Download as pdf or txt
Download as pdf or txt
You are on page 1of 36

ISTS 2015-d-49

GALLOP: A Low-Thrust Trajectory


Optimization Tool for Preliminary and
High Fidelity Mission Design

Chit Hong Yam (a.k.a. Hippo), and


Yasuhiro Kawakatsu

JSPS Postdoctoral Research Fellow


Institute of Space and Astronautical Science (ISAS),
Japan Aerospace Exploration Agency

30th ISTS, 9 July, 2015


Background: Low-Thrust Trajectory Optimization Tool

Earth-Mars-Vesta Flyby Mission


The Sims-Flanagan direct
transcription method (1999)
• Electric propulsion operates on
a significant time of the
trajectory
• Indirect methods can be
sensitive to initial guess
• Direct method: willing to accept
lower accuracy for a quick and
robust algorithm

JPL’s preliminary low-thrust • Discretize low-thrust arcs into


mission design tool: MALTO impulsive V
Purdue University: GALLOP
2
Background: Low-Thrust Trajectory Optimization Tool

Earth-Mars-Vesta Flyby Mission


Main Developers
MALTO: Jon A. Sims, Paul
A. Finlayson, Edward A.
Rinderle

GALLOP: Troy McConaghy,


Hippo Yam

JPL’s preliminary low-thrust


mission design tool: MALTO
Purdue University: GALLOP
3
GALLOP Software Layout

Initial Guess Tool: Modified solution file


STOUR-LTGA, PaGMO

GALLOP (written Optimizer:


Input File Solution File
in C and Matlab) SNOPT

0.7

Maximum
0.6 Solution
V magnitude (km/s)

0.5

0.4 Post-Processing Scripts


(written in MATLAB)
0.3

0.2

0.1

0
0 5 10 15 20 25 30 35 40 45 50 55 60
Segment

4
Example: Nuclear Electric Propulsion Mission to Pluto

0 E-3
V-2 • Gravity-assists at Venus,
-5
Earth, and Jupiter
J-4
E-1
• Earth Launch: May 8,
-10
To Jupiter
2015
y, AU

-15 • TOF = 10.5 yrs


-20 • Initial mass = 20,000 kg

-25
• Final Mass = 9,196 kg
P-5

Arrival V∞ = 0
-30
-10 -5 0 5 10 15 20 25
x, AU

5
GALLOP: Gravity Assist Low-thrust Local Optimization Program

NEP mission to Jupiter


• Transforms Optimal Control
Problem into Nonlinear
Programming Problem
• Improve on the original model
while maintaining fast and
robust
• Apply to preliminary mission
planning and high fidelity
mission design

Investigate new ways to improve on


- Control variables formulation
- Dynamical system propagation
6
Trajectory Model

Δ Matchpoint Objective: Max. final mass


● Segment midpoint Segments are●inSegment midpoint
Impulsive ΔV equal durationsImpulsive ΔV Constraints:
| Segment boundary | Segment boundary

• V magnitude is limited by
ΔVN
the max. thrust of the
spacecraft

ΔV
V33
Vi ≤ Vmax i = 1,2,…,N
Max. Final
Mass ΔV
V22 • Equality constraints on the
ΔV
V11
mismatches at matchpoint
 r ,  r ,  r ,  v ,  v ,  v ,  m = 0
x y z x y z
Low-Thrust Trajectory Model

7
Trajectory Model

Δ Matchpoint
Variables
● Segment midpoint Segments are●inSegment midpoint
Impulsive ΔV equal durationsImpulsive ΔV • Departure and arrival times
| Segment boundary | Segment boundary
• Departure V∞
ΔVN
• Flyby conditions (zero-
sphere of influence)

Max. Final
ΔV
V33 • V magnitudes and
Mass ΔV
V22 directions at each segment
ΔV
V11
• 3N variables
N nonlinear constraints
Low-Thrust Trajectory Model (+7 x Leg)

8
Formulation of the Control: Delta-V

• Formulation of the V
vectors can influence the
convergence behavior
• McConaghy and Longuski
studied 4 formulations:
Cartesian, Spherical,
Magnitude and Cartesian,
Magnitude and Direction
• Conclusions:
Without good initial guess –
Cartesian is more robust
With good initial guess –
Spherical is faster

9
The N-Vector Formulation

Earth-Jupiter Flyby Mission

0.7 200
Maximum Thrusting Segment
Solution 150 Coasting Segment
0.6

100
V magnitude, km/s

Clock Angle  , deg


0.5
50
0.4
0
0.3
-50

0.2
-100

0.1 -150

0 -200
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Segment Segment

Variables: ΔV1, ΔV2, …., ΔVN Variables: θ1, θ2, …., θN

10
The ΔV Angles as Chebyshev Series

θ θ(x,c0,c1,…,cn) = c0T0(x) + c1T1(x) +…+ckTk(x)


where Tk(x) is the Chebyshev
θ2
Polynomial of degree k
θ1
θN
T0 ( x)  1
T1 ( x)  x
T2 ( x)  2 x 2  1
Time, x T3 ( x)  4 x 3  3 x
T4 ( x)  8 x 4  8 x 2  1
Variables: {c0, c1,…,ck}θ
{c0, c1,…,ck}ψ for x  [1, 1]

11
The ΔV Magnitude as On/Off-Node Formulation

Earth-Jupiter Rendezvous
4
Assume Max. Thrust
3 or Null Thrust
Off-Node

2
y, AU

1
On-Node
On
Earth
0
Jupiter

-1 Off
-2
-5 -4 -3 -2 -1 0 1 2
x, AU

Variables: {x, y, z, vx, vy, vz, m, t}on, off

12
Earth-Mars-Vesta Flyby Test Case

• SEP Spacecraft, m0 = 545 kg


0.5
Thrust-On
M-2
Thrust-Off
Node
Node
E-1 • Maximize mf
0

• C-T-C from Earth to Mars, T-


y, AU

-0.5
C from Mars to Vesta
-1
• Use 4th degree Chebyshev
-1.5
series on θ and ψ
-2 Vesta-3

-1.5 -1 -0.5 0 0.5 1


x, AU

Characteristics N-Vector On/Off-Node Cheby. Node +


Cheby.
# of Variables 202 110 98 78
# of Constraints 76 37 76 37

13
Earth-Mars-Vesta Flyby Test Case

Launch Period Study


for 120 days with step size = 5 days, 24 runs
540

520
Formulation Total Run
Time, sec
500
N-Vector 62
Final Mass, kg

480 Node 27
~ 1 % difference in mf
460 Chebyshev 34
440 Node + Cheby. 15
420
N-Vector
400 Node
Cheby
380 Node + Cheby

360
0 20 40 60 80 100 120
Launch Date (days since July 25, 2009)

14
Earth-Mercury Rendezvous Test Case

Launch Period Study


2
1

Final Mass minus 370.0 kg


1.8

0.5 1.6
Node + Cheby.
Mercury 1.4
y, AU

0
1.2
0.2 % difference in mf
1
-0.5
0.8 N-Vector

-1 Earth
0 20 40 60 80 100
-1 -0.5 0 0.5 1 Launch Date (days since 09-Apr-2007)
x, AU
• 15 On/Off pulses; 2nd to 4th degree Formulation Total Run
Chebyshev series on θ and ψ. Time
• Run time is reduced by an order of N-Vector 219 min
magnitude and results are accurate Node + Cheby. 7 min
within 0.2%.

15
New Dynamical Model: Continuous Thrust Propagation

• Original: Impulsive V connected by conics


• Continuous Thrust Model: thrust are constant (inertial or rotating
frame) within a segment and are integrated numerically
• A fast and accurate Taylor integration method is employed
• Removed N constraints by using normalized thrust (e.g. Fx/Fmax)
No need to have Vi ≤ Vmax
16
Earth-Mercury Rendezvous: Impulsive vs Continuous thrust

Initial mass = 660 kg, Thrust = 92mN, Isp = 3300s


1 1

0 .5 0 .5

y, AU
y, AU

0 0

-0 .5 -0 .5

-1 -1
-1 -0.5 0 0 .5 1 -1 -0 .5 0 0 .5 1
x, AU x, AU

Impulsive V model Continuous thrust


model
mf = 392.9 kg mf = 382.4 kg
17
Adaptive Mesh 1: Sundman Transformation

Adaptive mesh: What if the segment length changes with


distance from the center?
r  rx2  ry2  rz2
r  r r r
x
2
y
2
z
2

rx  rvx
rx  v x
ry  v y dt ry  rv y

rz  v z
ds  rz  rvz

v x   rx / r  u x
3
r vx  rx / r 2  ru x
v y   r y / r  u y
3 v y  ry / r 2  ru y

v z   rz / r  u z
3 vz  rz / r 2  ru z
t  r
• Sundman transformation: time is slower when closer to the center
• Propagation is changed from equally-spaced time-domain to s-domain
18
Examples of s-domain sampling

Start with a circular orbit, propagate with a constant thrust in the x-direction

Sampled equally in t-domain s-domain

constant
tangential
thrust

19
Continuous thrust t-propagation vs s-propagation

Continuous thrust Continuous thrust


t-propagation s-propagation
mf = 382.4 kg mf = 387.0 kg

20
Adaptive Mesh 2: Propagate with True Anomaly

transverse
radial
(local horizon)

Advantage: the mesh is uniform


w.r.t. the orbital geometry or angle
Limitation: singularity at e or h = 0

21
Example: Trajectory with Many Revolutions

Sample Orbit Raising


Trajectory

JAXA Mission Concept:


DESTINY

22
DESTINY Orbit Raising: Time vs θ Sampling

Initial Orbit: Period = 8.4 hours, Spacecraft: m0 = 400 kg,


Thrust = 0.04 N. Propagate for 100 revs.
4
4
x 10
x 10 1
1

0.5
0.5

0
0

-0.5 -0.5

-1
Red: thrust -1

-1.5 Blue: coast -1.5

-2 -2

-2.5
-2.5

-3
-1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 -3
4 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5
x 10 4
x 10

Time-sampling θ-sampling
Segment length = 2.1 hrs Segment length = 90o 23
High Fidelity Dynamics Example: Micro-Spacecraft PROCYON

• Dynamics changed from 2-


body to N-body
• Continuous Thrust Model with
equal duration in N-body system
• Gravity assist: formulate as
closest approach state (or as a
asteroid’s pseudo V∞)
orbit
asteroid’s
orbit
• Under testing, not fully
Earth’s orbit implemented yet
2000
DP107

May 2016 Launch: Dec 2014


Flyby: Dec 2015

24
Conclusions and Future Work

• GALLOP is a low-thrust trajectory optimization tool targeted for


preliminary and potentially high fidelity mission design.
• GALLOP has extended the original model to include the following
capabilities:
• On/Off and Chebyshev approximation on the controls
• Continuous thrust (instead of impulsive ΔV)
• A fast and accurate Taylor integrator
• Adaptive mesh using Sundman transformation
• Adaptive mesh using true anomaly
• GALLOP has been applied to design NEP missions, Earth-Mars
cyclers, Jupiter Moons Tour (GTOC6), and PROCYON.
Future Work
• Benchmarking various example problems
• Implementing full N-body dynamics
• Releasing as an open-source tool
25
Q&A

Contact
Chit Hong Yam (a.k.a. Hippo)
Email: <chithongyam@gmail.com>
<chithong.yam@ac.jaxa.jp>

JSPS Postdoctoral Research Fellow


Institute of Space and Astronautical Science (ISAS),
Japan Aerospace Exploration Agency
Backup Slides
Trajectory Model of GALLOP
(Gravity-Assist, Low-thrust, Local Optimization Program)

Δ Matchpoint
● Segment midpoint
• Trajectory is divided into legs
● Segment midpoint
Impulsive ΔV Impulsive ΔV – Legs begin and end with planets
| Segment boundary | Segment boundary– One matchpoint on each leg

• Legs are subdivided into segments


ΔVN

• Impulsive V at each segment


– Magnitude limited by duration of
segment
ΔV
V33
• Conic propagation between impulses
ΔV
V22
– Propagate forward and backward
ΔV
V11
from planets to matchpoint
• Discontinuities at matchpoints are
GALLOP constrained to be zero
After the Sims and Flanagan Model

28
Hippo Yam
Variables and Constraints
• Objective: maximize arrival mass or
minimize launch mass
● Segment midpoint
• Variables
Impulsive ΔV
– Bodies
| Segment boundary
 Time
 Mass (also the objective if launch or
ΔVN arrival)
 V∞
 Flyby altitude and angle
– Segments
 V magnitude and direction
ΔV
V33
ΔV
V22 • Constraints
– Matchpoints
ΔV
V11
 position, velocity, mass = 0
– Segments
 V magnitude ≤ Vmax

• First derivatives are calculated analytically

29
Hippo Yam
Taylor Integration Method

• Replacing Keplerian
propagation with numerical
integration means a significant
degrade in speed
• Use Taylor integration method
instead of “standard” Runge-
Kutta integration scheme
• Based on Taylor’s expansion
and automatic differentiation
Online Taylor integrator,
by Jorba and Zou

30
Comparing Taylor Integration Method with RKF

r  rx2  r y2  rz2 • Integrate with 10,000 random


initial conditions, controls, and
rx  v x times
ry  v y • Integrate forward then backward to
rz  v z check the accuracy of the solutions,
report the worst case as the max.
v x   rx / r 3  u x error
v y   r y / r 3  u y • The speed gained by using Taylor
v z   rz / r 3  u z is about an order of magnitude

31
How does the number of segments affect results?

Earth-Mercury
Rendezvous
380

378
N = 40
Mercury

376
Final Mass, kg

Earth

374

372

370
Assumes mf approaches an
asymptotic limit when N is large
368

mf,N→∞  mf,Nmax
0 50 100 150 200 250 300 350
Number of Segments

Hippo Yam 32
Error and Segment Duration

Relative Error  Segment Duration 


  m f  m f ,N max
m f , N max   TOF N
-0.5

-1 log10( ) = 2.085*log10() - 6.576


Earth-Mercury Rendezvous
-1.5

-2
Relative Suggested Segment
Error Duration, days
-2.5
 = 10-1
log10( )

-3
472
-3.5  = 10-2 157
-4
2
 = 10-3 52
R = 0.782
-4.5  = 10-4 17
-5  = 10-5 6
-5.5
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6
log10(),  in days

Hippo Yam 33
Guidelines in Choosing the Segment Duration

Trajectory  = 10-1  = 10-2  = 10-3  = 10-4  = 10-5


Earth-Neptune flyby
480 205 88 37 16
(TOF = 3474 days)
Earth-Neptune flyby
1001 376 141 53 20
(TOF = 4000 days)
Earth-Saturn rendezvous
569 212 79 29 11
(TOF = 2700 days)
Earth-Mercury rendezvous
472 157 52 17 6
(TOF = 2327 days)

Relative Error  Segment Duration 


  m f  m f ,N max
m f , N max   TOF N
Hippo Yam 34
Example: Earth-Mercury Rendezvous Mission

0 .5

Mercury

y, AU
0

Earth
-0 .5

similar to Deep Space’s 1 engine


-1
-1 -0 .5 0 0 .5 1
x, AU

35
4
x 10
3.75

3.7 Time
mprop = 0.32 kg
mprop = 0.97 kg
Apogee (km)
3.65

θ
3.6

3.55

3.5
6500 6600 6700 6800 6900 7000 7100 7200 7300
Perigee (km)

You might also like