Bradley J. Wall*
Embry-Riddle Aeronautical University, Prescott, AZ, 86301
Mission design using low-thrust propulsion requires a method for approximating the
spacecraft’s trajectory and its cost. This can be for the purpose of evaluating a large search
space or for providing a suitable initial guess to be used with more-accurate trajectory
optimizers that are more likely to converge when given an initial guess satisfying the
equation of motion constraints and the terminal boundary conditions. A new low-thrust 3D
shape-based approximation method is developed and tested on two sample problems that
require interception and rendezvous trajectories. The method quickly locates near-optimal
minimum fuel solutions for the examples presented.
Nomenclature
α = thrust pointing angle (x-y plane)
β = thrust pointing angle (out-of-plane)
γ = flight path angle
θ = angle between the initial and terminal position vectors
θ = angular velocity in rad/sec
θf = total transfer angle
µ = gravitational constant
a, b, c, d, e, f, g = 7 inverse polynomial coefficients for the 2D shape-based method
AU = astronomical unit (1.496·108km)
a z, b z, c z, d z = 4 polynomial coefficients for the z-direction of the 3D shape-based method
cexh = low-thrust exhaust velocity
m = mass
n = thrust along (0) or against (1) the velocity vector
Nrev = number of full revolutions about the sun
q = exponent of the highest order out-of-plane term
r = radial position
s = distance from the spacecraft to the sun in 3D cylindrical coordinates
Ta = total thrust acceleration
Tain = thrust acceleration in the x-y plane
Taz = thrust acceleration in the z-direction
v∞ = hyperbolic excess velocity
Subscripts
I. Introduction
M ission design using low-thrust propulsion requires a method for approximating the spacecraft’s trajectory and
its cost. This can be for the purpose of pruning a large search space or for providing a suitable initial guess to
be used with more-accurate trajectory optimizers. Such optimizers, for example direct transcription methods in
*
Assistant Professor, Aerospace and Mechanical Engineering, 3700 Willow Creek Rd., and AIAA Member.
1
American Institute of Aeronautics and Astronautics
092407
Copyright © 2008 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
which the continuous problem is converted into a nonlinear programming problem, are more likely to converge
when given an initial guess satisfying the equation of motion constraints and the terminal boundary conditions.
Many approximation methods have been used to find near-optimal low-thrust spacecraft trajectories in significantly
less time than locating a true optimal solution.
Recently, Petropoulos and Longuski1 have developed a shape-base approach for low-thrust interception
trajectories using an exponential sinusoid. While this method is novel and has been proven to provide near optimal
low-thrust trajectory approximations, it does have some disadvantages.
1. The four coefficients defining the exponential sinusoid allow for the initial radial position, initial velocity
components, and terminal radial position boundary constraints to be satisfied but the transfer time
constraint must be satisfied by changing the departure or the arrival times. Thus, this method does not
result in a low-thrust interception trajectory for any combination of departure and arrival times.
2. The coefficients are not easily determined; the solution of two nonlinear equations of two variables is
required.
3. The number of coefficients limits the method to low-thrust interception trajectories and not the more
complicated case of low-thrust rendezvous trajectories.
De Pascale and Vasile2 have developed a three-dimensional shape-based method using pseudoequinoctial
elements capable of low-thrust rendezvous trajectories. While this method is also novel and has been proven to
provide near optimal low-thrust trajectory approximations, it too has a disadvantage. Three additional unknowns
associated with the shape are required in addition to the unknowns of the Lambert problem, i.e. the number of
heliocentric revolutions, the departure date, and the arrival date, meaning that for a given set of Lambert problem
boundary conditions there is not a unique solution.
This paper discusses a new shape-based technique for low-thrust rendezvous trajectories using a 6th degree
inverse polynomial for the in-plane motion3,4 and a polynomial with four coefficients for the out-of-plane motion.
Using these functions, the boundary constraints associated with the rendezvous problem, the initial position and
velocity components, terminal position and velocity components, and the transfer time, may all be satisfied. The
equations of motion are also satisfied using a shape-based technique by solving for the required Ta and α time
histories such that the spacecraft follows the given shape. By allowing the terminal velocity components to be
optimization parameters, this method may also be used for low-thrust interception trajectories.
1
r= (1)
a + bθ + cθ 2 + dθ 3 + eθ 4 + f θ 5 + gθ 6
A brief derivation of the constraints on the parametric function is shown below. A more detailed derivation is
shown in Ref. 3.
The rendezvous problem of interest is depicted in Fig. 1 where the total transfer angle is:
θ f = θ + 2π ⋅ Nrev (2)
2
American Institute of Aeronautics and Astronautics
092407
The angle θ is the angle between the position vectors of body 1 and body 2 measured counter-clockwise, as shown
in Fig. 1, and Nrev is the number of revolutions about the sun, an additional unknown that obviously applies to
multiple revolution trajectories. From Fig. 1 it is clear that θ is zero when the spacecraft is at position 1. Thus the
constraints on the coefficients a, b, and c are:
1 tan γ 1 1 µ
a= , b=− , c= 3 2 − 1 (3)
r1 r1 2r1 r1 θ1
r2 θ
r1 γ, direction
of
θf first point
in Aries
where µ is the gravitational constant, γ1 is the flight path angle and θ1 is the angular velocity at departure, all known
quantities. Using the fact that θ is equal to θf when the spacecraft is at position 2, expressions for coefficients e, f,
and g may be written as:
1
− ( a + bθ f + cθ f2 + dθ 3f )
r
e 30θ 2f −10θ 3f θ f4 2
f= 1 tan γ
2θ 6 −48θ f 18θ f2 −2θ 3f − 2
− ( b + 2cθ f + 3dθ f2 ) (4)
f r2
g 20 −8θ f θ f2
µ − 1 + 2c + 6 dθ
r 4θ 2 r f
2 2 2
where coefficient d is iterated until the time of flight constraint, shown in Eq. (5), is also satisfied.
tf θf 4
r (θ )
∫ dt = ∫
0 0
µ
(1 r (θ ) ) + 2c + 6dθ + 12eθ 2 + 20 f θ 3 + 30 gθ 4 dθ (5)
The integration of the right-hand side of Eq. (5) is accomplished via trapezoidal rule to facilitate faster computation
times with respect to more accurate integration schemes, e.g. ode45 in MATLAB5. The iteration of coefficient d is
done using the MATLAB function fzero and is quite robust using a simple guess of zero.
The thrust magnitude is required such that the spacecraft following the inverse polynomial trajectory satisfies the
equations of motion, shown in Eq. (6).
3
American Institute of Aeronautics and Astronautics
092407
µ
r − rθ 2 +
= Tain sin α
r2
(6)
1 d 2
r dt
( )
r θ = Tain cos α
where Tain is the thrust acceleration in the x-y plane and α is the thrust pointing angle for the low-thrust spacecraft.
For this derivation it is convenient to assume that the thrust direction is either along or against the velocity vector,
i.e.:
α = γ + nπ (7)
where n is 0 or 1 for along or against the flight path angle respectively. It can be shown that the thrust
acceleration, Tain , is.
where r is from Eqn. (1) and the flight path angle γ is determined by Eq. (9).
r
tan γ = = −r ⋅ ( b + 2cθ + 3dθ 2 + 4eθ 3 + 5 f θ 4 + 6 gθ 5 ) (9)
rθ
Assuming that the flight path angle γ is always between –π/2 and π/2 rad, i.e. that there are no retrograde
trajectories, the cosγ term in Eqn. (8) is easily determined.
µ
r − rθ 2 = − 3 r + Tain sin α
s
1 d 2
r dt
( )
r θ = Tain cos α (10)
µ
z=−
z + Taz
s3
The only reason this particular shape was chosen was because it was known a priori that the coefficients could
be solved for via a linear system of equations. Given the initial and terminal z position and velocity, the four
coefficients defining the function z(θ) may be determined from the 4th order linear system:
az = z1
z1
bz = (13)
θ1
z − a − bθ f
cz 1 qθ f −θ f2 2
d = q z2
z θf − ( q − 1) θ f θ − b
2
With the coefficients known, the thrust acceleration in the z direction, Taz, may be determined from the equation
of motion in the z-direction, Eq. (10);
µ
Taz = z − cz ( q − 1)( q − 2 ) θ q −3 + d z q ( q − 1) θ q − 2 θ 2 + bz + cz ( q − 1) θ q − 2 + d z qθ q −1 θ (14)
r3
where θ and θ are from Eqs. (15) and (16) respectively.
µ 1
θ 2 = 4 (15)
r (1 r ) + 2c + 6dθ + 12eθ 2 + 20 f θ 3 + 30 gθ 4
The total thrust acceleration is then determined from Eqs. (8) and (14):
Note that it is assumed that z is small relative to r and therefore one may assume that s is equal to r. This is done
so that the two-dimensional shape-based method can be used in its entirety with no modifications, therefore
retaining the convenience of all of the shape-based coefficients being solved for via a linear system of equations
except for coefficient d that satisfies the final time boundary condition. Note that this does exclude this method
from arbitrary inclination changes but is still valid for inclination changes less than 15° where the relative error
between r and s is less than 5%. This limitation still includes most space missions to other celestial objects.
The exponent variable q is intended to be any integer equal to or greater than three although the formulation does
not require this, i.e. q may be any real number. Equations (13) and (14) are derived for an arbitrary value of q
because different values of m may be better suited for specific space maneuver applications. For the example
problems in this paper a value of 7 was chosen because, when tested on simple circle-to-circle transfers, this value
resulted in a high final mass while not having an excessively high maximum thrust acceleration value. It is obvious
that the maximum thrust acceleration value is arbitrary and dependent on the spacecraft’s propulsion system thus it
is recommended to first find the best value of q for your spacecraft before using an arbitrary value. Also, it is
unwise to let m be an optimization parameter because as q increases the final mass tends to decrease however the
maximum thrust acceleration increases, albeit for a shorter time interval.
5
American Institute of Aeronautics and Astronautics
092407
The remaining text in this section is from Ref. 4 and is written again here for convenience. It is often required to
maximize the final spacecraft mass, i.e. minimize propellant consumption for the mission, therefore an expression
for final mass is also derived. First, it is assumed that the spacecraft has an engine with constant specific impulse.
Then the thrust acceleration, Ta is simply:
T −m ⋅ cexh
Ta = = (17)
m m
where T is the thrust magnitude and cexh is the constant exhaust velocity. Rearranging Eq. (17) yields an expression
for m :
Ta ⋅ m
m = − (18)
cexh
thus:
mf θf
dm Ta
∫
mo m
= −∫
0 exh ⋅ θ
c
dθ (19)
where Ta is from Eq. (16) and θ is from Eq. (14). By integrating Eq. (19) the final mass mf may be determined.
The minimization of fuel consumption may also be expressed as the minimization of the low-thrust ∆V given
by Eq. (20).
m
∆V = cexh ⋅ ln o (20)
m
f
θf
mf 1 Ta (θ )
ln
m
o
=
cexh
∫ θ (θ ) dθ
0
(21)
where Ta(θ) and θ (θ ) are from Eqs. (16) and (14) respectively. Substituting this expression into Eq. (20) yields:
θf
Ta (θ )
∆V = ∫ θ (θ ) dθ
0
(22)
This expression for the ∆V of the shape-based trajectory is independent of the exhaust velocity cexh. This means
that for the performance index of minimizing fuel consumption, or equivalently minimizing the ∆V, an optimizer
implementing this shape-based method with constant cexh, will converge to the same trajectory independent of what
engine parameters are chosen.
To summarize, given the boundary conditions, one can solve for all of the coefficients associated with the in-
plane and out-of-plane shapes defined by Eqs. (3-5) and (13) respectively. Then, using these coefficients, one can
solve for the required mass or ∆V such that the spacecraft follows the given shapes.
IV. Examples
th
The 6 degree inverse polynomial has been successfully used to find near-optimal 2D low-thrust rendezvous
trajectories via the use of a genetic algorithm3. This paper uses the new 3D shape-based technique to solve an Earth-
Mars rendezvous mission found in Ref. 2 and an Earth-Mars-Ceres example problem similar to that found in Ref. 1.
6
American Institute of Aeronautics and Astronautics
092407
A. Earth to Mars Rendezvous
Table 1. List of the free parameters and their bounds for the Earth-Mars-Ceres test case.
Optimization Parameter Bounds
Launch date Jan. 1, 1990 to Dec. 31, 2049
v 0.75 to 2.0 km/sec
Magnitude of ∞ at Earth departure
v
In-plane angle of ∞ at Earth departure -π to π rad
v
Out-of-plane angle of ∞ at Earth departure -π to π rad
Time of flight from Earth to Mars 0 to 3 years
Magnitude of v∞ at Mars arrival 0 to 3 km/sec
In-plane angle of
v∞
at Mars arrival -π to π rad
v
Out-of-plane angle of ∞ at Mars arrival -π to π rad
Periapse radius of Mars flyby 200 to 10,000 km
Out-of-plane angle of Mars flyby plane -π to π rad
Time of flight from Mars to Ceres 0 to 1133 days
After running for 100 generations and 26 minutes, the resulting solution has a propellant mass fraction of 0.290.
To find the true optimal solution, this shape-based solution was used as an initial guess for a direct method, as done
previously for the Earth-Mars mission. Fifteen nodes were used for each of the two trajectories, Earth-Mars and
Mars-Ceres. This, along with the eight unknowns from Table 1 (not including the v∞ at Mars variables) resulted in
778 variables. The difference between this example and that presented in Ref. 1 is that the thrust acceleration is
assumed constant, 1.6·10-4 m/s2, here where in Ref. 1 the thrust was a function of the available power supplied by
solar panels and therefore was proportional to the inverse square of the distance to the sun. The optimal trajectory,
shown in Fig. 5, has a propellant mass fraction of 0.252, a difference in final mass of only 3.8% from the shape-
based approximation solution.
0.6
z, AU
0
-0.6
-3 -2 -1 0 1 2 3
x, AU
3
Earth Departure
2 Trajectory Arc 1
1 Mars Flyby
y, AU
Trajectory Arc 2
0
Ceres Arrival
-1 Earth
-2 Mars
Ceres
-3
-3 -2 -1 0 1 2 3
x, AU
Figure 5. Optimal Earth-Mars Flyby-Ceres trajectory.
With a launch v∞ of 2.0 km/s, the time of flight from Earth to Mars is 392 days and 741 days from Mars to Ceres
for a total time of flight of 1133 days, the maximum number of days allowed. The Mars flyby occurs at an altitude
8
American Institute of Aeronautics and Astronautics
092407
of 4343 km at a B-plane angle of 34°. Again it is shown, in Fig. 6, that although the thrust acceleration time history
of the shape-based solution exceeds the allowable thrust acceleration, the direct method is robust in locating an
optimal solution.
-4
x 10
5
4.5 Optimal Solution
4 Shape-Based Solution
V. Conclusion
The MATLAB genetic algorithm utilizing the 3D shape-based approximation method is capable of locating
near-optimal solutions for intercept and rendezvous trajectories, assuming modest inclination changes. This method
has proven useful as an initial guess generator for a direct method optimizer. The time required for a shape-based
approximation is on the order of 10-100 times faster than for the true optimal solutions, while providing a solution
within 5% of the optimal solution. This work was done on a single core 3.2GHz personal computer. By introducing
this new low-thrust approximation technique capable of accomplishing both interception and rendezvous
trajectories, mission planning problems with millions of possible missions may be approximated quickly. Those
approximations may then be used for further evaluation and optimization.
References
1
Petropoulos, A., Longuski, J., “Shape-Based Algorithm for Automated Design of Low-Thrust, Gravity-Assisted
Trajectories,” Journal of Spacecraft and Rockets, Vol. 41, No. 5, pp. 787-796, 2004.
2
De Pascale, P., Vasile, M., “Preliminary Design of Low-Thrust Multiple Gravity-Assist Trajectories,” Journal of Spacecraft
and Rockets, Vol. 43, No. 5, pp. 1065-1076, 2006.
3
Wall, B., “Technology for the Solution of Hybrid Optimal Control Problems in Astronautics,” Ph.D. Dissertation, Aerospace
Engineering, University of Illinois, Urbana, IL, 2007.
4
Wall, B., Conway, B., “Shape-Based Approach to Low-Thrust Rendezvous Trajectory Design,” Journal of Guidance,
Control, and Dynamics, Submitted January 2008.
5
www.mathworks.com, 2008.
6
Enright, P., Conway, B., “Discrete Approximations to Optimal Trajectories Using Direct Transcription and Nonlinear
Programming,” Journal of Guidance, Control, and Dynamics, Vol. 15, No. 4, pp. 994-1002.
9
American Institute of Aeronautics and Astronautics
092407