Acta Astronautica 138 (2017) 301–317

Contents lists available at ScienceDirect

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

Surrogate assisted multidisciplinary design optimization for an all-electric

GEO satellite
Renhe Shi, Li Liu, Teng Long *, Jian Liu, Bin Yuan
Key Laboratory of Dynamics and Control of Flight Vehicle, Ministry of Education, School of Aerospace Engineering, Beijing Institute of Technology, Beijing 100081, PR
China

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

Keywords: State-of-the-art all-electric geostationary earth orbit (GEO) satellites use electric thrusters to execute all propulsive
All-electric GEO satellite duties, which significantly differ from the traditional all-chemical ones in orbit-raising, station-keeping, radiation
Multidisciplinary design optimization damage protection, and power budget, etc. Design optimization task of an all-electric GEO satellite is therefore a
Surrogate-based analysis and optimization
complex multidisciplinary design optimization (MDO) problem involving unique design considerations. However,
Disciplinary modeling
solving the all-electric GEO satellite MDO problem faces big challenges in disciplinary modeling techniques and
efficient optimization strategy. To address these challenges, we presents a surrogate assisted MDO framework
consisting of several modules, i.e., MDO problem definition, multidisciplinary modeling, multidisciplinary anal-
ysis (MDA), and surrogate assisted optimizer. Based on the proposed framework, the all-electric GEO satellite
MDO problem is formulated to minimize the total mass of the satellite system under a number of practical
constraints. Then considerable efforts are spent on multidisciplinary modeling involving geosynchronous transfer,
GEO station-keeping, power, thermal control, attitude control, and structure disciplines. Since orbit dynamics
models and finite element structural model are computationally expensive, an adaptive response surface surrogate
based optimizer is incorporated in the proposed framework to solve the satellite MDO problem with moderate
computational cost, where a response surface surrogate is gradually refined to represent the computationally
expensive MDA process. After optimization, the total mass of the studied GEO satellite is decreased by 185.3 kg
(i.e., 7.3% of the total mass). Finally, the optimal design is further discussed to demonstrate the effectiveness of
our proposed framework to cope with the all-electric GEO satellite system design optimization problems. This
proposed surrogate assisted MDO framework can also provide valuable references for other all-electric spacecraft
system design.

1. Introduction launched by SpaceX Falcon 9 rocket in 2015. It was reported that

Since the dawn of the space era, the satellites in geosynchronous earth
orbit (GEO) have gained great interests because of the merits in com-
munications, earth observation, navigation, etc. State-of-the-art GEO
satellites are capable of using electric thrusters to perform all propulsive
tasks like orbit-raising and station-keeping [1]. Compared with conven-
tional GEO satellites using the all-chemical or hybrid propulsion systems,
i.e., using both chemical and electric thrusters, all-electric GEO satellite
can save considerable amount of propellant owing to the superior effi-
ciency of electric propulsion (EP) system, which results in significant
reduction of the launch cost and additional payloads. In recent years,
successful development of several all-electric GEO satellites like ABS-3A
and Eutelsat-115B have been reported [2]. Those satellites were based on
Boeing BSS-702SP all-electric platform with four XIPS-25 thrusters and
BSS702SP only needs 5 kg propellant per year for station-keeping using
xenon ion electric propulsion system (XIPS), which produces consider-
able economic benefits [3].
Although all-electric GEO satellites consume much less propellant
than the competitive chemical ones, it requires extremely long transfer
time (i.e., several months) due to the low thrust (generally from tens to
hundreds of mN) produced by electric thrusters. In view of the prolonged
transfer time within the Van Allen belts, the satellite is thereby exposed
to serious radiation which results in the damage of devices like solar
arrays [4]. There have been numerous studies on continuous low-thrust
orbit transfer problems over the last a few decades [5–11]. Most of the
existing studies only focused on optimizing low-thrust trajectories using
direct method (the problem is solved by gradient based optimization
procedures) or indirect method (the problem is solved by shooting

* Corresponding author.
E-mail address: tenglong@bit.edu.cn (T. Long).

http://dx.doi.org/10.1016/j.actaastro.2017.05.032
Received 24 December 2016; Received in revised form 12 April 2017; Accepted 27 May 2017
Available online 1 June 2017
0094-5765/© 2017 IAA. Published by Elsevier Ltd. All rights reserved.
R. Shi et al.
procedures) to minimize the transfer time or radiation damage [8].
However, it is not sufficient to design a satellite system only from an
optimized orbit transfer trajectory, because designers must make trade-
offs among different subsystems (disciplines) of a GEO satellite. The
design of geosynchronous transfer orbit, station-keeping strategy, power,
attitude control, thermal control, and structure subsystems, etc. should
be considered simultaneously to better model the all-electric GEO sat-
ellite system. Multidisciplinary design optimization (MDO) is therefore
preferred to deal with the satellite system design problems. MDO was
originally proposed by Sobieski [12,13], which was defined as “a
methodology for the design of complex engineering systems and sub-
systems that coherently exploits the synergy of mutually interacting
phenomena” by NASA's Langley Research Center [14]. In the literature,
there were many studies on the applications of MDO in aerospace system
design. For instance, Wu et al. [15] used collaborative optimization (CO)
method to solve the MDO problem of a remote sensing satellite involving
attitude control, power, thermal control, and structure subsystems.
Francesco et al. [16] quantitatively assessed the MDO models for the
early design phase of expendable launch vehicles, and pointed out that
the use of MDO approaches could improve the design quality. Hwang
et al. [17] used a gradient-based approach to solve a small satellite MDO
problem involving over 25,000 design variables considering the coupling
among seven disciplines. Based on Isight environment, Huang et al. [18]
developed a design optimization system integrating a non-nested
collaborative optimization method to solve the multidisciplinary
modeling and optimization problem for a satellite with maneuver capa-
bility. Kwon et al. [19] combined the fields of Hall Effect Thruster (HET)
design and low-thrust trajectory optimization to develop a simultaneous
design optimization environment for space mission analysis. Spangelo
et al. [20] established the systems-level model of a CubeSat including
propulsion, orbit, energy, batteries, and radiation for design optimization
to address the feasibility of using CubeSat Ambipolar Thruster (CAT) for
propulsive duties. More applications of MDO in aerospace engineering
were also reported in Refs. [21–24]. Although MDO techniques have
been widely employed in aerospace system design according to the
aforementioned researches, no research work on all-electric GEO satellite
MDO problems has been reported yet to the authors' knowledge.
Fig. 1. Surrogate assisted MDO framework defined for all-electric GEO satellite design optimization.
302
Acta Astronautica 138 (2017) 301–317
To apply MDO in practical spacecraft system design, two challenges
have to be addressed, i.e., accurate disciplinary modeling techniques and
efficient global optimization strategy [15]. With the development of
computer hardware and software, high fidelity analysis models, e.g.,
computational fluid dynamics (CFD) models, finite element analysis
(FEA) models, have been widely employed in spacecraft design to
improve the modeling accuracy. However, the computationally expen-
sive cost of these high fidelity analysis models still significantly limits the
application of MDO in engineering practice. To reduce the computational
cost in solving MDO problems with expensive functions, surrogate-based
analysis and optimization (SBAO) technologies have been widely
employed [25–28]. In SBAO, a surrogate model is constructed to repre-
sent the true computationally expensive analysis model or multidisci-
plinary analysis (MDA) process for simulation-based optimization. In
recent years, a number of SBAO methods have been developed [29–34],
and many have been successfully applied in spacecraft system design like
GEO satellite truss structure optimization [35] and earth observation
satellite multidisciplinary design optimization [36]. More details about
SBAO can be referred to Refs. [25–28].
This paper aims at establishing a surrogate assisted satellite MDO
framework, and it is believed to be the first time to apply MDO tech-
niques for all-electric GEO satellites system design. The outline of the
paper is organized as follows. The surrogate assisted MDO framework is
presented in Section 2, and the studied all-electric telecommunication
satellite (denoted as AETS) MDO problem is defined in Section 3. Section
4 presents the multidisciplinary modeling for geosynchronous transfer,
GEO station-keeping, power, thermal control, attitude control, and
structure disciplines. Section 5 gives the fixed-point iteration based MDA
process. In Section 6, an adaptive response surface method using intel-
ligent space exploration strategy (denoted as ARSM-ISES) is used to solve
the AETS MDO problem, and the optimization results are discussed in
detail. Conclusions and future work are given in Section 7.
2. Surrogate assisted MDO framework
A surrogate assisted MDO framework is developed to handle the
complex modeling, simulation, data interaction, and optimization for all-
R. Shi et al.
Fig. 2. AETS GEO communications satellite model.
electric GEO satellite design optimization as shown in Fig. 1. The
framework consists of several modules, i.e., MDO problem definition,
multidisciplinary modeling, multidisciplinary analysis, and surrogate
assisted optimizer.
(1) MDO problem definition
In the proposed framework, the all-electric GEO satellite MDO
problem is first defined including the disciplines, design variables, con-
straints, objective, etc. The definition of studied satellite MDO problem is
detailed in Section 3.
(2) Disciplinary modeling
The disciplinary models are established in terms of the defined MDO
problem. The modeled disciplines are geosynchronous transfer, GEO
station-keeping, power, thermal control, attitude control, and structure.
The details of these models are presented in Section 4.
(3) Multidisciplinary analysis
Given the design structure matrix (DSM) of the all-electric GEO sat-
ellite, a multidisciplinary analysis process is conducted to ensure the
system consistency (i.e., the interdisciplinary capability condition is
satisfied). In this paper, the MDA process is solved by a fixed-point
iteration method detailed in Section 5.
(4) Surrogate assisted optimizer
Since the all-electric GEO satellite MDO problem involving many
computationally expensive models (e.g., FEA model, dynamics model,
etc.), the choice of the MDO strategy is critical. To reduce the compu-
tational cost, a surrogate-based optimizer is provided to efficiently solve
Fig. 3. DSM of AETS MDO problems.
303
Acta Astronautica 138 (2017) 301–317
the optimization problem to obtain the optimal design. Currently, our
previously proposed surrogate assisted optimization algorithms have
been integrated in this framework, such as FCR [33], ARSM-ISES [34],
SRBF [35]. In this work, ARSM-ISES optimizer is used to solve AETS MDO
problems, as detailed in Section 6.
3. AETS MDO problem definition
Our studied all-electric telecommunication satellite (AETS) is a state-
of-the-art all-electric GEO communication satellite with dimensions of
2400(X)
2200(Y)
3500(Z) mm. AETS comprises payload module,
service module, and solar arrays. It is fitted with four shaped commu-
nication antennas and 34 transponders including 16 C-band, 14 Ku-band,
and 4 Ka-band transponders. AETS uses four ion thrusters mounted on
the bottom of the satellite to execute geosynchronous transfer and GEO
station-keeping maneuvers. The ion thrusters can provide a maximum
thrust of 200 mN with 4.5 kW power and 4000s specific impulse. A
rendering of AETS is shown in Fig. 2.
Compared with chemical transfer, the electric orbit-raising process of
AETS is significantly prolonged and results in the degradation of solar
arrays caused by the radiation damage. AETS also has to execute the GEO
station-keeping maneuvers in a different way from that of using chemical
thrusters due to the slight velocity increment provided by the ion
thrusters. Moreover, AETS needs to deploy the solar arrays during the
transfer, which requires the considerations for thermal control and atti-
tude control subsystems. The mass and stiffness distribution of AETS
structure are also different from its competitive all-chemical GEO satel-
lite because AETS is much lighter and smaller owing to the propellant
saved during orbit transfer and station-keeping.
In view of the aforementioned characteristics of all-electric GEO
satellite, the modeled disciplines involve geosynchronous transfer, GEO
station-keeping, solar power, thermal control, attitude control, and
structure for AETS MDO problem. The coupling relationship for the MDO
problem is organized in design structure matrix as shown in Fig. 3. The
DSM shows the execution flow and the exchange of data among different
disciplines. The terms above the diagonal constitute feed-forward infor-
mation while the terms below constitute backward information. The
DSM shows that feedbacks of geosynchronous transfer, station-keeping,
thermal control, and attitude control disciplines are from structure
discipline, which requires an iterative loop to close the design circle. The
variables and parameters in DSM are defined in Table 1.
An eternal struggle of the space industry is to minimize the mass
because it means reducing launch cost and increasing flexibility in choice
of launch vehicles [37]. Therefore, the objective of AETS MDO problem is
defined as the total mass of the satellite, i.e., y61 output from the structure
discipline in Fig. 3. Based on the aforementioned DSM, the MDO problem
is formulated in Eq. (1)
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Table 1 4. Multidisciplinary modeling for AETS

Definition of the DSM's parameters for different disciplines.

Symbol Definition Symbol Definition The multidisciplinary modeling for all-electric GEO satellite concep-
x1 Thrust angles in geosynchronous y23 Power requirement in GEO tual design is a particularly challenging task considering the complexity
transfer orbit (GTO) operation of satellite system. This section describes the analysis modeling for those
x2 Position of thrusters y26 Mass of fuel used in GEO major disciplines in all-electric GEO satellite MDO problems.
station-keeping
x3 Area of solar arrays and capacity of y34 Output power of solar arrays
storage battery 4.1. Geosynchronous transfer discipline modeling
x4 Area of heat radiator y35 Area of solar arrays
x5 Angular momentum of reaction y36 Mass of power subsystem
wheel The geosynchronous transfer discipline determines the geosynchro-
x6 Thickness of the structural y46 Mass of thermal control nous transfer orbit (GTO) to accomplish low continuous thrust GEO
composite material subsystem insertion of the satellite. For the given starting injection orbit and the
p1 Thrust, specific impulse of the y56 Mass of attitude control electric thrusters, a two-stage electric propulsion transfer strategy is
thrusters, initial orbital parameters subsystem
employed to determine the GTO [38]. In each transfer stage, the thrust
of GTO
p2 Thrust, specific impulse of the y61 Total mass of the satellite, vector is fixed in the corresponding coordinates depicted in Fig. 4. The
thrusters, initial orbital parameters objective coordinates are defined as follows.
of GEO station-keeping
p3 Non-ionizing energy loss (NIEL) of y62 Center of gravity height
(1) Coordinate RTN: The origin point is the center of mass of the
solar panel material, parameters of
material displacement damage satellite. The R-axis locates in the orbital plane and passes through
equation, parameters of radiation the center of the earth. The T-axis locates in the orbital plane and
belt, glass cover thickness, is perpendicular to the R-axis. The N-axis is the right-handed third
parameters of solar power model direction.
p4 Parameters of heat network model, y64 Total mass of satellite, objective
(2) Coordinate PQH: The origin point is the center of the earth. The H-
initial temperature of satellite
p5 Fixed mass and power of attitude y65 Moments of inertia axis points in the same direction from the orbital angular mo-
control subsystem mentum vector. The P-axis locates in the orbital plane and points
p6 Mechanical parameters of y1 Transfer time of GTO process, to the perigee. The Q-axis is the right-handed third direction.
structural material constraint
y12 Initial satellite mass in GEO y2 Accuracy of GEO station-
keeping, constraints
The electric thrust is regarded as the perturbation force and the
y13 Transfer information and flying y3 Depth of discharge, beginning- Gaussian perturbation dynamics of GTO is formulated in Eq. (2) [39]
time during the radiation belt, of-life power, and ending-of-life
power requirement of the satellite power, constraints 8 da 2
>
> ¼ pffiffiffiffiffiffiffiffiffiffiffiffi2ffi ðfR e sin f þ fT ð1 þ e cos f ÞÞ
in GTO >
> dt 1 e
y14 Extreme external heat flux y4 >
>
n
Internal temperature of satellite, >
>
constraint >
> p ffiffiffiffiffiffiffiffiffiffiffiffi
ffi
>
> 2
y15 Transfer information y5 Residue of Angular momentum, > de ¼ 1  e ðf sin f þ f ðcos f þ cos EÞÞ
>
>
> R T
constraint >
> dt na
y16 y6 >
>
Mass of fuel used in GTO Natural frequency of the electric >
>
GEO satellite, constraints > di
> r cos u
>
> pffiffiffiffiffiffiffiffiffiffiffiffiffifN
y23 Power requirement in GEO > dt ¼ 2
>
>
operation < na 1  e2
xi (design variables of discipline i, row vector or scalar), pi (fixed parameters discipline i), yij >
> dΩ r sin u
(coupling variables from discipline i to j), and yi (output of discipline i). >
> ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> fN
>
> dt na 1  e2 sin i
2
(2)
>
>
>
>
>
> pffiffiffiffiffiffiffiffiffiffiffiffiffi    
>
> dω 1  e2 dΩ
>
> ¼  þ 1 þ
r
 cos i
>
> fR cos f fT sin f
min8Msatellite ¼ y61 ðXÞ; X ¼ ½x1 ; x2 ;x3 ;x4 ;x5 ; x6  >
> dt nae p dt
>
>
½y1 ; y12 ; y13 ; y14 ; y15 ;y16  ¼ D1 ðx1 ;p1 ;y 61 Þ >
>
>
> >
> 2     
>
> ½y2 ;y 23 ;y 26  ¼ D2 ðx2 ;p2 ;y12 ;y62 Þ >
: dM ¼ n  1  e r r
>
>  fR cos f  2e þ fT 1 þ sin f
>
> ½y 3 ;y34 ;y35 ;y36  ¼ D3 ðx3 ;p3 ; y13 ; y23 Þ dt nae p p
>
>
>
> 8
>
> ½y4 ;y46  ¼ D4 ðx4 ;p4 ;y 14 ;y 34 ;y 64 Þ > fR ¼ FR þ AR
>
> >
<
>
> ½y5 ;y56  ¼ D5 ðx5 ;p5 ;y 15 ;y 35 ;y 65 Þ
< fT ¼ FT þ AT
½y6 ;y61 ;y62 ;y64 ;y65  ¼ D6 ðx6 ;p6 ;y 16 ; y26 ; y36 ; y46 ; y56 Þ >
s:t >
:
>
> C1 ðx1 ;p1 ;y 61 Þ  0 fN ¼ FN þ AN
>
>
>
> C2 ðx2 ; p2 ;y12 ;y62 Þ  0
>
>
>
> C3 ðx3 ; p3 ;y13 ;y23 Þ  0 where a, e, i, Ω, ω, M are the Keplerian elements, p ¼ a(1-e2), u ¼ ω þ f,
>
>
>
> C4 ðx4 ;p4 ; y14 ; y34 ;y64 Þ  0
>
> and f is the true anomaly. r is the distance from the satellite to the earth
>
> C5 ðx5 ;p5 ;y15 ;y35 ;y65 Þ  0
: center. F ¼ [FR,FT,FN] and [AR,AT,AN] are the thrust accelerations and
C6 ðx6 ; p6 ;y16 ;y 26 ;y 36 ;y 46 ;y 56 Þ  0 disturbance accelerations respectively expressed in RTN coordinate sys-
(1) tem. The GTO model only considers the disturbance accelerations due to
the first four zonal harmonics of non-spherical gravitational potential,
where X is the design variables of the satellite MDO problem, Di ðxi ; yi ; which are given in the Appendix.
y ij Þ and Ci ðxi ; y i ; y ij Þ are the disciplinary analysis model and constraint It is noteworthy that the electric thrusters are incapable of operating
function of the i-th discipline respectively. Eq. (1) depicts that the design during eclipses because the solar arrays cannot generate power in the
variables are optimized to obtain a compatible and feasible satellite shadow of the earth. An earth conical shadow model illustrated in Fig. 5
design with the lowest total mass. The disciplinary models and the so- is employed to address the eclipses problem during orbit transfer. It is
lutions of the MDO problem will be presented in following sections. assumed that the thrust acceleration F ¼ 0 when the satellite enters the
umbra or penumbra according to Eq. (3) [39]

304
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Fig. 6. Illustration of thrust angle.

Fig. 4. Coordinates used in GTO model.
Table 2
Design variables and constraint of geosynchronous transfer discipline.
(
ψ1 < ψ < ψ2 penumbra Parameters Type Symbol Unit Range

Yaw angle in first-stage of GTO Design variable α [0,45]
ψ 2  ψ umbra Pitch angle in first-stage of GTO Design variable β 
[0,45]
8 a    Pitch angle in second-stage of GTO Design variable φ 
[30,45]
>
> e Rs þ ae
>
> ψ 1 ¼ π  arcsin  arcsin (3) Total orbit transfer time Constraint tf Day 180
< r ρ
> a   
>
> R þ ae
>
: ψ 2 ¼ π  arcsin e þ arcsin s
Given the starting injection orbit and the EP system parameters, the
r ρ GTO is finally determined by the aforementioned two-stage orbit transfer
strategy. Although the obtained GTO is not theoretical optimal, it is easy
where ψ is the field angle for the earth center between the satellite and to be implemented in practical engineering because no frequent attitude
sun, ψ 1 and ψ 2 are the field angles when the satellite are located on the maneuvers are needed during the transfer. The total orbit transfer time
boundaries of penumbra and umbra respectively, ae ¼ 6387 km is the and the propellant consumption could be computed by solving the orbit
equivalent radius of the earth considering the effect of atmosphere, transfer dynamics. The propellant consumed in GTO is exported as the
Rs ¼ 696000 km is the radius of the sun. coupled input of the structure discipline. The design variables x1 ¼ [α, β,
The satellite is initially launched into the great-elliptic starting in- φ] and constraint y1 ¼ tf of geosynchronous transfer discipline are
jection orbit, i.e, a ¼ 24328–28461 km, e ¼ 0.7096–0.7296, summarized in Table 2.
i ¼ 18.5–23.5 . In the first stage of orbital transfer, the control parame- A simulation-based validation is performed to validate the afore-
ters are the yaw angle α and pitch angle β of thrust acceleration in the mentioned two-stage geosynchronous orbit transfer model. In the simu-
RTN coordinate depicted in Fig. 6(a). The absolute values of α and β are lation example, the initial mass of the satellite is 2000 kg. Two 200 mN
not changed, while the ± sign of β is switched when u is 90 or 270 . ion thrusters (specific impulse ISP ¼ 4000s) are utilized to accomplish the
Affected by such thrust acceleration, the semi-major axis is increased orbit transfer. The Keplerian elements of the starting injection orbit are
while the inclination is decreased according to Eq. (2). When the semi- a ¼ 24328 km, e ¼ 0.7296, i ¼ 23.5 , ω ¼ 180 , Ω ¼ 90 , M ¼ 0 . The
major axis of GTO is the same with that of GEO, i.e., 42166 km, the starting injection orbit is illustrated in Fig. 7. The control parameters of
first stage of the orbit transfer terminates. In the second stage of orbit transfer are α ¼ 0 , β ¼ 30 , φ ¼ 30 . The simulation results of GTO are
transfer, the thrust vector is fixed in QoH plane with the pitch angle φ to exhibited in Fig. 8. It is clear that the satellite is finally inserted into the
the negative direction of Q-axis depicted in Fig. 6(b). The ± sign of φ is geostationary earth orbit with 0 eccentricity, 0 inclination, and
switched when u is 90 or 270 . Affected by such thrust acceleration, the 42166 km semi-major axis. Moreover, the Keplerain elements vary
mean value of the semi-major axis remains unchanged while the incli- smoothly during the orbit transfer. The simulation results illustrate that
nation and eccentricity are gradually decreased to zero to accomplish the two-stage orbit transfer strategy can successfully accomplish the all-
GEO insertion. Noted that Eq. (2) becomes singular when i and e are zero, electric satellite GEO insertion task.
the orbit propagation process is forced to terminate when i  0.01 and It should be noticed that the implementation of all-electric orbit
e  0.0001. It is acceptable for the conceptual design of the satellite, transfer can save considerable propellant, which could lead to lighter
because the orbital acquisition process will eliminate those errors in
practical engineering.

Fig. 5. Illustration of earth shadow model. Fig. 7. Illustration of the starting injection orbit.

305
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Fig. 8. GTO Keplerian elements simulation results.

satellites and lower launch cost. However, the mass benefit of using EP 8
system comes at the cost of much longer orbit transfer time (often several > dT
> kT ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
ffi
months) compared with traditional chemical propulsion transfer. For this >
> d T þ d 2N þ z2c
2
>
>
validation example, the propellant consumption is only 164.68 kg, while >
>
>
< dN
the total orbit transfer time is 187.30 days. A much longer orbit transfer kN ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (4)
time might bring additional problems to the design of solar arrays >
> d T þ d 2N þ z2c
2
>
>
considering the effect of radiation damage, which will be discussed in >
>
>
> zc
detail in Section 4.3. >
> k ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
: R
d T þ d 2N þ z2c
2

4.2. GEO station-keeping discipline modeling

Based on the given thruster configuration, a low-thrust GEO station-
keeping strategy is proposed to implement the on-orbit station keeping
In geostationary earth orbit, the satellite is influenced by various
with EP system illustrated in Fig. 10. Different from traditional chemical
perturbation factors (e.g., luni-solar perturbation, non-sphericity earth
station-keeping, a completed EP station-keeping period lasts for two
potential perturbation), which tend to make the satellite drift from its
weeks with seven short periods of two days. In each short station-keeping
expected orbital position. Therefore, on-orbit station keeping is required
period, the thrusters work in order of Thruster 1- Thruster 4- Thruster 2-
to ensure the satellite to remain in specified latitude and longitude. The
Thruster 3. Thruster 1 and Thruster 2 work at the same orbital position,
GEO station-keeping discipline models the north/south station- keeping
while Thruster 3 and Thruster 4 have the same working position which
(NSSK) and east/west station-keeping (EWSK) maneuvers implemented
differs about 180 longitude from the former ones.
by EP system to determine the thruster installation configuration. Zhang
The NSSK is performed to control the periodic latitude oscillations
et al. [40] proposed a low-thrust station-keeping strategy considering the
because of the luni-solar perturbations on the inclination, while the
failure mode of the thrusters, which is favorable for the engineering
EWSK is implemented to control the longitude evolution due to the drift
applications. Based on this approach, a simplified GEO station-keeping
caused by the non-sphericity earth potential perturbation and the non-
model is established to design the positions of the electric thrusters.
zero eccentricity caused by the solar radiation pressure. As for tradi-
For the all-electric GEO satellite in this paper, four ion thrusters are
tional chemical station-keeping, the NSSK and EWSK are implemented
mounted symmetrically on the bottom of satellite as shown in Fig. 9.
independently because the required velocity increment for correction
Coordinate TNR is defined to describe the position of the onboard
maneuvers can be easily produced by the chemical thrusters. On the
thrusters. The mass center of the satellite is taken as the origin point. The
contrary, the NSSK and EWSK must be performed simultaneously for EP
T-axis is tangent to the orbit. The R-axis passes through the center of the
station-keeping due to the slight velocity increment provided by electric
earth and points in the opposite direction from the earth. The N-axis is
thrusters. The linearized impulse dynamics for the EP station-keeping
perpendicular to T and R axis. For the given configuration design vari-
maneuvers is formulated in Eq. (5) [40]
ables dT, dN, and the height of mass center zc, the coordinates of the
thrusters in TNR are denoted as Thruster 1(-dT, dN, zc), Thruster 2(dT, dN,
zc), Thruster 3(-dT, -dN, zc), and Thruster 4 (dT, -dN, zc). The thruster vectors
are configured to pass the mass center of the satellite, thus the absolute
direction cosines of the thruster vectors in TNR denoted as kT, kN, and kR
are computed through Eq. (4).

Fig. 9. Illustration of electric thrusters on the bottom of the satellite. Fig. 10. Illustration of GEO station-keeping strategy.

306
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Table 3 the GEO satellite is 120 E with the initial orbit elements of
Design variables and constraints of GEO station-keeping discipline. a ¼ 42166.2 km, i ¼ 0.01 , e ¼ 0.0001, λ ¼ 120 . The simulation results
Parameters Type Symbol Unit Range of latitude and longitude evolution are exhibited in Fig. 11. It shows
T position of thruster Design variable dT mm [500,1180] that the NSSK and EWSK control accuracy, i.e., the maximum station-
N position of thruster Design variable dN mm [800,1050] keeping deviations, are about 0.008 and 0.041 respectively. The

EWSK accuracy Constraint λmax 0.05 simulation results depict that the proposed GEO station-keeping model

NSSK accuracy Constraint imax 0.05 is able to remain the satellite within specified limits in latitude
and longitude.

8
> 3 4.3. Power discipline modeling
>
> ΔD ¼  ΔVT
>
> a
>
>
s
>
>
>
> ΔV R Since all-electric propulsion orbit transfer has a long transit time
>
> Δλ ¼ 2
>
> Vs within the Van Allen radiation belts, the degradation of the solar arrays
>
>
>
> ΔVR ΔVT
caused by the radiation damage must be considered. The power disci-
>
>
> Δex ¼
< sin l þ 2 cos l pline designs the area of solar arrays and the capacity of battery to pro-
Vs Vs
(5) vide sufficient available power, i.e., no less than the required power
>
> ΔVR ΔVT
>
> Δey ¼  cos l þ 2 sin l usage of the satellite.
>
>
>
>
V s Vs In view of the slight radiation damage due to electrons, the solar
>
>
>
> ΔVN power model mainly looks into the damage caused by the trapped pro-
>
> Δix ¼ cos l
>
> Vs tons. As for AETS, the solar arrays are shielded with 3-mil fused silica
>
>
>
> cover glass, which can block the protons with the energies up to 2.8 MeV
>
> ΔVN
: Δiy ¼ sin l [41]. It is assumed that the accuracy of older NASA AP8 model is suffi-
Vs
cient enough for the conceptual design of the satellite in this paper, hence
where D is the orbital drift ratio, l ¼ Ωþ ωþ M is the mean right as- an approximation model of the radiation flux based on NASA AP8 model
    developed in Ref. [1] is used to compute the omnidirectional radiation
e
cension of thruster work position, x ¼ e cosðΩþ ωÞe sinðΩþ ωÞ is flux of protons in terms of McIlwain coordinate L and latitude l formu-
ey
    lated in Eq. (8)
ix i cosðΩÞ
the eccentricity vector, ¼ is the inclination vector, ΔVR ,
iy i sinðΩÞ 2
ψðL; l; EÞ ¼ aðL; EÞebðL;EÞl
ΔVT , and ΔVN are the velocity increments in the direction of R,T,N axises,
r
Vs is the geosynchronous velocity. L¼
According to Eq. (4) and Eq. (5), the low thrust GEO station keeping R cos2 l (8)
2
control equation is formulated in Eq. (6). aðL; EÞ ¼ a0 ea1 Eþa2 ða3 þLÞ
KΔV ¼ Δe (6) bðL; EÞ ¼ b0 þ b1 E þ b2 L þ b3 EL þ b4 EL þ b4 L2 þ b5 L3

In Eq. (6), K is the control matrix which is determined by the thrusters where R is the radius of the earth, and r is the distance between satellite
configuration given in Eq. (7) and the center of earth. The constants in the equation are presented in
Table A2 in the Appendix.

2

 3T
kN =Vs  kR sin lomg þ 2kT cos lomg Vs kR cos lomg þ 2kT sin lomg Vs 3=as kT
6 k =V


6 N s  kR sin lomg  2kT cos lomg Vs kR cos lomg  2kT sin lomg Vs 3=as kT 77
K¼6

 7 (7)
4 kN =Vs kR sin lomg  2kT cos lomg Vs  kR cos lomg  2kT sin lomg Vs 3=as kT 5



kN =Vs kR sin lomg þ 2kT cos lomg Vs  kR cos lomg þ 2kT sin lomg Vs 3=as kT

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
where Δe ¼ ½ Δi2x þ Δi2y ; Δex ; Δey ; ΔDT is the orbital elements variation The displacement damage Dd for the solar arrays is given as [41]
in the orbit evolution, ΔV ¼ ½ΔV1 ; ΔV2 ; ΔV3 ; ΔV4 T is the required ve-
locity increment provided by each thruster for the correction maneuvers Dd ðEÞ ¼ ϕðEÞNIELðEÞ (9)
in a short position keeping period, lomg is the working right ascension of
where ϕðEÞ is the fluence of protons with energy E, NIEL(E) is the non-
Thruster 1.
ionizing energy loss for the solar cell material caused by protons with
When the control matrix K is invertible, ΔV could be calculated as
energy E. As for the all-electric GEO satellite in this paper, the solar arrays
ΔV ¼ K 1 Δe. The on-orbit longitude and latitude evolution of the GEO
are made of gallium arsenide (GaAs) material. A table of NIEL experi-
satellite is able to be computed through the orbital dynamics in Eq. (5),
mental data of different energy levels for gallium and arsenide is avail-
and the propellant consumed for station keeping is exported as the
able in Ref. [42]. The NIEL of GaAs is computed by taking the weighted
coupled input of the structure discipline. The design variables x2 ¼ [dT,
average NIEL of gallium and arsenide according to their atomic weights.
dN] and constraints y2 ¼ [λmax ,imax ] of GEO station-keeping discipline are
A piecewise linear interpolation with analytic expression is used to fit the
summarized in Table 3.
experimental data as shown in Fig. 12. The NIEL of GaAs of any energy
A 15 years' simulation from Jan 1st, 2020to Jan 1st, 2035 is per-
level can be fitted with the interpolation.
formed to validate the proposed GEO station-keeping model. The initial
The power degradation coefficient pr of the solar arrays is computed
mass of the satellite is 2000 kg, the area-mass ratio is 0.05, and the
through Eq. (10) [43]
sunlight pressure ratio is 1.5. The thruster configuration variables are
set as dT ¼ 1180 mm, dN ¼ 1050 mm. The assigned orbital position of

307
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Fig. 11. GEO station-keeping control accuracy of 15 years simulation results.

  Note that the power required by the satellite must not be larger than
Dd the power available produced by the solar arrays, the solar power should
pr ¼ K log 1 þ (10)
Dx satisfy the constraints as

where solar cell material coefficients used in this paper are
Dx ¼ 3.52
109 and K ¼ 0.135. It is noted that since the effect of
shielding on slowing down of the encountered protons is not considered
in power discipline modeling, the actual power degradation coefficient pr
may be higher than that computed by Eq. (10).
The beginning-of-life power PBOL is defined as the initial power of
solar arrays when the satellite begins to work on-orbit after the transfer.
P0 is the power provided by the solar arrays before orbit transfer. PBOL
and P0 can be written as [44]
PBOL ¼ ð1  pr ÞP0
(11) DOD ¼ C=Cs
P0 ¼ S0 XXs Xe X0 Asa ηFc βp ΔT þ 1 cos χ
where S0 ¼ 1353 W/m2 is the solar constant, X ¼ 0.95, Xs ¼ 0.9637,
Xe ¼ 1, X0 ¼ 0.98 are the correction factors, Asa is the area of the solar
arrays, η ¼ 0.28 is the photoelectric conversion efficiency of the given
GaAs material, Fc ¼ 0.98 is the solar array loss coefficient, βp is the power
temperature coefficient and βp ΔTþ 1 ¼ 0.826, χ is the incident angle
of sunlight.
The ending-of-life power PEOL of the solar arrays is computed through
Eq. (12) [44]
L msolar ¼ ρsa Asa
PEOL ¼ PBOL 1  dy t
where Lt ¼ 15 years is the on-orbit lifetime of the satellite, and dy ¼ 1.4%
is the annual power degradation of the solar arrays in GEO.
PBOL  PGTO
(13)
PEOL  PGEO
where PGTO is the required power usage of the satellite in GTO, and PGEO
is the required power usage of the satellite in GEO. As for AETS,
PGTO ¼ 14.15 kW, PGEO ¼ 11.90 kW.
When the satellite is in the shadow of the earth, the storage battery
group of the satellite has to provide the required power. The depth of
discharge (DOD) for the battery is defined as the ratio of its discharged
capacity C to its rated capacity Cs as shown in Eq. (14) [18,44]
Te (14)
C ¼ Pe ⋅
VEOL
where Pe ¼ 6.6 kW is the power usage of the satellite in eclipse,
Te ¼ 72min is the maximum eclipse time in GEO, VEOL ¼ 67.2 V is the
end-of-life voltage of the battery. The maximum DOD should be no more
than 80% to prolong the life of battery.
The buffer mass of the solar arrays and battery group are exported as
the coupled input of the structure discipline as shown in Eq. (15)
(12) (15)
mbattery ¼ Cs ⋅VBOL =γ b
where ρsa ¼ 2:83kg=m2 is the areal density of GaAs solar cell,
γ b ¼ 39.6 kW/kg is the specific energy of the battery, the VBOL ¼ 96 V is
the begin-of-life voltage. The design variables x3 ¼ [Asa, Cs] and con-
straints y3 ¼ [PBOL, PEOL, DOD] of power discipline are summarized
in Table 4.

4.4. Thermal control discipline modeling

When the electric thrusters and other components operate, the tem-
perature of the satellite rises. In view of the continuous thruster operation

Table 4
Design variables and constraints of power discipline.

Parameters Type Symbol Unit Range

Area of solar arrays Design variable Asa m2 [60,90]

Capacity of battery Design variable Cs Ah [60,90]
Beginning-of-life power Constraint PBOL kW 14.15
Ending-of-life power Constraint PEOL kW 11.90
Depth of discharge Constraint DOD – 0.8
Fig. 12. GaAs NIEL experimental regression curve.

308
R. Shi et al. Acta Astronautica 138 (2017) 301–317

requirement and high thermal power of EP system, the temperature Table 5

control (TC) of all-electric GEO satellite is stricter than that of the Design variable and constraint of thermal control discipline.

competitive all-chemical one. From thermal balance, the thermal control Parameters Type Symbol Unit Range
discipline designs the area of radiators to restrict the temperature of the Area of radiators Design variable Ar m2 [5,10]
satellite in a certain range. Steady-state temperature Constraint T K [267,328]
As for AETS, the radiators are mounted on the south/north panels of
the satellite. Other surfaces of the satellite are encapsulated by multilayer
insulation (MIL) blankets to reduce the influence of external heat flux.

mr ¼ ρr σAr ε1 T14 þ ε2 T24 (19)

Considering the extreme orbital heat flux and the internal heat source, a
thermal network work model is established to calculate the steady-state
where ρr ¼ 20 kg/kW is the power density of the radiator. The design
temperature inside the satellite. The thermal network model consists of
variable x4 ¼ Ar and constraint y4 ¼ T of thermal control discipline are
seven nodes, i.e., inside part of the satellite (Node 0), south panel (Node
summarized in Table 5.
1), north panel (Node 2), top panel (Node 3), bottom panel (Node 4), east
panel (Node 5), and west panel (Node 6) as shown in Fig. 13. It is
assumed that only Node 0 has heat conduction to the other nodes, and the 4.5. Attitude control discipline modeling
space background radiation is neglected. Without considering the ther-
mal capacity of MIL, the thermal balance equation of the seven nodes is The attitude control (AC) subsystem of AETS utilizes four reaction
formulated in Eq. (16) [45]. wheels for attitude hold and adjustment under different satellite external
disturbance torques. The disturbance torques depend on the orbit
Ai λi ðTi  T0 Þ transfer trajectory, moments of inertia, and sunshine area, i.e., the di-
εi σAi Ti4 þ  αi qi Ai ¼ 0 (16) mensions of the solar arrays and the satellite. As for all-electric GEO
δi
satellites, the solar arrays are deployed at the beginning of transfer,
where σ ¼ 5:67
108 W⋅m2 ⋅K4 is the Stefan–Boltzmann constant, Ti which magnifies the external disturbance torques due to the increased
is the temperature of i-th node, εi and αi are the surface emissivity and moment of inertia and area of sunlit surface. Hence, the needed angular
absorption factor of i-th node respectively, Ai is the surface area of i-th momentums of reaction wheels should be carefully designed to absorb
node, noted that A1,2 ¼ Ar is the area of each radiator, λi is the equivalent the accumulated angular momentum in an unloading cycle. Considering
thermal conductivity between i-th node and Node 0, δi is the surface the altitude of GTO, the moments of aerodynamic force and earth mag-
thickness of i-th node, qi is the average heat flux on surface of i-th node. netic are ignored.
The parameters are detailed in Table A3 in the Appendix. 20 3
The thermal balance equation inside the satellite is formulated in Eq. 2 3
(17) [45] MgX 6 μ 7
6 7 6 7
6 3 3 ðIZZ  IXX Þθ 7
Mg ¼ 6 M 7 ¼
4 gY 5 6 r 6 7
dT0 X 6
Ai λi ðTi  T0 Þ 7
¼ þ Qh 4 5
cm (17) MgZ μ
dt i¼1
δi 3 3 ðIZZ  IYY Þϕ (20)
r
2 3
where c ¼ 300 J/K/kg is the specific heat of the satellite, m is the satellite MsX  
6 7 Fe
mass, Qh is the internal power dissipation. As for AETS, Qh is 50% of the Ms ¼ 6 M
4 sY 5
7 ¼ lp
2 A⋅n
power usage of the satellite output by the power discipline. c
The steady-state temperature inside the satellite, i.e., T0, can be MsZ
calculated according to Eq. (16) and Eq. (17), which is the local The external disturbance torques caused by gravity gradient and solar
constraint of the thermal control discipline as shown in Eq. (18). pressure expressed in body-fixed coordinates system of the satellite, i.e.,
Mg and Ms respectively, are exhibited in Eq. (20) [18]. In the equation,
267:0K  T0  323:0K (18)
μ ¼ 398600.5km3s-2 is the earth's gravitational constant, IXX, IYY, and IZZ
The buffer mass of the radiators are calculated by Eq. (19) as the are the moments of inertia in body-fixed coordinates system respectively,
coupled input of the structure discipline θ and ϕ are the pitch angle and roll angle respectively, A is the area of
sunlit surface, Fe is the solar flux constant 1358 W/m2, c is the light speed,
lp is radius vector form mass center to solar pressure center, n is the unit
normal vector of sunlit surface.
The momentum dumping cycle Td of the reaction wheels is one day
during the transfer. The momentum of the reaction wheels are dumped
by the control moment provided by the electric thrusters. Due to the
superior efficiency of EP system, the trivial propellant consumption for
momentum dumping during the transfer is not considered in this paper.
The accumulated angular momentums of disturbance torques are
formulated in Eq. (21) [18].
 
T
HwX ¼ ð1 þ ηÞ∫ 0 d MgX þ MsX dt
 
T
HwY ¼ ð1 þ ηÞ∫ 0 d MgY þ MsY dt (21)
 
T
HwZ ¼ ð1 þ ηÞ∫ 0 d MgZ þ MsZ dt

where η ¼ 20% is the design tolerance of reaction wheels.

The needed angular momentum of a single reaction wheel Hw should
Fig. 13. Thermal network work model of the satellite. satisfy the angular momentum residue constraint cAC as shown in Eq.
(22), and the buffer mass of the reaction wheels mw are calculated by Eq.

309
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Table 6
Design variable and constraint of attitude control discipline.

Parameters Type Symbol Unit Range

Angular momentum of reaction wheel Design variable Hw N⋅m⋅s [25,50]

Angular momentum residue Constraint cAC N⋅m⋅s 0

(23) [18] as the coupled input of the structure discipline. The design
variable x5 ¼ Hw and constraint y5 ¼ cAC of attitude control discipline are
summarized in Table 6.

cAC ¼ Hw  maxðHwX ; HwY ; HwZ Þ  0 (22)

mw ¼ 4
ð5:8 þ 17ðHw  6Þ=90Þ þ 90:4 (23)

Fig. 15. Structural FE model of the satellite (MPCs are erased).

4.6. Structure discipline modeling

The structure discipline establishes the structural FEA model of the

satellite based on the given configuration to obtain the mass center,
moments of inertia, and natural frequency. The satellite structure is a
cuboid consisting of a communication cabin, a service cabin with four
fuel tanks inside, and a Φ 1200 mm central cylinder. The configuration of
the satellite structure is shown in Fig. 14.
The structural FEA model of the all-electric GEO satellite is shown in
Fig. 15, which includes 6235 elements and 5991 nodes. In the FEA
model, the bottom plate, service cabin plates, communication cabin
plates, middle plate, and top plate are made of aluminum alloy honey-
comb sandwich material. The clipboards and central cylinder are made of
carbon fiber reinforced composite material to enhance the stiffness of the
structure. The mass of other subsystems and propellant is modeled by
nonstructural mass (NSM) or lumped mass linked to the structure and Fig. 16. Illustration of the moment of inertia considering the solar arrays configuration.
fuel tanks. Specifically, the solar arrays are modeled as the lumped mass
on the south/north structural plates for modal analysis, while the geosynchronous transfer discipline, GEO station-keeping discipline, and
configuration of the deployed solar arrays is considered when computing attitude control discipline. The design variables x6 ¼ [SH, CH, TBH, SP,
the moment of inertia in Z axis of body-fixed coordinate system as shown CSP, TBP] and constraints y6 ¼ [fX, fY] of structure discipline are sum-
in Fig. 16. marized in Table 7.
The design variables of structural discipline are the honeycomb core
thickness and the ply thickness of the structural sandwich plates and
5. Multidisciplinary analysis approach
central cylinder. The stiffness of the whole satellite must satisfy the en-
gineering requirements, i.e., the first order rotational model frequencies
In view of the feedback coupling state variables y61 and y64 (i.e., total
around body-fixed X,Y axis are no less than 14 Hz respectively. Mean-
mass of satellite), y62 (i.e., center of gravity height), y65 (i.e., moments of
while, the total mass, the mass center height, and the moments of inertia
inertia) according to the DSM in Fig. 3, a MDA process is conducted to
of the satellite are transferred back as the coupled input of the
ensure the system consistency through a fixed-point iteration method.
During the MDA process, the disciplinary analysis is repeated until y61,
y62, y64, and y65 converge to satisfy the interdisciplinary compatibility.
The algorithm of fixed-point iteration based MDA approach is exhibited
in Table 8, and details of the algorithm are presented as follows.

Table 7
Design variables and constraints of structure discipline.

Parameters Type Symbol Unit Range

Core thickness of service cabin SN\EW plates Design SH mm [17,25]

variable
Core thickness of communication cabin SN\EW Design CH mm [17,25]
plates variable
Core thickness of central cylinder Design TBH mm [17,25]
variable
Ply thickness of service cabin SN\EW plates Design SP mm [0.7,0.13]
variable
Ply thickness of communication cabin SN\EW Design CSP mm [0.7,0.13]
plates variable
Ply thickness of bearing cylinder Design TBP mm [0.7,0.13]
variable
First order rotational modal frequency round X Constraint fX Hz 14
First order rotational modal frequency round Y Constraint fY Hz 14
Fig. 14. Structural configuration of the satellite.

310
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Table 8
Algorithm of the fixed-point iteration based MDA process.

Step 1. The initial value of the coupling state variables are given as
[y(0) (0) (0) (0)
61 , y64 , y62 , y65 ]. Set the iteration counter n ¼ 0 and start the MDA
X
15 X
15 X
14 X
15
by ¼ β0 þ βi xi þ βii x2i þ βij xi xj (25)
process. i¼1 i¼1 i¼1 j¼iþ1
Step 2. Given the design variables [x1, x2, x3, x4, x5, x6] and [y(n) 61 ,
y(n) (n) (n)
62 , y64 , y65 ], the state equations of the satellite system are solved to where β are the coefficients determined by least square method, and xi
update the coupling state variables as [y(nþ1) (nþ1) (nþ1) (nþ1)
61 , y64 , y62 , y65 ]. means the i-th design variable in the MDO problem. A threshold
Step 3. If the termination criterion as shown in Eq. (24) is satisfied, objective value is chosen as a “cutting plane” with which a reduced
the MDA process terminates and output the compatible design (i.e., design space is obtained through calling two global optimization pro-
the design variables and the coupling state variables satisfy the state cesses to find the lower and upper bounds for each variable. The sur-
equations); otherwise the iteration counter is increased n ¼ nþ1, and rogate modeling and space reduction process continues till
the process jumps back to Step 2. convergence.

 ðnþ1Þ 
y ðnÞ ðnþ1Þ
61  y 61 y
ðnÞ ðnþ1Þ
62  y 62 y
ðnÞ ðnþ1Þ
64  y 64 y
ðnÞ
65  y 65
max ; ; ;  10
3
(24)
yðnÞ 61 yðnÞ 62 y ðnÞ 64 yðnÞ 65

In ARSM-ISES, the cutting plane approach for ARSM is replaced by

6. Surrogate assisted design optimization of AETS MDO problem
In engineering practices, it is essential and attractive to improve the
overall performance of the AETS via performing conceptual optimiza-
tion, i.e., optimizing the key variables involved in the aforementioned
disciplinary models. Due to use of the time-consuming models (e.g., orbit
dynamics model, structural FEA model, etc.) and the natural iterative
process of AETS MDA, this AETS MDO problem is extremely computa-
tionally expensive. Recently, surrogate-based design optimization tech-
niques (especially the adaptive surrogate-based optimization methods)
have been widely employed to solve expensive engineering design
optimization problems [25–28]. In order to reduce computing expense,
our developed adaptive response surface method with intelligent space
exploration strategy (ARSM-ISES) [34] is used to efficiently solve this
satellite MDO problem. The methodology of ARSM-ISES is briefly
reviewed in Section 6.1. Then the optimization results of the satellite
MDO problem are presented. In the end, the impact of the optimization
is discussed.
6.1. Review of ARSM-ISES
ARSM is based on a second order response surface model (RSM)
described in Eq. (25)
the intelligent space exploration strategy (ISES). ISES inherits the basic
concept of significant design space (SDS) that is a relatively small hy-
percube sub-region where local or global optima are probably located to
implement space reduction without performing global optimization.
Consisting of iterative Maximin sequential Latin hypercube design (IMS-
LHD) sampling scheme, SDS identification algorithm, and tailored
termination criteria, ISES provides ARSM-ISES with a good global
exploration capability to reduce the probability of missing global optima.
Fig. 17 illustrates the flowchart of ARSM-ISES, and the overall procedure
of the algorithm is presented as follows.
Step 1. Given the optimization model, the tuning parameters are set
to configure ARSM-ISES, including the number of initial and newly-
added sample points, the maximum number of sample points, the
maximum constraint violation, etc. Set the iteration counter k ¼ 1.
Step 2. Sample points are generated through a maximin LHD sam-
pling method in the initial design space. The number of initial sample
points is equal to two times of the number of RSM's unknown
coefficients.
Step 3. MDA process is invoked to evaluate responses at initial or
newly-added samples. All the sample points and corresponding re-
sponses are stored in the design library for later use.
Step 4. RSM is constructed or refitted to approximate the augmented
objective F(x) as shown in Eq. (26) according to the samples and their
responses in current design space

311
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Fig. 17. Flowchart of ARSM-ISES.

Step 9. SDS for the next iteration is constructed to reduce the design
X
10

space. SDS is a hypercube sub-region which is determined according
FðxÞ ¼ f ðxÞ þ λ max gj ðxÞ; 0 (26)
j¼1
to the information of X* and X(k)
opt. The SDS identification algorithm is
detailed in the Appendix. Inside the renewed SDS, a set of sequential
where f(x) is the true objective (i.e., total satellite mass), gj(x)0 is the j- samples are collected by an iterative Maximin sequential LHD sam-
th constraint, and λ is the penalty factor. Ten constraints are imposed in pling scheme (IMS-LHD) to refine the RSM, which leads the search
this satellite MDO problem, which are detailed in Table 5. converging to the local or global optima efficiently. Then, the itera-
tion counter k ¼ kþ1, and the process jumps to Step 3 and continues.
Step 5. Sequential quadratic programming (SQP) is employed to
perform RSM-based sub-optimization in the current design space to
find a pseudo optimum X*. The MDA process is invoked to evaluate
Table 9
the true response at the pseudo-optimum, which is then recorded in
Initial and optimal design variables.
the design library.
Design variable Symbol Unit Range Initial Optimal
Step 6. The best design point with the lowest augmented objective
design design
value in the current design library is elected as the potential optimum

X(k) (k) Yaw angle in first-stage of GTO α
opt. Although Xopt may differ from the true global optimum, it is
[0,45] 30 10.19

Pitch angle in first-stage of GTO β [0,45] 30 29.24
recognized to be the closest to the desired solution. As the optimi- 
Pitch angle in second-stage of GTO φ [30,45] 30 43.48
zation proceeds, the accuracy of RSM will be gradually improved, and T position of thruster dT mm [500,1180] 900 659.18
X(k)
opt is expected to be finally identical to the true global optimum. N position of thruster dN mm [800,1050] 1050 1022.28
Step 7. If the termination criteria are satisfied, ARSM-ISES stops to Area of solar arrays Asa m2 [60,90] 90 71.30
Capacity of battery Cs Ah [60,90] 80 75.34
output the last potential global optimum X(k) opt as the best solution.
Area of radiators Ar m2 [5,10] 10 9.39
Otherwise, the process jumps to Step 8 and continues. Angular momentum of reaction Hw N⋅m⋅s [25,50] 50 32.29
Step 8. If X(k)
opt is not feasible, the penalty factor λ is enlarged using the wheel
present optimization information in terms of Eq. (27) Core thickness of service cabin SH mm [17,25] 20 24.02
SN\EW plates


min ηλðkÞ ; λmax smax  stol Core thickness of communication CH mm [17,25] 20 19.88
λðkþ1Þ ¼ (27) cabin SN\EW plates
λðkÞ smax < stol Core thickness of central cylinder TBH mm [17,25] 25 20.09
Ply thickness of service cabin SN\EW SP mm [0.07,0.13] 0.1 0.094
where η > 1 is the predefined increment factor, smax is the maximum plates
constraint violation, stol is the user-defined tolerance, λmax is the Ply thickness of communication CSP mm [0.07,0.13] 0.1 0.095
cabin SN\EW plates
maximum penalty factor. Then the augmented objective values of all
Ply thickness of bearing cylinder TBP mm [0.07,0.13] 0.13 0.082
sample points are updated.

312
R. Shi et al. Acta Astronautica 138 (2017) 301–317

Table 10
Initial and optimal objective.

Objective Initial design Optimal design

Total mass/kg 2543.9 2358.6

ARSM-ISES has been proved to perform better than the state-of-the-

art SBAO algorithms including MPS, EGO, and MSEGO in terms of
global convergence and efficiency. Moreover, ARSM-ISES is able to deal
with the expensive constraints through an adaptive penalty function
method, which significantly enhances the practicality of the algorithm in
solving real-world constrained engineering design optimization prob-
lems. For more details of ARSM-ISES, the reader is recommended to refer
to Long [34].

6.2. Optimization results

In this work, ARSM-ISES is employed for solving the satellite MDO Fig. 18. History curves of objective and maximum constraint violation.
problem. The number of initial sample points is 272, the number of
newly-added sample points at each iteration is 132, and the maximum
constraint violation is 0.05 [34]. ARSM-ISES terminates when the
maximum number of sample points (i.e., 1200) reaches.
The initial design variables are determined according to engineering
experience as shown in Table 9. For the initial design, α, β, and φ of GTO
are set to be 30 ; the four thrusters are mounted near the corners of the
bottom plate; the area of solar arrays, capacity of battery, angular mo-
mentum of reaction wheel, and thickness of bearing cylinder are
respectively the maximum within the limitations to ensure the feasibility
of the initial design; other design variables are approximately set as the
medium values between corresponding lower and upper bounds.
The optimization is executed on HP Compaq dc 7900 computer
equipped with Core 2 Quad CPU (2.83 GHz) and 8G memories. The
computational time is about 34.3 h for the entire optimization process.
The optimization results are listed in Tables 9 and 10 respectively. The
constraints of the initial design and the optimal design are displayed in Fig. 19. Initial and optimized mass of each discipline.
Table 11. Fig. 18 plots the iteration history of objective and maximum
constraint violation.
spends the initial three iterations trying to improve the objective. After
As shown in Table 9, most of the design variables change significantly
that, it mainly focuses on improving the feasibility.
after optimization, and the impact of the changes for the satellite will be
To illustrate the merits of ARSM-ISES, we also employed genetic al-
further discussed in the next section. From Table 11, the optimal design is
gorithm based multidisciplinary feasible method (GA-MDF) to solve this
clearly feasible because all of the constraints are satisfied, and several
all-electric GEO satellite MDO problem. To make a fair comparison, GA-
constraints (e.g., first order rotational modal frequency and ending-of-life
MDF was enforced to terminate and output the history optimum when
power) become active. Besides finding a feasible design, the optimization
the number of MDA evaluations exceeds 1200. The optimized mass from
also yields 185.3 kg decrease in total mass, i.e., about 7.3% of the satellite
GA-MDF is 2396.6 kg, which is 38 kg larger than that from ARSM-ISES. It
total mass, as shown in Table 10. The reduced mass would represent
indicates that ARSM-ISES provides better global convergence than GA-
lower launch cost or more payloads carried on the satellite, which will
produce considerable economic benefits from the customer perspective.
Fig. 18 also depicts that the objective is improved and the constraints are
satisfied when the optimization terminates. The ARSM-ISES optimizer

Table 11
Constraints of initial design and optimal design.

Constraint SymbolUnit Range Initial Optimal

design design

Total orbit transfer time tf Day 180 175.63 160.10


EWSK accuracy λmax 0.05 0.049 0.046

NSSK accuracy imax 0.05 0.008 0.008
Beginning-of-life power PBOL kW 14.15 18.51 15.02
Ending-of-life power PEOL kW 11.90 14.99 12.16
Depth of discharge DOD – 0.8 0.7366 0.7822
Steady-state temperature T K [267,328]316.24 304.38
Angular momentum residue cAC N⋅m⋅s0 3.97 5.69
First order rotational modal frequency fX Hz 14 16.40 14.18
round X
First order rotational modal frequency fY Hz 14 16.29 14.15
round Y
Fig. 20. Optimized orbit transfer trajectory.

313
R. Shi et al.
Fig. 21. Initial and optimal electric thruster positions.
MDF with the same number of MDA evaluations for the AETS
MDO problem.
In summary, our proposed surrogate assisted MDO framework is
proved to be feasible and effective to cope with the all-electric GEO
satellite engineering design optimization issues, because a better design
is obtained and the total mass of the satellite is reduced after optimiza-
tion. The optimization results also demonstrate the effectiveness and
practicality of ARSM-ISES in solving sophisticated satellite system
MDO problems.
6.3. Discussions on optimal design
To quantitatively assess the impact of the optimization, the initial and
optimized mass of each discipline are exhibited in Fig. 19.
The yaw angle α in the first-stage of GTO changes from 30 to 10.19
and the pitch angle in the second stage changes from 30 to 43.48 after
optimization. It indicates that the optimal design prefers to increase the
semi-axis in the first stage of the transfer and decrease the inclination in
the second stage of the transfer, which reduces the total transfer time by
about two weeks and saves over 13 kg propellant. Fig. 20 plots the
optimized orbit transfer trajectory.
The configuration of thrusters also changes significantly after opti-
mization depicted in Fig. 21. Compared with the initial design, dT is
decreased from 900 mm to 659.18 mm, and dN is decreased about
Fig. 22. Rotational modal shapes of X,Y axis for the satellite.
314
Acta Astronautica 138 (2017) 301–317
27.7 mm. Due to the fact that NSSK consumes much more propellant than
EWSK, it is beneficial to improving the efficiency of NSSK maneuvers
from a system perspective by increasing the proportion of thrust in the
normal direction of GEO. As for the optimized thruster configuration, the
thrust component normal to GEO is increased by 9.65%, which saves
over 7 kg propellants for station keeping maneuvers within the lifetime.
In the optimal design, the area of solar arrays and capacity of battery
are decreased by 20.8% and 5.8% respectively, which reduce about
75.5 kg buffer mass of the power discipline. Moreover, the decreased size
of solar arrays means much smaller moments of inertia and the thermal
power of the satellite. Hence, the needed angular momentum of reaction
wheel is only 64.6% of that in the initial design, and the area of radiators
is also decreased by 6.1%, which significantly reduce the buffer mass of
thermal control and attitude control disciplines as shown in Fig. 19.
As for the structure discipline, the plate thickness of service cabin
where TC subsystems and AC subsystems are located is increased by
4 mm to enhance the stiffness of the satellite. On the contrary, the
thickness of other structural components are decreased respectively to
reduce the structure mass. Fig. 16 shows that the total structure mass of
the optimal design is 22 kg lower than that of initial design. Moreover,
the natural frequency constraints are satisfied, and the first order model
shapes after optimization are displayed in Fig. 22. It indicates that the
mass and stiffness distribution of the structure in optimal design is more
reasonable.
7. Conclusions and future work
To efficiently implement the state-of-the-art all-electric GEO satellite
system design, the paper presents a novel surrogate assisted MDO
framework. Based on the proposed framework, we concentrated on the
multidisciplinary modeling and surrogate-based optimization aspects,
which is a rather new endeavor considering the lack of study on new-
generation all-electric GEO satellite system multidisciplinary design
optimization. A two-stage orbit transfer strategy is employed to deter-
mine the GTO, and the GEO station-keeping maneuvers model imple-
mented by EP system is developed based on linearized impulse orbital
dynamics. The degradation of solar arrays is considered in building the
power model based on NASA AP8 approximation model and NIEL
experimental data, and the capacity of the battery is designed to satisfy
the depth of discharge constraint. The thermal network model is estab-
lished to compute the steady-state temperature inside the satellite as a
local constraint, and the needed angular momentums of reaction wheels
are calculated to absorb the accumulated angular momentum in an
unloading cycle considering the effect of deployed solar arrays during the
R. Shi et al. Acta Astronautica 138 (2017) 301–317

transfer. The structural FEA model of the satellite is established to attain
the structure mass and frequency information. Besides, a surrogate-based
optimization method is investigated to reduce the computational cost of
the satellite MDO problem. A detailed discussion of the optimization
results is presented. It shows that a feasible design is obtained and the
total mass yields a 185.3 kg decrease after optimization, which accounts
for about 7.3% of the satellite total mass. The reduced mass leads to a
lighter satellite with lower launch cost, and it could also be dedicated to
additional payloads which means more revenue from customers'
perspective. The optimization results illustrate that the proposed surro-
gate assisted MDO framework is feasible and effective to improve the
quality and efficiency of all-electric GEO satellite system design. It is
believed to be the first time to apply MDO techniques for all-electric GEO
satellites system design, and the work could be referred for further all-
electric spacecraft system research. In future work, we will try to apply
this proposed framework to other spacecraft systems design and
optimization.
Acknowledgement
This work was supported by National Natural Science Foundation of
China (Grant No. 51105040, 51675047, 11372036), Aeronautic Science
Foundation of China (Grant No. 2015ZA72004), and Fundamental
Research Fund of Beijing Institute of Technology (Grant No.
20130142008).

Appendix

The accelerations due to the first four zonal harmonics of non-spherical gravitational potential in earth center fixed (ECI) coordinates are displayed
as follows [46].
      T
3J2 μR2e 5z2 5z2 5z2
aJ2 ¼  x 1  ; y 1  ; z 3  (A1)
2r 5 r2 r2 r2

      T
5J3 μR3e 7z3 7z3 2 7z3 3r 2
aJ3 ¼  x 3z  ; y 3z  ; 6z   (A2)
2r 7 r2 r2 r2 5

      T
15J4 μR4e 14z2 21z4 14z2 21z4 70z2 21z4
aJ4 ¼ x 1  þ ; y 1  þ ; z 5  þ (A3)
8r 7 r2 r4 r2 r4 3r 2 r4

where [x,y,z] is the position of the satellite in ECI, Re ¼ 6378 km is the radius of the earth, J2~J4 are the zonal harmonics coefficients exhibited
in Table A1.
Table A1
First four zonal harmonics coefficients

Harmonics coefficient Value

J2 1.082,626
103
J3 2.535,635
106
J4 1.623,360
106

Table A2
Constants of the NASA AP8 approximation model

Symbol Value

a0 2.094
108 cm2s1

a1 1.673 MeV2
a2 2.07
a3 2.825
b0 0.00971
b1 0.0000982 MeV1
b2 0.01484
b3 0.0001561 MeV1
b4 0.004581
b5 0.0004356
The data in the table is cited from Dutta et al. [1].
Table A3
Parameters of the thermal network model

Symbol Value

ε1 ,ε2 0.86
ε3 ~ε6 0.67
α1 ,α2 0.27
α3 ~α6 0.35
A3, A4 5m2
A5, A6 8.5m2
λ1 ,λ2 1.7 W m1 K1
λ3 ~λ6 1.47
104 W m1 K1
δ1 ,δ2 20 mm
δ3 ~δ6 5 mm

The data in the table is cited from Wu et al. [47]. SDS is a relatively small

315
R. Shi et al. Acta Astronautica 138 (2017) 301–317

hypercube sub-region where local or global optima probably locate. SDS is determined by two components, namely, center (C) and size (L). During the
optimization process, SDS is automatically moved, contracted, or enlarged according to the known information, such as the size of the current design
space, position of the best sample, and fitting quality of the current RSM. The algorithm of SDS is exhibited in Table A4.
Table A4
Algorithm of SDS identification.

References [8] G. Genta, P.F. Maffione, Optimal low-thrust trajectories for nuclear and solar
electric propulsion, Acta Astronaut. 118 (2016) 251–261.
[9] J. Shan, Y. Ren, Low-thrust trajectory design with constrained particle swarm
[1] A. Dutta, P. Libraro, N. J. Kasdin, et al., Design of next generation all-electric
optimization, Aerosp. Sci. Technol. 36 (2014) 114–124.
telecommunication satellites, In: AIAA International Communication Satellite
[10] D.C. Byers, J.W. Dankanich, Geosynchronous-earth-orbit communication satellite
Systems Conference, Florence, Italy, 2013.
deliveries with integrated electric propulsion, J. Propuls. Power 24 (6) (2008)
[2] Boeing Defense, Space & Security, 702SP spacecraft. http://www.boeing.com/
1369–1375.
assets/pdf/defense-space/space/bss/factsheets/702/bkgd_702sp.pdf.
[11] A. Dutta, P. Libraro, N.J. Kasdin, et al., Minimum-fuel electric orbit-raising of
[3] Gunter Dirk Krebs, Hughes / Boeing: HS-702 / BSS-702, HS-GEM / BSS-GEM
telecommunication satellites subject to time and radiation damage constraints, in:
(geomobile), 2016 http://space.skyrocket.de/doc_sat/hs-702.htm 2016 (accessed
2014 American Control Conference, IEEE, 2014, pp. 2942–2947.
16.12.20).
[12] J. Sobieszczanski-Sobieski, Multidisciplinary design optimization: an emerging new
[4] A. Dutta, S. Sreesawet, S. Vijayan, et al., On the design of power and propulsion
engineering discipline, in: Advances in Structural Optimization, Springer,
subsystems of all-electric telecommunication satellites, in: 32nd AIAA International
Netherlands, 1995, pp. 483–496.
Communications Satellite Systems Conference, 2014, p. 4243.
[13] J. Sobieszczanski-Sobieski, R.T. Haftka, Multidisciplinary aerospace design
[5] D.J. Benney, Escape from a circular orbit using tangential thrust, J. Jet. Propuls. 28
optimization: survey of recent developments, Struct. Optim. 14 (1) (1997) 1–23.
(3) (1958) 167–169.
[14] I. Sobieski, Multidisciplinary Design Optimization: Attempt at Definition, Industry
[6] D.F. Lawden, Optimal escape from a circular orbit, Acta Astronaut. 4 (3) (1958)
university workshop on multidisciplinary aircraft design, 1993, pp. 23–48.
218–233.
[15] W. Wu, H. Huang, S. Chen, et al., Satellite multidisciplinary design optimization
[7] T. Edelbaum, L. Sackett, H. Malchow, Optimal low thrust geocentric transfer, in:
with a high-fidelity model, J. Spacecr. Rocket. 50 (2) (2013) 463–466.
10th Electric Propulsion Conference, 1973, p. 1074.

316
R. Shi et al.
[16] F. Castellini, M.R. Lavagna, A. Riccardi, et al., Quantitative assessment of
multidisciplinary design models for expendable launch vehicles, J. Spacecr. Rockets
51 (1) (2013) 343–359.
[17] J.T. Hwang, D.Y. Lee, J.W. Cutler, et al., Large-scale multidisciplinary optimization
of a small satellite's design and operation, J. Spacecr. Rocket. 51 (5) (2014)
1648–1663.
[18] H. Huang, H.C. An, W.R. Wu, et al., Multidisciplinary design modeling and
optimization for satellite with maneuver capability, Struct. Multidiscip. Optim. 50
(5) (2014) 883–898.
[19] K. Kwon, G. Lantoine, R.P. Russell, et al., A study on simultaneous design of a Hall
Effect Thruster and its low-thrust trajectory, Acta Astronaut. 119 (2016) 34–47.
[20] S. Spangelo, B. Longmier, Optimization of CubeSat system-level design and
propulsion systems for earth-escape missions, J. Spacecr. Rocket. 52 (4) (2015)
1009–1020.
[21] Y.C. Choi, K.H. Noh, J.W. Lee, et al., Mission and system design of air-Launching
rocket using multidisciplinary optimization approach, in: 47th AIAA/ASME/ASCE/
AHS/ASC Structures, Structural Dynamics, and Materials Conference 14th AIAA/
ASME/AHS Adaptive Structures Conference 7th, 2006, p. 1722.
[22] N. Yokoyama, S. Suzuki, T. Tsuchiya, et al., Multidisciplinary design optimization of
space plane considering rigid body characteristics, J. Spacecr. Rocket. 44 (1) (2007)
121–131.
[23] R.D. Braun, A.A. Moore, I.M. Kroo, Collaborative approach to launch vehicle design,
J. Spacecr. Rocket. 34 (4) (1997) 478–486.
[24] N.F. Brown, J.R. Olds, Evaluation of multidisciplinary optimization techniques
applied to a reusable launch vehicle, J. Spacecr. Rocket. 43 (6) (2006) 1289–1300.
[25] G.G. Wang, S. Shan, Review of metamodeling techniques in support of engineering
design optimization, J. Mech. Des. 129 (4) (2007) 370–380.
[26] A.I.J. Forrester, A.J. Keane, Recent advances in surrogate-based optimization, Prog.
Aerosp. Sci. 45 (1) (2009) 50–79.
[27] N.V. Queipo, R.T. Haftka, W. Shyy, et al., Surrogate-based analysis and
optimization, Prog. Aerosp. Sci. 41 (1) (2005) 1–28.
[28] F.A.C. Viana, T.W. Simpson, V. Balabanov, et al., Special section on
multidisciplinary design optimization: metamodeling in multidisciplinary design
optimization: how far have we really come? AIAA J. 52 (4) (2014) 670–690.
[29] D.R. Jones, M. Schonlau, W.J. Welch, Efficient global optimization of expensive
black-box functions, J. Glob. Optim. 13 (4) (1998) 455–492.
[30] F.A. C. Viana, R.T. Haftka, L.T. Watson, Efficient global optimization algorithm
assisted by multiple surrogate techniques, J. Glob. Optim. 56 (2) (2013) 669–689.
317
Acta Astronautica 138 (2017) 301–317
[31] L. Wang, S. Shan, G.G. Wang, Mode-pursuing sampling method for global
optimization on expensive black-box functions, Eng. Optim. 36 (4) (2004) 419–438.
[32] H.G. Zhu, L. Liu, T. Long, et al., Global optimization method using SLE and adaptive
RBF based on fuzzy clustering, Chin. J. Mech. Eng. 25 (4) (2012) 768–775.
[33] Y.L. Li, L. Liu, T. Long, et al., Metamodel-based global optimization using fuzzy
clustering for design space reduction, Chin. J. Mech. Eng. 26 (5) (2013) 928–939.
[34] T. Long, D. Wu, X.S. Guo, et al., Efficient adaptive response surface method using
intelligent space exploration strategy, Struct. Multidiscip. Optim. 51 (6) (2015)
1335–1362.
[35] L. Peng, L. Liu, T. Long, et al., An efficient truss structure optimization framework
based on CAD/CAE integration and sequential radial basis function metamodel,
Struct. Multidiscip. Optim. 50 (2) (2014) 329–346.
[36] R.H. Shi, L. Liu, T. Long, et al., An efficient ensemble of radial basis functions
method based on quadratic programming, Eng. Optim. 48 (7) (2016) 1202–1225.
[37] S.A. Feuerborn, D. Neary, J. Perkins, Finding a way: Boeing's all electric propulsion
satellite, in: 49th Joint Propulsion Conference, AIAA-2013-4126, 2013.
[38] X. Ma, D. Han, L. Tang, On the electric-propulsion-based geostationary transfer and
space environmental analysis, Aerosp. Control Appl. 41 (1) (2015) 31–35.
[39] J.X. Yang, Spacecraft Orbit Dynamics and Control, China Astronautic Publishing
House, Beijing, 2001.
[40] J.R. Zhang, S.G. Zhao, Z.C. Zhou, et al., Optimal station keeping for XIPS thrusters
in failure mode under eclipse constraints, J. Aerosp. Eng. (2016) 04016041.
[41] S.R. Messenger, E.M. Jackson, J.H. Warner, et al., Scream: a new code for solar cell
degradation prediction using the displacement damage dose approach, in: 2010
35th IEEE Photovoltaic Specialists Conference, 2010.
[42] I. Jun, M. Xapsos, S.R. Messenger, E.A. Burke, R.J. Walters, G.P. Summers,
T. Jordan, Proton nonionizing energy loss (NIEL) for device applications, IEEE
Trans. Nucl. Sci. 50 (6) (2003) 1924–1928.
[43] S.R. Messenger, E. A Burke, M. A Xapsos, et al., Proton displacement damage and
ionizing dose for shielded devices in space, IEEE Trans. Nucl. Sci. 44 (6Pt1) (1997)
2169–2173.
[44] Z.C. Zhou, Telecommunication Satellite Engineering, China Astronautic Publishing
House, Beijing, 2014.
[45] J.R. Tsai, Overview of satellite thermal analytical model, J. Spacecr. Rocket. 41 (1)
(2004) 120–125.
[46] A. Ghosh, V. Coverstone, Optimal cooperative CubeSat maneuvers obtained
through parallel computing, Acta Astronaut. 107 (2015) 130–149.
[47] W.R. Wu, H. Huang, Analysis and optimization of SSO satellite thermal control
subsystem, Spacecr. Eng. 21 (2) (2012) 44–49.