Вы находитесь на странице: 1из 219

Resolved Motion Control of Mobile

Hydraulic Cranes
by
Marc E. M unzer
Dissertation submitted to the Faculty of Engineering & Science
at Aalborg University
in partial fullment of the requirements for the degree of
Doctor of Philosophy in Electrical Engineering
Aalborg University, Denmark
Institute of Energy Technology
December, 2002
Aalborg University
Institute of Energy Technology
Pontoppidanstrde 101
DK-9220 Aalborg East
Copyright c Marc E. M unzer, 2003
Printed in Denmark by Arco Grask A/S
First print, February 2003
Second print, August 2004
ISBN 87-89179-44-7
Typeset in L
A
T
E
X2

.
Preface
This thesis is submitted to the Faculty of Engineering and Science at Aalborg University
in partial fullment of the requirements for the Ph.D. degree in Electrical Engineering.
The research has been conducted at the Department of Electrical Energy Conversion
which is part of the Institute of Energy Technology (IET), Aalborg University.
The project has been followed by my supervisor Peder Pedersen. I would like to thank
him for his supervision, patience, help and suggestions.
I would like to thank Sauer-Danfoss for supplying the funding to make this project pos-
sible and Hjbjerg Maskin Fabrik for donating the test crane used in the experimental
tests.
I would also like to thank the technical sta for helping with the experimental setup.
Especially Walter Neumayr for his patience in answering all my electrical questions
and Jan Christiansen for his help with the frequent changes necessary in the hydraulic
setups.
Also thanks to my family who gave me great support during the three years I was away
from home. A special thanks to my girlfriend Christina, for giving me the motivation
that I needed to actually nish this work.
Aalborg, February 2003
Marc E. M unzer
marc@munzer.net
iii
Abstract
This thesis deals with resolved motion control of mobile hydraulic manipulators. With
currently available hydraulic manipulators the operator is required to independently
control the individual joints. With a resolved motion control scheme the operator
controls the tool centre directly and a computer coordinates the motion of the individual
joints. Resolved motion control is therefore also often called tool centre control, end
eector control, or coordinated motion control.
The type of manipulator discussed in this thesis has 4 degrees of freedom: a rotation
about the base, a shoulder joint, an elbow joint and a telescopic extension section.
This type of manipulator typically has a range of greater than four metres and can lift
loads of up to hundreds of kilograms. This type of manipulator is commonly referred
to as a Large Scale Manipulator in order to dierentiate it from the typical indus-
trial robots. The term mobile refers to the transportable nature of the manipulator,
dierentiating it from industrial robots which are usually xed to a certain location.
The control structure used in this thesis is distributed joint control. A central con-
troller coordinates the motion of the individual joints based on the operators input
and the measured state of the manipulator. The central controller sends a reference
joint velocity to the individual joint controllers mounted at each joint. The individual
joint controllers can be programmed with the ow characteristic of the valve, the ge-
ometry of the joint, and a velocity control algorithm. This control structure supports
the current trend in industry of mobile hydraulic proportional valves with embedded
micro-controllers. The joint control algorithm can be programmed in the actual valve.
Communication between the individual joint controllers and the coordinating controller
can occur over a bus system, such as CAN bus.
The joint controller presented in this thesis uses a single angular position sensor for
feedback control of the joint velocity. A further sensor is added to add articial damping
to the joint motion. Two sensor strategies are implemented, the rst based on a pressure
sensor, the second based on a strain gauge. Since the motion of one joint aects the
other joints in the system, an analytic stability analysis was performed taking into
account the interaction between the joints.
The central controller presented in this thesis implements ow sharing, deection com-
pensation, gain scheduling, and redundancy control. Flow sharing means that the
reference speed is decreased if the ow demands are higher than the ow limits of the
v
vi Abstract
pump. Deection compensation estimates the deection in the telescopic boom sec-
tion and compensates for it. Gain scheduling is implemented to take care of the large
parameter changes which occur in the crane. Redundancy control determines how the
extra degree of freedom of the manipulator system is utilized.
A core part of this thesis is the experimental implementation of all the ideas on a real
mobile hydraulic crane.
Dansk resum e
Denne afhandling omhandler kranspids styring (tool center control) af en mobil hy-
draulisk manipulator. Med de nuvrende hydrauliske manipulatorer, er det ndvendigt
for operatren at koordinere styreinput, idet hver af de enkelte bevgelser (friheds-
grader) styres ved et manuelt styreinput. Med kranspids styring, kan operatren styre
kranspidsen direkte, idet en computer koordinerer de enkelte frihedsgrader, saledes at
den nskede bevgelse af kranspidsen opnas.
Denne type af manipulatorer, som beskrevet afhandlingen, har 4 frihedsgrader, hvo-
raf de tre rotoriske frihedsgrader er et drejeled ved fundamentet, et skulderled og et
albueled. Den sidste translatoriske frihedsgrad er udformet som et udskudssystem.
Et eksempel pa denne type af manipulatorer er mobilkraner, som typisk har en rkke-
vidde pa mere end re meter og kan lfte en last pa ere hundrede kilogram. Denne
klasse af manipulator kaldes ofte teleopererede manipulatorer for at skabe kontrast til
de typiske industrielle robotter. Termen mobil beskriver at manipulatoren er ytbar og
ikke kseret til et bestemt sted. Dette indebrer, at manipulatoren typisk har sin egen
mobile kraftforsyning.
I denne afhandling er behandlet et distribueret styresystem, idet hver enkelt frihedsgrad
udstyres med sit eget styresystem. En central styrecomputer srger for koordinerende
input til de enkelte styresystemer. De enkelte styresystemer er programmeret under
anvendelse af a priori viden om ventilkarakteristik, kinematiske bevgelses relationer
og en hastighedsstyrings algoritme.
Denne Styrestruktur understttes af en ny udvikling i industrien af ventiler med ind-
bygget mikrocomputer, som indebrer, at styrealgoritmen kan programmeres direkte
i ventilen. Kommunikation mellem de enkelte styresystemer og den centrale styrecom-
puter kan ske via et bus system, som for eksempel CAN bus systemet.
Styresytemet for de enkelte frihedsgrader, som er udviklet i denne afhandling, udnytter
tilbagekobling af en positionsmaling for at styre hastigheden pa de enkelte mekaniske
elementer. En ekstra sensor er endvidere indfrt for at skabe aktiv dmpning af de
enkelte mekaniske bevgeelementer. To sensor strategier er undersgt med henblik pa
at opna den nskede dmpeeekt. Den frste strategi er baseret pa en trykmaling,
mens den anden baseret pa en strain gauge maling.
vii
viii Dansk resume
For at behandle den gensidige pavirkning af de enkelte styresystemer er en analytisk
stabilitetsanalyse udviklet, som tager hensyn til koblingen mellem de enkelte elementer.
De centrale algoritmer, som er udviklet i denne afhandling, inkluderer en lsning af
ere problemer som volumenstrms deling (Flow Sharing), udbjnings-kompensation,
forstrkningsmanipulering (gain scheduling), og redundans styring.
Volumenstrms deling betyder at operatrens input er reduceres, hvis pumpen ikke kan
levere en volumenstrm, som kan tilfredsstille det samlede behov. Udbjnings kompen-
sation estimerer udbjningen af udskud systemet ved kranspidsen og kompenserer for
denne. Forstrkningsmanipulering ndrer forstrkningen i de enkelte reguleringssljfer
som funktion af mobilkranens tilstands parametre. Redundans styringen bestemmer,
hvorledes den overtallige frihedsgrad i systemet udnyttes.
En vsentlig del af afhandlingen er eksperimentel implementering og verikation af
teoretisk behandlede lsningsforslag pa en rigtig mobil hydraulisk kran.
List of abbreviations
ADC Analog to Digital Conversion
CAN Controller Area Network
DAC Digital to Analog Conversion
DSP Digital Signal Processor
FFT Fast Fourier Transform
GUI Graphical User Interface
HMF Hjbjerg Maskin Fabrik
LHV Load Holding Valve
LS Load Sensing
LSM Large Scale Manipulator
OCV Over Centre Valve
PID Proportional Integral Derivative
SMISMO Separate Meter In Separate Meter Out
ix
Contents
Preface iii
Abstract v
Dansk resum e vii
List of abbreviations ix
Contents xi
List of Figures xiii
1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Resolved Motion Basics . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Contributions of this Work . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Restrictions in the Scope of this Work . . . . . . . . . . . . . . . . . 13
1.6 Description of the Individual Chapters . . . . . . . . . . . . . . . . 14
2 Model Development 17
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.1 Previous Modelling Work . . . . . . . . . . . . . . . . . . . . 19
2.1.2 Introduction to the System . . . . . . . . . . . . . . . . . . . 20
2.1.3 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Hydraulic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Valve Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 Positive Joint Velocity . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Negative Joint Velocity . . . . . . . . . . . . . . . . . . . . . 25
2.2.4 The Extension Beam . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Mechanical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 Extension Beam Length . . . . . . . . . . . . . . . . . . . . . 28
xi
xii Contents
2.3.2 Modelling the Flexibility . . . . . . . . . . . . . . . . . . . . . 29
2.3.3 Dynamic Model of the Structure . . . . . . . . . . . . . . . . 34
2.3.4 Connecting the Linear Cylinder and the Angular Beam . 37
2.4 The Complete Model . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5 Verication of the Model . . . . . . . . . . . . . . . . . . . . . . . . 40
3 Joint Controller Design 43
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Structure of the Software Layers . . . . . . . . . . . . . . . . . . . . 47
3.3 High Level Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3.1 Open Loop Joint Velocity Control Scheme . . . . . . . . . 48
3.3.2 Closed Loop Joint Velocity Control Scheme . . . . . . . . . 50
3.4 Low Level Control Layer . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4.1 Increased Damping via Pressure Feedback . . . . . . . . . 52
3.4.2 Increased Damping via Strain Gauge Feedback . . . . . . 62
3.4.3 Gain Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . 63
4 Stability Analysis 67
4.1 Model of the Complete System . . . . . . . . . . . . . . . . . . . . 69
4.2 Kharitonovs Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Stability Analysis Results . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.1 Gain Scheduling Rules Used . . . . . . . . . . . . . . . . . . 74
4.3.2 Global Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5 Deection Compensation 79
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Deection Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3.1 Deection Estimation . . . . . . . . . . . . . . . . . . . . . . 86
5.3.2 Deection Compensation . . . . . . . . . . . . . . . . . . . 87
5.4 Conclusions on Deection Compensation . . . . . . . . . . . . . . 88
6 Handling Redundancy 89
6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.1 Previous Research . . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Basic Redundancy Control Concept . . . . . . . . . . . . . . . . . 92
6.2.1 Extra Utilization of the Redundancy . . . . . . . . . . . . . . 93
6.3 Different Strategies Tested . . . . . . . . . . . . . . . . . . . . . . . . 93
6.3.1 Minimum Norm in Joint Space . . . . . . . . . . . . . . . . . 93
6.3.2 Minimum Norm in Actuator Space . . . . . . . . . . . . . . 94
6.3.3 Minimum Force . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3.4 Elbow and Extension Only Strategy . . . . . . . . . . . . . . 95
Contents xiii
6.4 Comparing Different Strategies . . . . . . . . . . . . . . . . . . . . 96
6.4.1 Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4.2 Graphical Verication . . . . . . . . . . . . . . . . . . . . . . 99
6.4.3 Test Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.5 Manual Control of the Shoulder Joint . . . . . . . . . . . . . . . . . 104
6.6 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.6.1 Automatic Joint Limit Avoidance Strategy . . . . . . . . . . 105
6.6.2 Manual Shoulder Joint Control . . . . . . . . . . . . . . . . . 106
6.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7 Experimental Tests 109
7.1 Operator Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.1 Trajectory of the Tool Centre . . . . . . . . . . . . . . . . . . 112
7.1.2 Velocity of the Tool Centre . . . . . . . . . . . . . . . . . . . 113
7.1.3 Bandwidth of the Operator/Crane Combination . . . . . . 114
7.1.4 Conclusion on Operator Tests . . . . . . . . . . . . . . . . . 114
7.2 Resolved Motion Control . . . . . . . . . . . . . . . . . . . . . . . . 115
7.2.1 Rectangular Motion . . . . . . . . . . . . . . . . . . . . . . . 115
7.2.2 Triangular Motion . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.2.3 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . 123
7.3 Accuracy as a Function of Speed . . . . . . . . . . . . . . . . . . . 123
7.4 Gain Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
8 Conclusion 127
8.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.2 Evaluation of the Final Strategy . . . . . . . . . . . . . . . . . . . . 131
8.3 Future Research Work . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Appendices 133
A Introduction to Current Hydraulic Systems 135
A.1 Applications for Mobile Manipulators . . . . . . . . . . . . . . . . . 137
A.2 A Typical Mobile Crane . . . . . . . . . . . . . . . . . . . . . . . . . 139
A.3 System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
A.3.1 Cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
A.3.2 Valves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
A.3.3 Flow and Pressure Control Valves . . . . . . . . . . . . . . . 144
A.3.4 Load Holding Valves . . . . . . . . . . . . . . . . . . . . . . . 147
A.3.5 Power Supply . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
A.4 Problems with Current Industrial Systems . . . . . . . . . . . . . . . 151
xiv Contents
A.4.1 Efciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.4.2 Flow Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . 152
A.4.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
A.4.4 Environmental . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
A.4.5 Lack of Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . 153
A.5 Differences between Mobile and Stationary Hydraulics . . . . . . 154
B Hydraulic Decoupling 157
B.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
B.2 Pump Side Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
B.3 Pressure Compensator . . . . . . . . . . . . . . . . . . . . . . . . . . 163
B.4 Experimental Verication . . . . . . . . . . . . . . . . . . . . . . . . 165
C Angular Joint Actuation 167
C.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
C.2 Hydraulic Actuator Control Topologies . . . . . . . . . . . . . . . . 170
C.3 Different Options for Valve-Based Control . . . . . . . . . . . . . . 173
C.3.1 Type of Orice . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
C.3.2 Valve Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . 174
C.3.3 Control Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
D Review of Different Controllers 177
D.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
D.2 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
E Flow Sharing 183
E.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
E.1.1 Cylinder Velocity . . . . . . . . . . . . . . . . . . . . . . . . . 186
F Description of Laboratory Facilities 187
F.1 HMF Crane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
F.2 Development System . . . . . . . . . . . . . . . . . . . . . . . . . . 190
F.3 Tool Centre Position Measurement System . . . . . . . . . . . . . . 191
Bibliography 193
List of Figures
1.1 Typical mobile hydraulic crane. . . . . . . . . . . . . . . . . . . . . 4
1.2 A system with a Smart valve. . . . . . . . . . . . . . . . . . . . . . 5
1.3 Typical manipulator structure with the describing variables. . . . 6
1.4 Independent joint control scheme. . . . . . . . . . . . . . . . . . . 6
1.5 Cylindrical coordinate system. . . . . . . . . . . . . . . . . . . . . . 7
1.6 Same tool centre position - two different manipulator congura-
tions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7 Experimental test system used in this thesis. . . . . . . . . . . . . . 13
2.1 Complete Model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Reference and response for the spool position. . . . . . . . . . . . 23
2.3 Ramp and step response of the valve. . . . . . . . . . . . . . . . . 23
2.4 Flow response of the valve. . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Hydraulic set-up when raising a load. . . . . . . . . . . . . . . . . . 25
2.6 Block diagram for positive joint velocity. . . . . . . . . . . . . . . . 25
2.7 Hydraulic set-up when lowering a load. . . . . . . . . . . . . . . . 26
2.8 Block diagram of the negative joint velocity case. . . . . . . . . . 26
2.9 Hydraulic schematic of the extension beam. . . . . . . . . . . . . 27
2.10 Extension beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.11 Cylinders locked into position. . . . . . . . . . . . . . . . . . . . . . 30
2.12 Contribution of base deection. . . . . . . . . . . . . . . . . . . . . 30
2.13 Test set-up for the deection tests. . . . . . . . . . . . . . . . . . . . 31
2.14 Deection versus torque at 3m extension. . . . . . . . . . . . . . . 32
2.15 Deection of a simple beam. . . . . . . . . . . . . . . . . . . . . . . 32
2.16 Stiffness as a function of length. . . . . . . . . . . . . . . . . . . . . 33
2.17 Structure of the mechanical model. . . . . . . . . . . . . . . . . . 35
2.18 Block diagram of the coupled system. . . . . . . . . . . . . . . . . 36
2.19 Geometry of the elbow joint. . . . . . . . . . . . . . . . . . . . . . . 37
2.20 Block diagram of the complete model. . . . . . . . . . . . . . . . 38
2.21 Linear equivalent block diagram of the complete model. . . . . 39
2.22 FFT of crane motion for impulse input. . . . . . . . . . . . . . . . . . 41
2.23 Comparison of calculated and measured natural frequency. . . 42
xv
xvi List of Figures
2.24 Frequency estimation error. . . . . . . . . . . . . . . . . . . . . . . . 42
3.1 Proposed structure of distributed joint control. . . . . . . . . . . . 46
3.2 Actual implementation of distributed joint control.. . . . . . . . . 46
3.3 Structure of the software in the valve controller. . . . . . . . . . . 47
3.4 Open loop joint velocity control scheme. . . . . . . . . . . . . . . 48
3.5 Valve ow characteristic. . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6 Hysteresis on dead-band compensation. . . . . . . . . . . . . . . 49
3.7 Basic form of the controller. . . . . . . . . . . . . . . . . . . . . . . . 50
3.8 Modied controller for constant loop gain. . . . . . . . . . . . . . 51
3.9 Position based feedback controller. . . . . . . . . . . . . . . . . . . 51
3.10 Two different forms of pressure feedback. . . . . . . . . . . . . . . 53
3.11 Effect of pressure and PDot feedback. . . . . . . . . . . . . . . . . 54
3.12 Comparison of pressure and PDot feedback. . . . . . . . . . . . . 54
3.13 Schematic of the positive velocity case. . . . . . . . . . . . . . . . 55
3.14 Block diagram of the positive velocity case. . . . . . . . . . . . . . 55
3.15 Stability boundary of the valve loop. . . . . . . . . . . . . . . . . . 57
3.16 Diagram of a typical overcentre valve system. . . . . . . . . . . . 57
3.17 Block diagram of typical over-centre valve system. . . . . . . . . 58
3.18 Contour plots of the coefcients of the Routh Array. . . . . . . . . 59
3.19 Experimental verication of pressure feedback. . . . . . . . . . . 60
3.20 Increasing the lter time constant of the pressure feedback. . . . 60
3.21 Effect of rate-limited motion of valve. . . . . . . . . . . . . . . . . . 61
3.22 Pressure feedback effect in negative velocity direction. . . . . . 62
3.23 Straingauge feedback. . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.24 Direct strain feedback. . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.25 Gain scheduling rules for the shoulder joint. . . . . . . . . . . . . . 65
3.26 Gain scheduling rules for the elbow joint. . . . . . . . . . . . . . . 66
4.1 Overall block diagram of coupled system. . . . . . . . . . . . . . . 69
4.2 Linear equivalent overall block diagram of coupled system. . . . 70
4.3 Standard coupling networks. . . . . . . . . . . . . . . . . . . . . . . 71
4.4 Roots of the Kharitonov polynomials. . . . . . . . . . . . . . . . . . 73
4.5 Kharitonov image set. . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.6 Gain-scheduling rules for the damping controllers. . . . . . . . . . 75
4.7 Root plot of the transfer function F11. . . . . . . . . . . . . . . . . . 76
4.8 Root plot of the transfer function F12. . . . . . . . . . . . . . . . . . 76
4.9 Root plot of the transfer function F21. . . . . . . . . . . . . . . . . . 77
5.1 Denition of the virtual angle. . . . . . . . . . . . . . . . . . . . . . 81
5.2 Deection compensation. . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Feed forward deection compensation. . . . . . . . . . . . . . . . 82
List of Figures xvii
5.4 Location of the strain gauge and pressure transducers. . . . . . . 83
5.5 Estimation of torque . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.6 Deection versus measured strain. . . . . . . . . . . . . . . . . . . . 85
5.7 Deection slope versus length of extension beam. . . . . . . . . . 86
5.8 Deection estimation test for 100Kg load. . . . . . . . . . . . . . . 87
5.9 Deection estimation test for 200Kg load. . . . . . . . . . . . . . . 87
5.10 Deection compensation results. . . . . . . . . . . . . . . . . . . . 88
6.1 Joint limit avoidance strategy. . . . . . . . . . . . . . . . . . . . . . 96
6.2 Model used to compare the redundancy strategies. . . . . . . . 97
6.3 Graphical model of the crane. . . . . . . . . . . . . . . . . . . . . . 99
6.4 Tool centre test trajectories. . . . . . . . . . . . . . . . . . . . . . . . 100
6.5 Trajectory Dbackwards with minimumnormstrategy in joint space.101
6.6 Trajectory Dbackwards with minimumnormstrategy in actuator
space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.7 Trajectory D backwards with minimum force strategy. . . . . . . . 102
6.8 Trajectory D backwards for elbow and extension strategy. . . . . 102
6.9 Total energy used by the different strategies. . . . . . . . . . . . . 103
6.10 Automatic joint limit avoidance when moving towards the base. 105
6.11 Automatic joint limit avoidance when moving away from the
base. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.12 Manual shoulder joint control - raising. . . . . . . . . . . . . . . . . 107
6.13 Manual shoulder joint control - Lowering . . . . . . . . . . . . . . . 107
7.1 Denition of the operator tests. . . . . . . . . . . . . . . . . . . . . . 112
7.2 Tool centre trajectory for 400kg rigidly connected load. . . . . . 112
7.3 Straightline motion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.4 Tool centre velocity. . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
7.5 FFT of tool centre motion. . . . . . . . . . . . . . . . . . . . . . . . . 114
7.6 Rectangular path following with no load. . . . . . . . . . . . . . . 116
7.7 Rectangular path following with a 200kg load. . . . . . . . . . . . 117
7.8 Rectangular path following with a 400kg load. . . . . . . . . . . . 118
7.9 Triangular path following with no load. . . . . . . . . . . . . . . . . 120
7.10 Triangular path following with a 200kg load. . . . . . . . . . . . . . 121
7.11 Triangular path following with a 400kg load. . . . . . . . . . . . . . 122
7.12 Straightline motion at different velocities. . . . . . . . . . . . . . . 123
7.13 Controller gains tuned with beam retracted. . . . . . . . . . . . . 124
7.14 Controller gains tuned with beam extended. . . . . . . . . . . . . 124
7.15 Testing the gain scheduling algorithm. . . . . . . . . . . . . . . . . 125
A.1 Mobile manipulator as a crane. . . . . . . . . . . . . . . . . . . . . 137
A.2 Mobile manipulator with grab bucket. . . . . . . . . . . . . . . . . 138
xviii List of Figures
A.3 Mobile manipulator with personnel basket. . . . . . . . . . . . . . 138
A.4 Mobile manipulator mounted on a forest machine. . . . . . . . . 138
A.5 Schematic diagram of laboratory crane. . . . . . . . . . . . . . . 139
A.6 Mobile hydraulic crane installed in laboratory. . . . . . . . . . . . 139
A.7 Typical current actuator subsystem. . . . . . . . . . . . . . . . . . . 140
A.8 Typical hydraulic differential cylinder. . . . . . . . . . . . . . . . . . 141
A.9 Four different quadrants of operation . . . . . . . . . . . . . . . . . 142
A.10 Typical three position spool valve. . . . . . . . . . . . . . . . . . . . 142
A.11 Typical pressure vs ow characteristic for a spool valve. . . . . . . 143
A.12 Typical ow/pressure characteristic for a pressure control valve. . 144
A.13 Typical ow/pressure characteristic for a ow control valve. . . . 145
A.14 Pressure control valve . . . . . . . . . . . . . . . . . . . . . . . . . . 146
A.15 Flow control valve . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
A.16 Cross section of a typical load holding valve. . . . . . . . . . . . . 148
A.17 Load sensing with a pressure regulator. . . . . . . . . . . . . . . . . 150
A.18 Open center valves. . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
A.19 Variable displacement system. . . . . . . . . . . . . . . . . . . . . . 151
A.20 Efciency of different power supplies. . . . . . . . . . . . . . . . . . 152
A.21 Figures for different applications . . . . . . . . . . . . . . . . . . . . 154
B.1 Feedback action of load sensing system and pressure compen-
sator. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
B.2 Pump side module. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
B.3 Block diagram of pump side module . . . . . . . . . . . . . . . . . 161
B.4 Block diagram of the pump side module reorganized . . . . . . . 161
B.5 Time constant of the large volume systemas the operating points
are varied. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
B.6 Time constant of the small volume systemas the operating points
are varied. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
B.7 Pressure compensator. . . . . . . . . . . . . . . . . . . . . . . . . . . 163
B.8 Block diagram of pressure compensator. . . . . . . . . . . . . . . . 164
B.9 Reduced block diagram of pressure compensator. . . . . . . . . 164
B.10 Time constant of pressure compensator system at different op-
erating points. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
B.11 Comparison of the pump and load pressures. . . . . . . . . . . . 165
C.1 Rotary hydraulic actuator. . . . . . . . . . . . . . . . . . . . . . . . 170
C.2 Separate meter in separate meter out control (SMISMO). . . . . . 171
C.3 Pump based control. . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
C.4 Simple seat valve. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
F.1 Schematic diagram of laboratory crane. . . . . . . . . . . . . . . 189
List of Figures xix
F.2 Mobile hydraulic crane installed in laboratory. . . . . . . . . . . . 189
F.3 Diagram of the development system setup. . . . . . . . . . . . . . 190
F.4 Tool centre position measurement system. . . . . . . . . . . . . . . 191
xx List of Figures
Chapter 1
Introduction
Chapter Contents
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Resolved Motion Basics . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Equations of Motion . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Contributions of this Work . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Restrictions in the Scope of this Work . . . . . . . . . . . . . . . . 13
1.6 Description of the Individual Chapters . . . . . . . . . . . . . . . 14
1
1.1. Overview 3
1.1 Overview
This thesis deals with resolved motion control of mobile hydraulic manipulators. Re-
solved motion control can also be called tip control, tool centre control, end eector
control, or coordinated motion control. With a resolved motion control scheme the op-
erator can input the desired motion of the tip, or tool centre, of the manipulator to a
controller which then coordinates the motion of the individual joints of the manipulator
so that the tool centre of the manipulator follows the desired trajectory.
This is in contrast to the current method where the operator controls the motion of the
joints independently. In the current method of control most tasks require simultaneous
motion of two to four actuators. Since the actuators move the joints, the operator is
forced to think in joint coordinates rather than world coordinates.
A simple analogy is a human arm. A human arm has a shoulder and an elbow joint.
If a human being were to control his or her arm as current mobile manipulators are
controlled, the human would have to independently control both the shoulder and the
elbow joint so that the hand reaches a desired position. Instead of this inecient mode
of control, the human brain has developed its own version of resolved motion control,
so that a small part of the brain automatically controls the shoulder and elbow joint
depending on a hand position reference.
The current control method of manipulator control is a complex task which requires a
fair amount of practise and training time before an operator is considered skilled. Even
after an operator is considered skilled, the current control method is still stressful for the
operator because he or she constantly has to convert between the world coordinates
and the joint coordinates. With resolved motion control the manipulator control
becomes more logical and therefore simpler for the operator. An unskilled operator is
then able to accomplish complex tasks without a large amount of time spent training.
In addition, tasks can be performed with less stress.
An interesting paper on this topic is written by Oshina (1997) which presents an in-
dustry perspective of the future of mobile hydraulic machines. As the author points
out, the trend for mobile manipulators is to become multi-purpose machines designed
to become an extension of the human operator. Therefore these machines should be
able to be operated in a manner which is logical for the human operator. Joint control
is not logical and disrupts the operators logical thought progression. Resolved motion
control on the other hand is logical and gives a seamless interface between the operator
and the machine.
The type of manipulator discussed in this thesis is a commonly available mobile hy-
draulic manipulator and is shown in g. 1.1. This type of manipulator has 4 degrees
of freedom: a rotation about the base, a shoulder joint, an elbow joint and an
extension section. In this thesis the tip of the crane will be referred to as the tool
centre. The tool centre is usually called the end eector in the robotics literature. The
term mobile refers to the fact that this type of manipulator is not xed to a certain
location, but can move from one location to another. This is in contrast to industrial
4 Chapter 1. Introduction
robots which are usually xed to a certain location.
Figure 1.1 Typical mobile hydraulic crane.
The terms shoulder and elbow are used to refer to the manipulator joints since they
are easy to understand. Other literature often refers to the shoulder joint as the main
joint and the elbow joint as the jib joint.
The reader who is not very familiar with hydraulics or mobile manipulators is referred
to Appendix A which presents an introduction to hydraulics and mobile manipulators.
A key requirement for doing resolved motion control is accurate control of the motion
of the individual joints. In this thesis a distributed joint control approach was taken.
In a distributed joint control scheme, each joint has a local joint controller. A central
controller coordinates the overall motion of the manipulator by sending each joint
controller a reference dependent on an operators input. The individual joint controllers
then control the joints so that the joints follow the reference specied by the central
controller.
This approach was taken due to the emergence of new mobile hydraulic valves where
a micro controller is directly embedded into the valve. Systems with the new Smart
valves will have a structure as shown in g. 1.2. Communication between dierent
valves will occur over a bus system such as CAN bus. In addition it will be possible to
connect external inputs, such as sensors, via the micro-controllers built-in Analog to
1.2. Resolved Motion Basics 5
Digital (AD) converters.
Figure 1.2 A system with a Smart valve.
The micro-controller will have two layers of software: a low level layer and a high level
layer. The low level layer is programmed by the valve manufacturer and will include the
spool position control. At some point in the future when the spool position controller
dynamics of the valves are increased then ow control or pressure control options can
be embedded in the low level layer as well.
The high level layer will be programmed by the system manufacturer. This could, for
example, be a low pass lter to eliminate high frequency reference signals input by the
operator or a hysteresis lter to eliminate unnecessary crossing of the deadband area.
In this thesis, an angular velocity controller for a manipulator joint is programmed
into the high level layer. A joint velocity reference is given to the valve and based on
feedback from the sensors mounted at the joint, the joint velocity is controlled.
1.2 Resolved Motion Basics
The manipulator used in this thesis is one of the most common types of mobile manipu-
lators. The manipulator was shown in g. 1.1. A schematic diagram of the manipulator
showing the variables used to describe the position of the crane is shown in g. 1.3.
The shoulder joint angle is described by , the elbow joint angle by and the linear
extension section length by L
2
. The shoulder beam length, L
1
is xed. Due to space
restrictions in the laboratory, the rotation degree of freedom was locked. The tool
centre of the manipulator therefore moves in a vertical plane. The vertical motion is
described by y and the horizontal motion is described by x.
In this thesis, the operator species the desired velocity of the tool centre. This is in
contrast to traditional robotics, where the input is typically the desired position. Based
on the tool centre velocity reference given by the operator and the measured orientation
6 Chapter 1. Introduction
of the crane, the resolved motion controller calculates the angular velocity necessary
at the individual joints in order for the manipulator to follow the operators desired
trajectory. The function of a resolved motion rate controller is shown in g. 1.4.
Figure 1.3 Typical manipulator structure with the describing variables.
Figure 1.4 Independent joint control scheme.
1.2. Resolved Motion Basics 7
Coordinate System
The coordinate system used most frequently in robotics is the xed Cartesian coordi-
nate system. This system is based on the traditional x, y, z coordinate system and an x,
y, z velocity reference is therefore input to the controller. In an industrial environment
where the robot is in a xed reference frame, a Cartesian coordinate system makes the
most sense.
However on many types of mobile machines the operator sits on a seat connected to the
base of the arm. As the arm rotates about the base of the manipulator, the operator
rotates with the arm. With many other mobile manipulators, the operator is not phys-
ically connected to the machine and controls the manipulator from a remote position
via a remote control device. In these two cases, the best choice for coordinate system is
a cylindrical coordinate system. In this case, the operator controls the horizontal and
vertical component of the end eector in a vertical plane and then rotates the plane
by rotating the arm about the base. Such a cylindrical coordinate system is shown in
g. 1.5.
Figure 1.5 Cylindrical coordinate system.
A cylindrical system is easier for an operator to understand when the operators ref-
erence frame is constantly changing. By looking at the manipulators position, the
operator can easily see which way the manipulator will move when the tool centre is
given a reference signal to move outwards or inwards. In a Cartesian system, the op-
erator would have to keep track of the base coordinate frame in order to know which
way the tool centre will move when given an x or y velocity reference.
The type of coordinate system used in this thesis is therefore a cylindrical coordinate
system. In this thesis the radius is described by x and the height by y.
8 Chapter 1. Introduction
1.2.1 Equations of Motion
In order to implement resolved motion control, the forward kinematics equations
need to be found. The forward kinematics equations describe the relationship between
the joint angles and the tool centre position. Standard techniques for nding the
forward kinematics equations are available in any introductory robotics text book such
as Craig (1986). Since the typical mobile manipulator is a simple manipulator, a
simplied geometric approach was used to nd the forward kinematics equations.
The forward kinematics equations for the case of the typical mobile manipulator, as
shown in g. 1.3, are given by equations 1.1 and 1.2.
x = L
1
cos + L
2
cos ( + ) (1.1)
y = L
1
sin + L
2
sin ( + ) (1.2)
Since, it is not the position that is of interest in this thesis, but the velocity, equations
1.1 and 1.2 are dierentiated giving 1.3 and 1.4.
x = L
1
sin

L
2
sin ( + )(

+

) +

L
2
cos ( + ) (1.3)
y = L
1
cos

+ L
2
cos ( + )(

+

) +

L
2
sin ( + ) (1.4)
Writing equations 1.3 and 1.4 in matrix form gives:
_
x
y
_
=
_
L
1
sin L
2
sin ( + ) L
2
sin ( + ) cos ( + )
L
1
cos + L
2
cos ( + ) L
2
cos ( + ) sin ( + )
_
_

L
2
_

_
(1.5)
The matrix in equation 1.5 is called the Jacobian of the manipulator. As shown in 1.5
the Jacobian is dependant on the orientation of the manipulator. A simplied form of
1.5 is given in 1.6.
v = J(q) q (1.6)
J(q) is the Jacobian of the system, q is the vector of the joint angles and extension
length vector, v is the tool centre velocity vector, and q is the velocity vector of the
joints and extension section.
Since it is desired to nd the required joint velocities for a given tool centre velocity
reference, equation 1.6 is solved algebraically for the joint velocities q giving equation
1.2. Resolved Motion Basics 9
1.7.
q = J
1
(q)v (1.7)
This simple equation is the foundation for the implementation of resolved motion con-
trol. Given a desired tool centre velocity v, a required joint velocity reference q is
generated based on the current manipulator joint angles. However, implementation of
this equation is not that simple. Besides the obvious need for accurate joint control,
the exibility and the redundancy in the structure need to be handled in order to
successfully implement resolved motion control.
Flexibility in the Mechanical Structure
On most mobile manipulators, the demand for high power to weight ratios means
that the beams of the manipulator are made as light as possible. This results in the
beams being quite exible. In particular the extension beam on a mobile crane is very
exible. This means that when loads are applied to the tool centre, the beam bends
which results in a deection of the tool centre. As an example, a 400Kg load applied
to the tool centre of the crane used in the experimental tests of this thesis will produce
a deection of the tool centre of up to 0.5m. The exibility also decreases the natural
frequency of the system which results in oscillations in the tool centre motion.
Chapter 2 studies the nature of the exibility in more detail, chapter 3 presents control
techniques to increase the damping in the system and chapter 5 presents a scheme to
compensate for the deection.
Redundancy
Since the manipulator is a three degree of freedom manipulator operating in a two
degree of freedom space, the tool centre can be positioned at a certain location with
an innite number of dierent manipulator joint angles. This is illustrated in g. 1.6
where the same tool centre position is achieved with two, of many possible, dierent
joint angle congurations.
Redundancy adds extra complexity to the system but is usually considered a positive
feature of the manipulator because it introduces the possibility for optimization. One
example could be to optimize cylinder pressures. Since a certain tool centre position
can be arrived at with dierent joint angle congurations, a controller which optimizes
the cylinder pressures can choose the joint angles which results in the lowest cylinder
pressures.
Chapter 6 discusses redundancy in more detail and presents some dierent options to
10 Chapter 1. Introduction
handle the redundancy.
Figure 1.6 Same tool centre position - two dierent manipulator congurations.
1.3 Previous Work
Resolved motion control was rst applied to stationary robots. Stationary robots, or
industrial robots as they are often called, are typically powered with electric motors
at the joints, are mounted on a rigid base, and have relatively sti links. The idea
of resolved motion control was to make it possible to control a robot in Cartesian
coordinates instead of joint coordinates. This is generally described as control in World
Space instead of control in Joint Space. This means that instead of specifying the joint
angles to put the robot into a certain position, an operator species world coordinates
and a resolved motion controller converts those coordinates into the required joint
angles.
The basics of resolved motion control were presented in section 1.2. All introductory
textbooks on basic robotics explain the theory in more details. A good example is Craig
(1986). Using the geometry of the manipulator, it is possible to relate the positions and
velocities of the joints to the position and velocity of the tool centre. By manipulating
these relationships, it is possible to nd the joint motions which give a desired tool
centre motion.
The rst research into resolved motion control of the type of manipulators discussed
in this thesis, often called large scale manipulators, started in the late 1980s. One
of the rst references in the area was a patent granted to MacMillan Bloedel Limited
titled Articulated Arm Control, awarded to Lawrence and Ross (1991). This patent
presented a scheme for controlling the rate of the end eector using a computer. The
1.3. Previous Work 11
strategy was meant to apply mainly to structures like an excavator with a boom and
stick function. In the strategy, the operator determined an end eector velocity via
a joystick. The input signal was converted into a position requirement by integrating
the signal. The x, y, z position requirement was converted into a joint requirement
via a geometric solution to the inverse problem. No mention of how the joints were
controlled is made in the patent.
The inventors of this particular patent worked at the University of British Columbia
(UBC) and they continued their work, both through the PhD work of Nariman Sepehri
(Sepehri, 1990) and through cooperation between UBC, the University of Manitoba (U
of M) and various other research groups. One reference which is of particular interest
is written by Lawrence et al. (1993). This paper presents the results of some human
factors experiments which compare the performance of log loader operators using both
a traditional manual controller and a resolved motion controller. Experiments were
based on practical implementation on a real excavator. As expected, results showed
that new operators were more ecient with the resolved motion controllers than they
were with traditional controllers. More surprisingly however, results also showed that
within 10 hours of training on the new controller it was possible for a skilled operator
with up to 25 years of experience on the manual controller, to match his performance
with the manual controller. These were very promising results. The controller used in
this study was the resolved mode controller presented by Lawrence and Ross (1991)
where the individual joint position controllers were PD controllers with tuneable gains
for both P and D terms.
Sepehri presented some more details of the controllers in Sepehri et al. (1994). The
strategy used is based on a feed-forward load compensating scheme. The feed-forward
component is based on measurements of the hydraulic line pressures and an online
model of the crane. The use of a model allows minimization of cross couplings between
links, limitations in power, etc. The feed forward component is corrected via a closed
loop PD controller. The work presented was based on an excavator chassis, but used
as a log loader. Further work at UBC and U of M continued with work being done
on more advanced control strategies like force feedback (Parker et al., 1993) and fuzzy
logic control (Sepehri and Lawrence, 1998).
Meanwhile, other researchers were also working in the same area. Two papers (Krus
and Palmberg, 1992) (Krus and Gunnarsson, 1993) from Linkoping University in Swe-
den presented some results on experiments on vector control of mobile cranes using
mobile electro-hydraulic valves. Niemela and Virvalo (1994) present Fuzzy logic as-
sisted control of a hydraulic crane. In this article, a fuzzy logic control strategy is
presented which controls one degree of freedom of the crane automatically so that it
frees the operator from controlling two joysticks simultaneously. Touminen and Vir-
valo (1993) presents a strategy based on distributed control of the resolved motion
controller. In the work of Sato et al. (1993) a master slave controller for an excavator
system is presented.
The early strategies were all relatively simple. However as computer power increased so
did the complexity of the control algorithms used. Mattila and Virvalo (1997a) present
12 Chapter 1. Introduction
a Computed Force Control scheme for mobile cranes which uses an online model to
calculate the forces on the individual joints. The controller uses the force information
to control the pressures in the individual cylinders to drive the crane. Mattila and
Virvalo (2000) present a continuation of this work with more emphasis on separate
meter in separate meter out valves. Lee and Chang (2001) presents an algorithm based
on sliding mode control to control the straight-line motion of an excavator.
The Ph.D. of Elfving (1997) presents a distributed control strategy which could decou-
ple the velocity and force in the actuators so that more energy ecient control could
be realized. The strategy made use of separate meter in separate meter out valves so
that the two sides of the cylinder could be controlled independently.
Researchers have also started to apply the research work in traditional robotics in
compensating for exible links to hydraulic cranes. An example from Finland is by
Rouvinen and Handroos (1997) who applied neural networks to solve static deection
of a mobile hydraulic log loader crane. Work from Sweden by Nilsson et al. (1999)
uses a range sensing camera mounted in the gripper of a large scale manipulator to
dampen out vibrations and reduce the deection due to exibility in the beams of the
manipulator.
The most complete work found by the author on resolved motion control was a PhD
written by Linjama (1998) from Tampere University of Technology. This thesis dis-
cusses modelling of a two DOF manipulator and resolved motion control of the manip-
ulator using proportional valves. An eective controller was developed and shown to
be robust to parameter variations.
However, currently there are not many commercial implementations of resolved motion
schemes. One example is a manipulator used for tree pruning around power-lines and
is described by Goldenberg et al. (1995).
1.4 Contributions of this Work
This thesis builds on the previous work by presenting and implementing a resolved
motion control scheme for a manipulator with a telescopic boom section. Most of the
previous works in the area of resolved motion control of mobile hydraulic manipulators
have dealt with two degree of freedom structures where the beam lengths were xed.
This thesis also introduces a gain scheduling algorithm which is based on an online
estimation of the natural frequency of the system. The gain scheduler helps to maintain
performance and stability over the large change in the inertia of the system as the
telescopic beam changes length and the crane works with dierent loads.
This thesis also introduces a deection compensation algorithm to compensate for the
exibility in the telescopic beam. In addition, two dierent schemes are presented to
handle the redundancy of the manipulator.
1.5. Restrictions in the Scope of this Work 13
Furthermore, an independent joint controller is presented which can be programmed
directly into a mobile hydraulic valve with an embedded micro controller. The con-
troller uses a single angular position sensor to do the control. In order to introduce
damping into the system two methods are tested, one based on pressure feedback and
one based on strain feedback. The presented joint controller can be used together with
a resolved motion scheme or as a stand-alone component in a traditional manipulator
control scheme.
Since the manipulator structure leads to mechanical coupling between the cylinders, a
stability analysis based on a parameter space approach was used to show the stability
of the system of the coupled system over the entire parameter space. Kharitonovs
theorem was used to reduce the size of the parameter space.
The nal solution is implemented on the mobile crane shown in g. 1.7. No changes
were made to the factory installed hydraulic system on the manipulator. The laboratory
set-up is described in more detail in appendix F.
Figure 1.7 Experimental test system used in this thesis.
1.5 Restrictions in the Scope of this Work
This work was limited to valve based hydraulic joint control options. The eld of joint
control has so many dierent options that it is dicult to cover them all completely. A
14 Chapter 1. Introduction
brief review of some dierent options is presented in Appendix C. The decision to base
the thesis on valve-based solutions was based on the emergence of commercially avail-
able Smart valves. This leads to a solution which will be industrially implementable
in the near future.
In order to present a stability proof, a number of assumptions needed to be made. The
biggest assumption is that the load sensing system has no eect on the stability. This
is the similar to many papers in the state of the art, (Linjama and Virvalo, 1999),
(Mattila and Virvalo, 2000), (Honegger and Corke, 2001). Arguments for verifying this
assumption for the experimental system which has a load sensing system based on a
xed displacement pump and pressure regulator are presented in appendix B. However,
no mention of LS systems based on variable displacement pumps was made.
Another assumption made was that the controller knows the mass of the load. This
could be implemented via a strain gauge at the tool centre. In this thesis, the load
mass was manually entered by the operator via the Graphical User Interface (GUI).
A limitation was also made with regards to the length sensor of the extension beam.
The laboratory crane is equipped with a length measurement on each section of the
telescopic boom. However, it was decided that a strategy which is dependant on four
sensors is not economically feasible so the four independent measurements were summed
together to give a single measurement of the length of the overall beam. This assump-
tion introduces errors into the frequency estimation and the deection estimation. How-
ever, it is assumed that the increased accuracy with four sensors does not justify the
added cost.
1.6 Description of the Individual Chapters
Chapter 2 - Model Development
This chapter discusses the development of a linear model used in the design of the
controller. The model takes into account the coupling between the dierent cylinders.
A key feature is modelling the exibility in the extension beam as a simple torsion
spring. Static and dynamic verications of the model are presented.
Chapter 3 - Controller Design
This chapter presents the controller used at a single joint of the distributed joint con-
troller. A structure for the software is proposed which separates the control into a low
level layer and a high level layer. The low level layer contains the position control of
the spool and some optional articial damping units. The high level layer is modiable
by the system developer. Two methods of implementing the damping units are also
presented, one based on pressure feedback and one based on straingauge feedback.
Chapter 4 - Stability
This chapter analyses the stability of the entire system when independent joint control
is used. A parameter space approach was used. Kharitonovs theorem was used to
1.6. Description of the Individual Chapters 15
reduce the computational demands.
Chapter 5 - Deection Compensation
This chapter introduces a simple method of doing deection compensation based on a
strain gauge measurement.
Chapter 6 - Handling Redundancy
This chapter discusses how to handle the redundancy. It includes a review of the
previous work in the area, simulation of a number of dierent strategies, comparison
of the energy consumption of dierent strategies, and experimental verication of the
strategy.
Chapter 7 - Experimental Tests
This chapter presents the results of tests performed on an HMF 680 mobile crane.
Chapter 8 - Conclusion
The main contributions of this thesis are reviewed and ideas for future work are pre-
sented.
Appendix A - Introdution to Mobile Manipulators and Current Hydraulic Systems
This appendix contains an introduction to current mobile hydraulic systems, what the
problems are, and how the dierent components work. If the reader is new to mobile
hydraulics this is a good section to read.
Appendix B - Hydraulic Decoupling Assumption
This appendix presents the arguments for decoupling the hydraulic cylinders in the
model of the hydraulic system.
Appendix C - Options for Joint Control
This appendix contains a discussion of dierent joint control topologies. The focus is
not on the controllers used, rather it is on the physical actuation of the joints, i.e. the
actual actuators used and the actual control elements used.
Appendix D - Review of Dierent Controllers
This appendix contains a brief review of some of the controllers typically used in the
control of hydraulic position servos. The focus is more on the software side than the
hardware side.
Appendix E - Flow Sharing
This appendix contains details of the ow-sharing algorithm used.
Appendix F - Laboratory Facilities
This appendix contains a description of the laboratory facilities used for this Ph.D.
work. The crane, the software development environment, and the tool centre position
measurement system are discussed.
Chapter 2
Model Development
Chapter Contents
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.1 Previous Modelling Work . . . . . . . . . . . . . . . . . . . 19
2.1.2 Introduction to the System . . . . . . . . . . . . . . . . . . 20
2.1.3 Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Hydraulic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 Valve Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.2 Positive Joint Velocity . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Negative Joint Velocity . . . . . . . . . . . . . . . . . . . . 25
2.2.4 The Extension Beam . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Mechanical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3.1 Extension Beam Length . . . . . . . . . . . . . . . . . . . . 28
2.3.2 Modelling the Flexibility . . . . . . . . . . . . . . . . . . . . 29
2.3.3 Dynamic Model of the Structure . . . . . . . . . . . . . . . 34
2.3.4 Connecting the Linear Cylinder and the Angular Beam 37
2.4 The Complete Model . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5 Verication of the Model . . . . . . . . . . . . . . . . . . . . . . . 40
17
2.1. Overview 19
2.1 Overview
This chapter develops a model of a standard manipulator system. In this thesis the
model should fulll two criteria. The rst criteria, is that the model can be used
to quantitatively analyse the stability of the system. The second criteria is that the
model can be programmed in a micro-controller to estimate the natural frequency of
the system in real-time. The estimate can then be used to gain schedule a controller.
The approach taken in this chapter was to develop a linear model by analysing the
dominant dynamics of the system. The dynamics included in the linear model are the
mechanical coupling between the shoulder and elbow cylinder, the dynamics of the
over-centre valves, the dynamics of the proportional valves, the pressure dynamics in
the cylinders, and the exibility of the structure. The main non-linear feature of the
crane system is the large variations in the system parameters such as mechanical inertia,
mechanical stiness, ow gain, geometry, and hydraulic capacitance. The parameters
are slowly changing so the dynamics of the change are not as important as the actual
parameter values. Therefore the linear model is adapted to include all these eects via
variable model parameters.
This is in contrast to the traditional approach to modelling where a non linear model
of the system is developed and veried against the real system. If a linear model of
the system is desired, the non-linear model is linearized about an operating point.
However, the type of hydraulic manipulators which are discussed by this thesis are
very complex and a model which accurately captures all the dynamics becomes very
large and complex. There have been a number of student projects at The Institute
of Energy Technology, Aalborg University, which have developed realistic models of
hydraulic manipulators. A common feature of these models is high complexity and
high computational demands.
The traditional modelling approach does not fulll the two criteria presented above.
Firstly, it is impossible to make a quantitative analysis of stability on a complex model.
The only option is to simulate the system at many dierent operating points and
thereby verify stability. However this is a time consuming process given the computa-
tional demands of the model. In addition, there is a risk that a critical operating point
might be missed from the simulation study. A linear model developed by linearizing
the non-linear model will result in a complex linear model which will suer from the
same problems. Secondly, if the model should be implemented in real time on a mi-
cro controller, a very powerful micro controller would be needed. It is unlikely that
a micro controller powerful enough to run a realistic model and cheap enough to be
implemented industrially will be available in the next few years.
2.1.1 Previous Modelling Work
There have been a number of references which develop models of two degree of freedom
hydraulic cranes where the beams are assumed to be sti. Refer to Ellman et al.
20 Chapter 2. Model Development
(1996), Sepehri et al. (1990), and Krus et al. (1991). Also a number of models have
been developed which take the exibility of the beams into account. Mikkola and
Handroos (1996) used a combination of ADAMS and ANSYS to develop an oine
model of a log loader crane. Makinen et al. (1997) developed a mathematical approach
to the modelling of a single exible beam coupled with a hydraulic system. Linjama
and Virvalo (1997) used a combination of nite elements and assumed modes coupled
with Lagrange equations to simulate the dynamics of a exible crane in a non real-
time environment. Later on Linjama and Virvalo (1999) developed a linear state space
model of the exible crane so that advanced control techniques could be used.
The work which is most similar to this thesis is the thesis of Matti Linjama, from
Tampere University of Technology, Linjama (1998), which also presented a linear model
of the system which took the exibility of the beams into account.
The model presented in this chapter builds on the previous work by including both
a exible telescopic boom section and an over-centre valve on both the shoulder and
elbow joints.
2.1.2 Introduction to the System
Fig. 2.1 shows a schematic diagram of the complete system where the dierent func-
tional components are shown. The system is quite complex with many interacting
components. Motions of one cylinder aect the other cylinders through the mechanical
structure. In addition pressure uctuations in one cylinder aect the other cylinders
through the load sensing system.
Figure 2.1 Complete Model.
2.1. Overview 21
The model was split into a hydraulic model and a mechanical model. The hydraulic
model is presented in section 2.2 and included the eects of the control valves, the
over centre valves, and the cylinders. The mechanical model is presented in section 2.3
and included the dynamic eects of moving the mechanical structure. The connection
between the two models was made via geometric transformations.
The valves used in the model were hydraulic-mechanical ow control valves. The
specic valves used on the manipulator in the laboratory were Sauer-Danfoss PVG
32 valves with 25L/min ow capacity. The numerical examples refer to these valves,
however the model can be adapted to any ow control valve.
The load sensing system discussed in this thesis is a xed displacement pump with a
hydraulic-mechanical pressure regulator. This type of Load Sensing system is charac-
terized by a fast response time. If a load sensing system with variable displacement
pump was to be used the model would have to be changed in order to take into account
the slower dynamics.
The telescopic boom is modelled as a single beam whose length is controlled by a single
cylinder. The consequences of this assumption are discussed in more detail in section
2.2.4 and section 2.3.1.
2.1.3 Assumptions
The main assumption is that the three cylinders are decoupled hydraulically. This is
based on the observation that the LS system and the pressure compensators have much
faster dynamics than the rest of the system. Therefore the assumption is made that the
pressure drop across the main spools is constant. This means that the three cylinders
can be decoupled from a hydraulic point of view. This assumption is discussed in in
more detail in appendix B.
The assumption that the load sensing system is much faster than the rest of the system
means that the dynamics of the LS system can be removed from the model without
signicantly aecting the results. This is the typical approach taken in most of the
papers on the subject of resolved motion control of hydraulic manipulators. Note that
this does not imply that the LS system is stable, rather it makes the assumption that
the LS system is stable to being with.
The second assumption is that the extend and retract motion of the telescopic beam
has a low dynamic inuence on the rest of the system. This is due to the dierence in
the forces exerted by the extension cylinder and the joint cylinders. The linear motion
of the extension section acts on the inertia of a mass, whereas the angular motion acts
on the rotational inertia of a mass at the end of a long beam. The dynamic model of
the structure is therefore developed as a two degree of freedom model. The change in
length of the extension beam is included in the inertia matrix of the system.
A third assumption is to model the load as being rigidly attached at the tool centre.
22 Chapter 2. Model Development
Typical loads are swinging loads which have dynamics of their own. However, the
model verication shows that the eect of the swinging load is not very high.
The nal assumption is to describe the exibility of the extension beam as a torsion
spring lumped at the elbow joint. This assumption is discussed in section 2.3.2.
2.2 Hydraulic Model
The hydraulic system can be split into three actuator subsystems: the shoulder system,
the elbow system, and the extension system. From a functional perspective the shoulder
and elbow cylinder are the same and the same model can be used for both joints.
However, due to the direction dependant behaviour of the over centre valve, dierent
models for the lowering and raising case are presented for the shoulder and elbow
systems.
This section starts with a model of the control valve. Then the models for the shoul-
der and elbow systems are presented. Finally a discussion of the extension beam is
presented.
2.2.1 Valve Model
The valve used in this study was a Sauer-Danfoss PVG 32 mobile hydraulic proportional
valve. As is typical with valves used in mobile hydraulic systems, the valve has a large
dead-band, a rate limited spool motion, and a non linear ow gain.
The dead-band is due to both a delay in the electrical actuation and the overlap on
the spool. The describing function for a dead-band shows that the gain of the system
is reduced due to dead-band. Reduced gain has a stabilizing eect on the system.
Therefore for the linear stability analysis the dead-band is neglected. The rate-limited
motion of the spool is also neglected. The eect of the rate limiter will only be noticed
for large steps. In the typical operation of the crane there wont be any large steps to
the control valves due to the ltering in the controller.
q
V
ref
= K
f

2
n
s
2
+ 2
n
s +
2
n
(2.1)
The valve dynamics are approximated by a second order system with a variable gain
as shown in equation 2.1. V
ref
is the voltage input to the valve, q is the ow output of
the valve, K
f
is the ow gain,
n
is the natural frequency, and is the damping ratio.
The model parameters were determined by experimental analysis of the real valve. The
experiments were performed on the crane in the laboratory. Four tests were performed.
In each test the test was started with a ramp input to the valve followed by a step input
2.2. Hydraulic Model 23
of diering magnitude. The spool reference and the measured spool position for the
four tests is shown in g. 2.2
Figure 2.2 Reference and response for the spool position.
The speed of the valve can be determined from the lag between the spool position and
the ramp input reference. The damping can be seen from the step response. These two
responses are enlarged in g. 2.3(a) and g. 2.3(b).
(a) Ramp response (b) Step response
Figure 2.3 Ramp and step response of the valve.
From g. 2.3(b) it can be seen that the valve is critically damped. From g. 2.3(a),
the lag is around 80 milliseconds. From previous experiments it is known that the
electrical actuation has a delay of around 30 milliseconds, so the actual delay due to
the dynamics of the valve is around 50 milliseconds. This can be approximated by a
24 Chapter 2. Model Development
critically damped second order system with a natural frequency of around 40 rads/sec.
Figure 2.4 Flow response of the valve.
The ow output of the previous four tests is shown in g. 2.4. As can be seen from
the gure, for large step inputs, the ow response is not as well damped as the spool
position response due to extra dynamics in the pressure compensator and LS system.
Therefore the damping ratio of the model is reduced to 0.5. Even though large step
inputs are avoided in the real system, assuming a lower damping ratio will give a more
conservative model. It can also be seen from the ow response to the largest step, that
the frequency is around 50 rads/sec which matches well with the ramp tests. Remember
that a critically damped system will have a larger time lag to a ramp input than an
under-damped one.
2.2.2 Positive Joint Velocity
As was discussed in the introduction, the direction dependant behaviour of the over-
centre valve means that two models need to be developed, one for positive joint velocity
and one for negative joint velocity. The hydraulic diagram of the system for the posi-
tive velocity case is shown in g. 2.5. Note that the hydraulic cylinder for the shoulder
joint is mounted upside down on this particular crane. However, functionally there is
no eect of changing the orientation of the cylinder.
It is assumed that there is no pressure drop across the orice connecting the top of
the cylinder to the tank. That is to say, it is assumed that the top of the cylinder is
connected directly to tank. This is a conservative estimate since the pressure drop over
the orice will have a damping eect on the system.
The block diagram of the hydraulic system is shown in g. 2.6. If this model is not
familiar refer to appendix A for more information. The model is implemented so that
2.2. Hydraulic Model 25
it determines a force which is applied to the mechanical model. The mechanical model
then returns a velocity.
Figure 2.5 Hydraulic set-up when raising a load.
Figure 2.6 Block diagram for positive joint velocity.
2.2.3 Negative Joint Velocity
When lowering the load, the ow out of the cylinder is controlled by the over-centre
valve. The hydraulic diagram is shown in g. 2.7
The block diagram of this system is shown in g. 2.8. It is assumed that the over-centre
valve is much faster than the rest of the system. Therefore the internal dynamics of the
over-centre valve can be neglected. The orice opening of the overcentre spool is given
directly by the pressures acting on the spool. The pressure dependant component of
26 Chapter 2. Model Development
the linearized ow equation is neglected since it has a damping eect on the system.
Figure 2.7 Hydraulic set-up when lowering a load.
Figure 2.8 Block diagram of the negative joint velocity case.
2.2.4 The Extension Beam
The extension beam is a rather complex hydraulic unit of four cylinders connected in
parallel. The hydraulic schematic of the system is shown in g. 2.9.
The system is connected to a regenerative over-centre valve. The function of the re-
generative over-centre valve is to amplify the ow when the cylinder is extending. The
2.2. Hydraulic Model 27
ow from the rod side of the cylinder is fed back to the piston side. This increases the
speed of the cylinder without requiring more ow from the pump. The eect of this is
a higher ow gain in the extending case.
Figure 2.9 Hydraulic schematic of the extension beam.
In order to simplify the overall system model, the extension beam dynamics were
neglected from the model. This was considered acceptable because the extension beam
section has no noticeable stability issues.
The result is not being able to verify positional control of the extension beam. However,
the positional accuracy demands of the extension beam are much lower than the angular
joints. This is because the extension beam motion directly aects the tool centre. For
example 1cm motion of the extension beam results in 1cm motion of the tool centre.
The angular joints, in contrast, have a large positional gain. A 1cm motion of the
shoulder joint can have up to a 28cm eect on the tool centre. Therefore it is more
important to control the angular joints accurately than the extension section.
The only point where the extension system has an inuence is for very slow speeds
where the stiction in the cylinder and the non-linear behaviour of the over centre
valves could be noticed via limit cycles in the extension section. Limit cycles were
removed by decreasing the controller gains.
In the overall control scheme of the manipulator, the extension section position control
was not the limiting factor. However, if a system is developed with higher positional
requirements, it would be necessary to examine the control of the extension section in
more detail.
28 Chapter 2. Model Development
2.3 Mechanical Model
There are three parts to this section. The rst part discusses the eect of the telescopic
boom section. The second part presents a model of the exibility of the extension beam.
The third part models the dynamics resulting from the motion of the beams.
The exibility is examined via static tests where dierent loads are applied to the beam.
The dynamics are derived via a Lagrange-Euler approach.
2.3.1 Extension Beam Length
The extension beam is constructed as ve hexagonal sections which slide inside each
other, much like a telescope.
(a) Innermost beam sliding rst (b) Outermost beam sliding rst
Figure 2.10 Extension beam
As discussed in section 2.2.4, the hydraulic cylinders actuating the individual sections
are connected together in parallel. This means that when hydraulic pressure is applied
to the extension section, all the individual sections get an equal force. The order in
which they slide is dependant on the friction in the individual segments of the extension
section. The friction is dependant on the amount of contamination in the section, the
load being lifted, the angle of the boom, etc... This means that it is impossible to
predict which extension section will slide rst.
This has a large eect on how the beam behaves. For example if the thinnest beam
should slide rst, as in g. 2.10(b), the stiness would be low. If the main beam were
to slide rst, as in g. 2.10(a), the stiness would be much higher. This means that
there will always be an error on an estimation of the spring stiness. This should be
taken into account when using the stiness estimation. However the error is not very
large and can be accepted.
Another eect of not knowing how the beams slide relative to each other is that the
Centre of Gravity (COG) of the beam and the rotational inertia of the beam are not
able to be determined. If the little link slides rst, as in g. 2.10(b), then the COG
is closer to the joint and the rotational inertia is less. If the large link slides rst, as
2.3. Mechanical Model 29
in g. 2.10(a), then the COG moves away from the base and the rotational inertia is
increased. This will have an eect on the frequency estimation.
Having full control of the manner in which the extension beam extends would be
benecial for more precise modelling. This would require more complex hydraulics or
more complex electronics. At the current time the cost of such extra hydraulics is
not justied by the extra positional accuracy gained. However, in the future, when
manipulators will be used in tasks were positional accuracy is more important, these
types of hydraulic systems will most likely become available.
2.3.2 Modelling the Flexibility
The hypothesis proved in this section, is that the exibility in the extension beam
can be modelled as a simple angular spring lumped at the elbow joint. The models
proposed in the literature are dicult to use for controller design because they hide the
basic behaviour of the beam under complex mathematics. Linjama and Virvalo (1999)
were one of the few to reduce a complex non linear model into a linear state space form
which could be used for controller design. Note that the experimental results presented
in this section are specic to the crane in our laboratory, but it is expected that other
manipulators will behave in the same way.
The use of a simple angular spring model makes the controller design much more
intuitive. It also helps with implementing deection compensation. In order to verify
the hypothesis, tests were carried out on the crane in the laboratory.
An angular spring has a proportional relationship between torque and angular deec-
tion.

d
= KT (2.2)
where
d
is the angular deection, K is the spring constant, and T is the torque. In
order to verify the model statically, the deection versus torque curves at dierent
operating points were found.
The deection was measured by applying a controlled load to the tip of the crane
and measuring the tip deection via the tool centre measurement system described in
appendix F. In order to eliminate the eect of the hydraulic stiness of the shoulder
and elbow cylinders, both cylinders were locked into position with steel inserts. The
set-up is shown in g. 2.11.
It was however noticed that the base was also exible and contributed a large amount
to the total tool centre deection. Therefore the base angle deection was calculated
and subtracted from the total deection in order to get the actual extension beam
30 Chapter 2. Model Development
deection. The contribution of the base angle deection, , is shown in g. 2.12.
Figure 2.11 Cylinders locked into position.
Figure 2.12 Contribution of base deection.
The equation for the vertical deection of the tool centre position can be written as:
y = y
b
+ L
2
sin ( +
d
) (2.3)
This can be solved for the angular deection via:

d
= arcsin
_
y y
b
L
2
_
(2.4)
2.3. Mechanical Model 31
To apply dierent loads in a controlled manner, the crane tip was attached to the
ground with a chain hoist. Using a tension meter and the chain hoist, it was possible
to apply well-dened loads to the crane without changing physical masses. The test
setup is shown in g. 2.13.
Figure 2.13 Test set-up for the deection tests.
The torque on the beam is calculated via:
T = L
beam
F
applied
(2.5)
The angular deection due to the extension beam was plotted versus the applied torque.
The results of measuring the deection characteristic at an extension length of 3m is
shown in g. 2.14
The test was performed by rst increasing the tension in the chain hoist slowly, and
then slowly decreasing the tension. The tension was varied between 0 kN and 4 kN.
As can be seen from the gure, the relationship between the angular deection and
the torque is linear. Therefore it is assumed that the linear spring model holds for this
case. The spring constant for this length is given by the slope of the line.
Fig. 2.14 also shows that there is some hysteresis in the deection curves. This is most
likely due to the way that the extension beam slides. However the error caused by
neglecting the hysteresis is quite low and is therefore neglected from the model.
The experiment was repeated for dierent lengths of extension beam and in each case,
the torque versus deection curves were linear as in g. 2.14. It is therefore concluded
that the extension beam section can be modelled as a simple torsion spring at each
32 Chapter 2. Model Development
particular length of the extension beam section.
Figure 2.14 Deection versus torque at 3m extension.
Length Dependence of the Spring Constant
For a simple beam in deection as shown in g. 2.15, it is possible to, according to any
introductory textbook on the mechanics of materials, write the relationship of the tip
deection as:
y =
Fl
3
3EI
(2.6)
where y is the tip deection, F is the force applied, l is the length of the beam, E is
Youngs modulus of the beam and I is the cross sectional inertia.
Figure 2.15 Deection of a simple beam.
This equation can be rewritten in angular terms as:
=
Tl
3EI
(2.7)
using the relationships:
2.3. Mechanical Model 33
= sin
1
(
y
l
)

=
y
l
(2.8)
T = Fl (2.9)
Equation 2.7 can be written in the standard spring equation form as:
T = K (2.10)
Where:
K =
3EI
l
(2.11)
It is therefore expected to see a linear relationship between K and 1/l. As expected,
plotting the measured spring constants versus the inverse of the length, results in a
linear plot. The plot is shown in g. 2.16.
Figure 2.16 Stiness as a function of length.
The main reason for the errors is due to measurement errors. The fact that the beam
is quite sti means that small position measurement errors result in large stiness
errors. However another inuence is the way that the extension beam is built. This
was discussed in section 2.3.1 in more detail.
Stiffness of the Shoulder Beam
The deection of the shoulder beam is small and cant be measured very easily. How-
ever, it has a signicant eect on the frequency estimation when the extension beam
34 Chapter 2. Model Development
is fully retracted. Therefore the stiness of the beam can be estimated from the cross
sectional inertia, the length of the beam and the material properties of carbon steel.
The cross sectional inertia for a rectangular beam is given by:
I =
1
12
bh
3

1
12
(b 2w)(h 2w)
3
(2.12)
where b is the width of the beam, h is the height of the beam, and w is the wall
thickness of the beam. The spring constant is given by equation 2.11.
Deection of the Base
In most of the literature that the author is aware of, the crane is assumed to be mounted
on a sti base. (Linjama, 1998) was one of the few who did include the base stiness
in his model. This thesis neglects the eect of base exibility from further analysis. In
our laboratory set-up it is estimated that the base stiness is around ten times larger
than the manipulator stiness.
However, in real world applications the base is often not very sti and has a much
larger eect on the overall system dynamics. An example of a real world application is
a manipulator used in the forestry industry. This manipulator is typically mounted to
a large all-terrain-vehicle which has low stiness tires and is driving on spongy ground.
It would be relatively simple to add the base exibility to the model presented in this
chapter. The problem is that in the real world, the base stiness is unknown. Therefore
there is little point in including it in the current analysis. However, it would be an
interesting future research work to take into account the base stiness along with a
method to estimate the stiness online so that the controller can adapt itself.
2.3.3 Dynamic Model of the Structure
As mentioned in section 2.1.3, it is assumed that the motion of the extension beam
has little eect on the shoulder and elbow cylinders. As discussed in section 2.3.2, it is
assumed that the exibility of the extension beam can be lumped at the elbow joint.
The shoulder beam exibility is lumped at the shoulder joint. The base and support
are assumed to be sti. These assumptions give the model structure shown in g. 2.17.
In order to simplify the model, the load was assumed to be rigidly attached to the
crane. This is similar to the approach taken in (Linjama, 1998).
The angles and describe the angle between the joint and the end of the beam, the
angles
e
and
e
describe the actual joint angle, and the angles
d
and
d
describe the
deection of the joint. As shown in g. 2.17, the torques from the cylinders are assumed
to act directly at the joints. This assumption is possible because the beam between
2.3. Mechanical Model 35
the joint and the cylinder attachment point is short and relatively sti in comparison
to the rest of the exibility in the system.
Figure 2.17 Structure of the mechanical model.
Note that there is no damping in this model. Damping in hydraulic circuits is dicult
to model accurately, so it is usually approximated by viscous damping. The model
used in this thesis has no damping which is a conservative assumption. Friction in the
system will have a damping eect which will make the real system more stable than
the model. Two good references which discuss modelling of the damping in hydraulic
cylinders are (Virvalo, 99) and (Bonchis et al., 1999).
In order to develop the dynamic model of this structure, a Lagrange approach was
used. The kinetic and potential energy terms were found for each link and the Lagrange
equations solved. Since it was desired to develop a linear model, the gravity terms could
be neglected from the analysis. Therefore the strain energy due to the deections were
the only source of potential energy in the system. The energy of the two links and the
mass are given by equations 2.13 to 2.17.
KE
1
=
1
6
M
1
L
2
1

2
(2.13)
KE
2
=
1
2
M
2
_
(

L
1
)
2
+
1
3
L
2
2
(

+

)
2
+ L
1
L
2

(

+

) cos()
_
(2.14)
KE
L
=
1
2
M
L
_
(

L
1
)
2
+ L
2
2
(

+

)
2
+ 2L
1
L
2

(

+

) cos()
_
(2.15)
PE1 =
1
2
K
m

2
d
(2.16)
PE2 =
1
2
K
e

2
d
(2.17)
36 Chapter 2. Model Development
Lagranges equation is given by:
d
dt
_
KE
q
j
_

KE
q
j
+
PE
q
j
= Q
j
(2.18)
Where KE is the sum of all the kinetic energies, PE is the sum of all the potential ener-
gies, q
j
are the generalized coordinates, and Q
j
are the generalized forces. Since there
are three generalized coordinates, the Lagrange approach results in three equations
which can be written in standard matrix form as:
M q + Kq = Q (2.19)
where M is the mass (or inertia) matrix, K is the stiness matrix, q is the vector of
generalized angles, and Q is the vector of generalized torques. The equation for the
crane is given by:
_

_
M
11
0 M
13
0
0 0 0 0
M
31
0 M
33
0
0 0 0 0
_

_
_

d
_

_
+
_

_
0 0 0 0
0 K
m
0 0
0 0 0 0
0 0 0 K
e
_

_
_

d
_

_
=
_

_
T
1
T
1
T
2
T
2
_

_
(2.20)
Equation 2.20 can also be written in transfer function form and drawn as a block
diagram. The transfer functions are given by equations 2.21 to 2.24. The block diagram
is shown in g. 2.18.
Figure 2.18 Block diagram of the coupled system.
2.3. Mechanical Model 37
=
1
M
11
s
2
T
1

M
13
s
2
M
11
s
2
(2.21)
=
1
M
11
s
2
T
1

M
13
s
2
M
11
s
2
(2.22)

d
=
T
1
K
m
(2.23)

d
=
T
2
K
e
(2.24)
2.3.4 Connecting the Linear Cylinder and the Angular Beam
The relationship between the linear motion of the cylinder and the angular motion
of the joint is a non-linear relationship which depends on the joint angle. Also the
relationship between the force of the cylinder and the torque at the joint is dependant
on the joint angle in a non-linear way. Fig. 2.19 shows the geometry of the elbow joint.
Figure 2.19 Geometry of the elbow joint.
The relationship between the cylinder length (L
3
) and the inside joint angle (e) is given
by the cosine law:
L
2
3
= L
2
2
+ L
2
1
2L
1
L
2
cos e (2.25)
Dierentiating this equation gives the relationship between the cylinder velocity and
the angular joint velocity. The factor which relates cylinder velocity to angular velocity
38 Chapter 2. Model Development
is dened as K
g
, the geometry term.

L
3
=
L
1
L
2
L
3
sin(e)

= K
g

(2.26)
The torque and force are related by the same constant, K
g
. The torque due to the
force is given by:
T = FL
2
sin(d) (2.27)
Using the sine rule it can be rewritten as:
T = F
L
2
L
1
L
3
sin(e) = FK
g
(2.28)
2.4 The Complete Model
Combining the hydraulic model, the geometry relationships, and the mechanical model
gives the block diagram shown in g. 2.20. Note that due to the assumption that the
extension beam has low dynamic eect on the rest of the system, the system model is
a two degree-of-freedom model. Note that the subscript m refers to the main joint
and the subscript e refers to the elbow joint.
Figure 2.20 Block diagram of the complete model.
2.4. The Complete Model 39
The model can also be converted into a linear motion equivalent model by combining
the mass terms and the geometric terms. The reason for this will be seen in the chapter
on the stability analysis. The linear equivalent block diagram is shown in g. 2.21.
Figure 2.21 Linear equivalent block diagram of the complete model.
where:
M
11
q
=
M11
K
2
mg
(2.29)
M
33
q
=
M11
K
2
eg
(2.30)
M
13
q
=
M11
K
mg
K
eg
(2.31)
M
31
q
=
M11
K
mg
K
eg
(2.32)
K
eq
=
K
e
K
2
eg
(2.33)
K
mq
=
K
m
K
2
mg
(2.34)
(2.35)
40 Chapter 2. Model Development
2.5 Verication of the Model
Since the real system is non linear and the linear model is only valid at a certain
operating point it is impossible to use output curves of the model and compare them
to measurements on the real crane as a method of verifying the model. Instead, the
natural frequency of the model is calculated and compared to the measured frequency
of the real crane. This experiment is repeated for a number of dierent orientations and
dierent masses in order to check the model validity over the entire operating space of
the crane.
To calculate the natural frequency of the model, the inputs are assumed to be zero and
the hydraulic cylinders are replaced with springs whose hydraulic spring constant, K
h
,
is given by:
K
h
=
A
2
p

V
p
+
A
2
r

V
r
(2.36)
Where the subscripts p and r refer to the piston and rod sides of the cylinder respec-
tively. V is the volume, is the bulk modulus of the oil, and A is the cross sectional
area.
Going back to equation 2.20, the input torques can be replaced by the torque due to
the hydraulic spring constant and the angular deection at the joint.
_
M
11q
M
13q
M
31q
M
33q
__

_
+
_
0 0
0 0
__

_
=
_
K
mh
(
d
)
K
eh
(
d
)
_
(2.37)
Since
d
is a function of and
d
is a function of , equation 2.39 can be rewritten as
a mass and a stiness matrix as:
_
M
11q
M
13q
M
31q
M
33q
__

_
+
_
(K
1
mh
+ K
1
mq
)
1
0
0 (K
1
eh
+ K
1
eq
)
1
__

_
=
_
0
0
_
(2.38)
The natural frequencies can then be calculated via:
_

1

2
_
=

_
_
(K
1
mh
+ K
1
mq
)
1
0
0 (K
1
eh
+ K
1
eq
)
1
_
_
M
11q
M
13q
M
31q
M
33q
_ (2.39)
2.5. Verication of the Model 41
The two frequencies give the frequencies of the two modes of the system. The lowest
frequency has the most inuence on the system dynamics and can be measured directly
from the motion of the real crane. To measure the frequency of the system, the crane
was excited via an impulse input. The frequency was then calculated via a Fast Fourier
Transform (FFT) analysis of a reading of a strain gauge mounted on the extension
beam. A sample FFT for a case where the crane is stretched out roughly horizontally
with a load mass of 100Kg and an extension beam length of 3m is shown in g. 2.22.
Note that since the damping in the system is so low there is very little dierence
between the natural frequency and the damped frequency.
Figure 2.22 FFT of crane motion for impulse input.
As can be seen from the FFT, there is one dominant frequency which corresponds to the
rst mode. The second mode is almost not detectable. This changes as the extension
beam retracts. Since the extension beam, when it is fully extended is quite exible,
its spring constant dominates the dynamics. As the extension beam is retracted, the
stiness increases which means that the stiness of the other links in the system as
well as the stiness of the base play a more important role. This can be seen in
g. 2.23 and g. 2.24 since the frequency error increases with decreasing extension beam
length. Fig. 2.23 shows the estimated and the measured frequency at dierent extension
beam lengths. Fig. 2.24 shows the dierence between the frequency estimation and
measurement as a function of length.
As can be seen the model is quite accurate. At short extension beam lengths the
calculated frequency is up to 20% higher than the measured frequency. This is due to
the assumption that the base is sti. By including the eects of the base stiness the
frequency at short extension beam lengths is decreased. However, the error is assumed
to be low enough to neglect it.
At large extension lengths the estimated frequency is lower than the measured fre-
quency. This is due to the non-uniform cross section of the extension beam. The model
42 Chapter 2. Model Development
assumes that the extension beam has a uniform cross sectional area which means that
the COG should be at the centre of the beam. However, the actual beam becomes
thinner the farther out it goes. Therefore the COG is not at the centre of the beam,
rather it is closer to the base. This means that the rotational inertia is less than the
model predicts and hence the frequency of the real system is higher than the model.
Figure 2.23 Comparison of calculated and measured natural frequency.
Figure 2.24 Frequency estimation error.
Chapter 3
Joint Controller Design
Chapter Contents
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Structure of the Software Layers . . . . . . . . . . . . . . . . . . . 47
3.3 High Level Interface . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3.1 Open Loop Joint Velocity Control Scheme . . . . . . . . 48
3.3.2 Closed Loop Joint Velocity Control Scheme . . . . . . . . 50
3.4 Low Level Control Layer . . . . . . . . . . . . . . . . . . . . . . . . 52
3.4.1 Increased Damping via Pressure Feedback . . . . . . . . 52
3.4.2 Increased Damping via Strain Gauge Feedback . . . . . 62
3.4.3 Gain Scheduling . . . . . . . . . . . . . . . . . . . . . . . . 63
43
3.1. Overview 45
3.1 Overview
As was mentioned in the introduction, a distributed joint control approach is used in
this thesis. A distributed joint control approach means that each joint has its own local
velocity controller. A central controller gives velocity references to each of the joint
controllers. The core of each joint controller is a mobile hydraulic proportional valve
with an embedded micro controller.
Joint velocity control on mobile hydraulic manipulators both in the past and up to the
present time, has almost always been performed with a hydraulic-mechanical based
control system. In most systems, a hydraulic pressure regulator ensures that the pres-
sure drop across the main spool remains constant. The ow out of the valve, and to
a large extent the velocity of the cylinder, is then proportional to the opening of the
main spool and independent of the load. This type of control system is described in
more detail in appendix A.
With the decreasing cost of electronics, valve manufacturers started to embed electron-
ics into the valves. Newer mobile hydraulic control valves usually have the option of
electronic actuation of the main spool. However, due to power limitations and cost fac-
tors, the dynamics of the main spool actuation are not so high. Therefore ow control
is still done via hydraulic mechanical compensators. A good example of this type of
valve is the successful Sauer-Danfoss PVG 32 series.
The electronic actuation of the main spool is typically controlled via analog electronic
components. However, some new hydraulic proportional valves are now using embed-
ded micro-controllers to control the main spool actuation. The new micro-controllers
are powerful and have extra resources which are not being used by the main spool
position controller. Therefore a layer of high level control can be added by the system
manufacturer which can customize the function of the valve.
In this thesis it is assumed that it is possible to purchase a valve similar to the Sauer-
Danfoss PVG32 valve with the addition of an embedded micro controller. It is assumed
that the micro controller has an embedded bus interface, such as a CAN bus, and two
inputs for analog to digital conversion of sensor readings. The proposed structure for
the overall control system is shown in g. 3.1.
However, since mobile proportional valves with built in micro controllers were not
readily available at the time of writing, the implementation of the controllers in this
thesis was done on a single central controller. However, the joint controller code was
kept separate from the central controller code and can therefore easily be moved to
individual micro-controllers once they become available. The actual implementation is
shown in g. 3.2. Due to the speed of CAN bus systems, the proposed system will not
46 Chapter 3. Joint Controller Design
be noticeably dierent than the implemented system.
Figure 3.1 Proposed structure of distributed joint control.
This chapter presents the implementation of the joint control part of the system. The
software in the micro controller is split into two layers, a low level layer which would
be programmed by the valve manufacturer at the factory and a high level layer which
could be programmed by the system manufacturer in the eld. The rst part of this
chapter discusses the structure of the software in more detail. The second part presents
the high level layer in more detail and shows how joint velocity control for a mobile
manipulator is implemented. The third part of this chapter discusses the low level
layer programmed by the manufacturer. The nal part discusses gain scheduling the
controller and a tuning procedure for the system.
Figure 3.2 Actual implementation of distributed joint control..
3.2. Structure of the Software Layers 47
3.2 Structure of the Software Layers
The software on the micro controller is split into three layers, the rst is a low level
layer which is programmed by the valve manufacturer before the valve leaves the fac-
tory. The second is a user programmable layer programmed by the purchaser of the
valve. Communication between the two layers is kept to a minimum and occurs via
the interface layer. The proposed structure is shown in g. 3.3.
Figure 3.3 Structure of the software in the valve controller.
The low level layer contains the Analog to Digital Conversion (ADC) routines, the spool
position controller, and an articial damping unit. The interface layer contains the ow
characteristics of the spool, the gains of the damping unit, and four registers. The rst
of the four registers is used as an input and contains the spool position reference. The
three remaining registers are outputs and contain the actual spool position and the two
converted analog values. The high level layer contains the communication program plus
a user programmable section.
The details of the spool position controller wont be discussed here since the Sauer-
Danfoss PVG32 valve used in this thesis had a built in spool position controller. The
analog to digital routines convert the sensor signals on the two ADC ports on the valve
and put the converted signals into the two registers in the interface layer. This means
that the user program can read the converted values without having to touch the low
level.
The articial damping unit uses a force sensor of some sort (in this thesis both pressure
sensors and strain gauges were tested) to control the ow coming out of the valve so
that damping is introduced in the system. The gains of the feedback are dependant
on the system being controlled so they can be tuned by the end user via the interface
layer. The damping can be turned o by setting the gains to zero. The damping unit
is described in detail in section 3.4.
The ow characteristic is stored in the interface level and not in the low level layer since
it is benecial for the user program to have control over how the ow characteristic is
used. Especially dead-band compensation can be used in dierent ways depending on
the application.
48 Chapter 3. Joint Controller Design
The high level layer contains the code which controls how the valve behaves. When the
valve leaves the factory, the user program memory is pre-programmed with a simple
routine which catches spool position commands from the CAN bus and puts them
into the spool position register. The damping gains are set to zero at the factory.
This makes the valve appear to work just like a currently available mobile hydraulic
proportional valve with the addition of a CAN interface.
3.3 High Level Interface
The job of the high level layer is to take commands from the central controller and
convert them into a spool position reference to send to the low level layer. In this
section, the joint controller used in this thesis is implemented using the structure as
discussed in the previous section.
Two dierent controllers are presented in this section. The rst is an open loop con-
troller which uses the static ow characteristic of the valve to implement open loop
joint velocity control. The second adds an integral controller to correct for static er-
rors in the ow characteristic. The controller structure is based on the well-known
PID control structure. Due to the low bandwidth of the available proportional valves,
a control strategy was adopted which minimized the control action. A discussion of
more advanced controllers used in linear hydraulic cylinder control when faster valves
are available is presented in appendix D.
3.3.1 Open Loop Joint Velocity Control Scheme
A typical high-level layer that might be programmed is an open loop joint velocity
control scheme. This is possible because the valve used is a ow control valve where
the pressure drop over the main spool is maintained constant.
The angular joint velocity reference is converted into a linear velocity reference to
the cylinder via the geometry of the crane as presented in section 2.3.4. The linear
velocity reference is converted into a ow reference via the area of the cylinder. The
ow reference is converted into a voltage reference to the spool position controller. The
block diagram of this controller is shown in g. 3.4.
Figure 3.4 Open loop joint velocity control scheme.
3.3. High Level Interface 49
The geometry of the joint and the area of the cylinder would be programmed by the
manipulator manufacturer. The ow characteristic is directly related to the orice
area of the spool. The orice area characteristic of the spool is known by the valve
manufacturer and could therefore be programmed directly when the spool is inserted
into the valve. In the future it may be possible to insert a ram chip into the spool
during manufacturing with the spool prole.
In this thesis the orice area characteristic was not known so the ow characteristic
was measured by applying dierent voltages to the valve and measuring the ow. The
result is shown in g. 3.5. A third order polynomial was tted to the data.
Figure 3.5 Valve ow characteristic.
One problem with this system is that chattering in the spool position can occur if the
reference signal swings through zero due to the operators input. This is due to the
dead-band compensation combined with the slow dynamics of the spool. Therefore the
dead-band compensation can be taken out of the ow characteristic and added to the
valve reference via a hysteresis lter. This is shown in g. 3.6.
Figure 3.6 Hysteresis on dead-band compensation.
50 Chapter 3. Joint Controller Design
Therefore the ow characteristic stored in the interface layer is split into two sections.
The rst section stores the dead-band values for the positive and negative direction
cases. The second section stores the ow characteristic for the positive and negative
cases. The ow characteristic can be stored as a cubic polynomial, so only four coe-
cients are required for each direction. All ten constants are needed to fully dene the
ow characteristic.
3.3.2 Closed Loop Joint Velocity Control Scheme
The controller presented in the previous section is the simplest and is guaranteed to
be stable. However, there will always be errors in an open loop scheme. Small errors
in the velocity of one joint can have large consequences for the tool centre motion in a
resolved motion scheme. Therefore the open loop scheme is augmented with a closed
loop integral controller. This is shown in g. 3.7.
Figure 3.7 Basic form of the controller.
However, the eective gain of the controller as shown in g. 3.5 is not constant since
the joint system is non linearly dependant on the spool position and the geometry of
the crane. To make the loop gain independent of the spool position, the inverse model
of the valve and the geometry term is moved after the summing point. To keep the
valve from crossing the dead-band due to the controller action at low velocities, the
dead-band compensation is connected to the angular velocity reference. This is shown
in g. 3.8.
All resolved motion strategies require a measurement of position and therefore joint
angle sensors are already mounted on the crane. Velocity feedback either requires
dierentiation of the position signal or a dedicated velocity sensor, neither of which
are desired solutions. However, since the control action is integral control, the position
signal can be used directly instead of the velocity signal. The velocity reference is
integrated and compared with the position signal directly. The modied controller is
3.3. High Level Interface 51
shown in g. 3.8.
Figure 3.8 Modied controller for constant loop gain.
Figure 3.9 Position based feedback controller.
The controller can therefore be thought of as either an open loop velocity controller with
integral feedback or a proportional position controller with a velocity feed-forward term.
However, the correct way is the rst one, since the idea is to decrease the necessary
control action via the open loop model. Reduced control action is necessary since the
control valve is slow and has high dead-band. In this thesis, the control action was kept
low enough so that xed gains could be used on the controller. Increasing the control
gains introduced oscillations into the system. However, especially at long extension
lengths, a more accurate controller would be benecial.
It is expected that at some point in the near future, valves with higher bandwidth will
be available. At that point, it will be possible to increase the complexity of the control
to improve the performance.
52 Chapter 3. Joint Controller Design
3.4 Low Level Control Layer
There are three tasks in the proposed low level layer. The rst is the position control
of the main spool in the valve, the second is the conversion of the analog sensor signals,
and the third is the addition of articial damping.
The control of the spool position is already implemented in the Sauer-Danfoss PVG32
valve and therefore it wont be discussed in this thesis. However, it can be expected that
in the next few years, the switching frequency of the pilot stage on/o valves will be
increased due to higher performance demands. The increase in switching frequency and
the use of a micro controller will make it possible to implement new control strategies
which will improve the dynamics and accuracy of the main spool position controller.
Control strategies based on a switching valve such as sliding mode control will be
especially interesting to test.
Since most micro controllers come with built in analog to digital converters, the analog
to digital conversion task is also simple to implement. The basic concept is that the
low level layer converts the sensor signals at a xed sample rate and puts the converted
values into the registers on the interface layer. Both the low level and the high level
then have access to the converted values. The high level layer is responsible for putting
the converted values onto the CAN bus if the central controller requires access to the
sensor signals.
The damping unit uses some sort of force feedback to increase the damping in the
system. Since the gains of the feedback action are dependant on the system parameters
such as mass and stiness, the gains are put in the interface layer so that the end user
of the valve can tune the feedback action. The damping unit can be disconnected by
setting the gains in the interface layer to zero. Two types of damping are implemented
in this chapter, one based on pressure sensor feedback and one based on straingauge
feedback.
3.4.1 Increased Damping via Pressure Feedback
It is well known that pressure feedback can have benecial eects. (Edge, 1997) makes
mention of some examples from the early 1970s. There are two options, the rst is
feeding back the pressure directly, the second is feeding back the derivative of the
pressure. These two feedback ideas for a simple cylinder/Mass/Damper system are
shown in g. 3.10.
3.4. Low Level Control Layer 53
(a) Pressure feedback
(b) Pressure derivative feedback
Figure 3.10 Two dierent forms of pressure feedback.
Reducing the pressure feedback loop in g. 3.10(a), gives the transfer function shown
by equation 3.1. As can be seen the pressure feedback adds damping to the system.
v(s)
q(s)
=
A
mcs
2
+ K
p
s + A
2
(3.1)
Reducing the pressure derivative (PDot) feedback loop in g. 3.10(b) gives the transfer
shown by equation 3.2. As can be seen the PDot feedback reduces the natural frequency
in the system.
v(s)
q(s)
=
A
m(c + K
p
)s
2
+ A
2
=
A
s
2
+
A
2
m(c+K
p
)
(3.2)
The two eects can easily be seen from the bode plots of the two systems with dierent
gains. g. 3.11(a) shows the eect of increasing the pressure feedback gain. Fig. 3.11(b)
shows the eect of increasing the PDot feedback gain.
54 Chapter 3. Joint Controller Design
(a) Pressure feedback (b) PDot feedback
Figure 3.11 Eect of pressure and PDot feedback.
However, it is not possible to implement direct pressure feedback because otherwise the
static pressures would be fed back and result in static ow control errors. Therefore
the pressure feedback is high-pass ltered. The pressure feedback signal is shown by
g. 3.12(a). It is also impossible to feedback the pressure derivative directly due to
noise from the pressure sensor reading. As is well known, dierentiating noise results
in an amplication of the noise. Therefore the pressure signal is low-pass ltered before
dierentiation to remove the high frequency noise. The Pdot feedback signal is shown
in g. 3.12(b).
(a) Pressure feedback (b) PDot lter
Figure 3.12 Comparison of pressure and PDot feedback.
As can be seen from the two gures, the actual implementation of the two strategies is
identical. The only dierence is where the location of the lter time constant is chosen.
If the lter break frequency is chosen above the system frequency, then PDot feedback
is implemented. If the lter break frequency is chosen below the system frequency,
then pressure feedback is implemented.
In the next two sections, the positive velocity and negative velocity case are examined.
3.4. Low Level Control Layer 55
These two cases are dierent due to the direction dependant behaviour of the over-
centre valve. It will be shown that in the positive velocity case the lter break frequency
is chosen below the natural frequency of the system. This increases the damping due to
the pressure feedback action. In the negative velocity case, the lter break frequency
is chosen above the system natural frequency to implement PDot feedback. PDot
feedback stabilizes the over-centre valve.
Positive Velocity Case
To analyse the stability of the feedback action, the example system in g. 3.13 is
analysed.
Figure 3.13 Schematic of the positive velocity case.
The linear model for this system is shown in g. 3.14.
Figure 3.14 Block diagram of the positive velocity case.
56 Chapter 3. Joint Controller Design
This system is stable without any feedback. The rst location where the system can
become unstable is the loop with the valve dynamics and the pressure feedback. There-
fore this loop is analysed via Rouths Stability criterion. Rouths stability criterion is
discussed in more detail in any introductory text on feedback control.
Rouths Stability criterion says that the number of positive roots of a system is given
by the number of sign changes in the rst column of the Routh Array. The Routh
Array for the transfer function given by equation 3.3 is given by table 3.1.
G(s) =
b
2
s
2
+ b
1
s + b
0
a
4
s
4
+ a
3
s
3
+ a
2
s
2
+ a
1
s + a
0
(3.3)
a
0
a
2
a
4
a
1
a
3
0
b
1
b
2
0
c
1
0 0
Table 3.1 Rouths Array for the transfer function given by equation 3.3
where:
b
1
=
a
1
a
2
a
0
a
3
a
1
(3.4)
b
2
=
a
1
a
4
a
0
a
5
a
1
(3.5)
c
1
=
b
1
a
3
a
1
b
2
b
1
(3.6)
To guarantee stability, the coecients in the rst column all have to be positive. In the
system shown in g. 3.14 the only coecient that can be negative is c
1
. The coecient,
c
1
, for the above system is given by:
c
1
= 2Cz + 4Cw
n
z
2
t + 2Ct
2
w
2
n
z tK
p
w
n
(3.7)
Setting c
1
to zero results in a two dimensional curve of the stability boundary. This
curve is plotted in g. 3.15. The shaded area is the area of stability.
Each valve and cylinder combination will have a dierent gure. The gure above is
specic to the shoulder joint of the crane in our laboratory. The parameters for the
valve dynamics were identied in section 2.2.1. The natural frequency of the valve was
3.4. Low Level Control Layer 57
approximately 50 rads/sec and the damping ratio was approximately 0.5.
Figure 3.15 Stability boundary of the valve loop.
Negative Velocity Case
Due to the dynamics of the over-centre valve the negative velocity case gets more
complicated than the positive velocity case. A diagram of the system discussed in this
section is shown in g. 3.16. To make the analysis easier the positive velocity direction
is assumed to be in the down-wards direction.
Figure 3.16 Diagram of a typical overcentre valve system.
The block diagram of this system is given by g. 3.17. For more details of the over-
58 Chapter 3. Joint Controller Design
centre valve model refer to section 2.2.3.
Figure 3.17 Block diagram of typical over-centre valve system.
In contrast to the positive direction case which was stable without pressure feedback,
this system is unstable without any feedback. Therefore it is not enough to examine
the stability of the valve loop to tune the gains. The stability of the entire loop needs
to be examined. As in the positive direction case, Rouths Stability Criterion is used.
Since this system is a sixth order system, there are six rows in the Routh Array for this
system. The coecients in the rst column for the rst two rows are always positive.
However, the coecients in the rst column of the last four rows can be negative.
Therefore the coecients of the Rouths array were calculated algebraically and solved
for dierent values of gain and lter time constant. Fig. 3.18 shows the contour plots
for the three dimensional plots of the Routh array coecients. The shaded areas are
the areas where the coecients are positive. Therefore to make the system stable, it
is necessary to choose a gain and lter time constant which makes all four coecients
positive.
As can be seen the coecient B is positive for all values of gain and time constant.
The coecient C is negative for low values of gain and time constant. This makes
sense since without any feedback the system is unstable. By combining the coecients
D and E, the area to the upper left in g. 3.18(d) is the only area which makes the
system stable. This is in contrast to the positive velocity case where large values of
time constant permitted larger values of gain. It also shows that in the negative case,
it is PDot feed-back which is implemented, not P feedback.
These gures are dependent on the hydraulic capacitance, mass, valve dynamics, cylin-
der geometry, and over-centre valve geometry. The valve dynamics, the cylinder ge-
ometry, and the over-centre valve geometry are constant. However, the mass and the
3.4. Low Level Control Layer 59
hydraulic capacitance changes. Therefore choosing a gain which is stable at all the
dierent values of mass and capacitance is an iterative process.
(a) Coecient B1 (b) Coecient C1
(c) Coecient D1 (d) Coecient E1
Figure 3.18 Contour plots of the coecients of the Routh Array.
Experimental Verication
The damping eect of pressure feedback is veried experimentally on the HMF 680
crane described in appendix F. Step inputs were applied to the shoulder joint with
dierent values of feedback time constant and gain. The tool centre velocity was
measured via the tip measurement system described in appendix F. Two sets of tests
were performed, one for the raising case and one for the lowering case. In addition the
dierences between the theoretical model and the physical model are discussed.
The test for the raising case was started with a load of 200kg and with the extension
beam fully extended in a horizontal position. Fig. 3.19 shows the increased damping
60 Chapter 3. Joint Controller Design
eect of pressure feedback.
Figure 3.19 Experimental verication of pressure feedback.
However, there are many non-linear eects which cause the system to behave dierently
than one would expect. The rst is that the maximum amount of ltering is limited
due to noise eects. The model suggests that large lter time constants can be used
if the gain is simultaneously increased. However, in reality, if the signal is ltered too
much then the signal gets so small that it gets lost in the noise. This is shown in
g. 3.20. Fig. 3.20(a) shows that as the lter time constant is increased, the damping
eect decreases. Fig. 3.20(b) shows that even if the gain is increased, the damping is
not as eective as with a smaller time constant.
(a) Increasing the lter time constant (b) Increasing the gain
Figure 3.20 Increasing the lter time constant of the pressure feedback.
Another non-linear eect is the rate limiter in the valve. The rate limiter has the eect
of introducing a phase lag which is proportional to the size of the input signal. Large
input signals create large phase lags. Small input signals, have no phase lag. This puts
3.4. Low Level Control Layer 61
a limit on the maximum pressure feedback gain that can be introduced. If the phase lag
becomes too large, then small oscillations are introduced due to the feedback. Fig. 3.21
shows the eect of increasing the gain. The two signals shown in the plot are the spool
position reference and the actual spool position. At the start of the motion, there is
a large phase lag which gets smaller as time progresses and the signal size decreases.
However, if the gain is too high, then oscillations are introduced since the phase lag
never decreases. Another eect is that in the initial opening of the valve there is no
damping because the pressure build up is too fast. Therefore the rst overshoot in the
velocity cant be damped with the valve used.
(a) Low gain (b) Medium gain
(c) Good value of gain (d) High gain
Figure 3.21 Eect of rate-limited motion of valve.
Another non-linear eect is due to the crack pressure of the over-centre valve. When
the system is at rest, the pressure in the bottom of the cylinder is higher than the
pressure at the valve. When the valve opens, the pressure rst has to build in order to
match the pressure in the bottom of the cylinder. Therefore there are two possibilities
for the pressure sensor location, either at the cylinder or at the valve. It was chosen to
put the pressure transducer at the valve since the current trend suggests that pressure
sensors might be integrated into valves in the future.
62 Chapter 3. Joint Controller Design
Fig. 3.22 shows the eect of adding damping to the negative velocity direction. As
can be seen, pressure feedback stabilizes the system and gives extra damping. The
same non-linear eects apply to the lowering case. The interesting eect of the rate
limit eect is that it is not possible to implement actual Pdot feedback due to the slow
response of the valve. The ltering eect of the valve, makes the Pdot feedback into
pressure feedback. Also, as in the raising case, the size of the lter constant is limited
by the noise level in the system.
Figure 3.22 Pressure feedback eect in negative velocity direction.
3.4.2 Increased Damping via Strain Gauge Feedback
Straingauge feedback is a type of force feedback. This is similar to direct pressure
feedback in the raising case. From the linear block diagram of the complete system
presented in chapter 2 the part of interest is shown in g. 3.23.
Figure 3.23 Straingauge feedback.
T(s)
q(s)
=
AKK
g
(
sf
s + 1)
(AK
g
K
sf
+ Kc)s + Kc
sf
s
2
(3.8)
As can be seen the only dierence between equations 3.1 and 3.8 are the gains. The
dynamics are the same. Therefore the eects of changing the feedback gain and the lter
3.4. Low Level Control Layer 63
time constant are the same for the two systems. Also, the same stability relationships
as shown in the raising case are valid for strain feedback. The only dierence is the
numerical numbers. Therefore only the experimental verication in both the positive
and negative velocity direction is shown here.
Unfortunately the strain gauge set-up used to measure the strain was not very sensitive
and was plagued by noise. Therefore when the extension beam was all the way retracted
with a load of 100Kg, the signal to noise ratio was not large enough for the strain signal
to be used for damping. Therefore the strategy was veried with the extension beam
fully extended. This resulted in higher forces on the beam and therefore higher strains.
Again two sets of tests were performed, one in the positive direction and one in the
negative direction. Fig. 3.24 shows the results of adding strain feedback.
(a) Positive velocity (b) Negative velocity
Figure 3.24 Direct strain feedback.
3.4.3 Gain Scheduling
In the manipulator system discussed in this thesis there is a large change in the natural
frequency of the system due to the change in inertia and stiness of the system as the
extension beam extends and the load changes. It is, therefore, benecial to change the
gains of the damping units as a function of the natural frequency.
In order to do this, the natural frequency needs to be estimated online. It is not
possible to measure the natural frequency since the controllers will have an eect on the
actual frequency of the system. Therefore to estimate the natural frequency the model
developed in section 2.5 was programmed in the micro controller. Using a measurement
of the length of the beam, the angles of the beam and the load mass, an estimate of
the natural frequency could be found. As shown in section 2.5, the natural frequency
estimation is quite accurate. The frequency estimation would be implemented in the
central control which would then gain schedule the individual valves. For systems
such as excavators where the inertia change is smaller, xed gains can be used in the
64 Chapter 3. Joint Controller Design
damping units without any form of gain scheduling.
Deriving the gain scheduling rules is quite simple. The basic idea is that each joint is
tuned independently at a number of dierent manipulator congurations. In all the
tests the integral gain is set to zero.
The tuning rules are quite simple. In the raising case for the shoulder joint and both
directions for the elbow joint, the lter cut-o frequency is set to below the natural
frequency of the system. If the lter frequency is set too far below the systems natural
frequency, then the feedback signal will become too small and will be lost in noise.
In the lowering case for the shoulder joint, the lter time constant is chosen so that
the cut-o frequency is above the natural frequency of the system. However, for high
frequencies this is not quite possible since dierentiation of the pressure signal would
produce too much noise if the lter cut-o frequency is too high. The lter time
constants usually need to be adjusted slightly during the test.
A series of step inputs is then applied to the joint, slowly increasing the damping gain
until a nicely damped response is found. The test is repeated for both the up and down
directions at each joint.
The tuning is then repeated at a number of dierent congurations of the crane. In
the example system, four points was enough to give a good result. From the tests a set
of tables can be derived which will then give the gain scheduling rules.
For the HMF680 manipulator in the laboratory, the following tables were derived. The
gains for the shoulder are based on pressure in Pa to m
3
/s going into the cylinder. The
gains for the elbow are from measured strain in voltage to spool reference to the valve
in volts.
Frequency (Hz) (sec) K (m
3
/Bar)
3.790 0.05 0.0833e-10
2.501 0.20 0.2083e-10
1.672 0.30 0.2917e-10
1.163 0.40 0.3333e-10
Table 3.2 Shoulder joint positive velocity case.
Frequency (Hz) (sec) K (m
3
/Bar)
3.59 0.05 0.0417e-10
2.40 0.08 0.1042e-10
1.60 0.1 0.1250e-10
1.09 0.12 0.1458e-10
Table 3.3 Shoulder joint negative velocity case.
3.4. Low Level Control Layer 65
Frequency (Hz) (sec) K (V
valve
/V
strain
)
2.73 0.1 0.05
2.00 0.2 0.1
1.27 0.3 0.3
0.95 0.5 0.5
Table 3.4 Elbow joint positive and negative velocity case.
From these tables, the gain scheduling rules can be approximated. The gain scheduling
rules for the elbow joint in the raising and lowering case were very similar, so the same
gains were used for both cases.
For the HMF680 crane in our laboratory, a second order curve was found which gave
a good t to the four tuned gains. The gures below give the gain scheduling rules.
As can be seen the elbow joint has the same gains and time constants in the up and
down case. Note that due to the lack of sensitivity on the strain measurement, at
low inertias, the deection is so small that the strain gauge cannot measure it. A low
inertia means a high natural frequency, therefore above a certain frequency; the strain
feedback is disabled as can be seen in the gures.
Figure 3.25 Gain scheduling rules for the shoulder joint.
66 Chapter 3. Joint Controller Design
Figure 3.26 Gain scheduling rules for the elbow joint.
Chapter 4
Stability Analysis
Chapter Contents
4.1 Model of the Complete System . . . . . . . . . . . . . . . . . . . 69
4.2 Kharitonovs Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Stability Analysis Results . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.1 Gain Scheduling Rules Used . . . . . . . . . . . . . . . . . 74
4.3.2 Global Stability . . . . . . . . . . . . . . . . . . . . . . . . . 75
67
4.1. Model of the Complete System 69
4.1 Model of the Complete System
Since the system is a coupled system, the stability of the overall loop cant be guaran-
teed by testing the stability of the individual joint controllers, so a coupled model is
needed. The model used in the stability analysis is shown in g. 4.1. This is the same
model presented in chapter 2 with the addition of the two integral controllers and the
damping units.
Figure 4.1 Overall block diagram of coupled system.
The terms which have a subscript m are related to the Main (shoulder) cylinder.
The terms which have subscript e are related to the Elbow cylinder. The terms with
the subscript h refer to the estimated parameters. So for example the term K
mhg
is
the estimate of the geometry relation for the main joint. When the above diagram is
converted into the linear motion equivalent as discussed in chapter 2 then all the
estimated terms match up with their actual terms. For example every K
f
in the block
diagram is multiplied by 1/K
hf
. Refer to g. 4.2. K
f
/K
hf
could be interpreted as
an estimation error term. If the estimation were perfect then this quantity would be
1. If there is a 20% error in the estimation, then this term would be 1.25 or 0.8333
depending of the direction of the error. Therefore, instead of analysing the stability at
each ow gain and each geometry gain, an error range in estimation can be specied
70 Chapter 4. Stability Analysis
and tested. This can be thought of as a kind of condence limit.
Figure 4.2 Linear equivalent overall block diagram of coupled system.
The form of the above block diagram is good for understanding the physical system but
it is dicult to analyse the stability due to the coupled feedback nature of the system.
So the block diagram is rst reduced into the form shown in g. 4.3(a). Then via the
algebraic transformations given by equations 4.1 to 4.4, the form is changed into the
form shown in g. 4.3(b).
F11 =
G
11
1 G
11
G
12
G
21
G
22
(4.1)
F22 =
G
22
1 G
11
G
12
G
21
G
22
(4.2)
F12 = G
21
G
22
(4.3)
F21 = G
11
G
12
(4.4)
The new form has four transfer functions: F
11
, F
22
, F
12
and F
21
. In the new form,
the four transfer functions work on the same two inputs. The outputs of the transfer
functions are added together. There are no feedback paths. Therefore if it can be
shown that the four transfer functions are stable, then the stability of the entire system
is proven. It can also be shown that the denominators of F
11
and F
22
in this system
are equal. Since the denominator determines stability, the stability is determined by
analysing the denominators of F
11
, F
12
and F
21
which are all interval polynomials.
4.2. Kharitonovs Theorem 71
(a) Feedback form (b) Feedforward form
Figure 4.3 Standard coupling networks.
4.2 Kharitonovs Theorem
The stability method used in this thesis is based on a parameter space approach which
was developed by Kharitonov in the late 1970s.
Robustness and performance usually work against each other. It could be said that a
very robust controller will not have high performance and vice versa. This is reected
in most control approaches. For example a controller synthesis based on pole placement
fullls a performance specication with no regard to robustness. On the other hand, a
backstepping controller guarantees stability but makes no guarantees on performance.
This results in two dierent control design approaches. In the rst approach a controller
is designed which meets performance demands after which stability is checked. In the
second approach a controller is designed which is stable and then the performance
is checked. Both are iterative procedures. In this thesis the rst approach is used.
Controller gains are chosen rst and stability checked afterwards. If the gains are not
stable, new gains are chosen and the stability re-examined. For this approach to be
eective the stability analysis method needs to be fast.
In the system under consideration, there are two issues of large importance. The rst
is that the system parameters change a large amount. The second is that the system
is coupled which results in a high order transfer function. Since the parameters dont
change very quickly, the dynamics of the parameter changes are of little importance
and therefore a parameter space approach was used.
The basic idea behind the parameter space approach is to prove that all the poles of the
closed loop transfer function have negative real parts. A simple way to test this is to
72 Chapter 4. Stability Analysis
plot the roots for every single operating point and verify that they are all in the left half
plane. However due to the number of operating points, this has a large computational
eort.
A more ecient approach is to use a theorem proposed by Kharitonov. (Bhattacharyya
et al., 1995) presents the background and application of Kharitonovs theorem. Kharitonovs
Theorem is applied to interval polynomials of the form:
(s) =
o
+
1
s +
2
s
2
+
3
s
3
+
4
s
4
+
5
s
5
+ .... (4.5)
where the coecients of the polynomial
i
are bounded by x
i
and y
i
so that:

i
[x
i
, y
i
] (4.6)
A polynomial of this form has a large root space. However, Kharitonov proved that
the extremal roots of an interval polynomials roots space are given by the roots of the
following four polynomials:
K
1
(s) = x
o
+ x
1
s + y
2
s
2
+ y
3
s
3
+ x
4
s
4
+ x
5
s
5
+ y
6
s
6
+ ... (4.7)
K
2
(s) = x
o
+ y
1
s + y
2
s
2
+ x
3
s
3
+ x
4
s
4
+ y
5
s
5
+ y
6
s
6
+ ... (4.8)
K
3
(s) = y
o
+ x
1
s + x
2
s
2
+ y
3
s
3
+ y
4
s
4
+ x
5
s
5
+ x
6
s
6
+ ... (4.9)
K
4
(s) = y
o
+ y
1
s + x
2
s
2
+ x
3
s
3
+ y
4
s
4
+ y
5
s
5
+ x
6
s
6
+ ... (4.10)
Therefore in order to prove that an interval polynomial is stable, all that is required
is to show that the four Kharitonov polynomials are stable. The extremal roots given
by the Kharitonov polynomials will always be the rst roots to cross the stability
boundary. Fig. 4.4 shows the root space of an example crane system. The root space
was calculated by varying the coecients of the characteristic polynomial and plotting
the roots. The roots of the four Kharitonov polynomials are shown by diamonds. As
can be seen the Kharitonov roots are the extremal roots of the root space.
Another way to use the Kharitonov polynomials is to plot the frequency response of
the system. The principle is similar to a closed loop Nyquist plot. However, instead
of plotting the frequency plot for every single parameter variation, the frequency re-
sponse of the four Kharitonov polynomials are plotted. As can be seen in g. 4.5,
the four Kharitonov polynomials form the corners of a rectangle at each frequency.
The frequency response of all the dierent interval polynomials is contained within the
squares. This image set can again be used to prove stability. If the origin is excluded
from the Kharitonov boxes at every frequency, then the system is stable. The distance
from the origin to the image set gives a quantitative measure of the robustness of the
4.2. Kharitonovs Theorem 73
system.
Figure 4.4 Roots of the Kharitonov polynomials.
Figure 4.5 Kharitonov image set.
One problem with Kharitonovs theorem is that it is conservative. Kharitonov theorem
says that the coecients of the characteristic polynomial perturb independently, but
this is not true in most systems. The coecients typically perturb interdependently
because the coecients of the characteristic polynomial are functions of the parameters.
74 Chapter 4. Stability Analysis
There are two ways to handle interdepdent parameters. The rst is using the edge
theorem and the second is the Generalized Kharitonov Theorem (GKT). The edge
theorem states that the root set of the system can be found by plotting the root sets
of the edges of the system. The system roots will all lie within the roots sets of the
edges. The problem with this approach is that it is still quite number intensive. The
crane system has 11 parameters which would have 22528 edges.
The other approach is the Generalized Kharitonov Theorem (GKT), presented in
(Bhattacharyya et al., 1995). The GKT divides the characteristic equation into xed
and unknown polynomials. Then the Kharitonov equations are used on each of the
unknown polynomials. Since the Kharitonov theorem condenses to a stability region
with 4 points, independent on the number of unknowns, the GKT depends only on the
number of polynomials, not the number of unknowns. This means that the GKT is
much more compact than the Edge theorem.
In this thesis a modied version of the Kharitonov theorem is used. The parameters
which have the largest inuence on stability are the mass and the orientation of the
crane. Therefore the Kharitonov roots were plotted for the interval polynomial at
dierent orientations and dierent masses. This is still conservative, however, it is
number ecient and shows stability for the crane system.
In this thesis these techniques are used in the analysis of the controller. However, these
techniques can also be extended to synthesis of a controller. One reference for synthesis
using this method is (Djaferis, 1995) and another is (Bhattacharyya et al., 1995).
4.3 Stability Analysis Results
4.3.1 Gain Scheduling Rules Used
Using the transfer functions developed in the previous section it is possible to simulate
the performance and analyse the stability of the system at each combination of orien-
tation and load. In order to nd the gain scheduling rule, the gains were tuned at ten
dierent load/orientation combinations. This resulted in the values of the gains at ten
dierent values of natural frequency. The integral gains were kept constant, so only
the damping terms are gain scheduled.
The performance was analysed via the step response curve of the shoulder cylinder. The
stability was analysed by checking that the roots of the three characteristic polynomials
had negative real parts. The uncertainty in the estimations was included via the
Kharitonov theorem. Uncertainty in the two geometry terms, the ow gain, and the
four controller parameters was included. These seven variable parameters could be
combined via the Kharitonov theorem so that only 4 polynomials had to be solved at
each orientation/load combination.
The tuned controller parameters are shown as functions of natural frequency in g. 4.6.
4.3. Stability Analysis Results 75
The lines plotted on the gures show the gain-scheduling rule chosen.
Figure 4.6 Gain-scheduling rules for the damping controllers.
By limiting the dynamics of the gain scheduler to less than 5 times the system dynamics,
it is assumed that the gain scheduler dynamics do not inuence the system signicantly.
The dynamics can be limited with a lter on the frequency estimation.
4.3.2 Global Stability
To check the stability of the developed gain scheduling controller, the root space was
analysed at each dierent orientation and load combination possible with the crane.
The shoulder angle was varied between -10 degrees and 80 degrees, the elbow joint was
varied between -20 degrees and -150 degrees, the extension beam was varied between
2m and 7m, and the mass was varied between 0 and 400kg. Using embedded for loops
and plotting the roots of the four Kharitonov polynomials the root plots of g. 4.7,
g. 4.8, and g. 4.9 were found. As can be seen for all three diagrams, the roots all
76 Chapter 4. Stability Analysis
have negative real parts which indicate stability.
Figure 4.7 Root plot of the transfer function F11.
Figure 4.8 Root plot of the transfer function F12.
4.3. Stability Analysis Results 77
Figure 4.9 Root plot of the transfer function F21.
Chapter 5
Deection Compensation
Chapter Contents
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Deection Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.3.1 Deection Estimation . . . . . . . . . . . . . . . . . . . . . 86
5.3.2 Deection Compensation . . . . . . . . . . . . . . . . . . 87
5.4 Conclusions on Deection Compensation . . . . . . . . . . . . . 88
79
5.1. Overview 81
5.1 Overview
As shown in chapter 2, the extension beam is quite exible and can deect a large
amount with the application of large loads. The deection causes errors in the tool
centre position which is undesired. Handling exibility on stationary robots is an es-
tablished eld. However, mobile manipulators and stationary robots have very dierent
requirements. For example, 5cm accuracy in deection compensation in a mobile ma-
chine operating with a 10m long extension section is acceptable whereas 5cm accuracy
in a stationary robot which is doing pick and place operations at high speed would be
unacceptable. This means that in stationary robotics components and controllers can
be made quite complicated and expensive because the demand for positional accuracy
is high. In mobile machines, the demand for positional accuracy is lower however the
demand for simple and robust components is higher.
There has not been much research in deection compensation in mobile machines up to
this point. (Nilsson et al., 1999) used a laser range sensor at the tool centre to measure
the actual deection and correct the position. This approach was also used to dampen
oscillations in the tool centre position. (Rouvinen and Handroos, 1997) used a neural
net to correct for static deection.
In this thesis, an estimate of the deection is used by the controller to compensate for
the deection. The deection estimation,
d
, is added to the joint angle measurement,

e
, to get a virtual angle, which is then controlled by the controller. The denition
of the virtual angle, , is shown in g. 5.1.
Figure 5.1 Denition of the virtual angle.
82 Chapter 5. Deection Compensation
The idea in block diagram form is shown in g. 5.2
Figure 5.2 Deection compensation.
In order to reduce the feedback eect the deection estimation has on the system, the
deection feedback estimate is ltered with a slow low pass lter. This means that
the dynamic oscillations are ltered out and only the static deection is compensated.
The lter frequency is chosen at least ve times lower than the main system dynamics
and therefore the feedback action is assumed to have a low inuence on the system
dynamics. The deection compensation action can therefore be implemented as a feed-
forward term which has no dynamic eect on the system. This is shown in g. 5.3.
Figure 5.3 Feed forward deection compensation.
5.2 Deection Estimation
This section presents a strategy to estimate the deection of the extension beam. As
was mentioned before, a virtual control angle is dened which is a combination of the
actual joint angle () and the deection angle (
d
). The joint controller attempts to
control the virtual angle instead of the actual joint angle which results in deection
compensation.
5.2. Deection Estimation 83
In chapter 2 it was shown that the telescopic extension beam can be modelled as a
angular spring lumped at the elbow joint. The deection can therefore be expressed
as:

d
=
T
K(l)
(5.1)
where
d
is the deection, T is the torque, and K(l) is the length dependant spring
constant. Therefore an estimation of the torque needs to be found. If the mass of the
load, the length of the beam and the joint angles are known then the torque could be
calculated directly. Another way is to estimate the torque via a strain gauge on the
beam or via pressure transducers on the joint cylinder. Both strategies are described
and tested. However, the strain gauge approach is shown to be superior to the pressure
sensor approach.
Torque Estimation
Fig. 5.4 shows the locations of the straingauge and pressure transducers for the two
strategies. The strain gauge is mounted on the xed segment of the extension beam.
This beam segment is always at a xed distance from the main crane body which
means that the sensor can be connected with xed length wiring. It is expected that
in the future valves will come with the built in pressure transducers, P
T2
and P
B2
, as
is the case with the new valve from Ultronics, refer to (Ultronics, 1999). However, it is
impossible to measure the force output by the cylinder with these sensors because of
the overcentre valve. Therefore a third pressure sensor would need to be added P
B1
. If
the hoses are very long and the pressure drop across them is signicant, then a fourth
pressure transducer needs to be added P
T1
. In the tests P
T1
and P
B1
were used.
Figure 5.4 Location of the strain gauge and pressure transducers.
84 Chapter 5. Deection Compensation
The strain gauge solution is the more simple, more eective, and cheaper solution.
Only a single strain gauge bridge is needed to directly measure the torque on the
beam. Two pressure transducer readings plus the joint angle are needed to calculate
the torque based on pressure information. A strain gauge is also cheaper than two
pressure transducers. A strain gauge can be packed in a small package which is bolted
to the crane so it is also easy to replace.
However, the biggest argument in favour of a strain gauge is that the strain gauge
measures the torque directly whereas the pressure transducer is aected by friction
eects in the cylinder. This can be seen in g. 5.5(a) and g. 5.5(b). Fig. 5.5(a) shows
the relationship between the torque estimated by the cylinder pressures and the applied
torque. The applied torque was calculated based on the measured load applied to the
tool centre times the distance between the joint and the tool centre. The torque was
estimated using the measured pressure values, the areas of the cylinder, and the angle
between the cylinder and the beam:
T
p
= (P
b
A
b
P
t
A
t
) sin() (5.2)
where P
b
and P
t
are the pressures in the bottom and top of the cylinder respectively,
A
b
and A
t
are the cylinder areas, and is the angle between the beam and the piston.
Fig. 5.5(b) shows the relationship between the strain and the applied torque. The
strain is proportional to the applied torque via the beams spring constant.
(a) Pressure versus torque relationship (b) Strain versus torque relationship
Figure 5.5 Estimation of torque
As can be seen the pressure plot has signicant hysteresis due to friction in the cylinder.
In addition the friction in the cylinder and in the joints works to support the load so
that the torque estimation due to the pressure readings is lower than the real case. The
strain gives a very linear relationship to the applied torque. Therefore a straingauge
was used in this thesis to estimate torque.
5.2. Deection Estimation 85
Deriving a Relationship between Strain and Deection
It was shown in the previous section that the measured strain is proportional to the
applied torque. This leads to the conclusion that the deection is proportional to the
strain. Remembering, from chapter 2, that the spring constant is inversely proportional
to the beam length gives:

d
= K
s
(l) (5.3)
where is the measured strain and K
s
(l) is the length dependant constant relating
deection and strain. The constant K
s
(l) can convert directly between voltage from
the strain gauge and deection. To nd this relationship the same experimental set-up
was used as in section 2.3.2. As dierent torques were applied to the manipulator, the
deection and strain were measured and plotted. A test with an extension beam length
of 3m is shown in g. 5.6. The results are linear.
Figure 5.6 Deection versus measured strain.
The hysteresis in the gure is due to the extension beam and how the extension sections
slide on each other. The test was repeated at dierent extension beam lengths and the
slope measured at each dierent length. The slope versus the extension length is shown
in g. 5.7.
Therefore knowing the length of the extension beam, the slope of the deection ver-
sus strain curve can be found. Then together with the strain, the deection can be
86 Chapter 5. Deection Compensation
estimated.
Figure 5.7 Deection slope versus length of extension beam.
The variation in the slope values is due to measurement errors. The deection at
short beam lengths is quite small which results in errors in the angle measurement. In
addition, the strain gauges used did not have such a high sensitivity and therefore, the
measurement results were not very accurate. It was desired to repeat the experiments
with a more precise strain gauge set-up but time was unfortunately not available.
5.3 Experimental Results
5.3.1 Deection Estimation
The rst tests show how accurately the deection can be estimated. A straight-line
motion test was performed rst with a load of 100Kg and then with a load of 200Kg.
The extension beam was extended slowly and the deection measured and estimated
at dierent lengths. The position plot for the 100Kg load is shown in g. 5.8(a) and
the error is shown in g. 5.8(b).
The position plot for the 200Kg load is shown in g. 5.9(a) and the error is shown in
g. 5.9(b).
The maximum error of 5cm is deemed to be acceptable due to the low position accuracy
requirements of mobile crane applications. Also as can be seen from both tests, the
maximum error occurs in the mid range of the extension beam section. This is due to
the way that the extension beam is built and was discussed in section 2.3.1.
5.3. Experimental Results 87
(a) X-Y position of tool centre (b) Error in estimation
Figure 5.8 Deection estimation test for 100Kg load.
(a) X-Y position of tool centre (b) Error in estimation
Figure 5.9 Deection estimation test for 200Kg load.
5.3.2 Deection Compensation
This section shows the eect of deection compensation. Resolved motion control
was implemented on the manipulator where the shoulder joint was locked in position
so that the redundancy was eliminated. Therefore only the elbow and the extension
beam moved to follow the programmed trajectory. The manipulator was programmed
to follow a straight line with a 400Kg load. The tool centre speed was 0.1m/s.
Two issues can be seen from the plots. The rst issue is related to the deection model
developed in chapter 2. As can be seen the deection model is quite good until the last
meter of beam motion. The assumption made in the deection estimation is that the
beam is of uniform cross sectional area which gives it a uniform inertia. However, the
beam is not of uniform cross sectional area, rather it gets smaller as the length gets
88 Chapter 5. Deection Compensation
longer. Therefore the deection is also not uniform. The deection component due to
the small beams is larger than the deection component due to the large beams. The
deection model could be improved with a beam model which takes into account the
reduction in cross sectional area due to increasing length.
Figure 5.10 Deection compensation results.
The second issue is again the error at the start of motion. At the start of motion,
there is a rather large error which gets xed after a short period of time. New valves
with lower dead-band and faster dynamics will allow higher controller gains which will
improve this issue.
5.4 Conclusions on Deection Compensation
This chapter presented a brief overview of deection compensation. Due to time lim-
itations, deection compensation was not covered in as much detail as was originally
planned. However, the basic idea of the extension beam deection being approximated
by a simple beam is shown to be relatively accurate. However, problems were encoun-
tered due to the length dependence of the spring constant. As was shown, the spring
constant can be expressed as a function of the extension beam length. In this chapter,
this function was not determined with high precision. However, even with this low
precision, the deection could be estimated with an accuracy of around 5cm. This is
quite good considering the length of the manipulator is around 10m. However, it would
have been interesting to attempt to model this length dependant function with more
accuracy. Of special interest was to model this function with a neural network. This
will be left for future research.
Chapter 6
Handling Redundancy
Chapter Contents
6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.1 Previous Research . . . . . . . . . . . . . . . . . . . . . . . 91
6.2 Basic Redundancy Control Concept . . . . . . . . . . . . . . . . 92
6.2.1 Extra Utilization of the Redundancy . . . . . . . . . . . . . 93
6.3 Different Strategies Tested . . . . . . . . . . . . . . . . . . . . . . . 93
6.3.1 Minimum Norm in Joint Space . . . . . . . . . . . . . . . . 93
6.3.2 Minimum Norm in Actuator Space . . . . . . . . . . . . . 94
6.3.3 Minimum Force . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.3.4 Elbow and Extension Only Strategy . . . . . . . . . . . . . 95
6.4 Comparing Different Strategies . . . . . . . . . . . . . . . . . . . . 96
6.4.1 Simulation Model . . . . . . . . . . . . . . . . . . . . . . . . 97
6.4.2 Graphical Verication . . . . . . . . . . . . . . . . . . . . . 99
6.4.3 Test Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.5 Manual Control of the Shoulder Joint . . . . . . . . . . . . . . . . 104
6.6 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.6.1 Automatic Joint Limit Avoidance Strategy . . . . . . . . . 105
6.6.2 Manual Shoulder Joint Control . . . . . . . . . . . . . . . . 106
6.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
89
6.1. Overview 91
6.1 Overview
This chapter examines some dierent options to handle the redundancy in the ma-
nipulator structure. The discussion is based mainly on an energy eciency point of
view.
As was previously discussed, a mobile crane is a redundant system, since there is more
than one solution to the inverse kinematics problem. Redundancy makes the control
algorithm more complex, but makes the controller more exible since the same tool
centre trajectory can be accomplished with many dierent joint trajectories.
In traditional robotics, redundancy is usually handled by nding the joint trajectory
which minimizes the overall energy consumed for a specic tool centre trajectory. This
is possible because in typical robotic applications, the trajectory is predened. In
human operated mobile machines, the trajectory is not predened, so traditional re-
dundancy schemes from robotics cannot be applied.
Another important dierence is that a mobile crane has a hydraulic control system. A
typical hydraulic control system with a load sensing system does not use energy like
an electrical system. A hydraulic system has a single power source which produces
pressure and ow. The pump pressure is matched to the highest loaded cylinder in the
system. Lighter loaded cylinders throttle the pressure which results in large losses.
The dierent strategies presented in this chapter are implemented on a computer model
of a mobile hydraulic manipulator. Typical work cycles are dened and simulated on
the model. During the simulation the overall energy used is calculated. By comparing
the energy used by the dierent strategies it can be seen which strategy is the most
ecient.
6.1.1 Previous Research
There has not been much work done in handling redundancy in mobile hydraulic ma-
nipulators. However, three references of interest are (Singh et al., 1995), (Beiner, 1997),
and (Beiner and Mattila, 1999).
Singh et al. (1995) discusses a solution which uses the Moore-Penrose pseudo inverse of
the Jacobian to nd the minimum norm solution. The minimum norm solution results
in a solution to the inverse problem with the minimum joint velocities. In addition
they also present a gradient projection component which prevents the actuators from
approaching their joint limits.
Beiner (1997) develops a solution which tries to minimize the force on the cylinders.
Beiner starts with the goal of minimizing the norm of the actuator forces in a qua-
sistatic case and develops a set of simplied rules which solve the inverse kinematics
of the manipulator in order to minimize the forces on the cylinders. Beiner presents
four general rules: (i) maximizing the moment arm of the shoulder actuator, (ii) fully
92 Chapter 6. Handling Redundancy
extending the elbow actuator, (iii) fully retracting the extension, and (iv) keeping the
shoulder beam in a horizontal position. He splits the cranes operating space into dif-
ferent zones where each of these rules are most optimum. Depending on which zone
the crane tip is in, a dierent inverse kinematics equation set is used.
Beiner and Mattila (1999) present two new solutions to the pseudoinverse problem. The
solutions are presented in actuator coordinates instead of joint coordinates which makes
it possible to use in control strategies using actuator position feedback. Two solutions
are presented; the rst minimizes the kinetic energy of the structure by including the
inertia matrix of the crane in a weighted pseudo inverse solution. The second solution
optimizes the initial position of the crane depending on the desired trajectory.
6.2 Basic Redundancy Control Concept
As was presented in chapter 1, the Jacobian of the manipulator is a 2x3 matrix. The
problem in nding the inverse kinematics solution is that there is no inverse of a non
square matrix. However, the pseudo-inverse of the Jacobian can be used instead.
q = J
+
x (6.1)
Where the pseudo inverse of the Jacobian is dened as
J
+
= J
T
(JJ
T
)
1
(6.2)
The interesting property of using the pseudo inverse is that the solution to equation
6.1 is the minimum norm solution. It minimizes the length of the velocity vector:
min
_
(

2
+

2
+

L
2
2
) (6.3)
Using the minimum norm of the joint velocity vector results in the lowest possible joint
velocities for the particular desired tool centre velocity and manipulator conguration.
Note that the cylinder velocity is not linearly related to the joint velocity. This means
that minimizing the joint speeds does not necessarily minimize the cylinder speeds.
Since the quantity that should be minimized is ow, it is the cylinder velocity which is
of interest. The same minimum norm strategy can be applied to the jacobian relating
the tool centre velocity to cylinder speeds.
Using the minimum norm method to nd a solution to the inverse problem is of interest
because it results in low joint or cylinder velocities and there exist computationally
ecient methods to calculate it as shown in (Singh et al., 1995). However the solution
takes no account of the fact that this is a hydraulic system being controlled. Due to
large energy losses in hydraulic actuators, having three hydraulic actuators working at
6.3. Dierent Strategies Tested 93
once is most likely not the most ecient solution. Another problem is that the solution
does not take into account the geometry of the manipulator. For example, having the
shoulder joint rotating at high velocity requires high energy, whereas extending the
boom section at high velocity takes less energy due to the dierence in the inertias.
6.2.1 Extra Utilization of the Redundancy
Since the manipulator is redundant there are certain joint velocity vectors which will
not cause any motion of the tool centre. These vectors are dened using:

= (I J
+
J)H (6.4)
where H is any vector and (I J
+
J) is the null space projection matrix. H can be
chosen according to dierent strategies. (Singh et al., 1995) uses a gradient projection
method which minimizes an objective function.

= k(I J
+
J)H (6.5)
Here k is a constant which when positive maximizes H and when negative minimizes
H. The size of k determines how fast H is adjusted. The size of k is limited by actuator
limitations.
In (Singh et al., 1995) H is chosen to avoid the actuators from reaching their end stops.
The function H is dened as a function of joint position travel. As the joint gets nearer
to its limits, H increases. (Singh et al., 1995) cites some more references of the same
procedure where the chosen objective function is dierent.
This strategy could be used in other ways as well. One example is to minimize the
gravity vector. Another example is to minimize the inertia matrix. The problem with
this strategy is that it requires simultaneous motion of all three actuators which uses
extra energy.
6.3 Different Strategies Tested
Four strategies were tested. The rst one attempts to minimize the joint velocities.
The second one tries to minimize the cylinder velocities. The third one attempts to
minimize forces in the cylinders. The fourth one attempts to limit the motion to two
cylinders. These are only four of the many dierent strategies possible.
6.3.1 Minimum Norm in Joint Space
The rst strategy implemented is a copy of the strategy presented in (Singh et al.,
1995). This strategy computes the minimum norm solution in joint space. A gradient
projection scheme is added to avoid the joint limits.
94 Chapter 6. Handling Redundancy
As was mentioned in section 6.2.1, it is possible to use the null space to optimize some
criterion. The form for the inverse kinematics solution with null space optimization is:

= J
+
x + k(I J
+
J)H (6.6)
In (Singh et al., 1995), H was dened as:
H =
n

j=1
cosh
_

j,c

j,max

j,min
_
(6.7)
The gradient components are therefore given by:
H

j
= sinh
_

j,c

j,max

j,min
_
(6.8)
The strategy was implemented with k set to 0.00005 and set to 15. These values
were chosen via trial and error so that the joint limits were avoided.
6.3.2 Minimum Norm in Actuator Space
The second strategy is the minimum norm in actuator space. In the previous strategy
the joint space jacobian was used. However, since it is desired to minimize ow to the
cylinders, it is the actuator velocity which should be minimized. Therefore the previous
section is repeated with the actuator jacobian instead of the joint jacobian. The same
gradient projection approach is used as above to avoid the joint limits.
_
x
y
_
=
_
L
1
sin L
2
sin ( + ) L
2
sin ( + ) cos ( + )
L
1
cos + L
2
cos ( + ) L
2
cos ( + ) sin ( + )
_

(6.9)

_
2L
cyl
1
2L
a
1
L
b
1
sin (+
o
)
0 0
0
2L
cyl
2
2L
a
2
L
b
2
sin (+
o
)
0
0 0 1
_

L
cyl
1

L
cyl
2

L
cyl
3
_

_
The rst step is to nd the minimum cylinder velocities using the pseudo inverse of
the actuator jacobian given by equation 6.9. Then the cylinder velocities are converted
into joint velocities and the gradient projection joint limit avoidance strategy added.
The strategy was implemented with k set to -0.005 and set to 10. These values were
chosen via trial and error so that the joint limits were avoided.
6.3. Dierent Strategies Tested 95
6.3.3 Minimum Force
This solution was inspired by a combination of (Singh et al., 1995) and (Beiner, 1997).
In (Beiner, 1997) Beiner proposes four rules which minimize the forces on the cylinders.
These rules are (i) maximize the moment arm of the shoulder actuator, (ii) maximize
the length of the elbow actuator, (iii) minimize the length of the extension section,
and (iv) keep the shoulder beam horizontal. The rst rule can be achieved by holding
the shoulder joint at a specic angle dependant on the crane geometry as shown in
(Beiner, 1997). Obviously, it is impossible to full both (i) and (iv). Therefore (i) was
chosen as the favoured rule. The other two rules are straightforward.
To implement this strategy, a gradient projection scheme was used which tried to
minimize the dierence between the actual position and the desired position. The
component for the shoulder position is shown in equation 6.10.
H
F,
=
F

c

Max

min
(6.10)
In addition to force minimization, the joint avoidance strategy gradient projection
scheme from the previous two strategies was again used resulting in the motion equa-
tion:

= J
+
x + k
f
(I J
+
J)H
F
+ k
L
(I K
+
J)H
L
(6.11)
The strategy was implemented with k
L
set to -0.001,
L
set to 10, k
F
set to -0.005, and

F
set to 50. These values were chosen via trial and error so that the joint limits were
avoided.
6.3.4 Elbow and Extension Only Strategy
The nal strategy presented is based on the idea that it would be best to only move 2
cylinders at a time instead of all three simultaneously. This reduces the throttling losses
over the valves. The choice was made to use only the elbow and extension cylinders
since the shoulder cylinder will be the highest loaded cylinder and will therefore have
the highest pressure. Another benet of this strategy is that by reducing the number of
joints moving simultaneously, the position error is reduced. In particular, small errors
on the shoulder joint mean large errors on the tool centre position error.
To avoid joint limits a strategy was developed which would move the shoulder cylinder
as one of the other joints approached its joint limit. Refer to g. 6.1 for a graphical
explanation. As a joint moved towards its joint limit the shoulder cylinder would move
in such a way as to reduce the motion necessary by the aected cylinder. For example,
if the extension beam was being retracted and was nearing its limit, then the shoulder
beam would start to rotate upwards. When the shoulder beam rotates upwards, the
96 Chapter 6. Handling Redundancy
tool centre moves inwards. Since the extension beam retraction and the shoulder beam
upwards rotation both result in the tool centre moving inwards, the extension beam
can retract more slowly. If the shoulder beam moves fast enough, the extension beam
can stop its motion. The shoulder cylinder only moves in the case where the aected
joint is moving toward its limit.
Figure 6.1 Joint limit avoidance strategy.
A particular attraction of this strategy is that it has low computational demands. It
rst uses the inverse of the 2x2 jacobian for the elbow and extension section to calculate
the desired joint speeds. If one of the joints is within its restricted area and moving
closer to its joint limits, a velocity reference for the shoulder cylinder is generated
according to the rules shown in g. 6.1. The eect of the motion of the shoulder
cylinder is then subtracted from the desired velocity references and the inverse of the
2x2 jacobian used again to calculate the elbow and extension velocities.
6.4 Comparing Different Strategies
There are many dierent methods of utilizing the redundancy. However, it is dicult to
compare the strategies against each other due to diculties in implementation on the
real machine. Therefore in this thesis a non linear model of the system was developed
and each strategy implemented on the model. By specifying some standard tests and
then comparing the energy usage of the dierent strategies, it is possible to draw some
conclusions as to the energy eciency of the dierent strategies. This concept was
inspired by (Liang et al., 1999).
6.4. Comparing Dierent Strategies 97
6.4.1 Simulation Model
The model was developed in Matlab/Simulink. The larger code functions were imple-
mented as S-functions in m code. There were two parts to the model, a dynamic model
of the crane structure and a model of the hydraulic system. The model structure is
very similar to that developed in chapter 2. However the model developed in chapter 2
was used as a means of understanding the system dynamics and a means of controller
design. The model developed in this chapter is built on the same foundation, but
includes non linear eects like friction, gravity, etc... to make it more realistic.
The purpose of the model was to make a benchmark which could compare dierent
redundancy strategies. Since all the strategies, were tested on the same model, it was
not necessary to make the model 100% accurate. Therefore some assumptions were
made which simplied the model. The importance being that one strategy was not
penalized by the assumptions made in the model development.
The parameters for the model were taken from the HMF 680 crane in our laboratory.
The model structure is similar to that developed in chapter 2. The model used is shown
in g. 6.2. The block called Resolved Motion Rate Control was changed according to
the strategy tested.
Figure 6.2 Model used to compare the redundancy strategies.
Dynamic Model of the Crane Structure
The crane was assumed to be a sti system since it was assumed that the exibility
in the system has a negligible eect on the eciency. The inputs to the model were
torques. The outputs were joint angles and velocities. Geometric transformations
98 Chapter 6. Handling Redundancy
converted between joint coordinates and actuator coordinates.
The extension beam was modelled as ve identical beams (same length and weight)
which extended an equal amount.
A Lagrange approach was taken where the kinetic and potential energy of each beam
was calculated using standard equations for beams which are rotating and translating
in 2D space. Using the standard form of the Lagrange equation the dynamics could be
derived. The generalized forces applied to the structure were two torques (elbow and
shoulder) and one force (extension). The dynamics were written in the form:
M(q) q + C(q, q) + G(q) = Q (6.12)
Due to the low speeds of the manipulator, the coriolis terms could be neglected. Equa-
tion 6.12 was rewritten in the forward dynamics simulation form and viscous friction
added to make the solution more numerically stable. The accelerations could then be
integrated twice to get positions.
q =
Q F q G(q)
M(q)
(6.13)
For more details on this derivation refer to an introductory text on Rigid Body Dy-
namics. For a similar derivation taking the exibility of the crane into account please
refer to chapter 2 or (M unzer and Pedersen, 2001).
Model of the Hydraulic System
The valves were modelled as perfect ow sources, i.e. the ow is proportional to the
input signal. This is a fair assumption since the pressure compensators in ow control
valves such as the PVG 32 from Sauer-Danfoss are fast. No valve dynamics were
included since the reference signals in a typical work cycle are slow in comparison to
valve dynamics.
Over-centre valves were modelled on the elbow and shoulder joints, but not on the
extension system. Innite pilot ratio was assumed which results in high eciency but
stability problems. The stability was improved with Pdot feedback as described in
(Pedersen and M unzer, 2001).
Friction in the cylinders was modelled as coulomb and viscous friction. Coulomb friction
was assumed zero in the joint cylinders and assumed constant at 3,000N in the extension
cylinder. Since all strategies are compared on the same model the assumption was
decided to be acceptable. Viscous friction was assumed at 50,000N/m/s in all three
cylinders.
The load sensing system was modelled as a load sensing system with variable displace-
6.4. Comparing Dierent Strategies 99
ment pump and no dynamics. The pump pressure was assumed to be the maximum
working pressure of all the working cylinders plus 10bar. A cylinder with a ow ref-
erence of zero was decoupled from the load sensing system. Flow from the pump was
modelled as the sum of all the ows to the cylinders.
The total work done by the hydraulic system was measured by integrating the total
pump power output (pressure times ow).
6.4.2 Graphical Verication
To verify that the tasks were completed correctly, the actuator positions could be input
to a graphical model of the crane implemented in OpenGL see g. 6.3. The model
showed the motion of the crane which made it possible to verify that the strategies
worked as desired. The Simulink model output a le of joint positions which could be
run on the graphical model in real time.
Figure 6.3 Graphical model of the crane.
6.4.3 Test Trajectories
The redundancy strategies were implemented on the simulation model and forced to
follow the four dierent trajectories shown in g. 6.4(a). The acceleration and de-
celeration was dened to be maximum 1m/s
2
. This resulted in the velocity proles
of g. 6.4(b). The crane stopped at each corner in the path. The trajectories were
followed in both directions in two separate tests. This resulted in eight tests for each
strategy.
100 Chapter 6. Handling Redundancy
(a) Position
(b) Velocity
Figure 6.4 Tool centre test trajectories.
Example Runs of the Different Strategies Tested
To see how the dierent strategies follow a path, the four strategies were implemented
on the model and trajectory D in the reverse direction was input as the reference.
An example run of the minimum norm strategy is shown in g. 6.5. As can be seen
6.4. Comparing Dierent Strategies 101
from the gure all the cylinders are constantly in motion.
Figure 6.5 Trajectory D backwards with minimum norm strategy in joint space.
A sample run of the minimum actuator norm is shown in g. 6.6.
Figure 6.6 Trajectory D backwards with minimum norm strategy in actuator space.
102 Chapter 6. Handling Redundancy
A sample run of the minimum force strategy is shown in g. 6.7.
Figure 6.7 Trajectory D backwards with minimum force strategy.
A sample run of the elbow and extension only strategy can be seen in g. 6.8. As can
be seen the shoulder cylinder does not move except for the case where the extension
cylinder approaches its minimum length limit.
Figure 6.8 Trajectory D backwards for elbow and extension strategy.
6.4.4 Results
Each test resulted in plots of actuator positions, joint positions, energy input to the
system, and work done over time. The result of interest was the sum of the work
done during a task. Therefore the total energy of each test was measured for the four
6.4. Comparing Dierent Strategies 103
strategies over the eight dierent trajectories and the results shown in g. 6.9.
Figure 6.9 Total energy used by the dierent strategies.
As can be seen from the results, there is not that much dierence between the strategies.
Adding together the energy used to complete all the trajectories gives 878,190 joules
for the elbow/extension strategy, 1,009,100 joules for the minimum norm strategy,
1,041,700 joules for the minimum norm actuator strategy, and 1,064,200 joules for the
force minimization strategy, which shows how little dierence there is between the
strategies.
One explanation of why all the strategies have similar energy usages is that no strategy
is able to operate in the way it was designed to. For example, the minimum norm
solution based on actuator speeds is not able to move as it wants because of joint
limits. The same is true for the minimum norm in the joint space. The elbow/Extension
strategy is designed to only move two joints, but still needs to move three joints at
certain points due to joint limits.
It can also be said that if only 2 cylinders are moved, then the ows for those two
cylinders are higher so the total ow will be higher than in a minimum norm solution.
In a minimum norm solution, the ows might be lower, but three cylinders are in use
so the throttling losses will be higher. Therefore there is little dierence between the
strategies.
It should be noted that the initial positions of the beams have a strong eect on the
energy used. For example, in certain positions of the shoulder actuator, the elbow-
extension strategy never approaches joint limits and therefore the shoulder cylinder is
never used. Therefore initial angles were chosen which forced the shoulder cylinder to
104 Chapter 6. Handling Redundancy
move at some point in the trajectory. Also, initial joint angles were chosen in the mini-
mum norm solution which were already in their equilibrium positions. Therefore when
the test started, there was no initial motion due to the gradient projection component.
It should also be noted that the results can be changed by changing the crane pa-
rameters. This has interesting applications in parameter optimization. By changing
the crane parameters systematically, the cranes physical parameters can be chosen so
that eciency is optimized. As an example, consider coulomb friction in the extension
cylinder. Decreasing this friction, decreases the forces, and therefore the pressures.
This makes the elbow/extension strategy much more attractive.
6.5 Manual Control of the Shoulder Joint
Giving the operator manual control of the shoulder joint is another way to solve the
redundancy problem. This is a viable alternative for situations where the crane is
operating in a constrained workspace. If there are obstacles around the crane, then
it would be benecial to be able to lock the shoulder joint and control it manually.
This makes sure that the shoulder joint wont move unexpectedly.
In order to keep the motion of the shoulder joint from interfering with the tool centre
trajectory, the motion due to the shoulder joint was compensated by the elbow joint
and extension section. This meant that the operator could control the tool centre
trajectory and shoulder joint velocity independently.
The shoulder joint velocity was controlled via a thumbwheel mounted on the joystick,
see appendix F. The tool centre motion due to the shoulder joint was subtracted from
the tool centre reference velocity to give a modied tool centre trajectory. The elbow
joint and extension section were then controlled to follow the modied trajectory. A
good example of how this strategy works is to consider the case where the desired tool
centre trajectory is zero, that is the tool centre is still. When the shoulder joint is
moved down, the elbow and extension move to keep the tool centre velocity zero.
Given the shoulder reference velocity,

, the tool centre velocity can be determined
using the forward kinematics equation 6.14.
_
x
s
y
s
_
=
_
L
1
sin L
2
sin ( + )
L
1
cos + L
2
cos ( + )
_

(6.14)
Given the desired tool velocity [ x
r
, y
r
] and subtracting the tool centre motion induced
by the shoulder joint gives a modied trajectory. The modied trajectory is then used
together with the 2x2 jacobian for the elbow joint and extension section to give the
6.6. Experimental Results 105
desired elbow joint and extension section speed.
_

L
2
_
=
_
L
2
sin ( + ) cos ( + )
L
2
cos ( + ) sin ( + )
_
1
_
x
r
x
s
y
r
y
s
_
(6.15)
6.6 Experimental Results
This section presents experimental results of the two redundancy strategies. First the
automatic joint limit control strategy and afterwards the manual shoulder joint control
strategy is presented.
6.6.1 Automatic Joint Limit Avoidance Strategy
As discussed in section 6.3.4 this strategy locks the shoulder joint and controls just
the elbow and extension cylinder. When the elbow or extension cylinder nears a joint
limit the shoulder joint is moved to keep the joint nearing its limit from reaching its
limit. The rst case presented is when the tool centre is moving towards the base of the
manipulator. A negative velocity reference along the x-axis is given to the controller.
At some point, the extension section nears its shortest length, so the shoulder beam is
raised.
(a) Tool centre motion (b) Joint angles and extension length
Figure 6.10 Automatic joint limit avoidance when moving towards the base.
The second case is when the tool centre is moving away from the base. A positive
velocity reference along the y-axis is given to the controller. At some point the extension
beam nears its longest length at the shoulder beam is lowered to increase the reach of
the manipulator.
106 Chapter 6. Handling Redundancy
(a) Tool centre motion (b) Joint angles and extension length
Figure 6.11 Automatic joint limit avoidance when moving away from the base.
As can be seen from both these gures, as soon as all three beams begin moving
together, there is an error. This is to a large part due to the motion of the shoulder
joint. A small error in the motion of the shoulder joint results in a large error in the
motion of the tip.
In addition, since the resolved motion strategy is an independent joint control strategy,
errors in the motion of one joint are not corrected by other joints. For example,
consider a case where the shoulder beam is moving upwards and the elbow beam is
moving downwards. If the ratio between the speeds is correct, then the tool centre will
move in a straight horizontal line. However, if the shoulder joint is a bit too slow, then
the tool centre will start to move downwards. If the shoulder joint is too fast, the tool
centre will move upwards. The elbow joint will continue to move with no respect for
the error being caused by the shoulder joint.
The eect of this issue can be reduced with a higher performance joint controller.
Higher speed proportional valves are already available which will make it possible to
develop more accurate joint controllers.
6.6.2 Manual Shoulder Joint Control
This section presents experimental results of the manual shoulder joint control as pre-
sented in section 6.5. The idea of the strategy is to control the velocity of the shoulder
joint while keeping the velocity of the tool centre constant. The easiest way to show
the eectiveness of this strategy is to give the tool centre a reference velocity of zero
and track its position as the shoulder joint moves. Two tests are presented, one where
the shoulder beam is raised and one where it is lowered.
6.7. Conclusions 107
(a) Tool centre motion (b) Joint angles and extension length
Figure 6.12 Manual shoulder joint control - raising.
(a) Tool centre motion (b) Joint angles and extension length
Figure 6.13 Manual shoulder joint control - Lowering
As can be seen from the gures the tool centre is kept within a less than 10cm square
area while the shoulder joint is being moved.
6.7 Conclusions
As can be seen from the results of this present set of experiments, there is not one
strategy which is clearly much better than any other from an eciency point of view.
Also it can be seen that the results depend to a certain extent on the parameters of the
crane. Therefore it is not possible to make any rm conclusions as to which strategy
is the most optimal based on these current results. To make a concrete conclusion
would require experimental testing on a real system with real operators.
108 Chapter 6. Handling Redundancy
However, from an implementation point of view, it is the opinion of the author that
the elbow/extension strategy is the most simple and most robust. The tests also show
evidence that it is the most ecient strategy. It also has the highest position accuracy
since the shoulder joint, which has a large eect on the tip position, is not moved.
The other option, manual control of the shoulder joint, is also a viable alternative and
it is the authors opinion that this will be the one most used in industry. This is mostly
due to the fact that the operator is sure of the motion of the crane. There are no
surprise motions in this strategy.
In the current implementation of both redundancy strategies, the elbow and extension
section move in relation to the shoulder reference velocity. However, since the shoul-
der joint has slow dynamics, it is often the case that the elbow and extension section
respond to a shoulder reference before the shoulder starts to move. A better strat-
egy would be to use the measured shoulder velocity instead of the reference velocity.
However, the position signals from the sensors on the crane were not good enough to
dierentiate so the strategy could not be tested properly.
Chapter 7
Experimental Tests
Chapter Contents
7.1 Operator Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.1.1 Trajectory of the Tool Centre . . . . . . . . . . . . . . . . . 112
7.1.2 Velocity of the Tool Centre . . . . . . . . . . . . . . . . . . 113
7.1.3 Bandwidth of the Operator/Crane Combination . . . . . 114
7.1.4 Conclusion on Operator Tests . . . . . . . . . . . . . . . . 114
7.2 Resolved Motion Control . . . . . . . . . . . . . . . . . . . . . . . . 115
7.2.1 Rectangular Motion . . . . . . . . . . . . . . . . . . . . . . 115
7.2.2 Triangular Motion . . . . . . . . . . . . . . . . . . . . . . . . 119
7.2.3 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . 123
7.3 Accuracy as a Function of Speed . . . . . . . . . . . . . . . . . . 123
7.4 Gain Scheduling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
109
7.1. Operator Tests 111
This chapter presents experimental results of the overall strategy implemented in this
thesis. The experimental results were performed on an HMF 680 crane which is de-
scribed in appendix F. The controllers were implemented on a digital signal processor
and the operator interface was controlled via a Graphical User Interface developed in
C++ Builder from Borland. The development system is also described in appendix F.
In most cases, the tool centre position is measured directly using a direct tool centre
measurement system which is described in appendix F. This is in contrast to most
currently available reports on resolved motion where the joint trajectories are mea-
sured. Measuring the tool centre position directly shows if there are errors due to
beam exibility or calibration.
The chapter starts with some tests of a skilled operator using the manual control system
to give a baseline with which to compare the results of the resolved motion control
operated by an unskilled operator. Then tests are presented for the resolved motion
strategy. The rst two tests show the accuracy of the resolved motion scheme. First a
rectangle, and afterwards a triangle, are followed by the tool centre. Afterwards tests
of straight-line motion at dierent speeds are presented. Finally some results showing
the eectiveness of gain scheduling are presented.
7.1 Operator Tests
The tests presented in this section measure how a skilled human operator, using a cur-
rent mobile hydraulic crane, completes a common task. The goal is to nd a benchmark
which can be used to compare the resolved motion controller developed by this thesis
to currently available crane systems.
A local skilled crane operator was asked to come to the university and control the
mobile crane in our laboratory. The crane is an HMF 680 crane and is described in
more detail in appendix F. The task was started with the load sitting on a pallet 1m
from the base of the crane. The operator then moved the load over a 1 metre high
obstacle and onto a second pallet 3 metres from the base. The operator then moved
the load over a second 1.5 metre high obstacle onto a pallet 8 metres from the base.
Refer to g. 7.1. The task was repeated with two dierent loads, a 200kg load hanging
from a rope and a 400kg load rigidly attached to the crane.
The operator had manual control of three levers which controlled the shoulder joint,
the elbow joint, and the boom extension. The rotation joint was locked. Sauer-Danfoss
PVG 32 ow control valves were tted on all cylinders. The tool centre position was
112 Chapter 7. Experimental Tests
measured directly via the system described in appendix F.
Figure 7.1 Denition of the operator tests.
7.1.1 Trajectory of the Tool Centre
The rst objective of the benchmark was to measure how accurate the operator was
in following a trajectory. Therefore the trajectory of the tool centre was measured
and plotted as the operator completed the task described above. The trajectories are
shown in g. 7.2. As can be seen from the gure, the trajectories are not optimal.
Optimal is a not-so-well-dened term, but any person with experience in robotics would
be shocked by such a non-smooth trajectory. However, any construction site boss
watching the motion of the crane as it completed the above task would not be at all
concerned. This clearly illustrates a major dierence between industrial robots and
mobile manipulators.
Figure 7.2 Tool centre trajectory for 400kg rigidly connected load.
7.1. Operator Tests 113
In another task the operator was asked to move the tool centre in a straight line. The
speed of the tool centre during the task was between 0.1 and 0.2 m/s. The task was
repeated with dierent loads. The trajectory plot presented in g. 7.3 was the result
of the trajectory which was the most straight. As can be seen the operator keeps the
tool centre within 10cm over the 6 m of travel once the motion is started, but at the
start and at the end of the trajectory, the error is much larger.
Figure 7.3 Straightline motion.
7.1.2 Velocity of the Tool Centre
The second objective of the benchmark was to measure the speed of the tool centre as
the operator completed a task. This was calculated by dierentiating the position of
the tool centre as the operator completed the task described above.
(a) Entire workspace (b) Close to target
Figure 7.4 Tool centre velocity.
114 Chapter 7. Experimental Tests
A plot of the tool centre speed versus distance to target is shown in g. 7.4(a). As can
be seen the maximum speed of around 0.4m/s is attained when the load is far from the
pickup and drop o points. When the load gets closer to the drop o point, then the
speed is decreased to around 0.1m/s, refer to g. 7.4(b).
7.1.3 Bandwidth of the Operator/Crane Combination
The third test was to measure the bandwidth of the operator/machine system. The
bandwidth will be limited by the operator, the hydraulic system modes, the mechanical
structure modes, and the swinging load dynamics. The bandwidth was measured by
measuring the position of the tool centre during the three tests. An FFT was performed
on the measured tool centre motion. The FFT of one test is shown in Fig 10. There
was very little dierence between the tests. As can be seen the frequency is very
low, even though the particular test shown had the highest frequency limit. So it will
be concluded that the bandwidth of this particular crane/operator combination was
around 1Hz.
Figure 7.5 FFT of tool centre motion.
7.1.4 Conclusion on Operator Tests
This section presented a very brief attempt to establish a benchmark which could be
used to compare the currently available systems with the resolved motion controller
presented by this thesis. It is the authors opinion that this is a topic which is missing
from the literature available on resolved motion control. Researchers are very involved
in nding new control topologies and more advanced controllers, but there is something
missing which can be used to compare the dierent strategies. It could be compared to
having a cake baking contest where the contestants all made wonderful cakes but at the
end, there was no way of comparing the cakes to each other. It is the authors opinion
that it would be very benecial for the research community if a common method
7.2. Resolved Motion Control 115
of comparing controllers were to be presented. The approach taken by this thesis
could be further developed to come up with concrete methods of comparing a resolved
motion controller to a skilled operator controlling a manual system. Then dierent
control topologies could be given a rating which gives their performance in relation to
a currently available manually controlled system.
7.2 Resolved Motion Control
This section presents how accurate the resolved motion control is at short extension
beam lengths. As discussed in chapter 6, the strategy used in this thesis was to move
just the elbow joint and the extension section. The main beam was locked during these
tests. Chapter 6 presents experimental results of how the joint limits were handled.
7.2.1 Rectangular Motion
The rst test presented is how accurately the manipulator follows a rectangle. The test
was repeated three times, rst with no load, then 200kg, then 400kg. The tool centre
speed was 0.1m/s. In each case, the motion was started at the lower left corner. The
tool centre was then given a positive step in the x direction, followed by a positive step
in the y direction, followed by a negative step in the x direction, followed by a negative
step in the y direction. The motion is stopped between each direction change. Note
that the velocity signals were all derived from dierentiated position signals. Hence the
large amounts of noise in the signals. Increased ltering leads to nicer signals however
with the negative eect of higher phase lags.
116 Chapter 7. Experimental Tests
No Load
(a) Tool centre trajectory.
(b) Tool centre x-axis velocity. (c) Tool centre y-axis velocity
(d) Elbow angle velocity (e) Extension velocity
Figure 7.6 Rectangular path following with no load.
7.2. Resolved Motion Control 117
200kg Load
(a) Tool centre trajectory.
(b) Tool centre x-axis velocity. (c) Tool centre y-axis velocity
(d) Elbow angle velocity (e) Extension velocity
Figure 7.7 Rectangular path following with a 200kg load.
118 Chapter 7. Experimental Tests
400kg Load
(a) Tool centre trajectory.
(b) Tool centre x-axis velocity. (c) Tool centre y-axis velocity
(d) Elbow angle velocity (e) Extension velocity
Figure 7.8 Rectangular path following with a 400kg load.
7.2. Resolved Motion Control 119
7.2.2 Triangular Motion
In this test, the tool centre was programmed to follow a triangular path. The purpose
of this test is to show that the manipulator can move in any direction, not just vertical
or horizontal. Again, the tests were performed with 3 dierent loads, no load, 200kg
and 400kg. The tool centre speed was 0.1m/s. In each case, the test was started with
the tool centre at the lower left corner. The tool centre was then programmed with
a positive step in both the x and y directions, followed by a positive step on x and a
negative step on y, followed by a negative step on x. The motion is stopped between
each direction change.
120 Chapter 7. Experimental Tests
No Load
(a) Tool centre trajectory.
(b) Tool centre x-axis velocity. (c) Tool centre y-axis velocity
(d) Elbow angle velocity (e) Extension velocity
Figure 7.9 Triangular path following with no load.
7.2. Resolved Motion Control 121
200kg Load
(a) Tool centre trajectory.
(b) Tool centre x-axis velocity. (c) Tool centre y-axis velocity
(d) Elbow angle velocity (e) Extension velocity
Figure 7.10 Triangular path following with a 200kg load.
122 Chapter 7. Experimental Tests
400kg Load
(a) Tool centre trajectory.
(b) Tool centre x-axis velocity. (c) Tool centre y-axis velocity
(d) Elbow angle velocity (e) Extension velocity
Figure 7.11 Triangular path following with a 400kg load.
7.3. Accuracy as a Function of Speed 123
7.2.3 Discussion of Results
There are two issues which the experimental results show. The rst is the eect of the
joint angle calibration. As can be seen the gures all seem tilted a bit. This is due to
the calibration of the joint angles. Small errors in the joint angle calibration will lead
to this tilting. The reason for the small errors is due to the joint angle transducers
which were used in the experimental set-up. It was discovered that the repeatability of
the sensor was not very good and the calibration would change as the crane was used.
The other issue of interest is the error when the motion rst starts. Each time the
motion starts, there is a directional error in the tool centre motion. However, after
a short time, the tool centre velocity is correct. This is due to a combination of
the low controller gains and errors in the dead-band compensation. If the dead-band
compensation is too large, then there is an overshoot in the velocity response which
leads to oscillations. Therefore the dead-band compensation was reduced. This means
that when the motion rst starts the velocity is a bit less than desired. The controller
can bring it back to the correct value, but since the controller gains are low, the
correcting action is slow. Increasing the controller gains will increase the accuracy but
with an increase in the risk of instability or oscillations. It was decided that the small
position errors are acceptable.
7.3 Accuracy as a Function of Speed
Fig. 7.12 shows the eect of increasing the tool centre reference velocity. As is expected,
as the speed is increased, the accuracy decreases. This is due to the low gains used on
the controllers. Increasing the gains will result in more accuracy but will also increase
the risk for instability. When faster valves are available, it will be possible to work
with higher gains and get better accuracy at high speeds.
Figure 7.12 Straightline motion at dierent velocities.
124 Chapter 7. Experimental Tests
7.4 Gain Scheduling
This section shows the eectiveness and importance of gain scheduling in the damping
unit. The response of the tool centre to step inputs at the shoulder joint are measured.
Note that in these tests, the dead-band in the spool valve was overcompensated which
leads to the large overshoots on the response. The overshoot leads to oscillations which
make it possible to show the damping eects. In the nal controller implemented on
the crane, the dead-band was under compensated to remove the large overshoot.
In the rst test the extension beam was fully retracted and the gains tuned to give a
good response, see g. 7.13(a). The extension beam was then fully extended and a step
response was measured without changing the gains, see g. 7.13(b).
(a) Retracted beam (b) Extended beam
Figure 7.13 Controller gains tuned with beam retracted.
(a) Extended beam (b) Retracted beam
Figure 7.14 Controller gains tuned with beam extended.
In the second test, the extension beam was fully extended and the gains tuned to give
7.4. Gain Scheduling 125
a good response, see g. 7.14(a). The extension beam was then fully retracted and a
step response was measured without changing the gains, see g. 7.14(b).
By using gain scheduling, good responses at all extension beam lengths can be retained.
The gain-scheduling rule was tested by measuring the step response at three dierent
positions of the extension beam, fully extended, half extended, and fully retracted. The
results in g. 7.15 show that the strategy is eective at all extension lengths.
(a) Fully retracted (b) Half extended
(c) Fully extended
Figure 7.15 Testing the gain scheduling algorithm.
Chapter 8
Conclusion
Chapter Contents
8.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.2 Evaluation of the Final Strategy . . . . . . . . . . . . . . . . . . . . 131
8.3 Future Research Work . . . . . . . . . . . . . . . . . . . . . . . . . 132
127
8.1. Overview 129
8.1 Overview
This thesis presented an analysis of resolved motion control on mobile hydraulic manip-
ulators. The goal of the work was to investigate all the dierent issues which are a part
of resolved motion control and propose solutions to them. The main issues touched on
by this thesis were:
Model development
Distributed joint control
Articial damping
Robustness analysis
Deection compensation
Redundancy
Solutions to each of these issues were presented and veried on a real mobile hydraulic
crane.
Model Development
The goal of the model development was to develop a model which could be used both
to analyse stability and as a component of a gain scheduling controller. The important
factors in the model development were:
Low order so that it was ecient
Accurately capture the dynamics
Capture the coupling between dierent components
Show the eect of parameter variations
Linear so that transfer function analysis tools could be used
The transfer function matrix of the system was used in the controller design in order
to verify stability. The model was also used online to estimate the natural frequency
so that the controller gains could be gain scheduled.
Distributed Joint Control
The overall structure of the control used in this system was a distributed joint control
where each joint had its own joint controller. The joint controller was based on a
currently available mobile proportional valve with the addition of an embedded micro-
controller. Due to the dynamics of the valves, an open loop controller with low integral
feedback action was implemented. The results show that there are limitations as to the
positional accuracy of the system due to the slow bandwidth and large dead-band of
130 Chapter 8. Conclusion
the valves. Increased speed in the valves would have a positive eect on the controller
used.
Articial Damping
Two forms of articial damping were added to the structure, one based on pressure
feedback and one based on strain gauge feedback. To eliminate the static osets of
pressure and strain, the ltered derivative of the pressure and strain were used. The
pressure feedback was implemented on the shoulder joint and the strain gauge feedback
was implemented on the elbow joint.
Robustness Analysis
The robustness of the system was analysed with respect to the large parameter varia-
tions which occur during typical manipulator operation. A gain scheduling controller
was implemented in order to maintain performance and stability at all operating points.
The stability of the overall controller was tested via analysis of the root space of the
system model. In order to reduce the computational burden, Kharitonovs theorem
was used.
Deection Compensation
Compensation for deection of the extension beam was implemented. The deection
of the beam was estimated via a strain gauge on the extension beam and the length
measurement of the beam. The deection was compensated directly by the joint con-
troller. A slow lter was used on the deection estimation so that there would be no
dynamic eects from the deection feedback.
Redundancy
A number of dierent redundancy strategies were presented. In order to compare
between them, they were all implemented on a non-linear model of the crane system.
By summing the energy usage over a number of predened tasks, the energy usage
of the dierent strategies could be compared. This led to the choice of one strategy
which was implemented experimentally. The strategy locked the main beam in position
until the elbow or extension reached a joint limit. Reducing the number of cylinders
in motion reduces the throttling losses in the system. It also increases the positional
accuracy.
A second redundancy strategy was implemented experimentally where the operator
was given manual control of the shoulder joint angle. This is good for tasks where the
manipulator is operating in a constrained workspace and there is danger of hitting an
8.2. Evaluation of the Final Strategy 131
obstacle. With manual control of the shoulder joint, the operator has better control of
the crane.
8.2 Evaluation of the Final Strategy
It is the authors opinion that the strategies implemented on the mobile crane in the
laboratory are ready for industrial implementation at the present time. The author
bases this opinion on the fact that the strategy was implemented on a standard mobile
hydraulic crane where the only changes made to the crane were the addition of com-
monly available sensors. In addition, the control strategies used can be implemented
on cheap, commonly available micro controllers.
In terms of eectiveness of the strategy, it is signicantly easier for the author to
complete common tasks using the presented resolved motion control scheme than the
traditional manual control scheme. With the resolved motion strategy, the stress of
performing a task is much less due to the more logical control system. Also, for an
unskilled operator such as the author, the accuracy of the trajectory is much more
accurate with the new system than the old system.
It is obvious from the experimental results that the positional accuracy and the dynam-
ics could be improved with higher speed valves and/or more advanced joint controllers.
However, the question remains what is necessary? It is the authors opinion that higher
dynamics and higher positional accuracy are not justied by more advanced controllers
and faster valves since the operator is the limiting factor in the system, not the con-
troller. It is therefore very important to get feedback of the strategy from actual crane
operators, using the system in actual work situations.
The implementation of the resolved motion controller on the mobile hydraulic crane
has been presented to members of the industry who were quite impressed.
The only feature which is not quite ready for industrial implementation is the redun-
dancy control. It is expected that the strategy which will be most interesting for
industry will be the manual shoulder joint control due to fact the operator has direct
control over the motion of the crane. This strategy suers slightly at the moment, since
the strategy is based on a predicted shoulder motion, instead of the actual shoulder
motion. A slight error between the shoulder motion and the predicted shoulder motion
means that there is an error in the tool centre motion when the redundancy control is
activated. This would be solved by using the measured shoulder velocity instead of a
prediction. However, in the current set-up, the shoulder joint sensor was not accurate
enough to be used in this way.
132 Chapter 8. Conclusion
8.3 Future Research Work
There are many opportunities for future work in resolved motion control. One issue
which is of large importance is a more detailed requirements analysis. This can only be
done via testing of actual operators, either with a virtual reality model or implemen-
tation on an actual machine. Of interest is to nd actual performance demands that
the operator makes on the controller. Also of interest is to analyse the Man Machine
Interface the operator uses to interact with the manipulator.
Another idea is to nd a way to estimate the beam length without measuring it directly.
This would be very benecial as the beam length sensor at the moment is the single
largest cost in the resolved motion strategy. There is also has a large risk of its failure
due to its fragile nature.
A further idea is to investigate a direct tool centre control scheme instead of an inde-
pendent joint control scheme. The accuracy of an independent joint control scheme is
only as good as the accuracy of the individual joint controllers. It could be thought of
as an open loop tool centre control algorithm. A small constant error in a joint velocity
results in an error in the tool centre velocity. For the most part, the errors are small
enough to be ignored. However for large extension beam lengths, the positional error
becomes large enough to be signicant. Therefore current research in resolved motion
control is usually focussed on developing very accurate joint controllers. In contrast, a
direct tool centre control scheme uses the tool centre velocity error directly in the cen-
tral controller to calculate the joint references. There is therefore the possibility that
a controller based on this scheme will have higher positional resolution. However, the
coupling between the cylinders will be increased, leading to a more complex problem.
Appendices
--
There are six appendices in this thesis. Datasheets, etc were not included. For
more details please contact the author directly. Refer to the start of the thesis to
nd the authors email address.
Appendix A presents an introduction to current mobile hydraulic systems.
For those with a limited knowledge of hydraulic systems this is a good chapter to
read rst.
Appendix B presents an argument for decoupling the cylinders hydraulically
due to the action of the LS system.
Appendix C presents a brief discussion of some dierent options for joint
motion actuation.
Appendix D presents a brief state of the art review of some dierent con-
trollers used in the control of linear hydraulic cylinders.
Appendix E presents a brief discussion of ow-sharing.
Appendix F presents a description of the lab facilities.
--
Appendix A
Introduction to Current Hydraulic
Systems
Chapter Contents
A.1 Applications for Mobile Manipulators . . . . . . . . . . . . . . . . 137
A.2 A Typical Mobile Crane . . . . . . . . . . . . . . . . . . . . . . . . 139
A.3 System Components . . . . . . . . . . . . . . . . . . . . . . . . . . 140
A.3.1 Cylinders . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
A.3.2 Valves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
A.3.3 Flow and Pressure Control Valves . . . . . . . . . . . . . . 144
A.3.4 Load Holding Valves . . . . . . . . . . . . . . . . . . . . . . 147
A.3.5 Power Supply . . . . . . . . . . . . . . . . . . . . . . . . . . 149
A.4 Problems with Current Industrial Systems . . . . . . . . . . . . . . 151
A.4.1 Efciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.4.2 Flow Limitations . . . . . . . . . . . . . . . . . . . . . . . . . 152
A.4.3 Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
A.4.4 Environmental . . . . . . . . . . . . . . . . . . . . . . . . . . 153
A.4.5 Lack of Flexibility . . . . . . . . . . . . . . . . . . . . . . . . 153
A.5 Differences between Mobile and Stationary Hydraulics . . . . . 154
135
A.1. Applications for Mobile Manipulators 137
This appendix gives the reader an overview of current mobile hydraulic manipulators.
The overview starts with some applications of mobile manipulators, followed by a
description of a typical crane. Then all the major components are presented along
with an explanation of how they work. The chapter nishes with an overview of the
dierent problems that current systems experience.
A.1 Applications for Mobile Manipulators
Mobile manipulators are used in many dierent applications. The most common one
is as a mobile crane. A typical mobile hydraulic crane is shown in g. A.1. This type
of manipulator is usually used as shown in the gure, mounted on the back of a truck
to help with transferring loads. A similar application mounts a grab bucket to the end
of the manipulator to load and unload bulk materials, refer to g. A.2. Manipulators
can also be mounted with a personnel basket, refer to g. A.3. Then there are more
specialized uses of manipulators such as forest machines, refer to g. A.4. This thesis
focuses mainly on mobile cranes, however, the results can be applied to the other types
of mobile manipulator as well. Excavators are another class of mobile manipulators.
Excavators are characterized by sti beams, lower parameter variations, and more
interaction with the environment. Much of this thesis could be applied to excavators
as well, keeping in mind the dierences.
Figure A.1 Mobile manipulator as a crane.
138 Appendix A. Introduction to Current Hydraulic Systems
Figure A.2 Mobile manipulator with grab bucket.
Figure A.3 Mobile manipulator with personnel basket.
Figure A.4 Mobile manipulator mounted on a forest machine.
A.2. A Typical Mobile Crane 139
A.2 A Typical Mobile Crane
Figs. A.5A.6 show a detailed view of the mobile crane mounted in the laboratory at
Aalborg University. This is the crane which was used in the experimental tests. The
crane is built by Hojbjerg Maskin Fabrik (HMF) A/S, a Danish crane manufacturer. Its
range is about 10m with a lifting capacity of 400kg at that range. The hydraulic system
for cranes however is similar no matter what the size of the crane. The components
get larger, the ows and pressures higher, but the functionality and the requirements
are similar.
Figure A.5 Schematic diagram of laboratory crane.
Figure A.6 Mobile hydraulic crane installed in laboratory.
Fig. F.1 shows a schematic representation of the crane with all the major actuators
listed. The crane has a base (1), a support beam (8), a shoulder beam (19) and an
elbow beam (13). The crane has four actuators, an actuator (5) for rotating the crane
perpendicular to the base (1), an actuator (7) for actuating the shoulder joint (9), an
140 Appendix A. Introduction to Current Hydraulic Systems
actuator (11) for actuating the elbow joint (12), and a set of extension cylinders (15)
for extending the extension beam sections (14). Support arms (3) and feet (4) give the
crane stability. The crane has four Degrees of Freedom (DOF), a rotation about the
base, a shoulder joint, an elbow joint, and an extension section. Since the tool centre
moves in a three dimensional space, the crane is a redundant manipulator.
The actuators on this crane are linear hydraulic cylinders which work by putting pres-
surized hydraulic uid in the two sides of the cylinder. This uid creates a force which
moves the piston up or down, depending on the dierence in pressure on the two sides
of the cylinder. See section A.3.1. A valve controls the uid ow into and out of the
cylinder. The human operator has direct control of the valve, refer to section A.3.2.
For the lifting actuators a further component is introduced called a load holding valve.
This is a safety feature which will keep the crane from collapsing if there were a hose
rupture, refer to section A.3.4. The power for the system is supplied by a power supply.
The power supply is made up of a pump and a way of powering the pump, usually a
internal combustion engine in mobile machines, refer to section A.3.5. A typical actu-
ator system including a load holding valve is shown in g. A.7. The actuator for the
rotating DOF is the same as that shown in g. A.7 but without the OCV.
Valve
User Input
OCV
Tank Supply
Figure A.7 Typical current actuator subsystem.
A.3 System Components
This section presents the components commonly used on mobile cranes, an explanation
of how they work and what some of their problems are. For those with experience in
mobile hydraulic systems, this section can be skimmed.
A.3.1 Cylinders
The cylinder is the motion inducing component in the system. A schematic of a cylinder
is shown in gure g. A.8. This type of cylinder is called a dierential cylinder due to
the dierence in the areas of the two sides of the piston. It has two ports, labelled A
A.3. System Components 141
and B in the gure to which hydraulic uid is regulated. When uid enters the port,
the pressure of the uid increases. When uid leaves the port, the pressure decreases.
Pressure acting on a surface creates a force and from Newtons law, a force will generate
a motion. The two pressures from ports A and B will both create forces on the piston
which will result in an overall force which will move the piston.
F = P
A
A
A
P
B
A
B
F
L
oad (A.1)
Figure A.8 Typical hydraulic dierential cylinder.
Four Quadrant Control
An actuator has four main modes it can operate in depending on the direction of the
load force and the velocity. Fig. A.9 shows the four quadrants of operation. Special care
must be taken when designing a controller otherwise it is possible to get run away load
problems. Run away load problems occur when the load force acts in the same direction
as the velocity. If the controller is not properly designed, then the manipulator cant
be stopped and the load runs away. This is a problem for the operator as well as the
components. In a hydraulic system this problem leads to cavitation which causes wear
142 Appendix A. Introduction to Current Hydraulic Systems
in the components.
Figure A.9 Four dierent quadrants of operation
A.3.2 Valves
A valve controls the amount of uid going into the ports of the cylinder via a variable
orice. The most common type of valve is called a spool valve. A spool valve has a
spool which slides back and forth either opening or closing a port, refer to g. A.10.
The spool valve shown in g. A.10 is called a three position valve. This is because there
are three main positions the valve can be in. Either Port A is connected to Pressure
and Port B is connected to Tank, both ports closed, or Port B is open to pressure and
Port A is opened to Tank. This is the most common valve used on mobile hydraulic
machines. Referring to g. A.8, if it is desired to move the cylinder up, the spool is
moved to the left. If it is desired to move the cylinder down the spool is moved to the
right.
Figure A.10 Typical three position spool valve.
The equation for ow through an orice is given by Bernoullis equation, refer to
A.3. System Components 143
equation A.2.
q = K
o
x
s
_
P
s
P
p
(A.2)
Where K
o
is a constant which depends on the geometry of the valve and the uid
used, x
s
is the opening of the spool, P
s
is the power supply pressure and P
p
is the port
pressure. From this equation it is obvious that the ow is dependent on the pressure
at the port. Therefore since heavier loads produce higher port pressures, the ow will
be lower for heavier loads and higher for lighter loads for the same spool position. Or
looking at it another way, you could say that for dierent ows the pressure drop will
change. Typical pressure vs. ow characteristics for dierent valve openings are shown
in g. A.11.
Figure A.11 Typical pressure vs ow characteristic for a spool valve.
Deadband
In mobile hydraulic valves, there is a large dead-band due to the overlap on the spool.
The reason for the overlap is to reduce leakage. Due to the harsh working environment
of mobile hydraulic machines, the oil is often contaminated. In order to prevent the
spools from becoming clogged up, the tolerances on mobile hydraulic valves are not as
precise as on servo valves. The reduced tolerances force a larger overlap on the spools
in order to maintain low leakage.
New valve designs using poppet (seat) valves will be an interesting answer to this
problem since they have zero leakage when they are closed. However at the moment
the high ow gains of poppet valves, their ow characteristics, and their high ow
forces make them dicult to control.
144 Appendix A. Introduction to Current Hydraulic Systems
Bandwidth
Because most mobile hydraulic systems make use of a load sensing system which
matches the supply pressure to the pressure of the heaviest loaded cylinder, the supply
pressure is constantly changing. In order to guarantee constant speed on the main
spool, the pilot pressure is limited to the minimum pressure of the system, usually
around 20 Bar. This low pressure results in low pilot ows and forces. Therefore these
low dynamics are a feature of mobile hydraulic valves and need to be compensated
for. Systems with separate power supplies for the pilot pressure might be a solution
to low dynamics but increase the system cost and complexity. The reason for the high
bandwidth of servo valves is that their pilot stage can use the supply pressure directly.
The supply pressure is usually high pressure which results in high forces and high ows.
A.3.3 Flow and Pressure Control Valves
To make control of the system easier, valve manufacturers have developed two other
types of valves, Pressure control valves and Flow control valves. In a pressure control
valve, the input from the operator controls the pressure in the system, irrespective of
the demanded ows. A typical ow/pressure characteristic for a pressure control valve
at dierent orice openings is shown in g. A.12.
Figure A.12 Typical ow/pressure characteristic for a pressure control valve.
In a ow control valve the input from the operator controls the ow out of the valve,
irrespective of the load pressure on the cylinder. A typical ow/pressure characteristic
A.3. System Components 145
for a ow control valve with dierent orice openings is shown in g. A.13.
Figure A.13 Typical ow/pressure characteristic for a ow control valve.
Pressure Control Valves
Pressure control valves are normally used in applications where some sort of force
feedback is required by the operator. One example is excavators. In this type of work
it is good for the operator to control the force of the shovel. By controlling the force,
the operator notices what type of ground is being worked by the shovel. It would be
possible to notice the presence of a rock or pipe or other obstruction. As previously
mentioned, pressure control valves control the pressure across the valve, independent
of the ow.
A good example of a pressure control valve is the PVG 32 valve manufactured by
Sauer-Danfoss. The PVG 32 series of valves is congurable as both a ow or a pressure
control valve, just by switching the spool. The ow control mode of the PVG 32 is
presented in the next section.
For the following discussion please refer to g. A.14(a) and g. A.14(b). The operator
controls the position of the spool directly. In the gure the spool is oset to the left
with ow coming out of port B. The LS-A port is closed and the LS pressure comes
from LS-B. The pressure in LS-B is controlled by the variable orice between port B
and the LS-B port. As the valve is moved further to the left, the orice is closed more
and the pressure in LS-B rises, thereby increasing the supplied pressure. As the spool is
moved to the right, the orice opens more, letting more ow to tank, thereby dropping
the pressure in LS-B.
146 Appendix A. Introduction to Current Hydraulic Systems
(a) Cross section (b) Schematic
Figure A.14 Pressure control valve
A typical Flow-Pressure curve for a pressure control valve was shown in g. A.12. As
can be seen from the gure, the valve is not ideal. An ideal pressure control valve would
have completely vertical lines, signifying that the ow has no eect on the pressure at
all.
Flow Control Valves
Flow control valves work by keeping the pressure dierence across the main spool
constant. Therefore they have a second spool called a pressure compensator, refer to
g. A.15(a), which regulates the supply pressure. When the load pressure rises, the
pressure compensator opens more and gives a higher supply pressure, thereby keeping
the pressure drop across the spool orice constant. This means that the ow out of
the valve will be independent of the load pressure.
The following explanation makes reference to g. A.15(a) and the schematic in g. A.15(b).
When the main spool is completely closed, the area A1 is completely open and area A2
is completely closed. As the spool is opened, the pressure between A1 and A2 decreases
which moves the spool to the right. As it moves it opens area A2, which permits uid
to ow through the pressure compensator. As the pressure on the load (Pa or Pb)
rises, the spool is pushed further to the right, opening area A2 even more. Area A1
always remains open unless the pump is disconnected.
In crane applications ow control is usually desired. The operators job becomes easier if
it is known what input corresponds to what speed. However in most crane applications,
the combination of ow control valves and a Load Holding Valve (LHV) (see section
A.3.4) leads to an unstable system. Therefore many cranes are equipped with pressure
control spools.
A.3. System Components 147
For more discussion about dierent types of valves used in mobile applications with
more technical details refer to Andersson and Ayres (1997) and (Paoluzzi et al., 1997).
(a) Cross section (b) Schematic
Figure A.15 Flow control valve
A.3.4 Load Holding Valves
Load holding valves are also sometimes called Over Centre Valves (OCV) or counter
balance valves. The purpose of the load holding valve (LHV) is three fold:
Provide protection against hose rupture
Provide protection against the load running away
Provide protection against the load dropping under no power.
As shown in g. A.7, the LHV is mounted at the cylinder on the load bearing side. A
more detailed gure of a typical LHV is shown in g. A.16. The connection between the
LHV and the cylinder is rigid so that hose rupture between the LHV and the cylinder
148 Appendix A. Introduction to Current Hydraulic Systems
is not possible.
Figure A.16 Cross section of a typical load holding valve.
The LHV works only when the cylinder is moving downwards. When the cylinder is
moving upwards, the LHV is bypassed by the check valve labelled CV in the gure.
When the cylinder is moving downwards uid ow is prevented by the LH Spool. To
open the spool, the pressure on the top port of the cylinder has to be big enough to
open the LH spool.
In the case of a hose rupture (hose to port B on valve) the operator would see the oil
leak and stop the ow to the cylinder via the control valve. This would decrease the
pressure in the top of the cylinder which would close the LH spool and the cylinder
would stop. The LH spool is also designed to give zero leakage when it is closed.
The LHV also prevents run away load. Run away load occurs when the load is in the
same direction as the velocity. Since most ow control valves are meter in ow control
valves, the return port line is usually connected directly to tank via an orice. To make
the system most ecient, this orice is usually designed to be large to prevent unnec-
essary throttling. In the case where the load and velocity are in the same direction,
the loaded side of the cylinder is the one which is connected to tank. If the LHV were
not there, the load would run away. However, the LHV prevents this by making sure
that the top cylinder port pressure never drops below a certain pressure. If the cylinder
were to start to move too quickly, the pressure in the top of the cylinder would drop
and the LHV would close slightly, thus keeping the velocity in control.
As mentioned above, the addition of a LHV to a crane system equipped with ow
control valves, usually makes the system unstable. However, LHVs are a requirement
imposed by the legislation. Therefore, all systems must have an LHV, or some other
device which provides the safety functions of the LHV. For more discussion of the
stability problem with LHVs refer to Persson (1989) and Pedersen and M unzer (2001).
A.3. System Components 149
Alternatives to the LHV
One alternative is to include an electrically operated on/o valve instead of a load hold-
ing valve. This would eliminate the problem of stability but there would be problems
in getting such a system approved by the certifying authorities. There are too many
uncertainties in an electrical control scheme to rely on it for life critical functions. In
addition, some extra form of run away load protection would have to be added.
Another idea is to replace the over-centre valve with a pilot operated check valve, where
the pilot pressure comes from the top cylinder volume. This would take care of the hose
rupture protection. But then a system of preventing run away load would be required.
The new Separate Meter In Separate Meter Out systems seem to be a good way of
performing this function. Also systems utilizing pump based control strategies would
eectively handle this feature. These systems are described in more detail in appendix
C.
However, when using conventional three position spool valves with a single pressure
compensator, the over-centre valve is still a useful, cheap and simple device to provide
both run away load protection and hose rupture protection.
In order to remove the instability, pressure feedback can be introduced which results
in extra damping to the system. The idea is described in more detail in chapter 3.
A.3.5 Power Supply
The power supply has to supply enough power to move all of the actuators at their
desired operating conditions. However, since most of the time, the system is not using
all the actuators at once, there would be a lot of waste if the system were to run under
full operating conditions all the time. Therefore, most mobile systems employ load
sensing power supplies. A load sensing power supply, senses the load and adjusts its
output to provide the right amount of power. Most current load sensing systems sense
the highest load pressure and then adjust the output of the supply, to be a bit higher
than the measured load. There are currently two main types of systems employed in
industry, those with xed displacement pumps and those with variable displacement
pumps. Currently most systems are based on hydro-mechanical regulators. Some
companies have experimented with electro-hydraulic regulators but there are not many
commonly available commercial products yet.
The systems employing a xed displacement pump need a way of sending extra uid to
the tank without raising the system pressure. One way is to use a pressure regulator,
such as the Danfoss Pump Side Module shown in g. A.17. The idea is that the highest
pressure in the valve bank is fed back to the regulator and the regulator adjusts itself
so that the system pressure is slightly higher than this sensed pressure. Referring to
g. A.17, when P
ls
is low, then the spool is all the way to the left and lots of uid
goes to tank, therefore the pressure is low. When P
ls
gets higher it pushes the spool
150 Appendix A. Introduction to Current Hydraulic Systems
to the right, thereby decreasing the opening to tank and increasing the pressure in the
system.
Figure A.17 Load sensing with a pressure regulator.
The other option when using a xed displacement pump is to use open-centre valves.
In this case, when the spool is completely closed, a port inside the valve is opened to
tank which permits the extra uid to get out. Refer to g. A.18. This idea is not
possible with the closed-centre type of spools (see g. A.10) which are the more simple
and common ones.
Figure A.18 Open center valves.
The other way of doing load sensing is to use a variable displacement pump. A typical
set-up can be seen in g. A.19. This is the most ecient way, but it is also the most
expensive and most likely to be unstable. In order to limit problems with instability,
A.4. Problems with Current Industrial Systems 151
most variable displacement pump systems have rather low dynamics.
Figure A.19 Variable displacement system.
Currently load-sensing systems work well. However in some conditions there is a pos-
sibility of unstable operation. This is covered in more detail in Lantto (1994) and
Persson (1989).
A.4 Problems with Current Industrial Systems
Current systems have a number of problems. Four main sections of problems will be
looked at:
Eciency - Low eciency means lots of wasted energy, high running costs, and
oversized components
Power supply limitations - Flow limits in the power supply mean that maintain
a desired tool centre velocity is impossible.
Stability - Unstable systems lead to dangers for machinery, surrounding environ-
ment and operators.
Environmental - Noise and leakage
Lack of exibility - Systems are mechanically complex and are therefore dicult
to modify and tune.
A.4.1 Efciency
Eciency is an important parameter from the environments perspective and from the
operators perspective. Low eciency results in extra polution, extra running costs,
and larger power supplies.
One example of how the design of the system can increase eciency is with Load
Sensing systems. Refer to g. A.20. In the past, systems were powered by a xed
displacement pump and a xed pressure regulator. The introduction of load sensing
152 Appendix A. Introduction to Current Hydraulic Systems
systems which adjust the pressure of the supply greatly improved eciency. Power
supplies based on variable displacement pumps further improve the eciency.
Figure A.20 Eciency of dierent power supplies.
In (Liang et al., 1999) a total eciency of 27.26% is calculated, based on a two degree
of freedom mobile crane using standard Load Sensing valves and a xed displacement
pump. This gure is based just on the eciency of the valves, neglecting things like
leakage, pipe loss, friction, etc. As well counter balance valves and check valves were
not included. Therefore the gure for current mobile manipulator eciency is more
than likely lower than 25%. Eciency of mobile cranes is discussed further in (Kappi
et al., 2001) and (Liang and Virvalo, 2001).
Current systems also have a tendency to instability (see section A.4.3). To counteract
this instability, system manufacturers have worked with higher pressures to provide
damping. In a dierential cylinder a pressure is generated in the side opposing motion.
This generates damping, however it also means the pressure in the side generating the
motion must be higher. This raises the demands on the power supply. Higher pressures
also leads to increased forces on seals which leads to increased friction in components.
A.4.2 Flow Limitations
Due to the limited ow output of the pump it is possible that the demanded ow by
the operator is greater than the ow available from the pump. This results in some
actuators moving slower than desired. Unfortunately, the system does not scale the
ow to all the cylinders linearly, rather the cylinder with the highest load gets cut o
rst. In order to improve the performance of the system, ow sharing systems are
being developed. Flow-sharing is discussed in more detail in appendix E.
A.4. Problems with Current Industrial Systems 153
A.4.3 Stability
Stability is a big issue because instability is dangerous for the machine, the operator,
and the surrounding environment. In current systems stability is assured by increasing
the pressure via extra orices. Higher pressures lead to higher damping, however,
higher pressures simultaneously lead to decreased eciency. Another way to reduce
the risk of instability is to reduce the working speed of the actuator so that unstable
modes are not excited. A third way of reducing the risk of instability is to eliminate
certain functions, such as ow control, which are known to give problems with stability.
All these options have drawbacks.
The instability is due in part to the low natural damping in the system. In addition the
structure is very exible which leads to structural oscillations. In general it is dicult
to eliminate the risk of instability entirely using current hydro-mechanical systems.
A.4.4 Environmental
Hydraulic systems are generally criticized for being noisy. This comes from pressure
uctuations in components and hoses. Pressure uctuations come from low frequency
pumps or motors. Also valves make noise when throttling the pressure.
Hydraulic systems are also prone to leakage. In an outdoor environment, oil leakage
can damage plants and lead to poisoning of animals and shes.
A.4.5 Lack of Flexibility
Since current systems are mainly hydro-mechanical, changing system characteristics
requires new components to be made. Since hydraulic machines have many dierent
purposes and operating conditions, there are a large number of dierent variants. Valve
manufactures have a dicult job in keeping track of all the variants and providing them
on demand.
For example something as simple as changing the ow characteristic of a valve requires
manufacturing a new spool. For an example, g. A.21 shows three dierent ow output
curves for three dierent system manufacturers. Manufacturer A needs to have a large
dead-band followed by a high ow gain. Manufacturer B needs a small dead zone and
then a low ow gain. And manufacturer C needs a small dead-band followed by a
154 Appendix A. Introduction to Current Hydraulic Systems
non-linear ow gain.
Figure A.21 Figures for dierent applications
If more advanced options are required such as ow control or pressure control, the valves
get complicated. Refer back to sections A.3.2. The spool is complex, the valve body
is complex, the ow/pressure interaction is complex, modelling the valve is complex,
etc... In addition there are many tolerances that have to be accurate which drive up
manufacturing costs.
Then there are demands from the hydraulic system customers for more complex options.
A good example is ow sharing. In a mobile machine there is usually limited ow from
the power supply. Often it the actuators require more power than the power supply
can provide. In this case, it is dicult to control which actuator gets how much of the
supply. Flow sharing allows the user to control how the system splits the ows to the
dierent actuators. This is very dicult to implement mechanically.
A.5 Differences between Mobile and Stationary Hy-
draulics
Mobile hydraulic systems and industrial hydraulic systems are often compared because
they use similar components, however, they are very dierent from each other. Table
A.1 presents a comparison between mobile and industrial hydraulic systems.
A.5. Dierences between Mobile and Stationary Hydraulics 155
Industrial Mobile
Process Periodic Changes
Power Unlimited Limited
Space/Mass Unlimited Limited
Type of Valve Servo Proportional
Operating Conditions Controlled Uncontrolled
Sensors Unlimited Limited
Table A.1 Comparison between mobile and industrial systems
Because the task to be performed is not periodic, there is no way of knowing what
the task will be ahead of time. In typical robotic applications, such as pick and place
operations, the task is well dened, i.e. the workspace, the initial coordinates, the nal
coordinates, the object type etc... are all dened. This means that the manipulator
can be pre programmed with all this information to make the control more accurate.
In an industrial setting, the manipulator is usually connected directly to a separate
power supply. For example an electrically driven robot is connected directly to the
power grid which could be assumed as having an unlimited amount of power. A mobile
machine on the other hand, has to carry around its own power supply. Therefore there
is a need to limit the size of the power supply and the controller needs to take into
account the limited amount of power.
An industrial machine is typically xed in place, where the only cost of increased size,
is more oor space, which is not so bad. A mobile manipulator is limited to the amount
of mass and space it can take up since it has to carry around all that mass. Also for
the example of a mobile crane, the arm has to fold up and be stored in a small amount
of space.
Industrial machines use high performance servo valves with high speed and low dead-
band. Mobile hydraulics use lower performance proportional valves with large dead-
bands and low bandwidth. Proportional valves are used due to the lowered cost, ef-
ciency requirements, low leakage requirements, and robustness (clean uid) require-
ments.
Industrial machines operate in a controlled environment where the components can be
kept clean and temperature controlled. In a mobile environment, there is no control
of the environment. Temperature can change from a cold winter day of -30 degrees C
to a hot summer day of +30 degrees C. The weather can be dusty, rainy, snowy, etc,
therefore the components have to be more robust.
Industrial robotics have an unlimited choice of sensors. In mobile hydraulics, due to
the harsh environment and undened workspace, the sensor choice is more limited. In
addition mobile manufacturers are unable to add unlimited sensors to the manipulator
due to robustness and cost concerns.
Appendix B
Hydraulic Decoupling
Chapter Contents
B.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
B.2 Pump Side Module . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
B.3 Pressure Compensator . . . . . . . . . . . . . . . . . . . . . . . . . 163
B.4 Experimental Verication . . . . . . . . . . . . . . . . . . . . . . . 165
157
B.1. Overview 159
B.1 Overview
The load sensing system, in this thesis a Sauer-Danfoss pump side module, and the
pressure compensator work together to ensure that the pressure drop across the main
spools remain constant. The principle is shown in g. B.1.
Figure B.1 Feedback action of load sensing system and pressure compensator.
The gure shows a block diagram of a simple cylinder moving a mass. The ow into
the cylinder is controlled by an orice. The linearized ow equation over an orice is
given by:
q = K
x
x + K
p
P (B.1)
were K
x
and K
p
are operating point dependant constants. They relate the ow through
the orice to the orice opening and the pressure drop across the orice respectively.
The goal of the load sensing system and the pressure compensator is to keep the pump
pressure equal to the load sensing signal pressure. Since this is a linear analysis, osets
are removed, and therefore the signal DeltaP is approximately zero if the LS system
and the pressure compensator are signicantly faster than the rest of the system. If
the signal is zero the pressure feedback loops can be removed from the analysis.
This section presents models of both the pump side module and the pressure compen-
sator used on the experimental system. The models are solved algebraically and the
dierent time constants are isolated. The time constants are then plotted for dierent
operating points. As will be shown the time constants are all signicantly faster than
the dynamics of the rest of the system. The dynamics of the rest of the system are
measured in section 2.5.
160 Appendix B. Hydraulic Decoupling
If the operation of the pump side module and the pressure compensator are unfamiliar,
please refer to appendix A which presents a description of the operation of these two
units.
B.2 Pump Side Module
The typical load sensing system with xed displacement pump (as used on many mobile
manipulators) can be described by g g. B.2(a). The hydraulic functionality is shown
in g. B.2(b).
(a) Physical construction (b) Schematic
Figure B.2 Pump side module.
The linear model of this system can be specied by the following equations.
P
p
=

V
p
s
(Q
p
Q
v
Q
t
) (B.2)
P
1
=
1
1 +
V
1
s
K
1
P
l
s (B.3)
x =
P
p
A P
ls
A F
flow
K
s
(B.4)
Q
t
= K
x
x + K
p
P
p
(B.5)
F
flow
= K
fx
x + K
fp
P
p
(B.6)
Where:
B.2. Pump Side Module 161
K
x
= (C
d
(2)
_
2

)
_
P
po
(B.7)
K
p
= (C
d
(2)
_
2

)
x
o
2
_
P
po
(B.8)
K
fx
= (2C
d
(2) cos 69) P
po
(B.9)
K
fp
= (2C
d
(2) cos 69) x
o
(B.10)
x
o
and P
po
are the operating points for the spool position and Pump pressure respec-
tively.
The block diagram of the system is given by g. B.3.
Figure B.3 Block diagram of pump side module
Reorganizing the block diagram gives:
Figure B.4 Block diagram of the pump side module reorganized
162 Appendix B. Hydraulic Decoupling
As can be seen from g. B.4 there are two time constants. One for the pressure build-up
in the hoses between the pump and the pump side module and one for the pressure
build up in the small volume in the Pump Side module.
Fig. B.5 shows how the time constant of the large volume system changes for dierent
values of the operating point. The valve opening operating point was varied between
0.1mm and 5mm. The pressure operating point was varied between 10Bar and 200Bar.
Figure B.5 Time constant of the large volume system as the operating points are varied.
Fig. B.6 shows how the time constant of the small volume system changes for dierent
values of the operating point. The volume operating point was varied between 0cm
3
and 1.4cm
3
. The pressure drop operating point was varied between 1Bar and 200Bar.
Figure B.6 Time constant of the small volume system as the operating points are varied.
As can be seen the slowest time constant is around 4.3ms. As this is signicantly
faster than the dynamics of the system, it can be assumed that the pump side module
is eective in decoupling the cylinders hydraulically.
B.3. Pressure Compensator 163
B.3 Pressure Compensator
The pressure compensator makes sure that the pressure drop across the main spool is
constant. It takes the pump pressure and throttles it down to the correct pressure for
the main spool. Fig. B.7(a) shows a functional diagram of the pressure compensator
and g. B.7(b) shows a schematic diagram of the pressure compensator.
(a) Physical Construction (b) Schematic
Figure B.7 Pressure compensator.
Assuming that the spool is light enough to neglect its inertia, the only time constant
comes from the orice in the LS signal line. The system equations can be given by:
Q = K
x
x + K
p
(P
p
P
k
) (B.11)
F
flow
= K
fx
x + K
fp
(P
p
P
k
) (B.12)
F = 0 = P
2
A P
k
A F
flow
+ K
spring
x (B.13)
Q
ls
= K
ls
(P
ls
P
2
) (B.14)
x =
_
Q
ls
A
dt (B.15)
Solving equations B.11 and B.13 for P
k
and P
2
respectively we can make up the fol-
164 Appendix B. Hydraulic Decoupling
lowing block diagram.
Figure B.8 Block diagram of pressure compensator.
Reducing the block diagram into a more useable form results in:
Figure B.9 Reduced block diagram of pressure compensator.
Fig. B.10 shows a plot of the time constant of the above system for dierent operating
points. The operating points are dierent pressure drops across the orice and dierent
orice openings. As can be seen the system gets faster as the pressure drop increases.
The system also gets faster as the orice area decreases. Note that it is not the LS
B.4. Experimental Verication 165
orice that is discussed here but the spool orice area.
Figure B.10 Time constant of pressure compensator system at dierent operating points.
B.4 Experimental Verication
It is dicult to experimentally verify the dynamics of the load sensing system since
it is impossible, without modications to the components, to measure the pump side
module and pressure compensator spool positions. However, a good approximation
which shows the eectiveness of the load sensing system is to measure the dierence
between the actual load pressure and the LS system output pressure. Since the LS
system tries to match the LS system output pressure to the actual load pressure, the
time delay between the two pressure signals shows how fast the LS system is. A test was
performed where the shoulder cylinder was given a ow step in the positive direction.
This resulted in oscillations in the pump and load pressures. The measured results are
shown in g. B.11(a).
(a) Overall results (b) Zoomed in and shifted vertically
Figure B.11 Comparison of the pump and load pressures.
166 Appendix B. Hydraulic Decoupling
Since the Pump pressure is always 10bar higher than the load pressure, 10 bar was
subtracted from the pump pressure so that the results could be compared directly. As
can be seen from g. B.11(b) there is a delay between the load pressure and the pump
pressure. By using Matlabs System Identication Toolbox and assuming a rst order
model, a time constant of around 7 milliseconds was found. This matches well with
the assumption that the load sensing system is much faster than the system dynamics.
Appendix C
Angular Joint Actuation
Chapter Contents
C.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
C.2 Hydraulic Actuator Control Topologies . . . . . . . . . . . . . . . 170
C.3 Different Options for Valve-Based Control . . . . . . . . . . . . . 173
C.3.1 Type of Orice . . . . . . . . . . . . . . . . . . . . . . . . . . 173
C.3.2 Valve Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . 174
C.3.3 Control Mode . . . . . . . . . . . . . . . . . . . . . . . . . . 175
167
C.1. Overview 169
C.1 Overview
This appendix discusses some dierent options for manipulator joint actuation. Tra-
ditionally joint control has been done with a linear hydraulic cylinder controlled by a
proportional valve. However, in the last few years a number of new ideas have started
to emerge which are promising. Due to the goal of this PhD to develop a solution
which could be implemented on current mobile machines, these new strategies were
not covered in the body of this thesis. They are mentioned here to give completeness.
There are two main possibilities in actuating manipulator joints. The rst and most
traditional way is via a linear actuator which applies a force to the beam a certain
distance from the joint. The other is via a rotary actuator which directly applies a
torque to the joint. The benet of a rotary actuator is that the torque and velocity
gain are constant whereas a linear actuators torque and velocity gain are constantly
changing due to the changing geometry of the joint.
A linear actuator is typically a dierential hydraulic cylinder, but could also be an
electric motor driving a worm gear or a rack and pinion type linear actuator. Ro-
tary actuators are typically electric gear motors, but could also be rotary hydraulic
actuators, see g. C.1 in the next section.
Electric Actuation
Electric actuation of mobile cranes is not very common. Electric drives suer from
low power to weight ratio as compared to hydraulic drives. Direct joint activation via
rotary electric machines is not feasible due to the large size of the electric machines
necessary to drive the high loads in mobile crane applications. New linear drives which
use a rack and pinion type system or a worm gear system are also being experimented
with, refer to (Widner and Hamel, 1998).
However, it is the authors opinion that electric drives will never be commercially suc-
cessful in mobile crane applications. The popularity of electric drives in stationary
robotics is mostly due to the fact that electrical power is freely available via the elec-
tric power net. Hydraulic systems always require a separate power supply. However
in a mobile application, there is no electrical power available therefore an electrical
generator is needed. Electrical machines would make the mobile machine heavier than
is necessary. In addition electrical machines are more complicated than hydraulic sys-
tems. In environments such as forestry machines which experience dirty environments
and large temperature variations, the simpler and more robust the machine the better.
However, with the increased promise of commonly available fuel cells, this might change
at some point in the future.
170 Appendix C. Angular Joint Actuation
Rotary Hydraulic Actuators
Rotary hydraulic cylinders are also not very common in mobile cranes. They are
relatively new and havent yet been proved commercially. A good source for more
information on systems using rotary types of actuators is the Technical University of
Hamburg-Harburg. A good example being (Grabbel, 2002).
Figure C.1 Rotary hydraulic actuator.
Rotary hydraulic actuators have the very positive feature that the torque gain and the
velocity gain is a constant, independent of joint angle. This is in contrast to linear
hydraulic cylinders whose velocity and torque gain are dependant non-linearly on the
joint angle. This makes rotary cylinders easier to control. However, rotary hydraulic
actuators typically suer from relatively high leakages and large size.
C.2 Hydraulic Actuator Control Topologies
There are a number of dierent options to control hydraulic actuators. The most
common way is via a valve group mounted at the base of the manipulator. The valve
group has a number of separate directional control spools which control the ow to the
dierent cylinders in the system. One side of the cylinder is connected to the tank and
the other to the pressure source. Varying the opening of the spool valve controls the
amount of ow going to the cylinder. This type of control was used in this thesis and
is introduced for the new reader in appendix A.
However, in the last few years a few new topologies have been developed which show
much promise. In each case, the main benet over traditional hydraulics is the im-
proved eciency. The rst strategy presented is SMISMO valves, the second is pump-
based control, the third is switching valves, and the fourth is valves based on electro-
rheological uids.
C.2. Hydraulic Actuator Control Topologies 171
Separate Meter In Separate Meter Out Valves
In the last few years, interest has developed in the eld of separate meter in Separate
meter out (SMISMO) valves. Refer to Ultronics (1999) and Caterpillar (1999) for
two patents in the area. The idea is that a separately controlled metering device is
connected to each port of a cylinder, refer to the right part of g. C.2. The left part of
g. C.2 shows the conventional valve set-up. A SMISMO set-up allows the ow to be
independently controlled to the two sides of the cylinder. In traditional spool valves the
ow to the two sides of a cylinder is coupled and cannot be controlled independently.
Separating the control gives many new possibilities of control.
Figure C.2 Separate meter in separate meter out control (SMISMO).
The biggest result is that the eciency can be improved. The problem with coupled
ports on a cylinder is that the pressure level and the speed are coupled. In current
hydraulics it is often the case that there will be 100 Bar on one side of the cylinder
and 80 Bar on the other side. This means that the cylinder develops an eective force
based on a pressure dierence of 20 Bar. The force would be exactly the same if one
side had 20 bar and the other had 0 Bar. The second case would be much more energy
ecient because the power supply would only have to supply 20 Bar instead of 100
Bar.
(Elfving, 1997) addresses this problem with the development of a controller which
decouples the pressure level in the actuator and the velocity of the actuator. The work
was completed at Linkoping University in Sweden which started their experiments with
SMISMO valves in the very early 1990s. (Linyi and Qingfeng, 2001) presents another
example.
Another problem with traditional valves is that they are typically set up as meter in
valves. This is a problem for cases where the load is in the same direction as the velocity,
since the result is a run away load condition. A SMISMO valve can be programmed
to change its operating mode depending on the direction of the load and the desired
velocity. A review of the subject co-written by the author is presented in (Andersen et
al., 2001).
172 Appendix C. Angular Joint Actuation
Pump-Based Control
Another option which shows much promise is pump-based control. In this strategy
each actuator has a separate pump. The pump is directly controlled to supply the ow
required by the actuator, refer to g. C.3. This eliminates the need for valves. The
pump-based solution has the benet that it results in higher eciency. In typical valve
based solutions, there is one pump which supplies the pressure which is required by the
heaviest loaded actuator. The valves to the other actuators in the system, throttle the
high pump pressure down to the pressure level required by the actuator. This results
in high losses.
Figure C.3 Pump based control.
Other benets of the pump based solution are built in run-away-load protection and
energy storing. Run-away-load is a typical problem in traditional single spool valves
due to the fact that the meter in and meter out port are coupled. Most valves are
meter in control valves with the meter out side connected to tank. When the load and
the velocity act in the same direction, then the loaded side of the cylinder is connected
directly to tank. This causes run away load. In most cases, over centre valves are used
to handle this case. In a pump based control solution, the pump can be run as a motor
for the reverse loading case. This also makes it possible to put the reverse energy,
for example lowering a boom, back into the system. In traditional systems based on
hydraulic proportional valves and a load holding valve, the energy is throttled and
wasted to heat.
However the hydraulics industry is rather conservative and the idea of multiple pumps
and no valves is scary for most hydraulic manufacturers. Therefore, pump based
control systems are still quite rare in practice. For some good references on pump based
control refer to (Grabbel and Ivantysynova, 2001), (Rahmfeld and Ivantysynova, 2001),
and (Grabbel, 2002).
C.3. Dierent Options for Valve-Based Control 173
Switching Valves
Recently the use of high speed switching valves for hydraulic control has become in-
creasingly popular in the literature. In traditional hydraulic circuits, the pressure is
throttled in order to get lower pressure. However, this results in large losses. Switching
valves are based on a similar concept to Pulse Width Modulated (PWM) control known
from electric drives. A high-speed valve is switched on and o quickly with a certain
pulse width. The ow uctuations are ltered by the hydraulic circuit to give a smooth
velocity or pressure. An example of this idea is given by (Becker, 1997).
The problem with this idea is that hydraulic systems are much stier than electrical
systems. Due to the almost incompressible nature of hydraulics, ow uctuations result
in large pressure uctuations. This results in noise and wear on the components. In
order to increase the performance, high switching frequencies are needed. This means
that high speed components are needed which are expensive and use much power.
However, as the valve technology gets better, the performance will go up and cost
come down which will make this strategy more commercially viable.
Electro Rheological Fluids
Electro rheological uids are the most revolutionary idea of the four presented here.
In electro-rheological uids, the uid can be made sti by passing a magnetic eld
through it. This means that all orices can be eliminated. A valve in an electro-
rheological uid system, is just a tube with a coil around it. Electro-rheological uids
are not new and have been around for a while however they are much to radical for
the conservative hydraulics industry. However, the idea is good since the components
can be made much simpler with no moving parts. Also the control exibility is much
greater since valves can be inserted at many points in the system. A reference to this
type of system is (Yokota et al., 1996).
C.3 Different Options for Valve-Based Control
There are many dierent options for valve-based control. The rst option is which type
of valve to use. Traditionally this was a choice between servo valves and proportional
valves. However in the last couple of years, the distinction between servo and propor-
tional valves has become less distinct. So instead of specifying servo or proportional,
it is more interesting to discuss which features of the valves are important.
C.3.1 Type of Orice
There are two main types of orices used in valves, seat valves and spool valves. Tra-
ditionally most mobile hydraulic valves were equipped with spools because it was the
174 Appendix C. Angular Joint Actuation
simplest. A single spool can control 4 orices simultaneously. See g. A.10. A seat
valve can only control one orice at a time. See g. C.4. However the big advantage
with seat valves is that they have zero leakage. Spool valves are plagued by leakage due
to the tolerances between the spool and the valve body. To reduce leakages with a spool
valve it is necessary to have a large overlap which results in a dead-band in the ow
output. In systems with manual control, dead-band is not such a problem. However
when doing automatic control, dead-band is a problem and needs to be compensated
for. Seat valves also have very high ow gain, i.e. small openings result in large ows.
This makes the seat valve more dicult to control. With spool valves on the other
hand it is very easy to make dierent ow gains, by machining the spool dierently.
Therefore it is expected that for the next few years at least spool valves will remain
the orice of choice in mobile hydraulics.
Figure C.4 Simple seat valve.
It is expected that in the coming few years, the position control of seat valves will be
improved and seat valves will start to be more common in the market place.
C.3.2 Valve Bandwidth
The bandwidth of the valve is another issue of importance. Servo valves use high
pilot pressures to control the spool motion. This results in high forces and high ows
which results in high bandwidth. In mobile hydraulics it is not possible to have a high
pilot pressure since this would mean running a power supply at high pressure levels.
Instead mobile hydraulic valves use a low pilot pressure, typically around 20 bar. This
results in both force and ow limits. The force limits, limit the acceleration of the
spool. However, due to the light weight of the spool, the acceleration limits are not of
signicant importance. More important is the ow limit. In order to move the spool
a certain amount, a certain amount of uid has to be input to the system. The ow
input is dependant on the pressure drop across the control orice. Since the pressure
drop is limited the maximum speed of the spool is also limited.
The spool position controller of typical electrically activated proportional valves has a
rather low positional accuracy. Therefore a low ow gain is used together with large
spool travels. Increasing the positional accuracy means that higher ow gains could be
used. Higher ow gains mean that the overall spool travel can be reduced. A reduction
C.3. Dierent Options for Valve-Based Control 175
in the spool travel will increase the bandwidth of the valve.
In the case of the Sauer-Danfoss PVG 32 valve, the limit on spool positional accuracy
is the rather slow switching frequency of the pilot stage. Increasing the switching
frequency requires increasing the speed of the switching valve. This is possible due
to the decreasing cost of commonly available switching valves from the automotive
industry.
C.3.3 Control Mode
Typical proportional valves used on mobile manipulators come with either pressure
control or ow control functionality. Flow control is used in applications where it is
desired to control the cylinder velocity independent of load. A good example is a mobile
crane. Pressure control valves are used in applications where it is desired to control
the force of the manipulator independent of the velocity. A good example being an
excavator. Hydraulic-mechanical solutions to ow and pressure control are discussed
in Appendix A.
Recently, due to improved bandwidth in the main spool and a reduction in both sensor
and micro-controller costs, electronic versions of ow and pressure control are becoming
more common. In computed ow control, the pressure drop across the main spool
is measured using built in pressure transducers and the main spool orice opening
controlled to maintain a constant ow. This is discussed in more details in (Pedersen
and M unzer, 2001). A further good reference is (Backe and Feigel, 1990). This type of
ow control valve is commercially available from Ultronics Ltd.
Computed pressure control can also be implemented in valves with a fast main spool
and a built in pressure transducer. There are many dierent possibilities of controlling
the pressure. Two good references for this type of controller are (Eriksson, 1996) and
(Mattila and Virvalo, 1997b).
Appendix D
Review of Different Controllers
Chapter Contents
D.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
D.2 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
177
D.1. Review 179
D.1 Review
With the addition of a micro-controller and a faster main spool, many dierent options
of control become possible. This section presents a brief overview of some of the many
options available.
The most popular industrial controller is the PID (Proportional, Integral, Derivative)
controller. This type of controller uses the error between a reference value and the
measured system response to drive the system. A good overview of this type of con-
troller is (Virvalo, 2001). PID controllers are popular due to their ease of tuning and
robustness. However, the performance of PID controllers is limited due to xed gains
and limited feedback.
A more powerful method of control is based on state feedback control, often called pole
placement control. This type of control uses feedback from the dierent states of the
systems in order to place the poles of the closed loop system at desired locations. A
good example of this type of controller is presented in (Virtanen, 1993). The drawback
of this type of controller is that more information about the system is required. In the
typical cylinder position servo example, position, velocity, and pressure feedback are
neccessary. Some strategies dierentiate on the position signal which can cause noise
problems. (Virvalo et al., 1997) presents some dierent approaches to state feedback
control which reduce the need for dierentiation. (Backe and Boes, 1991) presents
a scheme for computing the gains of the feedback matrix depending on an identied
model of the plant. Observers can be used to estimate the states which cannot be
directly measured. Observers, however, have dynamics of their own which limit the
performance of the system.
As more eort was applied to understanding state feedback control the eld of optimal
control emerged. In optimal control, the gains of a state feedback matrix are calculated
in order to optimize some weighting function. The weighting function is usually spec-
ied as the square of some state variable, hence optimal controllers are often termed
LQ (Linear Quadratic) controllers. An LQG (Linear Quadratic Gausian) controller is
an extension of the LQ Controller where the state variables are found via an optimal
observer (Kalman lter).
One drawback of state feedback schemes is that they are based on a static model of
the plant. When the plant parameters change the gain matrix is no longer correct.
A solution to this problem are adaptive controllers. An adaptive controller adapts
the controller gains as the operating point of the plant changes. Some adaptive con-
trollers such as (Krus and Gunnarsson, 1990) estimate the plant dynamics online using
system identication techniques and adjust the gains accordingly. In many adaptive
control schemes, the eort is on reducing sensor count in order to lower the sensor
costs. However, the dynamic performance is limited because the controller needs to
adjust its online model after a parameter change. This is a standard problem for adap-
tive controllers with low sensor counts. Stability is also a problem for some adaptive
controllers.
180 Appendix D. Review of Dierent Controllers
Another type of controller is one where the dynamics of the plant are forced to follow a
reference model. This type of controller is sometimes called a Linear Model Following
controller, where the reference signal is tted to a linear model before it is compared
with the actual system output. Another type is called a Model Reference Adaptive
controller. The principle is the same as a linear model following controller however,
the error is used to tune the gains of a state feedback controller. An example of this
type of controller is discussed in (Tochizawa and Edge, 1999).
Another type of control which is very popular in the robotics literature is often called
computed torque control. It is based on a dynamic model of the structure where the
forces required to move the manipulator along a desired trajectory are calculated and
fed forward to a force control servo. Two examples of force control servos are (Virvalo
et al., 1999) and (Eriksson and Wikander, 1994). Errors in the feed-forward force
component are compensated by closed loop velocity or position controllers. The main
idea is that the feed-forward term reduces the control eort necessary by the closed
loop part. This improves the dynamics and compensates for inertial loads. Some
good examples of this type of strategy are (Mattila and Virvalo, 1997a), (Sepehri and
Lawrence, 1992), and (Zhou, 1995). The main problem with these types of strategies
is that using an incorrect model will result in large errors in the feed-forward part
resulting in large control actions which increase the risk of instability.
Recently there has been a large amount of interest in direct non-linear control schemes.
Ideas such as input output feedback linearization have been experimented with. An
example is given by (Hahn et al., 1994). In such a controller the controller attempts to
linearize the non linearities in the system with a special linearization. Traditional linear
control design methods can then be used. The problem however with input/output
feedback linearization controllers is that they require an accurate model since they
stabilize the model, not the actual plant. If the model is o, then there is no guarantee
on performance. These types of controllers usually also demand many feedback signals.
There are also a number of new techniques based on articial intelligence which show
a lot of promise. Controllers designed with Neural Networks such as (Hountras et
al., 1999) and (Fok and Ong, 1993), use a training algorithm to train the neural network.
This means that the controller parameters are tuned automatically. One major problem
with neural network controllers is that they are dicult to analyse. They are dependant
on their training set and if the training set is not constructed properly, they can exhibit
undesired behaviour.
Fuzzy logic is another technique which is based on articial intelligence. A fuzzy
controller uses a set of linguistic rules to dene a controller. Some examples of fuzzy
controllers in uid power systems are (Sepehri and Lawrence, 1998), (Corbet et al.,
1996), and (Zhao and Virvalo, 1993). Fuzzy controllers are in eect a gain scheduling
controller. This is very useful in hydraulic systems where the parameter variation is so
large. However, as with all gain scheduling controllers, it is dicult to verify stability
of a Fuzzy controller analytically.
D.2. Robustness 181
D.2 Robustness
Of special importance in controller design is the robustness of the controller. The ro-
bustness refers to how immune the controller is to both disturbances and to parameter
variations. Robustness analysis is especially important in the controller design of hy-
draulic systems such as mobile manipulators because the parameter change on a typical
manipulator is extremely large. Disturbances due to other joints and outside forces can
also be high.
Robust controllers based on non-linear control schemes are often based on a Lyapunov
analysis of the system and nding a control law which guarantees global asymptotic
stability. An example of this is (Sohl and Bobrow, 1999). Lyapunov analysis has
also resulted in some synthesis approaches like backstepping control and sliding mode
control. An example of a backstepping controller is given in (Alleyne, 1998). Sliding
mode control is a special case of backstepping control.
One method which has received a lot of attention over the last few years is H

con-
trol. In its most basic form, H

control is an extension of the LQG controller, where


instead of minimizing the square of the error, it minimizes the maximum error. This
results in a more conservative controller than the standard LQG controller. The LQG
controller is also often called the H
2
controller. The interest in the H

controller is
that it guarantees a bound on the maximum error given a known uncertainty on the
measurement or parameters. The H

technique is used together with classical loop


shaping techniques to give the best results. An example of an H

controller is given by
(Njabeleke et al., 1997). Another example of H

control is (Cheng and Moor, 1994).


H

is a synthesis technique, given a model and a set of uncertainties, applying the


technique results in a set of controller gains. Another approach to robust control is
one based on analysis of the parameter space. A parameter space approach tests the
stability of a designed controller given the known parameter variations. Two examples
of this approach are (Pritschow et al., 1997) and (Grabbel, 2002).
Appendix E
Flow Sharing
Chapter Contents
E.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
E.1.1 Cylinder Velocity . . . . . . . . . . . . . . . . . . . . . . . . 186
183
E.1. Overview 185
E.1 Overview
Since there is a limited amount of ow to the actuators, it is possible that in some cases,
there is not enough ow to keep all the actuators moving at the reference velocity. The
handling of this case is called Flow Sharing in the hydraulic terminology. This topic is
covered in more detail by Sepehri et al. (1992) and Andersen and Hansen (2001). The
solution adopted in this thesis is similar to that presented by Sepehri et al. (1992).
The problem is that hydraulic systems do not linearly decrease the ow to each actuator
as the system goes into ow limits. Instead, the cylinders will slow down one by one,
starting with the heaviest loaded joint. This results in the tool centre changing its speed
and direction. Since it is impossible to maintain both the speed and the direction of
the tool centre in a ow limit case, the speed is decreased so that the direction is
maintained.
The problem is dened as follows. The maximum possible ow output of the system,
q
m
is known. The ow references to all the actuators in the system, q
1
for the shoulder
joint, q
2
for the elbow joint, and q
3
for the extension system are also known. Therefore
it is possible to nd when the system goes into ow limits by the following relationship.
q
m
(q
1
+ q
2
+ q
3
) (E.1)
In the case where the required ow exceeds the available ow, the ow inputs can be
scaled by a scaling factor. The scaling factor can be dened as how much extra ow is
required versus how much is actually available:
q
1
actual
= Kq
1
ref
(E.2)
where:
K =
q
m
q
1
ref
+ q
2
ref
+ q
3
ref
(E.3)
It can be seen that scaling the joint velocities by a constant factor also scales the tool
centre velocity by a constant factor and maintains the direction of the tool centre. This
can be seen by looking at the forward kinematics equation.
x = J(k

) = kJ

(E.4)
186 Appendix E. Flow Sharing
E.1.1 Cylinder Velocity
However, note that the joint velocities are not directly proportional to cylinder speeds
since the joint speed is a function of both cylinder speed and joint position. The joint
angle and cylinder position relationship can be written as:
L
2
cyl
= L
2
b
+ L
2
a
2L
a
L
b
sin ( +
o
) (E.5)
Taking the derivative of this and solving for

gives:

=
2L
cyl
2L
a
L
b
sin ( +
o
)

L
cyl
(E.6)
The system jacobian can then be rewritten in terms of cylinder velocities instead of
joint velocities:
_
x
y
_
=
_
L
1
sin L
2
sin ( + ) L
2
sin ( + ) cos ( + )
L
1
cos + L
2
cos ( + ) L
2
cos ( + ) sin ( + )
_

(E.7)

_
2L
cyl
1
2L
a
1
L
b
1
sin (+
o
)

L
cyl
1
2L
cyl
2
2L
a
2
L
b
2
sin (+
o
)

L
cyl
2

L
cyl
3
_

_
From this equation, it can be seen that multiplying the cylinder velocities by a constant
scaling factor, will also scale the tool centre velocity by a constant factor and therefore
the tool centre direction will be maintained while the tool centre speed is reduced.
Appendix F
Description of Laboratory Facilities
Chapter Contents
F.1 HMF Crane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
F.2 Development System . . . . . . . . . . . . . . . . . . . . . . . . . . 190
F.3 Tool Centre Position Measurement System . . . . . . . . . . . . . 191
187
F.1. HMF Crane 189
This appendix describes the main laboratory facilities used in this thesis. The rst
section describes the hydraulic crane used in the experimental work. The second sec-
tion describes the development environment and the software used. The third section
describes the system used to measure the actual tool centre position.
F.1 HMF Crane
Figs. F.1F.2 show a detailed view of the mobile crane used for the experimental
implementation of the strategies presented in this thesis. The crane is built by Hojbjerg
Maskin Fabrik (HMF) A/S, a Danish crane manufacturer. Its range is about 10m with
a lifting capacity of 400kg at that range.
Figure F.1 Schematic diagram of laboratory crane.
Figure F.2 Mobile hydraulic crane installed in laboratory.
The crane has four hydraulic actuators, one for rotating the crane, one for the boom, one
for the elbow, and one for extending the arm. Fig. F.1 shows a schematic representation
190 Appendix F. Description of Laboratory Facilities
of the crane with all the major actuators listed. These actuators give the crane four
Degrees of Freedom (DOF). Due to space limitations in the laboratory, the rotation
degree of freedom was locked.
The crane was equipped with Sauer Danfoss PVG32 valves (www.sauer-danfoss.com).
These valves had ow control spools with a maximum ow of around 25L/min. One di-
rection over-centre valves were mounted at the elbow and shoulder joint. A regenerative
two direction over-centre valve was mounted on the telescopic extension beam.
A xed displacement pump was used for the power supply. Extra ow was dumped to
tank via the Sauer-Danfoss pump side module in the valve block. The power supply
could deliver a maximum of 30L/min at 300 Bar.
F.2 Development System
The development system used to implement the control algorithms in this thesis was
a dual processor system. A PC with a Microsoft Windows operating system was used
to provide a graphical user interface to the operator. A Blacktip PCI Digital Signal
Processor (DSP) card from Bittware Inc (www.bittware.com) was used to implement
the actual control algorithms. An Analog to Digital (AD) and Digital to Analog (DA)
converter card was connected to the DSP card. The sensors were connected to the
converter card via a signal conditioning box. Dual port RAM on the DSP card could
be accessed by both the DSP and the PC. The set-up is shown in g. F.3.
Figure F.3 Diagram of the development system setup.
The strength of this system is that the DSP can be programmed to give full real-time
control while retaining the possibility for a full graphical user interface on the PC. The
Dual Port RAM allows the PC to read and write from the memory of the DSP without
disturbing the operation of the DSP.
F.3. Tool Centre Position Measurement System 191
When the system is started, the DSP starts running the control code and the PC starts
running the user interface. The user interface allows the user to change constants
in the controller or read data from the controller. The drawback of most graphical
user interface systems is that the graphical user interface makes large demands on the
processor and makes real time operation impossible.
The graphical user interface was programmed in Borland C++ Builder which allowed
the creation of a very exible and powerful user interface. The DSP was programmed in
C. For more details of the development system please contact the author. The authors
email address is given at the start of this thesis.
F.3 Tool Centre Position Measurement System
The tool centre was measured directly via two wire displacement transducers. A wire
displacement transducer is made up of a spool onto which a thin wire is rolled. An en-
coder is connected to the spool. As the wire is pulled out, the spool turns. The encoder
keeps track of the rotation of the spool. The rotation of the spool is proportional to the
amount of wire pulled out. A rotational spring pulls the wire back in. The transducers
used gave 0.25mm accuracy and could measure up to 15m. Errors due to wire sag were
neglected due to the low weight of the wires and the sti spring in the transducer. The
transducers were placed around 10m apart on the ground with the wires connected to
the tool centre. This formed a triangle. Refer to g. F.4.
Figure F.4 Tool centre position measurement system.
Since the base length and the lengths of the two sides of the triangle are known, the x
192 Appendix F. Description of Laboratory Facilities
and y position of the tool centre can be calculated. Refer to the following equations:
x =
D
2
1
D
2
2
+ L
2
B
2L
B
(F.1)
y =
_
D
2
1
x
2
(F.2)
At certain positions the errors due to the curvature of the pulley wheels becomes a
problem, so an iterative solution was developed which reduced the errors to acceptable
levels. Please contact the author if more details are required.
Bibliography
Alleyne, Andrew (1998). A systematic approach to the control of electrohydraulic ser-
vosystems. In: Proc. of the American Control Conference. Philadelphia, Pennsyl-
vania.
Andersen, Torben Ole and Michael Rygaard Hansen (2001). Evaluation of velocity
control concepts involving counter balance valves in mobile cranes. In: Proceedings
of the Fifth International Conference on Fluid Power Transmission and Control
(ICFP). Hangzhou, China. pp. 311315.
Andersen, Torben Ole, Marc M unzer and Michael Rygaard Hansen (2001). Evaluations
of control strategies for separate meter-in separate meter-out hydraulic boom ac-
tuation in mobile applications. In: 17th International Conference on Hydraulics
and Pneumatics.
Andersson, Bo R. and John L. Ayres (1997). Load sensing directional valves, cur-
rent technology and future development. In: The Fifth Scandinavian International
Conference on Fluid Power. Vol. 1. pp. 99119.
Backe, W. and C. Boes (1991). Automatic controller conguration and adjustment
for servohydraulic plants. In: IFAC Computer Aided Design in Control Systems.
Swansea, UK. pp. 371374.
Backe, W. and H.-J. Feigel (1990). New possibilities with electrohydraulic load sensing.
In: O+P

Olhydraulik und Pneumatik 34.
Becker, Uwe (1997). The behaviour of position controlled actuator with switching valves
and a fuzzy controller. In: The Fifth Scandinavian International Conference on
Fluid Power, SICFP97. Vol. 2. Linkoping, Sweden. pp. 149157.
Beiner, L. (1997). Minimum force redundancy control of hydraulic cranes. Mechatronics
7(6), 537547.
Beiner, L. and J. Mattila (1999). An improved pseudoinverse solution for redundant
hydraulic manipulators. Robotica 17, 173179.
Bhattacharyya, S.P., H. Chapellat and L.H. Keel (1995). Robust Control, a Parametric
Approach. Prentice Hall PTR.
193
194 Bibliography
Bonchis, Adran, Peter I. Corke and David C. Rye (1999). A pressure based, velocity
independant, friction model for asymmetric hydraulic cylinders. In: Proceedings
of the 1999 IEEE Int. Conf. on Robotics and Automation. Detroit, Michigan.
Caterpillar (1999). System and method for controlling an independent metering valve.
United States Patent, 5,497,140, and 5,960,695.
Cheng, Yi and Bart L.R. De Moor (1994). Robustness analysis and control system
design for a hydraulic servo system. In: IEEE Trans. On Control Systems Tech-
nology. Vol. 2, No. 3. pp. 183197.
Corbet, T., N. Sepehri and P.D. Lawrence (1996). Fuzzy control of a class of hydrauli-
cally actuated industrial robots. In: IEEE Trans. On Control Systems Technology.
Vol. 4, No. 4. pp. 419426.
Craig, J.J. (1986). Introduction to Robotics: Mechanics and Control. Addison-Wesley.
Reading, Mass.
Djaferis, Theodore E. (1995). Robust Control Design - A Polynomial Approach. Kluwer
Academic Publishers.
Edge, K. A. (1997). The control of uid power systems - responding to the challenges.
Proceedings of the Institution of Mechanical Engineers 211(Part I), 91110.
Elfving, Magnus (1997). On Fluid Power Control, A Concept for a Distributed Con-
troller of Fluid Power Actuators. PhD thesis. Dept. of Mech. Eng., Linkoping
University, Linkoing, Sweden.
Ellman, A., T. Kappi and M. Vilenius (1996). Simulation and analysis of hydraulically-
driven boom mechanism. In: Ninth Bath Intern. Fluid Power Workshop.
Eriksson, B. (1996). Optimal Force Control to Improve Hydraulic Drives. PhD thesis.
Royal Inst. of Tech., Stockholm, Sweden.
Eriksson, Bengt and Jan Wikander (1994). Force controlled hydraulic cylinder towards
a modular servo unit. In: 20th International Conference on Industrial Electronics,
Control and Instrumentation. IEEE. pp. 16451649.
Fok, S.C. and E.K. Ong (1993). Neural network and uid power control. In: The Third
Scandinavian International Conference on Fluid Power. Vol. 2. pp. 139153.
Goldenberg, A.A:, J. Wiercienski, P. Kuzan, C. Szymczyk, R.G. Fenton and B. Shaver
(1995). A remote manipulator for forestry operation. IEEE Transactions on
Robotics and Automation.
Grabbel, Jorg (2002). Robust controller design for hydraulic joint servo actuators. In:
Int. FPNI PhD Symposium. Modena, Italy.
Grabbel, Jorg and Monika Ivantysynova (2001). Control strategies for joint integrated
servo actuators in mobile machinery. In: The 7th Scandinavian International Con-
ference on Fluid Power.
Bibliography 195
Hahn, H., A. Piepenbrink and K.D. Leimbach (1994). Input/output linearization con-
trol of an electro servo-hydraulic actuator. In: Proceedings of the Third IEEE
Conference on Control Applications. IEEE. pp. 9951000.
Honegger, Marcel and Peter Corke (2001). Model based control of hydraulically actu-
ated manipulators. In: Proceedings of the 2001 IEEE International Conference on
Robotics and Automation. IEEE. Seoul, Korea. pp. 25532559.
Hountras, A., I. Antoniadis and A. Kanarachos (1999). Implementation of n-step-ahead
neurocontrol on a 3-axes heavy duty hydraulic manipulator. Mechatronics 9, 235
270.
Kappi, Timo, Asko Ellman and Matti Vilenius (2001). Simulation study of a mo-
bile machine with special reference to energy eciency. In: Proceedings of the
Fifth International Conference on Fluid Power Transmission and Control (ICFP).
Hangzhou, China. pp. 378382.
Krus, P., A. Jansson and J.-O. Palmberg (1991). Real-time simulation of hydraulic
control systems with complex mechanical loads. In: IFAC Computer Aided Design
in Control Systems. Swansea, UK. pp. 351357.
Krus, P. and S. Gunnarsson (1990). Adaptive control of hydraulic actuators with exi-
ble mechanical loads. In: Fluid Power - Proceedings of the 9th International Sym-
posium. Oxford, UK. pp. 149158.
Krus, P. and S. Gunnarsson (1993). Adaptive control of a hydraulic crane using on-line
identication. In: Elfving (1997). pp. 363387.
Krus, Petter and Jan-Ove Palmberg (1992). Vector control of a hydraulic crane. In:
SAE Technical Paper Series.
Lantto, B. (1994). On Fluid Power Control - with Special Reference to Load-Sensing
Systems and Sliding Mode Control. PhD thesis. Dept. of Mech. Eng., Linkoping
University, Sweden.
Lawrence, Peter D. and Robert V. Ross (1991). Articulated arm control. In: US Pat.
Nr. 5,062,755.
Lawrence, Peter D., Farrokh Sassani, Brent Sauder, Nariman Sepehri, Ulrika Waller-
steiner and Jack Wilson (1993). Computer-assisted control of excavator-based ma-
chines. In: International O-Highway and Powerplant Congress and Exposition.
Milwaukee, Wisconsin.
Lee, Sung-Uk and Pyung Hun Chang (2001). Control of heavy-duty robotics excavator
using time delay control with switching action with integral sliding surface. In:
Proc. of the 2001 IEEE Int. Conf. on Robotics and Automation.
Liang, Xingui and Tapio Virvalo (2001). Energy reutilization and balance analysis in
a hydraulic crane. In: Proceedings of the Fifth International Conference on Fluid
Power Transmission and Control (ICFP). Hangzhou, China. pp. 306310.
196 Bibliography
Liang, Xingui, Tapio Virvalo and Matti Linjama (1999). The inuence of control valves
on the eciency of a hydraulic crane. In: The Sixth Scandinavian International
Conference on Fluid Power SICFP99. Tampere, Finland. pp. 381394.
Linjama, Matti (1998). The Modelling and Actuator Space Control of Flexibile Hy-
draulic Cranes. PhD thesis. Tamper University of Technology. Tampere, Finland.
Linjama, Matti and Tapio Virvalo (1997). Modelling of a exible hydraulic crane.
In: The Fifth Scandinavian International Conference on Fluid Power. Linkoping,
Sweden. pp. 3348.
Linjama, Matti and Tapio Virvalo (1999). State-space model for control design of multi-
link exible hydraulic cranes. In: The Sixth Scandinavian International Conference
on Fluid Power, SICFP99. Tampere, Finland. pp. 981998.
Linyi, Gu and Wang Qingfeng (2001). Research on noninterference control of multiple
actuators load sensing system with high inertia in construction machinery. In:
Proceedings of the Fifth International Conference on Fluid Power Transmission
and Control (ICFP). Hangzhou, China. pp. 373377.
Makinen, Jari, Asko Ellman and Robert Piche (1997). Dynamic simulations of exi-
ble hydraulic-driven multibody systems using nite strain beam theory. In: The
Fifth Scandinavian International inproceedings on Fluid Power, SICFP97. Vol. 2.
Linkoping, Sweden. pp. 119134.
Mattila, J. and T. Virvalo (1997a). Computed force control of hydraulic manipulators.
In: 5th Scandinavian Inter. Conf. On Fluid Power. pp. 139154.
Mattila, Jouni and Tapio Virvalo (1997b). Computed force control of hydraulic ser-
vodrive with novel LS system. In: International Conference on Fluid Power.
Hangzhou, China. pp. 97102.
Mattila, Jouni and Tapio Virvalo (2000). Energy ecient motion control of a hydraulic
manipulator. In: International inproceedings on Robotics and Automation. San
Fransisco, CA.. pp. 30003006.
Mikkola, A. and H. Handroos (1996). Modelling and simulation of a exible hydrauli-
cally driven log crane. In: Ninth Bath International Fluid Power Workshop - Fluid
Power Systems.
M unzer, Marc and Peder Pedersen (2001). Real time simulation model of exible mobile
crane for machine-operator interaction testing. In: 2nd International Workshop on
Computer Software for Design, Analysis, and Control of Fluid Power Systems.
Niemela, Esa and Tapio Virvalo (1994). Fuzzy logic assisted manual control of joystick
operated hydraulic crane. In: IEEE International Conference on Fuzzy Systems.
Vol. 1. IEEE. pp. 642647.
Nilsson, Bernt, Jonas Nygards, Ulf Larsson and

Ake Wernersson (1999). Control of
exible mobile manipulators: Positioning and vibration reduction using an eye-in-
hand range camera. Control Engineering Practice 7, 741751.
Bibliography 197
Njabeleke, I.A., R.F. Pannet, P.K. Chawdhry and C.R. Burrows (1997). H innity
control in uid power. In: IEEE Colloquium on Robust Control: Theory, Software
and Applications.
Oshina, Morio (1997). The tendency of servo control for hydraulic excavator. In: The
Fifth Scandinavian International Conference on Fluid Power. Vol. 3. pp. 271285.
Paoluzzi, R., G.L. Zarotti and M. Poldi Allaj (1997). Flow control systems for mo-
bile applications. In: The Fifth Scandinavian International Conference on Fluid
Power. Linkoping, Sweden. pp. 365380.
Parker, N.R., S.E. Salcudean and P.D. Lawrence (1993). Application of force feed-
back to heavy duty hydraulic machines. In: Proceedings of the IEEE International
Conference on Robotics and Automation. IEEE. pp. 375381.
Pedersen, Peder and Marc E. M unzer (2001). Strategies for stabilization of ow control
systems with counter balance valves. In: Drives and Controls.
Persson, Tomas (1989). On Fluid Power Control - With Special Reference to Over-
Center Valves. PhD thesis. Dept. of Mech. Eng., Linkoping University, Linkoping,
Sweden.
Pritschow, G., S.E. McCormac, B. Hiller and J. Zeiher (1997). Improving the dynamic
behaviour of a servodriven electro-hydraulic robotic manipulator. In: Proceedings
of 14th International Symposium on Automation and Robotics in Construction
(ISARC).
Rahmfeld, Robert and Monika Ivantysynova (2001). Displacement controlled linear ac-
tuator with dierential cylinder - a way to save primary energy in mobile machines.
In: Proceedings of the Fifth International Conference on Fluid Power Transmis-
sion and Control (ICFP). Hangzhou, China. pp. 296301.
Rouvinen, Asko and Heikki Handroos (1997). Deection compensation of a exible
hydraulic manipulator utilizing neural networks. Mechatronics 7(4), 355368.
Sato, S., H. Konno, K. Kobayashi and M. Takasu (1993). Master-slave manipulation
system for a backhoe excavator. In: The Third Scandinavian International Con-
ference on Fluid Power. Vol. 2. pp. 349361.
Sepehri, N. (1990). Dynamic Simulation and Control of Teleoperated Heavy Duty Hy-
draulic Manipulators. PhD thesis. Vancouver, B.C., Canada.
Sepehri, N. and P.D. Lawrence (1992). Model-based sensor-based velocity control
of teleoperated heavy-duty hydraulic machines. In: Proceedings of the 1992
IEEE/RSJ International Conference on Intelligent Robots and Systems. Raleigh,
NC. pp. 859864.
Sepehri, N. and P.D. Lawrence (1998). Fuzzy logic control of a teleoperated log loader
machine. In: Proc. Of the 1998 IEEE/RSJ Intl. Conf. On Intelligent Robots and
Systems. Victoria, B.C.. pp. 1571 1577.
198 Bibliography
Sepehri, N., F. Sassani and P.D. Lawrence (1990). On numerical simplication of robot
structure dynamics for fast simulation. In: Proceedings of the 9th IASTED Inter-
national inproceedings on Modelling, Identication and Control. Innsbruck, Aus-
tria. pp. 311315.
Sepehri, N., P.D. Lawrence, F. Sassani and R. Frenette (1994). Resolved-mode tele-
operated control of heavy-duty hydraulic machines. Journal of Dynamic Systems,
Measurement, and Control 116, 232240.
Sepehri, Nariman, Real N. Frenette and Peter D. Lawrence (1992). Proportional hy-
draulic control. In: US Pat. Nr. 5,167,121.
Singh, N., H. Zghal, N. Sepehri, S. Balakrishnan and P.D. Lawrence (1995).
Coordinated-motion control of heavy duty industrial machines with redundancy.
Robotica 13, 623633.
Sohl, Garett A. and James E. Bobrow (1999). Experiments and simulations on the
nonlinear control of a hydraulic servosystem. In: IEEE Trans. On Control Systems
Technology. Vol. 7 No. 2. pp. 238247.
Tochizawa, M. and K.A. Edge (1999). A comparison of some control strategies for
a hydraulic manipulator. In: Proceedings of the American Control inproceedings.
San Diego, CA. pp. 744748.
Touminen, Pasi and Tapio Virvalo (1993). The control of hydraulic mobile crane using
partly and fully distributed control system. In: The Third Scandinavian Interna-
tional inproceedings on Fluid Power. Vol. 2. pp. 105120.
Ultronics (1999). Electrohydraulic proportional control valve assemblies. European
Patent EP 0 809 737 B1.
Virtanen, A. (1993). The design of state controlled hydraulic position servo system.
In: The Third Scandinavian International Conference on Fluid Power. Vol. 2.
pp. 193206.
Virvalo, Tapio (2001). PI and PID controllers in a hydraulic position servo system -
what it is all about. In: Proceedings of the Fifth International inproceedings on
Fluid Power Transmission and Control (ICFP). Hangzhou, China. pp. 130134.
Virvalo, Tapio (99). On the damping of a hydraulic cylinder drive. In: The sixth Scan-
dinavian International Conference on Fluid Power. Tampere, Finland.
Virvalo, Tapio, Esa Makinen and Matti Vilenius (1999). Force control of a water hy-
draulic cylinder drive. In: 6th Scandinavian Int. Conf. on Fluid Power. Tampere,
Finland.
Virvalo, Tapio, Matti Linjama and Jouni Mattila (1997). Comparing dierent con-
trollers of an electrohydraulic position servo. In: International inproceedings on
Fluid Power. Hangzhou, China. pp. 215219.
Widner, T.C. and W.R. Hamel (1998). Fundamental control concepts for implementa-
tion of transmission based actuators in robotics and automation. In: Proc. Of the
Bibliography 199
1998 IEEE Int. Conf. On Robotics and Automation. Leuven, Belgium. pp. 3594
3600.
Yokota, S., Y. Kondoh and K. Yoshida (1996). Pressure control valves in application
of an electro-rheological uid (application to drive a one-link manipulator). In:
Ninth Bath International Fluid Power Workshop. pp. 229 245.
Zhao, Tienan and Tapio Virvalo (1993). Fuzzy control of a hydraulic position servo
with unknown load. In: Second IEEE International Conference on Fuzzy Systems.
pp. 785788.
Zhou, Jianjun (1995). Experimental evaluations of a kinematic compensation control
method for hydraulic robot manipulators. Control Eng. Practice 3(5), 675684.

Вам также может понравиться