You are on page 1of 167

C D  H P

C S

A Thesis Submitted to
the College of Graduate Studies and Research
in Partial Fullfillment of the Requirements
for the Degree of Doctor of Philosophy
in the Department of Mechanical Engineering
University of Saskatchewan
Saskatoon, Canada

By
Yang Lin

Supervisors: Prof. Richard Burton & Prof. Yang Shi


November 2011
A

In recent years, a great deal of interest has been oriented towards hydraulic systems
which are energy efficient, responsive (tracking), and accurate. The traditional approach to
achieve responsive and accurate positioning performance is to use a servo valve actuator
and position and/or velocity feedback. An alternate positioning system is an electrohy-
draulic actuation system (EHA), in which the fluid from the hydraulic motor is directed
back to the inlet of the pump. Changing the swashplate angle or varying the prime mover
shaft speed varies the flow to the hydraulic actuator (linear or rotary) which in turn is used
to control the positioning or speed of the load. Because there are no major losses associ-
ated with throttling of the fluid, power losses are minimized. In earlier EHA systems, the
actuator was limited to that of a rotary system because of the requirement for symmetry in
the flow to and from the motor. Recent design changes to linear single rod actuators have
expanded the EHA applications to linear positioning. In addition, a specially designed
EHA linear actuator system was shown to be able to position a load to 200 nanometers.
However, the ability to track a desired input path was not extensively studied and as such,
algorithms to control this high precision EHA system were required; hence this was the
motivation for this study.

The main objective of the thesis was to develop high performance control schemes
for (1) a valve controlled hydraulic positioning control system (HPCS) and (2) a specific
precision positioning EHA system and verify their position tracking performance.

Control methods that were applied to HPCSs in the past decade were comprehensively
reviewed in this dissertation. Many successful control algorithms have been developed
for hydraulic transmission systems, however, certain problems such as slip-stick friction,
uncertainty and nonlinearity in hydraulic actuators, pumps and valves are not fully ad-
dressed. Three control algorithms are considered in this study: (1) H2 -optimal control,
(2) H∞ PI plus feedforward control, and (3) robust sliding mode control. The design pro-

i
cesses of these three algorithms were based on discrete-time system models. The first
two algorithms were based on linear models of the systems while the third applied non-
linear actuator friction in the system model. These three different control algorithms are
developed and implemented using simulations and experiments; in addition, their control
performance in terms of position tracking and bandwidth performance are examined.
The original contributions of the research are:

1. Developing a comprehensive review of the control methods applied to HPCS sys-


tems during the past decade.

2. For the first time, applying the discrete-time H2 -optimal control algorithm on an
HPCS system. The applicability of the discrete-time H2 -optimal control for the
HPCS was verified.

3. Developing a new framework (SOF) from a PI plus feedforward control framework.


The feedback and feedforward gains were explicitly solved through H∞ optimization
technique instead of traditional tuning.

4. Designing a novel controller called robust sliding mode controller (RSMC) with
the sliding mode surface designed considering the parameter uncertainties in the
nonlinear friction model.

ii
C

Abstract i

Contents iii

List of Figures vi

List of Abbreviations ix

1 Introduction 1
1.1 Hydraulic Position Control Systems . . . . . . . . . . . . . . . . . . . . 1
1.2 Valve Controlled Hydraulic Position Control System (HPCS) . . . . . . . 3
1.3 Pump Controlled Electrohydraulic Actuator System (EHA) . . . . . . . . 4
1.4 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Review of Existing Control Methods Applied to Hydraulic Control Systems 9


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Adaptive Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Sliding Mode Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 H∞ Optimal Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Fuzzy Logic Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.7 Other Control Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7.1 Model predictive control . . . . . . . . . . . . . . . . . . . . . . 22
2.7.2 Quantitative feedback control . . . . . . . . . . . . . . . . . . . . 23
2.8 Concluding Remarks and Discussions . . . . . . . . . . . . . . . . . . . 24

3 Summary of Manuscripts 26
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Experimental Apparatus and Procedures . . . . . . . . . . . . . . . . . . 26
3.2.1 Experimental test apparatus of the valve controlled HPCS . . . . . 27
3.2.2 Experimental test apparatus of the EHA . . . . . . . . . . . . . . 28
3.3 Paper #1: Discrete-Time H2 -Optimal Control for a Servo Hydraulic Posi-
tion Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.2 Design procedures . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.3 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.4 Conclusions and contributions . . . . . . . . . . . . . . . . . . . 34
3.4 Paper #2: H∞ PI Plus Feedforward Control Design for an ElectroHydraulic
Actuator System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.2 Design procedures . . . . . . . . . . . . . . . . . . . . . . . . . 36

iii
3.4.3 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.4 Conclusions and contributions . . . . . . . . . . . . . . . . . . . 40
3.5 Paper #3:Modeling and Robust Discrete-Time Sliding Mode Control De-
sign for a Fluid Power ElectroHydraulic Actuator (EHA) System . . . . . 42
3.5.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5.2 Design procedures . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.5.3 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5.4 Conclusions and contributions . . . . . . . . . . . . . . . . . . . 48

4 Comparisons of the Different Control Methods on a Model of the EHA Sys-


tem 49
4.1 Quasi-Step Tracking Response . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 Sine Wave Tracking Response . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

5 Conclusions, Contributions and Recommendations 58

References 69

A Introduction of Control Theories Applied to Hydraulic Systems 70


A.1 PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
A.2 Sliding Mode Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
A.2.1 Sliding surface design . . . . . . . . . . . . . . . . . . . . . . . . 73
A.2.2 Sliding mode control design . . . . . . . . . . . . . . . . . . . . 74
A.3 H∞ Optimal Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
A.3.1 H∞ norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
A.3.2 Lyapunov stability . . . . . . . . . . . . . . . . . . . . . . . . . 77
A.3.3 Linear matrix inequalities . . . . . . . . . . . . . . . . . . . . . . 78
A.3.4 H∞ control design with LMIs . . . . . . . . . . . . . . . . . . . . 78
A.4 Fuzzy Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
A.4.1 Fuzzy set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
A.4.2 Fuzzification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.4.3 System knowledge and fuzzy rule base . . . . . . . . . . . . . . . 83
A.4.4 Decision making . . . . . . . . . . . . . . . . . . . . . . . . . . 84
A.4.5 Defuzzification . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

References 89

B Discrete-Time H2 -Optimal Output Tracking Control for an Experimental


Hydraulic Positioning Control System (HPCS) 90
B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
B.2 Experimental Test Facility . . . . . . . . . . . . . . . . . . . . . . . . . 92
B.3 Modeling and Identification of HPCS . . . . . . . . . . . . . . . . . . . . 93
B.3.1 Transfer function of the open-loop system . . . . . . . . . . . . . 93
B.3.2 Standard sampled-data state-space model . . . . . . . . . . . . . 95

iv
B.4 Discrete-Time H2 -Optimal Output Tracking Controller Design and Simu-
lation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
B.4.1 Discrete-time H2 -optimal output tracking control . . . . . . . . . 100
B.4.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . 102
B.5 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
B.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

References 107

C Modeling and H∞ PID Plus Feedforward Controller Design for an Electro-


hydraulic Actuator (EHA) System 108
C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
C.2 Modeling of the EHA System . . . . . . . . . . . . . . . . . . . . . . . . 112
C.2.1 Linear symmetrical actuator . . . . . . . . . . . . . . . . . . . . 112
C.2.2 Hydraulic pump . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
C.2.3 Pump/actuator connection & overall hydraulic model . . . . . . . 114
C.3 Discrete-Time PI Plus Feedforward Controller with H∞ Performance . . . 116
C.3.1 Transforming the PI plus feedforward controller into an SOF con-
troller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
C.3.2 H∞ optimization . . . . . . . . . . . . . . . . . . . . . . . . . . 118
C.4 Simulation Studies and Experimental Tests . . . . . . . . . . . . . . . . . 120
C.4.1 Simulation study . . . . . . . . . . . . . . . . . . . . . . . . . . 120
C.4.2 Experimental test . . . . . . . . . . . . . . . . . . . . . . . . . . 122
C.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

References 127

D Modeling and Robust Discrete-Time Sliding Mode Control Design for a


Fluid Power Electrohydraulic Actuator (EHA) System 128
D.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
D.2 Nonlinear Model of the EHA System . . . . . . . . . . . . . . . . . . . . 134
D.2.1 Related works . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
D.2.2 Nonlinear model with uncertainties . . . . . . . . . . . . . . . . . 136
D.2.3 Model transformation for the SMC design . . . . . . . . . . . . . 139
D.3 Robust Discrete-Time Sliding Mode Control Design for the EHA Model . 139
D.3.1 Sliding surface design . . . . . . . . . . . . . . . . . . . . . . . . 140
D.3.2 Robust sliding mode control design . . . . . . . . . . . . . . . . 142
D.3.3 Design procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 146
D.4 Simulation Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
D.5 Experimental Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
D.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

References 157

v
L  F

1.1 Valve controlled hydraulic position control system. . . . . . . . . . . . . 4


1.2 Schematic of the EHA system. . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Experimental friction in the particular EHA of interest and the identified
nonlinear friction model. . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1 Schematic of the adaptive control algorithm. . . . . . . . . . . . . . . . . 11


2.2 Schematic of the sliding mode control algorithm. . . . . . . . . . . . . . 15
2.3 Schematic of the H∞ control algorithm. . . . . . . . . . . . . . . . . . . 18
2.4 Schematic of the fuzzy logic control algorithm. . . . . . . . . . . . . . . 20

3.1 Hydraulic positioning control system. . . . . . . . . . . . . . . . . . . . 27


3.2 Experiment setup of the EHA control system. . . . . . . . . . . . . . . . 28
3.3 Experimental open-loop frequency responses (magnitude and phase). . . . 30
3.4 Model reference feedback control system.. . . . . . . . . . . . . . . . . . 30
3.5 Generalized control system diagram for the packed plant. . . . . . . . . . 31
3.6 Simulation tracking response of the H2 optimal. . . . . . . . . . . . . . . 33
3.7 Experimental tracking response of the H2 optimal. . . . . . . . . . . . . . 34
3.8 Experimental tracking response with the feedforward. . . . . . . . . . . . 39
3.9 Experimental tracking response without the feedforward. . . . . . . . . . 39
3.10 Experimental tracking errors. . . . . . . . . . . . . . . . . . . . . . . . . 40
3.11 Experimental friction in the EHA and the identified nonlinear friction model:
(1) ∗ points: Experimentally measured friction force; (2) Solid curve: Es-
timated curve for the friction force; (3) Dashed dotted line: Approximated
linear friction force. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.12 Nonlinear uncertain friction band. . . . . . . . . . . . . . . . . . . . . . 45
3.13 Simulation tracking response of the EHA system with uncertainties using
the proposed method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.14 Experimental tracking response of the EHA system with model uncer-
tainty using the proposed method. . . . . . . . . . . . . . . . . . . . . . 47

4.1 Quasi-step tracking response of the EHA system using H2 -optimal control 51
4.2 Quasi-step tracking response of the EHA system using PI H∞ plus feed-
forward control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 Quasi-step tracking response of the EHA system using robust sliding mode
control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.4 Sine wave tracking response of the EHA system using H2 -optimal control. 54
4.5 Sine wave tracking response of the EHA system using H∞ PI plus feedfor-
ward control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.6 Sine wave tracking response of the EHA system using robust sliding mode
control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.7 Frequency response of the EHA system with the proposed controllers. . . 57

vi
A.1 Schematic of a traditional PID control algorithm. . . . . . . . . . . . . . 70
A.2 Schematic of the sliding mode control algorithm. . . . . . . . . . . . . . 71
A.3 H∞ control with generalized model. . . . . . . . . . . . . . . . . . . . . 79
A.4 Membership function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
A.5 Fuzzy sets for car speed. . . . . . . . . . . . . . . . . . . . . . . . . . . 83
A.6 Schematic of the fuzzy logic control algorithm. . . . . . . . . . . . . . . 84
A.7 Fuzzy sets for car cruise control. . . . . . . . . . . . . . . . . . . . . . . 85
A.8 Aggregation, composition and accumulation processes of decision making. 86
A.9 Defuzzification using center of gravity. . . . . . . . . . . . . . . . . . . . 87

B.1 Hydraulic positioning control system. . . . . . . . . . . . . . . . . . . . 93


B.2 Experimental open-loop frequency responses (magnitude and phase). . . . 94
B.3 Simulated open-loop frequency responses (magnitude and phase). . . . . 95
B.4 Sampled-data model reference control system. . . . . . . . . . . . . . . . 96
B.5 Sampled-data control system diagram for the general plant. . . . . . . . . 97
B.6 Standard sampled-data control system. . . . . . . . . . . . . . . . . . . . 98
B.7 Discrete-time control system. . . . . . . . . . . . . . . . . . . . . . . . . 99
B.8 Simulation results of the position output signal. . . . . . . . . . . . . . . 103
B.9 Control signal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
B.10 Experimental results of the position output signal. . . . . . . . . . . . . . 104

C.1 Schematic of the whole hydraulic circuit of the system. . . . . . . . . . . 109


C.2 Bond graph of the symmetrical actuator. . . . . . . . . . . . . . . . . . . 112
C.3 Bond graph of the hydraulic pump. . . . . . . . . . . . . . . . . . . . . . 113
C.4 PI plus feedforward controller platform. . . . . . . . . . . . . . . . . . . 116
C.5 Simulation tracking response with the feedforward. . . . . . . . . . . . . 120
C.6 Simulation tracking response without the feedforward. . . . . . . . . . . 121
C.7 Simulation tracking errors. . . . . . . . . . . . . . . . . . . . . . . . . . 121
C.8 Experimental setup of the EHA system. . . . . . . . . . . . . . . . . . . 122
C.9 Experimental tracking response with the feedforward. . . . . . . . . . . . 123
C.10 Experimental tracking response without the feedforward. . . . . . . . . . 123
C.11 Experimental tracking errors. . . . . . . . . . . . . . . . . . . . . . . . . 124

D.1 Experimental friction in the EHA and the identified nonlinear friction model
[Chinniah, 2004]: (1) ∗ points: Experimentally measured friction force;
(2) Solid curve: Estimated curve for the friction force; (3) Dashed dotted
line: Approximated linear friction force [Chinniah, 2004]. . . . . . . . . . 131
D.2 Schematic of the hydraulic circuit of the system. . . . . . . . . . . . . . . 133
D.3 Nonlinear uncertain friction band. . . . . . . . . . . . . . . . . . . . . . 138
D.4 Simulation tracking response of the EHA system with uncertainties using
the proposed method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
D.5 Simulation tracking response of the EHA system with uncertainties using
the method in [Wang et al., 2008]. . . . . . . . . . . . . . . . . . . . . . 148
D.6 Simulation tracking error curves using the proposed method and the method
in [Wang et al., 2008]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

vii
D.7 Simulation tracking response of the EHA system with model uncertainty
and transducer noise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
D.8 Simulation tracking error of the EHA system with model uncertainty and
transducer noise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
D.9 Experimental setup of the EHA system. . . . . . . . . . . . . . . . . . . 152
D.10 Experimental tracking response of the EHA system with model uncer-
tainty using the proposed method. . . . . . . . . . . . . . . . . . . . . . 152
D.11 Experimental tracking error of the EHA system with model uncertainty
using the proposed method. . . . . . . . . . . . . . . . . . . . . . . . . . 153
D.12 Experimental tracking response of the EHA system with model uncer-
tainty using the method in [Wang et al., 2008]. . . . . . . . . . . . . . . 153
D.13 Experimental tracking error of the EHA system with model uncertainty
using the method in [Wang et al., 2008]. . . . . . . . . . . . . . . . . . . 154
D.14 Experimental tracking response of the EHA system using PID controller. . 154

viii
L  A

HPCS hydraulic position control system


EHA electro-hydraulic actuator
LMI linear matrix inequality
SMC sliding mode control
RSMC robust sliding mode control
LMI linear matrix inequality
LTI linear time-invariant
LQR linear quadratic regulator
MIMO multiple-input and multiple-output
PID proportional-integral-derivative
SOF static output feedback
BMI bilinear matrix inequality

ix
C 1
I

1.1 Hydraulic Position Control Systems


Hydraulic positioning systems have been widely used in a large number of fields by virtue
of their small size to power ratios and the ability to apply very large forces and torques
[Merritt, 1967]. These systems play an important role in transportation, earth moving
equipment, aircraft and industry machinery with heavy duty applications.
Traditional hydraulic transmission control systems are mainly based on valve con-
trolled hydraulic systems [Merritt, 1967; Habibi et al., 1994; Ziaei and Sepehri, 2001].
These types of systems use proportional valves to vary the flow rate. If the valve is “pres-
sure compensated” or if flow/velocity feedback is employed, very accurate velocity and
position control can be achieved. However, this control comes at the expense of pressure
losses across the proportional valve which can reduce the power efficiency of the circuit
significantly [Manring and Luecke, 1998].
In recent years, pump controlled systems which do not use proportional valves have
been the subject of significant research. Such pump controlled hydraulic systems, are often
categorized as “hydrostatic systems”. This configuration has higher control efficiency
because flow to the load is controlled by changing the “stroke” of the pump or by adjusting
the input shaft speed. In addition, the actuation of hydrostatic systems is achieved by
direct coupling of the pump with the actuator (pressure and return lines). So, few losses
are incurred in between the two. It should be noted that the term “hydrostatic” can refer to
a category of systems which are “closed”, meaning that fluid from the actuator is returned
directly to the inlet of the pump. The term “hydrostatic” is also used to describe similar
systems in which the return flow is directed to a reservoir first and then to the pump inlet.

1
It is the former definition that is adopted in this thesis.

There are mainly two ways to build a hydrostatic system. The first is to use a variable
displacement constant-speed-piston pump; the pump can regulate the flow by changing
the angle (stroke) of its swash plate. However, this type of hydrostatic circuit can be rel-
atively energy inefficient since the pump rotates continuously regardless of the movement
of the actuator [Manring and Luecke, 1998; Watton, 1989]. A second approach of flow
modulation in a hydrostatic system is to use a fixed displacement pump and then to reg-
ulate the flow by changing the rotational speed and direction of the pump/motor [Habibi
and Goldenberg, 2000; Bobrow and Desai, 1990]. The second configuration of hydrostatic
systems will be one of the focuses of this study.

The dynamics of hydraulic systems are highly nonlinear [Merritt, 1967]. As a con-
sequence, control of hydraulic systems is always a challenge due to the presence of non-
linearities which arise from fundamental properties such as fluid compressibility (due to
entrained air, mechanical compliance, dependency on pressure, and temperature), complex
flow properties of hydraulic valves (such as pressure losses, transient and turbulent flow
conditions), and nonlinear friction characteristics in hydraulic actuators (due to the com-
bined properties of static, coulomb and viscous friction). Furthermore, the system may be
subject to non-smooth and discontinuous nonlinearities due to the directional change of
the pump rotation or valve opening, valve overlap and pump/motor deadband. In addition
to these nonlinear characteristics which often demand the development of advanced con-
trol schemes, “model uncertainty” is a challenge for high performance hydraulic control
problems.

Conventional hydraulic positioning control designs are mainly based on linear approx-
imations of the system dynamics and in many cases, a very crude approximation of com-
ponents and fluid parameter values. Some essential characteristics, such as nonlinear actu-
ator friction, leakage, disturbances and model uncertainties have not been fully considered
in the design of controllers [Armstrong-Helouvry et al., 1994].

In order to achieve high precision control performance, advanced control schemes


which consider the real dynamic characteristics of the hydraulic system (nonlinear actu-
ator friction, compressibility of the hydraulic oil, hysteresis, and sensor disturbance) are

2
desirable for many applications. Many recent research studies have focused on advanced
controller design for hydraulic actuation systems. It is well known that adaptive control
and sliding mode control are the most often applied algorithms for hydraulic motion con-
trol systems [Guan and Pan, 2008a]. Other control methods such as fuzzy logic control,
adaptive control and advanced PID control have also been thoroughly studied and devel-
oped for hydraulic systems. A comprehensive review of hydraulic motion control designs
which have been developed in the past few decades will be considered in Chapter 2.
In summary, many advanced control schemes have been used to improve the perfor-
mance of position and velocity controlled systems. Of particular interest in this work is
the control of a traditional valve controlled system and of a special high precision “closed”
hydrostatic system called an electrohydraulic actuation (EHA) circuit. For clarity and to
facilitate the statement of the research objectives, these two systems will be introduced in
more detail in the following sections.

1.2 Valve Controlled Hydraulic Position Control System


(HPCS)
The first HPCS to be considered in this study was a valve controlled unit. The HPCS sys-
tem is a typical servo valve (proportional) controlled hydraulic position control system.
This configuration is the most commonly applied hydraulic actuation device in the indus-
try. The main advantage of using a valve controlled HPCS arrangement is that it is capable
of driving multiple actuation devices with a single power supply. This configuration has
been very common for mobile hydraulic systems where multiple implements are operated
using a single power source. The main disadvantage of using a valve controlled HPCS is
that the power efficiency can be very low. The major source of power losses are: (1) con-
tinually running pump even at standby situation to maintain circuit pressure; (2) pressure
drop across the servo valve especially at low actuation speeds and low load pressures.
The particular HPCS system that was studied in the author’s first paper [Lin et al.,
2009a] was based on such a valve controlled system. The schematic of the HPCS is
presented in Figure 1.1. This HPCS consists of a pressure compensated pump (which is

3
used to set a constant pump pressure), a proportional servo valve, a symmetric actuator,
a position transducer, interfacing devices and a controlling computer. The flow to the
actuator is modulated by varying the resistance of the proportional servo valve. However,
associated with this resistance is a pressure drop across the valve (essentially a varying
orifice) which translates to a power loss. If the duty cycle is such that the load pressure
is low, then losses during this portion of the cycle can be significant. With the pump
continually running, the speed and position of the load is controlled by the proportional
servo valve that is attached with the pump and the control algorithm programmed into the
computer.

Figure 1.1: Valve controlled hydraulic position control system.

1.3 Pump Controlled Electrohydraulic Actuator System


(EHA)
The second HPCS to be considered in this study is a pump controlled electrohydraulic
actuator (EHA) system and is the application in two of the author’s publications [Lin et al.,
2007, 2009a]. The EHA system is a typical pump controlled hydraulic system. The main
advantage of using a pump controlled hydraulic system is its higher efficiency property

4
compared with valve controlled system [Manring, 2005]. There are some disadvantages
of using pump controlled hydraulic systems compared with valve controlled systems:

1. They need an auxiliary hydraulic circuit to maintain a positive pressure in the low-
pressure line (two check valves, an accumulator, a relief valve).

2. Only a single actuator can be operated at a time.

3. For “closed” systems, a symmetric actuator is required since the flow from the ac-
tuator to the inlet of the pump must be the same as the flow from the pump into the
actuator.

The EHA system of interest in this study is based on the closed circuit hydraulic trans-
mission system and its schematic is shown in Figure 1.2. Of particular interest was a high
precision EHA that was designed by [Habibi and Goldenberg, 2000]. This EHA used
a single rod symmetric actuator to drive the load as illustrated in Figure 1.2. An inner
loop velocity controller was placed around the electric motor which, in conjunction with
the selected pump and designed actuator, resulted in a high precision control system. In-
deed, this position control system was capable of moving a 20kg sliding mass with a high
accuracy of 200 nanometers over a maximum displacement of 12cm.

Figure 1.2: Schematic of the EHA system.

In other studies on both EHA systems and valve controlled HPCS, it has been ascer-
tained that hydraulic friction in the actuator was a critical issue affecting system perfor-
mance in terms of tracking accuracy and repeatability of the actuator performance [Owen

5
and Croft, 2003]. In [Armstrong-Helouvry et al., 1994], seven types of friction models
for mechatronic systems were comprehensively studied and summarized. However, trying
to clearly identify which model best describes the physical system under consideration is
often very difficult.

Because a particular experimental EHA was examined in this study, it is useful to


summarize some of the research that has been conducted on this HPCS. As mentioned
earlier, this EHA system was developed by [Habibi and Goldenberg, 2000]. A linearized
model of the EHA was developed by only considering the linear viscous friction in the
hydraulic actuator. Based on this linear model, gain scheduling control was applied to
the prototype EHA system [Sampson, 2005]. In [Chinniah, 2004], experimental results
showed that static, coulomb and viscous friction (slip-stick) co-exist in the EHA’s linear
actuator. A typical friction characteristic of the EHA is illustrated in Figure 1.3. In this
figure, a curve fitting polynomial equation was obtained to approximate the actual mea-
sured friction characteristics. However, some concerns were raised by researchers about
the characteristics at high and at very low velocities, and the rate of change of velocity.
In addition, hysteresis was noted but not reflected in the model. In order to compensate
for the nonlinear actuator friction, a discrete-time sliding mode controller was designed
for the EHA system [Wang et al., 2008]. The control performance outperformed the PID
control and gain scheduling control that was designed in [Sampson, 2005], in both track-
ing and robust performance. In this dissertation, robustness refers to the persistence of a
system’s characteristic behavior under perturbations or conditions of uncertainty.

In many of the aforementioned studies on the EHA, maintaining the precision posi-
tioning ability was the main objective. In [Sampson, 2005], [Wang et al., 2008] and [Lin
et al., 2009b], the focus shifted to tracking performance for various types of inputs. In
general, the responses were acceptable but the nonlinear friction certainly limited the per-
formance over a large operating range. The original overall objective of research on the
aforementioned EHA was to investigate the ability of the system to move a sliding load
with the same positional accuracy using different control algorithms. New control algo-
rithms are needed which deal with some of the uncertainties inherent within the EHA.
Thus this became the motivation for this particular study.

6
50

40

Friction Force (N)


30

20

10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Piston Velocity (m/s)

Figure 1.3: Experimental friction in the particular EHA of interest and the
identified nonlinear friction model.

1.4 Objective

As discussed above, the presence of nonlinear friction in both the valve controlled actua-
tor and the EHA, combined with other nonlinear fluid properties posed many challenges
for control design. Instead of pursuing a more accurate model for the nonlinear friction
and then designing an explicit controller to compensate for the nonlinear friction, it was
believed that the friction characteristics and other nonlinearities could be modeled with
uncertainties for the controller design. This observation motivated the work in this thesis.
With the aforementioned philosophy in mind, the main objective of the thesis was
to design high performance control schemes for first a valve controlled HPCS and then
for the EHA system and verify their position tracking and bandwidth performance using
different input signals.
Three control algorithms are considered in this study: (1) H2 -optimal control, (2) H∞
PI plus feedforward control, and (3) robust sliding mode control. The design processes
of these three algorithms were based on discrete-time system models. The first two al-
gorithms were based on linear models of the systems while the third applied nonlinear
actuator friction in the system model. These three different control algorithms are devel-
oped and implemented on modeled and experimental systems; in addition, their control
performance in terms of position tracking and bandwidth aspects are examined.

7
The manuscript is a “paper based” thesis in which three important publications are
considered. The first paper is based on the valve controlled HPCS system and the next two
papers are based on the pump controlled EHA system. Both systems were presented in
the last two sections. In the final part of the thesis, a comparison study of the performance
(positioning, tracking and bandwidth) of three algorithms on a model of the EHA system
is presented. Inputs to the model are of the quasi-step and sine wave form. It was believed
that this study would provide a “test-bench” to compare the performance of the controllers.
The format of the remaining dissertation is scheduled as follows: A comprehensive
review of the control methods that were designed for specifically, hydraulic systems is
given in Chapter 2. In Chapter 3, the major manuscripts that have been published are
summarized and the contributions of each paper presented; experimental apparatuses that
are associated with the papers and experimental procedures used to collect data are intro-
duced. Comparisons of the proposed control methods to a model of an EHA system are
given in Chapter 4. Finally, concluding remarks and a summary of the overall contribu-
tions are presented in Chapter 5. The appendices include full manuscripts of the papers
as well as a tutorial on the control approaches discussed in the main body of the thesis and
in the papers. The papers have been modified in format only to be consistent with the rest
of the manuscript.

8
C 2
R  E C M A 
H C S 1

2.1 Introduction
Hydraulic positioning systems have been widely used in a large number of fields by virtue
of their small size to power ratios and their ability to apply very large forces and torques
[Merritt, 1967]. These systems play an important role in transportation, earth moving
equipment, aircraft and industry machinery with heavy duty applications. The behavior of
hydraulic components and systems is highly nonlinear, a consequence of the fundamental
behavioral properties such as fluid compressibility, complex flow properties of hydraulic
valves, and nonlinear friction characteristics in hydraulic actuators.
There are essentially two approaches that are followed in trying to compensate for
the nonlinear behaviors. The first is to redesign the components/systems or to design and
operate them in regions where the effects of the nonlinear behavior are minimized. The
second approach is to recognize the existence of the nonlinearities and to design con-
trollers which try to compensate by adjusting the control signal to the system according to
some algorithms. This approach has been very popular over the years and has led to some
very interesting types of controllers. The objective, then, of this Chapter is to summarize
some of the research and development that has occurred in the area of controller design
for fluid power systems specifically. It will be assumed that the readers of this Chapter
are familiar with control theories and definitions and as such, no attempt will be made to
present the theory or terminology. For readers who wish a more “tutorial” approach to this

1
A paper based on this Chapter has been submitted to International Journal of Fluid Power for publication

9
review which, indeed, does define such terms and methodology, they are encouraged to
refer to Appendix A of this dissertation. The format of the paper is as follows. Research
on basic PID controllers is considered first followed by publications which apply adaptive
control to fluid power systems. Sliding mode control is then examined followed by H∞
control and finally Fuzzy controllers. A brief summary of the research results and areas
where research needs to continue is presented.
In a review such as this, many manuscripts have been published in non archival jour-
nals or conference proceedings and have been very difficult to obtain. As such some
manuscripts do not appear for this reason but it is believed that the papers discussed in
this Chapter do cover most of the major contributions to controls as applied to fluid power
systems.

2.2 PID Control

There are many research studies which develop theories for “advanced” control for hy-
draulic motion systems. Many of these studies compare the performance of their ap-
proaches with the performance of conventional proportional-integral-derivative (PID) con-
trollers trained using the well established Ziegler-Nichols PID tuning method [Lee and
Cho, 2003].
In practice, PID control has been the most applied control method for industrial ap-
plications; indeed, it has been estimated that over 90% of the controllers in use today are
PID controllers, even though other advanced control theories and practical design methods
exist [Aström and Hägglung, 2001]. These controllers have been proven to be reliable and
perhaps more important, understandable by the general practitioner.
Many research works have been focused on developing refined and best tuning meth-
ods adjusting PID parameters for hydraulic motion control problems. In [Varseveld and
Bone, 1997], a discrete-time PID controller was designed for a pneumatic positioning sys-
tem operated by on/off solenoid valves; both friction compensation and position feedfor-
ward were added in the final control in order to reduce the steady-state error and tracking
error respectively. In [Liu and Daley, 1999, 2000], the researchers developed an online

10
optimal tuning PID controller for hydraulic positioning systems; the developed controller
was a semi-adaptive controller with the model parameters estimated when the operating
point changes, and the optimal controller gains were generated using the Optimization
Toolbox in MATLAB. Similar to the approach of [Liu and Daley, 1999, 2000], a PID
controller with the gains optimized by the LQR algorithm was developed for a hydraulic
loading system in [Huang et al., 2003].
In other publications on PID control of hydraulic systems, the basic PID control con-
cept has been combined with other advanced control algorithms, for example, sliding
mode PID control [Eker, 2006], fuzzy PID control [Zheng et al., 2009] and H∞ PID control
[Lin et al., 2010]. Further discussion on these topics will be presented in later sections.

2.3 Adaptive Control

Adaptive control is a very successful control strategy that has been applied to hydraulic
control systems. Consider Figure 2.1, the main idea of adaptive control is that the model
parameters of the plant are estimated online, and then the controller parameters are up-
dated based on the new model. The first application of adaptive controllers for a hydraulic
system occurred about 40 years ago [Porter and Tatnall, 1970]. However, poor computa-
tional techniques at the time made the control design performance very limited. With the
rapid development of computer and electronics technologies over the past two decades,
adaptive control methods now can be easily implemented on hydraulic actuated systems.

Figure 2.1: Schematic of the adaptive control algorithm.

11
In the early 1990s, most adaptive control designs applied to hydraulic systems were
based on conventional linear control theories. In [Plummer and Vaughan, 1996], [Bobrow
and Lum, 1995] and [Tsao and Tomizuka, 1994], recursive least squares identification
methods were applied to estimate and update the model parameters of the servo hydraulic
systems, online. The main difference in these studies was the controller designs which
were all based on standard linear control design methods. For example, in [Plummer
and Vaughan, 1996], conventional pole-placement control method was applied to update
the adaptive controller parameters online. The proposed adaptive controller outperforms
the conventional fixed pole-placement controller under conditions of large changes in the
load. In [Bobrow and Lum, 1995], LQR control was adopted to solve the control problem
of a servo single rod hydraulic cylinder system. Smoothed step and sine wave tracking
responses were tested on the system; experimental responses showed that the controller
could track the sine wave at lower frequencies (up to 9Hz).

Adaptive schemes have been widely applied to friction compensation problems for
high accuracy motion control and this is often referred to as model based friction compen-
sation. The slip-stick friction characteristic in the hydraulic actuators is a significant issue
in high precision motion control problems because of the complex behavior of the fric-
tion especially when the system is running in a low speed. In [Frieland and Park, 1991],
a simple nonlinear friction estimator was designed by using a simple Coulomb friction
model; it was argued that even in a more realistic slip-stick friction condition, the pre-
sented estimator could do a good job. Using the estimation method from [Frieland and
Park, 1991], the performance of friction compensation control for a hydraulic fish process
machine was tested in [Tafazoli et al., 1998b]. In [Xie, 2007], a LuGre friction model
was applied to a servo actuator; a sliding mode observer was designed to estimate fric-
tion parameters in the LuGre model. Based on the estimated friction, an adaptive friction
compensation controller was developed. Both simulation and experimental results based
on a pseudorandom sine wave response showed that the proposed control method was
very promising. The combination of the sliding mode friction observer and the adaptive
friction compensation controller together take advantage of both algorithms in the area of
robust and nonlinear control. These research works achieved good control performance;

12
however, all the control design studies were based on a linear model of the system. Some
important dynamic information is lost as a result of linearization which can reduce the
applicability of these methods to other general systems. Thus it is important to design
nonlinear adaptive control which would not require linearization.

In many published studies, researchers used a Hammerstein system to model the non-
linearities caused by the deadzone properties of the hydraulic valves and actuators. It
was assumed that the linear part and the nonlinear part could be separated independently.
Many adaptive control schemes have been proposed to deal with the hydraulic systems
using this Hammerstein model. For example, in [Knohl and Unbehauen, 2000], an artifi-
cial neural network method was used to train the model parameters of the nonlinear part
of the hydraulic model online. Then, an inverse of the nonlinear model was developed to
compensate for the nonlinear part. A standard LQ controller was designed to control the
linear part of the hydraulic system. The control performance of the proposed method was
much better than a nominal LQ controller after a learning phase of 50ms. However, this
can be considered as an “indirect” nonlinear control design example for hydraulic control
systems (one inverse controller for the nonlinear part and the other for the linear part).

Since the early 1990s, researchers introduced the backstepping technique into the non-
linear adaptive controller designs, [Krstic et al., 1995]. Backstepping is an advanced con-
trol method which requires a special type of model, defined as the “strict feedback system”.
Each state variable is stabilized by its previous state variable by the Lyapunov method un-
til the process reaches the final step associated with the control input signal. At this point
the entire system is considered to be stabilized by the control input.

In recent years, researchers have developed many nonlinear adaptive controllers for
hydraulic systems using the backstepping method and Lyapunov technique. In [Alleyne
and Hedrick, 1995], nonlinear adaptive control schemes were developed for the force con-
trol of an active suspension system driven by a double-rod cylinder. It was shown that
the adaptive controller could achieve much better performance than conventional linear
design methods; parametric uncertainties were considered in this work. Similar design
approaches also appeared in [Alleyne and Liu, 2000], [Liu and Alleyne, 2000], [Choux
and Hovland, 2010], [Sohl and Bobrow, 1999] and [Zeng and Sepehri, 2008] for control

13
design with hydraulic motion control systems.

It is well known that hydraulic systems are highly nonlinear and complex; many ill
defined or uncertain terms exist in the systems. Conventional adaptive controls assume
constant parameter uncertainties; uncertain nonlinearities such as complex nonlinear fric-
tion forces and disturbances have usually not been addressed. Design problems are chal-
lenged by the coupled relationship between the robust feedback control and the parameter
estimation process. To further improve the performance of the nonlinear adaptive con-
trol algorithms, novel robust nonlinear adaptive control algorithms were developed for
hydraulic systems.

In [Yao, 1997], a robust adaptive controller was designed for nonlinear systems with
the control law being divided into two parts – nominal stabilization and robust attenuation.
Yao introduced a discontinuous projection adaptation law so that the interaction between
the robust feedback control design and the parameter estimation process could be avoided.
The backstepping technique was also incorporated into the nonlinear feedback control
design.

Following up on this approach, researchers have designed advanced robust adaptive


controllers for servo hydraulic motion control systems, [Yao et al., 1998], [Yao et al.,
2001], [Liu and Yao, 2003], [Yao, 2009], [Duraiswamy and Chiu, 2003]. For example,
many adaptive control schemes developed for hydraulic systems assume that the origi-
nal fluid volume in the hydraulic line is known; however, in the study of [Guan and Pan,
2008b], they assumed that the control volume was unknown. A nonlinear adaptive con-
troller was designed to compensate for the uncertain nonlinear parameters introduced by
the uncertain fluid volume. In order to compensate for uncertainty issues, a novel-type
Lyapunov function was developed to construct an asymptotically stable adaptive controller
and adaptation laws. The robust stability of the controlled system was established. This
method was applied on a single rod electro-hydraulic actuator system with the control vol-
ume uncertain by changing the length of the fluid circuit. The performance showed better
robustness than the conventional adaptive controller without considering the uncertainties.

In the past few years, researchers have developed modified versions of the basic adap-
tive controller. For example, in [Cho and Burton, 2011], an output feedback simple adap-

14
tive control algorithm was developed for a high performance electro-hydraulic actuator
system. The proposed control algorithm was compared with a conventional PID controller;
experimental results showed that the proposed method outperforms the conventional PID
controllers.

2.4 Sliding Mode Control


Sliding model control (SMC) first appeared in the context of variable structure systems.
SMC has become an efficient tool in the control of complex systems with uncertainties due
to its low sensitivity to disturbances and parameter variations, [Utkin et al., 1999]. The
basic concept is depicted schematically in Figure 2.2. The main idea behind the SMC is to
use a discontinuous control input to force the state trajectory to a certain well determined
sliding surface (S = 0 in Figure 2.2) and to remain on this surface over time, the so called
“bang-bang control” concept.

Figure 2.2: Schematic of the sliding mode control algorithm.

The design of sliding mode control consists of two main steps: (1) develop a realistic
but stabilized sliding surface for the system states; (2) develop control laws to force the
system dynamics to follow the desired sliding surface. The dynamic characteristics of
the resulting closed-loop control system will be mainly determined by the design of the
sliding surface. The most significant advantage of SMC is that once the states of the

15
system reach the predefined sliding surface, the system behavior depends neither on the
system parameters nor the disturbances. This property meets with the need of designing
feedback control for dynamic systems under uncertainty conditions.

Although SMC has its outstanding characteristics in compensating for the nonlineari-
ties and disturbances of electromechanical systems, a drawback is that the controller con-
tains discontinuous functions and can cause chattering phenomena, which in turn, can
affect the performance of the control system. This problem is normally compensated by
using a smoothing function to replace the discontinuous switching function; this, however,
does degrade the tracking performance, [Slotine, 1985].

In [Utkin and Yang, 1978], several classical methods to design the sliding surface for
both continuous-time and discrete-time systems were proposed, and the design, analysis,
and application of SMC to electromechanical systems were elegantly presented in [Utkin,
1993]. Since then, many research studies have been devoted to the development of ad-
vanced SMC schemes for systems with uncertainties and nonlinearities [Tang and Misawa,
2000], [Janardhanan and Bandyopadhyay, 2006], new sliding surface design [Gao et al.,
1995], and other practical constraints [Choi, 1998, 1999]. Based on these more general
publications, SMC methods have been widely used in electro-hydraulic control systems
due to their natural ability to handle parameter uncertainties and disturbances. Some of
this research will now be considered.

In [Liu and Handroos, 1999], a standard sliding mode controller was designed for a
hydraulic servo system with a flexible load which was a connection between two sliding
load by a spring. A special sliding surface which was a first-order weighted error func-
tion was introduced. The experimental results were quite promising; however, the sliding
surface parameters were chosen based on experience and tuning, and as such, no explicit
solution of the optimal sliding surface parameters was provided. In [Ha et al., 2000], a
sliding mode friction observer and controller with friction compensation were designed
for motion control systems with nonlinear friction. As a follow-up to this study, the fric-
tion compensation sliding mode controller was applied to a servo hydraulic actuator sys-
tem and achieved much better control performance than conventional PD controllers, [Ha
et al., 1999], [Bonchis et al., 2001].

16
Integral compensation sliding mode control has been developed in order to decrease
the steady state error of conventional sliding mode controls. This concept was inspired
by the integral term in the PID control which reduces steady state error. In [Chern and
Wu, 1991, 1992], this integral sliding mode controller was designed and applied to servo
hydraulic tracking systems and achieved accurate final position. A similar approach was
applied to the control of a hydraulic suspension system, [Sam et al., 2004]. However,
improvement on the final positioning accuracy resulted in degradation in the speed of the
tracking process. In [Fung et al., 1997], to compensate for this problem yet maintain
the final positioning accuracy, a switching proportional-integral sliding mode control was
designed. In this approach only a conventional proportional sliding mode controller was
functional at the beginning of the motion, and a switch would bring the integral sliding
mode controller online once the actuator approached the end of the motion.
Recently, advanced sliding mode control algorithms have been designed for hydraulic
actuator systems for the purpose of high accuracy positioning. In [Wang et al., 2006] and
[Wang et al., 2008], a discrete-time sliding mode controller was designed for a high ac-
curacy pump controlled electro-hydraulic actuator (EHA) system. A linear sliding mode
surface was designed using the well known LQR optimization technique to make the slid-
ing surface easier to follow for the desired states trajectory. Using the same high accuracy
EHA system, a robust sliding mode controller was designed in [Lin et al., 2009b]; the op-
timal sliding surface was designed using the Lyapunov method and the optimization was
solved by the linear matrix inequality (LMI) technique.

2.5 H∞ Optimal Control

Modern H∞ control theory with explicit state space solutions was first developed by [Doyle
et al., 1989]. The basic idea behind H∞ control is to minimize the system ∞-norm of the
objective signal by solving the matrix Riccati equations; the explicit solutions of solving
Riccati equations for H∞ controllers were implemented. H∞ control is a linear control
strategy which needs a state space “packed” model of the controlled system, shown in
Figure 2.3. The dynamic characteristics of fluid power systems necessitated modeling

17
with uncertain dynamic and unknown disturbances; thus robust H∞ control strategies were
found to increase safety, reliability and stability of hydraulic actuators [Chen and Francis,
1995].

Figure 2.3: Schematic of the H∞ control algorithm.

One area in which the H∞ algorithm has been applied is hydraulic vehicle suspension
control systems, because the vehicle suspension control problem can be readily treated
as a disturbance attenuation problem. In [Yamashita et al., 1990], an H∞ controller was
designed for a quarter-car model in order to attenuate the human body acceleration in
the car. In [Hirata et al., 1995], the H∞ design method was applied to a railroad vehicle
suspension control. It was observed that the proposed H∞ controller achieved very good
suspension performance. The robustness of the H∞ controller was examined and compared
with an LQR controller. The LQR controller was found to be much more sensitive to
the high frequency disturbance such as noise. In [Du et al., 2005], an output feedback
H∞ controller was designed for a semi-active suspension with magneto-rheological (MR)
dampers. The inverse model of the MR damper was achieved and used to compensate
for the hysteresis characteristic and the H∞ control was designed for the nominal system
without considering hysteresis.
H∞ control algorithms also work very well in other systems using hydraulic compo-
nents, such as earth moving, power train and industrial manipulator control systems. In

18
[Zhang et al., 2002], robust controls were applied to a MIMO earthmoving powertrain sys-
tem. It was found out that the controller had better nominal and robust performance than
the LQR controller. In [Fales and Kelkar, 2008], [Fales and Kelkar, 2005], robust control
was designed for a wheel loader. The robustness of the control system was improved by
applying the feedback linearization method. A similar approach (feedback linearization
based H∞ control) was applied to a servo hydraulic aircraft brake system in [Tunay et al.,
2001]. In [Kim and Tsao, 2000] and [Cheng and Moor, 1994], the state-space H∞ control
approach was applied to servo hydraulic systems. The bandwidth performance was ex-
amined in both papers. A “mixed sensitivity” technique was applied to the control design
which treated the control objective in a filtered (weighted) form. The frequency response
showed that the higher order output filter and controller resulted in a better bandwidth
performance.

In 1994, Gahinet introduced a linear matrix inequality (LMI) approach to solve the H∞
optimization control problem, which deals with the conventional Riccati equation solving
problem by solving several linear matrix inequality problems, [Gahinet, 1994]. LMI opti-
mization problems can be easily handled by the LMI Toolbox in MATLAB.

Further research studies were focused on robust H∞ control design for different sys-
tems based on the LMI technique because of its natural ability to deal with matrix uncer-
tainties, such as polytopic uncertainties and norm bounded uncertainties. This property
meets the need of designing robust controllers for hydraulic systems which consists of
unknown nonlinearities and uncertainties. In [Dean and Fales, 2007], a robust H∞ con-
troller was designed for a pressure compensated variable displacement hydraulic pump. In
this study, uncertain bulk modulus and control valve coefficients were considered as pa-
rameter uncertainties and were integrated into polytopic uncertainty matrix. In [Yu et al.,
2003], researchers employed a gain scheduling type controller to control a robotic ma-
nipulator. The controller was “scheduled” between two H∞ output feedback controllers
whose designs were based on two different operating regions. The pole-placement tech-
nique was combined with the H∞ synthesis in the optimization. The proposed controller
achieved 60% less tracking error on the manipulator joint angle than a conventional PID
gain scheduling controller.

19
In [Njabeleke et al., 2000], a speed tracking controller was designed based on a ro-
bust H∞ method which considered model uncertainties. In [Lin et al., 2009a], a discrete-
time H2 -optimal controller which is a very similar control design algorithm with the H∞ -
optimal control was designed for a servo hydraulic position control system. Both sim-
ulation and experimental results showed the effectiveness of the proposed controller. In
[Chen et al., 2005], [Du et al., 2003] and [Son et al., 2001], robust H∞ control was applied
to car suspension systems which treated the suspension control problem as a disturbance
attenuation problem with hard constraints on the control signal limitations. The control de-
sign processes were implemented by applying the LMI technique to the H∞ optimization
control.

2.6 Fuzzy Logic Control

The fuzzy logic theorem was developed by Lotfi Zadeh during the 1960s [Zadeh, 1965,
1968, 1972]. The application of fuzzy logic to various control industries was first initiated
in the pioneering studies by [Mamdani and Assilian, 1975; King and Mamdani, 1977]. The
basic concept is illustrated in Figure 2.4 which treated the control design as a “linguistic”
decision making process. The design procedure is similar to a human thinking process and
so as to make it easier to be understood by control engineers.

Figure 2.4: Schematic of the fuzzy logic control algorithm.

20
In [Lee, 1990], a thorough review on fuzzy logic control applications to different in-
dustrial and science areas was provided. The first application of fuzzy logic to the control
of hydraulic actuated system was published in [Zhao and Virvalo, 1993] by introducing
a simple velocity fuzzification set to change the state controller gains due to the velocity
of the cylinder. This was done because the velocity signal was easier to obtain and car-
ried different loading conditions with it, making the controller less sensitive to different
load conditions. The proposed controller showed superior performance than a static state
feedback controller especially when the load of the system changed. Following the idea
of Zhao and Virvalo, in [Corbet et al., 1996], researchers introduced a combined posi-
tion and velocity fuzzy set and control rule to control a hydraulic two degree-of-freedom
robotic arm. Both step and ramp signal responses were achieved and it was proved that
the proposed controller could achieve fast response and robustness against varying loading
conditions. Similar fuzzy control approaches were also used in [Lee and Kopp, 2001] and
[Yuan et al., 2010] to control servo hydraulic positioning systems.

In [Lee and Kopp, 2001], a fuzzy controller was designed for a high precision in-
dustrial forging machine which considered the moving direction, position and material
of the forging process into the fuzzy set design. Higher precision was achieved using
the proposed controller and discussions about the simplicity and easy implementation of
fuzzy controllers were addressed. In [Chen et al., 2008], a dual-cylinder dual-actuation
hydraulic synchronous motion system was controlled by a fuzzy logic controller which
combined the fuzzy tracking controller with a feedforward compensation. Both hydraulic
nonlinearities and uncertainties were considered, and experimental results showed that a
synchronization error less than 4mm was achieved.

The implementation of fuzzy rules for the fuzzy controllers is highly dependant on the
designers’ experience with the system. There is no systematic rule to determine the num-
ber of fuzzy rules, number of membership functions and the width of each membership
function. Because of the nonlinear properties of the hydraulic motion systems and fuzzy
controllers, steady state error is normally an issue with fuzzy logic controls for hydraulic
systems. This initiated the idea of combination of fuzzy control with the PID control
scheme to eliminate the steady state error. In [Zheng et al., 2009], a fuzzy PID controller

21
was designed for a servo hydraulic press drive volume control system. The fuzzy ranges
of the error and change of error were separated into 7 sub-regions, and a 7 × 7 fuzzy rule
decision table was implemented to decide the PID gains according to different fuzzy sets.
The designed fuzzy PID controller was compared with a conventional PID controller and
obtained 50% less position error for both step and fixed frequency sinusoidal responses.
In [Shao et al., 2005], the researcher designed both a self-tuning fuzzy PID controller
and a hybrid fuzzy controller with a PID controller switched online when the position
approached the end of the stroke. Only the position error was considered and divided into
7 fuzzy regions for the fuzzy set. Simulation results showed that the hybrid of fuzzy and
PID control had a faster response than the self-tuning fuzzy PID control and the same level
of steady state error; however, the experimental results showed the opposite situation.

2.7 Other Control Methods


The previous five sections have reviewed the most commonly used and applied control
methods that have been designed for hydraulic systems. It should be noted that there are
some other control methods that have been successfully designed for certain hydraulic
systems yet not been widely adopted in applications. Some of them will be discussed in
this section.

2.7.1 Model predictive control

Model predictive control (MPC) have been proven to be an effective control design algo-
rithm for industrial process solutions, [Camacho and Bordons, 2004]. This approach uses
the model of the controlled system to predict the future output of the system for a set of
future desired outputs. A cost function which is based on the quadratic combination of the
future errors and future control signals is defined and optimally minimized to generate the
future control signals. Only the first control signal (current control signal) is sent to the
system. This process is repeated at each sampling interval. However, it was observed that
MPC is not adequately developed for hydraulic motion control applications.
In [Wu et al., 1998] and [Sepehri and Wu, 1998], a generalized predictive controller

22
(GPC) was designed for servo hydraulic actuator systems. Experimental tests showed that
the proposed controller achieved quick response and accurate tracking performance based
on both large and small input signals together with loading condition changes.

In [Yu et al., 2011], the researchers developed a network based modified GPC for
a servo hydraulic position control system. It was observed experimentally that with a
conventional GPC, the system can be driven unstable in the presence of random signal
transmission time-delay from sensor to controller and controller to actuator. The proposed
controller provided extra compensation on the time-delay and improved the stability of
the controlled system. Experimental results show the proposed controller successfully
compensated for the instability caused by the time-delay.

2.7.2 Quantitative feedback control

Quantitative feedback control is a frequency domain design algorithm which supplies fre-
quency compensation to the controlled system in the presence of bounded parameter un-
certainties, [Horowitz, 1963]. A set of models based on different operating conditions of
the system are identified, and bode frequency responses based on these models with pa-
rameter uncertainties are provided. Feedback compensations are designed to adjust the
shape of the closed loop frequency response to the desired region. It is an easily under-
stood and straight forward robust feedback control design method which has been applied
to process control, engine speed control and flight control, [Houpis and Rasmussen, 1999];
however, this type of control has not been widely applied to the design of controllers for
hydraulic systems. The exception to this were the studies conducted by Sepheri et al. In
[Niksefat and Sepehri, 2000] and [Niksefat and Sepehri, 2001], quantitative feedback con-
trol was applied to a hydraulic force control system. Both load changing and bulk modulus
variation were considered as parameter uncertainties. Experimental step and frequency re-
sults showed good robustness performance of the proposed control system when subjected
to the load, environmental stiffness and pump pressure variations. A similar approach also
appeared in [Karpenko and Sepehri, 2010] for the control of a servo hydraulic positioning
system with similar success.

23
2.8 Concluding Remarks and Discussions

This Chapter has summarized some of the research that has been conducted on the applica-
tion of controls to fluid power systems. In general, the tracking performance and precision
of positioning systems have improved significantly when compared to traditional PID con-
trollers. In addition, because of the nature of the control algorithms, the stability of the
systems has been assured (within the accuracy of the models used in their development).
The literature has shown that many of the control approaches require some form of
linearization. For fluid power systems, this is a significant issue because of the degree of
nonlinear behavior. The introduction of techniques that treat nonlinearities and/or inexact
parameter values as disturbances or uncertainties has certainly enhanced the application of
controls to hydraulics as well as other areas. It remains a challenge to continue to develop
controllers which require models that do not require stringent conditions on the accuracy.
Further, there exists a need to develop explanations of how these controllers work such
that general practitioners can understand the “physics” behind the ideas. However, with
all controllers, the increased “overhead” associated with these modern control approaches
has necessitated more sophisticated electronic computing systems. In addition, the abil-
ity to “fix” any problems in an application situation is significantly limited to those who
have expertise in the area. Given the many benefits of advanced controllers, perhaps the
positives outweigh the negatives.
It has been observed that fuzzy controllers have been applied to industrial hydraulic
application because there is no need for an exact model of the system. Furthermore, fuzzy
controller concepts are easier to understand by experienced control engineers because of
the linguistic property of the scheme. The implementation of the controller is also straight
forward and can be simply implemented in microcontrollers by only adding in a few lines
of codes. If similar conditions could exist for other sophisticated control designs, perhaps
more designers might be inclined to use these controllers rather than reverting back to the
conventional PID controller.
This chapter has provided a comprehensive review of control algorithms applied to hy-
draulic systems. In this dissertation, not all the reviewed control algorithms were applied.

24
Three controllers were developed for the hydraulic system in the thesis.
As stated previously, slip-stick friction, modeling uncertainty and nonlinearity are crit-
ical issues in high performance control for hydraulic systems. In previous research, it was
discovered that hydraulic position control systems demonstrate certain phase delays due
to actuator friction, oil compressibility etc. Previous control design works are mostly
focused on feedback frameworks. In this dissertation, a feedback PID framework is com-
bined with a feedforward loop to compensate the phase delays. The optimized feedback
and feedforward gains are explicitly solved using H∞ optimization.
The behavior of the hydraulic actuator friction (slip-stick friction) is nonlinear and
unstable especially when the actuator is moving at very low speeds, which is a major
issue in high precision positioning control. However, slip-stick friction is a very complex
phenomenon that could not be easily modeled. In this dissertation, instead of developing
an accurate model of the friction, the parameters of the friction model were considered to
be varied in a bounded range. The varying ranges were transformed into norm bounded
uncertainties. A robust sliding mode control algorithm was designed considering the norm
bounded uncertainties.
In order to test the control performance of different controllers, a comparison study
of the performance (positioning, tracking and bandwidth) of the proposed algorithms on
a model of the EHA system is presented. This study would provide a “test-bench” to
compare the performance of the controllers.

25
C 3
S  M

3.1 Introduction
This section will present a summary of the three manuscripts that have been published in
Journals or Proceedings and represent the major contributions of this dissertation. Two
other publications have been submitted but are not included here. For each paper, the
source of the publication, the objective, the design procedures, the main results, the orig-
inal contributions, and final concluding statements are presented. It should be noted that
these three papers are all based on the discrete-time control system design.

3.2 Experimental Apparatus and Procedures


Experimental results that were published in the three papers were based on two test se-
tups, one for the servo valve controlled HPCS and the other for the pump controlled EHA
system respectively. The designed discrete-time controllers were realized using the Real-
Time Target Toolbox in MATLAB. The input signals were also generated using MATLAB.
In paper #1, the sampling time chosen in the experiment was 0.003s (satisfying Nyquist-
Shannon sampling theorem and hardware limitation). The supply pressure was set to be
500psi. The zero position of the actuation was at the middle of the entire stroke. The ex-
periments were all done within a room temperature. The experiment for each input signal
was repeated and verified for more than 10 tests.
In paper #2 and #3, the sampling time chosen in the experiments were 0.001s (satis-
fying Nyquist-Shannon sampling theorem and hardware limitation). The pressure of the
make up accumulator was set to be 50psi. The zero position of the actuation can be any

26
position within the entire stroke range. The experiments were all done within a room tem-
perature. The experiment for each input signal was repeated and verified for more than 10
tests.

3.2.1 Experimental test apparatus of the valve controlled HPCS

The hydraulic positioning system is presented in Figure 3.1. The actuator (SHEFFER) was
controlled by a flow servo valve (MOOG 15-010) which was connected with a Pentium
IV PC computer. A linear variable differential transducer (DCLVDT LUCAS SHAEVITZ
5000 DC-E) was used to measure the position information of the load with an accuracy
of 0.078 volts/mm. The analog measurement signal (cylinder position) was fed back to
the computer through a 12 bit A/D and D/A board (National Instruments PCI-6025E).
The supplied pressure is 4MPa (600 psi). The load on the actuator is a 2.63 kg mass.
Slip-stick friction characteristics existed in the actuator, and thus a linear model was only
an approximation of the nonlinear real system. The hypothesis was that if the controller
could work well in a system that had known nonlinear behavior using a linear model, then
the “computational expense” of having to incorporate nonlinear characteristics into the
proposed controller design may not be necessary for some applications.

Figure 3.1: Hydraulic positioning control system.

27
3.2.2 Experimental test apparatus of the EHA

A fixed displacement, variable speed, bidirectional gear pump was used to drive the actu-
ator. The experimental EHA system under study is shown in Figure 3.2. The hardware-in-
the-loop experimental test bench included the following components: Pentium IV com-
puter, PCI-DAS1602/16 Analog & Digital I/O Board, Gurley LE18 Linear Encoder and
the designed bidirectional hydraulic circuit.

Figure 3.2: Experiment setup of the EHA control system.

3.3 Paper #1: Discrete-Time H2-Optimal Control for a


Servo Hydraulic Position Control System
This work has been presented at the ASME International Mechanical Engineering Congress
and Exposition in Seattle, 2007; an extended version was published in the International
Journal of Advanced Mechatronics. The complete manuscript is included in Appendix B.

28
3.3.1 Objective

In this study, a discrete-time linear H2 -optimal controller was designed for a servo hy-
draulic position control system achieving output tracking performance. Physically, H2 -
optimal controller was designed for the purpose of minimizing the energy of the output
signal. A linearized model was obtained and verified experimentally. Based on this model,
a discrete-time linear H2 -optimal controller was derived and applied to the experimental
system. Model uncertainties and nonlinearities were not considered in the design process.
The objectives of this paper were three-fold:

• To develop a model of the hydraulic position control system (HPCS) using the fre-
quency domain model identification technique.

• To achieve a discrete-time state-space standard optimization control model for the


feedback HPCS control system with a reference following model.

• To design a discrete-time H2 -optimal controller for the HPCS system using the lin-
ear matrix inequality (LMI) technique. What this means is, instead of designing a
conventional continuous-time controller and then transforming it into a discrete-time
controller, the discrete-time design method will be used to design the discrete-time
controller directly.

3.3.2 Design procedures

1. Modeling of the system: The model of the system was identified by an open-loop
system frequency response generated by a signal analyzer, shown in Figure 3.3. A
third-order linear transfer function was derived and the parameters of the transfer
function were determined by the magnitude and phase Bode plot as shown in (3.1).
The Bode plot of the achieved linear model was also shown in Figure 3.3 (solid line)
to compare with the experimental results.

422000
G(s) = . (3.1)
s3 + 141s2 + 12100s

29
Figure 3.3: Experimental open-loop frequency responses (magnitude and
phase).

2. Transformation:

The closed-loop control system block diagram (shown in Figure 3.4) was “packed”
and transformed into a generalized optimal control system form. This can be ob-
served in Figure 3.5. The model of the closed-loop system (except for the controller)
was reformatted into one packed block.

Figure 3.4: Model reference feedback control system..

The generalized plant G of the control system was developed as:

30
Figure 3.5: Generalized control system diagram for the packed plant.

x˙g (t) = Ag xg (t) + Bg1 ω(t) + Bg2 u(t),


eg1 (t) = Cg1 xg (t) + Dg11 ω(t) + Dg12 u(t), (3.2)
eg2 (t) = Cg2 xg (t) + Dg21 ω(t) + Dg22 u(t).

3. Discretization:

Since the design approach of this paper was based on the discrete-time framework,
the model of the system was transformed into a discrete-time one. The continuous-
time generalized plant was then discretized using the step-invariant transformation
[Chen and Francis, 1995]:

ZT s
Agd = eT s Ag , [Bg1d Bg2d ] = [B1 B2 ] eτAg dτ. (3.3)
0

The discrete-time linear time-invariant model of the generalized plant is now given
by:

31
xg (k + 1) = Agd x(k) + Bg1d ω(k) + Bg2d u(k),
eg1 (k) = Cg1 x(k) + Dg11 ω(k) + Dg12 u(k), (3.4)
eg2 (k) = Cg2 x(k) + Dg21 ω(k) + Dg22 u(k).

4. H2 optimal control design:

The last step was to design an optimal controller to minimize the 2-norm ||e||2 of
the closed-loop control system using the LMI design technique. The discrete-time
controller to be designed has the following state-space form:

xk (k + 1) = Ak xk (k) + Bk eg2 (k),


(3.5)
u(k) = Ck xk (k) + Dk eg2 (k).
Then, the closed-loop system is:

xcl (k + 1) = Acl xcl (k) + Bcl ω(k),


(3.6)
eg1 (k) = Ccl xcl (k) + Dcl ω(k).
Here, xcl is the state vector of the closed-loop control system, Acl , Bcl , Ccl , Dcl are the
system matrices of the closed-loop control system,
 
 x (k) 
 g  ,
xcl (k) = 
 x (k) 
k
   
 A + B D C   B + B D 
 gd g2d k g2 Bg2d C k   g1d g2d g21 
Acl :=   , Bcl :=   ,
 BkCg2 Ak Bk Dg21
  h i
Ccl := Cg1 + Dg12 DkCg2 Dg12Ck , Dcl := Dg11 + Dg12 Dk Dg21 .

Then, using the following theorem, the discrete-time H2 -optimal controller can be
designed:

Theorem 1: There exists a controller in form such that the inequality ||Zωe1 (ς)||22 < µ
holds if, and only if, the following LMIs hold,

trace(W) < µ
 
 W Cg1 X + Dg12 L Cg1 + Dg12 RCg2 
 
 
 ∗ X + XT − P I + ST − J  > 0
 
 
∗ ∗ Y + YT − H

32
 
 P Agd X + Bg2d L Agd + Bg2d RCg2 Bg1d + Bg2d RDg21 
J
 
 
 ∗ H Q Y Agd + FCg2 Y Bg1d + FDg21 
 
 T  > 0
 ∗ ∗ X+X −P I+S −JT
0 
 
 ∗ ∗ ∗ Y + XT − H 0 
 
 
∗ ∗ ∗ ∗ I
Dg11 + Dg12 RDg21 = 0

where the matrices X, L, Y, F, Q, R, S , J are symmetric matrices and P, H, W are


the variable matrices.

3.3.3 Main results

From the simulation and experimental results (Figures 3.6 and 3.7) it can be observed that
the output of the experimental HPCS can track the reference output signal with a tracking
error less than 10% by applying the discrete-time H2 -optimal output tracking control. This
is larger than the maximum tracking error observed in the simulated results (4%). Upon
reflection, the differences were attributed to the fact that: (1) the nonlinear behavior of the
physical system was not incorporated into the model, (2) uncertainties were not considered
in the controller design.

Figure 3.6: Simulation tracking response of the H2 optimal.

33
Figure 3.7: Experimental tracking response of the H2 optimal.

3.3.4 Conclusions and contributions

This paper focuses on the output tracking control design for the HPCS. First, the model
of an experimental HPCS was set up, and then an H2 -optimal model reference controller
was designed. Finally, experimental tests on an HPCS were conducted. Both simulation
and experimental studies illustrated that the H2 -optimal model reference control scheme
could be used for tracking of a continuous input for a hydraulic positioning systems with a
tracking error less than 10% of the full stroke. For some applications such as off highway
applications, this is certainly acceptable but in more precision type situations such as pre-
cision machining, this is not acceptable. It should be recalled that the original hypothesis
was “if the controller could work well in a system that has known nonlinear behaviour
using a linear model, then the “computational expense” of having to incorporate nonlinear
theory into the proposed controller design could be avoided”. It is clear that for high-
precision tracking, this hypothesis was not valid for this controller design.
The contribution of this paper is as follows:

• This paper was the first work applying the discrete-time H2 -optimal algorithm on an
electro-hydraulic system. The applicability of the discrete-time H2 -optimal control
for the hydraulic system was verified.

34
3.4 Paper #2: H∞ PI Plus Feedforward Control Design
for an ElectroHydraulic Actuator System

This work has appeared in the proceedings of the 6th FPNI PhD Symposium in Purdue,
West Lafayette, 2010. The complete manuscript is presented in Appendix C. This research
builds on the discrete design approach that was presented in section 3.3.

3.4.1 Objective

The proportional-integral-derivative (PID) control has been the most adopted control method
for industrial applications. It has been estimated that over 90% of the controllers in use
today are PID controllers, even though other advanced control theories and practical de-
sign methods exist. One of the most famous PID tuning methods proposed by Ziegler and
Nichols (Z-N method) has been extensively used by control engineers over the past sev-
eral decades. However, when high precision performance is required, the traditional PID
controllers may not be adequate to deal with disturbance, noises and model uncertain-
ties. Thus, advanced PID tuning or a new design method is required for high performance
control purpose.
One motivation for this work was to design a new PID design method which can deal
with the aforementioned problems. In many studies, the PID control design was trans-
formed into a static output feedback (SOF) problem; in this approach, an SOF scheme
is determined so that the closed-loop system can achieve the desired characteristics. It
should be noted that SOF problem is one of the most important open research topics in
control theory and applications. To date, there is still no general solution to solving this
problem for all type of systems.
In this study, LMI conditions with nonlinear constraints are developed to find an H∞
performance PID controller for an electrohydraulic actuator (EHA) system. In many con-
trollers that have been developed for this hydraulic system, state estimators were designed
to estimate all the states which made the design process complicated. SOF controller de-
sign does not require all the states of the system, and hence, makes it much easier to be

35
realized in practice.
In the proposed PID control design, not only the PID feedback control design is aug-
mented into an SOF problem, a new feedforward term is also introduced into the system
to compensate for the tracking error during the transient period. In this work, the con-
trol system containing both the feedback PID controller and the feedforward control is
augmented into a new type of SOF control design problem by introducing new state and
output vectors. The SOF controller design is transformed into a bilinear matrix inequality
(BMI) based optimization problem. As such, then, the main objectives of this work were
three-fold:

• To introduce a feedforward term into the traditional PID feedback controller frame-
work and to transform the problem into an SOF problem by defining new state and
output vectors.

• To solve the augmented SOF design problem by satisfying a set of LMI conditions.

• To determine and demonstrate how a traditional PI plus feedforward controller can


be successfully applied to the EHA system for a tracking process.

3.4.2 Design procedures

This section briefly describes the procedures that were used in achieving the controller
design objectives.

1. EHA system modeling:

As a first step, it was necessary to achieve a physical model of the EHA system and
discretize it to achieve a discrete-time model. The discrete-time model of the EHA
system is represented by the following linear system:

xk+1 = Ad xk + Bd (uk + v1k ), (3.7)

yk = Cd xk + v2k ,

where k is the sampling instant, xk ∈ Rn is the state vector of the EHA system,
yk ∈ R p is the measurement output, uk ∈ Rm is the control input signal, v1k is the

36
disturbance or noise at the input and v2k represents the measurement noise. The PI
plus feedforward control law is given by:

X
k−1
uk = K p ek + Ki ei + K f f (rk−1 − rk ). (3.8)
i=0

Here, K p and Ki are the feedback proportional and integral gains, K f f is the feedfor-
ward gain to be designed.

2. Transforming the PI plus feedforward controller into an SOF controller:


Pk−1
1) Define a new state vector x̄k = [xkT , i=0 eTi , rk−1
T
]T and a new output sequence
P
ȳk = [eTk , k−1 T T T
i=0 ei , G(rk−1 − rk ) ] .

2) Here, ek = rk − yk is the controlled tracking error of the EHA system, G is a


weighting factor. Then the augmented system is written by the following form:

x̄k+1 = Ā x̄k + B̄uk + B̄1 ωk , (3.9)

ȳk = C̄ x̄k + D̄k ,

where
   
 Ad 0 0   Bd 
   

Ā =  −Cd  , B̄ = 
 
I 0  
 0  ,
   
   
0 0 0 0
   
 0 Bd 0   −Cd 0 0 
   
B̄1 =  I
  , C̄ =  
0 −I   0 I 0  ,
   
  
I 0 0 0 0 GI
   
 I 0 −I   rk 
   
D̄ =  0
  , ω =  
 .
0 0  k  v1k
   
  
−GI 0 0 v2k

3. H∞ optimization

The controlled output sequence is chosen as

z̄k = Exk + Fuk , (3.10)

37
where E = R [0 I 0], R and F are another two weighting matrices. The closed-
loop control system for H∞ optimization can be represented by the following form

x̄k+1 = [Ā + B̄K C̄] x̄k + [ B̄K D̄ + B̄1 ]ωk , (3.11)

z̄k = [E + FK C̄] x̄k + FK D̄ωk .

Then the H∞ optimization problem for the SOF controller design leads to the fol-
lowing theorem.

Theorem 1: The PI plus feedforward controller guarantees that the closed loop
system is stable and ||T ωz̄ ||∞ < γ if there exist symmetrical matrices P > 0, Q > 0
and matrix K, with appropriate dimensions, which satisfy the following LMIs
 
 −Q [ Ā + B̄K C̄] [ B̄K D̄ + B̄ ] 0 
 1 
 T 

 ∗ −P 0 [E + FK C̄] 
  < 0, (3.12)
 
 ∗ ∗ −γI [FK D̄]T

 
∗ ∗ ∗ −γI
 
 P I 

  > 0, (3.13)
I Q 
and a constraint
PQ = I. (3.14)

The conditions of (3.12), (3.13) and (3.14) are in fact an LMI problem with the non-
convex constraint. It can be conveniently solved by using the cone complementarity
linearization (CCL) algorithm. The results of solving the LMI problem give rise to
the matrix K, and then PI plus feedforward controller gains can be determined.

3.4.3 Main results

The experimental studies for this proposed PI plus feedforward controller are shown in
Figures 3.8, 3.9 and 3.10 to confirm the observations made by simulations. For the pur-
pose of comparison, a Z-N tuning PI controller was designed for the EHA system. The
experimental tracking error is also shown in Figure 3.10.
From the experimental results, it can be summarized that:

38
0.03
reference signal
0.025 tracking response

Piston Position (m)


0.02

0.015

0.01

0.005

−0.005
0 1 2 3 4 5 6
Time (s)

Figure 3.8: Experimental tracking response with the feedforward.

0.03
reference signal
0.025 tracking response
Piston Position (m)

0.02

0.015

0.01

0.005

−0.005
0 1 2 3 4 5 6
Time (s)

Figure 3.9: Experimental tracking response without the feedforward.

1. From Figure 3.10, the trend of the experimental tracking error using the proposed PI
plus feedforward controller is similar comparing with the one without a feedforward
loop. However, the tracking error did not decrease as much as was predicted by sim-
ulation. It was suspected that because the experimental EHA system is inherently
nonlinear, the linear EHA model used for the controller design would “miss” some
of the dynamic terms of the system.

2. Despite its simple formulation, the proposed PI plus feedforward controller was able
to achieve a reasonably accurate tracking performance (in which the tracking error
is less than 3.6% of the full tracking displacement).

39
−3
x 10
4

3
H∞ PI controller Z-N PI controller
without feedforward
2
Tracking Error (m)
1

−1
H∞ PI controller
−2 with feedforward

−3

−4
0 1 2 3 4 5 6
Time (s)

Figure 3.10: Experimental tracking errors.

3. As observed in Figure 3.10, the H∞ PI controller outperforms the traditional Z-N PI


tuning method.

3.4.4 Conclusions and contributions

This work has proposed a discrete-time PI plus feedforward controller design for an EHA
system. The controller formulation was transformed into an SOF problem with H∞ perfor-
mance. LMI optimization technique was applied to solve the controller design problem.
Simulation studies and experimental tests on the EHA system verify the effectiveness of
the proposed method for position tracking from the application perspective. The extra
feedforward term did improve the tracking performance significantly. Despite its sim-
ple framework, the proposed PI plus feedforward controller achieved very good tracking
performance.
The contribution of this paper is that:

1. A feedforward term was added in the feedback framework of the PI feedback control
system. The problem was successfully transferred into an SOF control framework.

2. The proposed method outperforms the conventional Z-N PID tuning method ex-
perimentally. The tracking performance was improved with the extra feedforward

40
control loop.

3. This work was the first of its kind in solving a PI plus feedforward controller using
the LMI optimization technique.

41
3.5 Paper #3:Modeling and Robust Discrete-Time Sliding
Mode Control Design for a Fluid Power ElectroHy-
draulic Actuator (EHA) System
This work was presented at the ASME Dynamic System Control Conference in Hollywood,
2009; an extended version was published in the IEEE/ASME Transactions on Mechatron-
ics. The complete manuscript is presented in Appendix D. As was done in the first two
papers, the discrete-time design approach was also used in this work.

3.5.1 Objective

Friction compensation for the position control of a mechatronic system has received much
attention. For the particular EHA system under study, it was ascertained that hydraulic
friction in the actuator was a critical issue affecting system performance in terms of accu-
racy and repeatability of the actuator performance. Trying to clearly identify which model
best describes the physical system under consideration is often very difficult and this is
still an open research area. Instead of pursuing an accurate model for the nonlinear fric-
tion, an appropriate scheme to compensate for the nonlinear friction uncertainties can be
designed – this observation motivates the work of this study.
In [Chinniah, 2004], experimental results showed that static, coulomb and viscous
friction (slip-stick) co-exist in the EHA’s linear actuator. A typical friction characteristic of
the EHA of interest is illustrated in Figure 3.11. In this figure, a curve fitting equation was
used to approximate the actual measured friction characteristics. However, some concerns
were raised by researchers about the characteristics at high and at very low velocities, and
the rate of change of velocity. In addition, hysteresis was noted but not reflected in the
model.
Sliding model control (SMC) first appeared in the context of variable structure sys-
tems. Due to its low sensitivity to disturbances and parameter variations, SMC has become
an efficient tool in the control of complex systems with uncertainties [Utkin et al., 1999]. It
appears that a robust discrete-time SMC (DT-SMC) design for systems with uncertainties

42
50

40

Friction Force (N) 30

20

10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Piston Velocity (m/s)

Figure 3.11: Experimental friction in the EHA and the identified nonlinear
friction model: (1) ∗ points: Experimentally measured friction force; (2)
Solid curve: Estimated curve for the friction force; (3) Dashed dotted line:
Approximated linear friction force.

has not been fully investigated from the application perspective, which indeed, motivates
this work. The development of a robust DT-SMC for the EHA system from an application
perspective is the focus of the paper. The main objectives of this work are three-fold:

• To further analyze the effects of the uncertain friction in the EHA system, and to
model them as uncertain system matrices with norm bounds. The bounded region
covers all nonlinear friction behavior for the EHA system considered.

• To design an appropriate sliding surface for the EHA system and to determine its
parameters by efficiently solving a set of LMIs .

• To develop a DT-SMC for the EHA system such that the reaching motion satisfies
the discrete-time sliding mode reaching condition.

3.5.2 Design procedures

This section briefly describes the procedures that were used in achieving the controller
design objectives.

1. Nonlinear modeling with uncertainties:

43
It has been observed from the experimental testing on the EHA that the variations of
friction parameters (a1 , a2 , a3 in (3.15)) would result in changes to the EHA system
model.

F f = a1 ẋ2 + a2 ẋ + a3 ẋ > 0, (3.15)

F f = −a1 ẋ2 + a2 ẋ − a3 ẋ < 0.

Therefore, it was proposed to model the friction uncertainties as norm-bounded un-


certainties on the system model. This uncertainty is characterized by a ∆A matrix in
the following state-space model of the EHA system:
 
 x1 (k + 1) 
 
 
X(k + 1) =  x2 (k + 1)  (3.16)
 
 
x3 (k + 1)
 
 x1 (k) 
 

= (A + ∆A)  x2 (k) 
 (3.17)
 
 
x3 (k)
 
 w1 (k) 
 
 
+B[u(k) + ∆ f (x(k), k)] + T s  w2 (k)  ,
 
 
w3 (k)

In general, ∆A can be expressed in a general norm-bounded uncertainty form, that


is: ∆A = GD(k)H1 , where G, H1 are known real constant matrices with compatible
dimensions, D(k) is unknown norm-bounded and generally satisfies D(k)T D(k) ≤ I.
Specifically for the EHA system, the uncertainty of the system matrix A arises from
the uncertainty of a2 : ∆a2 . By analyzing the experimental data, it is recognized that
10% variations on a1 , a2 , and a3 can reasonably capture the real friction variation.
For the nominal value a2 = −1450, there exists a variation ∆a2 satisfying |∆a2 | ≤
0.1|a2 |.

Further, as stated, ∆A can be expressed in a general norm-bounded uncertainty form:

∆A = GDH1

44
 
 0 
   
 
=  0  D(k) 0 0.111 0.073 (3.18)
 
 
0.1

with D(K) in this specific case, a scalar satisfying −1 ≤ D(k) ≤ 1.

The uncertainties introduced by time-varying friction parameters are illustrated in


Figure 3.12: (1) When the parameter variations are 10% on a1 , a2 , and a3 , simulta-
neously, that is, 0.1ai , (i = 1, 2, 3), then it results in the upper bound friction curve;
(2) when the parameter variations are −0.1ai , (i = 1, 2, 3), then the lower bound
friction curve is obtained.

50
upper bound of
Friction Force Ff (N)

40 the friction

30

20 lower bound of
the friction
10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Piston Velocity (m/s)

Figure 3.12: Nonlinear uncertain friction band.

2. Model transformation for the SMC design:

A transformation matrix J was introduced to facilitate the following SMC design.


This transformation is highly desirable in that one of the state equations will become
independent of the control input and system nonlinearities.

3. Sliding surface design:

A Lyapunov stability technique and LMI approach were applied to design the sliding
surface. A theorem was developed with LMIs to calculate the coefficients of the
sliding surface:

45
Theorem 1: The system is robustly asymptotically stable if there exist symmetrical
matrices P > 0, Y > 0, ψ > 0 and matrix C, with appropriate dimensions, which
satisfy the following LMI
 
 −P ∗ ∗ 
 
 
 Ā11 − Ā12C ψU2TGGT U2 − Y ∗  < 0,
 
 
H1 U2 − H1 U1C 0 −ψ

and a constraint
PY = I.

4. Sliding mode control design: a sliding mode control theorem was designed based on
the developed sliding surface to design the sliding mode control signal. The sliding
mode reaching condition and the sliding mode bound were satisfied.

Theorem 2:Assuming that the LMIs in Theorem 1 have feasible solutions P, Y, ψ


and C. With the designed sliding surface, the sliding mode reaching condition can
be satisfied with the following control law:

u(k) = −B−1
2 [C̄ Āze (k) − (1 − T s q) · S (k) + εT s · sgn(S (k))

+ (α0 + β0 ) + (αδ + βδ ) · sgn(S (k))],

where

αU + α L αU − αL
α0 = , αδ = ,
2 2
βU + βL βU − βL
β0 = , βδ = , Ā = JAJ −1 .
2 2

3.5.3 Main results

The same tracking input signal was applied both to a model of the EHA and to the exper-
imental system. The simulation and experimental tracking response of the EHA system
are shown in Figure 3.13 and Figure 3.14.
There exists a small error during the transient period; indeed, a time-delay during the
transient portion is observed (indeed, it was also observed in [Wang et al., 2008]). To
try to improve this issue, a pure time-delay about 0.03 second was added into the system

46
model. However, once the loop was closed and the designed controller was implemented,
the system went unstable. This was also observed using a PID controller. At this point,
a plausible explanation for this issue is not available. It is suspected that the nonlinear
model is missing some dynamic terms which need to be accounted for in the controller
design. Future research will lead to new challenges to design a controller based on the
same design philosophy which can account for these new additional dynamics.

0.025

0.02
Piston Position (m)

0.015

0.01

0.005 simulation tracking response


tracking reference signal

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure 3.13: Simulation tracking response of the EHA system with uncer-
tainties using the proposed method.

0.025

0.02

tracking reference
Piston Position (m)

0.015

0.01

0.005 experimental
tracking response

−0.005
0 0.2 0.4 0.6 0.8 1
Time (s)

Figure 3.14: Experimental tracking response of the EHA system with


model uncertainty using the proposed method.

47
3.5.4 Conclusions and contributions

This paper concluded that the proposed upper and lower boundary curves do encompass
measured experimental friction forces. Based on the Lyapunov theory and LMI technique,
the asymptotically stable sliding surface can be conveniently designed based on the fea-
sible solutions of the derived LMI with a constraint. Further, the robust DT-SMC law
guarantees the sliding mode reaching condition. Simulation and experimental results on
the EHA system verify the effectiveness and the performance improvement by applying
the proposed method from the application perspective.
The contribution of this paper is that:

1. A discrete-time nonlinear model of the EHA systems was developed considering the
nonlinear friction model of the actuator. Parameter uncertainties were considered
and incorporated in the nonlinear model.

2. An original sliding surface design method using the LMI technique was developed.
Parameter uncertainties were incorporated into the actuator friction force. A theo-
rem regarding to the robust sliding surface design was achieved in this paper.

3. A sliding mode control law guaranteeing the sliding mode reaching condition was
developed.

48
C 4
C   D C M 
 M   EHA S

In the previous Chapter, three different control methods which were proposed in three
papers were briefly introduced. In this Chapter, in order to compare the positioning, track-
ing and bandwidth performance of the three controllers, the control responses to quasi-step
and sine wave inputs are examined. In addition, the frequency responses of the EHA model
and controllers are presented. It should be noted, because some of the experimental equip-
ment has failed at present, it was not possible to do the comparison in an experimental
system. However, a fully developed and verified large signal nonlinear model of the EHA
system which included effects of bulk modulus, leakages and nonlinear actuator frictions
was used as the physical plant of the EHA system. The time domain tracking responses
are presented in the following order: the tracking response followed by the tracking error
and the control signal.

4.1 Quasi-Step Tracking Response

The quasi-step tracking response using the proposed control methods are shown in Figure
4.1(a), Figure 4.2(a) and Figure 4.3(a). The tracking errors are shown in Figure 4.1(b),
Figure 4.2(b) and Figure 4.3(b). From the tracking errors, it can be concluded that the ro-
bust sliding mode control (RSMC) possessed the best tracking performance. The control
signal with the RSMC is shown in Figure 4.3(c). It has a large “kick” at the beginning of
each step, which quickly forces the tracking response to follow the tracking reference. The
H2 -optimal control and H∞ PI plus feedforward control are both linear control methods.

49
They both showed some oscillation on the tracking response at the end of each step stroke
where the actuator friction force is in a very nonlinear region. The H∞ PI plus feedfor-
ward control has better tracking performance than the H2 -optimal control; the time-delay
phenomenon observed in the H2 -optimal control was significantly decreased.

4.2 Sine Wave Tracking Response

The sine wave tracking responses using the proposed control methods are shown in Fig-
ure 4.4(a), Figure 4.5(a) and Figure 4.6(a). The trend of the sine wave tracking response
demonstrated similar performance with the quasi-step tracking response. The robust slid-
ing mode control possessed the best tracking performance, followed by the H∞ PI plus
feedforward control and the H2 -optimal control.

4.3 Frequency Response

The frequency responses of the controlled EHA system using the three proposed con-
trollers are shown in Figure 4.7. It can be observed that the RSMC control system has a
flat frequency response up to 220(rad/s) and suddenly drops down. H∞ PI plus feedforward
control and H2 -optimal control both possess a resonance peak on the frequency response.
However, if a ±3db wide tolerance band is considered for the frequency response, the H∞
PI plus feedforward control would have the widest bandwidth.

50
−3
x 10
15

tracking reference
10 tracking response using H2−optimal control

Piston Position (m)


5

−5
0 0.5 1 1.5 2 2.5 3
Time (s)

(a) Tracking response.


−3
x 10
4

2
Position Error (m)

−2

−4

−6
0 0.5 1 1.5 2 2.5 3
Time (s)

(b) Tracking error.

10
Control Signal (Volt)

−5

−10
0 0.5 1 1.5 2 2.5 3
Time (s)

(c) Control signal.

Figure 4.1: Quasi-step tracking response of the EHA system using H2 -


optimal control

51
−3
x 10
15
tracking reference
tracking response using H PI control

10

Piston Position (m)


5

−5
0 0.5 1 1.5 2 2.5 3
Time (s)

(a) Tracking response.


−3
x 10
4

2
Position Error (m)

−1

−2

−3

−4
0 0.5 1 1.5 2 2.5 3
Time (s)

(b) Tracking error.

10
Control Signal (Volt)

−5

−10
0 0.5 1 1.5 2 2.5 3
Time (s)

(c) Control signal.

Figure 4.2: Quasi-step tracking response of the EHA system using PI H∞


plus feedforward control.

52
−3
x 10
15
tracking reference
tracking response using RSMC
10

Piston Position (m)


5

−5
0 0.5 1 1.5 2 2.5 3
Time (s)

(a) Tracking response.


−3
x 10
4

2
Position Error (m)

−1

−2

−3

−4
0 0.5 1 1.5 2 2.5 3
Time (s)

(b) Tracking error.

10
Control Signal (Volt)

−5

−10
0 0.5 1 1.5 2 2.5 3
Time (s)

(c) Control signal.

Figure 4.3: Quasi-step tracking response of the EHA system using robust
sliding mode control.

53
tracking reference
0.025 tracking response using H −optimal control
2

Piston Position (m)


0.02

0.015

0.01

0.005

−0.005
0 0.5 1 1.5 2 2.5 3
Time (s)

(a) Tracking response.


−3
x 10
3

2
Position Error (m)

−1

−2

−3
0 0.5 1 1.5 2 2.5 3
Time (s)

(b) Tracking error.

4
Control Signal (Volt)

−2

−4

−6
0 0.5 1 1.5 2 2.5 3
Time (s)

(c) Control signal.

Figure 4.4: Sine wave tracking response of the EHA system using H2 -
optimal control.

54
tracking reference
0.025 tracking response using H∞ PI control

Piston Position (m)


0.02

0.015

0.01

0.005

−0.005
0 0.5 1 1.5 2 2.5 3
Time (s)

(a) Tracking response.


−3
x 10
3

2
Position Error (m)

−1

−2

−3
0 0.5 1 1.5 2 2.5 3
Time (s)

(b) Tracking error.

4
Control Signal (Volt)

−2

−4

−6
0 0.5 1 1.5 2 2.5 3
Time (s)

(c) Control signal.

Figure 4.5: Sine wave tracking response of the EHA system using H∞ PI
plus feedforward control.

55
tracking reference
0.025 tracking response using RSMC

Piston Position (m)


0.02

0.015

0.01

0.005

−0.005
0 0.5 1 1.5 2 2.5 3
Time (s)

(a) Tracking response.


−3
x 10
3

2
Position Error (m)

−1

−2

−3
0 0.5 1 1.5 2 2.5 3
Time (s)

(b) Tracking error.

4
Control Signal (Volt)

−2

−4

−6
0 0.5 1 1.5 2 2.5 3
Time (s)

(c) Control signal.

Figure 4.6: Sine wave tracking response of the EHA system using robust
sliding mode control.

56
20

Magnitude (db)
−20

−40

−60 H2−optimal control


H PI control

−80
RSMC

−100 1 2 3
10 10 10
Frequency (rad/s)

Figure 4.7: Frequency response of the EHA system with the proposed con-
trollers.

4.4 Summary
These results are obtained based on a nonlinear model of the EHA system. However,
the simulation model did reflect the nonlinear characteristics of the EHA system. The
time domain response and frequency response presented in the previous sections proved
that the robust sliding mode control would produce the best position tracking performance
compared with the other two controllers. The H∞ PI plus feedforward controller possesses
the widest bandwidth if a ±3db tolerance band is used as the criterion.

57
C 5
C, C  R

In this dissertation, the basic principles and concepts of advanced control algorithms
(PID control, adaptive control, sliding mode control, fuzzy control, H∞ control, ect.) de-
signed for hydraulic systems were comprehensively reviewed. The basic strategies of the
reviewed controllers were also briefly introduced using simple examples.
The overall objective of this research work was to develop advanced integrated feed-
back control algorithms, with applications to tracking control of a valve controlled hy-
draulic position control system (HPCS) and an electrohydraulic actuator (EHA) system.
For the two particular HPCS systems studied in this dissertation, three control algorithms
were developed to achieve position tracking performance: (1) H2 -optimal control, (2) H∞
PI plus feedforward control, (3) robust sliding mode control. Three papers that have been
published based on these works were summarized. Comparison of the proposed control
algorithms were supplied based on a nonlinear model of the EHA system using different
types of tracking input signals.
Discrete-time control algorithms were designed for the two HPCS systems studied in
the papers. With advancements in the industrial computer technology, control algorithms
for many practical applications can now be implemented in industrial control computers in
the discrete-time domain. Moreover, the designed controllers can be directly implemented
on micro-controllers to accomplish embedded control design.
Based on this research, the following conclusions are drawn as such:

1. It is concluded that: Discrete-time H2 -optimal control can be successfully applied


to a hydraulic servo valve driven positioning control system but only with marginal
accuracy.

58
2. It is concluded that: A PID plus feedforward control framework can be transformed
into a static output feedback (SOF) framework with proper augmentation of the sys-
tem vector. The LMI optimization technique can be applied to this SOF framework
to optimize the controller gains. The feedforward term successfully contributed to
improve the tracking and bandwidth performance significantly.

3. It is concluded that: The LMI optimization technique can be successfully applied


to a nonlinear system to determine the switching hyperplane for RSMC. A discrete-
time controller based on the SMC concept which used this LMI technique was ap-
plied to a particular nonlinear EHA system and proved robustness and high accuracy
with nonlinear friction in the actuator.

The major contributions of the research can be summarized as follows:

1. This research work has presented a comprehensive review of the control methods
developed for HPCS systems during the past decade.

2. This research work has, for the first time, applied the discrete-time H2 -optimal algo-
rithm on an HPCS system. The applicability of the discrete-time H2 -optimal control
for the HPCS was verified.

3. This research work has developed a new framework (SOF) from a PI plus feed-
forward control framework. The feedback and feedforward gains were explicitly
solved through H∞ optimization technique instead of traditional tuning.

4. This research work has introduced a novel controller called robust sliding mode con-
troller (RSMC) with the sliding mode surface designed considering the parameter
uncertainties in the nonlinear friction model.

Some possible future works are recommended and listed:

1. Friction estimation: In [Tafazoli et al., 1998b], the authors developed a friction es-
timator which estimated the friction of the system online. However, this friction

59
estimator was developed based on a simple Coulomb friction model. The perfor-
mance of the estimation can be improved if a more realistic model (slip-stick) of
the friction is used. It has been argued by researchers that the friction of hydraulic
actuators is associated with both velocity and acceleration of the actuator. If a math-
ematical model of hydraulic actuators based on the above argument can be derived
and applied, the friction estimation performance can be significantly improved.

2. Sliding mode control with friction estimation: Using the friction estimator men-
tioned above, the nonlinear friction in the actuator can be compensated using fric-
tion feedback. Meanwhile, the other parameters of the hydraulic system can be
controlled using a sliding mode controller. This can be considered as a form of dual
control strategy.

60
R
Alleyne, A., Hedrick, J.K., 1995. Nonlinear adaptive control of active suspension. IEEE
Transactions on Control Systems Technology 3, 94–101.

Alleyne, A., Liu, R., 2000. A simplified approach to force control for electro-hydraulic
systems. Control Engineering Practice 8, 1347–1356.

Armstrong-Helouvry, B., Dupont, P., de Wit, C.C., 1994. A survey of models, analysis
tools and compensation methods for the control of machines with friction. Automatica
30, 1083–1138.

Aström, K., Hägglung, T., 2001. The future of PID control. Control Engineering Practice
9, 1163–1175.

Bashash, S., Jalili, N., 2009. Robust adaptive control of coupled parallel piezo-flexural
nanopositioning stages. IEEE/ASME Transactions on Mechatronics 14, 11–20.

Bobrow, J.E., Desai, J., 1990. Experimental Robotics I. Springer Berlin / Heidelberg.
chapter Modeling and Analysis of a High Torque Hydrostatic Actuator for Robotic Ap-
plications. pp. 215–228.

Bobrow, J.E., Lum, K., 1995. Adaptive, high bandwidth control of a hydraulic actuator, in:
Proceedings of American Control Conference, Seattle, Washington, USA. pp. 71–75.

Bonchis, A., Corke, P.I., Rye, D.C., Ha, Q.P., 2001. Variable structure methods in hy-
draulic servo systems control. Automatica 37, 589–595.

Boyd, S., Ghaoui, L.E., Feron, E., Balakrishnan, V., 1994. Linear Matrix Inequalities In
System and Control Theory. SIAM.

Camacho, E.F., Bordons, C., 2004. Model Predictive Control. Springer.

Chen, C., Liu, L., Cheng, C., Chiu, G.T., 2008. Fuzzy controller design for synchronous
motion in a dual-cylinder electro-hydraulic system. Control Engineering Practice 16,
658–673.

Chen, H., Liu, Z., Sun, P., 2005. Application of constrained H∞ control to active suspen-
sion systems on half-car models. ASME Journal of Dynamic Systems, Measurement,
and Control 127, 345–354.

Chen, T., Francis, B.A., 1991. H2 -optimal sampled-data control. IEEE Transactions on
Automatic Control 36, 387–397.

61
Chen, T., Francis, B.A., 1995. Optimal Sampled-Data Control Systems. Springer-Verlag,
London.

Cheng, Y., Moor, B.L.R.D., 1994. Robustness analysis and control system design for a
hydraulic servo system. IEEE Transactions on Control Systems Technology 2, 183–197.

Chern, T.L., Wu, Y.C., 1991. Design of integral variable structure controller and applica-
tion to electrohydraulic velocity servosystems. IEE Proceedings D, Control Theory and
Applications 139, 439–444.

Chern, T.L., Wu, Y.C., 1992. An optimal variable structure control with integral com-
pensation for electrohydraulic position servo control system. IEEE Transactions on
Industrial Electronics 39, 460–463.

Chinniah, Y.A., 2004. Fault Detection in the Electrohydraulic Actuator Using Extended
Kalman Filter. Ph.D. thesis. University of Saskatchewan.

Cho, S., Burton, R., 2011. Position control of high performance hydrostatic actuation
system using a simple adaptive control (SAC) method. Mechatronics 21, 109–115.

Choi, H.H., 1998. An explicit formula of linear sliding surfaces for a class of uncertain
dynamic systems with mismatched uncertainties. Automatica 34, 1015–1020.

Choi, H.H., 1999. On the existence of linear sliding surfaces for a class of uncertain
dynamic systems with mismatched uncertainties. Automatica 35, 1707–1715.

Choux, M., Hovland, G., 2010. Adaptive backstepping control of nonlinear hydraulic-
mechanical system including valve dynamics. Modeling, Identification and Control 31,
35–44.

Corbet, T., Sepehri, N., Lawrence, P.D., 1996. Fuzzy control of a class of hydraulically
actuated industrial robots. IEEE Transactions on Control Systems Technology 4, 419–
426.

Dean, P.T., Fales, R.C., 2007. Modern control design for a variable displacement hydraulic
pump, in: Proceedings of American Control Conference, New York City, USA. pp.
3535–3540.

Dieulot, J.Y., Colas, F., 2009. Robust PID control of a linear mechanical axis: A case
study. Mechatronics 19, 269–273.

Doyle, J.C., Glover, K., Khargonekar, P.P., Francis, B.A., 1989. State-space solutions to
standard H2 and H∞ problems. IEEE Transactions on Control Systems Technology 34,
831–847.

Du, H., Lam, J., Sze, K.Y., 2003. Non-fragile output feedback H∞ vehicle suspension
control using genetic algorithm. Engineering Applications of Artificial Intelligence 16,
667–680.

62
Du, H., Sze, K.Y., Lam, J., 2005. Semi-active H∞ control of vehicle suspension with
magneto-rheological dampers. Journal of Sound and Vibration 28, 981–996.

Duraiswamy, S., Chiu, G.T.C., 2003. Nonlinear adaptive nonsmooth dynamic surface
control of electro-hydraulic systems, in: Proceedings of American Control Conference,
Denver, CO. USA. pp. 3287–3292.

Eker, I., 2006. Sliding mode control with PID sliding surface and experimental application
to an electromechanical plant. ISA Transactions 45, 109–118.

Fales, R., Kelkar, A., 2005. Robust control design for a wheel loader using mixed sensitiv-
ity H∞ and feedback linearization based methods, in: Proceedings of American Control
Conference, Portland, OR, USA. pp. 4381–4386.

Fales, R., Kelkar, A., 2008. Robust control design for a wheel loader using H∞ and
feedback linearization based methods. ISA Transactions 48, 312–320.

Frieland, B., Park, Y.J., 1991. On adaptive friction compensation, in: Proceedings of the
30th Conference on Decision and Control, Brighton, England. pp. 2899–2902.

Fung, R.F., Wang, Y.C., Yang, R.T., Huang, H.H., 1997. A variable structure control
with proportional and integral compensations for electrohydraulic position servo control
system. Mechatronics 7, 67–81.

Gahinet, P., 1994. A linear matrix inequality approach to H∞ control. International Journal
of Robust and Nonlinear Control 4, 421–448.

Gao, W., Wang, Y., Homaifa, A., 1995. Discrete-time variable structure control systems.
IEEE Transactions on Industrial Electronics 42, 117–122.

Grewal, M.S., Amdrews, A.P., 2001. Kalman filtering theory and practice using MAT-
LAB. JOHN WILEY & SONS.

Guan, C., Pan, S., 2008a. Adaptive sliding mode control of electro-hydraulic system with
nonlinear unknown parameters. Control Engineering Practice 16, 1275–1284.

Guan, C., Pan, S., 2008b. Nonlinear adaptive robust control of single-rod electro-hydraulic
actuator with unknown nonlinear parameters. IEEE Transactions on Control Systems
Technology 16, 434–446.

Ha, Q.P., Bonchis, A., Rye, D.C., Durrant-Whyte, H.F., 2000. Variable structure systems
approach to friction estimation and compensation, in: Proceedings of IEEE Interna-
tional Conference on Robotics & Automation, San Francisco, USA. pp. 3543–3548.

Ha, Q.P., Rye, D.C., Durrant-Whyte, H.F., 1999. Fuzzy moving sliding mode control with
application to robotic manipulators. Automatica 35, 607–616.

Habibi, S.R., Goldenberg, A.A., 2000. Design of a new high-performance electrohydraulic


actuator. IEEE/ASME Transactions on Mechatronics 5, 158–165.

63
Habibi, S.R., Richards, R.J., Goldenberg, A.A., 1994. Hydraulic actuator analysis for
industrial robot multivariable control, in: Proceedings of American Control Conference,
Baltimore, Mariland, USA. pp. 1003–1007.

Habibi, S.R., Sigh, G., 2000. Derivation of design requirements of optimization of a high
performance hydrostatic actuation system. International Journal of Fluid Power 2, 11–
27.

Hirata, T., Koizumi, S., Takahashi, R., 1995. H∞ control of railroad vehicle active suspen-
sion. Automatica 31, 13–24.

Horowitz, I., 1963. Synthesis of Feedback Systems. Academic Press, New York.

Houpis, C.H., Rasmussen, S.J., 1999. Quantitative Feedback Theory: Fundamentals and
Applications. Marcel Dekker Inc., New York.

Huang, Y.C., Chan, M., Hsin, Y.P., Ko, C.C., 2003. Use of PID and iterative learning
controls on improving intra-oral hydraulic loading system of dental implants. JSME
International Journal Series C 46, 1449–1455.

Janardhanan, S., Bandyopadhyay, B., 2006. Discrete sliding mode control of systems
with unmatched uncertainty using multirate output feedback. IEEE Transactions on
Automatic Control 51, 1030–1035.

Karpenko, M., Sepehri, N., 2010. On quantitative feedback design for robust position
control of hydraulic actuators. Control Engineering Practice 18, 289–299.

Kim, D.H., Tsao, T.C., 2000. A linearized electrohydraulic servovalve model for valve
dynamics sensitivity analysis and control system design. ASME Journal of Dynamic
Systems, Measurement, and Control 122, 179–187.

King, P.J., Mamdani, E.H., 1977. The application of fuzzy control systems to industrial
processes. Automatica 13, 235–242.

Knohl, T., Unbehauen, H., 2000. Adaptive position control of electrohydraulic servo
systems using ANN. Mechatronics 10, 127–143.

Krstic, M., Kanellakopoulos, I., Kokotovic, P., 1995. Nonlinear and Adaptive Control
Design. Wiley, New York.

Laurent, E.G., Francois, O., Mustapha, A., 1997. A cone complementarity linearization
algorithm for static output feedback and related problems. IEEE Transactions on Auto-
matic Control 42, 1171–1176.

Lee, C., Salapaka, S.M., 2009. Fast robust nanopositioning-a linear-matrix-inequalities-


based optimal control approach. IEEE/ASME Transactions on Mechatronics 14, 414–
422.

Lee, C.C., 1990. Fuzzy logic in control systems: fuzzy logic controller – parts I & II.
IEEE Transactions on Systems, Man, and Cybernetics 20, 404–433.

64
Lee, S.Y., Cho, H.S., 2003. A fuzzy controller for an electro-hydraulic fin actuator using
phase plane method. Control Engineering Practice 11, 697–708.

Lee, Y.H., Kopp, R., 2001. Application of fuzzy control for a hydraulic forging machine.
Fuzzy Sets and Systems 118, 99–108.

Lin, Y., Shi, Y., Burton, R., 2007. Discrete-time H2 -optimal control for hydraulic posi-
tion control systems, in: Proceedings of ASME International Mechanical Engineering
Congress and Exposition, Seatle, US.

Lin, Y., Shi, Y., Burton, R., 2009a. Discrete-time H2 -optimal output tracking control for an
experimental hydraulic positioning control system. International Journal of Advanced
Mechatronic Systems 1, 168–174.

Lin, Y., Shi, Y., Burton, R., 2009b. Modeling and robust discrete-time sliding mode control
design for a fluid power electrohydraulic actuator (EHA) system, in: Proceedings of
ASME Dynamic System and Control Conference, Hollywood, USA.

Lin, Y., Zhang, H., Shi, Y., Burton, R., 2010. H∞ PI plus feedforward controller design
for an electrohydraulic actuator (EHA) system, in: Proceedings of the 6th FPNI PhD
Symposium, Lafayette, USA.

Liu, G.P., Daley, S., 1999. Optimal-tuning PID controller design in the frequency domain
with application to a rotary hydraulic system. Control Practice Engineering 7, 821–830.

Liu, G.P., Daley, S., 2000. Optimal-tuning nonlinear PID control of hydraulic systems.
Control Practice Engineering 8, 1045–1053.

Liu, R., Alleyne, A., 2000. Nonlinear force/pressure tracking of an electro-hydraulic


actuator. ASME Journal of Dynamic Systems, Measurement, and Control 122, 232–
237.

Liu, S., Yao, B., 2003. Indirect adaptive robust control of electro-hydraulic systems driven
by single-rod hydraulic actuators, in: Proceedings of IEEE/ASME International Con-
ference on Advanced Intelligent Mechatronics, pp. 296–301.

Liu, Y., Handroos, H., 1999. Technical note sliding mode control for a class of hydraulic
position servo. Mechatronics 9, 111–123.

Mamdani, E.H., Assilian, S., 1975. An experiment in linguistic synthesis with a fuzzy
logic controller. International Journal of Man-Machine Studies 7, 1–13.

Manring, N., 2005. Hydraulic Control Systems. John Wiley & Sons.

Manring, N.D., Luecke, G.R., 1998. Modeling and designing a hydrostatic transmission
with a fixed-displacement motor. ASME Journal of Dynamic Systems, Measurement,
and Control 120, 45–49.

Marton, L., Lantos, B., 2009. Control of mechanical systems with stribeck friction and
backlash. Systems & Control Letters 58, 141–147.

65
Merritt, H.E., 1967. Hydraulic Control Systems. John Wiley & Sons.

Misawa, E.A., 1997. Discrete-time sliding mode control: the linear case. ASME Journal
of Dynamic Systems, Measurement, and Control 119, 819–821.

Niksefat, N., Sepehri, N., 2000. Design and evaluation of a robust force controller for
an electro-hydraulic actuator via quantitative feedback theory. Control Engineering
Practice 8, 1335–1345.

Niksefat, N., Sepehri, N., 2001. Designing robust force control of hydraulic actuators
despite system and environmental uncertainties. IEEE Control Systems Magazine 21,
66–77.

Njabeleke, I.A., Pannett, R.F., Chawdhry, P.K., Burrows, C.R., 2000. Design of H∞ loop-
shaping controllers for fluid power systems. Proceedings of the Institute of Mechanical
Engineering, Part C: Journal of Mechanical Engineering Science 214, 483–500.

Oliveira, M.C., Geromel, J.C., Bernussou, J., 2002. Extended H2 and H∞ norm char-
acterizations and controller parameterizations for discrete-time systems. International
Journal of Control 75, 666–679.

Owen, W.S., Croft, E.A., 2003. The reduction of stik-slip friction in hydraulic actuators.
IEEE/ASME Transactions on Mechatronics 8, 362–371.

Plummer, A.R., Vaughan, N.D., 1996. Robust adaptive control for hydraulic servo sys-
tems. ASME Journal of Dynamic Systems, Measurement, and Control 118, 237–244.

Porter, B., Tatnall, M.L., 1970. Performance characteristics of an adaptive hydraulic servo-
mechanism. International Journal of Control 11, 741–757.

Rotea, M.A., Khargonekar, P.P., 1989. Stabilization of uncertain systems with norm
bounded uncertainty – a control Lyapunov function approach. SIAM Journal on Control
and Optimization 27, 1462–1476.

Sam, Y.M., Osman, J.H.S., Ghani, M.R.A., 2004. A class of proportional-integral sliding
mode control with application to active suspension system. System & Control Letters
51, 217–223.

Sampson, E.B., 2005. Fuzzy Control of the ElectroHydraulic Actuator. Master’s thesis.
University of Saskatchewan.

Sepehri, N., Wu, G., 1998. Experimental evaluation of generalized predictive control
applied to a hydraulic actuator. Robotica 16, 463–474.

Shao, J., Chen, L., Sun, Z., 2005. The application of fuzzy control strategy in electro-
hydraulic servo system, in: Proceedings of IEEE International Conference on Mecha-
tronics and Automation, Niagra, Canada. pp. 2010–2016.

Slotine, J.J.E., 1985. The robust control of robot manipulators. International Journal of
Robotics Research 4, 49–63.

66
Sohl, G.A., Bobrow, J.E., 1999. Experiments and simulations on the nonlinear control
of a hydraulic servo system. IEEE Transactions on Control Systems Technology 7,
238–247.
Son, H.Y., Jeong, S.G., Choi, J.Y., Kim, J.K., Cheon, Y.S., Paek, Y.I., Kwon, S.I., Lee,
M.H., 2001. A robust controller design for performance improvement of a semi-active
suspension systems, in: Proceedings of IEEE International Symposium on Industrial
Electronics, Pusan, Korea. pp. 1458–1461.
Tafazoli, S., Silva, C.W., DLawrence, P., 1998a. Tracking control of an electrohydraulic
manipulator in the presence of friction. IEEE Transactions on Control Systems Tech-
nology 6, 401–411.
Tafazoli, S., de Silva, C.W., Lawrence, P.D., 1998b. Tracking control of an electrohy-
draulic manipulator in the presence of friction. IEEE Transactions on Control Systems
Technology 6, 401–411.
Tan, U.X., Latt, W.T., Shee, C.Y., Riviere, C.N., Ang, W.T., 2009. Feedforward
controller of ill-conditioned hysteresis using singularity-free prandtl-ishlinskii model.
IEEE/ASME Transactions on Mechatronics 14, 598–605.
Tang, C.Y., Misawa, E.A., 2000. Discrete variable structure control for linear multivariable
systems. Journal of Dynamic Systems, Measurement, and Control 122, 783–792.
Toufighi, M.H., Sadati, S.H., Najaif, F., 2007. Modeling and analysis of a mechatronic
actuator system by using bond graph methodology, in: Proceedings of IEEE Aerospace
Conference, pp. 1–8.
Tsao, T.C., Tomizuka, M., 1994. Robust adaptive and repetitive digital control and ap-
plication to hydraulic servo for non-circular machining. ASME Journal of Dynamic
Systems, Measurement, and Control 116, 24–32.
Tunay, I., Rodin, E.Y., Beck, A.A., 2001. Modeling and robust control design for aircraft
brake hydraulics. IEEE Transactions on Control Systems Technology 9, 319–329.
Utkin, V., Guldner, J., Shi, J., 1999. Sliding Mode Control in Electromechanical Systems.
CRC Press.
Utkin, V., Yang, K.D., 1978. Methods for constructing discontinuity planes in mutidimen-
sional variable structure systems. Automation and Remote Control 39, 1466–1470.
Utkin, V.I., 1993. Sliding mode control design principles and applications to electric
drives. IEEE Transactions on Industrial Electronics 40, 23–36.
Varseveld, R.B., Bone, G.M., 1997. Accurate position control of a pneumatic actuator
using on/off solenoid valves. IEEE/ASME Transactions on Mechatronics 2, 195–204.
Wang, S., Habibi, S., Burton, R., Sampson, E., 2006. Sliding mode control for a model of
an electrohydraulic actuator system with discontinuous nonlinear friction, in: Proceed-
ings of American Control Conference, Minneapolis, Minnesota USA. pp. 5897–5903.

67
Wang, S., Habibi, S., Burton, R., Sampson, E., 2008. Sliding mode control for an elec-
trohydraulic actuator system with discontinuous non-linear friction. Proceedings of the
Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering
222, 799–815.

Watton, J., 1989. Fluid Power Systems. NJ: Prentice-Hall.

Wit, C.C., Olsson, H., Astrom, K.J., Lischinsky, P., 1995. A new model for control of
systems with friction. IEEE Transactions on Automatic Control 40, 419–425.

Wu, G., Sepehri, N., Ziaei, K., 1998. Design of a hydraulic force control system us-
ing a generalised predictive control algorithm. IEE Proceedings - Control Theory and
Applications 145, 428–436.

Xie, L., 1996. Output feedback H∞ control of systems with parameter uncertainty. Inter-
national Journal of Control 63, 741–750.

Xie, W.F., 2007. Sliding-mode-observer-based adaptive control for servo actuator with
friction. IEEE Transactions on Industrial Electronics 54, 1517–1528.

Yamashita, M., Fujimori, K., Uhlik, C., Kawatani, R., Kimura, H., 1990. Control of
an automotive active suspension, in: Proceedings of 29th Conference on Decision and
Control, Honolulu, HI. pp. 2244–2250.

Yan, M., Mehr, A.S., Shi, Y., 2008. Discrete-time sliding-mode control of uncertain sys-
tems with time-varying delays via descriptor approach. Journal of Control Science and
Engineering 28.

Yan, M., Shi, Y., 2008. Robust discrete-time sliding mode control for uncertain systems
with time-varying state delay. IET Control Theory Application 2, 662–674.

Yao, B., 1997. High performance adaptive robust control of nonlinear systems: A general
framework and new schemes, in: Proceedings of 36th Conference on Decision and
Control, San Diego, CA USA. pp. 2489–2494.

Yao, B., 2009. Desired compensation adaptive robust control. ASME Journal of Dynamic
Systems, Measurement, and Control 131, 1–7.

Yao, B., Bu, F., Chiu, G.T.C., 1998. Nonlinear adaptive robust control of electro-hydraulic
servo systems with discontinuous projections, in: Proceedings of 37th Conference on
Decision and Control, Tampa, Florida USA. pp. 2265–2270.

Yao, B., Bu, F., Chiu, G.T.C., 2001. Non-linear adaptive control of electro-hydraulic
systems driven by double-rod actuators. International Journal of Control 74, 761–775.

Yi, J., Chang, S., Shen, Y., 2009. Disturbance-observer-based hysteresis compensation for
piezoelectric actuators. IEEE/ASME Transactions on Mechatronics 14, 456–464.

68
Yu, B., Shi, Y., Huang, J., 2011. Modified generalized predictive control of networked
systems with application to a hydraulic position control system. ASME Journal of
Dynamic Systems, Measurement, and Control 133, 031009:1–9.

Yu, Z., Chen, H., Woo, P.Y., 2003. Polytopic gain scheduled control for robotic manipu-
lators. Robotica 21, 495–504.

Yuan, X., Fang, Z., Li, H., 2010. Research on fuzzy control for the electro-hydraulic servo
system, in: Proceedings of Seventh International Conference on Fuzzy Systems and
Knowledge Discovery, Yantai, China. pp. 918–921.

Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8, 338–353.

Zadeh, L.A., 1968. Fuzzy algorithm. Information and Control 12, 394–102.

Zadeh, L.A., 1972. A rational for fuzzy control. ASME Journal of Dynamic Systems,
Measurement, and Control 94, 3–4.

Zeng, H., Sepehri, N., 2008. Tracking control of hydraulic actuators using a lugre friction
model compensation. ASME Journal of Dynamic Systems, Measurement, and Control
130, 014502–1–014502–7.

Zhang, R., Alleyne, A., Prasetiawan, E., 2002. Modeling and H2 /H∞ MIMO control of an
earthmoving vehicle powertrain. ASME Journal of Dynamic Systems, Measurement,
and Control 124, 625–636.

Zhao, T., Virvalo, T., 1993. Fuzzy control of a hydraulic position servo with unknown
load, in: Proceedings of 2nd IEEE International Conference on Fuzzy Systems, San
Francisco, USA. pp. 785–788.

Zheng, J.M., Zhao, S.D., Wei, S.G., 2009. Application of self-tuning fuzzy PID controller
for a SRM direct drive volume control hydraulic press. Control Engineering Practice
17, 1398–1404.

Ziaei, K., Sepehri, N., 2001. Design of a nonlinear adaptive controller for an electrohy-
draulic actuator. ASME Journal of Dynamic Systems, Measurement, and Control 123,
449–456.

69
A A
I  C T A  H-
 S

A.1 PID Control


Proportional-integral-derivative (PID) controller is the most commonly used feedback
control algorithm in industries. The controller tries to minimize the error of the con-
trolled process by adjusting three control parameters (K p , Ki and Kd ). The basic idea of
PID control is shown in Figure A.1 as follows:

Figure A.1: Schematic of a traditional PID control algorithm.

1. The proportional term K p makes part of the control signal proportional to the current
error signal; a high proportional gain results in a large change to the control signal
for a given error, aiming at a faster response of the control system. However, if the
proportional gain is too high, the control system may become unstable or can have
a large overshoot in the system response. In contrast, if the gain is decreased, there
could be a steady-state error in the final response.

2. The integral term Ki makes part of the control signal proportional to the accumu-
lation of the past error signal; the past error is collected over time and produces an
accumulated control signal that should have been sent to the plant previously. The

70
main contribution of the integral term is to eliminate or reduce the steady-state error
that could occur.

3. The derivative term Kd senses the changing rate of the error signal and is used to
reduce the magnitude of the overshoot produced by the integral and proportional
terms.

A.2 Sliding Mode Control


Sliding mode control (SMC) has become an efficient tool in the control of complex sys-
tems with uncertainties due to its low sensitivity to disturbances and parameter variations.
The basic concept of sliding mode control is depicted schematically in Figure A.2 . The
main idea behind SMC is to use a discontinuous control input to force the state trajectory
onto a certain well determined sliding surface (S = 0 in Figure A.2) and to remain on this
surface over time.

Figure A.2: Schematic of the sliding mode control algorithm.

In all control schemes, the objective is to find an appropriate control signal (U(t))
which will force the output to follow the input in a stable manner, when applied to the
plant. In sliding mode control, this is accomplished by forcing the error, e, (see Figure
A.2) to a surface called the sliding surface “S ” and to keep the error on that surface until
it is zero.

71
To elaborate this idea, consider the specific surface

S = ė + ae, (A.1)

where “a” is a pre-defined time constant. The choice of “a” is in fact a major challenge
in designing a stable but responsive surface. For S = 0 this surface implies that if the
error has some initial value, the error will converge to the origin of the error state diagram
(Figure A.2) in a first-order fashion; that is:

ė + ae = 0. (A.2)

The choice of actual form of the sliding surface S is to be designed. If the error and
its derivative (error states) do not lie on the S = 0, then the controller must be designed to
force the error states to somewhere on the S = 0 surface and then to stay on this surface
as it progresses towards the origin.
In a simple type of sliding mode controller, the control signal can be related to the
error by the following relationship:

U(t) = −k · sign(S ), (A.3)

where k is a designed constant and S the value of the sliding surface. Since the particular
sliding surface that is desired is given by S = 0, the next step is to find an optimal value
of k which will accomplish this. In this case, if a model of the plant is known, then
substituting the plant and controller equations into (A.2), k can be solved explicitly. Thus
for any initial point S , the error will be forced on the S = 0 sliding surface and stay there
(albeit in an oscillatory fashion). This oscillatory action due to the sign(S ) term, and thus
SMC is sometimes referred to as “bang-bang” control.
The design of sliding mode control consists of two main steps: (1) To develop a real-
istic but stabilized sliding surface S for the system error states; (2) to develop a control
law to force the system dynamics to follow the desired sliding surface. The dynamic char-
acteristics of the resulting closed-loop control system will be mainly determined by the
design of the sliding surface. One valuable advantage of SMC is that once the states of the
system reach the pre-defined sliding surface, the system behavior depends neither on the
system parameters nor the disturbances. This property meets with the need of designing
feedback control for dynamic systems with uncertainties explicitly.

72
A.2.1 Sliding surface design

Sliding surface design is an important procedure in SMC since the dynamic performance
of the control system will mostly depend on the sliding surface once the state trajectory
reaches the surface. Consider the control design for a linear system:

ẋ = Ax + Bu, (A.4)

where A and B are the state and input matrices, x is the state vector and u is the control
signal. Mathematically, it is a common practice to transform the system model into a
different form in order to design the sliding mode controller. The purpose of this transfor-
mation is to make one of the state equations to become independent of the control input
and system nonlinearities (no nonlinearity in this example). After transformation, this new
model form for sliding mode control would appear as:

x˙1 = A11 x1 + A11 x2 ,


(A.5)
x˙2 = A21 x1 + A22 x2 + B2 u.

The x1 state equation is independent of the control input and hence is associated with
the sliding surface design; x2 is then associated with the sliding mode control design and
is a function of the input. Consider a simple linear sliding surface as an example:

S = Cx1 + x2 . (A.6)

It can be derived that when sliding motion appears on the sliding surface S = 0 (x2 =
−Cx1 ), the system performance can be governed by the augmented state equation:

x˙1 = (A11 − A12C)x1 . (A.7)

The objective of coming up with an optimized surface S is to determine an appropri-


ate value of C to satisfy (A.7) in an optimized fashion. One approach is to appropriately
assign eigenvalues of the feedback system. Other optimization methods such as LQR op-
timization and the Lyapunov method could also be used to calculate the optimized matrix
C.

73
A.2.2 Sliding mode control design

The second stage of the design process is the implementation of the discontinuous control
signal to force the state trajectory of the system to stay close to the pre-designed surface
(S = 0). In order to guarantee sliding motion for the control system, the following dy-
namic projection on the S space has to be satisfied:

 .




 S < 0, if S > 0


 .

 S = 0, if S = 0 (A.8)




 .
 S > 0, if S < 0
Taking derivative of (A.6) and substituting (A.5) to it:

Ṡ = C x˙1 + x˙2

= (CA11 + A21 )x1 + (CA12 + A22 )x2 + B2 u (A.9)

The optimal design objective of the control system is:


.
S = 0 and S = 0. (A.10)

The first expression (S = 0) was used in the sliding surface design. The second ex-
pression Ṡ = 0 will be used in the control design. Therefore, a continuous “equivalent”
control symbolized as ueq could be obtained as follows.

.
S = C x˙1 + x˙2

= (CA11 + A21 )x1 + (CA12 + A22 )x2 + B2 u = 0, (A.11)

ueq = −B−1
2 [(CA11 + A21 )x1 + (CA12 + A22 )x2 ] .

However, this “equivalent” control could only guarantee that the state trajectory can
reach the desired sliding surface. It can not guarantee sliding motion around the sliding
surface (S = 0). In order to achieve sliding motion, a discontinuous control, being switch-
ing the sign according to the sign of S is combined with the “equivalent” control to make
a combined sliding mode control algorithm.

74
u = ueq + ∆u
(A.12)
= ueq + K · sign(S).
Then, the control signal shown in (A.12) will guarantee that the sliding motion of the
states of the signal will occur across the sliding surface. This concludes the introduction
of sliding mode control. Readers can refer to [Utkin et al., 1999] for more details on the
sliding surface and controller design.

75
A.3 H∞ Optimal Control

Modern optimal control theory with explicit state space solutions was first developed by
[Doyle et al., 1989]. One very important optimal control approach is called H∞ optimal
control. The basic idea behind control is to minimize the system H∞ -norm of the objective
signal (normally the error signals) by solving the matrix Riccati equations; the explicit
solutions of solving Riccati equations for H∞ controllers can then be implemented. Some
basic definitions of the H∞ control with brief mathematical derivations will be presented in
this section by giving the physical meaning of the control and some other optimal design
terminologies.

A.3.1 H∞ norm

H∞ optimal control is a mathematical optimization method which minimizes the “ H∞ -


norm” of the closed-loop control systems. Consider the following linear system as an
example. The state-space model of the system to be controlled is given by:

ẋ = Ax + Bu,

y = Cx + Du, (A.13)

where x denotes the state of the linear system, u is the input and y is the output. The
Y(s)
transfer function of the system T (s) = U(s)
can be represented as T (s) = C(I s − A)−1 B + D.
First, assume the system is stable (the eigenvalues of A are in the left half plane). The
purpose of the H∞ control is to find a control input u which will minimize the H∞ norm of
the system. A general definition of the H∞ norm of the transfer function (||T ||∞ ) is given
as follows:
||y||2
||T ||∞ = sup ,
||u||2
where ||y||2 represent the root-mean-square (RMS) value of the output and ||u||2 the RMS
value of the input. The term “sup” can be interpreted as determining the upper bound of
||T ||∞ in the frequency domain. To give a physical picture of what this actually means,
one could recall that the objective of the controller design is to generate an appropriate

76
input (control) signal to the system to force the output to follow the desired input. One
way of visualizing this is to examine the frequency response of a second-order transfer
function. Ideally, the ratio of the output to the desired input should be a constant over
a wide frequency range (the band width of the system). However, due to the dynamics
of the system, the output can be larger than the input as the frequency approaches the
natural frequency of the system. If this increase in the magnitude ratio can be reduced,
then the bandwidth of the system can be extended thus improving the performance of the
closed-loop system. Essentially the ||T ||∞ norm evaluates this “upper bound” and the H∞
controller is designed in an optimal fashion to reduce this upper bound.

A.3.2 Lyapunov stability

Before discussing the H∞ control design, it is necessary to briefly introduce the Lyapunov
stability theory. Consider a simple linear autonomous (no input) system:

ẋ = Ax. (A.14)

If one considers some function which reflects the energy of the system (the Lyapunov
function), and if the energy always reduces to a stable value if the system is perturbed, then
the system is considered to be stable. A Lyapunov function (positive quadratic function)
V > 0 is defined as V(x) = xT Px, where P is a symmetric positive definite matrix. Taking
the derivative of V(x) yields:

dV(x)
= xT AT Px + xT PAx
dt
= xT (AT P + PA)x. (A.15)

Lyapunov theory states that if the derivative of the Lyapunov function V is negative
the system in (A.14) is said to be stable. Mathematically, this means AT P + PA < 0, P < 0.
This inequality can be converted to a Riccatti equation. In order to guarantee the stability,
it is necessary to determine the optimal value P for a given A (which is known from the
plant equations). This is done by solving the Ricatti equation. Once P is known, then it
can be used to derive the control input. Further details can be found in [Boyd et al., 1994].

77
A.3.3 Linear matrix inequalities

In 1994, Gahinet introduced a “linear matrix inequality (LMI) approach” to solve the
H∞ optimization control problem. This approach solves the matrix Riccati equations by
solving several linear matrix inequality problems, for example, AT P + PA < 0, P < 0
[Boyd et al., 1994].
Consider again AT P + PA < 0, P < 0. This example was based on a very simple linear
system. However, most applications are far more complex and may contain nonlinearities.
Thus the Riccatti equations involving the Lyapunov function can be very complex. It is
necessary to transform the Riccati equations to a new form which is easier to be solved.
This is called the linear matrix inequality (LMI) approach. As before, P can be solved
from the LMI approach.

A.3.4 H∞ control design with LMIs

In order to fully understand H∞ control design theory, some mathematical derivation and
examples are necessary and will be given in this section. In H∞ control problems, it is
necessary that the traditional mathematical “form” of the system such as system described
in (A.13) needs to be re-formalized (called “packed”) into a standard generalized model
for H∞ optimization control design. This is a mathematical procedure and as such is not
expanded upon here.
Consider Figure A.3, the generalized state-space model of the control system is

ẋ = Ax + Bω ω + Bu u,

y = Cx + Dω ω + Du u, (A.16)

where x denotes the state of the system, u is the control signal and y is the output and ω is
the desired external input signal. The objective of the H∞ control design problem in this
example is to design an optimized state-feedback controller u = K x, so that the norm of
Y(s)
the system (T (s) = ω(s)
) is minimized. To do this a few mathematical steps are required.
Define a Lyapunov function for the system as V(x) = xT Px, P > 0. For a given γ > 0,
yT y − γ2 ωT ω is equivalent to ||y||22 /||ω||22 ≤ γ2 . In order to satisfy the Lyapunov stability

78
Figure A.3: H∞ control with generalized model.

condition and relate the control problem to the H∞ -norm of the system, the Lyapunov
function must satisfy
dV(x)
≤ γ2 ωT ω − yT y, (A.17)
dt
for all x, y and ω satisfying (A.16).
If (A.17) is satisfied, the H∞ -norm of the control system is less than γ. To prove this,
integrate (A.17) assuming zero initial conditions:
Z T
V(x(T )) + (yT y − γ2 ωT ω)dt ≤ 0. (A.18)
0

Since V(x(T )) > 0 (P is positive definete), the second part of (A.18) indicates that:
Z ∞ Z ∞
(y y)dt <
T
(γ2 ωT ω)dt. (A.19)
0 0
R∞
These integrals are in fact the squares of the 2-norms, that is ||τ||22 = 0
τT τdt. The
inequality in (A.19) can be written as:
||y||22
2
≤ γ2
||ω||2

||y||2
≤ γ (A.20)
||ω||2
||y||2
Recall that the H∞ -norm is ||T ||∞ = sup ||u|| 2
, that is, if γ can be minimized, then H∞ -
norm of the system is also minimized; the maximum amplitude ratio is minimized and the
bandwidth will be larger.

79
Because this is a state-feedback control problem, by substituting into (A.16), the
closed-loop generalized system can be written as:

ẋ = (A + Bu K)x + Bω ω,

y = (C + Du K)x + Dω ω. (A.21)

Now, Equation (A.17) can be rewritten as:


 
xT (A + Bu K)T )P + P(A + Bu K) x + [(C + Du K)x + Dω ω]T [(C + Du K)x + Dω ω]

−γ2 ωT ω ≤ 0 (A.22)

It is very difficult to obtain for P and K from this inequality as it stands. However, it
can be shown that applying the “Schur Complement” transformation, Equation (A.22) can
be transformed into the following form:
  
 xT   (A + B K)T )P + P(A + B K) + (C + D K)T (C + D K) PB   
   u u u u ω 


    x ω ≤ 0,
ωT   BTω P −γ2 I 
(A.23)
which is equivalent to:
 
 (A + B K)T )P + P(A + B K) + (C + D K)T (C + D K) PB 
 u u u u ω
 ≤ 0. (A.24)
 2 
Bω P
T
−γ I 
Equation (A.24) is not an LMI, because P and K are both unknown variables. How-
ever, it can be easily concluded that the existence of K and P > 0 satisfying (A.24) is
equivalent to the existence of K and Q > 0 satisfying
 
 Q(A + B K)T + (A + B K)Q + B BT Q(C + D K)T 
 u u ω ω u 
  ≤ 0. (A.25)
(C + Du K)Q −γ2 I
Introduce a new matrix variable Y = KQ, Equation (A.25) can be rewritten as
 
 AQ + QAT + B Y + Y T BT + B BT (CQ + D Y)T 
 u u ω ω u 
  ≤ 0. (A.26)
CQ + D Y u −γ2 I
Now, the problem is converted to an LMI problem in (A.26). For a given γ, Q and Y
can be determined and K = Y Q−1 . Thus the control input signal can be determined using
the equation u = K x. This procedure will be repeated with a smaller γ > 0 given each
time, until the LMI (A.26) is no longer feasible. Then the smallest γ > 0 that can make
(A.26) feasible will be the minimized H∞ -norm of the state-feedback control system.

80
A.4 Fuzzy Control
The fuzzy logic theory was proposed by Lotfi Zadeh during the 1960s. The application of
fuzzy logic to control industries was first initiated in the pioneering research by Mamdani
and his colleagues in [Mamdani and Assilian, 1975] and [King and Mamdani, 1977].

A.4.1 Fuzzy set

In fuzzy control theory, a new term – “fuzzy set” was introduced to describe the control
design process in a linguistic way rather than a numerical way. The main idea of fuzzy set
is that one “element” (for example, velocity of a moving mass) can be partially contained
in a pre-defined set (called membership set in fuzzy logic theory). In contrast, traditional
set theory only allows elements to be either completely contained in a set or completely
not contained in a set. Fuzzy sets therefore allow for vague expressions with respect to
membership sets, which reflects the design problems from a more human thinking per-
spective.
The fuzzy membership function for a certain fuzzy set ∆ is normally represented by
symbol µ. For a certain input α1 , if µ(α1 ) = 0, it means that is completely not a member of
set ∆; if µ(α1 ) = 1, it means that is completely a member of set ∆. However, when µ(α1 )
is a value between [0, 1], α1 is a partial member of set ∆.
Hence, a membership function of a fuzzy set can be described as:



 1 if α ∈ ∆,





f (n) = 
 (0, 1) if α is partial member of ∆,





 0 others.

An example of membership function is shown in Figure A.4. The peak of the function
where µ = 1 is called the “singleton” of the fuzzy set. The element that contains in
the region of [α2 , α3 ] has partial or full membership of the fuzzy set. In Figure A.4, if
the element α is velocity, then within the region, [α2 , α3 ] the membership function can
have values between 0 and 1, and is 0 outside [α2 , α3 ]. For simplification, most fuzzy
membership functions are approximated by linearized functions.

81
Figure A.4: Membership function.

Consider the classification of the speed of a car as an example. If it is defined that a


speed higher than 80km/h is fast, then a conventional numerical set would strictly con-
sider that a speed of 79km/h as “not fast”. This would not make sense according to normal
human sense. However, fuzzy set theory would consider that the speed of the car is “some-
what” or “partially” fast. As shown in Figure A.5, the speed of the car is divided into three
distinct fuzzy sets as: slow, medium and fast. Considering a speed of 70km/h, it can be ob-
served that from a fuzzy set perspective, 70km/h is a partial membership of both medium
and fast fuzzy set but with membership in different degrees (0.4 and 0.65).
The basic idea of fuzzy logic control is presented as follows and drawn in Figure A.6:

1. Fuzzification: based on the expert knowledge of the controlled system, the reference
input and the controlled output signal are translated into “linguistic” values – fuzzy
sets.

2. Decision making: simulate the human decision making based on the pre-designed
fuzzy sets and output fuzzy control actions.

3. Defuzzification: transform the fuzzy control actions into nonfuzzy control actions.

Each of these steps will be explained respectively using a car cruise control as an example.

82
Figure A.5: Fuzzy sets for car speed.

A.4.2 Fuzzification

Fuzzification is a process that transforms the inputs of the fuzzy controller e(t) into linguis-
tic values using the membership functions; physically this means categorizing the inputs
into each fuzzy set and determining how much they belong to these sets. Consider a car
cruise control as an example. The “designer” creates five triangular fuzzy sets for the
speed “error” of the car: zero, positive small, positive large, negative small and negative
large. As shown in Figure A.7, the speed error of 7km/h has a membership of 0.3 in the
zero set and a membership of 0.65 in the positive small set.

A.4.3 System knowledge and fuzzy rule base

As discussed before, fuzzy control design doesn’t have an explicit or systematic solution
to calculate the control signal. Expert knowledge for the controlled system is demanded.
This expert knowledge is then formalized into control languages of what kind of control
output should result from a certain feedback of the system. It often takes form of:
IF <combination of conditions>, THEN<consequence>
In the car cruise control example, the rules can be formulated as:

IF speed error is Positive Large, THEN brake

83
Figure A.6: Schematic of the fuzzy logic control algorithm.

IF speed error is Zero OR Positive Small, THEN turn the engine to Low power

IF speed error is Zero OR Negative Small, THEN turn the engine to Medium power

IF speed error is Negative Large, THEN turn the engine to High power

It should be noted that the AND/OR expressions in the fuzzy control theory will be
different from the traditional logical expressions. For example, AND will be modified
to take the minimum of two values while OR will be modified to take the maximum of
two values (this modification will be used in this example). Such definitions of the logic
operators could appear to be different in other literatures.

A.4.4 Decision making

Decision making is the crucial part of fuzzy control theory. It could be divided into three
steps: “aggregation”, “composition” and “accumulation”.

1. Aggregation: in the aggregation process, each of the IF statements in the fuzzy rule
base is calculated using augmented inputs (output of the fuzzification process).

2. Composition: in this step, each of the THEN statement in the fuzzy rule base is
calculated using the result that was calculated in the “aggregation” step. It should

84
Figure A.7: Fuzzy sets for car cruise control.

be noted that the output of the “composition” step is another “formalized” fuzzy set
rather than a numerical value. The formalization of the output fuzzy set is governed
by the results of the aggregation step.

3. Accumulation: in this final decision making step, the “formalized” fuzzy sets achieved
from the “composition” step will be accumulated into a single fuzzy set. This ac-
cumulated fuzzy set will be sent to the defuzzification step to translate the fuzzified
control signal into numerical control signals.

Again, consider the car cruise control example used earlier. A speed error of 7km/h
has a membership of 0.3 in the zero set, 0.65 in the Positive Small set and 0 in other
sets. There are OR statements in both Low Power and Medium Power fuzzy rule base
statements.

IF speed error is Zero OR Positive Small, THEN turn the engine to Low power

IF speed error is Zero OR Negative Small, THEN turn the engine to Medium power

IF speed error is Zero OR Positive Small, THEN turn the engine to Low power

85
IF speed error is Zero OR Negative Small, THEN turn the engine to Medium power

Then, through the aggregation step, the value of the IF statement for the Low Power
rule is equal to the maximum of 0.3 and 0.65 (Positive Small set). So, the result of state-
ment for the Low Power rule set is 0.65. Following the same idea, the result of statement
for the Medium Power rule set will be the maximum of 0.3 (Zero set) and 0 (Negative
Small set), which equals to 0.3. This completes the aggregation step.
The output fuzzy sets are truncated using the results calculated from the aggregation
stem respectively. The truncated fuzzy sets are combined so as to get a single fuzzy set as
shown by the shaded area in Figure A.8.

Figure A.8: Aggregation, composition and accumulation processes of de-


cision making.

86
A.4.5 Defuzzification

The decision making process has generated fuzzified control command; however, this
fuzzified control command cannot be directly used as the control signal of the system. A
defuzzification procedure has to be completed to transfer the fuzzified control command
into numerical control signals.
There are many defuzzification methods developed in fuzzy control design literatures.
One of them called the Center of Gravity (COG) method will be introduced and used in
the car cruise control example here.

Figure A.9: Defuzzification using center of gravity.

Using the COG defuzzification method, the gravity center of the area described by the
combined output fuzzy sets can be calculated and defined as (xc , yc ). The value of xc is the
exact control action that needs to be sent to the engine to control the speed of the car. The
gravity centers of both sub-areas, A1 and A2 , are at xc1 = 30% and xc1 = 57% which are
shown in Figure A.9. Then, the gravity center of the combined area can be calculated by
the following equation:

A1 xc1 + A2 xc2
xc = = 35% (A.27)
A1 + A2

87
Then, the control action for the car cruise control when the speed error equals to 7km/h
would be 35% of the maximum engine power. It should be emphasized that there are many
different fuzzification and defuzzification methods reported in the literature. Please refer
to [Zadeh, 1965] and [Zadeh, 1968] for more details.

88
R
Boyd, S., Ghaoui, L.E., Feron, E., Balakrishnan, V., 1994. Linear Matrix Inequalities In
System and Control Theory. SIAM.

Doyle, J.C., Glover, K., Khargonekar, P.P., Francis, B.A., 1989. State-space solutions to
standard H2 and H∞ problems. IEEE Transactions on Control Systems Technology 34,
831–847.

King, P.J., Mamdani, E.H., 1977. The application of fuzzy control systems to industrial
processes. Automatica 13, 235–242.

Mamdani, E.H., Assilian, S., 1975. An experiment in linguistic synthesis with a fuzzy
logic controller. International Journal of Man-Machine Studies 7, 1–13.

Utkin, V., Guldner, J., Shi, J., 1999. Sliding Mode Control in Electromechanical Systems.
CRC Press.

Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8, 338–353.

Zadeh, L.A., 1968. Fuzzy algorithm. Information and Control 12, 394–102.

89
A B
D-T H2-O O T C-
   E H P C-
 S (HPCS)

Yang Lin, Yang Shi and Richard Burton

Abstract

This paper explores the application of discrete-time H2 -optimal output tracking con-
trol for a hydraulic positioning control system (HPCS). By minimizing the H2 -norm of
the system, the discrete-time H2 -optimal control both stabilizes the plant and minimizes
the root-mean-square of the servo position error simultaneously. To facilitate computa-
tion of the H2 -optimal controller, linear matrix inequalities (LMIs) technique is applied.
Computer simulations illustrate the design procedure and the effectiveness of the proposed
method. Experimental tests on a real hydraulic positioning system for a tracking appli-
cation are also conducted and the results show that the method is suitable for practical
applications.

B.1 Introduction
Output tracking control, also called model reference control, has wide applications in in-
dustry, biology and medicine [Gao and Chen, 2008]. Its main objective is to force the con-
trolled output to track the output of a given reference model as close as possible. Model
reference control has been widely used in robot control [M.Dawson et al., 1992], flight
control [Liao et al., 2002] and motor control [Hu et al., 1995] etc. Hydraulic systems
have been used in industry in a large number of fields by virtue of their small size to
power ratios and the ability to apply very large force and torque [FitzSimons and Palaz-
zolo, 1996]. Hydraulic systems are extensively used in machine tools, material devices,

90
transportation and other mobile equipments. Hydraulic and pneumatic power and con-
trol are needed more than ever to solve force, torque, and speed problems in the fields of
aerospace, transportation, agriculture, mining, construction, manufacturing, marine, pro-
cessing, oil exploration, and power generation. Hydraulic cylinders and motors have the
unique capability of exerting and precisely controlling large forces and torques in cramped
quarters. However, hydraulic systems also have a number of characteristics which com-
plicate the development of high-performance control. The dynamics of hydraulic systems
are highly nonlinear and have model uncertainties [Merritt, 1967]. Uncertainties, such
as external disturbances, leakage, and friction, cannot be modelled exactly in general.
Therefore, how to develop the accurate hydraulic position control system is a challenge
for applied hydraulic systems. Design of the hydraulic positioning control system (HPCS)
has received great attention in the past decade. In existing literature, most of the works
on the HPCS design are developed in the continuous-time domain. In [FitzSimons and
Palazzolo, 1996], a continuous-time position-velocity-acceleration (PVA) controller was
designed for a hydraulic servo system. Linear robust adaptive control was applied to a hy-
draulic servo system in [Plummer and Vaughan, 1996]. In [Vossoughi and Donath, 1995],
dynamic feedback linearization control was applied for an electro-hydraulic actuated sys-
tem. In [Alleyne and Hedrick, 1995], nonlinear adaptive control was applied to the force
control of an active suspension driven by a double-rod cylinder. All the above mentioned
methods are developed in the continuous-time domain. However, with advancements of
the industrial computer technology, control algorithms for many practical applications are
implemented in industrial control computers in the discrete-time domain. In practice, the
continuous-time plant is controlled by a discrete-time controller with sample and hold
devices, which leads to a sampled-data system [Chen and Francis, 1995]. Sampled-data
control system design in the area of hydraulic systems has not yet received much attention,
which is the focus of this paper. On the other hand, many studies on hydraulic positioning
systems have been focused on the step tracking problem. However, in some complicated
applications, the output of a hydraulic system is expected to track a smooth curve that
can be generated by a reference model. To the best of the authors’ knowledge, very few
studies focus on the output tracking control for hydraulic positioning systems. This has

91
been the motivation for the application of the discrete-time H2 -optimal output tracking
control to an experimental hydraulic positioning control system which is reported in this
paper. Discrete-time H2 -optimal control is optimal with respect to a specified quadratic
performance index or cost function, and it can be synthesized based on the solutions of
two Riccati equations [Chen and Francis, 1991]. In [Doyle et al., 1989], the solution to a
continuous-time H2 -optimal control problem was derived using the state-space approach.
This gives a “clean” treatment of the problem and provides compact formulas for the op-
timal control. In [Chen and Francis, 1991] and [Chen and Francis, 1995], a state-space
solution to the discrete-time H2 -optimal control problem was derived by solving two alge-
braic Riccati equations. In [Lin et al., 2007], discrete-time H2 -optimal control was applied
for a HPCS to achieve the objective of step tracking. In this paper, a new methodology
based on the solution of linear matrix inequalities (LMIs) is applied to design a discrete-
time controller for a HPCS to achieve the output tracking objective. The paper is organized
as follows. Section B.2 introduces the experimental facility of the system. In Section B.3,
the open-loop model of the hydraulic system is identified based on the experimental data;
experimental verification is conducted to test the accuracy of the model. In Section B.4, a
discrete-time H2 -optimal controller is designed based on the LMI optimization to achieve
the output tracking objective. In Section B.5, experimental results of applying the designed
discrete-time H2 -optimal controller are given. Finally, Section B.6 offers some concluding
remarks and points out possible extensions.

B.2 Experimental Test Facility

The hydraulic positioning system is shown in Figure B.1. The actuator (SHEFFER) is
controlled by a flow servo valve (MOOG 15-010) which is connected with a Pentium IV
PC computer. A linear variable differential transducer (DCLVDT LUCAS SHAEVITZ
5000 DC-E) is used to measure the position information of the load with an accuracy of
0.078 volts/mm. The analog measurement signal (cylinder position) is fed back to the
computer through a 12 bit A/D and D/A board (National Instruments PCI-6025E). The
supplied pressure is 4MPa (600 psi). The load on the actuator is a 2.63 kg mass. Slip

92
stick friction characteristics existed in the actuator, and thus a linear model is only an
approximation of the nonlinear real system. The hypothesis was that if the controller
could work well in a system that has known nonlinear behavior using a linear model, then
the “computational expense” of having to incorporate nonlinear characteristics into the
proposed controller design may not be necessary for some applications.

Figure B.1: Hydraulic positioning control system.

B.3 Modeling and Identification of HPCS

In this section, the frequency response plot is used to identify the model of the open-loop
system, and to produce the standard form of the sampled-data state-space representation
of the HPCS.

B.3.1 Transfer function of the open-loop system

With a swept sinusoidal input signal, the frequency response or bode plot of the open-loop
system was obtained using a spectrum analyzer, as shown in Figure B.2. The final value of

93
the phase was determined to be −3π/2, and the higher frequency slope of the magnitude
plot −60db per decade implying that the system was a third-order one. The phase bode
plot starts from −π/2, so there should be a single s appearing in the denominator of the
open-loop transfer function, and hence, the open-loop transfer function is a type-1 system.
Finally, the open-loop system transfer function can be approximated by
Kω2n
G(s) =
s(s2 + 2ζωn s + ω2n )
where G(s) = y(s)/V(s), y and V are in volts.

Figure B.2: Experimental open-loop frequency responses (magnitude and


phase).

Using classical frequency domain identification techniques [Nise, 2000], the natural
frequency is estimated to be ωn = 110rad/s, the damping ratio, and the gain ζ = 37.2. The
open-loop transfer function of the plant (HPCS) is therefore:

422000
G(s) = . (B.1)
s3 + 141s2 + 12100s

94
To verify the identified model, the model response and the experimental response to the
same input signal are compared. Figure B.2 and Figure B.3 are the experimental and model
response to a swept sinusoidal input respectively. It is observed that the identified model
can represent the dynamics of the experimental open-loop hydraulic system extremely
well for frequencies less than 200Hz.

Figure B.3: Simulated open-loop frequency responses (magnitude and


phase).

B.3.2 Standard sampled-data state-space model

The control objectives of the HPCS design is to force the output to track a reference signal
that is generated by a reference model, as shown in Figure B.4. In this figure, S is an
ideal sampler, Kd the digital controller, H a zero-order-hold, and P the plant (open-loop
hydraulic system).

95
Figure B.4: Sampled-data model reference control system.

According to the identified transfer function model of the hydraulic system, the state-
space model can be represented by

x˙p = A p x p + B p u,

y = C p x p + D p u.

Here, x p , y and u are the states, output and input of the plant respectively , and A p , B p , C p , D p
are the state space matrices of the plant with appropriate dimensions.
In the following analysis, with respect to Figure B.4, the desired output is yr and the
plant output is y. The generalized plant G has inputs ω, u and outputs e, e. The system
block diagram is re-organized to be a standard sampled-data control system form as shown
in Figure B.5.
Further, the above diagram can be simplified to a standard sampled-data control system
shown in Figure B.6.
The objective is to design a controller such that the output y of the closed-loop control
system can track the reference signal to meet the required tracking performance. Suppose
that the reference signal yr is generated by a general reference model:

ẋr = Ar xr + Br ω,

y = Cr xr ,

where xr , yr and ω are the states, output and input of the plant respectively and Ar , Br , Cr
are the state space matrices of the reference model.

96
Figure B.5: Sampled-data control system diagram for the general plant.

The state-space equations are now re-organized to reflect the generalized plant G:

         
 ẋ p   A x + B u   A 0   x   0 B   ω 
   p p p 
 =  p   p   p  
  =       +   
  ,
ẋr Ar xr + Br ω 0 Ar xr Br 0   u
     
 e1   y − y   C x − C x 
   r   r r p p 
  =   =  

e2 yr − y Cr xr − C p x p 
     
 −C C   x   0 0   ω 
 p r  p 
 +    
=   
 −C C   x   0 0   u 
    
 

p r r

In the standard sampled-data control system, e1 is the signal to be minimized, and e2


is the signal to be input into the controller. In many applications, e1 and e2 are different.
In this specific output tracking control problem, the tracking error and the input signal to
the sampler/controller are the same. The generalized system G now can be written as a
standard linear time-invariant model as:

97
Figure B.6: Standard sampled-data control system.

x˙g (t) = Ag xg (t) + Bg1 ω(t) + Bg2 u(t),


eg1 (t) = Cg1 xg (t) + Dg11 ω(t) + Dg12 u(t),
eg2 (t) = Cg2 xg (t) + Dg21 ω(t) + Dg22 u(t),

where xg (t) is the state vector of the generalized plant of the system, and Ag , Bg , Cg , Dg are
the state space matrices of the generalized plant,
     
 x   A 0     0 B 
 p   p   p 
xg (t) =   , Ag =   , Bg1 Bg2 =   ,
 x    0 A    B 0 
r r r

       
 C   −C C   D   0 0 
 g1

 
 = 
p r
 ,  g11 Dg12  
 = 

 .
 
Cg2 −C p Cr   Dg21 Dg22 0 0

In both simulation and experimental studies, the reference model is chosen as:




 ẋr = −xr + ω



 yr = 0.5xr

98
Substituting parameters obtained from the experimental model (B.1) yields:
     
 −141 −12100 0 0   0   1 
     
     
 1 0 0 0   0   0 
Ag =   , Bg1 =   , Bg2 =   ,
 0     
 1 0 0   0   0 
     
0 0 0 −1 1 0
Cg1 = [00 − 4220000.5] ,
Cg2 = [00 − 4220000.5] .

According to the specifications of the hardware in the experimental HPCS, the sam-
pling time is chosen as T s = 0.03s. The continuous-time generalized plant can then be
discretized using the step-invariant transformation [Chen and Francis, 1995], and the re-
sulting discrete-time system matrices are

ZT s
T s Ag
Agd = e , [Bg1d Bg2d ] = [B1 B2 ] eτAg dτ.
0

C1 , C2 and D do not change in the discretization process [Chen and Francis, 1995]. The
discrete-time setup of the standard sampled-data control system is shown in Figure B.7.
Gd and Kd are the discretized forms of G and K, respectively.

Figure B.7: Discrete-time control system.

The discrete-time linear time-invariant model of the generalized plant is now given by:

99
xg (k + 1) = Agd x(k) + Bg1d ω(k) + Bg2d u(k)
eg1 (k) = Cg1 x(k) + Dg11 ω(k) + Dg12 u(k) (B.2)
eg2 (k) = Cg2 x(k) + Dg21 ω(k) + Dg22 u(k)
 
 x (k) 
 p 
where xg (k) =  is the discrete-time state vector and ω(k), u(k) the inputs of the
 x (k) 
r
generalized plant.
 
 1.4008 281.31 0 0 
 
 
 0.0232 4.6790 0 0 
Agd =  

 0.0003 0.0661 1 0 
 

0 0 0 0.9701

 
 0 2.4907 
 
 
h i  0 0.0039 
Bg1d Bg2d = 10−3 ×  

 0 0 
 

2.9955 0

The objective is now to design an H2 -optimal controller Kd which will force the output
position y to follow the reference signal yr .

B.4 Discrete-Time H2-Optimal Output Tracking Controller


Design and Simulation
In [Oliveira et al., 2002], an effective H2 output-feedback stabilization control design
scheme was proposed by solving a set of linear matrix inequalities (LMIs). In this section,
the method in [Oliveira et al., 2002] is extended and applied to solve the output tracking
control for the HPCS.

B.4.1 Discrete-time H2 -optimal output tracking control

The controller to be designed has the following state-space form:

100
xk (k + 1) = Ak xk (k) + Bk eg2 (k),
(B.3)
u(k) = Ck xk (k) + Dk eg2 (k).
Here, xk (k) is the state vector of the controller, and Ak , Bk , Ck , Dk are the state space
matrices of the controller.
By combining (B.2) and (B.3), the closed-loop system is

xcl (k + 1) = Acl xcl (k) + Bcl ω(k),


(B.4)
eg1 (k) = Ccl xcl (k) + Dcl ω(k),

where xcl (k) is the state vector of the closed-loop control system, Acl , Bcl , Ccl , Dcl are the
state space matrices of the closed-loop control system,
 
 x (k) 
 g 
xcl (k) =  ,
 x (k) 
k

   
 A + B D C   B + B D 
 gd g2d k g2 Bg2d C k   g1d g2d g21 
Acl :=  , B
 cl :=   ,
 BkCg2 Ak BD
k g21
  h i
Ccl := Cg1 + Dg12 DkCg2 Dg12Ck , Dcl := Dg11 + Dg12 Dk Dg21 .

Let the symbol Zωe1 (ς) denote the transfer function from the input ω to the output e1 . ς
is the discrete-time transfer function operator. Then the objective is to design a controller
to achieve the closed-loop stability and minimizing ||Zωe1 (ς)||2 .
It is now possible to apply the output feedback H2 -optimal control design method
[Oliveira et al., 2002] together with model reference control to the hydraulic position con-
trol system to achieve output tracking control. It is worth mentioning that the model
reference control is a new feature that has been incorporated into the work in [Oliveira
et al., 2002].
Define the packed controller matrices into a block matrix as:
 
 A B 
 k k 
K :=   .
 C D 
k k

Theorem 1 [Oliveira et al., 2002]: There exists a controller in form of equation (B.4)
such that the inequality ||Zωe1 (ς)||22 < µ holds if, and only if, the LMIs

101
trace(W) < µ (B.5)
 
 W Cg1 X + Dg12 L Cg1 + Dg12 RCg2 
 
 
 ∗ X + XT − P I + S T − J  > 0 (B.6)
 
 
∗ ∗ Y + YT − H
 
 P J Agd X + Bg2d L Agd + Bg2d RCg2 Bg1d + Bg2d RDg21 
 

 
 ∗ H Q Y Agd + FCg2 Y Bg1d + FDg21

  > 0
 ∗ ∗ X + X T − P I + ST − J 0  (B.7)
 
 
 ∗ ∗ ∗ Y + XT − H 0 
 
 ∗ ∗ ∗ ∗ I
Dg11 + Dg12 RDg21 = 0 (B.8)

hold, where the matrices X, L, Y, F, Q, R, S , J are symmetric matrices and P, H, W are


the variable matrices. In symmetric block matrices, we use ∗ to represent a term that is
induced by symmetry. This convex optimization problem can be numerically solved very
efficiently using MATLAB LMI Toolbox.
Then the controller can be obtained by:
 
 A B 
 k k 
K =  
 C D 
k k
   
 V −1 −V −1 Y B   Q − Y A X F   U −1
0 
 g2d
  gd
  
=      
 0 I L R −Cg2 XU −1
1 

UV = S −Y X, U and V can be conveniently obtained by the singular value decomposition.

B.4.2 Simulation results

To verify the effectiveness of the discrete-time H2 -optimal output tracking controller, the
input signal: 



 1, 1 ≤ t ≤ 4
ω(t) = 
 (B.9)

 0, otherwise
is chosen to evaluate the output tracking performance. The simulation results of the output
response and control action are shown in Figure B.8 and Figure B.9, respectively. It can

102
be seen that the output signal can track the desired one within 4% for the input chosen.

Figure B.8: Simulation results of the position output signal.

Figure B.9: Control signal.

B.5 Experimental Results

The designed controller was also applied in the experimental HPCS. The experimental
results are shown below in Figure B.10.

103
Figure B.10: Experimental results of the position output signal.

From the experimental results it can be observed that the output of the experimental
HPCS can track the reference output signal with tracking error less than 10% of the full
stroke by applying the discrete-time H2 -optimal output tracking control. This is larger
than the maximum tracking error observed in the simulated results (4%). Upon reflection,
the differences were attributed to the fact that (1) the nonlinear behaviour of the physical
system was not incorporated into the model, and (2) uncertainties were not considered in
the controller design.

B.6 Conclusions

This paper focuses on the output tracking control design for the HPCS. It is felt that this
is the original contribution of this research since this has not been fully investigated in
existing literature. First, the model of an experimental HPCS was set up, and then an
H2 -optimal model reference controller was designed. Finally, experimental tests on a
HPCS were conducted. Both simulation and experimental studies so illustrate that the H2 -
optimal model reference control scheme could be used for tracking of a continuous input
for a hydraulic positioning systems with tracking error less than 10% of the full stroke.
For some applications such as off highway applications, this is certainly acceptable but
in more precision type situations such as precision machining, this is not acceptable. It

104
should be recalled that the original hypothesis was “if the controller could work well in a
system that has known nonlinear behavior using a linear model, then the “computational
expense” of having to incorporate nonlinear theory into the proposed controller design
could be avoided”. It is clear that for high-precision tracking, this hypothesis was not
valid for this controller design. To further improve the output tracking performance, it is
necessary to incorporate the inherent nonlinear characteristics of hydraulic systems and
uncertainties/disturbances into the controller design. This approach is currently being
pursued by the authors.

Nomenclature

Ap, Bp, C p, Dp plant model matrices


Ag , Bg , Cg , Dg generalized plant model matrices
Ak , Bk , Ck , Dk controller model matrices
Agd , Bgd , Cgd , Dgd discretized generalized plant model matrices
Acl , Bcl , Ccl , Dcl closed-loop control system model matrices
Acl , Bcl , Ccl , Dcl closed-loop control system model matrices
||e||2 2-norm of the signal
y plant output
yr output of the reference model
u control input to plant
Kd discrete-time controller
S idealized sampler
H zero-order hold
e error signal
G packed generalized system
Gd discretized packed generalized system
xp states of the plant
xr states of the reference model
xg states of the generalized plant

105
xk states of the controller
xd states of the closed-loop control system
ω system input signal
V input volt-signal
ωn natural frequency of the plant
ξ damping ratio
X, L, Y, F, Q, R, S , J, variable matrices of related LMIs
P, H, W symmetric variable matrices of related LMIs

106
R
Alleyne, A., Hedrick, J.K., 1995. Nonlinear adaptive control of active suspension. IEEE
Transactions on Control Systems Technology 3, 94–101.
Chen, T., Francis, B.A., 1991. H2 -optimal sampled-data control. IEEE Transactions on
Automatic Control 36, 387–397.
Chen, T., Francis, B.A., 1995. Optimal Sampled-Data Control Systems. Springer-Verlag,
London.
Doyle, J.C., Glover, K., Khargonekar, P.P., Francis, B.A., 1989. State-space solutions to
standard H2 and H∞ problems. IEEE Transactions on Control Systems Technology 34,
831–847.
FitzSimons, P.M., Palazzolo, J.J., 1996. part i: Modeling of a one-degree-of-freedom
active hydraulic mount; part ii: Control. ASME Journal of Dynamic Systems, Mea-
surement, and Control 118, 439–448.
Gao, H., Chen, T., 2008. Network-based H∞ output tracking control. IEEE Transactions
on Automatic Control 53, 655–667.
Hu, J., Dawson, D., Qian, Y., 1995. Position tracking control of an induction-motor via
partial state-feedback. Automatica 31, 989–1000.
Liao, F., Wang, J., Yang, G., 2002. Reliable robust flight tracking control: An LMI ap-
proach. IEEE Transactions on Control Systems Technology 10, 76–89.
Lin, Y., Shi, Y., Burton, R., 2007. Discrete-time H2 -optimal control for hydraulic posi-
tion control systems, in: Proceedings of ASME International Mechanical Engineering
Congress and Exposition, Seatle, US.
M.Dawson, D., Qu, Z., Carroll, J.J., 1992. Tracking control of rigid-link electrically-
driven robot manipulator. International Journal of Control 56, 991–1006.
Merritt, H.E., 1967. Hydraulic Control Systems. John Wiley & Sons.
Nise, N.M., 2000. Control system engineering. John Wiley & Sons.
Oliveira, M.C., Geromel, J.C., Bernussou, J., 2002. Extended H2 and H∞ norm char-
acterizations and controller parameterizations for discrete-time systems. International
Journal of Control 75, 666–679.
Plummer, A.R., Vaughan, N.D., 1996. Robust adaptive control for hydraulic servo sys-
tems. ASME Journal of Dynamic Systems, Measurement, and Control 118, 237–244.
Vossoughi, R., Donath, M., 1995. Dynamic feedback linearization for electro-
hydraulically actuated control systems. ASME Journal of Dynamic Systems, Measure-
ment, and Control 117, 468–477.

107
A C
M  H∞ PID P F C
D   E A (EHA)
S

Yang Lin, Yang Shi and Richard Burton

Abstract

This work studies the modeling and design of a PID plus feedforward controller for
a high precision electro-hydraulic actuator (EHA) system. The high precision position-
ing EHA system is capable of achieving a very high accuracy positioning performance.
Many sophisticated control schemes have been developed to address these problems. How-
ever, in industrial applications, PID control is still the most popular control strategy used.
Therefore, the main objective of this work is to design a PID controller for the EHA sys-
tem, improving its performance while maintaining and enjoying the simple structure of
the PID controller. An extra feedforward term is introduced into the PID controller to
compensate for the tracking error especially exist during the transient period. The PID
plus feedforward control design is augmented into a static output feedback (SOF) control
design problem and the SOF controller is designed by solving an H∞ optimization problem
with bilinear matrix inequalities (BMIs).

C.1 Introduction
Hydraulic positioning systems play an important role in transportation, earth moving
equipment, aircraft and industry machinery with heavy duty applications [Gomis-Bellmunt
et al., 2008; Rovira-Mas et al., 2007]. Traditional hydraulic transmission control systems
are mainly valve controlled [Yanada and Furuta, 2007]. Pump controlled hydraulic sys-
tems, known as hydrostatic systems, are used as an alternative for valve controlled systems

108
in applications in which higher energy efficiencies are preferred. One configuration for hy-
drostatic systems uses a fixed displacement pump which regulates the flow in the hydraulic
circuit by changing the rotational speed and direction of the pump [Habibi and Golden-
berg, 2000]. In this study, a particular electrohydraulic actuator (EHA) system, shown in
Figure C.1, is considered in which high precision position control has been achieved.

Figure C.1: Schematic of the whole hydraulic circuit of the system.

Because the EHA is a closed hydraulic system, the modeling for the EHA is more
complex than the case of traditional open pump/motor hydrostatic systems. This paper
introduces such a model which was developed using the power bond graph (PBG). PBG
technique has been widely used in modeling and control of mechanical systems because
of its ability to handle causality issues for complex modeling problems. In [Toufighi et al.,
2007], similar approach was applied to analyze the dynamic characteristics of different
mechatronic systems.
For this particular EHA system, various control methods such as fuzzy control [Samp-
son, 2005], sliding mode control [Wang et al., 2008] and robust sliding mode control [Lin
et al., 2009] were proposed and acceptable tracking performance were achieved. But the
controller setups for these advanced algorithms were complicated and difficult to imple-
ment physically. Once problem occurs, only the expert in certain areas can fix it. Thus,
the first motivation of this paper was to design an easy understood and easy implemented
controller for the EHA system.
The proportional-integral-derivative (PID) control has been the most adopted control
method for industrial applications. It has been estimated that over 90% of the controllers
in use today are PID controllers, even though other advanced control theories and prac-

109
tical design methods exist [Aström and Hägglund, 2001]. One of the most famous PID
tuning methods proposed by Ziegler and Nichols (ZN method) has been extensively used
by control engineers over the past several decades [Hang et al., 1991]. However, when
extremely high control performance is required, the traditional PID controllers may not be
adequate to deal with disturbance, noises and model uncertainties. Thus, advanced PID
tuning or design method is required for high performance control purpose. For example,
in [Dieulot and Colas, 2009], a robust PID controller was designed for a mechanical axis
by adding extra flexible pole-placement constraints on the model. In [Thanh and Ahn,
2006], a nonlinear PID controller was designed for a pneumatic artificial muscle (PAM)
actuator system by combining neural network PID tunning. Thus, a second motivation for
this research was to design a new type of PID controller which can also deal with these
problems.

In many studies, PID control design was transformed into a static output feedback
(SOF) problem [Bianchi et al., 2008; Ge et al., 2002; Zheng et al., 2002]. It should be noted
that SOF problem is one of the most important open research topic in control theory and
applications [Bara and Boutayeb, 2005]. The linear matrix inequality (LMI) technique was
applied to solve the SOF problems. Because of the nonconvex property of the SOF control,
there have been many research studies during the past decade dedicated to trying to solve
this problem. However, to date, there is still no general solution solving this problem for
all type of systems. In this paper, LMI conditions with nonlinear constraints are developed
to find an H∞ performance SOF controller for the EHA system. An additional problem
occurs with the controllers designed in [Wang et al., 2008; Lin et al., 2009], all the states
of the system are required to design controllers for the EHA system. State estimators were
designed to estimate all the states which makes the design process even more complicated.
SOF controller design does not require all the states of the system and hence, makes it
much easier to be realized in practice.

In the proposed control design, not only the PID feedback control design is augmented
into a SOF problem, a new feedforward term is also introduced into the system to compen-
sate for the tracking error that appeared in [Wang et al., 2008; Lin et al., 2009]. In [Leva
and Bascetta, 2006; Visioli, 2004; Ge and Jouaneh, 1996; Wang et al., 2009], different

110
types of PID plus feedforward controllers have been presented. In this work, the control
system containing both feedback PID controller and the feedforward control is augmented
into a new type of SOF control design problem by introducing new state and output vec-
tors. The SOF controller design is transformed into bilinear matrix inequality (BMI) opti-
mizations. As such, then, the main objectives of this work are three-fold:

1. To build a model for the EHA system physically using the power bond graph (PBG)
technique.

2. To introduce a feedforward term into the traditional PID feedback controller setup
and transform the problem into a SOF problem by defining new state and output
vectors.

3. To solve the augmented SOF design problem with H∞ performance by solving a set
of linear matrix inequalities (LMIs).

The chapter is organized as follows. The mathematic model of the EHA system is
developed using the power bond graph technique in Section C.2. In Section C.3, the
feedback PID plus feedforward controller design is transformed into a SOF problem, LMI
conditions solving the H∞ optimization problem for the SOF is developed. Simulation
studies and experimental results for the EHA system position tracking are illustrated in
Section C.4. Finally, conclusion remarks are offered in Section C.5.
The notation used throughout this paper is fairly standard. The superscript T stands
for the matrix transposition, Rn denotes the n-dimensional Euclidean space, and Rm×n is
the set of all m × n real matrices. The notation P > 0 means that P is real symmetric and
positive definite, and I and 0 represent identity matrix and zero matrix respectively with
appropriate dimensions. In symmetric block matrices or long matrix expressions, the as-
terisk (∗) represents a term that is induced by symmetry. Matrices, if their dimensions are
not explicitly stated, are assumed to be compatible for algebraic operations. The physical
meanings of some important variables involved are elaborated in the nomenclature at the
end of the chapter.

111
C.2 Modeling of the EHA System

The layout of the EHA system is shown in Figure C.1. The actuator is driven by a bi-
directional fixed displacement gear pump. A special symmetrical actuator is connected
with the load and the motion of the load is controlled by varying the speed of the electric
motor. In this section, the hydraulic circuit is studied from a bond graph point of view.
The bond graph of each part of the hydraulic system is presented. To accommodate the
linearized model, approximations to several nonlinear equations are presented.

C.2.1 Linear symmetrical actuator

The bond graph of the hydraulic actuator including the load is given in Figure C.2.

Figure C.2: Bond graph of the symmetrical actuator.

The flow in and out of the linear symmetrical actuator are described by the following
equations:

112
V0 + Ax dP1
Q1 = A ẋ + + Le P1 + Li (P1 − P2 ), (C.1)
β dt
V0 − Ax dP2
Q2 = A ẋ − − Le P2 − Li (P2 − P1 ). (C.2)
β dt

Here, V0 is the mean pipe plus actuator chamber volume, Q1 represents the flow goes into
the actuator, Q2 is the flow goes out from the actuator, P1 and P2 are the pressures in the
actuator chambers, Le is the actuator external leakage coefficient and Li is the actuator
internal leakage coefficient.

C.2.2 Hydraulic pump

For the hydraulic pump, the effect of the case drain leakage was ignored, so only the effect
of the cross-port leakage was taken into account. The PBG for the hydraulic pump is given
in Figure C.3.

Figure C.3: Bond graph of the hydraulic pump.

113
The pump flow can be modeled as:

VU dPU
QU = D p ω p − KLCP (PU − PD ) − , (C.3)
β dt
VD dPD
QD = D p ω p − KLCP (PU − PD ) + , (C.4)
β dt

where, D p is the fixed pump displacement, ω p is the pump angular velocity, KLCP is the
pump cross-port leakage coefficient, QU and QD are the flow coming out and going into
the bidirectional pump, PU and PD are the pressures at the outlet and inlet port of the
pump.

C.2.3 Pump/actuator connection & overall hydraulic model

In this EHA system, the hydraulic actuator is connected to a horizontal movement sliding
mass, so the motion equation of the hydraulic actuator is:

F = (P1 − P2 )A = M ẍ + B ẋ, (C.5)

where M is the sliding mass load, F is the force pushing the sliding mass and B is the
actuator viscous friction coefficient.
The pump and actuator pipe connection can be modeled as a pressure drop [Habibi and
Goldenberg, 2000] which is represented by P pipe ≈ 2K pipe D2p (ω2P )+Pele (using Darcy’s pipe
flow equation). This pressure drop can be approximated by linearizing this equation at the
operating point ωPop :
∆P pipe ≈ 2K pipe D2p ωPop (∆ω p ). (C.6)

The relationship between the pump and actuator port pressure can be approximated as
P1 = PU − P pipe and P2 = PD + P pipe . Assuming that the flow from the accumulator to
be zero then Q1 = QU and Q2 = QD . Because of the symmetry of the pump and actuator,
under steady state conditions, the exiting flow from the pump equals the inlet flow to the
actuator, and the flow exiting the actuator equals the flow back into the pump. Then the
load flow QL can be approximated to be:

Q1 + Q2 QU + Q D
QL = = .
2 2

114
Substituting the pump and actuator flow equations (C.1), (C.2), (C.3) and (C.4):
V0 dPU dPD Ax dPU dPD
D p ω p − KLCP (PU − PD ) −
( − )− ( + )
2β dt dt 2β dt dt
V0 dP1 dP2 Ax dP1 dP2 Le
= A ẋ + ( − )+ ( + ) + (Li + )(P1 − P2 ).
2β dt dt 2β dt dt 2
Because the pressure drop across the pipe line is very small compare to the pump and
actuator port pressures, (dPU /dt) ≈ (dP1 /dt) and (dPD /dt) ≈ (dP2 /dt). Also due to the
symmetry of the actuator, (dP1 /dt) ≈ −(dP2 /dt). Defining a new parameter Lie = 2Li + Le ,
the approximated model of the pump and actuator connection is obtained as:
V0 dP1 dP2 Lie
D p ω p = A ẋ + ( − ) + (KLCP + )(P1 − P2 ) + 2KLCP ∆P pipe . (C.7)
β dt dt 2
Substituting (C.5) into (C.7) and take Laplace Transform on the equation:
Dp
x(s) =  MV  L   A2 +BL 
ie M/2+KLCP M+BV0 /β ie /2+BKLCP
s3 βA
0
+ s2 A
+s A
2KLCP P pipe (s)
−  MV  L   A2 +BL /2+BK  .
ie M/2+KLCP M+BV0 /β
s3 βA
0
+ s2 A
+ s ie
A
LCP

Recall (C.6), the transfer function of the hydraulic part of the system is:
∆x(s)
G H (s) = =
∆ω p (s)
D p (1 − 2K pipe KLCP D p ωPop )
 MV     A2 +BL /2+BK  .
s3 βA0 + s2 Lie M/2+KLCP
A
M+BV0 /β
+ s ie
A
LCP

Because 2K pipe KLCP D p ωPop  1 is negligible, the approximated hydraulic system transfer
function can be presented as:
∆x(s)
G H (s) = =
∆ω p (s)
Dp
 MV  L   A2 +BL .
ie M/2+KLCP M+BV0 /β ie /2+BKLCP
s3 βA
0
+ s2 A
+s A

A discrete-time state space model is required for this PID plus feedforward controller
design. Define states for the state space model: x = [x1 x2 x3 ] = [x ẋ ẍ] and generate
the discrete-time state space model by using the approximation given by [Grewal and
Amdrews, 2001]:
xi (k + 1) − xi (k)
ẋi (t) = , (C.8)
Ts

115
where T s is the sampling period. Then the discrete-time state space model can be repre-
sented by:

xk+1 = Ad xk + Bd uk ,

yk = Cd xk ,

where

 
 1 Ts 0 
 
 
Ad =  0 1 Ts  ,
 
 2 
0 − T s β[A +B(K
MV0
LCP +Lie /2)]
1− T s [βM(KLCP +Lie /2)+BV0 ]
MV0
 
 0 
 
 
Bd =  0  , Cd = [1 0 0].
 
 βAD p T s 
MV0

C.3 Discrete-Time PI Plus Feedforward Controller with


H∞ Performance
In this section, the design of a discrete-time PI plus feedforward controller for the EHA
system is presented. The derivative term of the PID controller is neglected in order to avoid
the derivative kicks and reduce the computational load of the optimization calculation. The
PI plus feedforward control platform is shown in Figure C.4.

Figure C.4: PI plus feedforward controller platform.

116
The discrete-time model of the EHA system is represented by the following linear
system:

xk+1 = Ad xk + Bd (uk + v1k ), (C.9)

yk = Cd xk + v2k ,

where k is the sampling instant, xk ∈ Rn is the state vector of the EHA system, yk ∈ R p is
the measurement output, uk ∈ Rm is the input control signal, v1k is the disturbance or noise
at the input and v2k represents the measurement noise. The PI plus feedforward control
law is given by
X
k−1
uk = K p ek + Ki ei + K f f (rk−1 − rk ). (C.10)
i=0

Here, K p and Ki are the feedback proportional and integral gains, K f f is the feedforward
gain to be designed.

C.3.1 Transforming the PI plus feedforward controller into an SOF


controller

In order to formulate the problem of PI plus feedforward controller design into a problem
of static output feedback (SOF) control, a new state vector and a new output sequence
Pk−1 T Pk−1 T
x̄k = [xkT , T T T
i=0 ei , rk−1 ] , ȳk = [ek ,
T T
i=0 ei , G(rk−1 − rk ) ] are defined. Here,

ek = rk − yk is the controlled tracking error of the EHA system, G is a weighting factor.


Then the augmented system is written by the following form:

x̄k+1 = Ā x̄k + B̄uk + B̄1 ωk , (C.11)

ȳk = C̄ x̄k + D̄k ,

where
   
 Ad 0 0   Bd 
   

Ā =  −Cd  , B̄ = 
 
I 0  
 0  ,
   
   
0 0 0 0

117
   
 0 Bd 0   −Cd 0 0 
   
B̄1 =  I
  , C̄ =  
0 −I   0 I 0  ,
   
  
I 0 0 0 0 GI
   
 I 0 −I   rk 
   
D̄ =  0
  , ω =  
 .
0 0  k  v1k
   
  
−GI 0 0 v2k

The control signal uk is defined as the SOF of the augmented system, which is:

uk = K ȳk , (C.12)

where K = [K p Ki K f f ].

C.3.2 H∞ optimization

Now the design problem of the PI plus feedforward controller with H∞ performance is
investigated. The SOF H∞ control problem is to find a controller of the form as (C.12)
such that the closed-loop control system is stable, and to minimize the ∞-norm (|| · ||∞ ) of
the system from the external input ωk to the controlled output z̄k (T ωk z̄k ).
The controlled output sequence is chosen as

z̄k = Exk + Fuk , (C.13)

where E = R[0 I 0], R and F are another two weighting matrices. Substituting (C.12)
into (C.11) and combining (C.13), the closed-loop control system for H∞ optimization can
be represented by the following form

x̄k+1 = [Ā + B̄K C̄] x̄k + [ B̄K D̄ + B̄1 ]ωk , (C.14)

z̄k = [E + FK C̄] x̄k + FK D̄ωk .

Using the well known Bounded Real Lemma for discrete-time system [Gahinet and
Apkarian, 1994] and the congruence transformation, the H∞ optimization problem for the
SOF controller design leads to the following theorem.

118
Theorem 2 Given the plant (C.9), the PI plus feedforward controller (C.10) guarantees
that the closed-loop system (C.14) is stable and ||T ωz̄ ||∞ < γ if there exist symmetrical
matrices P > 0, Q > 0 and matrix K, with appropriate dimensions, which satisfy the
following LMIs
 
 −Q [Ā + B̄K C̄] [ B̄K D̄ + B̄1 ] 0 
 
 
 ∗ −P 0 [E + FK C̄]T 
  < 0, (C.15)
 
 ∗ ∗ −γI [FK D̄]T 
 

∗ ∗ ∗ −γI
 
 P I 
 
  > 0, (C.16)
I Q 

and a constraint
PQ = I. (C.17)

Proof 1 From the closed-loop system represented by (C.14), apply the discrete-time Bounded
Real Lemma [Gahinet and Apkarian, 1994], the H∞ controller exist if and only if there ex-
ist positive symmetric matrix P and matrix K satisfying
 
 −P P[Ā + B̄K C̄] P[ B̄K D̄ + B̄1 ] 0 
 
 T 

 ∗ −P 0 [E + FK C̄] 
  < 0. (C.18)
 
 ∗ ∗ −γI [FK D̄]T 
 

∗ ∗ ∗ −γI

This is a bilinear matrix inequality problem which is very complicated to be solved. Taking
congruence transform on both sides of (C.18) by diag[P−1 I I I]T and diag[P−1 I I I],
and define a new LMI matrix Q = P−1 . Then (C.18) is transformed into

 
 −Q [Ā + B̄K C̄] [ B̄K D̄ + B̄1 ] 0 
 
 T 

 ∗ −P 0 [E + FK C̄] 
  < 0. (C.19)
 
 ∗ ∗ −γI [FK D̄]T 
 

∗ ∗ ∗ −γI
This completes the proof.

119
Remark 1: The conditions of (C.15), (C.16) and (C.17) are in fact an LMI problem with
non-convex constraints. It can be conveniently solved by using the cone complementar-
ity linearization (CCL) algorithm [Laurent et al., 1997]. The results of solving the LMI
problem give rise to the matrix K, and then PI plus feedforward controller gains can de-
termined.

C.4 Simulation Studies and Experimental Tests


According to the Theorem 1 presented in the previous section, the CCL algorithm is em-
ployed to solve the LMI problem in (C.15), (C.16) and (C.17) and as a result, an optimized
solution of K is obtained by minimizing the γ as:

K = [2428.6 20.2 2.8 × 104 ].

Then the SOF control signal equals to:

X
k−1
uk = K[eTk , eTi , G(rk−1 − rk )T ]T .
i=0

C.4.1 Simulation study

The simulation tracking response of the EHA system using the proposed PI plus feedfor-
ward controller is shown in Figure C.5.

tracking response
0.06
reference signal
Piston Position

0.04

0.02

−0.02
0 1 2 3 4 5 6
Time(s)

Figure C.5: Simulation tracking response with the feedforward.

120
For the purpose of comparison, the simulation tracking response of the EHA system
using the proposed H∞ PI controller without including the feedforward pass is also shown
in Figure C.6.

reference signal
0.06
tracking response
Piston Position (m)

0.04

0.02

−0.02
0 1 2 3 4 5 6
Time (s)

Figure C.6: Simulation tracking response without the feedforward.

It’s hard to tell a big difference from the two simulation tracking responses. The track-
ing error of these two simulation results are shown in Figure C.7.

−3
x 10
1.5

1
Tracking Error(m)

0.5

−0.5
PI control
−1 PI control
with feedforward
without feedforward
−1.5
0 1 2 3 4 5 6
Time(s)

Figure C.7: Simulation tracking errors.

Remark 2: The H∞ PI controller without the feedforward term was developed using
the same optimization method in Theorem 1. Using the proposed PI plus feedforward
controller, the tracking error gets significantly smaller than the one without a feedforward
term.

121
C.4.2 Experimental test

The experimental EHA system under study is shown in Figure C.8. The hardware-in-the-
loop experimental test system includes the following components: Pentium IV computer,
PCI-DAS1602/16 Analog & Digital I/O Board, Gurley LE18 Linear Encoder and the de-
signed bidirectional hydraulic circuit. Experimental tests were performed to confirm and
verify the observations obtained from simulation studies. Same tracking tests are done

Figure C.8: Experimental setup of the EHA system.

on the experimental EHA system as the simulation results. Experimental studies for this
proposed PI plus feedforward controller are shown in Figure C.9, Figure C.10 and Fig-
ure C.11 to confirm the observations made by simulations. The stroke of the experiment
test is chosen to be smaller than the simulation test because of the physical limit of the
EHA system. For the purpose of comparison, a Z-N tuning PI controller was designed for
the EHA system. The experimental tracking error is also shown in Figure C.11.
From the experimental results, it can be summarized that:

1. The experimental tracking error looks very close to the simulation tracking errors

122
0.03
reference signal
0.025 tracking response

Piston Position (m)


0.02

0.015

0.01

0.005

−0.005
0 1 2 3 4 5 6
Time (s)

Figure C.9: Experimental tracking response with the feedforward.

0.03
reference signal
0.025 tracking response
Piston Position (m)

0.02

0.015

0.01

0.005

−0.005
0 1 2 3 4 5 6
Time (s)

Figure C.10: Experimental tracking response without the feedforward.

using the same controllers.

2. From Figure C.11, the experimental tracking error using the proposed PI plus feed-
forward controller did decrease comparing with the one without a feedforward loop.
However, the tracking error didn’t decrease that significant as was shown in the sim-
ulation in Figure C.7. It is suspected that the main reasons caused this phenomena
is: the practical EHA system is naturally a nonlinear system, the linear EHA model
used for the controller design must have missed some dynamic characteristics of the
system;

3. Despite its simple formulation, the proposed PI plus feedforward controller can still
achieve a very accurate tracking performance (tracking error is less than 3.6% of the

123
−3
x 10
4

3
H∞ PI controller Z-N PI controller
without feedforward
2

Tracking Error (m)


1

−1
H∞ PI controller
−2 with feedforward

−3

−4
0 1 2 3 4 5 6
Time (s)

Figure C.11: Experimental tracking errors.

full tracking displacement);

4. The H∞ PI controller outperforms the traditional Z-N PI tuning method which can
be easily seen in Figure C.11.

C.5 Conclusions

This work has proposed a discrete-time PI plus feedforward controller design for an elec-
trohydraulic actuator system. The controller formulation was transformed into a SOF
problem with H∞ performance, LMI optimization technique was applied to solve the con-
troller design problem. To the author’s best knowledge, this work is the first of its kind
that solving a PID plus feedforward controller using the LMI optimization technique.
Simulation studies and experimental tests on the EHA system verify the effectiveness
of the proposed method for position tracking from the application perspective. The extra
feedforward term did improve the tracking performance significantly. Despite its simple
setup, the proposed PI plus feedforward controller achieved very good tracking perfor-
mance. However, there is no guarantee that there exists a general form of solution for all
type of control systems. If one can find such a general solution, maybe we don’t have
to turn to other complex control design algorithms when high control performances are

124
preferred.

Nomenclature
M Mass of the load. 20Kg
Ap Pressure area in symmetrical actuator. 5.05 × 10−4 m2
Dp Pump displacement. 1.6925 × 10−7 m3 /rad
β Bulk modulus of the hydraulic oil. 2.1 × 108 Pa
CT Lumped leakage coefficient. 5 × 10−13 m3 /sPa
KLCP Pump cross-port leakage coefficient.
Le External leakage coefficient.
Ts Sampling period. 0.001s
V0 Mean volume of the hydraulic actuator.
P1 , P2 Upstream and downstream actuator chamber pressure.
ωp Angular velocity of the hydraulic pump.
B Viscous friction coefficient. 760N s/m
P, Q, K LMI variable matrices.
x System states.
x1 , x2 , x3 Position, velocity and acceleration of the load.

125
R
Aström, K., Hägglund, T., 2001. The future of PID control. Control Engineering Practice
9, 1163–1175.
Bara, G.I., Boutayeb, M., 2005. Static output feedback stabilization with H∞ performance
for linear discrete-tiem systems. IEEE Transactions on Automatic Control 50, 250–254.
Bianchi, F.D., Mantz, R.J., Christiansen, C.F., 2008. Multivariable PID control with set-
point weighting via bmi optimization. Automatica 44, 472–478.
Dieulot, J.Y., Colas, F., 2009. Robust pid control of a linear mechanical axis: A case study.
Mechatronics 19, 269–273.
Gahinet, P., Apkarian, P., 1994. A linear matrix inequality approach to H∞ control. Inter-
national Journal of Robust and Nonlinear Control 4, 421–448.
Ge, M., Chiu, M.S., Wang, Q.G., 2002. Robust PID controller design via LMI approach.
Journal of Process Control 12, 3–13.
Ge, P., Jouaneh, M., 1996. Tracking control of a piezoceramic actuator. IEEE Transactions
on Control Systems Technology 4, 209–216.
Gomis-Bellmunt, O., Campanile, F., Galceran-Arellano, S., Montesinos-Miracle, D., Rull-
Duran, J., 2008. Hydraulic actuator modeling for optimization of mechatronic and
adaptronic systems. Mechatronics 18, 634–640.
Grewal, M.S., Amdrews, A.P., 2001. Kalman filtering theory and practice using MAT-
LAB. JOHN WILEY & SONS.
Habibi, S.R., Goldenberg, A.A., 2000. Design of a new high-performance electrohydraulic
actuator. IEEE/ASME Transactions on Mechatronics 5, 158–165.
Hang, C.C., Aström, K., Ho, W.K., 1991. Refinements of the ziegler-nichols tuning for-
mula. IEE proceedings. Part D. Control Theory and Applications 138, 111–118.
Laurent, E.G., Francois, O., Mustapha, A., 1997. A cone complementarity linearization
algorithm for static output feedback and related problems. IEEE Transactions on Auto-
matic Control 42, 1171–1176.
Leva, A., Bascetta, L., 2006. On the design of the feedforward compensator in two-degree-
of-freedom controllers. Mechatronics 16, 533–546.
Lin, Y., Shi, Y., Burton, R., 2009. Modeling and robust discrete-time sliding mode control
design for a fluid power electrohydraulic actuator (eha) system, in: Dynamic System
Control Conference, Hollywood, USA.
Rovira-Mas, F., Zhang, Q., Hansen, A.C., 2007. Dynamic behavior of an electrohydraulic
valve: typology of characteristic curves. Mechatronics 17, 551–561.

126
Sampson, E., 2005. Fuzzy control of the electrohydraulic actuator. Master’s thesis. Uni-
versity of Saskatchewan.

Thanh, T.D.C., Ahn, K.K., 2006. Nonlinear PID control to improve the control perfor-
mance of 2-axes pneumatic artificial muscle manipulator using neural network. Mecha-
tronics 16, 577C587.

Toufighi, M.H., Sadati, S.H., Najaif, F., 2007. Modeling and analysis of a mechatronic
actuator system by using bond graph methodology, in: Proceedings of IEEE Aerospace
Conference, Big Sky, MT. pp. 1–8.

Visioli, A., 2004. A new design for a PID plus feedforward controller. Journal of Process
Control 14, 457–463.

Wang, J., Wu, J., Wang, L., You, Z., 2009. Dynamic feed-forward control of a parallel
kinematic machine. Mechatronics 19, 313C324.

Wang, S., Habibi, S., Burton, R., Sampson, E., 2008. Sliding mode control for an elec-
trohydraulic actuator system with discontinuous non-linear friction. Proceedings of the
Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering
222, 799–815.

Yanada, H., Furuta, K., 2007. Adaptive control of an electrohydraulic servo system utiliz-
ing online estimate of its natural frequency. Mechatronics 17, 337C343.

Zheng, F., Wang, Q.G., Lee, T.H., 2002. On the design of multivariable PID controller via
LMI approach. Automatica 38, 517–526.

127
A D
M  R D-T S M
C D   F P E
A (EHA) S

Yang Lin, Yang Shi and Richard Burton

Abstract

This paper studies the design of a robust discrete-time sliding mode control (DT-SMC)
for a high precision electro-hydraulic actuator (EHA) system. Nonlinear friction in the
hydraulic actuator can greatly influence the performance and accuracy of the hydraulic
actuators; however, it is difficult to accurately model nonlinear friction characteristics. In
this paper, it is proposed to characterize frictions as an uncertainty in the system matrices.
Indeed, the effects of variations of the nonlinear friction coefficients are considered as
norm bounded uncertainties that span a bounded region to cover a wide range of real
actuator friction. For such a discrete-time dynamic model for the EHA system with system
uncertainty matrices and a nonlinear term, a sufficient condition for existence of stable
sliding surfaces is proposed by using the linear matrix inequality (LMI) approach. Based
on this existence condition, a DT-SMC is developed such that the reaching motion satisfies
the discrete-time sliding mode reaching condition for uncertain systems. Simulation and
experimental studies on the EHA system illustrate the effectiveness and applicability of the
proposed method.

D.1 Introduction
High precision position control for mechatronic systems has drawn great attention. Vari-
ous control schemes have been applied to mechatronic systems to compensate for nonlin-
earities, model uncertainties, and disturbances [Tan et al., 2009; Yi et al., 2009; Bashash

128
and Jalili, 2009; Lee and Salapaka, 2009]. Hydraulic positioning systems play an impor-
tant role in transportation, earth moving equipment, aircraft and industry machinery with
heavy duty applications. Traditional hydraulic transmission control systems are mainly
based on valve controlled hydraulic systems [Merritt, 1967]. Pump controlled hydraulic
systems, known as hydrostatic systems, are used as an alternative for valve controlled
systems in applications in which higher energy efficiencies are preferred. One configura-
tion for a hydrostatic system used a fixed displacement pump which regulates the flow by
changing the rotational speed and direction of the pump [Habibi and Goldenberg, 2000;
Chinniah, 2004]. In this paper, the focus is on a new positioning control design for a
particular electrohydraulic actuator (EHA) system.

Friction compensation for the position control of mechatronic systems has received
much attention. In [Tafazoli et al., 1998], the importance of friction modeling and com-
pensation for its effect was emphasized, and an adaptive control algorithm was designed
to compensate for the nonlinear friction in a hydraulic actuator to achieve tracking perfor-
mance; however, stability analysis was not included in this work. In [Marton and Lantos,
2009], an LQ control algorithm was designed for an electric motor system with stribeck
friction and backlash; the nonlinear friction dynamics were considered as a set of linear
models in different partitions of the state space by linearizing the friction model on differ-
ent portions. In [Wit et al., 1995], a new friction model was designed for mechatronic
systems; stribeck effect, hysteresis, spring-like characteristics for stiction and varying
break-away force were all considered in a simple nonlinear differential equation model,
and a simple observer based friction compensator was designed. A LuGre friction model
was applied to a hydraulic actuator system, and the adaptive observer and controller were
designed for its position tracking control [Zeng and Sepehri, 2008]. However, to the au-
thors’ best knowledge, there are no research work dealing with the friction compensation
problem from the perspective of robust control.

For the particular EHA system under study in this paper, it was ascertained that hy-
draulic friction in the actuator was a critical issue affecting system performance in terms
of accuracy and repeatability of the actuator performance [Owen and Croft, 2003]. In
[Armstrong-Helouvry et al., 1994], seven types of models of friction for mechatronic sys-

129
tems were comprehensively studied and summarized. However, trying to clearly identify
which model best describes the physical system under consideration is often very difficult.
Instead of pursuing an accurate model for the nonlinear friction, an appropriate scheme
to compensate for the nonlinear friction uncertainties can be designed – this observation
motivates the work in this paper.

A linearized model of the EHA was developed by only considering the linear viscous
friction in the hydraulic actuator [Habibi and Goldenberg, 2000]. Based on this linear
model, gain scheduling control was applied to the prototype EHA system [Habibi and
Sigh, 2000]. In [Chinniah, 2004], experimental results showed that static, coulomb and
viscous friction (slip-stick) co-exist in the EHA’s linear actuator. A typical friction char-
acteristic of the EHA of interest is illustrated in Figure D.1. In this figure, a curve fit
polynomial equation was used to approximate the actual measured friction characteristics.
However, some concerns were raised by researchers about the characteristics at high and
at very low velocities, and the rate of change of velocity. In addition, hysteresis was noted
but not reflected in the model. To compensate for the modeling uncertainties of this non-
linear friction, a sliding mode controller was designed and applied to this nonlinear EHA
model [Wang et al., 2008]. This form of controller was chosen because of its ability to
compensate for model uncertainties and disturbances.

Sliding model control (SMC) first appeared in the context of variable structure sys-
tems. Due to its low sensitivity to disturbances and parameter variations, SMC has be-
come an efficient tool in the control of complex systems with uncertainties [Utkin et al.,
1999]. The main idea behind the SMC is to use a discontinuous control input to force the
state trajectory to a certain sliding surface and to remain on this surface over time. The dy-
namic characteristic of the resulting closed-loop control system will be mainly determined
by the design of the sliding surface. In [Utkin and Yang, 1978], several classical methods
to design the sliding surface for both continuous-time and discrete-time systems were pro-
posed. Further, the design, analysis, and application of SMC were elegantly presented in
[Utkin, 1993]. Since then, many research works have been devoted to the development of
advanced SMC schemes for systems with uncertainties, nonlinearities, new sliding surface
design, and other practical constraints. In [Choi, 1998] and [Choi, 1999], continuous-time

130
50

40

Friction Force (N) 30

20

10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Piston Velocity (m/s)

Figure D.1: Experimental friction in the EHA and the identified nonlinear
friction model [Chinniah, 2004]: (1) ∗ points: Experimentally measured
friction force; (2) Solid curve: Estimated curve for the friction force; (3)
Dashed dotted line: Approximated linear friction force [Chinniah, 2004].

design for systems with mismatched uncertainties was considered by using the linear ma-
trix inequality (LMI) technique. In [Gao et al., 1995], a discrete-time SMC design with an
improved sliding surface was proposed. In [Tang and Misawa, 2000], a type of state feed-
back variable structure controller for linear multivariable continuous-time systems with
unmatched uncertainties was studied. An output feedback SMC design was proposed
for discrete-time systems by using a multirate scheme [Janardhanan and Bandyopadhyay,
2006]. SMC was also developed for systems with time-varying state delays [Yan and Shi,
2008; Yan et al., 2008]. However, it appears that a robust discrete-time SMC (DT-SMC)
design for systems with uncertainties has not been fully investigated from the application
perspective, which indeed, motivates this work. The development of a robust DT-SMC for
the EHA system from an application perspective is the focus of the paper.
The main objectives of this work are three-fold.

1. To further analyze the effects of the uncertain friction in the EHA system, and to
model them as uncertain system matrices with norm bounds. The bounded region
covers all nonlinear friction behavior for the EHA system considered.

2. To design an appropriate sliding surface for the EHA system and to determine its

131
parameters by efficiently solving a set of linear matrix inequalities (LMIs).

3. To develop a DT-SMC for the EHA system such that the reaching motion satisfies
the discrete-time sliding mode reaching condition.

The remainder of this paper is organized as follows. In Section D.2, the EHA system
configuration and its discrete-time nonlinear uncertain model are presented. The sliding
surface and the SMC are developed in Section D.3. Simulation studies and experimental
tests for the EHA system are illustrated in Section D.4 and Section D.5, respectively.
Finally, conclusion remarks are offered in Section D.6.
The notation used throughout this paper is fairly standard. The superscript T stands for
the matrix transposition, Rn denotes the n-dimensional Euclidean space, and Rm×n is the
set of all m × n real matrices. The notation P > 0 means that P is real symmetric and
positive definite, and I and 0 represent identity matrix and zero matrix respectively with
appropriate dimensions. In symmetric block matrices or long matrix expressions, the as-
terisk (∗) represents a term that is induced by symmetry. Matrices, if their dimensions are
not explicitly stated, are assumed to be compatible for algebraic operations. The physical
meanings of some important variables involved are elaborated in the nomenclature.

nomenclature

M Mass of the load. 20Kg

Ap Pressure area in symmetrical actuator. 5.05 × 10−4 m2

Dp Pump displacement. 1.6925 × 10−7 m3 /rad

a1 , a2 , a3 Coefficients of the nonlinear actuator friction.

βe Bulk modulus of the hydraulic oil. 2.1 × 108 Pa

CT Lumped leakage coefficient. 5 × 10−13 m3 /sPa

ξ Pump cross-port leakage coefficient.

L External leakage coefficient.

132
Ts Sampling period. 0.001s

V0 Mean volume of the hydraulic actuator.

P1 , P2 Upstream and downstream actuator chamber pressure.

ωp Angular velocity of the hydraulic pump.

η Viscous friction coefficient. 760N s/m

Ff Modeled Hydraulic friction.

∆A Uncertainty term for the system matrix.

G, D, H1 Norm bounded uncertainty term.

q, ε Adjustable parameters in the DT-SMC control design.

P, Y, ψ LMI variable matrices.

x System states.

z Regular transformed states.

∆f Nonlinear function.

x1 , x2 , x3 Position, velocity and acceleration of the load.

ω1 , ω2 , ω3 Lumped system noises and disturbances.

J State transformation matrix.

Figure D.2: Schematic of the hydraulic circuit of the system.

133
D.2 Nonlinear Model of the EHA System
In this section, results on the modeling of the EHA system are presented. How to in-
sightfully and effectively characterize the nonlinear friction in terms of the norm-bounded
system uncertainty matrices is the focus of this section.

D.2.1 Related works

The layout of the EHA system is shown in Figure D.2. The actuator is driven by a bi-
directional fixed displacement gear pump. A special symmetrical actuator is connected
with the load and the motion of the load is controlled by varying the speed of the electric
motor. A simplified model for the hydraulic system was developed in [Habibi and Sigh,
2000], which was derived by assuming that only viscous friction exists in the actuator.
The equation relating the input angular speed to the actuator velocity and load pressure
differential was given in [Habibi and Goldenberg, 2000]:

V0 Cep
D p ω p = A p ẋ + (Ṗ1 − Ṗ2 ) + (ξ + )(P1 − P2 )
2βe 2
L
+(2ξ + Cep )P pipe + (P1 − P2 ). (D.1)
2

The EHA system is connected to a horizontal sliding mass, and the equation of motion for
the whole system was given in [Habibi and Goldenberg, 2000]:

(P1 − P2 )A p = M ẋ + F f , (D.2)

where F f comprises three parts: static, coulomb and viscous friction. In [Habibi and
Goldenberg, 2000], for ease of the model derivation and for the convenience of the control
design, the static and coulomb friction were neglected and only the viscous friction was
considered when modeling the actuator friction, that is:

F f = η ẋ, (D.3)

where η is the viscous friction coefficient. A linearized model of the EHA system was
developed based on this assumption. To fully model the actuator friction, a nonlinear

134
model was developed in [Chinniah, 2004], where both coulomb and static friction were
incorporated, that is:

F f = a1 ẋ2 + a2 ẋ + a3 ẋ > 0, (D.4)

F f = −a1 ẋ2 + a2 ẋ − a3 ẋ < 0.

Here, a1 , a2 , a3 are the nonlinear friction coefficients. In [Wang et al., 2008], a nonlinear
model of the hydraulic part of the EHA system was developed as:

x1 (k + 1) = x1 (k) + T s x2 (k) + T s w1 (k), (D.5)

x2 (k + 1) = x2 (k) + T s x3 (k) + T s w2 (k), (D.6)


" #
a2 V0 + MβeCT
x3 (k + 1) = 1 − T s ( ) x3 (k) (D.7)
MV0
(2A2p + a2CT )βe
−T s x2 (k)
MV0
2a1 V0 x2 (k)x3 (k)
−T s sgn(x2 (k))
MV0
h i
βeCT a1 (x2 (k))2 + a3
−T s sgn(x2 (k))
MV0
2A p D p βe
+T s u(k) + T s w3 (k),
MV0

where x1 , x2 , x3 are the position, velocity and acceleration of the load, and T s is the sam-
pling period, u(k) represents the control signal which is the rotation speed of the bi-
directional hydraulic pump. The discretization approximation

xi (k + 1) − xi (k)
ẋi = (D.8)
Ts

was used to transform the continuous model to a discrete-time one. For this nonlinear
model, a sliding mode controller was designed in [Wang et al., 2008].
In the aforementioned studies, uncertainties on the system matrices introduced by
time-varying friction parameters were not incorporated into the model. Two challenges
were thus identified:

• How to explicitly incorporate the uncertainties on the system matrices into the
model?

135
• How to design a robust control scheme to further improve the positioning control
performance?

In the next subsection, the first challenge will be addressed and the “quest” for the second
one will be elaborated in Section D.3.

D.2.2 Nonlinear model with uncertainties

It has been observed from the experimental testing on the EHA that the variations of
friction parameters would result in changes to the EHA system matrices. Therefore, it was
proposed to model the friction uncertainties as norm-bounded uncertainties on the system
matrices.
This uncertainty is characterized by a ∆A matrix in the following state-space model of
the EHA system:
 
 x1 (k + 1) 
 
 
X(k + 1) =  x2 (k + 1)  (D.9)
 
 
x3 (k + 1)
 
 x1 (k) 
 

= (A + ∆A)  x2 (k) 
 (D.10)
 
 
x3 (k)
 
 w1 (k) 
 
 
+B[u(k) + ∆ f (x(k), k)] + T s  w2 (k)  ,
 
 
w3 (k)

where X(k) ∈ Rn , u(k) ∈ Rm , and


 
 1 Ts 0 
 
 
A =  0 1 Ts

 T s (a2 V0 +Mβe CT ) 
] 
2
 0 − T s (2A p +a2CT )βe [1 −
MV0 MV0
 
 1 0.001 0 
 
=  0 1 0.001  ,
 
0 −78.10 1.07

136
  
 0   0 
   
B =  0  =  0  ,
 2T s A p D p βe   
MV0 1.07

h i
2a1 V0 x2 (k)x3 (k) + βeCT a1 (x2 (k)2 ) + a3
∆ f (x(k), k) = −
2A p D p βe
·sgn(x2 (k)).

In general, ∆A can be expressed in a general norm-bounded uncertainty form [Rotea and


Khargonekar, 1989], that is: ∆A = GD(k)H1 , where G, H1 are known real constant matri-
ces with compatible dimensions, D(k) is unknown norm-bounded and generally satisfies
D(k)T D(k) ≤ I.
Specifically for the EHA system, the uncertainty of the system matrix A arises from
the uncertainty of a2 : ∆a2 . By analyzing the experimental data, it is recognized that
10% variations on a1 , a2 , and a3 can reasonably capture the real friction variation. For
the nominal value a2 = −1450, there exists a variation ∆a2 satisfying |∆a2 | ≤ 0.1|a2 |.
Therefore, ∆A can be written as:
 
 0 0 0 
 
∆A =  0 0 0  (D.11)
 0 − T s (∆a2CT )βe − T s (∆a2 V0 ) 
MV0 MV0
 
 0 0 0 
 
=  0 0 0 .
 0 7.66 × 10−5 ∆a 5.03 × 10−5 ∆a 
2 2

Further, as stated, ∆A can be expressed in a general norm-bounded uncertainty form:

∆A = GDH1
 
 0 
   


=  0  D(k) 0 
 (D.12)
0.111 0.073
 
 
0.1

with D(K) in this specific case, a scalar satisfying −1 ≤ D(k) ≤ 1.


The uncertainties introduced by time-varying friction parameters are illustrated in Fig-
ure D.3: (1) When the parameter variations are 10% on a1 , a2 , and a3 , simultaneously, that

137
is, 0.1ai , (i = 1, 2, 3), then it results in the upper bound friction curve; (2) when the param-
eter variations are −0.1ai , (i = 1, 2, 3), then the lower bound friction curve is obtained.
Clearly, the actual friction force (star points) can be fully covered within the boundary
spanned by the upper and lower curves representing the norm-bounded uncertainties.

Remark 1: The experimental friction force (stars) could be all lumped within the friction
bound generated from the ±10% bounds. This is a raw measurement; of course, a little
bit smaller bound can also include all the blue star friction forces; here, the bounds are
enlarged in case of possible measurement errors.

Remark 2: The proposed nonlinear model with norm-bounded uncertainties can fully
characterize the actual variation in friction parameters. Further, the “tighter” the boundary
is to the actual friction force, the less conservative the model would be.

Remark 3: It is desirable to design a controller that would be robust against all possible
system uncertainties falling within the boundary spanned by the norm-bounded uncertain-
ties.

50
upper bound of
Friction Force Ff (N)

40 the friction

30

20 lower bound of
the friction
10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Piston Velocity (m/s)

Figure D.3: Nonlinear uncertain friction band.

138
D.2.3 Model transformation for the SMC design

To facilitate the following SMC design, a state transformation z = Jx, is introduced where
J ∈ Rn×n is a real nonsingular matrix satisfying
 
 0 
 (n−m)×m 
JB =   , (D.13)
 B 2

and J is not unique.


Then, the system model can be transformed into the following regular form

z1 (k + 1) = (Ā11 + ∆Ā11 )z1 (k) + (Ā12 + ∆Ā12 )z2 (k),

z2 (k + 1) = (Ā21 + ∆Ā21 )z1 (k) + (Ā22 + ∆Ā22 )z2 (k)

+B2 [u(k) + ∆ f (z(k), k)], (D.14)

where J T = [U2 U1 ], Ā11 = U2T AU2 , ∆Ā11 = U2TGDH1 U2 , Ā12 = U2T AU1 , ∆Ā12 = U2TGDH1 U1 , Ā21 =
U1T AU2 , ∆Ā21 = U1TGDH1 U2 , Ā22 = U1T AU1 , ∆Ā22 = U1TGDH1 U1 .

This transformation is highly desirable in that one of the state equations (z1 (k + 1))
will become independent of the control input and system nonlinearities. This facilitates
the design of the SMC.

D.3 Robust Discrete-Time Sliding Mode Control Design


for the EHA Model
Generally, robust stability, meaning that the stability can still be guaranteed even if the
system is subject to uncertainties, is of practical importance. In this section, the design of
robust DT-SMC for the EHA model with uncertainties is to be presented. Two require-
ments of the DT-SMC design are as follows:

1) The discrete-time sliding surface S (k) must be asymptotically stable;

2) The robust DT-SMC law u(k) must guarantee the sliding mode reaching condition.

139
D.3.1 Sliding surface design

In state tracking or trajectory following problems, the sliding surface is defined in terms
of the error. Since the state in the transformed model is z, the error with respect to zd will
be considered. The control problem becomes one of constraining the states z to follow a
prescribed trajectory zd . Assuming that the desired states satisfy the nonlinear model for a
“hypothetical” control input [Misawa, 1997], such that:

z1d (k + 1) = (Ā11 + ∆Ā11 )z1d (k) + (Ā12 + ∆Ā12 )z2d (k),

then,

z1e (k + 1) = z1d (k + 1) − z1 (k + 1) (D.15)

= (Ā11 + ∆Ā11 )z1e (k) + (Ā12 + ∆Ā12 )z2e (k).

As a first step, a linear sliding surface S (k) is to be designed such that the sliding
surface S (k) = 0 is asymptotically stable. The linear sliding surface is defined as:

S (k) = C̄ze (k) = [C Im ]ze (k) = Cz1e (k) + z2e (k) = 0. (D.16)

Here, C is a real matrix to be calculated. Substituting z2e (k) = −Cz1e (k) into (D.15), then

z1e (k + 1) = [Ā11 + ∆Ā11 − (Ā12 + ∆Ā12 )C]z1e (k), (D.17)

which represents the sliding motion of the transformed system in (D.14).


Before proceeding further, a lemma which plays an important role in the derivation of
the sliding surface design is introduced.

Lemma 1 [Xie, 1996]: Given matrices Σ1 = ΣT1 , Σ2 and Σ3 with appropriate dimensions,
the following inequality
Σ1 + Σ2 ∆(k)Σ3 + ΣT3 ∆T (k)ΣT2 < 0

holds for all ∆(k) satisfying ∆(k)T ∆(k) ≤ I, if and only if there exists some λ > 0 such that

Σ1 + λ−1 Σ2 ΣT2 + λΣT3 Σ3 < 0.

140
As a second step, the corresponding parameters involved in the sliding surface are to
be obtained by using the LMI technique. The main results regarding the sliding surface
design is summarized in Theorem 1.

Theorem 1: The system in (D.14) is robustly asymptotically stable if there exist symmet-
rical matrices P > 0, Y > 0, ψ > 0 and matrix C, with appropriate dimensions, which
satisfy the following LMI
 
 −P ∗ ∗ 
 
 
 Ā11 − Ā12C ψU2TGGT U2 − Y ∗  < 0, (D.18)
 
 
H1 U2 − H1 U1C 0 −ψ

and a constraint
PY = I. (D.19)

Here, ‘∗’ represents the blocks that are inferred by symmetry.


Proof: For the system in (D.14), define a Lyapunov function candidate

V(k) = zT1e (k)Pz1e (k), (D.20)

where P is positive definite (P > 0).


The finite difference ∆V(k) of the Lyapunov function must be such that:

∆V(k) = V(k + 1) − V(k)

= zT1e (k)[[(Ā11 + ∆Ā11 ) − (Ā12 + ∆Ā12 )C]T P

·[(Ā11 + ∆Ā11 ) − (Ā12 + ∆Ā12 )C] − P]

·z1e (k) < 0. (D.21)

Apply the Schur complement [Boyd et al., 1994], (D.21) can be transformed into:
 
 −P ∗ 
 
  < 0 (D.22)
(Ā + ∆Ā ) − (Ā + ∆Ā )C −P−1 
11 11 12 12

then
 
 −P ∗ 
 
 
Ā11 − Ā12C −P−1

141
 
 0   
 
+   D H1 U2 − H1 U1C 0
 U TG
2
 T
 T  0 
 
+ H1 U 2 − H1 U 1 C 0 DT   < 0. (D.23)
 U TG
2

Apply Lemma 1, (D.23) can be satisfied if and only if there exists some λ > 0 such that
 
 −P ∗ 
 

Ā11 − Ā12C −P−1 
 
 0 0 
 
+ 
 0 λ−1 U TGGT U 
2 2
 
 U T H T − C T U T H T   
 2 1 1 1 
+ 
  λ H1 U2 − H1 U1C 0 < 0. (D.24)
0

Again apply Schur complement to (D.24):


 
 −P ∗ ∗ 
 
 
 Ā11 − Ā12C λ−1 U2TGGT U2 − P−1 ∗  < 0. (D.25)
 
 
H1 U2 − H1 U1C 0 −λ−1

Redefine Y = P−1 , ψ = λ−1 . Then, it can be readily shown that ∆V(k) < 0 if the LMI
(D.18) and the constraint (D.19) are satisfied. Based on the Lyapunov stability theory
[Boyd et al., 1994], the system in (D.14) is asymptotically stable. This completes the
proof. ¤

Remark 4: The conditions of the asymptotic stability for the system in (D.14) are in fact
an LMI problem with a non-convex constraint. It can be conveniently solved by using the
cone complementarity linearization (CCL) algorithm [Laurent et al., 1997]. The results of
solving the LMI problem give rise to the matrix C, and then the sliding surface is readily
determined.

D.3.2 Robust sliding mode control design

After designing the sliding surface according to Theorem 1, the robust DT-SMC control
signal u(k) can be derived to satisfy the sliding mode reaching condition. The control law

142
takes the following form:

u(k) = u+ (k), when S (k) > 0,

u(k) = u− (k), when S (k) < 0.

Such a control law can force the controlled trajectory to move toward the sliding surface
and then oscillate around the sliding surface in a “chattering” type motion. The control
signal guarantees that once the state trajectory crosses the sliding surface for the first time,
it will cross the surface again in each sampling period. This type of state motion will
be included inside a certain band – the so-called quasi-sliding mode band (QSMB) [Gao
et al., 1995]. In general, the smaller the band, the better the controlled performance. The
proposed design of the DT-SMC law is summarized in the following theorem.

Theorem 2: For the transformed model of the EHA system in (D.14), assume that the LMI
(D.18) with a constraint (D.19) has feasible solutions P, Y, ψ and C. Then with the designed
sliding surface given in (D.16), the sliding mode reaching condition can be satisfied with
the following control law

u(k) = −B−1
2 [C̄ Āze (k) − (1 − T s q)S (k) + εT s · sgn(S (k))

+(α0 + β0 ) + (αδ + βδ ) · sgn(S (k))], (D.26)

where
αU + α L αU − αL
α0 = , αδ = ,
2 2
βU + β L βU − βL
β0 = , βδ = , (D.27)
2 2
Ā = JAJ −1 .

αU and αL are the upper and lower bounds of α(k), βU and βL are the upper and lower
bounds of β(k) such that:

αL ≤ α(k) = C̄J∆AJ −1 ze (k) ≤ αU , (D.28)

βL ≤ β(k) = B2 ∆ f ≤ βU . (D.29)

Proof: Design of the DT-SMC for the EHA system model with uncertainties must guar-
antee that the system dynamics reach the QSMB in limited time. Now apply the following

143
reaching condition to the discrete-time system [Gao et al., 1995]:



 S (k + 1) − S (k) ≤ −εT s sgn(S (k)) − qT s S (k),








 if S (k) > 0
∆S (k) = 



 S (k + 1) − S (k) ≥ −εT s sgn(S (k)) − qT s S (k),






 if S (k) < 0

where T s is the sampling period, q > 0 and ε > 0 are two adjustable parameters and
1 − T s q > 0.
The incremental change of S (k) can be calculated by:

∆S (k) = C̄ze (k + 1) − C̄ze (k) (D.30)

= C̄(Ā + ∆Ā)ze (k) + B2 [u(k) + ∆ f (z(k), k)] − C̄ze (k),

where ∆Ā = J∆AJ −1 .


Further, substituting the designed DT-SMC control law (D.26) into (D.30) gives rise
to

∆S (k) = α(k) + βk − T s qS (k) − εT s · sgn(S (k)) (D.31)

−(α0 + β0 ) − (αδ + βδ ) · sgn(S (k)).

It is observed that

α(k) ≤ α0 + αδ · sgn(S (k)), S (k) > 0,

α(k) ≥ α0 + αδ · sgn(S (k)), S (k) < 0,

β(k) ≤ β0 + βδ · sgn(S (k)), S (k) > 0, (D.32)

β(k) ≥ β0 + βδ · sgn(S (k)), S (k) < 0.

The sign of the incremental change ∆S (k) in (D.31) is opposite to the sign of S (k), inde-
pendent of the values of α(k) and β(k). Hence, the reaching condition is satisfied. This
completes the proof. ¤

Remark 5: In order to satisfy the crossing condition of the quasi-sliding mode, the sign of
S (k) changes back and forth as it crosses S (k) = 0 . Assume that the following condition

144
satisfies:
sgn(S (k)) = −sgn(S (k + 1)) = sgn(S (k + 2)). (D.33)

From (D.31), we have

S (k + 2) = (1 − T s q)S (k + 1) − εT s · sgn(S (k + 1))

+α(k + 1) + β(k + 1) − (α0 + β0 )

−(αδ + αδ ) · sgn(S (k + 1))

= sgn(S (k))[(1 − T s q)2 |S (k)| + T s qεT s + T s q(αδ + βδ )]

+(1 − T s q)[α(k) + β(k) − α0 − β0 ]

+[α(k + 1) + β(k + 1) − α0 − β0 ]. (D.34)

In order to get the quasi-sliding mode sgn(k + 2) = sgn(k), the following condition has to
be satisfied:
T s q[εT s + (αδ + βδ )] > (2 − T s q)(αδ + βδ ). (D.35)

Also, for assurance of the QSMB, we have sgn(S (k + 1)) = −sgn(S (k)). From (27), if
assume S (k) < 0, then

S (k + 1) = (1 − T s q)S (k) + εT s + (αδ + βδ ) (D.36)

+α(k) + β(k) − α0 − β0 > 0.

Since αδ + βδ + α(k) + β(k) − α0 − β0 ≤ 0, the following inequality holds:

−εT s
(1 − T s q)S (k) > −εT s ⇒ S (k) > . (D.37)
1 − T sq

If it is assumed that S (k) > 0, then

εT s
S (k) < . (D.38)
1 − T sq

Now the sliding mode band can be obtained as:

εT s
|S (k)| < . (D.39)
1 − T sq

From (D.39), it can be observed that the QSMB decreases when the sampling period de-
creases.

145
D.3.3 Design procedure

The design procedure of the proposed robust DT-SMC is summarized in the following
steps:

1. From the experimental friction force data, determine the appropriate uncertainties
on the system matrix (∆A).

2. Apply the CCL algorithm to solve the LMI problem in (D.18) and (D.19), and obtain
C to construct the sliding surface as in (D.16).

3. Calculate the robust DT-SMC scheme by using (D.26).

D.4 Simulation Studies


In the simulation studies, the proposed robust DT-SMC design procedure is applied to the
EHA system. Consider the uncertainties on the nonlinear friction force coefficients by
introducing the uncertainty term ∆A. The measured friction values fall within the uncer-
tainty region spanned by the upper and lower bound curves, as was shown in Figure D.3.
According to the design procedure presented in the previous section, the CCL algo-
rithm is employed to solve the LMI problem in (D.18) and (D.19) and as a result, a feasible
solution of C is obtained as:
 
C= 1378.1 170.3 .

Then the designed sliding surface is

S (k) = Cz1e (k) + z2e (k) = 0. (D.40)

Further, based on (D.26) in Theorem 2, the robust DT-SMC is then designed as:

u(k) = −B−1
2 [C̄ Āze (k) − (1 − T s q)S (k) + εT s · sgn(S (k))

+(α0 + β0 ) + (αδ + βδ ) · sgn(S (k))], (D.41)


 
where B2 = 1, C̄ = 1378.1 170.3 1 , T s = 0.001, ε = 5, q = 100, and α0 , β0 , αδ , αδ
can be calculated from (D.27), (D.28) and (D.29) at each step of the simulation.

146
For the EHA system model with uncertainties, assume that the friction parameters in
(D.4), a1 , a2 , and a3 , will be varying between ±10% around their nominal values, re-
spectively. Applying the designed robust DT-SMC to the uncertain model, the reference
tracking performance is illustrated in Figure D.4. For the purpose of comparison, the
reference input signal is similar to that used in [Wang et al., 2008].



 5t2 , if 0 < t ≤ 0.1,





r(t) = 
 −5t2 + 2t − 0.1, if 0.1 ≤ t ≤ 0.2,





 0.1, if t ≥ 0.2.

0.025

0.02
Piston Position (m)

0.015

0.01

0.005 simulation tracking response


tracking reference signal

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure D.4: Simulation tracking response of the EHA system with uncer-
tainties using the proposed method.

Then, the SMC method proposed in [Wang et al., 2008] is applied to the same EHA
model with uncertainties and the resulting step tracking performance is shown in Fig-
ure D.5. The tracking error curves for both cases are depicted in Figure D.6. It is apparent
that the tracking performance of the proposed method in this work is better than that re-
ported in [Wang et al., 2008]. This is expected because the control design method used
in [Wang et al., 2008] did not explicitly incorporate the uncertainties into the design.
Transducer noise (white noise with upper and lower magnitude bounds ±0.004) is also
considered in the simulation study to further test the robustness performance of the pro-
posed controller. The tracking response and the tracking error are shown in Figure D.7

147
0.025
simulation
tracking response
0.02

Piston Position (m)


tracking reference
0.015

0.01

0.005

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure D.5: Simulation tracking response of the EHA system with uncer-
tainties using the method in [Wang et al., 2008].

and Figure D.8 (dotted line). It is observed that the tracking error is still within an ac-
ceptable level even considering the transducer noise. The method in [Wang et al., 2008]
was also applied to the EHA system subject to the same type of transducer noise, and the
tracking error is shown in Figure D.8 (solid line). The comparison in Figure D.8 clearly
demonstrates the improved robustness against the transducer noise by the proposed robust
DT-SMC design.

D.5 Experimental Tests

The experimental EHA system under study is shown in Figure D.9. The hardware-in-the-
loop experimental test system includes the following components: Pentium IV computer,
PCI-DAS1602/16 Analog & Digital I/O Board, Gurley LE18 Linear Encoder and the de-
signed bidirectional hydraulic circuit. Experimental tests were performed to confirm and
verify the observations obtained from simulation studies. In practical EHA experimental
tests, only the position signal could be measured. But for the controllers presented in this
work, full feedback states are required. Therefore, the extended Kalman filter [Grewal and
Amdrews, 2001] was applied to estimate the unmeasured state information in the experi-
mental tests. In this paper, the extended Kalman filter designed by a former researcher in

148
−4
x 10
2

Position Error(m)
step response error
using the proposed method
−2

−4 step response error


using the method in [15]

−6
0 0.2 0.4 0.6 0.8 1
Time (s)

Figure D.6: Simulation tracking error curves using the proposed method
and the method in [Wang et al., 2008].

the group was used to do the simulation and experimental test; the details of the Kalman
filter design can be found in [Chinniah, 2004]. The tracking response and the tracking
error for the EHA system are shown in Figure D.10 and Figure D.11, respectively.
For the purpose of comparison, the same experimental test procedure is also applied
to the EHA system based on the control and estimation method employed in [Wang et al.,
2008]; the identical EKF is employed to estimate the unmeasured states. The step re-
sponse and the tracking error are shown in Figure D.12 and Figure D.13, respectively. It
is observed that the tracking response using the proposed method can reach the steady
state tracking much faster than that using the method in [Wang et al., 2008]. For further
comparison studies, well tuned PID controller was also applied to the EHA system and
the tracking response is shown in Figure D.14. It can be easily seen that the best tuned
PID controller response needs much longer time to reach the steady state and has a larger
tracking error than the proposed method.
In summary, from the experimental results, it can be seen that:

• To quantitatively compare the positioning performance, the 2-norm of the error sig-
nals in Figure D.11 and Figure D.13 was calculated, respectively. The results are:
0.0012 for the method proposed in this paper, and 0.0014 for the method in [Wang
et al., 2008]. By applying the proposed method, the high-precision positioning per-

149
0.025

0.02

Piston Position (m)


0.015

0.01

0.005 simulation tracking response


tracking reference signal
0
0 0.2 0.4 0.6 0.8 1
Time (s)

Figure D.7: Simulation tracking response of the EHA system with model
uncertainty and transducer noise.

formance has been improved by 14.3%.

• The steady-state tracking performance is of high-precision with the tracking error


close to 0.

• There exists a small error during the transient period; indeed, a time-delay during the
transient portion is observed (indeed, it was also observed in [Wang et al., 2008]).
To try to improve this issue, a pure time-delay about 0.03 second was added into
the system model. However, once the loop was closed and the designed controller
was implemented, the system went unstable. This was also observed using a PID
controller and that used in [Wang et al., 2008]. At this point, a plausible explanation
for this issue is not available and a study has been initiated to identify possible causes
of this delay. It is suspected that the nonlinear model is missing some dynamic terms
which really need to be accounted for in the controller design. Future research will
lead to new challenges to design a controller based on the same design philosophy
which can account for these new additional dynamics.

150
−4
x 10
2

0
tracking error using

Position Error (m)


the proposed method

−2

tracking error using


−4 the method in [15]

−6

−8
0 0.2 0.4 0.6 0.8 1
Time (s)

Figure D.8: Simulation tracking error of the EHA system with model un-
certainty and transducer noise.

D.6 Conclusions
This paper has proposed a robust DT-SMC design for a nonlinear EHA system model with
uncertainties. The norm-bounded uncertainty has been introduced to account for the effect
of varying friction parameters. It is concluded that the proposed upper and lower bound-
ary curves do encompass measured experimental friction forces. Based on the Lyapunov
theory and LMI technique, the asymptotically stable sliding surface can be conveniently
designed based on the feasible solutions of the derived LMI with a constraint. Further, the
robust DT-SMC law guarantees the sliding mode reaching condition. Simulation studies
illustrate that the proposed method outperforms the SMC approach employed in [Wang
et al., 2008] for the EHA system. Experimental results on the EHA system further ver-
ify the effectiveness and the performance improvement by applying the proposed method
from the application perspective.

151
Figure D.9: Experimental setup of the EHA system.

0.025

0.02

tracking reference
Piston Position (m)

0.015

0.01

0.005 experimental
tracking response

−0.005
0 0.2 0.4 0.6 0.8 1
Time (s)

Figure D.10: Experimental tracking response of the EHA system with


model uncertainty using the proposed method.

152
−3
x 10
3

2.5

2
Position Error (m)

1.5

0.5

−0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure D.11: Experimental tracking error of the EHA system with model
uncertainty using the proposed method.

0.025

0.02
Piston Position (m)

0.015 tracking
reference

0.01
experimental
tracking response

0.005

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure D.12: Experimental tracking response of the EHA system with


model uncertainty using the method in [Wang et al., 2008].

153
−3
x 10
2.5

1.5
Position Error (m)

0.5

−0.5

−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure D.13: Experimental tracking error of the EHA system with model
uncertainty using the method in [Wang et al., 2008].

0.025

0.02
Piston Position (m)

0.015
Kp = 600, Ki = 0, Kd = 3 × 10−3

0.01 Kp = 1000, Ki = 5 × 10−4 , Kd = 3 × 10−3

Kp = 2500, Ki = 2 × 10−5 , Kd = 0
0.005

0
0 0.2 0.4 0.6 0.8 1
Time (s)

Figure D.14: Experimental tracking response of the EHA system using PID
controller.

154
R
Armstrong-Helouvry, B., Dupont, P., de Wit, C.C., 1994. A survey of models, analysis
tools and compensation methods for the control of machines with friction. Automatica
30, 1083–1138.

Bashash, S., Jalili, N., 2009. Robust adaptive control of coupled parallel piezo-flexural
nanopositioning stages. IEEE/ASME Transactions on Mechatronics 14, 11–20.

Boyd, S., Ghaoui, L.E., Feron, E., Balakrishnan, V., 1994. Linear matrix inequalities in
system and control theory. Society for Industrial and Applied Mathematics, Philadel-
phia.

Chinniah, Y.A., 2004. Fault Detection In the Electrohydraulic Actuator Using Extended
Kalman Filter. Ph.D. thesis. University of Saskatchewan.

Choi, H.H., 1998. An explicit formula of linear sliding surfaces for a class of uncertain
dynamic systems with mismatched uncertainties. Automatica 34, 1015–1020.

Choi, H.H., 1999. On the existence of linear sliding surfaces for a class of uncertain
dynamic systems with mismatched uncertainties. Automatica 35, 1707–1715.

Gao, W., Wang, Y., Homaifa, A., 1995. Discrete-time variable structure control systems.
IEEE Transactions on Industrial Electronics 42, 117–122.

Grewal, M.S., Amdrews, A.P., 2001. Kalman filtering theory and practice using MAT-
LAB. JOHN WILEY & SONS.

Habibi, S.R., Goldenberg, A.A., 2000. Design of a new high-performance electrohydraulic


actuator. IEEE/ASME Transactions on Mechatronics 5, 158–165.

Habibi, S.R., Sigh, G., 2000. Derivation of design requirements of optimization of a high
performance hydrostatic actuation system. International Journal of Fluid Power 2, 11–
27.

Janardhanan, S., Bandyopadhyay, B., 2006. Discrete sliding mode control of systems
with unmatched uncertainty using multirate output feedback. IEEE Transactions on
Automatic Control 51, 1030–1035.

Laurent, E.G., Francois, O., Mustapha, A., 1997. A cone complementarity linearization
algorithm for static output feedback and related problems. IEEE Transactions on Auto-
matic Control 42, 1171–1176.

Lee, C., Salapaka, S.M., 2009. Fast robust nanopositioning-a linear-matrix-inequalities-


based optimal control approach. IEEE/ASME Transactions on Mechatronics 14, 414–
422.

155
Marton, L., Lantos, B., 2009. Control of mechanical systems with stribeck friction and
backlash. Systems & Control Letters 58, 141–147.

Merritt, H.E., 1967. Hydraulic Control Systems. JOHN WILEY & SONS.

Misawa, E.A., 1997. Discrete-time sliding mode control: the linear case. ASME Journal
of Dynamic Systems, Measurement, and Control 119, 819–821.

Owen, W.S., Croft, E.A., 2003. The reduction of stick-slip friction in hydraulic actuators.
IEEE/ASME Transactions on Mechatronics 8, 362–371.

Rotea, M.A., Khargonekar, P.P., 1989. Stabilization of uncertain systems with norm
bounded uncertainty – a control Lyapunov function approach. SIAM Journal on Control
and Optimization 27, 1462–1476.

Tafazoli, S., Silva, C.W., DLawrence, P., 1998. Tracking control of an electrohydraulic
manipulator in the presence of friction. IEEE Transactions on Control Systems Tech-
nology 6, 401–411.

Tan, U.X., Latt, W.T., Shee, C.Y., Riviere, C.N., Ang, W.T., 2009. Feedforward
controller of ill-conditioned hysteresis using singularity-free prandtl-ishlinskii model.
IEEE/ASME Transactions on Mechatronics 14, 598–605.

Tang, C.Y., Misawa, E.A., 2000. Discrete variable structure control for linear multivariable
systems. Journal of Dynamic Systems, Measurement, and Control 122, 783–792.

Utkin, V., Guldner, J., Shi, J., 1999. Sliding Mode Control in Electromechanical Systems.
CRC Press.

Utkin, V., Yang, K.D., 1978. Methods for constructing discontinuity planes in mutidimen-
sional variable structure systems. Automation and Remote Control 39, 1466–1470.

Utkin, V.I., 1993. Sliding mode control design principles and applications to electric
drives. IEEE Transactions on Industrial Electronics 40, 23–36.

Wang, S., Habibi, S., Burton, R., Sampson, E., 2008. Sliding mode control for an elec-
trohydraulic actuator system with discontinuous non-linear friction. Proceedings of the
Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering
222, 799–815.

Wit, C.C., Olsson, H., Astrom, K.J., Lischinsky, P., 1995. A new model for control of
systems with friction. IEEE Transactions on Automatic Control 40, 419–425.

Xie, L., 1996. Output feedback H∞ control of systems with parameter uncertainty. Inter-
national Journal of Control 63, 741–750.

Yan, M., Mehr, A.S., Shi, Y., 2008. Discrete-time sliding-mode control of uncertain sys-
tems with time-varying delays via descriptor approach. Journal of Control Science and
Engineering 28.

156
Yan, M., Shi, Y., 2008. Robust discrete-time sliding mode control for uncertain systems
with time-varying state delay. IET Control Theory Application 2, 662–674.

Yi, J., Chang, S., Shen, Y., 2009. Disturbance-observer-based hysteresis compensation for
piezoelectric actuators. IEEE/ASME Transactions on Mechatronics 14, 456–464.

Zeng, H., Sepehri, N., 2008. Tracking control of hydraulic actuators using a lugre friction
model compensation. ASME Journal of Dynamic Systems, Measurement, and Control
130, 014502–1–014502–7.

157

You might also like