the combined force of wind load and
dead load comes out to be -284.2N and –568.4N at the top exterior nodes and top interior
nodes respectively.
Case 2
138
Calculation of wind force was carried out with these values of C pe and C pi by the method
mentioned in Case-1. After calculation, combined force of the wind and the dead load
Table C in appendix shows the forces induced in different elements of tensegrity grid
structure due to different case of wind loading described earlier. In both Case 1 and 2, the
maximum and the minimum forces in the top cable occur in element 8 and 10
respectively. The maximum leg force is observed in element 26 in both Case 1 and Case
2, but the minimum leg force is observed in element 31 and 29 in case 1 and case 2
respectively. The maximum and the minimum force in the strut is observed in elements
46 and 45 in Case 1 and in element 59 and 60 in Case 2 respectively. In the vertical wind
force analysis, it is observed that maximum force for Case 3 and minimum force for Case
4 occurs in the same member and vice versa. In all cases of wind analysis, it is found that
there is no stress reversal in any element, though there is decrease in force in some
The deflections at various nodes for the different cases of wind load analysis are listed in
Table 4.6. The positive sign indicates downward deflection where as the negative sign
indicates upward deflection. Under horizontal wind load, node 3 has the maximum
downward deflection and node 19 the maximum upward deflection in both Case 1 and 2.
However, the central bottom node is deflected up by 0.225 mm in Case 1 and is moved
because of net upward force on all the nodes. All nodes deflect downwards in Case 4 due
139
to net downward vertical force in the joints. However, the deflections are within
permissible limit i.e. L/250 as per codal provision (IS 800: 1984). By increasing the
vertical live load up to 750 N/m2, the net loads in Case 3 will be reduced and at some
stage it will be downward. Hence, the upward deflection will be reduced and the structure
WIND LOAD
In the previous section, a 2m×2m tensegrity grid was analysed for wind loads as per
Indian Standards of practice. However, in real field application, larger size grids might be
used for easy and quick construction. Hence in this section, an 8m×8m tensegrity grid
was checked for the wind load for Delhi region. For design, the rigidity ratio and height
spacing on the periphery and restrained in the vertical direction only. The structure was
subjected to 750 N/m2 live load and wind loads as per Indian code of practice. The cross
sectional area and Young’s modulus of struts and cables were same as considered for 2 m
grid. The wind load was calculated as described in Section 4.4.1 and the values were
same in Case 1 and Case 2. The wind load calculated for Case 3 and Case 4 is described
below.
Case 3
140
Table 4.6 Deflection in mm due to wind load
Node
number Case1 Case 2 Case 3 Case 4
1 0 0 0 0
2 0.867 1.016 -2.6 0.26
3 1.357 1.592 -5.19 0.511
4 1.042 1.244 -6.48 0.64
5 0 0 0 0
6 0.94 0.733 -6.48 0.64
7 0.815 0.786 -3.51 0.35
8 -0.397 -0.713 -2.6 0.26
9 0.71 0.713 -5.19 0.511
10 0.206 0.225 -3.51 0.35
11 -0.225 0.054 -3.02 0.3
12 -0.214 -0.203 -3.51 0.35
13 -0.707 -0.723 -5.19 0.511
14 0.586 0.288 -2.6 0.26
15 -0.791 -0.833 -3.51 0.35
16 -0.807 -1.04 -6.48 0.64
17 0 0 0 0
18 -1.155 -0.973 -6.48 0.64
19 -1.49 -1.283 -5.18 0.511
20 -0.947 -0.812 -2.6 0.26
21 0 0 0 0
141
Case 4
The node numbers and the loading for case 1 are shown in Fig. 4.8 (a) and 4.8 (b)
respectively. The structure was analysed and the maximum deflection and forces in
elements for different cases are presented in tabular form in Tables 4.8, 4.9 and 4.10.
The maximum deflections obtained in analysis for different load case are shown in Table
4.7. Maximum vertical deflection is observed in Case 4 for the central bottom node and
maximum deflection in horizontal direction at node 18 for Case 2. As per the codal
practice, the value should be within span/250, which is 32 mm. Hence, the maximum
deflection is 40% of allowable limit. Similarly, the maximum forces in elements for
different load case are shown in Table 4.8 and 4.9. The results show that the maximum
values are less than maximum carrying capacity. Though some of the cables are loosen
due to less force than the prestressed equilibrium value, the structure will remain stable
142
(a)
(b)
Figure 4.8 a) 8m grid with node numbers
b) Loading for case 1
143
Table 4.7 Maximum deflection due to combined live load and wind load for 8m×8m
grid
144
Table 4.8 Member force range for case 1 & 2 due to combined live load and wind
Case 1 Case 2
Node Node Force Node Node Force
Element 1 2 (N) Element 1 2 (N) Remarks
Top
cable 12 32 20 1624.295 3 19 28 888.3033 Minimum
Top
cable 226 184 201 1008.688 8 30 19 1768.686 Maximum
Leg
cable 258 3 19 1334.5 258 3 19 1351.169 Minimum
Leg
cable 508 221 206 1869.078 506 197 207 1932.938 Maximum
Bottom
periphery
cable 513 1 3 1909.101 529 225 223 248.85 Minimum
Bottom
periphery
cable 529 225 223 344.34 536 211 209 1956.3 Maximum
Bottom
interior
cable 569 105 107 2055.8 625 9 35 2205.1 Maximum
Bottom
interior
cable 632 191 217 1685.9 565 87 89 1500.5 Minimum
Strut 657 1 19 1825.157 657 1 19 1832.525 Minimum
Strut 907 223 206 2783.827 895 217 203 2811.519 Maximum
145
Table 4.9 Member force range for case 3 & 4 due to combined live load and wind
Case 3 Case 4
Node Node Force Node Node Force
Element 1 2 (N) Element 1 2 (N) Remarks
Top
cable 45 47 34 4140.618 11 21 32 53.02465 Minimum
Top
cable 160 146 129 204.9414 160 146 129 2933.232 Maximum
Leg
cable 270 9 22 4871.493 274 11 23 3835.107 Minimum
Leg
cable 271 35 34 74.70585 276 35 22 63.0304 Maximum
Bottom
periphery
cable 514 3 5 35.85003 514 3 5 2130.502 Minimum
Bottom
periphery
cable 516 7 9 3042.414 517 9 11 624.5654 Maximum
Bottom
interior
cable 566 89 91 11.72602 549 35 37 1228 Maximum
Bottom
interior
cable 569 105 107 3492.367 569 105 107 7700.277 Minimum
Strut 657 1 19 211.0879 669 7 22 772.6405 Minimum
Strut 675 37 22 6017.199 673 9 23 6301.024 Maximum
146
4.6 CONCLUDING REMARKS
In this chapter, the tensegrity grid structures are analysed for live load, lack of fit in
cable, lack of fit in strut, thermal load and the combined loads. The effects of all these
forces on the maximum deflection, the maximum and the minimum forces on different
elements were also studied separately for 2m×2m grid. The rigidity ratio and height of
structure were assumed as 10 and 0.5m respectively. It is found that the effect of
temperature is predominant on deflections and member forces and lack of fit has
marginal effect on results. In all the analysis, no reversal of stress was found in the
elements. The maximum deflection and member forces were within permissible value.
The structure is also analysed for wind load as per Indian standard of practice are found
safe. Some cable elements were slackened, but posing no damage to the structure due to
inherent redundancy.
Finally, 8m×8m tensegrity grid was designed for vertical load of 750 N/m2 and wind
load. The rigidity ratio and height of structure is taken as 10 and 0.8m respectively.
The next chapter describes a neural network based approach for design of tensegrity
structures.
147
CHAPTER 5
NEURAL NETWORK
5.1 INTRODUCTION
processing units called neurons or nodes. Each processing unit is capable of only a few
stores the knowledge it has learnt as strength of the connections between its processors.
The large number of the processing units and interconnections, similar to the neural
structure of human brain, give the neural network its capability. ANNs can also be
organized into layers. They can produce meaningful solutions to problems even when
Neural networks model the simplest and the most basic characteristics of the biological
neurons. The idea is not to replicate their size and complexity but to exploit their essential
neuron in the human nervous system has unique capabilities to receive process and
transmit electrochemical signals over the neural pathways that comprise the brain’s
communication systems.
148
The structure of a pair of typical biological neurons is shown in Fig. 5.1. Dendrites
extend from the cell body to other neurons where they receive signals at a connection
point called synapse. On the receiving side of the synapse, these inputs are transfered to
the cell body. In the cell body, the inputs (signals) are summed, some inputs tending to
excite the cell, other tending to inhibit its firing. When the cumulative excitation in the
cell exceeds a threshold, the cell fires and sends a signal down the axon to other neurons.
This basic functional outline has been used in the ANN concept.
ANNs have been successfully applied for solving a wide range of problems in various
fields. However, their applications on tensegrity structures are very limited. Tensegrity
structures are associated with problems like form finding, analysis and design. The design
of tensegrity structures is complex and iterative process. In the present study, an attempt
149
has been made to explore the possibility of applying the ANN for designing tensegrity
5.2.1 Introduction
Neural networks analyse data by passing it through several simulated processors that are
are then multiplied by a set of weights, w. The neurons then, nonlinearly transform the
sum of the weighted inputs, by means of an activation function, f, into an output value, a.
The output of a neuron, thus, depends on the neuron’s input and on its activation function.
Sometimes a bias, b, is also added to the network. The bias is then regarded as a weight,
a= f ( ∑ wp ) or a = f ( ∑ wp + b ) (5.1)
Though McCulloch and Pitts (1943) introduced simplified neurons quite early,
application of the ANNs in civil engineering started in late 1980s by Flood (1989). Since
then, ANNs have been applied to various problems such as determining the loads on the
axles of fast moving trucks (Gagarine et al., 1992), construction simulation (Flood, 1990),
estimating construction costs (Moselehi et al., 1991) and the selection of vertical
formwork systems (Kamarthi et al., 1992). ANN models can be divided into two groups
propagation neural nets based on feed forward architectures are examples of supervised
150
Adaptive resonance theory (ART) networks (Carpenter and Grossberg, 1988), Hopfield
networks (Hopfield, 1982) and Mean field annealing (MFA) networks (Peterson and
Anderson, 1987).
Multilayer feed forward networks are one of the corner stones of research in ANN. This
section discusses the use of artificial neurons as building blocks for multilayer feed
forward networks, together with learning algorithms needed to train such structures. The
neuron has n number of inputs xi, each of which input is connected to the neuron by a
weighted link wi . The neuron sums up the net input according to the equation:
n
net = ∑ x i w i (5.2)
i =1
To calculate the output, an activation function f , is applied to the net input of the neuron.
like logsig, tansig etc. The artificial neuron is an abstract model of the biological neuron.
The strength of a connection is coded in the weight. The intensity of the input signal is
artificial neuron works in discrete time steps; the inputs being read and processed at one
moment in time.
There are several possible learning methods for a single neuron. Most of the supervised
methods are based on the idea of changing the weight in a direction that the difference
between the calculated output and the desired output is decreased. Examples of such rules
are the perception learning rule, the gradient descent learning rule etc. as discussed by
151
Haykins (2002). A single layer network is a simple structure consisting of m neurons,
each having n inputs. The system performs a mapping from the n-dimensional input
space to the m-dimensional output space. To train the network, the same learning
algorithms as for a single neuron can be used. This type of network is widely used for
linear separable problems. However, unlike a neuron, single layer networks are not
capable of classifying non-linear separable data sets. One way to tackle this problem is to
Multilayer networks solve the classification problem for non-linear sets by employing
hidden layers, whose neurons are not directly connected to the output. The additional
the separation capacity of the network. This new architecture introduced the way to train
the hidden units for which the desired output is not known. The back propagation
algorithm offered a solution to this problem. In back propagation approach, the training
occurs in a supervised style. The basic idea is to present the input vector to the network,
calculate the output of each layer in the forward direction and then the final output of the
network. For the output layer, the desired values are known and therefore the weights can
be adjusted as for a single layer network according to the gradient decent rule. To
calculate the weight changes in the hidden layer, the error in the output layer is back
propagated to these layers according to the connecting weights. This process is repeated
for each sample in the training set. One cycle through the training set is called an epoch.
The number of epochs needed to train the network depends on various parameters,
152
especially on the error calculated in the output layer as reported by Fauselt (1994) and
Patterson (1996).
Among all the models, the back propagation network (BPN) is generally found successful
structural problems have been solved using BPN training and reported in literature. Since
it is a well-established model to solve different kinds of problem, a few papers are briefly
described below.
Hajela and Berke (1992) used BPN to determine the displacement and stress response in
static analysis using force displacement analysis. Murkhejee and Despande (1995)
explored the application of ANNs in the design of a single span reinforced concrete beam.
The inputs were the span of the beam, type of steel, grade of concrete and applied loads,
where as outputs were area of tensile steel, width and depth of beam and the moment
capacity of beam. VanLuchene and Sun (1990) used feed forward BPN to simulate the
structural analysis of a simply supported rectangular plate. Jenkins (1995) applied BPN to
a simple six-storey structural steel frame with rigid joints to approximate structural
analysis. Tang (1996) studied the application of a feed forward ANN with an adaptive
Bhatnagar et al. (1997) explored the possibility of the application of feed forward, non-
recurrent and multilayred BPN to carry out an energy efficient building design. The input
parameters were thermal transmittance, time lag and decrement factor, where as output
153
parameters were the material and section properties of the building elements. Deng et al.
(2005) presented ANN based reliability analysis methods i.e. ANN based Monto-Carlo
simulation, first order reliability methods and second order reliability methods. They used
functions.
Training algorithms are the algorithms, which use the gradient of the performance
function to adjust the weights, minimizing the performance. The gradient is determined
backwards through the network. Hagan and Menhaj (1994) derived the back-propagation
In the basic back-propagation training algorithm, the weights are moved in the direction
updates the network weights and biases in the direction in which the performance
function decreases most rapidly - the negative of the gradient. One iteration of this
where x k is a vector of current weights and biases, g k the current gradient and α k the
learning rate. Training is of two types: batch training and conjugate gradient training.
154
5.2.4 Application of ANN on Tensegrity Structures
Though ANNs are widely used in predicting the target outputs in structural analysis, their
application on tensegrity structure are very much limited. Domer et al. (2003) combined
They used Stuttgart neural net simulator (SNNS) to calculate the nodal displacements of
tensegrity structures using various activation functions. Panigrahi et al. (2005) used ANN
MATLAB for finding the coordinates of the tensegrity structure using force density
method and used the data for training and testing of the network based on back
propagation algorithm.
The basic components of a neural network, such as processing elements, inputs and
outputs, weighting factors, input functions, activation functions, and the learning
A single neuron evaluates the input signals, determines the strength of each one,
calculates a total for the combined input signals, compares the total to some threshold
level and finally determines the output. All the input signals reach the processing element
threshold level. The weighting factors of the connection are filters in the form of
multiplicative coefficients of the output sent from one processor to another, and may
155
A single processing element can be connected with another processing element to make a
layer of the processing elements (or nodes). Inputs can be connected to several nodes
with various weights. Several layers can be interconnected to get multiple layers. The
layer, which receives the inputs, is called the input layer. The network outputs are
generated from the output layer. Any other layers are called hidden layers because they
are internal to the network and have no direct contact with the external environment. A
network may consist of zero to several hidden layers. The output signal from a node may
be passed on as an input to other nodes, or even sent back as an input to itself. When
output node is not an input node on the same layer or the preceding layer, the network is
called as a forward network. When outputs of a node are directed back as inputs to
previous or same level nodes, the network is called as feedback network. Input/output is
a vector consisting of the values for inputs/ outputs. Input and output patterns should be
normalized to improve the proportions of the problem domain and the validity of the
outputs. The delta rule is based on continuously modifying the strength of the
connections to reduce the difference (the delta), or the error, between the desired output
and the current output of a processing element. The rule is also referred as a least mean
square learning rule, because it minimizes the mean squared error (MSE).
Generally, two transfer functions, the input function and the activation function operate
over the inputs I1, I2………In, as shown in Fig. 5.2. The input signals are first processed
by the input function (generally the summation function) and the results netj (t) are then
activated by an activation function and /or output, which produce the output signal.
156
Though in the beginning, only sigmoid activations were used, now a range of different
activation functions, each suited for a different problem, are in use. Some of commonly
1
F (x) = f (x) = [
− Net (t ) + b ] (5.4)
1+ e j
[
f ( x ) = Tanh Net j (t ) ] (5.5)
I1
W1
NET
I2 OUT
W2
=F (NET)
Wn
ARTIFICAL NEURON
Figure 5.2 Artificial Neuron with input Function and Output Function
157
f (x)
x
Figure 5.3 Sigmoid function
Quite often, depending upon the requirement, the input function and output neuron may
The above components are the variables in a neural network training system and lead to
various types of networks. In the present study, a multilayerd feed forward back
158
5.3.2 Back-Propagation Neural Networks
A back propagation neural network is often constructed from three or more layers of
neurons; one input layer, one output layer and one or more hidden layers. The first phase
of operation of the feed forward-back propagation ANN is called “feed forward”. During
this operation, for the given input vector I (t ) , for i th training pattern, the output H h (t ) of
H h (t ) = f [Net h (t )] (5.7)
Where,
Oo (t ) = f [Net o (t )] (5.9)
Where,
⎡ ⎤
Net o (t ) = ∑ Woh ⋅ f ⎢∑ Whi (t ) ⋅ I i (t )⎥ (5.10)
h ⎣ i ⎦
The second phase of operation is called the “error back propagation”. The error function,
E (o) is defined by the sum of squares of the difference between the desired output,
E (O ) =
1
∑ [To (t ) − Oo (t )]2 (5.11)
2
159
Where, the summation extends over the number of training patterns. For the hidden layer
to output layer connections, the adaptive rule of the weight Woh can be determined by
where,
∂E (O)
ΔWoh = −η = −η ∑ Δ o (t ).H h (t ) (5.13)
∂Woh h
df ( Neto )
Δ o (t ) = [To (t ) − Oo (t )] (5.14)
dNeto
the coefficient, η is called the learning rate. Similarly, the adaptive rule for the input layer
∂E (o)
ΔWhi = −η = −η ∑ Δ h (t ).I i (t ) (5.15)
∂Whi h
df ( Neth )
Δ h (t ) =
dNeth o
∑Who .Δo (t ) (5.16)
Applying the differentiation process successively, the error back propagation rules can
be extended to the networks with any number of hidden layers. Their weights will be
160
continuously adjusted until the outputs to reach to a desired accuracy. The accuracy of
the trained network depends upon 1) number of hidden layers; 2) number of neurons in a
From practical experience, the main criterion for design of the tensegrity structures is the
deflection control. The deflection of the structure mainly depends upon the cross
sectional area of the tension members (mild steel cable), the cross sectional area of
compression members (galvanized iron pipes), grid size, height of the structure, loads and
supports provided on periphery of the grid. The influence of these parameters has
algorithm was used in MATLAB 7 environment to obtain the cross sectional area of
cable. The cross sectional area of the strut was kept constant. The input consist of five
neurons i.e. load/m2, the grid size, the height of the structure, the support condition and
the deflection. The output was one node i.e. cross sectional area of cable. For training and
testing, data were generated by static analysis of different grids using FEM. The load
considered was 400 N/m2 with a step of 5 N/m2. The grid size was taken as 4m, 6m and
8m and the height of the structure was varied from 0.5m to 0.8 m with an interval of 0.1m.
Keeping the strut area fixed, the cross sectional area of cable was obtained corresponding
to the rigidity ratio of 10, 20, 30, 40 and 50 (i.e. 34.514, 17.252, 11.505, 8.629 and 6.903
mm2 respectively). In first case, the structure was supported on four corners with three
degrees of translation in each support locked. In input file, this condition was denoted by
161
0. In second case, keeping all other conditions same as in previous case; additional
support was provided on bottom central nodes lying on periphery of the grid structure. In
input file, this condition was denoted by 1.The maximum deflection value on the central
bottom node was found from FEM analysis. For testing, the data were generated for a
load step of 6.25 N/m2 with all other conditions as done in case of training. Also, the test
data were generated for 5m-grid size that is not included in training. The maximum
deflection value considered was span/100. Values higher than span/100 were ignored for
The architecture used for the present study is shown in Fig. 5.4. One hidden layer with 17
neurons was found to give the best results after a number of trials. All data were
normalized between 0 and 1. The input data were shuffled randomly before training. Log
sigmoid was used as activation function for input layer and hidden layer, where as purelin
is used for output layer. The mean square error is shown in Fig. 5.5.
The proposed network was trained for 8567 data and tested for 7409 data. The validations
of 178 typical data are shown in Table 5.1. The percentage of deviation for the cross
sectional area of cable between the predicted ANN value and the target value was
calculated. The pattern test is shown in Fig. 5.6. The pattern test shows the percentage
deviation lies within 10%, which is insignificant. In most cases, the deviation percentage
is less than 1%. However, it is also observed that percentage of deviation is as high as
40% in case of very small loads, where deflection is also very small i.e. less than 0.6 mm.
Further, these small deflections are insignificant for design of tensegrity structures and
162
can be ignored. The network also predicted very accurately for 5m-grid size for which
2
1
3
2
3
1
5
17
163
Figure 5.5 Mean Square Error of the network
Table 5.1 Validation of trained and tested data for tensegrity grid structure
Area
Target Area Predicted
Load Height Grid Support Deflection of Cable by ANN
2 2
(N/m ) (m) Size (m) Condition (mm) (mm ) (mm2) Error (%)
6.25 0.5 4 0 0.0863451 34.514 32.334 6.317
12.5 0.5 4 0 0.1725306 34.514 34.256 0.747
18.75 0.5 4 0 0.2587435 34.514 34.417 0.280
31.25 0.5 4 0 0.4312518 34.514 33.927 1.702
37.5 0.5 4 0 0.5175472 34.514 33.878 1.843
43.75 0.5 4 0 0.6038702 34.514 33.956 1.616
368.75 0.5 4 0 5.1309942 34.514 34.671 -0.456
164
381.25 0.5 4 0 5.3066319 34.514 34.775 -0.756
387.5 0.5 4 0 5.3944933 34.514 34.832 -0.921
237.5 0.6 4 0 2.507334 34.514 34.476 0.111
243.75 0.6 4 0 2.5711216 34.514 34.424 0.262
256.25 0.6 4 0 2.698735 34.514 34.326 0.544
262.5 0.6 4 0 2.7625607 34.514 34.283 0.670
268.75 0.6 4 0 2.8263992 34.514 34.243 0.784
293.75 0.8 4 0 2.1290113 34.514 34.586 -0.209
306.25 0.8 4 0 2.2090941 34.514 34.620 -0.307
312.5 0.8 4 0 2.2491414 34.514 34.642 -0.370
318.75 0.8 4 0 2.2891925 34.514 34.666 -0.442
331.25 0.8 4 0 2.3693064 34.514 34.724 -0.609
337.5 0.8 4 0 2.4093692 34.514 34.757 -0.703
343.75 0.8 4 0 2.4494359 34.514 34.791 -0.804
356.25 0.8 4 0 2.5295809 34.514 34.865 -1.018
343.75 0.6 4 0 6.8975751 17.257 17.543 -1.658
356.25 0.6 4 0 7.144149 17.257 17.529 -1.578
362.5 0.6 4 0 7.2675061 17.257 17.523 -1.541
368.75 0.6 4 0 7.3909101 17.257 17.517 -1.506
381.25 0.6 4 0 7.6378586 17.257 17.506 -1.443
387.5 0.6 4 0 7.7614032 17.257 17.501 -1.415
393.75 0.6 4 0 7.8849947 17.257 17.497 -1.389
368.75 0.8 4 0 5.0230756 17.257 17.402 -0.843
381.25 0.8 4 0 5.176959 17.257 17.407 -0.868
387.5 0.8 4 0 5.2539218 17.257 17.409 -0.878
393.75 0.8 4 0 5.3308987 17.257 17.410 -0.887
6.25 0.5 4 0 0.2422417 11.505 12.455 -8.256
12.5 0.5 4 0 0.4837506 11.505 10.731 6.731
18.75 0.5 4 0 0.7254726 11.505 11.108 3.453
31.25 0.5 4 0 1.2095571 11.505 11.520 -0.133
37.5 0.5 4 0 1.4519202 11.505 11.446 0.514
43.75 0.5 4 0 1.6944977 11.505 11.326 1.552
56.25 0.5 4 0 2.1802974 11.505 11.129 3.264
193.75 0.7 4 0 4.8477919 11.505 11.451 0.473
200 0.7 4 0 4.98787 11.505 11.453 0.449
165
206.25 0.7 4 0 5.1280005 11.505 11.455 0.434
212.5 0.7 4 0 5.2681834 11.505 11.455 0.431
218.75 0.7 4 0 5.4084188 11.505 11.455 0.438
225 0.7 4 0 5.5487066 11.505 11.453 0.456
381.25 0.8 4 0 10.133926 8.629 8.298 3.833
387.5 0.8 4 0 10.284887 8.629 8.285 3.984
393.75 0.8 4 0 10.435902 8.629 8.272 4.138
381.25 0.8 4 0 12.60629 6.903 6.803 1.444
387.5 0.8 4 0 12.794423 6.903 6.788 1.672
393.75 0.8 4 0 12.982638 6.903 6.772 1.902
365 0.5 5 0 11.643194 34.514 38.109 -10.416
370 0.5 5 0 11.805695 34.514 38.078 -10.326
375 0.5 5 0 11.968278 34.514 38.049 -10.242
380 0.5 5 0 12.130945 34.514 38.022 -10.165
385 0.5 5 0 12.293695 34.514 37.998 -10.094
380 0.5 5 0 23.743091 17.257 17.688 -2.499
385 0.5 5 0 24.067549 17.257 17.702 -2.576
390 0.5 5 0 24.392335 17.257 17.717 -2.668
395 0.5 5 0 24.71745 17.257 17.736 -2.773
400 0.5 5 0 25.042894 17.257 17.756 -2.893
245 0.5 5 0 22.546944 11.506 11.696 -1.652
250 0.5 5 0 23.024003 11.506 11.679 -1.502
255 0.5 5 0 23.501772 11.506 11.663 -1.362
260 0.5 5 0 23.980253 11.506 11.648 -1.232
210 0.8 5 0 15.637048 6.903 7.240 -4.884
215 0.8 5 0 15.972802 6.903 7.218 -4.563
220 0.8 5 0 16.308777 6.903 7.195 -4.228
225 0.8 5 0 16.644974 6.903 7.171 -3.882
230 0.8 5 0 16.981392 6.903 7.146 -3.526
6.25 0.5 6 0 1.2534991 34.514 35.026 -1.485
12.5 0.5 6 0 1.6865116 34.514 34.412 0.296
18.75 0.5 6 0 2.1200292 34.514 34.606 -0.267
356.25 0.8 6 0 11.709202 34.514 34.165 1.010
362.5 0.8 6 0 11.901065 34.514 34.144 1.072
368.75 0.8 6 0 12.092992 34.514 34.125 1.128
166
231.25 0.6 6 0 23.34754 17.257 17.450 -1.117
237.5 0.6 6 0 23.963705 17.257 17.459 -1.171
243.75 0.6 6 0 24.580714 17.257 17.472 -1.245
156.25 0.7 6 0 18.274996 11.505 11.645 -1.213
162.5 0.7 6 0 18.950553 11.505 11.622 -1.019
168.75 0.7 6 0 19.626979 11.505 11.601 -0.833
381.25 0.8 6 0 34.491235 11.505 11.422 0.720
387.5 0.8 6 0 35.040063 11.505 11.428 0.667
393.75 0.8 6 0 35.589396 11.505 11.435 0.612
31.25 0.7 6 0 6.349604 8.629 8.684 -0.639
37.5 0.7 6 0 7.2179397 8.629 8.795 -1.919
43.75 0.7 6 0 8.0877088 8.629 8.879 -2.901
231.25 0.8 6 0 28.262938 8.629 8.505 1.443
237.5 0.8 6 0 28.976784 8.629 8.495 1.555
243.75 0.8 6 0 29.691484 8.629 8.486 1.662
206.25 0.8 6 0 31.524494 6.903 6.742 2.336
212.5 0.8 6 0 32.412781 6.903 6.742 2.331
218.75 0.8 6 0 33.302385 6.903 6.744 2.311
381.25 0.6 8 0 56.451772 34.514 34.579 -0.187
387.5 0.6 8 0 57.425168 34.514 34.573 -0.171
393.75 0.6 8 0 58.400404 34.514 34.567 -0.154
6.25 0.7 8 0 1.2050416 34.514 32.705 5.240
12.5 0.7 8 0 1.8641808 34.514 33.051 4.239
18.75 0.7 8 0 2.5240377 34.514 33.363 3.335
31.25 0.5 8 0 12.114203 17.257 17.563 -1.774
37.5 0.5 8 0 14.577695 17.257 17.529 -1.578
43.75 0.5 8 0 17.055327 17.257 17.437 -1.044
337.5 0.7 8 0 74.732595 17.257 17.414 -0.912
343.75 0.7 8 0 76.184305 17.257 17.432 -1.015
356.25 0.7 8 0 79.098293 17.257 17.469 -1.229
81.25 0.6 8 0 58.701487 6.903 7.405 -7.276
87.5 0.6 8 0 63.373213 6.903 7.406 -7.291
93.75 0.6 8 0 68.08762 6.903 7.370 -6.769
31.25 0.5 4 1 0.1017029 34.514 34.760 -0.713
37.5 0.5 4 1 0.1220414 34.514 34.529 -0.044
167
43.75 0.5 4 1 0.1423803 34.514 34.324 0.550
381.25 0.6 4 1 0.9275241 34.514 34.990 -1.380
387.5 0.6 4 1 0.9422408 34.514 35.033 -1.504
393.75 0.6 4 1 0.9569578 34.514 35.076 -1.630
231.25 0.8 4 1 0.4226226 34.514 34.335 0.520
237.5 0.8 4 1 0.4317956 34.514 34.335 0.520
243.75 0.8 4 1 0.4409688 34.514 34.338 0.511
131.25 0.5 4 1 1.211565 11.505 11.282 1.935
137.5 0.5 4 1 1.2692712 11.505 11.298 1.801
143.75 0.5 4 1 1.3269815 11.505 11.310 1.695
56.25 0.5 4 1 0.6867646 8.629 8.478 1.744
62.5 0.5 4 1 0.762975 8.629 8.392 2.745
68.75 0.5 4 1 0.8391925 8.629 8.331 3.453
331.25 0.8 4 1 2.1463845 8.629 8.740 -1.286
337.5 0.8 4 1 2.1805337 8.629 8.741 -1.303
343.75 0.8 4 1 2.2146848 8.629 8.742 -1.308
81.25 0.6 6 1 1.202418 34.514 34.704 -0.551
87.5 0.6 6 1 1.2802106 34.514 34.651 -0.397
93.75 0.6 6 1 1.3580068 34.514 34.606 -0.268
306.25 0.7 6 1 3.1024127 34.514 34.699 -0.535
312.5 0.7 6 1 3.1616435 34.514 34.654 -0.405
318.75 0.7 6 1 3.220877 34.514 34.606 -0.266
6.25 0.6 6 1 0.3994044 17.257 17.726 -2.715
12.5 0.6 6 1 0.5476056 17.257 17.164 0.538
18.75 0.6 6 1 0.69582 17.257 17.389 -0.766
381.25 0.8 6 1 5.7908522 17.257 16.454 4.654
387.5 0.8 6 1 5.8804561 17.257 16.419 4.857
393.75 0.8 6 1 5.970067 17.257 16.385 5.052
31.25 0.6 6 1 1.3896711 11.505 11.923 -3.633
37.5 0.6 6 1 1.6079397 11.505 11.933 -3.719
43.75 0.6 6 1 1.8262365 11.505 11.913 -3.546
231.25 0.8 6 1 5.3088529 11.505 11.594 -0.774
237.5 0.8 6 1 5.440441 11.505 11.584 -0.687
243.75 0.8 6 1 5.5720443 11.505 11.572 -0.583
31.25 0.5 6 1 2.2478359 8.629 9.344 -8.283
168
37.5 0.5 6 1 2.6517597 8.629 9.274 -7.470
381.25 0.8 6 1 11.144234 8.629 8.317 3.610
387.5 0.8 6 1 11.318543 8.629 8.298 3.838
393.75 0.8 6 1 11.492878 8.629 8.279 4.059
281.25 0.8 6 1 10.351028 6.903 6.929 -0.376
287.5 0.8 6 1 10.567073 6.903 6.910 -0.098
293.75 0.8 6 1 10.783159 6.903 6.890 0.184
206.25 0.6 8 1 14.85578 17.257 17.258 -0.004
212.5 0.6 8 1 15.303345 17.257 17.217 0.230
218.75 0.6 8 1 15.750944 17.257 17.174 0.479
6.25 0.5 8 1 0.9404795 11.505 10.788 6.235
12.5 0.5 8 1 1.8783775 11.505 10.981 4.559
18.75 0.5 8 1 2.8162622 11.505 11.923 -3.632
156.25 0.5 8 1 23.444539 11.505 11.340 1.431
162.5 0.5 8 1 24.38187 11.505 11.355 1.306
281.25 0.8 8 1 18.12455 11.505 11.150 3.082
287.5 0.8 8 1 18.519323 11.505 11.134 3.221
293.75 0.8 8 1 18.914165 11.505 11.122 3.327
381.25 0.6 8 1 67.503113 6.903 6.951 -0.690
387.5 0.6 8 1 68.61042 6.903 6.922 -0.281
393.75 0.6 8 1 69.717871 6.903 6.895 0.111
6.25 0.7 8 1 1.3239006 6.903 6.688 3.109
12.5 0.7 8 1 2.148066 6.903 6.442 6.673
18.75 0.7 8 1 2.9724501 6.903 6.670 3.376
381.25 0.8 8 1 40.510097 6.903 7.175 -3.935
387.5 0.8 8 1 41.168259 6.903 7.226 -4.683
393.75 0.8 8 1 41.826608 6.903 7.281 -5.473
169
8.000
6.000
4.000
2.000 Key data points
Deviation (%)
0.000
-2.000 0 50 100 150 200
-4.000
-6.000
-8.000
-10.000
-12.000
The comparison between the target value of cable cross sectional area and the same
predicted by ANN output for 8m×8m grid size with intermediate supports along the
periphery is shown in Figs. 5.7, 5.8 and 5.9 by varying the height of structure, the loading
intensity and deflection respectively. The same is also repeated for 4m×4m grid size
without intermediate supports along the periphery and is shown in Figs. 5.10, 5.11 and
5.12 by varying the loading intensity, the height of structure and deflection respectively.
The data for 8m×8m grid and 4m×4m grid are trained in neural network, where as the
data for 5m×5m grid without intermediate supports along the periphery are tested using
the same network. Fig 5.13 and 5.14 shows the variation between the target value of
cable cross sectional area and the same predicted by ANN output for 5m×5m grid by
varying the deflection and the loading intensity respectively. A good match is found
170
between the target and the predicted value in all the cases including the grid size, which
was not trained. This shows that the optimal design of tensegrity structures can be
18
16 Target value
Cable area (mm2)
ANN output
14
12
10
6
0.4 0.5 0.6 0.7 0.8 0.9
Height (m)
Figure 5.7 Comparison between target value and ANN output for
8m×8m grid for same loading intensity and deflection
171
20
18
Target value
Load (N/m2)
40
35
Target value
Cable area (mm2)
30
ANN output
25
20
15
10
5
0
0 10 20 30 40 50 60
Deflection (mm)
172
20
18 Target value
Cable area (mm2)
16 ANN output
14
12
10
8
6
4
100 150 200 250 300 350
Load (N/m2)
20
Target value
15
Cable area (mm2)
ANN output
10
0
0.4 0.5 0.6 0.7 0.8 0.9
Height (m)
173
40
35
Target value
40
35
Target value
Cable area (mm2)
30 ANN output
25
20
15
10
5
0
0 10 20 30 40 50
Deflection(mm)
174
40
35
Target value
20
15
10
0
0 100 200 300 400
Load (N/m2)
A new design approach for tensegrity grids for roof structures using ANN has been
presented. The ANN was trained from the data generated by analysing different
tensegrity based 3D grid structure, using FEM, by varying the loads, heights, rigidity
ratios and the support conditions. Using log sigmoid as activation function for one hidden
layer with 17 neurons predicts the target cable cross sectional area very accurately.
Testing of the ANN shows good performance, with the error in reasonable limits.
The tensegrity structures are well known to exhibit non-linear behaviour resulting from
geometric non-linearity. Hence, the ability of the ANN to model tensegrity structures
175
the present study, the ANN can also be trained for other configurations. Use of ANN can
reduce the analysis and design effort and thus can be a valuable aid to the structural
engineers.
The next chapter will describe monitoring of tensegrity structures using advance sensors.
176
CHAPTER 6
6.1 INTRODUCTION
Structures are constructed to serve the purpose for which they are intended throughout
their design life span. After construction, structures need evaluation periodically to detect
any incipient damage so as to save life and property of the public. But localized damages
are likely to occur in the structure due to normal usage, natural calamities (such as
section or exfoliation. Farrar and Jauregui (1998) defined four distinct objectives of
severity and the remaining useful life of the structure. The main aim of monitoring
structures is to enable them monitor their own integrity while in operation and throughout
The various techniques available till date can be classified into following categories.
177
¾ Global techniques
testing, magnetic field analysis, penetrant dye testing and X-ray analysis)
¾ Techniques using smart materials and smart system concept (optical fibres, piezo-
sensors that are passive and bulky. Further, conventional sensors do not provide direct
information about damages and rather extract secondary information such as load and
strain history (Giurgiutiu et al., 2000). With the growth of technology, smart materials
and smart system concept becomes the centrepiece for structural health monitoring
Brownjohn et al (2003) monitored highway bridges using strain monitoring system and
accelometers for one month to study their performance. In the present study, piezo
sensors and strain gauges are used for monitoring of tensegrity structures.
The word ‘piezo’ is derived from a Greek word meaning pressure. Pierre and Paul-
178
centro symmetric crystals, such as quartz (SiO2), Lithium Niobate (LiNbO3), PZT [Pb
(Zr1-xTix) O3)] and PLZT [(Pb1-xLax)(Zr1-yTiy) O3)], in which electric dipoles (and hence
surface charges) are generated when the crystals are loaded with mechanical
deformations. The same crystals also exhibit the converse effect; that is, they undergo
mechanical deformations when subjected to electric fields. Among the various types of
low-cost, small size and good dynamic performance. Besides, it exhibits large range of
linearity, fast response, long-term stability and high-energy conversion efficiency. The
PZT patches can be manufactured in any shape, size and thickness at relatively low-cost
as compared to other smart materials and can be easily used over a wide range of
due to brittleness and low tensile strength. Due to high stiffness, the PZT sheets are good
actuators. They also exhibit high strain coefficients, due to which they can act as good
sensors also. These features make the PZT materials very suitable for use as actuators and
sensors.
The EMI technique is very similar to the conventional global dynamic response
techniques. The major difference is only with respect to the frequency range employed,
which is typically 30-400 kHz in EMI technique, against less than 100 Hz in the case of
the global dynamic techniques. In the EMI technique, a PZT patch is bonded to the
surface of the monitored structure using a high strength epoxy adhesive, and electrically
179
excited via an impedance analyzer. In this configuration, the PZT patch essentially
behaves as a thin bar undergoing axial vibrations and interacting with the host structure,
as shown in Fig. 6.1(a). The PZT patch-host structure system can be modeled as a
mechanical impedance (due the host structure) connected to an axially vibrating thin bar
(the patch), as shown in Fig. 6.1(b). The patch in this figure expands and contracts
3 (z) 2 (y)
1 (x) Alternating electric
PZT Patch field source 3
2
l l Host E3
1
structure
Point of Z w Z
mechanic h
al fixity Structural
l l
Impedance
(a) (b)
Figure 6.1 Modeling PZT-structure interaction (Bhalla, 2004)
(a) A PZT patch bonded to structure under electric excitation
(b) Interaction model of PZT patch and host structure
The patch has half-length ‘l’, width ‘w’ and thickness ‘h’. The host structure is assumed
sectional properties (area and moment of inertia) lumped along their neutral axes.
Therefore, the vibrations of the PZT patch in direction ‘2’ can be ignored. At the same
time, the PZT loading in direction ‘3’ is neglected by assuming the frequencies involved
180
to be much less than the first resonant frequency for thickness vibrations. The vibrating
patch is assumed infinitesimally small and to possess negligible mass and stiffness as
compared to the host structure. The structure can therefore be assumed to possess
uniform dynamic stiffness over the entire bonded area. The two end points of the patch
can thus be assumed to encounter equal mechanical impedance, Z, from the structure, as
shown in Fig. 6.2 (b). Under this condition, the PZT patch has zero displacement at the
mid-point (x= 0), irrespective of the location of the patch on the host structure.
Using the impedance approach of Liang et al. (1994), the electromechanical admittance
(the inverse of electro-mechanical impedance) was derived across the PZT patch as
(Bhalla, 2004)
wl ⎡ T ⎛ Za ⎞ 2 E ⎛ tan κl ⎞⎤
Y = 2ωj ⎢(ε 33 − d 31 Y ) + ⎜⎜ ⎟⎟d 31 Y ⎜
2 E
⎟⎥ (6.1)
h ⎣ ⎝ Z + Za ⎠ ⎝ κl ⎠ ⎦
damage detection. Z is a function of the structural parameters- the stiffness, the damping
and the mass distribution. Any damage to the structure will cause these structural
parameters to change, and hence alter the drive point mechanical impedance Z. Assuming
that the PZT parameters remain unchanged, the electromechanical admittance Y will
undergo a change and this serves as an indicator of the state of health of the structure.
Measuring Z directly may not be feasible, but Y can be easily measured using any
181
6.3.2 Details of PZT Patches
In the EMI technique, the same PZT patch serves the actuating as well as the sensing
functions. Fig. 6.2 shows a typical commercially available PZT patch suitable for this
particular application (PI Ceramic, 2007). The characteristic feature of the patch is that
the electrode from the bottom edge is wrapped around the thickness, so that both the
electrodes are available on one side of the PZT patch, while the other side is bonded to
the host structure. PZT patches of sizes ranging from 5mm to 15mm and thickness from
0.1mm to 0.3mm are best suited for most structural materials such as steel and RC. Such
thin patches usually have thickness resonance frequency of the order of few MHz. Hence,
the frequency response signature is relatively flat in 30-400 kHz frequency range. PZT
actuator/ sensor patch is bonded to the surface of the structure (whose health is to be
monitored) using high strength epoxy adhesive. The conductance signature of the patch is
acquired over a high frequency range (30-400 kHz). This signature forms the
benchmark for assessing the structural health. At any future point of time, when it is
desired to assess the health of the structure, the signature is extracted again and compared
10mm
10mm
182
6.4 STRAIN GAUGES
Strain gauges are very versatile sensors for structural behaviour monitoring. Strains on
structural surfaces are caused by complex member deformations, resulting from bending,
torsion, shearing and elongation/ contraction. Hence, strain measurements can capture an
element’s behaviour quite well (Sanayei and Saletnik, 1996). For efficient performance, it
is desirable that the gauge should be stable with respect to both time and temperature. In
addition, it should have minimum dimensions and inertia and should exhibit linear
response over the strain range of interest (Dally et al., 1984). Commercial strain gauges
VWSG consists of a pretensioned stainless steel wire whose ends are fixed to lugs that
are spot welded to the monitored component. A sensor coil determines the frequency of
the wire, which is proportional to the strain in the component. VWSGs are structurally
strong and quite robust for use in underground structures. They are especially suitable for
long term monitoring since vibration wires do not undergo any decay with time
(Oosterhout, 2003). They can be easily spot welded to the reinforcement bars of the
for underground structures, primarily based on VWSGs, spanning five years. This clearly
183
demonstrates the robustness and longevity of VWSGs. The main drawback of VWSGs is
that they are only suitable for measuring static strains, susceptible to extraneous noise in
the form of ambient vibrations and need special protection to prevent damage from
ESGs are based on the principle that under mechanical stress, the electrical resistance of a
conductor changes in proportion to the load induced strain. They essentially consist of
thin metallic foil grids, bonded to a thin, tough and flexible polyimide plastic film, which
can be adhesively bonded to the surface of the monitored component, as shown in Fig.
6.3.(a). When the structural component is loaded, its strain is transferred to the foil grid
and its resistance changes accordingly. The relative change in resistance, ∆R/R of the
ΔR
= Sgε (6.2)
R
where Sg is called the gauge factor or the calibration constant of the ESG. The output
∆R/R of a strain gauge is converted into voltage signal by means of a Wheatstone bridge
circuit, as shown in Fig. 6.3 (b), for which the output voltage Vo is given by
1 ⎛ ΔR ⎞ 1
Vo = Vi ⎜ ⎟ = Vi S g ε (6.3)
4 ⎝ R ⎠ 4
ESGs need considerable care during installation due to their fragile nature. Further,
electrical noise is very frequently associated with ESGs since the output voltage from a
Wheatstone bridge is of the order of few milli volts only. Fortunately, electrical noise can
184
(Dally et al., 1984). In addition, ESGs are very prone to deterioration by water. This
problem was experienced during the monitoring of rock bolts in the underground caverns
in Singapore (Zhao et al., 1999, 2002). Hence, they must be properly sealed if used in the
general, ESGs tend to be less stable as compared to VWSGs over long periods of time.
Because of this reason, majority of the long-term studies reported in the literature have
Metal grids
ESG, R
Ro
Voltage
Polyimide
Vo recording
device
plastic film
Ro Ro
Vi
(a) (b)
Optical fibres, which are thin fibres (few μm to few hundred μm in diameter) of glass and
silica, utilize fibre properties to generate optoelectronic signals indicative of the external
structures by Moyo (2002) through static and dynamic tests indicated good possibility of
185
its use on civil structures. FBG sensors are small, lightweight, corrosion resistant and
durable. VWSGs and ESGs require cables for recording data, which, for long distance
monitoring, suffer from electro-magnetic interference and electrical noise. FBG sensors,
on the other hand, are immune to EMI and can be multiplexed, thus eliminating long
cables. However, they are very fragile as compared to VWSGs and ESGs. For this
reason, efforts to install FBG sensors on civil structures often result in high rate of sensor
failure due to the harsh environment in the construction industry (e.g. Storoy et al.,
1997). The measurement system and the sensors themselves are relatively expensive as
6.5 ACCELEROMETERS
response can be either harmonic (e.g. dynamic vibration tests) or transient (e.g.
to a base (which is attached to the host structure) by means of a spring and a viscous
damper. The working principle of an accelerometer is depicted in Fig. 6.4, which shows
the seismic mass m connected to a base mass mb by a spring (of spring constant k) and a
damper (with a damping constant c). Let x, x& and &x& denote the displacement, velocity
and acceleration respectively of the base and y, y& and &y& represent the corresponding
terms for the seismic mass. If the base motion is harmonic in nature, i.e. x = xo e jωt , the
186
mω 2 xo
z= e j (ωt −φ ) (6.4)
k (1 − r ) + (2rξ )
2 2 2
where r denotes the ratio of the excitation frequency and the natural frequency, i.e.
of z behind x, as given by
2rξ
tan φ = (6.5)
1− r2
F(t)
y(t)
m Seismic
mass
k c
x(t)
mb Base mass
⎛ − e − jφ ⎞
z=⎜ ⎟ &x& (6.6)
⎜ ω 2 (1 − r 2 ) 2 + (2rξ ) 2 ⎟
⎝ n ⎠
where ω n = k / m is the natural frequency of the transducer system. For the limiting
case of very small frequency ratios (i.e. ω << ω n ⇒ r →0 and φ→0), Equation 6.6 can be
simplified as
187
&x& ≈ − zω n2 (6.7)
Hence, the acceleration of the host system can be determined by measuring the relative
displacement, z, between the seismic mass and the base. In order to ensure the necessary
condition ω << ω n , the system has to be equipped with very stiff springs and a small
displacement sensors due to their small mass, high stiffness and low damping. The main
In this research, electrical strain gauges (ESGs), LVDTs and piezo sensors were used for
monitoring the tensegrity structure. Piezo sensors were used in place of accelometers for
dynamic response measurement due to their small size and higher signal to noise ratio, as
reported by Bhalla et al. (2006). ESGs are used for measuring the level of prestress in the
struts and the stranded cables and as well as the forces developed during the destructive
test. Application of strain gauges in cables is not reported in literature though it is very
important to measure the prestress level to analyze the structure. Further, it is already
tensegrity structure to determine the forces developed during the loading. LVDTs were
Any tensegrity structure is vulnerable to damage due to its high flexibility. Further,
tensegrity structures used for temporary shelter purpose are susceptible to damage during
their use. Hence, damage detection in tensegrity structure is essential. Piezo sensors are
188
used for this purpose. Two different principles are utilized for comprehensive monitoring
– one based on local monitoring and the other on global monitoring. Both employ the
The details of EMI technique have already been covered in earlier sections. Peairs et al.
(2004) developed a low cost electrical admittance measurement method using FFT
analyzers. Fig. 6.5 (a) show the electrical circuit used by Peairs et al.. The current I
V0
I= (6.8)
R
where V0 is the voltage across a sensing resistor R ,which is in series with the PZT patch
I V
A= = 0 (6.9)
Vi RVi
The impedance analyzer such as HP 4194 costs nearly $40,000 and is not available in all
laboratories. On the other hand, the FFT analyzer costs nearly $10000 and also available
In this research, Peairs’ method outlined above has been further improved. In this study,
function generator and oscilloscope were used in place the FFT analyzer. Agilent 33220A
function generator was employed to generate the voltage signal Vi and Agilent 54622D
189
mixed signal oscilloscope to record the output V0. The electrical circuit and used for this
study is shown in Fig. 6.5 (b). This system not only has greater accuracy than the FFT
based method, but at the same time, costs $5000 only. Oscilloscope and function
generators are generally available in most structural laboratories Fig. 6.6 shows the
complete experimental set up, including the instrument and the test specimen.
FFT
V0 (Output voltage)
Analyzer
Vi R
I=Current
(a)
Function
Generator Oscilloscope
Vi V0
(Output
R Voltage)
I=Current
(b)
Figure 6.5 (a) Circuit for measuring PZT admittance by Peairs et al. (2004)
(b) Circuit for measuring PZT admittance used for experiment
190
Mixed signal oscilloscope
Function generator
Specimen
PZT Patch
A PZT patch was bonded to the surface of the G.I. pipe (compression member) using
RS850-940 epoxy as shown in Fig. 6.6. A sinusoidal input signal of 1V r.m.s. value at a
particular frequency was generated using the function generator. It was spilt into two
parts using a T connector. One part was applied across the reference channel of the
oscilloscope and the other part was applied as V0 across the circuit shown in Fig. 6.5 (b).
The resistance R used was 20 ohms. The output voltage V0 was measured across the
resistance and fed to the test channel of the oscilloscope. The process was repeated in the
entire frequency range 80-100 kHz at 100 Hz intervals. The signature was obtained for
the undamaged and three damage cases. The pipe was damaged near the centre along the
circumference by drilling 5mm holes. Damage 1 refers to one hole, damage 2 to two
holes and damage 3 three holes. Hence, damage severity gradually increased from
damage 1 to damage 3. Fig. 6.7 shows the change in conductance (real part of
191
admittance) signature resulting from three damages. The prominent effects of damage on
the conductance signature are the appearance of new peaks in the signature, and lateral
and vertical shifts of the peaks. These changes are the general indicators of damage.
Among the various techniques available to quantify the changes occurring in the
frequency response function of structures due to damage, the root mean square deviation
(RMSD) algorithm reported to be very robust (Bhalla, 2001). The RMSD deviation in
signatures is defined as
∑ (G )
N
2
1
j − G 0j
j =1
RMSD = (6.10)
∑ (G )
N
0 2
j
j =1
where G 1j is the post-damage conductance at the jth measurement point and G 0j is the
corresponding pre-damage severity value. The RMSD histogram is shown in Fig. 6.8 and
it is evident from the chart that with increase the damage, the RMSD is increasing.
Hence, component level damage can be easily identified using the EMI technique.
Further, the EMI signatures can be calibrated for damage severity either using the RMSD
Piezo-impedance transducers do not measure any direct physical parameter like stresses,
strains or temperatures. Rather, they extract a signature of the host structure to identify
structural damages. In the present study, PZT is used as a sensor to extract the signature.
192
0.003
undamaged
Conductance (S) 0.0025 1st damage
0.002
0.0015
0.001
0.0005
80 85 90 95 100
Frequency ( kHz)
(a)
0.004
undamaged
0.0035
2nd damage
Conductance (S)
0.003
0.0025
0.002
0.0015
0.001
0.0005
80 85 90 95 100
Frequency (kHz)
(b)
0.0035
undamaged
0.003
3rd damage
Conductance (S)
0.0025
0.002
0.0015
0.001
0.0005
80 85 90 95 100
Frequency (kHz)
(c )
Figure 6.7 Change in conductance signature with damage progression
(a) Damage 1 (b) Damage 2 (c) Damage 3
193
55
53.25
RMSD IN PERCENTAGE 52.14
50
Damage 3
45 Damage 2
43.43
Damage 1
40
DAMAGE CASE
Figure 6. 8 Variation in RMSD with increasing damage severity
The direct effect (i.e. stress induced electrical) charge for piezoelectric materials, under
D 3 = d 31T1 (6.11)
where d31 is the normal strain in the direction 1 corresponding to an electric field along
the poling direction 3, T1 is the stress and D3 is the electric displacement .PZT is used as a
sensor based on the direct effect and it generates proportional charge in response when
the sensor is exposed to a stress field. Multiplying cross sectional area (A) on both side of
T1 = YS1 (6.13)
194
Q = CV (6.14)
εA
C= (6.15)
h
⎛ d Yh ⎞
V = ⎜ 31 ⎟ S1 (6.16)
⎝ ε ⎠
⎛ d Yh ⎞
Replacing ⎜ 31 ⎟ as a constant term K, Equation 6.16 can be expressed as
⎝ ε ⎠
V = KS1 (6.17)
where Q is the charge, C is the capacitance, ε is the second order dielelectric permittivity,
h is the thickness, Y is the Young’s modulus of elasticity (at constant electric field), V is
the voltage, S1 is the stress and K is a constant. Hence voltage is proportional to the stress.
A four bar tesegrity single module i.e. a halfcuboctahedron fabricated in cable mode of
deployment (Refer Fig.3.10) was placed on four concrete blocks to minimize the effect of
any external vibration source. A PZT patch was bonded in one strut near the joint as
shown in Fig. 6.9. The patch was connected to Agilient 34411A digital multimeter as
shown in Fig. 6.10 to collect the time domain analysis data. To obtain the modal
frequencies of the undamaged structure, the structure was excited by hammering to the
joint near the bonded PZT patch. Fig. 6.11 shows the response of the excited structure
and the voltage amplitude gradually dies down with time due to damping.
195
Damage in any tensegrity structure may occur either due to damage in cable or damage in
strut. In the present damage detection study, the damage was introduced in one of the
cables, near to the strut in which the PZT was bonded. The damage was induced by
cutting the wires gradually in two stages as shown in Fig. 6.12. The structure was excited
by hammering at the joint after each stage of damage and the data were recorded in the
digital multimeter in time domain. The time domain data was transferred to frequency
domain using fast Fourier transformation (FFT) in MATLAB 7.0 environment and the
plot was made voltage against frequency to compare the change in frequencies of the
196
Figure 6.10 Multimeter connected to PZT patch
0.08
0.06
0.04
0.02
0
Voltage (V)
-0.02 0 5 10 15
-0.04
-0.06
-0.08
-0.1
-0.12
-0.14
Tim e (s)
197
Fig. 6.13 shows the modal frequencies for the damaged and undamaged structure. The
peaks are marked and the corresponding modal frequencies are highlighted. It is observed
that the first modal frequency for undamaged structure occurs at 33.5Hz and the same for
the first and second level of damages occurs at 33Hz and 31.5 Hz respectively. For higher
modal frequencies, the effect of damage is more clearly evident as the modal frequency
for undamaged, first level damage and second level damage structure occurs at 134.5Hz,
130Hz and 127Hz respectively. With increase of damage, the change in frequency is
more due to relation between modal frequencies and stiffness of the structure. The
structure.
Damaged
198
Damage detection study was also conducted by introducing damage in the strut, on which
the PZT patch was bonded. The strut was cut gradually by a saw keeping all other
members intact. As before, the structure was excited by hammering at the joint in two
stages and data were collected and transferred to frequency domain. The damage in the
strut is shown in Fig. 6.14. The modal frequencies for the damaged and undamaged
structure are shown in Fig. 6.15. The peaks are marked in the plot highlighting the
corresponding frequency. These peaks correspond to the modal frequencies. The first
modal frequency for undamaged structure occurs at 31 Hz and the same for the first and
second level of damages occurs at 30.5 Hz and 24 Hz respectively. The modal frequency
for undamaged, first level damage and second level damage structure occurs at 133.5 Hz,
128 Hz and 121.5 Hz respectively. This indicates the severity of damage in the strut. Fig.
6.16 shows the change in frequency in the strut and in the cable. It is observed that
greater change in frequency occurs in case of damage in the strut than the cable. This is
The dynamic behavior of tensegrity structure is very much less understood than their
the main area of research. This section describes dynamic characterization of grid
structures using piezo sensors and evaluates the performance of damage localization
algorithm.
199
6 1
33 70.5
73.5
5 Undamaged 0.8 Undamaged
1st damage 1st damage
FFT of Voltage
4 2nd damage
FFT o f V o lta g e
2nd damage 0.6
74
3 31.5
0.4
2
33.5 0.2
1
0 0
25 30 35 65 70 75
Frequency(Hz) Frequency(Hz)
1 1
Undamaged 82.5
Undamaged
1st damage 0.8 134.5
0.8 1st damage
2nd damage
2nd damage
F F T o f V o lta g e
F F T o f V o lta g e
0.6 0.6
85.5
0.4 0.4
79 130
0.2 127
0.2
0
0
120 125 130 135 140
75 80 85 90
Frequency(Hz)
Frequency(Hz)
200
Figure 6.14 Damage induced in the strut
201
7 2
Undamaged Undamaged
6 1st damage 31
1st damage 84
2nd damage
1.5 2nd damage
5
F F T o f V o lta g e
FFT of Voltage
30.5
4
1
3 85
78.5
24
2 0.5
0
0
20 25 30 35 75 80 85 90
Frequency(Hz) Frequency(Hz)
Undamaged
1st damage
0.8
2nd damage
133.5
FFTof Voltage
0.6
121.5
0.4
128
0.2
0
115 120 125 130 135 140
Frequency(Hz)
202
8
Frequency difference(Hz)
7 Damage in cable
6 damage in strut
5
4
3
2
1
0
1 2 3
Mode
(a)
14
12 Damage in cable
Frequency difference(Hz)
Damage in strut
10
0
1 2 3
Mode
(b)
203
The experimental data was obtained exciting the tensegrity grid structure described in
Chapter 3 by means of a hammer and dynamic response data was obtained by recording
the voltage across a piezoelectric-ceramic (PZT) patch bonded on the structure using
Agilent 34411A digital multimeter. The PZT patch bonded on one of the compression
member is shown in Fig. 6.17. The experimentally identified resonance modes and the
corresponding frequencies were compared with the updated finite element based model
covered in Chapter 3. Fig. 6.18 represents the comparison of the frequency response
function (FRF) before and after damage of the structure. Destructive test for the structure
has already been described in Chapter 3 and reproduced in Fig. 6.19. The damage in the
structure changes its stiffness and hence the modal parameters such as modal frequencies
204
0.6
before damage
0.5
after damage
0.4
V oltage
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70 80
Frequency (Hz)
Figure 6.18 Signature of the grid structure obtained before and after damage
12
4
16
11
Failure of top joint
205
The modal analysis of the tensegrity structure was carried out by using the updated FEM
model for both undamaged and damaged conditions. During the testing of the grid
structure, the cable joining node 11 and 12 was detached from joint 12 due to failure of
the U hook of joint 12 as shown in Fig.6.19. This damage is introduced in the finite
element model by removing the cable joining node 11 and 12 as shown in Fig. 6.20. The
Table 6.1. It is true that there are number of peaks in Fig. 6.18. It is not possible to use
this figure in isolation. The relevant modal analysis done with the help of finite element
Member deleted
206
Table 6.1 Comparison of frequency before and after damage
5 - - 43.518 42.1
The results in Table 6.1 indicate satisfactory match between the resonance frequencies of
numerical and experimental model for both undamaged and damaged structure. First
three mode shapes obtained by numerical analysis are shown in Figs 6.21, 6.22 & 6.23.
The corresponding mode shapes of the damaged grid structure are shown in Figs. 6.24,
The change in frequency is the indicator of damage in the structure. This concept can be
implemented for monitoring tensegrity structures in real field application. FRF will be
obtained in different time intervals and from the change in frequencies, the damage will
be detected. In laboratory test, sensors were connected by wires, which may not be
207
practicable in real life structure. The solution is to adopt wireless technology to obtain
signatures.
208
Figure 6.23 Third mode of undamaged 2m×2m grid structure
209
Figure 6.25Second mode of undamaged 2m×2m grid structure
210
6.8 DAMAGE LOCATION IN TENSEGRITY STRUCTURE
Naidu and Soh (2004) used EMI technique integrated with a finite element model for
damage location identification in beams using the higher vibration modes. As the higher
modes generally corresponds to local damages and it is not practical to obtain higher
pair to find the frequency changes of the damaged structure with the help of EMI
technique. The damage location was identified by considering the natural frequency
changes obtained experimentally and the mode shapes of the undamaged structure
In this research, Naidu and Sohs’ method, originally proposed for beams, has been
modified for tensegrity structure. In the original method, the damage indicator or damage
∑ ΔE i
p Δf i
DI p = i =1
m
× 100 (6.18)
∑ Δf
i =1
i
where ΔE is the element deformation parameter, p the element number, I the number
mode number, Δf the frequency shift between the undamaged and the damaged structure
for a particular mode and m is the number of modes considered. Experimentally, seven
modal frequencies were extracted, which matched with FEM results. Then three highest
frequencies were considered. In order to adapt this method for tensegrity structures, ΔE is
replaced by nodal displacement parameter D i.e. the displacement mode shape value in
vertical direction at particular node. Further, nodes have been considered in place of the
elements. Thus, unlike original approach, the damage metric values will correspond to
211
nodes of the structure, rather than elements. Fig. 6.27 shows the damage metric values for
the different nodes of the tensegrity grid structure which was tested earlier. From this
figure, the maximum damage metric is found to at node 16. Experimentally, this node
was associated with very large deformation as evident from Fig. 6.19. The plot also
shows high DI in node 2, 4 and 13. This is also similar to some extent to the experimental
observations. However, this method is not able to identify the damaged element i.e.
element connecting node 11 and 12. Both these nodes have DIs much below the
threshold. Hence, although this method does not provide complete damage identification
in terms of member, it does give some information about damage with respect to nodes.
Further studies are required to explore this method to identify the damages in the
30
Damage metric (%)
28
26
24
22
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Node number
212
6.9 CONCLUDING REMARKS
monitoring has been presented and the possibility of monitoring tensegrity grids using
piezo-transducers has been described. A low cost method has been developed to obtain
the admittance signature of the structure using function generator and mixed signal
oscilloscope, which are commonly available in structural laboratories. The method was
found very suitable for component level damage detection. Using the updated model, the
dynamic analysis of tensegrity grid structure was carried out and checked with the
Tensegrity structures in real life can be monitored online using wireless technology and
the damage can be detected by comparing the frequencies in intervals. It is observed the
damage in both cables and struts can be detected by bonding a single PZT patch on a strut
member. The frequencies undergo greater change for damage in the strut than the cable.
Low frequency technique is suitable for detecting global damage in the single module as
well as grid structure where as high frequency technique (EMI technique) is able to detect
localized damage i.e. on individual member. Hence, it is possible to detect damage in the
tensegrity structure in field by bonding a single piezo sensor. A single piezo sensor
bonded to a strut also provides information regarding damage occurrence with respect to
nodes of the tensegrity structure. Further theoretical and experimental studies are required
213
CHAPTER 7
WORK
7.1 CONCLUSIONS
This thesis reviews the definitions, different form finding methods, static and dynamic
analysis, design and deployment of tensegrity structures in detail. Further, the main
have been highlighted. The research mainly highlights on fabrication and testing of
assemble/dismantle and does not require lifting machine and equipments. Further, no
skilled labour is required for prestressing. Hence, the proposed structural system is ideal
for field deployment. The structure requires less space for storage and is easy to transport.
The tensegrity grids reported so far in the literature require mechanization in field
application, which is likely to bring up the overall operational cost. In addition, the
proposed method reduces the number of cables. Artificial neural network is employed as
an alternate design approach for the proposed structural system. To the best knowledge of
the researcher, this is the first time such a structure has been comprehensively monitored
214
The main conclusions and contributions of this research are summarized as follows:
been fabricated both in the strut and the cable mode and tested to obtain prestress
levels, deflection at nodes and the member forces. The structures have been
analysed using FEM with the experimental prestress considered as the initial
direction only. Although the results are comparable, small error between the
To match the experimental values with numerical values, the numerical model has
been updated. The updated model is found to be different for the strut and the
cable modes. A good agreement has been obtained between the updated model
and experimental results for both the strut and the cable mode.
During experiment, the failures of the structure in both the strut and the cable
mode is due to failure of strut (due to buckling) and no sign of failure in either
joints or any cable was seen. The variation of deflection and member forces is
found to be nonlinear.
The load carrying capacity of the structure is much higher in the cable mode than
the strut mode. Further, the fabrication, assembling and dismantling is simpler and
easy in case of cable mode of arrangement. Hence, the structure in cable mode is
215
For a single module, from the parametric study, it is observed that with decrease
in rigidity ratio, the deflection decreases substantially at higher loads where as the
strut force increases. With increase in the prestress level, the deflection values
tend to decrease nonlinearly and the strut force increases. However, the relative
A 2m×2m tensegrity grid structure has been fabricated by joining four single units
cohesively and tested and analysed. The comparison of deflection and various
member forces obtained experimentally and numerically for the grid structure
show reasonable agreement. Loosening of some cables do not cause the structure
unstable.
Numerical analysis shows that stress reversal occurs in some cables at loads
has little or no effect on maximum deflection but increases the maximum strut
force. With increase in rigidity ratio, the deflection decreases substantially but has
no or little effect on the maximum strut force. Increase in the height of structure
216
The effect of temperature, lack of fit in cable and strut, live loads and their
combination on grid structure has been studied. It is found that the effect of
has marginal effect on the results. In all the analysis, no reversal of stress is found
in either the strut or the cable. All the deflections and member forces are within
permissible value.
The grid structure has been analysed for wind loads for Delhi region. An 8m grid
structure has been designed for the combination of wind load and live load of 750
N/m2.
A multilayer feed forward back propagation ANN has been proposed to design
the tensegrity grids. The training and testing data are taken from separate
analysis. Using log sigmoid as activation function for one hidden layer with 17
neurons, the network predicts the cable cross sectional area very accurately. The
pattern test shows the deviation between the target and the ANN output lies
within 10%. ANN also predicts the cross sectional area for grid size not trained.
structure.
sensors has been explored. Using function generator and mixed signal
method has been developed to obtain the admittance signature of the structure at
217
high frequencies. Further, in comparison to FFT analyser, the frequency range in
the new method is higher and the result is as accurate as impedance analyser. A
The tensegrity structure as a whole is also monitored using PZT patches for global
Damage in both cable and strut is detected by bonding a single PZT patch on a
In this research, joints have been designed and fabricated only conceptually. The
A large grid structure should be fabricated to study the behaviour of the structure
using shake table and its application in earthquake prone areas should be
explored.
218
The effect of lack of fit should be studied in detail and its limitations/ tolerance
should be determined.
Theoretical and experimental studies should be carried out for locating damaged
The erection and packaging should be automated for rapid use in industry and
219
REFERENCES
Adriaenssens S M L and Barnes M R 2001 Tensegrity spline beam and grid shell
Barnes M R 1977 Form finding and analysis of tension space structures by dynamic
Batten M, Boorman R and Leiper Q 1999 Use of vibrating wire strain gauges to measure
loads in tubular steel props supporting deep retaining walls Proceedings: Institution of
Bhalla S 2001 Smart System Based Automated Health Monitoring of Structures M.Eng.
220
Bhalla S 2004 A mechanical impedance approach for structural identification, health
University
structures National Conference on High Rise Buildings: Materials and Practices New
Delhi 563-69
systems for energy efficient building design Architectural Science Review 40 53-59
221
Brownjohn J M W 2003 Ambient vibration studies for system identification of tall
upgrading by dynamic testing and finite element model updating Journal of Bridge
Calladine C R 1978 Buckminster Fullers tensegrity structures and Clerk Maxwells rules
for the construction of stiff frames International Journal of Solids and Structures 14 161-
72
Carpenter G A and Grossberg S 1988 The ART of adaptive pattern recognition IEEE
Computer 21 77-88
finite element analysis John Wiley & Sons (ASIA) Pte Ltd Singapore
222
Connelly R and Whitely W 1996 Second order rigidity and prestress stability for
Crane C D, Duffy J and Correa J C 2005 Static analysis of tensegrity structures Journal
Communications 30 311-16
Defossez M 2004 Mechanical response of a tensegrity structure close to its integrity limit
Deng J, Gu Desheng, Li Xibing and Yue Z Q 2005 Structural reliability analysis for
implicit performance functions using artificial neural networks Structural Safety 27 25-48
Domer B, Fest E, Lalit V and Smith I. 2003a Combining dynamic relaxation method with
223
Domer B, Raphael B, Shea K and Smith I F C 2003b A study of two stochastic search
methods for structural control Journal of Computing in Civil Engineering ASCE 17 132-
41
Domer B and Smith I F C 2005 An active structure that learns Journal of Computing in
Fest E, Shea K, Domer B and Smith I F C 2003 Adjustable tensegrity structures Journal
224
Fest E, Shea K and Smith I F C 2004 Active tensegrity structures Journal of Structural
Flood I 1990 Simulating the construction process using neural networks Proceedings: 7th
Fuller R B 1962 Tensile integrity structures United States Patent 3 063 521
Publishers London
225
Gagarine N Flood I and Albrecht P 1992 Weighing trucks in motion using Gaussian-
Giurgiutiu V, Redmond J, Roach D and Rackow K 2000 Active Sensors for Health
October 90-117
Hagan M T and M Menhaj 1994 Training feed forward networks with the Marquardt
Hajela P and Berke L 1992 Neural networks in structural analysis and design: An
Hanaor A 1988 Prestressed pin-jointed structures- flexibility analysis and prestress design
Hanaor A and Liao M K 1991 Double-layer tensegrity grids: Static load response I:
226
Hanaor A 1991 Double-layer tensegrity grids: Static load response II: Experimental study
Hanor A 1994 Geometrically rigid double layer tensegrity grids International Journal of
Hanaor A and Levy R 2001 Evaluation of deployable structures for space enclosures
Asia Singapore
Hopfield J J 1982 Neural networks and physical systems with emergent collective
IS: 1239 1990 Mild Steel Tubes, Tubulars and Other Wrought Steel Fittings -
Specification - Part 1: Mild Steel Tubes Bureau of Indian Standards New Delhi
IS: 1161 1988 Steel Tubes for Structural Purposes – Specification Bureau of Indian
227
IS: 1835 1976 Specification for Round Steel Wire for Ropes Bureau of Indian Standards
New Delhi
IS: 800 1984 Code of practice for general construction in steel Bureau of Indian
IS: 875 1987 Code of practice for design loads (other than earthquake) for buildings and
IS: 875 1987 Code of practice for design loads (other than earthquake) for buildings and
IS: 3459 1977 Specification for Small Wire Ropes Bureau of Indian Standards New
Delhi
IS: 10942 2000 Ferrule- Specification Bureau of Indian Standards New Delhi
Jenkins W M 1995 Neural network based approximation for structural analysis, In:
Topping BHV (Ed.) Developments in Neural networks and Evolutionary Computing for
228
Kamarthi S Sanvido V and Kumara R 1992 Neuroform- neural network system for
Kenner H 1976 Geodesic math and how to use it Berkeley California University of
California press
tensegrity system subjected to follower wind loading Computers and Structures 81 2199-
2217
System Energy Transfer Journal of Intelligent Material Systems and Structures 5 12-20
229
Masic M, Skeleton R E and Gill P E 2005 Algebraic tensegrity form finding
2110-25
Maurin B and Motro R 1997 Density methods and minimal forms computation IASS
32
McCulloch W S and Pitts W 1943 A logical calculus of the ideas immanent in nervous
Cambridge UK Draft
230
Moselhi O, Hegazy T and Fazio P 1991 A hybrid neural network methodology for cost
International Conference, Mobile and Rapidly Assembled Structures (MARS 96) Seville
Southampton UK 101-09
Motro R, Najari S and Jouanna P 1986 Static and dynamic analysis of tensegrity systems
Motro R 2003 Tensegrity structural systems for the future Kogan Page Science UK
London
231
Murakami H 2001a Static and Dynamic analyses of tensegrity structures Part-I:
3613
Murakami H 2001b Static and Dynamic analyses of tensegrity structures, Part-II: Quasi-
Murakami H and Nishimura Y 2001a Initial shape finding and modal analyses of cyclic
Naidu A S K and Soh C K 2004 Identifying damage location with admittance signatures
627-42
232
Oppenheim I J and Williams W O 2001a Vibration of an elastic tensegrity structure
network for form finding of tensegrity structures Proceedings: 2nd Indian International
Patterson D W 1996 Artificial neural netwoks: Theory and application Prentice Hall
Singapore
Peairs Daniel M, Park G and Inman Daniel J 2004 Improving Accessibility of the
Peterson C and Anderson J R 1987 A mean field learning algorithm for neural networks
233
PI Ceramic 2007 Product Information Catalogue Lindenstrabe Germany
http://www.piceramic.de
USA
Quirant J, M N Kazi-Aoual and Motro R 2003 Designing tensegrity systems: the case of
Sakantamis K and Larsen O P 2004 The Cocoon method: A physical modeling tool for
Sanayei M and Saletnik M J 1996 Parameter estimation of structures from static strain
Schodek and Daniel L 1993 Structure in sculpture Cambridge Mass (USA) MIT press
Sheck H J 1974 The force density method for form finding and computation networks
234
Snelson K 2004 Kenneth Snelson (online) New York (USA)
Skeleton R E, Jean P P and Mingori D L 2001 Dynamics of the shell class of tensegrity
Stern I P 1999 Development of design equations for self deployable N- strut tensegrity
Storoy H, Saether J and Johannessen K 1997 Fiber optic condition monitoring during a
full scale destructive bridge test Journal of Intelligent Material Systems and Structures 8
633-43
Sultan C and Skelton R E 1998 Tendon control deployment of tensegrity structures In:
Proceedings: SPIE 5th Symposium on Smart Structures and Materials 3323 455-66
5223-52
235
Sultan C, Corless M and Skelton R E 2002b Linear dynamics of tensegrity structures
Sultan C and Skeleton R E 2004 A force and torque tensegrity sensor Sensors and
Tang Yu 1996 Active control of single degree freedom systems using artificial networks
Tibert A G and Pellegrino S 2002 Deployable tensegrity reflectors for small satellites
Tran T M 2002 Reverse displacement analysis for tensegrity structures M.S. Thesis
University of Florida
236
VanLuchene R D and Sun R 1990 Neural networks in structural engineering
Vassart N and Motro R 1999 Multi parametered form finding method application to
Vishay Micro-Measurements 2005 P O box 27777 Raleigh North Carolina 27611 USA
www.vishaymg.com
In: Proceedings of 8th International Conference on Steel, Space and Composite structures
145-52
237
Wang B B and Li Y Y 2003 Novel cable strut grids made of prisms: part1 Basic theory
Programme London
Zhang J Y and Ohsaki M 2006 Adaptive force density method for form-finding problem
Zhang J Y and Ohsaki M 2007 Stability condition for tensegrity structures International
Zhao J, Liu Q, Lee K W, Choa V and Teh C I 1999 Underground cavern development in
the Jurong sedimentary rock formation Tunnelling and Underground Space Technology
14 449-59
Zhao J, Cai J G, Tunbridge L, Song H W, Zhang X H and Zhao X B 2002 Results of rock
cavern monitoring and assessment of cavern stability Technical Report for NTU-DSTA
238
PUBLICATIONS
JOURNALS
1. Panigrahi, R., Gupta, A. and Bhalla, S., “ Dismountable steel tensegrity grids as
2. Panigrahi, R., Gupta, A. and Bhalla, S., “Design of tensegrity structure using
3. Panigrahi, R., Gupta, A. and Bhalla, S., “A low cost variant of electro-mechanical
CONFERENCES
239
3. Gupta, A., Bhalla, S. and Panigrahi, R., “ Behaviour of foldable tensegrity
4. Panigrahi, R., Gupta, A., Bhalla, S. and Arora, K., “Application of artificial neural
India, pp 1950-1962
Steel, Space & Composite Structures, May 15-17, 2006, Kuala Lumpur, Malaysia,
pp 495-502
7. Panigrahi, R., Gupta A. and Bhalla S., “A new design approach for tensegrity
8. Pande, H., Prakash, A., Panigrahi, R., Bhalla, S. and Gupta, A., “Fabrication and
9. Panigrahi, R., Gupta A. and Bhalla S., “ Behaviour of tensegrity grid structure
240
Engineering (RACE-2007), College of Engineering and Technology, BPUT,
10. Panigrahi, R., Gupta A. and Bhalla S., “ Parametric study on tensegrity grid
241
BIO DATA
Ramakanta Panigrahi was born on 13th June 1966 in Cuttack district in Orissa, India. He
received his B.Sc. Engineering with Honours in May, 1989 from University College of
Engineering, Burla under Sambalpur University. He did his Master of Technology from
Department of Civil Engineering IIT Delhi. His research interests are tensegrity
242
APPENDIX
243
Table A Member forces in Newton
244
Table B Member forces in Newton due to load combinations
245
Table C Member forces in Newton due to wind load
246