International Journal of Solids and Structures 49 (2012) 973–988

Contents lists available at SciVerse ScienceDirect

International Journal of Solids and Structures

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

Steady-state dynamics of a 3D tensegrity structure: Simulations and experiments

J. Michielsen ⇑, R.H.B. Fey, H. Nijmeijer
Department of Mechanical Engineering, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands

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

Article history: This paper considers a modeling and analysis approach for the investigation of the linear and nonlinear
Received 27 July 2011 steady-state dynamics of a base excited 3D tensegrity module carrying a top mass. The tensegrity module
Received in revised form 7 November 2011 contains three compressive members, which may buckle and six cables (tendons). First, a dynamic model
Available online 11 January 2012
of the system is derived using Lagrange’s equation with constraints. The buckling modeling of the com-
pressive members is based on the assumed-mode method with a single mode discretization. The tendons
Keywords: are modeled as piecewise linear springs, which can only take tensile forces. This research focusses on the
Tensegrity structure
dynamic stability of the tensegrity structure by defining the geometrical and material properties in such a
Steady-state analysis
Stability
way that the system is just below the static stability boundary. Static and linear dynamic analysis is per-
Buckling formed. In the nonlinear steady-state analysis, frequency-amplitude plots, power spectral density plots,
Piecewise linear bifurcation point continuation diagrams, and Poincaré maps are presented. A tensegrity structure is
Path-following designed and manufactured and an experimental set-up is realized in order to validate the model by
Bifurcations comparing experimentally and numerically obtained responses. In the validation stage, the numerical
Experiments results are based on an amplifier-shaker-tensegrity structure model. It can be concluded that the numer-
ical results match partly quantitatively and partly qualitatively with the experimentally obtained
responses.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction structures are also used in fast robotic applications because of their
low mass, see e.g. Aldrich (2004) and Masic and Skelton (2004). But
Tensegrity structures originate from sculptures designed by the advantage of tensegrity structures goes beyond mass effi-
Snelson (1973). This class of structures consists of a set of pin- ciency. The usage of tendons provides the possibility to fold a
jointed bodies (compressive members) and cables (tensile mem- tensegrity structure in a small volume of space for transportation
bers), also called tendons, which ensure structural integrity by purposes. This key feature is of particular interest in aerospace
defining a stable volume in space. Inspired by these sculptures, applications (Furuya, 1992). A review of self-deploying tensegrity
Buckminster Fuller first patented these types of structures by com- structures is presented in Duffy et al. (2000). Moving tensegrity
bining the words tension and integrity to tensegrity in 1962 (Fuller, mechanisms can be found in several applications, such as a flight
1962). A historical survey of tensegrity structures is provided by simulator (Sultan and Corless, 2000), a space telescope (Sultan
Motro (1992). et al., 1999), and a smart sensor (Sultan and Skelton, 2004). Con-
Due to a favorable high stiffness/mass ratio, tensegrity struc- trolled tensegrity systems provide the possibility to make build-
tures have been introduced as potential alternatives in civil, aero- ings responsive to earthquakes, severe winds, and thermal loads.
space, and mechanical engineering. Suspension bridges are The control of a tensegrity structure with three compressive mem-
classical examples of tensegrity based structures because the ten- bers and nine tendons is studied in Kanchanasaratool and William-
sile components are often cables. More recently, designs similar to son (2002).
the sculptures of Kenneth Snelson are found in bridge designs. The Despite these advantages, tensegrity structures, just like other
Kurilpa bridge in Brisbane, Australia, is a perfect example. In addi- structures, may become unstable due to additional dynamic loads.
tion, new applications in civil engineering are tensegrity grid struc- In the worst case, this may result in failure of the system. The influ-
tures used in roof designs, see Adriaenssens and Barnes (2001) and ence of dynamic loads on tensegrity structures can be investigated
Quirant et al. (2003). Hypar tensegrity roofs can, for example, be by developing sufficiently accurate mathematical models and anal-
found in sports stadiums in Florida, Georgia, and Seoul. Tensegrity ysis strategies. In general, in order to apply tensegrity structures in
structural designs, it is vital to be able to predict, understand, and
⇑ Corresponding author. Tel.: +31 623508134. eventually optimize the static and dynamic stability of such sys-
E-mail address: [email protected] (J. Michielsen). tems. Design criteria in structural design often involve a low mass,

0020-7683/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijsolstr.2011.12.011
974 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

Fig. 1. The definition of the geometry, beams, struts, tendons, and nodes of the tensegrity structure (upper figure) and the reference frames (lower figure). In both figures, the
top mass is not depicted.

a high stiffness, and a high residual strength. For tensegrity structures, forces are characterized by network theory and the kinematics are ex-
these design criteria are mainly influenced by two important design pressed in terms of the compressive member vectors. This approach
parameters, namely the thickness of the compressive members and can only be applied to Class 1 tensegrity structures, i.e. the compressive
the amount of pretension in the tendons. When resonances occur, members are not allowed to be in contact with each other.
the compressive members of the tensegrity structure may dynamically This paper deals with the static and dynamic stability of a tensegrity
buckle, although the system is statically stable. Static and dynamic structure with three compressive members and six tendons carrying a
buckling can be avoided by adopting a relatively high bending stiffness top mass (and a top mass supporting structure). The compressive
of the compressive members but this may also lead to an increase of the members of the tensegrity module are allowed to buckle statically
mass of the tensegrity structure. A high pretension in the tendons, in and dynamically in a predefined direction. A dynamical model of the
general, leads to high stiffness and avoidance of slackening of the ten- tensegrity structure is derived using Lagrange’s equation of motion
dons when resonances occur. But it also leads to lower residual with constraints. The resulting equations of motion are used to study
strength and decreased static and dynamic stability of the structure be- the static and (nonlinear) dynamic responses by loading the system
cause the allowed additional load of the compressive members de- with a top mass and, in the dynamic case, by additional periodic base
creases. This all leads to trade-offs in design choices. One can excitation. In the second part of this paper, the model, which is first ex-
conclude that, constructing a tensegrity structure just below the static tended by an amplifier-shaker model, is validated by comparing
stability boundary, in order to avoid superfluous use of material, only numerically and experimentally obtained responses. From this analy-
makes sense if there is enough safety margin to avoid dynamic buck- sis, it can be concluded that the modeling approach results in a model
ling of compressive members and slackening of tendons, which may that matches partly quantitatively and partly qualitatively with the
be caused by additional dynamic loads. Nonlinear dynamic analysis dynamical behavior of the experimental set-up. Consequently, the
of tensegrity structures can be used to predict dynamic buckling due model can be used as a starting point to examine the origin of nonlinear
to nonlinear resonances. Additionally, in the event of dynamic buckling dynamic response phenomena of tensegrity structures in general.
of one or more compressive members, it can be used to evaluate if the This paper is organized as follows. The geometric description of the
structure does or does not globally collapse. tensegrity structure is discussed in Section 2. Buckling modeling of the
The Lagrangian approach has been frequently used for dynamic compressive members of the tensegrity module is presented in Sec-
modeling of tensegrity structures. A linear dynamical model of a tenseg- tion 3. The equations of motion of the tensegrity structure are derived
rity structure with three compressive members and six tendons is stud- in Section 4. The geometrical and material properties are defined in
ied to investigate the energy dissipation efficiency by solving initial Section 5. In Section 6, static response results are presented and ana-
value problems in Oppenheim and Williams (2001). The dynamics of lyzed to determine a suitable top mass. Modal analysis results and Fre-
a tensegrity structure with three compressive members and a six-mod- quency Response Functions (FRFs) are presented in Section 7 and in
ule tensegrity structure are examined using the eigenmodes of the line- Section 8, respectively. This is followed by a detailed numerical study
arized dynamical model in Murakami (2001a). In addition, the nonlinear of nonlinear dynamic steady-state responses, initialized by periodic
equilibrium equations are used to estimate the critical loads for static base excitation, in Section 9. The results from Sections 6–9 lay the foun-
compressive member buckling and slackening of tendons in Murakami dations for the design of an experimental set-up. In Section 10, the
(2001b). The static stability due to external loads/moments of a tenseg- experimental set-up is introduced and the corresponding amplifier-
rity structure with three compressive members and six tendons is stud- shaker-tensegrity model is defined. In Section 11, parameter identifica-
ied in Lazopoulos (2005) by evaluating the global instability of the model tion is carried out. The model of the tensegrity structure is validated by
and local Euler buckling of the compressive members. Skelton (2006) comparing experimentally and numerically obtained linear and non-
suggested to describe the dynamics in a matrix differential equation in- linear dynamic responses in Section 12. Finally, the conclusions of
stead of a vector differential equation. When using this method, the this paper are presented in Section 13.
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 975

2. Description of the geometry ular to the horizontal plane) are defined to express the body-fixed
reference frame ~ e2x1 ~
e21 ¼ ½ ~ e2z1 T of strut B1C1 in terms of the
e2y1 ~
Fig. 1 shows the geometry and the reference frames of the reference frame ~e0
tensegrity structure under consideration. Here, compressive mem-
e21 ¼ A1 ðuÞA0 ðh1 Þ~
~ e0 ð6Þ
bers are indicated by black lines and tendons by gray lines. A Carte-
sian reference frame ~ e0 ¼ ½~
e0x ~e0y ~e0z T is connected to the absolute Here, again the direction cosine matrices of Eq. (2) are used, but
origin OA with position vector ~ r OA ¼ ½0 0 0~ e0 . The position vector now based on h1 and u. The position vectors of the center of mass
of the relative origin OR, which is connected to the centre of the of strut B1C1, the top mass, and node C1 are respectively defined by
vertically moving base of the structure containing nodes Ai for
~ e0
rcm2 ¼ ½x2 ðtÞ y2 ðtÞ uðtÞ þ z2 ðtÞ~ ð7Þ
i = 1, 2, 3, is given by ~ rOR ¼ ~r OR  ~rOA ¼ ½0 0 uðtÞ~ e0 , where u(t) is
the (periodic) vertical motion of the base. ~ e0
rcm3 ¼ ½0 0 uðtÞ þ z3 ðtÞ~ ð8Þ
Three compressive members are located between nodes Ai and Bi e0
rC 1 ¼ ½a3 cos h1 a3 sin h1 uðtÞ þ z3 ðtÞ~
~ ð9Þ
for i = 1, 2, 3. These compressive members will be called beams in the
remainder of the paper because they are allowed to buckle. The uni-
form, homogeneous, pinned–pinned beams AiBi with rectangular 3. Buckling modeling
cross section have length L1, height H, width W, Young’s modulus
E, and mass density q1. The buckling direction of beams AiBi is Fig. 2 shows the perfect and buckled geometry of beam A1B1 and
predefined in the ~ e1zi direction. Three other compressive members the Cartesian reference frame ~ e11 . The origin of this reference frame
are located between nodes Bi and Ci for i = 1, 2, 3. These rigid com- coincides with node A1.
pressive members are called struts. The uniform, homogeneous The axial displacement field la(t, x1) of beam A1B1 is defined in
struts BiCi are cylindrically shaped with length L2, radius R, and mass e1x1 -direction and the transversal displacement field ha(t, x1) in
the ~
density q2. A top mass (not visible in Fig. 1), modeled by a point mass e1z1 -direction. Coordinate x1 is the beam centerline coordinate.
the ~
m3, is located at the centre of an imaginary circle through points C1, The displacement field in ~ e1y1 -direction is neglected due to
C2, and C3. It is assumed that the top mass m3 can only translate in H  W. It is assumed that the centerline of beam A1B1 initially
vertical direction due to the design of the experimental set-up. has a geometrical shape imperfection ha0(x1). Note that due to
Due to gravity g, the system is statically loaded in vertical direction the assumption of symmetry, it is also assumed that all imperfec-
by the weights of beams AiBi, struts BiCi, and by the weight of top tions in beams AiBi are identical. The centerline of beam A1B1 is de-
mass m3. As a result of the model assumptions, the system has a form scribed by the curve Xðt; x1 Þ~e1x1 þ Zðt; x1 Þ~e1z1 , where
of symmetry, because revolving the structure over 120° around the
Xðt; x1 Þ ¼ x1 þ la ðt; x1 Þ ð10aÞ
e0z -axis results in an identical structure. Consequently, only one third
~
of the structure needs to be modeled. Zðt; x1 Þ ¼ ha0 ðx1 Þ þ ha ðt; x1 Þ ð10bÞ
The absolute position vector of node Ai depends on radius a1 The transversal displacement field is discretized based on the
and the position of the base u(t) Assumed-Mode Method (Tongue, 1996) by using separation of vari-
rAi ¼ ½ a1 cos a0i
~ a1 sin a0i
T 0
uðtÞ  ~
e ð1Þ ables using a time varying amplitude h(t) and a shape function /(x1)
ha ðt; x1 Þ ¼ hðtÞ/ðx1 Þ ð11Þ
where a0i ¼ 23p ði  1Þ for i = 1, 2, 3 defines the angle of node Ai in the
horizontal plane. For each beam AiBi, two rotations ai(t) = a0i + a(t) The initial geometrical shape imperfection of beam A1B1 is discret-
(defined in the horizontal plane through nodes Ai) and b(t) (defined ized in an analogous way
in the plane through nodes Ai and Bi perpendicular to the horizontal
ha0 ðx1 Þ ¼ h0 /ðx1 Þ ð12Þ
plane), are used to express the reference frame ~ e1xi ~
e1i ¼ ½ ~ e1zi T
e1yi ~
of beam AiBi in terms of the reference frame e ~0 It is assumed that beams AiBi are pinned–pinned at nodes Ai and Bi.
2 32 3 Consequently, the transversal displacement and the reaction mo-
cos b 0 cos aisin b sin ai 0 ments at the beam’s ends are zero, i.e. ha(t, 0) = ha(t, L1) = 0 and
6 76 7 0
e1i ¼ 4
~ 0 0 54  sin ai
1 cos ai e ¼: A1 ðbÞA0 ðai Þ~
0 5~ e0 o2 ha ðt; 0Þ=ox21 ¼ o2 ha ðt; L1 Þ=ox21 ¼ 0. Now, the following shape func-
 sin b 0 cos b 0 0 1 tion is admissible because it obeys the latter kinematic and dynamic
ð2Þ boundary conditions and it can be differentiated infinitely many
times
The position vector of node Bi is defined by  
p x1
/ðx1 Þ ¼ sin ð13Þ
r Bi ¼ ~
~ e1i
r Ai þ ½ LðtÞ 0 0 ~ ð3Þ L1
where L(t) is the distance between nodes Ai and Bi. This distance
changes if buckling of beams AiBi occurs. Note that Eqs. (1) and
(3) define the position vectors of all nodes Ai and Bi, which can be
conveniently used to compute the tendon lengths. Expressions of
the cross tendon length lc(t), e.g. of the tendon located between
nodes A2 and B1, and the horizontal tendon length lh(t), e.g. of the
tendon located between nodes B2 and B1, are then found to be
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffi ffi
lc ¼ L2 ðtÞ þ 2 3a1 LðtÞ sinða02  aÞ cos b þ 3a21 ð4Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
lh ¼ 3L2 ðtÞ cos2 b þ 6a1 LðtÞ cos a cos b þ 3a21 ð5Þ

For the model, due to symmetry, it is sufficient to consider the posi-

tion of strut B1C1 solely. Therefore, two other rotations,
h1(t) = h01 + h(t) (defined in the horizontal plane through nodes Bi) Fig. 2. Geometrical description of perfect beam A1B1 (upper figure) and the bucked
and u(t) (defined in the plane through nodes B1 and C1 perpendic- beam A1B1 (lower figure).
976 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

Axially, the beam is assumed to be inextensible. The axial displace- In Eq. (20), V is the total potential energy function, T is the total ki-
ment field la(t, x1) is kinematically related to the transversal dis- netic energy function, R is a Rayleigh dissipation function, and k is a
placement field ha(t, x1) by the inextensibility constraint column with Lagrange’s multipliers, which are related to the forces
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi and torques working on the system to ensure the satisfaction of the
 2
ola oha0 oha oha kinematic constraint equations, on velocity level represented by
ðt; x1 Þ ¼ 12  1 ð14Þ
ox1 ox1 ox1 ox1 WT q_ ¼ 0.

see e.g. Koiter (1945) and Fey et al. (2011). The axial displacement
4.1.1. Energy functions
field is obtained by integrating Eq. (14)
The total potential energy function V is the sum of the axial
Z x1
ola ðt; x01 Þ 0 strain energy of one cross tendon V ;c , one horizontal tendon V ;h ,
la ðt; x1 Þ ¼ dx1 ð15Þ
0 ox01 the gravitational energy V g , and the bending strain energy V b of
beam A1B1
The expressions of the axial and transversal displacement field, to-
gether with the angles a and b, fully define the position vector of a V ¼ V ;c þ V ;h þ V g þ V b ð22Þ
point of beam A1B1 in terms of the reference frame ~ e0 The tendons are modeled as piecewise linear springs, which can
2 3T only take tensile forces. Therefore, the axial strain energy function
x1 þ la ðt; x1 Þ
6 7 1 0 V ;i of tendon i (i=c,h) becomes
~
r 1 ðx1 ; y1 ; z1 ; tÞ ¼ ~
r A1 þ 4 y1 e0
5 A ðbÞA ða1 Þ~ (
z1 þ ha0 ðx1 Þ þ ha ðt; x1 Þ 0 if li ðtÞ  l0i < 0
V ;i ¼ ð23Þ
ð16Þ 1
k ðl
2 i i
 l0i Þ2 if li ðtÞ  l0i > 0
The latter expression is valid for 0 6 x1 6 L1, W/2 6 y1 6 W/2, and where li(t) is the time-dependent tendon length, see Eqs. (4) and (5),
H/2 6 z1 6 H/2. The distance between nodes Ai and Bi, used in Eq. l0i is the stress-free tendon length, and ki is the tendon stiffness. The
(3), is equal to total gravitational energy function V g is based on the mass of beam
LðtÞ ¼ L1 þ la ðt; L1 Þ ð17Þ A1B1, strut B1C1, and one third of the top mass
ZZZ
The material type of beams AiBi is assumed to be linearly elastic. The V g ¼ q1 g e0z dx1 dy1 dz1 þ m2 gðuðtÞ þ z2 Þ
r 1 ðx1 ; y1 ; z1 ; tÞ  ~
~
exact curvature j(t, x1) of beam A1B1 follows from Koiter (1945) V

    1
ðoX=ox1 Þ o2 Z=ox21  o2 X=ox21 ðoZ=ox1 Þ þ m3 gðuðtÞ þ z3 Þ ð24Þ
jðt; x1 Þ ¼ ð18Þ 3
 32
ðoX=ox1 Þ2 þ ðoZ=ox1 Þ2 where V ¼ ½ðx1 ; y1 ; z1 Þ 2 R3 j0 6 x1 6 L1 ; W=2 6 y1 6 W=2; H=2 6
z1 6 H=2 specifies the volume of beam A1B1, m2 = pq2L2R2 is the
It is important to note that 5th-order Taylor series expansions are mass of strut B1C1, and g is the gravitation acceleration constant.
used to approximate Eqs. (14) and (18) because these expressions Note that the mass of beam A1B1 is equal to m1 = q1L1H W. The
can not be integrated symbolically as will be necessary later on to strain energy due to bending of beam A1B1 follows from
derive the energy functions of the system. Note further that, in ab- Z Z
sence of geometrical imperfections and transversal loading, i.e. E1 I 1 L1
P c L21 L1
Vb ¼ ðj  j0 Þ2 dx1 ¼ ðj  j0 Þ2 dx1 ð25Þ
when there is no gravity, static buckling of beams AiBi occurs if 2 0 2p 2 0

the critical axial Euler buckling load Pc is exceeded, where where j0 describes the curvature of beam A1B1 in the strain-free sit-
p2 EI 1 uation due to the geometrical imperfection ha0(x1). As mentioned
Pc ¼ with I ¼ WH3 : ð19Þ before, the latter integral is solved by taking 5th-order Taylor series
L21 12
approximations of the beam curvature j, see also Eq. (18).
The total kinetic energy function T is the sum of the kinetic en-
4. Equations of motion ergy T m1 of beam A1B1, the kinetic energy T m2 of strut B1C1, and the
kinetic energy T m3 of one third of the top mass
Due to the complexity induced by the 3D geometry and by T ¼ T m1 þ T m2 þ T m3 ð26Þ
buckling modeling, the equations of motion are first conveniently
derived for a set of nine dependent DOFs in Section 4.1. Due to The kinetic energy function of beam A1B1 includes the translational
symmetry, only one third of the structure will be taken into ac- and rotational energy of the beam and is computed by solving
count in the energy expressions. In Section 4.2, the equations of ZZZ
1
motion are reduced to a formulation of the dynamics in terms of T m1 ¼ q ~ r_1 dx1 dy1 dz1
r_1  ~ ð27Þ
2 1 V
a set of three independent DOFs.
where V is as specified above. The kinetic energy function of strut
4.1. Constrained equations of motion B1C1 equals
 T 
The equations of motion are derived by applying Lagrange’s r_cm2  ~
T m2 ¼ 12m2 ð~ r_cm2 Þ þ 12x e21 Jcm2 ~
~ ~ e21  x
~ ð28Þ
equation of motion with constraints
( where x~ is the angular velocity vector, which is expressed in terms
d
dt
ðT ;q_ Þ  T ;q þ V ;q ¼ R;q_ þ ðWkÞT of the body fixed reference frame ~ e21 by using the direction-cosine
T
ð20Þ matrices of Eq. (6)
W q_ ¼ 0
T

where q is a column containing the nine dependent DOFs, which de- x

~ ¼ ½0 u e21 þ 0 0 h_ 1 A1 ðuÞ~
_ 0 ~ e21 ð29Þ
scribe the deformed geometry of the (symmetric) tensegrity struc- T
and e21 Jcm2 ~
~ e21
is the inertia tensor of strut B1C1. For the cylindrically
ture, i.e.
shaped strut B1C1, the diagonal elements of the diagonal inertia ma-
q ¼ ½ a b h x2 y2 z2 h u z3 
T
ð21Þ trix Jcm2 are equal to J x2 x2 ¼ 12 m2 R2 and J y2 y2 ¼ Jz2 z2 ¼ 12 m2 ð3R2 þ L22 Þ.
Substituting the first time derivative of Eq. (7) and the angular
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 977

velocity vector of Eq. (29) in Eq. (28) leads to the kinetic energy The constraint equations of Eq. (35) establishes the relation be-
function of strut B1C1 tween the dependent DOFs q, see Eq. (21), and the independent
  DOFs qm by q = q(qm(t)). The velocity and acceleration of the depen-
_ 2 þ 2uðtÞ
T m2 ¼ 12m2 x_ 22 þ y_ 22 þ z_ 22 þ uðtÞ _ z_ 2 dent DOFs in terms of the independent DOFs are q_ ¼ Tðqm ðtÞÞq_ m and
2 € ¼ Tðqm ðtÞÞq
q _ q_ m ðtÞ; q ðtÞÞq_ m respectively, where T(qm(t)) =
€ m þ Tð
þ 12ð J x2 x2 sin u þ J z2 z2 cos2 hÞh_ 2 þ 12 J y2 y2 u
_2 ð30Þ m
oq(qm(t))/oqm. Now, the equations of motion with constraints, i.e.
The kinetic energy of one third of the top mass, which is assumed to Eq. (36), can be reduced to a formulation of the dynamics in terms
be only translating in vertical direction, is equal to of the independent DOFs, see e.g. Skelton and de Oliveira (2009)
 
_ 2 þ 2uðtÞ
T m3 ¼ 16m3 z_ 23 þ uðtÞ _ z_ 3 ð31Þ € m þ Gðq_ m ; qm Þ þ Cq_ m þ Hðqm Þ ¼ Bðqm Þu
Mðqm Þq € ðtÞ ð39Þ

where M ¼ TT MT; G ¼ TT MT_ q_ m þ TT G; C ¼ TT CT; H ¼ TT H, and B ¼

The generalized damping forces, based on a linear viscous damping
TT B.
model, are taken into account by the Rayleigh dissipation function
Later on, unless stated otherwise, results are presented for the
R ¼ 12ðc11 a_ 2 þ c22 b_ 2 þ c33 h_ 2 Þ þ c12 a_ b_ þ c13 a_ h_ þ c23 b_ h_ ð32Þ following output variables: (1) rotation a(t), (2) the transversal dis-
placement measured halfway the central axis of beams AiBi, i.e.
Note that only time derivatives of DOFs a, b, and h play a role in Eq. hmid(t) = Z(t, L1/2), see Eqs. (10b) and (3) the height difference be-
(32), because these DOFs will be chosen as the three independent tween the base (horizontal plane through nodes Ai) and the hori-
DOFs in Section 4.2. zontal plane through nodes Ci, denoted by hz ¼ ~ e0z  uðtÞ.
r C 1 ðtÞ  ~

4.1.2. Constraint equations

The kinematic constraint equations relate the dependent DOFs 5. Geometrical and material properties
x2, y2, z2, h, u, and z3 to the independent DOFs a, b, and h. The first
three kinematic constraint equations follow by equating Eq. (7) to The parameter values of the undamped tensegrity structure are
the following expression for the position vector of the center of presented in Table 1. The corresponding critical axial buckling load
mass of strut B1C1 of the steel beams AiBi follows from Eq. (19) and is equal to
Pc = 66.3 N for h0 = 0 mm. Parameter D~loc will be introduced later.
~
rcm2 ¼ ~ e11 þ 12½ L2
r A1 þ ½ L 0 0 ~ e21
0 0 ~ ð33Þ As stated before, all results presented later on are based on 5th-or-
der Taylor series approximations of the inextensibility constraint,
The final three kinematic constraint equations are found by equat-
Eq. (14), and the beam curvature, Eq. (18). In this way, accurate re-
ing two expressions for the position vector ~
r C1 of node C1,
sults are obtained in the displacement range of interest, while
~ e11 þ ½ L2
r A1 þ ½ L 0 0 ~ e21 ¼ ½ a3 cos h a3 sin h uðtÞ þ z3 ~
0 0 ~ e0 ð34Þ keeping computation times acceptable.
It is assumed that the stiffness of the tendons is so high, that the
Working out Eqs. (33) and (34) results in six holonomic, scleronom-
elongation of the tendons will always remain small compared to
ic kinematic constraint equations
their lengths. Therefore, initially, the stress-free tendon lengths
2 3
x2  L cos a cos b  12 L2 cos h cos u are computed by using an imaginary reference configuration q0m ,
6 y  L sin a cos b  1 L sin h cos u 7 in which buckling, beam imperfections, gravity, and elongation of
6 2 2 2 7
6 7 the horizontal tendons are neglected, i.e. L(t) = L1, h0 = 0, g = 0,
6 z 2  L sin b  1
L 2 sin u 7
hc ¼ 6
6
2 7¼0
7 ð35Þ and the position of nodes Bi in the horizontal plane is known and
6 h  arctan Z 1 7
6 7 located on a circle with radius a2. In this case, the total potential
4 u  arccos Z 2 5 energy function only contains the strain energy of the cross ten-
z3  L sin b  L2 sin u dons. The cross tendon length lc is now described by one DOF,
namely a. Minimizing this energy function to zero and taking into
with Z1 = (Lsinacosb)/(a1 + Lcosacosb) and Z 2 ¼ ða3 
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi account the geometrical boundary conditions, i.e. beams AiBi and
2
L cos b þ 2a1 L cos a cos b þ a21 Þ=L2 .
2
cross tendons should not interfere, results in the imaginary refer-
Substitution of Eqs. (22), (26), and (32) in Eq. (20) results in the ence configuration q0m . For chosen values of a1, a2 (see Fig. 1), and
equations of motion and a set of constraint equations L1 this leads to
( 2 3
MðqÞq€ þ Gðq; € ðtÞ þ ðWkÞT
_ qÞ þ Cq_ þ HðqÞ ¼ BðqÞu
11
12
p
T
ð36Þ 6  pffiffiffi  12 7
W q_ ¼ 0 q0m ¼ 6 2
4 arcsin L11 L1  3a1 a2  a21 þ a22
7
5 ð40Þ
where the holonomic constraint equations of Eq. (35) are written on 0
velocity level by
For this (imaginary) reference configuration, the corresponding
ohc ðqÞ stress-free tendon lengths ~loc and loh are found by substituting
h_ c ¼ WT q_ ¼ 0 with WT ðqÞ ¼ ð37Þ
oq
Table 1
Parameters values of the tensegrity structure.
4.2. Unconstrained equations of motion
Parameter Value Unit Parameter Value Unit
a1 8.75 cm L1 30.0 cm
The equations of motion with constraints, i.e. Eq. (36), are re-
a2 8.75 cm W 2.0 cm
duced to a formulation of the dynamics in terms of independent a3 1.50 cm H 1.2 mm
DOFs. In this formulation, the explicit use of the constraint equa- g 9.81 m/s2 q1 7850 kg/m3
tions, and thus the Lagrange’s multipliers k, is avoided. As men- kc 1.403  105 N/m E1 2.1  1011 N/m2
tioned before, the tensegrity structure configuration is fully D~l
oc
3.56  105 m Pc 66.3 N

defined by a, b, and h, implying a set of independent DOFs loc 25.195 cm h0 1.0 mm

kh 2.332  105 N/m L2 10.0 cm
T loh 15.155 cm R 7.5 mm
qm ¼ ½ a b h  ð38Þ
q2 2700 kg/m3
978 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

Eq. (40) in Eqs. (4) and (5) respectively. In the static equilibrium po-
sition of the unloaded system (m3 = 0 and u(t) = 0), a small level of
prestress in the tendons is desired and a substantial amount of com-
pressive stress in the beams is required to be near the static stability
boundary of the system. Therefore, in the actual model, the real
stress-free cross tendon length loc is specified by shortening ~loc by
choosing D~loc ¼ 3:56  105 m

loc ¼ ~loc  D~loc ð41Þ

Shortening ~loc can be interpreted as a way to define a pretension in

the cross tendons of F 0c ¼ kc D~loc ¼ 5:0 N. Note that this pretension
will increase the pretension caused by the gravity field, acting on
beams AiBi, struts BiCi, and top mass m3.

6. Static responses

The top mass is replaced by a vertical force, equal to Fex = m3g, in

order to compute the static load path of the system. Here, Fex > 0
represents a downward force and consequently Fex < 0 represents
an upward force. In Fig. 3, F c is the static cross tendon force, F h is
the static horizontal tendon force, and F b is the static compressive
force between nodes Ai and Bi. The latter force is computed by solv-
ing the equilibrium equations of node A1. Fig. 3 shows that for
Fex = 0, there is pretension in the tendons and a compressive force
in the beams, which are not perfectly straight due to gravity and a
geometrical shape imperfection. Fig. 4. The stable static equilibrium points for varying external loads (Fex = m3g).
Fig. 4 shows the corresponding branches of stable static equi-
librium points qm . Note that the stable static equilibrium points
are the solutions of V;qm ¼ 0, provided that V ;qm qm ðqm Þ > 0. Static 7. Modal analysis
buckling of beams AiBi becomes more dominant when increasing
the compressive load Fex or m3. In a tensile test (Fex < 0), slacken- The nonlinear equations of motion of Eq. (39) are linearized
ing of the horizontal tendons occurs at Fex = 54.94 N. Thus, be- around a stable static equilibrium point qm for modal analysis
low this value, indicated by a circle ’’, structural integrity is lost. and to compute FRFs of the system. Throughout this paper, (small)
Slackening of the horizontal tendons is further illustrated by dynamical responses of the linearized equations of motion are de-
looking at the horizontal tendon force F h , which becomes zero noted by Dqm.
at Fex = 54.94 N, see Fig. 3. All internal equilibrium forces have
a minimum (in an absolute sense) at Fex = 2.26 N. Here, gravity, 7.1. Undamped eigenfrequencies
acting on beams AiBi and struts BiCi, and the pretension of the
cross tendons due to D~loc is canceled out to some extent by The system’s undamped eigenvalues k0k = ± jx0k (j2 = 1) and
the external tensile load. For the tensile test, the horizontal ten- corresponding eigenmodes u0k are found by solving the eigenvalue
don force F h has a maximum at Fex = 17.66 N, whereas, for the problem
compressive test, the cross tendon force F c shows a maximum at ðHl  x20k Ml Þu0k ¼ 0 ð42Þ
Fex = 44.16 N.
where Ml and Hl are respectively the mass and stiffness matrix of
the linearized equations of motion, x0k = 2pf0k is the kth undamped
angular eigenfrequency, and u0k ¼ ½ u1k u2k u3k T is the corre-
sponding kth (real) eigenmode. The resulting undamped eigenfre-
quencies f0k are shown in Fig. 5 for a varying top mass m3. The
first undamped eigenfrequency f01 shows a maximum at
m3 = 1.5 kg. The second undamped eigenfrequency f02 starts to in-
crease when static buckling becomes significant and the third un-
damped eigenfrequency f03 shows a minimum at m3 = 4.8 kg.
From this point on, the top mass is fixed to m3 = 3.0 kg, which
corresponds to Fex = 29.4 N in Section 6. For this choice, all tendons
are clearly under tension and still no severe static buckling occurs
in the stable static equilibrium configuration.

7.2. Undamped eigenmodes

The kth undamped eigenmode u0k is normalized with respect to

the spatial position of node B1 evaluated at the stable static equi-
librium point qm corresponding to a top mass of m3 = 3.0 kg. The
Fig. 3. The internal equilibrium forces for varying external loads (Fex = m3g). kth eigenmode, denoted by du0k ¼ ½ dak dbk dhk T , is normalized
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 979

interpreted as the normalized eigenmode, in which deformation

of the horizontal tendons dominates the response.
The undamped angular eigenfrequencies x0k = 2pf0k are stored
in diagonal matrix X0 and the corresponding eigenmodes column-
wise in matrix U0 in order to compute the proportional damping
matrix (for m3 = 3.0 kg)
 
C ¼ 2UT
0 UT0 Ml U0 NX0 U1
0 ð44Þ

where diagonal matrix N contains the rather arbitrarily chosen

dimensionless modal damping parameters n1 = 0.02, n2 = 0.0175,
and n3 = 0.0225. Consequently, the modes are clearly undercritically
damped.

8. Frequency response functions

The FRFs H(f) of the tensegrity structure follow from the state
Fig. 5. The undamped eigenfrequencies for a varying top mass. space formulation of the linearized equations of motion

Hðf Þ ¼ Ct ðAt  j2pf IÞ1 Bt ð45aÞ

Table 2 where I is the identity matrix, f is the excitation frequency, At is the

The undamped eigenfrequencies and the corresponding normalized eigenmodes for system matrix, Bt is the input column, and Ct is the output matrix,
m3 = 3.0 kg.
i.e.
f (Hz) du01 du02 du03
0 I 0
f01 = 10.4 da 0.1105 0.0949 0.1168 At ¼ ; Bt ¼ ;
f02 = 34.5 db 0.0185 0.0115 0.0012 M1 l Hl M 1
l C M 1
l Bl
ð45bÞ
f03 = 104.1 dh 0.0139 0.0342 0.0094 o T
Ct ¼ ½ a hmid hz 
oqm qm ¼qm

Again, the matrices of the linearized equations of motion are de-

in such a way that the absolute displacement with respect to
noted by the subscript l.
r B1 ðqm ðm3 ¼ 3ÞÞ is equal to Dr = 2.0 cm by
~
Fig. 7 shows three FRFs of the tensegrity structure. The vertical
    dashed lines indicate the eigenfrequencies f0k. A clear anti-reso-
r B1 qm  ~
~ r B1 qm  du0k ¼ Dr ð43Þ nance in jHDhz Du€ j is located between f02 and f03. Note that the mod-
ulus of HDhz Du€ is only weakly influenced by the second eigenmode,
in which buckling of beams AiBi is dominant.
Fig. 6 shows the resulting normalized eigenmodes du0k, correspond-
ing to the undamped eigenfrequencies f0k. The values of the
undamped eigenfrequencies and normalized eigenmodes are pre- 9. Nonlinear steady-state analysis
sented in Table 2. The first normalized eigenmode du01 is domi-
nated by a rotation around the vertical axis (large da) and In this section, the nonlinear steady-state responses of the
buckling of beams AiBi is most dominant in the second normalized tensegrity structure are considered for a prescribed harmonic base
eigenmode du02 (largest dh). The first and third normalized eigen- acceleration
mode show similar behavior. However, the third normalized eigen- € ðtÞ ¼ U a sinð2pftÞ
u ð46Þ
mode du03 is characterized by less buckling of beams AiBi and the
sign of db is positive in contrast to a negative db value for the first where Ua is the acceleration amplitude, f is the excitation frequency,
normalized eigenmode. Physically, the third eigenmode can be and t is the time.

Fig. 6. Normalized eigenmodes corresponding to the undamped eigenfrequencies for m3 = 3.0 kg. Dashed lines refer to the equilibrium situation.
980 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

Fig. 7. Bode plot of the FRFs for m3 = 3.0 kg.

9.1. Frequency-amplitude plots by circles ’’, and Secondary Hopf (SH) bifurcations, also called Nei-
mark-Sacker bifurcations, will be indicated by triangles ’M’.
Frequency-amplitude plots are computed by continuation of Fig. 8 shows the frequency-amplitude plot in the frequency
periodic solutions obtained by solving two point boundary value interval 2.5 6 f 6 117.5 Hz for two acceleration amplitudes,
problems using the software package AUTO97 (Doedel et al., namely Ua = 1.0 m/s2 (black lines) and Ua = 6.0 m/s2 (gray lines).
1998). The local stability of the periodic solutions is obtained and Note that for Ua = 1.0 m/s2, the shapes of the frequency-amplitude
co-dimension one bifurcations are detected using Floquet theory, curves resemble the shapes of the FRFs in Fig. 7 to a large extent.
see e.g. Thomsen (2003). The amplitudes shown in the fre- Apparently, nonlinear effects play a minor role for Ua = 1.0 m/s2.
quency-amplitude plots of Fig. 8 are the peak-to-peak values (indi- This especially holds for higher excitation frequencies, where re-
cated by a tilde) of the steady-state solution in terms of a(t), hmid(t), sponse amplitudes are small. However, for Ua = 1.0 m/s2, two PD
and hz(t). In addition, stable periodic solutions are depicted with bifurcations are encountered at the first harmonic resonance near
solid lines, unstable periodic solutions with dashed lines, and the f01 = 10.4 Hz, see enlargement A. The unstable harmonic solution
locations of the undamped eigenfrequencies are indicated by branch between these PD bifurcations, located at f = 9.82 Hz and
squares ’h’. Period Doubling (PD) bifurcations will be indicated f = 10.60 Hz, shows weak softening effects, characterized by two
by diamonds ’}’, Cyclic Fold (CF) bifurcations will be indicated CF bifurcations. Later on, the type of steady-state responses inside

Fig. 8. Frequency-amplitude plots for m3 = 3.0 kg and two acceleration amplitudes, namely Ua = 1.0 m/s2 (black lines) and Ua = 6.0 m/s2 (gray lines).
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 981

enlargement B. In this case, the superharmonic resonance peak is

related to the second harmonic resonance.
The nonlinear dynamical behavior dramatically increases
when the acceleration amplitude is set to Ua = 6.0 m/s2. Near
the first harmonic resonance, neither a stable nor an unstable
periodic solution branch can be computed in the frequency
interval enclosed by two PD bifurcations at f = 6.5 Hz and
f = 13.5 Hz, see enlargement A. Next to the second superharmon-
ic resonance near f01/2 = 5.2 Hz, now showing softening behavior
(indicated by two CF bifurcations), also a third superharmonic
resonance near f01/3 = 3.47 Hz and a 1/2 subharmonic resonance
near f = 2f01/3 are present in all frequency-amplitude plots but
these are only clearly visible in enlargement A. The second har-
monic resonance peak related to the second eigenfrequency at
f02 = 34.5 Hz shows softening effects, which is caused by dynamic
buckling of beams AiBi. Three stable periodic solutions coexist for
the small frequency interval 27.86 6 f 6 27.91 Hz, of which two
are visible in enlargement C. Near f = 2f02 = 70 Hz a softening
1/2 subharmonic resonance related to the second eigenfrequency
occurs, initiated by two PD bifurcations. Based on observations
in Fey et al. (2011), this resonance is very probably due to para-
metric excitation. The third harmonic resonance peak near
f03 = 104.1 Hz also exhibits softening behavior. In this case, soft-
ening is caused by slackening of the horizontal tendons. Finally,
two SH bifurcations, located at f = 79.92 Hz and f = 97.47 Hz, are
encountered on the branch related to the third harmonic reso-
nance. Again, the steady-state solutions in the frequency interval
enclosed by these SH bifurcations will be investigated later by
means of Poincaré maps.

Fig. 9. PSD plots of the periodic solutions at f = 4.0 Hz (left plots) and f = 3.3 Hz 9.2. PSD plots
(right plots) for Ua = 6.0 m/s2.
Now, the Power Spectral Density (PSD) plots of two nearby peri-
odic solutions are used to study the origin of one superharmonic
the frequency interval enclosed by these PD bifurcations will be resonance peak in a more profound manner. The frequency F/f on
examined by constructing several Poincaré maps of steady-state the horizontal axis of the PSD plot has been normalized by the
solutions. Now, first, the remaining nonlinear features of the fre- excitation frequency f. Undamped eigenfrequency f01 is indicated
quency-amplitude plots for Ua = 1.0 m/s2 are discussed. A small by dark-grey dashed vertical lines.
second superharmonic resonance peak, corresponding to the first For Ua = 6.0 m/s2, the PSD plots of the periodic solutions
harmonic resonance, is observed near f f01/2 = 5.2 Hz. In addition, at f = 4.0 Hz and f = 3.3 Hz are presented in Fig. 9. When compar-
a small third superharmonic resonance peak is visible in the ing the PSD plots of the periodic solutions at f = 4.0 Hz with
frequency-amplitude plot of h ~
mid near f f02/3 = 11.5 Hz, see the PSD plots of the superharmonic resonance at f = 3.3 Hz, for

Fig. 10. Poincaré maps in the subspace spanned by hz and h_ z for different excitation frequencies f and Ua = 1.0 m/s2.
982 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

Fig. 11. Poincaré maps in the subspace spanned by hz and h_ z for different excitation frequencies f and Ua = 6.0 m/s2.

Fig. 12. Two parameter continuation diagram near the first harmonic resonance Fig. 14. Two parameter continuation diagram near the third harmonic resonance
(f01 = 10.4 Hz). (f03 = 104.1 Hz).

9.3. Poincaré maps

Initial value problems are solved in order to examine possible

a-periodic steady-state responses in the frequency interval en-
closed by the PD bifurcations at f = 9.82 Hz and f = 10.60 Hz
respectively for Ua = 1.0 m/s2. This is also carried out for the fre-
quency interval enclosed by the SH bifurcations at f = 79.92 Hz
and f = 97.47 Hz respectively for Ua = 6.0 m/s2. For both cases,
starting at the stable side of the right bifurcation point, the exci-
tation frequency is stepwise decreased with a frequency step of
Df = 0.05 Hz. The effect of transient responses is minimized by
disregarding the solution for the first 15 s after each time the
excitation frequency is changed. The resulting steady-state re-
Fig. 13. Two parameter continuation diagram near the second harmonic resonance sponses are used to construct Poincaré maps, by taking snap-
(f02 = 34.5 Hz). shots of the solution at t = T,2T, . . . , NpT, where T = 1/f, and
plotting the result in a two-dimensional subspace spanned by
hz and h_ z . Each Poincaré map contains Np = 2000 points.
Fig. 10 shows several Poincaré maps in the frequency interval
the latter case a sudden increase in the magnitude can be ob- 9.81 6 f 6 10.63 Hz and for an acceleration amplitude of
served at F/f = 3 for all responses. Indeed, F = 3f 10 Hz approx- Ua = 1.0 m/s2. The harmonic solution at f = 10.63 Hz experiences a
imately coincides with the first undamped eigenfrequency f01. transition to a 1/2,1/4, and 1/8 subharmonic response if the
Consequently, this is a third superharmonic resonance peak, excitation frequency is incrementally decreased. This behavior
related to the first eigenfrequency. Using this approach, the ori- suggests an infinite cascade of PD bifurcations, also known as the
gin of the remaining superharmonic resonance peaks is also Feigenbaum route to chaos (Feigenbaum, 1983), which results in
validated. a solution with chaotic behavior at f = 10.54 Hz. Solutions with a
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 983

Fig. 15. The experimental set-up (upper picture) and an enlargement of the shaker, tensegrity structure, and top mass (lower picture).

chaotic nature remain visible up to the frequency close to the left CF bifurcations if 5.38 < Ua 6 5.43 m/s2, and five CF bifurcations if
PD bifurcation at f = 9.82 Hz. Note that in Mallon et al. (2008, 5.43 < Ua 6 5.66 m/s2. The 1/2 subharmonic resonance peak re-
2010), where the dynamic stability of a cylindrical shell carrying lated to f01 is encountered if Ua P 5.43 m/s2, for which values
a top mass is investigated, similar phenomena are encountered. two PD bifurcations exist. Fig. 14 shows that the third harmonic
Also there, the first harmonic resonance peak becomes unstable resonance peak near f03 = 104.1 Hz starts to exhibit softening ef-
for an increasing amplitude of the harmonic base excitation and fects if Ua P 1.79 m/s2. The 1/2 subharmonic (parametric) reso-
a-periodic responses are found. nance near f = 2f02 starts to grow for Ua P 5.79 m/s2.
For Ua = 6.0 m/s2, Fig. 11 shows several Poincaré maps at fre-
quencies within the frequency interval enclosed by the SH bifurca- 10. Experimental set-up
tions at f = 79.92 Hz and f = 97.47 Hz. Harmonic responses are
found just outside this frequency interval. The Poincaré maps at The remainder of this paper focusses on validation of the model,
f = 95.17 Hz and f = 80.77 Hz show low order subharmonic re- derived in Sections 2–4, by comparing experimentally and numer-
sponses, characterized by a discrete number of Poincaré points. ically obtained steady-state responses. The values of a1, a3, H, W, L1,
Quasi-periodic solutions (closed curves) are visible in the remain- L2, loc, and loh, presented in Table 1, are adopted in the tensegrity
ing Poincaré maps. structure design. The experimental set-up and an enlargement of
the tensegrity structure are shown in Fig. 15.
9.4. Two-parameter continuation diagrams An electrodynamic shaker (a) is fixed to the upper part of table
(b), which is rigidly connected to the floor. The shaker contains a
The change of the loci of bifurcations of the periodic solutions massive exciter housing and a moving armature. It is driven by
can be studied by performing parameter continuation in a two an amplifier operating in voltage-mode. A prescribed (harmonic)
parameter space, see Doedel (2007). In this way, upper bounds amplifier input voltage
on the acceleration amplitude can be found so that nonlinear re-
V 0 ðtÞ ¼ v d sinð2pftÞ ð47Þ
sponse phenomena due to CF and PD bifurcations (sudden jumps
in response amplitudes, 1/2 subharmonic resonances, and a-peri- where vd is the excitation amplitude in [V] and f is the excitation
odic behavior) can be avoided. frequency in Hz, results in an amplifier output voltage V(t), which
For base excitation amplitudes Ua 6 6.0 m/s2, the two-parame- serves as the input for the shaker. Voltage V(t) results in a current
ter continuation diagram is divided in four frequency regions, i.e. I(t) through the coil of the shaker. This current generates a vertically
in the vicinity of the three eigenfrequencies f0k. The results are de- directed force working on the shaker armature, which results in
picted in Figs. 12–14. In these figures, the loci of CF bifurcations are acceleration ü(t) of the armature. Input signal generation and
depicted with black curves and the loci of PD bifurcations with data-acquisition is performed using a laptop with MATLAB2008a/
gray curves. The left plot of Fig. 12 shows that the second superhar- Simulink in combination with two TUeDACS Microgiant devices
monic resonance peak near f f01/2 starts to exhibit softening ef- (Franken, 2008).
fects if Ua P 4.22 m/s2. The 1/2 subharmonic resonance peak Part E of the tensegrity structure (containing nodes Ci) is fixed to
near f 2f01/3 emanates if Ua P 4.14 m/s2. The two PD bifurcations connector part F, which allows rotations around the vertical axis by
near the first harmonic resonance become visible if Ua P 0.56 m/s2, two integrated ball bearings placed in parallel. Pure vertical trans-
see the right plot of Fig. 12. Recall that in the frequency domain en- lation of triangular block G is realized by a guide mechanism with
closed by these PD bifurcations, solutions with a chaotic nature are six air bearings (d) in the center of plate (e). It is stressed that plate
most probably present if Ua P 1.0 m/s2. Fig. 13 shows that the sec- (e) with a weight of approximately 1000 kg, is isolated from vibra-
ond harmonic resonance peak contains two CF bifurcations if tions of the environment and vibrations of the shaker by using
4.88 6 Ua 6 5.38 m/s2, three CF bifurcations if Ua > 5.66 m/s2, four three active air bearings (f).
984 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

In the design of the tensegrity structure, the buckling direction Fig. 15), (2) the angular velocity h_ i ðtÞ (measured with a gyroscope
of beams AiBi matches with the buckling direction of the model, all sensor, integrated in connector part F), (3) the transversal displace-
compressive and tensile members are (dominantly) loaded in the ment hmid(t) of beam A1B1 measured halfway the length of this beam
axial direction, and the members are connected to each other by (indirectly measured with a calibrated strain-gauge (n)), and (4) the
multiple ball bearings in order to keep friction forces as low as pos- absolute velocity of the top mass, indicated by h_ zþu ðtÞ (measured
sible. Base plate A is rigidly fixed to the shaker armature. Three in- with laser vibrometer (o)). Note that hz(t) = hz+u(t)  u(t). The veloc-
sets, each connected with a ball bearing to base plate A, are used to ity signals are integrated over time to obtain the corresponding dis-
enable rotation ai(t) of beams AiBi. The combined mass of one inset placement/rotation signals. To avoid drift during time integration,
and ball bearing is denoted by mAi . The lower ends of the slender the measured velocity signals are passed through a second order
beams AiBi (H  W  L1) with rectangular cross sections are at- high-pass frequency filter with a cut-off frequency of fc = 1.5 Hz.
tached to (three) parts B. Parts B are coupled to the insets in base The effect of transient responses in stepped frequency-sweep exper-
plate A by pin-joints. The upper ends of beams AiBi are rigidly at- iments is minimized by neglecting the responses of the first 75 exci-
tached to (three) parts C. Parts C are connected by pin-joints to tation periods for each excitation frequency. Finally, the position
nodes Bi. As a result, the buckling direction of the pinned–pinned signals are used to compute the average peak-to-peak values of a
beams AiBi is predefined in the ~ e1zi -direction. Each node Bi has a con- steady-state solution, containing the responses of 150 excitation
nection structure consisting of three tendon fixation points to en- periods, for each excitation frequency.
sure that the virtual intersection point of the tendons is located in
nodes Bi. The connection between parts C, the connection structure 10.1. Amplifier-Shaker-Tensegrity structure model
of nodes Bi, and parts D can be interpreted as a cardan joint with
two rotational DOFs. Rotation of parts D around the axes of struts In the experiments, the acceleration ü(t) of the base is not pro-
BiCi is enabled by integrating two ball bearings placed in parallel portional to the prescribed input voltage V0(t) because the shaker
inside parts D. has its own dynamics. Consequently, the dynamics of the amplifier
The geometrical imperfections of the three beams AiBi are and shaker have to be taken into account, resulting in an amplifier-
measured and equal to h0 = 1.1 mm, h0 = 0.9 mm, and shaker-tensegrity structure model. For a detailed description of the
h0 = 0.85 mm respectively. Indeed, the geometrical shape imper- modeling of the amplifier-shaker-structure system is referred to
fections have approximately the shape as defined in Eq. (11). The Mallon (2008, 2010) and Fey et al. (2011). The equations of motion
different geometrical shape imperfections result in a slightly asym- of the shaker in combination with the structure under investiga-
metric tensegrity structure. This asymmetry can be removed to a tion are derived by extending the nine dependent DOFs q of Eq.
large extend by altering the amount of pretension in the three (21), with two independent DOFs describing the dynamics of the
cross tendons. To realize this, three bolts H, see Fig. 15, are located electrodynamic shaker, namely the charge q(t), i.e. the integral
at the bottom side of base plate A and three additional bolts are _
over time of the current IðtÞ ¼ qðtÞ, and the position u(t) of the sha-
used to secure bolts H in order to avoid that the initial pretension ker armature, i.e. the base. In general form, the equations of motion
level of the cross tendons changes during dynamical experiments. of the amplifier-shaker-tensegrity structure are
The masses of some parts of the tensegrity structure are listed
in Table 3. It is assumed that the total mass of the base plate mb Lc€I þ Rc I þ jc u_ ¼ VðtÞ ¼ P a ðV 0 ðtÞ þ ba V_ 0 ðtÞÞ ð48Þ
is the sum of mA, the mass of plate A, and 3mAi , the masses of the ms u€ þ cs u_ þ ks u ¼ jc I þ F ten ð49Þ
three ball bearings in the horizontal plane plus the corresponding
where Lc is the coil inductance, Rc is the coil resistance, jc is the current-
insets, which support parts B, so mb ¼ mA þ 3mAi . Beams AiBi are
to-force constant, Pa and ba are the amplifier coefficients, ms is the trans-
manufactured from spring steel. Each tendon is flexible and con-
lating mass of the shaker, cs is the mechanical damping constant of the
tains 140 steel twisted thin wires. The remaining components
shaker, ks is the spring stiffness of the shaker, and Ften is the vertical force
are made from aluminium. The total top mass m3 contains the
exerted on the shaker mass by the tensegrity structure.
mass of part E, the mass of connector part F, the mass of triangular
block G, the mass of an angular velocity sensor, and a possible
additional mass, so that m3 P 1.210 kg. Note that approximately 11. Parameter identification
0.4 kg of the top mass can both rotate around and translate in
the vertical direction and that the remaining part only can trans- The parameter values of the amplifier-shaker model are identi-
late in vertical direction. The translating part of the top mass can fied in Section 11.1 by using experiments on the bare shaker. In
be increased by mounting the additional mass on top of the verti- Section 11.2, some of the parameter values of the tensegrity struc-
cally moving triangular block G. ture are identified by using measurements on the total experimen-
The following quantities are measured during the experiments: tal set-up.
_
(1) the base velocity uðtÞ (measured with laser vibrometer (m), see
11.1. Amplifier-shaker identification

Based on static experiments on the bare shaker, it can be con-

Table 3
cluded that the shaker stiffness ks is linear for the displacement
Masses of some parts the experimental tensegrity structure. The masses of parts B, C,
D, and E include ball bearing masses. range of interest. The remaining parameters

Description Value Unit cs ¼ ½ ms cs Lc jc Rc Pa ba T ð50aÞ

Base plate mb = 1.356 kg
Inset + (horizontal ball bearing) mAi ¼ 27 þ 33 ¼ 60 g
are identified by minimizing functional J, representing a sum of
Part B + beam AiBi+part C mAi Bi ¼ 28 þ 42 þ 32 ¼ 102 g squared differences between the real and imaginary parts of the
Connection structure in node Bi mBi ¼ 15 g
b Du_ DV of the bare shaker
numerical FRF HDu_ DV 0 and the measured FRF H 0
Part D + strut BiCi mBi C i ¼ 64 þ 42 ¼ 106 g
Cross tendon mlc ¼ 8 g Jðcs Þ ¼ rðcs ÞT rðcs Þ ð50bÞ
Horizontal tendon mlh ¼ 2 g
Part E mE = 40 g
where r(cs) is the weighted residue column specified at N discrete
frequency points in the frequency interval 2.5 6 f 6 120 Hz
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 985

2 3
b Du_ DV ðf1 ÞÞ
w1 ReðHDu_ DV 0 ðf1 ; cs Þ  H three independent DOFs). In the eigenmode corresponding to k2,k,
6 0
7 resonance of the shaker armature/base plate is dominant. In addi-
6 . 7
6 .. 7 tion to these eigenvalues and eigenmodes, which occur in complex
6 7
6 b Du_ DV ðfN ÞÞ 7
6 wN ReðHDu_ DV 0 ðfN ; cs Þ  H 7 conjugate pairs, one real eigenvalue and corresponding eigenmode
6
r¼6 0 7 ð50cÞ
b 7 is strongly related to the electric part of the shaker model. In the
6 w1 ImðHDu_ DV 0 ðf1 ; cs Þ  H Du_ DV 0 ðf1 ÞÞ 7
6 7 latter eigenmode, the current is dominant.
6 .. 7
6 . 7 It appears that n3 and n4 are approximately linearly dependent
4 5
b on the top mass, whereas n1 is approximately independent of the
wN ImðHDu_ DV 0 ðfN ; cs Þ  H Du_ DV 0 ðfN ÞÞ
top mass. These observations are used by introducing two linear
Weighting functions wn ¼ 1=j H b Du_ DV ðfn Þj for n = 1, . . . , N are intro- relations for n3 and n4, which depend on the top mass
0

duced to equally distribute the relative error of the residue vector n3 ðm3 Þ ¼ n30 þ n31 m3 ð51Þ
over the whole frequency range.
n4 ðm3 Þ ¼ n40 þ n41 m3 ð52Þ
Responses of the bare shaker resulting from white noise excita-
tion are used to estimate the experimental FRF H b Du_ DV , where Du_ is
0 Note that the mechanical damping parameter of the shaker, corre-
the velocity of the base and DV0 is the white noise amplifier input sponding to n2, already has been identified in Section 11.1.
voltage. The influence of disturbance noise is reduced by averaging Next, a number of unknown parameters ct of the amplifier-sha-
over 225 records, where each record contains 214 points. The mass ker-tensegrity structure model, linearized around the static equi-
of the shaker armature is increased by fixing an additional mass of librium points for the eleven top masses, is estimated using
1.381 kg to it. In this way, the effect of the mass of the base plate of measured FRFs. To be more specific, the parameter values of the
the tensegrity structure, mb = 1.356 kg, is taken into account in the initial stress-free tendon length loc, the cross and horizontal tendon
identification procedure of the amplifier-shaker model. The identi- stiffness kc and kh, the critical axial buckling load Pc, the effective
fied parameter values cs are presented in Table 4. mass m1e = q1eWHL1 of beams AiBi, the effective mass
m2e = q2epR2L2 of struts BiCi, and the dimensionless damping
11.2. Tensegrity structure identification parameter related to ni with i = 1,3,4 are estimated. Note that iden-
tification of effective masses m1e and m2e in fact means that effec-
For identification purposes, FRFs H b l;k of the experimental set-up tive mass densities q1e and q2e are identified. In this way,
are measured for k = 1, . . . , Nm = 11 top masses. Here, subscript l re- additional masses in the experimental set-up, which are absent
fers to a specific input–output relation, namely l = 1 corresponds to in the model, are taken into account. The unknown parameters
b DuDV ; l ¼ 2 corresponds to H
H b DhDV ; l ¼ 3 corresponds to H b Dh DV , are collected in
0 0 mid 0

and l = 4 corresponds to H b Dh DV . The relation between subscript k

z 0

and the top mass is depicted in Table 5. Each FRF corresponding ct ¼ ½loc kc kh Pc m1e m2e n1 n30 n31 n40 n41 T ð53Þ
to a certain top mass is obtained by averaging over 210 records,
The identification method is divided into two steps due to the com-
where each record contains 216 points.
plex linearized dynamics of the amplifier-shaker-tensegrity struc-
The measured FRFs are, first of all, used to determine the modal
ture model. In the first step, the squared difference between the
parameters of the experimental set-up based on a multi-DOF mod-
numerical and experimental eigenvalues, ki,k(ct) = li,k + mi,k j and
al parameter fit in the frequency domain with four dominant ^
ki;k ¼ l
^ i;k þ m
^i;k j respectively, is minimized in order to determine a
modes, indicated by subscript i. The experimental eigenvalues
^ suitable initial guess c0t of the unknown parameter values. In the
ki;k ¼ l^ i;k þ m
^i;k j are used to derive the experimental dimensionless
second step, the initial guess c0t is used to identify the final param-
damping coefficients for the eleven top masses based on the rela-
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi eter values ct of the tensegrity structure. This step is based on a
tion ^ ni;k ¼ l ^ i;k = l^ 2i;k þ m ^2i;k . In the eigenmodes corresponding to
similar optimization method as presented in SubSection 11.1. In
eigenvalues k1,k, k3,k, and k4,k, resonances in the tensegrity structure
this case, the optimization method includes all measured FRFs for
are dominant (note that three eigenmodes indeed correspond to
all top masses, i.e. l = 1, . . . , 4 and k = 1, . . . , Nm. The resulting identi-
fied parameter values are listed in Table 6.
Table 4
Identified parameter values cs of the amplifier-shaker model.
12. Model validation
Parameter Value Unit
ms 1.7295 kg Numerical results based on the identified amplifier-shaker-
ks 4.30  104 N/m tensegrity structure model are compared with experimentally ob-
cs 162.33 kg/s tained eigenvalues in Section 12.1, experimentally obtained FRFs in
Lc 2.258  103 H
Section 12.2, and experimentally obtained frequency-amplitude
jc 9.0076 N/A
Rc 0.920 X plots based on stepped frequency sweep-up and sweep-down
Pa 74.97 – measurements in Section 12.3. Recall that in the experimental
ba 1.515  103 s1 set-up, a small part of the top mass can both rotate and translate,
whereas rotational inertia was not modelled. The justification of

Table 5 Table 6
The relation between subscript k and the top mass m3. Identified parameter values ct of the tensegrity structure.

k m3 Unit k m3 Unit Parameter Value Unit Parameter Value Unit

1 1.442 kg 7 4.450 kg loc 25.167 cm n1 0.0638 –
2 1.933 kg 8 4.935 kg kc 95.8 kN/m2 n30 0.0341 –
3 2.421 kg 9 5.423 kg kh 112.9 kN/m2 n31 0.0035 kg1
4 2.994 kg 10 6.019 kg Pc 63.8 N n40 0.0553 –
5 3.482 kg 11 6.507 kg m1e 40 g n41 0.0045 kg1
6 3.962 kg m2e 232 g
986 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

this model simplification has been checked. The translational iner- ical and experimental eigenvalues are in reasonable good agree-
tial force of the top mass has dominant influence on the dynamic ment. It can be concluded that the linearized amplifier-shaker-
response compared to the (small) rotational inertial moment of tensegrity structure model can be used to predict the location of
the lower part of the top mass. Moreover, rotational inertia of the eigenvalues of the system for top masses in the range of
the remainder of the (moving part of the) structure dominates 1.442 6 m3 6 6.507 kg.
the neglected rotational inertia of the lower part of the top mass.
12.2. Frequency response functions

12.1. Eigenvalues Fig. 17 shows Bode plots and coherence functions of the mea-
sured FRFs H b Dh;DV and H b Dh ;DV for the frequency interval
0 z 0
Fig. 16 shows the experimental eigenvalues ^ ki;k ðm3 Þ (black 2.5 6 f 6 120 Hz and a top mass of m3 = 2.994 kg. In this figure,
markers) obtained from the multi-DOF modal parameter fit in black lines indicate measured FRF, whereas gray lines show the
the frequency domain and the numerical eigenvalues ki;k ðct ; m3 Þ FRFs of the linearized identified amplifier-shaker-tensegrity struc-
(gray markers) computed with linearized dynamic models of the ture model.
identified amplifier-shaker-tensegrity structure. The (real) eigen- Five resonance peaks can be distinguished in the Bode plots of
value corresponding to the eigenmode dominated by the electric the measured FRFs. The heavily damped resonance peak at
part of the shaker is not shown and the damped eigenfrequencies f2 13.7 Hz corresponds to the eigenmode dominated by shaker
are expressed in Hz by fi = Im (ki)/2p = mi/2p. resonance and the small resonance peak at f5 102.0 Hz is due
The following observations hold for the experimental as well as to the finite stiffness of the shaker support structure. The latter is
the numerical eigenvalues. The first eigenfrequency near concluded based on the measured FRFs of the bare shaker. The
f1 7.0 Hz, corresponding to an eigenmode dominated by rotation remaining resonance peaks near f1 = 7.0 Hz, f3 = 35.7 Hz, and
around the ~ e0z axis, slightly decreases for increasing top masses, f4 = 86.4 Hz correspond to eigenmodes dominated by the tenseg-
whereas the second eigenfrequency near f2 14.0 Hz (related to rity structure. It is stressed that the resonance peaks of the tenseg-
the shaker resonance) is barely influenced by changes in the top rity structure can clearly be distinguished in all Bode plots for the
mass. The third eigenfrequency f3, corresponding to the eigen- eleven top masses. In general, the behavior of the coherence func-
mode, in which buckling of beams AiBi is dominant, increases from tion is similar for all FRFs. Coherence values drop in frequencies
31.5 Hz for m3= 1.442 kg to 47.7 Hz for m3= 6.507 kg. Finally, the intervals with anti-resonances. From Fig. 17, two conclusions can
fourth eigenfrequency f4 has a minimum at m3 4.0 kg. The obser- be drawn:
vations for f1, f2, and f4 match with the modal analysis results of the
tensegrity structure when the shaker is excluded, see Fig. 5 in The measured FRFs are approximated (reasonably) well in the
Section 7. frequency interval 2.5 6 f 6 80 Hz by their numerical
The imaginary parts mi,k of the numerical and experimental counterparts.
eigenvalues correspond very well. The real parts li,k of the numer- For the frequency interval 80 6 f 6 120 Hz, differences in the
modulus and the phase between the experimental and numer-
ical FRFs are observed mainly due to the small resonance in the
experimental FRFs near 102 Hz, which obviously cannot be pre-
dicted by the current model.

Overall, it can be concluded that the FRFs of the amplifier-sha-

ker-tensegrity structure model show a satisfactory correspondence
with the experimental FRFs for all top masses.

12.3. Nonlinear steady-state responses

The nonlinear steady-state responses of the system, experimen-

tally generated with stepped frequency sweeps with a harmonic
input voltage according to Eq. (47), are used to determine the
experimental frequency-amplitude plots. During each experiment,
a frequency sweep-down is carried out (by stepwise decreasing the
excitation frequency with Df = 0.1 Hz), which is followed by a fre-
quency sweep-up (by stepwise increasing the excitation frequency
with Df = 0.1 Hz). The nonlinear dynamic steady-state responses
near the first and third harmonic resonance frequencies of the sys-
tem are studied by using two top masses, i.e. m3 = 1.442 kg and
m3 = 2.994 kg.
For the frequency range 5 6 f 6 18 Hz, the experimentally ob-
tained frequency-amplitude plots of ~ ~z (black and gray dots)
h and h
for m3 = 1.442 kg and vd = 0.055 V are presented in Fig. 18 together
with the numerically obtained frequency-amplitude plots (solid
and dashed black lines) based on the identified amplifier-shaker-
tensegrity structure model. The first harmonic resonance peak, pre-
dicted by the model and related to the first eigenfrequency near
f = 7.0 Hz, corresponds well to the measured first harmonic reso-
nance peak. In Fig. 18, the softening 1/2 subharmonic resonance near
Fig. 16. Experimental eigenvalues (black markers) and the numerical eigenvalues 14 Hz is a combination resonance due to 2f1 f2 14 Hz. In the
of the tensegrity structure. model, it occurs in a somewhat smaller frequency range than in
 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988 987

Fig. 17. Bode plot and coherence function of HDhDV 0 (upper plots) and HDhz DV 0 (lower plots).

Fig. 18. Experimental and numerical frequency-amplitude plots for m3 = 1.442 kg and vd = 0.055 V.

Fig. 19. Experimental and numerical frequency-amplitude plots for m3 = 2.994 kg and vd = 0.0325 V.

the experiment. In the experiment, the occurrence of frequency hyster- and a voltage amplitude of vd = 0.0325 V. Near f = 33 Hz, a
esis is much clearer than in the simulation. Indeed, in the frequency harmonic resonance peak is found, which is related to the third
sweep-up, a sudden increase in the amplitude is visible at 14.2 Hz, eigenfrequency at f03 = 35.7 Hz (recall that the corresponding
whereas a sudden decrease in amplitude occurs at 13.5 Hz. In the sim- eigenmode is dominated by buckling of beams AiBi). The experi-
ulations, the two CF bifurcations, which are responsible for the sudden mental peak-to-peak values of h ~z are overestimated by the model
jumps, are much closer to each other. Note that very probably the 1/2 near the resonance peak near 33 Hz and underestimated near the
subharmonic resonance near f = 2f01 = 14 Hz is absent in Fig. 8 due to anti-resonance between near 27 Hz. The main reason for this is
the absence of shaker dynamics there. Summarizing, in Fig. 18, despite asymmetrical buckling behavior of beams AiBi in the experiment
some quantitative differences, the global nonlinear steady-state due to the different geometrical shape imperfections. Globally,
dynamics in the experiment are qualitatively comparable to the ~ z , the experimental solution branch near the resonance peak
for h
numerical responses. related to the third eigenfrequency is in good agreement with
Fig. 19 shows the experimentally and numerically obtained fre- the numerical solution branch. For ~ h, the numerical and experi-
quency-amplitude plots of ~ ~ z for a top mass of m = 2.994 kg
h and h mental obtained solution branches are in good agreement.
3
988 J. Michielsen et al. / International Journal of Solids and Structures 49 (2012) 973–988

13. Conclusions are compared with experimental frequency-amplitude plots ob-

tained by stepped frequency sweep experiments. Globally, the mea-
In this paper, the numerical and experimental linear and nonlinear sured nonlinear dynamic steady-state responses can be predicted by
steady-state dynamics of a base excited 3D tensegrity structure carry- the model.
ing a top mass (and a top mass supporting structure) are examined. The It can be concluded that for a tensegrity structure near the static
equations of motion are derived by using Lagrange’s equation of mo- stability boundary, linear and nonlinear dynamic steady-state re-
tion with constraints. In this model, the compressive members (beams sponses, caused by additional periodic loads, can be predicted by
Ai Bi) of the tensegrity module are allowed to buckle and the tendons the model developed in this study. Therefore, this model structure
are modeled by piecewise linear springs, which can only take tensile may be used as a starting point to efficiently examine the static and
forces. Top mass, pretension in the tendons, and geometrical and mate- dynamic stability of other tensegrity structures.
rial properties of the tensegrity structure are chosen in such a way that
the system is close to the static stability boundary. The effect of addi- Acknowledgements
tional harmonic base excitation is investigated in order to study the dy-
namic stability of the structure. Bram de Jager is gratefully acknowledged for fruitful discus-
Static responses show that static buckling of beams AiBi be- sions and Pieter van Hoof and Peter Hamels for the manufacturing
comes more dominant for relatively large vertical compressive of the tensegrity structure.
loads, whereas slackening of the horizontal tendons is encountered
for a certain vertical tensile load. For larger loads, the static re- References
sponses are characterized by geometric nonlinearities (large
Adriaenssens, S., Barnes, M., 2001. Tensegrity spline beam and grid shell structures.
changes in the geometrical equilibrium configuration occur), espe-
Eng. Struct. 23, 29–36.
cially with respect to the rotation around the vertical axis. Aldrich, J., 2004. Control synthesis for a class of light and agile robotic tensegrity
The undamped eigenfrequencies and corresponding eigen- structures. Ph.D. thesis, University of California.
modes are studied by linearizing the equations of motion around Doedel, J., 2007. Numerical Continuation Methods for Dynamical Systems.
Department of Computer Science, Concordia University, Montreal.
stable static equilibrium points for a range of top masses and by Doedel, E., Paffenroth, R., Champneys, A., Fairgrieve, T., Kuznetsov, Y., Oldeman, B.,
solving the corresponding eigenvalue problems for the undamped Sandstede, B., Wang, X., 1998. AUTO97: Continuation and bifurcation software
system. In addition, for a specific top mass, a linearized model is for ordinary differential equations (with HOMCONT). Technical report,
Concordia University.
used to examine the linear dynamic responses of the system by Duffy, J., Rooney, J., Knight, B., Crane, C., 2000. Review of a family of self-deploying
evaluating three FRFs of the system. tensegrity structures with elastic ties. Shock Vib. 32, 100–106.
Nonlinear steady-state responses of the tensegrity structure are Feigenbaum, M., 1983. Universal behavior in nonlinear systems. Physica 7D, 16–39.
Fey, R., Mallon, N., Kraaij, C., Nijmeijer, H., 2011. Nonlinear resonances in an axially
investigated by prescribing periodic base excitation with a constant excited beam carrying a top mass: simulations and experiments. Nonlinear Dyn.
acceleration amplitude high enough to introduce nonlinear dynamic 66 (3), 285–302.
phenomena. Frequency-amplitude plots are obtained by solving two- Franken, J., 2008. TUeDACS MicroGiant. Eindhoven University of Technology.
Fuller, R., 1962. Tensile integrity structures. United States Patent 3063521.
point boundary value problems in combination with a continuation Furuya, H., 1992. Concepts of deployable structures in space applications. Int. J.
method for periodic solutions. Multiple types of nonlinear dynamic Space Struct. 7, 143–152.
resonances are encountered such as harmonic resonance peaks Kanchanasaratool, H., Williamson, D., 2002. Modeling and control of class nsp
tensegrity structures. Int. J. Control 75, 123–139.
(mostly characterized by softening effects), superharmonic resonance
Koiter, W., 1945. Over de stabiliteit van het elastisch evenwicht. Ph.D. thesis, Delft
peaks, and subharmonic/parametric resonance peaks. In addition to University of Technology.
these periodic responses, Poincaré maps suggest that quasi-periodic Lazopoulos, K., 2005. Stability of an elastic tensegrity structure. Acta Mech. 179, 1–
and chaotic solutions are also encountered near some resonance peaks. 10.
Mallon, N., 2008. Dynamic stability of thin-walled structures: a semi-analytical and
Finally, two parameter continuation diagrams of the loci of cyclic fold experimental approach. Ph.D. thesis, Eindhoven University of Technology.
and period doubling bifurcations of periodic responses are presented to Mallon, N., Fey, R., Nijmeijer, H., 2008. Dynamic stability of a thin cilindrical shell
assess upper bounds for the harmonic excitation amplitude in order to with top mass subjected to harmonic base-acceleration. Int. J. Solids Struct. 45
(6), 1587–1613.
prevent dynamic instability of the tensegrity structure. Mallon, N., Fey, R., Nijmeijer, H., 2010. Dynamic stability of a base-excited thin
A tensegrity structure has been designed and manufactured and orthotropic cylindrical shell with top mass: similations and experiments. J.
an experimental set-up has been realized in order to validate the Sound Vib. 329 (15), 3149–3170.
Masic, M., Skelton, R., 2004. Open-loop controll of class-2 tensegrity towers. Proc.
dynamic model of the tensegrity structure. The design of the SPIE 5383, 298–308.
tensegrity structure incorporates a predefined (static and dynamic) Motro, R., 1992. Tensegrity systems: the state of the art. Int. J. Space Struct. 7 (2),
buckling direction of the compressive members of the tensegrity 75–83.
Murakami, H., 2001a. Static and dynamic analyses of tensegrity structures. Part I.
module and it ensures that the members are (dominantly) loaded Nonlinear equations of motion. Int. J. Solids Struct. 38, 3599–3613.
in the axial direction. Murakami, H., 2001b. Static and dynamic analyses of tensegrity structures. Part II.
To be able to compare the experimentally obtained responses to Quasi-static analysis. Int. J. Solids Struct. 38, 3615–3629.
Oppenheim, I., Williams, W., 2001. Vibration of an elastic tensegrity structure. Eur. J.
their numerical counterparts, an amplifier-shaker-tensegrity struc-
Mech. A 20, 1023–1031.
ture model is derived because the dynamics of the amplifier and Quirant, J., Kazi-Aoual, M., Motro, R., 2003. Designing tensegrity systems: the case of
shaker, used to excite the base of the tensegrity structure in the a double layer grid. Eng. Struct. 25, 1121–1130.
experimental set-up, interact with the dynamics of the tensegrity Skelton, R., 2006. Dynamics of Tensegrity Systems: Compact Forms. In: Proceedings
of the 45th IEEE Conference on Decision and Control.
structure with top mass. Skelton, R., de Oliveira, M., 2009. Tensegrity Systems. Springer, Dordrecht.
Numerical FRFs, computed with a linearized dynamical model of Snelson, K., 1973. Tensegrity Masts. Shelter Publications.
the amplifier-shaker-tensegrity structure, are in good correspon- Sultan, C., Corless, M., 2000. Tensegrity flight simulator. J. Guid. Control Dyn. 23,
1055–1064.
dence with experimentally obtained FRFs. In addition, linearized dy- Sultan, C., Skelton, R., 2004. A force and torque tensegrity sensor. Sensors Actuators
namic models are used to compute the numerical eigenvalues of the A 112, 220–231.
system for several top masses. These eigenvalues are in good agree- Sultan, C., Corless, M., Skelton, R., 1999. Peak to peak control of an adaptive
tensegrity space telescope. In: Proceedings of SPIE the International Society for
ment with the experimental eigenvalues, which are determined by a Optical Engineering 3667, pp. 190–201.
multi-DOF modal parameter fit in the frequency domain using Thomsen, J., 2003. Vibrations and Stability; Advanced Theory, Analysis, and Tools,
measured FRFs. Finally, frequency-amplitude plots based on the 2nd ed. Springer, Berlin.
Tongue, B., 1996. Principles of Vibration. Oxford University Press Inc., New York.
nonlinear dynamic amplifier-shaker-tensegrity structure model