Vous êtes sur la page 1sur 12

Sensors and Actuators A 112 (2004) 220–231

A force and torque tensegrity sensor


Cornel Sultan a,∗ , Robert Skelton b
a Harvard University, Boston, MA 02115, USA
b Department of Mechanical and Aerospace Engineering, University of California, La Jolla, CA 92093-0411, USA
Received 24 November 2003; received in revised form 28 January 2004; accepted 31 January 2004

Abstract

Tensegrity structures represent a special class of flexible structures, whose members can simultaneously perform the functions of strength,
sensing, actuating, and feedback control. In this article we show how these structures intrinsic properties can be exploited to construct a
smart sensor for simultaneous measurement of six different quantities: three orthogonal forces and three orthogonal torques. The static
and dynamic characteristics of the sensor are computed and the influence of friction and prestress upon these characteristics is analyzed.
An optimal estimator design is presented and its performance is evaluated through numerical simulations. These simulations indicate that
the tensegrity sensor is capable of simultaneously providing correct estimates of the six quantities of interest.
© 2004 Elsevier B.V. All rights reserved.

Keywords: Tensegrity; Sensor; Optical fiber

1. Introduction fibers and can be used to estimate the quantities of interest


(here three external forces and three external torques). Also,
A major challenge in the field of sensors technology is because the number of candidate sensing elements is large,
the development of smart sensors. These should be sens- various combinations can be used to provide more exact as
ing devices, easy to calibrate and adapt to new operating well as fault tolerant and redundant approximations of the
conditions and requirements, and capable of accommo- measured quantities.
dating complex algorithms for local processing to remote The article is organized as follows. First, a detailed de-
controllers and computers. A smart sensor should also be scription of the tensegrity sensor is given. Next, the static
able to compensate algorithmically for the inherent noise. and dynamic characteristics of the sensor are analyzed. The
Another important demand for smart sensors is to be capa- article shows how these characteristics can be easily cali-
ble of simultaneously providing redundant, highly reliable brated through pretension adjustment. Then an optimal es-
information about different physical signals. timator is designed, based on the linearized mathematical
This article presents a smart sensor which enables the si- model of the structure’s dynamics and assumed models of
multaneous measurement of six quantities: three orthogonal the external quantities to be estimated. Finally the design is
forces and three orthogonal torques. The device consists of a evaluated through numerical simulations.
tensegrity structure (Fig. 1) composed of 18 elastic tendons,
six rigid bars, and a rigid top, whose intrinsic properties are
exploited to make it a smart sensor. 2. Tensegrity sensor description
One of the most important advantages of tensegrity struc-
tures is that the elastic members (e.g. tendons) provide ex- A perspective view of a tensegrity sensor is given in Fig. 1.
cellent opportunities for sensing functions while also acting Its components are: a triangular fixed base (A11 A21 A31 ); a
as structural elements. Measurements of the geometry of triangular rigid top (B12 B22 B32 ); three bars attached through
these tendons are easy to obtain through embedded optical ball and socket joints to the base (Ai1 Bi2 ); three bars simi-
larly attached to the rigid top (Ai2 Bi2 ); 18 tendons connect-
ing the end points of the bars. The tendons are classified as
∗ Corresponding author. Present address: Guidance Navigation and Con-
saddle: S, (Aj2 Bi1 ); vertical: V, (Aj1 Bi1 and Aj2 Bi2 ); diag-
trol, Scientific Systems Company Inc., Suite 3000, 500 West Cummings
Park, Woburn, MA 01801, USA. Tel.: +1-781-933-5355x251;
onal: D, (Aj1 Ai2 and Bj1 Bi2 ) tendons, respectively.
fax: +1-781-938-4752. For the tensegrity sensor, the rigid top plays the role of
E-mail address: cornel sultan@yahoo.com (C. Sultan). the proof element. The sensing task consists in estimating

0924-4247/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.sna.2004.01.039
C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231 221

^t
3 B
22
B O
12
t ^t
1 B
32

^
b B
3 11
Bij A
12 B
B 31 21
A
32
A
^ 22
δ
b
3 ^
ij b
A 1
^ 21
α
b
ij 1 O A
31
A A 11
ij

Fig. 1. Two stage tensegrity structure.

the three orthogonal forces and three orthogonal torques act- (see [12]). Tensegrity structures research was initiated by
ing on the rigid top. In order to accomplish this task, some Fuller [4] and Pugh [23]. Rigorous investigations have been
of the tendons will act as sensing elements and provide real carried out rather recently, starting, interestingly enough,
time information about their lengths. These individual mea- with the fine contribution of the celebrated literary critic
surements are the inputs to a dynamic state estimator which of last century’s modernism, Kenner [9]. Calladine [1],
produces the desired estimates. In this design the sensing Motro et al. [11], Pellegrino and Calladine [22], Sultan
tendons are not only structural elements, but members of the et al. [31], made important contributions to the general the-
intelligent system which accomplishes the estimating task. ory of these structures statics. Sultan et al. [31], Murakami
The tensegrity structure, with the estimator designed to meet and Nishimura [13,14] and Nishimura and Murakami [18]
certain performance specifications, represents a smart, inte- published a series of results regarding analytical solutions
grated sensor, capable of providing simultaneous informa- of the statics problem, while Hanaor [5], Kebiche et al. [8],
tion about six different quantities. Vassart and Motro [38], developed numerical methods of
large generality. The field of tensegrity structures dynamics
2.1. Tensegrity structures research was pioneered by Motro et al. [11]. Significant
theoretical advances have been later made by Murakami
Tensegrity structures represent a class of space structures [15] and Sultan et al. [32,33] who developed nonlinear and
composed of a set of soft members and a set of hard mem- linearized dynamics models for tensegrity structures. Using
bers. The soft members cannot carry other significant loads simpler models, Connelly and Whiteley [2] proved some
except for tensile ones. The representative example is an important results regarding the stability and rigidity of
elastic tendon which cannot be compressed for all practical equilibrium configurations, while Oppenheim and Williams
purposes but can carry significant tension. Because of this [20,21] discovered interesting properties of their vibration
property we shall also refer to these members as tensile and damping characteristics. Control design studies have
members. On the other hand, the hard members are char- been pioneered by Skelton and Sultan [26], followed by
acterized by the fact that they can carry any type of load. contributions of Sultan and Skelton [28], Djouadi et al. [3],
The representative example is a bar which can carry signif- and Kanchanasaratool and Williamson [7]. Applications
icant and comparable tension, compression forces, bending of tensegrity structures have been proposed, ranging from
torques, etc. tensegrity domes [6,39], to antennas [3], space telescopes
A structure composed of soft and hard elements as de- [27], flight simulators [30], deployable structures [35,36].
scribed above is a tensegrity structure if it has the property
of prestressability. This property consists of the structure’s 2.2. Sensing mechanisms
ability to maintain an equilibrium shape with all tensile
members in tension and in the absence of external forces or The problem of choosing the sensing mechanism of the
torques. Tensegrity structures integrity is guaranteed by the sensing tendons deserves some discussion. The small size,
tensile members in tension, hence their denomination, tense- large dynamic bandwidth and ease of embedding make fiber
grity, an acronym of tension-integrity coined by R.B. Fuller. optic sensors extremely attractive for use as sensing elements
The origins of tensegrity structures can be pin-pointed in a tensegrity sensor. Fiber optic sensors also present sev-
to a constructivist artist’s (Loganson) sculpture in 1921 eral inherent advantages over conventional (resistive) sen-
222 C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231

sors which include: electro-magnetic and hazardous envi- opportunity for compensation during strain measurement.
ronment insensitivity, the absence of Joule heating effect, Disadvantages are mainly related to the signal process-
high response bandwidth and low cost and weight per unit ing complexity and to the difficult remote operability and
length. lead insensitivity. The fabrication of modal interferom-
Optical fiber sensors are classified as being extrinsic or eters is also somehow difficult. Their sensitivity is two
intrinsic. In intrinsic sensors the optical energy in the fiber or three orders of magnitude lower than for standard
is affected directly by the phenomenon being measured and interferometers.
changes in the output intensity give an indication of the Polarimetric sensors use specialized high birefringence
magnitude of the disturbance. Typical intrinsic fiber sensors fiber, containing two well-defined “eigenaxes” (birefringent
are interferometric sensors such as Fabry–Perot (see e.g. axes) which are orthogonal and have significantly differ-
[40,41]), Mach–Zehnder (see [16]), Michelson (see [37]), ent effective refractive indices. Light energy entering either
polarimetric sensors (see [10]) and modal interferometers “eigenmode” will remain in that mode as the fiber undergoes
such as two-mode elliptic core [17] and twin core [37] ones. temperature or strain variations, but the relative propagation
In extrinsic sensors the fiber only serves to carry optical speed of energy in these two modes will be affected. Changes
power to, and sensing information from, the region of inter- in relative delay between components of the light in these
est. The sensing mechanism of the tensegrity sensing ten- two modes have been shown to be proportional to longitu-
dons should be intrinsic in nature since the fiber must be dinal strain [19]. This relative phase delay can be measured
embedded in the tendon whose characteristics (e.g. lengths) by observing the output light through a correctly oriented
variations are to be measured. polarizer. Polarimetric sensors are single fiber—thus easy to
Optical fiber sensors are also classified according to the embed—exhibit excellent lead-in insensitivity, remote oper-
transduction mechanism which brings about a change in ability, are relatively easy to manufacture. On the other hand,
some property of optical power passing through the fiber, they are expensive, since they require specialized fiber, fiber
such as intensity, phase, modal content, polarization, etc., polarizers, etc. The signal processing complexity associated
the corresponding sensors being known as intensity, phase with polarimetric sensors is also significant.
or interferometric based, modal interferometric and polari- Each of the sensing mechanisms previously described rep-
metric sensors. resent viable options for tensegrity sensing tendons. The
Intensity sensors consist of an optical power source, an op- purpose of this paper is to discuss the mechanics and com-
tical fiber, and a photo-detector or spectrum analyzer. They putational features of the sensor and to present a feasible
are simple and keep the signal processing complexity at a design, which is done next.
low level. The main disadvantage of intensity sensors is their
relatively low sensitivity compared to interferometric or po-
larimetric sensors. 3. Modeling assumptions for the tensegrity sensor
Interferometric sensors are more complicated, consisting
of a reference fiber which is isolated from the perturbation The assumptions made for the mathematical modeling of
being measured and a sensing fiber, extremely susceptible to the tensegrity sensor are: the top is a rigid, homogeneous
this perturbation. The outputs from the two fibers interfere at plate, the bars are rigid, axially symmetric, of negligible lon-
the photo-detector, the magnitude of the interference being gitudinal inertia, the tendons are massless and linear elastic.
dependent on the perturbation. The Mach–Zehnder, Michel- The bars are attached to the fixed base or top through ball
son, Fabry–Perot interferometric sensors, employ the clas- and socket joints. Damping torques which are proportional
sical corresponding interferometer schemes. Their main ad- to the relative angular velocity between the bars and the base
vantage is an excellent strain sensitivity, while the main dis- or top act at the joints. The influence of the gravitational
advantage is that the coherence length of the optical source field is neglected.
limits the maximum sensor length (typically 1 m or less). For The inertial system of reference, b̂1 , b̂2 , b̂3 , is a dextral
most of the interferometric sensors the input/output curve set of unit vectors, whose center coincides with the geo-
is sinusoidal, requiring further post-processing electronics metric center of the base triangle, A11 A21 A31 . Axis b̂3 is
which increases the signal processing complexity. Of the orthogonal to plane A11 A21 A31 pointing upward while b̂1
interferometric sensors mentioned here, the Mach–Zehnder is parallel to A11 A31 . The top reference frame, t̂1 , t̂2 , t̂3 ,
and Michelson interferometers are usually larger in size is a central principal system of the homogeneous top plate
(since they are two-fiber sensors), which could make them B12 B22 B32 .
difficult to embed in the sensing tendon. Fabry–Perot, on the The 18 independent generalized coordinates necessary to
other hand is single fiber, has excellent remote operability describe the configuration of this system are ψ, φ, θ, the
properties with insensitive leads and, in principle, sensors Euler angles for a 3–1–2 sequence to characterize the ori-
of virtually any length can be manufactured. entation of the top reference frame in the inertial one, X,
Modal interferometers are generally single fibers, lending Y , Z, the top mass center inertial Cartesian coordinates, δij ,
themselves to embedment. They can, in principle mea- αij , the declination and the azimuth of the axis of symme-
sure strain and temperature simultaneously providing the try of bar Aij Bij , measured with respect to the inertial frame
C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231 223

(Fig. 1). Hence the vector of generalized coordinates is prestressable configurations (see Fig. 2). These configura-
tions are characterized as follows: all bars are identical of
q = [δ11 α11 δ21 α21 δ31 α31 δ12 α12 δ22 mass m and length l; the top and base are equal equilateral
triangles of side b; all bars have the same declination (δ); the
α22 δ32 α32 ψ φ θ X Y Z]T (1)
vertical projections of points Ai2 , Bi1 , i = 1, 2, 3, onto the
base make a regular hexagon. For simplicity, it is assumed
that all saddle tendons are identical (of rest-length S0 and
4. Statics: prestressable configurations stiffness kS ), all verticals are identical (of rest-length V0 and
stiffness kV ), all diagonals are identical (of rest-length D0
Under no external forces and torques the structure can and stiffness kD ).
attain equilibrium configurations with all tendons in ten- Sultan et al. [31] showed that these configurations proper-
sion, called prestressable configurations. The corresponding ties can be described using three parameters, α, the azimuth
mathematical conditions are (see [31]) of bar A11 B11 , δ, and P, a positive scalar called the preten-
sion coefficient. In this paper we shall consider that under
A(q)T = 0, Tj > 0 (2) no applied forces or torques the sensor yields a symmetrical
where A(q) is the equilibrium matrix given by A[n, j] = prestressable configuration characterized by α = 50◦ , δ =
∂lj /∂qn , n = 1, . . . , 18, j = 1, . . . , 18, lj and Tj are the 30◦ , l = 0.4 m, b = 0.27 m. The corresponding generalized
length and tension of tendon j, respectively. Sultan et al. coordinates values are
[31] published closed form solutions of Eq. (2) for a partic- qe = [30 50 30 290 30 170 30 170 30
ular class of prestressable configurations called symmetrical
50 30 290 300 0 0 0 0 0.58]T (3)
B B B where the angles are given in degrees and the lengths in me-
12 32 22 ters. The tensions in all saddle, vertical, and diagonal ten-
dons are, respectively, equal to TS , TV , TD . The compression
force in all bars is the same, C0 . The state of tension and
compression of the structure is characterized by
B B
31 A
32
11 [ TS TV TD ] = [ 0.29 0.09 0.27 ]P, C0 = 0.37P
A B (4)
22 21
A
12
The bars are designed for buckling and the tendons for max-
imum stress as shown in [27,34], yielding:
  
4l 2 λC
m = ρb πl R2 − R4 − 3 
0
(5)
A
11
A
21
A
31
π Eb
Frontal View
E∗ T∗ γ
k∗ = (6)
B
σ∗max
A 11
32 Here k∗ is the stiffness of tendon ∗, λ and γ are the safety
A
21 coefficients, ρb , Eb , and R the density, Young’s modulus,
B
and exterior radius of a bar (assumed pipe), respectively,
22 whereas E∗ and σ∗max are the Young’s modulus and maxi-
B
12
mum allowed stress in tendon ∗, respectively.
A
^ 12
B b ^
31 2 b
1
O
b
5. Static characteristics
A
α 31
In order to gain further insight into this sensor’s prop-
A
11 erties, we shall investigate its static characteristics when it
is subjected to constant external forces or torques acting
B
32 on the rigid top. For several values of the pretension co-
B
A 21 efficient, P, the tendons are designed for maximum stress
22 according to Eq. (6).
Top View
Consider that a force along the t̂3 axis, F3 , acts on the rigid
Fig. 2. Symmetrical prestressable configuration. top. The plot of vertical displacement of the top mass center
224 C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231

30

20

10
F (N)

0 P = 200
3

P = 150
P = 25 P = 50 P = 100
–10

–20

–30
–0.03 –0.02 –0.01 0 0.01 0.02 0.03
∆Z (m)

Fig. 3. Load deflection characteristic.

vertical coordinate (&Z) versus the amplitude of F3 is called to the following equations of motion (see [33]):
the load–deflection characteristic. If a torque, M3 , along the

6 
4
t̂3 axis acts on the rigid top, the plot of ψ variation (&ψ) mi Mi q̃¨ + dCq̃˙ + ki Ki q̃ + Df + w = 0 (7)
versus M3 is called the torsional characteristic. In order to i=1 i=1
compute these characteristics a tangent predictor–corrector
continuation procedure has been applied to solve the nonlin- wherem 6
i ∈ {Mt , J1 , J2 , J3 , m, J}, ki ∈ {P, kS , kV , kD }
4
ear equations of equilibrium which result from the applica- and i=1 m i M i , dC, i=1 ki Ki are the mass, damping,
tion of the virtual work principle (see [27,29,31] for details). and stiffness matrices, respectively. Detailed formulas for
The results obtained for α = 50◦ , δ = 30◦ , λ = γ = 4, l = these matrices are given in [33]. In the above, q̃ = q − qe ,
0.4 m, b = 0.27 m, E∗ = 1010 N/m2 , σ∗max = 109 N/m2 Mt is the mass of the top, J1 , J2 , J3 the principal mo-
are shown in Figs. 3 and 4. ments of inertia of the top, J the longitudinal moment
At all equilibria on these characteristics, the tendons were of inertia of a bar, d > 0 is the friction coefficient at all
tested for maximum stress (considering γ = 1.5) and slack- joints, f = [ M1 M2 M3 F1 F2 F3 ]T the vec-
ness conditions and the bars for buckling (with λ = 1.5). tor of external forces and torques acting on the rigid
None of these tests failed for the investigated ranges of F3 top, given on t̂i , and w accounts for the inherent plant
and M3 . noise.
We ascertain that the static characteristics vary with the The linearized equations of motion are used to eval-
pretension coefficient, P. As P varies, the shapes of these uate the influence of friction and pretension upon the
curves change from a very nonlinear, hardening spring-like dynamics of the structure, evaluated by plotting the min-
characteristic, to a more linear one. Also the slopes of these imum and maximum natural frequencies (ωmin and ωmax ,
curves at the origin (sensor’s sensitivity) can be modified respectively) versus P and d. Static design constraints,
through pretension adjustment. Eqs. (5) and (6), are employed and the same numerical
values as before are used. In addition the following char-
acteristics of the bars and rigid top are specified: Eb =
7 × 1010 N/m2 ; ρb = 2800 kg/m3 ; R = 0.01 m; Mt =
6. Dynamic characteristics 0.1 kg; J1 = 0.06 kg m2 ; J2 = 0.06 kg m2 ; J3 = 0.1 kg m2 .
The results are shown in Figs. 5 and 6.
The nonlinear equations of motion derivation has been For the investigated range of P and d, we ascertain that
presented in [27,32]. Their linearization around a member slightly damped structures exhibit only oscillatory modes,
of the symmetrical prestressable configurations class leads which, as the friction coefficient increases, transform into
C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231 225

1.5

0.5
(Nm)

0 P = 300
3
M

P = 250
–0.5
P = 200

–1 P = 150
P = 100

–1.5

2
–0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6

∆ψ (rad)

Fig. 4. Torsional characteristic.

Fig. 5. Minimum natural frequency (ωmin ) variation.

pure exponentially decaying ones (Fig. 5). However, the 7. Estimator design
oscillatory nature of the linearized motions does not com-
pletely disappear (ωmax = 0, see Fig. 6). The pretension in- In order to estimate the external forces acting on the rigid
fluence is opposite: natural frequencies increase with P. Fig. top, we cast Eq. (7) in first-order form and consider the
6 shows that for small P the maximum natural frequency linearized measurement equations:
is small; the range of P over which this parameter remains
small increases with d. ẋp = Ap xp + Dp f + wp , z = M p xp + v (8)
226 C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231

Fig. 6. Maximum natural frequency (ωmax ) variation.

Here xp = [ q̃T q̃˙ T ]T , f represents the vector of forces Chebyshev polynomial of first kind and degree j − 1. Hj
and torques acting on the rigid top, to be estimated using are orthogonal over t ∈ [0, τ] with weight g(τ) = τ/2(t(τ −
the measurements z, which consist of some of the tendons t))−1/2 :
lengths, wp = [0 − wT ]T and v represent plant and measure-  τ 
πτ 2 if j = 1
ment noise, respectively. Hj (τ)g(τ)Hk (τ) dτ = &jk (11)
0 4 1 if j = 1
For the purpose of f estimation we choose to represent its
elements as some arbitrary collection of prescribed functions &jk being the Kronecker delta symbol.
of time, which are orthogonal over an interval of length τ. Of particular interest is the case when the Chebyshev
The coefficients of the orthogonal functions are to be auto- system can generate arbitrary linear combinations of Hj ,
matically updated by the use of the real time measurements. j = 1, . . . , h. The next result is a direct consequence of
One possibility is to approximate each element of f by the properties of Chebyshev systems proved by Skelton and
specifying a system of the form: Likins [24].
ṡi = Qi si + Bi &(t), fi = NiT si (9)
Theorem 1. The Chebyshev system in Eq. (9) can generate
where the choices for Qi depend upon the particular orthog- an arbitrary sum of Chebyshev polynomials of degree h−1
onal functions. Here &(t) is the Dirac impulse. The response if and only if the last element of Ni is nonzero.
of the system is equivalent to that of the unforced system,
with initial conditions si0 = si0 + Bi . It follows that there exists an initial conditions vector (si0 )
Various Qi matrices can be constructed to generate as such that h of the coefficients of the Chebyshev polynomial
eigenfunctions of Eq. (9) different orthogonal polynomials. representation of fi can take on arbitrary values if and only
For example, if if Ni (h) = 0. On the strength of this fact we can claim that
  any particular coefficients which are appropriate for fitting
0 0 0 0 0
 1 0 0 0 0 the actual fi can be automatically determined by employing
 
2 0 4 0 0 0 state estimation techniques.
Qi =   (10)
τ 3 0 6 0 0
 Following these observations, for each of the six elements
 0 8 0 8 0 of f we construct a Chebyshev system of degree h − 1:
... h×h
ẋf = Af xf + Bf &(t), f = Cf xf , Af = diag[Qi ],
then any solution of Eq. (9) (arbitrary initial conditions si0 ) Bf = vec[Bi ], Cf (i, h × i) = 0 (12)
yields fi as a sum of Chebyshev polynomials fi = NiT H,
Hj = cos[(j − 1)cos−1 (µ)], µ = 2t/τ − 1 with Hj the with dim(Qi ) = h, ∀i = 1, . . . , 6.
C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231 227

The augmented system which results using the linear mod- is minimized. Here ei (t) is the response of the system given
els of the structure, Eq. (8), and of the external forces and by Eq. (16) when the disturbance (impulse or initial state) is
torques, Eq. (12), is applied only in the ith channel (n∗ is the dimension of the
∗ vector). Under fairly general conditions, this deterministic
ẋa = Aa xa + Ba wa , f = C a xa , z = Ma xa + v, treatment is mathematically equivalent with the stochastic
xaT = [ xpT xfT ], wTa = [ wT &(t) ] T
(13) one in which the disturbances are zero mean, white noise
stochastic processes (see [25]).
  The solution to this optimal estimation problem is given
Ap Dp Cf
Aa = , Ma = [ Mp 0 ], Ca = [ 0 Cf ], by
0 Af
  G = E0 MaT Va−1 (18)
018×18 0
 
Ba =  −I18 0  (14) Aa E0 + E0 ATa − E0 MaT Va−1 Ma E0 + Ba Wa BaT + E0 = 0
0 Bf (19)
The estimation task is accomplished with the aid of a state
where Wa = diag(· · · w2ai · · · ), Va = diag(· · · v2i · · · ), E0 =
estimator designed for the system given by Eq. (13):
diag(· · · e2i0 · · · ), wai , vi being the intensity of the impulsive
x̂˙ a = Aa x̂a + G(z − Ma x̂a ), f̂ = Ca x̂a (15) disturbance applied in the ith channel, and ei0 the ith initial
The error estimation system is condition.

ė = (Aa − GMa )e + Ba wa − Gv, e = xa − x̂a (16)


8. Numerical example
xa0 and x̂a0 being the initial conditions of the system and of
the estimator. Ideally a sensors selection algorithm should be integrated
Under the assumption of observability on the pair with the system design in order to choose an appropriate set
(Aa , Ma ) the poles of the error system can be arbitrarily of sensing tendons out of all possible combinations (see [28]
placed. Since the plant is affected by disturbances (wa , v) for an example of integrated structure design, control system
we design G such that design, and actuators selection). In this article we assume
nxa +nwa +nv  ∞ that previous considerations (e.g. technological ones) have

V= ei (t)T ei (t) dt (17) led to the decision that the sensing tendons are the six saddle
i=1 0 ones.

500
h =5
400
Estimation Error

4
300
3
2
200

100

0
0 5 10 15 20 25 30 35 40
τ
3.5

3 h=2

2.5 3
Decay Rate

2
4
1.5
5
1

0.5

0
0 5 10 15 20 25 30 35 40

τ
Fig. 7. Chebyshev parameters influence upon estimator performance.
228 C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231

– = Exact; – – = Estimated
1.4
M1 exact = 1
1.2

1
M1 (Nm)

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
Time( s)
4

M2 exact = 1+t
3
M2 (Nm)

0
0 0.5 1 1.5 2 2.5 3
Time( s)

Fig. 8. M1 and M2 estimator evaluation.

The structure and estimator design procedures are applied are designed for maximum stress using λ = 4 in Eq. (6).
assuming that the bars, made of Al (for which Eb = 7 × The inertial properties of the proof element (rigid top)
1010 N/m2 , ρb = 2800 kg/m3 ), have an exterior radius R = were assumed to be: Mt = 0.1 kg; J1 = 0.06 kg m2 ; J2 =
0.01 m, are designed for buckling (with γ = 4) according to 0.06 kg m2 ; J3 = 0.1 kg m2 . These values correspond to
Eq. (5), and the tendons (assumed to have all the same me- those used to generate the static and dynamic characteristics.
chanical properties: E∗ = 1010 N/m2 ; σ∗max = 109 N/m2 ) The pretension and friction coefficients are chosen from

– = Exact; – – = Estimated
1.4

1.2 M3 exact = log(1+t)

1
M3 (Nm)

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
Time( s)

F1 exact = 1t 2+0.1 t3
0
F1 (N)

6
0 0.5 1 1.5 2 2.5 3
Time( s)

Fig. 9. M3 and F1 estimator evaluation.


C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231 229

– = Exact; – – = Estimated
1.4
F2 exact = sin(t)
1.2

1
F2 (N)

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
Time( s)

0.5
F3 exact = t
1.5
F3 (N)

0.5

0
0 0.5 1 1.5 2 2.5 3
Time( s)

Fig. 10. F2 and F3 estimator evaluation.

static and dynamic considerations. We choose P = 200 maximum natural frequency has a small value (see Figs. 5
such that the static characteristics in Figs. 3 and 4 look lin- and 6). For estimator design we choose ei0 = 0, wai = 10,
ear for the investigated ranges of F3 and M3 and their slopes vi = 10−5 , BiT = [ 10 0 0 ], Cf (i, j) =&j,h×i (&j,h×i
are not very large, since large slopes result in low sensitivi- being the Kronecker delta symbol).
ties. The friction coefficient is given a value d = 0.6 which In order to choose the Chebyshev system parameters (h
assures that the minimum natural frequency is zero and the and τ) we evaluate their influence upon the estimation error,

= Exact; = Estimated
1.5

0.5
F3(N)

0. 5

1
Exact F3 = sin(1.5 t)
1. 5
0 1 2 3 4 5 6 7 8 9 10
Time (s)

1.5

0.5
F3(N)

0. 5

1
Exact F3 = sin(3 t)
1. 5
0 1 2 3 4 5 6 7 8 9 10
Time (s)

Fig. 11. Estimator evaluation for sinusoidal F3 excitation.


230 C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231

defined by Y = trace(Ca E0 CaT ), and the decay rate, β = ones. The pretension has an opposite effect, its increase
min(|Re(eig(Aa − GMa ))|). These variations are plotted in leading to higher natural frequencies.
Fig. 7. Let us first analyze the influence of h. For fixed τ, fast The estimating task is accomplished by an optimal state
estimation (high decay rate) is obtained for small h. The es- estimator whose inputs are the measurements provided by
timation error also decreases with decreasing h. Apparently, some of the tendons. For estimator design the excitations
a small h would be desirable. On the other hand a small h are represented as some arbitrary collections of Chebyshev
results in complicated functions being approximated by sim- polynomials. The influence of the maximum degree of the
ple functions, which works if the time interval over which Chebyshev polynomials and of the length of the interval
the approximation is computed is small. For fixed h, small τ over which these are orthogonal upon the estimator perfor-
results in fast estimation. But the estimation error increases mance is evaluated by plotting the estimation error and the
with decreasing τ. Thus, a trade-off must be made between decay rate variations. The decay rate increases with decreas-
the speed of the estimator and accuracy. For our design and ing the maximum degree and with decreasing the interval
the following simulations, we choose h = 3 and τ = 1. of orthogonality. The estimation error also decreases with
Comparisons between exact values of the quantities of in- the decreasing of the maximum degree but increases with
terest and those provided by the estimator, when all six ex- decreasing the length of the interval of orthogonality.
ternal forces and torques act on the rigid top, are shown in Numerical simulations of the exact applied forces and
Figs. 8–10. The tensegrity sensor simultaneously provides torques and those estimated by the tensegrity sensor show
fast and accurate estimations of all six quantities. Fig. 11 the validity of the proposed approach. The estimation is fast
shows the numerical simulations of the estimations of F3 enough (settling time less than 1.5 s) and accurate for all six
compared with the exact values, when only a sinusoidal ex- quantities. As with all estimators, previous knowledge of the
citation, F3 , acts. It is easily seen that, as the frequency of maximum possible frequency of the estimated quantity can
external excitation increases, the estimator looses its capa- help in the decision regarding poles allocation.
bility to provide accurate information. The problem can be
fixed by changing the estimator poles locations and making
it fast enough. One way to do this is, as we saw, by changing
the Chebyshev system parameters (e.g. decreasing h and τ). References
On the other hand, if more knowledge about the excitations
to be estimated is available, the Chebyshev system might be [1] C.R. Calladine, Buckminster Fuller’s tensegrity structures and Clerk
replaced with other types of orthogonal filters. For example, Maxwell’s rules for the construction of stiff frames, Int. J. Solids
Struct. 14 (1978) 161–172.
if it is known that the excitations have a certain frequency, a [2] R. Connelly, W. Whiteley, Second-order rigidity and prestress stabil-
harmonic system may be used instead of the Chebyshev one. ity for tensegrity frameworks, SIAM J. Discrete Math. 9 (3) (1996)
We remark that our linear and nonlinear simulations of this 453–491.
tensegrity sensor mathematical models indicated that there [3] S. Djouadi, R. Motro, J.C. Pons, B. Crosnier, Active control of
is practically no difference between the linear and nonlinear tensegrity systems, ASCE J. Aerospace Eng. 11 (2) (1998) 37–44.
[4] R.B. Fuller, Synergetics, Explorations in the Geometry of Thinking,
models responses for the investigated range of parameters
Collier Macmillan, London, 1975.
and disturbances. This is generally the case when the tenseg- [5] A. Hanaor, Prestressed pin-jointed structures—flexibility analysis and
rity structure is highly prestressed (see [33] for details). prestress design, Comput. Struct. 28 (6) (1988) 757–769.
[6] A. Hanaor, Aspects of design of double layer tensegrity, Int. J. Space
Struct. 7 (2) (1992) 101–113.
9. Conclusions [7] N. Kanchanasaratool, D. Williamson, Modeling and control of class
NSP tensegrity structures, Int. J. Contr. 75 (2) (2002) 123–139.
[8] K. Kebiche, M.N. Kazi-Aoual, R. Motro, Geometrical nonlinear
Through the design and performance analysis of a analysis of tensegrity systems, Eng. Struct. 21 (9) (1999) 864–876.
tensegrity sensor for simultaneous measurement of three or- [9] H. Kenner, Geodesic Math and How to Use It, University of Cali-
thogonal forces and three orthogonal torques, we illustrate fornia Press, Berkeley, 1976.
the opportunities tensegrity structures present for sensor [10] C.E. Lee, H.F. Taylor, A.M. Markus, E. Udd, Optical fiber
technology. Fabry–Perot embedded sensor, Opt. Lett. 14 (1989) 1225–1227.
[11] R. Motro, S. Najari, P. Jouanna, Static and dynamic analysis of
The static properties reveal the fact that the pretension
tensegrity systems, in: Proceedings of the ASCE International Sym-
coefficient can be used as a tuning factor of the static char- posium on Shell and Spatial Structures: Computational Aspects,
acteristics: their shape can change from a very nonlinear, Springer-Verlag, New York, 1986, pp. 270–279.
hardening spring-like curve, to an approximately linear one [12] R. Motro, Structural morphology of tensegrity systems, Int. J. Space
as the pretension increases. Also the slopes at the origin (sen- Struct. 11 (1/2) (1996) 25–32.
[13] H. Murakami, Y. Nishimura, Initial shape finding and modal analysis
sitivities) can be modified through pretension adjustment.
of cyclic right-cylindrical tensegrity modules, Comput. Struct. 79
Dynamic characteristics can be tuned through friction and (2001) 891–917.
pretension coefficients adjustment. The natural frequency [14] H. Murakami, Y. Nishimura, Static and dynamic characterization of
decreases with increasing friction and the low frequency regular truncated icosahedral and dodecahedral tensegrity modules,
oscillatory modes transform into pure exponential decaying Int. J. Solids Struct. 38 (2001) 9359–9381.
C. Sultan, R. Skelton / Sensors and Actuators A 112 (2004) 220–231 231

[15] H. Murakami, Static and dynamic analysis of tensegrity structures. [33] C. Sultan, M. Corless, R.E. Skelton, Linear dynamics of tensegrity
Part 1. Nonlinear equations of motion, Int. J. Solids Struct. 38 (2001) structures, J. Eng. Struct. 24 (2002) 671–685.
3599–3613. [34] C. Sultan, R.C. Skelton, Tensegrity structures prestressability inves-
[16] K. Murphy, J.C. Duke Jr., A rugged optical fiber interferometer tigation, Int. J. Space Struct. 18 (1) (2003) 15–30.
for strain measurements inside a composite material laminate, J. [35] C. Sultan, R.E. Skelton, Deployment of tensegrity structures, Int. J.
Compos. Technol. Res. 10 (1988) 11–15. Solids Struct. 40 (18) (2003) 4637–4657.
[17] K.A. Murphy, M.S. Miller, A.M. Vengsarkar, R.O. Claus, [36] A.G. Tibert, S. Pellegrino, Deployable tensegrity reflectors for small
Elliptical-core, two mode, optical fiber sensor implementation meth- satellites, J. Spacecraft Rockets 39 (5) (2002) 701–709.
ods. IEEE J. Lightwave Technol. 8 (11) (1990) 1688–1696. [37] R.D. Turner, T. Valis, W.D. Hogg, R.M. Measures, Fiber-optic strain
[18] Y. Nishimura, H. Murakami, Initial shape finding and modal analysis sensors for smart structures, J. Intell. Mater. Syst. Struct. 1 (1990)
of cyclic frustum tensegrity modules, Comput. Methods Appl. Mech. 26–49.
Eng. 190 (2001) 5795–5818. [38] N. Vassart, R. Motro, Multiparametered formfinding method: ap-
[19] Y. Ohtsuka, T. Ando, Y. Imai, M. Imai, Modal birefringence mea- plication to tensegrity systems, Int. J. Space Struct. 14 (2) (1999)
surements of polarization-maintaining single-mode fibers without and 147–154.
with stretching by optical heterodyne interferometry, J. Lightwave [39] B.B. Wang, X.L. Liu, Integral tension research in double-layer tenseg-
Technol. 5 (1987) 602–607. rity grids, Int. J. Space Struct. 11 (4) (1996) 349–362.
[20] I.J. Oppenheim, W.O. Williams, Vibration of an elastic tensegrity [40] A. Wang, S. Gollapudi, K.A. Murphy, R.G. May, R.O. Claus,
structure, Eur. J. Mech. A: Solids 20 (6) (2001) 1023–1031. Sapphire-fiber-based intrinsic Fabry–Perot interferometric, Opt. Lett.
[21] I.J. Oppenheim, W.O. Williams, Vibration and damping in a 3 bar 17 (14) (1992) 1021–1032.
tensegrity structure, ASCE J. Aerospace Eng. 14 (3) (2001) 85–91. [41] A. Wang, S. Gollapudi, R.G. May, K.A. Murphy, R.O. Claus, Ad-
[22] S. Pellegrino, C.R. Calladine, Matrix analysis of statically and kine- vances in sapphire-fiber-based interferometric sensors, Opt. Lett. 17
matically indetermined frameworks, Int. J. Solids Struct. 22 (4) (21) (1992) 1544–1557.
(1986) 409–428.
[23] A. Pugh, An Introduction to Tensegrity, University of California Biographies
Press, Berkeley, 1976.
[24] R.E. Skelton, J. Likins, Orthogonal filters for model error compen- Cornel Sultan received a PhD in Aeronautics and Astronautics, a MS
sation in the control of nonrigid spacecraft, J. Guidance Contr. Dyn. in Mathematics, both from Purdue University in 1999, and a MS in
1 (1) (1978) 41-49. Aerospace Engineering from Bucharest Polytechnic University in 1992.
[25] R.E. Skelton, T. Iwasaki, K. Grigoriadis, A Unified Algebraic Ap- He was a Research Fellow at Harvard University and recently joined
proach to Linear Control Design. Taylor & Francis, New York, 1997. Scientific Systems Company Inc., MA, as a Senior Research Engineer.
[26] R.E. Skelton, C. Sultan, Controllable tensegrity, a new class of His current research interests are in tensegrity structures, formation
smart structures, in: Proceedings of the SPIE Fourth Symposium flying spacecraft, reusable launch vehicles, and bio-engineering. He has
on Smart Structures and Materials, vol. 3039, 1997, pp. 166– published over 25 papers, holds one patent, and lectured extensively on
177. tensegrity structures.
[27] C. Sultan, Modeling, Design and Control of Tensegrity Structures
with Applications, PhD Dissertation, Purdue University, 1999, p. 200. Robert E. Skelton received a BS in electrical engineering from Clemson
[28] C. Sultan, R.E. Skelton, Integrated design of controllable tensegrity University, a ME in electrical engineering from the University of Alabama,
structures, in: Proceedings of the ASME International Congress and and a PhD from the University of California, Los Angeles, in mechanics
Exposition, vol. 54, 1997, pp. 27–37. and structures. He began his career with Lockheed Missiles and Space
[29] C. Sultan, M. Corless, R.E. Skelton, Peak to peak control of an Company and then moved to Sperry Rand, designing controllers for
adaptive tensegrity space telescope, in: Proceedings of the SPIE Sixth spacecraft. Between 1975 and 1997 he was a Professor of Aeronautics
Symposium on Smart Structures and Materials, vol. 3667, 1999, and Astronautics at Purdue University. He is a Fellow of AIAA and
pp. 190–201. IEEE, the 1991 Russell Severence Springer Professor at the University
[30] C. Sultan, M. Corless, R.E. Skelton, Tensegrity flight simulator, J. of California, Berkeley, a recipient of the Senior Scientist Award form
Guidance Contr. Dyn. 23 (6) (2000) 1055–1064. the Alexander von Humboldt Foundation, and a recipient of a research
[31] C. Sultan, M. Corless, R.E. Skelton, The prestressability problem of award from the Japan Society for the Promotion of Science. He served on
tensegrity structures. Some analytical solutions, Int. J. Solids Struct. the National Research Council’s Aeronautics and Engineering Board and
38/39 (2001) 5223–5252. on the External Independent Review Team for the servicing missions of
[32] C. Sultan, M. Corless, R.E. Skelton, Symmetrical reconfiguration of the Hubble Space Telescope. He has published three books and over 150
tensegrity structures, Int. J. Solids Struct. 39 (2002) 2215–2234. journal papers. He holds patents on the control of tensegrity structures.

Vous aimerez peut-être aussi