Professional Documents
Culture Documents
DSCC2014
October 22-24, 2014, San Antonio, TX, USA
DSCC2014-5979
1
q̇ = Ξ(q)ω (3)
2
where
q4 I3×3 + [ρ×]
Ξ(q) =
−ρ T
0 −a3 a2
FIGURE 1. Definitions of the coordinate frames [a×] = a3 0 −a1 , a ∈ R3
−a2 a1 0
are reviewed. In Sec.3, an optimal sliding manifold is presented
along with its optimality proof and stability analysis. Then, a The matrix Ξ(•) obeys the following properties,
Lyapunov stability analysis is used to derive an asymptotically
stabilizing sliding control law and a parameter adaptation law. ΞT (q)Ξ(q) = I3×3
A robust controller is designed to reject external disturbances.
In Sec.4, numerical simulations are shown to demonstrate the ΞT (a)a = 03×1 , ∀a ∈ R4 (4)
closed-loop performance of the proposed controller. d T
Ξ (q)q̇ = ΞT (q)q̈
dt
2 PRELIMINARIES From these properties, one can show that if the desired attitude
2.1 Coordinate frames trajectory is specified by qd = [ρdT , q4d ]T , the desired angular ve-
We define three coordinate frames of interest. The inertial locity ωd must obey,
frame of reference, in which Newton’s law is satisfied, is
denoted as Fr . We attach three mutually perpendicular axes ωd = 2ΞT (qd )q̇d
to the spacecraft, and call this the body-fixed frame Fb . The (5)
spacecraft is modeled as a rigid body actuated by either thrusters ω̇d = 2ΞT (qd )q̈d
or momentum wheels in three orthogonal directions. The body
axes are chosen to coincide the directions of actuations. The From the quaternion definition, one can see that q and −q repre-
desired spacecraft attitude is described by a frame denoted Fd . sent the same physical rotation. Hence compared with algebraic
The frame definitions are depicted in Figure 1. subtraction, the error calculated from quaternion multiplication
provides a better way because it resolves the sign ambiguity. The
quaternion error and multiplication are defined as,
T
2.2 Kinematics δρ Ξ (qd )q
δq = q ⊗ q−1 −1 −1
= Ξ(qd ), qd q = =
The unit quaternion is used to describe the spacecraft atti- d δ q4 qT qd
tude, (6)
−1 −ρd
qd =
qd4
ρ ê sin(θ /2)
q= = (1)
q4 cos(θ /2) δ q represents the rotation from Fd to Fb .
where ê is a unit vector representing the axis of rotation, θ is the 2.3 Dynamics
angle of rotation from Fr to Fb . It has to satisfy the following For a spacecraft having its three thrusters aligned with the
unity norm constraint, body axes, the dynamic equations of motion are given by
If the angular velocity of the spacecraft with respect to Fr , ex- where J is the positive definite inertial matrix of the spacecraft,
pressed in Fb , is denoted as ω, the quaternion kinematic equa- u is the torque generated by the thrusters.
3 METHODOLOGY
We apply sliding mode control for controller synthesis. The
idea of sliding mode control is to allow the transformation of a 3.1 Calculus of variations
controller design problem for a general n-th order system to a In [16], the necessary conditions for optimality are derived
simple stabilization problem with reduced order, i.e., stabilizing from calculus of variations. For the functional minimization
the dynamics associated with the switching function. Then for problem in Eqn.(9), the necessary conditions are summarized in
the equivalent reduced-order system, intuitive feedback control Table 1. It can be shown by direct substitution that the following
strategies can be applied. optimal angular velocity ω ∗ ,
∂ J∗
Proof : We expand ∂t (q,t) by the chain rule, We denote the column vector of the inertia parameters as J.
which shows that V̇ is negative semi-finite. Hence s and Jˆ are 3.6 Robust controller
bounded. Invoke Barbalat’s lemma, To take into account unexpected external disturbances in
practice, we slightly modify Eqn. (7) by adding a combined dis-
V̈ = −2sT K ṡ (30) turbance input d that can be from air drag, solar pressure, gravity
gradient, magnetic field, spherical harmonics,
The boundedness of ṡ can be seen by combing Eqns.(7), (24)
and (26). This implies the uniform continuity of V̇ , hence we J ω̇ = −[ω×]Jω + u + d (36)
conclude that V̇ → 0. Equivalently s → 0 as t → ∞. Furthermore,
to analyze the convergence of parameter estimation, we consider
Although d is unknown, but its magnitude has known bounds
D ∈ R3 ,
J ṡ + Ks = F J̃ (31)
|di (t)| ≤ Di , ∀t > 0, i = {1, 2, 3} (37)
All the terms except ṡ are uniformly continuous. Thus ṡ is uni-
formly continuous. From Barbalat’s lemma again, ṡ → 0. There-
fore, The robust sliding mode controller is given by,
To enforce the asymptotic parameter estimation, i.e. J̃ → 0, the where ki = Di + ηi for i = 1, 2, 3 and ηi ’s are non-negative con-
following persistent excitation condition must be satisfied, stants. With the same Lyapunov function used before, we can
show that the Lie derivative is now,
Z t+T
F T (δ q, ω, ωd , ω̇d )F(δ q, ω, ωd , ω̇d )dτ ≥ εI6×6 , ∀t ≥ to 3
t
(33) V̇ ≤ − ∑ ηi |si | − sT Ks (39)
i=1
where T,to , ε are some positive scalars.
For the momentum wheel model, the sliding control law Again, the state variables are guaranteed to reach the sliding
and parameter adaptation laws are, manifold regardless of unknown disturbances. To avoid control
chattering after reaching the sliding manifold, saturation func-
tions can replace sign functions [2].
u = −F Ĵ − GĴw − Ks
Ĵ˙ = Γ−1 F T s
(34) 4 Numerical Simulations
Ĵ˙ w = Γ−1 GT s In this section, we show the proposed controller perfor-
G = −[ω×]L(v) + L(ω̇d ) − L(rsgn[δ q4 ]δ ρ̇) mance through numerical simulations. The proposed controller
1.5
||s(t)||
φ̇ = 0.001745 rad/sec
θ = 0.3927 rad (40) 1
ψ̇ = 0.04859 rad/sec
0.5
Then φ , ψ can be obtained by integration. The desired quaternion
trajectory can be computed by converting Euler angle parameter- 0
0 5 10 15
ization to unit quaternions,
time (sec)
sin( θ2 ) cos( φ −ψ
2 ) FIGURE 2. Plot of the norm of switching funciton s(t) which con-
φ −ψ
sin( θ2 ) sin( 2 )
qd = (41) verges to 0
cos( θ2 ) sin( φ +ψ
2 )
cos( θ2 ) cos( φ +ψ
2 )
1 1
By numerically differentiating qd , we can compute q̇d , q̈d and
0.8
ωd ,ω̇d by Eqn. (5). We let the actual inertial matrix J be, 0.5
q1
q2
0.6
0
20 5 1 0.4
J = 5 17 3 (42) 0.2 −0.5
1 3 15 0 10 20 30 0 10 20 30
time (sec) time (sec)
0.8 1
Our proposed controller does not require knowledge of the inertia
ˆ of the inertia
of the spacecraft, so we use an initial estimate J(0) 0.6
0.8
with 30% error,
q4
q3
0.4
0.6 qd (t)
0.2
26 1.6 1.4
q(t)
0 0.4
ˆ = 1.6 13 1.2
J(0) (43) 0 10 20 30 0 10 20 30
time (sec) time (sec)
1.4 1.2 8.5
√ √ √
Furthermore, 90o error angle is used along [1/ 3, 1/ 3, 1/ 3]T FIGURE 3. Plot of q(t) and qd (t) showing asymptotic quaternion
as the initial condition, tracking
√1
3
sin(π/4)
√1 sin(π/4) and consequently the control objectives are achieved. The con-
⊗ qd (0)
3
q(0) = 1 (44) vergence of the switching function is shown in Figure 2. The
√ sin(π/4)
3 asymptotic quaternion and velocity tracking performance are
cos(π/4) plotted in Figures 3 and 4 respectively. The controlled thruster
torque is shown in Figure 5.
and the spacecraft is at rest initially ω(0) = [0, 0, 0]T . The
controller parameters are set to be, r = 3, K = 10 · I3×3 , Q = I6×6 .
4.2 Robust performance
If the spacecraft is subject to disturbances, the robust sliding
mode controller in Eqn. (38) should be used. In our simulation,
the following disturbance is used,
4.1 Nominal performance
Without external disturbance, we have shown that the state T
will be driven to the sliding manifold using Barbalat’s lemma, d(t) = sin(t), −1, cos(t) (45)
q2
q1
0.4 0.6
0
0.3 0.4
0.8 1
0
0.6
−0.1 0.8
0 5 10 15
q3
q4
0.4
time (sec) q d (t)
0.6
0.2 q (t)
0
0.6
u1 δω1
u(N · m)
−5 u2 0.5 δω2
u3 δω3
ω − ωd (rad/sec)
0.4
−10
0.3
−15
0.2
−20 0.1
0 5 10 15
time (sec)
0
−0.1
FIGURE 5. Plot of the controlled thruster torque 0 5 10 15
time (sec)
0
u(N · m)
−5
−10
−15
−20
0 5 10 15
time (sec)