Page 1
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 210
Direct Fuzzy Logic Controller for Nano-Satellite 1Mohammed Chessab Mahdi, 2Abdal-Razak Shehab, 3Mohammed. J. F Al Bermani
1Foundation of Technical Education Technical Institute of Kufa-Iraq 2University of Kufa - College of Engineering-Iraq 3University of Kufa - College of Engineering-Iraq
[email protected] ; [email protected] ; [email protected]
Abstract-In this paper, a direct fuzzy controller is applied to
the attitude stabilization control of a CubeSat. The Takagi-
Sugeno (T-S) model fuzzy controller is evaluated for attitude
control of magnetic actuated satellites based on the attitude
error including error in the angles and their rates. The
detailed design procedure of the fuzzy control system is
presented. Simulation results show that precise attitude
control is accomplished and the time of satellite maneuver is
shortened in spite of the uncertainty in the system.
Keywords- Fuzzy Logic Controller; PID Controller; Attitude
Control System; Nanosatellite; KufaSat.
I. INTRODUCTION
CubeSats are a class of research spacecraft called Nano
Satellites. The first CubeSats were launched in June 2003 on
a Russian Eurockot, and approximately 75 CubeSats have
been placed into orbit as of August 2012.
The Iraqi student satellite project Kufasat was started in
2012. The launch of the satellite is planned in the near
future. The main tasks for Kufasat will be to perform dust
density measurement and remote sensing. The project is
sponsored by the University of Kufa and it will be the first
Iraqi satellite to fly in space. Kufasat is a Nano-satellite
based on the CubeSat concept. In accordance with CubeSat
specifications, its mass is restricted to 1.3 kg, and its size is
restricted to a cube measuring 10×10×10 cm3. It also
contains 1.5 m long gravity gradient boom, which will be
used for passive attitude stabilization. The satellite attitude
control problem includes attitude stabilization and attitude
maneuver. Attitude stabilization is the process of
maintaining a desired attitude, and the attitude maneuver is
the re-orientation process of changing one attitude to
another [1]. In general, attitude stabilization systems are
classified as active or passive. Active attitude stabilization
requires power while passive do not require any power. The
simplicity and low cost of active magnetic control makes it
an attractive option for small satellites in Low Earth Orbit
(LEO).
A gravity gradient stabilized satellite has limited
stability and pointing capabilities (± 5º) so, magnetic coils
are added to improve both the three-axis stabilization and
pointing properties. Magnetic coils around the satellite's
XYZ axes can be fed with bidirectional constant current
electrical power to generate a magnetic dipole moment ,
which will interact with the geomagnetic field vector to
generate a satellite torque by taking the cross product [2] ,
which is used to control the attitude of the satellite.
Gravity gradient stabilization has been used in attitude
control since the early 1960s [3], but accurate three-axis
control cannot been achieved using gravity gradient
stabilization alone. Gravity gradient stabilization combined
with magnetic torquing has gained increased attention as an
attractive attitude control system (ACS) for small
inexpensive satellites and is also proposed in this satellite
[4].
Because both the direction and the strength of the
geomagnetic field vary as the satellite orbits Earth, the
magnetic control is both non-linear and time dependent.
Attitude control in two-axes only can be achieved when
using three magnetic torquers because the magnetic torques
are constrained on a plane perpendicular to the local
magnetic field. In this paper, a comparison between two
attitude control laws: Proportional-Integral-Derivative
controller (PID) and Fuzzy logic controller (FLC) are
presented. Two well-known fuzzy logic control methods are
Mamdani‟s Fuzzy Inference Systems (FIS) and Takagi-
Sugeno‟s T-S Method. Mamdani type fuzzy inference gives
an output that is a fuzzy set while Sugeno-type inference
gives an output that is either constant or a linear (weighted)
mathematical expression.
A novel approach is presented using direct fuzzy logic
control with the Takagi-Sugeno (T-S) model to control the
magnetic coil current directly.
II. DYNAMIC MODEL
The mathematical model of a satellite‟s attitude is
described by kinetic and kinematic equations of motion
[5]which are now presented. The kinetic equations of
motion for a satellite in LEO is
(1)
where
is the angular velocity of a body-fixed reference
frame to an inertial reference frame.
is the principal axis moment of inertia matrix with
respect to a body-fixed frame .
, where diag is a 3x3 diagonal matrix .
is the total external torque acting on a satellite
expressed in body-fixed reference frame components found
in equation (2) .
Page 2
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 211
(2)
where is the total gravity gradient torque, is
the total magnetic torque , and is the total
disturbance torques , all expressed as components in the
body-fixed frame, Fb.
Equation (1) can be broken down by components into
roll, pitch, and yaw dynamic equations respectively, as
shown in Figure 1 and restated as equations 3a, 3b, and 3c.
Fig. 1 Roll (φ), Pitch (θ) and Yaw (ψ) Angles in body frame
(3a)
(3b)
(3c)
where ωx, ωy, and ωz are angular velocity component of
expressed in the body-fixed frame, Ix, Iy, and Iz, are
the principal moments of inertia expressed in the body-fixed
reference frame, and Tx, Ty, and Tz are the components of
the total external torque, given in Eq.(2), expressed in the
body- fixed frame. Equations (3a, 3b, 3c) are known as
Euler‟s equations of motion for a rigid body [6]. If the Euler
angles ϕ, θ, and ψ are small in magnitude, the relationship
between body angular velocity and Euler angular rates may
be approximated [7] as
(4)
where ωo is the orbital angular velocity and
are the time rates of change of ϕ, θ, and ψ .
A. Gravity Gradient Torque:
The gravity gradient torque, using the aforementioned
small Euler angle approximation and taking principal axes
as reference axes, is given [3] by
(5)
where TGx , TGy, and TGz are the gravity torque (from
Eq. 2) components about the roll, pitch, and yaw axes,
respectively.
B. Magnetic Field Torque:
The magnetic coil produces a magnetic dipole when
currents flow through its windings, which is proportional to
the ampere-turns and the area enclosed by the coil. The
torque generated by the magnetic coils can be modeled
as
(6)
where is the generated magnetic moment vector inside
the body (and written as components in the Body-fixed
Frame, Fb) and is the local geomagnetic
field vector . Eq (7) shows the components of when
written in Fb.
(7)
where Nk is number of windings in the magnetic coil, ik is
the coil current, Ak is the span area of the coil, and k = x, y,
z (i.e. the x, y, z values for these quantities). The magnetic
torque can be represented as
(8)
where Tmx , Tmy, and Tmz are the magnetic torque
components about the roll, pitch, and yaw axes, respectively,
mx, my, and mz are the corresponding components of the
magnetic moments, and Bφ , Bθ , and Bψ is the Earth‟s
magnetic field expressed in body-fixed frame, Fb. After
combining Equations (3), (4), (5), and (8) the final
linearized attitude dynamic model of the satellite, including
the gravity gradient and magnetic coil torques written in
body frame components, becomes
(Roll) (9a)
(Pitch) (9b)
(Yaw) (9c)
Page 3
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 212
, and are the second time derivatives of ϕ, θ, and ψ.
Disturbance torques , magnitudes such as solar
radiation pressure torque and residual magnetic dipole
torque are negligible and therefore have been neglected.
If the system angular position and velocity states are
given as: ; (Roll), ;
(Pitch) ; (Yaw) ,
then the linear system can be expressed as a state space
model
(10)
Where
(11)
(12)
is the control signal defined by
(13)
III. PID CONTROLLER DESIGN
The Proportional-Integral-Derivative controller (PID
controller) is the most widely used feedback control
approach [8].It is also one of the simplest control algorithms.
In the absence of knowledge of the underlying process,
employing one or many PID controllers is often the best
choice to achieve a system designer‟s control objectives. A
typical PID control system structure is shown in Figure 2,
where Kp is the proportional gain, Kd is the derivative gain,
and Ki is the integral gain, where each gain is assumed to be
a positive scalar.
By appropriately adjusting these gains, the desired
output can be achieved while maintaining system stability.
It can be seen that in a PID controller, the error signal e(t) is
used to generate the proportional, integral, and derivative
with the resulting signals weighted and summed to form the
process defined as a „plant model‟ feedback error control
signal u(t) applied.
Fig. 2 PID Controlled System
A mathematical description of the PID controller per [9]
can be found as:
(14)
where u(t) is the plant model input signal and is
defined as the error signal, , where
is the input reference signal. In Equation (14), the
proportional action is related to the error present and is used
to reduce the system‟s rise time. The integral action based
on past error, is used to reduce the system‟s steady state
error, that is, make the system final value closely match its
desired value. Finally, the derivative action, related to
future behavior of error and it is used to improve system
stability. It is also used to drive system overshoot
performance error and improve transient response.
Parameters values used to dynamically simulate Kufasat
performance are listed in Table (1), which explain that the
moment of inertia for one axis, Iz, is significantly smaller
than the other two due to the deployed configuration.
TABLE 1 KUFASAT MOI, ORBITAL ANGULAR VELOCITY AND MASS OF THE GG
BOOM PARAMETERS
Ix
(kg m2)
Iy
(kg m2)
Iz
(kg m2) ωo (rad/sec)
Tip
Mass(g)
0.1043 0.1020 0.0031 1.083*10-3 40
The PID controller is tuned by selecting parameters KP,
Ki, and Kd that give an acceptable closed-loop response. A
Page 4
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 213
desirable response is often characterized by the measures of
settling time, overshoot, and steady state error to mention a
few. Many PID tuning methods have been proposed over
the years, ranging from the simple, but most famous
Ziegler-Nichols tuning method, to the more modern simple
internal model control (SIMC) tuning rules by Skogestad
[10]. In this work, all PID parameters are obtained using
Ziegler-Nichols tuning method. These parameter values are
listed in Table (2) and the post-tuning performance
parameters are listed in Table (3).
TABLE 2 PID CONTROLLER PARAMETERS
Controller type Proportional
gain (Kp)
Derivative
gain (Kd)
Integral
gain (Ki)
PID controller for Roll 0.00216 0.04 0.000004
PD controller for Pitch 0.00064 0.052 ------
PID controller for Yaw 0.0006 0.003 0.000003
TABLE 3 POST-TUNING PERFORMANCE PARAMETERS
Angle
Controller
Delay Time
(Td) sec
Rise Time
(Tr) sec
Peak Time
(Tp) sec
Settling
Time
(TS) sec
Peak
Overshoot
PO%
Steady State
Error %
Roll PID 3 5 10 50 10 0.0
Pitch PD 2.5 4.5 8 10 2.6 0.3
Yaw PID 2 2.5 5 15 21 0.1
The high level attitude control PID Controller (s) SIMULINK block diagram is shown in Figure 3 with further details
reflected in Figure 4.
Fig. 3 Simulink block diagram of satellite model with PID controller
Fig. 4 Complete Simulink diagram of satellite model with PID controller
Page 5
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 214
IV. FUZZY LOGIC CONTROLLER DESIGN
Fuzzy logic controller (FLC) system have been
successfully applied to a wide variety of problems [11]. For
nonlinear problems, many existing experiments have
demonstrated that FLC systems have good performance even
with additive noise [12].
The primary advantage of the FLC is the ability to easily
incorporate heuristic rule-based knowledge from experts in
the control strategy [12] . As a result, fuzzy control is usually
applied to a complex system whose dynamic model is not
well defined or not available at all.
FLC offers the following important benefits, compared to
conventional control techniques:
Developing a FLC is cheaper than developing PID
which need to a sensor for the feedback information.
FLCs are more robust than PID controllers because
they can cover a much wider range of operating
conditions than PID controllers.
FLCs are customizable, because it is easier to
understand and modify their rules, which are
expressed in linguistic terms [12], [13].
Two well-known fuzzy control methods are Mamdani‟s
Fuzzy Inference Systems (FIS) and Takagi-Sugeno‟s T-S
Method [14]. Mamdani‟s method is widely accepted for
capturing expert knowledge. It allows us to describe this
expertise in more intuitive, more like human manner.
However, the Mamdani method‟s fuzzy inference concept
entails substantial computational effort.
The T-S method, originally proposed by Takagi and
Sugeno [15], is computationally efficient and works well with
optimization and adaptive techniques. This makes it very
attractive to solving control problems, especially those using
dynamic nonlinear systems. The most fundamental difference
between the Mamdani and the T-S methods is the way the
crisp output is generated from the fuzzy inputs. While the
Mamdani method uses the fuzzy output "defuzzification", the
T-S method uses weighted average to compute the crisp
output [16]. So the T-S method is typically used when output
membership functions are either linear, constant, or both.
Three T-S Multi Input Single Output (MISO) FLCs, one
for each coil, are used to control each coil‟s current switching
(on, off) and polarity. Each FLC has nine rules and three
linguistic variables. Each FLC‟s three linguistic variables are
for its two inputs, the error (E) and change-in-error (CE), and
it‟s one output, (U) which represents the control action. Each
variable (E, CE, and U) is then represented by a membership
function, as show in Figure 5. Each FLC‟s output is used to
control the roll, pitch and yaw angles, respectively, through
the associated coil current. These variables per FLC are
mapped into three fuzzy set Positive (PO), Negative (NE) and
Zero (ZE) rules. The output variable indicates the desired
magneto-torquer polarity, (HI) for positive polarity and (LO)
for negative polarity. These rules are given in Table (4)
which allows describing the dynamics of the controller.
TABLE 4 RULE BASE FOR THE CONTROLLER OF ROLL, PITCH AND YAW ANGLES
(a) Membership Function for Ephi (Error)
(b) Membership Function for CEphi (Change of Error)
(c) output U
(d) output surface
Fig. 5 Membership functions used in FLC for (a) error E, (b) change of error
CE, (c) output U, and (d) output surface
Page 6
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 215
The attitude control/FLC set up SIMULINK block diagram is shown in Figure 6.
Fig. 6 Satellite model and Fuzzy Logic Controller SIMULINK Diagram
V. SIMULATION
In this section, several simulations of the proposed direct fuzzy control are presented. The FLC Scaling factors were
selected based on trial and error. The values of these scaling factors are listed in Table (5).
TABLE 5 SCALING FACTORS VALUES
Scaling factor GE
Phi
GCE
phi
GU
phi
GE
theta
GCE
theta
GU
theta
GE
psi
GCE
psi
GU
psi
Value 200 2 5 205 30 5 50 1.5 5
where GE is the error scaling factor, GCE is the change-in-error scaling factor, and GU is the output scaling factor. The coil
parameters values used for the Kufasat simulations are listed in Table (6):
TABLE 6 COIL PARAMETERS
Parameter Symbol Value Unit
Length a 85 mm
Width b 75 mm
Mass M 20 g
Wire diameter D 0.1016 mm
Wire resistivity 1.68×10-7 Ω/m
No of turns N 305 turn
Power at full load P 100 mW
Voltage at full load V 4.5 Volt
Coil resistance R 211 Ohm
Coil current I 21.5 mA
Min Temperature Tmin -60 Co
Max Temperature Tmax 80 Co
Page 7
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 216
A. Stabilization Test:
In this test, (0 to 1) rad step input was applied on two cases, first using the PID controller and the second using the FLC
controller. Figure 7 show the system response when using PID and FLC controllers, when the initial conditions of the roll,
pitch, and yaw angles are equal to (0, 0, 0) degrees.
Fig. 7 Attitude response using three PID and three FLC controllers with a 1 rad step input
Page 8
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 217
B. Attitude Control Maneuver (ACM) Test
The Direct Fuzzy Controller (DFC) is designed for any initial Euler orientation and any desired reference attitude. In this
section, the fuzzy controller is tested to achieve different orientations. Figures 8and 9 illustrate Kufasat attitude response to a
small and a large ACM with PID and FLC controllers.
Fig. 8 Response to a small ACM from [0° 5° 0°] to [5° 0° 10°] with three PID and three FLC controllers
Page 9
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 218
Fig. 9 Response to a large ACM from [-20° 50° -10°] to [+20° 0° 60°] with PID controller
The FLC has better performance in terms of percent
overshoot and rise time and minimal steady state error. Even
though the PID controller produces the response with lower
delay and rise times compared with the fuzzy logic
controller, it has a long settling time due high peak
overshoot. In addition to that FLC is controllable and more
stable than PID controller when the system is under effect of
Attitude Maneuver (AM). It is clear from results that the
settling time of FLC is less than the settling time of PID,
that mean the fuzzy controller consume shorter execution
time than the PID.
VI. CONCLUSION
In this paper, a simple direct fuzzy logic controller for
Kufasat attitude control is developed. Then its performance
compared with conventional PID controller. Several
fundamental observations were made based on simulation
and analysis of both controller types.
First, the FLC has good performance, in terms of
minimum overshoot, short rise time and minimal steady
state error. The three FLCs were built using simple direct
fuzzy controller logic with a reduced number of fuzzy rules.
Page 10
Journal of Control Engineering and Technology JCET
JCET Vol. 4 Iss. 3 July 2014 PP. 210-219 www.vkingpub.com © American V-King Scientific Publish 219
This simple structure can reduce the calculation time of the
control action, hence improving the reliability of system.
Second, the FLC has better performance in terms of
percent overshoot and rise time. It is observed that FLC is
controllable and more stable than PID controller when the
system is under effect of Attitude Maneuver (AM). Third,
because roll and yaw are strongly coupled, roll takes a
longer time to achieve acceptable appointing accuracy. Fourth, even though the PID controller produces the
response with lower delay and rise times compared with the
fuzzy logic controller, it has a long settling time due high
peak overshoot. Fifth, CubeSats operate on a strict power
budget due to tight power requirements and limitations.
These systems have relatively limited (small solar array area
available and, limited battery mass and volume allocation).
Kufasat uses two lithium ion cells in parallel as a 2W peak
power and 1.5W average secondary battery. Due to these
power limitations, only one magneto-torquer coil can be
switched on at a time. A control algorithm must be
modified in future to allow for the choice of the coil that
will achieve the best results, given the local geomagnetic
field vector.
REFERENCES
[1] Scrivener, S. L. and Thomson, R C,” Survey of Time-
Optimal Attitude Maneuvers” Journal of Guidance Control
and Dynamics, vol. 17, pp225-233,1994.
http://dx.doi.org/10.2514/3.21187
[2] W. H. Steyn,” Fuzzy Control for a Non-Linear MIMO Plant
Subject to Control Constraints” IEEE Transaction on
Systems, Man, and Cybernetics, Vol.24, No.10, October
1994. http://dx.doi.org/10.1109/21.310540
[3] Hughes, P. C, “Spacecraft Attitude Dynamics”. John Wiley,
NY, 1986.
[4] Skullestad, A and Olsen, K.” Control of GravityGradient
Stabilized Satellite Using Fuzzy Logic,Modeling,
Identification and Control”, VOL.22, NO. 141-152, 2001.
[5] M. Paluszek, P. Bhatta, P. Griesemer, J. Mueller and S.
Thomas, ”Spacecraft Attitude and Orbit Control”,Princeton
Satellite Systems, Inc, 2009.
[6] Marcel J. Sidi “Spacecraft Dynamics and Control APractical
Engineering Approach“, Cambridge University Press, 1997.
[7] Bryson, A. E.”Control of Spacecraft and Aircraft”, New
Jersey, USA: Princeton University press.
[8] Boiko, Igor.” Non-parametric Tuning of PIDControllers”,
Springer, 2013. http://dx.doi.org/10.1007/978-1-4471-4465-6
[9] Katsuhiko Ogata,”Modern Control Engineering”, Pearson
Education International, 2002.
[10] Skogestad, S and Grimholt, C.” The SIMC Method for
Smooth PID Controller Tuning, in PID Control in the Third
Millennium: Lessons Learned and NewApproaches” edited
by Ramon Vilanova,Antonio Visioli, Springer, pp. 147-175,
2012. http://dx.doi.org/10.1007/978-1-4471-2425-2_5
[11] Elmer P. Dadios. ” Fuzzy Logic - Controls,
Concepts,Theories and Applications”, InTech, 2012.
[12] K. M. Passino & Stephen Yurkovich ,”Fuzzy Control”,
Addison Wesley Longman, Inc.1998.
[13] Huaguang Zhang Derong Liu “Fuzzy Modeling andFuzzy
Control “,ISBN-10 0-8176-4491-1 2006.
[14] Yuanyuan Chai, Limin Jia, “Mamdani Model based Adaptive
Neural Fuzzy Inference System and itsApplication”,
International Journal of Computational Intelligence, pp. 22-
29, 2009.
[15] Takagi, T. & Sugeno, M.”Fuzzy Identification of Systems
and Its Applications to Modeling and Control”, Proceedings
of the IEEE Transactions on Systems, Man and Cybernetics,
Vol. 15, 1985.
[16] Haman .A, Geogranas.N.D,”Comparison of Mamdani and
Sugeno Fuzzy Inference Systems for Evaluating the Quality
of Experience of Hapto- Audio-Visual
Applications”,IEEEInternational Workshop on Haptic
Audio Visual Environments and their Applications, 2008.
BIOGRAPHY
Mohammed Chessab Mahdi had his B.Sc.
degree in control and system engineering
from University of Technology –Baghdad at
1984 and had his M.Sc. degree in space
technology from University of Kufa at 2013.
He is full time lecture in Technical Institute of
Kufa -Foundation of Technical Education–
Iraq and member of KufaSat team - space
research unit-Faculty of Engineering
University of Kufa. He has good skills in the design and modeling
of attitude determination and control systems using Matlab
program. He has been published more than 6 researches.
Dr. Abd AL-Razak Shehab body of receive
his B.Sc. from Baghdad University at 1987 ,
M.Sc. and Ph.D. from Saint Petersburg
polytechnic government university (Russia
federal) at 2000 and 2004
respectively .Currently he is full time lecture in
electrical engineering department (Head of
department since 2007) – Faculty of
Engineering –Kufa University–Iraqi ministry of
high education and scientific research. He has good skills in the
design and modeling of control systems and switched reluctance
motor (SRM). He has been published more than 9 researches.
Dr. Mohammed. J. F Al Bermani body of
receive his B.Sc. from Baghdad University at
1971, M.Sc. from Aston University,
Birmingham, UK at 1983 and Ph.D. from
Baghdad University at 1997. Currently he is
full time lecture and head of KufaSat team -
space research unit-Faculty of Engineering
University of Kufa. Iraqi ministry of high
education and scientific research. He has good
skills in Attitude determination and six degrees of freedom
dynamics of spacecraft. He has been published more than 25
researches