Nonlinear Control

Techniques for
Electro-Hydraulic
Actuators in Robotics
Engineering
 Nonlinear Control
Techniques for
Electro-Hydraulic
Actuators in Robotics
Engineering

Qing Guo and Dan Jiang

MATLAB is a trademark of The MathWorks, Inc. and is used with permission. The MathWorks
does not warrant the accuracy of the text or exercises in this book. This book’s use or discussion
of MATLAB software or related products does not constitute endorsement or sponsorship by The
MathWorks of a particular pedagogical approach or particular use of the MATLAB software.

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 2018 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-138-63422-0 (Hardback)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or here-
after invented, including photocopying, microfilming, and recording, or in any information storage or
retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.
copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC),
222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that
provides licenses and registration for a variety of users. For organizations that have been granted a
photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.

Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
 To my parents

and my daughter.

Qing Guo
Contents

Foreword. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xvii
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xix
Authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xxi
Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xxiii

1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1

2. Model Construction of Electro-Hydraulic Control System . . . . . . . . . . . . . .7

3. Linear PID Control Design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

4. Robust Control Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

5. Output Feedback Control Method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

6. Parametric Adaptive Control Method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

vii
Foreword

I am delighted to introduce the first book on electro-hydraulic system field.

When I undertook a project on the Lower Limb Exoskeleton in the year 2010,
I was very interested in the electro-hydraulic control system to drive this
complicated mechatronics plant. Hence I have tried to investigate many ref-
erences about the advanced control techniques used in the electro-hydraulic
system by many researchers. In these references, especially with the lack of
high-cited literature, it indicated that more and more engineers are required
to introduce new control ideas into mechanical control plants, not limited
to the traditional PI control method. Especially in certain terrible working
conditions, perhaps PI control has no favorable dynamic performance or
accuracy since its effective model is only two orders. If the control algorithm
is not changed, we must select a better sensor or actuator, and other elements
to replace the current hardware. This will lead to more cost and consumption
of more material. Thus, we need to update the control algorithm and look for
better control techniques to improve the performance of the electro-hydraulic
system. Based on this concept, now that I have my attention on it, it appears
to be a better idea to spread some advanced control techniques applied in the
electro-hydraulic control system, especially in typical fields such as robotics
engineering and fluid transmission engineering.
Electro-hydraulic servo actuator started gaining recognition in the 1960s
as a field. This actuator has many obvious advantages than the general
servo motor actuator, such as higher load-bearing and small size-to-power
ratio. Many mechatronics plants about multi-DOF manipulator, crane, space
manipulator, and lower limb of BigDog favor this actuator due to the exis-
tence of large dynamic external load beyond the capability of the servo
motor actuator. Furthermore, the electro-hydraulic servo actuator has high
control accuracy and fast response performance since the supply pressure
can be easily increased. The operation and reliability are also beneficial to
the automation control system. In addition, the energy saving of the electro-
hydraulic system can also be realized by some pump valve combined control
method.
For the researchers of automatic control field, the important purpose is to
apply several control methods into practice. However, different actual con-
trol plants have different characteristics. There is no generic control method
superiorly used in any plant. To address different control problems, the corre-
sponding control scheme should be designed and optimized. Although some

ix
x Foreword

control method does not have an ideal performance in special working con-
ditions, engineers cannot give up easily, that is, the issue is analyzed case by
case. Thus, in this book, many control methods handle many different con-
trol problems and conditions in electro-hydraulic system, which is only for
the reference of the reader.

Yan Shi
School of Automation Science and Electrical Engineering
Beihang University
Beijing, China
Preface

Electro-hydraulic servo systems (EHSs) are nowadays widely used in mecha-

tronic control engineering due to their higher load-bearing and small size-to-
power ratio. It was found that EHSs are beginning to be commonly applied
in some multi-DOF manipulator, crane, space manipulator, and lower limb
of BigDog. However, there exist several typical problems that are related
to EHS.
One of the fundamental difficulties in electro-hydraulic control is the unde-
sirable dynamic behavior of the designed controller due to the external load
disturbance that exists in EHS. This is caused by the driven force or torque of
the mechatronics plant. For different types of external load on the electro-
hydraulic actuator (EHA), it is often considered as zero [31] or unknown
constant [64], even a bounded uncertainty disturbance [7,60,62]. However,
Kim et al. [33] developed a high-gain disturbance observer (HGDO) with
backstepping control to compensate for the unknown external load while
guaranteeing the position tracking error within an acceptable level. Chen [9]
proposed a nonlinear disturbance observer integrated with a general non-
linear controller. In practice, the external load may be a largely unknown
structural disturbance of EHS, which should be compensated by the con-
structed controller. Even though the feedback control of EHS may be stable,
it is clear that the dynamic performance will be declined if the external load
increases beyond the maximum load capability of EHA.
The second problem in electro-hydraulic control system is degraded
dynamic and steady-state performance due to the hydraulic parametric
uncertainties existing in EHS. To handle this problem, several advanced
control methods have also been presented such as local model lineariza-
tions [5,38,49], robust H∞ controllers [20,42], quantitative feedback control
schemes [45], geometric control approach [55], output regulation control
[52], parametric adaptive controllers [1,18,23,24], state-constrained control
[26,27], robust controller with extended state observer [21,61], and distur-
bance observer [47]. These controllers usually adopted adaptive parametric
estimation law (APEL) to estimate the uncertainty parameter. It is to be noted
that the load disturbance and parametric uncertainty often lead to unex-
pected chatter, overshoot, and zero bias of tracking error. Thus, the designed
controller should be considered in EHS to guarantee not only the prescribed
accuracy of output tracking error but also the desirable dynamic responses
of system state.
In addition, as the common nonlinear control system with strict feed-
back form is usually handled by the well-known backstepping method [37],
many high-order derivatives of the virtual control variables are generated in
backstepping iteration, which easily result in the derivatives explosion [58],

xi
xii Preface

violent control, and saturation. To address this problem, the dynamic surface
control (DSC) was proposed to design a stable dynamic surface [54] instead
of a virtual control derivative. The advantage of DSC is to eliminate the
severe proliferation and singularity of the nonlinear system and guarantee
fast state convergence and desirable dynamic performance [50]. The dynamic
surface is often designed as a linear filter to transform high-order deriva-
tions of virtual control variables into a different stable dynamic surface. If
the DSC is not considered in the backstepping controller, Guo et al. [18] pro-
posed another computation method of virtual control variable to avoid the
derivatives explosion. This virtual control variable can be directly filtered by
a linear decayed memory filter to smooth the high-order derivatives.
In this book, based on the aforementioned contributions of many
researchers, the authors have tried their best to apply some typical linear and
nonlinear control techniques into EHAs, which drive a two-DOF robotic arm,
for example. Some control ideas and motion mechanism are partially similar
to many references. Both theoretical proof and simulation and experimental
results are given to verify the corresponding control method in detail. The
authors hope that these points of view can benefit the reader to study the
electro-hydraulic control system in depth.

MATLAB® is a registered trademark of The MathWorks, Inc. For product

information, please contact:
The MathWorks, Inc.
3 Apple Hill Drive
Natick, MA 01760-2098 USA Tel: 508 647 7000
Fax: 508-647-7001
E-mail: [email protected]
Web: www.mathworks.com
List of Figures

Figure 1.1 Some typical problems and the corresponding solutions

in the EHS model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2
Figure 2.1 Symmetrical and asymmetrical hydraulic cylinders. . . . . . . . . . . . .8
Figure 2.2 Linearized model of the electro-hydraulic actuator. . . . . . . . . . . . 14
Figure 3.1 Linear feedback control loop of the electro-hydraulic
system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Figure 3.2 Closed loop from the load disturbance FL to the
cylinder position Y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Figure 3.3 Change of pipe pressure loss ptube from valve to
cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Figure 3.4 Motion control mechanism of the two-DOF robotic arm. . . . . . 25
Figure 3.5 General framework for the mechanical movement. . . . . . . . . . . . 25
Figure 3.6 Two cylinder dynamic lengths and force arms. . . . . . . . . . . . . . . . . 30
Figure 3.7 Mechanical properties curve of the motor with load. . . . . . . . . . 31
Figure 3.8 Simulation model for solvering the dynamic pressure of
the cylinder with load. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Figure 3.9 Simulation result with the elbow joint θ 2 = 140°, the
sinusoidal input of the shoulder angle f u = 1.05 Hz. . . . . . . . . . . . 33
Figure 3.10 Simulation result with the shoulder joint θ 1 = − 70°, the
sinusoidal input of the elbow angle f f = 1.85 Hz. . . . . . . . . . . . . . . . 35
Figure 3.11 Simulation result with two sinusoidal inputs of the
shoulder angle f u = 0.6 Hz and the elbow angle f f = 1 Hz.. . . . . 37
Figure 3.12 Frequency domain characteristic of the open-loop
control system for the shoulder actuator for Equation 3.56. . . 38
Figure 3.13 Performance results of the PI controller design for the
shoulder actuator for Equation 3.56. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Figure 3.14 Frequency domain characteristic of the open-loop
control system for the shoulder actuator for Equation 3.57. . . 40
Figure 3.15 Performance results of the PI controller design for the
shoulder actuator for Equation 3.57. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

xiii
xiv List of Figures

Figure 3.16 Experimental equipment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

Figure 3.17 Position tracking results of the upper arm cylinder. . . . . . . . . . . . 43
Figure 3.18 Position tracking results of the forearm cylinder. . . . . . . . . . . . . . . 43
Figure 3.19 Dynamic control voltages of two servo valves.. . . . . . . . . . . . . . . . . 44
Figure 3.20 Two chamber pressures of two cylinders. . . . . . . . . . . . . . . . . . . . . . . . 44
Figure 4.1 Dynamic ranges of KFLu and KFLf in one motion duration. . . . 50
Figure 4.2 Robust model with parametric and structural
uncertainties. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure 4.3 Block diagram of the closed-loop system with robust
performance requirements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Figure 4.4 Frequency response of the open-loop system with
varying uncertainty parameters Kq , V t , b, and KFLu . . . . . . . . . . . . 55
Figure 4.5 Cross-linked feedback system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
 
Figure 4.6 Singular values of the inverse function μ−1 Wp and
 n
the nominal closed-loop system μ Gc . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Figure 4.7 Block diagram for the description of robust stability and
robust performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Figure 4.8 Maximum robust stability bound with varying
uncertainty parameters Kq , V t , b, and KFLu . . . . . . . . . . . . . . . . . . . . . . 60
Figure 4.9 Maximum robust performance bound with varying
uncertainty parameters Kq , V t , b, and KFLu . . . . . . . . . . . . . . . . . . . . . . 60
Figure 4.10 Frequency domain result of the designed robust
controller. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Figure 4.11 Cylinder position response in time domain by the
designed robust controller. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Figure 4.12 Designed controller result in time domain. . . . . . . . . . . . . . . . . . . . . . 63
Figure 4.13 Square response of two joint angles in simulation and
experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Figure 4.14 Control voltage of two servo valves in simulation and
experiment for square demand.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Figure 4.15 Sinusoidal response of two joint angles in simulation
and experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Figure 4.16 Control voltage of two servo valves in simulation and
experiment for sinusoidal demand. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
List of Figures xv

Figure 4.17 Square response of the experiment result by the two

control methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Figure 4.18 Sinusoidal response of the experiment result by the two
control methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Figure 4.19 Snapshots of the sinusoidal experiment process. . . . . . . . . . . . . . . 67
Figure 5.1 Step response experiment of the upper arm hydraulic
actuator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Figure 5.2 State estimations by the high-gain state observer in step
response experiment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
Figure 5.3 Sinusoidal response experiment of the upper arm
hydraulic actuator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Figure 5.4 State estimations by the high-gain state observer in the
sinusoidal response experiment.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Figure 5.5 Comparison result in condition (1). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Figure 5.6 Comparison result in condition (2). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Figure 5.7 The coordinated motion experiment results of the
robotic arm joints, sinusoidal demand input for the
shoulder actuator, and step demand input for the elbow
actuator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Figure 5.8 Experiment video of coordinated motion for the robotic
arm joints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Figure 6.1 Block diagram of the control system. . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Figure 6.2 Four state responses of the EHS by the proposed
controller. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Figure 6.3 Six estimation values by parametric adaptive estimation
laws. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Figure 6.4 Control voltages of two hydraulic actuators by the
proposed controller. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Figure 6.5 Comparison result for the demand input
xs1d = 14.5 sin(2πt) mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Figure 6.6 Comparison result for the demand input
xs1d = 29 sin(π t) mm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Figure 6.7 Simulation results of load disturbance estimation. . . . . . . . . . . . 118
Figure 6.8 Experimental results of the load disturbance estimation
on two EHAs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
xvi List of Figures

Figure 6.9 Simulation results of position tracking error by two

controllers, y1 —upper arm error, y2 —forearm
error. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Figure 6.10 Experimental results of position tracking error by two
controllers, y1 —upper arm error, y2 —forearm
error. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
List of Tables

Table 3.1 Servo Valve Parameters of Moog D633-R02K01M0NSM2 . . . . . 21

Table 3.2 Parameters of the Hydraulic Cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Table 3.3 Mechanical Parameters of the Two-DOF Robotic Arm . . . . . . . . . 26
Table 3.4 Parametric Analysis of the Shoulder Hydraulic Actuator . . . . . 34
Table 3.5 Parametric Analysis of Elbow Hydraulic Actuator . . . . . . . . . . . . . 36
Table 3.6 Parametric Analysis of Two Joint Angles Simultaneously
Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Table 3.7 Control Parameters and Performance Design with
Respect to Different Pressures of the Upper Arm
Cylinder with Different Loads in the Condition of
Extended State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Table 3.8 Control Parameters and Performance Design with
Respect to Different Pressures of the Upper Arm
Cylinder with Different Loads in the Condition of
Retracted State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Table 3.9 Hydraulic Parameters Used in Simulation and
Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Table 4.1 Hydraulic Parameters Used in Simulation and
Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
Table 6.1 Specific Parameters and Brand of Main Components . . . . . . . . . 110
Table 6.2 Hydraulic Parameters Used in Experiments . . . . . . . . . . . . . . . . . . . . 112

xvii
Acknowledgement

Qing Guo and Dan Jiang acknowledge Professor Andrew Plummer, director
of Power Center for Power Transmission and Motion Control, Department
of Mechanical Engineering, University of Bath for the funding assistance and
previous work. They thank Dr. Tian Yu and Dr. Can Du from Power Center
for Power Transmission and Motion Control for their collaborative study in
related electrohydraulic work.
During the writing of the book, the authors were assisted by many
researchers from electrohydraulic engineering regarding some innovative
methods and techniques. Although some theoretical methods have a long
way to go in the application of electrohydraulic engineering, scholars did
not give up using new control methods and techniques, which is much
appreciated by them.
The study was supported by UK Engineering and Physical Sciences
Research Council project (No. EP/H024190/1), the National Natural Science
Foundation of China (Nos. 61305092 and 51205045), the Postdoctoral Science
Foundation of China (No. 2013M542487), the Fundamental Research Funds
for the Central Universities, China (Nos. 20160250331 and ZYGX2016J160),
and the Open Foundation of the State Key Laboratory of Fluid Power &
Mechatronic Systems (No. GZKF-201515).

xix
Authors

Qing Guo earned his B.E. in automation from Harbin Institute of Technology
(Harbin, China) in 2003, and went on to earn his M.S. and Ph.D. there, in 2005
and 2008, respectively. In 2009, he became a lecturer at the School of Aero-
nautics and Astronautics, University of Electronic Science and Technology of
China (ChengDu, China), and in 2013, he was promoted to associate profes-
sor. From December 2013 to December 2014, he served as an academic visitor
at the Center for Power Transmission and Motion Control, Department of
Mechanical Engineering, University of Bath, United Kingdom. Dr. Guo’s
research interests include robust and adaptive control, mechatronic systems,
and rehabilitation robots.
Dan Jiang earned her B.E. in mechanical engineering (2002), and M.S. (2005)
and Ph.D. (2009) in fluid power transmission and control from the Harbin
Institute of Technology (Harbin, China). Since April 2009, she has been with
the School of Mechatronics Engineering, University of Electronic Science and
Technology of China (ChengDu, China) as a lecturer. Her research interests
include hydraulic control and microfluidic technology, and she has published
several research papers in these areas.

xxi
Symbols

Two-DOF Two-degree-of-freedom
EHS Electro-hydraulic system
EHA Electro-hydraulic actuator
Ksv Gain of the servo valve
Tsv Time constant of the servo valve
u Input control voltage of the servo valve
Cd Discharge coefficient
w Area gradient of the servo valve spool
pL Load pressure of the servo valve
ps Supply pressure
xv Spool position of the servo valve
ρ Density of the hydraulic oil
y Displacement of the cylinder
yd Demand displacement
Ctl Total leakage coefficient of the cylinder
βe Effective bulk modulus
Ap Annulus area of the symmetrical chamber
Aa Annulus area of the single-rod chamber
Ab Annulus area of the no-rod chamber
Vt Half-volume of the cylinder
m Load mass
K Load spring constant
b Viscous damping coefficient
FL External load of the hydraulic actuator
k Positive constant in the hyperbolic tangent function
sgn(.) The sign function
tanh(.) The hyperbolic tangent function
zi System state error
xi System state
ϑi Parametric uncertainty
ϑ̂i Parametric estimation
ϑ̃i Parametric estimation error
θi Joint angle of the robotic arm
d Equivalent disturbance
d̂ Disturbance estimation
d̃ Disturbance estimation error

xxiii
1
Introduction

CONTENTS
1.1 Parametric Uncertainty Problem of EHS . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Largely Unknown Load Disturbance of EHS . . . . . . . . . . . . . . . . . . . 4
1.3 Control Method Illustration in EHS . . . . . . . . . . . . . . . . . . . . . . . . . . 5

In mechatronic engineering, there exist three common actuators used in

automation system, that is, servo motor, electro-hydraulic, and pneumatic.
The servo motor system has high mechanical efficiency and control accuracy
than the other two actuators. However, its power is restricted by the element
size and structure of motor and driver. Pneumatic actuator is an environ-
mentally friendly actuator, which has no obvious pollution to environment.
But the pneumatic actuator has very low execution efficiency in engineering.
Many available energy dissipates as heat. Although the mechanical efficiency
of the electro-hydraulic actuator (EHA) is not higher than that of the servo
motor actuator, electro-hydraulic servo systems (EHSs) are nowadays more
widely used in mechatronic control engineering due to their higher load-
bearing and small size-to-power ratio [40] than the general servo motor
actuator. When the supply pressure is large, the dynamic response may also
be faster than the servo motor under some large external load and other
unknown disturbance. It was found that EHSs are beginning to be commonly
applied for large power equipment such as wheel loaders [13], load simula-
tors [63], insulator fatigue test devices [66], and exoskeletons [22]. EHA is
a favorable control executive part to realize some multi-DOF manipulator,
crane, space manipulator, and lower limb of BigDog due to the large dynamic
external load.

1
2 Nonlinear Control Techniques for EHAs in Robotics Engineering

In this book, four control methods are discussed in the electro-hydraulic

system such as linear classical control, robust control, nonlinear control,
and adaptive control methods. Each method has its own advantages to
address different control problems. The authors have investigated numer-
ous references about the hydraulic control of EHS. Generally, four typical
problems need to be addressed in EHS control (Figure 1.1). The first prob-
lem is parametric uncertainties such as effective bulk modulus β e , leakage
coefficient of the cylinder Ctl , and hydraulic oil density ρ. These param-
eters are often unknown constants in different working conditions. To
solve the unknown constant parameters, the parametric adaptive estima-
tion (PAE) law is adopted to estimate some parametric uncertainties. The
second problem is unmodeled uncertainties, such as nonlinear friction of
cylinder and viscous resistance of hydraulic oil [3,17]. These parameters
are dynamic variables, which are often adopted by extended state observer
(ESO) [61] or disturbance observer (DO) [59]. The third problem is output
feedback control, which assumes that some system states are not directly
measured from sensors. Thus, we need to construct the high-gain observer
[33–35] to estimate some unmeasured system states, which are used in the

x1 = x2
1
x2 = (–Kx1 – bx2 + Ap x3 – FL)
m
4βe Ap 4βe Ctl 4βe Cdw
x3 = x2 x3 + ps – tanh(kx4)x3x4 + Δf (t)
Vt Vt Vt ρ
Fo PAE—Parametric adaptive estimation
1 Ksvu δ(t) ur
x4 = – x + + EH prob ESO—Extended state observer
Tsv 4 Tsv Tsv S m lem HGOB—High-gain observer
od s in
el DOB—Disturbance observer
QFC—Quantitative feedback control
EHS control SPT—Singular perturbation theory

Problem 1: Problem 2:
Problem 3: Problem 4:
Parametric Unmodeled
uncertainty Unknown state External load
uncertainty

PAE ESO HGOB DOB

Geometric Output
Backstepping
PID, QFC,
Robust H∞ control regulation
SPT
approach control

FIGURE 1.1
Some typical problems and the corresponding solutions in the EHS model.
Introduction 3

controller design. The last problem is the external load on the EHA, which
is a largely unknown dynamic variable. To address the external load, a
simple method is adopted by the DO. In this book, we mainly consider
Problems 1 and 4, that is, parametric uncertainty and external load dis-
turbance. These two problems may degrade the dynamic behavior and
steady-state control accuracy of EHS and the robustness of the designed
controller.

1.1 Parametric Uncertainty Problem of EHS

Since some unknown parametric variation may be significant in differ-
ent working conditions (i.e., phenomenon such as oil temperature varia-
tions, pressure-flow characteristics, hysteresis in flow gain characteristics,
oil leakage, characteristics of valves near null) [42], many state/output
feedback controllers cannot be well established to guarantee the dynamic
performance of EHS. Thus, the parametric estimation is one available
method to obtain unknown parameters by state observer construction inte-
grated with other nonlinear controller. However, different from paramet-
ric uncertainties, the main disturbance is the largely unknown torque/-
force disturbance caused by external loadings on the hydraulic actuator.
So, further references have been more focused on disturbance rejection
of EHS. For instance, Reference 64 proposed a nonlinear controller in
which the external load is treated as an uncertain but bounded distur-
bance. It has been shown that the closed-loop stability can be directly
analyzed by Lyapunov technique. Yao and Bu [7,60] assumed that the
maximum relative uncertainty of the external load disturbance is bounded
by a known value and proposed a discontinuous projection-based adap-
tive backstepping controller. Kim et al. [33] presented a DO with propor-
tional integral (PI) control form to estimate a biased sinusoidal external
load. Then, Won et al. [59] developed a high-gain disturbance observer
(HGDO) with backstepping to compensate for the unknown external load
and guaranteed tolerance of the position tracking error. These references
denote that the external load with wide variations is an important fac-
tor to decline the dynamic response performance of the hydraulic con-
troller, especially in some critical condition where the external load of
the hydraulic actuator is close to the limitation. Therefore, to improve
the dynamic behavior of EHS, various advanced control approaches were
developed to estimate unknown parametric uncertainties and unmeasured
disturbance.
Recently, some robust H∞ control methods [19,20,28,29] and quantitative
feedback theories [45] have been presented to overcome parametric uncer-
tainties and to guarantee the robustness of the controller. A geometric control
4 Nonlinear Control Techniques for EHAs in Robotics Engineering

approach [55] was verified by numerical simulation to realize the dynamic

tracking position of a single-rod cylinder. In the past two decades, back-
stepping control was widely used in EHS [2,3,56]. If the model of EHS is
a strict feedback system [32,37], the backstepping controller based on state
feedback can be well implemented. Bu and Yao [8] presented a discontin-
uous projection-based adaptive backstepping controller to estimate some
unknown parameters of the asymmetric hydraulic actuator. Subsequently,
Guan and Pan [16,17] also constructed a parametric adaptive estimation
law to guarantee the asymptotic convergence of the backstepping con-
troller. Kaddissi et al. [31] proposed an equivalent parameter identification
method by the least squares and obtained better performance than the lin-
ear controller. Ahn et al. [1] presented an adaptive position control for a
pump-controlled EHA based on an adaptive backstepping control scheme.
To address some unmeasured physical states, Sun and Chiu [53] proposed a
perturbation observer to estimate the load pressure of a single-rod hydraulic
actuator. Pi and Wang [48] designed an observer-based cascade controller
to estimate the disturbance force in the hydraulic manipulator. In addition,
some output observers [34,35,44,61] were used to estimate hydraulic states
with less measured information than state feedback observers. These various
observers were verified in backstepping or other nonlinear control method.

1.2 Largely Unknown Load Disturbance of EHS

The other difficulty in EHS is the undesirable dynamic behavior of the
established controller due to the ignorance of the largely unknown distur-
bance caused by the dynamic external load (i.e., torque/force). As far as
the authors know, the external load was not well addressed in aforemen-
tioned studies. If the external load disturbance is treated as zero [56] or a
known constant [1], even bounded by a known value [64], many novel con-
trollers are designed conveniently by Lyapunov technique. Thus, Yao and Bu
[7,60] assumed that the maximum relative uncertainty of the external load
disturbance was bounded by a known value and proposed a discontinuous
projection-based adaptive backstepping controller. Yao et al. [62] presented
an adaptive robust controller to handle the nonlinear parametric uncertainty
in an auxiliary function. In this approach, the first and second derivatives
of the external load disturbance were bounded by two constants directly in
order to conveniently obtain the negative definite Lyapunov function. Then,
he also assumed that the modeling uncertainty and the external load dis-
turbance became zero after a finite time in Reference 61. Kaddissi et al. [31]
proposed an equivalent parameter identification method by the least squares
and obtained better performance than the linear controller. In this method,
Introduction 5

the load torque disturbance was still assumed as zero. However, in engineer-
ing practice, the external load is often the largely unknown disturbance of
EHS. Thus, the designed controller would not only eliminate system state
error but also suppress the unknown disturbance of the external load. Even
though the EHS may be stable, it is clear that the dynamic performance will
be declined if the external load increases beyond a definite boundary [25].
Subsequently, Kim et al. [36] proposed a flatness-based nonlinear con-
troller to improve the position tracking performance while assuming the
known constant external load. Then his research team developed a HGDO
with backstepping control to compensate for the unknown external load
and guaranteed the position tracking accuracy. In this novel approach, the
DO had two different forms. One was a second-order high-pass filter [33]
to estimate a sinusoidal disturbance with unknown frequency. The other
was a HGDO [59] to estimate the largely unknown disturbance caused by
the friction, the load force, and the parameter uncertainties. Both the sim-
ulation and the experimental results indicated that the external load can
be well compensated, as well as the extended system state error was ulti-
mate boundedness. Chen proposed a nonlinear DO integrated with general
nonlinear controller [9]. In practice, the external load may be the largely
unknown structural disturbance of EHS, which should be compensated by
the constructed controller.

1.3 Control Method Illustration in EHS

Recently, there have been several control methods applied in electro-
hydraulic control system. The classical approach to the control of EHSs is
proportional integral derivative (PID) control, which is easy to be used in
industry. Variable control parameters of PID controllers [6,43] are adopted
to suit the variable characteristics of the dynamic model. The advantage of
the linear PID controller is the simple structure and the feedback variable
is only the output information, not all the state or other indirect computed
variables. However, the linear controller cannot effectively compensate the
external load and has an obvious bias error that exists in tracking response
due to the parametric uncertainty. Thus, some advanced controllers are used
in EHS to improve the dynamic performance under the uncertain hydraulic
parameters and load disturbances. The robust controller can improve the
robustness of the feedback control system according to small gain theorem.
But the controller is relatively conservative. The parametric adaptive control
method often addresses the parametric uncertainty and has an obvious effect
of the parametric estimation. However, some unmodeled uncertainties are
not well addressed due to the unknown boundness of the hydraulic parame-
ter. Furthermore, the nonlinear model of EHS is assumed to be strict feedback
6 Nonlinear Control Techniques for EHAs in Robotics Engineering

form. Otherwise, the backstepping control method cannot be used to derive

the virtual control variable.
The feedback control method is often of two forms, that is, the state feed-
back and the output feedback. The former needs adequate sensors to measure
the state of EHS. In practice engineering, due to the cost constraints and
measurement reliability of engineering, many hydraulic states may not be
easy to obtain, such as pressures in different cylinder chambers, spool posi-
tion, and its velocity. Therefore, the output feedback control method should
be investigated to achieve a similar performance to the state feedback con-
trol. Some output state observers are used in EHS to estimate hydraulic
states, which can be used in backstepping or other nonlinear control method
[34,35,44]. These state observers are designed as proportional or PI struc-
ture. If the observer is convergence, the dynamic behavior can be obviously
improved. Apart from the above-mentioned observer, Sun presented a per-
turbation observer to estimate the different chamber pressure of the single-
rod hydraulic actuator [53]. Pi and Wang constructed a DO to estimate and
compensate the unknown disturbance in the hydraulic manipulator [48].
2
Model Construction of Electro-Hydraulic
Control System

CONTENTS
2.1 Hydraulic Cylinder Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7
2.1.1 Symmetrical Cylinder Model . . . . . . . . . . . . . . . . . . . . . . . . . .8
2.1.2 Asymmetrical Cylinder Model . . . . . . . . . . . . . . . . . . . . . . . .9
2.2 Servo Valve Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9
2.2.1 Load Flow Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 Spool Position Response Model . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Parametric Uncertainty and Load Disturbance . . . . . . . . . . . . . . . . . 11
2.4 Nonlinear State-Space Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Linearized Model of Electro-Hydraulic Actuator . . . . . . . . . . . . . . . 14

The general electro-hydraulic system model often includes four elements

such as hydraulic cylinder, servo valve, mechatronics plant, and hydraulic
parametric uncertainty. In terms of the model construction of the electro-
hydraulic actuator, the mechatronics plant can be omitted and is usually
replaced by the so-called external load disturbance. In this chapter, these
model elements are illustrated in detail.

2.1 Hydraulic Cylinder Model

In EHS, the hydraulic cylinder is one type of motion actuator to realize the
linear stretch of the plant. It often has two different styles, that is, symmetri-
cal hydraulic cylinder and asymmetrical hydraulic cylinder. The former is a
double-rod acting mechanism in both sides of the cylinder and the latter is
a single-rod acting mechanism in only one side of the cylinder as shown in
Figure 2.1.

7
8 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) Double-rod cylinder

Vt/2 Vt/2
Load disturbance
Ap Ap FL

Qa pa pb
Qb

Servo valve

Relief valve

ps pr
Pump

Tank

(b) Single-rod cylinder

Vb
Va Load disturbance
Aa Ab FL

pa pb
Qa Qb

Servo valve

Relief valve

ps pr

Pump

Tank

FIGURE 2.1
Symmetrical and asymmetrical hydraulic cylinders: (a) the double-rod acting mechanism of the
symmetrical hydraulic cylinder and (b) the single-rod acting mechanism of the asymmetrical
hydraulic cylinder.

2.1.1 Symmetrical Cylinder Model

By referring to the symmetrical cylinder model mentioned in References 40
and 41, the flow-pressure continuous equation of the hydraulic cylinder is

Vt
QL = Ap ẏ + Ctl pL + ṗL , (2.1)
4βe
Model Construction of Electro-Hydraulic Control System 9

where pL = pa − pb is the load pressure, QL is the load flow, y is the displace-

ment of the cylinder, Ctl is the coefficient of the total leakage of the cylinder, β e
is the effective bulk modulus, Ap is the annulus area of the cylinder chamber,
and V t is the half-volume of the cylinder.
When the cylinder starts to move, the spring force and the viscous friction
of hydraulic oil are two typical resistances of the electro-hydraulic system.
Especially in load condition, the hydraulic pressure has to overcome the
external load force to drive the plant motion. Thus, if the major viscous
friction of hydraulic oil is simplified as coulomb friction, the mechanical
dynamic equation driven by the hydraulic actuator can be constructed as
follows:
mÿ = pL Ap − Ky − bẏ − FL , (2.2)

where m is the load mass, K is the load spring constant, b is the viscous
damping coefficient, and FL is the external load on the hydraulic actuator.

2.1.2 Asymmetrical Cylinder Model

The flow-pressure continuous model of the hydraulic cylinder is given by

Aa ẏ + Ctl (pa − pb ) + (V0a + Aa y)ṗa /βe = Qa
, (2.3)
Ab ẏ + Ctl (pa − pb ) − (V0b − Ab y)ṗb /βe = Qb

where Qa and Qb are load flows with the spool position of the servo valve
xv ≥ 0 and xv < 0, respectively, pa and pb are the pressures inside the two
chambers of the cylinder, Aa and Ab are the ram areas of the two chambers,
and V 0a = V t /2 and V 0b = V t /2 are the initial total control volumes of the
two cylinder chambers, respectively.
Different from the symmetrical cylinder, the load pressure of the asym-
metrical cylinder pL = (pa Aa − pb Ab )/Aa , the mechanical dynamic equation
is described as follows:

mÿ = pa Aa − pb Ab − Ky − bẏ − FL (t). (2.4)

2.2 Servo Valve Model

The servo valve model includes the load flow model and the spool posi-
tion response model. In terms of the zero-opened four-way spool valve,
the load flow of the valve can be described differently for symmetrical and
asymmetrical cylinders, respectively.
10 Nonlinear Control Techniques for EHAs in Robotics Engineering

2.2.1 Load Flow Model

For the symmetrical cylinder, the load flow of the valve is simplified as
follow:

1
QL = Cd wxv (ps − sgn(xv )pL ), (2.5)
ρ

where xv is the spool position of the servo valve, ps is the supply pressure
of the pump, Cd is the discharge coefficient, w is the area gradient of the
servo valve spool, and ρ is the density of the hydraulic oil, and sgn(.) is the
sign function, that is, sgn(x) = 1, if x > 0, else if x < 0, sgn(x) = − 1, else
sgn(x) = 0.
For the asymmetrical cylinder, the load flow of the valve cannot be
described as QL , and the flow in the two directions Qa and Qb is given by
 
Cd wxv 2(ps − pa )/ρ xv ≥ 0
Qa =  ,
C wx 2(p − pr )/ρ xv < 0
 d v a (2.6)
Cd wxv 2(pb − pr )/ρ xv ≥ 0
Qb =  ,
Cd wxv 2(ps − pb )/ρ xv < 0

where Qa is the main load flow as xv ≥ 0, Qb is the main load flow xv < 0,
and pr is the return pressure of the tank.

2.2.2 Spool Position Response Model

The spool position response model describes the relationship between the
spool position and the input control voltage. Without loss of generality, this
model can be constructed as three forms, that is, a simplified model, first-
order linear dynamic model, and second-order dynamic model.
For the simplified model, since the cut-off frequency of the servo valve
is far greater than the control system bandwidth, the valve dynamics can
be neglected in model construction [33]. Thus, the spool position response
model is given by
xv = Ksv u, (2.7)

where Ksv is the gain of the servo valve and u is the control voltage of the
servo valve.
For the one-order linear dynamic model, it is considered one time constant
in the spool position response to denote the command delay of the servo
valve, which is constructed as follows:

Tsv ẋv + xv = Ksv u, (2.8)

where Tsv is the response time constant of the servo valve.

Model Construction of Electro-Hydraulic Control System 11

If both the delay and damping characteristics are considered in the spool
position response model of the servo valve, then the second-order dynamic
model is given by

2 2
ẍv + 2ζsv ωsv ẋv + ωsv xv = Ksv ωsv u, (2.9)

where ζ sv is the damping ratio and ωsv is the natural frequency.

2.3 Parametric Uncertainty and Load Disturbance

Electro-hydraulic systems often have some model uncertainties such as para-
metric uncertainty and external load disturbance. The former is mostly due
to unknown viscous damping, load stiffness, variations in control fluid
volumes, physical characteristics of the valve, bulk modulus, and oil temper-
ature variations [13], and the latter is caused by the driven force or torque
of the mechatronic plant. However, in practice, all the parametric uncer-
tainty and load disturbance are unknown. Without loss of generality, some
hydraulic parametric uncertainties are considered to be bounded by several
unknown or known constants as follows:

Cd = C̄d + Cd , w = w̄ + w, βe = β̄e + βe ,

(2.10)
Ctl = C̄tl + Ctl, ρ = ρ̄ + ρ, Vt = V̄t + Vt ,

where C̄d , w̄, β̄e , C̄tl , ρ̄, and V̄t are known normal hydraulic parameters, and
their uncertainties are constrained as follows:

Cd min ≤ Cd ≤ Cd max , wmin ≤ w ≤ wmax ,

βe min ≤ βe ≤ βe max , Ctl min ≤ Ctl ≤ Ctl max , (2.11)
ρmin ≤ ρ ≤ ρmax , Vt min ≤ Vt ≤ Vmax .

As shown in Equation 2.11, Cd min , Cd max , wmin , wmax , βe min , βe max ,
Ctl min , Ctl max , ρmin , ρmax , Vt min , and Vt max are unknown or known
boundaries of parametric uncertainties.

Remark 2.1
The above parametric uncertainties should be estimated in controller design.
Otherwise, the dynamic performance of the electro-hydraulic system will be
declined and the robustness of the designed controller will also be degraded.

The external load FL is considered to be a structural disturbance of the

electro-hydraulic system. Since FL is caused by the driven force or torque of
the mechatronic plant, it is reasonable to assume that the dynamic value of FL
12 Nonlinear Control Techniques for EHAs in Robotics Engineering

depends on the motion position, velocity, and acceleration of the hydraulic

cylinder, that is, y, ẏ, ÿ. Without loss of generality, FL can be described as the
following two forms.
The first form of FL (t) is bounded by
     
|FL (t)| ≤ c0 y + c1 ẏ + c2 ÿ , (2.12)

where c0 , c1 , c2 are known or unknown weighting boundaries.

Remark 2.2
By Equation 2.12, the boundary discussion of FL (t) can be transformed into
the direct state dynamics of y. Thus, the stability analysis of the electro-
hydraulic system model and the control design become easy.

On the other hand, the external load FL is constrained by

|FL (t)| ≤ FL , (2.13)

where FL is the known and unknown boundary constant.

Remark 2.3
The advantage of Equation 2.13 is very clear to denote the constraint of load
disturbance. However, if the boundary FL is unknown, the DO should be
used to estimate FL rather than to estimate the dynamic unknown vari-
able FL (t). Even though FL is a known value, the controller is relatively
conservative as FL is discussed as a direct uncertainty constraint in control
design.

2.4 Nonlinear State-Space Model

According to Equations 2.1, 2.2, and 2.7, the simplest electro-hydraulic
actuator model without the spool position response model is given by
⎧
⎪ ẋ1 = x2
⎪
⎪
⎪
⎪
⎨ 1
ẋ2 = (−Kx1 − bx2 + Ap x3 − FL (t))
m , (2.14)
⎪
⎪
⎪
⎪ 4β A 4β C 4β C wK 
⎪
⎩ ẋ3 = −
e p
x2 −
e tl
x3 +
e d
√
sv
ps − sgn(u)x3 u
Vt Vt Vt ρ

where [x1 , x2 , x3 ]T = [y, ẏ, pL ]T are three state variables of the electro-
hydraulic actuator and u is the control variable of the servo valve.
Model Construction of Electro-Hydraulic Control System 13

If the one-order spool position response model is considered, then the

electro-hydraulic actuator model is constructed as follows:
⎧
⎪
⎪ ẋ1 = x2
⎪
⎪
⎪
⎪ 1
⎪
⎪ ẋ2 = (−Kx1 − bx2 + Ap x3 − FL (t))
⎪
⎪
⎨ m
4βe Ap 4βe Ctl 4βe Cd w  , (2.15)
⎪
⎪ ẋ = − x − x + √ ps − sgn(x4 )x3 x4
⎪
⎪
3
V
2
V
3
V ρ
⎪
⎪
t t t
⎪
⎪
⎪
⎪ 1 K
⎩ ẋ4 = − x4 +
sv
u
Tsv Tsv

where [x1 , x2 , x3 , x4 ]T = [y, ẏ, pL , xv ]T are four state variables of the electro-
hydraulic actuator.
If the asymmetrical cylinder model is considered with the second-order
spool position response model, then, according to Equations 2.3, 2.4, 2.6,
and 2.9, the electro-hydraulic actuator model is given by
⎧
⎪ ẋ1 = x2
⎪
⎪
⎪
⎪
⎪
⎪ ẋ2 = (x3 Aa − x4 Ab − Kx1 − bx2 − FL )/m
⎪
⎪
⎪
⎪
⎪ ẋ3 = h1 (−Aa x2 − Ctl (x3 − x4 ))
⎪
⎪
⎪ 
⎪
⎪  
⎨ + h1 Cd wx5 2/ρ s1 ps − x3 + s2 x3 − pr
, (2.16)
⎪
⎪ ẋ4 = h2 (Ab x2 + Ctl (x3 − x4 ))
⎪
⎪
⎪
⎪   
⎪
⎪ − h2 Cd wx5 2/ρ s1 x4 − pr − s2 ps − x4
⎪
⎪
⎪
⎪
⎪
⎪ ẋ5 = x6
⎪
⎪
⎪
⎩ 2 2
ẋ6 = −2ζsv ωsv x6 − ωsv x5 + Ksv ωsv u

where [x1 , x2 , x3 , x4 , x5 , x6 ]T = [y, ẏ, pa , pb , xv , ẋv ]T , and

βe βe
h1 = , h2 = ,
V0a + Aa x1 V0b − Ab x1
(2.17)
1 + sgn(x5 ) 1 − sgn(x5 )
s1 = , s2 = .
2 2
Remark 2.4
The above three models (2.14), (2.15), and (2.16) are all used in practical con-
trol design. The difference is the model accuracy. In conventional condition,
the model (2.14) is enough to describe the dynamics of the electro-hydraulic
system. However, in largely unknown load disturbance and parametric
uncertainty, the models (2.15) and (2.16) are necessary to be used in advanced
control method.
14 Nonlinear Control Techniques for EHAs in Robotics Engineering

2.5 Linearized Model of Electro-Hydraulic Actuator

The linearized model of the electro-hydraulic actuator is necessary to be used
in the linear controller design such as PID, which is very widely adopted in
engineering (Figure 2.2). For the symmetrical cylinder, the linearized model
is derived from Equations 2.1, 2.2, 2.5, and 2.8.
At first, the load flow QL in Equation 2.5 is linearized as follows:

QL = Kq xv + Kc pL , (2.18)

where Kq and Kc are linearized coefficients from the derivatives of Equa-

tion 2.5 as follows:

∂QL ps − pL
Kq = = Cd w ,
∂xv ρ
(2.19)
∂QL Cd wxv
Kc = =  .
∂pL 2 ρ(ps − pL )

Second, from Equations 2.1, 2.2, 2.19, and 2.8, we can describe the linearized
model of the electro-hydraulic actuator as follows:
Thus, from Figure 2.2, the linearized model of the double-rod electro-
hydraulic actuator is given by

Kq Ksv 1 Vt
Ap (Tsv s+1) u(s) − A2p Kce + 4β e
s FL (s)
y(s) =     , (2.20)
Vt m 3 mt Kce Be Vt bKce Vt K KKce
s + + s + 1+ 2 +
2 s+
4βe A2p A2p 2
4βe Ap Ap 2
4βe Ap A2p

where y(s), u(s), and FL (s) are frequency domain transformations, and
Kce = Kc + Ctl .
In terms of the single-rod electro-hydraulic actuator, the load pres-
sure pL is defined as pL = pa − υpb , and the load flow QL is defined as
QL = (Qa + υQb )/(1 + υ 2 ), where υ is the annulus area ratio between the

FL(t)
Ksv xv QL 1 pL – y
u 1
Kq Vt Ap
Tsvs + 1 Kc + Ctl + s ms2 + bs + K
– 4βe

Aps

FIGURE 2.2
Linearized model of the electro-hydraulic actuator.
Model Construction of Electro-Hydraulic Control System 15

cylinder with rod and the cylinder without rod. Thus, the load flow QL is
linearized as follows:

QL = Kqi xv − Kci pL , i = a, b. (2.21)

From Equation 2.6, we can obtain that

⎧ 
⎪
⎪ 2(ps − pL )
⎪
⎪ C wx , xv ≥ 0
⎪
⎨ d v ρ(1 + υ 3 )
QL =  . (2.22)
⎪
⎪
⎪
⎪ 2(np s + pL )
⎪ Cd wxv
⎩ , xv < 0
ρ(1 + υ 3 )

The linearized coefficients in Equation 2.21 can be derived from Equa-

tion 2.22 as follows:


2(ps − pL ) Cd wxv
Kqa = Cd w , Kca =  ,
ρ(1 + υ 3 ) 2ρ(ps − pL )(1 + υ 3 )
 (2.23)
2(αps + pL ) Cd wxv
Kqb = Cd w , Kcb = −  .
ρ(1 + υ 3 ) 2ρ(υps + pL )(1 + υ 3 )

Then, like Equation 2.20, the linearized model of the single-rod electro-
hydraulic actuator is also obtained as follows:


Tsv s+1 Vt
Kqi Ksv u(s) − Aa Kcei + 2(1+υ 2 )βe
s FL (s)
y(s) = , (2.24)
Den

where i = a, b, Kcei = Kci + Ctl , and

  
Vt m 3 mKcei bVt
Den =Aa (Tsv s + 1) s + + s2
2(1 + υ 2 )βe A2a A2a 2(1 + υ 2 )βe A2a
  
bKcei Vt K KKcei
+ 1+ 2 + s + .
Aa 2(1 + υ 2 )βe A2a A2a
(2.25)
16 Nonlinear Control Techniques for EHAs in Robotics Engineering

Remark 2.5
The two linearized models (2.20) and (2.24) can be used in classical control
design such as PID, robust H∞ , where a linear controller is obtained. This lin-
ear controller is relatively simple to guarantee the stability of the closed-loop
system and basic dynamic behavior. However, there exist unknown paramet-
ric uncertainty and load disturbance in the electro-hydraulic system, which
lead to the existing model error in linearized models. Thus, the two nonlinear
models (2.15) and (2.16) are adopted in the nonlinear controller design. Many
nonlinear controllers such as backstepping, adaptive, slide mode, etc. should
guarantee that the transient and steady behavior of the electro-hydraulic
system achieve the prescribe performance. Furthermore, the designed con-
troller must guarantee the global convergence or ultimate boundedness of
the generalized system state, including hydraulic state variable, parametric
estimation error, and DO error.
3
Linear PID Control Design

CONTENTS
3.1 Linear Feedback Control Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Mechatronics Plant Model Description . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Servo Valve Model Construction . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Hydraulic Cylinder Model . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.3 Mechanical Motion Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 System Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.1 Motor Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.2 Output Pressure Analysis of Pump . . . . . . . . . . . . . . . . . . . . 31
3.3.3 Dynamic Pressure of Cylinder with Load . . . . . . . . . . . . . . . 32
3.4 PID Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.5 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

The linear PID control is a classical control method, which is very easy to be
understood. The PID controller is designed in frequency domain based on the
linear model of the electro-hydraulic system. The main evaluation indexes
for the linear PID control design are relative stability margin and bandwidth.
This design idea is derived from the frequency design method in classical
control principle. In this chapter, the linear PID control design method is
introduced and the system stability and dynamic behavior are discussed.

3.1 Linear Feedback Control Loop

According to Figure 2.2, the whole linear feedback control loop of the electro-
hydraulic system is shown in Figure 3.1. This is a typical cylinder position
feedback control loop. The actual cylinder position Y is controlled by the PID
controller to track the position demand Yexp . The servo valve and the cylinder
model can be considered as the plant model, which is similar to the linearized
model of the double-rod electro-hydraulic actuator (2.20). Here, the dynamics

17
18 Nonlinear Control Techniques for EHAs in Robotics Engineering

1 V FL
K + t s
Ap2 ce 4βe

Cylinder dynamics
Yexp Servo valve –
Control law Xv Kq/Ap Y
Gc(s) K Gcyd(s) =
– Gsv(s) = sv Vtmt 3 mtKce BeVt 2 B K KK KK
Tsv s+1 s + + s + 1 + e 2 ce + 2ce s + 2ce
4βe Ap2 Ap2 4βe Ap2 Ap Ap Ap
Y

Sensor model
Gsen(s)

FIGURE 3.1
Linear feedback control loop of the electro-hydraulic system.

of the position sensor model is not temporarily considered, that is, it is a unit
feedback, Gsen (s) = 1.
Consider the position demand Y exp as the system input, and the actual
cylinder position Y as the output. The open-loop transfer function of the
position feedback control loop is given by

Y(s)
Gol (s) = = Gc (s)Gsv (s)Gcyd (s)Gsen (s)
u(s)
Kq Ksv
Ap (Tsv s+1)
(3.1)
=     .
Vt m 3 mt Kce Be Vt bKce Vt K KKce
s + + s2 + 1 + + s+
4βe A2p A2p 4βe A2p A2p 4βe A2p A2p

Remark 3.1
When the position feedback control loop is mainly considered, the external
load FL need not be substituted into Equation 3.1, which is different from
Equation 2.20. In other words, FL is compensated by the designed control u.

Now the effect of the external load FL to the closed loop is analyzed. Con-
sider FL as the load disturbance input, and the actual cylinder position Y as
the output. The closed loop from FL to Y is shown as Figure 3.2.
The closed-loop transfer function from FL to Y is given by

Y(s)
GyFL (s) =
FL (s)
− 1
(Kce + Vt (3.2)
A2p 4βe s)Gcyd (s)
= .
1 + Gc (s)Gsv (s)Gcyd (s)Gsen (s)
Linear PID Control Design 19

1 Vt FL
Kce + s
Ap2 4βe

Cylinder dynamics
Servo valve Xv –
Control law Kq/Ap Y
Gc(s) Ksv Gcyd(s) =
– Gsv(s) = Vtmt mt Kce BeVt BeKce KKce KKce
Tsv s+1 s3 + + s2 +1+ + s+
Y 4βe Ap2 Ap2 4βe Ap2 Ap2 Ap2 Ap2

Sensor model
Gsen(s)

FIGURE 3.2
Closed loop from the load disturbance FL to the cylinder position Y.

If the static gain of the controller Gc (s) is obviously larger than 1, then
Equation 3.2 can be simplified as follows:

1 1
− K − K
A2p ce A2p ce
GyFL (s) = = , s → 0 ⇔ t → ∞, (3.3)
Gc (s)Gsv (s) kp Ksv

where kp is the static gain of Gc (s), that is, kp = Gc (0).

From Equation 3.3, owing to the constants Kce and Ksv , the negative effect of
FL to the system dynamic tracking performance is only declined by increas-
ing the static gain of the controller Gc (s). This denotes that a reasonable
controller u can suppress the external disturbance FL to a prescribed level.

Theorem 3.1
Consider the two transfer functions (3.1) and (3.2). There exists a linear
PID controller Gc (s) with sufficiently large static gain such that the closed-
loop feedback control system as shown in Figure 3.1 has enough global
stability margin and the external load disturbance FL (t) is suppressed into
a sufficiently small finite boundary.

Proof. The position feedback control system is a linear system as shown in

Figure 3.1. According to the global stability property of a linear control sys-
tem, if there exists an isolated linear controller to guarantee the stability of
the closed-loop feedback system, then the feedback control system is globally
stable.
To address the open-loop transfer function of the position feedback control
loop Gol (s), a suitable PID controller is designed as follows:

ki
Gc (s) = kp + + kd s, (3.4)
s
20 Nonlinear Control Techniques for EHAs in Robotics Engineering

where kp , ki , and kd are the proportional, integral, and derivative control

parameters, respectively.
If the three control parameters have no constraint, the closed-loop poles
of Gol (s) can be arbitrarily configured to any position in the negative half
plane. Thus, this PID controller is designed to guarantee the stability of the
closed-loop feedback control system as shown in Figure 3.1. Especially, some
reasonable PID parameters can guarantee that the open-loop transfer func-
tion Gol (s) has predefined amplitude margin, phase margin, and expected
bandwidth as follows:
⎧ exp
⎪
⎪ Am (Gol (s)) > Am dB
⎨
exp
Pm (Gol (s)) > Pm deg , (3.5)
⎪
⎪
⎩ exp
ωc > ωc
exp exp exp
where Am , Pm , and ωc are the predefined performances of the relative
stability margin.
On the other hand, from Equation 3.3, the sufficiently large static gain Kp of
Gc (s) can simultaneously reduce the disturbance effect caused by the external
load FL (t) as follows:
  Kce Kce
y(∞) = FL (∞) ≤ Fmax
L , (3.6)
A2p kp Ksv A2p kp Ksv

where Fmax
L is a finite boundary of FL (t). 

However, in practice, the control variable u is constrained by the control

saturation, that is, u ≤ umax . Thus, the PID control law Gc (s) should be appro-
priately designed to compromise the dynamic stable performance (3.5) and
the disturbance suppression (3.6).

3.2 Mechatronics Plant Model Description

In this section, a two-DOF robotic arm is considered to be a typical mecha-
tronics plant, which is driven by two electro-hydraulic actuators. The linear
PID control of the electro-hydraulic system is designed to illustrate the
motion control of the robotic arm.

3.2.1 Servo Valve Model Construction

Servo valve is the control element. It receives a voltage signal given by the
controller and be converted to the spool position to control the oil flow in
Linear PID Control Design 21

TABLE 3.1
Servo Valve Parameters of Moog D633-R02K01M0NSM2
No. Name Symbol Value Unit

1 Response time Tsv 12 ms

2 Rated flow qn 5 L/min
3 Rated pressure pn 70 bar
4 Maximal pressure pm 350 bar
5 Output current range Ictrl 4–20 mA
6 Control voltage range Uctrl −10 to 10 V
7 Spool diameter Dv max 7.9 mm
8 Servo valve pressure loss pvloss 1.4 bar
9 Pipe pressure loss ptube 7.5 bar

the hydraulic cylinder. From Equation 2.8, the math model is simplified as
follows:
xv Ksv
Gsv (s) = = , (3.7)
u Tsv s + 1

where xv is the spool position of the servo valve, u is the control voltage of
the servo valve, Ksv is the gain of the servo valve, and Tsv is the response time
constant of the servo valve.
According to the production of Moog D633-R02K01M0NSM2, some perfor-
mance parameters are shown in Table 3.1.
The maximal spool displacement is given by

xvm = Dvm /2 = 3.95 mm. (3.8)

The gain of the servo valve Ksv is computed by

xvm 3.95 × 10−3

Ksv = = = 3.95 × 10−4 m/V. (3.9)
Uctrl 10

Thus, the first-order transfer function model of this servo valve is con-
structed as
xv Ksv 3.95 × 10−4
= = . (3.10)
u Tsv s + 1 0.012s + 1

Owing to the servo valve hysteresis characteristics, the servo valve pres-
sure loss is computed by

pvloss = ps max × ς × 2 = 350 × 0.2% × 2 = 1.4 bar, (3.11)

where ps max is the maximum pressure and ς is the parameter hysteresis

characteristic.
22 Nonlinear Control Techniques for EHAs in Robotics Engineering

Seelen-Innenwand Smooth inner core

Druckmittel Hydraulikol nach DIN 51.524 Hydraulic fluid Hydraulic fluid to DIN 51.524
und DIN 51.525 and DIN 51.525
Schlauch Anschlüsse

ΔPAnschlüsse/connections
hose connections
ΔPSchlauch/hose(bar)

12 6 AnschluBtyp M

(bar)
Durchflu Bwiderstand
10 5
DurchfluB widerstand

Connection type M
ν = 55 mm2/s ν = 55 mm2/s
flow resistance for

AnschluBtyp 1
for 1 m schlauch

flow resistance
8 4
Connection type 1
1 m hose

6 3
AnschluBtyp 6
4 2 Connection type 6
2 1
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
DurchfluB Q (L/min) DurchfluB Q (L/min)
Flow rate Q (L/min) Flow rate Q (L/min)
Der gesamt-widerstand des schlauches errechnet sich nach folgender formel:
The total hose resistance is calculated by means of the formula:
ΔPsec = ΔPAnschlüsse 1 (bar) + ΔPSchlauch (bar) x L (m) + ΔPAnschlüsse 2 (bar)
ΔPtotal = ΔPConnection 1 (bar) + ΔPHose (bar) x L (m) + ΔPConnection 2 (bar)

FIGURE 3.3
Change of pipe pressure loss ptube from valve to cylinder.

According to the production of Hoerbiger-Hose-H3, the change of pipe

pressure loss ptube from valve to cylinder is shown in Figure 3.3.
Ignoring the pressure loss of the connector, the pipe pressure loss from the
proportional valve outlet to the inlet of the hydraulic cylinder is computed by

ptube = kp/Q × Qa × ltube , (3.12)

where kp/Q = 5 is the approximate slope of the pipe pressure loss with
respect to flow, Qa = 3 L/min is the average flow of this hydraulic system,
and ltube = 0.5 m is the pipe length from valve to cylinder. Thus, ptube is
computed approximately by

ptube = kp/Q × Qmax × ltube = 5 × 2.98 × 0.5 = 7.45 bar. (3.13)

If the maximal output pressure from the pump is the supply pressure ps ,
the maximal no-load flow of the servo valve is computed by

ps ps − pvloss − ptube
q0m = qn =5× L/min. (3.14)
pn 70

Thus, the no-load flow gain of the proportional valve is computed by

q0m 2
Kq0 = m /s. (3.15)
xvm
Linear PID Control Design 23

Then the load flow gain of the servo valve is computed by

ps − pl − pvloss − ptube
Kq = Kq0 , (3.16)
ps

where pl is the load pressure of the hydraulic cylinder, which is estimated by

the second-order Lagrangian math model.

3.2.2 Hydraulic Cylinder Model

The hydraulic cylinder is the actuator of the control loop to output the posi-
tion and pressure. In this example, the asymmetric hydraulic cylinder is
adopted as shown in Figure 2.1b. Some parameters of the hydraulic cylinder
“Hoerbiger LB6-1610-0080-4M” are shown in Table 3.2.
The cross-section of the rod chamber is computed by

πD2c π × (16 × 10−3 )2

Ap = = = 2.01 × 10−4 m2 . (3.17)
4 4
The total volume of the cylinder is as follows:

Vt = Ap Lc = 1.83 × 10−5 m3 . (3.18)

The pressure gain of the valve Kp is computed by

ps − pvloss − ptube
Kp = . (3.19)
xvm
The flow-pressure coefficient of the valve Kc and the coefficient Kce are
computed by
Kc = Kq /Kp ,
(3.20)
Kce = Kc + Ctl ,

TABLE 3.2
Parameters of the Hydraulic Cylinder
No. Name Symbol Value Unit

1 Cylinder length Lc 91 mm
2 Piston stroke dr max 79 mm
3 Cylinder diameter Dc 16 mm
4 Rod diameter Drod 10 mm
5 Load mass of No. 2 cylinder m2f 1.778 kg
6 Load mass of No. 1 cylinder m1f 3.550 kg
7 Load spring stiffness K 0 N/m
8 Load damping Be 50 Ns/m
9 Elastic modulus βe 7000 bar
24 Nonlinear Control Techniques for EHAs in Robotics Engineering

where Ctl is the total leakage coefficient of the cylinder.

From the linearized model of the single-rod electro-hydraulic actuator
(2.24), the load flow gain of the valve Kq and the flow-pressure coefficient
of the valve Kc are rewritten as follows:
⎧
⎪
⎪ 2
⎪
⎪ K = Kq ×
⎪
⎨ qa 1 + υ3
, (3.21)
⎪
⎪
⎪
⎪ 2(υ(p s − p vloss − ptube ) + pl )
⎪
⎩ Kqb = Kq × (p − p
vloss − ptube − pl )(1 + υ )
3
s

⎧
⎪
⎪ 2
⎪
⎪ K = Kc ×
⎪
⎨ ca 1 + υ3
, (3.22)
⎪
⎪
⎪
⎪ 2(ps − pvloss − ptube − pl )
⎪
⎩ Kcb = Kc × (υ(ps − pvloss − ptube ) + pl )(1 + υ 3 )

where υ ≤ 1 is the annulus area ratio of two chambers, and the coefficient Kce
is rewritten as Kcea = Kca + Ctl , Kceb = Kcb + Ctl .

3.2.3 Mechanical Motion Model

The motion control mechanism of the two-DOF robotic arm is shown in
Figure 3.4 [51]. The shoulder and elbow joints are driven by two single-rod
electro-hydraulic actuators.
At first, the general framework for the mechanical movement must be
given in Figure 3.5. Some mechanical parameters of the two-DOF robotic arm
are shown in Table 3.3.
The initial load mass is computed by

mf 0 = m2f 0 − m2 = 0.808 − 0.739 = 0.069 kg. (3.23)

If the load mass is mf , the moment of inertia of the load rotation from the
elbow is given by

Ifp2 = I2f − I2 − m2 (xc − P2 Pm20 )2 + mf (P2 P3 − xc )2 . (3.24)

If the load mass is mf0 , the centroid distance from the elbow to the forearm
(including initial load) xc0 is P2 Pm2 . So the moment of inertia of the initial
load rotation from the elbow is

Ifp20 = 0.0218 − 0.0145 − 0.730 × (0.122 − 0.103)2 + 0.069 × (0.33 − 0.122)2

= 0.010018 kgm2 .
(3.25)
Linear PID Control Design 25

Elbow θ2 >0
ε22

l2(θ2) b2 c2(θ2)
Shoulder cylinder θ1 < 0 θ1 > 0 Elbow cylinder
Va Vb d21 a2
Aa Ab
Pb ε21 Shoulder
Pa
d22
Qa Qb c1(θ1)
b1
ε11 a1 l1(θ1)
Servo valve d12 d13 Servo valve
d11

Relief valve 2-DOF robot arm

Ps Pr

Pump

FIGURE 3.4
Motion control mechanism of the two-DOF robotic arm.

P3
Forearm
Pm2

θ2 > 0 Y

P2 Cylinder 2
Elbow θ1 < 0

Upper arm εm1 Pm1

θ1 > 0

Cylinder 1 Shoulder
O
P1
X
P0

FIGURE 3.5
General framework for the mechanical movement.

If the load is a spherical particle, the moment of inertia of the load rotation
from the elbow is proportional to the mass of the load. So it is computed by

mf 1.039
Ifp2 = Ifp20 = × 0.010018 = 0.1508573 kgm2 . (3.26)
mf 0 0.069
26 Nonlinear Control Techniques for EHAs in Robotics Engineering

TABLE 3.3
Mechanical Parameters of the Two-DOF Robotic Arm
No. Name Symbol Value Unit

1 Forearm mass m2 0.739 kg

2 Upper arm mass with cylinder m1 1.772 kg
3 Load mass mf 1.039 kg
4 Moment of inertia for forearm around I2 0.0145 kgm2
the elbow
5 Moment of inertia for upper arm I1 0.0713 kgm2
around the shoulder
6 Forearm rotation range θ2 20–140 deg
7 Upper arm rotation range θ1 −70 to 50 deg
8 Forearm around the elbow inertia with I2f 0.0218 kgm2
load
9 Forearm mass with load m2f0 0.808 kg
10 Upper arm length P1 P2 0.35 m
11 Distance from centroid upper arm to P1 Pm1 0.164 m
shoulder
12 Forearm length P2 P3 0.33 m
13 Distance from centroid forearm to P2 Pm2 0.122 m
elbow
14 Distance from centroid forearm to P2 Pm20 0.103 m
elbow without load

When the current load is 1.039 kg, the distance from the centroid forearm
to the elbow (including the current load) is computed by

mf × P2 P3 + m2 × P2 Pm20
xc = = 0.23565 m. (3.27)
mf + m2

Substituting Ifp2 and xc into the above equation, the equivalent moment of
inertia of the forearm with the new load is computed by


I2f = 0.16911 kgm2 . (3.28)

The equivalent moment of inertia of the forearm with the new load rotated
at its centroid is computed by

 
I2fm = I2f − (m2 + mf )x2c = 0.07038 kgm2 . (3.29)

The equivalent moment of inertia of the upper arm with the cylinder
rotated at its centroid is computed by

I1m = I1 − m1 (P1 Pm1 )2 = 0.02364 kgm2 . (3.30)

Linear PID Control Design 27

According to Lagrange’s theorem, a two-link dynamic model is estab-

lished. The Lagrange function is defined as follows:

L = KE − V, (3.31)

where KE is the kinetic energy of the system and V is the potential.

Thus, the Lagrange equation of the two-DOF robotic arm is given by

∂ ∂L ∂L
T= − , (3.32)
∂t ∂ q̇ ∂q

where T is the generalized force vectors including force and torque and q is
the generalized coordinates vector of the system.
The coordinate X-O-Y is constructed in Figure 3.5. The shoulder joint P1 is
the origin point. A hinge point P1 is fixed with the bedframe. The position
vector of P1 Pm1 is as follows:
T
roPm1/o = P1 Pm1 sin(−θ1 − εm1 ) P1 Pm1 cos(−θ1 − εm1 )
(3.33)
T
= −P1 Pm1 sin(θ1 + εm1 ) P1 Pm1 cos(θ1 + εm1 ) .

The velocity vector of Pm1 is computed by

T
vPm1/o = −P1 Pm1 θ̇1 cos(θ1 + εm1 ) −P1 Pm1 θ̇1 sin(θ1 + εm1 ) . (3.34)

The position vector of P1 P2 is given by

T
roP2/o = P1 P2 sin(−θ1 − εm1 ) P1 P2 cos(−θ1 − εm1 )
(3.35)
T
= −P1 P2 sin(θ1 + εm1 ) P1 P2 cos(θ1 + εm1 ) .

The velocity vector of P2 is computed by

T
vP2/o = −P1 P2 θ̇1 cos(θ1 + εm1 ) −P1 P2 θ̇1 sin(θ1 + εm1 ) . (3.36)

The position vector of Pm2 is given by

roPm2/o = roP2/o + [−P2 Pm2 sin(θ1 + θ2 ), P2 Pm2 cos(θ1 + θ2 )]T

= [−P1 P2 sin(θ1 + εm1 ) − P2 Pm2 sin(θ1 + θ2 ), (3.37)
T
P1 P2 cos(θ1 + εm1 ) + P2 Pm2 cos(θ1 + θ2 )] .

The velocity vector of Pm2 is given by


−P1 P2 θ̇1 cos(θ1 + εm1 ) − P2 Pm2 (θ̇1 + θ̇2 ) cos(θ1 + θ2 )
vPm2/o = . (3.38)
−P1 P2 θ̇1 sin(θ1 + εm1 ) − P2 Pm2 (θ̇1 + θ̇2 ) sin(θ1 + θ2 )
28 Nonlinear Control Techniques for EHAs in Robotics Engineering

Then, the kinetic energy of the upper arm including cylinder 2 is com-
puted by

KEpm1 = 0.5m1 vPm1/o · vPm1/o + 0.5I1 θ̇1 · θ̇1 . (3.39)

The kinetic energy of the forearm including the load is computed by

KEpm2 = 0.5m2f vPm2/o · vPm2/o + 0.5I2f (θ̇1 + θ̇2 ) · (θ̇1 + θ̇2 ). (3.40)

The gravitational potential energy of the upper arm including cylinder 2 is

computed by

Vpm1 = m1 gP1 Pm1 cos(θ1 + εm1 ). (3.41)

The gravitational potential energy of the forearm including the load is

computed by

Vpm2 = m2f g(P1 P2 cos(θ1 + εm1 ) + P2 Pm2 cos(θ1 + θ2 )). (3.42)

So, the total energy of the two-link system is computed by

L = KEpm1 + KEpm2 − Vpm1 − Vpm2

= 0.5m1 P1 P2m1 θ̇12 + 0.5I1 θ̇12 + 0.5m2f [P1 P22 θ̇12 + P2 P2m2 (θ̇12 + θ̇22 )

+ 2P1 P2 · P2 Pm2 θ̇1 (θ̇1 + θ̇2 ) cos(θ2 − εm1 )] + 0.5I2f (θ̇1 + θ̇2 )2 (3.43)

− m1 gP1 Pm1 cos(θ1 + εm1 )

− m2f g[P1 P2 cos(θ1 + εm1 ) + P2 Pm2 cos(θ1 + θ2 )].

Using the Lagrange equation, if the angle vector q = [θ 1 , θ 2 ]T , then the kinetic
equation of the two-link system is given as follows:

H(q)q̈ + C(q, q̇)q̇ + G(q) = T, (3.44)

where

   
H11 H12 C11 C12 G1 Tu
H= , C= , G= , T= , (3.45)
H21 H22 C21 C22 G2 Tf
Linear PID Control Design 29

H11 = I1 + I2f + m1 P1 P2m1 + m2f P1 P22 + m2f P2 P2m2

+ 2m2f P1 P2 · P2 Pm2 cos(θ2 − εm1 ),
H12 = I2f + m2f P1 P2 · P2 Pm2 cos(θ2 − εm1 ),
H22 = I2f + m2f P2 P2m2 ,
C11 = −2m2f P1 P2 · P2 Pm2 θ̇2 sin(θ2 − εm1 ),
C12 = −m2f P1 P2 · P2 Pm2 θ̇2 sin(θ2 − εm1 ), (3.46)
C21 = m2f P1 P2 · P2 Pm2 θ̇1 sin(θ2 − εm1 ),
C22 = 0,
G1 = −m1 gP1 Pm1 sin(θ1 + εm1 )
− m2f g[P1 P2 sin(θ1 + εm1 ) + P2 Pm2 sin(θ1 + θ2 )],
G2 = −m2f gP2 Pm2 sin(θ1 + θ2 ).

After the kinetic model is constructed, according to the changes of the input
commands, a rate, and secondary rate of commands, two driving torques of
a two-link system are computed.
Then, two cylinder dynamic lengths are computed by

⎧ 
⎪
⎨ c1 (θ1 ) = a21 + b21 − 2a1 b1 cos(π/2 + θ1 + ε11 )
 , (3.47)
⎪
⎩ c (θ ) = a2 + b2 − 2a b cos(π − θ − ε − ε )
2 2 2 2 2 2 2 21 22

where a1 = a2 = 0.3219 m, b1 = b2 = 0.045 m, ε11 = 6.24°, ε21 = 8.04°, and

ε22 = 6°.
Thus, the two dynamic force arms are given by

⎧   
⎪
⎪ a21 + c1 (θ1 )2 − b21
⎪
⎪ l (θ ) = a1 sin arccos
⎪
⎨ 1 1 2a1 c1 (θ1 )
   , (3.48)
⎪
⎪ a22 + c2 (θ2 )2 − b22
⎪
⎪
⎩ l2 (θ2 ) = a2 sin arccos
⎪
2a2 c2 (θ2 )

where the ranges of two joint angles are − 70° ≤ θ 1 ≤ 50°, 20° ≤ θ 2 ≤ 140°.
In this robotic plant, the simulation results of two cylinder dynamic lengths
and force arms are as shown in Figure 3.6
30 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) (b)
0.37 0.045
0.36
0.35 0.04
0.34
0.035
0.33
c1 (m)

l1 (m)
0.32
0.03
0.31
0.3 0.025
0.29
0.02
–80 –60 –40 –20 0 20 40 60 –80 –60 –40 –20 0 20 40 60
θ1 (°) θ1(°)

(c) (d)
0.37 0.045
0.36
0.35 0.04
0.34
0.035
0.33
c2 (m)

l2 (m)

0.32
0.03
0.31
0.3 0.025
0.29
0.02
20 40 60 80 100 120 140 20 40 60 80 100 120 140
θ2 (°) θ2 (°)

FIGURE 3.6
Two cylinder dynamic lengths and force arms. (a) The dynamic length of the upper arm cylinder
c1 . (b) The dynamic force arm of the upper arm cylinder l1 . (c) The dynamic length of the forearm
cylinder c2 . (d) The dynamic force arm of the forearm cylinder l2 .

3.3 System Performance Analysis

The system performance should be considered due to the selected motor
and the pump. From Equation 2.12, the dynamic loads FL of the robotic arm
depend on the hydraulic cylinder position y, velocity ẏ, and acceleration ÿ.
Thus, this system performance includes motor performance, output pressure
analysis of the pump, and dynamic pressure of the cylinder with load.

3.3.1 Motor Performance

The maximal allowable speed of the pump is 3000 rpm. According to the
mechanical efficiency curve in the motor type “BSM63N-375,” the continu-
ous rated torque is 1.85 Nm as the motor is operated in 3000 rpm with load
Linear PID Control Design 31

8
70

650 V
10

565 V

7
D
60

C BU
D
C BU

6
8

50

S
320 V
Current (Amps)

5
DC B
Torque (lb-in.)

Torque (Nm)
160 V
40
6

US
DC B

4
30

US

3
4

20

2
2

10

1
0

0
0 1 2 3 4 5 6 7
Speed (rpm) × 1000

FIGURE 3.7
Mechanical properties curve of the motor with load.

as shown in Figure 3.7. So, at this operated point, the motor rated power is

1.85 × 3000 × 2 × π
Pm = = 581 W. (3.49)
60

3.3.2 Output Pressure Analysis of Pump

The fixed pump displacement ν = 3.14 mL/r. The continuous torque of
motor Tc = 2.09 Nm. The peak torque is Tpeak = 8.36 Nm. So, the rated
pressure and the maximum pressure of the pump is computed by

Tc × 2 × π × 10
Pn = × η = 39.7 bar,
ν
(3.50)
Tpeak × 2 × π × 10
Pmax = × η = 158.9 bar.
ν

where η = 0.95 is the volumetric efficiency coefficient of the pump.

32 Nonlinear Control Techniques for EHAs in Robotics Engineering

The peak motor torque can stick about 2 s. This indicates only a short time
for the pump to provide a maximum pressure 158.9 bar. So the operated pres-
sure is no more than 40 bar long time. The maximum output flow of the pump
is computed by
Qmax = 3000 × ν × η = 8.95 L/min. (3.51)

Considering the servo valve and pipe pressure loss, the actual pressure of
cylinders is computed by

Pan = Pn − pvloss − ptube = 39.7 − 1.4 − 7.5 = 30.8 bar,

(3.52)
Pa max = Pmax − pvloss − ptube = 158.9 − 1.4 − 7.5 = 150 bar.

Considering the full efficiency of the pump ς , the output power of the
pump is computed by

Pout = Pm × ς = 581 × 0.85 = 494 W. (3.53)

3.3.3 Dynamic Pressure of Cylinder with Load

The pressure of the cylinder with load is a dynamic value, which depends on
the mass of the load, joint rotational movement, angular velocity, and angular
acceleration. The motion model of a two-link mechanical structure with load
® ®
is built by MATLAB /Simulink shown in Figure 3.8. Two joint angles θ 1 , θ 2
are imported like these sinusoidal variation.
⎧
⎪ θmin + θmax θmax − θmin
⎪
⎪ θ (t) = + sin(2πft)
⎪
⎪ 2 2
⎪
⎨
θmax − θmin
θ̇(t) = × 2π × cos(2π ft) . (3.54)
⎪
⎪ 2
⎪
⎪
⎪
⎩ θ̈(t) = − θmax − θmin × 4π 2 × sin(2π ft)
⎪
2

Torque
Theta Theta Matrix Matrix Terminator

Force Force
Sub_Theta_Command Sub_Matrix
Theta
c_Theta c_Theta

Sub_Torque_Force Sub_Actuator_Flow

FIGURE 3.8
Simulation model for solvering the dynamic pressure of the cylinder with load.
Linear PID Control Design 33

(a) 150 (b) 60

50
100
40
50
30
Pu (bar)

Pf (bar)
0 20
10
–50
0
–100
–10
–150 –20
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)
(c) 700 (d) 1
0.8
600
0.6
500
0.4
Poweru (W)

400
Qf (L/min)

0.2
300 0
200 –0.2
–0.4
100
–0.6
0 –0.8
–100 –1
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)
(e) 4 (f) 1
0.8
3
0.6
0.4
2
Powerf (W)
Qu (L/min)

0.2
1 0
–0.2
0 –0.4
–0.6
–1
–0.8
–2 –1
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

FIGURE 3.9
Simulation result with the elbow joint θ 2 = 140°, the sinusoidal input of the shoulder angle
f u = 1.05 Hz. (a) The pressure of the upper arm cylinder with load pu . (b) The pressure of the
forearm cylinder with load pf . (c) The flow of the upper arm cylinder with load Qu . (d) The flow
of the forearm cylinder with load Qf . (e) The power consumption of the upper arm cylinder with
load Poweru . (f) The power consumption of the forearm cylinder with load Powerf .

Case 1: The pressure of the cylinder supporting the shoulder

If the elbow joint is fixed, θ2 = θ2 min = 140◦ and the load mf = 1.039 kg,
and the shoulder angle is sinusoidal input, the simulation result is acquired
as shown in Figure 3.9.
As shown in Figure 3.9, if the condition of the frequency of θ 1 is 1.05 Hz, the
flow of the upper arm cylinder with load is 150 bar. It reaches the maximum
motor torque value. But the power of the upper arm cylinder is 110 W, which
is less than the rated power 494 W. At this moment, the flow of the upper
34 Nonlinear Control Techniques for EHAs in Robotics Engineering

TABLE 3.4
Parametric Analysis of the Shoulder Hydraulic Actuator
Maximum Maximum Maximum Maximum
Condition Frequency Pressure Flow Power

θ 2 = 140° 1.05 Hz 150 bar 3.75 L/min 110 W

θ 2 = 20° 0.7 Hz − 149 bar 2.77 L/min 76 W

arm cylinder is 3.75 L/min, which is less than the output flow of the pump
8.95 L/min. These results indicate that the frequency of the shoulder angle
should not exceed 1.05 Hz. The parametric analysis of the shoulder hydraulic
actuator is shown in Table 3.4.
Case 2: The pressure of the cylinder supporting the elbow
If the shoulder joint is fixed, θ1 = θ1 min = −70◦ and the load mf = 1.039 kg,
and the elbow angle is sinusoidal input, the simulation result is acquired as
shown in Figure 3.10.
As shown in Figure 3.10, if the condition of the frequency of θ 2 is 1.85 Hz,
the flow of the forearm cylinder with load is 150 bar. It reaches the maxi-
mum torque value. But the power of the forearm cylinder is 226 W, which
is less than the rated power 494 W. At this moment, the flow of the fore-
arm cylinder is 6.6 L/min, which is less than the output flow of the pump
8.95 L/min. These results indicate that the frequency of the elbow angle
should not exceed 1.85 Hz. The parametric analysis of the elbow hydraulic
actuator is shown in Table 3.5.
Case 3: The pressure of the cylinder with two joints in simultaneously motion
If two joints change simultaneously and the load mf = 1.039 kg, the simu-
lation results are acquired as shown in Figure 3.11.
As shown in these figures, if the condition of the frequency of θ 1 is 0.6 Hz
and θ 2 is 1 Hz, the flow of the upper arm cylinder with load is 150 bar. It
reaches the maximum torque value. But the power of the upper arm cylinder
is 70 W, which is less than the rated power 494 W. Also, the power of the
forearm cylinder is only 47 W. At this moment, the total flow of two cylinders
is 5 L/min, which is less than the output flow of the pump 8.95 L/min. These
results indicate that the frequencies of shoulder and elbow angles should not
exceed 0.6 and 1 Hz respectively. The parametric analysis of two hydraulic
actuators is shown in Table 3.6.

3.4 PID Controller Design

The actual angles of two joints are measured by encoders. The errors between
actual angles and command angles are as input of the control system. The
controller inputs the control voltage into proportional valves and the spool
dynamic positions are regulated to drive the cylinder motion.
Linear PID Control Design 35

(a) (b)
200 250
200
150 150
100
100
Pu (bar)

Pf (bar)
50
0
50
–50

0 –100
–150
–50 –200
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)
(c) 1 (d) 7
0.8 6
0.6 5
0.4 4
Qf (L/min)
Qu (L/min)

0.2 3
0 2
–0.2 1
–0.4 0
–0.6 –1
–0.8 –2
–1 –3
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)
(e) 1
(f ) 200
0.8
0.6 0
0.4
Powerf (W)

–200
Poweru (W)

0.2
0 –400
–0.2
–600
–0.4
–0.6 –800
–0.8
–1 –1000
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

FIGURE 3.10
Simulation result with the shoulder joint θ 1 = − 70°, the sinusoidal input of the elbow angle
f f = 1.85 Hz. (a) The pressure of the upper arm cylinder with load pu . (b) The pressure of the
forearm cylinder with load pf . (c) The flow of the upper arm cylinder with load Qu . (d) The flow
of the forearm cylinder with load Qf . (e) The power consumption of the upper arm cylinder with
load Poweru . (f) The power consumption of the forearm cylinder with load Powerf .

Considering the dynamic external load on two hydraulic actuators, the

supply pressure is chosen as follows:
3
ps = pl max . (3.55)
2
36 Nonlinear Control Techniques for EHAs in Robotics Engineering

TABLE 3.5
Parametric Analysis of Elbow Hydraulic Actuator
Maximum Maximum Maximum Maximum
Condition Frequency Pressure Flow Power

θ 1 = − 70° 1.85 Hz 150 bar 6.6 L/min 226 W

θ 1 = 50° 1.95 Hz − 150 bar 7.0 L/min 255 W

The supply pressure ps is 150 bar. pl max is the maximal pressure of the cylin-
der with load. If the load mass is 3 kg, the value is 100 bar of the upper arm
cylinder, which satisfies this relation. So the load mass is chosen as 3 kg suffi-
ciently. Owing to the asymmetrical hydraulic cylinder, the open-loop transfer
function of this system is shown as follows:

1. If the hydraulic cylinder is extended, that is, ẏ ≥ 0, then

Kqa Ksv
Gc (s)
A T s+1
Gol =  a sv  . (3.56)
Vt mt m t Kcea Be Vt
s +
3 + s2
2(1 + υ 2 )βe A2a A2a 2(1 + υ 2 )βe A2a
 
Be Kcea Vt K
+ 1+ + s + KK2cea
A2a 2(1 + υ 2 )βe A2a Aa

2. If the hydraulic cylinder is retracted, that is, ẏ < 0, then

Kqb Ksv
Gc (s)
A T s+1
Gol =  a sv  , (3.57)
Vt mt mt Kceb Be Vt
s 3 + + s2
2(1 + υ 2 )βe A2a A2a 2(1 + υ 2 )βe A2a
 
Be Kceb Vt K KKceb
+ 1+ + s+
Aa 2 2(1 + υ )βe Aa
2 2 A2a

where Gc (s) is the PID control law.

The maximal supply pressure ps is 150 bar and the maximal load mass mf
is 3 kg. The control law is designed by MATLAB/SISO tool like this
 
ki 1
Gc (s) = kp + × , (3.58)
s 1 + b1 s

where kp and ki are the proportional and integral control parameters, and
b1 is the lag parameter. This hysteresis element 1/(1 + b1 s) can improve the
dynamic quality of the control variable u and suppress the control saturation
as the initial large error existing in the system.
Linear PID Control Design 37

(a) 150 (b) 200

100 150

50 100
Pu (bar)

Pf (bar)
0 50

–50 0

–100 –50

–150 –100
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)
(c) 2.5 (d) 4
2 3
1.5
2
Qu (L/min)

Qf (L/min)

1
1
0.5
0
0

–0.5 –1

–1 –2
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)
(e) 500 (f) 50
0
400 –50
–100
300
Powerf (W)
Poweru (W)

–150
200 –200
–250
100 –300
–350
0
–400
–100 –450
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
t (s) t (s)

FIGURE 3.11
Simulation result with two sinusoidal inputs of the shoulder angle f u = 0.6 Hz and the elbow
angle f f = 1 Hz. (a) The pressure of the upper arm cylinder with load pu . (b) The pressure of the
forearm cylinder with load pf . (c) The flow of the upper arm cylinder with load Qu . (d) The flow
of the forearm cylinder with load Qf . (e) The power consumption of the upper arm cylinder with
load Poweru . (f) The power consumption of the forearm cylinder with load Powerf .

The designed results of control design for the upper arm cylinder is shown
as follows:
1. If the hydraulic cylinder is extended, that is, ẏ ≥ 0, the con-
trol parameters and the lag parameter are kp = 178.5, ki = 21, and
b1 = 0.073.
38 Nonlinear Control Techniques for EHAs in Robotics Engineering

TABLE 3.6
Parametric Analysis of Two Joint Angles Simultaneously Motion
Joint Maximum Frequency Maximum Pressure Maximum Flow Maximum Power

Shoulder 0.6 Hz 145 bar 2.14 L/min 70 W

Elbow 1 Hz 70 bar 3.6 L/min 47 W

Root locus editor for open loop 1 (OL1) Open-loop bode editor for open loop 1 (OL1)
4000 200

2000 100

0 0

–100
–2000

–200 G.M.: 21.3 dB

–4000 Frequency: 32.9 rad/s
–4000 –2000 0 2000 4000
Stable loop
Bode editor for closed loop 1 (CL1) –300
0
–90

–200 –180

–400 –270
0
–180 –360
–360 P.M.: 57°
Frequency: 7.02 rad/s
–540 –450
100 102 104 10–5 100 105
Frequency (rad/s) Frequency (rad/s)

FIGURE 3.12
Frequency domain characteristic of the open-loop control system for the shoulder actuator for
Equation 3.56.

The frequency domain characteristic of the open-loop control sys-

tem for the upper arm actuator is as shown in Figure 3.12. In the
right Bode chart, the amplitude margin G.M. is 21.3 dB and the phase
margin P.M. is 57°, and the closed loop is a stable loop. Generally,
if G.M. is larger than + 6 dB and P.M. is larger than 45°, the stabil-
ity margin is enough. Furthermore, the open-loop root locus shows
that all the roots of the closed-loop control system are located in
the negative half plane. Thus, the feedback control system by the PI
controller Gc (s) is stable. The bandwidth of this closed-loop control
system is 7.02 rad/s, that is, 1.11 Hz. If the sinusoidal input demand
is less than 1.11 Hz, then the dynamic tracking performance is sat-
isfactory. Otherwise, the tracking performance will be declined. The
Linear PID Control Design 39

(a) Step response (b)

0.09 10
0.08
8
0.07
0.06 6
Amplitude

Amplitude
0.05
4
0.04
0.03 2
0.02
0
0.01
0 –2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Time (s) Time (s)

FIGURE 3.13
Performance results of the PI controller design for the shoulder actuator for Equation 3.56. (a)
The step response of the upper arm cylinder for the maximal stroke y. (b) The voltage control
with respect to step response u.

performance results of the PI controller design for the shoulder actu-

ator is as shown in Figure 3.13. As a step demand is selected as the
input, the actual position response of the upper arm cylinder is as
shown in Figure 3.13a. The step response for the maximal stroke is
regulated to the maximum value 79 mm by the PID controller after
0.8 s. The overshoot is less than 13% and the rising time is less than
0.3 s, which indicates the superior performance of the control design.
The voltage control u with respect to step response is shown as Figure
3.13b. The dynamic magnitude of u is less than the control saturation
umax = 10 V, which satisfies the engineering requirement.
2. If the hydraulic cylinder is retracted, that is, ẏ < 0, the control
parameters and the lag parameter are kp = 102, ki = 12, b1 = 0.073.
The frequency domain characteristic of the open-loop control sys-
tem for the upper arm actuator is as shown in Figure 3.14. The
amplitude margin G.M. is 18.6 dB, the phase margin P.M. is 49.5°, and
the bandwidth of this closed-loop control system is 9.08 rad/s, that is,
1.43 Hz, which is larger than case 1. The performance results of case 2
is shown as Figure 3.15. The steady-state time is 1 s, which has a little
weaker performance than case 1. However, the control magnitude is
less than 5 V, which denotes that the control consumption of case 2 is
more than case 1.

To address different external loads, the required supply pressure (3.55) is

different. Thus, from Equations 3.21 and 3.22, the model parameters Kqa ,
Kqb , Kcea , and Kceb are distinguished. To guarantee enough stability mar-
gin of Gol (s) and the similar dynamic performance, the control parameters
40 Nonlinear Control Techniques for EHAs in Robotics Engineering

Root locus editor for open loop 1 (OL1) Open-loop bode editor for open loop 1 (OL1)
4000 150

100
2000
50

0 0

–50
–2000
–100
G.M.: 18.6 dB
–4000 –150 Frequency: 33.3 rad/s
–4000 –2000 0 2000 4000 Stable loop
–200
Bode editor for closed loop 1 (CL1)

0 –90

–200 –180

–400 –270
0
–180 –360
–360 P.M.: 49.5°
Frequency: 9.08 rad/s
–540 –450
100 102 104 10–4 10–2 100 102 104
Frequency (rad/s) Frequency (rad/s)

FIGURE 3.14
Frequency domain characteristic of the open-loop control system for the shoulder actuator for
Equation 3.57.

(a) Step response (b) Step response

0.1 6
0.09
5
0.08
0.07 4
Amplitude
Amplitude

0.06 3
0.05
0.04 2
0.03 1
0.02
0
0.01
0 –1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s) Time (s)

FIGURE 3.15
Performance results of the PI controller design for the shoulder actuator for Equation 3.57. (a)
The step response of the upper arm cylinder for the maximal stroke y. (b) The voltage control
with respect to step response u.
Linear PID Control Design 41

TABLE 3.7
Control Parameters and Performance Design with Respect to Different Pressures of
the Upper Arm Cylinder with Different Loads in the Condition of Extended State
Load Pressure (bar) Kp /Ki Bandwidth (rad/s) Steady Time (s) Maximum Control (V)

20 150/10 8.92 1 7.42

40 150/10 8.37 1 7.53
60 150/10 7.73 1 7.68
80 170/10 7.74 1 8.70
100 179/21 7.02 0.9 9.34

TABLE 3.8
Control Parameters and Performance Design with Respect to Different Pressures of
the Upper Arm Cylinder with Different Loads in the Condition of Retracted State
Load Pressure (bar) Kp /Ki Bandwidth (rad/s) Steady Time (s) Maximum Control (V)

20 128/15 8.7 1 6.35

40 128/15 9.28 1 6.25
60 119/14 9.31 1 5.82
80 110/13 9.24 1 5.40
100 102/12 9.08 1 5.0

need to be adaptively regulated with the variable external load. Table 3.7
gives the control parameters and performance design with respect to differ-
ent pressures of the upper arm cylinder with different loads in the condition
of extended state.
As the hydraulic cylinder is controlled to be retracted, the control
parameters are also switched into the other designed interpolation table as
shown in Table 3.8. Obviously, the control parameters in the retracted condi-
tion are less than the corresponding parameters in the extended condition
because the external load of the retracted cylinder is negative where the
gravity of the load mass need not be compensated.
Similarly, the control parameters of the forearm cylinder is also designed
like the above two tables in the condition of extended and retracted states.

3.5 Experiment
To verify the PI controller, the experimental bench of the two-DOF robotic
arm driven by EHA is set up as shown in Figure 3.16 [11,12]. The two EHAs
include two servo valves (Moog D633-R02K01M0NSM2), two double-acting
cylinders (Hoerbiger LB6-1610-0080-4M), an axial piston pump (Takako TFH-
315), a servo motor (BALDOR BSM63N-375), and a relief valve. The angle
42 Nonlinear Control Techniques for EHAs in Robotics Engineering

11
1

3 10

8 9 4
5
6

FIGURE 3.16
Experimental equipment (1—robotic arm, 2—hydraulic cylinder, 3—servo valve, 4—pressure
gauge, 5—relief valve, 6—fixed displacement pump, 7—servo motor, 8—encoder, 9—pressure
sensor, 10—tank, 11—IPC).

TABLE 3.9
Hydraulic Parameters Used in Simulation and Experiment
Parameter Value Parameter Value

Cd 0.62 w 0.024 m
ps 40 bar Aa 2.01 cm2
Ab 1.25 cm2 ν 0.62
Vt 1.74 × 10−5 m3 βe 7 × 108 Pa
Ksv 7.9 × 10−5 m/V Lmax 58 mm
Tsv 12 ms umax 10 V
K 0 b 2200 N.s/m
Ctl 2.5 × 10−11 m3 /(s · Pa) ρ 850 kg/m3

of joint is measured by a relative encoder (AVAGO AEDA-3300-BE1), which

can obtain the cylinder position by trigonometry computation. The load pres-
sure on each EHA is measured by an optional tension/compression load cell
(Burster 8417-6005). The position derivatives is computed by forward Euler
method. The control law is executed by industrial personal computer (IPC).
The load mass is a disk on the forearm terminal. Some hydraulic parameters
of this experimental bench are shown in Table 3.9. The two cylinder position
demands are considered as y1d = 0.5Lmax sin(1.6π t), y2d = 0.5Lmax sin(2π t).
According to Tables 3.7 and 3.8, the PI control parameters are designed as
kp = 150, ki = 10, if ẏi ≥ 0. Otherwise, kp = 128, ki = 15.
Linear PID Control Design 43

Figures 3.17 and 3.18 show the position tracking results of the two
hydraulic cylinders. The magnitude of two sinusoidal demands is 29 mm,
and the two frequencies are 0.8 and 1 Hz. The two dynamic position errors
are less than 3 mm, that is, (yid − yi )/yid < 10%. The position errors of the
forearm cylinder is less than that of the upper arm cylinder, since the exter-
nal load of the upper arm hydraulic actuator is more violent than that of
forearm actuator from Figure 3.11. The two control variable u does not
exceed the control saturation ± 10 V, which means that the designed PI
controller can guarantee the stability of the closed-loop feedback control

(a) y1d y1 (b) Δy1

30 3
2.5
20
2
Position (mm)
Position (mm)

10 1.5
1
0
0.5
–10 0
–0.5
–20
–1
–30 –1.5
0 5 10 15 0 5 10 15
Time (s) Time (s)

FIGURE 3.17
Position tracking results of the upper arm cylinder. (a) The upper arm position response y1 . (b)
The upper arm position error y1 .

(a)
y2d y2 (b) Δy2
30 3

20 2
Position (mm)

Position (mm)

10 1

0 0

–10 –1

–20 –2

–30 –3
0 5 10 15 0 5 10 15
Time (s) Time (s)

FIGURE 3.18
Position tracking results of the forearm cylinder. (a) The forearm position response y2 . (b) The
forearm position error y2 .
44 Nonlinear Control Techniques for EHAs in Robotics Engineering

system of the electro-hydraulic actuator, and the dynamic tracking perfor-

mance of the cylinder position is satisfactory (Figure 3.19). Figure 3.20 shows
the two chamber pressures of the asymmetrical cylinder. These pressures
are switched from a small value into the approximate value of the supply
pressure, which is less than the supply pressure ps = 40 bar.

(a) 8 (b) 8 u2
u1
6 6
4 4

Control (V)
Control (V)

2 2
0 0
–2 –2
–4 –4
–6 –6
–8 –8
0 5 10 15
0 5 10 15
Time (s)
Time (s)

FIGURE 3.19
Dynamic control voltages of two servo valves. (a) The upper arm control variable u1 . (b) The
forearm control variable u2 .

(a) 50 (b) 40
pa pb pa pb
35
40
Chamber pressure (bar)

30
pressure (bar)

30
25
20 20
15
10
10
0 5
–10 0
0 2 4 8 10 12 0 2 4 8 10 12
Time (s) Time (s)

FIGURE 3.20
Two chamber pressures of two cylinders. (a) The two chamber pressures of the upper arm
cylinder pa , pb . (b) The two chamber pressures of the forearm cylinder pa , pb .
4
Robust Control Method

CONTENTS
4.1 Linearized Hydraulic Model Construction . . . . . . . . . . . . . . . . . . . . 46
4.2 Analysis of Parametric Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.3 Robust Model Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 Robust Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.4.1 Analysis of Open-Loop System . . . . . . . . . . . . . . . . . . . . . . . 54
4.4.2 Weight Function Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4.3 Robust Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4.4 Simulation and Experimental Result . . . . . . . . . . . . . . . . . . . 62

Some hydraulic parametric uncertainties exist in electro-hydraulic systems.

These uncertainties include uncertain linear/nonlinear parameter and uncer-
tain nonlinearity [17], which are caused by unknown viscous damping, load
stiffness, variations in control fluid volumes, physical characteristics of valve,
bulk modulus, and oil temperature variations [42]. If these uncertainties are
not addressed in the control design, the dynamic and steady performance of
EHS will be degraded and the robustness of the designed controller will also
be reduced. In this chapter, the robust control method of EHS is proposed to
handle the parametric uncertainties and compensate their negative effect in
real engineering [20]. Thus, the dynamic and steady behaviors are obviously
improved, especially the largely unknown load disturbances emerging in the
electro-hydraulic actuator.
In this chapter, first, the linear mathematical model of EHS is constructed
by flow-pressure model linearization. Second, the external load forces of the
two-DOF robot arm are modeled by Lagrange dynamic equation. A bounded
linear function is constructed to describe the relationship between load forces
and cylinder position. Then the H∞ suboptimal controller is designed in the
output feedback form with structural and parametric uncertainty. Both the
simulation and experiment are performed to validate the effectiveness of
the developed algorithm.

45
46 Nonlinear Control Techniques for EHAs in Robotics Engineering

4.1 Linearized Hydraulic Model Construction

In this section, for convenient discussion, the dynamic cylinder model of
EHA is adopted by the linearized hydraulic model mentioned in Equa-
tions 2.4 and 2.25. Thus, the linear model of the hydraulic cylinder is shown
as follows:

Kqi
 
1 Vt
Aa xv (s) − A2a Kcei + 2(1+υ 2 )βe
s FL (s)
y(s) =   , (4.1)
Vt mt mt Kcei bVt
s3 + + 2 s
2
2(1+υ 2 )βe A2a A2a 2(1+υ 2 )βe Aa
 
bKcei Vt K KKcei
+ 1+ + s+
A2a 2(1+υ 2 )βe A2a A2a

where y is the displacement of the piston, xv is the spool position of the servo
valve, Kqi , Kcei (i = a, b) are the flow gain of the servo valve and flow-pressure
coefficient, Aa and Ab are the annulus areas of the two chambers (υ = Aa/Ab ),
V t is the cylinder volume, β e is the effective bulk modulus, mt is the load
mass, K is the load spring constant, b is the viscous damping coefficient of
oil, and FL is the external load on the hydraulic actuator from the mechanical
structure of the two-DOF robotic arm.

Remark 4.1
From Equation 4.1, it is clear that the hydraulic cylinder model is simplified
as a third-order linear model. If there exist bounded uncertain parameters
such as V t , K, b, Kq , and Kce , this linear model is appropriately handled by the
robust control method, which obtains an H∞ suboptimal solution to preserve
a satisfactory dynamic control performance.

The dynamic characteristic of the servo valve can be described as a second-

order linear model mentioned in Equation 2.9 as follows:

2 2
ẍv + 2ζsv ωsv ẋv + ωsv xv = Ksv ωsv u, (4.2)

where ζ sv is the damping ratio, ωsv is the natural frequency, Ksv is the servo
valve gain, and u is the control voltage.
According to Equations 4.1 and 4.2, the whole state-space model of the
electro-hydraulic actuator is five. The control input variable is u and the
output cylinder position is y.
Robust Control Method 47

4.2 Analysis of Parametric Uncertainty

In the linear model of the cylinder (4.1), V t , b, and Kq are three paramet-
ric uncertainties. They are caused by operating conditions, environmental
variability, hydraulic characteristic variations, and the linearized model error.
The flow gain of the servo valve Kq changes with time due to variable load
pressure PL and component degradation variations. Because of the asymme-
try cylinder, the value of Kq is different in cylinder extension or retraction,
respectively.
Some hydraulic parameters used in simulation and experiments are shown
in Table 4.1. According to Equations 3.16 and 3.21, the load flow gain of valve
Kq for the single-rod electro-hydraulic actuator is rewritten as follows:


ps − pl − pvloss − ptube
Kq = Kq0 ,
ps

2
Kqa = Kq , (4.3)
1 + υ3

2(υ(ps − pvloss − ptube ) + pl )
Kqb = Kq .
(ps − pvloss − ptube − pl )(1 + υ 3 )

The load mass is 1 kg and the motion frequency of the shoulder arm is
assumed to be no more than 0.5 Hz. Thus, the load pressure pL range of the
shoulder cylinder can be estimated by Lagrange equation. The maximum
value of pL is 40 bar. If some known parameters are substituted into Equa-
tion 4.3, the flow gain ranges of the rod chamber and non-rod chamber are

TABLE 4.1
Hydraulic Parameters Used in Simulation and
Experiments
Parameter Value Parameter Value

xv max 0.79 mm α 0.62

Ps 40 bar Pr 2 bar
Aa 2.01 cm2 Kq0 0.015 m2 /s
Ksv 3.95 × 10−4 m/V βe 2 × 108 Pa
ωsv 353.6 rad/s ζ sv 0.707
pvloss 1.4 bar ptube 7.5 bar
Lc 157 mm dr max 79 mm
m1 1.772 kg m2 0.739 kg
48 Nonlinear Control Techniques for EHAs in Robotics Engineering

estimated as follows:
0.012 ≤ Kqa ≤ 0.02,
(4.4)
0.019 ≤ Kqb ≤ 0.027.

So the range of Kq can be considered as 0.012 < Kq < 0.027. The nominal
values K̄q of Kq is 0.02 m2 /s and the maximum relative uncertainty pKq is
0.375 with the relative variation. The flow gain of the servo valve Kq can be
represented as follows:

Kq = K̄q (1 + pKq δKq ), −1 ≤ δKq ≤ 1. (4.5)

Because the dynamic change of volume in two chambers is neglected in

Equation 2.1, the cylinder volume V t is not accurate. The actual volume can
be described as follows:

Va = Va0 + Aa y
. (4.6)
Vb = Vb0 − Ab y

If the following property holds:

Va0 = Vb0 = Vt /2,

    (4.7)
Aa y  Vt , Ab y  Vt ,

then flow-pressure continuous equation of the cylinder is available. How-

ever, the initial volumes of two chambers are unknown, which means per-
haps not in the center of the cylinder chamber. So the parameter V t can be
represented as a parametric uncertainty. The initial volumes of two chambers
are estimated as follows:

(Lc − dr max )Aa /2 ≤ Va0 ≤ (Lc + dr max )Aa /2
. (4.8)
(Lc − dr max )Ab /2 ≤ Vb0 ≤ (Lc + dr max )Ab /2

After the known parameters are substituted into Equation 4.8, the nom-
inal values V̄t of V t is 2.85e-5 m3 and the maximum relative uncertainty
pVt is 0.663 with the relative variation δ Vt . The cylinder volume V t can be
represented as follows:

Vt = V̄t (1 + pVt δVt ), −1 ≤ δVt ≤ 1. (4.9)

The viscous damping coefficient of the cylinder b can be estimated by the

experiment test. The value is 5000 or 2000 Ns/m in cylinder extension or
retraction, respectively. So the nominal values b̄ of b is 3500 Ns/m and the
Robust Control Method 49

maximum relative uncertainty pb is 0.428 with the relative variation δ b . The

viscous damping coefficient b can be represented as follows:

b = b̄(1 + pb δb ), −1 ≤ δb ≤ 1. (4.10)

Now the external load FL is the maximum disturbance in the dynamic char-
acteristic of the cylinder. It can be considered as an unmodeled disturbance.
In this chapter, the quantitative relationship between the external load FL and
the position of the cylinder is analyzed to describe the structural uncertainty.
According to Equation 3.43, two torques on the upper arm and forearm Tu
and Tf are obtained by Lagrange method. Then, two external load forces FLu ,
FLf that change with the variation of two joint angles are computed by
⎧ Tu (θ1 , θ2 )
⎪
⎨ FLu (θ1 , θ2 ) = l (θ )
⎪
1 1
, (4.11)
⎪
⎪ T (θ ,θ )
⎩ FLf (θ1 , θ2 ) = f 1 2
l2 (θ2 )

where li (θ i )(i = 1, 2) are the force arms of the external load forces.
To consider the effect of the external load in the linear model (4.1) of EHS,
two fictitious proportional gains are defined as follows:
⎧
⎪ FLu (θ1 , θ2 ) Tu (θ1 , θ2 )
⎪
⎪ KFLu (θ1 ) = =
⎪
⎪ c1 (θ1 ) l1 (θ1 )c1 (θ1 )
⎪
⎪
⎪
⎪ FLu (θ1 , θ2 ) Tu (θ1 , θ2 )
⎪
⎪
⎨ KFLu (θ2 ) = c2 (θ2 )
⎪ =
l1 (θ1 )c2 (θ2 )
, (4.12)
⎪
⎪ FLf (θ1 , θ2 ) Tf (θ1 , θ2 )
⎪
⎪ KFLf (θ1 ) = =
⎪
⎪ c1 (θ1 ) l2 (θ2 )c1 (θ1 )
⎪
⎪
⎪
⎪
⎪
⎪ F (θ , θ ) Tf (θ1 , θ2 )
⎩ KFLf (θ2 ) = Lf 1 2 =
c2 (θ2 ) l2 (θ2 )c2 (θ2 )

where ci (θ i )(i = 1, 2) are the dynamic lengths of the cylinder position.

The dynamic ranges of KFLu and KFLf in a duration are shown in Figure 4.1.
Here, two fictitious proportional gains KFLu (θ 1 ) and KFLf (θ 2 ) are larger than
the other two gains, respectively. It means that the external load FLu is
influenced by the shoulder angle θ 1 more than θ 2 . It is similar to FLf .
So the structural uncertainty caused by FL can be described as follows:

FLu = KFLu (θ1 )y1
. (4.13)
FLf = KFLf (θ2 )y2

The nominal values K̄FLu , K̄FLf of KFLu and KFLf are 0 N/m, which means
the no-load motion. The maximum relative uncertainties pKFLu and pKFLf are
50 Nonlinear Control Techniques for EHAs in Robotics Engineering

1500
KF (θ1)
Lu
KF (θ2)
Lu
KF (θ1)
Lf
1000
KF (θ2)
Lf

500
KF (N/m)
L

–500

–1000
0 0.5 1 1.5 2
t (s)

FIGURE 4.1
Dynamic ranges of KFLu and KFLf in one motion duration.

1200 and 300 with the relative variations δ KFLu and δ KFLf . The two fictitious
proportional gains and can be represented as follows:


KFLu = K̄FLu (1 + pKFLu δKFLu ), −1 ≤ δKFLu ≤ 1
. (4.14)
KFLf = K̄FLf (1 + pKFLf δKFLf ), −1 ≤ δKFLf ≤ 1

Remark 4.2
From Equations 4.5, 4.9, and 4.10, the three parametric uncertainties Kq , V t ,
and b are adopted by the multiplication uncertain expressions. However, KFL
is adopted by the addition uncertain expression.

4.3 Robust Model Construction

The parameters Kq , V t , and b may be represented as an upper linear fractional
transformation (LFT) [14,39], that is, Kq = Fu (MKq , δ Kq ), V t = Fu (MVt , δ Vt ),
b = Fu (Mb , δ b ), where
Robust Control Method 51

0 K̄q
MKq = ,
pKq K̄q
−pVt 1/V̄t
MVt = , (4.15)
−pVt 1/V̄t
0 B̄e
MBe = .
pBe B̄e

The structural uncertainty parameters KFLu , KFLf can be represented as

KFLu = Fu (MKFLu , δ KFLu ), KFLf = Ff (MKF , δ KF ), where
Lf Lf

0 1
MKFLu = ,
pKFLu K̄FLu
 (4.16)
0 1
MKF = p K̄FLf .
Lf KF Lf

Taking the shoulder hydraulic actuator for example, the robust model with
parametric and structural uncertainties is shown in Figure 4.2. If some vec-
tors are defined as the state variable vector X = [x1 , x2 , x3 , x4 , x5 ]T , the control
inputs u(t) is the voltage of the servo valve, the vector of exogenous inputs
w(t) = [w1 , w2 , w3 , w4 , w5 , w6 ]T , the vector of measurements y(t), the reg-
ulated output vector z(t) = [z1 , z2 , z3 , z4 , z5 , z6 ]T , then the linear uncertain
state-space model is described as follows:

⎧
⎪
⎪ Ẋ(t) = AX(t) + B1 W(t) + B2 u(t)
⎨
Z(t) = C1 X(t) + D11 W(t) , (4.17)
⎪
⎪
⎩
Y(t) = C2 X(t) + Yd (t)

where

⎡ ⎤
0 1 0 0 0
⎢ ⎥
⎢ −ωsv 2 −2ζsv ωsv 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 1 0 ⎥
A =⎢
⎢
⎥,
⎥
⎢ 0 0 0 0 1 ⎥
⎢ ⎥
⎢ ⎥
⎣4K̄q βe Aa 4βe K̄FLu Kce 4βe A2a K̄F −4Kce βe ⎦
0 − − − b̄βe Kce − m Lu − mb̄
V̄t m1f V̄t m1f V̄t m1f V̄ m 1f V̄t 1f
t 1f

(4.18)
52 Nonlinear Control Techniques for EHAs in Robotics Engineering

δK w4
w1 FLu
δKq w6
z1 δV MK
t z6 FLu
1/m1f
xv 4β A MKq + −
e a
m1f − x⋅ 3 x⋅ 2 x⋅ 1 x1
−
MV
t ∫ ∫ ∫
− −
− δB
e w5
z5
MB 1/m1f
e

4Kcβe
4βe Aa2
m1f
w3
δB
z3 e
βe Kc
MB
e m1f
δK w2
FLu
z2
MK 4βe Kc
FLu m1f

FIGURE 4.2
Robust model with parametric and structural uncertainties.

⎡ ⎤
0 0 0 0 0 0
⎢ 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 ⎥
B1 = ⎢
⎢ 0
⎥,
⎥ (4.19)
⎢ 0 0 0 0 0 ⎥
⎣ pKq pKFLu pBe ⎦
− − −pKFLu −pBe −pvt
V̄t V̄t V̄t
 T
B2 = 0 Ksv ωsv
2 0 0 0 , (4.20)

⎡ ⎤
4K̄q βe Aa
⎢ m 0 0 0 0 ⎥
⎢ 1f ⎥
⎢ 4βe Kc ⎥
⎢ 0 0 0 0 ⎥
⎢ ⎥
⎢ m1f ⎥
⎢ ⎥
⎢ b̄βe Kc ⎥
⎢ 0 0 0 0 ⎥
⎢ m1f ⎥
C1 = ⎢
⎢ 1
⎥ , (4.21)
⎥
⎢ 0 0 0 0 ⎥
⎢ m1f ⎥
⎢ ⎥
⎢ B̄e ⎥
⎢ ⎥
⎢ 0 0 0 0 ⎥
⎢ m1f ⎥
⎢ ⎥
⎣ 4K̄q βe A 4βe K̄FLu Kc 4βe A2a B̄e βe Kc −4Kc βe ⎦
0 − − −
V̄t m1f V̄t m1f V̄t m1f V̄t m1f V̄t
Robust Control Method 53

eu
Wu

Uncertainty model

Δ
z
w
+
r + u Ghyd y ep
K Wp
− −
+

d −
Gd

FIGURE 4.3
Block diagram of the closed-loop system with robust performance requirements.

⎡ ⎤
0 0 0 0 0 0
⎢ 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 ⎥
⎢ ⎥
D11 =⎢
⎢ 0 0 0 0 0 0 ⎥,
⎥ (4.22)
⎢ 0 0 0 0 0 0 ⎥
⎢ ⎥
⎣ pKq pKFLu pBe ⎦
− − 0 0 −pvt
V̄t V̄t V̄t
 
C2 = 0 0 1 0 0 . (4.23)

Here, the influence of load stiffness is neglected, that is, K = 0. The

measurement noise Yd (t) can be described in frequency domain as follows:

1 + 0.12s
Yd (s) = Gd (s)N(s), Gd (s) = 0.006 × , (4.24)
1 + 0.001s

where N(s) is the noise input.

The block diagram of the closed-loop system with robust performance
requirements is shown in Figure 4.3. In this figure, the linear uncertain state-
space model is described as block Ghyd , the block  = diag(δ Kq , δ Vt , δ Be , δ KFLu ),
r is the demand displacement of the cylinder input, d is the measure-
ment noise input, and K is the robust controller. The weighting functions
W p and W u are used to reflect the relative significance of the performance
requirement.

Remark 4.3
The block diagram of the closed-loop system (Figure 4.3) involves two inputs
r and d, and two outputs ep and eu , where ep is the tracking error of the
54 Nonlinear Control Techniques for EHAs in Robotics Engineering

cylinder position and eu is the control variable. The objective of the robust
controller is to design K to guarantee the tracking error ep = W p (r − y)
achieving the required performance under the control saturation constraint
of the control variable eu and the existed external disturbance d.

4.4 Robust Controller Design

The goal of the controller design is to satisfy the robust stability and perfor-
mance in the case of the parametric and structural uncertainties. In addition,
the controller should reduce the output sensitivity due to the presence of
measurement noise.

4.4.1 Analysis of Open-Loop System

If the structural uncertainty model FL (4.13) is embedded in the linear EHS
model (4.1) and (4.2), then the open-loop system of the EHS model from the
control variable u to the output cylinder position y is given by
Kqi Ksv ωsv
2
Y(s) Aa s2 +2ζsv ωsv s+ωsv2
=    ,
u(s) Vt mt
s +
3 mt Kcei
+ bVt
s2 +
2(1+υ 2 )βe A2a A2a 2(1+υ 2 )βe A2a
 Vt KFLi
   Kcei KFLi

1 + bKcei2 +
Vt K
2 + s + KK2cei / 1 +
Aa 2(1+υ )βe Aa
2 22Aa (1+υ )βe
2 Aa A2a

(4.25)

where KFLi (i = u, f ) are two fictitious proportional gains defined in Equa-

tion 4.12.
By referring to the open-loop control system analysis in Section 3.1, Gopen
needs to be analyzed since the three performance indexes, that is, the mag-
nitude margin Am , the phase margin Pm , and the system bandwidth ωc , are
reflected in the open-loop system. The frequency response of the open-loop
system with uncertainties is shown in Figure 4.4. The magnitude is little dif-
ferent between perturbed parameter and nominal parameter. But the phase is
significantly different, which is mainly caused by the structural uncertainty
parameter KFLu . Owing to the presence of KFLu , the stable phase of the open-
loop system will lead or lag approximately 40° compared to the nominal
model in the middle frequency stage. So the important objective of robust
controller design is to compensate the phase margin of the system.
It should be noted that, in a general condition, the external load FL (t) is a
typical disturbance input of the EHS. The linear EHS model (4.1) and (4.2) is
difficult to simplify into Equation 4.25 due to the two system inputs u and FL .
Thus, according to the analysis of the dynamic ranges of KFLu and KFLf in a
Robust Control Method 55

100
Log magnitude (dB)

Nominal
Perturbed

10−2

10−4
0.1 1 6.28 10 100

0
Phase (degrees)

−50

−100

−150

−200
0.1 1 6.28 10 100
Frequency (rad/s)

FIGURE 4.4
Frequency response of the open-loop system with varying uncertainty parameters Kq , V t , b, and
KFLu .

duration in Section 4.2, the simplified linearized model (4.13) can be substi-
tuted into Equation 4.1, and the open-loop system of the EHS model (4.25) is
obtained.

4.4.2 Weight Function Design

The weighting functions W p and W u are iteratively designed according to
robust performance requirement and control constraint. If W p is larger, the
dynamic tracking performance of EHS is required to be higher. But the robust
controller is more difficult to find a feasible solution in this case. Otherwise,
the robust controller is easily designed. If W u is larger, the control constraint
is rigorous, which means the control variable u is bounded in sufficiently
small value. In this case, the robust controller is also difficult to design. The
general form of W p is designed as follows [15]:

s/Ms + ωc
Wp = . (4.26)
s + ωc Ae

Here, Ae is the maximum steady accuracy requirement, which is selected as

0.01. It means the relative steady error between system output and demand
56 Nonlinear Control Techniques for EHAs in Robotics Engineering

input is less than 1% in the lower-frequency domain. Ms is the peak of

sensitivity, which is selected as 2. ωc is the prospective bandwidth of the
closed loop, which is selected as 2π according to the bandwidth range of the
demand position input. The weighting function W u is selected as 6 × 10 − 4 ,
which can satisfy the maximum control voltage ± 10 V of the servo valve in
the following iteration.

Theorem 4.1 ([67] Small Gain Theorem)

Consider the general feedback system in Figure 4.5, which includes two
cross-linked subsystems H1 and H2 . If there exist four constants γ 1 , γ 2 , c1 ,
and c2 such that

H1 e1  ≤ γ1 e1  + c1
, (4.27)
H2 e2  ≤ γ2 e2  + c2

where ∀e1 , e2 ∈ Lp , and γ 1 γ 2 < 1, then there exist two sufficiently large
constants γ and c such that
   
 e1   u1 
   
 e2  ≤ γ  u2  + c, (4.28)

where two inputs u1 , u2 ∈ Lp .

Proof. According to Figure 4.5, we can see that

y1 = H1 e1 , e1 = u1 − y2 ,
(4.29)
y2 = H2 e2 , e2 = u2 + y1 .

u1 + e1
H1
– y1

+
y2 u2
e2
H2
+

FIGURE 4.5
Cross-linked feedback system.
Robust Control Method 57

Thus, if two errors e1 and e2 are defined as the system outputs, and u1 , u2
are system inputs, then the following condition is satisfied:

e1 = u1 − H2 e2
. (4.30)
e2 = u2 + H1 e1

So, from triangular inequality rule, the following property holds:


H1 e1  = u2 − e2  ≥ e2  − u2 
. (4.31)
H2 e2  = u1 − e1  ≥ e1  − u1 

Substituting Equation 4.31 into Equation 4.27, we can see that


e2  ≤ γ1 e1  + u2  + c1
. (4.32)
e1  ≤ γ2 e2  + u1  + c2

To address Equation 4.32, if the two sides of two equalities are multiplied by
γ 1 , γ 2 , respectively, and are incorporated together, we can obtain
⎧ 1
⎪
⎨ e1  ≤ 1 − γ γ (u1  + γ2 u2  + c2 + γ2 c1 )
⎪
1 2
. (4.33)
⎪
⎪ 1
⎩ e2  ≤ (u2  + γ1 u1  + c1 + γ1 c2 )
1 − γ1 γ2

If two constants γ and c are defined as follows:

⎡ ⎤ ⎡ ⎤
 1 γ2   c 2 + γ2 c 1 
   
⎢ 1 − γ1 γ2 ⎥
1 − γ1 γ2 ⎥ ⎢ 1 − γ1 γ2 ⎥
γ = ⎢
⎣ γ 1

⎦ ,

c = ⎣⎢ ⎥ 
⎦ , (4.34)
 1   c 1 + γ1 c c 
   1−γ γ 
1 − γ1 γ2 1 − γ1 γ2 1 2

then Equation 4.28 is satisfied. 

Remark 4.4
Theorem 5.1 denotes that if the feedforward subsystem and feedback
subsystem H1 and H2 are L stable, and the gain γ 1 , γ 2 are bounded by
γ 1 γ 2 < 1, then the closed-loop system is a bounded-input-bounded-output
stable system.

According to Theorem 5.1 and  Remark

 4.4, if the nominal closed-loop sys-
tem of EHS Gnc is satisfied as Gnc  < 1, then the closed-loop system may be
58 Nonlinear Control Techniques for EHAs in Robotics Engineering

101

100
Magnitude

10−1

μ(W −1
p )
10−2
μ(G cn)

10−3
10−4 10−3 10−2 10−1 100 101 102
Frequency (rad/s)

FIGURE 4.6    
Singular values of the inverse function μ−1 Wp and the nominal closed-loop system μ Gnc .

stable. Furthermore, considering the weight performance function W p , if the

following rule holds, then the closed-loop system is stable:
     
Wp Gn  ≤ μ Wp μ Gnc < 1, (4.35)
c ∞

where μ(·) represents the singular values of the subsystem.

For convenient expression, Equation 4.35 is rewritten as follows:
   
μ Gnc < μ−1 Wp . (4.36)
 
Then the singular values of the inverse function μ−1 Wp and the nominal
 n
closed-loop system μ Gc are shown in Figure 4.6. As shown in Figure 4.6,
   
the dynamic variety μ Gnc is lower than μ−1 Wp . So Equation 4.35 is
satisfied.

4.4.3 Robust Controller Design

The robust controller will solve two problems: robust stability and robust
performance requirement [4,10], which can be described in Figure 4.7. The
left figure shows the robust stability of the controller K under the parametric
uncertainty . In this case, u and w are two inputs and the required perfor-
mance z, the position response y, and the control performance eu are three
Robust Control Method 59

Gic = Fu(Ghyd, Δ)
Δ
z Gc
w
u Ghyd y Gic
d ep
Gd Wp

K K

eu
Wu

FIGURE 4.7
Block diagram for the description of robust stability and robust performance.

outputs. After the robust stability is discussed, the robust performance of the
controller K is analyzed in the right figure. In this case, d is the input and the
dynamic tracking performance of position ep is the output. The robust sta-
bility of the designed controller is equivalent to the H∞ suboptimal control
problem as follows.  
The subsystem Gic = Fu Ghyd ,  with the parametric uncertainty is con-
structed as a cross-link together with the controller K. According to Theo-
rem 5.1, a stabilizing controller K needs to be found such that the H∞ norm
of the closed-loop transfer function is less than a given positive number γ ,
that is,
    
Fl (Gic , K)∞ = Fl Fu Ghyd ,  , K ∞ < γ , (4.37)

where γ > γ0 := minK Fl (Gic , K)∞ .

stabilizing
The robust performance requirement represents two performances, that is,
the tracking error of position response ep and the control variable eu . Both of
them are constrained by the weight functions W p and W u . Thus, if the sen-
sitivity and complementary sensitivity functions of the closed-loop feedback
control system satisfies the following constraint:

 
 Wp (I + Gic K)−1 Gd 
 
  < 1, (4.38)
 Wu K(I + G K)−1 
ic ∞

then the required robust performance is achieved.

The robustness analysis for different uncertainty parameters is given in
Figures 4.8 and 4.9. The maximum robust stability bound with varying uncer-
tainty parameters Kq , V t , b, and KFLu from the relative uncertainties  = − I
to  = I. Figure 4.8 shows the upper bound and lower bound of the singular
60 Nonlinear Control Techniques for EHAs in Robotics Engineering

100
Upper bound
Low bound

10−1
μ(GicK)

10−2

10−3
10−3 10−2 10−1 100 101 102
Frequency (rad/s)

FIGURE 4.8
Maximum robust stability bound with varying uncertainty parameters Kq , V t , b, and KFLu .

100 Nominal
Upper bound
Low bound

10−0.1
μ

10−0.2

10−1 100 101 102

Frequency (rad/s)

FIGURE 4.9
Maximum robust performance bound with varying uncertainty parameters Kq , V t , b, and KFLu .
Robust Control Method 61

value of Gic K, where the parametric uncertainty  is involved in Gic . From

this figure, μ(Gic K) < 1, which denotes that the feedback control system has
the robust stability. No matter how the uncertain parameters Kq , V t , b, and
KFLu changes, the closed-loop system is stable as shown in Figure 4.7. Differ-
ent from Figure 4.8, Figure 4.9 gives the singular values between μ(Gsen Gd )
and μ(1 − Gsen ), where Gsen is the sensitivity function of the closed-loop
system and 1 − Gsen is the complementary sensitivity function.
In Figure 4.8, the singular value of the closed loop perturbed by uncer-
tainty parameters is 0.37, which is less than 1. It shows that the closed-loop
system Gc is stable and its relative uncertainties  can be extended to 1/0.37.
However, in Figure 4.9, the maximum singular value of the upper bound
for μ(Gsen Gd ) or μ(1 − Gsen ) shows that the robust performance require-
ment is 1.06. It is a little more than 1 in the low frequency stage. It means
the steady trace accuracy may not be excellent. But in the middle fre-
quency stage 4–20 rad/s, the maximum singular value is less than 1. It
means that the dynamic trace performance and robustness margin meet the
requirement [46].
Then the frequency domain result of the designed robust controller is
shown in Figure 4.10. After iterative design, the robust controller is seven
orders. Taking the realistic project into account, the controller should be
reduced into lower order by the balanced truncation method [57]. From
Figure 4.10, we can see that reduced order controller is very similar to the

102
Log |u|

7th orders u
4th orders u

100
10−4 10−2 100 102 104
Frequency (rad/s)
100
7th orders u
Phase (degrees)

4th orders u

−100
10−4 10−2 100 102 104
Frequency (rad/s)

FIGURE 4.10
Frequency domain result of the designed robust controller.
62 Nonlinear Control Techniques for EHAs in Robotics Engineering

0.08

0.06

0.04
y (m)

0.02

0
Response
Demand
−0.02
0 2 4 6 8 10 12
Time (s)

FIGURE 4.11
Cylinder position response in time domain by the designed robust controller.

original seventh-order controller. Thus, the fourth-order control is reason-

able. At last, the controller is simplified to four orders as follows:

n4 s4 + n3 s3 + n2 s2 + n1 s + n0  
Ki (s) = yid − yi , i = 1, 2, (4.39)
d4 s + d3 s + d2 s + d1 s + d0
4 3 2

where n4 = 0.053, n3 = 1.049, n2 = 155.8, n1 = 5376.2, n0 = 7.6, d4 = 1,

d3 = 146.2, d2 = 1938.3, d1 = 6198.7, and d0 = 77.4, yid is the demand input,
and yi is the measurement output, which can be computed by triangle
geometry for the upper arm and forearm hydraulic actuators.
After the robust controller is obtained, the cylinder position response in
time domain is given by Figure 4.11. The step demand is a pulse input, where
the magnitude is ± 79 mm and the period is 4 s. The response time of the
cylinder position is 0.4 s and the steady-state error is less than 1 mm. This
denotes that the robust controller can guarantee the output tracking accuracy.
The control voltage of the servo valve is shown in Figure 4.12. Because the
control saturation is ± 10 V, the robust controller is designed to guarantee the
dynamic tracking accuracy of the cylinder position under the control input
constraint by the robust performance requirement (4.38).

4.4.4 Simulation and Experimental Result

The nonlinear simulation of Equations 4.1, 4.2, and 4.39 can be implemented
by MATLAB/Hydraulic and Simechanics modules. The maximum stroke of
the cylinder is 79 mm. So the range of the shoulder joint angle is from −60°
to 20° and the range of the elbow joint angle is from 40° to 130°. Furthermore,
Robust Control Method 63

15

10

5
u (V)

−5

−10
0 2 4 6 8 10 12
Time (s)

FIGURE 4.12
Designed controller result in time domain.

the relative variations δ Vt , δ b , δ KFLu , and δ KFLf are all selected as 1, which means
the maximum uncertainties. Then, two simulation results about the square
response and the sinusoidal response of the maximum stroke are shown in
Figures 4.13 through 4.16.
Shoulder angle (degrees)

20 Demand
0 Simulation
−20 Experiment
−40
−60
−80
0 5 10 15 20 25
Elbow angle (degrees)

120
100
80
60
40
0 5 10 15 20 25
t (s)

FIGURE 4.13
Square response of two joint angles in simulation and experiment.
64 Nonlinear Control Techniques for EHAs in Robotics Engineering

10
Shoulder control (V)

−5 Simulation
Experiment
−10
0 5 10 15 20 25

10
Elbow control (V)

−5

−10
0 5 10 15 20 25
t (s)

FIGURE 4.14
Control voltage of two servo valves in simulation and experiment for square demand.

The two square frequencies are 0.1 and 0.1 Hz and the sinusoidal frequen-
cies are 0.3 and 0.5 Hz, respectively. Since the load on the shoulder joint
is heavier than the elbow joint, the sinusoidal frequency for the shoulder
motion is not allowed too high. The steady errors of two joint angles are less
than 2, 0.5, respectively, shown in Figure 4.13, which means that the relative
position errors are less than 5%. Owing to the small dynamic mechanical load
on the elbow arm, the tracking accuracy is higher than the shoulder arm. The
control voltages of two servo valves are no more than the control saturation
± 10 V as shown in Figure 4.14.
Similarly for the sinusoidal response, the maximum dynamic errors of two
joint angles are less than 3°, 3°, respectively, shown in Figure 4.15. These
steady errors are no more than the relative error 5% of demand input. The
control voltages are also no more than ± 10 V as shown in Figure 4.16. So the
two robust controllers of two hydraulic actuators can be validated effectively
in simulation.
In this experiment, the supply pressure ps is 40 bar, and the motor revo-
lution is fixed at 1000 rpm. The joint angle is measured by an encoder and
the control voltage is measured by a serial port terminal of the servo valve.
The experiment results are very close to the corresponding simulation results
as shown in Figures 4.13 through 4.16. The steady errors of two joint angles
are less than 2°, 2°, respectively, in square response results. The maximum
Robust Control Method 65

Shoulder angle (degrees)

40
Demand
20 Simulation
0 Experiment
–20
–40
–60
–80
0 1 1.65 2 3 3.3 4 5 6 7 8 9 10

140
Elbow angle (degrees)

120
100
80
60
40
0 1 2 3 4 5 6 7 8 9 10
t (s)

FIGURE 4.15
Sinusoidal response of two joint angles in simulation and experiment.

dynamic errors of two joint angles are less than 4°, 4°, respectively, in sinu-
soidal response results, which are no more than the relative error 5% of
demand input. So the two robust controllers can guarantee the tracking posi-
tion accuracy and dynamic response performance in the case of the robust
stability. Furthermore, the control voltages of two servo valves are also no

2
Shoulder control (V)

–2
–3 Simulation
Experiment
–4
0 1 2 3 4 5 6 7 8 9 10

5
4
Elbow control (V)

–5
0 1 2 3 4 5 6 7 8 9 10
t (s)

FIGURE 4.16
Control voltage of two servo valves in simulation and experiment for sinusoidal demand.
66 Nonlinear Control Techniques for EHAs in Robotics Engineering

Shoulder angle (degrees)

20
–45
0 –50
–55 20
–20 –60 18
–65
–40 16 Demand
14 PI control
–60 10 Robust control
0 5 10 15 20 25
Elbow angle (degrees)

120
100
80
40
60 5
40
0 5 10 15 20 25
t (s)

FIGURE 4.17
Square response of the experiment result by the two control methods.

more than the control saturation, which is reflected in the robust performance
requirement eu as shown in Equation 4.38.
The traditional PI control methods can also be used in this EHS. The experi-
ment result comparison for the two control methods is shown in Figures 4.17
and 4.18. In Figure 4.17, although the step response by PI control is faster
than by robust control, the transient chatter emerges in two responses of the
shoulder, especially in the retraction process of the shoulder actuator. This
chatter phenomenon is the main cause by the different hydraulic parameters,
rapid changes of load forces between actuator extension and retraction. It
shows that the robust H∞ control method can eliminate the transient chatter
Shoulder angle (degrees)

20 –40 Demand
PI control
0 –50 Robust control
–20 –60
5
–40
–60
0 1 2 3 4 5 6 7 8 9 10

140
Elbow angle (degrees)

120
100
80 130
120
60
110
40 5
0 1 2 3 4 5 6 7 8 9 10

t (s)

FIGURE 4.18
Sinusoidal response of the experiment result by the two control methods.
Robust Control Method 67

(a) (b)

(c) (d)

(e) (f )

FIGURE 4.19
Snapshots of the sinusoidal experiment process. (a) Two cylinders retracted entirely (t = 0 s). (b)
The elbow cylinder extended to the maximum stroke (t = 1 s). (c) The shoulder cylinder extended
to the maximum stroke (t = 1.65 s). (d) The elbow cylinder retracted entirely (t = 2 s). (e) The elbow
cylinder extended to the maximum stroke once again (t = 3 s). (f) The shoulder cylinder extended
to the maximum stroke once again (t = 3.3 s).
68 Nonlinear Control Techniques for EHAs in Robotics Engineering

problem caused by the structural disturbances and parametric uncertainties.

The maximum steady error is 5° of the shoulder step response by PI control,
which is a little larger than by robust control. The steady error of the elbow
step response is 2°, less than that of the shoulder step because of the smaller
load forces. In Figure 4.18, the hysteresis of sinusoidal response by PI control
is more clearer than by robust control. It shows that the dynamic tracking
performance is obviously improved by robust control method.
The video sequences of sinusoidal experiment process are as shown in
Figure 4.19. The two-DOF robotic arm is controlled steady by two robust
controllers with 1 kg disk load. Double cylinders are all retracted at ini-
tial time t = 0 s. Approximately 1 s later, the elbow cylinder is extended to
reach its maximum stroke and the shoulder cylinder is extending gradually.
Then the elbow cylinder has turned to retract and the shoulder cylinder is
extended to reach its maximum stroke at t = 1.65 s. At t = 2 s, the elbow
cylinder is retracted entirely and the shoulder cylinder turns to retract grad-
ually. Then the elbow cylinder is extended to reach its maximum stroke once
again at t = 3 s, which means the duration of the elbow joint motion has
ended. But the shoulder cylinder is retracting gradually at this time. After
3.3 s, the shoulder cylinder is retracted entirely, which means the duration of
the shoulder joint motion has ended.
5
Output Feedback Control Method

CONTENTS
5.1 Output Feedback Control Model of EHS . . . . . . . . . . . . . . . . . . . . . 69
5.2 State Observer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.1 Full-State Observer Construction . . . . . . . . . . . . . . . . . . . . . . 72
5.2.2 Observer Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.3 Nonlinear Backstepping Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3.1 Backstepping Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3.2 Controller Design with Observer . . . . . . . . . . . . . . . . . . . . . . 84
5.3.3 Stability Discussion of EHS . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.4.1 Result of the Proposed Method . . . . . . . . . . . . . . . . . . . . . . . 88
5.4.2 Compared Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

In this chapter, an output feedback controller with a full-state observer is

presented for a single-rod hydraulic actuator [19]. The system states are not
measured except the cylinder displacement. Although this system is not a
strict feedback system, the proposed controller theoretically guarantees the
stability and convergence of EHS. In addition, the stable full-state observer
is designed to estimate some unknown hydraulic states, which can be used
in the backstepping controller. To verify this proposed backstepping control
based on the full-state observer, an experiment is carried out. Experiment
results are obtained from the hydraulic position control of the two-DOF
robotic arm motion. Experiment results highlight the better dynamic perfor-
mance of the proposed method in comparison to the conventional PI control
method in some specific critical condition with high response frequency and
large unknown external load.

5.1 Output Feedback Control Model of EHS

The electro-hydraulic system shown in Figure 2.1b comprises a servo valve, a
single-rod cylinder, a fixed-displacement pump, a relief valve, and a variable

69
70 Nonlinear Control Techniques for EHAs in Robotics Engineering

load. The variable load Pl drives the two-DOF robotic arm manufactured by
Italian Institute of Technology, which is referred as the Robotic BigDog. In this
section, the nonlinear dynamic model of the single-rod hydraulic actuator
will be discussed.
The dynamics of the servo valve is adopted by a second-order linear model
mentioned in Equation 2.9 as follows:

2 2
ẍv + 2ζsv ωsv ẋv + ωsv xv = Ksv ωsv u, (5.1)

where xv is the spool position of the servo valve, ζ sv is the damping ratio, ωsv
is the natural frequency, and Ksv is the gain of control voltage u.
According to Equation 2.6, the flow equations of the single-rod cylinder
can be described as follows:
 
Cd wxv 2(ps − pa )/ρ xv ≥ 0
Qa =  ,
Cd wxv 2(pa − pr )/ρ xv < 0
 
Cd wxv 2(pb − pr )/ρ xv ≥ 0
Qb =  (5.2)
Cd wxv 2(ps − pb )/ρ xv < 0

where Qa is the main load flow as xv ≥ 0, Qb is the main load flow as xv < 0,
pa and pb are the pressure inside the two chambers of the cylinder, ps is the
supply pressure of the pump, Cd is the discharge coefficient, w is the area
gradient of the servo valve spool, and ρ is the density of the hydraulic oil.
The two load flows Qa and Qb are handled like the following uniform form:
⎧ 
⎪
⎪ 1 + sgn(xv ) 2
⎪
⎪ Q = Cd wxv (ps − pa )
⎪
⎪ a
ρ
⎪
⎪ 2
⎪
⎪
⎪
⎪ 
⎪
⎪ 1 − sgn(xv ) 2
⎪
⎪ + (pa − pr )
⎪
⎪
Cd wxv
⎨ 2 ρ
, (5.3)
⎪
⎪ 
⎪
⎪ 1 + sgn(xv ) 2
⎪
⎪ Q = Cd wxv (pb − pr )
⎪
⎪ b
ρ
⎪
⎪ 2
⎪
⎪
⎪
⎪ 
⎪
⎪ 1 − sgn(xv ) 2
⎪
⎪ + (ps − pb )
⎩ 2
Cd wxv
ρ

where the sgn(.) function is defined as

⎧
⎨−1 x>0
sgn(x) = 0 x=0 . (5.4)
⎩
1 x<0
Output Feedback Control Method 71

In Equation 5.3, the function sgn(xv ) should be smoothed in the deriva-

tion of the backstepping control. So sgn(xv ) can be replaced with tanh(kxv ) as
follows [16]:

ekxv − e−kxv
sgn(xv ) ≈ tanh(kxv ) = , k  0. (5.5)
ekxv + e−kxv

According to Equation 2.3, the flow-pressure continuous model of the

hydraulic cylinder is shown as follows:

Aa ẏ + Ctl (pa − pb ) + (V0a + Aa y)ṗa /βe = Qa
, (5.6)
Ab ẏ + Ctl (pa − pb ) − (V0b − Ab y)ṗb /βe = Qb

where y is the displacement of the piston, Ctl is the coefficient of the total
leakage of the cylinder, β e is the effective bulk modulus, Aa and Ab are the
ram areas of the two chambers, and V 0a and V 0b are the initial total control
volumes of the two cylinder chambers, respectively.
From Equation 2.2, the mechanical dynamic equation is shown as follows:

mÿ = pL Ap − Ky − bẏ − FL , (5.7)

where m is the load mass, K is the load spring constant, b is the viscous damp-
ing coefficient, Ff is the stick-slip friction caused by the viscous damp of oil,
and FL is the external load on the hydraulic actuator from the mechanical
structure of the two-DOF robotic arm.
The stick-slip friction often includes two items: a stick phase Ffstatic occurs
when the velocity is within a small critical velocity range and a slip fric-
tion Ffslip , which is the same as the Coulomb friction model [30]. Here, Ff
is defined as

Ff (y, ẏ) = Ffstatic + b(1 + δqb )ẏ, −1 ≤ δ ≤ 1, (5.8)

where b is the viscous damping coefficient of oil, qb is the maximum relative

uncertainties of b, and δ is the relative variation.
According to Remark 2.2, the H∞ norm of FL is bounded by a constant FL
as follows:

FL (y, ẏ, ÿ) ∞

< FL . (5.9)

If these vectors are defined: the state vector X = [x1 , x2 , x3 , x4 , x5 , x6 ]T =

[y, ẏ, pa , pb , xv , ẋv ]T , the control inputs u(t), which is the voltage of the servo
valve, the displacement of piston y(t) indirectly measured by the encoder, the
72 Nonlinear Control Techniques for EHAs in Robotics Engineering

sixth-order nonlinear dynamic model of EHS is constructed as follows:

⎧
⎪ ẋ1 = x2
⎪
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ ẋ2 = (x3 Aa − x4 Ab − Kx1 − Ffstatic − b(1 + qb )x2 − FL )
⎪
⎪ m
⎪
⎪
⎪
⎪ βe Aa x2 βe Ctl
⎪
⎪ ẋ3 = − − (x3 − x4 )
⎪
⎪ +
⎪
⎪ V 0a A x
a 1 V 0a + Aa x1
⎪
⎪ ⎛  ⎞
⎪
⎪ 1 + tanh(kx5 ) 2
⎪
⎪ (p − x )
⎪
⎪ βe Cd wx5 ⎜ ρ
s 3 ⎟
⎪
⎪ ⎜ 2 ⎟
⎪
⎪ + ⎜   ⎟
⎪
⎪ V + A x ⎝ 1 − tanh(kx ) 2 ⎠
⎪
⎨
0a a 1
+
5
(x3 − pr )
2 ρ . (5.10)
⎪
⎪ β β
⎪
⎪ A
e b 2x C
e tl
⎪
⎪ ẋ = − (x3 − x4 )
⎪ 4
⎪ V0b − Ab x1 V0b − Ab x1
⎪
⎪ ⎛  ⎞
⎪
⎪
⎪
⎪ 1 + tanh(kx5 ) 2
⎪
⎪ (x4 − pr ) ⎟
⎪
⎪
⎪ βe Cd wx5 ⎜ ⎜ 2 ρ ⎟
⎪
⎪ − ⎜   ⎟
⎪
⎪ V − A x ⎝ 1 − tanh(kx ) 2 ⎠
⎪
⎪
0b b 1
+
5
(ps − x4 )
⎪
⎪ ρ
⎪
⎪ 2
⎪
⎪
⎪
⎪ ẋ5 = x6
⎪
⎪
⎪
⎩ 2 2
ẋ6 = −ωsv x5 − 2ζsv ωsv x6 + Ksv ωsv u

5.2 State Observer Design

Owing to the unknown states except x1 , a high-gain state observer is
designed to estimate the full state, which can be used in the design of
the backstepping control. In this section, different from the previous state
observer in Reference 16, a stable observer is constructed by more processing
steps. Then the global asymptotic convergence for the full-state observer is
proved.

5.2.1 Full-State Observer Construction

According to Equation 5.10, the nonlinear model of the EHS can be rewrit-
ten as

Ẋ = AX + φ(X) + Bu,
(5.11)
y = CX,
Output Feedback Control Method 73

where

⎡ ⎤
0 1 0 0 0 0
⎢ ⎥
⎢ Aa Ab ⎥
⎢0 0 − 0 0 ⎥
⎢ m m ⎥
⎢ ⎥
⎢ ⎥
⎢ βe Ctl βe Ctl ⎥
⎢0 0 − − 0 0 ⎥
⎢ V0a + Aa x1 V0a + Aa x1 ⎥
⎢ ⎥
A=⎢ ⎥, (5.12)
⎢ ⎥
⎢ βe Ctl βe Ctl ⎥
⎢0 0 − − 0 0 ⎥
⎢ V0b − Ab x1 V0b − Ab x1 ⎥
⎢ ⎥
⎢ ⎥
⎢0 0 0 0 0 1 ⎥
⎢ ⎥
⎣ ⎦
0 0 0 0 −ωv2 −2ζv ωv

 T
B= 0 0 0 0 0 Ksv ωv2 , (5.13)

 
C= 1 0 0 0 0 0 , (5.14)

⎡ ⎤
0
⎢ ⎥
⎢ 1 ⎥
⎢ (−Kx1 − Ffstatic − bx2 − bpb x2 − FL ) ⎥
⎢ m ⎥
⎢ ⎥
⎢ ⎛   ⎞ ⎥
⎢ 1 + tanh(kx5 ) 2 ⎥
⎢ (ps − x3 )+⎟ ⎥
⎢ ⎜ 2 ρ ⎥
⎢ βe Cd wx5 ⎜ ⎟ β A x
e a 2 ⎥
⎢ ⎜ ⎟ − ⎥
⎢ V0a + Aa x1 ⎝ 1 − tanh(kx )  2 ⎠ V0a + Aa x1 ⎥
⎢ 5 ⎥
⎢ (x3 − pr ) ⎥
φ(X) = ⎢
⎢
2 ρ ⎥.
⎥
⎢ ⎛   ⎞ ⎥
⎢ 1 + tanh(kx5 ) 2 ⎥
⎢ (x4 − pr )+⎟ ⎥
⎢ ⎥
⎢ βe Cd wx5 ⎜ ⎜ 2 ρ ⎟ βe Ab x2 ⎥
⎢− ⎟+ ⎥
⎢ V0b − Ab x1 ⎜⎝ 1 − tanh(kx )  2 ⎠ V0b − Ab x1 ⎥
⎢ 5 ⎥
⎢ (ps − x4 ) ⎥
⎢ 2 ρ ⎥
⎢ ⎥
⎢ ⎥
⎣ 0 ⎦
0
(5.15)
74 Nonlinear Control Techniques for EHAs in Robotics Engineering

Equivalent transformation is used in the third and fourth columns of

matrix A; the new matrix A0 can be defined as
⎡ ⎤
0 1 0 0 0 0
⎢ ⎥
⎢ K Aa Ab Aa ⎥
⎢− 0 − − 0 0 ⎥
⎢ m ⎥
⎢ m m m ⎥
⎢ ⎥
⎢ βe Ctl ⎥
⎢ 0 0 − 0 0 0 ⎥
3×(−1)+4 ⎢ V0a + Aa x1 ⎥
A0 = ⎢ ⎥,
⎢ ⎥
⎢ βe Ctl ⎥
⎢ 0 0 − 0 0 0 ⎥
⎢ V0b − Ab x1 ⎥
⎢ ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 1 ⎥
⎣ ⎦
0 0 0 0 −ωv2 −2ζv ωv
(5.16)

Since the leakage coefficient Ctl has a small value and hence little impact,
two items A0 (3, 3) and A0 (4, 3) are neglected in this chapter. Hence, A0 (3, 3) =
A0 (4, 3) = 0. To ensure the validation and convergence of the observer, two
items A0 (3, 5) and A0 (4, 5) are replaced by other nonzero items as follows:

βe 2
A0 (3, 5) = Cd w ps , (5.17)
V0a + Aa x1 ρ

βe 2λ
A0 (4, 5) = − Cd w ps , (5.18)
V0b − Ab x1 ρ

where 0 < λ < 1 is a constant.

Then, two elements A0 (6, 5) and A0 (6, 6) are moved into φ(X), which means
A0 (6, 5) = A0 (6, 6) = 0. So the new matrix A0 and φ(X) are defined as

⎡ ⎤
0 1 0 0 0 0
⎢ ⎥
⎢0 0 a23 −a23 − a24 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢0 0 ⎥
⎢ 0 0 0 a35 ⎥
A1 = ⎢
⎢
⎥,
⎥ (5.19)
⎢0 0 0 0 −a45 0 ⎥
⎢ ⎥
⎢ ⎥
⎢0 0 0 0 0 1 ⎥
⎣ ⎦
0 0 0 0 0 0
Output Feedback Control Method 75

⎡ ⎤
0
⎢ ⎥
⎢ 1 ⎥
⎢ (−Kx1 − Ffstatic − bx2 − bpb x2 − FL ) ⎥
⎢ m ⎥
⎢ ⎛  ⎞ ⎥
⎢ 1 + tanh(kx ) 2 ⎥
⎢ 5
(ps − x3 ) ⎥
⎢ ⎜ ρ ⎟ ⎥
⎢ βe Cd wx5 ⎜⎜ 
2 ⎟ ⎥
⎟ ⎥
⎢ 
⎢ V0a + Aa x1 ⎜ 2 ⎟ ⎥
⎢ ⎝+ 1 − tanh(kx5 ) 2
(x3 − pr ) − Cd w ps ⎠ ⎥
⎢ ρ ρ ⎥
⎢ 2 ⎥
⎢ ⎥
⎢ βe Aa x2 ⎥
⎢ − ⎥
φ1 (X) = ⎢
⎢ V + A x ⎥,
⎞⎥
0a a 1
⎢ ⎛  ⎥
⎢ 1 + tanh(kx5 ) 2 ⎥
⎢ ⎜ (x4 − pr ) ⎟ ⎥
⎢ βe Cd wx5 ⎜ 2 ρ ⎟⎥
⎢ ⎜   ⎟ ⎥
⎢− ⎟⎥
⎢ V0b − Ab x1 ⎜
⎝+ 1 − tanh(kx ) 2 2λ ⎥
ps ⎠ ⎥
5
⎢ (ps − x4 ) − Cd w
⎢ 2 ρ ρ ⎥
⎢ ⎥
⎢ βe Ab x2 ⎥
⎢ ⎥
⎢ + ⎥
⎢ V0b − Ab x1 ⎥
⎢ ⎥
⎣ 0 ⎦
−ωv2 x5 − 2ζv ωv x6
(5.20)
where
Aa
a23 = , (5.21)
m

Ab
a24 = , (5.22)
m

βe 2
a35 = Cd w ps , (5.23)
V0a + Aa x1 ρ


βe 2λ
a45 = Cd w ps . (5.24)
V0b − Ab x1 ρ

To estimate X, the high-gain observer is designed as follows:

˙
X̂ = A1 X̂ + φ1 (X̂) + Bu + (y − ŷ),
 T (5.25)
 = 1 2 3 4 5 6 , i > 0, i = 1, . . . , 6.

where  is the high gain of the observer, y is the measured value of

displacement, ŷ = x̂1 .
76 Nonlinear Control Techniques for EHAs in Robotics Engineering

5.2.2 Observer Convergence

If the designed observer is fast convergence, the next estimated state X̂(k +
1) can be computed by the current estimated state X̂(k) and current control
u(k) according to Equation 5.25. Now the stability and convergence of this
observer are analyzed. Since Equation 5.11 is subtracted by Equation 5.25,
the error state model of the observer is defined as

˙ = A X̃ + δ (X, X̃),
X̃ (5.26)
c φ

where

 T
X̃ = x̃1 x̃2 x̃3 x̃4 x̃5 x̃6
 T (5.27)
= x1 − x̂1 x2 − x̂2 x3 − x̂3 x4 − x̂4 x5 − x̂5 x6 − x̂6 ,

⎡ ⎤
− 1 1 0 0 0 0
⎢ ⎥
⎢− 0 a23 −a23 − a24 0 0⎥
⎢ 2 ⎥
⎢ ⎥
⎢− 3 0 0 0 a35 0⎥
Ac = A1 − LC = ⎢
⎢−
⎥, (5.28)
⎢ 4 0 0 0 −a45 0⎥
⎥
⎢ ⎥
⎢− 0 0 0 0 1⎥
⎣ 5 ⎦
− 6 0 0 0 0 0

δφ (X, X̃) = φ1 (X) − φ1 (X̂). (5.29)

The characteristic polynomial of error system matrix Ac is described as

det |sI − Ac | = s6 + 1s
5
+ 2s
4
+ [a23 3 + (a23 + a24 ) 4 ]s
3

2
+ [a23 a35 5 − (a23 + a24 )a45 5 ]s

+ [a23 a35 6 − (a23 + a24 )a45 6 ]s = 0. (5.30)

Theorem 5.1
If the high gain of the observer is satisfied to the following condition (5.31),
then the matrix Ac is Hurwitz after removing the only one eigenvalue 0. In
Output Feedback Control Method 77

other words, all the eigenvalues of Ac has a negative real part.

⎧
⎪ 1>0
⎪
⎪
⎪
⎪
⎪
⎪ 2 >0
⎪
⎪
⎪
⎪
⎪
⎪ a23 3 + (a23 + a24 ) 4 > 0
⎪
⎪
⎪
⎪
⎨ a23 a35 5 − (a23 + a24 )a45 5 > 0
.
⎪
⎪ a23 a35 6 − (a23 + a24 )a45 6 > 0
⎪
⎪
⎪
⎪
⎪
⎪ 1 2 − a23 3 − (a23 + a24 ) 4 > 0
⎪
⎪
⎪
⎪
⎪
⎪ 1 2 [a23 3 + (a23 + a24 ) 4 ] − [a23 + (a23 + a24 ) 4]
2
⎪
⎪
3
⎪
⎩
− 12 5 [a23 a35 − (a23 + a24 )a45 ] + 1 6 [a23 a35 − (a23 + a24 )a45 ] > 0
(5.31)

Proof. Equation 5.30 shows that one eigenvalue of the matrix Ac is 0 easily.
After removing the eigenvalue 0, Equation 5.30 can be simplified as

s5 + c4 s4 + c3 s3 + c2 s2 + c1 s + c0 = 0. (5.32)

According to Routh criterion, the necessary and sufficient conditions that all
characteristic roots have a negative real part must satisfy two properties as
follows:

I. All coefficients are greater than zero, which means ci > 0(i = 0, . . . ,4).
II. Arrange the coefficients ci in the following form:

s5 1 c3 c1
s4 c4 c2 c0
c 4 c 3 − c2 c 4 c 1 − c0
s3 0
c4 c4
c24 c1 − c4 c0
s2 c2 − c0 0
c 4 c 3 − c2 , (5.33)
c4 c 0 c 3 − c 2 c 0
c 4 c 1 − c0 c4
s1 − 0 0
c4 c c 1 − c4 c 0
2
c2 − 4
c 4 c 3 − c2
s0 c0 0 0

where all the first worksheet of Equation 5.33 should be greater

than 0.

According to property (II), if Equation 5.30 is substituted into the first

worksheet of Equation 5.33, then the high gains of the observer i (i = 1, . . . ,6)
are satisfied to five inequality conditions as shown in Equation 5.31. 
78 Nonlinear Control Techniques for EHAs in Robotics Engineering

According to the fourth and fifth inequalities of Equation 5.31, the follow-
ing property is satisfied:

a23 a35 > (a23 + a24 )a45 . (5.34)

When Equations 5.23 and 5.24 are substituted into Equation 5.34, the
constant parameter is satisfied as follows:

A2a
λ< . (5.35)
A2a + A2b

Now, the dynamic estimation error in Equation 5.26 is considered. If a small

positive constant ε is defined, then the new estimation error is described as
 T
η̃ = η̃1 η̃2 η̃3 η̃4 η̃5 η̃6
 T (5.36)
x̃ x̃2 x̃3 x̃4 x̃5
= 1 x̃6 .
ε4 ε3 ε2 ε2 ε

After η̃ is substituted into Equation 5.26, the new error state equation is
described as
εη̃˙ = Aη η̃ + δφ (η, η̃, ε), (5.37)

where
⎡ ⎤
− 1 /ε 1 0 0 0 0
⎢ ⎥
⎢− 2 /ε
2 0 a23 −a23 − a24 0 0⎥
⎢ ⎥
⎢ ⎥
⎢− 3 /ε
3 0 0 0 a35 0⎥
Aη = ⎢
⎢
⎥,
⎥ (5.38)
⎢− 4 /ε
3 0 0 0 −a45 0⎥
⎢ ⎥
⎢− 1⎥
5 /ε
4
⎣ 0 0 0 0 ⎦
− 6 /ε
5 0 0 0 0 0

⎡ ⎤
0
⎢ ⎥
⎢εδ2 (η2 , η̂2 , ε)⎥
⎢ ⎥
⎢ ⎥
⎢ δ3 (η5 , η̂5 , ε) ⎥
⎢
δφ (η, η̂, ε) = ⎢ ⎥. (5.39)
⎥
⎢ δ4 (η5 , η̂5 , ε) ⎥
⎢ ⎥
⎢ 0 ⎥
⎣ ⎦
εδ6 (η6 , η̂6 , ε)

Substituting i in Equation 5.28 with the first column of Equation 5.38,

respectively, the two characteristic polynomials of matrix Aη and Ac are
Output Feedback Control Method 79

identical, which means the properties of all characteristic roots for Aη is

equivalent to Ac . So, the matrix Aη is also Hurwitz after removing one eigen-
value 0 and the seven inequality conditions (5.31) are still available for the
new error state Equation 5.37.

Theorem 5.2
If φ 1 (X) is continuous and differentiable, and its first-order derivative is
bounded, Equation 5.39 satisfies the following condition:

δφ (η, η̃, ε) ≤ εμδ η̃ 2, (5.40)

where μδ is a positive constant.

Proof. According to Equation 5.20, φ 1 (X) is continuous and differentiable.

Because all the states described specified physical variables, the first-order
derivative of φ 1 (X) is bounded. According to Equation 5.36, the first-order
derivative of φ 1 (X) is still bounded, although its form derives from φ 1 (X).
Then Equation 5.39 can also be described as

⎡ ⎤
0
⎢ ⎥
⎢ εδ2 (η2 , η̂2 , ε) ⎥
⎢ ⎥
⎢ 2 ⎥
⎢ε δ3 (η5 , η̂5 , ε)⎥
⎢
δφ (η, η̂, ε) = φ1 (η) − φ1 (η̂) = ⎢ ⎥. (5.41)
⎥
⎢ε2 δ4 (η5 , η̂5 , ε)⎥
⎢ ⎥
⎢ ⎥
⎣ 0 ⎦
εδ6 (η6 , η̂6 , ε)

According to Lagrange mean value theorem, if δi (ηi , η̂i , ε)(i = 2, 3, 4, 6) is

continuous and differentiable, a point η0i exists such that

φ1i (ηi ) − φ1i (η̂i )

= δi (ηi0 ), ηi0 ∈ (ηi , η̂i ), (5.42)
ηi − η̂i

where δ i (η0i ) is the derivative at η0i .

If ηi and η̂i belong to Mi , which is a bounded neighborhood, η0i also belongs
to Mi and the following condition is satisfied:

φ1i (ηi ) − φ1i (η̂i )  

≤ sup δi (ηi0 ) . (5.43)
ηi − η̂i η0 ∈M
i i
80 Nonlinear Control Techniques for EHAs in Robotics Engineering

Thus, we can see that

⎡ ⎤
0 0
⎢ sup δ (η η̃2 ⎥
⎡ ⎤ ⎢ 0 2 2 ⎥
0 ⎢η2 ∈M2 ⎥
⎢  ⎥
⎢ εδ2 (η2 , η̂2 , ε) ⎥ ⎢ε sup δ (η0 η̃3 ⎥
⎢ 2 ⎥ ⎢ 3 3 ⎥
⎢ε δ3 (η5 , η̂5 , ε)⎥ ⎢η30 ∈M3 ⎥
δφ (η, η̃, ε) = ⎢ ⎥ =ε ⎢  ⎥
2 ⎢ε2 δ4 (η5 , η̂5 , ε)⎥ ⎢ε sup δ (η0 η̃4 ⎥
⎢ ⎥ ⎢ 4 4 ⎥
⎣ 0 ⎦ ⎢η0 ∈M4 ⎥
⎢4 ⎥
εδ6 (η6 , η̂6 , ε) ⎢ 0 ⎥
2 ⎢  ⎥
⎣ sup δ (η0 η̃6 ⎦
6 6
η60 ∈M6
2
⎡ ⎤ (5.44)
0
⎢ sup δ (η0 η̃2 ⎥
⎢ 0 2 2 ⎥
⎢η2 ∈M2 ⎥
⎢  ⎥
⎢ sup δ (η0 η̃3 ⎥
⎢ 3 3 ⎥
⎢η30 ∈M3 ⎥
<ε ⎢  ⎥ .
⎢ sup δ (η0 η̃4 ⎥
⎢ 4 4 ⎥
⎢η0 ∈M4 ⎥
⎢4 ⎥
⎢ 0 ⎥
⎢  ⎥
⎣ sup δ (η0 η̃6 ⎦
6 6
η60 ∈M6
2

If a positive constant is satisfied, the following condition:

⎛! !⎞
!  !!
!
μδ = max ⎝!! sup δi (ηi0 !!⎠ , (5.45)
!η0 ∈Mi !
i

and taking μδ into Equation 5.44, Equation 5.40 is satisfied. 

Theorem 5.3
In Equation 5.37, if the positive constant ε satisfies the following condition
(5.45), then the error vector η̃ converges to 0, which means η → η̂:

1
0<ε< , (5.46)
2λmax (Pη )μδ

where Pη is a positive definite matrix and λmax (Pη ) is the maximum eigen-
values of matrix Pη .

Proof. A Lyapunov candidate function is defined as

Vη = εη̃TPη η̃. (5.47)

Output Feedback Control Method 81

Since Aη is a Hurwitz matrix, a positive definite matrix Pη exists such that

 
I 0
ATη Pη + Pη Aη = − 5×5 . (5.48)
0 0

If Equation 5.37 is combined with Equations 5.47 and 5.48, then the
derivative of V η is

V̇η = −η̃T η̃ + 2η̃T Pη δφ (η, η̂, ε). (5.49)

So, V̇η satisfies the following inequality:

2
V̇η ≤ − η̃ 2 + 2η̃T Pη δφ (η, η̂, ε)
2 2
≤ − η̃ 2 + 2ελmax (Pη )μδ η̃ 2 (5.50)
2
≤ −(1 − 2ελmax (Pη )μδ ) η̃ 2.

If ε satisfies the condition shown in Equation 5.46, then V̇η < 0. Therefore,
the error vector η̃ converges to 0 and the system described by Equation 5.37
is exponentially stable. 

Since η → η̂, considering the relationship between η and X in Equa-

tion 5.36, we see that X → X̂.

5.3 Nonlinear Backstepping Control

The dynamic model of EHS in Equation 5.10 shows that the system order is
six. But the relative degree is only five. Therefore, this EHS is not a strict
feedback nonlinear system and then the recursive procedure of the back-
stepping control method cannot be used in Equation 5.10 directly. In this
section, the dynamic model is divided into two parts to design the backstep-
ping controller after a simple transformation of the pressure error from two
chambers.
82 Nonlinear Control Techniques for EHAs in Robotics Engineering

5.3.1 Backstepping Procedure

Equation 5.10 is considered to design the backstepping controller by recur-
sive procedure. At first, a pressure error is defined as

ep = p − pd ,
Aa x3 − Ab x4
p= ,
Aa (5.51)
Kx1d (t) Ffstatic FL
pd (t) = + + ,
Aa Aa Aa

where pd is the virtual command of the load pressure and p is the actual load
pressure, and x1d is the displacement command of the system input.
The error of x5 is defined as

e5 = x5 − x5d , (5.52)

and the virtual command x5d is satisfied as

1
x5d = − (k1 ep − ṗd + g1 ), (5.53)
g2

where k1 is a positive constant, the functions g1 and g2 are

βe Aa x2 βe Ctl βe A2b x2
g1 = − − (x3 − x4 ) −
V0a + Aa x1 V0a + Aa x1 Aa (V0b − Ab x1 )
βe Ab Ctl
+ (x3 − x4 ), (5.54)
Aa (V0b − Ab x1 )
⎛  ⎞
ekx5 2
⎜ ekx5 + e−kx5 Cd w ρ (ps − x3 ) ⎟
βe ⎜ ⎟
g2 = ⎜   ⎟
⎜
V0a + Aa x1 ⎝ ⎟
e−kx5 2 ⎠
+ kx Cd wx5 (x3 − pr )
e 5 + e−kx5 ρ
⎛  ⎞ (5.55)
ekx5 2
⎜ ekx5 + e−kx5 Cd w ρ (x4 − pr ) ⎟
βe Ab ⎜ ⎟
+ ⎜   ⎟.
⎜
Aa (V0b − Ab x1 ) ⎝ −kx ⎟
e 5 2 ⎠
+ kx −kx
Cd wx5 (p s − x4 )
e 5 +e 5 ρ

Then the Lyapunov candidate function V 3 is defined as

1 2
V3 = e . (5.56)
2 p
Output Feedback Control Method 83

Since the derivative of V 3 is


Aa ẋ3 − Ab ẋ4
V̇3 = ep ėp = ep − ṗd = ep (g2 x5 + g1 − ṗd ), (5.57)
Aa
we can see that
− k1 ep + g2 e5
= −k1 ep + g2 (x5 − x5d )

1 (5.58)
= −k1 ep + g2 x5 + (k1 ep − ṗd + g1 )
g2
= g2 x5 + g1 − ṗd .

After Equation 5.57 is substituted into Equation 5.56, the derivative of V 3

is described as
V̇3 = −k1 e2p + g2 ep e5 . (5.59)

The function V 3 is expanded and the Lyapunov candidate function V 4 is

defined as
1 2
V4 = V3 + e , (5.60)
2k2 5
where k2 is a positive constant.
Then the derivative of V 4 is
e5
V̇4 = V̇3 + ė5
k2
(5.61)
e5
= −k1 e2p + g2 ep e5 + (x6 − ẋ5d ).
k2
If the virtual command x6d is satisfied as

x6d = −k2 g2 ep + ẋ5d − e5 , (5.62)

then we see that

e25 e5 e 6
V̇4 = −k1 e2p − + , (5.63)
k2 k2
where the error of x6 is defined as

e6 = x6 − x6d . (5.64)

Similarly, V 4 is expanded and the Lyapunov candidate function V 5 is

defined as
1 2
V5 = V4 + e , (5.65)
2k3 6
where k3 is a positive constant.
84 Nonlinear Control Techniques for EHAs in Robotics Engineering

The derivative of V 5 is

e6
V̇5 = V̇4 + ė6
k3
e25 e5 e6 e6
= −k1 e2p − + + (ẋ6 − ẋ6d ) (5.66)
k2 k2 k3
e25 e5 e6 e6
= −k1 e2p − + 2
+ (−ωsv 2
x5 − 2ζv ωsv x6 + Ksv ωsv u − ẋ6d ).
k2 k2 k3

If the controller u is designed as


1 2 k3
u= ωsv x5 + 2ζsv ωsv x6 + ẋ6d − e 5 − e6 , (5.67)
Ksv ωsv
2 k2

then we see that

e25 e2
V̇5 = −k1 e2p − − 6 < 0. (5.68)
k2 k3

So far, the recursive procedure for the third-order equivalent model of EHS
ends. By the construction of three Lyapunov functions, the stability of the
variable states xi (i = 3, 4, 5, 6) is analyzed from the third equation to the sixth
equation in Equation 5.10.

5.3.2 Controller Design with Observer

Owing to unknown variable states x5 and x6 , the variables e5 , e6 , ẋ6d cannot
be computed and the controller u shown in Equation 5.57 cannot be used in
EHS directly. In this section, the backstepping controller is further processed
by state estimation x̂i (i = 1, . . . , 6).
At first, the first and second equations in Equation 5.10 can be described as
the second-order linear nonhomogeneous model as follows:

Aa p − Ffstatic − FL
ẍ1 + α1 ẋ1 + α2 x1 = , (5.69)
m

where
b(1 + pb ) K
α1 = , α2 = . (5.70)
m m

Second, the virtual command of load pressure pd is designed according to

different forms of the displacement command x1d as follows:
Output Feedback Control Method 85

1. If x1d = υ sin ωt (υ is the amplitude and ω is the angular frequency

of the sinusoidal function), pd is redesigned as

pd (t) = pds (t) + pdd (t),

Kυ sin ωt FL
pds (t) = + , (5.71)
Aa Aa
pdd (t) = −σ1 x2 (t),

where σ 1 is a positive constant. After the item pdd is involved in pd ,

the characteristics roots of model (5.69) can be reassigned to improve
the dynamic performance of x1 .
Compared to the third equation in Equation 5.51, the stick item
Ffstatic is not included in pd shown in the second equation in
Equation 5.71 since the actual displacement x1 will be regulated
dynamically.
2. If x1d is the step command, we consider the similar form

x1d = υ(1 − e−τ t ), (5.72)

where τ is an asymptotic time constant.

Then, pd is redesigned as

pd (t) = pds (t) + pdd (t),

Kυ(1 − e−τ t ) bυe−τ t FL
pds (t) = + + + Ffstatic , (5.73)
Aa Aa Aa
pdd (t) = −σ1 x2 (t).

After Equation 5.71 or 5.73 is substituted into Equation 5.69, the model can
be described by Laplace transform as follows:

s2 x1 (s) + (α1 + σ1 )sx1 (s) + α2 x1 (s) = p (s), (5.74)

where
Aa (pds + ep ) − Ffstatic − FL
p (s) = . (5.75)
m
If p (s) is considered as the system input, and x1 (s) is the system output,
since α 1 + σ 1 > 0, α 2 > 0, the two characteristics roots of system (5.74) have
all negative real parts, which means x1 (t) is asymptotically stable.
When t → ∞, s → 0 in Equation 5.74. According to the backstepping pro-
cedure in section A, the pressure error ep converges to 0. Then, from
Equation 5.74, we see that
Aa pds − Ffstatic − FL
x1 (t) → . (5.76)
α2 m
86 Nonlinear Control Techniques for EHAs in Robotics Engineering

Now Equation 5.71 or 5.73 is substituted into Equation 5.76, we see that

Kx1d + max (FL ) − FL Kx1d (t)

x1 (t) → = = x1d . (5.77)
m mα2

The above equation shows that the system output x1 (t) can be controlled to
track its command x1d by the equivalent system input p (s). The parameter σ 1
is designed to make x1 fast converge to x1d .
According to the Lyapunov analysis in section A, the pressure error ep
converges to 0 exponentially, which means the dynamic response of load
pressure p can be satisfied to the system requirement by the regulation of
parameter k1 . In the expression of pds (t) shown in Equations 5.71 and 5.73,
the dynamic external load error FL (y, ẏ, ÿ) and FL exist. But this error has
little impact on the convergence of x1 .
Finally, the actual states xi (i = 2, . . . ,6) are replaced by their estimation and
the backstepping controller can be described as

⎧ 
⎪ 1 k3
⎪
⎪
⎪ û = 2
ωsv x̂5 + 2ζsv ωsv x̂6 + x̂˙ 6d − ê5 − ê6
⎪
⎪ Ksv ωsv
2 k2
⎪
⎪
⎪
⎪ Aa x̂3 − Ab x̂4
⎪
⎪
⎪
⎪ êp = − pd
⎪
⎪ Aa
⎪
⎨
1
⎪ x̂5d = − (k1 êp − ṗd + g1 ) , (5.78)
⎪
⎪ g2
⎪
⎪
⎪
⎪
⎪
⎪ x̂6d = −k2 g2 êp + x̂˙ 5d − ê5
⎪
⎪
⎪
⎪
⎪
⎪ ê5 = x̂5 − x̂5d
⎪
⎪
⎩
ê6 = x̂6 − x̂6d

where x̂i (i = 3, 4, 5, 6) can be recursively computed by the observer Equa-

tion 5.25.

5.3.3 Stability Discussion of EHS

Now the stability of EHS should be analyzed after the full-state feedback
control is realized. In Equation 5.11, the controller u is replaced by û as
follows:

Ẋ = AX + φ(x) + Bû. (5.79)

Output Feedback Control Method 87

Combined with Equations 5.10, 5.52, 5.53, 5.62, 5.64, and 5.67, the three state
errors satisfy the following conditions:

ėp = −k1 ep + g2 e5 ,
ė5 = −e5 − k2 g2 ep + e6 ,
(5.80)
k3
ė6 = − e5 − e6 .
k2
According to Equations 5.74 through 5.76, the state error x1 equation is
satisfied:
Aa ep
ë1 (s) + (α1 + σ1 )ė1 (s) + α2 e1 (s) = . (5.81)
m
If the full-state error is defined as
 T
Z = z1 z2 z3 z4 z5
 T (5.82)
= e1 e 2 ep e5 e6 ,

and combined with Equations 5.11, 5.79 through 5.81, we see that

Ż = Ae Z + Be Z + B(u − û)
(5.83)
= Ae Z + Be Z + Bũ(X, X̃, x1d ),

where ⎡ ⎤
−1
⎢−α2 −(α1 + σ1 ) ⎥
⎢ ⎥
Ae = ⎢
⎢ −k1 ⎥,
⎥ (5.84)
⎣ −1 ⎦
−1
' (T
Be = 0 Aa
m ep g2 e5 −k2 g2 ep + e6 − kk3 e5 , (5.85)
2

ũ(X, X̃, x1d ) = u(X, x1d ) − û(X̂, x1d ). (5.86)

The observer error dynamics shown in Equation 5.37 is fast dynamics

adopted by the high-gain state observer. On the contrary, the tracking error
dynamics shown in Equation 5.83 is slow dynamics. Therefore, according to
Equation 5.25, the gain of the observer  and the estimation of controller û
can guarantee that the state estimation X̂ converge to X quickly, and then
û → u. This explanation is simply described as

˙
X̂ = A1 X̂ + φ1 ( X̂ ) + B û + (y − ŷ). (5.87)
↓ ↓ ↓ ↓
Ẋ X X u
88 Nonlinear Control Techniques for EHAs in Robotics Engineering

The controller u can guarantee that the state errors ep , e5 , and e6 are asymp-
totic convergence to 0 after the Lyapunov candidate functions V 3 , V 4 , and
V 5 are constructed. Moreover, the state errors e1 and e2 are also asymptotic
convergence to 0 on the basis of Equation 5.81. Obviously, the matrix Ae in
Equation 5.84 is Hurwitz. Therefore, the full-state error dynamics shown in
Equation 5.83 is a quasi-steady-state model as follows:

t>ς
Ż = Ae Z + Be Z + B(u − û) ⇔ Ż = Ae Z + Be Z, (5.88)

where ς is a definite time constant of t.

When t → ∞, according to Equation 5.88, the full-state error Z converges to
0 as quickly as possible.

5.4 Experiment
To test the proposed backstepping control method and the full-state
observer associated with the control of EHS, the experimental frame-
work of the model is presented as shown in Figure 3.16. Some of the
same hydraulic parameters are shown in Table 3.9. The other parameters
are V 0a = 1.74 × 10−5 m3 , V 0b = 8.66 × 10−6 m3 , Ksv = 7.9 × 10−4 m / V,
ωsv = 353.6 rad/s, ζ sv = 0.707, b = 2500 Ns/m, K = 1000 N/m, Ffstatic =
20 N, FL = 500 N, and pb = 20%. Since the stroke of the cylinder is 79 mm
and the shoulder angle of the upper arm is from − 70° to 50°, the range of
absolute coordinates position x1 is from − 0.0476 to 0.0303 m. However, in
the experiment, the displacement of the cylinder is not more than 58 mm to
avoid the boundary collision of the robotic arm.
The observer and control parameters are designed as follows:

1. The observer parameters: λ = 0.5, 1 = 1000, 2 = 10, 3 = 10−3 ,

−3
4 = 10 , 5 = 1000, 6 = 10.
2. The control parameters: k1 = 10−6 , k2 = 10−9 , k3 = 10−12 , σ 1 = 500.
3. In the function tanh(.), k = 1000.

5.4.1 Result of the Proposed Method

The full states are first estimated by observer Equation 5.25. Then, Equa-
tions 5.54, 5.55, and 5.71 or 5.73, and the estimated states are substituted
Output Feedback Control Method 89

into the controller (5.78) to compute the current control voltage. Here, the
displacement command is considered as two types: sinusoidal and step
demands. The amplitude of the sinusoidal command is 0.0289 m with fre-
quency 0.5 Hz. The step command also has the same amplitude. The actual
displacement can be obtained by trigonometry of the robotic arm according
to the angle measured by the relative encoder.
Figures 5.1 through 5.4 show the related results of the sinusoidal and
step experiment, respectively. In Figure 5.1a, after 1.27 s, the actual cylin-
der displacement has been close to the value 0.025 m when the steady error
is less than 5%. The maximum control voltage is 9 V, not more than its
saturation ± 10 V as shown in Figure 5.1b. When the displacement nears
the step command, the control voltage reduces to 0 quickly. The estima-
tion of x1 is a correct prediction for the actual cylinder displacement by
the designed observer as shown in Figure 5.2a. According to the estima-
tions of x3 and x4 as shown in Figure 5.2c and d, the dynamic estimation
is sharp due to the existing error of the high observer. Moreover, this
single-rod EHS has a one-dimensional internal dynamics, which results
in the rigid dynamic error of estimated state. This phenomenon is also
shown in the sinusoidal experiment, which means the estimation of load
pressure pL defined in Equation 5.51 is not smooth until the dynamic
response finishes. In step experiment, after approximately 3 s later, the esti-
mation of pL is smooth and is close to 30 bar shown in Figure 5.2c. The
estimation of spool position x5 is corresponded to the control voltage û,
which is not more than its saturation ± 7.9 mm shown in Figure 5.2e. In
the sinusoidal experiment, the maximum dynamic error of displacement
is less than 5.7 mm, which means the dynamic error is not more than

(a) (b)
0.08 10
Command
0.06 Actual 8
Error
Estimation of u (V)

0.04
0.0289 6
0.025
x1 (m)

0.02
4
0
2
–0.02
–0.04 0

–0.06 –2
0 1.27 2 3 4 5 6 7 8 9 10 0 2 4 6 8 10
t (s) t (s)

FIGURE 5.1
Step response experiment of the upper arm hydraulic actuator. (a) The position of the upper arm
cylinder. (b) The control voltage of the upper arm cylinder.
90 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) 0.04 (b) 0.1

0.0289

Estimation of x2 (m/s)
0.02 0.05
Estimation of x1 (m)

0 0

–0.02 –0.05

–0.04 –0.1

–0.06 –0.15
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
(c) × 106 (d) × 105
4 3.5
3
Estimation of x3 (Pa)

Estimation of x4 (Pa)

3 2.5
2
2
1.5

1 1
0.5
0 0
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
(e) (f) 1.5
× 10–3
5
Estimation of x6 (m/s)

1
Estimation of x5 (m)

3 0.5

2 0

1 –0.5

0 –1
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)

FIGURE 5.2
State estimations by the high-gain state observer in step response experiment. (a) Estimation of
x1 in step response. (b) Estimation of x2 in step response. (c) Estimation of x3 in step response.
(d) Estimation of x4 in step response. (e) Estimation of x5 in step response. (f) Estimation of x6 in
step response.

10% when the sinusoidal frequency is 0.5 Hz. Owing to the initial posi-
tion error of x1 and dynamic estimated error for the sinusoidal command,
the control voltage is saturation in a short duration. When the actual dis-
placement tracks the command well, the control voltage is less than its
saturation.
Output Feedback Control Method 91

(a) (b)
0.06 10
Command
Actual 8
0.04 Error 6

Estimation of u (V)
0.02 4
2
x1 (m)

0 0
–2
–0.02
–4
–0.04 –6
–8
–0.06 –10
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)

FIGURE 5.3
Sinusoidal response experiment of the upper arm hydraulic actuator. (a) The position of the
upper arm cylinder. (b) The control voltage of the upper arm cylinder.

5.4.2 Compared Result


The conventional PI control method u = kp (x1d − x1 ) + ki (x1d − x1 )dt is also
used in this experiment. To illustrate the problem, we compare the PI method
and the backstepping method in two different conditions:

1. The sinusoidal frequency is 0.5 Hz and the elbow is always retracted

as seen in Figure 5.8a, which means the required control bandwidth
is low and the dynamic external load is relatively small.
2. The sinusoidal frequency is 1 Hz and the elbow is always extended
as seen in Figure 5.8b, which is the worst control condition for the
shoulder hydraulic actuator.

Here, we choose the PI control parameters kp = 100 and ki = 15, which

can guarantee the fast response of displacement shown in Figure 5.5a. In fact,
the dynamic error of the PI method is smaller than the proposed method in
the first condition shown in Figure 5.5b. But when the required frequency of
command is increased and the external load nears the critical value, which is
the maximum bearing load of the shoulder hydraulic actuator, the dynamic
performance of the PI method becomes worse than the proposed method
shown in Figure 5.6a. The control voltage is more intense for the PI method
as shown in Figure 5.6c, which results into the dynamic displacement error
that is larger than the proposed method shown in Figure 5.6b. These com-
pared results show that the fixed control parameters of the PI method cannot
fit all work conditions. If the control parameters are chosen big enough, the
robustness performance will become weak in critical conditions, except most
normal conditions. Therefore, the proposed method is suitable for this robotic
92 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) 0.04 (b) 0.2

Estimation of x2 (m/s)
0.02 0
Estimation of x1 (m)

0 –0.2

–0.02 –0.4

–0.04 –0.6

–0.06 –0.8
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
(b) × 106 (d) × 106
4 4
Estimation of x3 (Pa)

Estimation of x4 (Pa)
3 3

2 2

1 1

0 0
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
(e) (f)
0.01 1.5
0.008
1
Estimation of x6 (m/s)

0.006
Estimation of x5 (m)

0.004 0.5
0.002
0 0
–0.002
–0.5
–0.004
–0.006 –1
–0.008
–0.01 –1.5
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t ( s)

FIGURE 5.4
State estimations by the high-gain state observer in the sinusoidal response experiment. (a)
Estimation of x1 in the sinusoidal response. (b) Estimation of x2 in the sinusoidal response.
(c) Estimation of x3 in the sinusoidal response. (d) Estimation of x4 in the sinusoidal response.
(e) Estimation of x5 in the sinusoidal response. (f) Estimation of x6 in the sinusoidal response.

arm in some critical conditions where the advantage of the PI method is

limited.
Here, two joints of this robotic arm can be driven simultaneously to realize
the coordinated motion. The hydraulic actuator of the shoulder is controlled
by the proposed method and the actuator of the elbow is controlled by the
Output Feedback Control Method 93

(a) 0.04

0.02

0
x1 (m)

–0.02

–0.04 Command
Backstepping
PI
–0.06
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
t (s)
(b) 0.06
Backstepping
0.05 PI

0.04
Error of x1 (m)

0.03

0.02

0.01

–0.01
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
t (s)
(c) 10
Backstepping
PI
5
Estimation of u (V)

–5

–10
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
t (s)

FIGURE 5.5
Comparison result in condition (1). (a) The cylinder position response of the sinusoidal demand.
(b) The cylinder position error of the sinusoidal demand. (c) The control voltage estimation of
the sinusoidal demand.
94 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) 0.02

0.01

–0.01
x1 (m)

–0.02

–0.03
Command
–0.04 Backstepping
PI
–0.05
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
t (s)

(b) 0.06
Backstepping
PI
0.04
Error of x1 (m)

0.02

–0.02
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
t (s)

(c) 10
Backstepping
PI
5
Estimation of u (V)

–5

–10
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
t (s)

FIGURE 5.6
Comparison result in condition (2). (a) The cylinder position response of the sinusoidal demand.
(b) The cylinder position error of the sinusoidal demand. (c) The control voltage estimation of
the sinusoidal demand.
Output Feedback Control Method 95

(a) 0.04

0.02

0
–0.005
–0.02
x1 (m)

–0.04

–0.06 Shoulder command

Shoulder actual
–0.08 Elbow command
Elbow actual
–0.1
0 1 1.3 2 2.5 3 4 5
t (s)

(b) 10
Shoulder control
Elbow control

5
u (V)

–5

–10
0 1 2 3 4 5
t (s)

FIGURE 5.7
The coordinated motion experiment results of the robotic arm joints, sinusoidal demand input
for the shoulder actuator, and step demand input for the elbow actuator. (a) The two cylinder
positions response of sinusoidal and step demands. (b) The two control voltages of sinusoidal
and step demands.

PI method. The two input displacement commands are sinusoidal and step,
respectively, which is the same command as section B. The displacement
response and the corresponding control voltage of two hydraulic actuators
are shown in Figure 5.7. At initial time 0 s, the two joints are all retracted.
Then, after 1.3 s, the joint angle of the elbow is extended to its maximum 130°
by the PI control and then the elbow joint will always keep this extended
state. After 1.5 s, the shoulder actuator is retracted to its minimum dis-
placement − 0.0289 m. After 2.5 s, the shoulder actuator is extended to its
96 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) (b)

(c) (d)

FIGURE 5.8
Experiment video of coordinated motion for the robotic arm joints. (a) t = 0 s. (b) t = 1.3 s.
(c) t = 1.5 s. (d) t = 2.5 s.

maximum displacement 0.0289 m. Then the sinusoidal law of the shoulder

joint is repeated by the proposed control method. The experiment video of
coordinated motion for the robotic arm joints is shown in Figure 5.8, which
is corresponding to the four sample points.
6
Parametric Adaptive Control Method

CONTENTS
6.1 Dynamic Model of EHS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2 Parametric Adaptive Backstepping Controller . . . . . . . . . . . . . . . . . 99
6.2.1 General Backstepping Control Design . . . . . . . . . . . . . . . . . 100
6.2.2 Decayed Memory Filter Design . . . . . . . . . . . . . . . . . . . . . . 104
6.2.3 Revised Parametric Adaptive Control Law . . . . . . . . . . . . . 106
6.3 Disturbance Observer Application . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.4 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.4.1 Result of the Proposed Controller . . . . . . . . . . . . . . . . . . . . 112
6.4.2 Comparison with Simplified Backstepping Controller . . . . . 114
6.5 Result of Disturbance Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

To the best of the authors’ knowledge, the aforementioned backstepping and

adaptive controllers need to deal with some derivatives of virtual control
variables, which exist in backstepping iteration. These derivatives can be
computed by the system state error model and parametric estimated law
established in recursive controller design [37]. Since this computed process
easily leads to derivative explosion, few research are focused on the quality
of these virtual control variables (i.e., data validation). If these virtual control
variables are not available, especially violent, both the stability and dynamic
performances of EHS will be significantly decreased. Thus, in this chapter, a
parametric adaptive backstepping control method [18] is presented based on
state feedback to estimate some unknown parameters in a hydraulic model.
Then, a decayed memory filter is proposed to compute the derivatives value
of virtual control variables in the backstepping control design. The effec-
tiveness of the proposed control is verified by a comparative experimental
study.

97
98 Nonlinear Control Techniques for EHAs in Robotics Engineering

6.1 Dynamic Model of EHS

For convenient illustration, the dynamic model of EHS is adopted by the
symmetrical hydraulic actuator. The load flow QL of the servo valve men-
tioned in Equation 2.5 is given by

1 
QL = Cd wxv ps − sgn (xv ) pL , (6.1)
ρ
where pL is the load pressure of the cylinder, xv is the spool position of the
servo valve, ps is the supply pressure of the pump, Cd is the discharge coef-
ficient, w is the area gradient of the valve spool, and ρ is the density of the
hydraulic oil.
The flow-pressure continuous equation of the hydraulic cylinder men-
tioned in Equation 2.1 is given by
dy Vt dpL
QL = Ap + Ctl pL + , (6.2)
dt 4βe dt
where y is the displacement of the cylinder, Ctl is the coefficient of the total
leakage of the cylinder, β e is the effective bulk modulus, Ap is the annulus
area of the cylinder chamber, and V t is the half-volume of the cylinder.
If the major viscous friction of the hydraulic oil is simplified as Coulomb
friction, the mechanical dynamic equation driven by the hydraulic actuator
mentioned in Equation 6.3 is shown as follows:
Ap pL = mÿ + bẏ + Ky + FL , (6.3)
where m is the load mass, K is the load spring constant, b is the viscous
damping coefficient, and FL is the external load on the hydraulic actuator.
The dynamics of the servo valve is adopted by one-order linear model
mentioned in Equation 2.8, which is given by
Tsv ẋv + xv = Ksv u, (6.4)
where Ksv is the gain of the control voltage u and Tsv is the response time
constant of the servo valve.

Remark 6.1
In Equation 6.1, the function sgn(.) should be smoothed in the derivation of
backstepping control, which is replaced by the hyperbolic tangent function
tanh(.) as follows [35]:
ekxv − e−kxv
sgn(xv ) ≈ tanh(kxv ) = , k  0, (6.5)
ekxv + e−kxv
where k is a positive constant.
Parametric Adaptive Control Method 99

If the state vector X = [x1 , x2 , x3 , x4 ] = [y, ẏ, pL , xv ]T , the output variable

y = x1 , and the control input u(t) is defined, the dynamics of EHS is con-
structed as the fourth state space model as follows:

⎧
⎪
⎪ ẋ = x2
⎪ 1
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪ ẋ2 = (−Kx1 − bx2 + Ap x3 − FL )
⎪
⎪ m
⎨
4βe Ap 4βe Ctl 4βe Cd w . (6.6)
⎪
⎪ ẋ3 = − x2 − x3 + √ ps − tanh(kx4 )x3 x4
⎪
⎪ Vt ρ
⎪
⎪ Vt Vt
⎪
⎪
⎪
⎪
⎪
⎪ 1 Ksv
⎩ ẋ4 = − x4 + u
Tsv Tsv

The external load FL (t) is divided into two elements, which is mentioned in
Equation 4.11 as follows:

⎧
⎪ Tu (θ1 , θ2 )
⎪
⎨ FLu (θ1 , θ2 ) = l1 (θ1 )
⎪
, (6.7)
⎪
⎪ Tf (θ1 , θ2 )
⎪
⎩ FLf (θ1 , θ2 ) =
l2 (θ2 )

where FLu is the load force on the shoulder hydraulic actuator, FLf is
the load force on the elbow hydraulic actuator, and the dynamic force
arms li (θ i ), (i = 1, 2) are computed by triangle geometry mentioned in
Equation 3.48.

Remark 6.2
Owing to measurement noise and uncertainty disturbance in engineer-
ing practice, FLu , FLf are difficult to be obtained. The computed value in
Equation 6.7 may deviate from the true value. Thus, these values should be
handled by a filter before being used in parametric estimation law.

6.2 Parametric Adaptive Backstepping Controller

All the states xi (i = 1, . . . , 4) are known for the controller design. If
these unknown parameters are defined as ϑ 1 = K, ϑ 2 = b, ϑ 3 = FL ,
√
ϑ 4 = ((4β e Ap )/V t ), ϑ 5 = ((4β e Ctl )/V t ), ϑ6 = ((4βe Cd w)/(Vt ρ)), then the
100 Nonlinear Control Techniques for EHAs in Robotics Engineering

dynamics of EHS can be described as follows [35,36]:

⎧
⎪
⎪ ẋ1 = x2
⎪
⎪
⎪
⎪ Ap ϑ1 ϑ2 ϑ3
⎪
⎪
⎨ ẋ2 = m x3 − m x1 − m x2 − m
⎪
. (6.8)
⎪
⎪ ẋ3 = −ϑ4 x2 − ϑ5 x3 + ϑ6 ps − tanh(kx4 )x3 x4
⎪
⎪
⎪
⎪
⎪
⎪ 1 Ksv
⎪
⎩ ẋ4 = − x4 + u
Tsv Tsv

6.2.1 General Backstepping Control Design

At first, the state error zi (i = 1, . . . , 4) and parametric estimated error ϑ̃i (i =
1, . . . , 6) are defined as follows:
⎧
⎪ z1 = x1 − x1d
⎪
⎨
zi = xi − αi−1 (i = 2, 3, 4) , (6.9)
⎪
⎪
⎩
ϑ̃i = ϑi − ϑ̂i , (i = 1, . . . , 6)

where x1d is the demand of cylinder displacement, α i is the ith virtual control,
and ϑ̂i is the estimation of ϑ i .

Theorem 6.1
Assuming that the unknown parameters ϑ i (i = 1, . . . , 6) are constant, if the
virtual control variables α i ∈ C1 (i = 1, 2, 3) and their derivatives α̇i (i = 1, 2, 3)
are smooth, then there exists a backstepping controller u, which guarantees
zi (t)(i = 1, . . . , 4) → 0, ϑ̃i (t)(i = 1, . . . , 6) → 0, as t → ∞.

Proof. The derivative of z1 is given by

ż1 = z2 + α1 − ẋ1d . (6.10)

Then the Lyapunov function V 1 is defined as

1 2
V1 = z . (6.11)
2 1
Since the derivative of V 1 is

V̇1 = z1 (z2 + α1 − ẋ1d ), (6.12)

the virtual control α 1 is designed as follows:

α1 = −c1 z1 + ẋ1d , (6.13)

where c1 is a positive constant.

Parametric Adaptive Control Method 101

Therefore, the derivative of V 1 is

V̇1 = −c1 z21 + z1 z2 . (6.14)

Since the derivative of z2 is

ż2 = ẋ2 − α̇1

Ap ϑ1 ϑ2 ϑ3
= x3 − x1 − x2 − + c1 x2 − c1 ẋ1d − ẍ1d (6.15)
m m m m
Ap Ap ϑ1 ϑ3 ϑ3
= z3 + α2 − x1 − x2 − + c1 x2 − c1 ẋ1d − ẍ1d ,
m m m m m

the Lyapunov function V 2 is defined as follows:

1 ) 1 3 2
V2 = V1 + z22 + ϑi − ϑ̂i , (6.16)
2 2ki
i=1

where k1 , k2 , and k3 are positive constants.

Therefore, the derivative of V 2 is

Ap Ap ϑ1 ϑ2 ϑ3
V̇2 = −c1 z21 + z1 z2 + z2 z3 + α2 − x1 − x2 −
m m m m m

ϑ̂˙ 1 ϑ̂˙ 2 ϑ̂˙ 3
+ c1 x2 − c1 ẋ1d − ẍ1d − ϑ̃1 − ϑ̃2 − ϑ̃3
k1 k2 k3

Ap Ap
= −c1 z21 + z2 z1 + z3 + α2 − ϑ̂1 x1 /m − ϑ̂2 x2 /m − ϑ̂3 /m
m m
    
ϑ̂˙ 1 ϑ̂˙ 2
+ c1 x2 − c1 ẋ1d − ẍ1d − ϑ̃1 + x1 z2 /m − ϑ̃2 + x2 z2 /m
k1 k2
 
ϑ̂˙ 3
− ϑ̃3 + z2 /m . (6.17)
k3

If the parametric adaptive estimation laws and the virtual control α 2 are
designed as follows:

ϑ̂˙ 1 = −k1 x1 z2 /m, ϑ̂˙ 2 = −k2 x2 z2 /m, ϑ̂˙ 3 = −k3 z2 /m, (6.18)

m c2 m m 
α2 = − z1 − z2 + ϑ̂1 x1 /m + ϑ̂2 x2 /m + ϑ̂3 /m − c1 x2 + c1 ẋ1d + ẍ1d ,
Ap Ap Ap
(6.19)
102 Nonlinear Control Techniques for EHAs in Robotics Engineering

we see that the derivative of V 2 is

Ap
V̇2 = −c1 z21 − c2 z22 + z2 z3 , (6.20)
m
where c2 is a positive constant.
Then the derivative of z3 is

ż3 = ẋ3 − α̇2 = −ϑ4 x2 − ϑ5 x3 + ϑ6 ps − tanh(kx4 )x3 x4 − α̇2 . (6.21)

If the Lyapunov function V 3 is defined as

1 ) 1 6 2
V3 = V2 + z23 + ϑi − ϑ̂i , (6.22)
2 2ki
i=4

then the derivative of V 3 is

Ap
V̇3 = −c1 z21 − c2 z22 + z2 z3
m
  
+ z3 −ϑ4 x2 − ϑ5 x3 + ϑ6 ps − tanh(kx4 )x3 (z4 + α3 ) − α̇2

ϑ̂˙ 4 ϑ̂˙ 5 ϑ̂˙ 6

− ϑ̃4 − ϑ̃5 − ϑ̃6
k4 k5 k6
 Ap  
2 2 z2 + ϑ̂6 ps − tanh(kx4 )x3 z4
= −c1 z1 − c2 z2 + z3 m

+ϑ̂6 ps − tanh(kx4 )x3 α3 − ϑ̂4 x2 − ϑ̂5 x3 − α̇2
     
ϑ̂˙ 4 ϑ̂˙ 5 ϑ̂˙ 6 
− ϑ̃4 + x2 z3 − ϑ̃5 + x3 z3 − ϑ̃6 − x4 z3 ps − tanh(kx4 )x3 ,
k4 k5 k6
(6.23)

where k4 , k5 , and k6 are positive constants.

If the parametric adaptive estimation laws and the virtual control α 3 are
designed as follows:

ϑ̂˙ 4 = −k4 x2 z3 , ϑ̂˙ 5 = −k5 x3 z3 , ϑ̂˙ 6 = k6 x4 z3 ps − tanh(kx4 )x3 , (6.24)

m
Ap z2 + c3 z3 + ϑ̂4 x2 + ϑ̂5 x3 + α̇2
α3 = −  , (6.25)
ϑ̂6 ps − tanh(kx4 )x3
we see that the derivative of V 3 is

V̇3 = −c1 z21 − c2 z22 − c3 z23 + z3 z4 ϑ̂6 ps − tanh(kx4 )x3 , (6.26)

where c3 is a positive constant.

Parametric Adaptive Control Method 103

If the Lyapunov function V 4 is defined as

1
V4 = V3 + z24 , (6.27)
2

then the derivative of V 4 is

V̇4 = V̇3 + z4 ż4 = −c1 z21 − c2 z22 − c3 z23

+ z3 z4 ϑ̂6 ps − tanh(kx4 )x3 + z4 (ẋ4 − α̇3 )


x4 Ksv
= −c1 z21 − c2 z22 − c3 z23 + z4 z3 ϑ̂6 ps − tanh(kx4 )x3 − + u − α̇3 .
Tsv Tsv
(6.28)

If the controller is designed as

Tsv x4 Tsv Tsv

u=− c4 z4 + + α̇3 − z3 ϑ̂6 ps − tanh(kx4 )x3 , (6.29)
Ksv Ksv Ksv Ksv

then the derivative of V 4 is

V̇4 = −c1 z21 − c2 z22 − c3 z23 − c4 z24 < 0, (6.30)

where c4 is a positive constant.

Thus, by the parametric adaptive estimation laws (6.18) and (6.24), the ele-
ments involving ϑ̃i (i = 1, . . . , 6) vanish in V̇4 . Simultaneously, V̇4 is negative
definite by the virtual control (6.13), (6.19), and (6.25) and the backstep-
ping controller (6.29). This iteration controller guarantees the parametric
estimation errors ϑ̃i (i = 1, . . . , 6) and the system state error zi (i = 1, . . . , 4)
converging to zero. 

From the assumption in Theorem 6.2, the derivative of α 2 , α 3 used in

Equations 6.19 and 6.25 can be simplified as follows:

m
α̇2 = − (z2 + α1 − ẋ1d ) − c2 z3 − c2 α2
Ap
c2 m c1 m ˆ
+ (−ϑ̂1 x1 /m − ϑ̂2 x2 /m − ϑ̂3 /m + c1 x2 − c1 ẋ1d − ẍ1d ) − ẋ2 ,
Ap Ap
(6.31)

z3 − m
Ap (c2 z2 + z1 ) + c3 (ẋˆ 3 − α̇2 ) + ϑ̂4 ẋˆ 2 + ϑ̂5 ẋˆ 3 + α̈2
α̇3 = −  , (6.32)
ϑ̂6 ps − tanh(kx4 )x3
104 Nonlinear Control Techniques for EHAs in Robotics Engineering

where ẋˆ 2 , ẋˆ 3 is estimated by the dynamic model of EHS as follows:

⎧ Ap
⎪
⎨ ẋˆ 2 = x3 − ϑ̂1 x1 /m − ϑ̂2 x2 /m − ϑ̂3 /m
m . (6.33)
⎪
⎩ˆ
ẋ3 = −ϑ̂4 x2 − ϑ̂5 x3 +ϑ̂6 ps − tanh(kx4 )x3 x4

6.2.2 Decayed Memory Filter Design

As shown in Equations 6.31 and 6.32, although the derivatives α̇2 , α̇3 are
computed in backstepping design, the algorithm is still very complicated.
If these derivatives are directly used in the controller u, the robustness of this
backstepping controller may decline. Therefore, a decayed memory filter is
designed to re-estimate these derivatives and filter the variable parameter ϑ 3
mentioned in Remark 6.1. The filter form is described as follows [65]:
⎧  
⎪ ˆ
⎪ Eφ (k) = φ(k) − φ̂(k − 1) + φ̇(k − 1) · Tc
⎪
⎪
⎪
⎨
ˆ − 1) · T + G · E (k) ,
φ̂(k) = φ̂(k − 1) + φ̇(k φ = α 2 , α3 , ϑ3 , (6.34)
c φ
⎪
⎪
⎪
⎪ H
⎪ ˆ
⎩ φ̇(k) ˆ − 1) + E (k)
= φ̇(k φ
Tc

ˆ
where φ(k) is the kth calculated value of φ, φ̂(k) is the estimation of φ(k), φ̇(k)
is the kth estimation of the derivative φ̇, Tc is the interval of the controller, the
filter parameters G = 1 − λ2 , H = (1 − λ)2 , and the range of the filter factor
λ is (0, 1).
Since the calculated values of α 2 , α 3 are obtained by Equations 6.19
and 6.25, the estimation values α̂i , α̇ˆ i (i = 2, 3) are obtained by Equation 6.34.
From Equation 6.7, the calculated value of ϑ 3 is described as follows:

⎧
⎪ FLu (θ1 , θ2 ) Tu (θ1 , θ2 )
⎪
⎪ ϑ31 = =
⎨ m1f m1f l1 (θ1 )
, (6.35)
⎪
⎪ FLf (θ1 , θ2 ) Tf (θ1 , θ2 )
⎪
⎩ ϑ32 = =
m2f m2f l2 (θ2 )

where ϑ 31 , ϑ 32 represent two calculated values of the parameter ϑ 3 for two

hydraulic actuators. After Equation 6.35 is substituted into Equation 6.34, the
estimation values ϑ̂ , ϑ̇ˆ (i = 1, 2) are also obtained.
3i 3i
Then the convergence of the proposed filter should be analyzed. If the vec-
 T
ˆ
tor χ (k) is defined as χ (k) = α̂(k) , α̇(k) , then the filter (6.34) is described
Parametric Adaptive Control Method 105

as follows:
⎡ ⎤ ⎡ ⎤

α̂(k)
 1−G (1 − G)Tc α̂(k − 1) G
= ⎣ H ⎦ + ⎣ H ⎦ α(k) . (6.36)
ˆ
α̇(k) − 1−H ˆ − 1)
α̇(k
Tc Tc

Theorem 6.2
The input α(k) is assumed to be the input of the discrete linear system (6.36),
ˆ
and α̂(k), α̇(k) are the outputs; then Equation 6.36 is input-to-state stable (ISS),
ˆ
and α̂(k) → α(k), α̇(k) → α̇(k), as k → ∞.

Proof.

1. If two constants are defined as λ1 = 1 − G, λ2 = 1 − H, then we

can see that λ1 < 1, λ2 < 1, λ1 < λ2 . The characteristic polynomial
of system matrix A in Equation 6.36 is described as follows:
⎡ ⎤
s − λ1 −λ1 Tc
|sI − A| = ⎣ λ2 − 1 ⎦ = s2 − (λ1 + λ2 )s + λ1 = 0. (6.37)
− s − λ2
Tc

Substituting λ1 = λ2 , λ2 = 2λ − λ2 into Equation 6.37, the eigen-

values of system matrix A is given by

λ 1 + λ2 ± (λ1 + λ2 )2 − 4λ1
s1,2 = = λ. (6.38)
2
Since the filter factor λ < 1, the eigenvalues norm is strictly less
than 1. According to the stability of the linear discrete system, if the
system input α(k) is bounded, then Equation 6.36 is ISS.
ˆ
2. Since the two outputs α̂(k), α̇(k) are ISS, the filter convergence can be
analyzed by two steps. First, from Equation 6.36, we can see that

ˆ − 1) + Gα(k) .
α̂(k) = (1 − G)α̂(k − 1) + (1 − G)Tc α̇(k (6.39)

When α̂(k) reaches its steady state, there exists a sufficiently large
ˆ
integer n0 , ∀k > n0 , α̇(k) → 0, and α̂(k) = α̂(k − 1). Substituting these
two conditions into Equation 6.39, we can obtain that α̂(k) → α(k).

ˆ
Second, from Equation 6.36, the dynamics of α̇(k) is given by

ˆ H ˆ − 1) + H α(k) .
α̇(k) = − α̂(k − 1) + (1 − H)α̇(k (6.40)
Tc Tc
106 Nonlinear Control Techniques for EHAs in Robotics Engineering

ˆ
When α̇(k) reaches its steady state, there exists a sufficiently large inte-
ˆ
ger n1 , ∀k > n1 , α̂(k − 1) → α(k − 1), and α̇(k) ˆ − 1). Simultaneously,
= α̇(k
we can see that (H/Tc )(α(k) − α(k − 1)) = Hα̇(k − 1). Substituting these three
ˆ
conditions into Equation 6.40, we can obtain that α̇(k) → α̇(k). 

6.2.3 Revised Parametric Adaptive Control Law

In Theorem 6.2, the unknown parameter ϑ 3 is assumed to be constant.
This assumption is not appropriate due to the variable external loads
on two hydraulic actuators. Thus, the aforementioned parametric adap-
tive backstepping controller should be revised to guarantee the EHS (6.8)
convergence.

Theorem 6.3
Assuming that the unknown parameters ϑ 3 is variable but its derivative ϑ̇3 is
bounded, and the other unknown parameters ϑ i (i = 1, 2, 4, 5, 6) are constants,
if the virtual control variables α i ∈ C1 (i = 1, 2, 3), and their derivatives α̇i (i =
1, 2, 3) are bounded, then there exists a revised backstepping controller u that
guarantees zi (t)(i = 1, . . . , 4) → 0, ϑ̃i (t)(i = 1, . . . , 6) → 0, as t → ∞.

Proof. The parametric adaptive estimation law of ϑ̂˙ 3 in Equation 6.18 can be
revised as follows:

 
ϑ̂˙ 3 = −k3 z2 + ϑ̇ˆ 3 +  ϑ̇3 max sgn(ϑ3 − ϑ̂3 ), (6.41)

 
where  ϑ̇3 max is the bound of the filter error ϑ̇3 (i.e., ϑ̇3 = ϑ̇3 − ϑ̇ˆ 3 ), ϑ 3
is the calculated value from Equation 6.35, and ϑ̂3 , ϑ̇ˆ 3 are the filter outputs.
Owing to ϑ̇3 = 0, the Lyapunov function V 2 is rewritten as follows:


Ap Ap
V̇2 = −c1 z21 + z1 z2 + z2 z3 + α2 − ϑ̂1 x1 − ϑ̂2 x2 − ϑ̂3 + c1 x2 − c1 ẋ1d − ẍ1d
m m

ϑ̂˙ 1 ϑ̂˙ 2 (ϑ̂˙ 3 − ϑ̇3 )

− ϑ̃1 − ϑ̃2 − ϑ̃3
k1 k2 k3

Ap Ap
= −c1 z21 + z2 z1 + z3 + α2 − ϑ̂1 x1 − ϑ̂2 x2 − ϑ̂3 + c1 x2 − c1 ẋ1d − ẍ1d
m m
     
ϑ̂˙ 1 ϑ̂˙ 2 ϑ̂˙ 3 − ϑ̇3
− ϑ̃1 + x1 z2 − ϑ̃2 + x2 z2 − ϑ̃3 + z2 . (6.42)
k1 k2 k3
Parametric Adaptive Control Method 107

Substituting Equation 6.41 and ϑ̂˙ 1 , ϑ̂˙ 2 in Equation 6.18 into Equation 6.42,
we can see that

Ap ϑ̃3
V̇2 = −c1 z21 − c2 z22 + z2 z3 + ( ϑ̇3 − | ϑ̇3 |max sgnϑ̃3 ). (6.43)
m k3

According to the convergence of the filter, ϑ̇3 is bounded by | ϑ̇3 |max .

This denotes that the last element in Equation 6.43 is less than 0.
Similarly, if the filter errors α̇2 , α̇3 (i.e., α̇2 = α̇2 − α̇ˆ 2 , α̇3 = α̇3 − α̇ˆ 3 )
are bounded by | α̇2 |max , | α̇3 |max , then the virtual control α 3 is revised as
follows:
m
Ap z2 + c3 z3 + ϑ̂4 x2 + ϑ̂5 x3 + α̇ˆ 2 − | α̇2 |max sgnz3
α3 = −  . (6.44)
ϑ̂6 ps − tanh(kx4 )x3

Substituting Equations 6.44 and 6.24 into V̇3 , the Lyapunov function V 3 is
rewritten as follows:

V̇3 = −c1 z21 − c2 z22 − c3 z23 + z3 z4 ϑ̂6 ps − tanh(kx4 )x3

(6.45)
ϑ̃3  
+ ( ϑ̇3 −  ϑ̇3 max sgnϑ̃3 ) − z3 ( α̇2 + | α̇2 |max sgnz3 ),
k3

where the last element is less than 0.

Finally, if the controller u is redesigned as follows:

Tsv x4 Tsv ˆ Tsv

u=− c4 z4 + + α̇3 − | α̇3 |max sgnz4
Ksv Ksv Ksv Ksv
Tsv
− z3 ϑ̂6 ps − tanh(kx4 )x3 , (6.46)
Ksv

then the derivative of V 4 is rewritten as follows:

ϑ̃3  
V̇4 = −c1 z21 − c2 z22 − c3 z23 − c4 z24 + ( ϑ̇3 −  ϑ̇3 max sgnϑ̃3 )
k3 (6.47)
− z3 ( α̇2 + | α̇2 |max sgnz3 ) − z4 ( α̇3 + | α̇3 |max sgnz4 ) < 0.

By the parametric adaptive estimation law (6.41), the sign of the dynamic
element ϑ̃3 is guaranteed to be negative in V̇4 . Similarly, the signs of z3 and
z4 become negative by the virtual control (6.44) and the revised backstep-
ping controller (6.47). This iteration controller guarantees ϑ̃i (i = 1, . . . , 6) and
zi (i = 1, . . . , 4) converging to zero. 
108 Nonlinear Control Techniques for EHAs in Robotics Engineering

FL

Revised backstepping controller u Electro-hydraulic systems y

(6.46) (6.6)

α
ˆ 2,αˆ 3
x
ϑ̂i, (i = 1,...,6) ˆ ,αˆ
α
α
ˆ 2,αˆ3 2 3
Parametric ˆ ,αˆ
α Filtering α2,α3 Virtual control
2 3
estimation laws estimation (6.13), (6.19),
ˆ ,ϑˆ
(6.18), (6.24), (6.41) ϑ (6.34) ϑ3 (6.44)
3 3

x z α1, α2, α3
Traditional ARC
x1d State errors
(6.9) z

FIGURE 6.1
Block diagram of the control system.

Remark 6.3
The constants | ϑ̇3 |max , | α̇2 |max , and | α̇3 |max need to be predefined
before the revised backstepping controller (6.46). From Equations 6.35
and 6.40, these three constants can be estimated by maxt→∞ {α̇id (t) − α̇ˆ i (t)},
max {ϑ̇ d (t) − ϑ̇ˆ (t)}, where α̇ d (t) = |α̇ |
t→∞ 3 3 isin 2πt for i = 2, 3, ϑ̇ d (t) =
i max 3
|ϑ̇3 |max sin 2πt. The constants |α̇i |max , |ϑ̇3 |max can be estimated in the follow-
ing simplified controller (6.49).

Therefore, the revised backstepping controller (6.46) guarantees zi (i =

1, . . . , 4) and ϑ̃i (i = 1, . . . , 6) converging to zero. This controller involves the
parametric estimation laws (6.18), (6.24), and (6.41), the state errors (6.9), the
virtual control laws (6.13), (6.19), and (6.44), and the filtering estimation (6.34)
as shown in Figure 6.1.

6.3 Disturbance Observer Application

For another way, the external disturbance FL (t) can be adopted by
the observer disturbance due to the dynamic characteristic of FL (t). If
f 21 (x3 ) = −Ap x3 /m, f 22 (x1 ) = −x1 /m, f 23 (x2 ) = −x2 /m, and d = −FL (t)/m
are defined, then the high-gain observer is given by

˙
d̂ = Kd (ẋ2 − f21 (x3 ) − ϑ̂1 f22 (x1 ) − ϑ̂2 f23 (x2 ) − d̂). (6.48)
Parametric Adaptive Control Method 109

In fact, integrating the above equation, the kth disturbance estimation of d

is given by

c
kT

d̂(k) = Kd x2 − Kd (f21 (x3 ) + ϑ̂1 f22 (x1 ) + ϑ̂2 f23 (x2 ) + d̂)dt. (6.49)
(k−1)Tc

According to Equation 6.8, we can see that

ḋ = ẋ2 − f21 (x3 ) − ϑ1 f22 (x1 ) − ϑ2 f23 (x2 ). (6.50)

If the disturbance observer error d̃ = d − d̂ is defined, then

˙ x1 ϑ̃1 x2 ϑ̃2
d̃ = −Kd d̃ + ḋ + Kd + Kd , (6.51)
m m
where ϑ̃1 = ϑ1 − ϑ̂1 , ϑ̃2 = ϑ2 − ϑ̂2 .
Integrating the above equation, we can obtain

t  
−Kd t −Kd (t−τ ) x1 (τ )ϑ̃1 (τ ) x2 (τ )ϑ̃2 (τ )
d̃(t) = e d̃(0) + e ḋ(τ ) + Kd + Kd dτ
m m
0

t
|ḋ|max (1 − e − Kd t ) x1 (τ )ϑ̃1 (τ ) + x2 (τ )ϑ̃2 (τ )
≤ e−Kd t d̃(0) + + Kd e−Kd (t−τ ) dτ .
Kd m
0
(6.52)

If the designed parametric adaptive estimation law Equation 6.18 can

guarantee two estimation errors ϑ 1 , ϑ 2 bounded where the corresponding
boundedness is arbitrarily small, then Equation 6.52 can be furthermore han-
dled. In other words, ∀ > 0, ∃t1 > 0, as the time t > t1 , ϑ i < for i = 1, 2.
We can assume that ϑ̃i ≤ |ϑ̃i |max . Thus, Equation 6.52 satisfies the following
condition:
t1
−Kd t |ḋ|max (1 − e−Kd t ) x1 (τ )ϑ̃1 (τ ) + x2 (τ )ϑ̃2 (τ )
d̃(t) ≤ e d̃(0) + + Kd e−Kd (t−τ ) dτ
Kd m
0

t
x1 (τ )ϑ̃1 (τ ) + x2 (τ )ϑ̃2 (τ )
+ Kd e−Kd (t−τ ) dτ
m
t1

|ḋ|max (1 − e−Kd t ) (|x1 |max |ϑ̃1 |max + |x2 |max |ϑ̃2 |max )e−Kd t
≤ e−Kd t d̃(0) + +
Kd m
ε(|x1 |max + |x2 |max )
+ . (6.53)
m
110 Nonlinear Control Techniques for EHAs in Robotics Engineering

Therefore, as t → ∞,

|ḋ|max ε(|x1 |max + |x2 |max )

d̃(t) ≤ + . (6.54)
Kd m

The large observer gain Kd can reduce the DO error d̃(t). Furthermore, t1 is
sufficiently large and is arbitrarily small, which means d̃(t) is arbitrarily
reduced by the observer and estimated gains parameters.

Remark 6.4
The DO d̂ (6.48) is not different from the parametric adaptive estimation law
ϑ 3 , since this DO is directly designed based on the state equations. How-
ever, the parametric adaptive estimation law is derived in the backstepping
controller iteration.

6.4 Experiment
In this chapter, the two-DOF robotic arm is employed to implement and
test the performance of the proposed control method. Meanwhile, several
working conditions will be considered when both robotic arms are driven
simultaneously, or either one joint is run instead. The specific parameters and
brands of the main components in the experimental architecture are listed in
Table 6.1.
Some hydraulic and mechanical parameters of this EHS are shown in
Table 6.2. Since the hydraulic parameters Cd , w, ρ, β e , K, b, and Ctl are
not obtained exactly, which have some uncertainties in different exper-
imental conditions, it is necessary to estimate the uncertain parameters

TABLE 6.1
Specific Parameters and Brand of Main Components
Element Type Marks Quantity

Servo motor BSM63N-375 BALDOR 1

Fixed displacement pump TFH-315 Takako 1
Servo valve D633-R02K01M0NSM2 Moog 2
Hydraulic cylinder LB6-1610-0080-4M Hoerbiger 2
Relative encoder AEDA-3300-BE1 AVAGO 2
Pressure sensor M3041-000006-350BG MEAS 4
Parametric Adaptive Control Method 111

ϑ i (i = 1, 2, 4, 5, 6) by the proposed parametric adaptive estimation law. How-

ever, the approximate values C̄d , w̄, ρ̄, β̄e , K̄, b̄, and C̄tl can be preset from some
hydraulic references as shown in Table 6.2. If the proposed parametric adap-
tive estimation law is not adopted, the approximate values ϑ̄i (i = 1, 2, 4, 5, 6)
without considering parametric uncertainties can also be used in the back-
stepping controller instead of the estimation values. From the parametric
definition in Equation 6.8, the known nominal values are given by

⎧
⎪ ϑ̄1 = K̄
⎪
⎪
⎪
⎪
⎪
⎪ ϑ̄2 = b̄
⎪
⎪
⎪
⎪
⎪
⎪ 4β̄e Ap
⎪
⎨ ϑ̄4 =
Vt . (6.55)
⎪
⎪
⎪
⎪ 4β̄e C̄tl
⎪
⎪ ϑ̄5 =
⎪
⎪ Vt
⎪
⎪
⎪
⎪
⎪
⎪ 4β̄ C̄ w̄
⎩ ϑ̄6 = e √d
Vt ρ̄

Thus, the simplified backstepping controller is redesigned as follows:

⎧ Tsv x4 Tsv Tsv

⎪
⎪ u=− c4 z4 + + α̇3 − z3 ϑ̄6 ps − tanh(kx4 )x3
⎪
⎪ Ksv Ksv Ksv Ksv
⎪
⎪
⎪
⎪ m
⎪
⎪ α̇2 = − (z2 + α1 − ẋ1d ) − c2 z3 − c2 α2
⎪
⎪
⎪
⎪ Ap
⎪
⎪
⎪
⎪ c2 m c1 m ¯
⎪
⎪ + (−ϑ̄1 x1 /m − ϑ̄2 x2 /m − ϑ3 /m + c1 x2 − c1 ẋ1d − ẍ1d ) − ẋ2
⎪
⎨ Ap Ap
m .
⎪
⎪ z3 − (c2 z2 + z1 ) + c3 (ẋ¯ 3 − α̇2 ) + ϑ̄4 ẋ¯ 2 + ϑ̄5¯˙x3 + α̈2
⎪
⎪ Ap
⎪
⎪ α̇3 = − 
⎪
⎪ ϑ̄6 ps − tanh(kx4 )x3
⎪
⎪
⎪
⎪
⎪
⎪ Ap
⎪
⎪ ẋ¯ 2 = x3 − ϑ̄1 x1 /m − ϑ̄2 x2 /m − ϑ3 /m
⎪
⎪ m
⎪
⎪
⎪
⎩
ẋ¯ 3 = − ϑ̄4 x2 − ϑ̄5 x3 + ϑ̄6 ps − tanh(kx4 )x3 x4
(6.56)

The stroke of the cylinder Smax is 79 mm, but in the experiment, the dis-
placement of the cylinder is not more than 58 mm to avoid the boundary
collision of the robotic arm.
The initial values of the estimated parameters are predefined as zero, that
is, ϑ̂i0 = 0, (i = 1, . . . , 6) to verify the convergence effectiveness of the para-
metric estimation laws. Some control parameters are designed as follows:
112 Nonlinear Control Techniques for EHAs in Robotics Engineering

TABLE 6.2
Hydraulic Parameters Used in Experiments
Parameter Value Parameter Value

C̄d 0.62 w̄ 0.024 m

xv max 7.9 mm Smax 79 mm
ps 40 bar Ap 2.01 cm2
Vt 1.74 × 10−5 m3 β̄e 2.2 × 108 Pa
Ksv 7.9 × 10−4 m/V Tsv 12 ms
K̄ 1000 N/m b̄ 2200 Ns/m
C̄tl 2.5 × 10−11 m3 /(s · Pa) ρ̄ 800 kg/m3
m1 1.772 kg m2 0.739 kg
mf 1 kg I1 0.071 kgm2
I2 0.015 kgm2 I2f 0.022 kgm2
m2f 1.739 kg m1f 3.511 kg
P1 P2 0.35 m P1 Pm1 0.16 m
P2 Pm2 0.12 m ε m1 7.9°

1. The constants: k1 = 10, k2 = 0.1, k3 = 10−3 , k4 = 10−6 , k5 = 10 − 9 ,

k6 = 10−5
2. The constants: c1 = 10, c2 = 1, c3 = 10−5 , c4 = 103
3. The constants: | ϑ̇3 |max = 5 × 103 , | α̇2 |max = 5 × 107 , | α̇3 |max =
0.01
4. The filter parameters: λ = 0.7, Tc = 10−3 s

6.4.1 Result of the Proposed Controller

In order to verify the proposed parametric adaptive backstepping controller
in Equation 6.46, two sinusoidal demands of the cylinder displacement
are selected as xs1d = 29 sin(0.6π t) mm and xe1d = 29 sin(π t) mm. The four
state responses, six parametric estimations, and two control voltages of the
proposed controllers are shown in Figures 6.2 through 6.4.
Figure 6.2a shows the good displacement tracking responses of two sinu-
soidal demands by the proposed controllers. The load pressures measured
by pressure sensors are not more than the supply pressure 40 bar shown in
Figure 6.2c. Owing to the larger dynamic load on the shoulder actuator, the
motion frequency of the shoulder should be lower than that of the elbow.
During time slice (0, 0.8 s), the control saturation arises since the initial track-
ing error is significant. After 0.8 s, the spool positions of two servo valves do
not exceed the saturation ± xvmax , which are similar to the dynamic charac-
teristic of two control voltages us , ue shown in Figure 6.4. In the experiment,
the robotic arm motion will generate resonance effect when the tracking ten-
dency of two joint angles approach each other. Especially in some time such
Parametric Adaptive Control Method 113

(a) 50 (b)
30 0.1

x 2s (m/s)
x1 (mm)

0
–30 s 0
–50 x1d
x1s –0.1
0 5 10 15 20 0 5 10 15 20
t (s) t (s)
50 0.2
30
x1 (mm)

x 2e (m/s)
0 0.1
–30 0
–50 xe1d
xe1 –0.1
0 5 10 15 20 0 5 10 15 20
t (s) t (s)
(c) (d) 10
60 8
5

x 4s (mm)
x 3s (bar)

40 0
20 –5
0 –10
0 5 10 15 20 0 5 10 15 20
t (s) t (s)
60 10
8
x 4e (mm)
x 3e (bar)

40 5

20 0

0 –5
0 5 10 15 20 0 5 10 15 20
t (s) t (s)

FIGURE 6.2
Four state responses of the EHS by the proposed controller. (a) Cylinder displacement response
x1 . (b) Cylinder velocity x2 . (c) Load pressure of cylinder x3 . (d) Spool position of servo valve x4 .

as 8, 11, and 18 s, two hydraulic actuators need simultaneously suffering

two maximum external loads; the instant control supplement is significant
as shown in Figure 6.4a and b.
The six uncertain parameters are estimated, respectively, by the parametric
adaptive estimation laws (6.18), (6.24), and (6.41) shown in Figure 6.3. The
parametric estimation ϑ̂1 approaches the nominal value ϑ̄1 , which denotes
that the load spring constant is almost consistent with the approximate preset
values. The other four parametric estimations ϑ̂2 , ϑ̂4 , ϑ̂5 , and ϑ̂6 also con-
verge to the respective steady-state values like ϑ̂1 . However, these parametric
estimations have obviously deviated from the approximate preset values,
respectively. This phenomenon indicates that the permanent parameters ϑ̄2 ,
ϑ̄4 , ϑ̄5 , and ϑ̄6 cannot show the actual hydraulic parameters Cd , w, ρ, β e , b,
and Ctl with some uncertainties in different working conditions. Different
from the steady characteristic of the other five parameters, ϑ̂3s , ϑ̂3e are dynamic
estimations, which described two variable external loads of the hydraulic
actuators caused by the two-DOF robotic arm motion. If the parametric esti-
mated errors of ϑ̂i (i = 2, 4, 5, 6) are significant, the static control bias emerges
in long time shown in the initial 1 s in Figure 6.5f. If the dynamic estimations
of ϑ̂3 are inaccurate, the dynamic control saturation emerges in Figure 6.6f.
114 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) 1200 (b) 3000

1000 2500
800 2000

ϑ̂2
ϑ1

600 1500
400 1000
200 ϑ̂1 500 ϑ̂1
– –
ϑ1 ϑ2
0 0
0 5 10 15 20 0 5 10 15 20
t (s) t (s)

(c) 1400 (d) × 109

12
1200
10
1000
8
800
ϑ̂3

ϑ̂4
600 6

400 4

200 2 ϑ̂4
–
ϑ4
0 0
0 5 10 15 20 0 5 10 15 20
t (s) t (s)

(e) × 1010
1500 (f ) 3

2.5
1000 2
ϑ̂6

1.5
ϑ̂5

500
1
ϑ̂5
– ϑ̂6
ϑ5 0.5 –
0 ϑ6
0 5 10 15 20 0
t (s) 0 5 10 15 20
t (s)

FIGURE 6.3
Six estimation values by parametric adaptive estimation laws. (a) Uncertainty parameter ϑ̂1 .
(b) Uncertainty parameter ϑ̂2 . (c) Uncertainty parameter ϑ̂3 . (d) Uncertainty parameter ϑ̂4 .
(e) Uncertainty parameter ϑ̂5 . (f) Uncertainty parameter ϑ̂6 .

6.4.2 Comparison with Simplified Backstepping Controller

To illustrate the problem, the comparison results for two different backstep-
ping controllers are given in two critical conditions where the external load
of the hydraulic actuator is close to the limitation. The proposed controller
(6.46) involves the parametric estimation laws and the filtering estima-
tion. The other simplified backstepping controller (6.56) has no parametric
estimation laws and no filtering estimation, where the approximate preset
parameters ϑ̄1 , ϑ̄2 , ϑ̄4 , ϑ̄5 , and ϑ̄6 are determined by Equation 6.55, ϑ 3 is
computed by Equation 6.35, without used Equations 6.34 and 6.41. Differ-
ent from the above experimental condition, the elbow joint is fixed and the
Parametric Adaptive Control Method 115

(a) (b)
10 10
8
6
4
5
2

ue (V)
us (V)

0
–2
–4 0

–6
–8
–10 –5
0 5 10 15 20 0 5 10 15 20
t (s) t (s)

FIGURE 6.4
Control voltages of two hydraulic actuators by the proposed controller. (a) The control voltage
of shoulder controller us . (b) The control voltage of elbow controller ue .

displacement demands of shoulder are selected as xs1d = 14.5 sin(2π t) mm,

xs1d = 29 sin(π t) mm. The comparison results are shown in Figures 6.5 and
6.6. The derivatives α̇2 , α̇3 obtained by the proposed filter are significantly
smaller than the computation values by Equation 6.56 as shown in Figures
6.5a–d. This experimental phenomenon denotes that the proposed filter has
more capability to suppress the violent derivative of virtual control than the
conservative approach based on model computation. Figure 6.5e shows that
the proposed controller has higher tracking performance, since the para-
metric estimation laws are used to adapt the actual hydraulic parameter
with uncertainties. On the contrary, the simplified backstepping controller
(6.56) with certain parametric assumptions decline the dynamic behavior
of the closed-control loop and results in the sharp controller as shown
in Figure 6.5f.
Especially in large stroke motion experiment of the shoulder actuator as
shown in Figure 6.6, the simplified backstepping controller could not obtain
the satisfactory tracking performance. However, to some extent, the pro-
posed filter still guarantees the dynamic response of the shoulder actuator.
These results show that the large stroke motion is easier to cause the saturated
and sharp controller than the high-frequency motion due to the dynamic
external load on the shoulder actuator.

6.5 Result of Disturbance Observer

The DO (6.48) can replace the estimation law (6.41) since the two external
loads on two hydraulic actuators are dynamic variables. In simulation, the
external loads FL1 and FL2 on two EHAs are computed by Lagrange equation
using some mechanical parameters of the robotic arm. This two values can
116 Nonlinear Control Techniques for EHAs in Robotics Engineering

(a) × 106 (b) × 106

6
6
4
4
αˆ2

α2
2 2
0 0
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)
× 108 2 × 10
9

1
1
0.5
0

α2
0
αˆ2

–0.5 –1
–1 –2
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)

(c) 1 × 10
–4 (d) 5 × 10
–4

0 0
α̂3

–1 α3 –5
–2 –10
–15
2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
t (s) t (s)
0.01
0.05
0.005
0
α̂3

0
α3

–0.005
–0.01 –0.05
2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
t (s) t (s)
(e) 20 (f ) 10 Proposed controller
15 Simplified controller
10 8
6
0 4
−10 2
x1 (mm)

us (V)

−15 0
−20 −2
−4
−30
−6
−40 Demand −8
Proposed controller
Simplified controller −10
−50
0 2 4 6 8 10 0 2 4 6 8 10
t (s) t (s)

FIGURE 6.5
Comparison result for the demand input xs1d = 14.5 sin(2π t) mm. (a) α 2 and α̇2 by the pro-
posed filter. (b) α 2 and α̇2 by Equation 6.56. (c) α 3 and α̇3 by the proposed filter. (d) α 3 and
α̇3 by Equation 6.56. (e) Cylinder displacement response x1 . (f) The control voltage of shoulder
controller us .

be selected as the demand of the DO. Then the external load disturbances
on two EHAs are estimated by the DO (6.48) as shown in Figure 6.7. This
high-gain observer guarantees the observer errors d̃1 and d̃2 convergence to
zero. If the observer error is obvious, the controller will degrade the dynamic
tracking performance.
The load disturbance estimations on two hydraulic actuators by the pro-
posed controller are shown in Figure 6.8, which shows that the external
Parametric Adaptive Control Method 117

(a) × 106 (b) × 106

6 6
4 4
α̂2

α2
2 2
0 0
0 5 10 15 20 0 5 10 15 20
t (s) t (s)
× 108 × 109
1 2
0.5
0
0
αˆ2

α2
–0.5 –2
–1
0 5 10 15 20 0 5 10 15 20
t (s) t (s)

(c) × 10–4 (d) × 103

1 2
0 0
–1 –2
α̂3

–2 α3 –4
–3 –6
–4 –8
2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20
t (s) t (s)
0.01
0.15
0.005 0.1
0
αˆ3

α3

0.05
–0.005 0
–0.01 –0.05
2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20
t (s) t (s)
(e) 40
(f ) Proposed controller
Simplified controller
30 10
20
10 5
us (V)
x1 (mm)

0
−10 0
−20
−30 –5
Demand
−40 Proposed controller
Simplified controller –10
−50
0 5 10 15 20 0 5 10 15 20
t (s) t (s)

FIGURE 6.6
Comparison result for the demand input xs1d = 29 sin(π t) mm. (a) α 2 and α̇2 by the proposed
filter. (b) α 2 and α̇2 by Equation 6.56. (c) α 3 and α̇3 by the proposed filter. (d) α 3 and α̇3
by Equation 6.56. (e) Cylinder displacement response x1 . (f) The control voltage of shoulder
controller us .

load on the upper arm is greater than that on the forearm, although the fre-
quency of the former is smaller than that of the latter. These two disturbance
estimations are similar to the simulation results.
The simulation results of position tracking error by two controllers are
shown in Figure 6.9. The proposed controller represents the proposed para-
metric adaptive controller (6.46) with the parametric estimation laws (6.18),
(6.24), and (6.41). The other controller combines the simplified backstepping
controller (6.56) with the DO (6.48) and the parametric estimation laws (6.18),
118 Nonlinear Control Techniques for EHAs in Robotics Engineering

200
d1
100
dˆ1
d1 (m/s2)

−100

−200
0 2 4 6 8 10
Time (s)
200
d2
100
dˆ2
d2 (m/s2)

−100

−200
0 2 4 6 8 10
Time (s)
10
d̃1
5
d˜ (m/s2)

d̃2
0

−5

−10
0 2 4 6 8 10
Time (s)

FIGURE 6.7
Simulation results of load disturbance estimation.

(6.24) as follows:
⎧
⎪ Tsv x4 Tsv Tsv
⎪
⎪ u=− c4 z4 + + α̇3 − z3 ϑ̂6 ps − tanh(kx4 )x3
⎪
⎪ Ksv Ksv Ksv Ksv
⎪
⎪
⎪
⎪ m
⎪
⎪ α̇2 = − (z2 + α1 − ẋ1d ) − c2 z3 − c2 α2
⎪
⎪
⎪
⎪ A p
⎪
⎪
⎪
⎪ c m c1 m ¯
⎪
⎪ +
2
(−ϑ̂1 x1 /m − ϑ̂2 x2 /m − d̂ + c1 x2 − c1 ẋ1d − ẍ1d ) − ẋ2
⎪
⎨ Ap Ap
.
⎪
⎪
⎪ z3 − Amp (c2 z2 + z1 ) + c3 (ẋ¯ 3 − α̇2 ) + ϑ̂4 ẋ¯ 2 + ϑ̂5 ẋ¯ 3 + α̈2
⎪
⎪ α̇3 = −
⎪
⎪

⎪
⎪ ϑ̂6 ps − tanh(kx4 )x3
⎪
⎪
⎪
⎪ Ap
⎪¯
⎪
⎪
⎪ ẋ2 = x3 − ϑ̂1 x1 /m − ϑ̂2 x2 /m − d̂
⎪
⎪ m
⎪
⎪
⎪
⎩ ẋ¯ = − ϑ̂ x − ϑ̂ x + ϑ̂ p − tanh(kx )x x
3 4 2 5 3 6 s 4 3 4
(6.57)
Parametric Adaptive Control Method 119

150

100
Disturbance estimation (m/s2)

50

−50

−100

−150 dˆ1
dˆ2
−200
0 2 4 6 8 10
Time (s)

FIGURE 6.8
Experimental results of the load disturbance estimation on two EHAs.

3
Controller 1
2 Controller 2
1
Δy1 (mm)

0
−1
−2
−3
0 2 4 6 8 10
Time (s)
3
Controller 1
2 Controller 2
1
Δy2 (mm)

0
−1
−2
−3
0 2 4 6 8 10
Time (s)

FIGURE 6.9
Simulation results of position tracking error by two controllers, y1 —upper arm error, y2 —
forearm error.
120 Nonlinear Control Techniques for EHAs in Robotics Engineering

Remark 6.5
It is to be noted that the controller (6.57) is different from the simplified
backstepping controller (6.56). In Equation 6.57, the parametric adaptive esti-
mation law can also be used to estimate ϑ i (i = 1, 2, 4, 5, 6). Furthermore, the
high-gain disturbance observer (6.48) can be adopted to estimate the distur-
bance d̂, which can be directly compensated in the backstepping design.

The performance of the backstepping controller with the DO (emax =

0.15 mm) is better than the proposed controller without the DO (emax =
1.13 mm) as shown in Figure 6.9. This means the DO is constructed to
improve the dynamic behavior of the designed controller. The controller can
eliminate the bias caused by the unknown dynamic external load.
Figure 6.10 shows that the dynamic tracking performance of the back-
stepping controller with the DO (emax = 1.17 mm, σ e = 0.59 mm) is also
better than the proposed controller without the DO (emax = 2.89 mm,
σ e = 1.53 mm). This result denotes that the tracking performance is declined
as the current external load disturbance on the upper arm approaches its
limitation (d̂1 = 150 mm/s2 ) as shown in Figure 6.8. Thus, the backstepping
controller with the DO constructs the HGDO to compensate largely dynamic
load disturbance and to avoid the obvious error of the position response.

4
Controller 1
Controller 2
2
Δy1 (mm)

−2

−4
0 2 4 6 8 10
Time (s)
4
Controller 1
2 Controller 2
Δy2 (mm)

−2

−4
0 2 4 6 8 10
Time (s)

FIGURE 6.10
Experimental results of position tracking error by two controllers, y1 —upper arm error, y2 —
forearm error.
References

1. K. Ahn, D. Nam, and M. Jin. Adaptive backstepping control of an elec-

trohydraulic actuator. IEEE/ASME Transactions on Mechatronics, 19(3):987–995,
2014.
2. A. Alleyne and R. Liu. A simplified approach to force control for electro-
hydraulic systems. Control Engineering Practice, 8(12):1347–1356, 2000.
3. A. Alleyne and R. Liu. Systematic control of a class of nonlinear systems with
application to electrohydraulic cylinder pressure control. IEEE Transactions on
Control System Technology, 8(4):623–634, 2000.
4. G. Balas and J. Doyle. Robustness and performance trade-offs in control design
for flexible structures. IEEE Transactions on Control Systems Technology, 2(4):352–
361, 1994.
5. J. Bobrow and K. Lum. Adaptive, high bandwidth control of a hydraulic actu-
ator. ASME Journal of Dynamic System, Measurement, Control, 118(4):714–720,
1996.
6. I. Boiko. Variable-structure PID controller for level process. Control Engineering
Practice, 21(5):700–707, 2013.
7. F. Bu and B. Yao. Observer based coordinated adaptive robust control of robot
manipulators driven by single-rod hydraulic actuators. In Proceedings of 2000
IEEE-ICRA, San Francisco, CA, pp. 3034–3039. IEEE, 2000.
8. F. Bu and B. Yao. Nonlinear model based coordinated adaptive robust control of
electro-hydraulic robotic arms via overparametrizing method. In Proceedings of
2001 IEEE-ICRA, Seoul, Korea, pp. 3459–3464. IEEE, 2001.
9. W. Chen. Disturbance observer based control for nonlinear systems. IEEE/ASME
Transactions on Mechatronics, 9(4):706–710, 2004.
10. J. Doyle, K. Glover, P. Khargonekar, and B. Francis. State-space solutions to
standard H2 and H∞ control problems. IEEE Transactions on Automatic Control,
34(8):831–847, 1989.
11. C. Du, A. Plummer, and D. Johnston. Variable pressure valve-controlled
hydraulic actuation for a robotic arm. In Proceedings of 9th International Fluid
Power Conference, Aachen, Germany, pp. 186–197, 2014.
12. C. Du, A. Plummer, and D. Johnston. Performance analysis of an energy-efficient
variable supply pressure electro-hydraulic motion control system. Control Engi-
neering Practice, 48:10–21, 2016.
13. R. Fales and A. Kelkar. Robust control design for a wheel loader using H∞ and
feedback linearization based methods. ISA Transactions, 48(3):313–320, 2009.
14. M. Fan, A. Tits, and J. Doyle. Robustness in the presence of mixed parametric
uncertainty and unmodeled dynamics. IEEE Transactions on Automatic Control,
36(1):25–38, 1991.
15. D. Gu, P. Petkov, and M.M. Konstantinov. Robust Control Design with MATLAB.
Springer-Verlag, London, 2005.
16. C. Guan and S. Pan. Adaptive sliding mode control of electro-hydraulic system
with nonlinear unknown parameters. Control Engineering Practice, 16(11):1275–
1284, 2008.

121
122 References

17. C. Guan and S. Pan. Nonlinear adaptive robust control of single-rod electro-
hydraulic actuator with unknown nonlinear parameters. IEEE Transactions on
Control System Technology, 16(3):434–445, 2008.
18. Q. Guo, P. Sun, J. Yin, T. Yu, and D. Jiang. Parametric adaptive estimation and
backstepping control of electro-hydraulic actuator with decayed memory filter.
ISA Transactions, 62(S1):202–214, 2016.
19. Q. Guo, T. Yu, and D. Jiang. High-gain observer-based output feedback con-
trol of single-rod electro-hydraulic actuator. IET Control Theory and Applications,
9(16):2395–2404, 2015.
20. Q. Guo, T. Yu, and D. Jiang. Robust H∞ positional control of 2-DOF robotic
arm driven by electro-hydraulic servo system. ISA Transactions, 59(11):55–64,
2015.
21. Q. Guo, Y. Zhang, B. Celler, and S. Su. Backstepping control of electro-
hydraulic system based on extended-state-observer with plant dynamics largely
unknown. IEEE Transactions on Industrial Electronics, 63(11):6909–6920, 2016.
22. Q. Guo, Y. Zhang, and D. Jiang. A control approach for human-mechatronic-
hydraulic-coupled exoskeleton in overload-carrying condition. International
Journal of Robotics and Automation, 31(8):272–280, 2016.
23. W. He, Y. Chen, and Z. Yin. Adaptive neural network control of an uncertain
robot with full-state constraints. IEEE Transactions on Cybernetics, 46(3):620–629,
2016.
24. W. He, Y. Dong, and C. Sun. Adaptive neural impedance control of a robotic
manipulator with input saturation. IEEE Transactions on Systems, Man, and
Cybernetics: Systems, 46(3):334–344, 2016.
25. W. He and S. Ge. Vibration control of a flexible string with both boundary
input and output constraints. IEEE Transactions on Control Systems Technology,
23(4):1245–1254, 2014.
26. W. He and S. Ge. Cooperative control of a nonuniform gantry crane with
constrained tension. Automatica, 66(4):146–154, 2016.
27. W. He and S. Zhang. Control design for nonlinear flexible wings of a robotic
aircraft. IEEE Transactions on Control Systems Technology, 25(1):351–357, 2017.
28. W. He, S. Zhang, and S. Ge. Adaptive control of a flexible crane system
with the boundary output constraint. IEEE Transactions on Industrial Electronics,
61(8):4126–4133, 2014.
29. W. He, S. Zhang, and S. Ge. Robust adaptive control of a thruster assisted
position mooring system. Automatica, 50(7):1843–1851, 2014.
30. C. Johnson and R. Lorenz. Experimental identification of friction and its com-
pensation in precise, position controlled mechanisms. IEEE Transactions on
Industry Applications, 28(6):1392–1398, 1992.
31. C. Kaddissi, J. Kenne, and M. Saad. Identification and real-time control of an
electrohydraulic servo system based on nonlinear backstepping. IEEE/ASME
Transactions on Mechatronics, 12(1):12–22, 2007.
32. H. Khalil. Nonlinear Systems (3rd edition). Prentice-Hall, Englewood Cliffs, NJ,
2001.
33. W. Kim, D. Shin, D. Won, and C. C. Chung. Disturbance-observer-based
position tracking controller in the presence of biased sinusoidal disturbance
for electrohydraulic actuators. IEEE Transactions on Control System Technology,
21(6):2290–2298, 2013.
References 123

34. W. Kim, D. Won, and C. Chung. High gain observer-based nonlinear posi-
tion control for electro-hydraulic servo systems. In Proceedings of 2010 American
Control Conference, Baltimore, MD, pp. 1440–1446, 2010.
35. W. Kim, D. Won, and C. Chung. Output feedback nonlinear control for electro-
hydraulic systems. Mechatronics, 22(6):766–777, 2012.
36. W. Kim, D. Won, and M. Tomizuka. Flatness-based nonlinear control for position
tracking of electrohydraulic systems. IEEE/ASME Transactions on Mechatronics,
20(1):197–206, 2015.
37. M. Krstic, I. Kanellakopoulos, and P. Kokotovic. Nonlinear and Adaptive Control
Design. John Wiley & Sons, Inc., New York, NY, 1995.
38. G. Liu and S. Daley. Optimal-tuning PID controller design in the frequency
domain with application to a rotary hydraulic system. Control Engineering
Practice, 7(7):821–830, 1999.
39. V. Lu, K. Zhou, and J. Doyle. Stabilization of uncertain linear systems: An LFT
approach. IEEE Transactions on Automatic Control, 41(1):50–65, 1996.
40. N. Manring. Hydraulic Control Systems. John Wiley & Sons, Inc., New York, NY,
2005.
41. H. Merritt. Hydraulic Control Systems. John Wiley & Sons, Inc., New York, NY,
1967.
42. V. Milić, Ž. Šitum, and M. Essert. Robust H∞ position control synthesis of an
electro-hydraulic servo system. ISA Transactions, 49(4):535–542, 2010.
43. M. Moradi. Self-tuning PID controller to three-axis stabilization of a satellite
with unknown parameters. International Journal of Non-Linear Mechanics, 49:700–
707, 2013.
44. P. Nakkarat and S. Kuntanapreeda. Observer-based backstepping force control
of an electrohydraulic actuator. Control Engineering Practice, 17(8):895–902, 2009.
45. N. Niksefat and N. Sepehri. Design and experimental evaluation of a robust
force controller for an electro-hydraulic actuator via quantitative feedback
theory. Control Engineering Practice, 8(12):1335–1345, 2000.
46. A. Packard, M. Fan, and J. Doyle. A power method for the structured singular
value. In Proceedings of the 27th Conference of Decision and Control, Austin, TX, pp.
2132–2137. IEEE, 1988.
47. H. Pan, W. Sun, H. Gao, and X. Jing. Disturbance observer-based adaptive track-
ing control with actuator saturation and its application. IEEE Transactions on
Automation, Science and Engineering, 13(2):868–875, 2016.
48. Y. Pi and X. Wang. Observer-based cascade control of a 6-DOF parallel hydraulic
manipulator in joint space coordinate. Mechatronics, 20(6):645–655, 2010.
49. A. Plummer and N. Vaughan. Robust adaptive control for hydraulic servo sys-
tems. ASME Journal of Dynamic System, Measurement, Control, 118(2):237–244,
1996.
50. Y. Qiu, X. Liang, and Z. Dai. Backstepping dynamic surface control for an anti-
skid braking system. Control Engineering Practice, 42:140–152, 2015.
51. C. Semini. The report of production display for “HyQ robot” made in the
Dynamic Legged Systems Lab, Istituto Italiano di Tecnologia (IIT), Genova, Italy,
http://www.iit.it/en/advrlabs/dynamic-legged-systems.html.
52. X. Song, Y. Wang, and Z. Sun. Robust stabilizer design for linear time-
varying internal model based output regulation and its application to an electro
hydraulic system. Automatica, 50(4):1128–1134, 2014.
124 References

53. H. Sun and G. Chiu. Nonlinear observer based force control of electro-hydraulic
actuators. In Proceedings of 1999 American Control Conference, San Diego, Califor-
nia, pp. 764–768, 1999.
54. D. Swaroop, P. Hedrick, J. Yip, and J. Gerdes. Dynamic surface control for a class
of nonlinear systems. IEEE Transactions on Automatic Control, 45(10):1893–1899,
2000.
55. I. Ursu, A. Toader, A. Halanay, and S. Balea. New stabilization and tracking
control laws for electrohydraulic servomechanisms. European Journal of Control,
19(1):65–80, 2013.
56. I. Ursu, F. Ursu, and F. Popescu. Backstepping design for controlling electrohy-
draulic. Journal of the Frankin Institute, 343(1):94–110, 2006.
57. W. Gawronski. Balanced systems and structures: Reduction, assignment, and
perturbations. Control and Dynamic Systems, 54:372–415, 1992.
58. D. Wang and J. Huang. Neural network-based adaptive dynamic surface con-
trol for a class of uncertain nonlinear systems in strict-feedback form. IEEE
Transactions on Neural Networks, 16(1):195–202, 2005.
59. D. Won, W. Kim, D. Shin, and C. Chung. High-gain disturbance observer-based
backstepping control with output tracking error constraint for electro-hydraulic
systems. IEEE Transactions on Control System Technology, 23(2):787–795, 2015.
60. B. Yao and F. Bu. Adaptive robust motion control of single-rod hydraulic
actuators theory and experiments. IEEE/ASME Transactions on Mechatronics,
5(1):79–91, 2000.
61. J. Yao, Z. Jiao, and D. Ma. Extended-state-observer-based output feedback non-
linear robust control of hydraulic systems with backstepping. IEEE Transactions
on Industrial Electronics, 61(11):6285–6293, 2014.
62. J. Yao, Z. Jiao, and D. Ma. High-accuracy tracking control of hydraulic rotary
actuators with modeling uncertainties. IEEE/ASME Transactions on Mechatronics,
19(2):633–641, 2014.
63. J. Yao, Z. Jiao, Y. Shang, and C. Huang. Adaptive nonlinear optimal compen-
sation control for electro-hydraulic load simulator. China Journal of Aeronautics,
23(6):720–733, 2010.
64. H. Yu, Z. Feng, and X. Wang. Nonlinear control for a class of hydraulic servo
system. Journal of Zhejiang University Science, 5(11):1413–1417, 2004.
65. P. Zarchan. Tactical and Strategic Missile Guidance. AIAA, Reston, VA, 1997.
66. J. Zhao, J. Wang, and S. Wang. Fractional order control to the electrohydraulic
system in insulator fatigue test device. Mechatronics, 23(7):828–839, 2010.
67. K. Zhou, J. Doyle, and K. Glover. Robust and Optimal Control. Prentice-Hall Inc.,
Englewood Cliffs, NJ, 1996.
Index

A D
Actuators, 1; see also Electro-hydraulic Decayed memory filter, 97
servo systems design, 104–106
Adaptive control, 2 Disturbance
Amplitude margin, 20 effect, 20
Arm hydraulic actuator rejection, 3
sinusoidal response of, 91 Disturbance observer (DO), 2
step response experiment, 89 with general nonlinear controller, 5
Asymmetrical cylinder model, 9 in parametric adaptive control
method, 115–120
DO, see Disturbance observer
B Double-rod acting mechanism, 8
Dynamic control voltages of two servo
Backstepping control, 4 valves, 44
controller, 86 Dynamic model of EHS, 98; see also
general, 100–104 Parametric adaptive control
procedure, 82 method
external load, 99
flow-pressure continuous equation of
C hydraulic cylinder, 98
fourth state space model, 99
Closed loop; see also Open-loop load flow of servo valve, 98
from load disturbance, 19 one-order linear model, 98
with robust performance sgn(.) function, 98
requirements, 53
transfer function, 18
Command delay, 10 E
Complementary sensitivity function, 61 EHA, see Electro-hydraulic actuator
Control method, 5; see also Parametric EHSs, see Electro-hydraulic servo
uncertainty problem systems
external load, 4–5 Electro-hydraulic actuator (EHA), 1
linear PID controller, 5 linearized model of, 14–16
output feedback control method, 6 Electro-hydraulic control system, 7; see
robust controller, 5 also Hydraulic cylinder model;
state feedback control method, 6 Servo valve
Control voltage of two servo valves nonlinear state-space model, 12–14
for sinusoidal demand, 65 parametric uncertainty and load
for square demand, 64 disturbance, 11–12
Coordinated motion experiment of Electro-hydraulic servo systems (EHSs),
robotic arm joints, 95, 96 1, 8, 69; see also Control method;
Cross-linked feedback system, 56 Parametric uncertainty
Cylinder problem
position feedback control loop; see adaptive control, 2
also Linear feedback control control methods, 5
loop external load on, 3
position response in time domain, 62 geometric control approach, 3
volume, 48 H∞ control methods, 3

125
126 Index

Electro-hydraulic servo systems (EHSs) High gain of observer, 75, 76

(Continued) Hydraulic cylinder model, 7, 23; see also
to improve behavior of, 3 Electro-hydraulic control
largely unknown load disturbance system model; Linear feedback
of, 4 control loop
linear classical control, 2 annulus area ratio and coefficient, 24
nonlinear control, 2 asymmetrical cylinder model, 9
nonlinear model of EHS, 5 cylinder volume, 23
problems and solutions in, 2 double-rod acting mechanism, 8
quantitative feedback theories, 3 flow-pressure coefficient of valve,
robust control, 2 23, 24
unmodeled uncertainties, 2 flow-pressure continuous equation
Equivalent parameter identification of, 98
method, 4 hydraulic parameters, 47
ESO, see Extended state observer load flow gain of valve, 24
Extended state observer (ESO), 2 mechanical dynamic equation, 9
External load, 4–5, 49 parameters of, 23
as disturbance to EHS, 54 pressure gain of valve, 23
in parametric adaptive backstepping rod chamber, 23
controller, 99 single-rod acting mechanism, 8
symmetrical cylinder model, 8–9
F
Feedback control method, 6 I
Fictitious proportional gains, 49–50 Initial load mass, 24
Flatness-based nonlinear controller, 5
Flow
equations of single-rod cylinder, 70 J
gain of servo valve, 48
Flow-pressure Joint angles, 49
coefficient of valve, 23, 24
continuous equation, 48, 98 K
continuous model, 71
Fourth state space model, 99 Kinetic energy of arm, 28
Full-state
error, 87
observer construction, 72–75 L
Lagrange equation of two-DOF robotic
G arm, 27
Largely unknown load disturbance, 4
General backstepping control design, LFT, see Linear fractional transformation
100–104 Linear classical control, 2
Geometric control approach, 3 Linear feedback control loop, 17, 18; see
Gravitational potential energy of arm, 28 also Hydraulic cylinder model;
Linear PID control design
amplitude margin, 20
H
closed loop from load disturbance, 19
H∞ control methods, 3 closed-loop transfer function, 18
HGDO, see High-gain disturbance disturbance effect, 20
observer open-loop transfer function, 18
High-gain disturbance observer phase margin, 20
(HGDO), 3 Linear fractional transformation
with backstepping control, 5 (LFT), 50
Index 127

Linearized model of electro-hydraulic N

actuator, 14–16
Nonlinear backstepping control, 81; see
Linear PID control design, 17; see also
also Output feedback control
Linear feedback control loop;
model of EHS; Parametric
Mechanical motion model;
adaptive backstepping
Proportional integral derivative
controller
control; System performance
backstepping controller, 86
analysis
backstepping procedure, 82
controller, 5
controller design with observer, 84–86
open-loop transfer function for, 19
coordinated motion of robotic arm
Linear uncertain state-space model, 51
joints, 95, 96
Load
experiment on, 88
disturbance estimation, 118
full-state error, 87
rotation, 24
Lyapunov candidate function, 82, 83
Load flow
observer error dynamics, 87
gain of valve, 24, 47
PI and backstepping method
model, 10 comparison, 91, 93, 94
of servo valve, 98 pressure error, 82
Load mass, initial, 24 sinusoidal response of upper arm
Lyapunov candidate function, 82, 83 hydraulic actuator, 91
stability discussion of EHS, 86
M state errors, 88
state estimations, 90, 92
Maximum relative uncertainties, 49–50 step response experiment of upper
Measurement noise, 53 arm hydraulic actuator, 89
Mechanical dynamic equation, 9, 71 Nonlinear control, 2
Mechanical motion model, 24; see also Nonlinear DO integrated with general
Two-link dynamic model nonlinear controller, 5
distance from centroid forearm to Nonlinear model of EHS, 5, 72
elbow, 26 Nonlinear state-space model, 12–14
equivalent moment of inertia of
forearm, 26
framework for, 25 O
gravitational potential energy of Observer; see also Disturbance observer;
arm, 28 State observer design
initial load mass, 24 backstepping controller with, 84–86
kinetic energy of arm, 28 convergence, 76–81
Lagrange equation of two-DOF error dynamics, 87
robotic arm, 27 extended state, 2
load rotation, 24 full-state observer construction, 72–75
mechanical parameters of robotic HGDO, 3, 5
arm, 26 high gain of, 75, 76
moment of inertia, 24 One-order linear model, 98
motion control mechanism, 24, 25 Open-loop; see also Closed loop;
position vector of hinge point, 27 Open-loop control system
total energy of two-link system, 28–29 frequency domain
two cylinder dynamic lengths, 29, 30 system analysis, 54
velocity vector, 27 system frequency response, 54, 55
Mechatronics plant model, 20; see also transfer function, 18, 36
Mechanical motion model; Open-loop control system frequency
Servo valve model construction domain
Moment of inertia, 24 for shoulder actuator, 40
Motion control mechanism, 24, 25 for upper arm actuator, 38
128 Index

Output Parametric analysis of shoulder

feedback control method, 6 hydraulic actuator, 34
flow of pump, 32 Parametric estimation, 3
power of pump, 32 Parametric uncertainties, 2
pressure analysis of pump, 31–32 and load disturbance, 11–12
Output feedback control model of EHS, Parametric uncertainty analysis, 47; see
69; see also Electro-hydraulic also Robust controller design;
control system; State observer Robust model construction
design cylinder volume, 48
dynamics of servo valve, 70 external load, 49
flow equations of single-rod fictitious proportional gains, 49–50
cylinder, 70 flow gain of servo valve, 48
flow-pressure continuous model, 71 flow-pressure continuous
mechanical dynamic equation, 71 equation, 48
Robotic BigDog, 70 hydraulic parameters, 47
sgn(.) function, 70–71 joint angles, 49
sixth-order nonlinear dynamic model load flow gain of valve, 47
of EHS, 72 maximum relative uncertainties,
stick-slip friction, 71 49–50
two load flows, 70 structural uncertainty, 49
viscous damping coefficient of
cylinder, 48–49
P Parametric uncertainty problem, 2, 3; see
PAE law, see Parametric adaptive also Control method
estimation law disturbance rejection, 3
Parametric adaptive backstepping parametric estimation, 3
controller, 99; see also Nonlinear unknown parametric variation, 3
backstepping control; Phase margin, 20
Parametric adaptive control PI controller, see Proportional integral
method controller
decayed memory filter design, PID, see Proportional integral derivative
104–106 Pneumatic actuator, 1
dynamics of EHS, 100 Position tracking
general backstepping control design, error by two controllers, 118, 120
100–104 results of arm cylinder, 43
revised parametric adaptive control Position vector of hinge point, 27
law, 106–108 Pressure error, 82
Parametric adaptive control method, 97; Pressure gain of valve, 23
see also Dynamic model of EHS; Pressure of cylinder
Parametric adaptive with load, 31–34, 38
backstepping controller supporting shoulder, 33–34
application, 108–110 supporting the elbow, 34, 35, 36
compared to simplified backstepping with two joints in simultaneous
controller, 114–115, 116, 117 motion, 34, 37, 38
decayed memory filter, 97 Proportional integral controller (PI
disturbance observer, 115–120 controller), 3, 41; see also
experiment on, 110–114 Proportional integral derivative
load disturbance estimation, 118 control
position tracking error by two and backstepping method
controllers, 118, 120 comparison, 91, 93, 94
virtual control variables, 97 design for shoulder actuator, 39, 40
Parametric adaptive estimation law dynamic control voltages of two
(PAE law), 2 servo valves, 44
Index 129

position tracking results of arm four orders in control, 62

cylinder, 43 frequency domain result of robust
two chamber pressures of two controller, 61
cylinders, 44 frequency response of open-loop
Proportional integral derivative (PID), 5 system with uncertainties,
Proportional integral derivative control, 54, 55
34; see also Proportional integral input and output, 53–54
controller linear uncertain state-space model, 51
frequency domain characteristic of measurement noise, 53
open-loop control system, with parametric and structural
38, 40 uncertainties, 52
linear, 5 robust performance with uncertainty
open-loop transfer function, 36 parameters, 60
performance for actuator, 39, 40 robust stability, 60
supply pressure, 35, 36 singular values of inverse function, 58
upper arm cylinder, 37, 41 Small Gain Theorem, 56–57
structural uncertainty parameters, 51
Q weight function design, 55–58
Robust model construction, 50–54; see
Quantitative feedback theories, 3 also Parametric uncertainty
analysis; Robust control;
R Sinusoidal response
Revised parametric adaptive control law,
106–108 S
Robotic arm
coordinated motion of joints, 95, 96 Servo motor system, 1
mechanical parameters, 26 Servo valve, 9; see also Electro-hydraulic
Robotic BigDog, 70 control system model
Robust control, 2, 45; see also Robust command delay, 10
controller design; Parametric dynamics, 70
uncertainty analysis load flow model, 10
external load forces of two-DOF robot load flow of, 98
arm, 45 spool position response model, 10–11
hydraulic parametric Servo valve model construction, 20; see
uncertainties, 45 also Mechanical motion model
linearized hydraulic model first-order transfer function model, 21
construction, 46 load flow gain of servo valve, 23
linear mathematical model of EHS, 45 maximal spool displacement, 21
Robust controller, 5 no-load flow gain of proportional
Robust controller design, 54, 58; see also valve, 22
Robust model construction pressure loss from valve
analysis of open-loop system, 54 to cylinder, 22
block diagram for robust stability and servo valve parameters, 21
performance, 59 servo valve pressure loss, 21
block diagram of closed-loop sgn(.) function, 70–71, 98
system, 53 Single-rod acting mechanism, 8
complementary sensitivity Singular values of inverse function, 58
function, 61 Sinusoidal response, 62–68; see also
cross-linked feedback system, 56 Robust controller design;
cylinder position response in time Square response
domain, 62 control voltage for sinusoidal
external load as disturbance to demand, 65
EHS, 54 control voltage for square demand, 64
130 Index

Sinusoidal response (Continued) dynamic pressure of cylinder with

cylinder position response in time load, 31–34
domain, 62 mechanical properties curve of motor
designed controller result in time with load, 31
domain, 63 motor performance, 30–31
experiment of upper arm hydraulic output flow of pump, 32
actuator, 91 output pressure analysis of pump,
of experiment result by two control 31–32
methods, 66 parametric analysis of shoulder
frequency domain result of robust hydraulic actuator, 34
controller, 61 pump output power, 32
steady error, 68 simulation model for dynamic
of two joint angles, 65 pressure of cylinder with
video sequences of, 67, 68 load, 38
Sixth-order nonlinear dynamic model of simulation result with elbow joint, 33
EHS, 72 volumetric efficiency coefficient of
Small Gain Theorem, 56–57 pump, 31
Spool position response model, 10–11
Square response T
in two control methods, 66, 66
of two joint angles, 63, 65 Two chamber pressures of two
State errors, 88 cylinders, 44
State estimations, 90, 92 Two-link dynamic model, 27; see also
State feedback control method, 6 Mechanical motion model
State observer design, 72; see also kinetic equation of, 28
Nonlinear backstepping control total energy of, 28–29
full-state observer construction, 72–75 two cylinder dynamic lengths and
high gain of observer, 75, 76 force arms, 30
nonlinear model of EHS, 72 two driving torques of, 29
observer convergence, 76–81 Two load flows, 70
Steady error, 68
Step response experiment of arm U
hydraulic actuator, 89
Stick-slip friction, 71 Unknown parametric variation, 3
Structural uncertainty, 49; see also Unmodeled uncertainties, 2
Linearized model of Upper arm cylinder control design,
electro-hydraulic actuator 37, 41
model, 54
parameters, 51 V
Supply pressure, 35, 36
Symmetrical cylinder model, 8–9 Velocity vector, 27
System performance analysis, 30; see also Virtual control variables, 97
Linear PID control design Viscous damping coefficient of cylinder,
actual cylinder pressure, 32 48–49
cylinder pressure in simultaneous Volumetric efficiency coefficient of
joints motion, 34, 37, 38 pump, 31
cylinder pressure supporting elbow,
34, 35, 36 W
cylinder pressure supporting
shoulder, 33–34 Weight Function Design, 55–58