Professional Documents
Culture Documents
ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans
level in the system, and this enables the possibility to compensate The paper is organized as follows. The next section states the
for destabilizing nonlinearities, while stabilizing nonlinearities can flexible spacecraft modeling and control problems. Using the
be exploited. Moreover, this approach essentially combines the input-to-state stability characteristic, the adaptive backstepping
adaptive schemes for faster convergence of the estimated param- sliding mode fault-tolerant control law is derived in Section 4.
eters and the tracking error. Several authors have considered the Next, the results of numerical simulations demonstrate various
design of an adaptive supervisory control system using the back- features of the proposed control law. Finally, the paper is com-
stepping method [18,19]. Using backstepping controllers with or pleted with some concluding comments.
without an adaptive mechanism has also been developed for atti-
tude stabilization and tracking of rigid spacecraft in Refs. [20–22] 2. Mathematical model of flexible spacecraft
and the references therein. However, most (if not all) of the previ-
ous research work can hardly be extended to a flexible spacecraft 2.1. Kinematic equation
system when the effect of the motion of the elastic appendages is
taken into account, especially without a priori knowledge of the The unit quaternion is adopted to describe the attitude of the
bound of the modal variables. spacecraft for global representation without singularities [29]. The
unit quaternion q̄ is defined by
Another important problem encountered in practice for space-
cos(Φ /2)
craft control system design is that of control failures. As is well q
q̄ = = 0 (1)
known, control failures can cause control system performance de- n sin(Φ /2) q
terioration and lead to instability and even catastrophic accidents.
where n is the Euler axis, Φ is the Euler angle, and q0 and q are the
Therefore, it is essential to maintain high reliability for space-
scalar and vector components of the unit quaternion, respectively;
craft control system design against possible control faults. A fault-
they are subject to the constraint qT q + q20 = 1. Then the kinematic
tolerant controller could be desirable to find a solution to meet the
equation in terms of the unit quaternion can be given by
desired objective in addition to delivering the desired moments.
Fault tolerant control (FTC) is an area of research that emerges to q̇0 1 qT
= ω (2)
increase availability by specifically designing control algorithms q̇ 2 q0 I + S (q)
capable of maintaining stability and performance despite the oc-
where ω ∈ R3 is the angular velocity of a body-fixed reference
currence of faults, and it has received considerable attention from
frame of a spacecraft with respect to an inertial reference frame
the control research community and aeronautical engineering in
expressed in the body-fixed reference frame, I ∈ R3×3 represents
the past couple of decades. In particular, in Ref. [23], Boškovic et al.
the identity matrix, and S (q) denotes a skew-symmetric matrix
used a multiple models method to detect and isolate actuator faults which is given by
for a spacecraft attitude control system. Based on a dynamically
0 −q3 q2
" #
driven recurrent neural network architecture, an FDI strategy was
proposed for the satellite’s attitude control system when thruster S (q) = q3 0 −q1 . (3)
failures occurred [24]. Chen and Saif [25] presented a fault diagno- −q2 q1 0
sis approach in a satellite system for identifying thruster faults by
using an iterative learning observer. In [26], a robust FDI method 2.2. Relative attitude error kinematics
based on neural state space models was applied to a satellite at-
T
titude control subsystem, and the robustness, sensitivity and sta- Let q̄e = q0e qTe denotethe relative attitude error from a
bility properties of this method were investigated. In Ref. [27], the desired reference frame to the body-fixed reference frame of the
authors use the dynamics inversion and time-delay theory to de- spacecraft; then one may have
sign a passive fault-tolerant controller for a rigid satellite with T
q̄e = q̄ ⊗ q̄− 1
qTe
four reaction wheels to achieve attitude tracking control. To take d = q0e (4)
into account the redundant thrusters, an indirect adaptive fault- where q̄d −1
denotes the inverse of the desired quaternion q̄d with
tolerant control for attitude tracking of rigid spacecraft is proposed T
the definition q̄− 1
= q0d −qTd and ⊗ is the operator for
in the presence of unknown uncertainties, disturbances and actu- d
ator failures, in which a bounded parameter of the lumped pertur- quaternion multiplication, which is defined by
bations is introduced to be updated on-line [28].
q0a q0b − qTa qb
In this paper, an attempt is made to provide an adaptive FTC q̄a ⊗ q̄b = (5)
q0a qb + q0b qa − S (qa )qb
strategy for a flexible spacecraft with redundant actuators, such
as four reaction wheels which are commonly used for attitude for any given two groups of quaternions of qa and qb . As a result,
the relative attitude error can be obtained by
control, which addresses the aforementioned issues. The proposed
control strategy is based on backstepping sliding mode control the- q̇0e 1 −qTe
= (ω(t ) − Rd ωd (t )) (6a)
ory and it is applied to a flexible spacecraft suffering from unknown q̇e 2 q0e I + S (qe )
faults of the reaction wheels, external disturbances, and unknown or
inertia matrix of the spacecraft. A key feature of the proposed strat-
egy is that the design of the FTC is done independently of the q̇0e 1 −qTe
= ω(t ) for ωd = 0 (6b)
information about the faults, and the measurements of the flexi- q̇e 2 q0e I + S (qe )
bility variables are not needed as well. A sufficient condition is also where Rd is the rotation matrix from the desired reference frame
provided such that the proposed fault-tolerant controller guaran- to the body-fixed reference frame, and ωd is the angular velocity of
tees that the closed loop is semi-global input-to-state stable by the desired reference frame with respected to the inertial reference
the Lyapunov-like stability analysis. The effects due to the flexi- frame expressed in the desired reference frame. Note that in this
ble elements of the spacecraft could also be treated provided the paper we consider the case ωd = 0, for convenience, to develop
robustness conditions on the control gains guarantee the satisfac- the control law. It is worth noting that the developed approach
tion, exactly or approximately, of the control requirements, under can be generalized to the case ωd 6= 0 by properly changing the
appropriate conditions on the partial system parameters. Finally, designed controller form (or by adding the terms related to ωd and
applications are carried out on an orbiting spacecraft with flexible ω̇d , respectively, as the feed-forward control inputs of the attitude
appendages. control system).
Y. Jiang et al. / ISA Transactions 49 (2010) 57–69 59
2.3. Flexible spacecraft dynamics Assumption 3. The control effectiveness matrix E and vector f are
also assumed to be unknown during the entire orbiting operation,
Under the assumption of small elastic displacements, the but kDf (t )k is bounded.
dynamic equations of a spacecraft with flexible appendages can be
found in Ref. [9] and the references therein, and are given by Remark 2. For Assumption 1, the structural parameters are sup-
posed to be poorly known, and are constant or can vary during
J ω̇ + δ η̈ = −ω × (J ω + δ η̇) + u(t ) + d(t )
T T
(7a) spacecraft operations. In both cases, since their variation is as-
η̈ + C η̇ + K η + δ ω̇ = 0 (7b) sumed to be slow with respect to the spacecraft dynamics, their
derivatives are or can be considered zero. Even if bounds are as-
where J is the symmetric inertia matrix of the whole structure, δ
sumed, they are not used in the control system design, which is a
is the coupling matrix between the elastic and rigid structure, η
reasonable assumption in practice. For Assumptions 2 and 3, they
is the modal coordinate vector, u(t ) is control torque acting on the
are feasible from the practical point of view for a certain orbiting
main body and generated by, for example, the reaction wheels, and
spacecraft. Note that the notation kxk in this paper denotes the Eu-
d(t ) is the external disturbance including environmental torques
clidean norm of vector x, and kX k is the induced two-norm of a
such as gravitational torque, torque as a result of solar radiation,
1 matrix X .
and magnetic effects, etc.; the matrices C = diag{2ζi Λi2 , i = 1,
2, . . . , N } and K = diag{Λi , i = 1, 2, . . . , N } are the damping 3. Derivation of an adaptive backstepping fault-tolerant con-
and stiffness matrices, respectively, in which N is the number of trol law
1
elastic modes considered, Λi2 is the natural frequency, and ζi is the
In this work, the control objectives are to achieve high precision
corresponding damping ratio.
of attitude tracking and vibration reduction of a flexible spacecraft
in the presence of external disturbances, parameter uncertainties
Remark 1. The above dynamics of the spacecraft are obtained by
and unknown actuator faults. For this, a robust adaptive backstep-
computing the kinetic and potential energies and then applying the
ping control strategy is investigated in this paper. Adaptive back-
Lagrange equations with the assumption of small elastic displace-
stepping is a recursive Lyapunov-based scheme and the idea of it is
ment approximation. This simplified equation is easy to manipu-
to design a controller recursively by considering some of the state
late and more suitable for control law design. Of course, the exact
variables as ‘virtual controls’ and designing intermediate control
model, which is time varying and more difficult to handle, can
laws for them. The advantage of adaptive backstepping compared
be used instead for verifying the effectiveness of the control law
with other control methods lies in its design flexibility, due to its
derived on the basis of the simplified model, to accomplish the ra-
recursive use of Lyapunov functions such that cancellations of use-
tional maneuver and vibration reduction for the closed-loop sim-
ful nonlinearities are avoided and often additional nonlinear terms
ulation later.
are introduced to improve the transient performance.
Let us consider that an actuator fault occurs. In particular, To give a clear idea of such controller design procedure, the
consider the situation in which the actuator loses complete or following variables are defined:
partial control power; a bias fault is also considered [27]. Then the
x1 = qe (10a)
general nonlinear spacecraft attitude dynamics model in Eq. (7a)
with four reaction wheels can be rewritten as x2 = ω − ϕ(qe ) (10b)
attitude parameters and angular velocity vector, a sliding surface where 2γ1 = min {1, λmin (J0 ), λmin (P ), 1/µ1 } and 2γ2 = min
equation in vector form is proposed as follows: {1, λmax (J0 ), λmax (P ), 1/µ1 }; note that λmin (·)(λmax (·)) denotes
the minimum (maximum) eigenvalues of matrix (·).
σ = x2 + β x1 (14)
Upon taking the derivative of V2 along the trajectories of
where β is a positive constant chosen by the designer. Then the x2 Eqs. (13) and (21), we obtain
subsystem is considered, and its time derivative is 1
V̇2 = σ T J0 σ̇ + V̇1 + ξ T P ξ̇ + α̃ α̃˙
ẋ2 = ω̇ − ϕ̇(qe ) = ω̇ + k1 q̇e . (15) µ1
1 1
For the controller design, the following variable is introduced, = − k1 xT1 x1 + xT1 x2 + σ T F (ω, qe0, qe )Θ
2 2
ψ = η̇ + δω, (16)
+ δ T K δ T C − S (ω)δ T ξ − S (ω)DJw (Ω + DT ω)
which simplifies the following developments. Note that
+ DEu(t ) + (Df + d) − δ T C δω
ψ̇ = η̈ + δ ω̇ = −C ψ − K η + C δω. (17)
0 I
−δ 1
+ξ P
T
ξ+ ω + α̃ α̃.
˙ (24)
Then we have −K −C Cδ µ1
h i
0 I −δ 0 I
ξ̇ = ξ+ ω (18) Since the matrix −K −C has all its eigenvalues in the left-hand
−K −C Cδ
plane, there exists a symmetric and positive-definite solution P of
T the Sylvester equation
with ξ = ηT ψ T , and Eq. (8a) can be rewritten as
0 I 0 −K
J0 ω̇ = −S (ω) (J − δ T δ)ω + DJw Ω + δ T ψ
P + P = −2Q (25)
−K −C I −C
+ δ T K δ T C ξ − δ T C δω + DEu(t ) + (Df + d)
(19)
for some positive-definite matrix Q . Therefore, using Eq. (25), and
with J0 , J − δ T δ − DJw DT . after some algebra, we have
Computing the derivative of σ and taking into account the dy- 1 1
namics Eq. (19), one obtains V̇2 = − k1 xT1 x1 − ξ T Q ξ + xT1 x2
2 2
J0 σ̇ = −S (ω)J0 ω + δ T K δ T C − S (ω)δ T ξ + σ T F (ω, qe0, qe )Θ + δ T K δ T C − S (ω)δ T ξ
1.5 6
5 0.05
1 4
3 0
0..5
2 -0.0.5
0 1 40 60 80 100
0 10 20 30 40 50 60 70 80 90 100 0
-1
0 10 20 30 40 50 60 70 80 90 100
0
-0.1 5
5
0
-0.2 0
-5
40 60 80 100 0.05
-0.3 -5
0 10 20 30 40 50 60 70 80 90 100
0
0.8 -10
10 -0.0.5
0.6 -15 40 60 80 100
5
0.4
-20
0 0 10 20 30 40 50 60 70 80 90 100
0.2
40 60 80 100
0
0 10 20 30 40 50 60 70 80 90 100 15
0.05
10
0
0
5 -0.0.5
-0.2 0
40 60 80 100
-2
-0.4 0
-4
40 60 80 100
-0.6 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
time (sec) time (sec)
(a) Time response of quaternion. (b) Time response of angular velocity.
-500 5
0
-1000 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
2 nd reaction wheels' angular speed (rad/sec)
500 5
0
0
-500
-1000 -5
-1500 -10
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
3 rd reaction wheels' angular speed (rad/sec)
500 5
0
0
-500
-1000 -5
-1500 -10
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
4 th reaction wheels' angular speed (rad/sec)
1500 10
1000
5
500
0
0
-500 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
time (sec) time (sec)
(c) Time response of reaction wheel’s velocities. (d) Time response of control torque.
Fig. 1. Attitude maneuvering using the proposed adaptive backstepping sliding mode control method. Case 1: fault-free (solid line); Case 2: fault scenario 1 (dashed–dotted
line).
1
−δ
T
+ kF k α̃ kσ k − α̃(µ1 kF k kσ k)
+
δ K δ T C
+ (k1 + β)
δ T
kωk kξ k
+
P
Cδ
µ1
1 1 + (k1 + β)
δ T K δ T C
kx1 k kξ k + kσ k kDf + dk
≤ − k1 xT1 x1 − ξ T Q ξ − k2 σ T σ + xT1 x2
2 2
62 Y. Jiang et al. / ISA Transactions 49 (2010) 57–69
1 st mode η1
2
1
0
-1
-2
0 10 20 30 40 50 60 70 80 90 100
2 nd mode η2
0.4
0.2
0
-0.2
-0.4
0 10 20 30 40 50 60 70 80 90 100
3 rd mode η3
0.4
0.2
0
-0.2
-0.4
0 10 20 30 40 50 60 70 80 90 100
Vibration energy E (N.m)
2
1.5
1
0.5
0
0 10 20 30 40 50 60 70 80 90 100
time (sec)
(e) Time response of vibration displacements.
Fig. 1. (continued)
α̃
x1 Definition. The closed-loop system (6b), (8) and (27) is semi-
≤ − α̃ xT1 xT2 ξ R1 x
T
globally input-to-state stable (ISS) with respect to state z (t ) and
2
input (d(t ) + Df ) if for any positive constants ∆z and ∆(d+Df )
ξ
there exist gains k1 , k2 and λ such that, given kz (0)k < ∆z and
kx 1 k supt ≥0 kd(t ) + Df (t )k < ∆(d+Df ) , the following bound on the state
" #
kξ k R2 kωk + kσ k k(Df + d)k
− kx 1 k kωk z (t ) is guaranteed to hold:
kξ k
2
≤ −λm kX k + kσ k kDf + dk (28) kz (t )k ≤ max χ (kz (0)k , t ) , φ sup k(d + Df )(τ )k ,
0≤τ ≤t
V̇2 ≤ −λm kX k2 + kω + (k1 + β) x1 k kDf + dk Remark 3. A continuous function φ : [0, a) → [0, ∞) belongs
(k1 + β) k(Df + d)k to class K if is strictly increasing and φ(0) = 0. It belongs to
0
≤ −λm kX k2 + kx1 k
kωk
0 1 k(Df + d)k K ∞ if a = ∞ and φ(r ) → ∞ as r → ∞. The function χ :
√ [0, a) × [0, ∞) → [0, ∞) belongs to class KL if for each fixed
≤ −λm kX k2 + 2κ k(Df + d)k kX k (29) s the mapping χ (r , s) belongs to class K with respect to r and for
each fixed r the mapping χ (r , s) is decreasing with respect to s and
where κ = max{1, (k1 + β)}. χ (r , s) → 0 as s → ∞. See more details in Ref. [30].
Then, for any parameter ρ ∈ (0, 1), the derivative of V̇2 can be Then, using similar arguments as in Ref. [30], we can show that
bounded by a strictly negative quadratic function as whenever the bounds Eqs. (23) and (30) hold in the restricted
√ region for V2 and V̇2 , they can be transformed into the following
V̇2 ≤ −ρλm kX k2 − (1 − ρ)λm kX k2 + 2κ k(Df + d)k kX k bound on the system states:
ρλm
≤ −ρλm kX k2 ≤ −
V2 (30) rγ
γ2 ρλ
2 − 2γm t
kz k ≤ max 2 e 2 kz k (0),
γ1
as long as
1
µ2 0 0 0
2
1 1 1
k2 β 2 + k2 β −
0 k1 0
R1 =
2 4 4
1 1
k2 β −
0 k2 0
4 2
1
0 0 0 Q
2
(k1 + β)
δ T K δ T C
1 1 k2 k1
2 k2 β 2 + k1 + k2 k21 − −
2 2 2
−δ
T
h i
P C δ
+ δ K δ T C
+ (k1 + β)
δ T
k2 k1 1
R2 =
− k2 −
2 2 2
−δ
T
h i
P C δ
+ δ K δ T C
+ (k1 + β)
δ T
(k + β)
δT K
δT C
1
1
− − λmin (Q )
2 2 2
Box I.
Theorem 1. Consider a spacecraft system involving actuator faults Corollary 1. Consider the flexible spacecraft system in Eqs. (6b) and
governed by Eq. (9) under Assumptions 1–3. Given any initial (8) satisfying Assumptions 1–3. With the application of a controller in
condition z (0), there exist large enough control gains k1 , k2 , λ and µ1 Eq. (36),
such that the closed-loop system (6b), (8) and (27) is semi-globally
DT
ISS from an external disturbance input (d(t ) + Df (t )) to the state u(t ) = −k2 σ − k3 +
S (ω)DJw (Ω + DT ω)
z (t ) with the bound restrictions ∆z and ∆(d+Df ) on the state z (t ) λ
and disturbance (d(t ) + Df (t )), respectively, even in the presence of +
δ T C δω
+ kF k α̂ sgn(σ )
(36)
unknown actuator faults.
and the adaptive law in Eq. (27b), if the gain k3 is selected such that
Remark 4. From the above analysis, because the matrix E is not k3 ≥ kd + Df k is satisfied, and with the other parametric constraints
used in the control scheme, there is no need to include a health given in Eq. (27), then we can ensure that x1 , x2 , σ , α̃ , ξ → 0 as
monitoring unit to identify or estimate which actuator is un- t → ∞.
healthy. Knowledge of the degree of failure for each actuator is
Proof. Consider the same Lyapunov function candidate V2 as used
not even needed. The wheel fault accommodation/compensation
earlier; the time derivative of V2 along the system trajectories is
is done automatically and adaptively by the proposed control al-
given, after some algebraic operations, as
gorithm. This feature is necessary to build affordable and effective
fault-tolerant spacecraft control schemes. 1 1
V̇2 ≤ − k1 xT1 x1 − ξ T Q ξ + xT1 x2
2 2
Remark 5. From Eq. (27), the above stability proof is ensured as
+
S (ω)DJw (Ω + DT ω)
+
δ T C δω
long as DEDT is positive definite. That is to say that it is required
+ kF k α + k(Df + d)k kσ k
that the number of active controls after failure should be greater
than or at least equal to 3, i.e. the remaining active wheels are able
−δ
T
+
δ K δ T C
+ (k1 + β)
δ T
to produce a efficient actuating torque vector for the spacecraft to +
P
Cδ
perform the given maneuvers.
× kωk kξ k + (k1 + β)
δ T K δ T C
kx1 k kξ k
Remark 6. To prevent parameter drift and avoid singularities in
1
the adaptive control law, several methods of robustifying the up- + σ T DEu(t ) + α̃ α̃˙
date laws have been suggested in the literature [31] over the years; µ1
the following modified update law is used in the later simulation ≤ −λm kX k2 ≤ 0. (37)
study:
Because V2 > 0 and V̇2 ≤ 0, we can easily show that x1 , x2 , σ ,
α̂˙ = −µ1 µ2 α̂ + µ1 kF k kσ k (34) α̃ , ξ ∈ L∞ . Further, from the integration of Eq. (37), it follows that
X ∈ L2 ∩L∞ , and therefore, from Barbalat’s lemma, we can conclude
where µ2 is designed positive constant.
that x1 , x2 , σ , α̃ , ξ → 0 as t → ∞. The proof is completed.
Remark 7. In practice, the sign function in the control law (27a)
can be replaced by a saturation-like function to reduce the 4. Simulation and comparison results
chattering as
The numerical application of the proposed control schemes
DT
to attitude control of a flexible spacecraft is presented using
u(t ) = − k2 σ −
S (ω)DJw (Ω + DT ω)
λ MATLAB/SIMULINK software, and the results are reported here.
The spacecraft is characterized by a nominal main body inertia
σ
matrix [9]
+
δ T C δω
+ kF k α̂
(35)
kσ k + ε "
350 3 4
#
by introducing a small boundary ε > 0. J = 3 270 10 kg m2 , and the coupling matrices
In the above analysis, we assume that the disturbance (d(t ) + 4 10 190
Df (t )) is simply bounded, i.e., (d(t ) + Df (t )) ∈ L∞ [0, ∞). We 6.45637
1.27814 2.15629
further suppose that (d(t ) + Df (t )) ∈ L2 [0, ∞) ∩ L∞ [0, ∞), which −1.25619 0.91756 −1.67264 1/2
is a rather restrictive hypothesis. According to Theorem 1, then the δ = kg m/s2 ,
1.11687 2.48901 −0.83674
following corollary can be stated: 1.23637 −2.6581 −1.12503
64 Y. Jiang et al. / ISA Transactions 49 (2010) 57–69
ωx (deg/sec)
q0
12
1 10 0.5
0.8 8
0.6 0
6
0.4
4
0.2 -0.5
0 2 40 60 80 100
0 10 20 30 40 50 60 70 80 90 100 0
q1
-2
0 10 20 30 40 50 60 70 80 90 100
0 × 10-3
5 ωy (deg/sec)
-0.1 10
0
-0.2 0
-5
40 60 80 100 0.5
-0.3 -10
0 10 20 30 40 50 60 70 80 90 100
q2 0
× 10-3 -20
0.8 5
0.6 -30 -0.5
0 40 60 80 100
0.4
-40
0.2 0 10 20 30 40 50 60 70 80 90 100
-5
0 40 60 80 100
ωz (deg/sec)
-0.2 30
0 10 20 30 40 50 60 70 80 90 100
25 0.5
q3
20
0 × 10-3 0
4 15
-0.2 2 10 -0.5
40 60 80 100
0 5
-0.4
-2 0
40 60 80 100
-0.6 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
time (sec) time (sec)
(a) Time response of quaternion. (b) Time response of angular velocity.
Fig. 2. Attitude maneuvering using PD control method. Case1: fault-free (solid line); Case2: fault scenario 1 (dashed–dotted line).
1 1 1 1
respectively; the first four elastic modes have been taken into
− −
account in the model used for simulating the spacecraft at ωn1 = 2 2 2 2
1 1 1 1
0.7681, ωn2 = 1.1038, ωn3 = 1.8733, ωn4 = 2.5496 rad/s with D = −
− .
(38)
damping ξ1 = 0.0056, ξ2 = 0.0086, ξ3 = 0.013, ξ4 = 0.025, while √2 √2 √2 √2
2 2 2 2
for designing the controller only the first three mode have been
involved. The inertia matrix Jw is selected as Jr = 10I4×4 kg m2 , 2 2 2 2
and the configuration matrix D of four reaction wheels is given as To test the controller performance with and without various
Y. Jiang et al. / ISA Transactions 49 (2010) 57–69 65
4
2
0
-2
-4
0 10 20 30 40 50 60 70 80 90 100
1
0.5
0
-0.5
-1
0 10 20 30 40 50 60 70 80 90 100
1
0.5
0
-0.5
-1
0 10 20 30 40 50 60 70 80 90 100
Vibration energy E (N.m)
8
6
4
2
0
0 10 20 30 40 50 60 70 80 90 100
time (sec)
(e) Time response of vibration displacements.
Fig. 2. (continued)
combinations of reaction wheel faults, the rest-to-rest maneuver is In addition, for the purpose of comparison, the conventional
considered in the simulation, and the initial conditions have been proportional-derivative (PD) without reconfigurable process is
set at q0 = 0.173648, q1 = −0.263201, q2 = 0.789603 and also performed at the same simulation conditions. In the following
q3 = −0.526402, i.e., a rotation of 160◦ is to be considered in the simulations, the control and adaptation gains were selected by
attitude maneuvering. Note that, due to the change in reference, trial and error until a good performance was obtained for the
the controller regulates the quaternion to the equilibrium point above cases. The controller parameters of the different methods
[1, 0, 0, 0], since this is the closest equilibrium point in terms of (proposed control law and PD) were determined so that all the
rotation path. In addition, the initial modal variables and the time settling times were almost the same in all the schemes for the fault-
derivatives ηi (0) and η̇i (0)(i = 1, 2, 3, 4) are supposed given by free condition.
ηi (0) = η̇i (0) = 0, i.e., the flexible appendages are undeformed. In
addition, in the simulation, the periodic disturbance torque 4.1. Attitude control under a fault-free condition
d(t ) = [0.3 cos(0.01t ) + 0.1 0.15 sin(0.02t ) + 0.3 cos(0.025t ) In this case, first, to show the effect of the proposed adaptive
backstepping sliding mode fault-tolerant controller in Eq. (27),
0.3 sin(0.01t ) + 0.1] T
(39)
the simulation was done under the given initial condition. The
is also considered. In this study, three different sets of simulations time histories of the quaternion of the spacecraft, angular velocity,
are conducted to demonstrate the effectiveness of the proposed modal displacement, required control torque and reaction wheel’s
approach: velocities are shown in Fig. 1 ((a)–(e), solid line). It is noted that
an acceptable desirable orientation response is achieved, and the
(1) Attitude control using the proposed control laws for a fault- spacecraft reached the required angle with a settling time less than
free condition. 30 s. Moreover, the elastic vibrations are significantly suppressed
(2) Attitude control with using proposed control law under fault by considering them in the design, and the oscillations settle within
scenario 1: 30 s. The last plot in Fig. 1 ((e), solid line) shows the vibration
energy response, which is described by E = η̇T η̇ + ηT K η, and it
f = 0.5, 5 ≤ t ≤ 25 s
2 can be observed that the energy has almost no oscillations after 30
e1 = 0.2, t ≥ 3 s
(40) s. This illustrates that the designed controller is capable of reducing
e3 = 0.6, t ≥ 5 s
the system vibration while maintaining the tracking capability of
e4 = 0.4, t ≥ 10 s. the spacecraft.
For the purpose of comparison, the system is also controlled
(3) Attitude control with using proposed control law under fault
by using the traditional PD control. The same simulation cases are
scenario 2:
repeated with this controller and the results of the simulation are
f1 = 0.3,
2 ≤ t ≤ 10 s shown in Fig. 2((a)–(e), solid line). For this case, it can be observed
f3 = 0.6,
5 ≤ t ≤ 20 s that the attitude rotational maneuver can be achieved, but severe
e2 = 0.2, t ≥ 10 s (41) oscillations are excited during maneuvering as demonstrated in
e3 = 0, t ≥5s the modal displacement and vibration energy responses as shown
e4 = 0.8, t ≥ 2 s.
in Fig. 4((e), solid line). Even after maneuvering, the vibrations
66 Y. Jiang et al. / ISA Transactions 49 (2010) 57–69
ωx (deg/sec)
1.5 6
5
1 4 0.05
3 0
0.5
2 -0.0.5
0 1 40 60 80 100
0 10 20 30 40 50 60 70 80 90 100 0
-1
0 10 20 30 40 50 60 70 80 90 100
0
ωy (deg/sec)
-0.1 0
5
-5
-0.2 0
-10
40 60 80 100
-0.3 -5
0 10 20 30 40 50 60 70 80 90 100 0.05
0.8 -10 0
0 5 -0.0.5
-0.2
40 60 80 100
-5
-0.4 0
-10
40 60 80 100
-0.6 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
time (sec) time (sec)
(a) Time response of quaternion. (b) Time response of angular velocity.
-1500 -10
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
3 rd reaction wheels' angular speed (rad/sec)
1000 5
0
0
-1000
-2000 -5
-3000
-10
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
4 th reaction wheels' angular speed (rad/sec)
1000 10
500 5
0 0
-500 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
time (sec) time (sec)
(c) Time response of reaction wheel’s velocities. (d) Time response of control torque.
Fig. 3. Attitude maneuvering using the proposed sliding mode control method. Case 1: fault-free (solid line); Case 2: fault scenario 2 (dashed–dotted line).
still exist. Despite the fact that there still exists some room for 4.2. Attitude maneuver control under faulty conditions
improvement with different design control parameter sets, there
is not much improvement in the attitude and velocity responses. Here two actuator fault scenarios are considered, as described
From the comparison between Figs. 1 and 2, the performance of the by Eqs. (40) and (41), respectively. The first case is that a bias fault
proposed design controller is better than that of the conventional occurs at the second wheel for 20 s after 5 s after the simulation
PD control even if these designs will adapt the system parameters starts; the first, third and fourth wheels lose 80%, 40% and 60%
under external disturbances. of the control power after 3 s, 5 s, and 10 s after the simulation
Y. Jiang et al. / ISA Transactions 49 (2010) 57–69 67
2
1
0
-1
-2
0 10 20 30 40 50 60 70 80 90 100
0.4
0.2
0
-0.2
-0.4
0 10 20 30 40 50 60 70 80 90 100
0.4
0.2
0
-0.2
-0.4
0 10 20 30 40 50 60 70 80 90 100
Vibration energy E (N.m)
2
1.5
1
0.5
0
0 10 20 30 40 50 60 70 80 90 100
time (sec)
(e) Time response of vibration displacements.
Fig. 3. (continued)
starts, respectively. Figs. 1 and 2 ((a)–(e), dashed–dotted line) for constraints on the control magnitude and elastic deflection. These
this case show the results of the same motion commands used in control approaches provides the theoretical basis for the practical
the previous subsections for the proposed and PD control methods, application of the advanced control theory to a flexible spacecraft
respectively. It can be seen that high control precision and a good attitude control system.
tracking process are still obtained for the proposed controller,
and no significant amount of vibration occurred, whereas the PD 5. Conclusions
controller shows the degradation of tracking performance after the
actuator fault. In addition, the PD controller also shows tracking A fault-tolerant adaptive backstepping sliding mode control
ability, to some degree, because of its own robustness. scheme has been developed for flexible spacecraft attitude ma-
The second case is that the third wheel loses complete control neuvering using redundant (four) reaction wheels in the presence
power after 5 s, a bias fault occurs at the first and third wheels for 8 of parametric uncertainty disturbances and even unknown faults.
s and 15 s after 2 s and 5 s respectively, and the second and fourth The proposed control design methods do not require the system
wheels loses 80% and 20% of the control power 10 s and 2 s after the identification process to identify the faults as well as the process
simulation starts, respectively. This kind of combination is a harder of fault detection and isolation. The control formulation is based
upon Lyapunov’s direct stability theorem; the semi-globally input-
fault scenario than the first case. As shown in Figs. 3 and 4(a–3,
to-state stability of the closed-loop system is ensured, and the ro-
dashed–dotted line), the proposed controller works effectively and
bustness to disturbances, unknown faults and elastic vibrations is
achieves the control target, while for the PD controller, it can be
also guaranteed provided that appropriate robustness conditions
observed that the system performance is significantly degraded
on the controller gains are satisfied. The control designs have been
after the actuator faults; moreover, severe vibrations result from
evaluated using numerical simulation comparisons between the
the control law as compared with the proposed methods. What is developed approach and other referenced schemes, and the ex-
more, the vibrations still exist during the tracking process, which pected performances have been shown to occur. In the simula-
deteriorates the performance of the attitude maneuvers. tions, several different types of reaction wheel failure scenarios
Summarizing all the cases (normal case and fault cases), it is were investigated. Based upon the results presented in the paper,
noted that the proposed controller design method can significant it is concluded that the fault-tolerant adaptive backstepping slid-
improve the normal performance over that of the PD method in ing mode control scheme successfully handles failures if one re-
both theory and simulations. Also, in the fault case, the proposed action wheel fails completely, or if the efficiency of one or several
methods have better results than those of conventional cases. It can reaction wheels decreases. While the simulation results presented
also be observed that as more and more faults are considered in the in this paper merely illustrate formulations for a particular atti-
design, the proposed controllers can still guarantee the tracking tude maneuver, further testing would be required to reach any con-
performance. In addition, extensive simulations were also done clusions about the efficacy of the control and adaptation laws for
using different control parameters, disturbance inputs and even a tracking arbitrary maneuvers. In addition, this fault-tolerant con-
combination of reaction wheel faults. These results show that in trol scheme places no restriction on the magnitude of the desired
the closed-loop system attitude control and vibration stabilization control, and a design explicitly considering the actuator limit is also
are accomplished in spite of these undesired effects in the system. being investigated. Future work is planned to study the digital im-
Moreover, the flexibility in the choice of control parameters can plementation of such a control scheme on hardware platforms for
be utilized to obtain desirable performance while meeting the attitude control experimentation.
68 Y. Jiang et al. / ISA Transactions 49 (2010) 57–69
ωx (deg/sec)
30
1 25 0.5
0.8 20
0.6
15 0
0.4
0.2 10
-0.5
0 5 40 60 80 100
0 10 20 30 40 50 60 70 80 90 100 0
-5
0 10 20 30 40 50 60 70 80 90 100
0
5 ωy (deg/sec)
-0.1 10
0
-0.2 0
-5
40 60 80 100 0.2
-0.3 -10
0 10 20 30 40 50 60 70 80 90 100
-20 0
0.8 × 10-3
2
0.6 -30 -0.2
40 60 80 100
0.4 0
-40
0.2 0 10 20 30 40 50 60 70 80 90 100
-2
0 40 60 80 100
ωz (deg/sec)
-0.2 30
0 10 20 30 40 50 60 70 80 90 100
25 0.1
20
0 15 0
5
-0.2 0 10
-0.1
-5
5 40 60 80 100
-0.4 0
-10
-0.6 40 60 80 100 -5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
time (sec) time (sec)
(a) Time response of quaternion. (b) Time response of angular velocity.
Fig. 4. Attitude maneuvering using the PD control method. Case 1: fault-free (solid line); Case 2: fault scenario 2 (dashed–dotted line).
4
2
0
-2
-4
0 10 20 30 40 50 60 70 80 90 100
1
0.5
0
-0.5
-1
0 10 20 30 40 50 60 70 80 90 100
1
0.5
0
-0.5
-1
0 10 20 30 40 50 60 70 80 90 100
Vibration energy E (N.m)
8
6
4
2
0
0 10 20 30 40 50 60 70 80 90 100
time (sec)
(e) Time response of vibration displacements.
Fig. 4. (continued)
References [16] Lee KW, Nambisan PR, Singh SN. Adaptive variable structure control of aircraft
with an unknown high-frequency gain matrix. Journal of Guidance, Control,
[1] Singh SN. Rotational maneuvers of nonlinear uncertain spacecraft. IEEE and Dynamics 2008;31(1):194–203.
Transactions on Aerospace Electronic and Systems 1988;24:114–23. [17] Krstic M, Kanellakopoulos I, Kokotovic P. Nonlinear and adaptive control
[2] Nagata T, Modi VJ, Matsuo H. Dynamics and control of flexible multi-body design. New York: Wiley; 1995.
systems part II: Simulation code and parametric studies with nonlinear [18] Ciliz MK. Adaptive backstepping control using combined direct and indirect
control. Acta Astronautica 2001;49:595–610. adaptation. Circuits, Systems and Signal Processing 2007;26(6):911–39.
[3] Karray F, Grewal A, Glaum M, Modi V. Stiffening control of a class of nonlinear [19] Kim KS, Kim Y. Robust backstepping control for slew maneuver using
affine system. IEEE Transactions on Aerospace Electronic and Systems 1997; nonlinear tracking function. IEEE Transactions on Control Systems Technology
33(2):473–84. 2003;11(6):822–9.
[4] Hosi SM, Maghami PG, Kelkar AG. Design of dynamic dissipative compensators [20] Zhou D, Shen TL, Tamura K. Nonlinear and adaptive nonlinear controllers for
for flexible space structures. IEEE Transactions on Aerospace Electronic and attitude stabilization and tracking of a spacecraft. Transactions of the Japan
Systems 1995;31(4):1314–24. Society for Aeronautical and Space Sciences 2005;48(159):7–12.
[5] Junkins JL, Bang H. Maneuver and vibration control of hybrid coordinate [21] Kristiansen R, Nicklasson PJ. Satellite attitude control by quaternion-based
systems using Lyapunov stability theory. Journal of Guidance, Control, and backstepping. In: Proceedings of the 2005 American control conference. 2005,
Dynamics 1993;16(4):668–76. p. 907–12.
[6] Hu QL, Ma GF. Variable structure control and active vibration suppression [22] Singh SN, Yim W. Nonlinear adaptive backstepping design for spacecraft
of flexible spacecraft during attitude maneuver. Aerospace Science and attitude control using solar radiation pressure. In: Proceeding of 41st IEEE conf.
Technology 2005;9:307–17. decision control. 2002, p. 1239–44.
[7] Iyer A, Singh SN. Variable structure slewing control and vibration damping of [23] Boškovic JD, Li S-M, Mehra RK. Intelligent control of spacecraft in the presence
flexible spacecraft. Acta Astronautica 1991;25(1):1–9. of actuator failures. In: Proceedings of the 38th IEEE conference on decision &
[8] Iyer A, Singh SN. Sliding mode of control of flexible spacecraft under
control. 1999. p. 4472–77.
disturbance torque. International Journal of Systems Science 1990;21(9):
[24] Li L, Ma LY, Khorasani K. A dynamic recurrent neural network fault diagnosis
1755–71.
and isolation architecture for satellite’s Actuator/thruster failures. Lecture
[9] Hu QL. Sliding mode maneuvering control and active vibration damping of
Notes in Computer Science 2005;3498(3):574–83.
three-axis stabilized flexible spacecraft with actuator dynamics. Nonlinear
[25] Chen W, Saif M. Observer-based fault diagnosis of satellite systems subject to
Dynamics 2008;52(3):227–48.
[10] Singh SN, Araujo AD. Adaptive control and stabilization of elastic spacecraft. time-varying thruster faults. Journal of Dynamic Systems, Measurement, and
IEEE Transactions on Aerospace and Electronic Systems 1999;35(1):115–22. Control 2007;129(3):352–6.
[11] Maganti GB, Singh SN. Simplified adaptive control of an orbiting flexible [26] Wu Q, Saif M. Robust fault diagnosis for a satellite large angle attitude system
spacecraft. Source: Acta Astronautica 2007;61(7–8):575–89. using an iterative neuron PID observer. In: Proceedings of the American control
[12] Singh SN, Zhang R. Adaptive output feedback control of spacecraft with flexible conference. 2006. p. 5710–5.
appendages by modeling error compensation. Acta Astronautica 2004;54(4): [27] Jin JH, Ko SH, Ryoo CK. Fault tolerant control for satellites with four reaction
229–43. wheels. Control Engineering Practice 2008;16:1250–8.
[13] Hu QL, Shi P, Gao HJ. Adaptive variable structure and commanding shaped [28] Cai WC, Xiao XL, Song YD. Indirect robust adaptive fault-tolerant control for
vibration control of flexible spacecraft. Journal of Guidance, Control, and attitude tracking of spacecraft. Journal of Guidance, Control and Dynamics
Dynamics 2007;30(3):804–15. 2008;31(5):1456–63.
[14] Zeng Y, Araujo AD, Singh SN. Output feedback variable structure adaptive [29] Sidi MJ. Spacecraft dynamics and control. Cambridge (UK): Cambridge
control of a flexible spacecraft. Acta Astronautica 1999;44(1):11–22. University Press; 1997.
[15] Hu QL, Ma GF. Adaptive variable structure maneuvering control and vibration [30] Khalil H. Nonlinear systems. 3rd ed. Upper Saddle River (NJ): Prentice-Hall;
reduction of three-axis stabilized flexible spacecraft. European Journal of 2002.
Control 2006;12(6):654–68. [31] Ioannou PA, Sun J. Stable and robust adaptive control. Prentice-Hall; 1995.