Академический Документы
Профессиональный Документы
Культура Документы
Abstract: In this paper, linear-quadratic regulator (LQR) method is proposed for the design of nonlinear spacecraft control systems. The
Downloaded from ascelibrary.org by Northeastern Univ Library on 04/22/13. Copyright ASCE. For personal use only; all rights reserved.
proposed design is based on the linearized spacecraft model that involves only three components of the quaternion. By using a simple and
special structure of the linearized reduced quaternion model, an analytical solution for the controller. It is shown that the analytical solution of
the state feedback matrix is an explicit function of the cost matrices Q and R. The analytic solution makes it convenient in design practice to
tune the feedback matrix and the cost matrices Q and R to balance the requirements between response performance and fuel consumption. It is
shown that the designed controller globally stabilizes the nonlinear spacecraft system, whereas it locally optimizes the spacecraft perfor-
mance. A design example is provided to show the effectiveness of the design method. DOI: 10.1061/(ASCE)AS.1943-5525.0000142.
© 2012 American Society of Civil Engineers.
CE Database subject headings: Spacecraft; Design; Control systems.
Author keywords: Quaternion; Spacecraft; Linearization; Nonlinear; Attitude control.
u ¼ ðu1 ; u2 ; u3 ÞT
∂g
¼ 03
be the control torques in body frame. Denote ∂q ω ¼ 0
2 3 q1 ¼ q2 ¼ q3 ¼ 0
0 ω3 ω2
ω× ¼ 4 ω3 0 ω1 5 Therefore
ω2 ω1 0 " # " #" # " #
ω_ 03 03 ω J 1
as a matrix operator that maps the cross product ω × x to a matrix ¼ 1
þ u ¼ Ax þ Bu ð7Þ
product ω × x. It is well-known that the spacecraft dynamics system q_ 2 I3 03 q 03
equation is given by (Wertz 1978)
where
" # " # " #
J ω_ ¼ ω × ðJωÞ þ u ð2Þ 03 03 ω J 1
A¼ 1
; x¼ ; and B¼ ð8Þ
It is assumed that the constant inertia matrix J is a diagonal ma- 2 I3 03 q 03
trix because this is approximately correct in most real spacecraft
designs. Let It is easy to verify that this linearized spacecraft system equation
is controllable. It will be shown in the next section that there exist
α α diagonal matrices D and K such that the state feedback
q0 ¼ cos ; q ¼ ½q1 ; q2 ; q3 T ¼ ^eT sin
2 2
u ¼ ½ D K x ¼ Gx ð9Þ
and
T is the LQR design for the linearized quaternion system Eq. (7), and
α α the LQR design globally stabilizes the nonlinear spacecraft system
q ¼ ½q0 ; qT T ¼ cos ; ^eT sin ð3Þ
2 2 described by Eqs. (2) and (4) .
be the quaternion that represents the rotation of the body frame rel-
ative to the inertial frame, where ^e = unit rotational axis; and α = LQR Design
rotational angle about the rotational axis. The nonlinear spacecraft
kinematics equations of motion can be represented by the quatern- Because the linearized spacecraft system (7) is fully controllable,
ion (Wie 1998; Wertz 1978; Sidi 1997) as follows: LQR design can be directly obtained (Athans and Falb 1966). The
advantage of the LQR design is obvious because we can find op-
q_ ¼ 12 ω × q þ 12 q0 ω timal K and D to achieve other design goals, such as minimizing the
ð4Þ control energy (required because of the restrictions on control au-
q_ 0 ¼ 12 ωT q
thority and saturation) and optimizing the response performance.
Yang (2010) showed that the nonlinear spacecraft kinematics Instead of solving nonlinear Lyapunov matrix equation to get Q
equations of motion [Eq. (4)] can be replaced by a set of indepen- and R matrices, analytical feedback formulas can be found because
dent nonlinear spacecraft kinematics equations of motion that leads of the simple and special structure of the linearized reduced
to a controllable linearized quaternion model. quaternion spacecraft model.
Lemma 1 (Yang 2010): For any given q, if α ≠ π, there exists a For the linearized reduced quaternion spacecraft system Eq. (7)
one-to-one mapping between ω and q. _ Moreover, let the linear quadratic cost function to be minimized is
Z
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 ∞ T
f ðqÞ :¼ q0 ¼ 1 q21 q22 q23 L¼ ½x Qx þ uT Rudt ð10Þ
2 0
F¼ ; Q¼ ð13Þ 2
F T12 F 22 03 Q2
¼ ωT Q1∕2
2 R1∕2 ω × Jω
1 1∕2
where the elements of F and Q in Eq. (13) = 3 by 3 matrices. Sub- ωT Q1∕2
2 Q1 þ ðJR1∕2 Q1∕2 1∕2 1∕2
2 þ Q2 R JÞ ωψ ð20Þ
2
stituting (8) and (13) into Eq. (12) and using simple manipulations
yield Because P, J, Q, and R are all diagonal positive definite matri-
" # ces, the second term of the last expression is negative definite. If
F 11 J 1 R1 J 1 F 11 F 11 J 1 R1 J 1 F 12 Q1 1
2 R ¼ cI, i.e., R ¼ cQ2 ; or Q2 R ¼ cJ, i.e., R ¼ cQ2 J, where c is
a constant, then the first term vanishes, therefore, dV∕dt is negative
F T12 J 1 R1 J 1 F 11 F T12 J 1 R1 J 1 F 12 semidefinite, and the nonlinear system described by nonlinear sys-
" #
2 ðF 12 þ F 12 Þ þ Q1
1 T 1 tem Eqs. (2) and (4) is globally stable with the optimal controller
2 F 22
¼ 1
ð14Þ given by Eq. (18). Actually, it can be shown that the closed loop
2 F 22 Q2 nonlinear system is asymptotically stable. Let S ¼ fxjVðxÞ _ ¼ 0g.
Because D and K are full rank matrices, clearly S ¼ fx ¼ ðω; qÞ ¼
Because Q and R are positive definite and F T12 ¼ F 12 , comparing ð0; qÞg: From Eq. (2) because u ¼ Dω Kq ≠ 0 if q ≠ 0, no
the (2,2) block on both sides of Eq. (14) yields solution can always stay in S except x ¼ ðω; qÞ ¼ ð0; 0Þ: By using
a well-known result in Khalil 1992 (Corollary 3.2), the origin is
1∕2 globally asymptotically stable. By definition (Khalil 1992,
F 12 ¼ JR1∕2 Q2 ð15Þ p. 111), the region of attraction of the nonlinear system is the whole
space spanned by x.
Because J, Q, and R are diagonal, substituting Eq. (15) into the In system design practice, if the performance and the local sta-
(1,1) block of Eq. (14) gives bility are the only design considerations, Q and R can be chosen
without any restriction; if the global stabilization is also required
1 1∕2 for nonlinear spacecraft system, some restriction, though it is mild,
F 11 ¼ JR1∕2 Q1 þ ðJR1∕2 Q1∕2 1∕2 1∕2
2 þ Q2 R JÞ ð16Þ has to be placed on Q and R, i.e., either R ¼ cQ2 or R ¼ cQ2 J,
2
where c is any positive constant.
To establish the relationship between the closed loop poles and
Substituting Eqs. (15) and (16) into the (2,1) block of Eq. (14)
the design matrices Q, and R, Eq. (18) can be simplified further as
gives
follows. Let Q1 ¼ diagðq1i Þ; Q2 ¼ diagðq2i Þ; and R ¼ diagðr i Þ, the
matrices D and K can be further simplified as
1∕2 1∕2 1∕2
F 22 ¼ 2Q2 Q1 þ JR1∕2 Q2 ð17Þ 1∕2
1
D ¼ R1∕2 Q1 þ ðJR1∕2 Q1∕2 2 þ Q 1∕2 1∕2
2 R JÞ ¼ diagðd i Þ
2
Eqs. (15)–(17) give a complete solution of Lyapunov matrix sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffi !
Eq. (12). Therefore, Eq. (11) can be rewritten as q1i q2i
¼ diag þ J ii ð21Þ
ri ri
uðtÞ ¼ R1 BT FxðtÞ ¼¼ ½R1 J 1 F 11 ; R1 J 1 F 12 x
with
¼ ½D; Kx ð18Þ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffi
q1i q2i
di ¼ þ J ii ð22Þ
Under some additional conditions, the LQR optimal control ri ri
given by Eq. (18) globally stabilizes the nonlinear system described
by Eqs. (2) and (4). Let and
rffiffiffiffiffiffi
q2i
K ¼ R1∕2 Q2
1∕2
P¼ Q1∕2 R1∕2 J ¼ diagðk i Þ ¼ diag ð23Þ
2 ri
Let the linear matrix transformation T ij ðCÞ be a matrix with the controller on the nonlinear structure are twofold. First, the LQR
following properties: (1) the (i,j) element of T ij ðCÞ is C, (2) the controller globally asymptotically stabilizes the nonlinear space-
diagonal elements are ones, (3) all the remaining elements are ze- craft system, i.e., for any initial attitude and for any initial angular
ros. It is well-known that the inverse of T ij ðCÞ is rate, the LQR controller will bring the nonlinear spacecraft to the
T 1
ij ðCÞ ¼ T ij ðCÞ. For i ¼ 1, 2, and 3, let origin ðω; qÞ ¼ 0. Second, when the nonlinear spacecraft is close to
the origin, the linearized model is a very good approximation of the
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi nonlinear spacecraft system; because LQR controller is an optimal
2
di
di
2 Jkiii pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi design for the linearized system, therefore it is a suboptimal control
J ii J ii
si s2i 2ti for the original nonlinear spacecraft system. To the best of the au-
Ci ¼ ¼ ð26Þ thor’s knowledge, there is no any other spacecraft controller
2 Jkiii 2t i
design method that satisfies simultaneously these two properties.
By using the transformation of T 36 ðC 3 ÞT 25 ðC 2 ÞT 14
ðC 1 ÞðA B½D; KÞT 14 ðC 1 ÞT 25 ðC 2 ÞT 36 ðC 3 Þ, the closed loop Design Examples
eigenvalues of linear system (7) using the LQR design are given
by, for i ¼ 1, 2, and 3 In this section, an example studied by Zhou and Colgren (2005) is
used to illustrate the design method. The spacecraft inertial matrix
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffi ffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffi is given by
J ii ri þ J 2 r Jq21ir J1ii qr2ii
1 q2i q1i
λ2i1 ; λ2i ¼
ii i ii i
ð27Þ 2 3
2 1;200 100 200
J ¼ 4 100 2;200 300 5 ð28Þ
Eq. (27) provides a lot of useful information for the LQR design. 200 300 3;100
First, as ri → 0, the corresponding pair of eigenvalues go to minus
infinity of the complex plane; as ri → ∞, the corresponding pair of
eigenvalues go to origin of the complex plane. Second, as long as
pffiffiffiffiffiffiffiffiffi Nonlinear linearization Linearized LQR design Designed
q1i > q2i r i J ii , the corresponding pair of eigenvalues are real and Spacecraft Spacecraft
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi System System
LQR
Controller
unequal because d i ∕J ii > ðd i ∕J ii Þ2 2ki ∕J ii , these two eigen-
pffiffiffiffiffiffiffiffiffi
values are always negative. Third, if q1i ¼ q2i r i J ii , these are
pffiffiffiffiffiffiffiffiffi Fig. 1. LQR controller design is based on the linearized spacecraft
two equal real negative eigenvalues. Fourth, if q1i < q2i r i J ii , this model
is a pair of complex eigenvalues with a negative real part. There-
fore, increasing q1i and decreasing q2i will increase the damping
ratio; otherwise, it will decrease the damping ratio. Finally,
increasing q2i and decreasing ri will increase the natural frequency;
otherwise, it will decrease the natural frequency. This information Designed Nonlinear
can be useful in spacecraft system design. LQR Spacecraft
By using the LQR design, the closed loop poles are implicitly Controller System
2 3
31:06637549427606 0 0
D¼4 0 41:71184140136478 0 0 5;
Downloaded from ascelibrary.org by Northeastern Univ Library on 04/22/13. Copyright ASCE. For personal use only; all rights reserved.
0 0 49:51151569716377
2 3
0:79056941504209 0 0
K¼4 0 0:79056941504209 0 5
0 0 0:79056941504209
Fig. 3. Monte Carlo runs of quaternion response of the nonlinear system with nondiagonal inertia matrix