2dof_isp
TRANSCRIPT
-
7/30/2019 2DOF_ISP
1/6
Robust control and modeling a 2-DOF Inertial
Stabilized Platform
Hamed Khodadadi
Department of Technical and EngineeringIslamic Azad University, Khomeini Shahr Branch,
Isfahan, [email protected]
Mohammad Reza Jahed Motlagh, Mohammad Gorji
Electrical Engineering DepartmentIran University of Science & Technology
Tehran, Iran
[email protected] , [email protected]
Abstract This paper presents an uncertain linear model for an
Inertial Stabilization Platform (ISP). The system has a 2 degree of
freedom (DOF) gimbal which will be attached to a moving vehicle
and the optical sensors mounted on stabilized gimbal. The main
purpose of an ISP is to eliminate the various disturbance and body
motion to make the line of sight (LOS), hold steady in an inertial
space. Due to dynamical model of the 2-DOF gimbal, friction, cable
restraint, noise, other disturbances from the outside environment
and the motions of the vehicles body (as a result of maneuvering or
vibration), the pointing and tracking accuracy of the gimbaled
system may strongly degrade. These disturbances are typically
nonlinear. Modeling and controlling the stabilization loop with
these nonlinearities are the main problem. The approach of this
paper is considering a linear doubted model. Linearity makes the
model simple and so linear classic controller can be used to meeting
design requirements. Therefore the cost of the design and
implementation stages will decrease. In addition uncertainty makes
the model practical and so the nonlinear nature of mentioned
disturbance can be interpreted by this model. So this built ISP by a
PI controller which is capable enough to overcome the disturbances
and to meet the implication requirements will be controlled. For
decreasing the effect of these uncertainties and other probable
variation in ISP's parameters on system performance, a determinedsimple structure for robust enhancer compensator is used and
designed.
KeywordsISP, gimbal, LOS, stabilization loop, nonlinearity,
uncertainty
I. INTRODUCTIONThe main purpose of the optical sensing equipment, (such as cameras,television, radars, lasers and navigation instruments) is to point thesensor from a dynamical platform to a fixed or moving surface. For
achieving accurate pointing, it is necessary to control the sensors LOS.In such an environment where the equipment is typically mounted on amovable platform, maintaining sensor orientation toward a target is a
serious challenge. An ISP is an appropriate way that can solve thischallenge [1, 2].
An ISP is a mechanism, which is used for controlling the inertialorientation of it's payload[1]. In fact, a LOS stabilized platform is a system that maintains thesightline of an optical sensor relative to another object or inertial space
while external disturbances such as accidental vehicle motion andmaneuvers and other additional disturbances subject it.In a multi-axis ISP, there are extra nonlinearities that make thestabilization task more difficult [2]. These nonlinearities arise not only
from the dynamics of gimbaled system but also from the nonlinear
behavior of other disturbance torques such as friction, spring or cable
restraint torques, mass imbalance, vehicle motions and maneuversnoise and Environmental torques. Among these undesired factors, the
gimbal friction, restoring torque and mass imbalance have the moseffect [3].Fortunately in the shaping of our implication mass imbalance does no
exist but the gimbal friction and restoring torque should be consideredprecisely. The papers which focus on ISP modeling or disturbancerejection in stabilization loop use various manners for consideringfriction. Some of them such as [4-6] represented a model by assumingfriction only as viscous or coulomb or combination of them while in
presented model by [2, 3, 7-8], friction assumed as nonlinear part thacan be estimated by using the Kalman filter or can be rejected by usingan adaptive multilayer feedforward neural network. Cable restraint has
been investigated in few papers such as [1, 3, 7 and 9].
Generally, studying and probing a linear differential equation with someuncertainties is simpler than a hardly nonlinear precise one. Forachieving to both benefits (simplicity of first group of paper andaccuracy of seconds), in this paper by using the most complete classica
model for friction in the static case (presented by Olsson et al [10]) andconverting it to the sum of coulomb and viscous models withuncertainty about both and generalizing this approach to other nonlinear
parts, an uncertainly linear model for a 2-DOF gimbaled ISP will bepresented. For reducing these uncertainties, a simple structure for robusenhancer compensator which has been introduced by Zhu et al [11] wil
be used and designed.
This paper is divided in 7 parts. The first is this introduction to thestructure and principle of 2-axis gimbaled ISP for stabilizing LOS. In
the next section, after an introduction of the geometric transformationsused to go from one to another of the coordinate frames, the cinematicrelation between them will be considered. Then dynamical equation o
the 2-DOF system for elevation and azimuth stabilizing gimbaAccording to Yoon and Lundberg [12] will be represented. Thirdsection is devoted to present a real and practical model by notice tomentioned nonlinearities especially friction and restoring torque.
Section four begins with representation a linear and practical model ofstabilization loop with some uncertainty on its parameters. Then, by
doing experimental test for identification of system and by comparingthe result and the presented model with numerical value, the model wil
be validated. Section five assigns to studying the performance ostabilization loop by using the designed PI controller while the externadisturbance and applied to it. Then with considering lower and uppe
band of uncertainty range of uncertain parameters, the performance of
two worse case systems will be shown. For decreasing theseuncertainties, the robustness enhancer feedforward compensator whichhas been introduced by Zhu et al [11] will be used. This compensatornot only attenuates the effect of uncertainty and decrease the existingdifference by a desired factor but also improve disturbance rejection
International Conference on Electrical, Control and Computer EngineeringPahang, Malaysia, June 21-22, 2011
978-1-61284-230-1/11/$26.00 2011 IEEE 223
-
7/30/2019 2DOF_ISP
2/6
This compensator will be designed for the desired implication of thisbuilt ISP in section 6. Finally the ability of disturbance rejection ofcompensated systems (nominal and worse case systems) will be
considered. In the seventh section, the conclusion of this paper will berepresented.
II. CINEMATIC AND DYNAMIC EQUATIONS1. kinematics relations
As we can see in fig. 1, to establish a 2-axis gimbal system threedifferent coordinate frames are used: The gimbal platform or base frame(B), followed by the gimbal mediator or outer frame (K), and finally theinner frame (A) [12]. The optical sensors can be placed on the innergimbal which provides LOS stabilization on yaxis or elevation gimbal.
LOS stabilization on zaxis or azimuth gimbal can be achieved by usingouter gimbal.
In this configuration, 1 is defined as the relative azimuth angle
between the mediator and base coordinate frames where 2 is defined
as the relative elevation angle between inner and mediator coordinateframes. Considering these rotations and by using Euler matrix,
transformation matrixes between the base and mediator frames (K
B C )
and (A
KC ) as the mediator and inner frames become
(1)
,
0
010
0
22
22
=
CS
SC
CAK
=
100
0
0
11
11
CS
SC
CKB
where
(2));cos( 11 =C );sin( 11 =S );cos( 22 =C )sin( 22 =S
by introducing the [ ]/
TB
B I
q r =G
,
[ ]/TK
K I k k kp q r =G
and [ ]/TA
A I a a aq r =G
as the
angular velocity of base, outer and inner frame to the inertial
space's angular velocity1and by using transformation matrixes,
/K
K IG
, /A
A IG
and the inertia matrixes of the outer and inner
gimbals are equal to:
(3)
+=
+=
+=
+=
=
1
11
11
///
GGG
rr
qCpSq
qSpCp
C
r
q
p
k
k
k
IB
BK
BBK
K
k
k
k
IK
K
(4)
+=
+=
=
+=
=
22
2
22
///
CrSprqq
SrCpp
Crq
p
kka
ka
kka
IK
KA
KBA
A
A
A
A
IA
A
GGG
2. Dynamical model of Elevation gimbalYoon and Lundberg [14] indicated that the dynamical equation ofelevation gimbal with attention to torque relationship about y-axis
(controllable axis) can be derived as:
1 Where the component ofIB
B
/G
i.e. p,q,r are traditional notations for roll,
pitch, and yaw in flight dynamics
(5)
2 2( ) ( )
( ) ( )
ay a y az ax a a xz a a
yz a a a xy a a a
I q T I I p r I p r
I r p q I p r q
= + +
+
In this formula, Ty is the total external torque about y-axis and can bereferred as the sum ofTelv (motor torque) and Td1 (external disturbance
torque). Suppose that the products of inertia can be neglected.
(60=== yzxzxy III
Figure 1.A two-axis gimbal typical configuration [9].
Assuming further that
(7)azax II =
Then Eq. (5) becomes
(81delvyaay TTTqI ==
So if Eq. (6) and Eq. (7) are met, the block diagram of elevationequation can be represented as fig. 2. It can be seen that the elevationangular velocity of inner gimbal is only a function of torque motor and
external disturbance and is not affected by the body or elevation gimbamotions.
3. Dynamical model of Azimuth gimbalAccording to Yoon and Lundberg [12], the dynamical equation for
azimuth gimbal can be derived:
(9)321 dldldlzkk TTTTrJ +++=
In this formulaJk is the total inertia of two-axis gimbal about z-axis(Jk= Iaz+ Jkz), Tz is the total external torque about azimuth gimbal andcan be referred as the sum of Taz (motor torque) and Td2 (externadisturbance torque), and finally Tdl1, Tdl2, and Tdl3 represent differen
gimbal inertia disturbances.The simplified model of Eq. (9) with two previous section
assumptions (i.e. Eq. (6) and Eq. (7)) and supposing that the products ofinertia of the azimuth gimbal can be put aside, i.e.
(100=== yzxzxy JJJ
can be represented as:
(11)(
)( 2
kyaxkxkk
akaydazkkzaz
JIJqp
qpITTrJI
++
=+
224
-
7/30/2019 2DOF_ISP
3/6
The block diagram of azimuth equation can be represented as fig. 3.Since the output which should be controlled is ra and not rk, the desiredoutput can be illustrated by using Eqs. (2) and (3) as Fig. 3.
III. REAL AND PRACTICAL MODELAs it was mentioned in pervious sections, total external implementedtorque to each axis can be modeled as sum of the desired torque for
stabilization (Telv or Taz) and disturbance torque (Td1 or Td2). In thissection, disturbance torque and it's various component that might affect
the stabilization loop will be explained. Among these undesired factors,the most complex nonlinear phenomenon is gimbal friction, restoringtorque and mass imbalance [3]. The complete and real model of gimbal
system can be achieved after studying these factors.
1. FrictionDue to the nonlinear friction in the gimbal, the disturbance torque
which has been introduced by Lin and Hsiao [3] is described by usingthe following formula
(12))( hf kfT
=
Modeling this nonlinear function has been mentioned in manyreferences. Olsson et al [10] present four classical models in the staticfriction case. The simplest of them considers the friction only ascoulomb friction (see fig. 4a) while the second of them combines viscousfriction (which is proportional to relative velocity between two movingsurfaces) with coulomb friction as shown in fig. 4b.
The more complete model can be defined by paying attention to thefact that friction force at rest is higher than coulomb friction level. As itcan be seen in fig. 4c, stiction is the static term that must be added to theformer terms.
The real and practical model is given in fig. 4d. Stribeck observedthat friction dose not decrease discontinuously and the velocitydependence is continuously.
As it mentioned briefly in introduction, the papers which focus oninertial stabilized platform modeling or disturbance rejection instabilization loop use various manners for considering friction. Some of
them such as Seong et al [4] represented a model by assuming frictiononly as viscous while [5] considered it as coulomb. In represented model
by Kwon et al [6] coulomb and viscous are the components of friction.Kennedy et al [7] divided friction in two parts: a linear part proportionalto velocity or viscous model and a nonlinear one. Li et al [8] and Orejas[2] used the coulomb friction model and dynamically estimated andidentified this value by using the Kalman filter. Lin and Hsiao [3]introduced a nonlinear friction model and tried to reject the disturbancetorque resulted from it by using an adaptive multilayer feedforwardneural network. Ji and Li [13] believe that the mechanical friction, as atypical nonlinear factor, can hardly realize compensate with the precisemodel, especially under the low velocity.
The approach that is chosen in this paper is to use the most completeclassical model (i.e. Stribeck friction) and to convert it to the sum ofcoulomb and viscous models with uncertainty about both. As it is shown
in fig. 5-d this or any method for continuously decreasing and thenaccelerating friction value can be defined as a combination of a coulombvalue and a velocity- related one with a variant value as:
Figure 2.Dynamical model of elevation gimbal.
Figure 3.Dynamical model of azimuth gimbal.
(13)vFvFT vcf
~
)sgn(
~+=
For example in low speed, the constant value is relatively high
where velocity-related one is negative and increasing in speed leads toreduction in friction quantity. Reversely in high speed, the constant valueis relatively low and velocity-related one is positive. Consequently speedgrowth results in increase of friction quantity. According to Li et al [8]"Friction between moving components not only varies with time but isnonlinear". This statement can encourage us to use an uncertain model.
2. Restoring torqueCable restraint or spring force is the other factor that may affect and
degrade the stabilization. Kennedy et al [7] introduced a model with alinear part related to angle position and a nonlinear one as:
(14)ncrpcrcr FFT +=
Similar to the approach which was represented in previous sectionTcr is expressed as the sum of a constant and a relative part withuncertainty on both parameters.
(15)pcrccrcr FFT~~
+=
3. Elevation practical modelBy combination of Eqs (13), (15) and (7), the dynamical equation ofelevation gimbal can be represented as:
(16)
1 1 1
1 10
sgn( ) ( )
( )
ay a elv di c a k v a k
t
ccr pcr a k
I q T T F q q F q q
F F q q d
=
whereTdi1 = other disturbance torque on elevation axis
1cF = uncertain coulomb part of friction on inner gimbal
1vF = uncertain viscous part of friction on inner gimbal
1ccrF = uncertain constant part of cable restraint on inner gimbal
1pcrF = uncertain relative part of cable restraint on inner gimbal
Figure 4.Four classical model of friction [10].
225
-
7/30/2019 2DOF_ISP
4/6
Figure 5.Practical model of elevation gimbal.
This model can be presented as fig. 5. Similarly, the dynamical equation
of azimuth gimbal can be represented.
IV. STABILIZATION LOOPIn an ISP the goal is to hold the payloads orientation steady in inertial
space [1]. In this case, a direct drive PMDC torque motor for generating
the required torque for each axis has been used. As it can be seen inmany references such as Dorf and Bishop [14], in an armature-controlled DC motor, the armature current generates productive torque.This torque can be spent for gimbal motion and disturbance rejection as
the Eq. (16) for elevation stabilization. The block diagram of speedcontrol (for example in elevation stabilization axis) is presented in fig.6.To simplify the block diagram, it is useful to obtain the transfer function
from motor speed to three inputs (armature voltage, base motion andexternal disturbance).
(17)
1
~)
~~()
~(
)()(
)(
111
2
1
312
den
sK
RFsFRLFKKsIRLFsIL
sKsH
sV
sq
t
apcrvaapcrtbayaavaya
tgs
a
a
=
++++++
==
(18)
1
)(
~)
~~()
~(
)()(
111
2
1
3
1
den
sRsL
RFsFRLFKKsIRLFsIL
sRsL
T
sq
aa
apcrvaapcrtbayaavaya
aa
d
a
+=
++++++
+=
(19)
1
)~~
)((
~)
~~()
~(
)~~
)(()(
11
111
2
1
3
11
den
FsFRsLsKK
RFsFRLFKKsIRLFsIL
FsFRsLsKK
q
sq
pcrvaatb
apcrvaapcrtbayaavaya
pcrvaatb
k
a
+++
=
++++++
+++
=
Since the main purpose of the implicated ISP is stabilization the optical
sensor's speed in order to inertial space, the angular velocity of sensorshould be measured and used in feedback loop. So, two rate gyros (onefor each axis) are installed on the inner gimbal in elevation and outergimbal in azimuth axes.
Fig. 7 schematically shows the components of a stabilization loop. In
this fig, Hg(s) is the transfer function of rate fiber optic gyro, ng is thegyroscope noise and Celv(s) is the elevation controller.By reaching to this point, the most complete and practical model of a
gimbaled 2-DOF ISP with considering some uncertainty on system'sparameters are presented. The most prominent advantage of the
presented model is that this model is completely linear. Linearity makesthe model simple and therefore the linear classic design algorithms can
be used for designing the controller for stabilizing the ISP.Consequently, the costs of design, building and implementation ofcontroller decrease considerably.For example in the next section a conventional PI controller will be
designed for the ISP. Therefore, if the controller is able to satisfy thedesign requirements despite the existent uncertainties, we achieve ourgoals.
Similar to the most ISP control configuration, in this implicationcontrol systems contains two loops: an inner rate or stabilization loopinside an outer track loop. The stabilization loop (which is focused in
this paper) compensates for disturbances and minimizes unnecessarymotion of vehicle. Meanwhile, the role of track loop is remaining thesensor LOS pointed toward the target [1].
For validating the presented model and numerical value of systemparameters, an experimental test has been implemented. In this test byusing an infrared receiver-transmitter and a scaled plate (as an
incremental shaft encoder which has been mounted on the rotating axisa voltage command apply to system. The infrared receiver-transmittersenses and measures the pulses which are result of turning of scaled
plate and send to a FPGA. After counting the pulses, every one micro
second, the counted value transfer to a PC and store in it. In this case astep order with +40v amplitude applied to system.As it can be seen in fig. 8 the recorded data indicate the angular positionand not angular velocity, so for having a fairly comparison between
model and plant, after putting the numerical value of implementedsystem's parameter in Eq. (17), an integrator must be added to it. Theangular position of model and plant due to +40v step command
illustrated in fig. 8. The conformity between them indicates that thepresented model is practical and can be used in modeling 2-DOF
system.
Figure 6.Real model of elevation axis by using a DC motor.
Figure 7.Complete model for elevation stabilization Loop.
Figure 8.Comparison between step response of model and plant.
226
-
7/30/2019 2DOF_ISP
5/6
V. PERFORMANCE OF STABILIZATION LOOPThe main purpose of an ISP is rejecting the various disturbance andfuselage motion to make the LOS, hold steady in inertial space.Moreover, because of the implication of this ISP, the stabilization loopshould have a good bandwidth. In other words, the performance of
stabilization loop as inner loop must be so fast that in inspection oftrack or outer loop the dynamic of inner loop can be ignored. This
requirement leads to having an ISP with maximum 35ms settling time.For a large class of high-precision stabilization loops, classical PI and
PID compensators are usually adequate [1]. As it was mentioned before,due to the achieved linearity of this modeling, a PI controller can beused. If a PI compensator with the transfer function as
( ) (331.8 6.48 ) /C s s s= + is used, a 350-Hz servo bandwidth will
obtained on the inner (elevation) axis with the nominal quantity of ISP's
parameters that have been introduced in previous section.Step response and Bode diagram of nominal system indicates that thestabilized system has the settling time less than 35ms and bandwidthhigher than 350Hz (see fig. 9).
To test the ability of ISP in disturbance rejection, the sum of a 10-Hzpulse with 0.06 amplitude (worse case amplitude of the coulombtorque), a 0.02 constant value (worse case constant part of cablerestraint), a 0.1 random signal as the external disturbance and the sum
of three sinusoidal values with the amplitude of10 rad/s and 0.1, 1 and
10 Hz frequencies (as 10 [sin(10 ) sin( ) sin(0.1 )]t t t + + ) as the
outer gimbal speed in y-axis (with considering these frequencies, theeffect of low, medium and high frequencies of body motion can be
pondered) and a 0.1 random signal as fuselage vibration was
implemented to the elevation stabilization loop. The gyroscope noisecan be assumed to be normally distributed with a zero mean and astandard deviation of0.01 rad/s.As it can be comprehended from fig. 10, the effect of the disturbance
and motion has been attenuated in angular velocity of LOS (system'soutput) considerably.By assuming the lower and higher band for parameters with uncertainty,and calculating the two worse cases that have the most differences with
nominal system, the step and disturbance response of them (with thesame controller of nominal system) are illustrated in figures 11 and 12respectively. Also the bandwidth of system in this case is less than 350
Hz. Comparisons between these figures indicate that disturbanceresponses are still pleasing, but the step responses although have a goodbehavior and this point can be considered as a confirmation of linearmodeling of the 2-DOF gimbal ISP, are different to nominal system andexceed the desired settling time.
To solve this problem, a simple structure robustness enhancerfeedforward compensator will be used. The compensator which has
been introduced by Zhu et al [11] has the ability of uncertainty anddisturbance reduction and can be augmented to main controller.
According to [11] for the design of the robustness enhancer, one can usemany advanced methods. But since the approach of this paper is
presenting a simple and linear model that decrease the design andimplementation cost without degrading in ISP performance, the simple
structure which has no extra cost is appropriate. Another reason forusing robust controller is the typical hazardous environment for most
ISP application. According to [13], the ISP usually works in atrociousconditions, which have many random disturbance factors. So having a
fast dynamic and high stabilization precision ISP which has theadaptability and robustness regarding the model parameter changing, isrequested. Thus in the next section, a robust enhancer controller will bedesigned for the ISP.
VI. ROBUST COMPENSATOR AND SIMULATION RESULTSAccording to the proposed structure in [12], W as the weight of
injected input can be considered as
Figure 9.Step response of nominal system.
Figure 10.Disturbances attenuations of nominal system.
Figure 11.Step response of 2 worse cases.
Figure 12.Disturbances attenuations of worse case System.
227
-
7/30/2019 2DOF_ISP
6/6
(20)12
)1025.0/(100 += mGsW
As it can be seen, Whas the gainK=100, and two fast dynamic for
pre-filter which make the Wproper multiplied to1
mG
as the inverse of
model of system. With the enhanced compensator, step responses of twoworse cases in comparison to nominal system are illustrated in fig. 13.
As it can be seen in fig. 13, the settling times of two worse cases areless than desired value and consequently bandwidths of two systems areas sufficient as high. Besides, the performance of worse case systems isvery similar to nominal system. This means that if the ISP parametersvary in different conditions, the total performance will not be verydifferent. In fact, the ability of uncertainty reduction of the enhancedrobustness compensator, justifies the act of assuming a linear uncertainmodel for stabilization loop.
Figure 13.Step response of nominal and 2 worse Cases.
Figure 14.Disturbances attenuations of compensated system
Taking the disturbance and fuselage motion similar to previoussection and implementing them to the compensated system, it can beseen that the disturbance and motion effects on the system same asuncertainty output are decreased (see fig. 14)
VII. CONCLUSIONSThe details of a 2-DOF gimbal ISP modeling have been considered inthis paper. The presented model is an uncertain linear model that by
considering all nonlinearities torque such as 2-DOF gimbal inertiadisturbances, friction, cable restraint, noise, other disturbances from theoutside environment and vehicle body motion, was represented. Since
the model is completely linear, linear classic controller can be used tosatisfy desired stabilization performance. Therefore the design andimplementation's cost will decrease. Validating the present model with
experimental data and results of testing the performance of the ISP fornominal and worse cases systems with the designed PI controllerindicated that assuming linear uncertain model for 2-DOF gimbal ISP isreasonable. Although a simple structure robustness enhance
compensator (which imposes little extra costs) for decreasing theseuncertainties, improving the performance of system and guarantyingthat act of ISP in hazardous conditions which will be used is pleasingyet has been enhanced to main controller. With this structure, effect o
any undesired variation of ISP's parameters on system performancewill degrade. In fact this paper presents an appropriate solution fodesigning and controlling ISPs. A PI controller that because of linearity
of the presented model can be response of design requirement plus asimple structure robust enhancer that can be decrease the effect of
uncertainties of model parameters and other disturbance on totaperformance.
REFERENCES
[1] Masten, M.K.: Inertially Stabilized Platform for Optical Imaging SystemsIEEE Control Systems Magazine, February 2008.
[2] Orejas, M.E.: UAV Stabilized Platform M.Sc. Thesis of Space Science &Technology Czech Technical University 2007.
[3]
Lin, C.L., Hsiao, Y.H.: Adaptive Feedforward Control for DisturbanceTorque Rejection in Seeker Stabilizing Loop IEEE Transactions on ControSystems Technology, Vol. 9, No. 1, January 2001.
[4] Seong, K.J., Kang, H.G., Yeol, B.Y., Lee, H.P.: The Stabilization LoopDesign for a Two-Axis Gimbal System Using LQG/LTR Controller, SICEICASE International Joint Conference 2006 Oct. 18-21, in Bexco, BusanKorea 2006.
[5] Prasatporn, W., Sangveraphunsiri, V.: Control of Inertial StabilizationSystems Using Robust Inverse Dynamics Control and Adaptive ControlThammasat Int. J. Sc. Tech., Vol. 13, No. 2, April-Jun 2008.
[6] Kwon, Y.S., Hwang, H.Y., Choe, Y.S.: Stabilization Loop Design onDirect Drive Gimbaled Platform With Low Stiffness and Heavy InertiaInternational Conference on Control, Automation and Systems Oct. 17-20in COEX, Seoul, Korea 2007.
[7] Kennedy, P.J., Kennedy, R.L.: Direct Versus Indirect Line of Sight (LOS)Stabilization, IEEE Transactions on Control Systems Technology, Vol. 11
and No. 1, January 2003.[8] Li, B., Hullender, D., DiRenzo, M.: Nonlinear Induced Disturbance
Rejection in Inertial Stabilization Systems, IEEE Transactions on ControSystems Technology. Vol. 6, No.3, May 1998.
[9] Hilkert, J.M.: Inertially Stabilized Platform Technology, IEEE ControSystems Magazine, February 2008.
[10] Olsson, H., Astrom, K.J., Canudas de Wit, C., Gafvert, M., Lischinsky, P.Friction Models and Friction Compensation. European Journal of Control4, pp. 176-195, 1998.
[11] Zhu, H.A., Teo, C.L., Hong, G.S., Poo, A.N.: A Robustness Enhancer forModel-Based Controllers, International Journal of Control, pp.lX3- 1261101. 56, No. 6, 1998.
[12] Yoon, S., Lundberg, J.B.: Equations of Motion for a Two-Axes GimbaSystem, IEEE Transactions on Aerospace and Electronic Systems, Vol. 37
No. 3, July 2001.
[13] Ji, W., Li, Q.: Design Study of Adaptive Fuzzy PID Controller for LOSStabilized System, International Conference on Intelligent Systems Designand Applications, IEEE 2006.
[14] Dorf, R.C., Bishop, R.H.: Modern Control systems, Perentice Hall Incninth Edition, 2002.
228