Professional Documents
Culture Documents
UC Santa Cruz
A University of California author or department has made this article openly available. Thanks to
the Academic Senate’s Open Access Policy, a great many UC-authored scholarly publications
will now be freely available on this site.
Let us know how this access is important for you. We want to hear your story!
http://escholarship.org/reader_feedback.html
Peer Reviewed
Title:
Automatic mass balancing of a spacecraft three-axis simulator: Analysis and experimentation
Journal Issue:
Journal of Guidance, Control, and Dynamics, 37(1)
Author:
Chesi, S
Gong, Q
Pellegrini, V
Cristi, R
Romano Prof, M
Publication Date:
01-01-2014
Series:
UC Santa Cruz Previously Published Works
Permalink:
http://escholarship.org/uc/item/7q4549j9
DOI:
http://dx.doi.org/10.2514/1.60380
Local Identifier:
1002686
Abstract:
Spacecraft three-axis simulators provide frictionless and, ideally, torque-free hardware simulation
platforms that are crucial for validating spacecraft attitude determination and control strategies. To
reduce the gravitational torque, the distance between the simulator center of mass and the center
of rotation needs to be minimized. This work proposes an automatic mass balancing system for
spacecraft simulators, which uses only the three sliding masses during the balancing process,
without need of further actuators. The proposed method is based on an adaptive nonlinear
feedback control that aims to move, in real time, the center of mass toward the spacecraft
Copyright Information:
All rights reserved unless otherwise indicated. Contact the author or original publisher for any
necessary permissions. eScholarship is not the copyright owner for deposited works. Learn more
at http://www.escholarship.org/help_copyright.html#reuse
the distance between the simulator center of mass and the center of rotation needs to be minimized. This work
proposes an automatic mass balancing system for spacecraft simulators, which uses only the three sliding masses
during the balancing process, without need of further actuators. The proposed method is based on an adaptive
nonlinear feedback control that aims to move, in real time, the center of mass toward the spacecraft simulator’s center
of rotation. The stability of the feedback system and the convergence of the estimated unknown parameter (the
distance between the center of mass and the center of rotation) are analyzed through Lyapunov stability theory. The
proposed method is experimentally validated using the CubeSat Three-Axis Simulator at the Spacecraft Robotics
Laboratory of the Naval Postgraduate School.
imbalance compensation based on estimation techniques is to the CR. Furthermore, due to the small entity of the masses and the
considered in Sec. III. Section IV presents the simulation results, low speed of their motion, it has been assumed that the effect of the
and the experimental results on the spacecraft simulator are presented motion of the balancing masses on the angular momentum can be
in Sec. V. Finally, Sec. VI states the conclusions of this work. neglected. In Eq. (4), ri × is a cross-product matrix defined as
2 3 23
II. Transversal Imbalance Compensation 0 −r3 r2 r1
r× 4 r3 0 −r1 5 with r 4 r2 5 (5)
A. System Dynamic Model
−r2 r1 0 r3
The spacecraft simulator kinematics and dynamics can be modeled
using two coordinate systems as shown in Fig. 2: the inertial
coordinate systems Xi , Y i , Zi and a coordinate system fixed to the To balance the spacecraft simulator, the Earth gravitational field gb in
simulator body xb, yb , zb . Both coordinate systems have their origins the body coordinate system must be computed using the simulator
in the center of rotation. The orientation of the body-fixed coordinate attitude through the direction cosine matrix
system with respect to the inertial coordinate system defines the
attitude of the spacecraft simulator. gb Rbi gi (6)
The spacecraft three-axis simulator can rotate freely, but cannot
translate, because the CR is a fixed point in the inertial coordinate where gi 0 0 −9.81 T m∕s2 is the gravitational field in the
system. Each balance mass is assumed to be an ideal mass particle inertial coordinate system, and Rbi is the rotation matrix defined as
Downloaded by UNIV OF CALIFORNIA - SANTA CRUZ on June 3, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.60380
vector ri represents the position of the ith balance mass with respect
where ri , the ith balance mass’s position with respect to the CR, is the
control input to be designed.** By considering Eq. (2) and taking into
consideration the effect of τ r , the system dynamics can be written as
**Notably, the effect of the displacement of the three equal masses, sliding
each in one direction, on the generation of the gravitational torque, is the same
that it would be obtained by using a single mass displaced in three dimensions
by the sum of the displacement vectors of the three masses. However, using
three masses has the advantage of leaving diagonal the matrix of principal
Fig. 2 Inertial and body coordinate systems. inertia with respect to the center of rotation.
200 CHESI ET AL.
where ωp represents the angular velocity perpendicular to the Earth’s Because the matrix −gb t× is always singular, r cannot be directly
gravitational field. found by inverting it. However, due to property (24), the designed
To design the state feedback control and the adaptive law for Θ,
^ the control torque τ r is guaranteed to be in the range of −gb t× for
following positive definite Lyapunov function is used: all t, so that Eq. (25) always has a solution for r. Indeed, a solution is
given by
Vq; ω; Θ
~ 1 ωT Jω 1 Θ
2 2
~ TΘ
~ 1 qT q
2
(16)
r kggb k×τ2 mr
b
p
(26)
The time derivative of the Lyapunov function along the motion of the
system leads to To see this, substituting Eq. (26) into Eq. (25), and using the condition
gb ⊥ τ r , ∀ t from Eq. (24), it can be found that
_ ωT Jω
_ Θ _~
~ TΘ
Vt qT q_
g b × gb × τ r gb · τr gb − gb · gb τ r
ω −ω × Jω ΦΘ τ r Θ
T _~
~ TΘ −mp −mp τr
kgb k2 mp kgb k2 mp
_~
ωT ΦΘ ωT τ r Θ
~ TΘ (17) (27)
where Eqs. (9) and (11) have been used, together with the hypothesis Therefore, the control applied to the mass balancing system is given
by Eq. (26) with τ r be designed according to Eq. (20).
of constant J. From Eqs. (12) and (17) became
J−1
xx · gz · Θy 0
b ~ − J−1
yy · gz · Θx 0 0 · Θz 0
b ~ ~ (37)
Since J−1 −1
xx , J yy , gz ≠ 0, Eq. (37) implies Θx Θy 0. Thus, rx ,
b ~ ~ off
roff
y have been correctly estimated and compensated. In fact, by
considering Eqs. (20) and (23), the right-hand side of Eq. (9) is equal
to zero and this implies that no gravitational torque is acting on the
spacecraft (i.e., the distance between CM and CR has been
compensated at least in two directions). But nothing can be claimed
about the third component of Θ. ~ This component will be estimated
using a Kalman filter as demonstrated in the following section.
can be obtained by replacing the control actions into Eq. (2) so that [18–21]. For the dynamic system described in Eq. (2), either UKF or
EKF can be used, and the performance is expected to be similar. In
this study, the UKF is used to estimate the offset roff
z by a standard
_ J−1 −ω × Jω ΦΘ
ω ~ − kp ωp (29) approach of augmenting this unknown parameter into the state
space [22].
The projection factor P p in Eq. (14), can be written as To write the system in state-space form, let us define the state
2 3 2 b 2 3 vector of the continuous system as x ωx ; ωy ; ωz ; roff T
z . Here, the
1 0 0 gx gbx gby gbx gbz off
unknown parameter rz is augmented into the state with the
Pp q 4 0 1 0 5 − 4 gby gbx gby 2 gby gbz 5 kg1b k (30) corresponding dynamics to be r_off
z 0. The disturbance torque due
0 0 1 gbz gbx gbz gby gbz 2 to the imbalance along the zb axis is
Because on the set Ω, ωp t ≡ 0, the spacecraft simulator is rotating roff × ms∕c gb −ms∕c gby roff b off
z ;s∕c gx rz ; 0
T
(38)
around the gravity field along the zb axis so that the components gbx ,
gby of the projection factor go to zero. As a consequence, Therefore, the spacecraft simulator dynamics in the state-space form
is given by
2 3
1 0 0 x_ fx dt (39)
Pp q 4 0 1 0 5 (31) where
0 0 0
2 0 2 31 3
−ms∕c · gby · roff
z
Premultiplying Eq. (29) by the projection factor P p leads to 6 J−1 @−ω × Jω 4 m · gb · roff 5A 7
f 6
4
s∕c x z 7
5 (40)
0
dP p ω
Pp ω
_ _ p Pp J −1 −ω × Jω ΦΘ
ω ~ − K p ωp (32) 0
dt
and dt represents the process noise. Let the measurement be the
where the time derivative of the projection factor is equal to zero only angular velocity given by the inertial measurement unit (IMU). Then
on the set fωp t ≡ 0g. Since ω ωg ωp on fωp ≡ 0g, Eq. (32) the output equation is given by y Hx vt, where
becomes
2 3
1 0 0 0
0 Pp J−1 −ωg × Jωg ΦΘ
~ (33) H 40 1 0 05 (41)
0 0 1 0
Because J is diagonal and ωg has only one component along the zb
axis, it can be shown that ωg × Jωg 0. Therefore, Eq. (33) and vt represents the measurement noise.
becomes To correctly estimate the offset in the last direction using the UKF,
the observability of the unknown parameter needs to be analyzed. In
0 Pp J−1 ΦΘ
~ (34) this paper, the definition of observability for the nonlinear system
introduced in [23] has been adopted. Consider a general nonlinear
where the matrix Φ can be written, from Eq. (11), as system
2 3 ξ_ ft; ξ; u; ξ ∈ Rn ; u ∈ Rm Y ht; ξ; u (42)
0 gbz 0
Φ ms∕c 4 −gbz 0 05 (35) where ξ is the state, ut is a continuous control input, and Y
0 0 0 represents the variable that can be directly measured by sensors.
Suppose z zt; ξ; u is a variable to be estimated. Let U represent
And so, Eq. (34) can be written as an open and connected set in the time-state-control space
R × Rn × Rm .
2 32 −1 3 32 3 Definition 1 (Observability) [23]: The function z zt; ξt; ut
1 0 0 Jxx 0 0 0 gbz 0 Θ~ x
is said to be observable in U if for any two trajectories t; ξi t; ui t,
0 ms∕c 4 0 1 0 54 0 J−1
yy 0 5 −gbz 0 0 54 Θ~ y 5
i 1; 2 in U defined on a same interval t0 ; t1 , the equality
0 0 0 0 0 J−1
zz 0 0 0 Θ~z
(36) hξ1 t; u1 t hξ2 t; u2 t; a:e: in t0 ; t1 (43)
zt; ξ1 t; u1 t zt; ξ2 t; u2 t (44) graduate School. The spacecraft simulator inertia matrix is
J diag0.0226; 0.0257; 0.0266 kg · m2 . The total mass is
almost everywhere in t0 ; t1 . Suppose for any trajectory t; ξt; ut ms∕c 4.2 kg with the balance masses mx my mz 0.3 kg.
in U there always exists an open set U1 ⊂ U so that t; ξt; ut is An offset is simulated with values roff 1 − 0.9 − 1.4T · 10−3 m,
contained in U1 and zt; ξ; u is observable in U1 . Then, z the initial angular velocity ω 0.0888 0.8229 1.3611 T rad∕s,
zt; ξ; u is said to be locally observable in U. and quaternion q 0 0 0 1 T . In the simulation, the sliding
Lemma 1 [23]: Consider a system without control masses are used as actuators. The feedback control law in Eq. (20) is
generated using the simulated data of angular velocity, gravity field,
ξ_ ft; ξ; ξ ∈ Rn Y ht; ξ (45) and quaternions. Furthermore, the simulation assumes that the data
related to the gravity field, angular velocity, and quaternions have the
Let U ⊂ R × Rn be an open set. Consider same sample time T s 0.05 s. The maximum translation distance
for each sliding mass is 28 mm providing a total offset com-
V Y T ; DY T ; : : : : : : ; Dl−1 Y T T (46) pensation capability of 2 mm in each direction. The uncertainty in
the sliding masses’ positions has been modeled using Gaussian white
for some l > 0, where D is the differentiation operator. If noise distribution with standard deviation σ 0.5 · 10−6 m. More
details on the CubeTAS simulator are presented in Sec. V.
∂V In the first step of the proposed automatic balancing procedure, the
rank n
∂ξ adaptive control designed in Sec. II is applied to eliminate the
Downloaded by UNIV OF CALIFORNIA - SANTA CRUZ on June 3, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.60380
Y ω 0.02
2 0 2 31 3 0.01
−ms∕c · · gby roff
z
6 −1 B 6 7C 7 0
_ 6
DY ω J B−ω × Jω 6 ms∕c · gbx · roff 7C 7 (47)
4 @ 4 z 5A 5 −0.01
−0.02
0 0 50 100 150 200 250 300 350 400
2.4 0.03
1.8
1.2 0.02
0.6
0 0.01
−0.6
0 50 100 150 200 250 300 350 400 0
0.01
2.4 0 5 10 15 20
1.8
1.2
−5
0.6 x 10
0 2
−0.6
0 50 100 150 200 250 300 350 400 1.5
1
2.4
1.8 0.5
1.2
0.6 0
0 0 5 10 15 20
Downloaded by UNIV OF CALIFORNIA - SANTA CRUZ on June 3, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.60380
−0.6
0 50 100 150 200 250 300 350 400
Fig. 7 Simulation results: Top graph shows the estimation of roff
z and
the real value. Lower graph shows the estimation error roff
e using UKF.
Fig. 6 Simulation results: magnitude of ω, ωp and ωg during
transversal balancing procedure.
estimated, and then it can be compensated to drastically reduce the
gravitational torque acting on the spacecraft simulator.
tuned in a way that the sliding mass travel distance is always less than
To experimentally verify the effectiveness of the automatic
its maximum value, as shown in Fig. 4.
balancing procedure proposed here, the following method will be
As shown in Fig. 5, during the mass balancing procedure, Θ~ x and
implemented on the CubeTAS. After the balancing procedure, an
Θ~ y converge to zero, implying that the offsets in xb and yb directions
arbitrary angular velocity will be applied to the system. The total
are balanced. Θ~ z does not converge to zero, because the offset in the
energy, which is the sum of the kinetic energy and the potential
third direction roff
z cannot be balanced with the proposed adaptive energy, is conserved, that is,
control. These simulation results agree with our analysis presented in
preceding sections. Figure 6 shows the behavior of the angular
Etot Er t Pg t constant (55)
velocity during the first step of the balancing procedure. The angular
velocity perpendicular to the gravity field ωp tends to zero and
ω → ωg . Thus, the spacecraft remains to spin along the gravity vector where Pg t represents the potential energy and Er t
2 ω tJωt represents the kinetic rotational energy. If the system
1 T
as proven in Sec. II using Lyapunov stability analysis.
Once the first step is completed, the offset in the directions is balanced, then Pg t ≅ 0. Therefore, the rotational kinetic energy
perpendicular to the gravity field has been compensated. The offset in is conserved during the motion, that is, Er t Er 0 constant.
the last direction roff Thus the variation of the rotational kinetic energy in time serves as a
z can be estimated in the second step. At the
beginning of the second step, an angular velocity with components in measure of the effectiveness of the mass balancing procedure.
all directions is applied to the spacecraft simulator. This guarantees Because the rotational kinetic energy can be evaluated by the
the observability of the unknown parameter roff gyroscope measurement, this quantity provides an easy way to
z as explained in
Sec. III. During the rotational motion, a UKF is adopted to estimated demonstrate the accuracy of the balance procedure experimentally.
roff Figure 8 shows the variations of the kinetic energy before and after
z . The associated measurement and process covariance matrices,
considering the hardware setup, have been chosen as the balancing procedure. When the system is unbalanced, the
variation in amplitude of the kinetic energy will be greater than the
amplitude variation with respect to the balanced system. As shown in
R EvvT diag4.87 × 10−5 2 ; 4.87 × 10−5 2 ;
51 Fig. 8, the kinetic energy of the unbalanced system has a standard
4.87 × 10−5 2 rad2 ∕s2
deviation σ unbalance 3.847 × 10−3 J, where the balanced system
has a standard deviation σ balance 4.5331 × 10−6 J; thus this value
Q EddT diag10−10 ; 10−10 ; 10−10 ; 1 × 10−4 2
(52)
×rad2 ∕s4 ; rad2 ∕s4 ; rad2 ∕s4 ; m2 ∕s2 0.02
0.012
The simulation results using the preceding gain matrices are shown in
Fig. 7, where the error between roff z and its estimation has been
0.01
defined as
e krz − r^z k
0.008
roff off off (54)
0.006
It can be seen that the filter converges to the correct value of roff
z in less
−6
than 20 s. Indeed, roff
e decreased to less than 10 m after the filter
0.004
converges. The small error in the estimation shown in Fig. 7 depends 0 5 10 15 20 25 30 35 40
on the UKF sample time T s 0.05 s. In fact, simulations show that a
reduction of the sample time leads to a reduction of roff e . Thus, it can Fig. 8 Simulation results: kinetic energy before Ebr and after Ear the
be claimed that the offset in the last direction has been correctly balancing procedure.
204 CHESI ET AL.
Downloaded by UNIV OF CALIFORNIA - SANTA CRUZ on June 3, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.60380
Fig. 9 CubeSat three-axis simulator at the Spacecraft Robotics Laboratory of the Naval Postgraduate School (Lab Director, Marcello Romano).
has been reduced by approximately 99.9%. The value of the standard of all the components were determined in three-dimensional (3-D)
deviation after balancing procedure is different from zero because the CAD software NX 7 by Siemens. The property of each component
uncertainties in the sliding masses’ positions have been modeled as was derived by the CAD file provided by the component’s
Gaussian white noise distribution. By adding noise into the sliding manufacturer. The CubTAS 3-D CAD model is shown in Fig. 10. The
masses’ positions, the system is subjected to a torque that prevents the upper part is mainly for three reaction wheels and a battery, whereas
kinetic energy from being a constant. However, the sensible the lower part is for PC-104 boards, power management boards, and
reduction in the kinetic energy standard deviation is an indication of balancing masses with relative linear actuators. The IMU provides
an effective balance of the system. information regarding angular velocity and magnetic and gravity
fields. A magnetometer, sun sensor, three magnetic coils, and three
reaction wheels are used for attitude estimation and control. During
V. Experimental Results the automatic mass balancing experiment, only the data relative to
A. Experimental Apparatus
angular velocity and gravity field are used.
CubeTAS is a CubeSat-scale three-axis hardware simulator
B. Automatic Mass Balancing Experiments
designed to recreate an ideally torque-free platform for testing
nanosatellites’ attitude determination and control algorithms in a lab In the first step of the balancing procedure, the offset along the xb
environment. The experimental apparatus shown in Fig. 9 has been and yb directions will be compensated. At the beginning of the first
developed at the Spacecraft Robotics Laboratory of the Naval experiment, an initial angular velocity of ω 0.0888 0.8229
Postgraduate School. 1.3611T rad∕s is applied to the spacecraft simulator. As the
The CubeTAS simulator contains an automatic balancing system, designed adaptive nonlinear control takes action, the IMU registers
consisting of three stainless steel masses driven by three linear the gravity field components and angular velocity values. Shown in
stepper motors along their axis, respectively. The sliding masses are Fig. 11 are the gravity field vector components during the rotational
custom made to encapsulate the stepper motors. The whole system is motion. It can be seen that, after t 200 s, the only component of the
supported by a spherical air bearing that allows it to rotate 50 deg gravity field that is different from zero is the one aligned to the zb axis.
about two axes in the horizontal plane and 360 deg along the vertical This is in agreement with our theoretical analysis. Figure 12 shows
axis [24]. A Helmholtz cage has been constructed around the the magnitude of the components of the angular velocity. Solid lines
spacecraft simulator to allow the simulation of the orbital magnetic are the experimental results and dashed lines are simulation results.
field. Attached to the Helmholtz cage, a high-accuracy motion As shown in Fig. 12, the experimental data relative to the magnitude
capture camera system measures the attitude. The main design goal of the angular velocity ωp tend to zero, so that, at the end of the first
for CubeTAS is to have a shape suitable to be fit in a sphere as shown step of the balancing procedure, the spacecraft simulator remains to
in Fig. 10. The mechanical design was made such that the offset spin only along the zb axis. This behavior was expected from the
between the center of mass and the center of rotation is minimum. To simulation results of the proposed balancing method as shown by the
meet these goals, the physical shape was designed and the locations dashed lines.
4
0
so that each diagonal element reflects the square of each experimental
−4 initialization error. By analyzing the gyroscope output signal in the
−8 steady-state condition, the measurement covariance matrix has been
−12 chosen to be
0 50 100 150 200 250 300 350 400
Fig. 11 Experimental results: gb components during transversal time should be as close as possible to zero, as shown in Fig. 13.
balancing procedure. The energy conservation method proposed in Sec. IV has been
implemented after the mass balancing procedure, to further verify the
effectiveness of the balancing method. Before the mass balancing
The discrepancies between experimental and simulation results are procedure, an angular velocity has been given to the system and the
particularly evident after t 250 s. In fact, the work of the kinetic rotational energy has been recorded. The procedure is
unmodeled dissipative torque acting on the spacecraft simulator, such repeated after the balancing procedure. Figure 14 shows that the
as aerodynamic drag, causes the system to lose energy and, variation in amplitude of the kinetic energy term is drastically
consequently, to reduce its angular velocity ω. For this reason, as reduced after the mass balancing procedure; thus, it is reasonable to
shown in Fig 12, the experimental value kωg k decreases in time conclude that the proposed mass balancing procedure has produced a
instead of converging to a constant. Another difference between significant reduction of the offset between the center of mass and the
experimental and simulation results appears at the beginning of center of rotation. Indeed, the standard deviation has changed from
the automatic balancing procedure, where the simulated and σ unbalance 2.8223 × 10−4 J to σ balance 8.2474 × 10−5 J, which is
experimental systems, under the same control law, react slightly a reduction of ≈71%. As expected, the reduction in the standard
different. Such difference may be caused by unmodeled dynamics. deviation of the kinetic energy is less than the one obtained in the
For example, the inner-loop control that adjusts the position of simulation. This is because all the measurements from the IMU
the moving masses to the desired location is not modeled in relative to angular velocity and gravity field are affected by noise that
the simulation. Nevertheless, these discrepancies do not affect the has not been modeled in the simulation. Furthermore, other
effectiveness of the balancing procedure because, in both cases, the dissipative torques that are difficult to simulate affect the spacecraft
angular velocity ωp → 0, as deduced analytically in Eq. (22). simulator. The aerodynamic drag is one of the major dissipative
In the second step of the automatic balancing procedure, to torques present during laboratory experimentation. The presence of
compensate the offset in the zb axis, an unscented Kalman filter is different wires and components generates an irregular shape that,
used to estimate the unknown offset. In the experimental validation, exposed to the air flow during the rotational motion, produces a
the UKF sample time has been set equal to onboard computer sample dissipative aerodynamic torque that is difficult to model.
time Ts 0.05 s. The initial error covariance matrix P0 has been Nevertheless, by considering low value of angular velocity, the
chosen as magnitude of the dissipative torque can be considered very small. The
presence of these unmodeled dynamics degrades the estimation
performance. For this reason, the reduction in the standard deviation
P0 diag0.22 ;0.22 ;0.22 ;2 × 10−3 2 rad2 ∕s2 ;rad2 ∕s2 ;rad2 ∕s2 ;m2
of the kinetic energy during the laboratory experimentation is less
with respect to the simulated one shown in Fig. 8. In particular, by
2.4
1.8 −3
x 10
1.2 2.5
0.6
0 2
−0.6
0 50 100 150 200 250 300 350 400
1.5
2.4
1
1.8
1.2
0.6 0.5
0
−0.6 0
0 50 100 150 200 250 300 350 400
−0.5
2.4
1.8
−1
1.2
0.6
0 −1.5
−0.6
0 50 100 150 200 250 300 350 400
−2
0 5 10 15 20 25 30 35 40
Fig. 12 Experimental and simulation results: Solid lines are
experimental data; dashed lines are simulated data of the magnitude of Fig. 13 Experimental results: roff
z estimation using UKF during vertical
ω, ωp , and ωg during transversal balancing procedure. balancing procedure.
206 CHESI ET AL.
0 10 20 30 40 50
Mexican Society of Instrumentation, Vol. 4, No. 5, 2000, pp. 76–81.
[10] Hatcher, N. M., and Young, R. N., “Automatic Balancing System for
Fig. 14 Experimental results: kinetic energy before Ebr and after Ear the Use on Frictionlessly Supported Attitude-Controlled Test Platforms,”
balancing procedure. NASA TN-D-4426, Langley Research Center, March 1968.
[11] Yang, Y., and Cao, X., “Design and Development of the Small Satellite
Attitude Control System Simulator,” AIAA Modeling and Simulation
observing the energy profile shown in Fig. 14, it is possible to see that Technologies Conference and Exhibit, AIAA Paper 2006-6124,
the energy tends to decrease in time. This decreasing behavior is due Keystone, CO, Aug. 2006.
to the work of the dissipative torques that affect the system. [12] Li, Y., and Gao, Y., “Equations of Motion for the Automatic Balancing
System of 3-DOF Spacecraft Attitude Control Simulator,” Third
International Symposium on Systems and Control in Aeronautics and
VI. Conclusions Astronautics (ISSCAA), IEEE, June 2010, pp. 248–251.
This paper presents an automatic mass balancing system for [13] Jung, D., and Tsiotras, P., “3-DoF Experimental Test-Bed for Integrated
spacecraft simulators, which allows real-time relocation of the center Attitude Dynamics and Control Research,” AIAA Paper 2003-5331,
of mass to the center of rotation using only sliding masses. Compared Aug. 2003.
[14] Schwartz, J. L., and Hall, C. D., “System Identification of a Spherical
with manual balancing, the proposed automatic balancing system can Air-Bearing Spacecraft Simulator,” AAS Paper 2004-122, Feb. 2004.
greatly reduce the labor involved in the balancing process and save [15] Woo, H., Perez, O. R., Chesi, S., and Romano, M., “CubeSat Three
design time. The use of sliding masses as the only actuator makes the Axis Simulator (CubeTAS),” Proceedings of the AIAA Modeling and
implementation relatively easy and the proposed automatic balancing Simulation Technologies Conference, AIAA Paper No. 2011-6271,
system suitable for small-size spacecraft simulators. To overcome the Aug. 2011.
physical limitation of the actuator (i.e., the torque generated by [16] Likins, P. W., Elements of Engineering Mechanics, McGraw–Hill, New
moving masses is constrained in the direction perpendicular to the York, 1973, pp. 438–439.
gravity field), a two-step design is proposed. The performance of the [17] Leine, R. I., and Wouw, N., “Stability and Convergence of Mechanical
proposed automatic balancing method is theoretically justified using Systems with Unilateral Constraints,” Lecture Notes in Applied
and Computational Mechanics, Vol. 36, Springer, New York, 2008,
Lyapunov stability analysis and is demonstrated at simulation level. pp. 145–148.
The experiment conducted on a CubeSat-scale simulator also shows [18] Crassidis, J., and Markley, F., “Unscented Filtering for Spacecraft
the efficacy of the proposed method; however, the presence of Attitude Estimation,” Journal of Guidance, Control, and Dynamics,
unmodeled dynamics (e.g., aerodynamic drag) and different sources Vol. 26, No. 4, 2003, pp. 536–542.
of noise degrades the performance. Further work is needed to doi:10.2514/2.5102
incorporate dissipative torques and to analyze the effect of various [19] Julier, J., and Uhlmann, J. K., “A New Extension of the Kalman Filter to
noises to improve the accuracy. Nonlinear Systems,” Signal Processing, Sensor Fusion, and Target
Recognition VI, Vol. 3068, 1997, pp. 182–193.
[20] Matthew, C. V., Schwartz, J. L., and Christopher, D. H., “Unscented
References Kalman Filtering for Spacecraft Attitude State and Parameter
[1] Kim, B. M., Velenis, E., Kriengsiri, P., and Tsiotras, P., “Designing a Estimation,” Proceedings of the AAS/AIAA Space Flight Mechanics
Low-Cost Spacecraft Simulator,” IEEE Control Systems, Vol. 34, No. 4, Conference, AAS Paper 04-115, 2005.
Aug. 2003, pp. 26–37. [21] Sekhavat, P., Gong, Q., and Ross, I. M., “NPSAT1 Parameter Estimation
[2] Schwartz, J. L., Peck, M. A., and Hall, C. D., “Historical Review of Air Using Unscented Kalman Filtering,” Proceedings of the American
Bearing Spacecraft Simulators,” Journal of Guidance, Control, and Control Conference, IEEE, Piscataway, NJ, July 2007, pp. 4445–4451.
Dynamics, Vol. 26, No. 4, 2003, pp. 513–522. [22] Simon, D., Optimal State Estimation: Kalman, H ∞ , and Nonlinear
doi:10.2514/2.5085 Approaches, Wiley, Hoboken, NJ, 2006, pp. 403–407.
[3] Smith, G. A., “Dynamic Simulators for Test of Space Vehicle Attitude [23] Kang, W., and Barbot, J.-P., “Discussions on Observability and
Control System,” Proceedings of the Conference on the Role of Invertibility,” Seventh IFAC Symposium on Nonlinear Control Systems,
Simulation in Space Technology, Part C, Virginia Polytechnic Inst. and Aug. 2007.
State Univ., Blacksburg, VA, Aug. 1964. [24] Chesi, S., Pellegrini, V., Gong, Q., Cristi, R., and Romano, M.,
[4] Fullmer, R. R., “Dynamic Ground Testing of the Skipper Attitude “Automatic Mass Balancing for Small Three-Axis Spacecraft Simulator
Control System,” 34th AIAA Aerospace Science Meeting and Exhibit, Using Sliding Masses Only,” Advances in the Astronautical Sciences,
AIAA Paper, 1996-0103, Jan. 1996. Vol. 142, No. 3, 2012, pp. 2547–2562.