Finite Set Control Transcription for Optimal Control Applications
-
Upload
belinda-marchand -
Category
Technology
-
view
116 -
download
1
description
Transcript of Finite Set Control Transcription for Optimal Control Applications
university-logo
BackgroundApplicationsConclusions
Finite Set Control Transcription for Optimal Control
Applications
Stuart A. Stanton1 Belinda G. Marchand2
Department of Aerospace Engineering and Engineering Mechanics
The University of Texas at Austin
19th AAS/AIAA Space Flight Mechanics Meeting, February 8-12, 2009Savannah, Georgia
1 Capt, USAF; Ph.D. Candidate2 Assistant Professor
The views expressed here are those of the author and do not reflect the official policy or position of
the United States Air Force, Department of Defense, or the U.S. Government.
Stanton, Marchand Finite Set Control Transcription 1
university-logo
BackgroundApplicationsConclusions
Outline
BackgroundSystem DescriptionFSCT Method Overview
ApplicationsLinear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Conclusions
Stanton, Marchand Finite Set Control Transcription 2
university-logo
BackgroundApplicationsConclusions
System DescriptionFSCT Method Overview
System Description
I Hybrid System Dynamics
y = f(t,y,u)
I Continuous States
y =ˆ
y1 · · · yny
˜T
yi ∈ R
I Discrete Controls
u = [u1 · · · unu ]T
ui ∈ Ui = {ui,1, . . . , ui,mi}
I ExamplesI Switched SystemsI Task Scheduling and Resource Allocation ModelsI On-Off Control SystemsI Control Systems with Saturation Limits
Stanton, Marchand Finite Set Control Transcription 3
university-logo
BackgroundApplicationsConclusions
System DescriptionFSCT Method Overview
Solving an Optimal Control Problem Numerically
Minimize J = φ(t0,y0, tf ,yf ) +R tft0
L(t,y,u) dt
subject toy = f(t,y,u),0 = ψ0(t0,y0),0 = ψf (tf ,yf ),0 = β(t,y,u)
↘ ?
Minimize J = F (x)
subject to
c(x) =h
cTy (x) cTψ0(x) cTψf
(x) cTβ (x)iT
= 0
↘X
NLP Solver
Stanton, Marchand Finite Set Control Transcription 4
university-logo
BackgroundApplicationsConclusions
System DescriptionFSCT Method Overview
FSCT Method Overview
I Parameter vector consists only of states and times
x = [· · · yi,j,k · · · · · · ∆ti,k · · · t0 tf ]T
I Control history is completely defined byI Pre-specified control sequenceI Control value time durations, ∆ti,k , between switching points
I Key parameterization factors
ny Number of Statesnu Number of Controlsnn Number of Nodesnk Number of Knotsns Number of Segments (ns = nunk + 1)
Stanton, Marchand Finite Set Control Transcription 5
university-logo
BackgroundApplicationsConclusions
System DescriptionFSCT Method Overview
FSCT Method Overview
x = [· · · yi,j,k · · · · · · ∆ti,k · · · t0 tf ]T
u1 ∈ U1 = {1, 2, 3},
u2 ∈ U2 = {−1, 1}.u
∗ =
»
1 2 3 1 2 3−1 1 −1 1 −1 1
–
1
2
3
1
2
-1
1
-1
1
-1
y1
y2
Seg.
u1
u2
tt0 tf
ny = 2
nu = 2
nn = 4
nk = 5
ns = 11
∆t1,1 ∆t1,2 ∆t1,3 ∆t1,4 ∆t1,5 ∆t1,6
∆t2,1 ∆t2,2 ∆t2,3 ∆t2,4 ∆t2,5 ∆t2,6
1 2 3 4 5 6 7 8 9 10 11
Stanton, Marchand Finite Set Control Transcription 6
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Two Stable Linear Systems
y = f (y, u) = Auy,
u ∈ {1, 2} ,
where
A1 =
»
−1 10−100 −1
–
, A2 =
»
−1 100−10 −1
–
−30 −20 −10 0 10 20 30
−30
−20
−10
0
10
20
y1
y 2
y =
[−1 10−100 −1
]y
(a) u = 1
−20 −10 0 10 20 30
−20
−15
−10
−5
0
5
10
15
20
y1
y 2
y =
[−1 100−10 −1
]y
(b) u = 2
Figure: Individually Stable Systems
Stanton, Marchand Finite Set Control Transcription 7
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Two Stable Linear SystemsI Several switching laws
(a) Unstable u =
1, y1y2 < 02, otherwise
(b) Stable u =
1, y1 > y2
2, otherwise
(c) Stable u =
1, yTP 1y < yP 2y2, otherwise
where P uAu + ATuP u = −I
−8000 −6000 −4000 −2000 0 2000−8000
−7000
−6000
−5000
−4000
−3000
−2000
−1000
0
1000
y1
y 2
y = Auy
u =
{1, y1y2 < 02, otherwise
(a)
−35 −30 −25 −20 −15 −10 −5 0 5 10 15
−30
−25
−20
−15
−10
−5
0
5
10
y1
y 2
y = Auy
u =
{1, y1 > y2
2, otherwise
(b)
−40 −30 −20 −10 0 10 20 30 40
−30
−20
−10
0
10
20
30
y1
y 2
y = Auy
u =
{1, yTP1y < y
TP2y2, otherwise
(c)
Figure: Three Switching Laws
Stanton, Marchand Finite Set Control Transcription 8
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Two Stable Linear Systems
I FSCT OptimizationJ = F (x) = tf − t0
yTf yf = 1
u∗
k =3
2+
1
2(−1)k
−20 −15 −10 −5 0 5 10 15 20
−30
−25
−20
−15
−10
−5
0
5
10
y1
y 2
(a)
−20 −15 −10 −5 0 5 10 15 20
−30
−25
−20
−15
−10
−5
0
5
10
y1
y 2
(b)
Figure: FSCT Locally Optimal Switching Trajectories
I Optimization implies the switching law
u =
1, − 1m
≤ y2y1
≤ m
2, otherwise
Stanton, Marchand Finite Set Control Transcription 9
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
2-Dimensional Lunar Lander
I Dynamics
y =
2
6
6
4
r1
r2
v1
v2
3
7
7
5
=
2
6
6
4
v1
v2
u1
−g + u2
3
7
7
5
,
I Controls
u1 ∈ {−50, 0, 50} m/s2,
u2 ∈ {−20, 0, 20} m/s2
I Initial and FinalConditions
r0 = [200 15]T km
v0 = [−1.7 0]T km/s
rf = 0
vf = 0
v0
r0, t0
rf ,vf , tf
Stanton, Marchand Finite Set Control Transcription 10
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
2-Dimensional Lunar Lander
−20 0 20 40 60 80 100 120
0
100
200
Minimum Timer
km
rrrrrrrrrrrrrrrrrrrrrrrrrrrr
−20 0 20 40 60 80 100 120−4
−2
0
2v
km
/s vvvvvvvvvvvvvvvvvvvvvvvvvvvv
−20 0 20 40 60 80 100 120−50
0
50 u1
m/s2
u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1
−20 0 20 40 60 80 100 120−20
0
20 u2
Time (s)
m/s2
u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2
(a)
−20 0 20 40 60 80 100 120 140 160
0
100
200
Minimum Fuelr
km
rrrrrrrrrrrrrrrrrrrrrrrrrrrr
−20 0 20 40 60 80 100 120 140 160−2
−1
0
1 v
km
/s
vvvvvvvvvvvvvvvvvvvvvvvvvvvv
−20 0 20 40 60 80 100 120 140 160−50
0
50 u1
m/s2
u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1u1
−20 0 20 40 60 80 100 120 140 160−20
0
20 u2
Time (s)
m/s2
u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2u2
(b)
Figure: Optimal Solutions for the Minimum-Time (a) and Minimum-Fuel (b)Lunar Lander Problem
Stanton, Marchand Finite Set Control Transcription 11
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Small Spacecraft Attitude Control: Fixed ThrustI Fixed thrust cold gas propulsion for
arbitrary attitude trackingI Reference trajectory defined by rqi0
and rωi(t)
I Minimize deviations between body frameand reference frame with minimumpropellant mass consumption
J = β1pf − β2mpf
pf−p0 =
Z tf
t0
p dt =
Z tf
t0
“
rqvb”T “
rqv
b”
dt.
l3
↗Thuster Pair
l1
r
l2
↙Thuster Pair
y =
2
6
6
6
6
6
6
6
6
6
4
bqi
bωi
mp
rqi
p
3
7
7
7
7
7
7
7
7
7
5
= f (t,y,u)
ui ∈ U = {−1, 0, 1}
I where ui indicates for each principalaxis whether the positive-thrustingpair, the negative-thrusting pair, orneither is in the on position
Stanton, Marchand Finite Set Control Transcription 12
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Small Spacecraft Attitude Control: Fixed Thrust
Actual Trajectory
Desired Trajectory
0 2 4 6 8 10 12 14 16 18 20−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Time (s)
Quaternions: Actual vs. Desired
(a)
0 2 4 6 8 10 12 14 16 18 20−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Time (s)
rad
/s
Angular Velocity: Actual vs. Desired
(b)
Figure: Fixed Thrust Attitude Control
Stanton, Marchand Finite Set Control Transcription 13
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Small Spacecraft Attitude Control: Variable Thrust
I Variable thrust cold gas propulsionI Valve rod modifies nozzle throat area
I Include additional states to modelvariable thrust
I Resulting dynamics are still hybrid
Nozzle Throat
30o
30o
29.5o
Valve Core Rod
Valve Core Motion
I States and Controls
y =
2
6
6
6
6
6
6
6
6
4
bqi
bωi
mp
d
vrqi
p
3
7
7
7
7
7
7
7
7
5
u =
»
w
a
–
wi ∈ {0, 1}ai ∈ {−1, 0, 1}
I wi indicates whether the ith thruster pair is on or offI ai indicates the acceleration of the valve core rods of the ith thruster pair
Stanton, Marchand Finite Set Control Transcription 14
university-logo
BackgroundApplicationsConclusions
Linear Switched SystemLunar LanderSmall Spacecraft Attitude Control
Small Spacecraft Attitude Control: Variable Thrust
Actual Trajectory
Desired Trajectory
0 2 4 6 8 10 12 14 16 18 20−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
Time (s)
Quaternions: Actual vs. Desired
(a)
0 2 4 6 8 10 12 14 16 18 20−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Time (s)
rad
/s
Angular Velocity: Actual vs. Desired
(b)
Figure: Variable Thrust Attitude Control
Stanton, Marchand Finite Set Control Transcription 15
university-logo
BackgroundApplicationsConclusions
Conclusions
I This investigation explores the range of applications of the FSCTmethod
I The applicability of the method extends to all engineering disciplines
I FSCT vs. Multiple Lyapunov FunctionsI Optimal control laws may be extracted whose performance exceeds
those derived using a Lyapunov argument
I Multiple independent decision inputs managed simultaneouslyI Solutions derived via the FSCT method are utilized in conjunction
with a hybrid system model predictive control schemeI Optimized control schedules can be realized in the context of potential
perturbations or other unknowns
I Some continuous control input systems may be more accuratelydescribed as systems ultimately relying on discrete decision variables
I Continuous control variables may often be extended into a set ofcontinuous state variables and discrete inputs
Stanton, Marchand Finite Set Control Transcription 16