AIAA Guidance, Navigation, and Control Conference 10 - 13 August 2009, Chicago, Illinois
AIAA 2009-6318
Linear Dynamics and PID Flight Control of a Powered Paraglider
Yoshimasa OchiTPF ^{1} FPT, Hiroyuki KondoTPF ^{2} F,PT and Masahito WatanabeTPF ^{3} National Defense Academy, Yokosuka, Kanagawa 239-8686, Japan
FPT
This paper describes analytical derivation of a linear dynamic model of a powered paraglider (PPG) from a nonlinear dynamic model, which was derived by the authors. The linear model indicates that the longitudinal dynamics are decoupled from the lateral- directional one. The linear model obtained on the canopy coordinates is transformed into the one on the payload coordinates by linear state transformation, since measurement units are mounted on the payload. The transformation matrix is obtained by linearizing the relations of the velocity and the angular velocity between the canopy and the payload, and also by linearizing the relation among the coordinate transformation matrices. The authors are developing a new design method of a proportional-integral-derivative (PID) control system, which is based on plant model reduction with ν-gap metric and the integral-type optimal servomechanism. This method is applied to flight controller design for three single-input- single-output systems of the PPG. A design example and computer simulation results show linear dynamic properties of the PPG and illustrate good control performance and desirable stability margins of the PID controllers.
I.
Introduction
A powered paraglider (PPG), shown in Fig. 1, can be a useful unmanned aerial vehicle (UAV) for land observation, surveillance, space vehicle retrieval, etc. Although it is subject to wind, its light and foldable wing
or canopy makes it portable equipment for both civil and military uses. Actually PPG-type UAVs have been developed and sold on a commercial basis. Such a UAV needs an autonomous/automatic flight control system, and in order to design a control system we need a dynamic model of a PPG, unless we take an empirical design approach, which is often employed for a black-box plant. However, a PPG is not a totally unknown plant. In fact, many papers have been reported on modeling of PPG or paraglider (PG) nonlinear/linear dynamics.P ^{1}^{-}^{1}^{0} P The authors also proposed a nonlinear dynamic model with eight degrees of freedom, where all the internal forces between the canopy and the suspended payload are analytically eliminated to obtain a model in the form of state equation.P ^{1}^{1} P Its validity was demonstrated through numerical simulation and comparison of the results with flight experiment data of a manned paraglider. In this paper, first we derive a linear model from the nonlinear model by analytical first-order approximation of Taylor series expansion as well as numerical partial differential. Since the original nonlinear model is described with the state variables of the canopy, the linear model also explicitly expresses motion of the canopy. However, measurement sensors such as accelerometers and gyros are mounted on the payload; hence, it would be more convenient in dynamics analysis and/or control system design, if the model is expressed by the state variables of the payload. This motivated us to derive a linear transformation matrix between the canopy states and the payload states. With the state transformation, we can obtain a payload-state linear model from a canopy one, and vice versa. Then, we consider designing a flight control system for the PPG using the linear model. Although a number of studies on PPG or PG flight control have been reported, many of them employ complicated control methods such as model predictive controlP ^{1}^{2} P or inversely consider just open-loop control.P ^{1}^{3}^{,} ^{1}^{4} P In our study, we employed PID (proportional- integral-derivative) control, which is most commonly used in industry. The authors are proposing a new design method of a PID controllerP ^{1}^{5} P based on integral-type optimal servomechanism (IOS), which is a derivative of the linear quadratic regulator or LQR. In the design method, first we reduce a given linear plant model to a second-order
^{1} PT Professor, Department of Aerospace Engineering, 1-10-20 Hashirimizu, Senior Member AIAA.
^{2} PT Graduate student, Department of Aerospace Engineering, 1-10-20 Hashirimizu.
TP
TP
TP
^{3} PT Graduate student, Department of Aerospace Engineering, 1-10-20 Hashirimizu.
1
American Institute of Aeronautics and Astronautics
Copyright © 2009 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
system so that ν-gapP ^{1}^{6} P between the original plant and the reduced plant becomes as small as possible. The plant parameters are determined by parameter-space search, since the number of parameters is just two or three. Of course, the three-term controller cannot always stabilize any plant; however, when stabilizable, it usually provides a set of control gains with good control performance and substantial stability margins, which would be attributed to the properties of the LQR. Moreover, the control gains can be adjusted through selection of weighting matrices for trade-off between control performance and robust stability. In application to a PPG, three PID controllers are designed to control the altitude by thrust, the forward speed by collective brake deflection of the canopy, and the heading angle by differential brake deflection. Time responses to appropriate reference outputs are computed as well as stability margins for each control loop, where other loops are closed. This paper is organized as follows. In the next section, we describe an outline of the linear model and transformation between the canopy-state model and the payload-state model. Then we present the design method of
a PID controller based on the IOS and the ν-gap metric. In Sections IV and V, numerical analysis and simulation results are presented, followed by summary and conclusions in Section VI.
Figure 1. Configuration of PPG
Figure 2. Coordinate systems
II. Linear Dynamic Model and State Transformation
A. Coordinate Systems and Degrees of Freedom of Motion
Figure 1 shows coordinate systems of the PPG. The inertial coordinate system ΣB _{I} B is defined as (XB _{I} B, YB _{I} B, ZB _{I} B). The
XB
The location of the
origin and the positive direction of the XB _{I} B-axis are appropriately chosen. Definition of the canopy coordinate system
ΣB _{c} B = (XB _{c} B, YB _{c} B, ZB _{c} B) and the payload coordinate system ΣB _{p} B = (XB _{p} B, YB _{p} B, ZB _{p} B) is also shown in Fig. 2. The origin OB _{c} B of the canopy-fixed coordinate system is chosen at the center of mass (CM) of the canopy and the ZB _{c} B-axis is chosen in the direction from OB _{c} B to OB _{m} B. The XB _{c} B-axis is perpendicular to the ZB _{c} B-axis in the symmetry plane of the canopy and the positive direction of the XB _{c} B-axis is taken forward. The YB _{c} B-axis is defined so that the coordinates form a right-hand coordinate system. The payload is assumed to be symmetric, and the origin is taken at the CM. The XB _{p} B-axis is taken forward along the thrust direction, the ZB _{p} B-axis is downward and perpendicular to the XB _{p} B-axis in the symmetry plane, and then the YB _{p} B-axis is defined to form a right-hand system. It is assumed that the XB _{c} BZB _{c} B-plane corresponds to the
_{I} BYB _{I} B-plane
is horizontal and the positive direction of the ZB _{I} B-
axis is taken vertically downward.
_{p} BZB _{p} B-plane
XB
in a trimmed
straight flight.
The payload is connected with the suspension lines of the canopy at two points, OB _{m}_{R} B and OB _{m}_{L} B. OB _{m} B is the middle point between OB _{m}_{R} B and OB _{m}_{L} B. It is assumed that the motion of the canopy has six degrees of freedom (DOF) and that the suspension lines are deformed only about the ZB _{c} B-axis. The payload then has two DOF, i.e., the relative yawing about the ZB _{c} B-axis and the relative pitching about the line OB _{m}_{R} BOB _{m}_{L} B. Thus, the PPG is modeled as a system with eight DOF.
2
American Institute of Aeronautics and Astronautics
Nonlinear and Linear State Equations Let the velocity of the canopy at OB _{c} B represented in ΣB _{c} B be VB _{c} B = [uB _{c} B vB _{c} B wB _{c} B]P ^{T} , and the angular velocity of the canopy
also in ΣB _{c} B be ωB _{c} B = [pB _{c} B qB _{c} B rB _{c} B]P ^{T} . Let the relative yaw angle and rate between the canopy and the payload be ψB _{p}_{c} B and rB _{p}_{c} B=dψB _{p}_{c} B/dt, respectively, and similarly let the relative pitch angle and rate be θB _{p}_{c} B and qB _{p}_{c} B=dθB _{p}_{c} B/dt, respectively. In addition, let the pitch and roll angles of the canopy be θB _{c} B and φB _{c} B, respectively. We then define the state vector xB _{c} B = [uB _{c} B
B.
P
P
^{P}
vB c B wB c B pB c B qB c B rB c B qB pc B rB pc B θB pc B ψB pc B φB c B θB c B]P T
= [VB _{c} PB
T
P
P
ωB c PB ωB pc PB θB pc B ψB pc B φB c B θB c B]P T
T
P
T
P
, where ωB _{p}_{c} B = [qB _{p}_{c} B rB _{p}_{c} B]P ^{T} . The actual control inputs are
P
thrust FB _{p}_{t}_{h}_{x} B, the right and left brake angles, δB _{R} B and δB _{L} B, which are deflection angles of the right and left rear-portions of the canopy. The right (left) brake angle is defined as dB _{R} B/cB _{c} B (dB _{L} B/cB _{c} B,), where dB _{R} B (dB _{L} B) is the pull-length of the right (left) control lines and cB _{c} B is the mean aerodynamic-chord length of the canopy. For convenience and for derivation of
linear state equation, we define collective deflection angle δB _{e} B = δB _{R} B + δB _{L} B, and differential one δB _{r} B = δB _{R} B −δB _{L} B. We then
define the control input vector as u =
is described by the nonlinear state equation,P ^{1}^{1}
_{[}_{F}_{B} _{p}_{t}_{h}_{x} B δB _{e} B δB _{r} B]P ^{T}
P
P
. With the state and control vectors, the 8-DOF motion of the PPG
(1)
where f(xB _{c} B) ∈ℜP ^{1}^{2}^{×}^{1} P and g(xB _{c} B) ∈ℜP ^{1}^{2}^{×}^{3} P are nonlinear functions of xB _{c} B. Let the equilibrium state and input vectors be xP ^{*} P and uP ^{*} , i.e., f(xB _{c} PB ) + g(xB _{c} PB )uP ^{*} P = 0.
Under the assumption of symmetry of the PPG with respect to the XZ-plane, the non-trivial equations in Eq. (2) are four equations corresponding to forward and downward force trim and pitching moment trim of the canopy and the
(2)
x &
c
=
f (x )
*
P
c
+
*
P
g(x )u
c
P
payload. Defining small perturbations of the state and control variables around the equilibrium point (xB _{c} PB , uP ^{*} ) as ΔxB _{c} B
^{*} P and Δu = u −uP ^{*} , and applying the first-order approximation of Taylor series expansion, we obtain the
linearized state equation,
(3)
where AB _{c} B∈ℜP ^{1}^{2}^{×}^{1}^{2} P is a constant matrix given by partial differentiation of f(xB _{c} B) + g(xB _{c} B)u with respect to xB _{c} B at the trim point (xB _{c} PB , uP ^{*} ), and BB _{c} B∈ℜP ^{1}^{2}^{×}^{3} P is given by g(xB _{c} PB ). We have analytically derived the linear state equation, although we have no space to show details here. Since the velocity of the PPG is small and its angle of attack is relatively large in trim flight, we approximate small perturbation of the angle of attack by
(4)
*
P
P
= xB c B −xB c PB
P
Δx & = A Δx + B Δu
c
c
c
c
*
P
,
Δ
α
c
=
(cos
*
α
c
)
2
u
*
c
(
Δ
w
c
−
*
P
P
tan
*
α
c
⋅Δ
u
c
)
,
where αB _{c} B = tanP ^{−}^{1} (wB _{c} B/uB _{c} B). This approximation provides a better correspondence of the system matrix AB _{c} B between the analytical linearization and the numerical one. Appropriately rearranging the elements of the state vector makes the matrices AB _{c} B and BB _{c} B block-diagonal, so that Eq. (3) is separated into two state equations, i.e., for the longitudinal motion we have
(5)
and for the lateral-directional motion
(6)
where ΔxB _{c}_{l}_{o}_{n}_{g} B = [ΔuB _{c} B ΔwB _{c} B ΔqB _{c} B ΔθB _{c} B ΔqB _{p}_{c} B ΔθB _{p}_{c} B]P ^{T} , ΔuB _{l}_{o}_{n}_{g} B = [ΔFB _{b}_{t}_{h} B ΔδB _{e} B]P ^{T} , ΔxB _{c}_{l}_{a}_{t} B = [ΔvB _{c} B ΔpB _{c} B ΔrB _{c} B ΔφB _{c} B ΔrB _{p}_{c} B ΔψB _{p}_{c} B ΔψB _{c} B]P ^{T} and ΔuB _{l}_{a}_{t} B = ΔδB _{r} B.
P
Δx&
= A
Δx
clong
clong
clong
+ B
Δu
clong
long
Δx& = A Δx + B Δu
clat
clat
clat
clat
P
lat
,
P
P
C. State Transformation
Since the canopy generates most of the aerodynamic forces and moments, using the canopy states makes the derivation of the dynamic equation easier and straightforward. However, sensors such as accelerometers, gyros, camera, etc. are mounted on the payload; hence, expressing the state equation in terms of the payload states, which are defined below, would be more convenient for analysis and synthesis of a flight control system. Although we
could derive a nonlinear payload-state equation, we choose to derive a linear payload-state equation, considering application of a linear control method. The following relations hold between the canopy states and the payload states. First, the velocity of the payload expressed in ΣB _{c} B is given by
(7)
In Eq. (7), when we define lB _{c} B (lB _{p} B) to be the distance between OB _{m} B and OB _{c} B (OB _{p} B), KB _{p}_{c}_{1} B and KB _{p}_{c}_{2} B are defined, respectively,
as
_{V}_{B} _{p}_{c} B = VB _{c} B + KB _{p}_{c}_{1} BωB _{c} B + KB _{p}_{c}_{2} BωB _{p}_{c} B.
3
American Institute of Aeronautics and Astronautics
K pc
1
⎡
⎢
= −
⎢
⎢
l
⎣
p
(
l
p
0 l
p
cos
θ pc
+
l
c
)
cos
θ pc
0
+
l
c
sin
θ ψ
pc
sin
pc
− l
p
sin
θ
pc
ψ
cos
pc
−
l
p
l
p
sin
pc
sin
θ
pc
θ
sin
ψ
pc
ψ
cos
pc
0
and
K
pc
2
= l
p
⎡ cos
⎢
⎢
⎢
⎣
cos
θ pc
θ pc
cos
sin
ψ
ψ
pc
pc
− sin
θ pc
− sin
θ
pc
sin
ψ
c
sin
θ
pc
ψ
cos
pc
0
⎤
⎥
⎥
⎥
⎦
.
Defining the velocity of the payload expressed in ΣB _{p} B as VB _{p} B = [uB _{p} B vB _{p} B wB _{p} B]P ^{T} , we can also write VB _{p}_{c} B as
P
VB _{p}_{c} B = TB _{p}_{c} BVB _{p} B
⎤
⎥
⎥
⎥
⎦
(8)
(9)
(10)
where TB _{c}_{p} B is a coordinate transformation matrix from ΣB _{p} B to ΣB _{c} B and defined as
⎡ cos
⎢
⎢
⎢
⎣
cos
pc
θ
pc
cos
sin
− sin
θ
pc
θ
ψ
ψ
−
sin
ψ
pc
ψ
cos
pc
pc
pc
T
cp
=
0
Eliminating VB _{p}_{c} B from Eqs. (7) and (10), we have the first relation:
VB c B = TB cp BVB p B − KB pc1 BωB c B − KB pc2 BωB pc B.
sin
sin
θ
pc
θ
pc
cos
sin
ψ
ψ
pc
pc
cos
θ
pc
⎤
⎥
⎥
⎥
⎦
.
Second, the relation between the angular velocities is given by
ωB c B = TB cp BωB p B − KB pc3 BωB pc B,
where ωB _{p} B = [pB _{p} B qB _{p} B rB _{p} B]P ^{T} is the angular velocity vector of the payload in ΣB _{p} B, and KB _{p}_{c}_{3} B is defined by
⎡− sin
⎢
0
⎣
⎢
⎢
P
ψ
pc
0 ⎤
0
⎦
1
⎥
⎥
⎥
K
pc 3
=
ψ
cos
pc
.
(11)
(12)
(13)
(14)
Third, from the definition of Euler angles and relative attitude angles between the canopy and the payload, we have the identical equation
(15)
where ψB _{c} B is the yaw angle of the canopy, and φB _{p} B, θB _{p} B, and ψB _{p} B are the roll, pitch, and yaw angles of the payload, respectively. TB _{p}_{I} B (TB _{c}_{I} B) is the coordinate transformation matrix from the inertial coordinates to ΣB _{p} B (ΣB _{c} B). Thus, we have obtained three relations, which are given by Eqs. (12), (13), and (15). Applying the first-order approximation of Taylor series expansion to the equations yields a linear relation between the canopy state vector and the payload one as follows. First, from Eq. (12) we have
(16)
, where the superscript ‘*’ means that the functions and the variables are evaluated at the trim states, and KB _{p}_{c}_{4} ^{*} PB P is defined as
TB cI B(φB c B, θB c B, ψB c B) =TB cp B(θB pc B, ψB pc B)TB pI B(φB p B, θB p B, ψB p B),
ΔVB c B + KB pc1 PB ωB c B + KB pc2 PB ωB pc B = TB pc PB ΔVB p B +KB pc4 PB
P
P
P
P
*
*
*
*
[ΔθB _{p}_{c} B
ΔψB _{p}_{c} B]P ^{T}
P
where _{α}_{B} _{p} ^{*} PB P = tanP ^{−}^{1}
*
P
(wB p PB /uB p PB
P
*
K
*
pc
4
*
p
= V
⎡ sin(
⎢
⎢
⎢
⎣
α θ
p
−
0
*
pc
*
*
)
pc
*
−
cos(
α θ
p
−
)
). Next, Eq. (13) is linearized as
P
P
ΔωB c B = TB pc PB ΔωB p B − KB pc3 PB
*
*
ΔωB _{p}_{c} B.
P
0
cos(
α θ
p
−
pc
*
*
0
)
⎤
⎥
⎥
⎥
⎦
,
(17)
(18)
Finally, from Eq. (15) the following linear equations are obtained:
_{Δ}_{θ}_{B} _{p} B
= ΔθB _{c} B+ΔθB _{p}_{c} B
⎡ Δ
⎢
⎣
φ
p
Δ
ψ
p
⎤
⎥
⎦
=
1
cos
*
θ
p
⎡ cos
sin
⎣
⎢
⎢
*
θ
c
*
θ
pc
sin
cos
*
θ
c
*
θ
pc
0
cos
*
θ
p
⎤
⎥
⎦
⎥
⎡ Δ
⎢
⎢
⎢ Δ
⎣
Δ
φ
c
ψ
pc
ψ
c
⎤
⎥
⎥
⎥
⎦
.
(19)
(20)
Adding ψB _{c} B to the state vector xB _{c} B, defining the payload state vector as xB _{p} B = [VB _{p} PB combining Eqs. (16), (18), (19), and (20), we obtain the relation ΔxB _{p} B = TΔxB _{c} B, where ΔxB _{p} B =xB _{p} B −xB _{p} ^{*} PB P and T ∈ℜP ^{1}^{3}^{×}^{1}^{3} P is a constant transformation matrix.
T
P
P
ωB p PB ωB pc PB θB pc B ψB pc B φB p B θB p B ψB p B]P T
T
P
T
P
, and
(21)
With Eq. (21), Eq. (3) is transformed into the equation expressed by the payload state as
4
American Institute of Aeronautics and Astronautics
(22)
where AB _{p} B = TAB _{c} BTP ^{−}^{1} P and BB _{p} B = TBB _{c} B. As Eqs. (5) and (6) resulted from Eq. (3), Eq. (22) is rewritten into the longitudinal state equation
(23)
and the lateral-directional state equation
(24)
where ΔxB _{p}_{l}_{o}_{n}_{g} B = [ΔuB _{p} B ΔwB _{p} B ΔqB _{p} B ΔθB _{p} B ΔqB _{p}_{c} B ΔθB _{p}_{c} B]P ^{T} and ΔxB _{p}_{l}_{a}_{t} B = [ΔvB _{p} B ΔpB _{p} B ΔrB _{p} B ΔφB _{p} B ΔrB _{p}_{c} B ΔψB _{p}_{c} B ΔψB _{p} B]P ^{T} . This means that the state transformation matrix itself becomes block-diagonal by rearranging the state variables.
Δx & = A Δx + B Δu
p
p
p
p
Δx&
plong
= A
plong
Δx
plong
,
+ B
Δu
plong
long
Δx& = A Δx + B Δu
plat
plat
plat
plat
P
lat
,
P
III. PID Control based on Integral-Type Optimal Servomechanism
A. Plant Representation Consider a second-order delay system described by a transfer function GB _{C} B(s), i.e. y(s) = GB _{C} B(s)u(s),
where y(s) is an output and u(s) is an input, and GB _{C} B(s) is defined as
s
G
C
(
) =
n
0
s
2
+
d s
1
+
d
0
.
(25)
(26)
The plant is obviously controllable and observable. To give a state-space representation of the system, state vector is
defined as x= [xB _{1} B xB _{2} B]P ^{T}
(27)
(28)
where
(29)
B =
PB
B=
[
y
y&
] ^{T}
. The system is then expressed as
x& = Ax + Bu
y = Cx,
A =
⎡ 0
−
⎣
⎢
d
0
−
1
d
1
⎤
⎥
⎦
,
⎡ 0 ⎤ ⎥ ⎦
⎣
⎢
n
0
, C = [1
0].
B. Integral-type Optimal Servo Controller
The first step is to design an integral-type optimal servo controller for the plant given by Eqs. (27) and (28). In this study, we employ a design method of an IOS controller by Smith and Davison.P ^{1}^{7} P With the definition of control
error e = r − y, where r is a constant reference output, we defined an augmented system as
(30)
d
dt
⎡ x & ⎤ ⎡ A
⎢
⎣
0
0
⎢
⎤
⎥ ⎦ ⎣ e ⎦
⎥
⎤
⎡ x &
⎡ B ⎤ ⎢ ⎣
0
⎥
⎦
⎥
⎦
=
⎢
⎣
+
u &
.
e
− C
For a quadratic cost function defined as
(31)
the optimal control law that minimizes J can be derived as
(32)
by using the LQR theory. Integrating Eq. (32) yields
(33)
Defining KB _{x} B = [KB _{P} B KB _{D} B]P ^{T} and KB _{I} B = KB _{e} B, we can rewrite the control law as
(34)
This is exactly the I-PD (proportional and derivative preceded integral) control law, which means that the PID gains are determined at a time as optimal state feedback gains. Once the I-PD control law is obtained, it can be arbitrarily converted to the PID, PI-D or generic 2-DOF form.P ^{1}^{8} P Note that any of the controller forms has the same closed-loop property such as stability. In practical use, the pure derivative of y is replaced with an approximate one s/(TB _{D} Bs+1), where TB _{D} B is an appropriate constant.
J
=
∫ ∞
(
T
&= K
u
u =
K
x
x
x
u = K
0
x&
Q
x
x&
+
Qe
x&+
K
e
e
+ K
e
∫
edτ
.
&
y + K
I
∫
y + K
P D
2
+
2
Ru & )dt
,
P
edτ
.
C. Plant Reduction based on ν-Gap Metric
As stated above, we can design an optimal PID controller, if the plant dynamics are given by Eq. (25). However, the order of a plant GB _{P} B(s) is generally higher than the second. For a higher-order plant, we need to reduce its transfer function to Eq. (26) and to achieve this we employ the model reduction technique based on the ν-gap metric.
5
American Institute of Aeronautics and Astronautics
Specifically, we search plant parameters of GB _{C} B(s) that minimize the ν-gap between GB _{P} B(s) and GB _{C} B(s). It is known that if the ν-gap is small, closed-loop properties for the two plants are close to each other.P ^{1}^{9}^{,} ^{2}^{0} P Reference 21 also uses this method in conjunction with LMI to find a reduced-order plant for controller design. In our study, parameter-space search is adopted, since the number of parameters is small. Before giving the definition of ν-gap, let us define the following function
Ψ
(
(
G s
P
),
G
C
(
s
))
=
(35)
and the conditions
(36)
(37)
where η(GB _{P} B(s)) denotes the number of poles of GB _{P} B(s) in the open right-half complex plane and ηB _{0} B(GB _{C} B(s)) the number of imaginary axis poles of GB _{C} B(s). ‘wno’ denotes the winding number valuated on the standard Nyquist contour
indented around any imaginary axis poles of GB _{P} B(s) and GB _{C} B(s). With the definitions, the ν-gap is defined as
(38)
1+GB _{C} B(−jω)GB _{P} B(jω) ≠ 0, ∀ω wno (1+GB _{C} B(−jω)GB _{P} B(jω))+η(GB _{P} B(s)) −η(GB _{C} B(s)) −ηB _{0} B(GB _{C} B(s)) = 0,
δ
ν
(
G
P
,
G
C
)
=
⎨
⎩
⎧
Ψ ( G
P
,
G
C
)
∞ if Eqs. (36) and (37) are satisfied.
1 otherwise.
From our design experience, we have found that dB _{0} B of GB _{C} B(s) can be set to zero for GB _{P} B(s) having integral element 1/s. This also reduces the number of parameters to be searched. Note that depending on GB _{P} B(s), there exists no GB _{C} B(s)
for which the ν-gap is small. In that case, we cannot design a PID controller that possesses good closed-loop properties or even stability. However, this is natural, since the three-term controller cannot stabilize all plants.
Thus, the design procedure is summarized as follows. Approximately cancel closely located poles and zeros of GB _{P} B(s), if necessary.
Design an integral-type optimal servo using GB _{C} B(s). Features of the design method are as follows.
1)
2) Determine the parameters of GB _{C} B(s) based on the ν-gap metric via parameter-space search.
3)
1) PID gains are obtained at a time.
2)
3) Trade-off among P, D, and I actions is also possible through weight selection of the diagonal elements of QB _{x} B and Q, respectively. 4) The controller has desirable properties as an LQR such as good time response and desirable stability margins, although some degradation may occur due to plant model reduction.
Trade-off between control performance and control effort can be performed through weight selection.
IV. Linear Dynamics Analysis
We use data of a manned paragliderP ^{2}^{2} P for numerical analysis and simulation. The span and the chord length of the canopy are 9.95 m and 3.05 m, respectively, and the weights of the canopy and the payload are 6.4 kg and 93.0 kg, respectively. The trim condition without control inputs is steady-state gliding, where the airspeed is 9.54 m/s, the angle of attack is 15.1 deg, the pitch angle of the canopy is 3.45 deg, and the relative pitch angle is −4.83 deg. The system and control matrices of the linear dynamic model were computed by analytical linearization as well as by numerical partial differentiation. The two methods gave almost the same results. The matrices of the state equations and state transformation matrices are given in Appendix. To illustrate the difference between the canopy-state system and the payload-state system, we computed time responses for the initial relative pitch angle ΔθB _{p}_{c} B(0) = 0.1 rad, where other initial states were zero. Figures 3 shows time responses of the pitch angle of the canopy, _{Δ}_{θ}_{B} _{c} B. Although the same initial states are given, the time responses of θB _{c} B are very different. The reason for this is that the initial pitch angle ΔθB _{c} B(0) for the payload-state system is −0.1 rad, whereas it is exactly zero for the canopy-state system, where ΔθB _{b} B(0) = 0.1 rad. Computing time responses of the lateral-directional motion for the initial relative yaw angle ΔψB _{p}_{c} B(0) = 0.1 rad, we observe the similar difference between the canopy-state and payload-state systems. Thus, we should consider physical meaning of the states, when we deal with the systems. Table 1 summarizes the eigenvalues of the linear models of the PPG and motion modes corresponding to the complex conjugate pairs. Figure 4 shows time responses of the pitch rate of the canopy computed using the payload- state model for _{Δ}_{θ}_{B} _{p}_{c} B(0) = 0.1 rad. We can see from the figure that the relative pitch rate has a faster, good-damping mode, and that the canopy pitch rate has slower, more oscillatory mode. Hence, the former mode corresponds to the
6
American Institute of Aeronautics and Astronautics
relative pitching motion and the latter to the pitching motion of the canopy. Figure 5 shows time responses of the yaw rate of the canopy computed using the payload-state model for ΔψB _{p}_{c} B(0) = 0.1 rad. We can also see from the figure that the relative yaw rate has a faster, oscillatory mode, and that the canopy yaw rate has a slower, less oscillatory mode. The former mode corresponds to the relative yawing motion and the latter to the Dutch-roll motion of the canopy.
time, s
Figure 3. Time responses of the canopy pitch angle for ΔθB _{p}_{c} B(0) = 0.1
Table 1 Eigenvalues of the linear systems (longitudinal and lateral-directional)
motion mode |
eigenvalues |
natural frequency, rad/s |
damping ratio |
first-order modes |
_{−}_{1}_{.}_{9}_{6}_{,} _{−}_{0}_{.}_{5}_{5}_{7} |
_{⎯} |
_{⎯} |
relative pitching |
_{−}_{3}_{.}_{3}_{6}_{±}_{1}_{.}_{9}_{7}_{i} |
3.89 |
0.863 |
canopy pitching |
_{−}_{0}_{.}_{2}_{8}_{5} _{±}_{0}_{.}_{9}_{4}_{1}_{i} |
0.983 |
0.290 |
first-order modes |
_{−}_{1}_{2}_{1}_{,} _{−}_{1}_{.}_{3}_{5}_{,} _{0} |
_{⎯} |
_{⎯} |
relative yawing |
_{−}_{0}_{.}_{1}_{5}_{0}_{±}_{5}_{.}_{0}_{3}_{i} |
5.03 |
0.0297 |
Dutch-roll |
_{−}_{0}_{.}_{2}_{2}_{4}_{±}_{0}_{.}_{7}_{2}_{1}_{i} |
0.755 |
0.297 |
time, s
Figure 4. Time responses of the canopy pitch rate for ΔθB _{p}_{c} B= 0.1 rad
time, s
Figure 5. Time responses of the canopy yaw rate for ΔψB _{p}_{c} B= 0.1 rad
7
American Institute of Aeronautics and Astronautics
V. Flight Control System Design and Simulation
A. Flight Control System Design The design method of a PID controller has been applied to the linear model of the PPG. Since the PPG has three control inputs, we design three PID controllers for respective SISO systems as follows. The transfer function from thrust, ΔFB _{p}_{t}_{h}_{x} B(s), to the altitude variation, Δh(s), is given by
(39)
(
s
)
=
7.1214 |
× |
10 |
− 5 |
( |
s |
+ 180.28)( |
s |
+ |
2.0410)( |
s |
+ 0.4408)( |
s |
2 |
+ 6.2294 |
s |
+ |
14.678) |
||||
( s s |
+ 1.9578)( s |
+ |
0.5567)( |
s |
2 |
+ |
6.7208 |
s |
+ |
15.170)( |
s 2 |
+ |
0.56976 s |
+ |
0.96647) |
G
P
1
We have then obtained a second-order plant with a ν-gap of 0.0177 as
G
C
1
( ) ^{=}
s
0.01061
.
s
2
+
s
(40)
Choosing the weighting matrices as QB _{x} B = 0B _{2}_{×}_{2} B, Q =100, and R =1 and the time constant as TB _{D} B =0.025 s, the resultant
PID control gains are KB _{P} B = −52.33, KB _{I} B = 10.00, and KB _{D} B = −42.67. The transfer function from collective brake deflection angle, ΔδB _{e} B(s), to the forward speed, ΔuB _{p} B(s), is given by
(41)
G
P
2
(
s
)
=
− |
0.19986( |
s |
+ |
69.954)( |
s |
+ 12.736)( |
s |
+ |
1.5482)( |
s 2 |
+ |
2.5840 s + |
4.0135) |
|||
( |
s |
+ |
1.9578)( |
s |
+ |
0.5567)( |
s |
2 + |
6.7208 |
s |
+ |
15.170)( |
s 2 |
+ |
0.56976 s + |
0.96647) |
We have then obtained a second-order plant with a ν-gap of 0.298 as
G
C
2
(
s
^{)}
− 69.24
^{=} s − s −
2
6
20
.
(42)
Choosing the weighting matrices as QB _{x} B = 0B _{2}_{×}_{2} B, Q =1, and R =1 and the time constant as TB _{D} B =0.025 s, we obtained PID control gains of KB _{P} B = 0.9671, KB _{I} B = −1.000, and KB _{D} B = 0.2749. The transfer function from differential brake deflection angle, ΔδB _{r} B(s), to the heading angle, ΔψB _{p} B(s) +ΔβB _{p} B(s), is given by
(43)
G
P 3
(
s
)
35.434( |
s |
+ 0.30072)( |
s 2 |
+ |
2.1269 |
s |
+ |
37.670)( |
s 2 |
+ |
1.8946 s |
+ |
8.5392) |
||||
+ 121.0)( |
s |
+ 1.347)( |
s |
2 |
+ 0.2990 |
s |
+ 25.35)( |
s 2 |
+ |
0.4488 |
s |
+ |
0.5698) |
^{=} s ( s
Before designing a PID controller, we first design a roll damper, whose control law is ΔδB _{r} B = −0.5ΔpB _{p} B. Figure 6 shows time responses of the canopy roll rate, ΔpB _{p} B, to the step input ΔδB _{r} B = 0.1 rad with and without the roll damper. It is obvious that the damping of the roll rate is improved.
time, s
Figure 6. Time responses of the roll rate to step input, ΔδB _{r} B = 0.1 rad
The transfer function of the closed-loop system then becomes
G
P dmp
3
(
s
)
35.434( |
s |
+ |
0.30072)( |
s 2 |
+ |
2.1269 |
s |
+ |
37.670)( |
s 2 |
+ |
1.8946 |
s |
+ |
8.5392) |
|||
+ 119.09)( |
s |
+ 2.6908)( |
s 2 |
+ |
0.79272 |
s |
+ 28.537)( |
s |
2 |
+ 0.47395 |
s |
+ |
0.26189) |
^{=} s ( s
(44)
For GB _{P}_{3}_{d}_{m}_{p} B(s), we have then obtained a second-order plant with a ν-gap of 0.310 as
1.288
2
+
0.9 s
.
G
C
3
(
s
) =
(45)
s We designed a PID controller for the reduced-order model of Eq. (45), choosing the weighting matrices as QB _{x} B = 10IB _{2} B, Q =10, and R =1 and the time constant as TB _{D} B =0.025 s. The obtained gains are KB _{P} B = −6.193, KB _{I} B = 3.162, and KB _{D} B =
−3.785.
8
American Institute of Aeronautics and Astronautics
Stability margins are summarized in Table 2. The margins are computed by opening each closed-loop, while other loops are closed. The controllers have large gain and phase margins. The margin of gain increase is ∞ for the speed controller.
Table 2
Stability margins
controller |
gain margin |
phase margin |
crossover frequency |
altitude |
28.63 dB |
89.09 deg |
0.5150 rad/s |
speed |
_{−}_{1}_{8}_{.}_{3}_{5} _{d}_{B} |
42.44 deg |
9.172 rad/s |
heading |
_{∞} |
85.05 deg |
2.597 rad/s |
In this design example, we use the PID control law
where e = yB _{r}_{e}_{f} B −y
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.