Professional Documents
Culture Documents
12
(3)
Two vectors that define the shape and orientation of the Impulse Minimization
initial and final orbits complete the transfer geometry:
Minimization of Eq. (9) by a steep descent technique
Downloaded by INST TEC DE AERONAUTICA (ITA) on May 22, 2018 | http://arc.aiaa.org | DOI: 10.2514/3.3363
ey = eyjcoscoy, sincoy cosi'y, sincoy siniy] y = 1, 2 (7) requires computation of the gradient. Upon differentiation,
Eq. (9) provides the following expression:
The true anomaly interval traversed in the transfer orbit
(A0) may be determined directly: dl = [(I2-dI2)/|!2 (20)
cosA0 - 0° < A0 < 180° (8) or
No generality is lost if the true anomaly interval is limited 57
to the first two quadrants. Although this does restrict the (21)
problem to "short transfers/' the "long transfers" (180° <
A0 < 360°) may be computed by changing the signs of the dl
velocity vectors in the transfer orbit. If sinA0 = 0, the (22)
plane of the transfer is no longer uniquely determined and
an alternate formulation is required to avoid this singularity.6 57 [I:
(23)
Impulse Computation
Terms in the above expressions may be expanded as follows:
The function to be minimized is the total impulse for the
two-impulse maneuver: bl/dp, - ±(e>Vtt/dp«) - (dVi/dp.) (24)
I = + I Is (9) (25)
where
11 = ±V a - (10) (26)
12 = V2 T V(2 (11) (27)
(When a double sign is used, the upper sign refers to a "short (28)
transfer.")
Velocity vectors in the initial and final orbits at the de- (29)
parture and arrival points (Vi and V2) and the corresponding Noting that dVi/dpf and dV2/dp« are each zero results in a
velocity vectors in the transfer orbit (V^ and V<2) are com- simplified expression for d//dp* that may be obtained through
puted as follows: several algebraic manipulations:
X (eHhUi) (12) I 2 -(v +
= d= (30)
X (e 2 -hU 2 ) (13)
X (e, -f uo (14) Additional expressions are obtained from Eqs. (26-29) by
direct differentiation of the vector equations:
X (e, -f U 2 ) (15)
5 cscA0 cscA0 c
The final impulse equations are obtained from Eqs. (10-15)
by substituting Eq. (6) and performing several algebraic " n 50i
manipulations : / M /
- cscA0—-1 ) -
V r2/\
l! = ±[ v + zUi] - Vi (16)
TT
I2 = V2 T [v - zU2] (17) 1
~
where
cosi, — sim'] (32)
v = [(/ip() 1/8 (r 2 - r O J / l r : X r 2 | (18)
avtt
* = (M/P«) 1/2 tan(A0/2) (19)
Impulses corresponding to "long" and "short" transfers
are compared by changing the double sign in Eqs. (16) 502
and (17), and the combination producing the lesser impulse
is used for the remaining computations. 5Vi/502 = 0 (34)
2330 G. A. McCUE AND D. F. BENDER AIAA JOURNAL
(36)
considered in order to satisfy rendezvous constraints.9
Rapid convergence was not considered a necessity. Instead,
it was desired to insure .a methodical search to the nearest
local minimum.
cscA0 The numerical results presented here were obtained by an
- cscAfl ^ ] -
adaptive descent process that conducts a step by step nu-
merical search employing Eqs. (1-47). The search is termi-
cscA0 nated when the following necessary conditions for a local
minimum have been achieved:
-| (37)
t>fa ' 502 < 6 < € <«
1/2
0 < €« 1 (48)
bV2/c>4>2 = (M/p2) [-cos4>2, -sin<£2, 0] (38)
During the nth step of the search the gradient vector is
The remaining terms in Eqs. (31-38) may be computed computed and the n + 1st coordinate vector is determined
from the following expressions: as follows:
= [—sinfa, cos fa cosi, cosfa sini] (39)
dU2/c>02 = [-sin02, coscfe, 0] (40) - 0 0
Since the necessary conditions for minimum impulse [setting Initial orbit pi = 5000 miles, ei = 0.2, «i = -90°0, i = 5°
Eqs. (21-23) equal to zero] yield an intractable set of equa- Final orbit p2 = 6000 miles, ez = 0.2, o>2 = +30°0
7094
tions, one is faced with the problem of obtaining numerical Opti- time,
solutions. Early attempts to minimize the impulse function mum 0i,° 0o,° Pt, miles Impulse, fps sec
considered here were successful but often required excessive
amounts of computer time.4'5 The scheme reported in 1 73 .8152 187.5568 6644 .8496 4902 .651223852 3.4
Hefs. 4 and 5 was a step wise steep descent procedure similar 2 40 .8343 298 .2634 6617 .7904 5343 . 148693477 2.5
3 177 .8114 73 .6465 4611 .8023 5393..781144757 3.0
to one of several methods discussed by Kelley.8 That is, 4 308..2034 37,.7403 4592..8574 5654. 191209679 2.8
the local negative gradient was computed, and a step-by-step
DECEMBER 1965 MINIMUM IMPULSE ORBITAL TRANSFER 2331
The control logic for the optimization process is as follows: n = 2000'. 1 = 2097.825
1) If the inequality
7(01, 02, P«)»+l < 7(01, 02, Pt)n (50) p = 1000 1= 2097.827
3120.0 -
is not satisfied, a is decreased by a prescribed percentage,
and a new coordinate vector (0i, 02, pt)n+i is computed. P
This nth stage of the process is repeated until Eq. (50) is MILES
n = 500 I - 2097.836
numerical demonstration of the nonoptimality of certain Fig. 7 Lawden spiral AF compared to optimum two-
Lawden spiral maneuvers was also noted in Ref. 16. Using impulse AF (percentage difference noted).
the adaptive descent program, these numerical results have
now been expanded to give a broad comparison of the two-
impulse maneuver and the Lawden spiral. All possible transfers may be generated by allowing S = sm\f/
In Refs. 13 and 14 Lawden develops formulae for an inter- to increase monotonically from 0.0 to 3~1/2. It is clear that
mediate thrust arc that locally satisfies the necessary condi- as S increases, \(/ and r also will increase monotonically from
tions for an optimal planar trajectory in an inverse-square 0.0° to 35°. 2 and from 0.0 to infinity, respectively.
gravitational field. The path is a spiral defined as follows: Orbits that osculate the spiral trajectory may be computed
from the following formulae13-14:
r = a0S6/(l - 3S2) (51) p -= 6
- 4S2)2/(3 - 5S2)2]
ao[S (3 (53)
where r is radius from the attracting center, and a0 is a 6 2 2 2 4 6
proportionality constant whose units are distance. In a = a0[S (3 - 5S ) /(9 - 72S + 169S - 126S )] (54)
Eq. (51), S = sini/' where \f/, the angle denoting thrust e = (1 - p/a)1/2 (55)
direction, is defined in Fig. 5. The central angle traversed
during a given Lawden spiral maneuver may be computed The magnitude of the acceleration at any point is given by
from the following relation: S*"|
6
°*4' (56)
.\l/ + 3 cot^f. (52) J
and the total AF to transfer between two points on the spiral
is given by
-3(1 - -152
(
= (!)"'[ (3 - 5S ) 2
1 (57)
By varying sin2^, one may construct a sequence of orbits
that osculate the Lawden spiral. Table 2 presents the
elements of a number of these osculating orbits that were
used in a comparison of optimum two-impulse transfers and
the corresponding Lawden spiral maneuver. The elements
were produced by a single precision computer program by
CO, True
relative to anomaly of
first orbit's osculation
SinV p, miles e perigee point
0.01 -
0.04 65.84162 0.4031143 0 103°. 5837
0.06 225.7066 0.4958521 147°. 9592 106°. 7799
0.08 543.9924 0.5752421 233°. 6340 109°. 5519
0.10 1081.600 0.6464052 290°. 3322 112°. 0700
0.001 0.002 0.004 0.01 0.02 0.04
0.12 1905.120 0.7120126 330°. 8360 114°. 4234
0.14 3088.218 0.7737022 361°. 2268 116°. 6671
A(SIN%) 0.16 4713.446 0.8325991 384°. 8082 118°. 8397
0.18 6874.618 0.8895526 403°. 5450 120°. 9707
Fig. 6 Lawden spiral AF compared to optimum two- 0.20 9679.999 0.9452619 418°. 6843 123°. 0848
impulse AF.
DECEMBER 1965 MINIMUM IMPULSE ORBITAL TRANSFER 2333
setting do equal to 106 miles. Velocities and impulses are rapidly as the distances separating the points of osculation
proportional to (ju/a0)1/2. For the calculations presented increases. This percentage difference also increases as
here /* was set equal to 95634.50100 miles3/sec2. sin2^ of the initial orbit decreases.
An IBM 7094 double-precision computer program was By plotting AF for the two maneuvers as a function of
utilized to compute optimum two-impulse transfers between A0 one may demonstrate that the Lawden spiral becomes
numerous pairs of orbits which osculate the spiral. Nu- more nearly optimal as the arc length traversed approaches
merical results pertaining to these computations are sum- zero. Figure 8, which was constructed from the numerical
marized in Table 3. The calculations were performed for results appearing in Table 3, indicates that AFz,s — AF2-imp
various initial orbits (sin2^ initial) and various separations increases as the 5th power of A0. Extrapolating this nu-
between the points of osculation, A (sin2^). The central merical result leads one to conclude that the Lawden spiral
angle A# traversed during each of the maneuvers is also can be optimal only over a differentially small arc length.
noted. The velocity changes AF required for various This is consistent with recent results reported by Robbins.17
Lawden spiral arcs (AVLs) are tabulated alongside the
impulse required for the corresponding optimum two-impulse Minimum Impulse Transfer between Inclined Orbits
transfer (AF 2 -i mp ). The calculations are accurate to the
number of significant figures tabulated. Table 3 also Although Contensou,18 Breakwell,19 and Lawden14 have
tabulates the difference in AF for the two maneuvers and demonstrated the existence of optimal one- and two-impulse
compares this difference to the AF for the spiral maneuver. maneuvers between coplanar noncoapsidal elliptical orbits,
In all cases computed, a two-impulse maneuver that required the authors are not aware of a corresponding proof for inclined
less AF than the corresponding Lawden spiral maneuver orbits. However, the analyses of Refs. 14, 18, and 19 could
was found. be extended to inclined orbits and one would expect to obtain
In Fig. 6 the difference in velocity change required for the results similar to those obtained for coplanar orbits.
maneuvers AF^s — AF2-imP is plotted as a function of For a two-impulse transfer between inclined orbits to be
position difference between the osculation points. A family optimal it is necessary that it be composed of two optimal
of curves was generated by varying sin2;/' of the initial orbit. one-impulse maneuvers. That is, the transfers from the
The single precision Lawden spiral calculations were not initial orbit to the transfer orbit and from the transfer orbit
sufficiently accurate to allow these comparisons to be ex- to the final orbit must each be optimal. To investigate
tended to smaller values of A(sin2^). All the curves pre- this necessary condition numerically the optimum two-
sented indicate that the difference in velocity change increases impulse transfer between several different inclined non-
approximately as the 4.7 power of A(sin2^). coapsidal elliptical orbit pairs was computed. The impulse
Figure 7 was constructed to allow rapid comparison of functions concerning transfers from the initial orbit to the
AFz,s and AVLs — AF2-imp. Several diagonal lines were transfer orbit and from the transfer orbit to the final orbit
drawn to denote the percentage difference between the AF were each examined in considerable detail using the adaptive
required for the two maneuvers. In this figure the different descent program. Table 4 contains orbital elements for
curves were generated by varying the separation between one such set of orbits. In all cases these impulse functions
the points of osculation on the Lawden spiral. The per- exhibited the flat bottomed valleys encountered with optimal
centage difference in AF for the two maneuvers increases one-impulse transfers between coplanar orbits. These
Table 4 Elements of initial, final, and optimum transfer orbit (two-impulse transfer)
100.0 References
1
McCue, G. A., "Optimum two-impulse orbital transfer and
rendezvous between inclined elliptical orbits," AIAA J. 1,
1865-1872 (1963).
2
Des Jardins, P. R., Bender, D. F., and McCue, G. A.,
"Orbital transfer and satellite rendezvous," North American
Aviation, Inc. Final Kept. SID 62-870 (August 31, 1962).
3
McCue, G. A., "Optimization and visualization of
Downloaded by INST TEC DE AERONAUTICA (ITA) on May 22, 2018 | http://arc.aiaa.org | DOI: 10.2514/3.3363