Professional Documents
Culture Documents
(Legal Code) Disclaimer
(Legal Code) Disclaimer
(Legal Code) Disclaimer
0 대한민국
Disclaimer
Ph.D. Dissertation
Daehee Won
August 2015
Robust Backstepping Control for
Electro-Hydraulic Servo Systems using
Nonlinear Observer
by
Daehee Won
August 2015
Copyright
c 2015 by Daehee Won
All rights reserved.
Robust Backstepping Control for
Electro-Hydraulic Servo Systems using
Nonlinear Observer
by
Daehee Won
Rae-Young Kim
(Co-Chair of Committee)
Sangdeok Park
(Member)
Wonhee Kim
(Member)
CONTENTS I
LIST OF FIGURES IV
SYMBOLS VI
ABSTRACT VIII
1 Introduction 1
I
Contents
II
Contents
6 Conclusions 108
APPENDIX 110
BIBLIOGRAPHY 120
III
List of Figures
IV
List of Figures
V
Symbols
VI
Symbols
Cel
Ctl total leakage coefficient (i.e., Ctl = Cil + 2 ) [m5 /Ns]
w area gradient of the servo-valve spool [m]
VA (xp ) volume of cylinder chamber A in xp m3
VB (xp ) volume of cylinder chamber B in xp m3
xp position of the piston m
App area of the piston m2
Ar area of the piston rod m2
VA0 initial volume of chamber A m3
VB0 initial volume of chamber B m3
Ap pressure area of the piston m2
PL load pressure (i.e., PL = PA − PB ) N/m2
FL external load force N
J0 rotational inertia of the mechanical system kgm2
J1 rotational inertia due to the payload kgm2
m mass of the piston kg
k load spring constant N/m
b viscous damping coefficient N/(m/s)
x1 position of the piston m
x2 velocity of the piston m/s
x3 load pressure N/m2
x4 spool position of the servo-valve m
x5 velocity of the servo-valve m/s
u current input mA
VII
Abstract
Robust Backstepping Control for Electro-Hydraulic
Servo Systems using Nonlinear Observer
Daehee Won
Department of Electrical Engineering,
Graduate School of HANYANG UNIVERSITY
Directed by Professor Chung Choo Chung
IX
To my wife Younghee
and to our children Yeseul and Jong-hyeok
X
CHAPTER 1
Introduction
1
Chapter 1. Introduction
Previous Works
Various control methods have been used to control the position or force of an EHS. The
adaptive linear control theory has been used to deal with parameter variations. However,
important dynamic information is lost in the linearization process, which favors nonlinear
control methods. Local linearization of the nonlinear dynamics about a nominal operating
point was proposed in [1]. However, the nonlinear behavior of the system requires the use
of a conservative loop gain that sacrifices controller performance in favor of robustness.
In [5, 6], the use of pressure feedback was proposed in order to improve the performance of
proportional integral derivative (PID) controllers in EHSs, but stability was not proven.
Adaptive control is a valid method to overcome systematic uncertainties, especially un-
certainties derived from uncertain parameters; hence, several kinds of nonlinear adaptive
control schemes have been proposed in hydraulic control systems such as feedback lin-
earization adaptive control. An indirect adaptive control incorporating a linear model of
an EHS was designed for position tracking [7]. However, the linear controllers using pole
placement design contain certain limitations to ensure tracking accuracy and robustness
of the controller, especially for highly nonlinear problems.
Therefore, nonlinear robust controllers, which ensure the stability and robustness of
the system, were developed owing to their primary advantage of manipulating nonlin-
earities in the system [8–12, 17, 19–24]. Two powerful robust control designs based on
Lyapunov functions emerged: variable structure control (VSC) and backstepping control.
VSC strategies have been studied for the control of EHSs [8–12]. However, chattering in
the control action, which is inherent in the VSC, can easily excite high frequency modes
and may result in system instability. Input-output (IO) linearization has also been used
in the design of EHSs [14, 15, 50, 51].
Various methods based on Lyapunov functions have been studied for EHSs where the
valve dynamics was considered to make EHS an input affine system. Backstepping con-
trollers, which provide a powerful control strategy for handling the nonlinearities of the
EHS system were introduced in [17, 19–24] and were applied to EHS for exponential
convergence of the desired position or load pressure. In [23], a special type of Lyapunov
function has been used to construct a Lyapunov-based controller and parameter update
law for nonlinear parameters. The controller combines the backstepping method, to pro-
vide the whole system controller, and all unknown parameter update laws. A simple
robust method was used to solve the problem of unmodeled uncertainties. The system
showed improved performance over controllers with only linear parameter adaptation.
Although long and complex, backstepping controllers can improve the position/force
tracking performance of the EHS [17].
2
Chapter 1. Introduction
The passivity-property has been used in many nonlinear physical domains to derive
useful and robust control laws. Using the mechanical systems’ physical energy func-
tions and their modification as Lyapunov functions, a passivity-property can be derived,
from which a whole class of fixed and adaptive control laws with rigorous analysis and
arbitrary gains has been obtained [18]. In the area of hydraulic systems, other than
approaches based on linearization and a linear system’ assumption, the typical nonlin-
ear control approach is based on backstepping [19, 22]. In these methods, the desired
force from the actuator is first designed, and then the actuator force is controlled via
backstepping. Usually, a simple quadratic term in the actuator force error is used in the
Lyapunov function, and nonlinearities are canceled out to preserve stability. In other
words, the physical energetic structure of the actuator pressure dynamics has not been
used. Passivity-based control is a design technique that uses passivation to achieve the
control objective [26, 27]. This technique has been successfully applied to a nonlinear
model and was shown to be very effective for the EHS [27–29]. These methods need to
measure the full state information of the system. Passivity-based control is simpler and
more straightforward than a backstepping algorithm, since the synthetic inputs in the
passivity-based control are decoupled. Thus, modeling error effects associated with the
controller are compartmentalized [28].
Although previous nonlinear control methods improved the position tracking perfor-
mance of EHSs, there are three primary issues in the EHS control:
• Output feedback
(1) Output Feedback: Many practical applications require estimation of the states
and parameters that can be used in designing a stable control algorithm; the unmea-
surable states and parameters are generally estimated based on the knowledge of the
physical system, such as a model, and the available measurements. The majority of pre-
vious methods require full state feedback [8–12, 14, 15, 17, 19–24, 26–29, 50, 51]. However,
it is not always possible to measure the full state of the EHS because of cost and space
limitations. Furthermore, measurement noise is associated with the pressure sensor and
it is impossible to measure the spool position of the servo-valve unless the position sen-
sor is embedded in the servo-valve. To overcome this problem, several observers were
designed to estimate the full state of active suspensions and EHS [32–35].
3
Chapter 1. Introduction
The extended Kalman filter (EKF) cannot be used to estimate the state from poor
a priori estimations or when large unmodeled perturbations occurs. This is because the
EKF relies on linearization to propagate the mean and covariance of the state. The
unscented Kalman filter (UKF) can give improved estimation performance for nonlinear
systems, since the EKF requires the computation of Jacobians and the UKF does not
use Jacobians. However, UKF is not practical for implementation in embedded control
systems, because this method requires heavy on-line computational effort.
The high gain observer has robust performance when poor a priori estimations or large
perturbations occurs [36–38]. Since only observer designs were studied in those studies,
only the stabilities of estimation error dynamics were proven. If the nonlinear control
method is used with the observer in the EHS, the closed-loop stability should be studied
while considering both tracking error dynamics and estimation error dynamics. Several
observer-based nonlinear controllers were proposed [39, 40]. In [39], only acceleration was
estimated by the observer. Alternatively, a proportional integral (PI) observer was de-
signed to estimate the full state [40]. Although the estimation performance was validated,
the closed-loop stability was not proven.
(2) Load force, Friction Disturbance and Parameter Uncertainties: Recently,
controllers were designed for use in the presence of friction and internal leakage [54]. This
method required knowledge of the friction. All of these methods improve the position
tracking performance without considering their disturbances, friction, or load torque.
However, when significantly affected by the disturbance, the position tracking perfor-
mance is degraded, since the force or torque available for the systems is diminished.
Therefore, a method for compensating for the disturbance is needed in order to improve
EHS performance.
In practice, since direct measurement of the disturbance is not possible, it is necessary
to estimate the disturbance using adaptation or estimation algorithms. Several distur-
bance estimation methods for EHS have been proposed [12, 55–58]. A constant friction
estimator was proposed, and was found to be effective for estimating non-constant friction
when the velocity is not zero [56]. Adaptation methods based on the Friedland method
were proposed for friction estimation [12, 57]. In another work, the constant load torque
was estimated by an observer based on a mathematical model and was compensated for
by a feed forward-loop [58].
If both sinusoidal load torque and constant friction are considered, the disturbance
becomes a biased sinusoidal signal. Although all of the methods are effective in com-
pensating for either a constant or pure sinusoidal disturbance, they cannot be used to
4
Chapter 1. Introduction
5
Chapter 1. Introduction
measurement noise. Low-pass filters are generally used to reduce the measurement noise.
Although the measurement noise of the measured signal is suppressed by filtering, the
measurement noise may be amplified when the measured signal’s derivative is used. Thus,
a disturbance observer is required that does not use the derivative of the measurement
signal. Recently, a high gain observer based method was studied to overcome this prob-
lem [72].
In Chapter 2, an overview of the system and its subsystems are presented. The
system is analyzed and modeled in three separate subsystems: the servo-valve subsystem,
hydraulic subsystem, and mechanical subsystem. Each subsystem has its own section in
this chapter. The mathematical models of three subsystems are introduced.
In Chapter 3, we present a position tracking control based on an output feedback
nonlinear control using the passivity-based control method. This dissertation solves the
problem for linear EHSs and focuses on the gain selection of the passivity-based con-
troller for the output feedback control and the proof of closed-loop stability. The high
gain observer is designed to estimate the full state, and the high gain technique is used to
reduce the effects of the nonlinear terms. The passivity-based controller is implemented
for position tracking. Although the passivity-based control is simpler and more straight-
forward than the backstepping algorithm, the passivity-based controller guarantees only
the asymptotic stability of the tracking error dynamics. Therefore, it is difficult for the
passivity-based controller to be designed with a high gain observer since the exponential
stabilities of the quasi-steady-state model and the boundary layer system are required
for the stability of the singular perturbed system [36]. We solve the problem by selecting
6
Chapter 1. Introduction
the controller gain such that the origin of the tracking error dynamics is exponentially
stable. We prove that the closed-loop system is asymptotically stable (under the selection
criteria) using the singular perturbation method. Therefore, the observer gain is selected
considering the controller gain.
In Chapter 4, we discuss the design and implementation of a nonlinear position
tracking controller with a disturbance observer (DOB) for EHSs in the presence of a
biased sinusoidal disturbance. The proposed method is designed for implementation on
a rotational joint driven by a linear EHS. In this type of EHS, the main disturbance
consists of constant friction and sinusoidal load torque and is thus a biased sinusoidal
signal with unknown frequency. The DOB, whose simplicity results from it requiring only
mechanical parameters, is designed as a second-order high pass filter for estimating the
disturbance without any additional algorithms. The nonlinear controller is designed to
track the desired position, operating as a near IO linearizing inner-loop load pressure
controller and a backstepping outer-loop position controller. Variable structure control
is implemented in order to compensate for errors in disturbance estimation. The desired
load pressure is generated using the differential flatness property of the EHS mechanical
subsystem. The disturbance includes not only a biased sinusoidal signal but also the
effects of modeling assumptions and the parameter uncertainties within the bandwidth
of the DOB; it can be canceled out using the proposed method.
In Chapter 5, we propose a high gain disturbance observer (HGDOB) based robust
backstepping control, with a position tracking error constraint, to improve the position
tracking performance in the presence of disturbances. The proposed methods guarantee
the tolerance for the position tracking error in the presence of disturbances. The proposed
method consists of a HGDOB and a backstepping controller. The HGDOB is designed to
estimate the disturbances, which include friction, load force, and parameter uncertain-
ties (i.e., unknown internal/external leakage, unknown volumes in the orifice equation,
and temperature dependent oil characteristics). Auxiliary state variables are proposed
in order to avoid amplification of the measurement noise in HGDOB. In order to com-
pensate for the disturbances while guaranteeing tolerance of the position tracking error,
a backstepping controller using a BLF is proposed. As a result, the proposed method
satisfies the output constraint in the presence of disturbances and improves the position
tracking performance. In addition, the effect of input saturation caused by disturbances
can be minimized using this method.
The performance of the proposed methods is verified through simulations and validated
through experiments. Compared to previous methods, the proposed method improves the
position tracking performance.
7
CHAPTER 2
Modeling of Electro-Hydraulic
Servo Systems
An overview of the electro-hydraulic servo system and its subsystems are presented.
The system is analyzed and modelled in three separate subsystems: servo-valve sub-
system, hydraulic subsystem, and mechanical subsystem. Each subsystem has its own
section in this chapter. The mathematical models of three subsystems are introduced.
2.1 Introduction
Hydraulic actuators transform hydraulic power into mechanical power based on Pas-
cal’s law, which states, "any change in pressure at any point of an incompressible and
confined fluid at rest is transmitted equally in all directions." Even though real fluids
are very slightly compressible, compared to their electrical counterparts, they also have
a high power-to-weight ratio [1, 2]. Furthermore, they are able to rapidly generate very
large forces. Thus, they are important in modern industrial automation and have been
used in many types of mechanizations, including hydraulically actuated robots, excava-
tors, and rolling mills. Unfortunately, hydraulic servo systems also have disadvantages
compared to electrical actuators. A significant disadvantage is that they behave nonlin-
early. Furthermore, hydraulic servo systems often have friction, oil leakage, and show
position dependent behavior.
A typical hydraulic system consists of a pump, one or more control valves, and a
hydraulic cylinder [1]. The flow of hydraulic fluid into the cylinder chambers, which are
connected to a servo-valve through cylinder ports, is controlled by the servo-valve and
is influenced by load pressure, which is the difference between the pressures of the inlet
and outlet cylinder. The force of the actuator is produced by the load pressure. In this
8
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
section, we introduce a nonlinear mathematical model for the EHS. An overview of the
system and its subsystems are shown in Figure A.1. The system is analyzed and modelled
in three separate subsystems, each having its own section in this chapter. The valve is
also a subsystem of the hydraulic circuit, but it is modelled separately.
• Servo-valve subsystem
• Hydraulic subsystem
Besides the three modeled subsystems, the system also includes a digital control unit.
This unit works in discrete time steps and introduces sampling effects. However, the
effects of sampling are ignored, since it is assumed that the control period (sampling
period) is small; therefore, all modeling and analysis is considered in continuous time.
The first subsystem of the model to be presented is the servo-valve subsystem.
9
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
The valve is the hydraulic component in charge of controlling the hydraulic power,
supplied by the pump, that goes to the actuators. The most common type of valve
employed on hydraulically actuated machinery is a flow-control servo-valve. These valves
regulate the output flow according to a control signal input. These valves have four ports
(P , T , A, and B), and the valve spool is hydraulically driven by a pilot stage. Port A
connects the valve to actuator chamber A, and port B connect the valve to actuator
chamber B, while port P connects the valve to the pump, and port T connects the valve
to the reservoir. The pressure at the valve ports A, B, P , and T are named as PA , PB , Ps ,
and Pr respectively. The valve input, i, is the input current of the torque motor. Flapper-
type M200 and M454 servo-valves manufactured by Star Hydraulic Ltd. [75], and a
schematic for the two-stage, four-way nozzle flapper type valve are depicted in Figure 2.2.
The dynamic response of a flow control servo-valve can be approximated in the frequency
10
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
where
qL load flow rate of the servo-valve [m3 /sec];
kf servo-valve static flow gain at zero load pressure drop [(m3 /sec)/mA];
τ servo-valve time constant [sec];
i input current of the torque motor [mA].
Standard flow control servo-valves are available in several sizes, with many internal de-
sign configurations. The value of servo-valve sensitivity kf depends on the rated flow
and input current. The appropriate time constant for representing servo-valve dynamics
will depend largely upon the flow capacity of the valve. If it is necessary to represent
servo-valve dynamics through a wider frequency range, a second-order response can be
used [41], as:
ωn2
qL
(s) = kf (2.2)
i s + 2ζωn s + ωn2
2
where
ωn natural frequency [rad/sec];
ζ damping ratio.
Figure 2.3 shows a hydraulic schematic of a valve spool. In a servo-valve, the spool
is a movable part located inside the valve housing. By sliding back and forth, it can com-
pletely or partially block the valve channels, setting the flow direction and magnitude.
The spool moves inside the valve housing according to the electrical input u applied to
the valve. Although nonlinear model of the mechanical behavior of servo-valve can be
developed from the relationship between the spool position, xv , and the input current,
i. The valve dynamics can be approximated as second order linear dynamics by
ωn2
xv
(s) = kv (2.3)
i s + 2ζωn s + ωn2
2
where
xv spool position of the servo-valve [m];
i input current of the torque motor [mA];
kv torque motor gain [m/mA].
11
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
The port flows are formed by the orifice flows through the equations
qs = qa (xv , Ps , Pa ) + qb (xv , Ps , Pb ),
qr = qc (xv , Pa , Pr ) + qd (xv , Pb , Pr ),
(2.4)
qA = qa (xv , Ps , Pa ) − qc (xv , Pa , Pr ),
qB = qd (xv , Pb , Pr ) − qb (xv , Ps , Pb )
where
Ps supply pressure of the valve [N/m2 ];
Pr return pressure of the valve [N/m2 ];
PA and PB pressure of the pressure area A and B [N/m2 ];
qs and qr supply and return flow [m3 /sec];
qA and qB flow out and in of the valve [m3 /sec];
qa , . . . , qd flow through the valve orifices [m3 /sec].
Figure 2.3 shows a hydraulic schematic of the valve spool for the two-stage, four-way
nozzle flapper type valve. In Figure 2.3, for steady state flow analysis by assuming that
the compressibility flows are zero, then we obtain:
qA = qa (xv , Ps , PA ) − qc (xv , PA , Pr ),
(2.5)
qB = qd (xv , PB , Pr ) − qb (xv , Ps , PB ).
We assume that the valve orifices with critical center are matched and symmetrical,
which is the case for most of the spool valves manufactured.
12
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
Flows through the valve orifices are described by the orifice equation.
r
2
qa (xv , Ps , PA ) = Cd wxv (Ps − PA ),
ρ
r
2
qb (xv , Ps , PB ) = Cd wxv (Ps − PB ),
ρ
r (2.6)
2
qc (xv , PA , Pr ) = Cd wxv (PA − Pr ),
ρ
r
2
qd (xv , PB , Pr ) = Cd wxv (PB − Pr ).
ρ
Substituting (2.5) into (2.6), the control flow equation of both chambers is given by
r r
2 2
qA = Cd w sgn (xv ) (Ps − PA ) + sgn (−xv ) (PA − Pr ) ,
ρ ρ
r r (2.7)
2 2
qB = Cd w sgn (xv ) (PB − Pr ) + sgn (−xv ) (Ps − PB )
ρ ρ
where
1 (xv > 0)
sgn(xv ) = 0 (xv = 0) (2.8)
−1
(xv < 0)
Ps is the pressure difference between the supply pressure of the pump Ppump and the
return line pressure Pr , i.e, Ps = Ppump − Pr . In general, the return line pressure can
be neglected because it is usually much smaller than the other pressures involved. The
position of the spool xv cannot be measured directly in this valve. If the orifices are both
13
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
qa (xv , Ps , PA ) = qd (xv , PB , Pr ),
(2.9)
qb (xv , Ps , PB ) = qc (xv , PA , Pr ).
Ps = PA + PB (2.10)
PL = PA − PB (2.11)
yields,
Ps + PL Ps − PL
PA = and PB = . (2.12)
2 2
Assume further that the valve geometry is ideal (i.e., the orifice edges are perfectly square
with no rounding and there is no radial clearance between the spool and the sleeve), then
the leakage flows (qa and qb when xv is positive, qa and qb when xv is negative) are zero,
then combining (2.9), (2.11), and (2.12) yields:
s
2 Ps − PL
qL = Cd Aa for xv > 0. (2.13)
ρ 2
For negative spool positions, qL = −qd , and (2.5) and (2.11) give:
s
2 Ps + PL
qL = −Cd Ab for xv < 0. (2.14)
ρ 2
Because the valve is assumed symmetrical, (i.e„ Axv = Aa (xv ) = Ab (−xv )), (2.13)
and (2.14) can be combined into a single relation:
s
xv 1 xv
qL = Cd |A| Ps − PL . (2.15)
|xv | ρ |xv |
14
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
X X dV0 V0 dP
qA − qB = + (2.17)
dt βe dt
where
V0 initial volume [m3 ];
xpmax maximum stroke of the cylinder [m];
βe effective bulk modulus [N/m2 ].
where
Cil internal leakage coefficient [m5 /Ns];
Cel external leakage coefficient [m5 /Ns];
VA (xp ) volume of cylinder chamber A [m3 ] in xp ;
VB (xp ) volume of cylinder chamber B [m3 ] in xp ;
xp position of the piston [m].
The volumes of the piston chambers, VA (xp ) and VB (xp ), can be expressed as
where
App area of the piston [m2 ];
Ar area of the piston rod [m2 ];
VA0 and VB0 initial volumes of chamber A and B [m3 ];
Ap pressure area of the piston [m2 ].
Assuming that the cylinder has a constant piston pressure area Ap throughout the op-
15
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
Remark 2.2. If the external leakage terms Cel PA and Cel PB are negligible, then we can
rewrite the load flow continuity equation as:
dPA βe dxp
= q A − Ap − Cil (PA − PB ) ,
dt VA (xp ) dt
(2.22)
dPB βe dxp
= −qB + Ap − Cil (PB − PA ) .
dt VB (xp ) dt
1
qL = (qA + qB ) . (2.23)
2
16
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
1
qL = (qA + qB )
2
1 VA (xp ) dPA VB (xp ) dPB dxp
= − + Cel PL + 2Cil PL + 2Ap (2.24)
2 βe dt βe dt dt
VB0 − Ap xp dPB
1 VA0 + Ap xp dPA dxp
= − + Cel PL + 2Cil PL + 2Ap .
2 βe dt βe dt dt
Remark 2.3. If the initial fluid volumes in both chambers are equal (i.e., VA0 = VB0 =
V0 ), then the total volume of fluid in both chambers is Vt = VA + VB = 2V0 . This means
xpmax
that the initial position of the piston is xp0 = 2 . Then, we can rewrite the load flow
continuity equation as:
Vt dPL Cel dxp
qL = + Cil + PL + A p . (2.25)
4βe dt 2 dt
Combining the control flow in (2.7) and the load flow continuity in (2.21), the fluid
dynamic equation of the actuator is given by
dPA βe AP dxp βe C t
=− − (PA − PB )
dt VA (xp ) dt VA (xp )
r r
βe C d w 2 2
+ sgn (xv ) (Ps − PA ) + sgn (−xv ) (PA − Pr ) xv ,
VA (xp ) ρ ρ
(2.26)
dPB βe AP dxp βe C t
= + (PA − PB )
dt VB (xp ) dt VB (xp )
r r
βe C d w 2 2
− sgn (xv ) (PB − Pr ) + sgn (−xv ) (Ps − PB ) xv .
VB (xp ) ρ ρ
Remark 2.4. If a servo-valve has matched and symmetrical orifices, and the initial
xpmax
position of the piston xp0 is 2 , then we can rewrite the fluid dynamic equation of the
actuator as:
r
dPL 4βe Ap dxp 4βe Cel 4βe Cd wxv 1
=− − Cil + PL + (Ps − sgn(xv )PL ). (2.27)
dt Vt dt Vt 2 Vt ρ
17
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
Finally, by applying Newton’s second law, the actuator’s force balance equation is
where
xp position of the piston [m];
m mass of the piston [kg];
k load spring constant [N/m];
b viscous damping coefficient [N/(m/s)];
FF friction force [N];
FL external load force [N].
18
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
θ+α+β =π
where
l1 length of the torque arm [m];
c length of the cylinder body [m].
θ (xp ) = π − α − β
b2 + l12 − (xp + c)2
a
−1
= π − cos − tan−1 (2.31)
2bl1 h
2 2 2 2
a + h + l1 − (xp + c)
a
= π − cos−1 √ − tan−1 .
2l1 a2 + h2 h
Using the joint angle θ as the coordinate, the kinetic energy of the rotational motion can
be expressed as:
1
T = (J0 + J1 ) θ̇2 (2.33)
2
where
J0 rotational inertia of the mechanical system [kgm2 ];
J1 rotational inertia due to the payload mounted at the end
of the mechanical system [kgm2 ].
Since the other two mechanisms are fixed, J0 and J1 are constant; J1 is unknown since
19
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
it depends on the unknown payload of the mechanical system. In terms of the piston
position xp , the kinetic energy of the rotational motion can be expressed as
2
1 ∂θ
T = (J0 + J1 ) ẋ2p (2.34)
2 ∂xp
where the fact that θ is uniquely related to xp has been used, i.e., θ(xp ). Using Euler-
Lagrange equation, the dynamics of the rotational motion can be described by
2
∂θ
(J0 + J1 ) ẍp = − kxp − bẋp + Ap PL − FF (ẋp )
∂xp
(2.35)
∂θ ∂ 2 θ 2
− (J0 + J1 ) ẋ .
∂xp ∂x2p p
2
∂θ
The effective mass driven by the hydraulic cylinder is (J0 + J1 ) ∂x p
, which could
∂θ
change greatly during operations since ∂x p
is a nonlinear function of the piston po-
sition xp . Furthermore, it is unknown due to the unknown rotational inertia J1 caused
by the payload. In (2.31) and (2.32), the relationship between xp and θ has significant
nonlinear terms. Thus, a hydraulically actuated mechanical system has significant non-
linear behavior. In this dissertation, all nonlinear terms caused by a complex mechanical
structure were allocated to the disturbance.
20
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
Ps Ps
Pr Pr
Ps
xv
qA qB
A
CelPA CelPB
PA PB
m
VA CilPL VB
b
xp
2.5 Summary
• The valve restriction areas are linearly proportional to the spool valve opening.
Vt = VA + VB = 2V0 .
21
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
Mechanical Subsystem
where
xp position of the piston [m];
m mass of the piston [kg];
k load spring constant [N/m];
b viscous damping coefficient [N/(m/s)];
PL load pressure [N/m2 ], i.e., PL = PA − PB ;
FF friction force [N];
FL external load force [N].
Hydraulic Subsystem
where
PL differential pressure between PA and PB [N/m2 ];
Ap pressure area of the piston [m2 ];
Cd discharge coefficient;
ρ density of hydraulic oil [kg/m3 ];
Cel
Ctl total leakage coefficient [m5 /Ns], i.e., Ctl = Cil + 2 ;
Cil internal leakage coefficient [m5 /Ns];
Cel external leakage coefficient [m5 /Ns];
w area gradient of the servo-valve spool [m];
Vt total actuator volume [m3 ];
βe effective bulk modulus of the system [N/m2 ].
22
Chapter 2. Modeling of Electro-Hydraulic Servo Systems
Servo-Valve Subsystem
where
xv spool position of the servo-valve [m];
ζ damping ratio [rad];
ωn natural frequency [rad/s];
kv torque motor gain [m/mA];
i input current of the torque motor [mA].
ẋ = f (x, u)
ẋ1 = x2
1
ẋ2 = m (−kx1 − bx2 + Ap x3 − FF − FL )
p
ẋ3 = −h1 x2 − h2 x3 + h3 Ps − sgn(x4 )x3 x4 (2.39)
ẋ4 = x5
ẋ5 = −ωn2 x4 − 2ζωn x5 + ωn2 kv u
y = x1
where
x1 position of the piston [m];
x2 velocity of the piston [m/s];
x3 load pressure [N/m2 ];
x4 spool position of the servo-valve [m];
x5 velocity of the servo-valve [m/s];
u current input [mA];
4βe Ap 4βe Ctl 4βe Cd w
h1 := Vt , h2 := Vt , and h3 := Vt ρ .
√
23
CHAPTER 3
Passivity-based Output Feedback
Control
3.1 Introduction
Various techniques have been used to control the position or force of an EHS. Var-
ious methods based on Lyapunov function have been studied for EHS and the valve
dynamics was considered to make EHS an input affine system. Backstepping approaches
were applied to EHS [17, 19–24] for exponential convergence of the desired position or
load pressure. Although backstepping controllers are long and complex [17], the back-
stepping controllers can improve the position/force tracking performance of the EHS.
Passivity-based control is a design technique that uses passivation to achieve the control
objective [27]. This technique has been successfully applied to a nonlinear model and was
shown to be very effective for the EHS [27–29]. The passivity-based control is simpler
and more straightforward than a backstepping algorithm since the synthetic inputs in
the passivity-based control is decoupled. Thus, modeling error effects associated with
the controller are compartmentalized [28]. Although previous nonlinear control methods
24
Chapter 3. Passivity-based Output Feedback Control
improved the position tracking performance of EHSs, all of these methods require full
state feedback. However, it is not always possible to measure the full state of the EHS
due to cost and space limitations. Furthermore, measurement noise is associated with
the pressure sensor and it is impossible to measure the spool position of the servo-valve
unless the position sensor is embedded in the servo-valve. Thus, the estimation of the
full state of the EHS is necessary. Several observers were designed to estimate the full
state of active suspensions and EHS [32–35].
The extended Kalman filter (EKF) can be difficult to tune and often gives unreliable
estimates if the system nonlinearities are served. In addition, this method cannot be used
to estimate the state from priori poor estimation or when large unmodelized perturba-
tions occur. This is because the EKF relies on linearization to propagate the mean and
covariance of the state. The unscented Kalman filter (UKF) can an give improved esti-
mation performance for nonlinear systems since the EKF required the computation of
Jacobians and the UKF dose not used Jacobians. But UKF is not practical to implement
in embedded control systems because this method required heavy on-line computational
effort. Since only observer designs were studied, only the stabilities of the observers
were proven. Because the designs of the nonlinear controllers, which are designed for the
strict feedback form, are not separated from those of the observers, closed-loop stability
should be studied. The high gain observer has a robust performance when poor a prior
estimation or large perturbations occurs and can be practical to implement because of
light on-line computational effort [36–38]. Several observer-based nonlinear controllers
were proposed [39, 40]. In [39] only acceleration was estimated by the observer. A pro-
portional integral (PI) observer was designed to estimate full state [40]. Although the
estimation performance was validated, the closed-loop stability was not proven.
The main contribution of this chapter is the design and implementation of a passivity-
based output feedback nonlinear control for position tracking in EHS. Our previous
research [26] studied for the rotary type EHS and was based on simulations with focus
on the design of the controller and the observer. This chapter solves the problem for the
linear type EHS, and focuses on the gain selection of the passivity-based controller for
the output feedback control and the proof of the closed-loop stability. The high gain ob-
server is designed to estimate the full state, and the high gain technique is used to reduce
the effects of the nonlinear terms. The on-line computational burden also can be signifi-
cantly reduced. The passivity-based controller [28] is implemented for position tracking.
Although the passivity-based control is simpler and more straightforward than the back-
stepping algorithm [28], the passivity-based controller guarantees only the asymptotic
stability of the tracking error dynamics. Therefore, it is difficult for the passivity-based
25
Chapter 3. Passivity-based Output Feedback Control
controller to be designed with a high gain observer since the exponential stabilities of the
quasi-steady-state model and the boundary layer system are required for the stability of
the singular perturbed system [36]. The problem is solved in this chapter by selecting
the controller gain such that the origin of the tracking error dynamics is exponentially
stable. We prove that the closed-loop system is asymptotically stable using the singular
perturbation method if the condition is satisfied. Therefore, the observer gain is chosen
with consideration of the controller gain. The performance of the proposed method is
validated through simulation and experiments.
26
Chapter 3. Passivity-based Output Feedback Control
Combining the control flow rate equation (2.16) and the load flow rate continuity
equation (2.25), the fluid dynamic equation of the actuator is given by
Remark 3.1. To smooth the function sgn(xv ), sgn(x4 ) is replaced with tanh(rx4 ) where r
is a sufficiently large positive constant. This modification is reasonable due to the leakage
of the valve spool [42–44]. ♦
Finally, by applying Newton’s second law, the actuator’s force balance equation is
where
m mass of the piston [kg];
k load spring constant [N/m];
b viscous damping coefficient [N/(m/s)];
Combining (3.1)-(3.3), the nominal dynamics of the EHS can be formulated as a state
space representation:
ẋ1 = x2
1
ẋ2 = (−kx1 − bx2 + Ap x3 )
m
p
ẋ3 = −h1 x2 − h2 x3 + h3 Ps − tanh(rx4 )x3 x4
(3.4)
ẋ4 = x5
ẋ5 = −ωn2 x4 − 2ζωn x5 + ωn2 kv u
y = x1
27
Chapter 3. Passivity-based Output Feedback Control
where
x1 position of the piston [m];
x2 velocity of the piston [m/s];
x3 load pressure [N/m2 ];
x4 spool position of the servo-valve [m];
x5 velocity of the servo-valve [m/s];
u current input [mA];
4βe Ap 4βe Ctl 4βe Cd w
h1 := Vt , h2 := Vt , and h3 := Vt ρ .
√
Assumption 3.1. In an actual EHS, all states are physically bounded. Therefore, the
initial values of all states are and remain in the compact set. ♦
p
Under Assumption 3.1, the nonlinear term, h3 Ps − tanh(rx4 )x3 x4 in EHS (2.39) is
locally Lipschitz in the compact set.
For position feedback, we design a high gain observer to estimate the full state and
reduce the nonlinearity of the EHS [26]. Equation (3.4) can be rewritten as
ẋ =Ax + φ(x) + Bu
(3.5)
y =Cx
where
0 1 0 0 0
0 0 Ap
m 0 0
√
A = 0 0 0 h3 Ps 0 ,
0 0 0 0 1
0 0 0 0 0
0
1
m (−kx1 − bx2 )
p √
φ(x) = −h1 x2 − h2 x3 + h3 ( Ps − tanh(rx4 )x3 − Ps )x4
0
−ωn2 x4 − 2ζωn x5
h iT
= 0 φ2 (x) φ3 (x) 0 φ5 (x) ,
28
Chapter 3. Passivity-based Output Feedback Control
h iT
B= 0 0 0 0 ωn2 kv ,
h i
C= 1 0 0 0 0 .
where x̂ is the estimation of states and L is the observer gain. We analyze the convergence
of the estimation error and stability of the closed-loop system. Let us define the estimation
error as
x̃1 x1 − x̂1
x̃2 x2 − x̂2
x̃ = x̃3 = x3 − x̂3 . (3.7)
x̃ x − x̂
4
4 4
x̃5 x5 − x̂5
where
−l1 1 0 0 0
−l2 Ap
0 m 0 0
√
Ao = A − LC = −l3 0 0 h3 Ps 0 ,
−l 0 0 0 1
4
−l5 0 0 0 0
0
φ2 (x) − φ2 (x̂)
h iT
δ(x, x̂) = φ(x) − φ(x̂) = φ3 (x) − φ3 (x̂) = 0 δ2 δ3 0 δ5 .
0
φ5 (x) − φ5 (x̂)
29
Chapter 3. Passivity-based Output Feedback Control
where is a small positive constant. Using the newly defined variable η, the estimation
error dynamics (3.8) is transformed into a fast dynamics in a singularly perturbed form
as
where
−α1 1 0 0 0
−α2 Ap
0 m 0 0
√
Aη = −α3 0 0 h3 Ps 0 ,
−α 0 0 0 1
4
−α5 0 0 0 0
0
2
δ2
∆(, x, x̂) = 3 δ3 .
0
5 δ 5
[26] Consider the estimation error dynamics (3.8). If is designed to satisfy the
condition
1
0<< (3.11)
2λmax (Pη )γ∆
30
Chapter 3. Passivity-based Output Feedback Control
Proof. The estimation error dynamics is given by (3.10). To study the convergence of
the equilibrium points η = 0 of (3.10), we define a Lyapunov candidate function
Vη = η T Pη η. (3.12)
The derivative of Vη is
we see that
1
If 0 < < 2λmax (Pη )γ∆ , then V̇η is negative definite. Therefore, the origin of (3.10) is
exponentially stable.
31
Chapter 3. Passivity-based Output Feedback Control
where xi ∀i ∈ [1, n], y, and u are the state, the output and the input of system,
respectively, and gi (x) ∀i ∈ [1, n] is nonzero for all x.
Remark 3.2. In practice, g3 is seldom zero, since |x3 | is seldom close to Ps . In the rare
case that g3 = 0, g3 is set to a small positive number to avoid the problem of dividing by
√
zero [22]. Therefore we assume that g3min ≤ g3 < h3 2Ps . ♦
In (3.4), we see that the EHS (3.4) can be substituted into the strict feedback form (3.17)
as follows
ẋ1 = |{z}
1 x2
g1
−kx1 − bx2 Ap
ẋ2 = − + x3
| m
{z m
} |{z}
f2 g2
p
ẋ3 = −h1 x2 − h2 x3 + h3 Ps − tanh(rx4 )x3 x4
| {z } | {z }
f3 g3
(3.18)
ẋ4 = |{z}
1 x5
g4
y = x1 .
|{z}
h
32
Chapter 3. Passivity-based Output Feedback Control
We define xd1 as the desired position. Then the tracking errors can be written as
We differentiate each tracking error to create the tracking error dynamics as follows
[26, 28] Consider the EHS (3.4). Given an arbitrary desired position xd1 , if the
desired states xd2 , . . . , xd5 and the control input u are defined as
1
xdi+1 = (−fi + ẋdi − ki ei ), ∀i ∈ [1, 4]
gi
(3.22)
1
u = (−f5 + ẋd5 − k5 e5 )
g5
where control gain ki > 0 ∀i ∈ [1, 5], then the origin of the tracking error dynamics
is asymptotically stable.
Proof. Substituting (3.22) into (3.21), a chain of interconnected tracking error dynamics
can be derived
1 2
Vi = e , ∀i ∈ [1, 4]. (3.24)
2 i
33
Chapter 3. Passivity-based Output Feedback Control
we obtain
Then (3.26) shows that the relationship between ei and ei+1 is output strictly passive
[36] and ėi = −ki ei ∀i ∈ [1, 4] is zero-state observable [13]. Therefore, each subsystem
is bounded input bounded output (BIBO) stable for ∀i ∈ [1, 4]. Serial interconnections
of BIBO stable system are also BIBO stable. Further, the 5th tracking error dynamics
becomes
Remark 3.3. The tracking performance of the proposed controller (3.22) requires pa-
rameters. Therefore, it is important to know the accuracy values of parameters. The
tracking performance of the proposed controller (3.22) is most sensitive to the ratio of
hydraulic parameters h1 /h3 [22]. If we obtain the information of full state, the param-
eters h1 and h3 can be easily estimated using the adaptation algorithms [22, 45]. More
detailed analysis is beyond the scope of this chapter, but is given in [46]. ♦
u = ϕ(x, xd1 )
u = ω21kv (ωn2 x4 + 2ζωn x5 + ẋd5 − k5 e5 )
n
xd2 = ẋd1 − k1 e1
(3.28)
xd3 = Amp ( m
k b
x1 + m x2 + ẋd2 − k2 e2 )
1
xd4 = √ (h1 x2 + h2 x3 + ẋd3 − k3 e3 )
h3 Ps −tanh(rx4 )x3
d
x5 = x5 + ẋd4 − k4 e4 .
34
Chapter 3. Passivity-based Output Feedback Control
From the control law (3.28), the tracking error dynamics becomes
ė = Ae e + Be (e, x3 , x4 ) (3.29)
where
e = [e1 , e2 , e3 , e4 , e5 ]T ,
−k1 0 0 0 0
0
−k2 0 0 0
Ae = 0 0 −k3 0 0 ,
0
0 0 −k4 0
0 0 0 0 −k5
h iT
Be (e, x3 , x4 ) = g1 e2 , g2 e3 , g3 (x3 , x4 )e4 , g4 e5 , 0 .
1
xd2 = (−f1 + ẋd1 − k1 e1 )
g1
1
xdi+1 = (−fi + ẋdi − ki ei − gi−1 ei−1 ), ∀i ∈ [2, 4] (3.30)
gi
1
u= (−f5 + ẋd5 − k5 e5 )
g5
is defined. Notice that the difference is the addition of ‘gi−1 ei−1 ’ to xdi ∀i ∈ [2, 4] for the
backstepping control law (3.30). With the backstepping control law (3.30), the tracking
error dynamics becomes
35
Chapter 3. Passivity-based Output Feedback Control
where
−k1 g1 0 0 0
−g1 −k2 g2 0 0
Aback (x3 , x4 ) = 0 −g2 −k3 g3 (x3 , x4 ) 0 .
0
0 −g3 (x3 , x4 ) −k4 g4
0 0 0 −g4 −k5
Remark 3.4. In backstepping control design, the additional term gi−1 ei−1 in xdi+1
makes Aback (x3 , x4 ) = Aback0 + Aback1 (x3 , x4 ) where Aback0 < 0 and Aback1 (x3 , x4 ) =
−ATback1 (x3 , x4 ). Therefore, the tracking error dynamics (3.31) becomes strict passive.
Unlike the origin of the tracking error dynamics (3.29) in the passivity-based controller,
the origin of the tracking error dynamics (3.31) in the backstepping controller is expo-
nentially stable since there a Pback exists that satisfies
It was reported that the passivity-based control is simpler and more straightforward
than the backstepping algorithm in [28]. In [28], the controller gains are chosen to be
ki = |gi | ∀i ∈ [1, 4] and k5 > 0. These gain selections guarantee that the origin of the
error dynamics (3.29) is asymptotically stable. Generally, when the high gain observer is
implemented in the feedback controller, the closed-loop system is in the form of singular
perturbation system. According to Theorem 11.4 [36], the origins of the reduced-order
system and the boundary-layer system should be exponentially stable for the closed-
loop stability of the singular perturbation system. Since the passivity-based controller
guarantees that the tracking error dynamics (3.29) is asymptotically stable unlike the
backstepping controller, it is difficult to design a high gain observer with a passivity-based
controller. However, in EHS (3.4), since g1 , g2 , g4 , and g5 are constant and g3 (x3 , x4 ) is
bounded, a positive constant γ exists such that kBe (e, x3 , x4 )k2 ≤ γkek2 .
36
Chapter 3. Passivity-based Output Feedback Control
Consider the EHS (2.39). Let us define the positive definite matrix, Pe as
If the controller gains ki0 s of the passivity-based controller (3.28) are designed to
1
satisfy γ < 2λmax (Pe ) , then the origin of the tracking error dynamics (3.29) is expo-
nentially stable.
Proof. Let us define the quadratic Lyapunov function Ve (e) = eT Pe e. The derivative of
Ve gives us
Pe depends on the design of controller gains ki0 s. If the controller gains ki0 s are designed
1
to satisfy γ < 2λmax (Pe ) , then the origin of the tracking error dynamics (3.29) is expo-
nentially stable.
1
In Lemma 3.1, the controller law should be designed to satisfy γ < 2λmax (Pe ) . Ideally,
it is better as the controller gains are higher. However, the higher control gain may
amplify the noise in the position sensor, since the controller (3.28) uses not only position
measured by the sensor but also the derivative of the measured position. Hence, the
controller gain should be determined by considering the trade-off between the tracking
performance in the absence of noise and the amplification of the noise.
37
Chapter 3. Passivity-based Output Feedback Control
Since the estimated states x̂ are used in the control law (3.28), except x1 , we replace
u with û such that
where ê2 = x̂2 − xd2 and êi = x̂i − x̂di ∀i ∈ [3, 5].
Remark 3.5. The control law u = ϕ(x, xd1 ) (3.28) is not Lipschitz. Thus, in û (3.35),
x1 is used instead of x̂1 . Consequently, since x1 , xd1 , and ẋd1 are known, xd2 is used in
(3.35). ♦
Therefore, the tracking error dynamics (3.29) with a high gain observer (3.6) becomes
where ξ = [0, 0, 0, 0, − g5 (u − û))]T . The tracking error dynamics (3.37) and the
estimation error dynamics (3.10) can be represented in the singularly perturbed form [36]
as follows
with the equilibrium point (e, η) = (0, 0). The tracking error dynamics (3.37) is slow
dynamics and the estimation error dynamics (3.10) is fast dynamics. Making = 0 implies
that η = 0, i.e., [x1 , x2 , x3 , x4 , x5 ]T = [x̂1 , x̂2 , x̂3 , x̂4 , x̂5 ]T . Substituting the equilibrium
38
Chapter 3. Passivity-based Output Feedback Control
ė = Ae e + Be (e, x3 , x4 ). (3.39)
d
η = Aη η (3.40)
dτ
t
where τ = and Aη is defined as in (3.10). The following theorem will explain the
stabilities of the equilibrium points e = 0 of (3.39) and η = 0 of (3.40).
Consider the quasi-steady-state model (3.39) and the boundary layer system (3.40).
The origins of (3.39) and (3.40) are exponentially stable.
Remark 3.6. From the perspective of the singular perturbation analysis, the estimation
error dynamics (3.10) is faster than the tracking error dynamics (3.37). Under the as-
sumption that = 0, the estimation error η and tracking error e are exponentially stable
if the position is measurable. ♦
In practice, it is not possible for a high gain observer (3.6) to have = 0. Thus it is
necessary to find a positive value of < ¯ such that the system is stable.
Consider the error dynamics (3.38) of the closed-loop system. If a continuous interval
Γ = (0, ¯) exists such that for all ∈ Γ
T1 1
¯ = min 2 , (3.41)
2T1 T3 + T2 2T3
where T1 , T2 , and T3 , are positive constants, then the origin (e, η) = (0, 0) of (3.38)
is asymptotically stable as t goes to infinity.
39
Chapter 3. Passivity-based Output Feedback Control
Vcl = eT Pe e + dη T Pη η. (3.42)
From Lemma 3.1, Theorem 3.1 and Lemma 3.2, differentiating Vcl with respect to time
t gives us
d
V̇cl ≤ − (1 − 2λmax (Pe )γ)kek22 + 2eT Pe ξ + (− + 2dλmax (Pη )γ∆ )kηk2 . (3.43)
In (3.43),
h iT
kξk2 =k 0 0 0 0 −g5 (u − û) k2 . (3.44)
Now we study the inequality of kξk2 . Since both xd3 and x̂d3 use ẋd2 , M3 exists such that
|xd3 − x̂d3 | ≤ M3 kx − x̂k2 . From (3.28) and (3.35), we get
1 1
xd4 − x̂d4 = − (−h1 x2 − h2 x3 − ẋd3 + k3 e3 ) + (−h1 x̂2 − h2 x̂3 − x̂˙ d3 + k3 ê3 )
g3 ĝ3
1 1
= − (−h1 x2 − h2 x3 − ẋd3 + k3 e3 ) + (−h1 x̂2 − h2 x̂3 − x̂˙ d3 + k3 ê3 ) (3.45)
g3 g3
1 1 d
− (−h1 x̂2 − h2 x̂3 − x̂˙ d3 + k3 ê3 ) + (−h1 x̂2 − h2 x̂3 − x̂˙ 3 + k3 ê3 ).
g3 ĝ3
√
Since 0 < g3min ≤ |g3 | < h3 2Ps ,
1
|xd4 − x̂d4 | ≤ |(−h1 x̃2 − h2 x̃3 − (ẋd3 − x̂˙ d3 ) + k3 x̃3 )|
g3min
1 1 (3.46)
+ − | − h1 x̂2 − h2 x̂3 − x̂˙ d3 + k3 ê3 | .
g3 ĝ3
| {z }
d4
Since all fi ’s are linear and ẍd2 cancels out in ẋd3 − x̂˙ d3 , M3d exists such that |ẋd3 − x̂˙ d3 | ≤
M3d kx − x̂k2 . Thus M4 also exists such that
1
From Theorem 3.1, if 0 < < 2λmax (Pη )γ∆ , x̂ converges to x. Thus x̂ and ê3 are bounded,
and g3 and ĝ3 are bounded. Therefore, d4 is bounded. Furthermore, since ĝ3 converges
to g3 as t → ∞, d4 converges to zero as t → ∞. Next we study the inequality of
¨d |, xd (3) also cancels out. Therefore, M4 exists such that |ẋd − x̂˙ d | ≤
|xd5 − x̂d5 |. In |ẍd3 − x̂ 3 2 d 4 4
40
Chapter 3. Passivity-based Output Feedback Control
Note that the function tanh(x4 ) consists of ex4 and e−x4 . Since x̂, ġ3 , and ĝ˙ 3 are bounded,
|d˙4 | is bounded. And |d˙4 | converges to zero as t → ∞. Using the same procedure, it can
be determined that Mu exists such that
1 ¨
|u − û| ≤ Mu kx − x̂k2 + |d4 |. (3.49)
ωn2 kv
Since x̂, g̈3 , and ĝ¨3 are bounded, |d¨4 | is bounded. And |d¨4 | also converges to zero as
t → ∞. Finally we conclude that γξ exists such that
h iT
kξk2 =k 0 0 0 0 −g5 (u − û) k2
1
V̇cl ≤ − (1 − 2λmax (Pe )γ) kek22 + 2 4 kPe k2 γξ kek2 kηk2
| {z } | {z }
T1 T2
(3.51)
d
+ 2 kPe k2 γd (d4 , ·) kek2 + − + 2d λmax (Pη )γ∆ kηk22 .
| {z } | {z }
D1 T3
1
Let us define d = 8 . Then if −(2T1 T3 + T22 ) + T1 > 0, it is guaranteed that T is positive
1
definite. Since D is bounded, X get into the bounded ball in a finite time. If 0 < < 2T3 ,
D goes to zero as t goes to infinity. Since there exists a continuous interval Γ = (0, ¯)
41
Chapter 3. Passivity-based Output Feedback Control
The condition (3.41) is only sufficient condition for closed-loop stability. Theorem 3.3
indicates that the proposed output feedback is not based on the separation principle.
Therefore, if the observer gains should be designed with the controller gains, then an
output feedback controller is constructed to track the desired position xd1 .
Remark 3.7. The convergence rate of a high gain observer is faster as the gain of the
observer approaches infinity ( → 0) if, as is ideal, no measurement noise is present. How-
ever, when noise is present, the noise is amplified by the high gain of the observer [48].
Hence, the observer gain should be determined by considering the trade-off between the
estimation performance in the absence of noise and the amplification of the noise. In
our EHS, the measurement noise is less of a factor since the noise level of the position
sensor is less than 0.035% of the full range (±0.026m). ♦
To investigate the performance of the proposed method, The parameters of the EHS,
the observer and the control gains are m = 0.5, Ap = 5.058×10−4 , k = 5.561×105 , b = 0,
h1 = 3.2569×1010 , h2 = 2.1456, h3 = 7.1693×109 , ωn = 950, ζ = 0.5, kv = 1.333×10−5 ,
k1 = 3000, k2 = 2500, k3 = 2300, k4 = 50, k5 = 100, l1 = 837.7153, l2 = 3.5088 × 105 ,
l3 = 8.5148 × 1010 , l4 = 0.3188, kp = 1500, ki = 20, and kf f = 100. In the simulations,
the position tracking performances of the proposed method and the estimation of the
high gain observer were studied. The performance of the proposed method was compared
with that of the PI control with the velocity feedforward that is often used in industrial
applications. The gains of the PI controller with velocity feedforward were well tuned for
step response and tracking the sinusoidal signal.The block diagrams of the two methods
are shown in Figure 3.1.
Matlab/Simulink was used for the simulations. For the implementation of the proposed
method, the S-function coded in C language was used. The measurement noises in the
42
Chapter 3. Passivity-based Output Feedback Control
position sensor and the input noise are 0.035% and 0.15% of the maximum ranges,
respectively.
The desired position used in the simulations is xd1 (t) = 0.015 sin(2πt). Figure 3.2
illustrates the position tracking performance, demonstrating that x1 tracked xd1 quite
well. Note that although asymptotic tracking error convergence is guaranteed by the
proposed method, the tracking error cannot converge to zero since the holding effect of the
control signal appeared during one sampling time of the controller in simulations. If the
control sampling rate gets faster, then the tracking error becomes smaller in simulations.
The estimation results are plotted in Figure 3.3, where we show that the estimated
states tracked the actual states quite well.
Step Response
To evaluate the step response of the proposed method, the proposed method was
compared with the PI controller with velocity feedforward used as industrial applications.
The step responses of both methods are shown in Figure 3.4 and Figure 3.5. The transient
behavior was improved by the proposed method compared with the PI controller with
velocity feedforward although a small overshoot was observed in the proposed method.
43
Chapter 3. Passivity-based Output Feedback Control
20
d
x1
Position [mm]
10 x1
−10
−20
0 0.5 1 1.5 2 2.5
time(s)
(a) Position tracking
2
Position error [mm]
−1
−2
0 0.5 1 1.5 2 2.5
time(s)
(b) Position tracking error
20
Current [mA]
−20
44
Chapter 3. Passivity-based Output Feedback Control
20
x1
Position [mm]
10 Estimated x1
−10
−20
0 0.5 1 1.5 2 2.5
time(s)
(a) Estimation of x1
300
Cylinder Velocity [mm/s]
x2
200
Estimated x2
100
0
−100
−200
−300
0 0.5 1 1.5 2 2.5
time(s)
(b) Estimation of x2
4
x 10
3
Load Pressure [N/m2]
x3
2
Estimated x3
1
0
−1
−2
−3
0 0.5 1 1.5 2 2.5
time(s)
(c) Estimation of x3
0.2
x4
Valve Position [mm]
0.1 Estimated x4
−0.1
−0.2
0 0.5 1 1.5 2 2.5
time(s)
(d) Estimation of x4
Valve Velocity [mm/s]
x5
20
Estimated x5
−20
(e) Estimation of x5
6
Position [mm]
4
0 0.5 1 1.5 2
time[s]
(a) Position tracking
6
Position error [mm]
0 0.5 1 1.5 2
time[s]
(b) Position tracking error
6
Position [mm]
0 0.5 1 1.5 2
time[s]
(a) Position tracking
6
Position error [mm]
0 0.5 1 1.5 2
time[s]
(b) Position tracking error
46
Chapter 3. Passivity-based Output Feedback Control
PC
CPU: Pentium Dual
Core 2.6GHz,
OS : Windows XP + RTX
DAC ADC
16-bit 16-bit
EHS x1
Servo valve
Actuator
LVDT
Servo valve
Piston rod
Actuator
The experiments were executed in order to evaluate the performance of the proposed
method. The implementation of the proposed method on the EHS test rig is depicted
in Figure 3.6. The EHS designed for the quadruped robot and the flapper type servo-valve
manufactured by Star Hydraulic Ltd. were used as shown in Figure 3.7. The effective
stroke of the piston range was ±0.026m, and the position feedback was obtained by
the linear variable differential transformer (LVDT). For the experiments, the control
47
Chapter 3. Passivity-based Output Feedback Control
algorithms (the proposed method and PI controller with velocity feedforward) coded in
C by an S-function were used in real-time operating systems. The derivatives in the
control law (3.35) are calculated by Tustin method. A 16-bit A/D converter and a 16-bit
D/A converter were used, and the sampling rate was 1 KHz. The two desired positions of
0.015 sin(2πt), and 0.015 sin(6πt) were used. The tracking performances of both methods
for a sine wave with a frequency 1 Hz and a 0.015m amplitude are depicted in Figs. 3.8 and
3.9, which show that the position tracking error of the proposed method was less than 40%
of that of PI with velocity feedforward although both methods required similar power as
shown in Figure 3.8 (c) and Figure 3.9 (c). The phase lag was decreased by the proposed
method, and the peaks of the position tracking errors in both methods appeared due to
static friction whenever the direction of motion as changed as shown in Figure 3.8 (b)
and Figure 3.9 (b). In the proposed method, the influence of friction was insignificant
due to compensation, as depicted in Figure 3.9 (c), although the proposed controller
was designed without friction. However, the position tracking error of PI with velocity
feedforward had large peaks. The estimated values results of x1 and x3 are plotted
in Figure 3.10. Since the actual position signal used in observer was noisy compared
with the position in simulations, the estimated load pressure was also noisy. Although,
the estimation error appeared in x3 due to the parameter uncertainties, sensor, noise,
and unmodelled dynamics, x̂1 and x̂3 tracked x1 and x3 well respectively. The tracking
performances of both methods for the fast reference 0.015 sin(6πt) are illustrated in
Figure 3.11. The flow rate of our EHS is limited and designed to track the sinusoidal
signal with low frequency. Furthermore, our reference signal had relative large amplitude
compared with the stroke of the piston range. Therefore, large tracking error appeared
due to the limitation of flow rate although the frequency of the sinusoidal reference
trajectory was only 3 Hz. Thus the control inputs were saturated as shown in Figure 3.12.
Although the poor tracking performance of both methods was apparent, the position of
the proposed method was larger than 120% of that of the PI controller with velocity
feedforward. This can be interpreted as an achievement of improved bandwidth with
the proposed method. The asymmetric behavior can be seen in the position tracking
performance of PI and feedforward controller of Figure 3.8 (b) and the load pressure
of Figure 3.3, unlike that in the simulations; this occurrence is unavoidable since the
actual structure of EHS is asymmetric.
48
Chapter 3. Passivity-based Output Feedback Control
20
d
x1
Position [mm]
10 x1
−10
−20
15 15.5 16 16.5 17 17.5
time[s]
(a) Position tracking
2
Position error [mm]
−1
−2
15 15.5 16 16.5 17 17.5
time[s]
(b) Position tracking error
20
Current [mA]
−20
49
Chapter 3. Passivity-based Output Feedback Control
20
d
x1
Position [mm]
10 x1
−10
−20
15 15.5 16 16.5 17 17.5
time[s]
(a) Position tracking
2
Position error [mm]
−1
−2
15 15.5 16 16.5 17 17.5
time[s]
(b) Position tracking error
20
Current [mA]
−20
50
Chapter 3. Passivity-based Output Feedback Control
20
x1
10 Estimated x1
Position[mm]
−10
−20
15 15.5 16 16.5 17
time[s]
(a) Estimation of x1
5000
x3
Load Pressure[N/m2]
Estimated x3
−5000
15 15.5 16 16.5 17
time[s]
(b) Estimation of x3
51
Chapter 3. Passivity-based Output Feedback Control
52
Chapter 3. Passivity-based Output Feedback Control
3.8 Summary
In this chapter, passivity-based output feedback nonlinear control was proposed for
position tracking in the EHS, and a high gain observer was designed for the estimation
of states. A high gain technique was used to reduce the effect of the nonlinear terms.
To design a passivity-based controller with a high gain observer, gain selection of the
designed controller was proposed in order to guarantee exponential stability. The stabil-
ity of the closed-loop system was demonstrated via singular perturbation analysis. The
tracking performance of the proposed method and the estimation results were studied
via simulation. Furthermore, we experimentally observed that the position tracking per-
formance of the EHS was improved by the proposed method. Although the proposed
controller was designed without friction, the influence of the friction was insignificant
due to compensation by the proposed method. The tracking error in the experiments
was asymmetric since the actual structure of the EHS was asymmetric.
53
CHAPTER 4
Biased Sinusoidal Disturbance
Estimation and Rejection Control
4.1 Introduction
Robust control methods under parameter uncertainties were proposed [52, 53]. Re-
cently, the controller design in the presence of friction and internal leakage was de-
signed [54]. This method required the knowledge of the friction. All of these methods
improve the position tracking performance without considering their disturbances or the
friction and load torque. However, when the disturbance significantly affects the position
tracking performance, the position tracking performance is degraded since the force or
torque available for the systems is diminished. Therefore, a method for compensating for
the disturbance is needed in order to improve the EHS performance. In practice, since
direct measurements of the disturbance are not possible, it is necessary to estimate the
54
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
55
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
The desired load pressure is designed to generate the desired load pressure using the
differential flatness property of the EHS mechanical subsystem. The disturbance, which
includes not only biased sinusoidal signal but effects of the assumptions for the modeling
and the parameter uncertainties, within the bandwidth of the DOB can be canceled by the
proposed method. The performance of the proposed method is validated via simulations
and experiments.
In many EHS applications, the valve dynamics is much faster than the dynamics of
the remaining parts of the system such that valve dynamics can be neglected without
significant reduction of control performance [14, 15, 50, 51]. Therefore, for the simplicity
we use the following approximation
xv = kv i (4.1)
where
xv spool position of the servo-valve [m];
kv torque motor gain [m/mA];
i input current of the torque motor [mA].
Combining the control flow rate equation (2.16) and the load flow rate continuity equation
(2.25), the fluid dynamic equation of the actuator is given by
Finally, by applying Newton’s second law, the actuator’s force balance equation is
where
m mass of the piston [kg];
k load spring constant [N/m];
b viscous damping coefficient [N/(m/s)];
d disturbance.
56
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
The goal is to make the piston position track the desired position with disturbance
compensation. In the problem to be studied as depicted in Figure 4.1, the load torque is
in the form of a sinusoidal signal. Generally, the main friction is the coulomb friction in
EHS [12]. Therefore, the disturbance, d, in our system is in the form of a biased sinusoidal
signal as
where
a constant friction coefficient;
mL amplitude of the load torque [Nm];
r rate of angle of the pendulum [rad/sec].
We assume that a, mL , and r are unknown, but the upper bound of mL , mLmax , is
known. The upper bound of the load torque frequency is also known. The assumption
that the upper bound of the load torque frequency is known is reasonable since the
frequency of the load torque is determined by the desired position profile.
Combining (4.1)-(4.4), the dynamics of the EHS can be formulated as the following state
space representation:
ẋ1 = x2
Mechanical
subsystem
1
ẋ =
m (−kx1 − bx2 + Ap x3 − d)
2 (4.5)
Hydraulic n p
ẋ3 = −h1 x2 − h2 x3 + h3 Ps − sgn(u)x3 u
subsystem
where
x1 position of the piston [m];
x2 velocity of the piston [m/s];
x3 load pressure [N/m2 ];
d disturbance;
u current input [mA];
57
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
The DOB is designed in the form of a second order high pass filter in order to estimate
a disturbance, d, that is biased on a sinusoidal signal with an unknown frequency. The
dynamics of x̂2 is defined as
1
x̂˙ 2 = (−kx1 − bx2 + Ap x3 − d)
ˆ (4.6)
m
where x̂2 and dˆ are the estimations of x2 and d, and x̂2 (0) = 0 and d(0)
ˆ = 0. The dˆ is
defined by
Z t
dˆ = −mkP (x2 − x̂2 ) − mkI (x2 − x̂2 )dt. (4.7)
0
Rt
Then, the estimation error is defined as d˜ = d − dˆ whereas d˜I is defined as 0 d˜ dt
with dˆI (0) = 0. Next, we will summarize the result of the estimation error boundedness
analysis.
˜
Theorem 4.1 (Boundness of d).
Consider the EHS (4.5). Suppose that DOB (4.6), (4.7) are used. If the cut-off
frequency ωc of the high pass-filter
s2
H(s) = (4.8)
s2 + kP s + kI
is above the maximum frequency of the disturbance, then d˜ is bounded and T > 0
exists such that
˜ ≤ d˜max = |H(s)|s=jω × ML ,
|d| for t > T. (4.9)
c max
Proof. We study the estimation error boundedness in the transient state and the steady-
˜ = 0. First, the estimation error
state. We assume that x1 (0) = 0. Then d(0) = 0 and d(0)
boundedness in the transient state is studied. Differentiating (4.7) with respect to time
gives
Z t
˙
d˙ = d˜ + kP d˜ + kI ˜
ddt. (4.10)
0
58
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
Since Ad is Hurwitz, γ exists such that keAd (t−τ ) k2 ≤ γe−λ(t−τ ) where λ is the minimum
singular value of Ad . Note that supt |d(t)| = a + MLmax . Thus,
Z t
ked k2 ≤ kBd k2 |d(t)| + (γkAd k2 eλi (t−τ ) kBd k|d(τ )|)dτ
0
γkAd k2 (4.13)
≤ sup |d(t)| + sup |d(t)|
t λ t
γkAd k2 (a + MLmax )
≤ a + MLmax + .
λ
˜ ≤ ked k2 ,
Since |d|
This means that d˜ is bounded in the transient state and the boundedness depends on
kP , kI . In order to analyze the upper bound on d˜ in steady-state, d˜ is studied from the
perspective of the frequency domain. Differentiating twice both on sides of (4.7) gives us
¨ ˙
d¨ = d˜ + kP d˜ + kI d.
˜ (4.15)
s2
d˜ = d (4.16)
s2 + kP s + kI
| {z }
H(s)
59
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
Bode Diagram
−20
−40
Magnitude (dB)
−60
−80
−100
−120
−1 0 1 2 3
10 10 10 10 10
Frequency (rad/sec)
where s is the Laplace operator. From (4.16), we see that the frequency response of the
estimation error is a typical second order high pass-filter. The cutoff frequency, ωc of
H(s), determined by the observer parameters, kP and kI , should be above the maximum
frequency of the disturbance. Then the estimation error d˜i is suppressed by high pass
filter (4.16).
Thus T exists such that
˜
|d(t)| ≤ d˜max = |H(s)|s=jωc × MLmax (4.17)
for t > T .
Figure 4.2 shows the frequency response of H(s) when kP and kI are 10π and (6π)2 ,
respectively. Although the DOB is able to suppress the estimation error of the disturbance
and perfectly reject the biased signal, it does not guarantee that the estimation error
of the disturbance asymptotically converges to zero. Therefore, compensation for the
estimation error is in need.
Remark 4.1. Actually, the assumptions for the modeling and the parameter uncertain-
ties exist in our EHS. Thus, d may include not only biased sinusoidal signal but effects
of the assumptions for the modeling and the parameter uncertainties. The proposed DOB
is in the form of second order high pass-filter. Thus, disturbance within the bandwidth of
the DOB can be estimated by the DOB. ♦
60
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
The controller is designed to track the desired profile xd1 . In EHS, the piston motion
is obtained by the load pressure. Therefore, the load pressure becomes the desired load
pressure by the current input, then the piston position moves to the desired position by
the load pressure. In the mechanical subsystem, we regard the position x1 and the load
pressure x3 as the output and the input, respectively. Let us define a flat output as
y = x1 . (4.18)
Using system (4.5) and (4.18), it can be shown that the states and the control input can
be expressed as function of flat output y and a finite number of its time derivatives
x1 = y
x2 = ẏ
ky + bẏ + my (2)
x3 = (4.19)
Ap
mφ y, ẏ, y (2) , y (3)
u= p q
Ap h3 Ps − sgn φ y, ẏ, y (2) , y (3) ky + bẏ + y (2)
where
kh
2 Ap h1 + bh2 + k b + mh2
φ y, ẏ, y (2) , y (3) = y+ ẏ + y (2) + y (3) .
m m m
Remark 4.2. A system is flat if we can find a set of outputs (equal in number to the
number of inputs) such that all states and inputs can be determined from these outputs
without integration [49]. More precisely, if the system has states x ∈ Rn , and inputs
u ∈ Rm then the system is flat if we can find outputs y ∈ Rm of the form
y = h x, u, u̇, ..., u(r) (4.20)
such that
x = ϕ y, ẏ, ..., y (q) ,
(4.21)
u = α y, ẏ, ..., y (q) .
61
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
Then since all variables in the mechanical subsystem can be parameterized in terms
of differential functions of the position and the velocity except the disturbance, the
mechanical subsystem is clearly be differentially flat as follows
Thus the use of flatness determines the desired load pressure xd3 such that
1
xd3 = mẍd1 + bẋd1 + kxd1 .
(4.23)
Ap
e1 = x1 − xd1 (4.24)
e2 = x2 − x∗2
e3 = x3 − x∗3
where x∗2 = ẋd1 − k1v (x1 − xd1 ), k1v > 0 and x∗3 is yet to be defined.
62
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
1
u= p [h1 x2 + h2 x3 − k3 e3 + ẋ∗3 ] (4.25)
h3 Ps − sgn(u)x3
where
1
x∗3 =xd3 + [−mk0 e0 + (k − bk1 + mk12v − mk1 )e1
Ap (4.26)
+ (b − mk1v − mk2 )e2 + dˆ − d˜max sgn(p0 e0 + p1 e1 + p2 e2 )].
Note that the term −sgn(p0 e0 + p1 e1 + p2 e2 )d˜max is implemented to compensate for the
estimation error of the disturbance using VSC.
p
Remark 4.4. In practice, Ps − sgn(u)x3 is seldom zero, since |x3 | is seldom close to
p p
Ps . In the rare case that Ps − sgn(u)x3 = 0, Ps − sgn(u)x3 is set to a small positive
number in order to avoid the problem of dividing by zero [22]. ♦
Remark 4.5. It is important to note that (4.25) cannot be solved ‘as is’, since it contains
the control variable, u, on both sides of an equation involving the sgn function. However,
u in the right side is used for only sgn function. We can see that sign of u is determined
p
by [h1 x2 + h2 x3 − k3 (x3 − x∗3 ) + ẋ∗3 ] since Ps − sgn(u)x3 > 0. Therefore, for the digital
implementation of the control law (4.25), the modified control law (4.27) written as
1
u= p [h1 x2 + h2 x3 − k3 e3 + ẋ∗3 ] (4.27)
h3 Ps − sgn(ub )x3
63
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
ė3 = − k3 e3 .
Consider the EHS (4.5). Suppose that the DOB (4.6), (4.7) and the controller
(4.25), (4.26) are used, and ωc of H(s) is greater than the maximum frequency of
the disturbance. If Am is Hurwitz and k3 > 0, and p0 , p1 , and p2 are chosen such
that
∗ ∗ p0
Pm =
∗ ∗ p1
p0 p1 p2
64
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
V = eTm Pm em (4.30)
If we define 2Pm Bm e3 as the input and em as the output in (4.32), (4.32) can be rewritten
as
Equation (4.33) shows that the relationship between em and 2Pm Bm e3 is output strictly
passive, and ėm = Am em + Ad (e) is zero-state observable. Therefore, (4.29) is bounded-
input bounded-output stable. e3 exponentially converges to zero. Therefore, em asymp-
totically converges to zero for t > T .
Remark 4.6. In (4.25), the derivative of x∗3 is used. The derivative of sgn(x) function
is infinite in zero point. Since sgn(x) function can be replaced with tanh(hx) where h is a
sufficiently large positive constant [43, 44], d˜max tanh(h(p0 e0 +p1 e1 +p2 e2 )) is used instead
of d˜max sgn(p0 e0 + p1 e1 + p2 e2 ) for the digital implementation of the control law (4.26).
♦
Figure 5.2 shows the block diagram of the controller structure. The desired load pres-
sure is generated using flatness concept (4.23) by the desired position. The disturbance
is estimated using the DOB (4.6) and (4.7). Then the current input is obtained by the
control law (4.25) and (4.26).
65
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
dˆ = asgn(x2 )
â = z − ka |x2 |µ (4.34)
1
ż = ka µ|x2 |µ−1 (−kx1 − bx2 + Ap x3 ) − âsgn(x2 ) sgn(x2 ).
m
The SimHydraulic model in SimScape was used as a hydraulic servo system. The param-
eters of the EHS and the control gains are m = 0.5, Ap = 5.058 × 10−4 , k = 0, b = 0,
Ps = 2.1 × 107 , h1 = 3.257 × 1010 , h2 = 2.146, h3 = 7.169 × 109 , kv = 1.333 × 10−5 ,
k0 = 1000, k1 = 200000, k2 = 2500, k3 = 600, k1v = 100, kP = 10π, kI = (6π)2 ,
and dLmax = 5. The desired profile, xd1 (t) = 0.008 sin(2πt) and disturbance, d(t) =
294 sin(62.83x1 ) + 20sgn(x2 ) were used. Simulations were performed for the following
three cases:
Since there was no disturbance in case 1, only position tracking controller (4.35) is
used as follows:
1
u= p [h1 x2 + h2 x3 − k3 (e3 ) + ẋ∗3 ]
h3 Ps − sgn(u)x3
1
x∗3 = xd3 + [−mk0 e0 + (k − mk1 )e1 + (b − mk2 )e2 ] (4.35)
Ap
1
xd3 = [mẍd1 + bẋd1 + kxd1 ].
Ap
Figure 4.4 shows the simulation results for Case 1. In Figure 4.4 (c), the force means
F = Ap x3 . When there was no disturbance, the position successfully tracked the desired
position using only the position tracking controller (4.35). Note that a small force was
required to move the actuator when there was no disturbance.
66
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
The simulation results for Case 2 is shown in Figure 4.5. In Case 2, only position
tracking controller (4.35) in the proposed method was also used. Due to the load, which
is heavier than the actuator mass, the force increased to overcome the disturbance as
compared with Case 1. Furthermore, the force increased near zero velocity due to the
friction. However, since the disturbance was not completely compensated for, the position
tracking error was greater than that of Case 1. Due to the Coulomb friction, the peak
appeared in the control input.
Figure 4.6 shows the simulation results for Case 3. In [56], it was shown that the
Friedland method is effective to estimate the time-varying disturbance when x2 6= 0
although the Friedland method was designed to estimate the constant friction. In (4.34)
ż = 0 when x2 = 0. Therefore, it was observed that the estimation error appeared when
the velocity was near zero. The estimation error resulted in the position tracking error
when the velocity was near zero. Thus, the peak and the oscillation appeared near zero
velocity points in the position and the control input.
Figure 4.7 shows the simulation results for Case 4. The disturbance estimation is
shown in Figure 4.7 (d). The cut-off frequency of the DOB was chosen to be much higher
than the frequency of the disturbance, which means that the disturbance was estimated
accurately. Therefore, the position tracking error was improved when compared with
that of Cases 2, 3 and the position tracking performance was recovered. Note that the
estimated disturbance was similar to the force since the required force to move the
actuator was much smaller than the required force to overcome the disturbance, but the
disturbance is not the same as the force.
67
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
10
xd
1
Position [mm]
5 x1
−5
−10
0 1 2 3 4 5 6
Time [s]
(a) Position tracking performance
1
Position error [mm]
0.5
−0.5
−1
0 1 2 3 4 5 6
Time [s]
(b) Position tracking error (xd1 − x1 )
0.5
Force [N]
−0.5
−1
0 1 2 3 4 5 6
Time [s]
(c) Force (Ap x3 )
2
Current [mA]
−2
−4
0 1 2 3 4 5 6
Time [s]
(d) Control input (u)
68
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
10
xd
1
Position [mm]
5 x1
−5
−10
0 1 2 3 4 5 6
Time [s]
(a) Position tracking performance
1
Position error [mm]
0.5
−0.5
−1
0 1 2 3 4 5 6
Time [s]
(b) Position tracking error (xd1 − x1 )
200
100
Force [N]
−100
−200
0 1 2 3 4 5 6
Time [s]
(c) Force (Ap x3 )
2
Current [mA]
−2
−4
0 1 2 3 4 5 6
Time [s]
(d) Control input (u)
69
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
10
xd
1
Position [mm]
5 x1
−5
−10
0 1 2 3 4 5 6
Time [s]
1
Position error [mm]
0.5
−0.5
−1
0 1 2 3 4 5 6
Time [s]
200
100
Force [N]
−100
−200
0 1 2 3 4 5 6
Time [s]
200
d
Estimated d
Disturbance [N]
100
−100
−200
0 1 2 3 4 5 6
Time [s]
ˆ
(d) Estimated disturbance (d)
2
Current [mA]
−2
−4
0 1 2 3 4 5 6
Time [s]
10
xd
1
Position [mm]
5 x1
−5
−10
0 1 2 3 4 5 6
Time [s]
1
Position error [mm]
0.5
−0.5
−1
0 1 2 3 4 5 6
Time [s]
200
100
Force [N]
−100
−200
0 1 2 3 4 5 6
Time [s]
200
d
Estimated d
Disturbance [N]
100
−100
−200
0 1 2 3 4 5 6
Time [s]
ˆ
(d) Estimated disturbance (d)
2
Current [mA]
−2
−4
0 1 2 3 4 5 6
Time [s]
Figure 4.7 Case 4: Position tracking performance of the proposed method with
disturbance compensation when there is disturbance
71
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
Serve-valve
Actuator
Load
The gains of the PI controller with velocity feedforward were well tuned for tracking the
sinusoidal signal. The EHS designed for the quadruped robot and the flapper type servo-
valve manufactured by Star Hydraulic Ltd. were used. The effective stroke of the piston
range is ±0.026 m. The position and force feedback were measured by the potentiometer
and load cell, respectively. The velocity was obtained by differentiating the position. The
derivatives are calculated by Tustin method. For experiments, the control algorithms
(the proposed method and PI controller with velocity feedforward) coded in C by an
S-function were used in real-time operating systems. The derivatives in the control law
formulation are calculated by Tustin method. The sampling rate of 1 KHz was used.
16-bit A/D and 16-bit D/A converters were used. A photograph of the EHS test rig is
shown in Figure 4.8. The position tracking performances of both the methods are shown
in Figure 4.9 and Figure 4.10. The forces of both methods were asymmetric since the zero
angle of the load and the zero position of the actuator were not identical. Furthermore,
the oscillations in the force appeared due to the structural vibration. Therefore, static
friction was obscured since the amplitude of the static friction was smaller than that of
72
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
the oscillations. Figure 4.10 (d) shows the disturbance estimation result. The estimated
disturbance includes biased sinusoidal signal, effects of the assumptions for the modeling
and the parameter uncertainties, and structural vibration whose frequencies are below
the cut-off frequency of DOB. Since the estimated disturbance was similar to the force,
we can conclude that the disturbance was estimated well according to the simulation
result. Although there were harmonics and mechanical oscillations in the disturbance,
the disturbance was estimated correctly since the proposed DOB is in the form of a
second order high pass filter. Therefore, the position tracking performance was improved
by a proposed method compared with PI controller with velocity feedforward.
73
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
10 Reference
Actual Position
Position [mm]
−5
−10
10 11 12 13 14 15 16
Time [second]
(a) Position tracking performance
1
Position error [mm]
0.5
−0.5
−1
10 11 12 13 14 15 16
Time [second]
(b) Position tracking error (xd1 − x1 )
200
100
Force
−100
−200
10 11 12 13 14 15 16
Time [second]
(c) Force (Ap x3 )
74
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
10 Reference
Actual Position
Position [mm]
5
−5
−10
10 11 12 13 14 15 16
Time [second]
(a) Position tracking performance
1
Position error [mm]
0.5
−0.5
−1
10 11 12 13 14 15 16
Time [second]
(b) Position tracking error (xd1 − x1 )
200
100
Force [N]
−100
−200
10 11 12 13 14 15 16
Time [second]
(c) Force (Ap x3 )
200
Disturbance [N]
100
−100
−200
10 11 12 13 14 15 16
Time [second]
ˆ
(d) Force (Estimated disturbance d)
75
Chapter 4. Biased Sinusoidal Disturbance Estimation and Rejection Control
4.7 Summary
In this chapter, the biased sinusoidal disturbance estimation and rejection control
method were proposed. The DOB was designed in the form of a second-order high pass
filter in order to estimate the disturbance. The nonlinear controller was designed to
track the desired profile and compensate for the estimation error of the disturbance
based on the flatness concept. In the experiments, although the disturbance was not a
purely biased sinusoidal signal, the disturbance was successfully canceled by the proposed
method. Therefore, the position tracking performance of the proposed method was better
than that of a PI controller with velocity feed forward.
76
CHAPTER 5
Robust Backstepping Control
with Output Error Constraint
High gain disturbance observer (HGDOB) based robust backstepping controller with
a position tracking error constraint is proposed to improve the position tracking per-
formance in the presence of disturbances. The proposed method consists of a high gain
disturbance observer and a backstepping controller. The HGDOB is designed to esti-
mate the disturbances which include the friction, the load force, and the parameter
uncertainties. Auxiliary state variables are proposed in order to avoid amplification of
the measurement noise in HGDOB. In order to compensate for the disturbances while
guaranteeing tolerance of the position tracking error, a backstepping controller using
barrier Lyapunov function (BLF) is proposed.
5.1 Introduction
Variable structure control (VSC) methods were proposed for the control of the EHS
in [8, 11, 12]. The chattering in the control action of VSC may excite high frequency
modes. Input-output linearization has been also used to compensate for the global non-
linearities of EHSs [50, 51, 61]. However, its control input signal may often have high
amplitude due to cancellation of the nonlinear terms. EHS dynamics is a strict feed-
back form. Thus, backstepping control methods using the EHS dynamics properties have
been proposed [20–23, 26]. Although all of these methods improved the position or force
tracking performance, the tolerance for tracking performance cannot be arbitrarily de-
termined by the previous methods. The high control gain may satisfy the tolerance for
the tracking performance, but it may make the system unstable since the ripple due to
the modeling error, parameter uncertainties, and measurement noise is amplified.
77
Chapter 5. Robust Backstepping Control with Output Error Constraint
In practical hydraulic systems, the effective bulk modulus and the original volumes
(mainly including pipelines and cylinder chamber) are uncertain but important, so un-
known nonlinear parameters, which enter the system dynamics in a nonlinear way, are
common, and previous adaptive schemes of hydraulic systems cannot solve the problem
from the unknown nonlinear parameters. Hence, constant approximate estimated values
of these volumes are always used, but this makes the performance of the control system
unsatisfactory in some cases.
Recently, a BLF was applied to the backstepping control for the output constraint [62].
Its applications for various systems have been proposed [63–65]. In the presence of dis-
turbances, existing barrier Lyapunov function based feedback controllers occurred input
saturation because the value of the barrier Lyapunov function goes to infinity when-
ever its states approach the boundary, thereby satisfying the output constraint. For BLF
based control application to EHSs, disturbances, such as the friction, the load force, and
the parameter uncertainties, should be considered. Thus, not only the tracking perfor-
mance but also compensation of the disturbance is required. However, it is difficult to
measure the disturbance in most cases due to the limitation of space and/or cost concern.
Thus, the disturbance is generally estimated with an assumption that the disturbance is
constant and/or its derivative is bounded using high observer gain since the disturbance
dynamics is unknown [66, 69–71]. These previous observers have used the derivative of
measurement signal with measurement noise. Low-pass filters are generally used to re-
duce the measurement noise. Although the measurement noise of the measured signal is
suppressed, the measurement noise may be amplified when measured signal’s derivative
is used. Thus the disturbance observer without the use of the derivative of measurement
signal is required.
The main contribution of this chapter is the design and implementation of a robust
backstepping position controller with output tracking error constraint in the presence
of disturbances. The proposed method consists of a HGDOB and a backstepping con-
troller. The HGDOB is designed to estimate the disturbances which include the friction,
the load force, and the parameter uncertainties, i.e., the effective bulk modulus and the
control volumes. In addition, HGDOB is proposed in order to minimize the effect of dis-
turbance that causes the input saturation. Auxiliary state variables are proposed in order
to avoid amplification of the measurement noise in HGDOB. In order to compensate for
the disturbances while guaranteeing tolerance of the position tracking error, a backstep-
ping controller using BLF is proposed. As a result, the proposed method satisfies the
output constraint in the presence of the disturbances and improve the position tracking
performance. The proposed method is validated via simulations and experiments.
78
Chapter 5. Robust Backstepping Control with Output Error Constraint
In many EHS applications, the valve dynamics is much faster than the dynamics of the
remaining parts of the system such that the valve dynamics is often neglected without a
significant reduction in the control performance [50, 51, 61]. Therefore, for simplicity we
use the following approximation
xv = kv i (5.1)
Combining (2.16) and (2.25), the fluid dynamic equation of the actuator is given by
where
m mass of the piston [kg];
k load spring constant [N/m];
b viscous damping coefficient [N/(m/s)];
FL load force [N];
FF friction force [N].
The goal is to make the piston position track the desired position with disturbance
compensation. We define ∆f2 and ∆f3 which include the parameter uncertainty terms
as
where
k b
∆km and ∆bm uncertainties of m and m;
∆h1 and ∆h2 uncertainty of h1 and h2 ;
4βe Ap 4βe Ctl
h1 := Vt , and h2 := Vt .
79
Chapter 5. Robust Backstepping Control with Output Error Constraint
Combining (5.1)-(5.4) with ∆f2 and ∆f3 , the dynamics of the EHS can be reformulated
as the following state space representation:
1 x2
ẋ1 = |{z}
g1
1 Ap FL FF
ẋ2 = − (kx1 + bx2 ) + x3 − − + ∆f2
| m {z m
} |{z} | m m
{z }
f2 g2 d2 (5.5)
p
ẋ3 = −h1 x2 − h2 x3 + h3 Ps − sgn (u) x3 u + ∆f3
| {z } | {z } |{z}
f3 g3 d3
y = x1
where
x1 position of the piston [m];
x2 velocity of the piston [m/s];
x3 pressure difference between the chamber A and B [N/m2 ];
u=i current input, i [mA];
y output;
4βe Cd wkv
h3 := Vt ρ ,
√ and all states are measurable.
1
d2 = ẋ2 + (kx1 + bx2 − Ap x3 ) ,
m (5.6)
p
d3 = ẋ3 + h1 x2 + h2 x3 − h3 Ps − sgn (u) x3 u.
Let us define the estimations of the disturbances, dˆ2 and dˆ3 . The estimation errors are
defined as
80
Chapter 5. Robust Backstepping Control with Output Error Constraint
1 1
where ε2 and ε3 are the observer gains. In general, prior information about the derivative
of the disturbances is unknown but bounded, at least locally [74]. In this chapter, we
consider the situation that satisfies the following Assumption 5.1.
In fact, FF is not differentiable at zero velocity since it includes the Coulomb friction.
Fortunately, the actual friction have finite jumps so that the friction model can be chosen
as any differentiable function which approximates the actual discontinuous Coulomb
friction (for example, replacing sgn(x2 ) in the Coulomb friction modeling by the smooth
function tanh(x2 ) [43]. Thus, Assumption 5.1 is physically reasonable [43]. Actually, all
of the states of EHS are physically bounded, i.e., |x| < xmax where xmax is constant [73].
Thus, ∆f2 and ∆f3 are bounded. In (5.5), ẋ2 and ẋ3 consist of state variables, FL , and
FF . Thus, the derivatives of ẋ2 and ẋ3 are also bounded. Consequently, the constants
d˙2max and d˙3max exit such that
To suppress the bounded derivatives of the disturbances, the high gains, i.e., low values
of ε2 and ε3 , are required. Actually, measurement noises appear in the sensors. The
dynamics of dˆ2 and dˆ3 (5.8) use the derivative of the state. If high observer gains are
used, the noise is amplified by the high gains. Thus, the observer is not practical to
implement. To avoid this problem, we use auxiliary state variables, ξ2 , ξ3 .
xi
ξi = dˆi − , ∀i ∈ [2, 3]. (5.10)
εi
1 kx1 + bx2 − Ap x3
˙ξ2 = − 1 ξ2 + x2
+ ,
ε2 ε2 ε2 m
(5.11)
˙ξ3 = − 1 ξ3 + x3 1 p
+ h1 x2 + h2 x3 − h3 Ps − sgn (u) x3 u .
ε3 ε3 ε3
1
− t
Then, |d˜i | ≤ e εi · |d˜i (0)| + εi ρi (t) for an envelope functions ρi (t) such that ρi (t) ≥
|d˙i |, ∀t ≥ 0, ∀i ∈ [2, 3].
81
Chapter 5. Robust Backstepping Control with Output Error Constraint
Proof. Differentiating the auxiliary state variables with respect to time gives
˙ ẋi
ξ˙i = dˆi − , ∀i ∈ [2, 3]. (5.12)
εi
Substituting (5.12) into the disturbance estimation error (5.7) with respect to time results
in
˙ ẋi
d˜i = d˙i − ξ˙i − , ∀i ∈ [2, 3]. (5.13)
εi
1 kx1 + bx2 − Ap x3
˜˙ ˙ 1 x2
d2 = d2 − − ξ2 + +
ε2 ε2 ε2 m
kx1 + bx2 − Ap x3
1
− d2 −
ε2 m
˙ 1 x2
= d2 − d2 − ξ2 + ,
ε2 ε2
(5.14)
˜˙ ˙ 1 x3 1 p
d3 = d3 − − ξ3 + + h1 x2 + h2 x3 − h3 Ps − sgn (u) x3 u
ε3 ε3 ε3
1 p
− d3 − h1 x2 − h2 x3 + h3 Ps − sgn (u) x3 u
ε3
1 x3
= d˙3 − d3 − ξ3 + .
ε3 ε3
From (5.7), (5.10), and (5.14), the following disturbance estimation error dynamics is
obtained as
˙ 1
d˜i = − d˜i + d˙i , ∀i ∈ [2, 3]. (5.15)
εi
1
− t
Therefore, |d˜i | ≤ e εi · |d˜i (0)| + εi ρi (t).
1
− t
From the result |d˜i | ≤ e εi · |d˜i (0)| + εi ρi (t) ∀i ∈ [2, 3], the upper bound of |d˜i (∞)|
becomes smaller as εi gets smaller.
Remark 5.1. Observer (5.11) with auxiliary state variable (5.10) does not require the
derivatives of ẋ2 and ẋ3 to obtain dˆ2 and dˆ3 . Thus, if (5.10) and (5.11) are used to
estimate the disturbances instead of (5.8), amplifying the measurement noise by the high
gain can be reduced so that it is negligible in practice. ♦
82
Chapter 5. Robust Backstepping Control with Output Error Constraint
Definition 5.1. A barrier Lyapunov Function is a scalar function V (x), defined with
respect to the system ẋ = f (x) on an open region D containing the origin, that is con-
tinuous, positive definite, has continuous first-order partial derivatives at every point
of D , has the property V (x) → ∞ as x approaches the boundary of D, and satisfies
V (x(t)) ≤ b, ∀t ≥ 0 along the solution of ẋ = f (x) for x(0) ∈ D and some positive
constant b [62]. ♦
Figure 5.1 shows the schematic illustration of a control and barrier Lyapunov function.
83
Chapter 5. Robust Backstepping Control with Output Error Constraint
Consider the EHS (5.5) and the DOB (5.10) and (5.11) with |z1 (0)| < kb . Suppose
the control law is given by
z1 = x1 − yd ,
α1 = −(kb2 − z12 )k1 z1 ,
z2 = x2 − ẏd − α1 ,
1 (2) z1 ˆ (5.17)
α2 = −f2 − k2 z2 − (g2 − 1)yd + α̇1 − 2 − d2 ,
g2 kb − z12
(2)
z3 = x3 − yd − α2 ,
1 (3)
u= −f3 − dˆ3 + yd + α̇2 − k3 z3 − g2 z2
g3
kb2
1
V1 = log (5.19)
2 kb − z12
2
where log(·) denotes the natural logarithm function. The derivative of V1 with respect
to time is given by
z1 ż1
V̇1 =
kb2 − z12
(5.20)
z1 (z2 + α1 )
= .
kb2 − z12
84
Chapter 5. Robust Backstepping Control with Output Error Constraint
z1 z2
V̇1 = −k1 z12 + .
| {z } kb2 − z12 (5.21)
Negative Term | {z }
Cross Term
(2)
ż2 = ẋ2 − yd − α̇1
(2)
(5.22)
= f2 + g2 x3 + d2 − yd − α̇1
δα1
where α̇1 = δz1 ż1 .
Let us define V2 as
1 1
V2 = V1 + z22 + d˜22 , (5.23)
2 2
then,
˙
V̇2 = V̇1 + z2 ż2 + d˜2 d˜2
(5.24)
˙
(2)
= V̇1 + z2 f2 + g2 x3 + d2 − yd − α̇1 + d˜2 d˜2 .
85
Chapter 5. Robust Backstepping Control with Output Error Constraint
1 1
where λ2 = ε2 − 4k2 .
(3)
ż3 = f3 + g3 u + d3 − yd − α̇2 (5.28)
1 1
V3 = V2 + z32 + d˜23 . (5.29)
2 2
˙
V̇3 = V̇2 + z3 ż3 + d˜3 d˜3 . (5.30)
˙
(3)
V̇3 = V̇2 + z3 f3 + g3 u + d3 − yd − α̇2 + d˜3 d˜3 . (5.31)
86
Chapter 5. Robust Backstepping Control with Output Error Constraint
= −k1 z12
2 2
1 ˜ 1 ˙ 1 ˙2
− k2 z2 − d2 − λ2 |d˜2 | − d2max + d
2k2 2λ2 4λ2 2max (5.32)
2 2
1 ˜ 1 ˙ 1 ˙ 2
− k3 z3 − d3 − λ3 |d˜3 | − |d3 | + |d3 |
2k3 2λ3 4λ3
1 1
where λ3 = ε3 − 4k3 .
3 2 2 !
1 ˜ 1
di + λi |d˜i | − d˙i
X
V̇3 ≤ −k1 z12 − ki zi −
i=2
2ki 2λi max
| {z }
Negative Terms
(5.33)
3
X 1 ˙2
+ d .
i=2
4λi imax
| {z }
Diturbance Term
If we take εi < 4ki ∀i ∈ [2, 3], then |z1 (t)| < kb , ∀t > 0, and zex (t) enters into the
bounded ball Br in the finite time t1 > 0, and stays within Br , ∀t > t1 .
Figure 5.2 shows the block diagram of the control system. HGDOB (5.10) and (5.11)
estimates the disturbances d2 and d3 using x and u. Then the current input u is obtained
by the BLF controller (5.17) using x, dˆ2 , dˆ3 and yd .
87
Chapter 5. Robust Backstepping Control with Output Error Constraint
Remark 5.2. From Theorem 5.2, the size of the boundedness ball Br mainly depends
3
1 ˙2
P
on 4λi dimax . Thus, the high observer gain, i.e., low values of ε2 and ε3 , can shrink
i=2
size of Br . In addition, if d2 and d3 are constants, d˙2 and d˙3 become zeros, from (31),
we see that V̇3 becomes
3 2 X3
1 ˜
λi d˜2i .
X
V̇3 = −k1 z12 − ki zi − di − (5.34)
i=2
2ki i=2
Thus it is clear that |z1 (t)| < kb , ∀t > 0 and zex (t) asymptotically converges to zero. ♦
(2)
Remark 5.3. The control law (16) includes α̇1 , α1 , and α̇2 . In the experiments, we
numerically approximated them with forward Euler method since analytic solutions are
very involved. In the numerical simulations, performance difference between the numerical
approximation and the analytic solution was hardly differentiated. ♦
88
Chapter 5. Robust Backstepping Control with Output Error Constraint
In the simulation and experiments, the following parameters were used: m = 10, k =
50, b = 1000, Cd = 0.6, Ap = 4.812 × 10−4 , βe = 1.8 × 109 , Vt = 6.2 × 10−5 ,
Cil = 2.48815 × 10−14 , Cel = 1.666 × 10−14 , w = 5.2 × 10−3 , ρ = 840, kv = 1.33 × 10−5 ,
Ps = 12.0 × 106 , k1 = 130, k2 = 700, k3 = 130, ε2 = 6.7 × 10−4 , ε3 = 1.0 × 10−3 ,
and kb = 1.0 × 10−3 . The position/output tracking error constraint, kb , is 1 mm. The
proposed method (5.17), i.e., BLF based backstepping control with HGDOB (BLF w/
HGDOB), was compared with the following two controllers:
1) The control Lyapunov function based backstepping control without HGDOB (CLF
w/o HGDOB):
1 (3)
u= −f3 − k3 z3 − g2 z2 + yd + α̇2 (5.35)
g3
where
(2)
z1 = x1 − yd , z2 = x2 − ẏd − α1 , z3 = x3 − yd − α2 ,
α1 = −k1 z1 ,
1 (2)
α2 = −f2 − k2 z2 − z1 − (g2 − 1)yd + α̇1 .
g2
2) The BLF based backstepping control (5.17) without HGDOB (BLF w/o HGDOB).
The sinusoidal desired position profile shown in Figure 5.3 was used. In the simulation,
since the main friction is generally the Coulomb friction in the EHS, the friction, FF =
sgn(x2 ) N, was used. In this simulations, to verify the performance of the proposed
method, the following three cases were processed using Matlab/Simulink:
• Case 2: Disturbance consists of load force and friction, i.e., d2 = − FmL − FmF , d3 = 0
where FL = −600 N and FF = sgn(x2 ) N.
89
Chapter 5. Robust Backstepping Control with Output Error Constraint
Position [mm]
20 yd
−20
0 1 2 3 4 5 6 7 8
Time [sec]
Figure 5.4 shows the position tracking errors of the three methods and the estimation
performances of disturbances. In Figure 5.4 (a), the green dashed lines indicate the
tolerance of the position tracking error, kb = 1 mm. Due to the parameter uncertainties,
the position tracking performance of the CLF without HGDOB was the worst among the
three methods and did not satisfy the position tracking error constraint, (|z1 | < 1 mm).
As shown in Figure 5.4 (b), BLF without HGDOB satisfied the position tracking error
constraint, (|z1 | < 1 mm) since the controller was designed based on BLF. However, the
performance of BLF without HGDOB was relatively poorer than that of the proposed
method shown in Figure 5.4 (c). Figure 5.4 (d) and (f) show that the disturbances were
well estimated so that the proposed method has the best position tracking performance
among the three methods. The three performances of the methods in Case 2 are shown
in Figure 5.5. As in Case 2, the position tracking performances of CLF without HGDOB
and of BLF without HGDOB were worse than that of the proposed method. Furthermore,
due to the load force and the friction, offset tracking errors appeared in both CLF without
HGDOB and BLF without HGDOB. On the other hand, the proposed method had almost
zero position tracking error since the load force and the friction were compensated for by
the proposed method. In Figure 5.5 (e), Figure 5.5 (g), Figure 5.6 (e), and Figure 5.6 (g),
small estimation errors in the form of impulse signal appeared because of the friction,
sgn(x2 ). The simulation results of Case 3 are shown in Figure 5.6. In Case 3, all of the
parameter uncertainties, the load force, and the friction were considered. As in Cases 1
and 2, the proposed method had the best position tracking performance among the three
methods while satisfying the position tracking error constraint, (|z1 | < 1 mm).
90
Chapter 5. Robust Backstepping Control with Output Error Constraint
5
Position error [mm]
yd − x1
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(a) CLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(b) BLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(c) BLF w/ HGDOB (proposed method)
91
Chapter 5. Robust Backstepping Control with Output Error Constraint
100
d2
d2 [m/s2 ]
50
dˆ2
0
−50
−100
0 1 2 3 4 5 6 7 8
Time [sec]
(d) Estimation performance of d2
5
d2 − dˆ2
d2 − dˆ2
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(e) Estimation error of d2
9
x 10
2
d3 [N/m2 s]
d3
dˆ3
0
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(f ) Estimation performance of d3
8
x 10
1
d3 − dˆ3
d3 − dˆ3
−1
0 1 2 3 4 5 6 7 8
Time [sec]
(g) Estimation error of d3
92
Chapter 5. Robust Backstepping Control with Output Error Constraint
5
Position error [mm]
yd − x1
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(a) CLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(b) BLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(c) BLF w/ HGDOB (proposed method)
93
Chapter 5. Robust Backstepping Control with Output Error Constraint
100
d2
d2 [m/s2 ]
50
dˆ2
0
−50
−100
0 1 2 3 4 5 6 7 8
Time [sec]
(d) Estimation performance of d2
5
d2 − dˆ2
d2 − dˆ2
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(e) Estimation error of d2
9
x 10
2
d3 [N/m2 s]
d3
dˆ3
0
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(f ) Estimation performance of d3
8
x 10
1
d3 − dˆ3
d3 − dˆ3
−1
0 1 2 3 4 5 6 7 8
Time [sec]
(g) Estimation error of d3
94
Chapter 5. Robust Backstepping Control with Output Error Constraint
5
Position error [mm]
yd − x1
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(a) CLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(b) BLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(c) BLF w/ HGDOB (proposed method)
95
Chapter 5. Robust Backstepping Control with Output Error Constraint
100
d2
d2 [m/s2 ]
50
dˆ2
0
−50
−100
0 1 2 3 4 5 6 7 8
Time [sec]
(d) Estimation performance of d2
5
d2 − dˆ2
d2 − dˆ2
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(e) Estimation error of d2
9
x 10
2
d3 [N/m2 s]
d3
dˆ3
0
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(f ) Estimation performance of d3
8
x 10
1
d3 − dˆ3
d3 − dˆ3
−1
0 1 2 3 4 5 6 7 8
Time [sec]
(g) Estimation error of d3
96
Chapter 5. Robust Backstepping Control with Output Error Constraint
10
yd
Position [mm] 5
−5
−10
0 1 2 3 4 5 6 7 8
Time [sec]
The simulations were tested with the near rectangular desired position as shown in Fig-
ure 5.7. Since the rectangular desired position cannot satisfy |z1 (t)| < kb when the rect-
angular desired position steps up, we used the near rectangular desired position. In this
simulations, to verify the performance of the proposed methods, the following case was
processed:
The performances of three methods were shown in Figure 5.8. Compared to the sinu-
soidal desired position, the near rectangular desired position was stiff. Thus, CLF without
HGDOB did not satisfy the constraint. BLF without HGDOB satisfied the constraint,
however, the performance of BLF without HGDOB was relatively poorer than that of
the proposed method. Although the near rectangular desired position was stiff, the dis-
turbances were estimated well. Thus, the performance of BLF with HGDOB was the
best among three methods.
• Case 6: Case 4 conditions with measurement noises, |x1n | < 50 µm and |x3n | <
50000 N/m2 where x1n is the position measurement noise and x3n is the load
pressure measurement noise.
97
Chapter 5. Robust Backstepping Control with Output Error Constraint
Actually, since the velocity was numerically approximated by differentiating the position,
the measurement noises for x1 and x3 were considered. Figure 5.9 shows the estimation
results of Case 5. Since the DOB (5.8) required derivatives of the measured signals to
estimate the disturbances, the nonzero initial desired velocity of yd and friction resulted
in the large ripple in the estimated disturbances of the DOB. On the other hand, the
HGDOB (5.10) and (5.11) did not have ripple in the estimated disturbances since the
HGDOB estimated the disturbances without derivatives. The estimation results of Case
6 are shown in Figure 5.10. Using the derivatives of the measured signals in DOB resulted
in the amplification of the measurement noises. Thus the estimated disturbances of the
DOB had large ripples. On the other hand, the HGDOB had small ripples in the estimated
disturbances since the measurement noises were not amplified.
98
Chapter 5. Robust Backstepping Control with Output Error Constraint
5
Position error [mm]
yd − x1
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(a) CLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(b) BLF w/o HGDOB
2
Position error [mm]
yd − x1
1
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(c) BLF w/ HGDOB (proposed method)
99
Chapter 5. Robust Backstepping Control with Output Error Constraint
100
d2
d2 [m/s2 ]
50
dˆ2
0
−50
−100
0 1 2 3 4 5 6 7 8
Time [sec]
(d) Estimation performance of d2
5
d2 − dˆ2
d2 − dˆ2
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(e) Estimation error of d2
9
x 10
d3 [N/m2 s]
5 d3
dˆ3
0
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(f ) Estimation performance of d3
9
x 10
2
d3 − dˆ3
d3 − dˆ3
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(g) Estimation error of d3
100
Chapter 5. Robust Backstepping Control with Output Error Constraint
20 BLF w/ DOB
d2 − dˆ2
BLF w/ HGDOB
0
−20
0 0.5 1 1.5
Time [sec]
(a) Estimation errors of d2
9
x 10
2
BLF w/ DOB
d3 − dˆ3
BLF w/ HGDOB
0
−2
0 0.5 1 1.5
Time [sec]
(b) Estimation errors of d3
101
Chapter 5. Robust Backstepping Control with Output Error Constraint
100
dˆ2 [m/s2 ]
50
0
BLF w/ DOB
−50
BLF w/ HGDOB
−100
0 0.5 1 1.5
Time [sec]
(a) Estimation performance of d2
20 BLF w/ DOB
d2 − dˆ2
BLF w/ HGDOB
0
−20
0 0.5 1 1.5
Time [sec]
(b) Estimation errors of d2
9
x 10
dˆ3 [N/m2 s]
5 BLF w/ DOB
BLF w/ HGDOB
0
−5
0 0.5 1 1.5
Time [sec]
(c) Estimation performance of d3
9
x 10
2
BLF w/ DOB
d3 − dˆ3
BLF w/ HGDOB
0
−2
0 0.5 1 1.5
Time [sec]
(d) Estimation errors of d3
102
Chapter 5. Robust Backstepping Control with Output Error Constraint
103
Chapter 5. Robust Backstepping Control with Output Error Constraint
load force, and the friction, the position tracking performance of the CLF without HG-
DOB did not satisfy the position tracking constraint, (|z1 | < 1 mm). Although BLF
without HGDOB satisfied the position tracking error constraint, (|z1 | < 1 mm), the
performance of BLF without HGDOB was relatively poorer than that of the proposed
method due to the disturbance. Furthermore, the disturbance resulted in offset position
tracking errors in both CLF without HGDOB and BLF without HGDOB. On the other
hand, the proposed method had the best position tracking performance among the three
methods while satisfying the position tracking error constraint, (|z1 | < 1 mm) since the
proposed method compensated for the disturbances. Figure 5.13 (a) shows the distur-
bance estimation result that included the parameter uncertainties, the load force, and
the friction. From Figure 5.5 (d) and Figure 5.6 (d) in the simulations, it was observed
that the dominant factor of d2 was the load force although the parameter uncertainties
were larger than in the actual case for the verification of the robustness to parameter
uncertainties. Thus, in order to verify the estimation performance, the estimated distur-
bance was compared to load force. The actual load force was shown in Figure 5.13 (b).
Note that the estimated disturbance was similar to the force since the required force to
move the actuator was much smaller than that required to overcome the disturbance
(the load force and the uncertainties) [61]. Ripples in the load force appeared because of
the structural vibration and measurement noises. Since the estimated disturbance dˆ2 was
similar to the measured force, we can conclude that the disturbance was well estimated
although there was some the structural vibration. Also, as discussed in Remark 5.1, since
we did not use the derivatives of xi to estimate the disturbances, the measurement noise
was suppressed significantly. Due to the estimations and the compensation of the distur-
bances, the position tracking error was satisfied and the position tracking performance
was improved using the proposed method.
104
Chapter 5. Robust Backstepping Control with Output Error Constraint
−5
0 1 2 3 4 5 6 7 8
Time [sec]
(a) CLF w/o HGDOB
20
Current [mA]
−20
0 1 2 3 4 5 6 7 8
Time [sec]
(b) Control input of CLF w/o HGDOB
2
Position error [mm]
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(c) BLF w/o HGDOB
20
Current [mA]
−20
0 1 2 3 4 5 6 7 8
Time [sec]
(d) Control input of BLF w/o HGDOB
105
Chapter 5. Robust Backstepping Control with Output Error Constraint
2
Position error [mm]
−1
−2
0 1 2 3 4 5 6 7 8
Time [sec]
(e) BLF w/ HGDOB (proposed method)
20
Current [mA]
−20
0 1 2 3 4 5 6 7 8
Time [sec]
(f ) Control input of BLF w/ HGDOB (proposed method)
106
Chapter 5. Robust Backstepping Control with Output Error Constraint
0
Load force [N]
−500
−1000
0 1 2 3 4 5 6 7 8
Time [sec]
(a) Estimated load force, F̂L ' −mdˆ2
0
Load force [N]
−500
−1000
0 1 2 3 4 5 6 7 8
Time [sec]
(b) Actual load force, FL
5.6 Summary
In this chapter, the HGDOB based robust backstepping controller with a position
tracking error constraint for EHSs was proposed in order to improve the position track-
ing performance in the presence of disturbances. The HGDOB was designed to estimate
the disturbances which include the friction, the load force, and the parameter uncertain-
ties while avoiding amplification of the measurement noise due to the introduction of the
auxiliary variables. The backstepping controller was proposed using the BLF to com-
pensate for the disturbances while guaranteeing the tolerance of the position tracking
error. In addition, the effect of input saturation caused by disturbances can be mini-
mized using this method. It was shown that the proposed methods improve the position
tracking performance and satisfy the position tracking error constraint via simulations
and experiments.
107
CHAPTER 6
Conclusions
In this dissertation, we presented three nonlinear position control methods for EHSs
using the Lyapunov method. First, we presented a position tracking control based on an
output feedback nonlinear control using the passivity-based control method. The high
gain observer was designed to estimate the full state, and the high gain technique is
used to reduce the effects of the nonlinear terms. The passivity-based controller was
implemented for position tracking. The problem is solved in this work by selecting the
controller gain such that the origin of the tracking error dynamics is exponentially stable.
We proved that the closed-loop system is asymptotically stable (under the selection
criteria) using the singular perturbation method.
Second, we studied the design and implementation of a nonlinear position tracking
controller with a disturbance observer for EHSs in the presence of a biased sinusoidal
disturbance. The DOB, whose simplicity results from it requiring only mechanical pa-
rameters, was designed as of a second-order high pass filter for estimating the disturbance
without any additional algorithms. The nonlinear controller was designed to track the de-
sired position, operating as a near IO linearizing inner-loop load pressure controller and a
backstepping outer-loop position controller. Variable structure control was implemented
in order to compensate for the error in the disturbance estimation. The desired load
pressure was generated using the differential flatness property of the EHS mechanical
subsystem. The disturbance includes not only a biased sinusoidal signal, but also the ef-
fects of the modeling assumptions and the parameter uncertainties within the bandwidth
of the DOB; it can be canceled out using the proposed method.
Finally, we proposed a high gain disturbance observer based robust backstepping con-
troller, with a position tracking error constraint, to improve the position tracking perfor-
mance in the presence of disturbances. The proposed methods guarantee the tolerance for
the position tracking error in the presence of disturbances. The proposed method consists
108
Chapter 6. Conclusions
of a HGDOB and a backstepping controller. The HGDOB was designed to estimate the
disturbances, which include friction, load force, and parameter uncertainties. In order to
compensate for the disturbances while guaranteeing tolerance of the position tracking er-
ror, a backstepping controller using BLF was proposed. As a result, the proposed method
satisfies the output constraint in the presence of disturbances and improves the position
tracking performance. In addition, the effect of input saturation caused by disturbances
can be minimized using this method.
The performance of the proposed methods is verified through simulations and validated
through experiments. Compared to previous methods, the proposed method improves the
position tracking performance.
In future work, the output feedback control method will be studied for the output
feedback, while accounting for disturbances. In the proposed output feedback controller,
the disturbance was not considered for position tracking. The DOB based control method
requires full state feedback. To overcome these problems, the position control method
will be designed to track the desired position, while accounting for disturbances.
109
APPENDIX A
Review of the Nonlinear Control
Theorem
The mathematical definition and the remark are from [76]. The emphasize is on the
interpretation and the remarks, the definition is included to complete the description.
Definition
Lipschitz-continuity comes in three different flavours.
Let f : Rm → Rm .
110
Appendix
Remark
Interpretation
• "Lipschitz just means f can’t be too steep, the bound on the difference quotient
being the Lipschitz constant" [76].
ẋ = f (x). (A.3)
Local Stability
If, in a ball BR0 there exists a scalar function V (x) with continuous first partial
derivatives such that
then the equilibrium point x = 0 is stable. If, actually, the derivative V̇ (x) is locally
negative definite in BR0 , then the stability is asymptotic [36, 77, 78].
Global Stability
Assume that there exists a scalar function V of the states x with continuous first order
derivatives such that
111
Appendix
then the equilibrium at the origin is globally asymptotically stable [36, 77, 78].
associated with a static feedback. To introduce the main definition we have to refer to a
class C of controllers. The main classes considered here are
Definition
Given a class of controllers C and a locally Lipschitz positive definite function Ψ (and
possibly a set P) we say that Ψ is a global control Lyapunov function (a Lyapunov
function outside P or a Lyapunov function inside P) if there exists a controller in C such
that [79]:
• for each initial condition x(0) there exists a solution x(t), for any admissible w(t),
and each of such solutions is defined for all t ≥ 0;
112
Appendix
In this case, we say that Ψ is a global control Lyapunov function (a Lyapunov function
outside P or a Lyapunov function inside P) if there exists a controller (in a specified
class C) such that, beside the conditions in above Definition, satisfies the constraints.
Note also that the problem with state constraints can be easily addressed. If we assume
that x(t) ∈ X is a hard constraint to be satisfied, we can immediately argue that as long
as N [Ψ, µ] = {x : Ψ(x) ≤ µ} ⊆ X , for some µ, and Ψ is a control Lyapunov function
(either global, inside or outside P), then the constraints can be satisfied satisfied by
means of a proper control action as long as x(0) ∈ N [Ψ, µ] = {x : Ψ(x) ≤ µ}.
ẋ = f (x) + g(x)ξ1
ξ˙1 = f1 (x, ξ1 ) + g1 (x, ξ1 )ξ2
ξ˙2 = f2 (x, ξ1 , ξ2 ) + g2 (x, ξ1 , ξ2 )ξ3
.. (A.5)
.
ξ˙k−1 = fk−1 (x, ξ1 , · · · , ξk−1 ) + gk−1 (x, ξ1 , · · · , ξk−1 )ξk
ξ˙k = fk (x, ξ1 , · · · , ξk ) + gk (x, ξ1 , · · · , ξk )u
where x ∈ Rn is the state and u ∈ R is the control input. There exist a continuously
differentiable feedback control law
∂V (x)
[f (x) + g(x)α(x)] ≤ −W (x) ≤ 0, ∀x ∈ Rn , (A.8)
∂x
113
Appendix
Remark
It is important to notice that:
Consider the strict-feedback systems (A.5), the Lyapunov function and the control law
will be constructed in a recursive manner [80].
Step 0
Design a continuously differentiable stabilizing function ξ1 = α(x) for the x subsystem;
i.e., construct a positive definite, radially unbounded function V (x) such that, with this
control law, its time derivative
∂V (x)
[f (x) + g(x)α(x)] ≤ −W (x) (A.9)
∂x
ẋ = f (x) + g(x)ξ1
(A.10)
ξ˙1 = f1 (x, ξ1 ) + g1 (x, ξ1 )ξ2 .
ξ1 = α(x) (A.11)
stabilizes the x subsystem. To take into account the deviation of the state variable ξ1
from the stabilizing function α1 (x), we define the error variable
z1 = ξ1 − α(x). (A.12)
114
Appendix
Then
∂α(x)
ż1 = ξ˙1 − ẋ
∂x (A.13)
∂α(x)
= f1 (x, ξ1 ) + g1 (x, ξ1 )ξ2 − (f (x) + g(x)(α(x) + z1 )) .
∂x
1
V1 = V (x) + z12 . (A.14)
2
We want to design a stabilizing function ξ2 = α1 (x, z1 ) such that the time derivative of
the Lyapunov function V1 is negative defnite.
∂V (x)
V̇1 ≤ − W (x) + g(x)z1
∂x
(A.16)
∂α(x)
+ z1 f1 (x, ξ1 ) + g1 (x, ξ1 )ξ2 − (f (x) + g(x)(α(x) + z1 )) .
∂x
It is clear that, if g − 1(x, ξ1 ) 6= 0, by choosing the stabilizing function for the virtual
control ξ2 as
ξ2 = α1 (x, z1 )
1
∂V ∂α(x)
(A.17)
= −k1 z1 − (x)g(x) − f1 (x, ξ1 ) + (f (x) + g(x)(α(x) + z1 ))
g1 ∂x ∂x
Step 2
115
Appendix
ẋ = f (x) + g(x)ξ1
ξ˙1 = f1 (x, ξ1 ) + g1 (x, ξ1 )ξ2 (A.19)
ξ˙2 = f2 (x, ξ1 , ξ2 ) + g2 (x, ξ1 , ξ2 )ξ3 .
z2 = ξ2 − α1 (X1 ). (A.21)
1
V2 = V1 (X1 ) + z22 . (A.22)
2
We can design a stabilizing function ξ3 = α2 (X1 , z2 ) such that the time derivative of the
Lyapunov function V2 is negative definite.
This recursive procedure will terminate at the k − th step, where the actual control
law for u will be designed.
Passivity Definition
The system
ẋ = f (x, u)
(A.23)
y = h(x, u)
116
Appendix
∂V
uT y ≥ V̇ = f (x, u), ∀(x, u) ∈ Rn × Rp . (A.24)
∂x
Moreover, it is said to be
• The system is input strictly passive if uT y ≥ V̇ + uT φ(u) and uT φ(u) > 0, for all
u 6= 0;
• The system is output strictly passive if uT y ≥ V̇ + y T ρ(y) and y T ρ(y) > 0, for all
y 6= 0;
• The system is strictly passive if uT y ≥ V̇ + ψ(x) for some positive definite function
ψ.
In all cases, the inequality should hold for all (x, u) [36].
Output Strictly Passive
• If the system
ẋ = f (x, u)
(A.25)
y = h(x, u)
Zero-state Observability
• The system
ẋ = f (x, u)
(A.26)
y = h(x, u)
117
Appendix
• If the system
ẋ = f (x, u)
(A.27)
y = h(x, u)
with f (0, 0) = h(0, 0) = 0 is passive with a positive definite storage function V (x),
then the origin of ẋ = f (x, 0) is stable.
– strictly passive or
– output strictly passive and zero-state observable.
then the origin x = 0 can be globally stabilized by u = −φ(y), where φ is any locally
Lipschitz function such that φ(0) = 0 and y T φ(y) > 0 for all y 6= 0 [36].
Differentially flat
m
XX (j+1) ∂
τ= yi (j) (A.28)
j≥0 i=1 ∂yi
118
Appendix
• Explicit Case. The explicit system ẋ = f (x, u) with m inputs is flat if and only
if there exists a flat output y of dimension m, two integers r and s and mappings
ψ from X × (Rm )s+1 to Rm , of rank m in a suitably chosen open subset, and
(φ0 , φ1 ) from R(m+2)r to Rn × Rm , of rank n + m in a suitable open subset, such
that y = (y1 , . . . , ym ) = ψ(x, u, u̇, . . . , u(s) ) implies that x = φ0 (y, ẏ, . . . , y (r) ),
dφ0
u = φ1 (y, ẏ, · · · , y (r+1) ), the differential equation dt = f (φ0 , φ1 ) being identically
dφ0
satisfied. It is easily verified that if x = φ0 (ȳ) and u = φ1 (ȳ) with dt = f (φ0 , φ1 )
2
dφ1 d φ1 dy (j) dj y
we have (x, ū) = Φ(ȳ) = (φ0 (ȳ), φ1 (ȳ), dt (ȳ), dt2 (ȳ), . . .), and dt = dtj =
(j+1)
y .
∂F
• Implicit Case. The implicit system F (x, ẋ) = 0 with rank ∂ ẋ = n − m, is
flat if and only if there exists an integer s and a mapping ψ from X × (Rn )s to
Rm , of rank m in a suitably chosen open subset, such that y = (y1 , . . . , ym ) =
ψ(x, ẋ, . . . , x(s) ) implies x = φ0 (y, ẏ, . . . , y (r) ) for a suitable integer r, the implicit
equation F (φ0 , dφ
dt ) = 0 being identically satisfied, with
0 dφ0
dt = Lτm φ0 .
In all cases, one can express all the system variables in function of the flat output and a
finite number of its successive derivatives [81].
119
Bibliography
[1] Herbert. E. Merrit, "Hydraulic Control System," Wiley & Sons, New York, 1967.
[2] Manring ND., "Hydraulic Control Systems," Hoboken, John Wiley & Sons, New
York, 2005.
[3] Clark, D., "Selection and performance criteria for electrohydraulic servodrives,"
in Proceedings of the 25th annual meeting of the national Conference on Fluid
Power, Reprint in Moog technical bulletin #122, 1969.
[6] Truong DQ, Ahn KK. "Force control for hydraulic load simulator using self-tuning
grey predictor-fuzzy PID," Mechatronics, vol. 19, no. 2, pp. 233-246, 2009.
[7] Yu WS, Kuo TS. "Continuous-time indirect adaptive control of the electrohy-
draulic servo systems," IEEE Transactions on Control Systems Technology, vol.
5, no. 2, pp. 163-177, 1997.
[8] T. Chen and Y. Wu, "An optimal variable structure control with integral compen-
sation for electrohydraulic position servo control systems," IEEE Transactions on
Industrial Electronics, vol. 39, pp. 460-463, Oct. 1992.
120
Bibliography
[10] Nguyen QH, Ha QP, Rye DC, Durrant-Whyte HF. "Force/position tracking for
electrohydraulic systems of a robotic excavator," in Proceedings of the IEEE Con-
ference on Decision Control, 2000, pp. 5224-5229.
[11] Jerouane M, Lamnabhi-Lagarrigue F. "A new sliding mode controller for a hy-
draulic actuators," in Proceedings of the IEEE Conference on Decision Control,
2001, pp. 908-913.
[12] Bonchis A, Corke PI, Rye DC, Ha QP. "Variable structure methods in hydraulic
servo systems control," Automatica, vol. 37, no. 4, pp. 589-595, 2001.
[13] C. Byrnes, A. Isidori, and J. C. Willems, "Passivity, feedback equivalent, and the
global stabilization of minimum phase nonlinear systems," IEEE Transactions on
Automation Control, vol. 36, no. 11, pp. 1228-1240, 1991.
[16] Shol GA, Bobrow JE. "Experimental and simulations on the nonlinear control
of a hydraulic servosystem," IEEE Transactions on Control Systems Technology,
vol. 7, no. 3, pp. 238-247, 1999.
[17] Sirouspour MR, Salcudean SE. "On the nonlinear control of hydraulic servo-
systems," in Proceedings of the IEEE International Conference on Robotics and
Automation, 2000, pp. 1276-1282.
[18] Sadegh, N., and Horowitz, R., "Stability and robustness analysis for a class of
adaptive controllers for robotic manipulators," International Journal of Robotics
Research, vol. 9, no. 3, pp. 74–92, 1990.
[19] Yao B, Bu F, Reedy J, Chiu, GTC. "Adaptive robust motion control of single-
rod hydraulic actuators: theory and experiments", IEEE/ASME Transactions on
Mechatronics, vol. 5, no. 1, pp. 79-91, 2000.
121
Bibliography
[21] Zeng H, Sepehri N. "Tracking control of hydraulic actuators using a LuGre fric-
tion model compensation," ASME Journal of Dynamic Systems Measurement and
Control, vol. 120, no. 1, pp. 014502-1-014502-7, 2008.
[22] A. Alleyne and R. Liu, "A simplified approach to force control for electro-hydraulic
systems," Control Engineering Practice, vol. 8, no. 12, pp. 1347-1356, 2000.
[25] Bu, F., Yao, B. "Integrated direct/indirect adaptive robust motion control of
single-rod hydraulic actuators with time-varying unknown inertia," in Proceedings
of 2001 IEEE/ASME International Conference on Advanced Intelligent Mecha-
tronics, 2001. pp. 624-629.
[26] Kim W, Won D, and Chung CC. "High gain observer based nonlinear control for
electro-hydraulic servo systems," in Proceedings of the American Control Confer-
ence, 2010, pp. 1440-1445.
[28] Alleyne A, Liu R. "Systematic control of a class of nonlinear systems with applica-
tion to electrohydraulic cylinder pressure control," IEEE Transactions on Control
Systems Technology, vol. 8, no. 4, pp. 623-634, 2000.
[29] Kugi A. "Non-linear control based on physical models", No. 260 in Lecture Notes
in Control and Information Sciences, Springer, 2000.
122
Bibliography
[32] Hedrick JK, Rajamani R, Yi K. "Observer design for electronic suspension appli-
cations," Vehicle System Dynamics, vol. 23, no. 1, pp. 413-440, 1994.
[33] Rajamani R, Hedric JK. "Adaptive observers for active automotive suspensions:
theory and experiment," IEEE Transactions on Control Systems Technology,
vol. 3, no. 1, pp. 86-93, 1995.
[34] Khan H, Abou SC, Sepehri N. "Nonlinear observer-based fault detection technique
for electro-hydraulic servo-positioning systems," Mechatronics, vol. 15, no. 9,
pp. 1037-1059, 2005.
[35] Girin A, Plestan F, Brun X, Glumineau A, Smaoui M. "High gain and sliding
mode observers for the control of an electropneumatic actuator," in Proceedings
of the IEEE International Conference on Control Application, 2006, pp. 3128-
3133.
[36] Hassan K. Khalil, "Nonlinear Systems(3rd ed.)," 3rd Edition, Upper Saddle River,
Prentice-Hall, New York, 2002.
[37] Dan Simon, "Optimal State Estimation: Kalman, H Infinity, and Nonlinear Ap-
proaches," Wiley & Sons, 2006.
[39] Bu F, Yao B. "Observer based coordinated adaptive robust control of robot ma-
nipulators driven by single-rod hydraulic actuators," in Proceedings of the IEEE
International Conference on Robotics & Automation, 2000, pp. 3034-3039.
[41] Thayer WJ. "Transfer Functions for Moog Servovalves", Moog Technical Bulleti,
1965.
[42] Hong Y, Zheng-jin F, Xu-yong W. "Nonlinear control for a class of hydraulic servo
system," Journal of Zhejiang University. Science 2004, vol. 5, no. 11, pp. 1413-
1417, 2004.
123
Bibliography
[43] Yao B, Bu FP, Chiu GTC. "Nonlinear adaptive robust control of electro-hydraulic
servo system with discontinuous projection," in Proceedings of the IEEE Confer-
ence on Decision Control, 1998, pp. 2265-2270.
[44] Do KD, Pan J. "Boundary control of transverse motion of marine risers with
actuator dynamics," Journal of Sound and Vibration, vol. 318, no. 4-5, pp.768-
791, 2008.
[46] Seiler, P, Alleyne A. "Dissipative adaptive control for strict feedback form sys-
tems," European Journal of Control, vol. 8, no. 5, pp. 435-444, 2002.
[47] Kokotovic PV, "The joy of feedback: Nonlinear and adaptive," IEEE Control
Systems Magazine, vol. 12, no. 3, pp. 7-17, 1992.
[49] Ph. Martin R.M. Murray P. Rouchon, "Flat Systems," The European Control
Conference (ECC) 97, 1997, pp. 211-218.
[51] B. Ayalew and B. T. Kulakowski, "Cascade tuning for nonlinear position con-
trol of an electro-hydraulic actuator", in Proceedings of the American Control
Conference, 2006, pp. 4627-4632.
124
Bibliography
system in the presence of friction and internal leakage," Int. J. Physical Sciences,
vol. 6, no. 14, pp. 3502-3517, 2011.
[55] K.-S. Kim, K.-H. Rew, and S. Kim, "Disturbance observer for estimating higher
order disturbances in time series expansion," IEEE Transactions on Automatic
Control, vol. 55, no. 8, pp. 1905-1911, Aug. 2010.
[56] B. Friedland and Y. J. Park, "On adaptive friction compensations," IEEE Trans-
actions on Automatic Control, vol. 37, no. 10, pp. 163-166, Oct. 1993.
[59] C. T. Chen, "Linear System Theory and Design, 3rd ed.," Oxford University Press,
New York, 1999.
[60] Yu, J., Chen, Z. and Lu, Y., "The variation of oil effective bulk modulus with
pressure in hydraulic systems," ASME J. Dynamic Systems, Measurement and
Control 116, pp. 146-150, 1994.
[61] W. Kim, D. Shin, D. Won, and C. C. Chung, "Disturbance observer based posi-
tion tracking and flatness controller in electro-hydraulic actuators," IEEE Trans-
actions on Control Systems Technology, vol. 21, no. 6, pp. 2290-2298, 2013.
[62] K. P. Tee, S. S. Ge, and, E. H. Tay, "Barrier Lyapunov functions for the control
of output-constrained nonlinear systems," Automatica, vol. 45, no. 4, pp. 918-927,
2009.
[64] B. V. E. How, S. S. Ge, and Y. S. Choo, "Control of coupled vessel, crane, cable,
and payload dynamics for subsea installation operations," IEEE Transactions on
Control Systems Technology, vol. 19, no. 1, pp. 208-220, 2011.
125
Bibliography
[65] Z. Li, C. Yang, N. Ding, S. Bogdan, and T. Ge, "Robust adaptive motion control
for underwater remotely operated vehicles with velocity constraints," Interna-
tional Journal of Control, Automation and Systems, vol. 10, no. 2, pp 421-429,
2012.
[67] H. Hammouri, G. Bornard, and K. Busawon, "High gain observer for structured
multi-output nonlinear systems," The Transactions on Automatic Control, vol. 55,
no. 4, pp. 987-992, 2010.
[68] F. Lafont, E. Busvelle, and J.-P. Gauthier, "An adaptive high-gain observer for
wastewater treatment systems," Journal of Process Control, vol. 21, no. 6, pp. 893-
900, 2011.
[69] S. Seshagiri, "Position control of permanent magnet stepper motors using condi-
tional servo-compensators," IET Control Theory and Applications, vol. 3, no. 9,
pp. 1196-1208, Apr. 2009.
[70] B.-Z. Guo and Z.-l. Zhao, "On the convergence of an extended state observer for
nonlinear systems with uncertainty," System & Control Letters, vol. 60, no. 6,
pp. 420-430, 2011.
[71] D. Shin, W. Kim and C. C. Chung, "Position control of a permanent magnet step-
per motor by MISO backstepping in semi-strict feedback form," in Proceedings of
the IEEE/ASME International Conference on Advanced Intelligent Mechatronics,
2011, pp. 808-813.
[72] V. Andrieu, L. Praly, and A. Astolfic, "High gain observers with updated gain and
homogeneous correction terms," Automatica, vol. 45, no. 2, pp. 422-428, 2009.
[73] R. L. Kosut, "Design of linear systems with saturating linear control and bounded
states," IEEE Transactions on Automatic Control, vol. 28, no. 1, pp. 121-124,
1983.
[75] Star Hydraulic Ltd., "Data sheet of M200 and M454 series servo valves".
126
Bibliography
[77] Shankar Sastry, "Nonlinear Systems - Analysis, Stability, and Control", Springer,
1999.
[78] Slotine, J.-J. E. and Li, W., "Applied Nonlinear Control", Prentice-Hall, 1991.
[80] Krstic, M., I. Kanellakopoulos, and P. Kokotovic, "Nonlinear and Adaptive Con-
trol Design", John Wiley & Sons, Inc., 1995.
[81] Jean Lévine, "Analysis and Control of Nonlinear Systems", John Wiley & Sons,
Inc., 1995.
127
연구 윤리 서약서
2015년06월01일
학위명 : 박사
학과 : 전기공학과
지도교수 : 정정주
년 월 일
성명 : 원대희 (서명)
한 양 대 학 교 대 학 원 장 귀 하
Declaration of Ethical Conduct in Research
Degree : Doctor 년 월 일