Acta Astronautica 153 (2018) 259–273

Contents lists available at ScienceDirect

Acta Astronautica
journal homepage: www.elsevier.com/locate/actaastro

On the optimal passive formation reconfiguration by using attitude control

D. Spiller a, *, Ko Basu b, Fabio Curti c, Christian Circi d
a
Department of Aerospace and Mechanical Engineering, Sapienza University, Via Eudossiana 18, Rome 00184, Italy
b
Department of Aerospace Engineering, The Pennsylvania State University, University Park, State College, PA 16802, USA
c
School of Aerospace Engineering, Sapienza University of Rome, Via Salaria 851, Rome 00138, Italy
d
Department of Astronautical, Electrical and Energy Engineering, Sapienza University, Via Salaria 851, Rome 00138, Italy

A R T I C L E I N F O A B S T R A C T

Keywords: The possibility to reconfigure satellite formations with utilization of the drag and the solar radiation pressure has
Inverse dynamics been established with previous works. Passive formation reconfiguration maneuvers may be achieved by con-
Particle swarm optimization trolling the attitude of the satellites and varying the area exposed to the perturbation forces. This paper deals with
Spacecraft formations the development of a simple but effective model to understand the possibilities offered by this kind of maneuvers.
Reconfiguration maneuvers Limits and advantages of the presented maneuvers are examined considering different reconfigurations maneu-
vers among close relative trajectories in Low Earth Orbits, Medium Earth Orbits and Geostationary Orbits. The
theoretical investigations are confirmed by numerical simulations where the Inverse Dynamics Particle Swarm
Optimization is employed. The relative dynamics is simulated with a high-fidelity orbital simulator considering all
the perturbations affecting the spacecrafts orbits and the gravity gradient torque influencing the attitude dy-
namics. Several test cases are reported to underline all the possibilities offered by the proposed maneuvering
strategy.

1. Introduction planning reconfiguration maneuvers using a high fidelity orbital simu-

This paper further develops the results reported in preliminary studies
considering the coupling between translational and rotational dynamics
introduced by perturbations for spacecraft formation flying (SFF) [1]. It
has been demonstrated that changing the atmospheric drag and the solar
radiation pressure by modifying spacecraft attitude allows the control of
the relative geometry of the formation. The mission scenario is inspired
by that of the JC2Sat Formation Flying Mission [2]. The motivation for
this study will contribute to the field of SFF missions [3,4].
The use of atmospheric drag and the solar radiation pressure for
formation maneuvers has been considered in a relatively small number of
works up until now. The Differential Drag (DD) has been mainly studied
for formation keeping [5–7] and rendezvous [8–12]. Also the solar ra-
diation pressure has been considered for station-keeping problems [13]
or formation maintenance problems [7,14,15]. Some reconfiguration
maneuvers have been presented in Ref. [16] using DD and in Ref. [17]
using the solar radiation pressure.
In Ref. [1] the problem of formation reconfiguration achieved by
means of drag and solar radiation pressure has been taken into account in
great detail. In that work the authors have developed a software for
lator to include all the most influent perturbations in low Eart orbit
(LEO), medium Earth orbit (MEO) and geostationary orbit (GEO). When
considering the atmospheric drag and the solar radiation pressure per-
turbations, the attitude and the orbital dynamics are coupled. As a result,
it has been shown that the Inverse dynamics Particle Swarm Optimiza-
tion (IPSO) [1,18–21] could successfully plan near minimum-time
reconfiguration maneuvers by means of attitude reorientation
maneuvers.
The goal of this paper is to give a valid theoretical analysis to better
understand the results reported in Ref. [1]. Moreover, new improved
results will be presented to further validate the analytical insight to the
problem. Some well-known close relative trajectories are the Projected
Circular Formation (PCF), the General Circular Formation (GCF) and the
Along-Track Formation (ATF), obtained as reported in Ref. [22] starting
from the linearized Hill-Clohessy-Whiltshire (HCW) equations [23]. In
Ref. [1] two maneuvers have been investigated: 1) the PCF to GCF ma-
neuver and 2) the Along Track reconfiguration. ATF will not be further
analyzed in this paper, however the reverse GCF to PCF maneuver will be
studied.
In this work the planning problem is numerically solved by the

* Corresponding author.
E-mail address: dario.spiller@uniroma1.it (D. Spiller).

https://doi.org/10.1016/j.actaastro.2018.01.052
Received 20 September 2017; Received in revised form 23 January 2018; Accepted 29 January 2018
Available online 6 February 2018
0094-5765/© 2018 IAA. Published by Elsevier Ltd. All rights reserved.
D. Spiller et al.
Figure 1. Inertial and Local reference frames.
software presented in Ref. [1], which rely on the IPSO to optimize the
result. The IPSO is a fast and validated optimization method based on the
PSO [24]. It is particularly suitable for the addressed problem as it
eliminates a costly integration process during the non-linear attitude
dynamic simulation [1]. Being based on a differentially flat imple-
mentation [25,26], the IPSO requires a small number of optimization
parameters thus making it possible to deal with very complex optimi-
zation problems. Furthermore, satisfied boundary conditions for the
attitude are guaranteed, thus reducing the complexity of the performance
index. B-spline curves are used to model the attitude kinematics of the
satellites. Some little modifications have been introduced to improve the
reliability, the robustness and the convergence ability of the software.
The introduction of the gravity gradient (GG) torque within the Euler's
equation will prove the efficiency of the proposed maneuvers considering
a more realistic attitude dynamic model than the one used in Ref. [1].
The PSO has been implemented using the local strategy [27], which is a
simple and effective way to escape from local minima.
A key distinction of this approach with respect to using thrusters is
that the spacecraft is not limited to a finite total control input (via pro-
pellant) over the mission lifetime. As such, minimum time reconfigura-
tion maneuvers are a viable objective. Since perturbation forces have a
small intensity compared to the gravitational field forces, the minimum
time planning guarantees to exploit the effect of the perturbation forces
at the maximum extent and to obtain reasonable maneuver times.
The main objective of this paper is to show the limits and possibilities
offered by the proposed maneuver approach by means of:
1. Theoretical investigations: starting from simple mathematical models
(HCW and Gauss' Variational Equations (GVE) [28]), the relative
motion variations induced by the drag and the solar radiation pres-
sure will be investigated to understand how they affect the proposed
trajectories.
2. Numerical simulations: accurate numerical experiments will prove
the effectiveness of the theoretical insights.
It is noteworthy that in this paper the IPSO method is utilized to solve
a very hard problem which requires an efficient work-station in order to
obtain the solution in a reasonable time, in contrast to [18–20] where the
IPSO has been presented as an advantageous method for possible
on-board implementation of an autonomous path planning.
The rest of the paper is organized as follows. In Sec. 2 the mathe-
matical optimization problem is discussed, and in Sec. 3 the reference
relative trajectories are introduced using the Relative Orbit Elements
(ROE) [29] parameterization. Sec. 4 describes properties and charac-
teristics of the proposed perturbation-based reconfiguration maneuvers
and Sec. 5 gives the necessary information concerning the numerical
optimization technique used to validate the theoretical investigations.
Results are reported in Sec. 6 and conclusions are given in Sec. 7.
260
Acta Astronautica 153 (2018) 259–273
2. Problem definition
Given an inertial reference frame L ¼ fX L ⇔ Y L ⇔ Z L g, a body-
fixed reference frame B and NF satellites in the formation with i 2
f1; 2…; NF g (see Fig. 1), the addressed optimization problem may be
briefly summarized as:
Find : min J ¼ tf
 (1)
subject to; 8t 2 t0 ; tf ;
X
ng X
ng
Orbital Dynamics : r€i ¼ ϕj;k ðr i Þ þ p; ðr i Þ þ p⋆; ðr i Þ
j¼1 k¼1
þpD ðr i ; ξi ; tÞ þ pSRP ðr i ; ξi ; tÞ
Attitude Dynamics : ξ_ i ¼ Ξðξi Þ ωi
ω_ i ¼ I 1 ðui þ M GG ðr i ; ξi Þ  ωi  I ωi Þ
Initial conditions : bðXðt0 Þ; Yðt0 ÞÞ ¼ 0
    
Final conditions : e X tf ; Y tf ¼ 0
Path Constraints : pðX; Y; U; tÞ  0
The above quantities and relationships are defined as follows:
 X ¼ ½x1 ; x2 ; …; xNF  2 ℝ6NF where xi ¼ ½r i ; r_i ; r i 2 ℝ3 is the inertial
positions and r_i 2 ℝ3 is the inertial velocity of each satellite,
expressed in L .
 Y ¼ ½y1 ; y2 ; …; yNF  2 ℝ6NF where y i ¼ ½ξi ; ωi ; ξi 2 ℝ3 is the attitude
parameterization (given by roll, pitch and yaw) and ωi 2 ℝ3 the body
angular velocity, expressed in B .
 U ¼ ½u1 ; u2 ; …; uNF  2 ℝ3NF where ui 2 ℝ3 is the vector of the
external torque for controlling the attitude of each satellite.
 b defines the initial conditions such that b : ℝ6NF  ℝ6NF →ℝNb ; e de-
fines the final conditions such that e : ℝ6NF  ℝ6NF →ℝNe ; p defines
the path constraints such that p : ℝ6NF  ℝ6NF  ℝ3NF  ℝ→ℝNp .
The values of Nb ; Ne ; Np 2 ℝ depend on the problem. Initial and final
conditions determine the satellite configuration in t0 and tf . Inside the
path constraints, the collision avoidance (satellites inter-distance must be
greater than a user-defined threshold) and the attitude control feasibility
(the control must be lower the maximum available torque) are consid-
ered. The orbital dynamics takes into account the gravitational har-
monics ϕi;j up to degree and order ng ¼ 20 (values consistent with what
reported in Ref. [30]), the third body perturbation of the Sun and the
Moon, p; and , the drag pD and the solar radiation pressure pSRP
perturbations. Moreover the perturbative effect of the gravity gradient
(GG) torque M GG ðr; ξÞ is properly modeled.
3. Reference relative trajectories
In this work, circular and near circular orbits will be taken into ac-
count. Let us consider the chief satellite (subscripted with c) centered at
the origin of the rotating EulerHill frame L ¼ fX L ; Y L ; Z L g as in Fig. 1.
Some reference trajectories of the deputy satellite (subscripted with d and
identified by the position vector ρd ¼ ½xd ; yd ; zd ) are defined in L by
means of the linear model described by the Hill-Clohessy-Wiltshire
(HCW) equations [23],
x€d  2nc y_d  3n2c xd ¼ 0; y€d þ 2nc x_d ¼ 0; z€d þ n2c zd ¼ 0; (2)
1=2
where nc ¼ ðμ=a3c Þ , the mean motion, is supposed to be constant. For a
close relative trajectory, we impose the energy matching condition [22]
given by δa ¼ ad  ac ¼ 0. Let us define the Relative Orbit Elements
(ROE) [29],
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Figure 2. Reference relative trajectories.

other ROE are zero. Defining y0 ¼ a0 δλ, the formation is then

described by ½xðtÞ; yðtÞ; zðtÞ ¼ ½0; y0 ; 0.
2. General Circular Formation (GCF): the relative trajectory lies on a plane
and is circular with radius R, centered in the position of the chief
satellite.
3. Projected Circular Formation (PCF): the relative trajectory lies on a
plane, is elliptical and the yz projection is a circle, centered in the
position of the chief satellite and with radius R.

Close relative trajectories can be represented in Cartesian coordinates

in L as

xd ðtÞ ¼ 0:5 R sinðnc t þ α0 Þ ¼ ac δ e sinðnc t þ α0 Þ;

yd ðtÞ ¼ y0 þ R cosðnc t þ α0 Þ ¼ ac δλ þ 2ac δ e cosðnc t þ α0 Þ; (7)
zd ðtÞ ¼ 0:5 Kzx R sinðnc t þ α0 Þ ¼ ac δ i sinðnc t þ α0 Þ;

where R is related to the dimension of the formation ellipse and α0 is the

initial phase angle. Comparing Eq. (4) and Eq. (7), given that sinðαÞ ¼
Figure 3. Dimension and inclination of the relative motion ellipse. cosðα þ π =2Þ and cosðαÞ ¼ sinðα þ π =2Þ 8α 2 ℝ, the following equalities
have been established,
2 3 2 3 π
δa a1
c ðad  ac Þ φ ¼ α0 þ ; θ ¼ α0 : (8)
6 δλ 7 6 ud  uc þ ðΩd  Ωc Þcosðic Þ 7 2
6 7 6 7
6 δex 7 6 ed cosðωd Þ  ec cosðωc Þ 7 GCF, PCF and ATF are then obtained as the following special case:
δα ¼ 6 7 6 7
6 δey 7 ¼ 6 ed sinðωd Þ  ec sinðωc Þ 7; (3)
6 7 6 7 pffiffiffi
4 δix 5 4 id  ic 5
 GCF: Kzx ¼ 3 (centered in the origin of L if y0 ¼ 0).
δiy ðΩd  Ωc Þsinðic Þ  PCF: Kzx ¼ 2 (centered in the origin of L if y0 ¼ 0).
 ATF: R ¼ 0 and y0 6¼ 0.
where æ ¼ ½a; e; i; Ω; ω; M are the classical orbital elements and
u ¼ M þ ω. Accordingly, the trajectory generating from (2) can be
In the general case, the xy-projection of Eqs. (4) and (7) is a 2 : 1
expressed as
ellipse centered at ð0; y0 Þ (see Fig. 8 for clarity). With reference to Fig. 3,
xd ðtÞ ¼ ac δ e cos ðnc t  φÞ; yd ðtÞ ¼ ac δλ þ 2ac δ e sinðnc t  φÞ; zd ðtÞ the ellipse axes cy and czx are

¼ ac δ i sinðnc t  θÞ; (4) qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

1=2 1=2
Where δe ¼ ðδe2x þ δe2y Þ , δi ¼ ðδi2x þ δi2y Þ , φ ¼ tan1 ðδey =δex Þ and
1
θ ¼ tan ðδiy =δix Þ. Eq. (4) is used to define the geometrical constraints of
the initial and final configurations. Note that Eq. (4) is the same as Eq. (4)
in Ref. [1], here reported for the sake of completeness
xd ðtÞ ¼ ac δe cosðnc t  δωÞ;
yd ðtÞ ¼ ac ðδω þ δM þ cos i δΩÞ þ 2ac δ e sinðnc t  δωÞ;
zd ðtÞ ¼ ac ð  δΩ sin ic cos nc t þ δ i sin nc tÞ;
which was written as a function of the differential orbital elements
δæ ¼ ½δa; δe; δi; δΩ; δω; δM
and was valid for ec ¼ 0 or ec →0 and δωc →0 (the sign of δω in Eq. (4) of
Ref. [1] is  instead of þ, and δa is zero). With reference to Fig. 2, the
following three formations are defined:
1. Along-Track Formation (ATF): the satellites are on the same orbit with
small differences in anomaly. Consequently, only δλ 6¼ 0, while all the
cy ¼ 2R; czx ¼ R K 2zx þ 1 (9)
and the inclination of the relative motion plane is
φzx ¼ tan1 Kzx : (10)
The mathematical models for close relative motion given in Eqs. (4),
(5) and (7) are linear approximations valid when no perturbations are
(5) considered. Actually, more accurate models exist, as the one described in
Ref. [31] where the ROE formalism in employed to introduce the J2
gravitational contribution and the differential drag. However, employing
such models, the simple and intuitive closed form solutions presented so
far to define the PCF, GCF and ATF configurations cannot be easily
(6)
recognized.
When introducing non-linearities, we can only impose such reference
orbits at the t ¼ t0 . As described in Ref. [22], initial conditions on δa may
be modified to take into account the along-track drift introduced by the
J2 , which represents one of the most significant non-keplerian effects.

261
D. Spiller et al.
Figure 4. Satellite dimensions, surfaces and angles considered for the
perturbation forces.
4. Perturbations as control inputs
In this section, opportunity to perform perturbation-based maneuvers
are introduced. Sec. 4.1 formulates the mathematical description of the
drag and solar radiation perturbations. Sec. 4.2 describes a preliminary
feasibility analysis about the opportunity to use perturbations as control
inputs. Sec. 4.3 focuses on the general features of perturbation-based
maneuvers, and in Sec. 4.4 the characteristics of minimum-time ma-
neuvers with in-plane forces are described.
4.1. Perturbations introduced by natural forces
The orbital dynamics of the satellites are modeled as in Ref. [1]. From
the analysis reported in Ref. [32], the lift/drag ratio is related to the
thermal accommodation coefficient and attains a maximum value of 0.05
when a diffuse re-emission model is employed with complete accom-
modation. In more recent studies, as in Ref. [33], an accommodation
coefficient close to 1 is recommended for circular orbit below 500 km. As
a consequence, in this work it is assumed that the lift component of the
atmospheric force is negligible with respect to drag.
With reference to Fig. 4, where the body reference frame B ¼
fxB ; yB ; zB g is shown, the drag perturbation pD is given by Ref. [34].
ϱðr; tÞ CD X   The proposed maneuvering approach is based on the distinction be-
pD ¼ pD ðr; ξ; tÞ ¼  Sj cosαj ðξÞ v2 b
v (11)
2m j:αj <π
2
where CD is the drag coefficient, m is the mass of the satellite and
 
αj ðξÞ ¼ cos1 bs j ðξÞ⋅bv : (12)
The unit vectors bs j ðξÞ and b
v represent the normal to each jth surface Sj
of the satellite and the normalized velocity of the satellite, respectively.
The density ϱ in Eq. (11) is provided by the Jacchia-Roberts model [30],
which is a good compromise between precision and implementation
complexity. In fact, uncertainties in the density estimation still remain
also in more sophisticated models [35] thus not justifying the imple-
mentation of more complicated algorithms in this work. The acceleration
components due to a non-fixed atmosphere are neglected since they
would not significantly affect the results.
Similarly to the case of Eq. (12), the angle βj in Fig. 4 that each surface
makes with the solar radiation flux direction b r sat; is given by
 
βj ðξÞ ¼ cos1 bs j ðξÞ⋅b
r sat; (13)
where br sat; is a unit vector directed from the satellite to the Sun.
Considering an ideal reflecting surface, the SRP perturbation pSRP is
262
Acta Astronautica 153 (2018) 259–273
modeled as [34].
 
S 0 r ;  X    
pSRP ¼ pSRP ðr; ξ; tÞ ¼ 2ζðtÞγ   Sj cos2 βj bs j (14)
mc r ;  r  j:β <π
j 2
where γ is the reflectivity coefficient, S 0 ¼ 1352:098 kg=s2 , c ¼
2:988⋅105 km/s. The shadow function ζðtÞ is equal to 1 when the satellite
is outside the Earth shadow cone, equal to 0 when it is inside and equal to
0.5 in penumbra. As reported in Ref. [34], the solar radiation pressure
contribution is not as strong as the drag contribution below altitudes of
about 800 km.
Differential accelerations are defined as vectorial differences between
perturbation accelerations acting on two distinct spacecrafts in the for-
mation. Hence, DD acceleration aDD and Differential Solar Radiation
Pressure (DSRP) acceleration aDSRP acting on the i-th satellite with
respect to the j-th satellite are
ði;jÞ   ði;jÞ  
aDD ¼ pD ðr i ; ξi ; tÞ  pD r j ; ξj ; t ; aDSRP ¼ pSRP ðr i ; ξi ; tÞ  pSRP r j ; ξj ; t ;
(15)
where i; j 2 f1; …; NF g, i 6¼ j. Perturbation accelerations may be used as
control inputs for the SFF reconfiguration illustrated by the well known
GVE [34]. Looking at Eq. (11), it can be seen that the drag force is
directed along  b v which may be confused with YL for very small ec-
centricities. As a consequence, the drag force may only be used to modify
a, e, ω and M. With the hypothesis of negligible lift, the drag force is a
dissipative force and, looking at Eqs. (3) and (4), only the in-plane x and y
components may be controlled (i.e., δa, δλ, δex and δey ). Conversely, the
SRP has both in-plane and out-of-plane effects, as stated in Eq. (14),
giving the possibility to vary all the orbital parameters. However, the
intensity of the force depends on the angle βðξÞ. In this case, a key role is
played by the orientation of the chief orbital plane, as it can vary the
intensity of the in-plane and out-of-plane components of the SRP
depending on the position of the Sun in L . In fact, for an equatorial orbit
the maximum value of the SRP is attained when β ¼ 23∘ 270 , i.e. when the
force is in the Ecliptic plane, thus having the in-plane component much
bigger than the out-of-plane component. Conversely, in a polar orbit the
maximum attainable magnitudes of in-plane and out-of-plane compo-
nents are switched.
4.2. Preliminary feasibility analysis of perturbation-based maneuvers
tween absolute perturbations and differential perturbations. The former
are those forces that are common to all the spacecrafts in the formation
and only depend on satellite position and environment (e.g. gravitational
harmonics) whereas the latter are forces that can be modified with regard
to magnitude and/or direction. Absolute perturbations can make unsta-
ble the PCF, GCF and ATF configurations or impact on the guidance while
performing a maneuver. For instance, the J2 perturbation sensibly
modifies the cross-track relative motion and the along-track separation if
the inclinations of the orbits are different, as shown in Ref. [31]. Atmo-
spheric drag and solar radiation pressure are differential perturbations
that can be used to perform reconfiguration maneuvers only if they can
counteract the other orbital perturbations affecting the formation.
An interesting graph comparing the magnitudes of all the perturba-
tion for all the relevant orbital regimes is reported in Ref. [30]. In LEO,
the most dominant perturbation is due to the J2 . A preliminary analysis of
the intensity of the gravity gradient acceleration rg J2 due to J2 can be
obtained using the linear approximation valid for near-circular orbits
reported in Ref. [22],
rg J2 ¼ κΛρd (16)
where
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Table 1 of ϱHP are

Reference density values from the Harris-Priester model (from Ref. [34]).
     
ðminÞ ðmaxÞ  ðmaxÞ   ðminÞ 
Altitude Minimum density, ϱHP Maximum density, ϱHP maxðkaDD kÞ ¼ aDD ϱHP ; minðkaDD kÞ ¼ aDD ϱHP : (21)
(km) (kg/m3) (kg/m3)
In Fig. 5, a comparison of the magnitudes of J2 gradient acceleration
200 2:557  1010 3:162  1010 and DD perturbations is shown. The orbital inclinations ic and the
300 1:708  1011 3:526  1011
exposed area ΔS have been considered as parameters. As can be seen, the
400 2:249  1012 7:492  1012
500 3:916  1013 2:042  1012
graph suggests that up to 400–500 km the magnitude of the DD pertur-
600 8:070  1014 6:390  1013 bation is bigger or similar to the magnitude of rg J2 (up to kρd k ¼ 1 km).
700 2:043  1014 2:185  1013 Therefore, it is expected that reconfiguration maneuvers can be accom-
800 7:069  1015 8:059  1014 plished in low LEOs. At higher LEOs, the differential drag is less than
rg J2 , suggesting that reconfiguration maneuvers cannot be performed
unless very large (and unreliable) exposed areas are considered. The
opportunity to perform DD-based reconfiguration maneuvers increases
2 3 with high values of ΔS (i.e., no maneuvers can be accomplished with
1  3s2ic s2θc s2ic s2θc s2ic sθc
6
7 near-spherical satellites). Considering the exposed areas of the satellite
6 72 1 1 1 7
6 s2ic s2θc s2ic sθc    s2ic cθc 7 model in Fig. 4, maneuvers can be reasonably performed up to
Λ¼6
6 4 2 4 4 7; κ
7 400–500 km.
6
7
4 1 5 1 3 5 Similar analysis can be performed in MEO and GEO, where other
s2ic sθc  s2ic cθc s2ic s2θc þ 
4 4 2 4 perturbations such as the J22 and the third body perturbation (from Sun

and Moon) must be considered. However, such preliminary analyses can
μR2e
¼ 6J2 ; sð⋅Þ ¼ sinð⋅Þ cð⋅Þ ¼ cosð⋅Þ: (17) only give an approximated order of magnitude for the required exposed
rc5
area, since the necessary model simplifications can affect the results. For
In Eq. (17), θc is the true anomaly of the chief spacecraft and example, the intensity of the J2 perturbation varies with the latitude, and
J2 ¼ 1:082  103 . An upper bound of the magnitude of rg J2 can be the in-plane and out-of-plane contributions can have helpful or disturb-
obtained employing the property (valid for every general or induced ing effects in different points of the orbit. As a consequence, for some
norm [36]) orbital regimes reconfiguration maneuvers might be accomplished even

kΛρd k  kΛkkρd k: (18)

Considering ρd ¼ 1 km and using the 2-norm to evaluate jjΛjj, the
maximum magnitude of the J2 gradient acceleration can be found
considering θc 2 ½0; 2π  for each inclination, i.e.
   
max rg J2  ¼ max rg J2 ðic ; rc Þ ¼ κðrc Þ max jjΛðic ; θc Þjj2 : (19)
ρd ¼1 km θc 2½0;2π 

A preliminary estimation of DD experienced by the formation can be

obtained considering CD ¼ 2:2 and m ¼ 100 kg for all the spacecrafts,
same orbital velocities (in magnitude and direction) and minimum/
maximum values of the atmospheric density from the Harris-Priester
ðminÞ ðmaxÞ
model, ϱHP and ϱHP , shown in Table 1 (same density for both the
spacecrafts). With these hypotheses, an upper bound of the DD magni-
tude can be estimated as

ϱHP CD 2
jjaDD jj ¼ v ΔS; (20)
2m

where ΔS is the difference between maximum and minimum exposed Figure 6. Increasing of the relative trajectory dimension with in-plane and
areas. Moreover, maximum and minimum values of kaDD k as a function out-of-plane forces.

Figure 5. Comparison between J2 gravity gradient

acceleration (kρd k ¼ 1 km) and DD.

263
D. Spiller et al.
Figure 7. Variation of the relative trajectory inclination with drag and SRP.
Figure 8. Projection of PCF and GCF with same value of zd;max .
if the average DD magnitude is lower than the average magnitude of the
J2 gradient acceleration. The proposed numerical approach is then a
useful instrument when a more reliable feasibility analysis is required to
estimate the opportunity to perform perturbation-based reconfigura-
tions.
4.3. Insights on perturbation-based maneuvers
For close relative trajectories described by Eq. (7), a relationship
between trajectory parameters R; Kzx and ROE can be found as
zd δi
R ¼ 2ac δe; Kzx ¼ ¼ : (22)
xd δe
Variations of R influence all the components of the deputy motion but
do not change the plane of the relative motion, i.e. the ratio zd =xd or φzx in
Fig. 3. From Eq. (22), a variation of R can be induced by a variation of ac
(also ad must change to preserve the energy matching condition) or a
variation of δe. The inclination of the relative motion plane is dictated by
the value of Kzx . However, Kzx also influences the dimension of the
relative trajectory as it modifies the ellipse axis czx as stated in Eq. (9),
stretching the trajectory along the Z L direction. As a consequence, both R
and Kzx can affect, in different ways, the dimension of the relative ellipse.
As a result of the GVE [28], out-of-plane forces can only modify i and Ω,
in-plane forces can modify a, e and M while ω can change with a force
264
Acta Astronautica 153 (2018) 259–273
with any direction. Since the ROE set is a combination of the orbital el-
ements, effects of perturbations on the relative motion change depending
on the direction of the perturbation forces.
With reference to Fig. 6, and considering out-of-plane forces, the
following maneuvers can be accomplished:
 The dimension of the relative motion ellipse can be increased, as
shown in Fig. 6(a), applying out-of-plane forces to the deputies. In this
way, the relative inclination is modified and the value of Kzx changes.
Accordingly, with reference to Eq. (9), the trajectory geometry is
modified as czx is affected.
 The inclinations of the deputy satellites can be modified, increasing or
decreasing the term aδi in the zd component of Eq. (4). Affecting the
term δi, this maneuver modifies Kzx (and φzx ) as can be seen from Eq.
(22) and as shown in Fig. 6(a).
In contrast to the previous case, using only along-track forces, the
following maneuvers can be accomplished:
 The dimension of the relative motion ellipse may be increased with
in-plane forces directed along Y L and applied to all the satellites, as
shown in Fig. 6(b). In this way ac is increased, as well as the semi-
major axes of the deputy satellites, enlarging the dimension of the
relative motion ellipse. However, as opposed to the previous point,
this maneuver does not modify the geometry of the relative ellipse, as
both the ellipse axes are modified by the same amount (see Eq. (9)).
 Variations of the relative motion plane inclination can be obtained
with in-plane forces. In fact, a variation of the angle φzx is obtained
when the orbital parameter e (dictating the form of the absolute orbit)
is changed. Such a maneuver affects δe and Kzx as stated in Eq. (22).
With regard to drag (only in-plane component) and solar radiation
pressure (both in-plane and out-of-plane components), a summary of the
maneuvering opportunities given by these perturbations is given below:
 As a consequence of the previous points, the drag force cannot in-
crease R with constant Kzx since it is an in-plane force directed along
Y L , opposite to the velocity direction. However, it may change the
inclination of the relative motion plane, as shown in Fig. 7(a). All the
formation satellites will decrease their value of the semi-major axis
because of the dissipative effect of the drag. As stated in Sec. 4.1, the
out-of-plane lift component is neglected.
 The SRP force might be used to increase the dimension of the relative
motion ellipse, as an out-of-plane component can be obtained
depending on the orbit geometry. Consequently, maneuvers as in
Fig. 7(b) can be addressed with the SRP. Taking advantage of the in-
plane and out-of-plane components of SRP, both R and Kzx can be
properly controlled. In the results section, the SRP will be used to
change Kzx and R modifying both the dimension and the inclination of
relative trajectory. If the maneuver is properly planned, the final
value of δe can be set to be equal to the initial value in order to modify
R keeping Kzx constant. With the SRP, it is not expected to denote a
semi-major axis decrease as in the case of the drag-induced maneuver.
4.4. Properties of minimum-time in-plane maneuvers
Let us consider maneuvers based on in-plane forces. Reconfiguration
from PCF to GCF, or vice-versa, are considered. With reference to Fig. 8,
let zd;max remain fixed during the reconfiguration maneuver, i.e.
maxðzd;PCF Þ ¼ maxðzd;GCF Þ: (23)
This hypothesis is almost true for short-time maneuvers lasting about
one orbit, and its validity will be verified by the reported simulations.
The maximum z displacement for the PCF and GCF cases is given by
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Figure 9. General scheme of the employed optimi-

zation process.

maxðzd;PCF Þ ¼ maxðRPCF sinðnc t þ α0 ÞÞ ¼ RPCF ; (24) u ¼ Iω

_ þ ω  I ω  M GG ðξ; rÞ: (29)

pffiffiffi
pffiffiffi Once the attitude history is known along with its time derivatives,
3 3 both the required torques and the angular velocity can be expressed in
maxðzd;GCF Þ ¼ max RGCF sinðnc t þ α0 Þ ¼ RGCF : (25)
2 2 closed form. The most important consequences of the inverse approach
Hence, considering a GCF to PCF, Eq. (23) leads to are that: 1) the integration of the attitude dynamics is avoided, and 2)
that the initial and final conditions of the attitude kinematics are imposed
pffiffiffi
  3 a priori. The attitude reorientation subproblem is shown in Fig. 9 by
RPCF tf ¼ RGCF ðt0 Þ 0:866 RGCF ðt0 Þ: (26) means of the first three blocks. The attitude histories of the formation
2
satellites are used to compute pD ðξÞ and pSRP ðξÞ. Once integrated the
Conversely, for the opposite PCF to GCF maneuver,
dynamics, the imposed constraints are calculated and the achieved for-
  2 mation reconfiguration is compared with the required one. Finally, IPSO
RGCF tf ¼ pffiffiffiRPCF ðt0 Þ 1:155 RPCF ðt0 Þ: (27) searches for the optimal solution which minimizes the maneuver time
3
satisfying the imposed constraints stated in Sec. 2.
It is noteworthy that for the GCF to PCF maneuver the dimensions of
the 2:1 ellipse in the xy plane have to be reduced. In the opposite case,
5.2. Attitude approximation with B-splines
the dimensions of the same ellipse must be increased.
When ξðtÞ 2 ℝ (i.e., one degree of freedom), we define the optimi-
5. Optimization strategy based on particle swarm optimization
zation parameters in the IPSO particles as

The optimization problem is solved using the IPSO algorithm h T iT

described in Ref. [1] (only minor modifications have been included
which do not modify the main original features). Hence, this section is
intended to describe the most important characteristics of IPSO. For
further details, refer to [1].
5.1. Inverse dynamics particle swarm optimization
The optimizer is based on the Particle Swarm Optimization (PSO), a
meta-heuristic optimization method based on the evolution of a fixed
number of solutions called particles [37]. The movement of the particle
within the feasible search space is dictated by a perturbation term named
velocity, that is a change in position. In this work, to avoid the risk of
converging to a local minima the local model of the PSO [27] is used to
evaluate the velocity.
The whole optimization process is depicted in the block diagram
shown in Fig. 9. The IPSO is based on a combination of PSO and the
inverse dynamics approach. The method relies on the flatness property
[25] typical of the attitude dynamics expressed by the Euler's equation
[26]. Accordingly, in this work the attitude reorientation problem has
been stated as an inverse dynamics sub-problem inside the generic
problem in Eq. (1). Since the angular velocity may be expressed as a
X ¼ ~ξ1 ; ~τ T1 ; …; ~ξNF ; ~τ TNF ; tf 2 ℝ2NF NP þ1 :
T
(30)
The vectors ~ξi and ~τ Ti contains the optimization parameters (NP for
T
each vector) defining the B-spline curve for the attitude maneuvers of the
satellites. When 2 or 3 attitude parameters for each satellite are consid-
ered, the number of independent optimization variables is 3NF NP þ 1
and 4NF NP þ 1, respectively (the time parameters vector is one for each
satellite).
B-splines are used taking full advantage of their capability to shape
curves [1,19–21]: accordingly, both the kinematic parameters (i.e. roll,
pitch and yaw) and the time are approximated with B-spline polynomials
using the coefficients in the ~ξ and ~τ T vectors, respectively (see Ref. [1]
T
i i
for further details). The B-spline approximation is defined upon a strictly
increasing independent variable 0  λ  1 and depends on the knot
vector K defined as
K ¼ fκ i ; i ¼ 0; :::; mjκ 0 ¼ 0; κ m ¼ 1; κ i  κ iþ1 g (31)
where m is defined by the user. The generic component ξ of the attitude
parameters vector can be approximated with a B-spline B ðλ; ~ξ; k Þ as
  NX
P 1

function of the attitude parameters, i.e. ω ¼ gðξ; ξ;_ €ξÞ, then in the body ξN ðλÞ ¼ B λ; ~ξ; k ¼ ~ξi N i;D ðλ; K Þ; (32)
reference frame B the sum of the external torques ΣM ext are related to i¼0

the kinematic by the Euler equation,

where the basis functions N i;d ðλ; KÞ are defined by the Cox-de Boor
  recursion formula (see Ref. [38]). To build a B-spline curve, a polyline
ΣM ext ¼ Iω _ €ξ :
_ þ ω  I ω ¼ f ξ; ξ; (28)
defined over the control points Uj ¼ ½~τj tf ; ~ξj ; j ¼ 0; …; NP  1; is intro-
Using the optimization algorithm of [1], the quantities ξ; _ €ξ are not duced and the time is
independent as they are analytically evaluated once the components of ξ
NX
P 1
are approximated with B-spline curves. Note that the GG torque may be
tN ðλÞ ¼ tf B ðλ; ~τ; K Þ ¼ tf ~τi N i;D ðλ; K Þ; (33)
taken into account after the dynamics integration. In fact, the required i¼0
control torque u is given as

265
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Table 2
where Nt is the number of time intervals where the constraint is evalu-
Satellites parameters.
ated, tj is a time instant, π i is a user-defined weight and
Parameter Value Parameter Value
 ðiÞ  
Mbody (kg) 90 Ix;B (kg⋅m2) 1:28e3 ðiÞ   0 if pk tj < Δi
pk tj ¼ (37)
Mpanel (kg) 40 Iy;B (kg⋅m2) 6:41e1 1 otherwise
CD 2.2 Iz;B (kg⋅m2) 1:23e3
γ 1.5 Mmax (Nm) 1e3 The parameter Δi is a decreasing tolerance, following the decreasing
scheme reported in Ref. [1]. As an example, the final position and ve-
locities of the deputies are imposed considering an additional orbital
Table 3 propagation for an orbital period after the execution of the reconfigu-
Initial conditions for the PCF-GCF maneuver. ration maneuver, similarly to what has been done in Ref. [39]. If the
r L ðt0 Þ r_L ðt0 Þ maneuver is consistent with the requirements, the deputies should have
position and velocity histories generating from the final conditions of the
x [km] y [km] z [km] x_ [km/s] y_ [km/s] z_ [km/s]
integration, i.e. ρðtÞ and ρ_ ðtÞ, consistent with the ones corresponding to
PCF to GCF maneuver the required configuration, ρ
ðtÞ and ρ_
ðtÞ. The comparison is carried out
SAT 1 0 0 0 0 0 0
considering the linearized dynamics of Eq. (2). As a consequence,
SAT 2 0 1 0 0.5n 0 n pffiffiffiffiffiffiffiffiffiffi
SAT 3 0.433 0.500 0.866 0.25n 0.866n 0.5n considering ti 2 ½tf ; tf þ T with T ¼ 2π a3 =μ, we define
SAT 4 0.433 0.500 0.866 0.25n 0.866n 0.5n
8     
ρ t ρ
d;k tj 
GCF to PCF maneuver >
<0  d;k j 
ðposÞ   if    < Δpos
SAT 1 0 0 0 0 0 0 pk tj ¼  R R  (38)
SAT 2 0 1 0 0.5n 0 0.866n >
:
SAT 3 0.433 0.500 0.750 0.25n 0.866n 0.433n 1 otherwise
SAT 4 0.433 0.500 0.750 0.25n 0.866n 0.433n
8     
ρ_ t ρ_
d;k tj 
>
<0  d;k j
ðvelÞ   if 
 nc R
  < Δvel
pk tj ¼ nc R  (39)
where ~τ ¼ ½~τi;1 ¼ 0; ~τi;2 > ~τi;1 ; …; ~τi;NP1 ¼ 1. Hence, the independent >
:
variable used for the attitude description is λ and the attitude time- 1 otherwise
history is given by the two-dimensional graph
 N   where R and nc are average radius and mean motion of the obtained
γ y ðλÞ ¼ t ðλÞ; ξN ðλÞ ; λ 2 ½0; 1 : (34) trajectory.
Time derivatives of ξN can be evaluated as reported in Ref. [1].
6. Numerical results

5.3. Performance index definition
The performance index, the penalty functions and the decreasing
tolerance technique are imposed following the criterion reported in
Ref. [1]. The performance index J is designed as
X
Ncon
J ¼ J0 þ Pi ;
i¼1
where the terms P i are the penalty functions associated to the constraints
described in Sec. 2 and evaluated as described in Sec. III.C of [1].
The term J0 is designed in slightly different ways depending if the
maneuver is performed with the drag or the SRP. In the former case,
J0 ¼ tf . With the drag, no out-of-plane force is detected and the results
are expected to confirm the speculations of Sec. 4.4 with regard to the
value of Rðtf Þ. On the other hand, with the SRP the out-of-plane
component can perturb the final result, influencing the value of the tra-
jectory radius R. To choose a specific strategy consistent with the
 
 
minimum-time maneuver we are searching for, J0 ¼ tf þ κ Rðtf Þ  R
f ,
where R
f assumes the values outlined in Sec. 4.4. Without the term in R,
different maneuvers may be found with different values of Rðtf Þ, hence
the proposed strategy allows to identify a unique maneuver. The same
term may be used for another kind of maneuver, i.e. for imposing other
values of the final radius of the relative trajectory. For instance, a GCF to
GCF imposing Rðtf Þ > Rðt0 Þ will be presented with this regard. Interme-
diate cases can be searched for with combinations of the two strategies,
which means that the SRP may be used to obtain slight variation of both
Kzx and R.
With regard to the penalty functions P i , they are evaluated as
X
Nt X
NF
ðiÞ  
P i ¼ πi pk tj
j¼0 k¼1
In this section some test cases are solved with the proposed technique.
Such results extend the findings reported in Ref. [1], as more precise
results are shown for difficult reconfiguration maneuvers. All the simu-
lations have been carried out on a personal computer with an Intel®-
processor CoreTM i7-2670QM CPU @2.20 GHz and with 6.00 GB of
RAM.
(35)
6.1. Parameters and problem setting
A SFF with NF ¼ 4 satellites is considered. Three different maneuvers
are taken into account:
1. Modify a PCF into a GCF.
2. Modify a GCF into a PCF.
3. Increase the radius of a close relative formation.
From the hypotheses and the analyses made in Sec. 4, case 3. can only
be attempted using the SRP perturbation. In fact, the lift component of
the atmospheric force is neglected and the drag is a dissipative force.
IPSO parameters have been chosen accordingly to [1], with the
exception of some parameters that have been slightly modified (in
particular, the maximum number of iterations has been increased to
achieve higher accuracies and the local model has been employed instead
of the unified model). The swarm is set to have NS ¼ 30 particles and the
B-splines of degree 5 are defined with NP ¼ 10 points. Weights and tol-
erances of the cost functions are given in Ref. [1].
Results will be reported considering near-equatorial circular orbits in
LEO and GEO and inclined circular orbits in MEO (the chief orbital pa-
rameters will be detailed in the following subsections). With regard to the
LEO case, results will be reported considering a single-axis maneuver
since we have only one degree of freedom as the drag force always points
(36)
in the opposite direction of the satellites velocity vector. However,
instead of a yaw maneuver around zB as in Ref. [1], here the maneuver is

266
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Table 4 around yB since the moment of inertia around this axis is lower than the
Chief orbital parameters in LEO. one around zB (see Table 2). The result of the maneuver is not affected as
Parameter Value Parameter Value the area exposed to the drag varies in a similar manner as in Ref. [1], but
a (km) 6778 ω (deg) 0 the control effort is reduced and a smaller value of the maximum torque
e 0 i (deg) 1 can be chosen. When dealing with the solar radiation pressure in MEO
Ω (deg) 10 M (deg) 0 and GEO, a two-axes maneuver around yB and zB is considered, as a
rotation around xB is useless in order to vary the exposed area. With
NF ¼ 4 and NP ¼ 10 and referring to Eq. (30), the number of optimi-
zation variables is 81 of the one-axis maneuver and 121 for the two-axes
maneuver.
Table 5 All the relevant physical and geometrical properties of the four sat-
Maneuver performances achievable in LEO with drag. ellites are shown in Fig. 4 and in Table 2. The inertia tensor referred to
Kzx ðt0 Þ Rðt0 Þ Kzx ðtf Þ tf Rðtf Þ the Body reference frame B of Fig. 4 is evaluated supposing a uniformly
Rðt0 Þ
distributed mass over the central body and the solar panels.
() (km) () (orbits) ()
Initial conditions for the PCF to GCF and GCF to PCF cases are re-
all deputies SAT2 SAT3 SAT4 all deputies ported in Table 3 expressed in L (velocities are a function of the mean
PCF to GCF 2.000 1.000 1.731 1.735 1.735 1.292 1.154 motion, varying among LEO, MEO and GEO). PCF and GCF are consid-
GCF to PCF 1.732 1.000 2.005 1.995 1.994 1.496 0.866 ered with a central satellite (SAT1) and three deputies: SAT2 is aligned
along the inertial velocity direction of the chief and SAT3 and SAT4 are
equally spaced in the circular trajectory. The simulation epoch begins at
the date January 2000, 00:00 a.m. UTC. In LEO, initial and final attitudes
are chosen with xB aligned with yL and yB normal to the orbital plane.
In MEO and GEO, instead, xB is aligned with the Sun-satellite direction
and yB is normal to the orbital plane. The attitude maneuvers in the LEO
scenario are referred to the LVLH reference system L , while in the MEO
and GEO scenarios they are referred to the ECI system L . In all cases,
given the geometrical and structural symmetry of the satellite model,
rotations are limited in ½90; 90 deg per axis.
The orbital integration of the satellites is carried out using a Gauss-
Jackson scheme [40]. As shown in Ref. [30], this method is faster than
standard Runge-Kutta schemes, guarantees great numerical accuracy
and, as a fixed step-size integrator, it allows a simple implementation of
the input attitude kinematics history inside the differential equation to be
integrated. Moreover, the Parallel Computing utility available in Matlab®
has been used to make the numerical computation more efficient. The
Gauss-Jackson integration is carried out with a 60-s step and relative and
absolute tolerances equal to 1013 and 1015 , respectively. IPSO stops
according to the following convergence criterion,

1 X k
Jgði1Þ  JgðiÞ
δJg ¼ <ε (40)
10 i¼k9 ðiÞ
Jg

where Jg is the performance index of the best particle in the swarm and
ε ¼ 1010 . A maximum number of 3000 iterations has been imposed in
order to reasonably limit the required computational time.
Figure 10. GCF to PCF maneuver in the LVLH Coordinate System L in LEO.
6.2. Case 1: reconfiguration in LEO

The chief orbital parameters are reported in Table 4. In Ref. [1], it has
been shown that the higher LEOs do not allow to obtain good results

Figure 11. GCF to PCF maneuver in the LVLH Co-

ordinate System L , xy projection in LEO.

267
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Figure 12. Exposed area (a) and z-axis torque (b) for
the GCF to PCF case in LEO.

Figure 13. Drag perturbation history for the GCF to

PCF case in LEO.

Figure 14. ROE history for the GCF to PCF case in

LEO (ROE components have been translated to zoom
all the deputies’ behavior).

268
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Table 6 Table 7
Chief orbital parameters in MEO. Maneuver performances achievable with SRP in MEO.

Parameter Value Parameter Value Kzx ðt0 Þ Rðt0 Þ Kzx ðtf Þ tf Rðtf Þ
Rðt0 Þ

a (km) 26378 ω (deg) 0

() (km) () (orbits) ()
e 0 i (deg) 60
fΩ1 ; Ω2 g (deg) f280; 10g M (deg) 0 all deputies SAT2 SAT3 SAT4 all deputies

i ¼ 60 deg, line of nodes along Earth-Sun direction

PCF to GCF 2.000 1.000 1.723 1.749 1.745 1.400 1.146
GCF to PCF 1.732 1.000 2.003 2.001 2.002 1.211 0.872

i ¼ 60 deg, line of nodes normal to Earth-Sun direction

PCF to GCF 2.000 1.000 1.753 1.733 1.732 1.015 1.138
GCF to PCF 1.732 1.000 1.999 2.000 2.000 1.008 0.874

orbit propagation after the end of the maneuver. The trajectory is

Figure 15. PCF to GCF maneuver in the L Reference System in MEO with Ω1 .
when considering PCF to GCF maneuvers (see Sec. 4.2 for further
details).
First of all, let us look at Table 5 to understand the performances
achievable with the drag perturbation. As can be seen, both the PCF to
GCF and the GCF to PCF maneuvers may be accomplished with very good
results. In both cases the time required for the maneuver is between one
and one and a half orbits. The inclination of the relative orbit plane have
been changed with great accuracy. Moreover, accordingly to the theo-
retical prediction stated in Sec. 4.4, the ratio between final GCF radius
and the initial PCF radius is 1.154 (instead of 1.155) and the ratio be-
tween final PCF radius and the initial GCF radius is 0.866, as expected.
This means that the maneuver time can be considered sufficiently short
such that the hypothesis in Sec. 4.4 hold.
To get a greater insight into the proposed maneuver, we report some
explanatory figures concerning the GCF to PCF maneuver (similar results
are obtained vice-versa). In Fig. 10, the maneuver trajectory is shown in
the L frame. The stars represent the initial positions of the satellites,
while the circles are the final positions. The dashed lines are the optimal
trajectories followed by the satellites and the solid lines represent one
modified starting from a GCF (which is the outer trajectory) and arriving
to a PCF (which is the inner trajectory) and the extremal values along the
z direction are the kept constant (which also means that the radius of the
GCF is quite the same as the semi-major axis of the finale relative ellipse
given by the PCF). The angular distances among the satellites are not
modified in a relevant way.
Fig. 11 reports the xy projection (in L ) of the maneuver proposed in
Fig. 10. The three deputy satellites start the transfer trajectory from the
outside ellipse and finish their maneuver when they arrive on the inner
ellipse. According to Eq. (7), the xy projection is always an ellipse with a
y:x ratio equal to 2:1.
The exposed area histories (due to the satellites' rotations about their
y-axis) are reported in Fig. 12(a). Given the symmetry of the considered
satellite model, symmetrical reorientation attitude maneuvers are
possible leading to the same reconfiguration results. Moreover, it should
be noted that such attitude maneuvers are consistent with the maximum
torque constraint, as can be seen from Fig. 12(b) where the normalized
torques are reported. Therefore, it is numerically verified that the
disturbance effect of the GG torque does not affect the results, validating
the results in this paper and in Ref. [1].
One very interesting result reported in Fig. 12 is that the exposed
areas (as well as the rotations, not reported for brevity) are likely to take
only two values, i.e. the maximum exposed area or the minimum exposed
area. This is particularly consistent with a minimum-time maneuver: if
we were looking at the linearized model of relative motion with the HCW
equations, it can be easily seen that reconfiguration maneuvers require a
bang-bang control. Accordingly, the fact that the exposed area, which is
proportional to the drag force, is mainly either minimum or maximum,
suggests the near-optimality of the proposed solution. As a consequence
of these attitude maneuvers, the drag force experienced by the satellites
is reported in Fig. 13, where the force components are referred to L . It
can be seen that the order of magnitude of the drag force as well of the
differential drag force is 108 km/s2.
Finally, it is noteworthy, looking at the ROE reported in Fig. 14, the
drag force only affects the in-plane elements, that is δa, δλ, δex and δey

Figure 16. PCF to GCF maneuver in the L Reference

System, xy projection in MEO with Ω1 .

269
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Figure 17. SRP perturbation history for the PCF to

GCF case in MEO with Ω1 .

Figure 18. ROE history for the PCF to GCF case in

MEO with Ω1 (ROE components have been translated
to zoom all the deputies’ behavior).

Figure 19. Rotation angles (a) and exposed areas (b)

for the PCF to GCF case in MEO with Ω1 .

270
D. Spiller et al. Acta Astronautica 153 (2018) 259–273

Table 8 which is consistent with the GVE and the fact that the drag is in the di-
Chief orbital parameters in GEO. rection of the velocity. The elements δix and δiy show periodic variations
Parameter Value Parameter Value due to the other conservative perturbations. It is noteworthy that δa and
a (km) 42168 ω (deg) 0 δλ converges to a zero final value, which is consistent with the expected
e 0 i (deg) 1 results (see Eq. (4)).
Ω (deg) 10 M (deg) 0

6.3. Case 2: reconfiguration in MEO

Table 9
Maneuver performances achievable with SRP in GEO.
The chief orbital parameters for the MEO cases are reported in
Table 6. The chief inclination has been set to 60 deg (value consistent
Kzx ðt0 Þ Rðt0 Þ Kzx ðtf Þ tf Rðtf Þ
Rðt0 Þ with real MEO missions such as the GNSS constellations) to get the op-
(-) (km) (-) (orbits) () portunity to obtain in-plane and out-of-plane components of the SRP with
similar magnitudes. With this regard, two different values of ascending
all deputies SAT2 SAT3 SAT4 all deputies
node have been chosen. With Ω1 , the line of nodes of the chief orbital
PCF-GCF maneuvers plane is aligned to the Earth-Sun direction, hence the maximum
PCF to GCF 2.000 1.000 1.728 1.729 1.716 0.960 1.139
GCF to PCF 1.732 1.000 1.963 1.979 1.984 0.936 0.862
component of the SRP is in plane. On the contrary, with Ω2 the line of
nodes is normal to the Earth-Sun direction and consequently the out-of-
GCF to GCF maneuver imposing R
f ¼ 1:2 km
plane component of the SRP is comparable with the in-plane one.
GCF to GCF 1.732 1.000 1.712 1.714 1.710 1.048 1.194
A PCF to GCF maneuver performed with Ω1 is depicted in Fig. 15 and
Fig. 16. In this case we are moving from the inner relative trajectory to
the outer one and the figures allow the reader to appreciate the ability to
perform the maneuver with the SRP. From Fig. 15 it can be see that the
maximum and minimum values along zL remain quite constant satisfying
the constraint imposed in the performance index.
Results reported in Table 7 for Ω1 and Ω2 show that in both the cases a
good reconfiguration maneuvers may be achieved. The Kzx coefficients
change achieving nice approximations of the expected values. The final
radius is quite close to the one expected when using only the in-plane
forces. Moreover, the time required for the cases with Ω1 is greater
than the time required when using Ω2 . This is consistent with the fact that
when the line of nodes is aligned with the Earth-Sun direction, the SRP
goes to zero when the satellites are in the Earth shadow, as can be seen in
Fig. 17. From this figure, moreover, the magnitude of the SRP can be
detected and compared to the drag intensity reported in Fig. 13. Even
though the SRP magnitude is less than the one provided by the drag the
maneuver can be accomplished since the gravitational field intensity in
MEO is less than in LEO.
The out-of-plane effect of the SRP on the reconfiguration maneuver
can be seen in Fig. 18, where all the six parameters are modified. It is
noteworthy that the variations in δix and δiy are no more periodical as
they were in Fig. 14 where their histories where induced by the con-
servative perturbations only. However, also in this case δa and δλ con-
verges to a zero final value, which is consistent with the expected results.
Finally, the same characteristics described in LEO with regard to
rotation angles and exposed area are detected in MEO, see Fig. 19. Also in
this case the rotation angles histories are such that the exposed area tends
Figure 20. GCF to GCF maneuver in the L Reference System in GEO with to have minimum or maximal values, thus representing a nice approxi-
R
f ¼ 1:2 km. mation of a bang-bang solution. For brevity, the torques required for the
attitude maneuvers are not shown. However, the necessary torques in
MEO are always lower than the ones in LEO since the forces ruling the

Figure 21. GCF to GCF maneuver in the L Reference

System, xy projection in GEO with R
f ¼ 2.

271
D. Spiller et al.
spacecrafts dynamics are very weak, the maneuver time is longer than in
LEO and no fast attitude maneuvers are required.
For investigation purposes, a test has been run using a chief orbit with
i ¼ 90 degree and Ω2 . In this case, the in-plane component of the SRP is
so small that the optimizer is cannot find a feasible reconfiguration ma-
neuver.
6.4. Case 3: reconfiguration in GEO
The chief orbital parameters are reported in Table 8 and the perfor-
mances obtained in GEO are reported in Table 9. In this case, a GCF to
GCF maneuver imposing an increased final radius of 1.2 km is also per-
formed. For all the reported cases, approximately one orbit is sufficient
for the maneuver. It is noteworthy that, as in the MEO case, the final
radius of the two trajectories after the maneuver is slightly different from
the expected one due to the presence of in-plane and out-of-plane
components.
Detailed results and figures are not reported for the PCF to GCF ma-
neuvers as they are quite similar to those already reported for the cases in
LEO and MEO. However, the GCF to GCF maneuver is reported through
Fig. 20 and Fig. 21. As described in Sec. 4.4, this maneuver can be
accomplished since the SRP perturbation has an out-of-plane component
that can be used to increase the radius of the relative trajectory. By means
of the reported figures, the reader can appreciate the performances
achieved with the proposed maneuver. The satisfaction of the end-point
constraints is achieved and errors on the values of Kzx and Rf are small.
Such errors might be due to the IPSO difficulty in reaching the optimal
maneuver or might be due to the simultaneous effects of the in-plane and
out-of-plane components of the SRP. Nonetheless, errors are so small that
in the worst case only a small amount of external control would be
needed to achieve the final desired state.
7. Conclusion
In this research we conclude with a major result, that is we are able to
obtain any orientation of the relative motion plane by modifying the
intensity of the drag and the solar radiation pressure effects with attitude
maneuvers. The results obtained computing the maneuver from the
projected circular formation to the general circular formation, and vice-
versa, confirm this finding.
In low Earth orbit, we further verify that the drag can only affect the
in-plane Relative Orbital Elements, making them vary accordingly to the
expected results and thus verifying the validity of the model. With added
improvements to the model, as well as refinements to the planning al-
gorithm based on the inverse dynamics particle swarm optimization, we
achieve closer to theoretical time optimal bang-bang solutions. The
presence of the gravity gradient torque does not affect the possibility to
perform the maneuvers. We further clarify the limitation of this passive
reconfiguration using perturbing forces in low Earth and geostationary
orbits, that is we do not have the ability to increase the major and minor
axes of the formation ellipse.
We characterize the differences in the nature of the reconfiguration
maneuvers using drag for low altitude orbits and solar radiation pressure
for high altitude orbits. We see the magnitude differences between the
perturbing forces but also the result of the geostationary reconfiguration
taking a single orbit and the low Earth case taking approximately an orbit
and a half. All the Relative Orbital Elements vary in the presence of the
solar radiation pressure as a consequence of the maneuver.
Given the great precision that can be achieved, the proposed ma-
neuvers may be used both as a default reconfiguration strategy and as a
backup system in the case of a failure in another control subsystem.
272
Acta Astronautica 153 (2018) 259–273
References
[1] D. Spiller, F. Curti, C. Circi, Minimum-time reconfiguration maneuvers of satellite
formations using perturbation forces, J. Guid. Contr. Dynam. 40 (5) (2017)
1130–1143, https://doi.org/10.2514/1.G002382.
[2] C. Lambert, A. Ng, Y. Nakamura, H. Horiguchi, Intersatellite separation mechanism
for the JC2Sat formation-flying missions, J. Guid. Contr. Dynam. 48 (4) (2011)
654–663, https://doi.org/10.2514/1.51896.
[3] D.P. Scharf, F.Y. Hadaegh, S.R. Ploen, A survey of spacecraft formation flying
guidance and control (part I): guidance, in: Proceedings of the 2003 American
Control Conference, 2, 2003, pp. 1733–1739, https://doi.org/10.1109/
ACC.2003.1239845.
[4] D. Scharf, F. Hadaegh, S. Ploen, A survey of spacecraft formation flying guidance
and control (part II): Control, Proc. Am. Contr. Conf. 4 (2004) 2976–2985.
[5] C. Leonard, W. Hollister, E. Bergmann, Orbital formation keeping with differential
drag, J. Guid. Contr. Dynam. 12 (1) (1989) 108–113, https://doi.org/10.2514/
3.20374.
[6] T. Reid, A.K. Misra, Formation flight of satellites in the presence of atmospheric
drag, J. Aero. Eng. 3 (1) (2011) 64, https://doi.org/10.7446/jaesa.0301.05.
[7] K.D. Kumar, A.K. Misra, S. Varma, T. Reid, F. Bellefeuille, Maintenance of satellite
formations using environmental forces, Acta Astronaut. 102 (2014) 341–354,
https://doi.org/10.1016/j.actaastro.2014.05.001.
[8] R. Bevilacqua, M. Romano, Rendezvous maneuvers of multiple spacecraft using
differential drag under J2 perturbation, J. Guid. Contr. Dynam. 31 (6) (2008)
1595–1607, https://doi.org/10.2514/1.36362.
[9] R. Bevilacqua, J.S. Hall, M. Romano, Multiple spacecraft rendezvous maneuvers by
differential drag and low thrust engines, Celestial Mech. Dyn. Astron. 106 (1)
(2010) 69, https://doi.org/10.1007/s10569-009-9240-3.
[10] M. Horsley, S. Nikolaev, A. Pertica, Small satellite rendezvous using differential lift
and drag, J. Guid. Contr. Dynam. 36 (2) (2013) 445–453, https://doi.org/10.2514/
1.57327.
[11] D. Perez, R. Bevilacqua, Differential drag spacecraft rendezvous using an adaptive
lyapunov control strategy, Acta Astronaut. 83 (2013) 196–207, https://doi.org/
10.1016/j.actaastro.2012.09.005.
[12] D. Perez, R. Bevilacqua, Lyapunov-based adaptive feedback for spacecraft planar
relative maneuvering via differential drag, J. Guid. Contr. Dynam. 37 (5) (2014)
1678–1684, https://doi.org/10.2514/1.G000191.
[13] C. Circi, Simple strategy for geostationary stationkeeping maneuvers using solar
sail, J. Guid. Contr. Dynam. 28 (2) (2005) 249–253, https://doi.org/10.2514/
1.6797.
[14] T. Williams, Z.-S. Wang, Uses of solar radiation pressure for satellite formation
flight, Int. J. Robust Nonlinear Control 12 (2–3) (2002) 163–183, https://doi.org/
10.1002/rnc.681.
[15] Y.-G. Hou, M.-J. Zhang, C.-Y. Zhao, R.-Y. Sun, Control of tetrahedron satellite
formation flying in the geosynchronous orbit using solar radiation pressure,
Astrophys. Space Sci. 361 (4) (2016) 1–11, https://doi.org/10.1007/s10509-016-
2732-1.
[16] S. Varmas, K. Kumar, Multiple satellite formation flying using differential
aerodynamic drag, J. Spacecraft Rockets 49 (2) (2012) 325–336, https://doi.org/
10.2514/1.52395.
[17] K. Shahid, K. Kumar, Multiple spacecraft formation reconfiguration using solar
radiation pressure, Acta Astronaut. 103 (2) (2014) 269–281, https://doi.org/
10.1016/j.actaastro.2014.05.021.
[18] D. Spiller, L. Ansalone, F. Curti, Particle swarm optimization for time-optimal
spacecraft reorientation with keep-out cones, J. Guid. Contr. Dynam. 39 (2) (2016)
312–325, https://doi.org/10.2514/1.G001228.
[19] D. Spiller, F. Curti, L. Ansalone, Inverse dynamics particle swarm optimization for
spacecraft minimum-time maneuvers with constraints, in: 23rd Conference of the
Italian Association of Aeronautics and Astronautics, 2015. Torino, Italy.
[20] D. Spiller, F. Curti, R. Melton, Inverse dynamics particle swarm optimization
applied to constrained minimum-time maneuvers using reaction wheels, in: 67th
International Astronautical Congress, 2016. Guadalajara, Mexico.
[21] D. Spiller, F. Curti, Inverse dynamics particle swarm optimization for nanosatellites
rendezvous via differential drag, in: 3rd IAA Conference, 2015. Rome, Italy.
[22] K. Alfriend, S. Vadali, Spacecraft Formation Flying, Elsevier Astrodynamics Series,
Butterworth-Heinemann, 2010, ISBN 978-0-75-068533-7. Ch. 4–8.
[23] W. Clohessy, R. Wiltshire, Terminal guidance system for satellite rendezvous,
J. Aero. Sci. 27 (9) (1960) 653–678, https://doi.org/10.2514/8.8704.
[24] J. Kennedy, R. Eberhart, Particle swarm optimization, in: Proceedings of the IEEE
International Conference on Neural Networks, 4, 1995, pp. 1942–1948, https://
doi.org/10.1109/ICNN.1995.488968.
[25] M. Fliess, J. Lvine, P. Martin, P. Rouchon, Flatness and defect of non-linear systems:
introductory theory and examples, Int. J. Contr. 61 (2007) 1327–1361, https://
doi.org/10.1080/00207179508921959.
[26] C. Louembet, Design of algorithms for satellite slew manoeuver by flatness and
collocation, in: American Control Conference, 2007, pp. 3168–3173, https://
doi.org/10.1109/ACC.2007.4282459.
[27] R. Eberhart, J. Kennedy, A new optimizer using particle swarm theory, Micro
Machine and Human Science, in: Proceedings of the Sixth International Symposium
on Micro Machine and Human Science, 1995, pp. 39–43, https://doi.org/10.1109/
MHS.1995.494215.
[28] R. Battin, An Introduction to the Mathematics and Methods of Astrodynamics, AIAA
Education Series, American Institute of Aeronautics & Astronautics, 1999, ISBN
9781600860263.
D. Spiller et al.
[29] G. Gaias, S. DAmico, Impulsive maneuvers for formation reconfiguration using
relative orbital elements, J. Guid. Contr. Dynam. (6) (2015) 1036–1049, https://
doi.org/10.2514/1.G000189.
[30] O. Montenbruck, E. Gill, Satellite Orbits, Springer-Verlag Berlin Heidelberg, 2000,
ISBN 978-3-642-58351-3. Ch. 3.
[31] G. Gaias, J.-S. Ardaens, O. Montenbruck, Model of j_2 perturbed satellite relative
motion with time-varying differential drag, Celestial Mech. Dyn. Astron. 123 (4)
(2015) 411–433, https://doi.org/10.1007/s10569-015-9643-2.
[32] G. Cook, The effect of aerodynamic lift on satellite orbits, Planet. Space Sci. 12 (11)
(1964) 1009–1020, https://doi.org/10.1016/0032-0633(64)90077-7.
[33] M.D. Pilinski, B.M. Argrow, S.E. Palo, Semiempirical model for satellite energy-
accommodation coefficients, J. Spacecraft Rockets 47 (6) (2010) 951–956, https://
doi.org/10.2514/1.49330.
[34] D. Vallado, W. McClain, Fundamentals of Astrodynamics and Applications,
Microcosm Press, 2007, ISBN 978-1-881883-14-2. Ch. 8.
273
Acta Astronautica 153 (2018) 259–273
[35] L. Mazal, D. Perez, R. Bevilacqua, F. Curti, Spacecraft rendezvous by differential
drag under uncertainties, J. Guid. Contr. Dynam. (2016) 1721–1733, https://
doi.org/10.2514/1.G001785.
[36] C.D. Meyer, Matrix analysis and applied linear algebra, other titles in applied
mathematics, Soc. Ind. Appl. Math. (SIAM, 3600 Market Street, Floor 6,
Philadelphia, PA 19104) (2000) 279–282. ISBN: 9780898719512.
[37] J. Kennedy, R. Eberhart, Swarm Intelligence, Elsevier Science, 2001, ISBN 1-55860-
595-9. Ch. 7–8.
[38] M.G. Cox, Practical spline approximation, in: Topics in Numerical Analysis,
Springer, 1982, pp. 79–112, https://doi.org/10.1007/BFb0063201.
[39] G.T. Huntington, A.V. Rao, Optimal reconfiguration of spacecraft formations using
the gauss pseudospectral method, J. Guid. Contr. Dynam. 31 (3) (2008) 689–698,
https://doi.org/10.2514/1.31083.
[40] M.M. Berry, L.M. Healy, Implementation of Gauss-Jackson Integration for Orbit
Propagation, 52, 2004, pp. 331–357. https://doi.org/10.1.1.618.7461.