Name of Author: Ryan R. Chladny
University of Alberta
Library Release Form
Name of Author: Ryan R. Chladny
Title of Thesis: Modeling and Control of Automotive Gas Exchange Valve
Solenoid Actuators
Degree: Doctor of Philosophy
Year this Degree Granted: 2007
Permission is hereby granted to the University of Alberta to reproduce single copies
of this thesis and to lend or sell such copies for private, scholarly, or scientific research
purposes only.
The author reserves all other publication and other rights in association with the
copyright in the thesis, and except as hereinbefore provided, neither the thesis nor any
substantial portion thereof may be printed or otherwise reproduced in any material
form whatever without the author’s prior written permission.
Ryan R. Chladny
7924 91 Ave
Edmonton, Alberta,
T6C 1R1, Canada
Submission Date
University of Alberta
Modeling and Control of Automotive Gas Exchange Valve Solenoid
Actuators
by
Ryan R. Chladny
A thesis submitted to the Faculty of Graduate Studies and Research in partial
fulfillment of the requirements for the degree of Doctor of Philosophy.
Department of Mechanical Engineering
Edmonton, Alberta
Fall 2007
University of Alberta
Faculty of Graduate Studies and Research
The undersigned certify that they have read, and recommend to the Faculty of Graduate Studies and Research for acceptance, a thesis entitled Modeling and Control of
Automotive Gas Exchange Valve Solenoid Actuators submitted by Ryan R. Chladny
in partial fulfillment of the requirements for the degree of Doctor of Philosophy.
Dr. C.R. (Bob) Koch
Dr. Jeff Pieper
Dr. Tongwen Chen
Dr. Alan Lynch
Dr. M. David Checkel
To my wife Melanie. Thank you for your continued love, understanding and
countless personal sacrifices that you have made in order for me to further my
education.
Abstract
A promising method for enhancing automotive internal combustion engine efficiency uses solenoid actuators to directly control the gas exchange
valves. Mitigation of valve seating velocities is challenging due to phenomena such as magnetic saturation and combustion gas force disturbances.
A comprehensive control strategy is presented for a hinged solenoid actuator. Gas forces on the exhaust valve are particularly problematic due to
the potential for large cycle-to-cycle variations. Soft seating is achieved
using a flatness-based landing algorithm with a nonlinear disturbance estimator. The estimator is used with an energy-based feedforward controller
to reject exhaust gas force disturbances. Feedback is provided through
the use of flux and current measurements and an accurate inductance
model. Overviews of the employed modeling and simulation techniques
and experimental testbench results are also presented. Both simulated and
experimental results indicate the proposed control methodology is capable
of compensating for the nonlinear magnetic dynamics and combustion gas
force disturbances experienced by exhaust valve solenoid actuators.
Acknowledgements
I would like to take this opportunity to acknowledge the following individuals and organizations who helped me during the course of this work.
Dr. Koch for his willingness to help, patience and understanding throughout my career as a graduate student. Your positive attitude, contribution
of time, insight and invaluable wealth of knowledge certainly made all of
my challenges much easier to cope with as they arose.
Dr. Lynch and Thomas Grochmal for their controls expertise and advice.
Dr. Kostiuk and the Department of Mechanical Engineering for continued financial aid and genuine concern during my mountain bike accident
recovery.
DaimlerChrysler AG for the generous donation of the solenoid actuators,
schematics and single cylinder head.
No amount of thanks (or donuts) can repay technicians, Bernie Faulkner,
Greg Miller and Rick Bubenko or machinists, Dirk Kelm, Albert Yeun
and Dave Pape for their expert skills, advice and the countless occasions
where they sacrificed time from their projects for my own.
Thanks are also deserved to Terry Nord and Bill Bizuk for their assistance
and support with the various custom electronic devices that were accrued
over the course of this work.
Thanks to my colleagues of room 4-28 for advice, support and otherwise
keeping moral levels up.
Finally, I would like to thank my parents for their continued love and
encouragement throughout all of my challenges, academic or otherwise.
Table of Contents
1 Introduction
1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.2 Variable Valve Timing . . . . . . . . . . . . . . . . . . . . . . . . . .
2
1.2.1
Electromagnetic Variable Valve Timing . . . . . . . . . . . . .
3
1.3 Problem Identification and Research Scope . . . . . . . . . . . . . . .
7
1.3.1
Contribution
. . . . . . . . . . . . . . . . . . . . . . . . . . .
2 VVT Actuator & Control Technology
8
11
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
2.2 Various Actuator Technologies . . . . . . . . . . . . . . . . . . . . . .
11
2.2.1
Pneumatic & Hydraulic Actuators . . . . . . . . . . . . . . . .
12
2.2.2
Rotary & Linear Motors . . . . . . . . . . . . . . . . . . . . .
12
2.2.3
Piezoelectric Actuators . . . . . . . . . . . . . . . . . . . . . .
14
2.2.4
Electroactive Polymer Actuators . . . . . . . . . . . . . . . . .
14
2.2.5
Solenoid Actuators . . . . . . . . . . . . . . . . . . . . . . . .
15
2.3 Feedback Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
2.4 Actuator Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
2.4.1
Lumped Parameter Models . . . . . . . . . . . . . . . . . . . .
19
2.4.2
Finite Element Analysis . . . . . . . . . . . . . . . . . . . . .
22
2.4.3
Mechanical System Modeling . . . . . . . . . . . . . . . . . .
23
2.5 Control
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
2.5.1
Landing Control
. . . . . . . . . . . . . . . . . . . . . . . . .
24
2.5.2
Feedforward Control . . . . . . . . . . . . . . . . . . . . . . .
26
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
3 Theory
32
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
3.2 Finite Element Modeling . . . . . . . . . . . . . . . . . . . . . . . . .
32
3.2.1
Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . .
33
3.2.2
Magnetic Vector Potential . . . . . . . . . . . . . . . . . . . .
38
3.2.3
Static Elements . . . . . . . . . . . . . . . . . . . . . . . . . .
43
3.2.4
Transient Elements . . . . . . . . . . . . . . . . . . . . . . . .
45
3.2.5
Element Shape Functions
. . . . . . . . . . . . . . . . . . . .
47
3.2.6
Solution Process
. . . . . . . . . . . . . . . . . . . . . . . . .
49
3.2.7
Matrix Assembly . . . . . . . . . . . . . . . . . . . . . . . . .
49
3.2.8
Static Model Solution . . . . . . . . . . . . . . . . . . . . . . .
52
3.2.9
Transient Model Solution . . . . . . . . . . . . . . . . . . . . .
53
3.3 Lumped Parameter Modeling . . . . . . . . . . . . . . . . . . . . . .
54
3.3.1
Reluctance Network
. . . . . . . . . . . . . . . . . . . . . . .
55
3.3.2
Nonlinear Induction
. . . . . . . . . . . . . . . . . . . . . . .
57
3.3.3
Electric Coupling . . . . . . . . . . . . . . . . . . . . . . . . .
58
3.3.4
Magnetic Co-energy
. . . . . . . . . . . . . . . . . . . . . . .
58
3.4 Simulink Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
60
3.5 Differential Flatness
. . . . . . . . . . . . . . . . . . . . . . . . . . .
61
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
62
4 Modeling and Simulation
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
65
65
4.2 3D Solid Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . .
66
4.3 Finite Element Modeling . . . . . . . . . . . . . . . . . . . . . . . . .
68
4.3.1
Static Modeling and Simulation . . . . . . . . . . . . . . . . .
70
4.3.2
Transient Modeling and Simulation . . . . . . . . . . . . . . .
72
4.4 Plant Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
73
4.4.1
Magnetic Subsystem . . . . . . . . . . . . . . . . . . . . . . .
73
4.4.2
Electric Subsystem . . . . . . . . . . . . . . . . . . . . . . . .
76
4.4.3
Magnetic Force Calculation . . . . . . . . . . . . . . . . . . .
78
4.4.4
Mechanical System . . . . . . . . . . . . . . . . . . . . . . . .
79
4.4.5
State Space Formulation . . . . . . . . . . . . . . . . . . . . .
81
4.5 Gas Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
82
4.5.1
Gas Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . .
82
4.5.2
Gas Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
4.6 Simulink Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
89
4.6.1
Power Supply and Electronics . . . . . . . . . . . . . . . . . .
89
4.6.2
Testbench Model . . . . . . . . . . . . . . . . . . . . . . . . .
89
4.6.3
Mechanical Model . . . . . . . . . . . . . . . . . . . . . . . . .
91
4.6.4
Gas Force Disturbance Model . . . . . . . . . . . . . . . . . .
92
4.6.5
dSPACE Controller Model . . . . . . . . . . . . . . . . . . . .
92
4.7 Computer Software & Hardware . . . . . . . . . . . . . . . . . . . . .
93
4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
94
5 Control Design
95
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
5.2 Controller Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
5.3 Initialization and Holding Control . . . . . . . . . . . . . . . . . . . .
99
5.3.1
Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . .
99
5.3.2
Holding Control . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.4 Closed-loop Landing Control . . . . . . . . . . . . . . . . . . . . . . . 100
5.4.1
Plant Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.4.2
Linear Full State Feedback - Linear Induction System . . . . . 101
5.4.3
Linear Full State Feedback - Nonlinear Induction . . . . . . . 106
5.4.4
Proportional-Integral Control . . . . . . . . . . . . . . . . . . 108
5.4.5
Flatness-based Voltage Control - Nonlinear Induction . . . . . 109
5.4.6
Preliminary Control Law Comparison . . . . . . . . . . . . . . 111
5.5 Reference Trajectory Design . . . . . . . . . . . . . . . . . . . . . . . 117
5.5.1
Nonlinear Constrained Problem . . . . . . . . . . . . . . . . . 118
5.5.2
Parametrization of the Flat Output Trajectory . . . . . . . . . 121
5.6 Flux-based Position Reconstruction . . . . . . . . . . . . . . . . . . . 123
5.6.1
Flux Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.7 Feedforward Controller . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.7.1
System Energy Derivation . . . . . . . . . . . . . . . . . . . . 129
5.7.2
Feedforward Current Input . . . . . . . . . . . . . . . . . . . . 132
5.8 Disturbance Observer . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
6 Experimental Setup
138
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.2 Material Testing Machine Experiments . . . . . . . . . . . . . . . . . 138
6.2.1
Static Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.2.2
Transient Evaluation . . . . . . . . . . . . . . . . . . . . . . . 141
6.3 Testbench Engine Emulator . . . . . . . . . . . . . . . . . . . . . . . 143
6.4 Preparation for Single Cylinder Engine Testing . . . . . . . . . . . . . 148
6.4.1
MicroAutobox Engine Controller and Interface Electronics . . 148
6.5 Equipment Description . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.5.1
Actuator Adapter and Load Rod . . . . . . . . . . . . . . . . 149
6.5.2
Circuit Protection . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.5.3
Computer Hardware and Software . . . . . . . . . . . . . . . . 151
6.5.4
Current and Voltage Sensing . . . . . . . . . . . . . . . . . . . 151
6.5.5
Cylinder Head . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.5.6
dSPACE DS1103 Controller . . . . . . . . . . . . . . . . . . . 153
6.5.7
Flux Sensor Integration Electronics . . . . . . . . . . . . . . . 154
6.5.8
Laser Position Sensor . . . . . . . . . . . . . . . . . . . . . . . 156
6.5.9
Laser and Pressure Sensor Mount . . . . . . . . . . . . . . . . 158
6.5.10 Load Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.5.11 Material Testing Machine . . . . . . . . . . . . . . . . . . . . 159
6.5.12 Pressure Regulator and Two-way Solenoid Valve . . . . . . . . 160
6.5.13 Pressure Transducer and Charge Amplifier . . . . . . . . . . . 160
6.5.14 Power Electronics . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.5.15 Power Supplies . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7 Results
167
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.2 Simulated and Measured Actuator Response . . . . . . . . . . . . . . 167
7.2.1
Static Performance Evaluation . . . . . . . . . . . . . . . . . . 168
7.2.2
Transient Performance Evaluation . . . . . . . . . . . . . . . . 170
7.2.3
Preliminary Flux Sensor Evaluation . . . . . . . . . . . . . . . 176
7.3 Simulated and Measured Controller Performance . . . . . . . . . . . . 180
7.4 Testbench System Model Validation . . . . . . . . . . . . . . . . . . . 181
7.4.1
Open-loop Feedforward Control . . . . . . . . . . . . . . . . . 183
7.4.2
Observer Convergence . . . . . . . . . . . . . . . . . . . . . . 185
7.4.3
Gas Disturbance Rejection . . . . . . . . . . . . . . . . . . . . 188
7.4.4
Simulated and Measured Control Performance . . . . . . . . . 190
7.5 Simulated Multi-cylinder Exhaust Manifold Pressure Disturbances . . 191
7.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
8 Conclusions and Further Research
200
8.1 Further Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
8.1.1
Engine Implementation . . . . . . . . . . . . . . . . . . . . . . 203
8.1.2
Control Development . . . . . . . . . . . . . . . . . . . . . . . 203
8.1.3
Alternative Actuators
Bibliography
A Supplemental Theory
. . . . . . . . . . . . . . . . . . . . . . 204
205
224
A.1 Vector Differential Calculus Operations & Notation . . . . . . . . . . 224
A.1.1 Gradient of a Scalar Function . . . . . . . . . . . . . . . . . . 224
A.1.2 Divergence of a Vector Field and the Laplacian Operator . . . 225
A.1.3 Curl of a Vector Field . . . . . . . . . . . . . . . . . . . . . . 226
A.2 Maxwell’s Equations Derived . . . . . . . . . . . . . . . . . . . . . . . 226
A.2.1 Coulomb’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . 226
A.2.2 Gauss’s Law of Electricity . . . . . . . . . . . . . . . . . . . . 227
A.2.3 Gauss’s Law for Magnetism . . . . . . . . . . . . . . . . . . . 228
A.2.4 Conservation of Charge . . . . . . . . . . . . . . . . . . . . . . 229
A.2.5 Ampére’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
A.2.6 Faraday’s Law of Induction . . . . . . . . . . . . . . . . . . . 230
A.3 Other Relations of Interest . . . . . . . . . . . . . . . . . . . . . . . . 231
A.3.1 Biot-Savart Law . . . . . . . . . . . . . . . . . . . . . . . . . . 231
A.3.2 Lenz’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
A.3.3 Lorentz Force . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
A.4 Magnetic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
A.5 Eddy Currents
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
A.6 The Lambert W Function . . . . . . . . . . . . . . . . . . . . . . . . 235
B Actuator Properties and Specifications
237
B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
B.2 Magnetic Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
B.3 Mechanical Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
B.3.1 Torsion Bar Force Measurement . . . . . . . . . . . . . . . . . 238
B.3.2 Testbench System Identification . . . . . . . . . . . . . . . . . 239
C Program and Data File Summary
245
C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
C.2 ANSYS Static Command and Result Files . . . . . . . . . . . . . . . 246
C.3 ANSYS Transient Command and Result Files . . . . . . . . . . . . . 247
C.4 Simulink Models and Parameter Files . . . . . . . . . . . . . . . . . . 247
C.5 Trajectory Optimization Files . . . . . . . . . . . . . . . . . . . . . . 248
C.6 Material Testing Machine Experimental Program and Data Files . . . 248
C.7 Testbench Experiment and Data Files . . . . . . . . . . . . . . . . . . 249
C.8 Primary Testbench Control Program Files . . . . . . . . . . . . . . . 250
C.9 Primary Analysis and Postprocessing Files . . . . . . . . . . . . . . . 251
List of Tables
4.1 Air gap and excitation operating points . . . . . . . . . . . . . . . . .
70
5.1 Reference trajectory constraints . . . . . . . . . . . . . . . . . . . . . 119
6.1 Experimental air gap and excitation operating points . . . . . . . . . 141
6.2 Material Testing Machine Experimental Equipment . . . . . . . . . . 165
6.3 Testbench Experimental Equipment . . . . . . . . . . . . . . . . . . . 166
B.1 Magnetic and Electric Lumped Model Parameters . . . . . . . . . . . 240
B.2 Mechanical Lumped Model Parameters . . . . . . . . . . . . . . . . . 244
C.1 ANSYS static command and result files . . . . . . . . . . . . . . . . . 246
C.2 ANSYS transient command and result files . . . . . . . . . . . . . . . 247
C.3 Simulink LPM files . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
C.4 MATLAB files used for generating landing control reference trajectories 248
C.5 Material testing machine experiment files . . . . . . . . . . . . . . . . 248
C.6 Testbench experimental and raw data files . . . . . . . . . . . . . . . 250
C.7 Primary testbench control program files . . . . . . . . . . . . . . . . . 250
C.8 Files and scripts used to analyze experimental results . . . . . . . . . 251
List of Figures
1.1 Schematic of prototype solenoid valve actuator . . . . . . . . . . . . .
4
1.2 Hinged electromagnetic gas exchange valve actuator . . . . . . . . . .
6
1.3 V-Cycle workflow of overall research objectives . . . . . . . . . . . . .
9
3.1 Magnetic vector potential interpretation . . . . . . . . . . . . . . . .
40
3.2 Static and transient model mesh and element types . . . . . . . . . .
44
3.3 Element Configurations . . . . . . . . . . . . . . . . . . . . . . . . . .
48
3.4 Cross-sectional schematic of hinged actuator indicating armature motion and magnetic flu
3.5 Schematic of Simulink modeling process
. . . . . . . . . . . . . . . .
60
3.6 Correspondence between flat output space and state space . . . . . .
63
4.1 3D solid model of the prototype actuator . . . . . . . . . . . . . . . .
66
4.2 3D solid model of the actuator, exploded view . . . . . . . . . . . . .
67
4.3 3D solid model of the actuator, cylinder head and custom engine cylinder 68
4.4 Modeled actuator flux path sections . . . . . . . . . . . . . . . . . . .
69
4.5 Static and transient model mesh and material types . . . . . . . . . .
71
4.6 Opener magnetic path . . . . . . . . . . . . . . . . . . . . . . . . . .
74
4.7 Valve flow area as a function of lift for valve seat diameters dos , dis . .
85
4.8 Comparison of the quadratic gas model to simulated, experimental testbench and engine†
4.9 Absolute error of the quadratic gas model with respect to experimental testbench results
4.10 Simulink Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
88
4.11 Simulink Model - Power Electronics . . . . . . . . . . . . . . . . . . .
90
4.12 Simulink Model - Testbench . . . . . . . . . . . . . . . . . . . . . . .
91
4.13 Simulink Model - Coil Dynamics . . . . . . . . . . . . . . . . . . . . .
92
4.14 Simulink Model - Mechanical Dynamics . . . . . . . . . . . . . . . . .
93
5.1 Control flowchart from closed to open and open to close . . . . . . . .
96
5.2 State machine flowchart [Chung, 2005] . . . . . . . . . . . . . . . . .
97
5.3 Control stages from closed to open position with respect to experimentally measured posi
5.4 Control flowchart from closed to open and open to close . . . . . . . .
99
5.5 Simulated linear time invariant landing control block diagram . . . . 106
5.6 Simulated proportion-integral current landing control block diagram . 109
5.7 Simulated flatness-based landing control block diagram . . . . . . . . 112
5.8 Landing control performance with 500 V source . . . . . . . . . . . . 113
5.9 Landing control performance with IC variation . . . . . . . . . . . . . 114
5.10 Simulated landing control performance with 42 V source . . . . . . . 115
5.11 Landing control performance with system spring variation . . . . . . 116
5.12 Landing control performance with system damping variation . . . . . 117
5.13 Optimized reference trajectories . . . . . . . . . . . . . . . . . . . . . 124
5.14 Reference trajectory coil voltage and current input
. . . . . . . . . . 125
5.15 Opener position with respect to current and flux . . . . . . . . . . . . 128
5.16 Simulated individual energy terms during an opening cycle . . . . . . 131
6.1 Hinged actuator performance evaluation experimental setup . . . . . 139
6.2 Single cylinder head test-bench setup . . . . . . . . . . . . . . . . . . 143
6.3 Cut-away view of testbench cavity . . . . . . . . . . . . . . . . . . . . 144
6.4 Sectional view of cylinder head test-bench cavity . . . . . . . . . . . . 145
6.5 Single cylinder head test-bench setup schematic . . . . . . . . . . . . 146
6.6 Ricardo Mark III single cylinder engine and test facility . . . . . . . . 149
6.7 dSPACE MicroAutobox and custom interface electronics . . . . . . . 150
6.8 Analog Drift of the RC Integration Circuit . . . . . . . . . . . . . . . 155
6.9 Laser sensor schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.10 Laser and pressure sensor mounting assembly . . . . . . . . . . . . . 158
6.11 Power electronic modes . . . . . . . . . . . . . . . . . . . . . . . . . . 162
7.1 FEA simulated opener force as a function of respective air gap and steady state current16
7.2 Experimentally measured opener force as a function of respective air gap and steady stat
7.3 FEA simulated closer force as a function of respective air gap and steady state current17
7.4 Opener magnet measured, FEA and LPM valve force as a function of armature position a
7.5 Opener magnet simulated and measured force error as a function of armature position an
7.6 Simulated and measured response of the opener magnet, 42 V, 0.50mm airgap173
7.7 Simulated and measured response of the opener magnet, 42 V, 0.75mm airgap174
7.8 Simulated and measured response of the opener magnet, 42 V, 1.00mm airgap175
7.9 Simulated and measured response of the opener magnet, 4 2V, 1.50mm airgap176
7.10 Simulated and measured response of the opener magnet, 42 V, 2.00 mm air gap177
7.11 Opener flux lines for a 1.50 mm airgap and 42 V step input . . . . . . 178
7.12 Opener flux density contour plots for a 1.50 mm airgap and 42 V step input179
7.13 Measured and estimated air gap during 10 Hz, 0.50 mm amplitude, 1.00 mm mean crossh
7.14 Measured and estimated air gap during 4 Hz, 1.50 mm amplitude, 2.5 mm mean crosshea
7.15 Simulated LPM-FEA and experimental free opening, 1 bar blowdown pressure182
7.16 Simulated LPM-FEA and experimental valve opening, 1 bar blowdown pressure183
7.17 Simulated LPM-FEA and experimental valve closing, 1 bar blowdown pressure184
7.18 Full cycle plots over a 0.25 to 4.5 bar pressure range, 250 rpm . . . . 185
7.19 Flatness voltage landing control via flux feedback - 3500 rpm, 1 bar blowdown, 6 cycles18
7.20 Simulated and experimental estimated position, velocity and pressure convergence, 1 bar1
7.21 Simulated and experimental estimated position, velocity and pressure convergence, 5 bar
7.22 Simulated and experimental position, pressure, current and desired feedforward current a
7.23 Simulated and measured opening cycles with cyclic pressure variations between 1 and 5 b
7.24 Simulated and measured opening cycles with cyclic pressure variations between 1 and 5 b
7.25 Simulated and experimental feedforward and landing results for 1 to 5 bar EVO pressures
7.26 Experimental impact velocity histogram at 1 bar, 200 cycles . . . . . 196
7.27 Experimental impact velocity histogram at 3 bar, 200 cycles . . . . . 197
7.28 Experimental impact velocity histogram at 5 bar, 200 cycles . . . . . 198
7.29 Simulated 1.5 bar disturbance response with secondary mid-stroke pressure disturbances1
A.1 Two real branches of the Lambert W function . . . . . . . . . . . . . 236
B.1 Induction curves of the various materials used in the hinged actuator magnetic path (larg
B.2 Induction curves of the various materials used in the hinged actuator magnetic path (sma
B.3 Torsion Bar Force Exerted on Valve Vs. Valve Position . . . . . . . . 241
B.4 Measured and predicted response during ‘cold’ operation . . . . . . . 242
B.5 Measured and predicted response after approximately 3000 cycles . . 243
Nomenclature
Note that bold typeface denotes a vector or matrix quantity.
ADC. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Analog to digital converter
α . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Saturable flux linkage model parameter [A−1 ]
At . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve position dependent effective
throat flow area [m2 ]
Av . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exhaust valve area [m2 ]
B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Magnetic field flux density [T]
BDC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Bottom dead centre
b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effective viscous damping coefficient
[ Ns
]
m
b̂ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viscous damping coefficient associated
with the armature [ Nms
]
rad
β . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flux linkage parameter [m/A]
bv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Viscous damping coefficient associated
with the valve [ Ns
]
m
E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Electric field [ mV2 ]
Cd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effective flow discharge coefficient
Cgf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effective gas force flow coefficient
DAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Digital to analog converter
ECU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Engine control unit
EMF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Electromotive force
F
ǫ0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Permittivity constant = 8.85 × 10−12 m
EVO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exhaust valve opening
FEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finite element analysis
FEM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Finite element method
Fm,c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic closer force on the armature
[N]
Fg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Gas force acting upon the valve [N]
Fgs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simplified gas force [N]
Fm,o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic opener force on the armature
[N]
Fv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve spring pre-load [N]
f1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Normalized gas force disturbance simplification
f2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Rate of change of the normalized gas
force disturbance simplification
γ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Initial gas force amplitude [N]
A
H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic field strength [ m
]
ICE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Internal combustion engine
IGBT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Insulated gate bipolar transistor
Io . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Armature and torsion bar moment of
inertia about the pivot point [kgm2 ]
i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coil current [A]
J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Current density [ mA2 ]
KVL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kirchoff voltage law
k1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Linear error dynamics controller gain
[1/s]
k2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Linear Error dynamics controller gain
[1/s2 ]
k3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Linear Error dynamics controller gain
[1/s3 ]
N
k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effective spring constant [ m
]
k̂ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Angular torsion bar spring constant
]
[ Nm
rad
κ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flux linkage parameter [m]
kt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Specific heat ratio of the in-cylinder
gas
N
kv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve spring constant [ m
]
L . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Inductance [H]
LHP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Left-hand plane
LPM. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Lumped parameter model
LTI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linear time invariant
LVDT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Linear
variable
differential
trans-
former
ℓv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Radial distance from the armature
pivot point to where the longitudinal
armature and valve axes intersect [m]
ℓm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Radial distance from the armature
pivot point to where the resultant
opener magnetic force acts on the armature [m]
λ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic flux linkage [Wb-turns]
M . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Local Mach number at the exhaust
valve throat
MMF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetomotive force [A-turns]
m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effective moving mass [kg]
mg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mass of the gas inside the cylinder [kg]
mv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Effective valve and moving spring
mass [kg]
H
µ0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Permeability constant = 4π × 10−7 m
µr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relative permeability
N . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Number of coil turns [turns]
Nf c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Number of turns of flux measurement
coil [turns]
Nec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Number of turns of excitation coil
[turns]
ν . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reluctivity [ m
]
H
ODE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Ordinatry differential equation
P . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cylinder pressure [Pa]
Patm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Atmospheric pressure [Pa]
Po . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Initial cylinder pressure [Pa]
PDE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Partial differential equation
PI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Proportional integral
PPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PowerPC
PSD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Position sensitive device
PWM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pulse width modulation
φ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Magnetic flux [Wb]
ψ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flux linkage model parameter [Wb]
QP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Quadradic programming
R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coil resistance [Ω]
RPM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Revolution per minute
ℜ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reluctance [ H1 ]
Rg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ideal gas constant for in-clinder gas
ρ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Electric charge density [ mC3 ]
S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Total valve stroke, 8 mm
SQP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sequential quadratic programming
σ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conductivity [S]
T . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . In-cylinder gas temperature [K]
TDC. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Top dead centre
θc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Crank angle [rad]
t0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Initial end-control time [s]
tf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Final end-control time [s]
u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Control input
V . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Electric scalar potential [V]
Vc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cylinder volume at valve opening [m3 ]
Vres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Combined cylinder head and crevice
volume [m3 ]
v . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Coil voltage [V]
VVT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Variable valve timing
Wc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Co-energy function [J]
x . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve displacement, ǫ{[−4mm, 4mm]}
xL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve lift ǫ{[−8mm, 0mm]}
xlc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Landing control engagement position
ẋ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve velocity [m/s]
ẍ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Valve acceleration [m/s2 ]
y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flat output [m]
yd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Desired position [m]
ẏd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Desired velocity [m/s]
ẏd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Desired acceleration [m/s2 ]
(3)
yd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Desired jerk [m/s3 ]
Chapter 1
Introduction
1.1
Background
R
ising fuel costs and greenhouse gas emissions (GHG) have garnered increas-
ing public and political concern in recent years. These trends are expected
to continue as developing nations such as India and China continue to rapidly industrialize while conventional oil reserves wane. A significant contribution to worldwide energy consumption and GHG emissions is the transportation sector, accounting for two thirds, or 8.4 million m3 (53 million US barrels) of crude oil a day
(4.6 million m3 of which are used for land transport of people) and 25% of global
GHG emissions [Heywood, 2006]. Adding to the transportation energy demand is
the increase in global vehicle production, which by the end of 2006, is expected to
reach approximately 70 million vehicles; an increase of 5% over the previous year
[Gottschalk, 2006]. Although Canada ranks as the eighth largest vehicle producer
with over 2.6 million vehicles in 2005 [OICA, 2005a, OICA, 2005b], Canada’s percapita oil consumption of 3 gallons per day ranks it as a leading energy consumer
[Energy Information Agency (EIA), 2006]. Similarly, Canada is a leading per-capita
GHG emitter with 26% coming from the transportation sector [Canada, 2005]. Consequently, even modest improvements in transportation engine efficiency could poten-
1
CHAPTER 1. INTRODUCTION
2
tially have profound economic and environmental effects at national and international
scales.
1.2
Variable Valve Timing
Without a feasible alternative to the internal combustion engine (ICE) foreseeable
within the next ten to fifteen years [Atkins and Koch, 2003], methods of improving
existing ICE efficiency have become increasingly important. One promising method
of improving ICE efficiency is through independent control of gas exchange valve timing. Traditional ICE gas exchange valve timing is mechanically fixed in proportion
to crankshaft position. This timing determines when valves open and close, thereby
affecting the air-fuel mixture and exhaust flow. At high engine speeds and loads, a
delayed closing of the intake valve is desired to take advantage of air-fuel momentum, otherwise known as the ram effect. Conversely, at low engine speeds and loads,
ram-charging is negligible and intake valve closing should occur early to maximize the
effective compression ratio and to avoid exhaust gas recirculation (EGR). Since timing
cannot be altered without significant engine modifications, a compromise between low
and high engine speed efficiency is assigned [Schechter and Levin, 1996]. Although
the variable valve timing concept is not new [Payne, 1899], a confluence of increasing
fuel prices, stringent emissions standards, efficient and affordable power-electronics
and microprocessor technologies have motivated automotive researchers to investigate
a variety of variable valve timing methods. Already available to consumers are engines with variable valve timing over limited ranges of crank angle. It is now common
to find manufacturers implementing mechanical devices capable of altering the valve
phase [Dugdale et al., 2005], duration [Borgmann et al., 2004, Hara et al., 2000] and
lift [Golovatai-Schmidt et al., 2004, Nakamura et al., 2001]. BMW’s Valvetronic,
Porsche’s Vario-Cam Plus and Honda’s IVTEC are just a few examples of mechan-
CHAPTER 1. INTRODUCTION
3
ical variable valve systems which are currently in production and offer some flexibility in valve actuation. Although such configurations do improve low to midspeed torque and fuel economy, the do not offer the ability to change the timing
or duration of the individual valve events. In each of the aforementioned production systems, all intake or exhaust valves will be changed to a new profile rather
than altering an individual valve trajectory. Independent valve control is desirable
for a multitude of reasons such as when performing a cylinder deactivation procedure, internal EGR management, or for a staggered intake valve opening operation
[Wilson et al., 1993]. In addition, an extensive host of engine operation modes also
become available with independent valve control, ranging from one cylinder starting
to regenerative engine braking [Turner et al., 2004, Schechter and Levin, 1996]. As a
result, the design and implementation of individual actuators which allow the ability to independently influence each valve irrespective of crank angle or other valves
is being actively pursued. Laboratory engines have already been fitted with hydraulic [Shen et al., 2006, Allen and Law, 2002, Barros da Cunha et al., 2000], pneumatic [Trajkovic et al., 2006, Richeson and Erickson, 1989], motor [Henry, 2001] and
electromagnetic actuators [Cope and Wright, 2006], [Niţu et al., 2005],
[Pischinger et al., 2000], [Lequesne, 1990] to demonstrate the various benefits of VVT
on engine performance (usually while disregarding valve actuation performance).
These studies have shown low to mid-speed torque output has been increased by
10%, fuel economy has been improved by 15% and a reduction of NOx emissions
by 20% [Pischinger et al., 2000, Moro et al., 2001, Lancefield et al., 1993] compared
with a conventional fixed timing valvetrain.
1.2.1
Electromagnetic Variable Valve Timing
A number of designs for variable valve actuators have been proposed. These include
valve control through electrical motors (linear and rotary) [A. Warburton et al., 2005,
4
CHAPTER 1. INTRODUCTION
Figure 1.1: Schematic of prototype solenoid valve actuator [Gladel et al., 1999]
Qiuz et al., 2004, Henry, 2001], piezoelectric [Weddle and Leo, 1998], pneumatic
[Richeson and Erickson, 1989, Gould et al., 1991], hydraulic [Allen and Law, 2002],
[Barros da Cunha et al., 2000], and solenoid actuators [Pischinger et al., 2000],
[Lequesne, 1990]. Many of these approaches cannot provide sufficiently fast, efficient and precise control of cylinder charge during engine transients.
Transient
cylinder charge control is of particular importance in the transition between combustion modes such as homogeneous charge compression ignition (HCCI) in spark
CHAPTER 1. INTRODUCTION
5
ICEs [Shahbakhti et al., 2007, Koopmans et al., 2003]. Although no clear actuation
method has been proven to be superior, solenoids are ideal in that they work through
non-contacting forces, provide sufficiently fast transition times, are relatively energy
efficient and are inexpensive to mass produce.
Typically, the solenoid actuator consists of a linear-moving armature with two coils
and two preloaded springs as described in [Gladel et al., 1999] and shown in Figure 1.1. The springs can achieve rapid flight times while minimizing electrical energy input and are essential in overcoming the significant combustion pressures imposed on the exhaust valve. The electromagnets are required for ‘catching’ the
armature at either stroke bound. In addition, they are used to overcome friction
and pressure disturbances. Permanent magnets have also been employed to “catch”
the armature at the stroke bounds with electromagnets providing a release force
[Lequesne, 1999, Theobald et al., 1994].
Other designs include hinged or clapper-type configurations [Kawase et al., 1991,
Montanari et al., 2004, Mianzo et al., 2005], such as the prototype used in this work.
An example of a typical hinged solenoid actuator is shown in Figure 1.2. In the
hinged armature design, the armature equilibrium is balanced between a torsion bar
and a linear compression spring. Pole geometry is considered ‘U’ shaped and made
of laminated steel for eddy current suppression.
A production solenoid actuator based variable valvetrain is estimated to increase
the parasitic engine load by 1% over a conventionally driven cam-roller valvetrain
[Flierl and Klüting, 2000]. However, there is a low to mid-speed improvement of 1520% in fuel economy through volumetric efficiency enhancement [Moro et al., 2001],
[Pischinger et al., 2000, Lancefield et al., 1993]. Similar performance and energy consumption characteristics are expected to be achieved using the prototype actuator
described in this work. In addition, higher efficiency of the actuator itself can be
attained through improved control strategies [Gunselmann and Melbert, 2003] and
6
CHAPTER 1. INTRODUCTION
Figure 1.2:
Hinged
[Stolk and Gaisberg, 2001]
electromagnetic
gas
exchange
valve
actuator
actuator design [Clark et al., 2005]. Arguably the greatest advantage of fully flexible
valvetrain over variable valve timing systems already in production is the ability to affect individual valves on a cycle-by-cycle basis. Such ability potentially enables a host
of advanced engine management strategies such as varying the effective compression
ratio, operating on alternative combustion cycles (HCCI, Miller) thereby enhancing
efficiency over a broad range of operating conditions.
CHAPTER 1. INTRODUCTION
1.3
7
Problem Identification and Research Scope
One of the challenges remaining in the implementation of solenoid actuators in production engines is the achievement of robust valve landing control subject to practical
constraints. These constraints have already been identified in literature
[Koch et al., 2002] and include:
• maximum available voltage of 42V
• maximum valve seating velocities of 0.1m/s [Wang et al., 2002] (for maintaining
acceptable engine acoustical noise levels and ensuring valve seating and wear
requirements are met)
• transition times of no more than 4.5ms [Peterson and Stefanopoulou, 2004] (in
order to meet maximum engine speeds of 5000 - 6000 RPM)
• practical feedback sensor technology
Performing efficient solenoid valve control is a challenging problem due to these stringent constraints and the nonlinear actuator characteristics. Depending on the actuator design, control may be complicated due to eddy currents, magnetic saturation,
limited range of authority, and sensitivity to parameter variations and combustion
gas force disturbances.
Given the above constraints, intermediate objectives are identified in relation to the
overall research scope. They are:
• Modeling and Simulation: An accurate yet simple model of the actuator system is derived and experimentally validated in order to design control strategies
and algorithms. The simulations offer an expeditious and cost effective method
for model and control performance refinement prior to implementation.
CHAPTER 1. INTRODUCTION
8
• Control Design: Identification, derivation and implementation of a control
methodology and design. Algorithms are evaluated in simulation using the
developed models.
• Experimental Testbench Control Validation: Experimental validation of
the models and control simulations on an idealized engine test platform. Performance is evaluated with respect to valve seating velocities, transition times
and energy efficiency while subject to the outlined constraints.
A “V-cycle” or rapid prototyping flow chart of these objectives within the context
of the overall VVT actuator design and implementation process is presented in Figure 1.3. The focus of this work is on refining and evaluating the developed control
algorithms on an experimental engine testbench simulator in preparation for implementation on an operational single cylinder engine. These aspects are indicated by
the dashed outline in Figure 1.3. Additionally, the modeling and simulation done
in previous work [Chladny, 2003] are extended to accommodate the latest prototype
and additional physical effects, such as gas force disturbances.
1.3.1
Contribution
The overall objective of this work is to provide the hardware and software required
for implementation of electromagnetic valvetrain actuators for use on an experimental
single cylinder ICE. In pursuing this goal, the following contributions have been made
related to control of electromagnetic VVT actuators:
• A method for position feedback using a finite element model and magnetic flux
measurement technique.
• The simplification of a compressible flow model for online cycle-by-cycle gas
force identification.
CHAPTER 1. INTRODUCTION
9
Figure 1.3: V-Cycle workflow of overall research objectives
• An experimental testbench method for exhaust gas disturbance emulation and
control performance evaluation.
• The development of a nonlinear observer for gas force disturbance identification.
• The design and implementation of feedforward and nonlinear landing controllers
subject to practical physical constraints and gas force disturbances using the
aforementioned feedback sensing and disturbance identification techniques.
• Demonstration of advanced nonlinear control on a highly-nonlinear system which
CHAPTER 1. INTRODUCTION
10
requires a detailed understanding of the actuator system physics, use of a novel
feedback sensor concept, implementation of a feed-forward and disturbance estimator to reject large disturbances on a cycle-by-cycle basis and integration of
the derived feed-forward and closed loop controllers.
In the following chapters, an overview of these contributions and the related supplementary material are provided within the context of relevant literature. The chapters
are organized as follows. A survey of relevant literature is presented in Chapter 2.
Proposed variable valve actuators, modeling, feedback, control and implementation
methods are compared and contrasted. Chapter 3 provides a summary of the related
modeling and control theory with respect to the solenoid variable valve actuator problem. Details of the modeling and simulations undertaken are provided in Chapter 4.
The overall proposed controller topology is presented in Chapter 5. In this chapter, the constituent algorithms that comprise the overall control method such as the
state estimation and landing controller are described in detail. Chapter 6 describes
the experimental procedures and equipment used to qualify the developed simulation
models and evaluate control performance. Experimental results and principal findings are discussed in Chapter 7. A summary of key conclusions and supplementary
reference material are provided in Chapter 8 and appendices, respectively.
Chapter 2
A Review Of Current Variable Valve Actuator
Technology, Modeling and Control
2.1
Introduction
C
urrently, a myriad of variable valve timing (VVT) actuators have been implemented on laboratory engines. Electric motor, pneumatic, hydraulic and
electromagnetic actuators have all been implemented and documented. Although
such actuators have demonstrated the benefits of VVT on engine performance, their
designs often neglect many of the issues which must be addressed prior to being implemented in a production vehicle. Means of preventing excessive valve seating velocity
and the resulting wear and acoustic emissions are of particular concern. Other considerations include methods of feedback, opening/closing (transition) time and power
consumption. This chapter highlights some of the current efforts to implement VVT
technology with respect to actuator types, modeling, feedback control and exhaust
gas disturbance rejection strategies.
2.2
Various Actuator Technologies
The following sections provide an overview of current and proposed VVT actuation
technology.
11
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
2.2.1
12
Pneumatic & Hydraulic Actuators
Presently, pneumatic and electro-hydraulic valve actuators for spark ignition engines
are not expected to be mass produced in a production vehicle due to the relatively
high unrecoverable energy input required to operate them [Henry, 2001]. In addition,
these actuators tend to require considerable maintenance and their performance and
accuracy are temperature sensitive. Pneumatic or hydraulic pumps, with coolers,
filters and all the related accessories required to deliver the air or fluid are necessary to allow these actuators to operate in an on-board vehicle environment. Some
hydraulic actuators rely on a traditional camshaft that provides pressure to small
cylinders located above the valve body. Timing and lift may be then varied with a
valve that controls the amount of fluid in the cylinders [Gecim, 1993, Kim et al., 1997,
Barros da Cunha et al., 2000]. Aside from the increased parasitic load on the engine,
such systems will increase associated engine manufacturing costs. Further costs may
be associated with the safety equipment required to protect personnel and the environment from any system failures due to the presence of fluid pressures in excess
of 20 MPa (3000 psi) [Sun and Cleary, 2003]. As well, advanced control systems
and electronics are required to coordinate appropriate valve motion. Pneumatic and
hydraulic actuators are also susceptive to the same constraints as electromagnetic
actuators such as cost, seating noise and speed requirements. It is for these reasons
that successful use of such actuators has been primarily limited to either laboratory
or to high output (and relatively low mileage) race engines.
2.2.2
Rotary & Linear Motors
Electric rotary motors which use permanent magnets have also been proposed for
use as valve actuators. They typically employ mechanical devices which transform
rotary motion into reciprocating linear motion. These systems appear relatively
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
13
easy to control since valve position can be considered fixed with respect to the motor displacement. However, they are far less developed than pneumatic and hydraulic devices [Giglio et al., 2002], perhaps due to the relatively large physical volumes required to achieve the necessary response and valve speeds. Due to the challenging task of overcoming exhaust gas force disturbances, there appears to be renewed interest in the motor driven desmodromic valve actuators [Parlikar et al., 2005,
A. Warburton et al., 2005]. Such actuators are relatively simple to control, and in
some cases, are capable of variable lift and zero holding current. However, the issues
of soft landing, transition time, under-hood volume and power consumption still appear problematic in some cases. In [Parlikar et al., 2005], zero engine load (no gas
disturbance) transition time is achieved within 3.5 ms however, full load (6000rpm,
open throttle) power consumption is estimated to be 160 W. Seating velocities of 0.15
to 0.27 m/s are reported. Power consumption in [A. Warburton et al., 2005] is lower
(100 W), with similar transition times and seating estimated to be below 0.24 m/s.
If further refitments can be made, the motor driven valve actuator may become a
viable flexible VVT candidate.
Linear motors have also been applied as variable valve actuators [Braune et al., 2006],
[Mercorelli et al., 2003]. They were initially predicted to be better suited to valve
strokes in excess of 20 mm when compared with solenoid electromagnetic actuators
[Lequesne, 1996]. This is because the high frequencies and the resulting duty cycle required in a VVT application cause significant power consumption and their
relatively slow response (due to large moving mass). In [Mercorelli et al., 2003], simulated results indicate landing speeds of approximately 0.01 m/s, however the transition time is approximately 5 ms and a voltage in excess of 100 V is used. Later
in [Braune et al., 2006], the same group provides considerably poorer experimental
results with transition times in excess of 15 ms (with zero pressure disturbances) and
holding currents (while open) in excess of 30 A. No landing speeds or velocity trajec-
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
14
tories are explicitly reported. The armature is separated from the valve stem when
fully closed (1 mm gap) which, aside from potential wear and acoustical noise caused
by internal impacts, results in an unusual opening lift trajectory. A potentiometer
is used for position feedback and the supply voltage used is not indicated. Susceptibility of permanent magnets to under-hood temperature variations is also of concern
[Trout, 2001]. Despite those shortcomings, there are few reported linear motor results
and in time, they may also become a viable VVT actuation option.
2.2.3
Piezoelectric Actuators
Actuators using piezoelectric crystals have been proposed in [Weddle and Leo, 1998].
The benefits of such a system would be through the exploitation of the fast response
and high forces inherent with piezoelectric systems. However, displacement is quite
limited from the crystals themselves, thus a lever arm and rack and pinion gearing
are used to amplify crystal deflection. To minimize system inertia, a ‘flapper’ or
butterfly type of valve is proposed instead of a typical valvetrain poppet valve. The
authors discuss several concerns that arose while testing an initial prototype. Aside
from flow and sealing concerns of a different gas exchange valve, variation of valve
displacement with frequency occurred as a result of the excitation of various modes.
As well, dimensional tolerances (and the additional expenses) are of concern due to
the mechanical amplification through the lever arm. To date, no further prototypes
have been documented.
2.2.4
Electroactive Polymer Actuators
Presently an emerging technology, electroactive polymer (EAP) actuators have also
been proposed for VVT applications [Ashley, 2003]. Such actuators consist of a dielectric polymer (acrylic or silicon for example) laminated between two conductive
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
15
sheets. A high voltage (1-5kV) is applied to the conducting layers, establishing an
electric field sufficiently strong enough to squeeze the polymer layer which in turn,
expands in area. The actuation technique offers large strain rates with respect to
force per unit density compared to magnetic and piezoelectric devices, however many
issues must be resolved before a production actuator could be viable. Aside from the
need to generate high voltages, these include fatigue and temperature susceptibility
/ sensitivity.
2.2.5
Solenoid Actuators
Another promising actuator configuration for independent valve control is the electromagnetic solenoid. Electromagnetic actuators for valve control were patented at least
as early as the 1970s [Longstaff and Holmes, 1975, Pischinger and Kreuter, 1984] and
are perhaps the most popular of the production fully flexible VVT actuator candidates. Unlike the aforementioned hydraulic and pneumatic actuators, electromagnetic
actuators are often designed with two springs which provide most of the necessary
energy for a given valve cycle. After the armature is released it accelerates to the
middle position, then the valve is decelerated with an opposing spring which stores
the kinetic energy for the the next valve event. In the non-powered rest position, the
two springs force the armature to the mid-stroke position. In this way, electromagnetic energy is only required to influence the valve behavior at either end of the valve
trajectory and to overcome any losses due to friction or gas forces. They are compact
in size, relatively inexpensive to mass produce and are not as temperature dependant as their hydraulic counterparts. Like the motor based actuators, they require
sophisticated electronic hardware and control software. A relatively large alternator /
permanent magnet induction motor must also be employed to supply approximately
70W of peak electrical power per actuator at maximum engine speed and load. However, the extra electrical power requirement (3% of total engine output) is expected
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
16
to be offset by the friction energy saved through the absence of camshafts (2% of
total engine output for a roller-bearing valve-train) [Flierl and Klüting, 2000]. One
potential drawback to solenoid actuators is that the springs often dominate valve
motion, especially near the mid-stroke position, making variable valve lift (VVL) a
challenging issue. However, even this limitation has been addressed with some success
in [Peterson et al., 2006, Wolters et al., 2003].
Successful implementation of electromagnetic actuators has also been generally limited to the laboratory, although several prototype vehicles have been built with an
electromagnetic valve-train [Pischinger et al., 2000]. Control with combustion pressure disturbances has been particularly challenging. Manually tuned feedforward control has been implemented within laboratory settings [Tai and Tsao, 2001]. However,
such systems are unable to account for the significant combustion pressure fluctuations. In order to implement a reliable control system it must be robust to variations
of the valve / actuator system. These variations include abrupt disturbances from
combustion pressure variations as well as the relatively gradual parameter changes
caused by component fatigue, friction and temperature variations. In addition to
these variations, the relatively high speed of the actuators, the non-linear magnetics
and the short distance over which control is feasible has motivated many to develop
closed loop control systems [Koch et al., 2002, Wang et al., 2002, Tai and Tsao, 2001,
W. Hoffmann and A. Stefanopoulou, 2001]. Due to the design goals and the necessity
for closed loop control, it has been determined that high armature position sensor resolution and frequency response are required. Such sensors are often expensive and difficult to calibrate in a mass production environment. However, efforts are being made
to develop alternative cost effective sensors or sensing methods with sufficient response
and resolution [Lynch et al., 2003, Rossi and Alberto, 2001, Takashi and Iwao, 1995,
Roschke and Bielau, 1995].
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
2.3
17
Feedback Sensors
The majority of experimental literature makes use of external valve and/or armaturebased position measurements for state reconstruction. These include position or velocity measurements via linear variable differential transducer (LVDT)
[Sun and Cleary, 2003], potentiometer [Braune et al., 2006], laser [Peterson, 2005],
[Tai and Tsao, 2003, Wang et al., 2002, Stubbs, 2000] or eddy current
[Peterson et al., 2006, Chung et al., 2007] displacement sensors.
Although these sensors provide sufficient precision, accuracy and response, efforts are
being made to develop alternative cost effective sensors or sensing methods with equivalent performance. These include the flux-based coil type [Chladny and Koch, 2006a,
Scacchioli, 2005, Montanari et al., 2004],
[Rossi and Alberto, 2001], [Rossi and Tonielli, 2001], [Roschke and Bielau, 1995],
observer-based [Lynch et al., 2003, Eyabi and Washington, 2006a, Eyabi, 2003] and
self inductive [Butzmann et al., 2000, Takashi and Iwao, 1995] schemes. In the latter
case, it is proposed that the driving coil itself be used to relate the measured rate of
change of induced coil current to the armature position and velocity. This may be
done by momentarily deactivating the drive coil and relating velocity-induced currents
to position. Drive currents are readily measurable through hall-effect devices and generally used in all control schemes. Using this method of sensing, landing speeds of
0.2m/s are claimed. However, the authors indicate that the system is limited by its
sensitivity to disturbances in addition to temporary loss of control authority while
sensing [Butzmann et al., 2000] and is therefore of limited practical use.
Observer-based state reconstruction makes use of partial state measurements and
estimated initial state conditions to predict plant output. A sliding mode observer
is used to predict valve position and velocity using only a measured current signal
in [Eyabi and Washington, 2006a]. The algorithm is implemented successfully on a
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
18
actuator with relatively soft springs (peak velocities ≤ 1.5 m/s, transition times ≥
25 ms) and gas force disturbances are not considered. In [Peterson et al., 2002] position and current are used to estimate velocity in a nonlinear reduced order observer.
Although observer-based feedback systems reduce the number or eliminate the need
for sensors, such schemes may be sensitive to initial state estimates and require high
gain for the relatively rapid estimation convergence needed for landing control. As a
result, landing control performance may be compromised when subjected to excessive
noise or disturbances.
Another feedback system demonstrated includes a microphone to adaptively improve
impact speeds from sound intensity measurement [Peterson and Stefanopoulou, 2004].
Although proven successful in laboratory testbench experiments, such a sensor scheme
is likely not practical in an engine environment due to multiple valves in operation
and other possible acoustic sources. The method also requires several cycles to converge from one operating condition to another and is thus perhaps more suitable for
slowly changing conditions.
In the case of [Montanari et al., 2004], using flux-based position reconstruction, two
functions are used to relate reluctance and excitation to air gap. Improved accuracy
is achieved with a similar measurement technique but through the use of a numerical
look-up table with finite element analysis (FEA) results in [Chladny and Koch, 2006a].
At this time, it appears that no other group has documented a FEA derived fluxbased position sensing technique. In both cases, the sensor is cost effective, practical
and capable of achieving landing speeds below 0.1 m/s.
2.4
Actuator Modeling
With many electromagnetic devices, simplified geometry and linear approximations
are often satisfactory in estimating device performance. In the context of variable
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
19
valve timing actuator design, control is to be applied in an operating region in which
the air gap is small and at excitation levels sufficiently large that material saturation
is significant, making linear theory unsatisfactory for accurately predicting actuator
performance [Chladny, 2003]. The following highlights some of the work that has
contributed to the advancement of solenoid actuator modeling.
2.4.1
Lumped Parameter Models
Nearly all solenoid VVT control algorithms described in literature are derived from
some variation of a lumped parameter model (LPM). Most use heuristic flux and force
relations that take on a form that neglect magnetic material saturation like below:
β
κ−x
βi2
Fmag (x, i) =
(κ − x)2
λ(x) =
(2.1)
where λ represents the magnetic flux linkage, x is the valve/armature position, i
is the coil current and Fmag is the resulting magnetic force. Coefficients β and κ
are determined by performing linear regression fits to experimentally obtained data
[Tai and Tsao, 2001, Peterson and Stefanopoulou, 2004]. A similar approach is used
in [Wang et al., 2002, Butzmann et al., 2000], but another analytic function is appended to approximate the saturated operating points (with constraints to prevent
discontinuities). Later, [Tai and Tsao, 2003] improved the model by adding a current
and position dependant gain to compensate for saturation effects. An inductance relation proposed by [Ilic’-Spong et al., 1987] that accounts for magnetic saturation is
used by [Chladny and Koch, 2006b, Chung et al., 2007, Koch et al., 2004] using two
different actuators. The authors fit the LPM parameters to FEA generated data and
demonstrate good agreement with static and transient experimental measurements.
Four methods for establishing LPMs to describe linear and rotary electromagnetic sys-
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
20
tems are investigated in [Melgoza and Rodger, 2002]. The methods are evaluated for
computational time and accuracy. It is acknowledged that it is not practical to fully
simulate the overall performance of the actuator through a discretization processes.
Rather, a more practical approach is the use of a simplified model that combines the
accuracy of off-line FEA results with the computational speed of a LPM.
A model that includes saturation and hysteresis is presented in
[Eyabi and Washington, 2006b]. Also considered is the effect of mutual inductance
between the opener and closer magnets. Here (as in [Vaughan and Gamble, 1996])
net coil current, i, is composed of dissipative current, id , and restoring current (or the
current related to the coenergy stored in the magnetic field), ir , such that i = ir + id
with ir and id defined by:
ir = f(x, λ) = g4 λ4 + g3 λ3 + g2 λ2 + g1 λg0
id = f1 (v) − τj
did
dt
(2.2)
(2.3)
where ir is a function of flux linkage λ and coefficients gi are dependent on position,
x, as follows:
gi = ni,6 x6 + ni,5 x5 + ni,4 x4 + ni,3 x3 + ni,2 x2 + ni,1 x + ni,0
(2.4)
Coefficients ni,j are fit to experimental data. Hysteresis is accounted for in the id
term which is a nonlinear function of voltage, v, and a dissipation term, τj , such that:
f1 (v) =


 d1,1 |v|d1,2 (x) sgn(v), v ≥ 0

 d2,1 |v|d22 (x) sgn(v),
v<0
(2.5)
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
21
and



τ1 , v > 0



τj =
τ2 , v < 0, x < xv




 τ3 , v < 0, x > xv
(2.6)
where parameters τj , di,j and xv are determined experimentally. Finally, a relationship
between magnetic force, Fm , position and flux is defined as:
Fm = m2 λ2 + m1 λ + m0 ,
(2.7)
with
mi = pi,8 x8 + pi,7 x7 + pi,6 x6 + pi,5 x5 + pi,4 x4 + pi,3 x3 + pi,2 x2 + pi,1 x + pi,0
(2.8)
where again, coefficients pi,j are fit to data (it is unspecified if it is simulated or experimentally generated). The model is contrasted with experimentally acquired results
with reasonably good agreement with a significant reverse magnetization effect due
to hysteresis losses. The work is relatively unique since others’ contributions have
found, including this work, that in practice, hysteresis is not significant and thus not
modeled.
Other LPMs are derived through the use of reluctance networks [Mercorelli et al., 2003,
Chillet and Voyant, 2001, Piron et al., 1999]. This technique parallels an electric circuit approach to predicting actuator static and transient performance. By discretizing
portions of the magnetic circuit into elements, each with their own permeability and
geometric identity, the same methods of solving electric circuits may be used to solve
the reluctance network. It is acknowledged in [Piron et al., 1999] that a preliminary
FEA analysis may need to be performed to accurately predict flux and eddy current
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
22
paths.
Other LPM magnetic models have been derived from fitting polynomial functions to
FEA data [Haskara et al., 2004] or experimental force and excitation data
[Vaughan and Gamble, 1996].
2.4.2
Finite Element Analysis
In [Chladny et al., 2005, Chladny, 2003] methods of FEA modeling and validation
of a solenoid VVT actuator are documented using two dimensional static and transient models. Comparable work has been documented in [Clark et al., 2005]. The
techniques used in [Clark et al., 2005] differ in that a 3D FEA model is used to investigate pole geometry on force response over a variety of operating ranges. The benefits
of using FEA to shape the pole face to improve stability (minimize reluctance and increase flux density) is highlighted in [Chladny, 2003, Lequesne, 1999, Lequesne, 1990].
Despite the greater computational load of a 3D model, there appears to be no gain
in model accuracy compared with existing results attained by 2D models. More commonly, FEA is used to evaluate reluctance network models [Chillet and Voyant, 2001,
Piron et al., 1999, Xiang, 2002] or for thermal rise due to coil heating [Stubbs, 2000].
Experimental evaluation of solenoid prototypes has been well documented for the
purposes of control design [Giglio et al., 2002, Wang et al., 2002, Tai and Tsao, 2001,
Stubbs, 2000]. Static force data as a function of excitation and position are typically
measured for fitting parametric functions to. However, FEA simulation and experimental results (static) documented in the same work are less common [Lequesne, 1999,
Clark et al., 2005, Hartwig et al., 2005]. At this time, there does not appear to be any
published results of transient FEA modeling in contrast with comparable experiments
other than [Chladny et al., 2005, Chladny, 2003, Koch et al., 2002]. In addition,
[Chladny et al., 2005] describes a method of parameterizing an eddy current model
by utilizing FEA results for a linear motion solenoid actuator.
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
2.4.3
23
Mechanical System Modeling
Most literature follows a similar mechanical system structure based on a one dimensional second order mass spring system [Peterson and Stefanopoulou, 2004],
[Montanari et al., 2004]. Others attempt to capture additional dynamics which take
place when a flexible lash adjuster is considered [Haskara et al., 2004],
[Lambrechts et al., 2004, Tai and Tsao, 2003]. The mechanical models that include
lash dynamics consist of a two mass-spring system representing a valve, armature
with additional relative stiffness and viscous damping coefficients. Although such a
model more accurately reflects expected valve and armature behavior, lash-armature
position mismatch is not predicted to be a critical design consideration if incorporated
into the trajectory design [Koch et al., 2004]. Other groups have modeled valve-seat
impacts
[Eyabi and Washington, 2006b] by modeling the magnetic pole faces as stiff massspring-damper systems in order to approximate a collision. According to those authors, bounce is not present provided landing speeds are below 0.3 m/s. Given the
low impact speed specification of 0.1m/s, the benefits of an impact model will thus
be most apparent in a landing control failure.
2.5
Control
In many proposed designs, motion control is performed in two stages
[Gunselmann and Melbert, 2003]. The first stage of motion is usually performed in
open-loop due to the relatively limited force authority at larger airgaps and low
bandwidth inherent to solenoid actuators (high inductance and restricted source voltage). A common approach is to use a feedforward controller to provide desirable
initial conditions for the landing controller. Feedforward controllers are also used
to overcome any gas forces that may be present, perhaps in combination with an
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
24
adaptive scheme [Peterson and Stefanopoulou, 2004] or engine load and speed maps
[Haskara et al., 2004]. Typically, such feedforward schemes are highly tuned to a particular operating condition [Chung, 2005] and not suited to accommodate potentially
large gas force cycle-to-cycle variations, due to a misfire for example. During the
second stage, closed loop feedback is used to achieve the stringent landing requirements. Irrespective of the control algorithm used, control performance variations can
be expected due to the various actuator designs, feedback techniques and voltage
sources. Even the power electronics and control thereof play a significant role in
overall actuator performance [Mercorelli and Liu, 2005, Mianzo et al., 2005]. Thus,
caution must be used when comparing one control algorithm to another unless done
with equivalent systems (simulation or experiments). Provided a sufficiently accurate model is used, nonlinear control (feedback linearization, backstepping, flatness,
sliding mode) generally results in superior performance and minimal control effort
compared to linear techniques. However, the algorithms may be computationally expensive, sensitive to disturbances and sensitive parameter variations when compared
to classical techniques.
2.5.1
Landing Control
A wide variety of control algorithms for the end-control or landing problem have
been proposed. They include: passive permanent magnets to ‘catch’ the armature
at either stroke bound with excitation currents applied to temporarily neutralize the
magnets during armature release [Lequesne, 1999], classical linearized model control [Konrad, 1998], an energy-based method [Schmitz, 1995] where mechanical system energy is regulated about a desired constant, proportional integral (PI) control [Tai and Tsao, 2001, Stubbs, 2000], linear quadratic regulator (LQR) control
[Tai and Tsao, 2003], sliding mode feedback linearization [Haskara et al., 2004],
[Eyabi, 2003], flatness-based control [Chladny and Koch, 2006b, Chung et al., 2007,
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
25
Chung, 2005, Koch et al., 2004, Mercorelli et al., 2003].
Also of interest is a sensorless, constant ratio approach that alleviates the need for
external position and velocity measurements [Butzmann and Melbert, 2003],
[Butzmann et al., 2000]. Here, armature motion is inferred from coil measurements
using the following relationship:
di
1
=
dt
1 + H0 x
vx
i ẋ
−
2
µ0 N Ac
x
(2.9)
where x is the air gap between armature and pole face, µ0 is the permeability constant,
H0 is a constant accounting for flux leakage, Ac is the core flux path area, i is the
coil current, v is the coil voltage and N is the number of coil turns. If the supply
voltage source, v, is set to zero (for measurement during landing control), and the
H0 x leakage term is neglected, the above relation simplifies to
di/dt
ẋ
=−
i
x
(2.10)
For landing control, a predefined desired velocity-position ratio is compared to the
measured
di/dt
i
signal. The coil is then switched to regulate the velocity-position ratio
to the desired constant.
Under laboratory testbench conditions, most methods of feedback are able to achieve
landing speeds of 0.2 m/s or less with variations in transition times between 3.5
and 20 ms. There is large variation reported among the voltage sources, feedback
techniques and power consumption used; all of which play a significant role in the
overall suitability for production implementation.
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
2.5.2
26
Feedforward Control
The approximate
1
x2
drop in magnetic force with air gap limits the effectiveness of the
closed loop controller to regions close to the pole face and is thus only used for landing
control [Tai and Tsao, 2002]. Due to the relatively wide range of typical ICE operating conditions imposed on solenoid VVT actuators, the control scheme must compensate both for slowly varying conditions and rapid disturbances. Changes occurring
over many valve opening and closing cycles are classified as slow changes. Examples of
slow changes are variations in coil resistance and friction as a function of temperature
or mechanical wear. Such variations can be accounted for through a cyclical adaptive approach as illustrated in [Wang et al., 2002, Peterson and Stefanopoulou, 2004,
Tai and Tsao, 2003, Butzmann et al., 2000]. However, rapidly varying disturbances,
such as those due to combustion gas forces (particularly those on the exhaust gas
valve), can vary significantly from one cycle to the next (over 250 N for the actuator
in this study). A feedforward controller will have to accommodate gas force disturbances because the majority of gas work occurs during the initial valve stroke and
must therefore adjust the imparted magnetic force accordingly to setup reasonable
initial conditions for the landing controller. The design is further complicated since
in-cylinder pressure sensors are not present in current production engines. As a result,
some form of online disturbance estimation scheme is required.
2.5.2.1 Cyclic-Adaptive Compensation
Controllers designed to compensate for parameter variations or disturbances occurring over many valve opening and closing cycles are classified as cyclic adaptive or
iterative learning controllers (ILC). Work such as [Peterson and Stefanopoulou, 2004,
Hoffmann et al., 2003, Tai and Tsao, 2003, Wang et al., 2002, Butzmann et al., 2000]
exploits the repetitive nature of the actuator operation to tune performance over many
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
27
cycles. For example, in [Peterson, 2005, Peterson and Stefanopoulou, 2004] impact
acoustical intensity is measured via a microphone to tune the controller over several
cycles via an extremum seeking controller:
Uc =
K1
K2
ẋ +
γ+x
β+x
(2.11)
where control output, Uc , is subject to tuning parameters K1 , K2 , γ and β. In practice,
only β is adjusted to reduce impact velocity based on impact sound intensity via the
following cost function:
Q = [Sdes − Smeas (k)]2
(2.12)
where the difference between the desired and measured sound intensity (Sdes and
Smeas (k) respectively) is is minimized over k cycles. Using this method impact velocities below 0.1 m/s and transition times of less than 4.0 ms are achieved with a 100 V
source. Convergence occurs in approximately 40 cycles. Velocity, ẋ, is estimated with
the nonlinear observer presented in [Peterson et al., 2002].
An iterative learning controller (ILC) is presented in [Hoffmann et al., 2003] that uses
observer output feedback. The magnitude of simulated transient force disturbances
on the order of 20 N (intake or low-load exhaust pressure disturbances) are varied by
2 N over 10 cycles to demonstrate how the ILC adjusts open-loop input based on the
actual and desired armature position weighted errors. A linear-quadratic regulator
(LQR) landing controller is used to track a trajectory in the latter part of the valve
stroke. Convergence is achieved after approximately 35 valve cycles with impact velocities of 0.04 m/s using a 200 V power source.
In [Chen et al., 2005], gas force disturbances are heuristically parameterized using an
exponential decay function in which the amplitude and time constant are estimated
over several combustion cycles via a hybridized multi-step least squares - Kalman
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
28
filter. Although the disturbance is parameterized as a function of time only (no coupling to engine speed, exhaust valve opening (EVO) angle or valve position) simulated
results appear promising. However as acknowledged by the authors, the technique is
not suited to the extreme pressure fluctuations experienced during transient engine
operation or misfires. Even when with a moderate change in initial exhaust valve
opening pressure, four consistent cycles are required for convergence.
2.5.2.2 Exhaust Gas Force Disturbance Compensation
Exhaust valve opening with solenoid actuators is a challenging control problem because the opening pressure can vary significantly from one cycle to the next. After
combustion, the engine piston is driven down and the EVO occurs usually 40 to 60
degrees of crank angle before bottom dead centre (BDC) [Heywood, 1988]. The pressure at valve opening varies depending on the combustion process and the crank angle
at which the valve opens. This is of concern as pressure significantly affects the force
required by the electromagnet for successful soft landing (while opening). In-cylinder
pressure sensors are not present in current production engines so this disturbance is
not known. Furthermore, since feedforward is used for control in the initial opening
of the valve, an online identification scheme is required. These rapid changes present
a challenging control problem [Hartwig et al., 2005] and have, at least in part, motivated researchers to develop alternative actuators such as the motor driven cam
actuators described in Section 2.2.2. One of the earliest incorporations of gas force
disturbances in valve control is in [Wang et al., 2000] where an experimental cylinder pressure trace was input in simulation. Valve force was related to gas pressure
through valve area times the difference in pressure between cylinder and exhaust port.
Later, [Hoffmann et al., 2003] simulates a similar version of the disturbance in an iterative learning controller. The disturbance was taken at a light engine load (peak
force of 20 N) and assumed to be unknown to the controller (varied by 2N every 10
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
29
cycles). In [Hartwig et al., 2005], it is proposed to use solenoid actuators only on the
intake valves, and driving the exhaust valves with a camshaft to avoid controlling the
exhaust opening process altogether. Similarly, automotive supplier Valeo announced
plans for a ‘half camless’ engine to be in production by 2009 [Alexander, 2006] where
solenoids are to be used only on the intake valves while exhaust valves will remain
cam-driven. However, no plans have yet been set for production of a fully camless
system, likely because of the added complexity of rejecting exhaust gas disturbances.
Indeed, there are very few published proposals for opening of solenoid exhaust valve
actuators while maintaining landing and transition criteria. In [Haskara et al., 2004],
the use of either a calibrated engine load-speed dependent feedforward current input
or a recursive estimation scheme is proposed. The mapping technique is ideal for
most operating conditions provided consistent maps and combustion can be maintained from one cycle to the next. This is not the case in the event of a misfire or
partial combustion. The speed-load (and presumably EVO timing) mapping technique is also time consuming to calibrate and may be unique from one engine to the
next. Reduction of calibration effort is currently an important priority for automotive manufactures [Ohata and Butts, 2005]. [Haskara et al., 2004] suggest the map
is updated online by adjusting the feedforward current input to maintain a specific
kinetic and potential energy at the point where the landing controller is engaged. The
disturbance, δ̂, is estimated using a sliding mode observer with the following error
dynamics:




h
i
 e 
 ė 
  = A   + B ff bk + δ̂
ė
ë
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
30
with




1 
 0
 0 
A=
 and, B = 

k
1
−m
− mb
m
(2.13)
and control input, ff bk , spring stiffness, k, moving mass, m, and friction, b parameterizing the mechanical response. Sampling discretization of the error dynamics is
approximating through a zero order hold resulting in:
ek+1 = Φek + Γ [ff bk,k + δk ]
(2.14)
where Φ = eAT and Γ = (eAT − I)A−1 B and δk represents the average disturbance
during a sampling interval. A sliding surface, sk , is defined as:


(2.15)
sk+1 = CΦek + CΓ [ff bk,k + δk ]
(2.16)
 ek 
sk = C 

ėk
a control input, vk , is defined as:
vk = −ff bk,k−1 + (CΓ)−1 Cek − (CΓ)−1 CΦek−1
(2.17)
with the disturbance estimate and compensated feedback given by:
dˆk = dˆk−1 + γ(vk − dˆk−1 )
ff bk,k = −dˆk
(2.18)
CHAPTER 2. VVT ACTUATOR & CONTROL TECHNOLOGY
31
where γ is a convergence tuning parameter. Disturbance dynamics are considered constant (un-modeled) between sampling intervals and thus, as reported by the authors,
the latter technique is ideal for slowly varying disturbances. Details of the technique
with respect to convergence performance or even if it was used in the results presented
are not given.
2.6
Summary
A brief overview of the progress towards designing and implementation of variable
valve actuators is given. Replacing the camshafts of traditional ICEs is a nontrivial
task, and as such, design considerations and constraints are numerous. Most published work focuses on the soft landing issue with little to no regard of exhaust gas
force disturbances. Some of the proposed control schemes are able to attain seating velocities of 0.1 m/s or less, often with the use of in-cylinder and / or expensive feedback
sensors, voltages exceeding 100 V or transition times exceeding 5ms (at low engine
load). Cyclically adaptive controllers or ILCs have also been demonstrated which are
capable of accommodating slowly time varying changes that occur over many opening/closing events. One of the most significant remaining challenges is exhaust valve
opening control under the influence of significant in-cylinder gas force disturbance
variations. The landing controller scheme proposed in this study addresses the issue
of exhaust gas force disturbances and landing control. The intake actuator gas disturbances are of lesser concern because their magnitudes and variations are less significant and more predictable. The proposed controller augments the flatness-based landing controller used in [Chung et al., 2007, Chladny and Koch, 2006b, Chung, 2005]
with exhaust gas disturbance estimation and feedforward capability using coil current and a practical magnetic flux measurement for feedback while using a 42 V
source.
Chapter 3
Theory
3.1
Introduction
D
ue to the multidisciplinary nature of electromechanical VVT actuators, a relatively large scope of physical phenomena must be examined when modeling
and controlling such devices. The following presents a brief overview of the modeling
and control techniques used with respect to the hinged solenoid actuator considered
in this study.
3.2
Finite Element Modeling
The hinged electromagnetic actuator used in this study has different opener and
closer geometry and thus current and force response. Static and transient finite
element models of the opener and closer magnetic paths are analyzed to obtain an
accurate characterization of actuator behavior for control design. In static analyses,
force and path flux values are generated as a function of current and armature (valve)
positions. These results constitute a data set which characterize the electrical and
mechanical response of the actuator for later use in lumped parameter model (LPM)
development. Transient analyses are used to estimate system response and the extent
to which it is inhibited by eddy currents. Furthermore, the finite element method
32
33
CHAPTER 3. THEORY
(FEM) results provide qualitative results which may offer insight into aspects such as
flux leakage and saturation regions. The following sections provide a brief overview
of the FEM process.
3.2.1
Maxwell’s Equations
James Clerk Maxwell recognized four general unifying relations which now make up
the foundation of all classical electromagnetic field theory [Maxwell, 1864]. These
equations provide a means of coupling time varying electric and magnetic fields
[Griffiths, 1999]. They are briefly derived in Appendix A and summarized below.
∇ · E(r, t) =
ρ
,
ǫ0
∇ × B(r, t) = µJ (r, t) + µǫ0
∇ × E(r, t) = −
(3.1)
∂E(r, t)
,
∂t
∂B(r, t)
,
∂t
∇ · B(r, t) = 0
(3.2)
(3.3)
(3.4)
where r is a position vector and t represents a time scalar. Equation 3.1 relates charge
density, ρ, to an electric field E and free space permittivity, ǫ0 , through Gauss’ law.
In Equation (3.2), Maxwell extended Ampére’s original expression relating magnetic
flux density, B, and current density, J , by adding the far right hand side term,
µǫ0 ∂E
, known as displacement current, to account for time varying electric fields
∂t
displacing electrons resulting in current. Equation (3.3) is a form of Faraday’s law
(through Stokes theorem) which states that a time-varying magnetic field produces
a corresponding electric field. The relationship is such that the electric field lines
produced tend to encircle the magnetic field lines. Equation (3.4), known as Gauss’
Law, indicates that no net flux may emanate from any given region of space (magnetic
CHAPTER 3. THEORY
34
monopoles do not exist). Thus, the net magnetic flux, φB , through an arbitrary closed
surface is always zero.
3.2.1.1 Simplifying Assumptions
These relations are general enough to account for electromagnetic waves ranging from
radio to gamma radiation. However, in the study of electromechanical devices, wave
phenomenon are often disregarded as the electrical excitation frequencies are usually
low enough that the system may be regarded as quasi-static, or slowly varying with
time. This means that although the time dependency of the magnetic induction term
− ∂B
of Equation (3.3) is of interest during transient analyses, the magnetic field may
∂t
be approximated by ignoring the displacement current term in Equation (3.2), as
nearly all the energy will be stored in the magnetic field rather than an electric field.
For example, electromagnetic devices are often not limited by the propagation speed
of electromagnetic fields traveling at the speed of light because their dimensions are
relatively small. As such, the general solution of a given problem without the quasistatic assumption will likely have some portion depending on a ratio between the
system geometry and the speed of light, and may be approximated as zero. The
result is an approximation that a slowly varying electric excitation produces only a
slowly varying magnetic field. Generally speaking, either the electric or magnetic
field will dominate over the other depending on the problem configuration. This fact
is used to simplify the above complex relations to more practical expressions. For
example, when a short circuit exists as in the case of a coil of wire, a large current
typically exists and a large static magnetic field results. Accordingly, the displacement
current is of negligible importance. In contrast, two isolated plates when excited with
a constant electric potential result in an electric field but not a magnetic one, as
the magnetic induction term is negligible. The quasi-static approximation allows the
assumption that only one field is dominant and thus it is possible to de-couple the
35
CHAPTER 3. THEORY
electric and magnetic relations and only consider the dominant field. The revised
Maxwell’s equations with the quasi-static limit imposed for the case of a dominant
magnetic field are shown below:
∇ · E(r, t) =
ρ
,
ǫ0
(3.5)
∇ × B = µJ ,
(3.6)
∇×E =−
(3.7)
∂B
,
∂t
∇·B =0
(3.8)
Note that the displacement current in Equation (3.2) is no longer present in Equation (3.6), yet the magnetic field still has a time dependency through Equation (3.7).
Material behavior is accounted for through the following constituent equations (neglecting temperature dependence, permanent magnets and relative motion):
B = µH
(3.9)
J = σE
(3.10)
where µ represents material magnetic permeability and σ the material conductivity.
When material saturation is considered (field intensity dependent) µ may be defined
by a relation between magnetic field intensity, H, and magnetic flux density, B.
Equation (3.10) represents the relationship between an electric field, E, and applied
current density, J . Material conductivity, σ, may be assumed orthotropic, and thus
a scalar quantity.
Equations (3.5) through (3.10) may be used to express the four vector fields B, H,
J and E as Poisson’s (or Laplacian) equations that may be numerically solved with
finite element methods.
36
CHAPTER 3. THEORY
Magnetic Field Intensity By substituting (3.9) into (3.6) and taking the curl of
both sides results in:
∇ × (∇ × H) = ∇ × J
(3.11)
With constant permeability, Equation (3.8) ⇒ ∇ · H = 0. Thus, with the use of the
identity:
∇ × (∇ × P ) = ∇(∇ · P ) − ∇2 P
(3.12)
where P is an arbitrary three dimensional field quantity. the left side of the above
equation may be simplified to:
∇ × (∇ × H) = −∇2 H
(3.13)
Using Equations (3.7), (3.10), the right side of (3.11) may be expressed as:
∇ × J = σ∇ × E = −σµ
∂H
∂t
(3.14)
which results in a partial differential equation (PDE) in Poisson’s form [Kreyszig, 1993]:
∇2 H = σµ
∂H
∂t
(3.15)
Electric Field Taking the curl of both sides of (3.7) and using the identity in (3.12)
results in :
∇ × (∇ × E) = ∇(∇ · E) − ∇2 E = −∇ ×
∂B
∂t
(3.16)
37
CHAPTER 3. THEORY
Assuming constant permittivity and negligible static charge implies ∇ · E = 0 and
thus:
∇2 E = ∇ ×
∂B
∂t
(3.17)
Using constituent Equations (3.6) and (3.10), the right hand side is simplified to:
∇×
∂
∂E
∂B
= µ J = µσ
∂t
∂t
∂t
(3.18)
resulting in the Poisson PDE for an electric field:
∇2 E = σµ
∂E
∂t
(3.19)
Current Density Substituting (3.10) into (3.7) and taking the curl of both sides
yields:
∇ × (∇ × J ) = −σ
∂
(∇ × B)
∂t
(3.20)
Again using the identity in (3.12) with the ∇ · J condition and substituting (3.6) into
the right side results in:
∇2 J = σµ
∂J
∂t
(3.21)
Magnetic Flux Density The Poisson PDE for magnetic flux density is obtained
by first substituting (3.10) into 3.6 yielding
∇ × B = σµE.
(3.22)
38
CHAPTER 3. THEORY
Solving for E, substituting the result into (3.7) and using (3.8) and (3.12) gives:
∇2 B = σµ
3.2.2
∂B
∂t
(3.23)
Magnetic Vector Potential
For the magnetic field solution of the actuator, 2D models are constructed with the
commercially available FEA software, ANSYS, by ANSYS Inc. In all of the model
types conducted, ANSYS utilizes the vector potential method for both static and
transient cases to solve the equations previously defined. Here a potential representing
the magnetic field is introduced as it is numerically easier to determine the vector
potential than the magnetic field directly. The magnetic field B may be expressed as
a magnetic vector potential A by observing that a magnetic field is divergence free as
shown in Equation (3.8) (may be proven with the Biot-Savart law) and thus it may
be stated that:
∇·B = 0⇒B = ∇×A
(3.24)
For 2D cases, the flux density B has only in-plane components B = Bx î + By ĵ and
current density, J = Jz k̂ and potential A = Az k̂ have only out of plane components.
Thus, magnetic flux density may be expressed as:
B = Bx î + By ĵ =
∂Az
∂Az
î −
ĵ
∂y
∂x
(3.25)
or
∂Az
î
∂y
∂Az
By ĵ = −
ĵ
∂x
Bx î =
(3.26)
(3.27)
39
CHAPTER 3. THEORY
Given ∇ × B = µJ , current density may be related to magnetic potential by:
∇ × ∇ × A = ∇(∇ · A) − ∇2 A = µJ
(3.28)
However, the vector potential does not offer uniqueness. Any function with zero curl,
or a gradient of a scalar function may be added to the potential A since the curl of a
gradient is always zero. Thus, the divergence of A, has no physical significance and
may be chosen to be zero:
∇·A=0
(3.29)
This stipulation is known as the Coulomb gauge condition. Thus, Poisson’s equation
for the magnetic potential is found using the above condition and the identity (3.12):
∇2 A = −µ0 J
(3.30)
Since there is no variation in the out of plane direction for 2D cases, Poisson’s equation
for the magnetic potential is found by equating out of plane terms,
∂ 2 Az
∂ 2 Az
ĵ +
ĵ = −µJz ĵ
∂x2
∂y 2
(3.31)
Physically, magnetic flux (per unit length) is the difference in magnetic potential
between any two points. Considering a 2D surface, S, and orthonormal contour, C,
as shown in Figure 3.1, flux, φ, and potential are related through:
φ=
Z
S
B · ds =
Z
S
∇ × A · ds =
I
C
A · dℓ
(3.32)
40
CHAPTER 3. THEORY
Figure 3.1: Magnetic vector potential interpretation as a the change in magnetic flux
per unit depth for 2D geometry
Since there is no net contribution to the circulation integral where Az is perpendicular
to the path, C, flux is expressed as:
φ = (Az1 − Az2 )L
(3.33)
where L represents the depth of the system. Because flux is defined by the difference
in potential, the potential must be explicitly defined (usually at the domain boundary)
through a boundary condition.
41
CHAPTER 3. THEORY
3.2.2.1 Time Dependance
The magnetic potential is also applicable for time dependant fields. In such cases, a
time varying electric potential (voltage) may be coupled to the magnetic domain. In
the quasi-static case, the electric field E is a vector field which is always irrotational
or has a zero curl component (proven with Coulomb’s Law) and therefore is not influenced by time varying magnetic fields. This property may be used to transform
the problem of solving for the vector E into a problem of solving for a scalar quantity known as the electric scalar potential, V . Since an irrotational vector may be
represented as the gradient of a scalar function the electric field may be expressed as:
E = −∇V
(3.34)
Substituting this expression into Equation (3.3) results in:
∇ · E = ∇ · (−∇V ) = −∇2 V =
ρ
ǫ0
(3.35)
Which is known as Poisson’s equation, or in non-conducting regions (ρ = 0), Laplace’s
equation. Thus the electric potential can be solved for with one differential equation
and then related to E rather than solving for both the divergence and curl of E.
Substituting Equation (3.24) into Faraday’s law, equation (3.7), yields:
∇×E =−
∂
(∇ × A)
∂t
(3.36)
or rearranging,
∂A
∇× E+
=0
∂t
(3.37)
42
CHAPTER 3. THEORY
When written as the gradient of a scalar potential, V
E = −∇V −
∂A
∂t
(3.38)
Equations (3.24) and (3.38) satisfy Maxwell’s equations (3.7) and (3.8). Substituting
Equation (3.10) into the expression for divergence free current density (∇ · J = 0)
results in:
σ∇ · E = 0
(3.39)
Substituting in Equation (3.38), the electric scalar potential yields:
∂A
=0
∇ · −∇V −
∂t
(3.40)
The finite element code (ANSYS) must also solve this equation.
Finally, the relations ∇ × H = J and Equations (3.9) and (3.10) are combined to
yield:
ν∇ × B = σE
(3.41)
Substituting in the magnetic and electric potentials results in:
∂A
ν∇ × (∇ × A) = σ −∇V −
∂t
(3.42)
This may be simplified using the identity in Equation (3.12) and the Coulomb gauge
condition to
−ν∇2 A + σ
∂A
+ σ∇V = 0
∂t
(3.43)
It should be noted that equations (3.30), (3.40) and (3.43) will vary depending on
the material properties. For conducting (steel and copper coil) regions, the following
43
CHAPTER 3. THEORY
equations result:
∇2 A = −µ(H)J
∇·
σ
∂A
− ∇V
∂t
=0
∂A
1
−
∇2 A + σ∇V = 0
∂t
µ(H)
(3.44)
(3.45)
(3.46)
For nonconducting, or air regions, the following equations apply:
∇2 A = −µ0 J
∇·
∂A
− ∇V
∂t
∇2 A = 0
3.2.3
(3.47)
=0
(3.48)
(3.49)
Static Elements
The commercially available finite element software ANSYS was used to solve the magnetic potential PDEs by discretizing the model domain into appropriately configured
elements. For the static models, two element types were used. The two dimensional
element type PLANE13 is used throughout the iron and air regions. Although this
element may be used in non-magnetic studies, the nodal degree of freedom (DOF)
that is solved for in the static analyses is the magnetic vector potential in the z or
normal direction. The PLANE13 quadrilateral element is defined by four nodes at
each corner [ANSYS Inc., 2005]. Nonlinear magnetic materials are permitted by associating an appropriate property table with the element type. Magnetic forces are
CHAPTER 3. THEORY
44
Figure 3.2: Static and transient model mesh and element types for the opener FEA
model
determined using both a Maxwell stress tensor calculation and virtual work calculation on surfaces in contact with air regions by using a macro that recognizes surfaces
that have been identified or ‘flagged’ for such force calculations. An example of the
meshed opener magnetic path is shown in Figure 3.2 with a complete description
of the modeling process provided in Chapter 4, Section 4.3. The elements making
up the armature region of the model are identified for such a force calculation. A
steady-state current density body load is applied to the elements which make up the
coil region. Note that this element does not possess voltage forced capability. During
simulation postprocessing, magnetic flux density is calculated with respect to the two
CHAPTER 3. THEORY
45
dimensions of the element co-ordinate system. Magnetic flux through the device is
calculated by first defining a two point path. A macro then defines a pre-specified
number of points along the path over which flux density is integrated. For all elements
used, a default coordinate system orientation was used. These default systems are
right-handed, orthogonal and parallel to the global cartesian coordinate system.
To model far-field decay without solving for a large amount of additional elements, a
single layer of INFI110 infinite boundary elements were used to surround the air region of the actuator model. These elements use four nodes and shape functions which
force the magnetic potential to zero at infinity [ANSYS Inc., 2005]. These elements
have both four node and eight node capability and only offer the magnetic potential DOF. When used with PLANE13 elements, ANSYS documentation recommends
using the four node option.
3.2.4
Transient Elements
The transient model elements used are similar to those of the static analyses, with
a few minor exceptions. In order to simulate the response of predetermined voltage
input, a model with a circuit coupled voltage source was used. Here, PLANE53 elements were used throughout the air, iron and coil regions instead of PLANE13. This
element is defined by eight nodes, each possessing up to four possible DOFs. These
include the magnetic vector potential, a time-integrated electric scalar potential, electric current, and electromotive force (EMF). It is these additional degrees of freedom
that make PLANE53 applicable to low frequency transient electromagnetic-circuit
coupled analyses. However, only the magnetic vector potential degree of freedom was
considered through the air and iron regions. In the coil region, the electric current
and EMF degrees of freedom were also activated to allow for nodal coupling to the
circuit domain. The circuit consists of three CIRCU124 elements, each of a different
configuration. CIRCU124 elements are based on Kirchhoff’s Current Law with stiff-
CHAPTER 3. THEORY
46
ness matrices based on a lumped circuit model. Independent voltage source, stranded
coil and resistor elements were meshed in series to represent the actuator circuit. The
resistor was typically set to a value close to zero as it was only added to enhance
model flexibility. The resistor only possesses a voltage degree of freedom. An independent voltage source was used to represent the actuator power supply. The voltage
source element has voltage and current degrees of freedom. The voltage degrees of
freedom are specified by terminating one node as a ground, or 0 V, and another node
as a piecewise linear voltage function of time. This allows for the ability to input a
simulated or measured voltage signal for direct comparison with other simulations or
experiments. Finally, the stranded coil element has voltage, current and EMF degrees
of freedom. The voltage degrees of freedom are coupled to the resistor and ground
nodes. The EMF and current DOFs are coupled to a node in the coil region in the
FEA domain. Similarly, all nodes in the coil region have the current and EMF DOFs
coupled to one another. Physical coil properties are implied by the elements from the
coil region FEA domain and specified material constants. As shown in Figure 3.2,
INFIN110 elements were used to surround the air region, however, with the 8 node
option activated, as recommended by ANSYS documentation for use with PLANE53
elements.
3.2.4.1 Eddy Currents
Eddy currents are characterized as local circulating currents which exist in the core
material. These are physically existing currents produced within the material due to
a time varying core flux. They may be thought of as a short circuit consisting of a coil
wrapped around the external core material path in that the change in flux induces a
current which in turn generates its own magnetic flux in the opposite sense (obeying
Lenz’s law) and ultimately opposes the change or rise of flux of the overall circuit.
Thus, the observed flux rise or magnetization curve will be lower than the that of
47
CHAPTER 3. THEORY
the static case. The energy difference between the static and rapid field buildup is
defined through resistive losses and hence heating. In summary, the two effects of
eddy currents are: an internal magnetomotive force (MMF) is generated which tends
to counteract the applied MMF and an irreversible heating loss of energy with the
i 2 Joule heating losses in the core. Thus, greater changes in flux tend to generate
more losses. A widening of the hysteresis loop is an indication of the eddy current
magnitudes. They may be minimized by using materials with low conductivity and
by laminating the core structure (through thin sheets) or sintered powder metallurgy
techniques. The laminations succeed by increasing the circulating path length and by
breaking the eddy current paths into many smaller loops with lower magnitude and
subsequently reduce the counter flux generated. In the actuator studied, the back
iron and armature are made from 0.3 mm silicon steel laminae to mitigate the effect
of eddy currents. Eddy currents are not to be confused with the electrodynamic
Amperian currents generated by electron spin which are used to explain material
magnetism.
3.2.5
Element Shape Functions
Figure 3.3 illustrates the PLANE13, PLANE53 and INFIN110 elements. Coordinates
s and t represent the local element nodal coordinate system. When the coordinates
are used with the shape functions, they are normalized, going from -1.0 on one side
of the element to +1.0 on the other. It should also be noted that s and t are not
necessarily orthogonal to one another. The shape functions for the 2D magnetic vector
potential (z component only) for each of the elements are as follows for PLANE13,
PLANE53 and INFIN110 respectively:
Az P LAN E13 =
1
(Az I (1
4
− s)(1 − t) + Az J (1 + s)(1 − t)
+Az K (1 + s)(1 + t) + Az L (1 − s)(1 + t))
(3.50)
48
CHAPTER 3. THEORY
Figure 3.3: Element Configurations
Az P LAN E53 =
1
(Az I (1
4
− s)(1 − t) + Az J (1 + s)(1 − t)
+AzK (1 + s)(1 + t) + Az L (1 − s)(1 + t))
(3.51)
+ 12 (Az M (1 − s2 )(1 − t) + Az N (1 + s)(1 − t2 )
+AzO (1 − s2 )(1 + t) + Az P (1 − s)(1 − t2 ))
Az IN F IN 110 =
1
(AzI (1
4
− s)(t2 − t) + Az J (1 + s)(t2 − t))
(3.52)
+ 12 (Az K (1 + s)(1 − t2 ) + AzL (1 − s)(1 − t2 ))
The shape functions for voltage are analogous, and can be formed by substituting
the scalar voltage with the magnetic vector potential terms. ANSYS assembles these
functions for each element in shape function matrices. Hence the magnetic potential
A and the scalar electric potential V may be represented by matrices NA and N V
49
CHAPTER 3. THEORY
respectively:
T
A = NA AN = NA
T
and






0
0
AzN






V = N TV VN
(3.53)
(3.54)
ANSYS then uses NA to calculate the flux density as follows:
B = ∇ × N AA N
(3.55)
Where AN and VN are the nodal magnetic and electric potentials.
3.2.6
Solution Process
The FEA discretization process results in a series of simultaneous nonlinear equations
that are numerically solved depending on the material properties (conductive, field intensity dependent) and if the system is time dependent or not. The following provides
a brief overview of the solution procedure (see [ANSYS Inc., 2005, Chladny, 2003] for
more details).
3.2.7
Matrix Assembly
After the 2D model is meshed and appropriate boundary conditions and loads are
applied, ANSYS solves equations of the following form:
[C]d˙ + [K]d = J
(3.56)
50
CHAPTER 3. THEORY
Where the degree of freedom vector is represented by


 Az 
d=

V
(3.57)
Az represents the magnetic vector potential in the Z direction (into the model/page)
as this is the only relevant potential direction for the 2D axis-symmetric case. Note
that this is a relatively large vector as it represents all elements. V is the time
R
integrated electric scalar potential, V = V dt which is input as a voltage excitation
(if one exists).
Matrices [C] and [K] are the coefficient matrices defined as:

L
N
G

 [K ] + [K ] + [K ] 0 
[K] = 

0
0
(3.58)
and,
R
L
K = V ol (∇ × NAT )T ν(∇ × NAT )dV,
R
N
K = V ol (∇ × NAT )T ν(∇ · NAT )dV,
G
R
T
T T
T
T
dν
K = 2 V ol d|B|
2 (B (∇ × NA )) (B (∇ × NA ))dV
Where the element shape function matrices are integrated over their respected volumes. If the model is axis-symmetric, the element positions are mapped to the global
coordinate system so that the appropriate volume can be derived for the entire actuator. As before, ν represents the reluctivity matrix, [µ]−1 , but for the case of
orthotropism, is considered a magnetic field intensity dependant scalar. The nonlinear input B-H cure is converted to a spline fit function of ν vs |B|2 from which the
51
CHAPTER 3. THEORY
derivative
dν
d|B|2
may be taken. The transient coefficient matrices are as follows:


AA Av C
 C

[C] =  
Av
vv
C
[C ]
(3.59)
where,
R
C AA = V ol NAσNAT dV,
Av R
C
= V ol NAσ∇ · N TV dV,
R
[C vv ] =
(∇ · N TV )T σ∇ · N TV dV
V ol
For static analyses (no time dependant potentials or fields), only the K matrices and
magnetic vector potential DOFs are required as Equations (3.44) through (3.49) will
be further simplified when B is time invariant.
The load vector is defined as:

R
T

 V ol JsNA dV 
J = R

T
J N dV
V ol t A
(3.60)
Where Js is the source current density vector (also referred to as current segments)
and Jt is the total current density vector. The total current density vector is equal
to the summation of the source currents, eddy currents, Je and induced velocity
currents, Jv , (not present as the armature is fixed in all simulations).
Jt = Je + Js + Jv
(3.61)
52
CHAPTER 3. THEORY
The eddy current density vector is solved for through the conductivity matrix, [σ]
and rate of change of magnetic potential:
∂Az
∂t
1
= −[σ] Σni=1 NAT Ae
n
Je = −[σ]
(3.62)
(3.63)
where n = 4 is the number of integration points for the quadrilateral elements and
Ae is the magnetic vector potential time derivative. The source current is related to
the electric scalar potential as:
Js = −[σ]∇ · V
1
= [σ] Σni=1 ∇ · N T VN
n
(3.64)
(3.65)
where N is the element shape functions for the electric scalar potential VN at the
integration points.
3.2.8
Static Model Solution
As shown in the previous section, the FEA discretization process results in a series
of simultaneous nonlinear equations as represented by Equation (3.56). For static
models, the time dependant magnetic potential vector and coefficient matrices can
be disregarded and an incremental Newton-Raphson method is used to solve nonlinear
systems by:
T
nr
[Kn,i
]∆Az,i = Jn − Jn,i
(3.66)
T
Where [Kn,i
] is the Dirichlet matrix for sub-step n, and iteration i. Az,i is the
magnetic potential vector at iteration i and ∆Az,i = Az,i+1 − Az,i. Jn is the applied
nr
current density vector for a given sub-step, n, and Jn,i
is referred to as the resisting
53
CHAPTER 3. THEORY
load vector which is calculated from element magnetic fluxes. The right-hand side
of Equation (3.66) is referred to as the residual or out-of-balance load vector and
represents the amount the system is out of equilibrium. A predetermined number of
sub-steps are required for solution convergence when the system is highly nonlinear
or path-dependent. These intermediate steps are performed so that the final current
density vector J is achieved by applying it in increments. At each sub-step the
Newton-Raphson procedure is performed by assuming a potential vector Az,o, which
is usually obtained from the last converged iteration, Az,i. An updated coefficient
T
nr
matrix, [Kn,i
] and load vector, Jn,i
are determined from the magnetic potential vector,
Az,i. Next ∆Az,i is solved for from Equation (3.66). Then Az,i+1 is computed by
adding ∆Az,i to Az,i or Az,o if it is the first iteration. These equilibrium iterations are
repeated until convergence is achieved for each sub-step and the current density vector
is fully applied. Convergence checking can be based on either magnetic potentials,
current segments, or both. However, for 2D models, ANSYS recommends convergence
to be determined by current segments as:
qX
(J − Jinr )2 < ζJ JRef
(3.67)
Where ζJ is a specified tolerance of a typical current segment value, JRef , which is
taken as kJ k. For magnetic potential convergence, ANSYS compares the change in
nodal potential values between successive equilibrium iterations to a similar criterion
as above.
3.2.9
Transient Model Solution
For the transient models, a generalized trapezoidal rule
Az,n+1 = Az,n + ∆tȦz,n+1
(3.68)
CHAPTER 3. THEORY
54
is used to numerically integrate Equation 3.56 over a series of pre-specified time
steps, ∆t = tn+1 − tn prior to the application of the incremental Newton-Raphson
method as discussed in the static analyses. A sufficiently small time step is required
to ensure the applied voltage load waveform is captured and for solution convergence.
The magnetic potentials at time tn are represented by Az,n and the respective time
derivative Ȧz,n is calculated at the previous time step. Substituting the trapezoidal
approximation at time tn+1 into Equation 3.56 results in:
1
1
[C] + [K] Az,n+1 = J +
[C]Az,n
∆t
∆t
(3.69)
This equation set is then solved in a similar fashion to the static models. However,
sub-steps are now replaced with a time step, (ie. the load is no longer ramped).
Therefore, to ensure convergence, a sufficiently small time step size must be used.
Upon calculation of Az,n+1 , Equation (3.68) is used to update Ȧz,n+1 .
3.3
Lumped Parameter Modeling
Although FEA techniques are ideal for providing accurate field solutions to relatively
complex nonlinear and coupled problems, the time and computational resources required are potentially untenable. Additionally, the models are not conducive to application of modern control design. As a result, a simplified lumped parameter model
is sought that sufficiently describes the actuator behavior yet is still computationally
simple enough to incorporate into a real-time control algorithm. The actuator may
be characterized by electromagnetic and mechanical subsystems as discussed in the
following sections and shown schematically in Figure 3.4.
CHAPTER 3. THEORY
55
Figure 3.4: Cross-sectional schematic of hinged actuator indicating armature motion
and magnetic flux
3.3.1
Reluctance Network
A common approach to modeling magnetic circuits is through construction of a reluctance network. Often, linear induction is assumed so that the relative permeabilities
of the backiron and armature materials are independent of field intensity. For most
magnetic devices, this assumption is reasonable for lower field intensities. Since the
linear permeability of steel is approximately 1000 times greater than air, the steel
material properties are expected to be of little significance in the inductance calculation at large air gaps (or prior to the onset of saturation). Upon material saturation,
the steel permeability is the same as free space, µ0 , or air, and hence the reluctance
of the steel flux path lengths become as significant as the air gaps. Saturation is
expected to occur at relatively large excitation levels and small air gaps, as it is at
those operating points where the magnetic field intensity will be greatest.
56
CHAPTER 3. THEORY
By taking advantage of the linear permeability relationship, B = µr H, between flux
density B, and applied field intensity, H, flux may be related to the material properties as:
φ = µr HApath
(3.70)
where Apath is the path cross sectional area. Note that for nonlinear or saturable
materials µr = µr (H). H can be related to the number of turns of the coil, N, the
current, i, and the length of the flux path, l by:
H=
Ni
M
=
l
l
(3.71)
where M is the magnetomotive force (MMF). Equation 3.70 can now be rearranged
as:
φ=
µr Apath Ni
l
(3.72)
This expression can be used to discretize the geometry and material of a linear magnetic device which results in a network of regions that can be solved in a similar
fashion as classical circuit analysis. Thus, flux for a device consisting of n different
linear materials in series can be expressed as:
φ=
n
X
µj Aj Nij
i=1
lj
(3.73)
In such cases, each of the elements are analogous to resistors in an electric circuit. In
magnetic terms, they are cumulatively known as the system reluctance, ℜ, where
ℜ=
M
Hl
Bl
l
=
=
=
φ
φ
µr φ
µA
or by definition of inductance, ℜ =
N2
.
L
(3.74)
Note that in devices with air gaps, it is often
possible to neglect the reluctance of materials with high permeability much in the
57
CHAPTER 3. THEORY
same way the resistance of wires in an electric circuit can often be neglected. Consequently, for devices such as solenoids, flux linkage, λ = λ(x, i), is highly dependant
on the air gap, or position of the armature, as well as current excitation or MMF.
Flux can finally be stated as:
φ(x, i) = M
n
X
1
ℜj
j=1
(3.75)
Where the magnetic system is discretized into n regions of unique reluctance elements,
ℜj .
3.3.2
Nonlinear Induction
As a means of providing an accurate inductance model, magnetic material saturation
is considered. Saturation effects will be present at high MMF values, particularly at
small armature/pole face air gaps and/or high current excitation [Chladny et al., 2005].
The following function is intended to approximate the net magnetic circuit flux linkage
response with magnetic material saturation as in [Ilic’-Spong et al., 1987]:
λ(x, i) = ψ(1 − e−ig(x) ),
(3.76)
where,
g(x) =
β
+ α.
κ−x
(3.77)
The parameters ψ, β, κ and α are identified with MATLAB using a nonlinear least
squares fit to the FEA model data. Specifically, the nlinfit.m m-function uses the
Gauss-Newton algorithm with Levenberg-Marquardt modifications for global convergence.
58
CHAPTER 3. THEORY
3.3.3
Electric Coupling
By making use of Kirchhoff’s second law and Faraday’s law, a differential equation
relating a DC source, coil resistance and flux linkage may be expressed as
vSource = Ri +
dλ(i, x)
dt
(3.78)
Eddy currents are not modeled as preliminary FEA modeling and experimental tests
indicated negligible contribution to overall current response 1 . A lumped parameter
method of accommodating eddy currents using FEA data was developed for a similar
actuator in [Chladny et al., 2005].
3.3.4
Magnetic Co-energy
When the actuator armature moves, energy is exchanged among three forms. Namely,
the mechanical system, the electrical system and the magnetic field. It is possible to
consider an energy balance of the entire system when observing the nature of the force
development so that a complex field analysis may be avoided [Schmitz and Novotny, 1965].
The general energy balance equation for the three systems may be expressed as:
∆We = ∆Wc + ∆Wm
(3.79)
where ∆We , ∆Wc , ∆Wm represent changes in electrical, magnetic and mechanical
energies respectively.
Since mechanical force may be expressed as a change of energy over a change in
1
The actuator back iron and armature are constructed from laminated silicon steel sheets. The
laminations succeed by breaking the eddy current paths into many smaller loops with lower magnitude and subsequently reduce the counter flux generated.
59
CHAPTER 3. THEORY
displacement, ∆x the average force on the armature may be stated as:
Favg ∆x = ∆Wm
(3.80)
Implicitly, it may be observed from Equation 3.79 that for any given position, the mechanical energy may be determined from either magnetic flux linkage, λ, or electrical
current, i. Thus, the average force may be expressed as:
Favg =
∆We (i, x) ∆Wc (i, x)
−
∆x
∆x
(3.81)
Favg =
∆We (λ, x) ∆Wc (λ, x)
−
∆x
∆x
(3.82)
or
For a simple air-coil-resistor circuit, Wc can be expressed as the stored energy in the
coil over a change in time, ∆t = t2 − t1 , as:
Wc =
Z
t2
iVCoil dt
(3.83)
t1
Using Faraday’s law of induction, VCoil =
Wc =
Z
t2
t1
dλ
,
dt
yields:
dλ
idt =
dt
Z
λ2
idλ
(3.84)
λ1
For such linear systems, the relationship between flux linkage, λ, and current, i is
often expressed as the coil self-inductance, L as:
L=
λ
i
(3.85)
Inductance is analogous to mechanical inertia or mass as it resists any change in
current due to an applied voltage, just as a mass resists a change in velocity due to
60
CHAPTER 3. THEORY
Figure 3.5: Schematic of Simulink modeling process
an applied force. If the system is excited from λ1 = 0, the stored field energy may be
expressed as:
Wc =
Z
0
3.4
λ
1
idλ = Li2
2
(3.86)
Simulink Model
As a means of hybridizing the accuracy of a FEA field solution with the expedient
solution time and flexibility of a LPM, a MATLAB-Simulink model of the actuator
system is designed. The intent of this model is to accurately represent the actuator
system for control algorithm evaluation. Perhaps the most significant aspect of the
modeling approach is the use of FEA generated data in look-up tables to simulate the
electromagnetic response. Here, flux and force FEA data are input with respect to armature position and excitation current. Faraday’s equation is numerically integrated
to derive flux linkage subject to applied coil voltage and coil resistance. As Figure 3.5
61
CHAPTER 3. THEORY
indicates, the model may be considered a LPM-FEA hybrid model as, technically, it
is neither but incorporates elements of both in an attempt to use the accuracy of a
field solution with the computational simplicity of solving a system of ODEs. Thus,
it has greater accuracy than a strictly ODE based model, but is not amenable to
analytic control design. Rather, its primary purpose is to simulate the experimental
testbench conditions for evaluating control performance with model-plant mismatch.
After models of the circuit and mechanical dynamics were implemented and experimentally validated (see Chapter 6), representations of the power electronics and gas
force dynamics were added. Descriptions of the sub-models are provided in Chapter
4. Due to the coupled multi-disciplinary nature of the model and the switched nature of the power electronics, a stiff solver was used. In such instances, MATLAB
recommends use of the multi-step solver ode15s based on numerical differentiation
formulas (NDFs) [Shampine and Reichelt, 1997].
3.5
Differential Flatness
The feedback control algorithm used in this work exploits a property of the actuator system known as “flatness” as first outlined in [Martin, 1992, Fliess et al., 1992]
using differential algebraic techniques. Most simply stated, a system may be considered (differentially) flat if the state(s) and input(s) may be expressed explicitly
as a function of the ‘flat’ output(s) and a finite number of output time derivatives
[Lévine, 2004, Martin et al., 1997]. More specifically, a system, ẋ = f (x, u), may be
considered flat provided the state variables, xǫRn , and inputs, uǫRm , can be parameterized by an m-dimensional flat output yǫRm of the form:
y = γ x, u, u̇, . . . , u(α) ,
(3.87)
62
CHAPTER 3. THEORY
satisfying:
x = φ y, . . . , y (β)
u = ψ y, . . . , y (β)
(3.88)
The outputs may correspond to physically measurable parameters or to some fictitious output. This endogenous feedback results in an equivalent linearization such
that an expression of the system dynamics can be related to the output without
the need for numerical integration [Mercorelli et al., 2003, Martin et al., 1997]. Figure 3.6 provides a graphical representation of how a trajectory may be designed in
flat output space, then one-to-one mapped (smoothly) to the original input and state
space. Consequently, the property is useful for solving motion planning and control
tracking problems since the problem of solving a dynamic system is reduced to an
algebraic (non-differential) one [Fliess et al., 1995]. Many mechanical systems may
be characterized as being differentially flat [Murray et al., 1995]. For example, linear
controllable systems are flat as well as input-output linearizable systems (by definition). Although no necessary and sufficient conditions have yet been determined, some
equivalence theorems exist [Rathinam and Sluis, 1995, van Nieuwstadt et al., 1994].
As a result, in most cases the only way to demonstrate that a system is flat is through
finding an appropriate output (or set thereof) that satisfy the conditions (3.87) and
(3.88). The flatness definition in (3.87) was later extended using differential geometry
techniques to include outputs with infinite coordinate dependance [Fliess et al., 1999]
and time transformations or orbital flatness [Fliess et al., 1994].
3.6
Summary
The preceding sections briefly summarize the key analytic and numerical tools used
during the course of the actuator modeling and control. Specifically, Maxwell’s general
CHAPTER 3. THEORY
63
Figure 3.6: Correspondence between flat output space and state space [Chung, 2005]
equations are subjected to the quasi-static approximation and the magnetic vector
and electric scalar potentials are used to provide differential equations which govern
air and iron regions of a 2D system. ANSYS applies these equations over discretized
regions and assembles them into a system of equations which are then solved using
a Newton-Raphson procedure. As well, methods of modeling magnetic circuits for
the purpose of control design are outlined. The hinged actuator is simulated using
the MATLAB-Simulink numerical solving environment for preliminary control performance evaluation. Finally, the concept of differential flatness is summarized which
provides a system characterization framework that allows for trajectory planning and
tracking development without the need for numerical integration. The remaining
CHAPTER 3. THEORY
64
chapters will expand on these techniques and the assumptions used in the context
of the hinged solenoid actuator system and in contrast to experimentally measured
results.
Chapter 4
Modeling and Simulation
4.1
Introduction
A
s part of the overall goal for implementation of the hinged actuator on a single
cylinder research engine, extensive modeling and simulation are conducted for
control design and visualization purposes. Solid modeling is performed as part of the
finite element analysis and testbench design. Finite element and lumped parameter
models are developed and experimentally validated for control design purposes. The
results of the these models are implemented in a MATLAB-Simulink model that also
includes representations of the power electronics and exhaust gas force disturbances
to further improve model scope and fidelity. A strong emphasis is placed on modeling
in part due to the nonlinear control algorithms used. Controllers derived through
feedback linearization techniques are model dependant and thus poor model fidelity
could adversely affect performance. All models are essential in the development,
evaluation and debugging of potential control algorithm candidates prior to hardware
implementation.
65
CHAPTER 4. MODELING AND SIMULATION
66
Figure 4.1: 3D solid model of the prototype actuator
4.2
3D Solid Modeling
An accurate geometric representation of the prototype actuator is desired for modeling, experimental design and visualization purposes. A model of a cylinder head is
also required for the testbench and custom single cylinder engine designs. The modeling package Pro/Engineer, by Parametric Technology Corporation (PTC), is used
to generate models of the actuator, cylinder head and related experimental hardware. The armature and back-iron geometry is exported from Pro/E to ANSYS to
CHAPTER 4. MODELING AND SIMULATION
67
Figure 4.2: 3D solid model of the actuator, exploded view
simplify the input during geometry definition. An example of the modeled actuator
unit is shown in Figure 4.1. An exploded view of the actuator and two actuator
units mounted to a cylinder head and cylinder are provided in Figures 4.2 and 4.3,
respectively. The intake and exhaust actuator units differ slightly to accommodate
differences between intake and exhaust valve requirements (such as spring stiffness
and valve geometries). Although there are two valve actuators per actuator unit (for
a four valve cylinder head), the actuation operations need not be coupled.
The model of the armature is also used in solid mechanics type finite element analysis
(using PTC’s Pro/Mechanica analysis software) to investigate the effect of torsion bar
deflection on nominal magnetic airgap.
CHAPTER 4. MODELING AND SIMULATION
68
Figure 4.3: 3D solid model of the actuator, cylinder head and custom engine cylinder
4.3
Finite Element Modeling
Two dimensional representations of the opener and closer are separately modeled to
minimize model complexity and computational time and because they have different
geometry. Due to flux path geometry, the magnetic field is expected to vary most
significantly in only two dimensions. The 2D simplification may be justified provided
the actuator has a sufficiently large depth, small air gap and minimal eddy currents
[Takahashi et al., 1991]. The actuator in this study has a depth over eight times
greater than the flux path width and is constructed from laminate silicon steel sheets
(0.3 mm thick) for eddy current suppression. The maximum air gap is approximately
CHAPTER 4. MODELING AND SIMULATION
69
Figure 4.4: Modeled actuator flux path sections
the same as the flux path width. Using 2D geometry for the 3D actuator assumes that
the more complex 3D eddy current paths during transients and fringing in the corners
of the back iron and armature are negligible. Aside from the added computational
load, full 3D analyses are also subject to potentially greater accuracy and convergence issues [Prieto et al., 2005]. Additionally, no flux path scaling or transformation
is required since the modeled and actual flux path areas are the same (since the device is not modeled as axis-symmetric) [Li and McEwan, 1993]. The simplification of
modeling the opener and closer components separately is justified through the high
permeability of the armature, relatively large distance between the two magnets, and
since in practice, only one coil is active at any time. Although similar data could be
gathered through experimental studies, a more expedient design process would be to
simulate the actuator response and control performance prior to prototype fabrication
and evaluation. Figure 4.4 illustrates the modeled opener and closer coil and steel
cross-sections that will most significantly influence the flux path. In both static and
transient studies, reference is made to air gaps or the distance from the armature to
CHAPTER 4. MODELING AND SIMULATION
70
Table 4.1: Air gap and excitation operating points
Air gaps [mm]
0.02 0.03 0.05
0.99 1.32 1.98
0.07 0.10 0.13 0.20 0.33 0.49 0.66
2.63 3.29 3.95 4.61 5.27 6.58 8.00
Coil excitations / MMF [Ampere-turns]
10
25
50 100 200 300 400 500 600 800
1000 1250 1500 1750 2000 2250 2500 3000 3750 5000
the respective magnet pole face. Due to the hinged nature of the actuator, the air
gap distance is ambiguous and therefore all air gap references are made with respect
to valve position. For example an opener air gap of 0.50 mm corresponds to a valve
position that is 0.50 mm away from being fully opened (or x = 3.50 mm). Similarly, a
closer air gap of 1.50 mm corresponds to a valve position that is 1.50 mm away from
being fully closed (or x = −2.50 mm). This is done to explicitly acknowledge the
affect the air gap has on the magnetic response while still referencing a state variable.
4.3.1
Static Modeling and Simulation
Static 2D models of the opener and closer are created using geometry generated
from Pro/Engineer and the ANSYS Parametric Design Language (APDL). For each
operating point, macros are called from a submitted batch file that assigns appropriate
geometry, mesh, material properties, boundary conditions and current excitations.
The resulting force and flux data are exported to file for further processing with
MATLAB. Results for the opener are determined for each of the operating points
listed in Table 4.1 for a total of 400 static solutions. The operating points were
selected to provide a relatively smooth force and flux relation as a function of air gap
and current. A higher number of data points are required for a smooth data set at
low air gaps due to the dramatic change in magnetic flux and force in these regions.
Similarly, due to material saturation, a higher resolution of data are required at
CHAPTER 4. MODELING AND SIMULATION
71
Figure 4.5: Static and transient model mesh and material types for the opener FEA
model
lower excitation. Mesh refinement is determined by inspection as well as by ensuring
force and flux convergence with respect to an increase in element density. In order
to prevent elements with poor aspect ratios, active mesh control is established in
the back iron, armature and air gap regions in addition to the model boundaries.
The default auto-mesh generator is used to mesh the remaining regions with the
finest mesh refinement possible. In order to ensure appropriate element densities and
shapes at extreme armature positions, a linear function is used to control the element
mesh in the air gap region over the 8 mm range of valve motion. For both iron and
air regions, 2D quadrilateral elements with a magnetic potential degree of freedom
CHAPTER 4. MODELING AND SIMULATION
72
are used. Figure 4.5 illustrates a full and close up view of a typical mesh over the
armature, air and opener back iron. A single layer of boundary elements is used
around the perimeter of the model to model far field decay. These infinite elements
use shape functions which require the magnetic potential to be zero at infinity. The
backiron region is divided into two material type regions to better represent the
anisotropic induction resulting from aligned grain structures of the laminate sheet.
Unlike the transient simulations, where coil current is coupled to circuit elements,
excitation is applied directly to the coil elements in the form of current density.
Validity of the model is assessed by comparing simulation results to experimental
measurements (see Chapter 6).
4.3.2
Transient Modeling and Simulation
The transient behavior of the model is determined by applying a voltage step at a constant air gap to a quasi-static transient FEA model. A step voltage is chosen as it is a
typical output waveform of driver circuits [Xiang, 2002, Amato and Meuller-Heiss, 2001,
Lequesne, 1990]. Quadrilateral elements with additional electromotive force (EMF)
and current degrees of freedom are used for circuital excitation and to account for
transient effects such as eddy currents. A voltage source, resistor, and stranded coil element are modeled to excite the actuator finite element domain as represented in Figure 4.5. The circuit elements are not part of the field solution. Rather, the stranded
coil element’s current and EMF degrees of freedom are coupled to the coil elements
in the actuator domain. The coil resistance, assumed independent of temperature,
is accounted for through the geometry of the FEA and the specified conductivity of
copper. Approximate step waveform voltages, measured from an actual experiments,
are applied to the FEA model to allow model validation and comparison with experimental data (see Chapter 6).
For further details of the finite element solution process, see [Chladny et al., 2005,
CHAPTER 4. MODELING AND SIMULATION
73
Chladny, 2003].
4.4
Plant Derivation
In the following sections, two types of lumped parameter models of the hinged actuator are presented for use in the development of control algorithms. In both cases,
the actuator system can be considered to be comprised of magnetic, electrical and
mechanical domains.
4.4.1
Magnetic Subsystem
Two magnetic models are presented for the hinged actuator. The first neglects
changes in magnetic path caused by material saturation and is derived from the material properties and through a reluctance network method. A second magnetic model
is presented based on the work of [Ilic’-Spong et al., 1987]. The latter model heuristically represents the net magnetic circuital response and is parameterized through
numerical fitting to a FEA data set.
4.4.1.1 Reluctance Network - Linear Induction
The following details the derivation of a plant model for the hinged actuator assuming
linear induction and the reluctance network method. A schematic of the magnetic
path for the hinged opener magnet is shown in Figure 4.6. In this case, the path has
been discretized into eight regions represented by various path lengths and the four
relative permeabilites. The net reluctance may be thus expressed (using a small angle
approximation) as:
ℜ=
1
(αℓ + δℓ + γℓ + ξℓ )
A
(4.1)
CHAPTER 4. MODELING AND SIMULATION
74
Figure 4.6: Opener magnetic path
where,
ℓarm1 + ℓarm2 + ℓarm3
µarm
ℓbkirn1 + ℓbkirn2
δℓ =
µirnlg
ℓbkirn3
γℓ =
µirnrt
ℓair1 + ℓair2
ξℓ =
µair
αℓ =
A = flux path cross-sectional area
(4.2)
CHAPTER 4. MODELING AND SIMULATION
75
Note that the air gap path lengths, ℓair1 and ℓair2 , only vary in relation to valve
position x. The derived reluctance may be related to inductance by:
L=
N2
N 2A
=
ℜ
αℓ + δℓ + γℓ + ξℓ
(4.3)
This expression can then take a more general form by combining the fixed path terms
and relating the changing airgap to valve position, x. Specifically,
L(x) =
βℓ
κℓ − x
(4.4)
where βℓ and κℓ are constants determined by the reluctance of the magnetic core,
air and armature flux path. For improved accuracy, these constants may be fit to
experimental or FEA simulated data. Recall from Chapter 3 Equation (3.85), that
inductance may also be defined as the ratio of path flux to applied current. The
expression for flux linkage is thus:
λℓ =
βℓ i
κℓ − x
(4.5)
4.4.1.2 Nonlinear Induction Model
As a means of providing a more accurate model of the physical system, magnetic
material saturation is considered. Saturation effects will be present at high magnetomotive force (MMF) values. Specifically, at small armature/pole face air gaps and
or high current excitation. Due to the soft landing requirements of this system, the
feedback control is active during such conditions (and thus requiring accurate modeling). The following function, proposed in [Ilic’-Spong et al., 1987], is intended to
CHAPTER 4. MODELING AND SIMULATION
76
heuristically approximate the flux linkage response with magnetic material saturation:
λ(x, i) = ψ(1 − e−ig(x) )
(4.6)
where
g(x) =
β
+α
κ−x
(4.7)
Using this form, the parameters ψ, β, κ and α are obtained with a nonlinear least
squares fit of collected experimental or numerically simulated force, position and
current excitation data and listed in Appendix B.
4.4.2
Electric Subsystem
The following sections provide details of the derivation of the electrical domain of the
actuator system in the case of the linear and nonlinear magnetic induction models.
4.4.2.1 Electric Subsystem - Linear Induction
The magnetic and electric domains are coupled through a source voltage and circuit
current which can be represented by an RL circuit described by Faraday’s law of
induction and the following KVL equation:
dλ(i, x)
dt
di
= iR + L(i, x) + Ψ(i, x)ẋ
dt
v = iR +
(4.8)
with applied voltage, v, coil current, i, λ(i, x) representing the magnetic flux linkage
of the electromagnet and L(i, x) the self-inductance of the electromagnet. In addition,
Ψ(i, x) represents the back EMF, R, the total resistance of the coil winding and, x,
CHAPTER 4. MODELING AND SIMULATION
77
the valve position (and thus relates to the effective air gap between the armature
and magnetic core pole face). Substituting the flux linkage model from (4.5) into the
defining KVL relation, (4.8), yields the following ODE:
dL(x)
ẋ
dx
β
di
βi
= iR + (
) +
ẋ
κ − x dt (κ − x)2
v = iR + L(x)i̇ + i
(4.9)
Solving for rate of change of current yields
di
i
(κ − x)
=−
ẋ +
(v − iR)
dt
(κ − x)
β
(4.10)
4.4.2.2 Electric Subsystem - Nonlinear Induction
For the nonlinear induction model, a similar procedure is used to derive the following
KVL equation which considers magnetic material saturation:
dλ(x, i)
dt
d
= iR +
ψ 1 − e−ig(x)
dt
ψg ′(x)
= iR + 2
1 − (1 + ig(x))e−g(x)i .
g (x)
v = iR +
(4.11)
where g ′(x) is
g ′ (x) =
β
(κ − x)2
(4.12)
CHAPTER 4. MODELING AND SIMULATION
78
Solving for rate of change of current yields
β
−βψiẋ + ei( κ−x +α) (κ − x)2 (v − iR)
di
=
dt
ψ(β + α(κ − x))(κ − x)
g(x)i
1
e
(v − iR)
′
=
− ig (x)ẋ
g(x)
ψ
4.4.3
(4.13)
(4.14)
Magnetic Force Calculation
A relationship between magnetic force, airgap and current is derived through coenergy, Wc . Here, the change in field energy is related to a change in armature
position [Woodson and Melcher, 1968]. To derive the co-energy of the system based
on the linear induction model, flux linkage, λℓ , is integrated with respect to current,
i:
Wcℓ (x, i) =
Z
i
λℓ (x, ξ)dξ
0
i2 β
=
κ−x
(4.15)
where Wcℓ represents coenergy of the linear induction model. Differentiating with
respect to x yields the expression for magnetic force:
dWcℓ (x, i)
dx
i2 βℓ
=−
(κℓ − x)2
Fmℓ (x, i) =
(4.16)
4.4.3.1 Magnetic Force - Nonlinear Induction
In a similar procedure, the nonlinear inductance model described in Section 4.4.1.2
is used to develop a relationship between magnetic force, airgap and current through
CHAPTER 4. MODELING AND SIMULATION
79
co-energy, Wc .
Wc (x, i) =
=
Z
Z
i
λ(x, ξ)dξ
0
i
0
β
−ξ( κ−x
+α)
ψ 1−e
β
dξ
β
ψe−i( κ−x +α) i( κ−x
+α)
=
κ+e
(βi + (−1 + αi)(κ − x)) − x)
β + α(κ − x)
(4.17)
Differentiating with respect to x yields the expression for magnetic force:
dWc (x, i)
dx
β
β
ψe−i( κ−x +α)
i( κ−x
+α)
=
−βi
+
(−1
+
e
−
αi)(κ
−
x)
(β + α(κ − x))2 (κ − x)
ψg ′ (x)
= 2
1 − (1 + ig(x))e−g(x)i
g (x)
Fm (x, i) =
4.4.4
(4.18)
Mechanical System
The mechanical subsystem is represented through application of Newton’s second
law relating the system forces to the armature and valve acceleration. Due to the
way in which the armature of the hinged actuator is constrained (see Figure 3.4),
the imparted magnetic force may be expressed as a moment. The moment can be
calculated by using the radial distance from the armature revolute joint to the position
where the distributed magnetic load may be equivalently resolved as a point load, ℓm .
Due to the relatively small change in angle, this parameter is assumed constant with
air gap. Valve force can be resolved with the radial distance from the armature
pivot point to where the longitudinal armature and valve axes intersect, ℓv . In this
way, Newton’s second law can be applied to derive valve motion as a function of
effective magnetic, torsion bar, spring, and viscous damping forces with respect to
80
CHAPTER 4. MODELING AND SIMULATION
valve position:
ẍ(mv +
b̂
k̂
Fm,j (i, x)ℓm
Io
)
+
ẋ(b
+
)
+
x(k
+
)
+
F
+
F
=
v
v
vpl
g
ℓ2v
ℓ2v
ℓ2v
ℓv
(4.19)
where: x is the valve position, mv is the valve and moving spring mass, bv is the friction
damping coefficient associated with the valve, b̂ is the viscous damping coefficient
associated with the armature, kv is the valve spring constant, k̂ is the angular torsion
bar spring constant, Fv is the valve spring pre-load, Fg is the gas force acting upon
the valve and Fm,j = −Fm,o , Fm,c is the magnetic force on the armature, with jǫo, c
to indicate the opener or closer magnet respectively. Typically only one magnet is in
operation at any given time. This may be further reduced to:
1
ẍ = −
m
with m = mv +
Io
ℓ2v
ℓm
ẋb + xk − Fv + Fg −
Fm,j (x, i)
ℓv
representing the effective system inertia, k = kv +
(4.20)
k̂
ℓ2v
the effective
spring constant and b = bv + ℓb̂2 the effective damping coefficient. The valve and armav
ture are considered to be rigidly coupled. Sufficiently small angles are assumed such
that sin θ ∼ θ, since θ is limited to ±6o . Armature and valve impacts are ignored since
in experimental results in this work and in [Eyabi, 2003] it is reported that impact
speed less than ≈ 0.2 m/s produced negligible armature or valve bounce. Thus, armature impacts are modeled simply by setting acceleration and velocity to zero when
the armature reaches the stroke bounds. Impact speeds of approximately 0.1m/s
or less are needed for successful control. The total valve stroke is represented by:
S = 8.00 mm. External forces such as those caused by gravity are considered negligible. Effective moving mass, spring pre-load, spring constant and friction parameters
are determined using a grey-box model system identification technique described in
Appendix B.
81
CHAPTER 4. MODELING AND SIMULATION
4.4.5
State Space Formulation
With the magnetic, electrical and mechanical relations now derived, the state vector
is defined as: x = [x1 x2 x3 ]T where x2 = κ − x for notational convenience.
4.4.5.1 State Space Model - Linear Induction
For the linear induction model, the resulting state space model is:



 ẋ1  

 
 ẋ  = 
 2  

 
ẋ3

x1 x3
x2
−x3
x3 b+(x2 −κ)k−Fg −Fv
m
+
x21 βℓm
mℓv x22

x2
β

 

 

+  0 u
 

 

0
(4.21)
where input u is defined as u = v − iR.
4.4.5.2 State Space Model - Nonlinear Induction
Again substituting x2 = κ − x and defining the state vector x = [x1 x2 x3 ]T results
in the following nonlinear state space model which includes saturation effects:




x1 g ′ (x2 )x3
g(x2 )
 ẋ1  

 
 ẋ  = 
−x3
 2  

 
− m1 −x3 b + (κ − x2 )k − Fv + Fg −
ẋ3
ℓm
F (x , x2 )
ℓv m 1

 
 
+
 
 
eg(x2 )x1 u
ψg(x2 )
0
0






Note that both systems are in control affine form [Isidori, 1997]:
ẋ = f (x) + g(x)u
ȳ = [0 1 0] x
(4.22)
CHAPTER 4. MODELING AND SIMULATION
4.5
82
Gas Model
Motion control of the valve subject to cycle-to-cycle gas force disturbances adds an
additional challenge to the soft landing objectives. In-cylinder pressures at exhaust
valve opening may change from over 5 bar to about 1 bar in consecutive cycles in
depending on the engine operating point (or even less in the event of a misfire).
Identifying and rejecting these disturbances is crucial to successful exhaust actuator
implementation in a real engine. Negative pressures such as those potentially encountered during regenerative braking are not considered and at any rate, will likely be
more predictable due to the change in operating mode.
4.5.1
Gas Pressure
To predict the transient pressure disturbance, a compressible flow model is used
assuming an initially sealed volume (cylinder with valves closed) which contains a
mass of exhaust gas at a given pressure and temperature. The gas is expelled through
the time varying valve opening. The cylinder volume varies with piston position which
is a function of engine crank position and time. Upon valve opening, mass transfer
takes place between the volume and atmosphere. The flow is considered isentropic
through a duct (exhaust port) with area, At , defined as a function of valve position.
The gas is assumed to be homogeneous (either pure air or exhaust depending on the
case simulated) and the Mach number at the minimum port area is dependent on
the in-cylinder and downstream exhaust pressure ratio. Using the aforementioned
idealizations, mass flow rate may be approximated as compressible flow through a
nozzle:
ṁg =
Cd At MP
p
kt /(Rg T )
(kt +1)
[Çengel and Boles, 1993]
(4.23)
(1 + (kt − 1)M 2 /2) 2(kt −1)
where: mg is the mass of the gas inside the cylinder, Cd is the effective flow discharge
coefficient, Rg is the ideal gas constant for the gas, T is the gas temperature, At is
83
CHAPTER 4. MODELING AND SIMULATION
the valve-position-dependent effective throat flow area, M is the local Mach number
at the effective valve throat, kt is the specific heat ratio of the gas (kt is 1.4 for air
and 1.3 for exhaust gas [C. R. Ferguson, 2000]). Local Mach number is expressed as a
function of heat ratio and pressure ratio (in this case, back pressure is assumed to be
atmospheric, Patm ). Mach number at the valve throat is limited to unity [Blair, 1999]
and is expressed as:
M(P ) =
v
u
u
t
2
kt − 1
Patm
P
−(kkt −1) −1
t
(4.24)
Mass flow rate and pressure flow rate are related using the ideal gas relation
P Vc = mRg T
(4.25)
Cylinder volume, Vc , varies as a function of engine crank angle, θc ,
πd2
Vc = − b
4
s
+
Ls
Ls
−Lr −
+ ( cos(θc )
2
2

2
Ls
2
Lr −
sin(θc )  + Vres
2
[Heywood, 1988]
(4.26)
where: θc varies from 0o to 360o with TDC at 0o , 360o and BDC at 180o , db is the
cylinder bore diameter, Lr is the connecting rod length, Ls is the piston stroke length,
Vres is the combined cylinder head and crevice volume.
The volume time derivative is
1
V̇c = d2b Ls π sin(θc )
8
Ls cos(θc )
p
4L2r − L2s sin2 (θc )
!
+ 1 θ̇c
(4.27)
84
CHAPTER 4. MODELING AND SIMULATION
Crank angle as a function speed, Ne (in revolutions per minute), initial crank angle
θo and time, t may be expressed as
θc = π sin(
πNe
t + θo )
30
(4.28)
The time derivative of crank angle is
πNe
π 2 Ne
cos(
t + θo )
30
30
θ̇c =
(4.29)
Thus, an approximate expression for pressure as a function of initial pressure, Po ,
valve lift, xL = (x + S/2) m, crank angle and time
Cd V
P = Po −
Rg T
Z
t
0
At (xL )M(P )P
p
kt /(Rg T )
(kt +1)
dt
(4.30)
(1 + (kt − 1)M(P )2 /2) 2(kt −1)
The discharge coefficient, Cd , of 0.9 is estimated based on other four-valve geometries
[Blair, 1999]. Valve curtain area, At , is valve lift dependant and may be represented
as a frustum of a cone [Blair, 1999]. For a valve seat angle of 45o, area may be
expressed by
At =











π(dos+dis )
2
πd
√
(d
2 is
q
(d −
dos −dis 2
)
2
is 2
+ ( dos −d
) , d ≤ dlim
2
(4.31)
+ d2 ), d > dlim
where dis and dos are the respective inner and outer valve seat diameters. For a valve
with a seat angle of φv = 45o , a lift limit beyond which the distance between the valve
and valve seat is no longer normal to the seat occurs at dlim =
Flow area as a function of valve lift is plotted in Figure 4.7.
dos −dis
sin 2φ
= dos − dis .
85
CHAPTER 4. MODELING AND SIMULATION
−4
6
x 10
A, x ≤x
t
L lim
x
At, xL ≤ xL lim
A , x >x
1
2
t
L
L lim
L lim
t
3
Valve curtain area A [m ]
5
4
L
At, xL > xL lim
3
2
1
0
0
3
4
Valve lift, x [mm]
5
6
7
8
L
Figure 4.7: Valve flow area as a function of lift for valve seat diameters dos , dis
4.5.2
Gas Force
Combustion gas pressure disturbances may be related to valve force through the
following:
Fg = Cgf P (x, Po, t)Av
(4.32)
where the valve face area is Av and Cgf is a gas force coefficient that approximates the
affect of flow losses behind the valve during opening. In [Schernus et al., 2002], Cgf
is found to range from approximately 0.85 to 0.7 over the valve stroke. For simplicity,
a constant value of Cgf = 0.8 is used throughout this study to approximate this
stagnation flow effect. As shown in Section 4.5.1, cylinder pressure, P , depends on
time, initial pressure, Po , and effective flow area, At . Upon investigation, it is found
that a quadratic functional form of gas force, Fgs , approximates this phenomenological
86
CHAPTER 4. MODELING AND SIMULATION
6
Simulated
Simplified
Testbench
Engine †
Pressure [bar]
5
4
3
2
1
0
Normalized Pressure [−]
−1
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
1
0.8
0.6
0.4
0.2
0
−0.2
Figure 4.8: Comparison of the quadratic gas model to simulated, experimental testbench and engine† measurements
gas flow model and may be expressed as:
Fgs = Cgf Po Av f1 (x) = γf1 (x)
(4.33)
f1 = c1 + c2 xL + c3 x2L
(4.34)
with
where the constants ci are obtained using least squares fit to simulated or measured
compressible flow pressure transients. The initial gas force magnitude is represented
by γ = Cgf Po Av and is considered constant throughout the opening cycle.
Com-
parisons between this simplified relation, the numerical gas flow model, testbench
experiments (see Section 6.3) and single cylinder engine tests are shown in Figure
4.8 over pressures ranging from 1 to 5 bar. Agreement between the simplified model
87
CHAPTER 4. MODELING AND SIMULATION
0.5
1 bar
2 bar
3 bar
4 bar
5 bar
0.4
Absolute Error [Bar]
0.3
0.2
0.1
0
−0.1
−0.2
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
Figure 4.9: Absolute error of the quadratic gas model with respect to experimental
testbench results over various pressure ranges
and experimental testbench measurements is good with a plot of the error between
the simplified model and measured testbench results over various pressures provided
in Figure 4.9.
Maximum error tends to occur at higher pressure levels (in excess
of 4 bar) at the 1 mm position with magnitudes of approximately 0.5 bar. Through
(4.32), errors of 0.5 bar translate to valve force errors of approximately 20 N. The relatively good agreement attainable with a relatively computationally simple gas force
model makes the quadratic gas model an ideal candidate to be used in disturbance
estimation.
Taking the time derivative of (4.33) results in an expression for the rate of change of
†
These data are provided by DaimlerChrysler AG from a single cylinder research engine equipped
with linear motion type solenoid actuators
88
CHAPTER 4. MODELING AND SIMULATION
Continuous
powergui
Gas Force
Opener Voltage
Compressed Air Disturbance
Closer Voltage
Sorensen DCS80−37
Position
Voltage
Velocity
Voltage
Opener Voltage
Opener Flux
State
Closer Flux
Results Output
Opener Curent
Closer Voltage
Switch High
Opener
Power Electronics
Switch Low
Opener
Closer Current
Testbench
Fmop, Fmcl, Acc
Bridge
Signals
Switch High
Closer
Swithch Low
Closer
Controler Outputs
dSPACE 1103
State
Figure 4.10: Simulink Model - top level
gas force:
Ḟgs = Cgf Po Av (c2 (ẋ) + 2c3 (x + S/2)ẋ) = γf2 (x, ẋ)
(4.35)
At higher engine speeds, gas forces caused by the time varying piston volume, V can
be approximated. In that case, f1 and f2 from (4.34) and (4.35) can be augmented
to reflect the crank position, θc , dependence:
f1c =
f1 Vo
V (θc )
(4.36)
f2c =
Vo f2
Vo f1 V̇
− 2
V (θc ) V (θc )
(4.37)
where cylinder volume, V and its time derivative are detailed in Section 4.5.1. Estimates for both Fgs and Ḟgs are required in the landing control law derived in Chapter 5.
CHAPTER 4. MODELING AND SIMULATION
4.6
89
Simulink Model
A flexible simulation model capable of incorporating the various physical domains
of the hinged actuator is developed using the MATLAB-Simulink environment. The
model is designed to simulate prospective control algorithms and to predict actuator
performance over a opening or closing cycle. The top-level of the model is shown in
Figure 4.10 which indicates the major sub-models which are briefly explained in the
following subsections. This model has been validated experimentally for the hinged
actuator in [Chladny and Koch, 2006b] and for a linear VVT actuator (including an
eddy current model, but without gas disturbances) in [Chladny et al., 2005].
4.6.1
Power Supply and Electronics
The actuator is excited through the use of a switched bridge amplifier circuit. In this
way, digital pulse width modulated (PWM) signals from the controller are amplified
as described in Chapter 6, Section 6.5.14. Through the use of one digital and one
PWM controller outputs, this ‘H-bridge’ circuit provides the ability to apply three
coil voltages: 42 V, 0 V and -42 V. The circuit is modeled using data from schematics
and component manufacturer data. The SimPower Toolbox is used to represent the
individual circuit components within the Simulink model. A representation of the
opener drive circuit is shown in Figure 4.11.
Presently, the 3 kW power supply used in experiments is modeled simply as a constant
source with a value of 42 V. This assumption is shown to be valid for the relatively
short pules such as those in Section 7.2.2.
4.6.2
Testbench Model
The testbench sub-model (shown in Figure 4.12) consists of the mechanical, opener
and closer coil dynamics models. Voltages from the power electronics are passed to
90
CHAPTER 4. MODELING AND SIMULATION
1
g
C
IGBT 1
IGBT 1
E
Diode 1
+
−
3
s
Current
v
Voltage
Measurement
+
1
Coil Voltage
−
+
Actuator
Bridge Voltage
R1
−
s
Controlled Voltage
Source
4
C2
C1
g
C
2
IGBT 2
IGBT 2
E
Diode
2
Ground
Figure 4.11: Simulink Model - Power Electronics (Opener)
the respective opener and closer coil dynamics which are described in the following
section.
4.6.2.1 Coil Dynamics Model
To accurately represent the magnetic response of the hinged actuator, flux and force
data from the FEA studies are implemented in lookup tables as shown in Figure
4.13. In this way, the difference in applied voltage and Ohmic losses is integrated and
related to flux linkage and position to coil current. Coil current and position are then
used to predict magnetic force through another data set. The coil dynamics models
are coupled to the power electronics sub-systems through the applied voltage and coil
91
CHAPTER 4. MODELING AND SIMULATION
x2o
−C−
3
Position
Demux
x3o
Mech
Initial Conditions
4
velocity
6
Closer Flux
9
Fmags, Acc
1
Gas Force
Mech Subsystem
3
Coil Voltage1
2
Closer Switched
Voltage
8
Closer Current
<x3>
Closer Coil Dynamics
1
5
Opener Switched
Voltage
Opener Flux
2
Coil Voltage
7
Opener Current
Opener Coil Dynamics
Figure 4.12: Simulink Model - Testbench
current. The system is also coupled to the mechanical system through armature position and magnetic force. Magnetic flux and current are output for control feedback
which are experimentally measured through analog integration of measurement coil
voltage and a hall effect sensor, respectively.
4.6.3
Mechanical Model
The actuator and valve mechanics are modeled using Newton’s Law as shown in
Figure 4.14. The model is coupled to the coil dynamics and gas disturbance models
through their respective forces and armature position. Position and velocity data
are output for feedback (in the case of full state feedback simulations) and analysis
purposes. If required, this model could be extended to include a valve / armature
92
CHAPTER 4. MODELING AND SIMULATION
1
Position
−K−
Gain4
<x1>
−C−
pos
2
Coil Voltage
N*dphi/dt
−K−
1
1
s
Integrator
1/N
1/N
N
i(x,lambda)
N
i
1/N
N
Load
Fmag(x,i)
i
R Opener
3
Current
2
Opener Flux
Figure 4.13: Simulink Model - Coil Dynamics (Opener)
impact model or a flexible element linking the armature and valve masses to represent
a hydraulic lash adjuster.
4.6.4
Gas Force Disturbance Model
To investigate the effects of time varying gas force disturbances on valve trajectory, the
numerical gas model presented in Section 4.5 is incorporated into the hinged actuator
system Simulink model. The model is coupled to the mechanical system through
valve position and gas force. Pressure changes caused by engine piston motion can
be selected as well (via engine speed and EVO crank angle input). Time dependent
experimental pressure traces can also be used from within this model, however such
simulation results are viewed as less reliable due to the lack of valve throat area
coupling.
4.6.5
dSPACE Controller Model
Given the myriad combinations of feedback, estimation and control algorithms available for evaluation, the control sub-system is implemented in modules. It contains
a series of logical operations, or state machine, that co-ordinate the various control
states and their respective algorithms at a simulated sample and execution rate of
93
CHAPTER 4. MODELING AND SIMULATION
3
Fmag
x3dot
k
Effective Spring Rate
b
2
x3=x2dot
vel
Effective Damping Rate
Fmag Op
4
3
1/m
FmagCl
5
~=
double
x3dot
1/m
Sign
Fgas
1
s
x3=x2dot
xo
Velocity
0
Fv
1
x2o
IC
preload
2
x3o
x3o
1
xos
x2
1
pos
Position
x2o
Switch
|u|
Abs
Figure 4.14: Simulink Model - Mechanical Dynamics
50 kHz. The controller model also has modules to co-ordinate what kinds of feedback
and landing controllers are to be used. Conversion of the controller voltage output
into a set of opener and closer PWM and logic signals for the power electronics is also
included to emulate the physical dSPACE 1103 control hardware output. Information such as armature position, coil currents and magnetic flux are input for control
calculation purposes while control output and feedback estimation data are output
for analysis and debugging.
4.7
Computer Software & Hardware
All modeling software suites were run on a 300W ATX desktop computer consisting
of an Intel Pentium 4 (2.80GHz) CPU, 1.0 Gbyte of RAM, 7200 rpm ATA hard disk,
using a Microsoft Windows XP operating system and MATLAB versions 6.5 to 7.1.
CHAPTER 4. MODELING AND SIMULATION
4.8
94
Summary
The preceding sections highlight the key modeling methods and techniques undertaken with specific reference to the hinged actuator system investigated. Static and
transient finite element analyses are conducted to establish and parameterize analytic
magnetic models. The analytic models are to be used in the derivation of control algorithms. Power electronics circuits, FEA data and mechanical and electric systems
are incorporated into a Simulink model intended to represent the complete hinged
actuator system for control system performance evaluation. Also provided is an idealized compressible flow model used to approximate the affects of gas pressure on
valve motion during exhaust valve opening. Where applicable, contrast of the various
models with equivalent experimental data are provided in Chapter 6.
Chapter 5
Control Design
5.1
Introduction
B
ased on derivation and validation of an actuator model and identification of
design constraints, a control strategy is formulated. In the following sections,
an overview of the proposed control methodology is presented. Auxiliary algorithms
for position feedback and exhaust gas disturbance force are also provided for use with
the proposed feedforward and feedback controllers. Also presented are the derivations
of classical linear and proportional integral (PI) landing controllers to contrast with
the proposed flatness-based algorithm. Simulated performance results comparing the
various feedback controllers for both the linear and and nonlinear induction models
are included to justify the use of a model that accounts for magnetic saturation. The
techniques used to derive the control reference trajectories are also discussed.
5.2
Controller Topology
As discussed in Chapter 1, the actuator and controller design must satisfy several
performance and physical constraints. These include maximum valve seating velocities of 0.1m/s, transition times (time from open to close or close to open) of no more
than 4.5ms, the use of practical feedback sensor technology and a maximum avail-
95
CHAPTER 5. CONTROL DESIGN
96
Figure 5.1: Control flowchart from closed to open and open to close
able voltage of 42V. Given these control constraints and actuator characteristics, the
design of the actuator controller is divided into the following primary areas:
• Initialization and holding routines
• Closed-loop landing controller
• Landing control reference trajectory design
• State and disturbance estimation
• Feedforward controller
The control system is a combination of feedforward and feedback landing controllers
with online disturbance estimation through measured current and flux-based state
CHAPTER 5. CONTROL DESIGN
97
Figure 5.2: State machine flowchart [Chung, 2005]
reconstruction. A block diagram of the above structure with respect to the actuator
plant and engine disturbances is presented in Figure 5.1. The individual control
stages are executed according to a set of logical conditions or state machine as shown
in Figure 5.2. Starting from an inactive state, an open-loop initialization routine is
called which moves the armature and valve into a closed position. Then, as shown
in Figure 5.3, a holding controller is engaged until a command is given to move the
valve to an open state. The closer holding controller is released and position, velocity
and disturbance estimation begins along with an energy-based feedforward controller.
98
CHAPTER 5. CONTROL DESIGN
3
2
−2
1
1
2
3
4
5
6
7
20
15
10
Opener holding 10
5
0
0
0
9
20
Closer
Opener 15
Landing
& Estimation
Feed−forward
&Estimation
Closer
holding
8
5
1
Velocity [m/s]
lc
4
0
−4
0
Closer Current [A]
x
2
Position
Velocity
2
3
4
5
Time [ms]
6
7
8
0
9
Opener Current [A]
Position [mm]
4
Figure 5.3: Control stages from closed to open position with respect to experimentally
measured position, velocity and coil current at 2 bar EVO
These routines are used to set up favorable initial conditions for the landing controller
which is engaged at position x = xlc . After this point, the feedforward controller is
disengaged and the closed-loop landing controller takes over to seat the armature and
valve with a low impact speed. When a pre-determined open position and velocity
are reached the valve is determined to be in the open state and the estimation and
closed-loop controller are disengaged and an opener holding routine is enabled. To
close, a similar procedure is used, although instead of an energy-based feedfoward
controller, a simpler position based routine is used instead (see [Chung, 2005]. This
is because minimal gas force disturbances are expected during the closing cycle. A
plot of the different control modes with respect to an experimental opening cycle at
an EVO pressure of 2 bar is provided in Figure 5.3. Figure 5.4 provides a flowchart
overview of the closed-to-open and open-to-closed control cycles and the following
sections describe in detail the individual routines.
CHAPTER 5. CONTROL DESIGN
99
Figure 5.4: Control flowchart from closed to open and open to close
5.3
Initialization and Holding Control
When the valve is not in transition it is either inactive, being initialized into the closed
position or being held open or closed. The initialization and holding processes have
relatively little impact on overall control performance, but are essential for practical
operation and are thus included for completeness.
5.3.1
Initialization
Prior to engine startup, or when the actuator is otherwise inactive, the armature
and valve equilibrium is at approximately a mid-stroke position. Due to the limited
force authority of the magnets and relatively high spring and torsion bar stiffness, a
CHAPTER 5. CONTROL DESIGN
100
time-based open loop controller alternatively pulses the opener and closer magnet to
‘swing’ the armature into the closed position utilizing the natural frequency of the
mass-spring system.
5.3.2
Holding Control
After initialization or at the end of a valve opening or closing cycle, the armature and
valve are held in an open or closed state. During this time, a pre-specified holding
current is regulated through a coil current measurement and holding controller until
a release command is issued by a software RPM generator (in the case of testbench
experiments) or ECU (in the case of a real engine). Current is typically regulated at
approximately 4.0 A through a simple ‘on-off’ controller that switches the holding coil
current on (+42 V mode) or off (0 V mode) if the actual current level is respectively
0.1 A less than or greater than the set-point. Switching is relatively inactive because
of the high inductance at these positions (minimal air gap). When a release command
is issued, the coil current is driven to 0 A with the -42 V mode to drive the holding
force to zero and minimize the time until motion is incipient.
5.4
Closed-loop Landing Control
To justify the use of a nonlinear controller and an induction model that accounts for
magnetic saturation, four closed loop landing controllers are derived and simulated
for comparison purposes. They are: two classical linear controllers based on both
the linear and nonlinear induction models, a PI controller and a flatness controller
based on the nonlinear induction model. The benefits of the flatness technique and
nonlinear induction model are apparent in simulation so only the flatness controller
is implemented with hardware.
101
CHAPTER 5. CONTROL DESIGN
5.4.1
Plant Models
With reference to the plant models derived in Chapter 4 Section 4.4, below are the
linear and nonlinear induction state space models. In both cases the state vector is
defined as x = [x1 x2 x3 ]T = [i x̄ ẋ]T where x̄ = κ − x and input u is defined as
u = v − iR for notational convenience.
5.4.1.1 Linear Induction Model
As introduced earlier, the resulting linear induction state space model is:



 ẋ1  

 
 ẋ  = 
 2  

 
ẋ3

x1 x3
x2
−x3
x3 b+(x2 −κ)k−Fg −Fv
m
+
x21 βℓm
mℓv x22

x2
β

 

 

+  0 u
 

 

0
(5.1)
5.4.1.2 Nonlinear Induction Model
The resulting state space model for the nonlinear induction model is:



x1 g ′ (x2 )x3
g(x2 )
 ẋ1  

 
 ẋ  = 
−x3
 2  

 
(−x3 b+(κ−x2 )k−Fv +Fg − ℓℓmv Fm (x1 ,x2 ))
ẋ3
−
m
5.4.2


 
 
+
 
 
eg(x2 )x1 u
ψg(x2 )
0
0






(5.2)
Linear Full State Feedback - Linear Induction System
In order to obtain an LTI compliant model, the nonlinear state equations in Section
5.4.1.1 are linearized about an equilibrium point xe = [x1e x2e x3e ].
It is apparent that the equilibrium velocity, x3e , must be zero from solving ẋ2 = 0.
The equilibrium current may then be solved for an arbitrary position, x2e = κ − xe .
This operating point is chosen to be xe = 3.25 mm, or 0.75 mm from the stroke
102
CHAPTER 5. CONTROL DESIGN
bounds, as that is the midpoint of the range over which control is typically executed.
x1e =
p
βℓv ℓm (Fg + Fv + k(κ − x2e ))x2e
βℓm
(5.3)
Solving for the equilibrium input voltage, v, is accomplished by setting ẋ1 = 0:
v=R
p
βℓv ℓm (Fg + Fv + k(κ − x2e ))x2e
βℓm
(5.4)
To generate the linearized A matrix, the Jacobian of the nonlinear system is computed
with respect to the state vector x at the derived equilibrium point. The result of which
is provided below:

0


A=

 √
2
0
0
−ℓm βℓv ((x2e−κ)k−F v−F g)
ℓv x2e m
√
0
k/m +
βℓm ℓv (κ−x2e )k+Fv +Fg )
βℓm
−1
2((x2e −κ)k−Fv −Fg )
x2e m
b
m






(5.5)
The corresponding linearized B matrix is:

x2e
β






B=
0




0
(5.6)
The tracking problem is solved locally through static state feedback. To simplify
the following analysis, the linearized system described by matrices (5.5) and (5.6) is
103
CHAPTER 5. CONTROL DESIGN
redefined as follows:

0 a13
 0

ẋ = 
0
1
 0

a31 a32 a33
y = (0 1 0)x




 b1 



x+  0 u






0
(5.7)
(5.8)
where
p
p
−βℓm ℓv ((x2e − κ)k − Fv − Fg )
2 −ℓm βℓv ((x2e − κ)k − F v − F g)
a13 = −
a31 =
βℓm
ℓv x2e m
2((x2e − κ)k − Fv − Fg )
b
a32 = k/m +
a33 = −
x2e m
m
x2e
b1 =
(5.9)
β
A change of coordinates and a state feedback control law is sought such that local exponentially stable error dynamics, and hence asymptotic tracking is achieved.
Successive derivatives of 5.8 are taken to determine the relative degree, ρ.
y = x2
ẏ = ẋ2
= x3
ÿ = ẋ3
= a31 x1 + a32 x2 + a33 x3
y (3) = a31 ẋ1 + a32 ẋ2 + a33 ẋ3
= a31 a13 x3 + a31 b1 u + a32 ẋ2 + a33 ẋ3
104
CHAPTER 5. CONTROL DESIGN
By inspection, the input appears in the third derivative. Therefore the relative degree
is well defined and equal to the system order. Furthermore, analysis of the transfer
function representation of (5.7),
ŷ(s)
b1 a31
=
û(s)
s(s2 − sa33 − a32 − a31 a13 )
(5.10)
confirms that unstable zeros do not exist and hence the tracking dynamics are locally
bounded input, bounded state. Therefore, the tracking problem is locally solvable by
static state feedback.
The coordinate transformation is defined by z = T x = [y ẏ ÿ]T , where

1
0
 0

T =
0
1
 0

a31 a32 a33






(5.11)
and the feedback law
u=
1
(ū − (a31 a13 + a32 )x3
b1 a31
(5.12)
−a33 (a31 x1 + a32 x2 + a33 x3 ))
where ū is an auxiliary input. The linearized system can thus be transformed into
Brunovsky controller form:




 0 
 0 1 0 

 


 
ż = 
 0 0 1  z +  0  ū

 

1
0 0 0
(5.13)
105
CHAPTER 5. CONTROL DESIGN
By choosing
ū = − k1 (z1 − yr (t)) − k2 (z2 − ẏr (t))
(5.14)
− k3 (z3 − ÿr (t)) + yr(3) (t)
yields the exponentially stable tracking error equation
e(3) + k1 e + k2 ė + k3 ë = 0
(5.15)
(i)
where e(i) = (y (i) − yr ) is defined for 0 ≤ i ≤ 3 and ki > 0. These error dynamics
guarantee local exponential tracking to a desired reference trajectory. Since the ref(i)
erence trajectories, yr , are assumed to be bounded, then so are the states z1 , z2 , and
z3 , whereby under transformation, the states, x1 , x2 , and x3 , are also bounded. Pole
placement is selected with respect to convergence rate and plant saturation (42 V and
stroke limitations). The composite control law in x-coordinates is derived through
combining (5.13) and (5.15) and applying the derived transformation z = T x as
shown below:
u=
1
(−k1 (z1 − yr (t)) − k2 (z2 − ẏr (t))
b1 a31
− k3 (z3 − ÿr (t)) + yr(3) (t) − (a31 a13 + a32 )x3
−a33 (a31 x1 + a32 x2 + a33 x3 ) z=T x
1
=
(−k1 (x2 − yr (t)) − k2 (x3 − ẏr (t))
b1 a31
− k3 (a31 x1 + a32 x2 + a33 x3 − ÿr (t)) + yr(3) (t)
− (a31 a13 + a32 )x3 −a33 (a31 x1 + a32 x2 + a33 x3 )
106
CHAPTER 5. CONTROL DESIGN
Figure 5.5: Simulated linear time invariant landing control block diagram
Finally, accounting for the substitution u = v −iR, the exponentially tracking control
law is:
v =x1 R +
1
(−k1 (x2 − yr (t)) − k2 (x3 − ẏr (t))
b1 a31
− k3 (a31 x1 + a32 x2 + a33 x3 − ÿr (t)) + yr(3) (t)
− (a31 a13 + a32 )x3 −a33 (a31 x1 + a32 x2 + a33 x3 )
Note that this controller requires either full state feedback or estimates thereof. For
the purposes of controller comparison simulations, it is assumed the full state is
available and that gas forces are negligible. A block diagram of the LTI end-controller
structure is shown in Figure 5.5
5.4.3
Linear Full State Feedback - Nonlinear Induction
An identical process to that described in Section 5.4.2 is used to derive an LTI model
based upon the nonlinear induction model. The equilibrium current is:
x1e = −
x2 e
(W−1 (z) + 1)
β + αx2e
(5.16)
107
CHAPTER 5. CONTROL DESIGN
where W−1 (z) is a real-valued branch of the Lambert’s function w(z)ew(z) (see Appendix A, Section A.6 or [Corless et al., 1996] for details)
W−1 (z) =
1
β 2 ℓv (Fg + Fv + k(κ − x2e )) + (αx2e )2 ℓv (Fg + Fv + k(κ − x2e ))
ψβeℓm
(5.17)
+ β(−ℓm ψ + 2αℓv x2e (Fg + Fv + k(κ − x2e ))))
Thus, the state space representation of the system about the equilibrium point is

0 a13
 0

ẋ = 
0 −1
 0

a31 a32 a33
y = (0 1 0)x




 b1 



x +  0 u






0
(5.18)
(5.19)
where
a13
(−
x1e β
=
x2e (β + αx2e
−x1e α+ xβ
2e
e
a32 =
ℓv mx32e (β + αx2e )3
b
a33 =
m
(x2e + α) xβx1e
b1 =
e 2e +α
ψβ
a31
x1e (β+αx2e )
)
x2e
ℓm βψx1e e
=
mℓv x22e
β
3 x1e α+ x2e
x2e e
ζ1 + ζ2
(5.20)
with
ζ1 = kℓv β 3 + 3αkℓv x2e β 2 + α 3αkℓv x22e − 2ℓm ψ β + α3 kℓv x32e
ζ2 =βℓm ψ −x21e β 3 − 2αx21e x2e β 2 − αx1e (αx1e − 2)x22e β + 2α(αx1e + 1)x32e
(5.21)
108
CHAPTER 5. CONTROL DESIGN
The system control law derivation and form is identical to that described in the case
of the linear induction controller with exception of the coefficients.
5.4.4
Proportional-Integral Control
A classical PI landing controller is presented as a benchmark comparison to the linear
and flatness-based controllers. The control technique is commonplace in industrial
applications due to the relative ease in design, tuning, robustness and computational
demands. Here, current is output in proportion to position error and integrated
position error as follows:
iP I = KP (y − yr ) + KI
Z
t
0
(y − yr ) (τ ) dτ
(5.22)
where iP I is the current control output, KP is the proportional gain and KI is the integral gain and yr is the desired reference trajectory. The gains are first approximated
using a Ziegler-Nichols method [Franklin et al., 1998] and then manually refined. PI
control enhances system response through increasing the response to output-command
error while the integration of such errors ensures steady state errors are eliminated.
Due to excessive sensor noise, a differential term is not included. To enhance the transient response, a feedorward current, iP If f , is introduced so that the desired current
output, id , is id = iP I + iP If f . Regulation about the desired current is done in a similar fashion as holding current regulation. The position based open loop feedforward
current is derived as shown in Section 5.5. This controller requires current, i, and
position, y, for feedback which is assumed to be available for simulation purposes.
Experimental performance of a similar controller is contrasted with a flatness-based
method for a linear type of actuator in [Chung, 2005]. A block diagram for the PI
landing controller with a feedforward current input is shown in Figure 5.6.
109
CHAPTER 5. CONTROL DESIGN
Figure 5.6: Simulated proportion-integral current landing control block diagram
5.4.5
Flatness-based Voltage Control - Nonlinear Induction
The landing controller proposed in this work uses a flatness-based landing controller as
described in [Chung et al., 2007, Chladny and Koch, 2006b] to achieve exponentially
convergent armature position tracking to a predetermined trajectory. Flatness based
control was chosen for landing as it allows the design of a trajectory which is subject
to both path and end constraints. Static state-feedback voltage control is obtained
by defining position, x, as a flat output, y = x. States velocity, ẋ and current, i are
related to the flat output through:
ẋ = ẏ
i=−
(5.23)
1
[W−1 (−η(y, ẏ, ÿ)/e) + 1]
g(y)
(5.24)
where W−1 is a real-valued branch of the Lambert’s function and
η(y, ẏ, ÿ) = 1 −
g 2 (y)(mÿ − A(y, ẏ, Fg ))
ψg ′ (y)
A(y, ẏ, Fg ) = −(ky + bẏ − Fv + Fg )
(5.25)
(5.26)
110
CHAPTER 5. CONTROL DESIGN
Voltage is related to the third time derivative of y as follows:
ℓm
1
ÿ = ẍ = −
ẏb + yk − Fv + Fg −
Fm (y, i)
m
ℓv
ℓm
1
(3)
y =−
ÿb + ẏk + Ḟg −
Ḟm (y, i) ,
m
ℓv
(5.27)
(5.28)
where,
1
g ′(y)
Ḟm (y, i) =2ẏFm (y, i)
−
(κ − y)
g(y)
′
ig (y)
(v − iR).
+
g(y)
(5.29)
Solving (5.28) and (5.29) for input voltage, v yields:
ℓv g(y)
g ′(y)
v=
ÿb
+
ẏ(k
+
2F
(y,
i)(
m
ℓm ig ′ (y)
g(y)
1
−
) + Ḟg + y (3) m + iR.
(κ − y)
(5.30)
The singularity at i = 0 is avoided through ensuring a non-zero bias current prior to
landing controller engagement. Non-zero current is also a requirement for flux-based
position reconstruction as described in Section 5.6.1. From (5.30) it is apparent
that the state (i, x, ẋ) and input, v, may be expressed as a finite number of time
derivatives of the output, y. Therefore, the system satisfies the flatness definition and
therefore an open-loop solution of input, v, may be expressed as a function of reference
trajectories as outlined in work such as [Fliess et al., 1999, Fliess et al., 1995]. To
compensate for deviations from desired and actual trajectories, additional feedback
can be implemented through defining tracking error as ỹ = y − yd , and linear error
111
CHAPTER 5. CONTROL DESIGN
dynamics:
k1 ỹ + k2 ỹ˙ + k3 ỹ¨ + ỹ (3) = 0
(5.31)
Coefficients k1 , k2 and k3 are chosen such that the characteristic equation P = s3 +
k3 s2 + k2 s + k1 = 0 is Hurwitz. Pole placement is selected with respect to convergence
rate and plant saturation (42 V and stroke limitations). Solving (5.31) for y (3) results
in:
(3)
y (3) = −k1 ỹ − k2 ỹ˙ − k3 ỹ¨ + yd
(5.32)
Substitution of (5.32) into (5.30) completes the expression for closed loop voltage control. Exponential tracking of y to yd will be achieved provided that the gains ki are
positive and chosen appropriately. As in the LTI case, the flatness controller requires
either full state feedback or estimates thereof. For the purposes of controller comparison simulations, it is assumed the full state is available. In practice however, the
state i is measured and y, ẏ and gas force Fg are estimated through an integrated flux
measurement and nonlinear observer (see Sections 5.6.1 and 5.8 respectively). Reference trajectories are designed subject to physical, practical and desired end condition
constraints as discussed in Section 5.5 as well as in [Chladny and Koch, 2006b] and
[Chung, 2005, Koch et al., 2004] for a linear motion actuator. A block diagram for
the simulated flatness-based landing controller is shown in Figure 5.7.
5.4.6
Preliminary Control Law Comparison
With four potential landing control laws derived, preliminary simulations are conducted to investigate the most promising candidate to be implemented in experiments. Using a simplified version of the Simulink LPM-FEA model described in
CHAPTER 5. CONTROL DESIGN
112
Figure 5.7: Simulated flatness-based landing control block diagram
Section 4.6, closer landing performance is investigated using specific initial current,
and velocity conditions. The simplified model does not involve gas forces, disturbance
feedforward controller, power electronics or state estimation (full state feedback is assumed). The landing controller is always engaged at a specific position, xlc , usually
1.5 mm away from the landing magnet pole face. Voltage sources of 500 V and 42 V
are simulated to predict landing performance with and without significant input saturation conditions. The LTI and flatness controllers have the same pole locations
of [−10000, −10000, −1500] for the purpose of comparison (and was sufficient for all
controllers to land the valve). Note that because the Simulink model coil and force
dynamics are based in part on the FEA simulations, model-plant mismatch is to be
expected for all controllers tested. Simulated landing position and velocity tracking is
provided in Figure 5.8 for initial conditions corresponding with the derived reference
trajectories and a 500 V source. By comparing the two LTI controllers, it is apparent
that even a linearized model that accounts for magnetic saturation provides a significant improvement with respect to impact velocity. The flatness controller has the
best performance with least tracking error and an impact velocity of 0.05 m/s. Based
on these results, it was apparent that the LTI linear induction controller would likely
prove unsuitable.
113
CHAPTER 5. CONTROL DESIGN
Position [mm]
−2.5
Reference
LTI − Linear Induction
LTI − Nonlinear Induction
PI
Flatness − Nonlinear Induction
−3
−3.5
−4
0
0.5
1
1.5
2
2.5
3
0.5
1
1.5
Time [ms]
2
2.5
3
Velocity [m/s]
0
−0.5
−1
−1.5
−2
0
Figure 5.8: Simulated landing control performance comparison with ideal initial conditions, 500 V
Next, deviations from the ideal initial conditions are introduced to provide insight to
performance under more realistic conditions. Figure 5.9 provides the landing results
under a initial velocity that is both 0.5 m/s greater and less than the initial reference
velocity. Beyond this range, all controllers except the flatness controller failed to land
completely. Due to the relatively high voltage of 500 V, initial current variations of up
to 10 A, caused insignificant performance variations and are not shown. In all large
initial condition variations, the flatness controller performed best respect to tracking
error and settling time. The LTI nonlinear induction controller had the best impact
velocity at 0.1 m/s (0.05 m/s less than the flatness controller), but was among the
longest with respect to settling time. The PI controller had the worst impact speed
but better settling time than the LTI controller.
To predict performance with a more realistic source, voltage was reduced to 42 V as
shown in Figure 5.10. Once again, the flatness-based controller performed best with
114
CHAPTER 5. CONTROL DESIGN
Position [mm]
−2.5
Reference
LTI − Nonlinear Induction
PI
Flatness − Nonlinear Induction
−3
−3.5
Velocity [m/s]
−4
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
0.5
1
1.5
2
2.5
Time [ms]
3
3.5
4
4.5
5
1
0
−1
−2
0
Figure 5.9: Simulated landing control performance comparison with ±0.5m/s initial
velocity deviation, 500 V
respect to impact speed and settling time. The PI controller was re-tuned to accommodate the reduced voltage with an impact velocity of 0.33 m/s and considerably
better tracking error than the LTI controller (which failed to land completely). Also
provided in Figure 5.10 is current response and input energy. In practice, the coil
current is limited to 40 A (peak) to prevent overloading. Reduced energy consumption is desired to minimize the associated parasitic engine load. In this respect, the
PI and flatness controllers preformed nearly equally well.
Landing robustness to parameter variations is also of concern given the wide range of
temperatures and long duration over which a vehicle engine must operate. In particular, the mechanical spring stiffness and damping are investigated as they are arguably
more susceptible to environmental and manufacturing variations. In Figure 5.12, PI
and flatness controller response to a ± 10% change in system spring stiffness, ksys ,
are shown. In the case where the system stiffness is 10% less than the value used in
115
CHAPTER 5. CONTROL DESIGN
Position [mm]
−2.5
Reference
LTI − Nonlinear Induction
PI
Flatness − Nonlinear Induction
−3
−3.5
−4
0
0.5
1
1.5
2
2.5
3
0.5
1
1.5
2
2.5
3
0.5
1
1.5
2
2.5
3
0.5
1
1.5
Time [ms]
2
2.5
3
Velocity [m/s]
1
0
−1
−2
0
Current [A]
40
20
Input Energy [J]
0
0
0.4
0.2
0
0
Figure 5.10: Simulated landing control performance comparison with ideal initial
conditions, 42 V
the control law (and reference and PI feedforward input derivations), landing impact
speeds are higher. The flatness and PI controllers had final speeds of 0.17 m/s and
0.35 m/s respectively. When the stiffness was increased, the PI controller failed to
land altogether and the flatness controller exhibited a steady state bias of 0.05 mm
as a result of the plant/model mismatch. Such biases are not uncommon with feedback linearization techniques where the tracking error causes a controller input that
exactly matches the model/plant discrepancy. In practice, such problems may be
116
CHAPTER 5. CONTROL DESIGN
Position [mm]
−2.5
Reference
PI
Flatness − Nonlinear Induction
−3
−3.5
ksys = k × 1.1
−4
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
2
2.5
Time [ms]
3
3.5
4
4.5
5
Velocity [m/s]
0
−0.5
ksys = k × 1.1
−1
−1.5
−2
0
0.5
1
1.5
Figure 5.11: Simulated landing control performance comparison with ideal initial
conditions and system spring variation of ± 10% at 42 V
overcome by introducing an additional integrator loop to eliminate any steady state
errors. Fortunately, spring stiffness is more likely to vary with component wear rather
than shorter term environmental changes, unlike viscous damping. To address performance alterations caused by viscous damping changes, system damping, bsys , was
varied by ± 500%. Due to the relatively weak force damping has, particularly at the
lower speeds encountered during the landing stages of control, only a slight change in
impact speed is observed for either controller. However, noticeable changes in settling
time are observed, particularly for the flatness controller. Despite that, the changes
in viscosity are likely to occur relatively slowly with respect to a given valve cycle.
In fact, both stiffness and damping forces are linearly dependent on the system state
they are relatively easy to recursively identify through on-line [Slotine and Li, 1991]
or off-line methods (see Appendix B).
The result of these simulations indicate the two most promising controllers are the
117
CHAPTER 5. CONTROL DESIGN
Position [mm]
−2.5
Reference
PI
Flatness − Nonlinear Induction
−3
bsys = b × 5
−3.5
−4
0
0.5
1
1.5
2
2.5
3
3.5
3
3.5
0
Velocity [m/s]
−0.5
bsys = b × 5
−1
−1.5
−2
0
0.5
1
1.5
2
2.5
Time [ms]
Figure 5.12: Simulated landing control performance comparison with ideal initial
conditions and system damping variation of ± 500% at 42 V
flatness-based and PI algorithms. Similar conclusions are documented in [Chung, 2005]
based on simulated and experimental results with a linear motion actuator. As a result, only the flatness-based landing controller is to be incorporated into the overall
control strategy and testbench experiments discussed in Section 6.3.
5.5
Reference Trajectory Design
To implement the flatness based or LTI landing control algorithms, a set of reference trajectories is required. These trajectories can also be used to generate a
corresponding feedforward current input for the PI controller. Optimal reference
trajectories are sought to move the armature from an initial state (provided by the
feedforward controller) to a open / closed state while subject to physical, practical
and desired end condition constraints. A nonlinear programming procedure described
in [Chung, 2005] and [Koch et al., 2004] for the closer of a linear motion actuator is
118
CHAPTER 5. CONTROL DESIGN
applied to both the opener and closer of the hinged actuator system used in this
work. The procedure consists of the formulation of a nonlinear constrained problem
to optimize a parameterized trajectory set that connects the prescribed initial and
final states. By using the flat mapping between the system output and input, the
method is able to account for the dynamics represented in the nonlinear induction
model such as magnetic saturation, armature motion and voltage input.
5.5.1
Nonlinear Constrained Problem
Nonlinear optimization problems can generally be formulated into the form:
min F (X)
(5.33)
X
subject to: Aeq X = Beq
AX ≤ B
(Linear Equality Constraints)
(5.34)
(Linear Constraints)
Ceq (X) = 0 (Nonlinear Equality Constraints)
(5.35)
C(X) ≤ 0 (Nonlinear Constraints)
(5.36)
where [A, Aeq , B, Beq , C(X), Ceq (X), X] ǫ Rn . The problem solution process minimizes an objective function, X, subject to a nonlinear cost function, F (X), while
subject to sets of linear and nonlinear constraints. The solution of such nonlinear constrained problem can be difficult, requiring many iterations and functional computations [Milam, 2003]. In the method used, and originally proposed in [Koch et al., 2004],
(5.34) is modified by substituting the minimization of F (X) with a nonlinear feasibility problem. Specifically, a solution will be found such that a parameterized trajectory, yd , will satisfy the imposed constraints. The constraints imposed are listed in
Table 5.1 where t0 and tf are the times when the landing controller are respectively
CHAPTER 5. CONTROL DESIGN
119
Table 5.1: Reference trajectory constraints
Position (Air Gap)
Velocity
yd (t0 ) = 1.50 mm
ẏd (t0 ) = ẏ0 m/s
yd (tf ) = 0.00 mm
ẏd (tf ) ≤ 0.1 m/s
ẏd (t) ≥ 0, t0 < t < tf
Acceleration
Voltage & Current
ÿd (t0 ) = −1/m (ẏd (t0 )b + yd (t0 )k − Fv v(t) ≤ 42V, t0 ≤ t ≤ tf
−ℓm Fm (yd (t0 ), i(t0 ))/ℓv ) m/s2
i(t0 ) = i0 A
ÿd (tf ) = 0 m/s2
ÿd (t) > −1/m (ẏd (t0 )b + yd (t0 )k − Fv ),
t0 < t < tf
engaged and disengaged. To ensure the armature moves from the initial position,
yd (t0 ) to the final open / closed state, yd (tf ), the following equality constraints are
imposed:
yd (t0 ) = 1.50 mm (Air gap)
(5.37)
yd (tf ) = 0.00 mm (Air gap)
(5.38)
Initial and final velocity constraints are also imposed:
ẏdc (t0 ) = 2.34 m/s
(5.39)
ẏdo (t0 ) = 2.58 m/s
(5.40)
ẏd (tf ) ≤ 0.10 m/s
(5.41)
Unique initial velocities for the opener, ẏdo (t0 ), and closer, ẏdc (t0 ), are prescribed to
account for the slight differences between the two magnets and spring preload bias. A
final landing velocity constraint, (5.41), is imposed to minimize wear and acoustical
noise as outlined in the original control objectives.
An initial coil current is also imposed to minimize the input voltage required upon
120
CHAPTER 5. CONTROL DESIGN
engagement of the landing controller as done in [Hoffmann et al., 2003].
io (t0 ) = 5.5 A
(5.42)
ic (t0 ) = 7.0 A
(5.43)
Where again, the subscripts ‘c’ and ‘o’ differentiate between the closer and opener
magnets, respectively. The imposed initial current ensures that an adequate flux level
is maintained in anticipation of the landing control effort, particularly because the
inductance is highest at the smaller air gaps inherent with the landing control.
The initial position and velocity constraints may be used to solve for corresponding
acceleration constraints for a predetermined initial current, i(t0 ), through substitution
into the expression for mass-spring dynamics, where:
ÿdo (t0 ) = −1/m (ẏdo (t0 )b + yd (t0 )k − Fv − ℓm Fmo (yd (t0 ), i(t0 ))/ℓv )
m/s2
(5.44)
ÿdc (t0 ) = −1/m (ẏdc (t0 )b + yd (t0 )k − Fv + ℓm Fmc (yd (t0 ), i(t0 ))/ℓv )
m/s2
(5.45)
ÿd (tf ) = 0.0
m/s2
(5.46)
A zero acceleration at tf is explicitly specified so that a constant velocity of ẏd (tf ) ≤
0.10 is maintained until seating occurs. This is done to ensure the valve seats at an
acceptable velocity even when the armature and valve are coupled through a flexible
lash adjuster. Additionally, acceleration constraints may be imposed to account for
the fact that the magnets are only able to impart attractive forces:
ÿd (t) > −1/m (ẏd (t0 )b + yd (t0 )k − Fv )
m/s2 ,
t0 < t < tf
(5.47)
121
CHAPTER 5. CONTROL DESIGN
An input voltage constraint may also be imposed through the flat mapping between
input voltage, v, and state provided in Equation (5.30):
|v(t)| ≤ 42 V,
t0 ≤ t ≤ tf
(5.48)
This constraint is imposed in accordance with the future 42 V on-board vehicle voltage
standard [Chang et al., 2002]. In all cases, realistic constraint values (based on simulated and experimental observations) are specified to ensure a physically realizable
trajectory while still satisfying the input and final state conditions.
5.5.2
Parametrization of the Flat Output Trajectory
With a nonlinear programming and constraint framework established, the trajectory
set must be parameterized for a particular solution to be found. As discussed in
[Chung, 2005], B-spline basis functions are used to mathematically describe a trajectory, yd . Spline functions are chosen because they are amenable to computing
continuous derivatives and are more numerically stable than higher order polynomials [van Nieuwstadt and Murray, 1995]. Sufficiently smooth reference trajectories are
required to ensure a diffeomorphic mapping between the input, output and state variables. In this case, three derivatives are required to relate the flat output to the states
and input (see Equation (5.30)). Therefore, the parameterized trajectory should have
a continuous third time derivative.
Using the methods shown in [Löewis, 2002], a desired trajectory may be parameterized with a spline basis function by:
yd (t) = θT Bk (t),
t = t0 , . . . , tf
(5.49)
CHAPTER 5. CONTROL DESIGN
122
where θ and Bk represent vectors of spline coefficients and B-spline basis functions of
order, k. Knots on an interval [t0 , tf ] represent where the basis functions are joined
and are specificed through a strictly increasing sequence of real numbers. In this way,
a trajectory, yd , over the interval [t0 , tf ] and subject to the constraints discussed in
the previous section may be determined provided Bk (t) 6= 0 [Boor, 1978].
The MATLAB Spline and Optimization Toolboxes are used to optimize the spline coefficients, by solving the nonlinear constrained problem via the fmincon.m function.
In order to achieve convergence, initial guesses of the coefficients, θ, must be provided
that are sufficiently close to the optimal solution. At each iteration, the fmincon.m
function uses a sequential quadratic programming (SQP) solution method to solve
a quadratic programming (QP) subproblem whose solution is used as a search direction for a line search procedure. Details of the SQP method can be found in
[Gill et al., 1981]. To ensure sufficient smoothness, five, fourth order B-splines (k =
5) are evenly spaced at six simple knots on [t0 = 0, tf ].
Optimization of θ is done separately for the opener and closer magnets. In doing so,
respective constraints (5.37) through (5.40) and (5.45) through (5.46) are passed to
fmincon.m as linear equality constraints. Additionally, conditions (5.41) and (5.47)
are imposed on θ as linear and nonlinear inequality constraints, respectively.
Convergence is knot position dependent which are in turn determined through choosing a suitable landing time from t0 = 0 to tf o = 2.35ms, and tf c = 2.60ms for the
opener and closer magnets, respectively. Given these constraints and knot positions,
a trajectory that satisfies the input constraint of |v| <42 V is obtained with the
123
CHAPTER 5. CONTROL DESIGN
following optimal spline coefficients:
=
θcT =
θoT
0.216 1.926 3.168 3.750 3.977 3.980 3.998 4.002 4.020
T
0.187 1.924 3.174 3.737 3.955 3.980 3.998 4.002 4.020
T
× 10−3
(5.50)
× 10−3
(5.51)
Figure 5.13 illustrates the resultant set of reference trajectories and Figure 5.14 represents the corresponding voltage and current input. Note that for the purposes of
comparison, the opener and closer reference trajectories are shown relative to their
respective magnet pole faces.
As mentioned previously, in practice the continuous
voltage input is approximated by a 50 kHz PWM via the dSPACE controller and
custom power electronics (see Section 6.5.14). The current input is used in the PI
controller as a feedforward input to improve convergence rate.
5.6
Flux-based Position Reconstruction
Due to the low-impact speed requirement of the valve and significant combustion
pressure fluctuations, a means of accurately sensing armature or valve position is required for feedback control. In [Koch et al., 2002] it was demonstrated that a sensor
accuracy of at least 10 µm is required during landing. Presently, a sensor with this
accuracy over a 8mm stroke length is either too expensive, unsuitable for under-hood
environments or otherwise unfeasible for on-board control. In addition, the sensor
response must not limit the control system and consequently stability. Therefore, the
sensor must have at least an equivalent response or bandwidth of the actuator system.
Methods of state reconstruction through external valve and/or armature-based position measurements have been documented. These include position or velocity mea-
124
CHAPTER 5. CONTROL DESIGN
−3
Air Gap [m]
0
1
Velocity [m/s]
0
0.5
1
1.5
2
2.5
0
0.5
1
1.5
2
2.5
6
0
x 10
0.5
1
1.5
2
2.5
0
0.5
1
1.5
2
2.5
0
−1
−2
−3
2
Closer
Opener
0.5
1.5
Acceleration [m/s ]
x 10
4000
2000
0
−2000
Jerk [m/s3]
2
0
−2
−4
Time [ms]
Figure 5.13: Optimized reference trajectories for the hinged actuator with nonlinear
induction model
surements via linear variable differential transformer (LVDT) [Sun and Cleary, 2003],
laser [Tai and Tsao, 2003, Wang et al., 2002, Stubbs, 2000] or eddy current displacement sensors [Peterson et al., 2006, Chung, 2005]. Another feedback system demonstrated uses a microphone to adaptively improve impact speeds from sound intensity measurement [Peterson and Stefanopoulou, 2004]. Although proven successful in
laboratory testbench experiments, such a sensor scheme is likely not practical in an
engine environment due to associated sensor cost, multiple valves in operation and
other acoustic sources.
These sensors provide sufficient precision, accuracy and response, efforts are being
125
CHAPTER 5. CONTROL DESIGN
40
Voltage [V]
20
0
−20
−40
−60
0
0.5
1
1.5
2
2.5
2
2.5
10
Closer
Opener
Current [A]
8
6
4
2
0
0
0.5
1
1.5
Time [ms]
Figure 5.14: Coil input corresponding to the optimized reference trajectories for the
hinged actuator and nonlinear induction model
made to develop alternative production amenable sensors or sensing methods with
equivalent performance. These methods include the flux-based coil type
[Chladny and Koch, 2006a, Scacchioli, 2005, Montanari et al., 2004],
[Ronchi and Rossi, 2002, Rossi and Alberto, 2001, Rossi and Tonielli, 2001],
[Roschke and Bielau, 1995], observer based [Lynch et al., 2003, Eyabi, 2003] and self
inductive [Butzmann et al., 2000, Takashi and Iwao, 1995]. In the latter cases, it is
proposed that the driving coil itself be used to relate the measured rate of change
of induced coil current to the armature position and velocity. This may be done by
momentarily deactivating the drive coil and relating velocity-induced currents to po-
CHAPTER 5. CONTROL DESIGN
126
sition. However, this method is susceptible to noise and signal processing challenges
in addition to temporary loss of control authority [Butzmann et al., 2000] and thus
potentially compromising to tracking performance.
Observer based state reconstruction that makes use of the measured current signal
and estimated initial state conditions to predict plant output has been implemented
in [Eyabi, 2003]. Using only a current measurement makes the position and velocity
estimates sensitive to initial conditions and requires high gain for the relatively rapid
estimation convergence required, making the estimation potentially prone to instability when subjected to excessive noise or disturbances. Thus, it is perhaps better
suited to compensate for relatively slow parameter variations.
5.6.1
Flux Sensor
Flux-based coil type sensors, such as the type used in this work and first proposed in
[Rossi and Tonielli, 2001], appear to be a promising method of achieving a low-cost
yet high-performance position measurement. The system incorporates a secondary
sensor coil concentric to each of the opener and closer magnet drive coils as shown
in Figures 1.2 and 3.4. These sensor coils are terminated across high impedance
analog integration circuits (see Section 6.5.7). The circuit output signal can thus be
related to magnetic flux. Using this signal with the drive current signal, position
may be predicted through an inductance model. In the case of devices with variable
air gaps such as solenoid actuators, inductance is highly dependant on armature
and hence valve position. Thus, any time-varying magnetic field produced by the
excitation coil will induce an electromotive force (EMF) in the secondary coil as
predicted by Faraday’s law of mutual inductance. The secondary coil is measured
with high impedance circuitry and thus the effect of resistance is assumed negligible.
As a result, the induced voltage, vf c , due to a change in excitation current or flux
127
CHAPTER 5. CONTROL DESIGN
may be expressed as:
vf c =Nf c
Nf c dλ(i, x)
dφ(i, x)
=
,
dt
Nec dt
where vf c is the induced voltage in the flux coil,
dφ
dt
(5.52)
is the flux induced potential and
Nf c and Nec are the respective number of turns for the secondary flux measurement
coil and excitation coil. In practice, the induced voltage is integrated (via analog
circuitry) and sampled. Analog drift is assumed linear with respect to time and
corrected for during online measurement through vf d = cf d t, where cf d is a measured
constant. Each channel is calibrated prior to excitation and the integration hardware
is reset externally when the armature is at the opposite pole face of the coil in use.
The reset of the integrator is particularly effective due to the periodic motion of the
valve. The integrators are reset when coil current is zero (releasing magnet) or when
the valve is in the open or closed holding states. The later reset requires an estimate
of the initial flux condition, φo , which is obtained by using the steady state holding
current measurement and the known position (since it is either fully open or closed).
The integrated flux linkage measurement may be expressed as:
λ =Nec (φo +
R
vf c dt − vf d
),
Nf c
(5.53)
where φo is the initial flux condition prior to integration. Using the derived nonlinear induction model, a relationship between current, flux and armature position is
expected to be of the form:
x=
βi
R
+κ
αi + ln(1 − Nec /ψ(φo + vf c dt/Nf c ))
(5.54)
which requires a non-zero current to maintain sufficient accuracy. This restriction
is also imposed in order to satisfy controllability conditions (Section 5.4.5). This
128
CHAPTER 5. CONTROL DESIGN
Valve Position [mm]
8
6
4
2
0
2
10
1.5
8
−3
Flux [Wbx10 ]
6
1
4 Coil Current [A]
0.5
2
0
0
Figure 5.15: Opener position with respect to current and flux
method of position reconstruction is similar to [Montanari et al., 2004], in which two
functions are used to relate reluctance and excitation to air gap. A plot of position
as a function of path flux and excitation current is provided in Figure 5.15 for the
opener magnet. Experimental efforts in this work have shown that improved performance can be attained over Equation (5.54) through use of a numerical look-up table
with FEA results relating flux and current to position [Chladny and Koch, 2006b,
Chladny et al., 2005].
129
CHAPTER 5. CONTROL DESIGN
5.7
Feedforward Controller
The approximate
1
x2
drop in magnetic force with air gap limits the effectiveness of a
closed loop landing controller to regions close to the pole face [Tai and Tsao, 2002].
For the exhaust actuator, the lack of force authority is exacerbated by in-cylinder
combustion gas forces, particularly at higher engine loads, which inhibit valve opening.
Initially, these forces can only be overcome by spring force, after which, open-loop
control is used to setup favorable valve motion and coil current conditions for the start
of the feedback or landing control. By using the position-based gas force relation
in Section 4.5.2 and a nonlinear disturbance observer (see Section 5.8), an energy
based method is used to determine the required coil current output for the open-loop
control. A detailed discussion of the individual system energy terms is provided in
the following section to clarify the proposed feedforward technique.
5.7.1
System Energy Derivation
An essential concept to the proposed feedforward design is system energy. The
armature-valve kinetic and valve spring / torsion bar potential energy may be expressed as a function of the work done to cause motion or displace the springs. The
general expression for work is the amount of force, F , applied over a given path, ds:
W =
I
F · ds
(5.55)
Given this general expression, the definitions for the spring and kinetic energy, as well
as gas force, magnetic closer and opener work can be derived. Kinetic energy, Wk for
an object with mass, m, velocity ẋ and momentum, p = mẋ is defined as (assuming
130
CHAPTER 5. CONTROL DESIGN
starting from rest):
Wk =
Z
F · ds =
Z
ẋ · dp =
Z
mẋdẋ =
mẋ2
2
(5.56)
Similarly, work due to potential spring energy is:
Ws =
Z
Fs dx
(5.57)
where Fs (x) = k(x − xo ) is the position dependent spring force. Thus:
Ws =
Z
x2
x1
k(x − xo )dx = (
kx2
− kxo x)|xx21
2
(5.58)
Force due to viscous friction is approximated as Ff = bẋ. The work done can thus
be related through:
Wf =
Z
bẋdx = b
Z
dx dt
ẋ
dx = b
dt dx
Z
ẋ2 dt
(5.59)
Using the model for simplified gas force, Fgs , from Section 4.5.2, work due to gas
forces, Wgs , over an opening cycle may be expressed as:
Wgs =
Z
Fg dx = Cgf Po Av (c1 xL +
c2 x2L c3 x3L
+
)
2
3
(5.60)
since the path dx is equivalent to valve lift, dxL . Recall, γ represents the combined
initial pressure, valve area and gas force coefficient, γ = Po Av Cgf , and is identified
using the online disturbance observer (see Section 5.8).
Work due to the opener, Wmo , or closer, Wmc , magnets is
Wm,o/c =
Z
Fmo/c dx =
Z
ℓm ψg ′ (x)
ℓm
−g(x)i
1
−
(1
+
ig(x))e
dx
=
Wc,o/c
ℓv g 2 (x)
ℓv
(5.61)
131
CHAPTER 5. CONTROL DESIGN
1.2
W
sys
1
W
g
Work [J]
0.8
W
s
0.6
0.4
W
W
mo
k
0.2
W
W
mc
0
−4
−3
−2
−1
0
Position [m]
1
2
f
3
4
−3
x 10
Figure 5.16: Simulated individual energy terms (5.56), (5.58), (5.59), (5.63), (5.61)
and (5.62) during an opening cycle (4 bar exhaust valve opening pressure)
where Wc is magnetic co-energy.
Finally, net mechanical system energy, Wsys , may be expressed as the sum of the work
done by the spring, kinetic energy, friction, gas force and opener and closer magnets:
Wsys = Wk + Ws + Wf + Wg + Wm,o + Wm,c
(5.62)
Provided all energy terms are accounted for, the system energy should remain constant
over a given opening or closing cycle. Figure 5.16 provides an example of these terms
for the valve system energy during a simulated opening cycle at a 4 bar EVO. Note
that the system energy remains constant, and the dominant terms are spring, kinetic
energy, magnetic opener and gas force work.
132
CHAPTER 5. CONTROL DESIGN
5.7.2
Feedforward Current Input
Typically, due to the characteristics of the valve and port of an engine, gas forces
occur primarily in the first half of the valve stroke as indicated in Figures 4.8 and
5.16. However, the magnetic force required to compensate for gas disturbances can
only be effectively applied in the later half of the valve stroke (see Figure 7.4). The
work done on the valve by gas forces, Wg , can be approximated by integrating (4.33)
with respect to valve lift:
Wgs (xL ) =
Z
xL1
0
Fgs dxL = γ(c1 L +
c2 x2L c3 x3L
+
)
2
3
(5.63)
To compensate for this disturbance, a desired restoration work trajectory, Wd , is
chosen as:
Z
c1 + c2 (xL − S) + c3 (xL − S)2 dxL
h
c
c3 2
=γxL c1 + xL
+ xL
2
3
i
−S(c2 + c3 xL ) + c3 S 2
Wd (xL ) =γ
(5.64)
A disturbance observer estimate is used to estimate the the initial gas force magnitude,
γ, which can then be used to adjust the above trajectory. Just as the majority of the
gas work occurs primarily over the first half of the valve stroke, this trajectory was
chosen to mirror the simplified gas force work about the mid-stroke position so that
the restoring magnetic force is applied in the latter half of the stroke. The advantages
of this strategy are: the opening magnet is only used when it is most effective, thereby
minimizing ohmic losses, and additionally, the disturbance estimate is not required
immediately, providing time for the observer to converge.
The above technique ensures that work performed on the valve by the gas force
133
CHAPTER 5. CONTROL DESIGN
is recovered using the magnetic force by the end of the valve stroke. However, to
expedite transition, (5.64) is adjusted so that the estimated gas work is recovered by
the time the landing controller is engaged (at position x = xlc ). Thus, the desired
magnetic work trajectory used with landing control, Wdlc , takes the form:
c
c3 Wgs (S) h
2
xL c1 + xL
+ xL
Wd (xlc )
2
3
i
−S(c2 + c3 xL ) + c3 S 2
Wdlc (d) =γ
(5.65)
The magnetic coenergy relation is used to convert work into a desired current output.
Using the flux relation in (3.76), the co-energy of the system Wc is:
Wc (x, i) =
Z
Z
i
λ(x, ξ)dξ
0
i
ψ(1 − e−ξg(x) )dξ
0
e−ig(x) − 1
=ψ i+
g(x)
=
(5.66)
To restore the energy lost to gas forces, the desired feedforward current, id , is calculated as a function of initial gas pressure and valve position by equating (5.66) and
(5.65) and solving for current:
−1−g(x)Wdlc
ψ
1
Wdlc W−1 (−e
id (x, Po ) =
+
+
g(x)
ψ
g(x)
)
(5.67)
Again, W−1 is a real-valued branch of Lambert’s W function. In the real system,
current is measured with a hall-effect sensor and regulated through the ‘H’-bridge
power electronics. As with the holding and PI controllers, the desired current, id ,
is obtained using a simple control law based on measured and desired current and
a PWM signal to the power electronics. The feedforward control is engaged upon
134
CHAPTER 5. CONTROL DESIGN
valve release and disengaged during landing control, holding, initialization and error
modes. Any discrepancy between the final feedforward current and the initial landing
control current appears to be minimal, or at least manageable by the landing control
algorithm as no method of current level transition is specified (or appears necessary).
5.8
Disturbance Observer
Since velocity and gas pressure are required by the feedforward and landing controllers
but are not measured, they are estimated online. Improved robustness and landing
performance are expected upon implementation of a gas force estimate and feedforward compensator compared to static position or time based feedforward controller.
Disturbance dynamics are characterized by assuming that gas force magnitude, γ,
may vary from one cycle to the next, however, it’s normalized trajectory, f1 , as a
function of position is typically predictable for all cycles. Therefore, all that must be
identified is the initial force magnitude, γ, which is used as a constant for the entire
cycle (but estimated for each new cycle). Velocity and initial force magnitude estimation are performed using a nonlinear Luenberger observer [Zeitz, 1987]. Defining
the estimated state as ŷ = [ŷ ŷ˙ γ̂], the observer structure is:

 
˙ŷ
   0
  
 ŷ¨  =  −k
   m
  
γ̂˙
0
1
−b
m
0

0 

−f1  
m 

0


ŷ 
 ℓob1



ŷ˙ 
 + ζ(y, iop, icl ) +  ℓob2


γ̂
ℓob3



 (y − ŷ)


(5.68)
where



ζ(y, iop, icl ) = 


0




1
(F
+
F
(y,
i
)
−
F
(y,
i
))
v
mop
op
mcl
cl

m

0
(5.69)
135
CHAPTER 5. CONTROL DESIGN
and the valve position output is represented through y = h(x).
The system is locally observable provided the term − f1m(y) 6= 0. Although the function
f1 (y) does become small towards the end of the valve stroke (f1 (4mm) = 0.01), the
gas force will have largely been dissipated by that time. Additionally, because the
gas force amplitude may be considered a constant over the whole cycle, the estimated
value at the time of landing control engagement (f1 (xlc )) may be used for the remainder of the valve stroke. However, simulated and experimental results have indicated
that an adequate estimate may be obtained over the entire valve stroke.
Assuming negligible magnetic force modeling errors, the observer structure error dynamics are:

 
˙ỹ
−ℓob1
  
  
 
−k
ẽ˙ o = 
 ỹ¨  =  −ℓob2 − m
  
γ̃˙
−ℓob3
1
0
−b
m
−f1 (y)
m
0
0


  ỹ 
 
  ỹ˙ 
 
 
γ̃
(5.70)
Observer gains Lo = [ℓob1 , ℓob2 , ℓob3 ]T are chosen to maintain LHP poles of the characteristic equation det(sI − (Ao − Lo Co )) = s3 + λ2 s2 + λ1 s + λ0 = 0 where:

 0

−k
Ao = 
 −m

0
1
0
−b
m
−f1 (y)
m
0
0






(5.71)
and Co = [1, 0, 0]. so that the observer gains may be expressed as:
m
ℓob1 = − mb + λ2 ℓob2 = − k+ℓmob1 b + λ1 ℓob3 (y) = − fλ10(y)
(5.72)
Note that ℓob3 is a function output injection, y, in order to maintain pole placement
in a similar fashion as [Löewis et al., 2000, Lévine et al., 1996].
CHAPTER 5. CONTROL DESIGN
136
The observer is implemented in real time by Tustin’s method of integration of the
estimated states:
Ts
(γ̇[n + 1] + γ̇[n])
2
˙ + 1] = ŷ[n]
˙ + Ts (ŷ[n
¨ + 1] + ŷ[n]
¨ + ℓob2 ỹ)
ŷ[n
2
Ts ˙
˙ + ℓob1 ỹ)
ŷ[n + 1] = ŷ[n] + (ŷ[n
+ 1] + ŷ[n]
2
γ[n + 1] = γ[n] +
(5.73)
(5.74)
(5.75)
with Ts representing the sample period and
γ̇[n] = ℓob3 (ŷ[n])ỹ
(5.76)
¨ = 1 (−bŷ[n]
˙ − ky[n] + f1 (y[n])γ[n] + ζ(y[n], iop[n], icl [n]))
ŷ[n]
m
(5.77)
Estimated gas force and the respective time derivative are input to the flatness control
law by:
F̂g = γ̂f1 (y)
(5.78)
˙
˙
F̂g = γ̂f2 (y, ŷ)
By including an approximation of the gas force dynamics via f1 and γ in the observer,
higher convergence rates can be expected than if gas force was to be estimated directly.
The estimation routine begins at the stroke bound, therefore initial estimates of position and velocity are known. Good performance over a wide operating range has be
found by taking the initial gas force estimate, γ̂0 , is taken to be 1 bar. This is because
the feedforward controller may input excessive magnetic force if an overestimate of
gas force is made, causing a collision between the armature and opener pole face. The
landing controller has a fundamental limitation in avoiding such a situation as the
CHAPTER 5. CONTROL DESIGN
137
magnets can only apply attractive forces.
For valve closing, the observer order is reduced to two by assuming γ, and hence
gas force, Fg and observer gain, ℓob3 are zero. Thus, aside from consideration of valve
preload and magnetic forces that consider material saturation, the structure is similar
to the observer proposed in [Peterson, 2005] while the valve is in transition from the
open to closed states.
5.9
Summary
A comprehensive valve control methodology is presented for a hinged solenoid valve
actuator. The key features are a flux-based feedback sensor, energy-based feedforward
controller, a nonlinear disturbance observer and flatness-based closed-loop landing
controller. Together, these components form a complete valve control system capable
of soft seating control and cycle-by-cycle gas force disturbance rejection while satisfying onboard and feedback sensor constraints. Simulated and experimental testbench
results of the system are presented in Chapter 7.
Chapter 6
Experimental Setup
6.1
Introduction
T
o ensure the finite element and lumped parameter models adequately capture
the significant physical dynamics, experimental validation is undertaken. Two
experiments are conducted, one involving the actuator in a material testing machine to
investigate the actuator performance (magnetic force and current response), another
to evaluate the control performance (impact velocities and disturbance rejection) on
a testbench engine emulator. Through system identification techniques, experiments
also serve to parameterize and further refine model fidelity, and thus control performance. In addition, preliminary work is undertaken to facilitate the implementation
of an engine control unit (ECU) that will manage the valve actuator controllers. The
following describes in detail the procedure and equipment used for the experimental
testing of a hinged electromagnetic prototype actuator.
6.2
Material Testing Machine Experiments
As a method of validating the developed FEA and analytic models, the opener magnet of the hinged VVT actuator was tested using a material testing machine and
associated apparatus. The actuator is constrained and the armature is manipulated
138
CHAPTER 6. EXPERIMENTAL SETUP
139
Figure 6.1: Hinged actuator performance evaluation experimental setup
to predetermined positions. At each position the magnetic force on the armature
is recorded for a series of steady state and transient current inputs. The material
testing machine also has a dynamic loading ability which is used as a preliminary
method of investigating the flux-based position measurement. A labeled photo of
the experimental apparatus is provided in Figure 6.1. A detailed description of the
equipment is provided in Section 6.5 with relevant specifications listed in Table 6.2.
In these experiments, only the opener portion of the actuator was tested. This is due
to the difficulty in fabricating a nondestructive load cell-to-actuator adaptor capable
of measuring the high tension loads that will be experienced at low closer air gaps.
These experimental tests are primarily intended to validate the the FEA modeling
CHAPTER 6. EXPERIMENTAL SETUP
140
procedure and not to evaluate the actuator performance in its entirety. Due to the
similarity between the opener and closer magnets, it is assumed that the validation of
the opener will demonstrate that the closer FEA is also valid. To mitigate the potential risk of damaging the actuator, the power supply current and overload protection
are used to artificially limit control input to a ‘dummy’ actuator. The actuator power
supply current limit and overload circuit protection is only then relaxed as necessary
and later, the real actuator is installed in the load frame.
6.2.1
Static Evaluation
As in the case of the FEA modeling, a steady state map of magnetic force as a
function of current and position is desired for performance characterization. Digital
current control is implemented in C on the dSPACE controller to regulate a desired
current amplitude for a pre-specified duration (25 ms) using pulse width modulation
(PWM) at a 50 kHz frequency. A Hall-effect current sensor is used for feedback. By
sending appropriate switching signals, the dSPACE controller allows the measured coil
current to rise to the specified level and then switches the power electronics to the
0 V mode. The power is switched on again when the current drops below a threshold
level. Controller switch signals, crosshead position, actuator coil voltage, coil current
and load cell signals are all recorded with the dSPACE controller at a sample rate of
50 kHz. A 5 V TTL logic signal is used to control the 42 V power transistors and is
also used to trigger data recording. Measured signals are saved as MATLAB binary
files (version 6.5). Table 6.1 lists the 19 current levels and 45 armature positions where
force is recorded to contrast with the performance predicted by the FEA and LPMs.
Because the sampling and PWM frequency are identical, voltage measurements are
monitored (via a scope) at a higher rate. The naming convention of all data files are
listed in Appendix C.
141
CHAPTER 6. EXPERIMENTAL SETUP
Table 6.1: Experimental air gap and excitation operating points
Air Gap [mm]
0.02
0.18
0.38
0.70
1.75
0.03
0.20
0.40
0.75
2.00
0.04
0.22
0.42
0.80
2.50
0.05
0.24
0.44
0.85
3.00
0.08
0.28
0.46
0.90
3.50
0.10
0.30
0.50
0.95
4.00
0.12
0.32
0.55
1.00
4.50
0.14
0.34
0.60
1.25
5.00
0.16
0.36
0.65
1.50
6.00
Coil Excitations [Ampere-turns]
50 100 150 200 250 300 350 400 450 500
600 700 800 900 1000 1250 1500 1750 2000
6.2.2
Transient Evaluation
Three types of transient excitation experiments are conducted to investigate the actuator response and flux-based position sensing technique. In all cases, coil voltage,
current, flux signals, crosshead postion and force data are recorded in MATLAB binary format for comparison to the transient FEA opener and LPM model results. See
Appendix C for a full listing.
6.2.2.1 Step Response
Aside from validating the transient performance of the FEA and LPM models, evaluating an actuator coil response to a step voltage input provides insight to the extent
which eddy-currents persist in the magnetic path material [Chladny et al., 2005]. For
that reason, an experiment is conducted using the dSPACE controller where the coil
current and magnetic force response to a step voltage input of a pre-specified amplitude over several positions is measured. To prevent coil damage, the step duration
is gradually increased to 1.74 ms which is sufficient to reach a current amplitude of
30 A at the largest air gap (lowest inductance) and voltage amplitude tested. After
the step duration period is exceeded, voltage polarity is effectively reversed (see the
CHAPTER 6. EXPERIMENTAL SETUP
142
power electronics Section 6.5.14) to reduce the flux and current as rapidly as possible.
Voltages of 24, 42 and 50 V are selected to investigate the effect on response. Armature position is varied between 0.50 and 2.00 mm in 0.25 to 0.50 mm increments. All
collected data are archived in MATLAB binary files as listed in Appendix C.
6.2.2.2 Sinusoidal Response
As an initial investigation of flux-based sensor performance, a sinusoidal current is
input at various predefined armature positions. The corresponding force, crosshead
position and integrated flux signals are recorded to reconstruct the position off-line.
A signal with an amplitude of 5 A and frequency of 750 Hz with mean excitations of
5, 10 and 15 A are tracked by the same current controller used in the steady state
experiments. As this experiment is primarily exploratory in nature, these tests were
also performed at voltages of 24, 42 and 50 V for air gaps between 0.50 and 2.00 mm
in 0.25 to 0.50 mm increments. All collected data are archived in MATLAB binary
files as listed in Appendix C.
6.2.2.3 Response with Armature Motion
The material testing apparatus shown in Figure 6.2 can manipulate the armature
position through changing the lower crosshead displacement with predefined time
varying functions. This feature is used to evaluate the flux sensor performance over
varying positions and current excitations. The armature position is varied sinusoidally
for various mean positions, amplitudes and frequencies over various step and sinusoidal current excitations while force, crosshead position and integrated flux signals
are recorded. Unfortunately, the maximum amplitude and frequency of the position
was limited to 1.50 mm and 10 Hz respectively, resulting in relatively slow speeds
(<0.1 m/s). This limitation is due to the material testing machine power and hydraulic pressure overload safeguards. As well, there is concern of destroying the
CHAPTER 6. EXPERIMENTAL SETUP
143
Figure 6.2: Single cylinder head test-bench setup
actuator because at higher frequencies, the stroke amplitude must be artificially increased beyond the actuator stroke bounds to compensate for the observed amplitude
attenuation. Although these armature speeds and position amplitudes are relatively
slow with respect to those experienced during actuation, they provide initial insight
into the flux sensor performance.
6.3
Testbench Engine Emulator
A techbench method of experimentally evaluating actuator control performance is
designed to avoid the added complexity of implementation on a working single cylinder engine. The designed system is located in 4-28 of the Mechanical Engineering
Building at the University of Alberta. It consists of an intake and exhaust actuator
CHAPTER 6. EXPERIMENTAL SETUP
144
Figure 6.3: Cut-away view of testbench cavity
module (four individual actuators) mounted to a prototype single cylinder engine
head. The cylinder head is donated by DaimlerChrysler AG specifically for use with
the prototype actuators and single cylinder engine. Three aluminum plates are designed to constrain the cylinder head to a custom work table by as shown in Figure
6.2 and a steel sub-frame (not shown). The front support plate has a centrally located
cylindrical hole with a diameter corresponding to the intended engine cylinder bore.
A 3D and 2D cross-sectional representation of the cavity and associated hardware
are provided in Figures 6.3 and 6.4, respectively. The cavity is pressurized with compressed air from the building supply. In order to produce realistic exhaust pressure
transients, cavity volume may be varied by affixing aluminum disks (of known length
and diameter) with press-fit permanent magnets to the unused valve faces. This is
CHAPTER 6. EXPERIMENTAL SETUP
145
Figure 6.4: Sectional view of cylinder head test-bench cavity
possible since only one actuator is tested at a time (and therefore up to three valves
not in use). The chamber is completed by a custom designed sensor mount-plate that
fits into the cavity and is secured with cap screws. This plate has ports to accommodate a compressed air inlet and a pressure transducer. The mount plate contains
two windows so that a position measurement can be made on any of the four valves
(through repositioning of the laser sensor). An air seal is provided with o-rings around
each of the windows and the cavity bore. Cavity pressure regulation is achieved by
an electronic pressure regulator and a two-way solenoid valve via signals generated
by the dSPACE controller. Opening pressures can be varied between 0 and 5 bar. A
48 L (11 gal) accumulator tank between the pressure regulator and two-way valve is
used to mitigate pressure pulsations. The air is expelled through the exhaust port
and muffler (upon actuator valve opening) as though it were exhaust gas. To accommodate the relatively continuous flow rates associated with higher emulated engine
speeds, a ball valve may be manually opened in parallel with the two-way solenoid
valve. As in the material testing machine experiments, actuator power is regulated
CHAPTER 6. EXPERIMENTAL SETUP
146
Figure 6.5: Single cylinder head test-bench setup schematic
through the same custom designed power electronics and a DC switched power supply
(also configurable through the dSPACE controller). A schematic representation of the
testbench system is provided in Figure 6.5. The testbench experiments are primarily
focused on controlling the exhaust actuator subject to gas disturbances as control of
the intake actuator is expected to be the same but without significant disturbances.
Valve control of a single valve was first considered as only one laser position sensor is
available. However, the laser is only used as a performance reference and is not used
by the controller. As a result of the torsion and valve spring forces being balanced at
a mid-stroke position, the valve lash adjusters were removed from the remaining ex-
CHAPTER 6. EXPERIMENTAL SETUP
147
haust and two intake valves so that the valve springs could hold the valves closed and
hence seal the combustion chamber during the experiments. The tests progressed in
complexity, initially with basic software and hardware checks followed by system identification of the mechanical parameters. A position-based feedforward controller and
flatness landing controller were then implemented (for both the opener and closer
magnets) initially using an externally mounted laser position sensor for feedback.
Later, the flux-based position algorithm replaced the laser measurement for feedback with the results documented in [Chladny and Koch, 2006a]. Next, a pneumatic
system used for the emulation of exhaust gas disturbances was installed. The feedforward algorithm was manually tuned to accommodate fixed opening pressures as
shown in [Chladny and Koch, 2006b]. Finally, feedforward and estimation algorithms
were implemented for the opener magnet to accommodate a random exhaust opening
pressure between zero and 5.0 bar as shown in [Chladny and Koch, 2007]. Using the
developed control strategy, landing impact velocities, transition time and energy consumption were evaluated for various pressure levels, engine speeds and voltages. All
data were acquired through the ControlDesk interface and saved in MATLAB binary
format (see Appendix C for a file listing and Chapter 7 for results).
Due to the relatively large number of system components and controller complexity,
this test-bench validation process is critical for debugging software, coordinating the
actuator controllers with the engine control unit and developing an associated software
interface. In addition, it is an opportunity for validating, evaluating and improving
the derived control strategies in a closely controlled environment and without risk
of compromising a working engine. Note that precautions should be taken to avoid
eye injury from the laser position sensor or hearing loss from expansion of the compressed air through the exhaust port. A detailed description of the equipment used
is provided in Section 6.5 with relevant equipment specifications listed in Table 6.3.
To achieve independent valve timing control for a two valve single cylinder engine,
CHAPTER 6. EXPERIMENTAL SETUP
148
the described exhaust actuator controller and an additional DS1103 controller for the
intake valve will be coordinated by a dSPACE 1404 MicroAutobox engine controller.
6.4
Preparation for Single Cylinder Engine Testing
Upon successful test bench testing, the cylinder head will be mounted to a custom
designed cylinder barrel and sleeve for experimental validation on a Ricardo Mark III
single cylinder research engine shown in Figure 6.6. These experiments will be used
to further validate and refine the proposed actuator control strategies by contrasting
performance with a modified production cam-driven cylinder head with similar combustion chamber and port geometry (not shown). Having a fully flexible valvetrain
will also serve as a significant extension of the present research engine test facility
capabilities. To this end, considerable work has been done to upgrade the present
engine control unit with a dSPACE MicroAutobox 1404 flexible engine controller and
associated interface electronics to coordinate two dSPACE DS1103 controllers (for an
intake and exhaust valve actuator) in addition to fuel injector(s) and spark timing
control.
6.4.1
MicroAutobox Engine Controller and Interface Electronics
A hardware interface between the dSPACE MicroAutobox 1404 engine controller and
engine hardware is presently in the final stages of testing. As part of this process,
a 91 conductor interface cable between the 1404 and proprietary electronic cam and
crank signal conditioners, analog and digital input/out conditioners and ignition and
fuel injector drivers (provided by Bazooka Electronics and Hitachi [Hitachi, 2003])
is designed and tested. A photo of the interface electronics and 1404 controller is
provided in Figure 6.7. At the time of writing, the interface and 1404 unit is operating
the aforementioned single cylinder engine with a conventional cam-driven valvetrain.
CHAPTER 6. EXPERIMENTAL SETUP
149
Figure 6.6: Ricardo Mark III single cylinder engine and test facility
6.5
Equipment Description
The following subsections provide detailed descriptions of the relevant equipment used
in the material testing and testbench experiments.
6.5.1
Actuator Adapter and Load Rod
An adapter that attaches the actuator to the lower material testing machine crosshead
was designed and fabricated from AISI 6061 T6 aluminum. The actuator was attached
to the adaptor using the socket-head cap screws and mounting locations provided for
normal cylinder head mounting. The adaptor is secured to the crosshead with steel
ready-rod and a locknut. To manipulate the armature position with the material
CHAPTER 6. EXPERIMENTAL SETUP
150
Figure 6.7: dSPACE MicroAutobox and custom interface electronics
testing machine crosshead, an aluminum load rod was fashioned to press against the
armature valve lash adjuster socket. In previous experiments [Chladny, 2003], load
cell measurements were observed to be affected by the actuator magnetic fields. The
rod was made of sufficient length so that no detectable interference was observed in
the load signal during excitation. Aluminum was chosen over steel as the construction
material to minimize a flux path to the load cell.
6.5.2
Circuit Protection
To prevent significant levels of current to flow unchecked through the actuator in the
event of a control or hardware failure, an overload protection circuit was constructed.
The circuit consists of a low-pass filter for the coil current which generates a signal
to the power supply when the low frequency current signal exceeds a variable shutoff
CHAPTER 6. EXPERIMENTAL SETUP
151
limit. The output lead from the primary power supply to the power electronics is
measured using a Hall-effect current sensor. The shutoff threshold is manually adjusted to allow a suitable current limit. It was also deemed prudent to protect the
expensive dSPACE controller from potential electrical spikes or surges that may originate from an actuator or power electronics failure. Thus, all switch signals from the
dSPACE controller to the power electronics are optically isolated using a 16 channel,
10 MBit/s opto-isolator circuit (comprised of several 6N137 HCPL-2601/2611 integrated circuits).
Both the overload circuit and opto-isolator require a ±15 V and 5 V DC power source
which is provided by a HP 6236B triple output power supply. Electrical schematics
of both devices are provided in [Chung, 2005].
6.5.3
Computer Hardware and Software
During the material testing machine experiments, a PC laptop running Microsoft
Windows XP and dSPACE ControlDesk 2.4 was used to communicate to a standalone dSPACE DS1103 PowerPC (PPC) target controller board via a PCMCIA card
and ethernet adaptor. The testbench experiments use a host PC running a Microsoft
Windows XP operating System and PCI card to optically link to the DS1103 PPC
target controller board. The ethernet and optical links provide a two-way communication between the PC host and controller for code uploads and data downloads. A
user interface developed with dSPACE ControlDesk software provides non-realtime
communication of parameters and transfer of real-time data acquisition between the
DS1103 controller and the PC.
6.5.4
Current and Voltage Sensing
Both coil voltage and coil current measurement signals are generated on-board the
custom power electronics (see Section 6.5.14). Coil voltage is measured by dividing
CHAPTER 6. EXPERIMENTAL SETUP
152
it with a differential operational amplifier circuit so that it is within the DS1103
analog to digital converter (ADC) input range. The value of the gain and resistors
are chosen such that the output voltage is one fifth of the actual voltage across the
coil. A schematic of the sensing system used is provided in [Chung, 2005].
Coil current is measured with a LEM LA55-P Hall-effect current sensor. The Halleffect current sensor works by directing the magnetic field produced by a flowing
current, given by B =
µo I
,
2πr
into a semiconducting material using a steel yoke. When
a semiconducting material is exposed to a magnetic field, it produces a measurable
current and voltage in proportion to the impinging field as predicted by the Halleffect phenomenon. In order to avoid the nonlinear behavior of the yoke and the
semiconducting material itself, an op-amp uses a coil that feeds back an opposing
magnetic flux in the yoke in proportion to the output of the semiconducting material.
Thus, any flux produced by the current intended to be measured is equally countered
by the flux produced by the op-amp driven coil. The counter flux is generated by
passing the op-amp current through a resistor and, N, external loops around the
yoke. The op-amp voltage can then be measured since it is proportional to the
current required to balance the magnetic flux produced by the line current. It is
then possible to relate the op-amp voltage to the current flowing in the conductor by
V =
IM easured R
.
N
6.5.5
Cylinder Head
A single cylinder engine head (donated by DaimlerChrysler AG) is incorporated into
the testbench apparatus to evaluate actuator control performance under realistic engine operating conditions but without the additional complications or risk of damaging a research test engine. The cast aluminum alloy cylinder head has four valves
and a bore of 97.00 mm with a combustion chamber and valve porting similar to the
Mercedes-Benz M273-E55 engine (used in the E, CLK, CL, CLS, GL and S 500/550
CHAPTER 6. EXPERIMENTAL SETUP
153
models, 2006 or newer). In the future, the cylinder head is to be removed from the
testbench and installed on a single cylinder engine.
6.5.6
dSPACE DS1103 Controller
For both the material testing machine and testbench experiments, dSPACE control
hardware is used. The system includes the DS1103 target board contained in a PX4
expansion box and connected to a CLP1103 connector/LED panel. The target board
is equipped with a master 400 MHz PPC Motorola 604e microprocessor and a 20 MHz
Texas Instruments TMS320F240 slave processor. Control at frequencies up to 50 kHz
are used through the target board which includes 16 analog to digital converters
(ADCs), 12-bit multiplexed channels equipped with four sample and hold ADC as
well as four parallel 16-bit channels each equipped with individual sample and hold
ADCs. Digital to analog output is available through 8, 14-bit channels in addition to
32-bit parallel I/O organized in four 8-bit groups. Each group can be programmed
to be either input or output. The slave processor provides an additional 16 10-bit
ADC inputs and 4 single phase PWM outputs. Complete specifications are available
in [dSPACE GmBH, 2003]. Measurements of the two coil voltages, currents, flux
linkage signals, laser position sensor and pressure sensor are sampled at a rate of 20 µs
(50 kHz) while executing the control algorithms. A second 100 µs background task is
established to accommodate lower priority processes such as an initialization routine,
RPM generator and processor temperature monitoring. Although programming is
available through the MATLAB Simulink / RTI environment, all algorithms are coded
in C language in part to improve real-time performance. All code is compiled and
downloaded from the host PC to the DS1103 board. Experiments are managed using
ControlDesk 2.4 software as a convenient method of interacting with the controller
and transferring data back to the PC.
CHAPTER 6. EXPERIMENTAL SETUP
6.5.7
154
Flux Sensor Integration Electronics
As shown in Section 5.6.1, magnetic path flux and coil current can be related to
armature position through a flux linkage model. However, due to the switching of
the coil current, potentially at the same rate as the sample frequency, magnetic flux
cannot be measured directly. Instead, the rate of change of flux is integrated with
custom analog integration circuits and then sampled through the DS1103 controller.
The rate of change of flux is physically measured as the mutually induced potential in
a measurement coil that is co-axially aligned with the primary excitation coil. Since
a magnetic flux must be present in order to infer a position measurement, a measurement coil is in both the opener and closer electromagnets since they are not energized
simultaneously. As a result, excitation current and position dependent logic is used
to change between the measurement coil signals during the valve transition to maintain a continuous position measurement. Despite such additional complications, the
feedback method is cost effective and practical since there are no moving parts and all
components are relatively inexpensive compared to other feedback sensing methods.
Integration is achieved through the use of a standard resistor-capacitor operational
amplifier circuit designed by DaimlerChrysler AG and fabricated by Bazooka Electronics.
6.5.7.1 Integration Drift
Op-amp integration circuits are subject to drift. Small non-zero bias currents cause
the feedback loop capacitor to accumulate charge until the op-amp output saturates.
The bias results from small amounts of current at each input terminal - usually drawn
from the ground plane [Chaniotakis and Cory, 2006]. The result of this bias when integrated is known as integration drift and has a tendency to affect the measured
signal. Fortunately, the time required for typical valve flight is on the order of 5 ms,
155
CHAPTER 6. EXPERIMENTAL SETUP
−3
20
x 10
18
Integrated Flux Signal [V]
16
14
12
10
8
6
0.50 mm
6.00 mm
7.00 mm
7.50 mm
Fit
4
2
0
0
2
4
6
8
Time [ms]
10
12
14
16
Figure 6.8: Analog Drift of the RC Integration Circuit
thus, the accumulated amount of drift over a measurement cycle is relatively small.
To manage drift (and prevent saturation), the integrators are reset by temporarily
grounding the measured capacitor that accumulates charge from the operational amplifier output when the valve is in the fully open or closed position (where position
is known). In addition, the flux sensor range is only appropriate over a limited valve
stroke range (and hence the time over which a measurement is affected by drift is
even less than any given flight time). Additionally, the drift may be assumed to be
linear with respect to time and therefore easily compensated for. During experimental
tests, the drift was measured at several positions by first resetting the integrator and
then sampling the ensuing drift. From these tests, a rate of drift was estimated and
used to correct all subsequent flux measurements. Figure 6.8 illustrates the results of
these tests and the resulting fit curve used to compensate all flux measurements. On
CHAPTER 6. EXPERIMENTAL SETUP
156
Figure 6.9: Laser sensor schematic
a production engine, this bias could be learned when the valve is in a known position
(such as open or closed) or could be filtered using a high-pass filter
6.5.8
Laser Position Sensor
A non-contact method of measuring valve displacement for validation of control performance was desired. A non-contacting sensor avoids complications arising from
physically attaching a sensor component to the valve such as altering the system
mass. Additionally, an optical method is ideal because of the compressed air pressure
CHAPTER 6. EXPERIMENTAL SETUP
157
transients within the cavity. A Micro-Epsilon LD1627-10 laser transducer is selected
to monitor valve position. By using a triangulation method and a semiconductor
optoelectronic position sensitive device (PSD) a position resolution of 6 µm (noise)
and frequency response of up to 37kHz over a 10 mm stroke is possible. The sensing principal works by projecting a point of laser light onto a target which is then
reflected diffusely. A lens then focusses the light on the PSD. The PSD then outputs
an analog voltage signal related to the impinging laser light position on a sensing
surface. Figure 6.9 indicates a schematic of the triangulation method. The following
indicates how a change in valve position, ∆x is related to a change in position of laser
light impingement on the PSD surface, ∆d [Song et al., 2006]:
∆x =
x2
∆d
b(a + c) cos α − b sin2 α
(6.1)
A gauge pressure change from atmospheric to 6 bar will alter the refractive index
of the air from approximately 1.0003 to 1.0015 [Picard and Fang, 2003]. According to the manufacturer, a change in refractive index causes a DC offset in position output and that the aforementioned change is insignificant (less than 100 ppm)
[Micro-Epsilon, 2007]. However, the tempered 3.0 mm plate glass window (refractive
index of 1.4740) does cause a static DC offset of approximately 0.2 mm and is compensated for prior to testbench experiments.
Compared with a charge coupled device (CCD), a PSD has higher sensitivity, response and a continuous sensing surface. However due to the analog nature of the
generated signal, it is more adversely affected by environment noise. It is important
to emphasize that the output position signal cannot by used in a real engine but
is used to validate the flux-based position sensors. To conserve the laser diode life,
power is switched off via the DS1103 controller when not in use.
CHAPTER 6. EXPERIMENTAL SETUP
158
Figure 6.10: Laser and pressure sensor mounting assembly
6.5.9
Laser and Pressure Sensor Mount
A laser sensor mount was designed to for the LD1627-10 to accommodate measurement of any four of the cylinder head valves. The incident beam on the valve face is
aligned with the valve axes to minimize measurement errors caused by misalignment.
The mount is easily removed and realigned through a high tolerance fit between the
mount plate and front support plate cavity. Sensor noise attributed to ambient light
variations is minimized through a relatively enclosed design and a black anodized surface finish. Compressed air inlet and pressure measurement ports are also provided.
Removable tempered glass windows are provided for optical access to the valve faces.
The complete assembly is shown in Figure 6.10.
CHAPTER 6. EXPERIMENTAL SETUP
6.5.10
159
Load Cells
To measure armature force in the material testing machine experiments, a Strainsert FL1U-2SGKT flat strain gauge load cell is installed between the load rod and
crosshead. The load cell strain gauge bridge is balanced and conditioned using a
Vishay 2100 strain gauge conditioner and amplifier system. Measurements from a
second load cell that was already installed in the load frame are also recorded. However, the second load cell is of a higher capacity and situated above the crosshead
and therefore less sensitive to lower excitation levels. This larger load cell is used to
calibrate the smaller capacity Strainsert as it had previously been calibrated with a
proving ring.
6.5.11
Material Testing Machine
The material testing machine used is a MTS 810 servohydraulic system with 318.10
load frame and SilentFlo 505.07 hydraulic power unit located in 3-26 of the Mechanical
Engineering Building at the University of Alberta. The axial displacement control
capability of the machine is used in all tests. The load frame is adjusted to allow
for ample clearance for the load rod, load cell and load cell adaptor to be installed.
After the rod, load cell and adaptors are installed into the load frame, the armature
position is set through the manual position control knob on the user interface control
unit. Upon sufficient change in crosshead displacement, the actuator armature stem
contacts the fixed load rod. Further change in the crosshead position overcomes the
torsion bar force and displaces the armature to a desired position. Great care should
be taken to not exceed the actuator 8 mm stroke limit to prevent the crosshead
from damaging the actuator. Specific positions are set and recorded by using the
crosshead position sensor and user display. The control unit also allows for cyclical
position control as discussed in Section 6.2.2.3.
CHAPTER 6. EXPERIMENTAL SETUP
6.5.12
160
Pressure Regulator and Two-way Solenoid Valve
Compressed air is regulated in the testbench cavity using a Wilkerson ER1 electromechanical pressure regulator and a 2-way ASCO 104R solenoid valve. The pressure
regulator has a built in pressure transducer and control hardware to regulate a pressure set-point via a 0-10 V analog signal input. In this way, pressure can be varied
through ControlDesk and the DS1103 hardware. Although this regulator is ideal
for setting a nominal pressure, it has a relatively slow response and is significantly
upstream from the testbench cavity. Thus, the solenoid valve is attached as closely
as possible to the cavity inlet port so as to more precisely control the pressure level
on a cycle-by-cycle basis. A simple control algorithm uses the pressure transducer
signal for feedback to regulate the internal cavity pressure at a preset level (specified
through ControlDesk) by opening and closing the solenoid. Through coordination of
the regulator and solenoid set points, a wide variety of exhaust valve opening pressures at different speeds can be tested. At higher emulated engine speeds, a ball valve
that is in parallel with the solenoid valve can also be adjusted to provide a higher
flow rate (at the sacrifice of having to use the slower pressure regulator).
6.5.13
Pressure Transducer and Charge Amplifier
To measure testbench cavity pressure transients, a Kistler 6061B pressure transducer
and Sundstrand 507 charge amplifier are used as shown in 6.5. Early tests used a
Kistler 6043A60 sensor mounted to the pressure mount. This was later exchanged for
a 6061B which is compatible with a measurement port provided in the cylinder head.
A pressure signal is required for cavity pressure regulation as well as to validate
and improve gas force disturbance models and estimation algorithms. A pressure
measurement is not used for actuator landing control or disturbance rejection because
production engines are not equipped with such sensors. A piezoelectric pressure
CHAPTER 6. EXPERIMENTAL SETUP
161
transducer is used because they exhibit superior frequency response, ruggedness and
insensitivity to magnetic fields when compared to other pressure sensing methods. A
change in pressure causes a thin membrane to apply force in a single direction on
a piezoelectric crystal. An electric charge is then produced in proportion to crystal
deformation. This charge is then measured and amplified through the charge amplifier
(as in an op-amp integrator circuit). The charge amplifier outputs an analog voltage
in proportion to accumulated charge from the piezoelectric pressure transducer. As
in the case of the flux integration circuits, the charge amplifier is subject to drift
over time and is thus externally reset by the DS1103 controller. The transducer and
charge amplifier were calibrated for a 10 V output corresponding to a pressure of
6 bar (90 psi) using a a Bundaberg 280L hydraulic pressure calibration stand.
6.5.14
Power Electronics
The relatively high current demands of the hinged actuator requires the control signals
from the DS1103 to be amplified. A switched H-bridge configuration for driving
the actuator coils is used because of the relative simplicity and low cost of such
circuits. A set of custom H-bridge power electronics were designed and built by
Bazooka Electronics Ltd. to supply the actuator with the necessary electrical power
in a coordinated and repeatable fashion. These circuits are designed to switch up
to 50 V up at frequencies up to 50 kHz. The actuator opener and closer coils,
opto-isolator signal lines, voltage and current measurement signal lines as well as a
3 kW and 35 W power supplies were connected to the custom power electronics. The
electronics provide three basic output modes of +42 VDC, 0 VDC and −42 VDC
to the coil. These three modes are achieved with two high speed voltage-controlled
insulated gate bipolar transistors (IGBT IRG4BC40W), T1 and T2, and two flyback
diodes, D1 and D2 . Inputs to the transistors for each of the three modes are shown
in the H-bridge power circuit representation provided in Figure 6.11. Here, a digital
CHAPTER 6. EXPERIMENTAL SETUP
162
Figure 6.11: Power electronic modes
high input closes the transistors.
In +42 V mode, transistors T1 and T2 are closed and current flows from the power
supply, through the actuator and to ground.
After the coil is energized, an effective potential of 0 V may be applied by setting
transistors T1 open and T2 closed. This results in the coil current to ‘free-wheel’
through diode D2 with a net potential difference across the actuator of 0 V.
Also available is a -42 V mode. After the coil is energized, T2 may be opened so that
the only current path available is through diodes D2 and D1. This mode is used to
drive the current down at a faster rate than the 0 V mode due to the effective change
CHAPTER 6. EXPERIMENTAL SETUP
163
in polarity. For example, it is used when releasing the armature in order to reduce the
magnetic force quickly and subsequently achieve faster valve travel time. It should
be noted that when T2 is open, the mode is the same regardless of the state of T1.
A 50 kHz PWM signal from the DS1103 controller board controls T1 (+42 and
0 V modes) while a digital TTL output signal regulates T2 . With sufficiently fast
switching frequencies and a relatively high inductance, it is reasonable to assume that
PWM duty cycle variation can regulate an average linear voltage output as done in
[Peterson and Stefanopoulou, 2004, Deng and Nehl, 1998, Pawlak and Nehl, 1988].
6.5.15
Power Supplies
Actuator power in both the material testing and testbench experiments is provided
with Sorensen programmable switching power supplies. In the material testing experiments a DCS60-18E 1 kW model was used and in the testbench setup, a larger
DCS80-37 3 kW supply is used. Both supplies convert a 60 Hz AC source into a variable DC power source. The testbench supply is remotely shutdown via the DS1103
controller and through a connection to an external overload circuit that monitors
the supply current output. Although the 1 kW and 3 kW power supply outputs are
respectively rated to maximum 18 A and 37 A of continuous current, it was observed
that both supplies are capable of providing short term (10 ms or less) currents in
excess of 50 A at 50 V (with a ‘dummy’ actuator load). Thus, actuator performance
is not expected to be limited by the supplies.
All other hardware accessories such as the electronic pressure regulator, opto-isolation,
analog integration, power electronic and current overload protection circuits are powered through a linear 35 W HP 6236B triple output supply.
CHAPTER 6. EXPERIMENTAL SETUP
6.6
164
Summary
The above sections provide a brief summary of the hardware and procedure used to
conduct the experimental studies. Using an axial material testing machine, custom
H-bridge power electronics are interfaced with a switching power supply and high
speed controller to investigate the static and transient force and current response of a
hinged prototype actuator. Actuator force is measured with a load cell and magnetic
path flux was inferred through analog integration electronics. Additional testbench
experiments are conducted to evaluate actuator control performance under emulated
engine conditions. Emphasis is placed on the study of the exhaust valve actuation to
address the issue of gas force disturbance rejection. Both experiments are essential in
validating developed models and qualifying and tuning the control algorithms prior to
implementation on a single cylinder research engine. The apparatus components and
relevant specifications for the material testing and testbench experiments are listed
in Tables 6.2 and 6.3, respectively.
165
CHAPTER 6. EXPERIMENTAL SETUP
Table 6.2: Material Testing Machine Experimental Equipment
Item
Prototype
Actuator
Actuator
Adaptor
Actuator-Load
Cell Rod
Description
Valve actuator RAM 113 050 02 77
Provided by DaimlerChrysler AG
Connects the actuator to the MTS
crosshead
Transfers load from the armature
to the load cell
Load Cell
Adaptor
Actuator Power
Supply
Accessory Power
Supply
Fixes the load cell to the MTS
load frame
Sorensen DCS60-18E programmable
switching power supply
Hewlett Packard 6236B triple
output linear supply for power
and miscellaneous electronics
Custom H-Bridge driver for
actuator current control with on
board current hall-effect sensor
Custom analog integration
circuits for position feedback
board current hall-effect sensor
Strainsert FL1U-2SGKT flat load
cell
MTS 661 20E-03 load cell
Power Electronics
Flux Measurement
Electronics
Load Cell
Load Cell
(in load frame)
Load Frame
Load Controller
Strain Gauge
Conditioner
Host PC
Controller
Misc. Electronics
Misc. Hardware
MTS 318.10 Axial Load Frame and
SilentFlo 505.07 hydraulic unit
MTS 810 servohydrauic system and
TestStar IIs controller
Vishay 2100 provides load cell
excitation and calibration
IBM ThinkPad notebook for data
collection / control interface
dSPACE DS1103 with PX4 enclosure
PCI card and 400 MHz processor
16 X 10 Mbit/s opto-isolator
Variable overload protection
Various connectors, leads and
fasteners to connect hardware
Relevant Specs.
8.00 mm (0.315”)
stroke
AISI 6061 T6 AL
plate
13mm (1/2”)φ x
300mm (12”) 6061
T6 Aluminum bar
6061 T6 Aluminum
0-60V, 0-18A
Max output: 1 kW
+6V, ±-20V
35 W
Switching ≤ 50 kHz
Max: 55V @ 50A
IGBT IRG4BC40W
Four channel with
external reset
Cap.: 200V @ 70 A
Cap.: 4448 N
Acc.: 1% FS
Cap.: 100 kN
Acc.: 1% FS
Cap.: 100 kN
(22 kip)
1-12V Excitation
Gain: 100-2100
1.2 GHz CPU
512 Mb RAM
16Bit sampling
to 50 kHz,
IC HCPL-2601
-
166
CHAPTER 6. EXPERIMENTAL SETUP
Table 6.3: Testbench Experimental Equipment
Item
Description
Prototype
Actuator
Hinged exhaust valve actuator
No. RAM 113 050 02 77 provided
by DaimlerChrysler AG
Model No. HAM 113 016 L003
provided by DaimlerChrysler AG
dSPACE DS1103 with PX4 enclosure
PCI card and 400 MHz processor
Custom analog integration
circuits for position feedback
Micro-Epsilon LD1627-10 laser
triangulation (PSD type) 10 mm
Kistler 6061B piezoelectric
transducer for M10x1 port
Single Cylinder
Head
Controller
Flux Measurement
Electronics
Valve Position
Sensor
Pressure Sensor
Charge Amplifier
Power Electronics
Actuator Power
Supply
Accessory Power
Supply
Pressure
Regulator
Host PC
Misc. Pneumatics
Misc. Electronics
Misc. Hardware
Sundstrand 507, adjustable
sensitivity of 0.01 to 110 pC/V
Custom H-Bridge driver for
actuator current control with on
board current hall-effect sensor
Sorensen DCS80-37 programmable
switching power supply
Hewlett Packard 6236B triple
output linear supply for power
and miscellaneous electronics
Wilkerson ER1, 1/2” NPT ports
Step response: 600 ms (nominal)
Windows XP desktop computer for
data collection and control
interface. Dual monitor output.
Parker F602 filter, 48L vessel
Bourdon tube indicator, flexible
hose (1/2”) and various fittings
16 X 10 Mbit/s opto-isolator
Variable overload protection
Various connectors, leads and
fasteners to connect hardware
Relevant
Specifications
8.00 mm (0.315”)
stroke
Four valve
Vol: 60.2 cm3
16 Bit sampling
to 50 kHz,
Four channel with
external reset
Freq. resp: 37 kHz
Res.:6 µm (noise)
Natural Freq: 90 kHz
Sensitivity: -25 pC/bar
Max pressure: 300 bar
Max Freq. 350 kHz
Switching ≤ 50 kHz
Max: 55V @ 50 A
IGBT IRG4BC40W
0-80 V, 0-37 A
Max output: 3 kW
+6V, ±-20 V
35 W
0-10 V input
Max pressure: 10.3 bar
AMD Athlon XP 1800
1.53 GHz CPU
1 GB Ram
IC HCPL-2601
-
Chapter 7
Results
7.1
Introduction
I
n the preceding chapters, finite element, Simulink and lumped parameter models
are provided for a prototype hinged solenoid variable valve timing actuator. The
models are used to predict the actuator steady state and transient response in part to
evaluate the performance of the various models. Experiments to validate the derived
models and a comprehensive control strategy are conducted. The following sections
highlight key simulated and experimental findings.
7.2
Simulated and Measured Actuator Response
Prior to control development, extensive simulation and experimental studies are conducted to characterize the actuator performance. The simulations include the finite
element analysis, lumped parameter-FEA hybrid model (Simulink) and state space
lumped parameter model (LPM) as discussed in Chapter 4. Associated experimental
measurements are made using a material testing apparatus described in Section 6.2.
Two significant types of studies are conducted, namely, steady state performance and
transient performance. The following sections briefly summarize these results.
167
CHAPTER 7. RESULTS
168
Figure 7.1: FEA simulated opener force as a function of respective air gap and steady
state current
7.2.1
Static Performance Evaluation
Static force and magnetic flux as a function of position and current are simulated using an FEA model (Section 4.3) and measured experimentally (6.2.1) over a relatively
wide range of armature positions and current levels. Figures 7.1 and 7.2 indicate the
simulated and measured steady state force response for the opener magnet. The
simulated force response of closer magnet is provided in 7.3. Figure 7.4 compares
measured and simulated armature valve force as a function of position and various
steady state excitation. Also plotted is the combined torsion bar and valve spring
force, kx, from (4.19), which in the range shown, opposes the magnetic force. A
plot of the difference between the simulated and measured results over the tested
operating points is provided in Figure 7.5. Good agreement between the measured
and simulated response is observed with a maximum deviation at low air gaps and
CHAPTER 7. RESULTS
169
Figure 7.2: Experimentally measured opener force as a function of respective air gap
and steady state current
higher current excitations. This is most likely due to the armature deflecting and
thereby reducing the air gap (assumed rigid in the FEA studies). In fact, at air gaps
less than 0.5 mm, one portion of the armature could be heard contacting the pole
face thus making such force measurements erroneous. This was also observed during
similar measurements performed for a linear actuator in [Chladny, 2003]. During normal operation, deflection is likely less significant because the armature is not rigidly
constrained as done in the static tests. In [Clark et al., 2005], similar experimental
and simulated discrepancies are observed despite a 3D FEA model being conducted.
Regions beyond 6 mm and currents below 10 A provided insufficient force to be detectable by the load cells. Both the simulated and measured response indicate the
effect of material saturation as the force does not continue to increase quadratically
with respect to current as a linear induction model suggests. Experimental results
for the closer magnet are not available as it is not readily possible to make force
CHAPTER 7. RESULTS
170
Figure 7.3: FEA simulated closer force as a function of respective air gap and steady
state current
measurements in a non-destructive fashion. However, due to the similarity between
the opener and closer magnets, it is expected that the simulated model is also capable
of closely representing the actual closer response.
Provided the FEA models have sufficient fidelity, the numerical data can be used
directly in a LPM-FEA hybrid model or to parameterize a LPM through fitting techniques as discussed in Chapter 4. As shown in the following sections, these models
(which use steady state data) can predict transient performance with good accuracy
and with considerably less computational demands than a FEA.
7.2.2
Transient Performance Evaluation
Although good steady state model performance may be demonstrated, transient force
and current response to a time varying voltage input is also important for model based
control design. In particular, the LPMs so far have neglected the effects of eddy cur-
171
CHAPTER 7. RESULTS
1000
Measured
FEA
LPM
Spring
900
800
700
Force [N]
600
35 A
500
400
20 A
300
10 A
200
6A
100
2A
0
0
0.5
1
1.5
2
Air Gap [mm]
2.5
3
3.5
4
Figure 7.4: Opener magnet measured, FEA and LPM valve force as a function of
armature position and various steady state currents
rents. Although these effects cannot be directly measured, their effect on coil current
response is apparent (if significantly present) when comparing actual response to a
model that does not account for them [Chladny et al., 2005]. To investigate if eddy
current effects are significant to the actuator performance, experiments are conducted
as described in Chapter 6, Section 6.2.2.1 where a relatively short step voltage is applied and then reversed until current returns to zero over various armature positions.
Transient FEAs are also conducted using the measured voltage input for comparison
purposes. The FEA models are capable of predicting eddy current response using
the material conductivity specification for the laminate sheet of which the actuator is
constructed. Therefore the FEA models provide insight to how significant eddy currents are for the actuator considered (and if necessary, may be used to parameterize
a secondary eddy current model [Chladny et al., 2005]).
Figures 7.6 through 7.10 compare the measured, FEA, LPM-FEA hybrid and LPM
CHAPTER 7. RESULTS
172
Figure 7.5: Opener magnet simulated and measured force error as a function of
armature position and various steady state currents
current and force response at respective air gaps of 0.50, 0.75, 1.00, 1.50 and 2.00 mm.
Also indicated are measured and modeled voltage inputs as an approximation is used
for the modeled input to simplify the data input, particularly for the FEA. These
results indicate an overall good agreement between all models and experimental measurements with a 14% over prediction of peak current by the LPM-FEA hybrid model
and a 15% peak force under prediction by the LPM model, both of which occur at a
0.50 mm air gap position. In fact, all models performed poorest at the lower air gaps
tested. This is likely due to the armature deflecting enough to reduce the nominal
air gap where as it is assumed rigid in the models. If this were the case, a slower
current response and higher peak force should be measured as a result of the higher
inductance associated with a smaller air gap. Indeed, this is the case as agreement
173
CHAPTER 7. RESULTS
50
Voltage [V]
Measured
Model Input
0
−50
0
0.5
1
1.5
2
2.5
3
3.5
Current [A]
20
FEA
FEA / LPM
LPM
Measured
10
0
0
0.5
1
1.5
2
2.5
800
Force [N]
3
3.5
Measured: Load Cell 1
Measured: Load Cell 2
FEA
FEA / LPM
LPM
600
400
200
0
−200
0
0.5
1
1.5
2
2.5
3
3.5
Time [ms]
Figure 7.6: Simulated and measured response of the opener magnet, 42 V, 0.50mm
airgap
between peak current and force improves with larger air gaps (smaller forces and
hence less deflection). Also apparent is the effect of the transient loading on the load
cell and frame structure. As mentioned in Chapter 6, the signals from two load cells
are recorded. Load cell 1 is a lower capacity unit located between the load rod and
material testing machine crosshead. Load cell 2 is a higher capacity (and less sensitive) unit situated between the load crosshead and frame. Their relative location is
apparent with load cell 2 having a smoother response, likely due to the inertia of the
crosshead acting as a mechanical filter. Similarly, the relatively large mass of the cross
head causes a temporary tensile (negative) load upon polarity reversal. Despite these
force measurement complications, the general trend across the different operating
points suggests good agreement. Of greater concern is agreement with transient current response since the steady state measurements have already demonstrated good
force-current model performance. Overall model performance is as expected with a
174
CHAPTER 7. RESULTS
50
Voltage [V]
Measured
Model Input
0
−50
0
0.5
1
1.5
2
2.5
3
3.5
Current [A]
20
FEA
FEA / LPM
LPM
Measured
10
0
0
0.5
1
1.5
2
2.5
600
Force [N]
3
3.5
Measured: Load Cell 1
Measured: Load Cell 2
FEA
FEA / LPM
LPM
400
200
0
−200
0
0.5
1
1.5
2
2.5
3
3.5
Time [ms]
Figure 7.7: Simulated and measured response of the opener magnet, 42 V, 0.75mm
airgap
general degradation in performance with respect to model simplicity. However, the
degradation is relatively modest suggesting the assumption of negligible eddy currents
is valid. Although the FEA provides the best overall current and force prediction,
the LPM-FEA hybrid and LPM models take only seconds to simulate (as opposed
to hours). Despite the LPM exhibiting the lowest overall agreement, it is the only
model that is amenable to analytical control design techniques.
Although the FEA has proven instrumental in providing data for LPM development,
it is the only method capable of producing potentially revealing qualitative results.
Figures 7.11a through 7.11f illustrate opener flux lines over time for the same step
voltage input shown in Figure 7.9 corresponding to an air gap of 1.50 mm. These
figures provide insight to the nature of the fringing and leakage air paths, something
that may assist in the development of reluctance network models. Additionally, due
to the transient nature of the study, appreciation of the eddy current skin effect (see
175
CHAPTER 7. RESULTS
50
Voltage [V]
Measured
Model Input
0
Current [A]
−50
0
0.5
1
1.5
2
2.5
3.5
FEA
FEA / LPM
LPM
Measured
20
10
0
0
0.5
1
1.5
2
2.5
3
3.5
Measured: Load Cell 1
Measured: Load Cell 2
FEA
FEA / LPM
LPM
400
Force [N]
3
200
0
0
0.5
1
1.5
2
2.5
3
3.5
Time [ms]
Figure 7.8: Simulated and measured response of the opener magnet, 42 V, 1.00mm
airgap
Appendix A.5) can also be made. Compared to similar transient studies conducted
in [Chladny, 2003], the observed skin effect is relatively modest. Similar to the flux
lines, contour plots of magnetic flux density are as shown in Figure 7.12. Such plots
are likely quite useful to an actuator path designer as they provide insight as to where
the material is most likely to saturate first (and hence offer no further use to field
manipulation or intensification). For example in Figure 7.12b, it is apparent that saturation is most likely to occur first in the back iron path corners (likely exacerbated
by the eddy current skin effect) and later (Figure 7.12e) in the pole face corners (due
to the relatively small amount of material through which fringing is likely). Thus,
it may be possible to further enhance the magnetic efficiency of the device through
the addiction of small radii at the inside of the back iron legs (where the inner coil is
wound) likely at an increase of manufacturing costs or overall actuator volume and
mass.
176
CHAPTER 7. RESULTS
50
Voltage [V]
Measured
Model Input
0
−50
0
0.5
1
1.5
2
2.5
3
3.5
Current [A]
30
FEA
FEA / LPM
LPM
Measured
20
10
0
0
0.5
1
1.5
2
2.5
3
3.5
Force [N]
400
Measured: Load Cell 1
Measured: Load Cell 2
FEA
FEA / LPM
LPM
200
0
0
0.5
1
1.5
2
2.5
3
3.5
Time [ms]
Figure 7.9: Simulated and measured response of the opener magnet, 4 2V, 1.50mm
airgap
7.2.3
Preliminary Flux Sensor Evaluation
As a preliminary method of evaluating the flux sensor performance, the material
testing machine crosshead position controller is set to induce a sinusoidal armature
motion. During this time, a step load is input through regulation of coil current
about a predefined set-point as discussed in Section 6.2.2.3. In this way, insight may
be gained to armature position measurements by comparing the crosshead position
output and the reconstructed position based on flux and coil current measurements
using the technique in Section 5.6.1.
Figure 7.13 shows the results at a crosshead
frequency of 10 Hz and amplitude of 0.50 mm. Minimum air gap is set to 0.50 mm.
Similarly, Figure 7.14 provides results of a test conducted at a frequency of 4 Hz but
with an amplitude of 1.50 mm and a minimum air gap of 1.00 mm. In both tests,
reasonable agreement is observed between the reconstructed position and measured
177
CHAPTER 7. RESULTS
50
Voltage [V]
Measured
Model Input
0
Current [A]
−50
0
0.5
1
1.5
2
2.5
3.5
FEA
FEA / LPM
LPM
Measured
30
20
10
0
0
0.5
1
1.5
0.5
1
1.5
2
2.5
Measured:3Load Cell 1
Measured: Load Cell 2
FEA
FEA / LPM
LPM
3.5
2
2.5
3
3.5
300
Force [N]
3
200
100
0
−100
0
Time [ms]
Figure 7.10: Simulated and measured response of the opener magnet, 42 V, 2.00 mm
air gap
crosshead position. Aside from initial field build up (at t< 1.5 ms), a maximum
deviation of approximately 0.7mm is observed for the 10 Hz case. It is unclear if the
discrepancy is a result of the reconstruction technique or an actual deviation as a result
of armature or load frame dynamics. The time required to establish a field sufficiently
strong enough to make an accurate measurement is significant and suggests that a
bias current will be required, particularly at low air gaps. Although these experiments
serve only as a preliminary indication that it is possible to reconstruct position from
flux and current measurements, they are not overly useful in that even at 10Hz, the
motion is relatively slow compared to a normal valve opening or closing cycle (which
typically occurs within 5 ms). Recording time is limited to approximately 70 ms due
to concerns of saturating the op-amp integration circuit and damaging the coil.
178
CHAPTER 7. RESULTS
(a) t=0.2 ms
(b) t=0.4 ms
(c) t=0.6 ms
(d) t=0.8 ms
(e) t=1.0 ms
(f) t=1.2 ms
Figure 7.11: Opener flux lines for a 1.50 mm air gap and 42 V step input at: 0.2, 0.4,
0.6, 0.8, 1.0 and 1.2 ms
179
CHAPTER 7. RESULTS
(a) t=0.2 ms
(b) t=0.4 ms
(c) t=0.6 ms
(d) t=0.8 ms
(e) t=1.0 ms
(f) t=1.2 ms
Figure 7.12: Opener flux density contour plots for a 1.50 mm air gap and 42 V step
input at: 0.2, 0.4, 0.6, 0.8, 1.0 and 1.2 ms
10
1.5
7.5
Air Gap [mm]
2
1
5
0.5
Measured Current [A]
180
CHAPTER 7. RESULTS
2.5
Estimated Position
Measured Position
Measured Current
0
0
10
20
30
40
50
60
0
70
Time [ms]
Figure 7.13: Measured and estimated air gap during 10 Hz, 0.50 mm amplitude,
1.00 mm mean crosshead motion
7.3
Simulated and Measured Controller Performance
Upon validation of the various magnetic models, the LPM-FEA hybrid model is
extended to include spring-mass and gas force dynamics as described in Section 4.6.
The model is instrumental in the development of a comprehensive control strategy
capable of addressing the issues of unknown gas force disturbances and soft seating
while subject to a 42 V source and production amenable feedback sensors. The
model is validated through comparison with relatively simple open-loop testbench
experiments. Upon successful simulated controller performance, the developed control
software is coded in C and compiled on a dSPACE DS1103 controller. The following
sections provide an overview of the simulated and measured testbench experiments.
Air Gap [mm]
2.5
10
2
7.5
1.5
5
1
Measured Current [A]
181
CHAPTER 7. RESULTS
2.5
Estimated Position
Measured Position
Measured Current
0.5
0
10
20
30
40
50
60
0
70
Time [ms]
Figure 7.14: Measured and estimated air gap during 4 Hz, 1.50 mm amplitude, 2.5 mm
mean crosshead motion
7.4
Testbench System Model Validation
Prior to feedback control implementation on the designed testbench (described in Section 6.3), a model of the complete actuator system is developed using the MATLABSimulink environment and acquired FEA data. The model accounts for bridge drive
circuits, mechanical and electromagnetic response. The model also approximates exhaust gas forces as a means to evaluate feedforward requirements under a variety of
blowdown pressures. The model is then validated through comparison to increasingly sophisticated testbench experiments. For example, Figure 7.15 contrasts the
simulated and experimental results of a pressurized release where the closer holding
current is simply switched off and the cavity depressurization and valve oscillation
are observed. Later, when a landing controller is implemented, the pressure trace
and digital outputs from the dSPACE controller are recorded and used as the model
182
CHAPTER 7. RESULTS
x 10
2
0
−2
−4
0
Measured (Valve)
Simulated
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
0.008
0.01
0.012
0.014
0.016
0.018
0.02
4
2
−2
Measured (Valve)
Simulated
−4
0
0.002
0.004
0.006
1
Measured Pressure
Gas Force
0.5
0
0
60
30
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
0.018
0.02
Simulated Gas Force [N]
0
Coil Current [A]
Gas Pressure [Bar]
Valve Velocity [m/s]
Valve Position [m]
−3
4
6
Closer − Measured
Closer − Simulated
4
2
0
0
0.002
0.004
0.006
0.008
0.01
0.012
Time [s]
0.014
0.016
0.018
0.02
Figure 7.15: Simulated LPM-FEA and experimental free opening, 1 bar blowdown
pressure
input (open-loop simulation) as shown in Figures 7.16 and 7.17. Here, gas force is
simulated as valve area times cavity pressure and gas force coefficient, Fg = Av P Cgf .
The simulation results suggest reasonable agreement given the simple gas force model
and the lack of a controller in the simulations. Specifically, a maximum of 1.5% displacement and velocity peak error among the various simulations is observed. The
primary discrepancy is due to a 0.25 ms delay (maximum) between the measured and
simulated results. This may be due to static friction force that is unaccounted for in
the model. Given that these results are obtained simply through using the same ini-
183
CHAPTER 7. RESULTS
x 10
Measured (Valve)
Simulated
2
0
−2
−4
0
2
4
6
8
−3
x 10
Measured (Valve)
Simulated
4
2
0
−2
0
2
4
6
8
Gas Pressure [Bar]
−3
x 10
1
Measured Pressure
Gas Force
0.5
0
Coil Current [A]
00
60
30
0
2
4
2
4
Time [ms]
15
10
6 Closer − Measured 8
−3
Closer − Simulated
x 10
Opener − Measured
Opener − Simulated
Simulated Gas Force [N]
Valve Velocity [m/s]
Valve Position [m]
−3
4
5
0
0
6
8
Figure 7.16: Simulated LPM-FEA and experimental valve opening, 1 bar blowdown
pressure
tial conditions and experimental switch signals (simulation is thus run in open-loop)
the model is expected to be valid for controller design purposes.
7.4.1
Open-loop Feedforward Control
Initially, only tuned open-loop feedforward controllers are used to setup appropriate
initial conditions for the flatness-based landing controller. Figure 7.18 demonstrates
typical landing performance (opening and closing) over a pressure range of 0.25 to 4.5
bar using the flux-based position and velocity estimation and open-loop feedforward
184
CHAPTER 7. RESULTS
Valve Position [m]
−3
4
Measured (Valve)
Simulated
2
0
−2
Valve Velocity [m/s]
−4
0
1
2
3
4
5
6
7
8
−3
x 10
0
−2
Measured (Valve)
Simulated
1
2
3
4
5
6
7
x 10 10
0.5
0.25
0
0
25
20
15
10
5
0
0
8
−3
Measured Pressure
Gas Force
1
2
5
3
4
5
6
0
8
7
−3
Simulated Gas Force [N]
−4
0
Coil Current [A]
Gas Pressure [Bar]
x 10
x 10
Closer − Measured
Closer − Simulated
Opener − Measured
Opener − Simulated
1
2
3
4
Time [ms]
5
6
7
8
−3
x 10
Figure 7.17: Simulated LPM-FEA and experimental valve closing, 1 bar blowdown
pressure
method. It is found that with tuning at each pressure level, landing velocities of
≤ 0.1m/s are consistently attainable. Figure 7.19 illustrates similar performance
is attainable at higher test-bench speeds as well. However, at speeds greater than
3000rpm, exhaust pressures are limited to approximately 1 bar due to compressed air
supply restriction on the testbench.
Although the manually tuned method appears acceptable, potentially large impacts
and / or actuation failure can result with even modest (±0.5 bar) changes in opening
pressure.
185
CHAPTER 7. RESULTS
−3
x 10
Position [m]
0
−2
−4
Increasing Pressure
−6
−8
0
10
20
30
Velocity [m/s]
5
40
Time [ms]
50
60
70
80
40
Time [ms]
50
60
70
80
0
Increasing Pressure
Pressure [Bar]
−5
0
10
30
4
2
0
0
Velocity [m/s]
20
−1
−2
−3
−4
Position [m]
−5
−6
−7
−3
x 10
5
Increasing Pressure
0
−5
−8
0
−1
−2
−3
−4
Position [m]
−5
−6
−7
−8
−3
x 10
Figure 7.18: Full cycle plots over a 0.25 to 4.5 bar pressure range, 250 rpm
7.4.2
Observer Convergence
Since velocity and gas pressure measurements are unavailable for control, they are
estimated through the observer discussed in Section 5.7. The observer structure
is first simulated and tuned independently of a feedback controller using measured
position data for simple testbench experiments as shown in 7.15. Later, the observer
is combined with a feedforward and landing controller first in simulation and then
coded and re-tuned on the testbench hardware.
Figures 7.20 and 7.21 indicate
typical position, velocity and gas pressure estimates at respective EVO pressures of
186
CHAPTER 7. RESULTS
−3
x 10
Position [m]
0
−2
−4
−6
−8
0
0.002
0.004
0.006
Time [ms]
0.008
0.01
0.012
0
0.002
0.004
0.006
Time [ms]
0.008
0.01
0.012
Velocity [m/s]
5
0
Pressure [Bar]
−5
1
0.5
0
0
−1
−2
−3
−4
Position [m]
−5
−4
Position [m]
−5
−6
−7
−3
5
Velocity [m/s]
−8
x 10
0
−5
0
−1
−2
−3
−6
−7
−8
−3
x 10
Figure 7.19: Flatness voltage landing control via flux feedback - 3500 rpm, 1 bar
blowdown, 6 cycles
1 and 5 bar. Although gas force is actually used in the feedforward and feedback
routines, the force estimate is converted to a pressure for direct comparison with
measured pressure (since valve gas force is not measured directly). Also shown is the
measured flux-based position signal in which all estimates (with current) are made.
Convergence performance in the case of the simulated results is significantly better
than the measured cases. This is largely due to a relatively noisy flux-based position
signal, particularly at the early stages of valve opening where a change in armature
position has only a moderate influence on path flux. This problem is intensified at
187
CHAPTER 7. RESULTS
Position [m]
4
Simulated
Laser
Flux−based
Estimated
2
0
−2
Velocity [m/s]
−4
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Simulated
Simulated Estimate
Measured
Estimated
4
2
0
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Pressure [bar]
3
Simulated
Simulated Estimate
Measured
Estimated
2
1
0
0
0.5
1
1.5
2
2.5
Time [ms]
3
3.5
4
4.5
5
Figure 7.20: Simulated and experimental estimated position, velocity and pressure
convergence, 1 bar
larger EVO pressures as the valve opens so slowly that a opener bias current is imposed
to maintain a position signal (albeit a relatively poor one). In the later half of the
valve opening stroke, the flux-based signal improves as do the respective estimates.
Despite the degraded estimation accuracy in the early opening stages, good control
performance is still possible in part because of the way in which the feedforward
controller is designed. Specifically, consideration is given to the low magnetic force
authority at distances of 4 mm away or greater from the pole face so that a relatively
low (and thus less noisy) feedforward current is demanded during the early stages of
valve opening. Figure 7.22 indicates desired and actual feedforward currents for EVOs
of 1 and 5 bar for both the simulated and measured cases. Upon release, a constant
188
CHAPTER 7. RESULTS
Position [m]
4
2
Simulated
Laser
Flux−based
Estimated
0
−2
−4
0
1
2
3
4
5
7
Simulated
Simulated Estimate
Measured
Estimated
3
Velocity [m/s]
6
2
1
0
0
1
2
3
4
5
6
7
Pressure [Bar]
6
Simulated
Simulated Estimate
Measured
Estimated
4
2
0
0
1
2
3
4
5
6
7
Time [ms]
Figure 7.21: Simulated and experimental estimated position, velocity and pressure
convergence, 5 bar EVO
bias current is applied in anticipation the requirement of a position measurement.
7.4.3
Gas Disturbance Rejection
Plots of simulated and measured position, velocity and gas pressure are shown in
Figure 7.23. Also shown are individual gas pressure estimates. In this figure, eight
consecutive cycles are shown where initial pressure is alternated between 1 and 5 bar to
demonstrate the observer and feedforward controller’s ability to compensate for large
cycle-to-cycle pressure variations. In both simulation and experiment, no changes
are made to the controller parameters, gains or initial gas force estimates between
cycles. Similar results are obtained for random initial pressure tests. Convergence
189
CHAPTER 7. RESULTS
Position [m]
4
2
0
Simulated
Measured
−2
−4
0
1
2
3
4
5
6
7
Pressure [bar]
6
Simulated
Simulated Estimate
Measured
Estimated
4
2
0
−2
0
1
2
3
4
5
Current [A]
Landing
Control
Engagement
10
0
0
1
2
3
4
7
Simulated − Closer
Simulated − Opener
Simulated − Desired
Measured − Closer
Measured − Opener
Desired
30
20
6
5
6
7
Time [ms]
Figure 7.22: Simulated and experimental position, pressure, current and desired feedforward current at 1 and 5 bar EVO
of γ is provided for the same cases, this time measured gas pressure is converted
into an equivalent initial pressure by dividing measured pressure by function f1 (x).
Overall, there is good agreement between simulated and experimental results. In
simulation, gas force estimation is initially very good with rapid convergence in the
first two millimeters of lift. However, deviation from the true value is observed in
the later portion of the stroke, likely due to model-plant mismatch bias and weak
observability. Gas pressure convergence is not as rapid in the experimental case due
to a poorer quality position signal (noise is not modeled in the simulation) and thus
lower gains are assigned. Position signal quality is poorer at high pressures as the
slower valve motion required the need to switch from the closer position coil (zero
CHAPTER 7. RESULTS
190
current) to the opener early on in the valve stroke (where the flux-based position
performance is most sensitive to noise). Despite this, convergence is rapid enough
to provide adequate feedforward current and the overall experimental performance is
very good. The phase plane plot in Figure 7.23 indicates a maximum opening impact
speed of 0.17 m/s in one of the 5 bar cycles. Such results are excellent considering
pressure variations from 1 to 5 bar are being applied alternatively for each of the
eight consecutive cycles to approximate an engine misfire.
7.4.4
Simulated and Measured Control Performance
The flux-based feedback sensor, observer, feedforward and landing controllers are implemented to provide a comprehensive valve actuation strategy. The routines are first
tested using the detailed actuator and power electronics Simulink model and then implemented in real-time on the testbench. Upon demonstration of successful landing
control at predefined EVO pressures, the proposed disturbance estimation and energybased feedforward method is implemented. Simulated and experimental performance
results of these tests (again using flux-based feedback) are shown in Figure 7.25 with
respect to position, velocity, pressure and coil current. These results illustrate the
comprehensive control performance in simulation and in experiment over a relatively
wide range of exhaust gas pressures. The effectiveness of the combined control strategy is experimentally demonstrated with an average impact velocity below 0.1 m/s
with a standard deviation of 0.06m/s (over two hundred cycles). Transition times
(defined by the time from when the control command is given to open to the point of
impact) vary significantly with pressure with typical maximum and minimum times
of 6.2 ms and 4.1ms for the 5 bar and 1 bar cases respectively. If transition time is
calculated as the time taken for the valve to move from the 98% closed to 98% open
position (-3.84mm to 3.84mm) [Peterson, 2005], then typical maximum and minimum
times are 4.5ms and 3.0ms for the respective 5 and 1 bar cases.
CHAPTER 7. RESULTS
191
To be sure acceptable landing performance is consistent and repeatable, impact speeds
are recorded from 200 cycles at three pressure levels. Due to data acquisition memory limitations, the recording time is limited to two seconds. As a result, the cycles
that occur during the time to download and save a two second interval (at most, 30
seconds) are not recorded. The engine speed (emulated) of the tests determines how
many recording intervals are required to total 200 valve cycles. Again, for higher
pressure EVOs, a slower speed is required to accommodate compressed air flow rate
limitations. Figures 7.26 through 7.28 show the opener and closer impact velocities
over 200 recorded cycles at 1, 3 and 5 bar EVO pressures, respectively. In all cases,
an average impact velocity of 0.1 m/s with an approximate standard deviation of
0.05 m/s are achieved with exception of the opener at 5 bar EVO pressure. Here, a
slightly higher average impact velocity is obtained with a significantly higher average
impact speed of 0.104 m/s and standard deviation of 0.09 m/s is observed. Also worthy of note is the difference in distribution between the 5 bar EVO case and all others.
The 5 bar opener distribution is somewhat polarized between relatively low impacts
and higher impacts while the other cases conform closer to a normal distribution.
It is suspected that this is a result of heightened sensitivity to variations in initial
conditions and estimation errors for the higher pressure cases. Or in other words, the
controller performance tends to either be exceptionally good, or in the event of even
moderate initial condition variations, higher impacts are observed.
7.5
Simulated Multi-cylinder Exhaust Manifold Pressure Disturbances
The gas disturbances considered thus far have been highly idealized to simplify the
estimation process while attempting to capture the main dynamic effect. It is acknowledged that multi-cylinder exhaust manifold pressure variations can be significant and difficult to measure and predict [Macián et al., 2004]. It is suspected that
CHAPTER 7. RESULTS
192
the proposed disturbance observer will be able to accommodate large variations in
back pressure provided they are slowly varying with respect to a valve event. However, if a relatively large pressure disturbance occurs mid-stroke of valve opening, it is
unknown how experimental landing performance may be affected. Several simulations
are provided to lend insight to performance under such conditions in Figure 7.29. The
simulations suggest that although performance is compromised with respect to impact
velocity and opening time, such disturbances may still be manageable (particularly
given the abrupt timing and magnitude simulated). One conceivable extension of the
testbench apparatus would be to inject volumes of compressed air into the exhaust
stream to investigate the subsequent control performance subject to exhaust manifold
variations.
7.6
Summary
The previous sections summarize and highlight the key results of the modeling and
control of a prototype solenoid valve actuator. The proposed modeling techniques
have been validated through quasi-static and testbench experiments. Feedback and
state estimates are made through a magnetic flux signal measurement and estimator.
This technique is sufficient for implementation in a production under-hood environment and cost effective compared to eddy current, LVDT and laser based sensors of
sufficient resolution and bandwidth. Using a disturbance observer and feedforward
controller, large EVO pressure variations may be identified and rejected on a cycle-bycycle basis ensuring soft seating over a wider range of operating modes. Additionally,
the flatness-based landing controller provides suitable tracking over a wide range of
initial conditions while restricted to a 42 V source.
193
CHAPTER 7. RESULTS
Position [m]
5
Cycles 1, 3, 5
and 7
0
−5
0
Simulated
Experiment
Cycles 2, 4, 6
and 8
1
2
3
4
Time [ms]
5
6
7
8
1
2
3
4
Time [ms]
5
6
7
8
4
Velocity [m/s]
3
2
1
0
Pressure [Bar]
−1
0
Estimated: Simulation
Estimated: Experiment
4
2
0
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
Velocity [m/s]
4
2
0
−2
−4
Figure 7.23: Simulated and measured opening cycles with cyclic pressure variations
between 1 and 5 bar
194
CHAPTER 7. RESULTS
Position [m]
5
0
Simulated
Experiment
−5
0
1
2
3
4
Time [ms]
5
6
7
8
1
2
3
4
Time [ms]
5
6
7
8
4
Velocity [m/s]
3
2
1
0
−1
0
Estimated: Simulation
Estimated: Experiment
γ [bar]
6
4
2
0
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
−4
−3
−2
−1
0
Position [mm]
1
2
3
4
Velocity [m/s]
4
2
0
−2
−4
Figure 7.24: Simulated and measured opening cycles with cyclic pressure variations
between 1 and 5 bar
195
CHAPTER 7. RESULTS
Position [mm]
4
2
0
Dashed line: Simulated
Solid line: Experimental
−2
Velocity [m/s]
−4
0
Pressure [bar]
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
4
5
6
7
4
Increasing pressure
2
0
0
Current [A]
1.0 bar
2.0 bar
3.0 bar
4.0 bar
5.0 bar
Increasing pressure
4
2
0
0
30
Opener
currents
Closer currents
20
10
0
0
1
2
3
Time [ms]
Figure 7.25: Simulated and experimental feedforward and landing results for 1 to 5
bar EVO pressures using flux-based feedback and observer disturbance rejection
196
CHAPTER 7. RESULTS
Opener
Closer
35
Opener Average impact
velocity = 0.098 m/s
Opener Std. Dev = 0.052 m/s
Closer Average impact
velocity = 0.091 m/s
Closer Std. Dev = 0.055 m/s
30
Frequency
25
20
15
10
5
0
0
0.05
0.1
Impact Velocity [m/s]
0.15
0.2
Figure 7.26: Experimental impact velocity histogram at 1 bar, 200 cycles
197
CHAPTER 7. RESULTS
Opener
Closer
45
Opener Average impact
velocity = 0.096 m/s
Opener Std. Dev = 0.044 m/s
Closer Average impact
velocity = 0.099 m/s
Closer Std. Dev = 0.065 m/s
40
35
Frequency
30
25
20
15
10
5
0
0
0.05
0.1
0.15
Impact Velocity [m/s]
0.2
Figure 7.27: Experimental impact velocity histogram at 3 bar, 200 cycles
0.25
198
CHAPTER 7. RESULTS
Opener
Closer
60
Opener Average impact velocity = 0.104 m/s
Opener Std. Dev = 0.094 m/s
Closer Average impact velocity = 0.091 m/s
Closer Std. Dev = 0.050 m/s
50
Frequency
40
30
20
10
0
0
0.05
0.1
0.15
0.2
Impact Velocity [m/s]
0.25
0.3
Figure 7.28: Experimental impact velocity histogram at 5 bar, 200 cycles
199
CHAPTER 7. RESULTS
Position [m]
4
2
Nominal
Moderate Disturbance
Large Disturbance
0
−2
−4
0
1
2
3
4
5
6
7
8
9
1
2
3
4
5
6
7
8
9
Velocity [m/s]
4
2
0
0
Pressure [Bar]
5
4
Large
3
2
1
Nominal
0
0
Opener Current [A]
Solid lines: Simulated
Dashed lines: Estimated
Moderate
1
2
3
4
1
2
3
4
5
6
7
8
9
5
6
7
8
9
20
15
10
5
0
0
Time [ms]
Figure 7.29: Simulated 1.5 bar disturbance response with secondary mid-stroke pressure disturbances
Chapter 8
Conclusions and Further Research
D
ue to the many stringent constraints and requirements inherent to using solenoid
variable valve timing actuators, the design, control and implementation of the
devices has proven to be exceptionally challenging. The multi-disciplinary nature
of the problem has also likely impeded development. Without considering the myriad challenging issues already facing any emerging automotive technology such as
ecconomics, mass production, environmental impact and safety, the successful implementation of solenoid gas exchange valve actuators involves insights of electricity,
magnetism, heat transfer, vibrations, fluid mechanics and control system theory. The
preceding work presents a comprehensive modeling and control strategy for such actuators that accounts for the electrical, mechanical and nonlinear magnetic aspects
of the actuators system. The control method is based on a foundation of extensive and accurate simulations as a means of maximizing control performance while
simultaneously reducing the debugging, tuning and risk associated with hardware
implementation. Specifically, 2D FEA models have been developed to characterize
actuator magnetic response and are used to parameterize lower-order lumped parameter models that are amenable for control design. All models have been validated
with equivalent experiments. Through careful analysis of simulated performance, a
control strategy has been designed that is capable of achieving the low impact velocity
200
CHAPTER 8. CONCLUSIONS AND FURTHER RESEARCH
201
constraint while rejecting variable gas force disturbances and being subject to realistic power source and feedback sensing techniques. Based on this work, the following
conclusions may be made:
1. Sufficiently accurate FEA models of the closer and opener magnets are developed to parameterize lower-order (albeit still nonlinear) dynamic lumped
parameter models that account for magnetic saturation effects and gas force
disturbances. All models are validated using experimental techniques of increasing complexity and operational realism.
2. As a means of simulating the dynamic response of the actuator system, the
electromagnetic FEA results are directly incorporated with a mechanical LPM
into a modular simulation environment. This LPM-FEA hybrid model provides
a means of investigating the effects of external parameters on the overall actuator performance while avoiding the computational demands of a FEA. For
example, the model is extended to include the influence of combustion gas forces
and switched voltage waveforms. This high fidelity model is capable of predicting an entire valve actuation cycle making it instrumental in development of a
comprehensive control strategy.
3. To simplify initial control hardware implementation, a testbench engine emulator was designed and fabricated. The apparatus incorporates a productionrealistic cylinder head and compressed air cavity that may be electronically
pressure regulated to produce transient gas force disturbances consistent with
an operating engine. This unique method of testing provides an intermediate
step to single cylinder engine experiments that allows for tuning and debugging in a controlled environment without the additional complications or risk
of working within a research engine test cell.
CHAPTER 8. CONCLUSIONS AND FURTHER RESEARCH
202
4. A method for position feedback through magnetic flux and current measurements is presented and demonstrated sufficient performance for use with an
estimator and closed-loop landing controller.
5. A one dimensional compressible flow model is further simplified as a strictly
valve position dependent phenomenon. In doing so, it may be used in a nonlinear observer capable of cycle-by-cycle gas force disturbance estimation. Once an
estimate of gas force magnitude is available, it may be compensated for through
an energy recovery based feedforward method. The combined estimation and
feedforward technique are demonstrated on a custom testbench experiment to
provide suitable initial landing control conditions for a closed-loop landing controller to successfully open an exhaust actuator.
6. A flatness-based closed loop landing controller is demonstrated both in simulation and experiment to exponentially track predefined reference trajectories.
The trajectories are parameterized using B-spline basis functions and optimized
subject to practical motion and voltage constraints. Through simulation, the
flatness-controller is shown to exhibit superior performance with respect to impact speed, transition time, parameter variations and input saturation when
compared to LTI and PI based controllers.
7. Both the feedforward and nonlinear landing controllers are successfully demonstrated while subject to subject to practical physical constraints and gas force
disturbances using the aforementioned feedback sensing and disturbance identification techniques.
8. The method of disturbance characterization, identification and control techniques are sufficiently generic to be applied to other valve actuator types.
CHAPTER 8. CONCLUSIONS AND FURTHER RESEARCH
8.1
203
Further Research
In order to further enhance the proposed control process and continue on the path of
engine implementation, the following highlights possible areas of future work.
8.1.1
Engine Implementation
1. Evaluation of the proposed control scheme on an operating single cylinder research engine. Work includes coordination of the intake and exhaust valve
controllers with the engine control unit, online diagnostics and error handling
and further performance evaluation.
8.1.2
Control Development
1. To address the issue of compensating for slowly varying system parameters (such
as friction and spring stiffness) a method of online parameter estimation could
be developed. The disturbance observer presented is not expected to necessarily
compensate for such changes due to the way such parameters interact with the
flatness based landing control law.
2. Provided sufficient computational resources are available, the presented control
strategy could be augmented with an adaptive scheme that adjusts feedback
gains, open-loop feedforward control parameters (closer) or compensates for
slowly time varying model parameters, such as changes in viscosity.
3. The landing controller used in this study tracks only one set of predefined trajectories. Improved landing performance and efficiency could likely be achieved
if a wider variety of trajectories were available or generated recursively online,
again subject to available computational resources.
CHAPTER 8. CONCLUSIONS AND FURTHER RESEARCH
8.1.3
204
Alternative Actuators
1. With recent advancements of motor driven and hydraulic valve actuators, a
survey and categorization of various potential prototype candidates could provide decisive insight for which technology is most likely to prove superior with
respect to efficiency, manufacturing / implementation cost and performance.
Bibliography
[A. Warburton et al., 2005] A. Warburton, L. F., Scott, J., Butler, N., and Wygnanki, W. (2005). Intelligent Valve Actuation (IVA) System for Gasoline and
Diesel Engines. SAE paper 2005-01-0772.
[Alexander, 2006] Alexander, D. (2006). Valeo pushes engine valves. SAE Automotive
Engineering International, page 42.
[Allen and Law, 2002] Allen, J. and Law, D. (2002). Production electro-hydraulic
variable valve-train for a new generation of I.C. engines. SAE paper 2002-01-1109.
[Amato and Meuller-Heiss, 2001] Amato, G. and Meuller-Heiss, M. (2001). Power
stage partitioning for E-VALVE applications. SAE paper 2001-01-0239.
[ANSYS Inc., 2005] ANSYS Inc. (2005). Ansys theory reference. Release 10.
[Ashley, 2003] Ashley, S. (2003). Artificial muscles. Scientific American.
[Atkins and Koch, 2003] Atkins, M. and Koch, C. (2003). A well-to-wheel comparison of several powertrain technologies. SAE paper 2003-01-0081.
[Barros da Cunha et al., 2000] Barros da Cunha, S., Hedrick, J., and Pisano, A.
(2000). Variable valve timing by means of a hydraulic actuation. SAE paper
2000-01-1220.
[Blair, 1999] Blair, G. (1999). Design and Simulation of Four-Stroke Engines. Society
of Automotive Engineers.
205
BIBLIOGRAPHY
206
[Boor, 1978] Boor, C. D. (1978). A Practical Guide to Splines. Springer-Verlag, New
York.
[Borgmann et al., 2004] Borgmann, K., Frhlich, K., Hall, W., Hofmann, R., Melcher,
T., and Bock, C. (2004). The High-Performance Powertrain of the X5 4.8is. SAE
paper 2004-08-0100.
[Braune et al., 2006] Braune, S., Liu, S., and Mercorelli, P. (2006). Design and control
of an electromagnetic valve actuator. 2006 IEEE CCA/CACSD/ISIC Conference,
pages 1657–1662.
[Butzmann and Melbert, 2003] Butzmann, S. and Melbert, J. (2003). Method for
controlling an electromechanical actuator. US Patent 6,648,297.
[Butzmann et al., 2000] Butzmann, S., Melbert, J., and Koch, A. (2000). Sensorless
control of electromagnetic actuators for variable valve train. SAE paper 2000-011225.
[C. R. Ferguson, 2000] C. R. Ferguson, A. T. K. (2000). Internal Combustion Engines. John Wiley and Sons, Inc.
[Canada, 2005] Canada, E. (2005). Canada’s Greenhouse Gas Inventory: 1990-2003.
Technical report.
[Çengel and Boles, 1993] Çengel, Y. and Boles, M. (1993). Thermodynamics: An
Engineering Approach. McGraw-Hill, 2nd edition.
[Chang et al., 2002] Chang, W. S., Parlikar, T. A., Seeman, M. D., Perreault, D. J.,
Kassakian, J. G., and Keim, T. A. (October 2002). A new electromagnetic valve
actuator. IEEE Workshop on Power Electronics in Transportation, Auburn Hills,
MI, pages 109–118.
BIBLIOGRAPHY
207
[Chaniotakis and Cory, 2006] Chaniotakis, M. and Cory, D. (2006). Op amp (cont.);
active filters; superdiode, log, antilog filters. In Course 6.071J, Introduction to
Electronics, Signals, and Measurement, number 30 in Op Amp Lecture Notes.
Massachusetts Institute of Technology OpenCourseWare.
[Chen et al., 2005] Chen, L., Mercorelli, P., and Liu, S. (2005). A Kalman Estimator
for Detecting Repetitive Disturbances. 2005 American Control Conference, pages
1631–1636.
[Chillet and Voyant, 2001] Chillet, C. and Voyant, J. (2001). Design-oriented analytical study of a linear electromagnetic acutator by means of a reluctance network.
IEEE Trans. Magn., 37:3004–3011.
[Chladny, 2003] Chladny, R. (2003). Modeling and simulation of automotive gas
exchange valve solenoid actuators. M.Sc. thesis, Dept. Mechanical Engineering,
University of Alberta.
[Chladny and Koch, 2006a] Chladny, R. and Koch, C. (2006a). A Magnetic FluxBased Position Sensor for Control of an Electromechanical VVT Actuator. 2006
American Control Conference, pages 3979–3984.
[Chladny and Koch, 2007] Chladny, R. and Koch, C. (2007). Flatness-Based Tracking of an Electromechanical VVT Actuator with Disturbance Observer FeedForward Compensation. Manuscript accepted from publication in IEEE Trans.
Contr. Syst. Technol.
[Chladny and Koch, 2006b] Chladny, R. and Koch, C. (October 2006b). FlatnessBased Tracking of an Electromechanical VVT Actuator with Magnetic Flux Sensor.
2006 IEEE CCA/CACSD/ISIC Conference, pages 1663–1668.
208
BIBLIOGRAPHY
[Chladny et al., 2005] Chladny, R., Koch, C., and Lynch, A. (2005). Modeling automotive gas-exchange solenoid valve actuators. IEEE Trans. Magn., 41:1155–1162.
[Chung, 2005] Chung, S. (2005). Flatness-based voltage end control of a gas exchange
solenoid actuator for IC engines. M.Sc. thesis, Dept. Mechanical Engineering,
University of Alberta.
[Chung et al., 2007] Chung, S., Koch, C., and Lynch, A. (2007). Flatness-based feedback control of an automotive solenoid valve. To be published in IEEE Trans. Contr.
Syst. Technol.
[Clark et al., 2005] Clark, R., Jewell, G., Forrest, S., Rens, J., and Maerky, C. (2005).
Design Features for Enhancing the Performance of Electromagnetic Valve Actuation Systems. IEEE Trans. Magn., 41:1163–1168.
[Cope and Wright, 2006] Cope, D. and Wright, A. (2006).
Electromagnetic Fully
Flexible Valve Actuator. SAE paper 2006-01-0044.
[Corless et al., 1996] Corless, R., Gonnet, G., Hare, D., Jeffrey, D., and Knuth, D.
(1996). On the Lambert W function. Advances in Computational Mathematics,
5:329–359.
[Deng and Nehl, 1998] Deng, F. and Nehl, T. W. (1998). Analytical modeling of
eddy-current losses caused by pulse-width-modulation switching in permanentmagnet brushless direct-current motors.
IEEE Transactions on Magnetics,
34(5):3728–3736.
[dSPACE GmBH, 2003] dSPACE GmBH (2003). DS1103 PPC Controller Board Features Release 4.0.
BIBLIOGRAPHY
209
[Dugdale et al., 2005] Dugdale, P., Rademacher, R., Price, B., Subhedar, J., and
Duguay, R. (2005). Ecotec 2.4L VVT: A Variant of GMs Global 4-Cylinder Engine.
SAE paper 2005-01-1941.
[Energy Information Agency (EIA), 2006] Energy Information Agency (EIA) (2006).
Global Oil Consumption. http://www.eia.doe.gov.
[Euler, 1779] Euler, L. (1989 (orig. date 1779)). De serie lambertina plurismique eius
insignibus proprietatibus. Leonhardi Euleri Opera Omnia, Ser. 1 Opera Mathematica, Bd 6.
[Eyabi, 2003] Eyabi, P. (2003). Modeling and sensorless control of solenoid actuators.
Ph.D. thesis, Dept. Mechanical Engineering, Ohio State University.
[Eyabi and Washington, 2006a] Eyabi, P. and Washington, G. (2006a). Modeling and
sensorless control of an electromagnetic valve actuator. Mechatronics.
[Eyabi and Washington, 2006b] Eyabi, P. and Washington, G. (2006b). Nonlinear
Modeling of an Electromagnetic Valve Actuator. SAE paper 2006-01-0043.
[Flierl and Klüting, 2000] Flierl, R. and Klüting, M. (2000). The third generation
of valvetrains - New fully variable valvetrains for throttle-free load control. SAE
paper 2000-01-1227.
[Fliess et al., 1992] Fliess, M., Lévine, J., Martin, P., and Rouchon, P. (1992). On
differentially flat nonlinear systems. Proceedings of the IFAC-Symposium NOLCOS
1992, Bordeaus, France, pages 408–412.
[Fliess et al., 1994] Fliess, M., Levine, J., Martin, P., and Rouchon, P. (1994). Nonlinear control and Lie-Backlund transformation: Toward a new differential standpoint.
Proc. 33rd IEEE Conf. Decision and Control, pages 339–344.
BIBLIOGRAPHY
210
[Fliess et al., 1995] Fliess, M., Lévine, J., Martin, P., and Rouchon, P. (1995). Flatness and defect of non-linear systems: Introductory theory and examples. Int. J.
Control, 61(6):1327–1361.
[Fliess et al., 1999] Fliess, M., Lévine, J., Martin, P., and Rouchon, P. (1999). A LieBäcklund approach to equivalence and flatness of nonlinear systems. IEEE Trans.
Auto. Contr., 44(5):922–937.
[Franklin et al., 1998] Franklin, G., Powell, D., and Workman, M. (1998). Digital
Control of Dynamic Systems. Addison Wesley.
[Gecim, 1993] Gecim, B. (1993). Analysis of a lost-motion-type hydraulic system for
variable valve actuation. SAE paper 930822.
[Giglio et al., 2002] Giglio, V., Iorio, B., Police, G., and di Gaeta, A. (2002). Analysis
of advantages and of problems of electromechanical valve actuators. SAE paper
2002-01-1105.
[Gill et al., 1981] Gill, P. E., Murray, W., and Wright, M. H. (1981). Practical Optimization. Academic Press, London.
[Gladel et al., 1999] Gladel, P., Kreitmann, F., Meister, O., Stolk, T., and Gaisberg,
A. V. (1999). Aktor zur elektromagnetischen Ventilsteuerung. German Patent
Application DE19961608 A1.
[Golovatai-Schmidt et al., 2004] Golovatai-Schmidt, E., Schwarzenthal, D., Geiger,
U., Haas, M., and Scheidt, M. (2004). Technologies for variable valve trains; a
contribution to modern S.I. engines. SAE paper 2004-34-0005.
[Gottschalk, 2006] Gottschalk, B. (2006). OICA President Gottschalk: Automotive
industry is a key sector with a high rate of innovation worldwide. International
Organization of Motor Vehicle Manufacturers (OICA).
211
BIBLIOGRAPHY
[Gould et al., 1991] Gould, L., Richeson, W., and Erickson, F. (1991). Performance
evaluation of a camless engine using valve actuators with programmable timing.
SAE paper 910450.
[Griffiths, 1999] Griffiths, D. (1999). Introduction to Electrodynamics. Prentice Hall.
[Gunselmann and Melbert, 2003] Gunselmann, C. and Melbert, J. (2003). Improved
robustness and energy consumption for sensorless electromagnetic valve train. SAE
paper 2003-01-0030.
[Hara et al., 2000] Hara, S., Hidaka, A., Tomisawa, N., Takemura, S., and Nohara,
T. (2000). Application of a variable valve event and timing system to automotive
engines. SAE paper 2000-01-1224.
[Hartwig et al., 2005] Hartwig, C., Josef, O., and Gebauer, K. (2005). Dedicated
intake actuator for electromagnetic valve trains. SAE paper 2005-01-0773.
[Haskara et al., 2004] Haskara, I., Kokotovic, V., and Mianzo, L. (2004). Control of
an elctro-mechanical valve actuator for a camless engine. International Journal of
Robust and Nonlinear Control, 14:561–579.
[Henry, 2001] Henry, R. (2001).
Single-cylinder engine tests of a motor-driven
variable-valve actuator. SAE paper 2001-01-0241.
[Heywood, 2006] Heywood, J. (2006). Fueling Our Transportation Future. Scientific
American, pages 60–63.
[Heywood, 1988] Heywood, J. B. (1988). Internal Combustion Engine Fundamentals.
McGraw-Hill.
[Hitachi, 2003] Hitachi (2003). Private Communication.
BIBLIOGRAPHY
212
[Hoffmann et al., 2003] Hoffmann, W., Peterson, K., and Stefanopoulou, A. (2003).
Iterative learning control for soft landing of electromechanical valve actuator in
camless engines. IEEE Trans. Contr. Syst. Technol., 11:174–184.
[Ilic’-Spong et al., 1987] Ilic’-Spong, M., Marino, R., Peresada, S., and Taylor, D.
(1987). Feedback linearizing control of switched reluctance motors. IEEE Trans.
Auto. Contr., 32(5):371–379.
[Isidori, 1997] Isidori, A. (1997). Nonlinear Control Systems. Springer Verlag.
[Kawase et al., 1991] Kawase, Y., Kikuchi, H., and Ito, S. (1991). 3-D nonlinear
transient analysis of dynamic behavior of the clapper type DC magnet. IEEE
Trans. Magn., 27(5):4238–4241.
[Kim et al., 1997] Kim, D., Anderson, M., Tsao, T., and Levin, M. (1997). Dynamic
model of a springless electrohydraulic valvetrain. SAE paper 970248.
[Koch et al., 2002] Koch, C., Lynch, A., and Chladny, R. (2002). Modeling and
control of solenoid valves for internal combustion engines. 2nd IFAC Conf. on
Mechatronic Systems, pages 317–322.
[Koch et al., 2004] Koch, C., Lynch, A., and Chung, S. (2004). Flatness-based automotive solenoid valve control. 6th IFAC Symp. Nonlin. Contr. Systems (NOLCOS),
pages 1091–1096.
[Konrad, 1998] Konrad, R. (1998). Verfahren zur Bewegungssteuerung für einen
ankers eines elektromagnetishen Aktuators. German Pat. Appl. 19834548 A1.
[Koopmans et al., 2003] Koopmans, L., Ström, H., Lundgren, S., Backlund, O., and
Denbratt, I. (2003). Demonstrating a SI-HCCI-SI Mode Change on a Volve 5Cylinder Electronic Valve Control Engine. SAE paper 2003-01-0753.
BIBLIOGRAPHY
213
[Kreyszig, 1993] Kreyszig, E. (1993). Advanced Engineering Mathematics. John Wiley & Sons.
[Lambert, 1758] Lambert, J. H. (1758). Observationses variae in mathesin puram.
Acta Helvetica, physicomathematico-anatomico-botanico-medica, Basel, 3:128–168.
[Lambrechts et al., 2004] Lambrechts, P., Boerlage, M., and Steinbuch, M. (2004).
Trajectory planning and feedforward design of electromechanical motion systems.
Control Engineering Practice, 13:145–157.
[Lancefield et al., 1993] Lancefield, T., Gayler, R., and Chattopadhay, A. (1993). The
practical application and effects of a variable event valve timing system. SAE paper
930825.
[Landau and Lifshitz, 1984] Landau, L. D. and Lifshitz, E. M. (1984). Electrodynamics of Continuous Media. 2nd edition.
[Lequesne, 1990] Lequesne, B. (1990). Fast acting, long-stroke solenoids with two
springs. IEEE Trans. Ind. Applicat., 26:848–856.
[Lequesne, 1996] Lequesne, B. (1996). Permanent magnet linear motors for short
strokes. IEEE Trans. Ind. Applicat., 32:161–168.
[Lequesne, 1999] Lequesne, B. (1999). Design and optimization of two-spring linear
actuators. European Trans. on Electrical Power, 9:377–383.
[Lévine, 2004] Lévine, J. (2004). On flatness necessary and sufficient conditions. 6th
IFAC Symp. Nonlin. Contr. Systems (NOLCOS), pages 125–130.
[Lévine et al., 1996] Lévine, J., Lottin, J., and Ponsart, J.-C. (1996). A nonlinear
approach to the control of magnetic bearings. IEEE Trans. Contr. Syst. Technol.,
4(5):524–544.
BIBLIOGRAPHY
214
[Li and McEwan, 1993] Li, E. and McEwan, P. (1993). 3d to 2d transformation solution of transient eddy current electromagnetic fields in an actuator. IEEE Transactions on Magnetics, 29:1733–1736.
[Ljung, 1987] Ljung, L. (1987). System Identification: Theory for the User. PrenticeHall, Englewood Cliffs, NJ.
[Ljung, 2004] Ljung, L. (2004). System Identification Toolbox for use with Matlab:
User’s Guide Version 6. MathWorks, Inc., Natick, Mass.
[Löewis, 2002] Löewis, J. (2002). Flachheitsbasierte Trajektorienfolgeregelung elektromechanischer Systeme. PhD thesis, TU-Dresden.
[Löewis et al., 2000] Löewis, J., Rudolph, J., and Woittennek, F. (2000). Discretetime flatness-based control of an electromagnetically levitated rotating shaft. Proc.
Math. Theory Networks Systems (MTNS 2000).
[Longstaff and Holmes, 1975] Longstaff, K. and Holmes, S. (1975). Internal combustion engines. U.S. Patent No. 3,882,833.
[Lynch et al., 2003] Lynch, A., Koch, C., and Chladny, R. (2003). Nonlinear observer
design for sensorless electromagnetic actuators. 3rd International Conference on
Dynamics of Continuous, Discrete and Impulsive Systems, May 2003.
[Macián et al., 2004] Macián, V., Luján, J. M., Bermúdez, V., and Guardiola, C.
(2004). Exhaust pressure pulsation observation from turbocharger instantaneous
speed measurement. Meas. Sci. Technol., pages 1185–1194.
[Martin, 1992] Martin, P. (1992). À l’étude des systèmes diffèrentielement plats. PhD
thesis, École des Mines de Paris.
[Martin et al., 1997] Martin, P., Murray, R., and Rouchon, P. (1997). Flat Systems.
BIBLIOGRAPHY
215
[Maxwell, 1864] Maxwell, J. C. (1864). A Dynamical Theory of the Electromagnetic
Field. Philosophical Transactions of the Royal Society of London, 155:459–512.
[Melgoza and Rodger, 2002] Melgoza, E. and Rodger, D. (2002). Comparison of table
models of electromagnetic actuators. IEEE Trans. Magn., 38:953–956.
[Mercorelli et al., 2003] Mercorelli, P., Lehmann, K., and Liu, S. (2003). Robust
flatness based control of an electromagnetic linear actuator using adaptive PID
controller. Proceedings of the IEEE Decision and Control Conference, pages 3790–
3795.
[Mercorelli and Liu, 2005] Mercorelli, P. and Liu, S. (2005). Model predictive control
of transistor pulse converter for feeding electromagnetic valve actuator with energy
storage. Proceedings of the 44th IEEE Conference on Decision and Control, and
the European Control Conference, pages 6794–6799.
[Mianzo et al., 2005] Mianzo, L., Newton, S., Z., and Popovic (2005). Integrated
control and power electronics for an electro-mechanical valve actuation system.
2005 IEEE/ASME International Conference on Advanced Intelligent Mechatronics,
pages 485–491.
[Micro-Epsilon, 2007] Micro-Epsilon (2007). Internal communication.
[Milam, 2003] Milam, M. (2003). Real-Time Optimal Trajectory Generation for Constrained Dynamical Systems. PhD thesis, University of Alberta.
[Montanari et al., 2004] Montanari, M., Ronchi, F., Rossi, C., and Tonielli, A. (2004).
Control of a camless engine electromechanical actuator: Position reconstruction
and dynamic performance analysis. IEEE Trans. Ind. Electron., 51:299–311.
BIBLIOGRAPHY
216
[Moro et al., 2001] Moro, D., Ponti, F., and Serra, G. (2001). Thermodynamic analysis of variable valve timing infuence on SI engine efficiency. SAE paper 2001-010667.
[Murray et al., 1995] Murray, R., Rathinam, M., and Sluis, W. (1995). Differential
Flatness of Mechanical Systems: A Catalog of Prototype Systems. 1995 ASME
Intl Mech Eng Congress and Expo.
[Nakamura et al., 2001] Nakamura, M., Hara, S., Yamada, Y., Takeda, K., Okamoto,
N., Hibi, T., Takemura, S., and Aoyama, S. (2001). A continuous variable valve
event and lift control device (VEL) for automotive engines. SAE paper 2001-010244.
[Niţu et al., 2005] Niţu, C., Grǎmescu, B., and Niţu, S. (2005). Application of electromagnetic actuators to a variable distribution system for automobile engines.
Journal of Materials Processing Technology, 161:253–257.
[Ohata and Butts, 2005] Ohata, A. and Butts, K. (2005). Towards a concurrent engine system design methodology. 2005 American Control Conference, pages 3296–
3302.
[OICA, 2005a] OICA (2005a). World Motor Vehicle Production by Country and
Type: Cars 2004 - 2005. International Organization of Motor Vehicle Manufacturers (OICA).
[OICA, 2005b] OICA (2005b). World motor vehicle production by country and type:
Light commercial vehicles 2004 - 2005. International Organization of Motor Vehicle
Manufacturers (OICA).
[Parlikar et al., 2005] Parlikar, T., Chang, W., Qiu, Y., Seeman, M., Perreault, D.,
Kassakian, J., and Keim, T. (2005). Design and experimental implementation of an
BIBLIOGRAPHY
217
electromagnetic engine valve drive. IEEE/ASME Transactions on Mechatronics,
10:482–494.
[Pawlak and Nehl, 1988] Pawlak, A. M. and Nehl, T. W. (1988). Transient finite
element modeling of solenoid actuators: The coupled power electronics, mechanical
and magnetic field problem. IEEE Transaction on Magnetics, 24(1):270–273.
[Payne, 1899] Payne, D. (1899). Electically controlled valve gear for gas or other
motors. U.S. Patent 623,821.
[Peterson, 2005] Peterson, K. (2005). Control methodologies for fast and low impact
electromagnetic actuators for engine valves. Ph.D. Thesis, University of Michigan.
[Peterson et al., 2006] Peterson, K., Grizzle, J., and Stefanopoulou, A. (2006). Nonlinear control for magnetic levitation of automotive engine val(v)es. IEEE Trans.
Contr. Syst. Technol., 14(2):346–354.
[Peterson and Stefanopoulou, 2004] Peterson, K. and Stefanopoulou, A. (2004). Extremum seeking control for soft landing of an electromechanical valve actuator.
Automatica.
[Peterson et al., 2002] Peterson, K., Stefanopoulou, A., Megli, T., and Haghgooie,
M. (2002). Output observer based feedback for soft landing of electromechanical
camless valvetrain actuator. 2002 American Control Conference, pages 1413–1418.
[Picard and Fang, 2003] Picard, A. and Fang, H. (2003). Methods to determine the
density of moist air. IEEE Trans. Instrum. Meas., 52:504–507.
[Piron et al., 1999] Piron, M., Sangha, P., Reid, G., Miller, T., D.Ionel, and Coles,
J. (1999). Rapid computer-aided design method for fast-acting solenoid actuators.
IEEE Trans. Ind. Applicat., 35:991–999.
BIBLIOGRAPHY
218
[Pischinger and Kreuter, 1984] Pischinger, F. and Kreuter, P. (1984). Electromagnetically operating actuator. U.S. Patent No. 4,455,543.
[Pischinger et al., 2000] Pischinger, M., Salber, W., van der Staay, F., Baumgarten,
H., and Kemper, H. (2000). Benefits of the electromechanical valve train in vehicle
operation. SAE paper 2000-01-1223.
[Prieto et al., 2005] Prieto, R., Oliver, J., and Cobos, J. (2005). Study of nonaxisymmetric magnetic components by means of 2D FEA solvers. IEEE Power
Electronics Specialists Conference (PESC), pages 1074–1079.
[Qiuz et al., 2004] Qiuz, Y., Parlikarz, T., Chang, W., Seemanz, M., Keimy, T., Perreaulty, D., J., and Kassakian (2004). Design and experimental evaluation of an
electromechanical engine valve drive. Proceedings of the IEEE Power Electronics
Specialists Conference, pages 4838–4843.
[Rathinam and Sluis, 1995] Rathinam, M. and Sluis, W. (1995). A test for differential
flatness by reduction to single input systems. CDS Technical Report - CDS95-018.
California Institute of Technology.
[Richeson and Erickson, 1989] Richeson, W. and Erickson, F. (1989). Pneumatic actuator with permanent magnet control valve latching. U.S. Patent No. 4,852,528.
[Ronchi and Rossi, 2002] Ronchi, F. and Rossi, C. (2002). Sensing device for camless
engine electromagnetic actuators. Proceedings of the IEEE International Conference on Industrial Electronics Control and Instrumentation (IECON), pages 1669–
1674.
[Roschke and Bielau, 1995] Roschke, T. and Bielau, M. (1995). Verfahren zur modellbasierten Messung und Regelung von Bewegungen an elektromagnetischen Ak-
219
BIBLIOGRAPHY
toren. Technische Universitaet Dresden German Patent Application DE 19544207
A1.
[Rossi and Alberto, 2001] Rossi, C. and Alberto, T. (2001). Method and device for
estimating the position of an actuator body in an electromagnetic actuator to
control a valve of an engine. Magneti Marelli European Patent Application EP
1152129 A1.
[Rossi and Tonielli, 2001] Rossi, C. and Tonielli, A. (2001). Method and device for
estimating the position of an actuator body in an electromagnetic actuator to
control a valve of an engine. European Patent Application EP 1152129 A1.
[Scacchioli, 2005] Scacchioli, A. (2005). Hybrid regulation of electromagnetic valves
in automotive systems. Ph.D. thesis, Dept. of Electrical Engineering, University of
L`Aquila.
[Schechter and Levin, 1996] Schechter, M. and Levin, M. (1996). Camless engine.
SAE paper 960581.
[Schernus et al., 2002] Schernus, C., van der Staay, F., Janssen, H., Neumeister, J.,
Vogt, B., Donce, L., Estlimbaum, I., Nicole, E., and Maerky, C. (2002). Modeling
of exhaust valve opening in a camless engine. SAE paper 2002-01-0376.
[Schmitz, 1995] Schmitz, G. (1995).
Verfahren zur Regelung der Ankerauftref-
fgeschwindigkeit an einen elektromagnetischen Aktuator durch extrapolierende Abschätzung der Energieeinspeisung. German Pat. Appl. 19807875 A1.
[Schmitz and Novotny, 1965] Schmitz, N. and Novotny, D. (1965). Introductory Electromechanics. The Ronald Press Company.
BIBLIOGRAPHY
220
[Shahbakhti et al., 2007] Shahbakhti, M., Lupul, R., and Koch, C. R. (2007). Predicting hcci auto-ignition timing by extending a modified knock-integral method.
SAE paper 2007-01-0222.
[Shampine and Reichelt, 1997] Shampine, L. and Reichelt, M. (1997). The MATLAB
ODE Suite. SIAM Journal on Scientific Computing, 18:1–22.
[Shen et al., 2006] Shen, Y., King, E., and Pfahl, U. (2006). Influences of Intake
Charge Preparations on HCCI Combustion in a Single Cylinder Engine with Variable Valve Timing and Gasoline Direct Injection. SAE paper 2006-01-3274.
[Slotine and Li, 1991] Slotine, J. and Li, W. (1991). Applied Nonlinear Control. Prentice Hall.
[Song et al., 2006] Song, H., Wang, X., Ma, L., Cai1, M., and Cao, T. (2006). Design
and Performance Analysis of Laser Displacement Sensor Based on Position Sensitive Detector (PSD). 4th International Symposium on Instrumentation Science
and Technology (ISIST), 48:217–222.
[Stolk and Gaisberg, 2001] Stolk, T. and Gaisberg, A. (2001). Elektromagnetischer
aktuator. German Patent Application DE 10025491 A1.
[Stubbs, 2000] Stubbs, A. (2000). Modeling and controller design of an electromagnetic engine valve. Master’s thesis, University of Illinois at Urbana-Champaign.
[Sun and Cleary, 2003] Sun, Z. and Cleary, D. (2003). Dynamics and control of an
electro-hydraulic fully flexible valve actuation system. Proceedings of the American
Control Conference, 4:3119–3124.
[Tai and Tsao, 2001] Tai, C. and Tsao, T. (2001). Quite seating control design of an
electromagnetic engine valve actuator. Proc. 2001 ASME Ineternational Mechanical Engineering Congress and Exposition.
221
BIBLIOGRAPHY
[Tai and Tsao, 2002] Tai, C. and Tsao, T. (2002). Control of an electromechanical
camless valve actuator. Proceedings of the American Control Conference, pages
262–267.
[Tai and Tsao, 2003] Tai, C. and Tsao, T. (2003). Control of an electromechanical
actuator for camless engines. Proc. 2003 American Control Conference.
[Takahashi et al., 1991] Takahashi, N., Nakata, T., Fujiwara, K., and Nishimura, T.
(1991). Factors affecting errors due to 2-D approximate analysis of 3-D magnetic
fields with eddy currents. IEEE Transactions on Magnetics, 27(2):5223–5225.
[Takashi and Iwao, 1995] Takashi, D. and Iwao, M. (1995). Valve drive device for
internal combustion engine and initial position setting method for valve element.
Japan Patent Application JP 0224624.
[Theobald et al., 1994] Theobald, M., Lequesne, B., and Henry, R. (1994). Control
of engine load via electromagnetic valve actuators. SAE paper 940816.
[Trajkovic et al., 2006] Trajkovic,
S.,
Milosavljevic,
A.,
Tunestål,
P.,
and
B.Johansson (2006). FPGA Controlled Pneumatic Variable Valve Actuation. SAE
paper 2006-01-0041.
[Trout, 2001] Trout, S. (2001). Material Selection of Permanent Magnets, Considering Thermal Properties Corretly. Electrical Insulation Conference and Electrical
Manufacturing and Coil Winding Conference, pages 365–370.
[Turner et al., 2004] Turner, J., Bassett, M., Pearson, R., Pitcher, G., and Douglas,
K. (2004). New Operating Strategies Afforded by Fully Variable Valve Trains. SAE
paper 2004-01-1386.
BIBLIOGRAPHY
222
[van Nieuwstadt et al., 1994] van Nieuwstadt, M., Rathinam, M., and Murray, R. M.
(1994). Differential flatness and absolute equivalence. Proceedings of the 33rd
Conference on Decision and Control, Lake Buena Vista, FL, pages 326–333.
[van Nieuwstadt and Murray, 1995] van Nieuwstadt, M. J. and Murray, R. M. (1995).
Approximate trajectory generation for differentially flat systems with zero dynamics. Proceedings of the 34th Conference on Decision and Control, New Orleans,
LA, pages 4224–4230.
[Vaughan and Gamble, 1996] Vaughan, N. and Gamble, J. (1996). The modeling and
simulation of a proportional solenoid valve. ASME Dynamic Systems, Measurment
and Control, 118:120–125.
[W. Hoffmann and A. Stefanopoulou, 2001] W. Hoffmann and A. Stefanopoulou
(2001). Valve position tracking for doft landing of electromechanical camless valvetrain. Int. Federation of Automatic Control (IFAC), Advances in Automotive
Control.
[Wang et al., 2002] Wang, Y., Megil, T., Haghgooie, M., Peterson, K., and Stefanopoulou, A. (2002). Modeling and control of electromechanical valve actuator.
SAE paper 2002-01-1106.
[Wang et al., 2000] Wang, Y., Stefanopoulou, A., Haghgooie, M., Kolmanovsky, I.,
and Hammoud, M. (2000). Modeling of an electromechanical valve actuator for
a camless engine. Proc. AVEC 2000. 5th Int. Symposium on Advanced Vehicle
Control.
[Weddle and Leo, 1998] Weddle, C. and Leo, D. (1998). Embedded actuation systems for camless engines. Proceedings of the International Conference on Adaptive
Structures and Technologies.
BIBLIOGRAPHY
223
[Wilson et al., 1993] Wilson, N., Watkins, A., and Dopson, C. (1993). Asymmetric
valve strategies and their effect on combustion. SAE paper 930821.
[Wolters et al., 2003] Wolters, P., Salber, W., Geiger, J., and Duesmann, M. (2003).
Controlled auto ignition combustion process with an electromechanical valve train.
SAE paper 2003-01-0032.
[Woodson and Melcher, 1968] Woodson, H. and Melcher, J. (1968). Electromechanical Dynamics Part I: Discrete Systems. John Wiley & Sons.
[Xiang, 2002] Xiang, J. (2002). Modeling and control of a linear electro-mechanical
actuator (LEMA) for operating engine valves. IEEE Ind. Applicat. Conf., 3:1943–
1949.
[Zeitz, 1987] Zeitz, M. (1987). The extended Luenberger observer for nonlinear systems. Systems & Control Letters, 9:149–156.
Appendix A
Supplemental Theory
A.1
Vector Differential Calculus Operations & Notation
The following briefly summarizes the notation used to denote mathematical operations commonly used in the electromagnetic discipline, especially with Maxwell’s
equations.
A.1.1
Gradient of a Scalar Function
The gradient allows the derivation of vector fields from scalar functions, the latter
of which are computationally easier to handle. For example, the gradient of a scalar
function s(x, y, z), results in
∇s =
∂s
∂s
∂s
i+
j+ k
∂x
∂y
∂z
(A.1)
where x, y and z are Cartesian coordinates. The geometric interpretation of the
gradient is such that ∇s points in the direction of maximum increase of s and the
magnitude k∇sk gives the slope or rate of increase along the maximum direction.
224
APPENDIX A. SUPPLEMENTAL THEORY
A.1.2
225
Divergence of a Vector Field and the Laplacian Operator
A useful operation in defining field quantities is the divergence operator, ∇·. Consider
a differentiable vector function v(x, y, z), where x, y and z are Cartesian coordinates
and the components of v are given by v1 , v2 and v3 as:
v(x, y, z) = v1 i + v2 j + v3 k
(A.2)
Then the divergence of v may be expressed as:
∇·v =
∂
∂
∂
∂v1 ∂v2 ∂v3
i+
j + k · (v1 i + v2 j + v3 k) =
+
+
∂x
∂y
∂z
∂x
∂y
∂z
(A.3)
Note that ∇ · v depends only on v and the points in space, not on the choice of
coordinate system.
If a scalar function s(x, y, z) is able to be differentiated twice, the divergence of a
gradient may be expressed as:
∇ · ∇s =
∂2s ∂2s ∂2s
+
+
∂x2 ∂y 2 ∂z 2
(A.4)
This operation is also known as the Laplacian of s(x, y, z) where the Laplacian operator is denoted by ∇2 and may be written as:
∇2 = ∆ = ∇ · ∇ =
∂2
∂2
∂2
+
+
∂x2 ∂y 2 ∂z 2
(A.5)
The geometric interpretation of divergence is the measure of how much a vector
spreads or diverges from a given point. A point of positive divergence can be thought
of as a ‘source’. Similarly, a negative point of divergence can be considered a ‘sink’.
226
APPENDIX A. SUPPLEMENTAL THEORY
A.1.3
Curl of a Vector Field
Again consider a differentiable vector function v(x, y, z) as before. The curl of v in a
right handed Cartesian coordinate system is then defined as:
∇×v =
∂v3 ∂v2
−
∂y
∂z
i+
∂v1 ∂v3
−
∂z
∂x
j+
∂v2 ∂v1
−
∂x
∂y
k
(A.6)
Although curl is defined in terms of coordinates, it is not dependant on the coordinate
system. The result of the curl operation can be considered a measure of how much
the vector v rotates about a specified point. The curl of a gradient is always zero.
For any scalar function s(x, y, z) that is twice differentiable
∇ × (∇s) = 0
(A.7)
Thus it may be observed that gradient fields are irrotational. Similarly, the divergence
of a curl is also always zero, ∇ · (∇ × v) = 0.
A.2
Maxwell’s Equations Derived
The following entails a brief overview of the empirical laws of electricity and magnetism in order to provide a very basic derivation of Maxwell’s equations.
A.2.1
Coulomb’s Law
Coulomb derived an inverse squared relation from measurements he made experimentally with an electrical torsion bar.
f=
qQr
4πǫ0 |r|3
(A.8)
APPENDIX A. SUPPLEMENTAL THEORY
227
This expression describes the force, f [Newtons], experienced by a point charge, q
[coulombs], when brought towards another point charge of like sign, Q. Where r[m], is
a position vector from one point charge to another, and ǫ0 is the free space permittivity
F
constant (8.854x10−12 m
). Similarly, the electric field intensity, E [volts per meter], of
charge Q may be expressed as:
E(r) =
Qr
4πǫ0 |r|3
(A.9)
Since electric charge occurs in multiples of the elementary unit of charge, 1.60x10−19 C,
charge density may be expressed as:
1 X
qi
δV →0 δV
i
ρe (r) = lim
(A.10)
which expresses a continuum model of a sum of charges at a point located by r
enclosed by a small volume, δV . Where δV is considered infinitesimally small when
compared to the considered system’s dimensions yet large enough to contain a large
number of charges, qi .
In such instances, one may assume that all charges on such a volume will experience
the same electric field. Thus, for the repulsion force of a body containing a number
of charges one may average the force to derive an electric force density, F. Where,
F = ρe E
A.2.2
(A.11)
Gauss’s Law of Electricity
An often more convenient way to relate charge density and electric field intensity is
through the integral form of Gauss’s Law, which is implied by Coulombs law,
228
APPENDIX A. SUPPLEMENTAL THEORY
I
S
ǫ0 E · nda =
Z
ρe dV
(A.12)
V
where n is the unit vector normally directed from area S which encloses volume V .
It can be seen that in the case of a point charge, Equation A.12 reverts to that of
Coulombs law, Equation A.8 where the surface S may be taken as a sphere centered at
the location of point charge Q. A differential form of Equation A.12 may be obtained
by applying the divergence theorem,
I
S
v · nda =
Z
∇ · vdV
(A.13)
(∇ · ǫ0 E − ρe )dV = 0
(A.14)
V
to result in the following:
Z
V
Since the integrated volume is arbitrary, one may state:
∇·E =
ρe
ǫ0
(A.15)
Which is in fact one of Maxwell’s equations. Both Oersted and Gauss experimented
with electromagnetism and its forces upon objects such as a compass needle.
A.2.3
Gauss’s Law for Magnetism
Gauss’s law for magnetism is a formalized means of stating the observation that unlike
electricity, free magnetic poles do not exist (at least they have yet to be observed).
This implies that the net magnetic flux passing through any closed surface, S, is
always zero.
229
APPENDIX A. SUPPLEMENTAL THEORY
ΦB =
I
S
B · nda = 0
(A.16)
Or, again recognizing the surface is arbitrary and applying the divergence theorem,
the relation may be expressed in differential format as,
∇·B =0
(A.17)
This relation represents another of Maxwell’s equations.
A.2.4
Conservation of Charge
Observed experimental evidence has shown that electric charge is always conserved.
For example, an equal and opposite charge is observed on a previously neutral atom
when an electron is removed. When this concept is applied to an arbitrary volume,
V , enclosed by a smooth surface, S, any flow rate of charge out of the surface must
be balanced by the rate of which charge decreases in the volume. A continuum
variable, current density, J [ mC2 s ] is often used to express this charge flow rate. The
sign indicates direction of flow of positive charge and its magnitude indicates the net
rate of charge flow per unit area. Mathematically this observation may be expressed
as:
I
d
J · nda = −
dt
S
Z
ρe dV
(A.18)
V
or, in differential format (again utilizing the divergence theorem) as:
∇·J+
∂ρe
=0
∂t
(A.19)
230
APPENDIX A. SUPPLEMENTAL THEORY
A.2.5
Ampére’s Law
As a means to relate the magnetic effect of time varying electric fields and currents.
Maxwell extended the original expression by adding the far right hand side term,
known as displacement current, to account for influence of time varying electric fields
displacing electrons and thus generating current.
∇ × B = µ0 J + µ0
∂ǫ0 E
∂t
(A.20)
The later format is another of Maxwell’s equations.
A.2.6
Faraday’s Law of Induction
Faraday discovered that time varying magnetic fields induce an emf. He infered
from this that time varying magnetic fields must produce time varying electric fields.
Experimentally this was performed by thrusting a permanent bar magnet through a
loop of wire and observing the induced current. The emf was equal to the change of
flux
ε=
I
E · dl = −
dΦ
dt
(A.21)
And E is related to a change B by the equation
d
E · dl = −
dt
C
I
Z
S
B · nda
(A.22)
Or by applying Stoke’s theorem, in differential form as:
∇×E=−
∂B
∂t
(A.23)
The latter format is known as Faraday’s Law and completes the set of Maxwell’s
equations.
APPENDIX A. SUPPLEMENTAL THEORY
A.3
231
Other Relations of Interest
The following describes other relations which may be referred to through the course
of this work.
A.3.1
Biot-Savart Law
The Biot-Savart Law is a means to assist in calculating the field produced by an arbitrary current density. As can be imagined, the solution to many of the aforementioned
equations may be challenging even for relatively simple geometry. The Biot-Savart
Law discritizes a current density into elements which may be individually solved to
calculate the field, dB, due to each current element, dl.
A.3.2
Lenz’s Law
Lenz’s Law is a means by which to predict the direction of induced currents by means
of energy conservation. Although, the same results may be achieved by scrutinizing
Faraday’s law, Lenz’s law is perhaps a more intuitive procedure for predicting currents
in a closed loop produced by a changing magnetic field. Essentially it states that the
induced current will appear in a direction that opposes the change that produced it.
A.3.3
Lorentz Force
The Lorentz force, f , is defined as the force imposed on a test charge, q, that is
moving with velocity, v while moving through electric field, E and magnetic field, B.
This relation is quantified by:
f = qE + qv × B
(A.24)
APPENDIX A. SUPPLEMENTAL THEORY
232
A single moving charge, qv may be expressed as a current. Thus the first term
is a force on a static charge and the second term is the force on a current. As
mentioned earlier, we are generally not concerned with electrostatics. Further, when
considering the average contribution the force may be expressed as current density
and the magnetic field as shown:
F = Jf × B
A.4
(A.25)
Magnetic Materials
When considering the design of electromagnetic actuators, how a material behaves
in the presence of a magnetic field is of great importance. The orbiting electrons of
a particular atom within any material cause small but significant electric currents,
which in turn cause small magnetic fields. The spin of the electron about its own axis
also contributes to this current. The way in which the atom’s magnetic fields interact
with each other determines the overall magnetic behavior. For example, the fields
produced by the electron spins of most materials are random and cancel out. Other
materials may have regions which have aligned spins and thus an overall net field
exists. The orientations are susceptible to external fields where the net magnetic field
may be decreased, or enhanced depending on the new orientation. Generally, there
are three major classifications of magnetic materials, diamagnetic, paramagnetic and
ferromagnetic. Diamagnetic materials are characterized by resisting an applied external field. The effect is slight, but still measurable. For example, a typical relative
permeability of a diamagnetic material may be 0.9. Paramagnetic materials align so
that there is a slight but measurable increase in the applied field. A typical relative
permeability may be on the order of 1.1. Ferromagnetic material possesses the ability
to greatly enhance an external field, with relative permeabilities ranging from 100 to
APPENDIX A. SUPPLEMENTAL THEORY
233
1,000,000. As the regions of common orbital spin align, the external field is intensified, after which, the material is said to be saturated and then behaves as though
no material were present, or takes on permeability of free space. This saturation
limit is of great concern when controlling solenoid actuators as the armature material
will most likely be saturated at low air gaps, where accurate control is most desired.
The effect is also significant as the transition from high permeability to free space
permeability is nonlinear, making accurate analytic performance estimates extremely
challenging. It should also be noted that once magnetized, the aligned regions do not
instantaneously return to their initially demagnetized state, nor does flux density follow the same magnetization path. This phenomena is known as magnetic hysteresis.
Although this may complicate even a low frequency analysis, it is often of greater
concern with the design of higher frequency devices such as transformers. Hysteresis
is often negligible in systems with air gaps, especially at low flux densities, as the
energy required to magnetize the gap is several orders greater than any losses incurred through magnetic domain realignment. The field strength required to return
a material to a demagnetized space is referred to as coercive force. Ferromagnetic
permeability is also sensitive to temperature, with a sharp drop at what is referred
to as the Curie temperature. This temperature typically corresponds with a phase
change (change in crystalline structure).
Despite the complex properties of ferromagnetic materials, their high permeability
allows flux to be constrained to a predefined path just as one would use wires to
connect an electrical circuit. In fact, in many circumstances magnetic systems may
be analyzed as a circuit problem rather than a field problem as discussed in Section
3.3.
APPENDIX A. SUPPLEMENTAL THEORY
A.5
234
Eddy Currents
Eddy currents are characterized as local circulating currents which exist in the core
material. These are physically existing currents produced within the material due to
a time varying core flux. They may be thought of as a short circuit consisting of coil
wrapped around the external core material path in that the change in flux induces a
current which in turn generates its own magnetic flux in the opposite sense (obeying
Lenz’s law) and ultimately opposes the change or rise of flux of the overall circuit.
Thus, the observed flux rise or magnetization curve will be lower than the that of
the static case. The energy difference between the static and rapid field buildup is
defined through resistive losses and hence heating. In summary, the two effects of eddy
currents are: an internal MMF is generated which tends to counteract the applied
MMF, an irreversible heating loss of energy with the i 2 losses in the core. Thus greater
changes in flux tend to generate more losses. A widening of the hysteresis loop is an
indication of the eddy current magnitudes. They may be minimized by using materials
with low conductivity and by laminating the core structure (through thin sheets or
sintered powder metallurgy techniques). In the actuator studied, laminations are used
to increase the circulating path length thereby breaking the eddy current paths into
many smaller loops with lower magnitude and subsequently reduce the counter flux
generated. Eddy currents are not to be confused with the electrodynamic Amperian
currents generated by electron spin which are used to explain material magnetism.
The eddy current counter flux tends to constrain the applied or useful flux path
density to the edge of the material, also know as the skin effect. The skin depth,
δ may be approximated as a function of the material conductivity, σ, excitation
frequency f , material permeability, µ through the following relation:
δ=√
1
πµσf
[Landau and Lifshitz, 1984]
(A.26)
APPENDIX A. SUPPLEMENTAL THEORY
A.6
235
The Lambert W Function
The solution of a feedforward current trajectory as well as the parametrization of the
coil current state, i, with the flat output, y, and respective time derivatives, requires
the algebraic solution of of the Lambert W function [Corless et al., 1996] as discussed
in Chapter 5. This expression was first introduced by Euler in 1779 [Euler, 1779]
after studying Lambert’s transcendental equation in [Lambert, 1758]. This multivalued relation is defined to be the inverse of the function w 7→ wew = z, z ∈ C. It
thus provides a solution to:
W (z)eW (z) = z
(A.27)
When restricted to only real, negative arguments, Equation (A.27) has a unique real
solution represented by the principal branch, W0 , as shown in Figure A.1. For all
real z in the range 0 > z > −1/e, the function is multi-valued with two real solutions
that are represented by the W0 and W−1 branches. For i ≥ 0, only the principal
real branch, W−1 (z), is considered since the argument z varies from -1 to -∞ in all
instances that it is invoked in this study.
APPENDIX A. SUPPLEMENTAL THEORY
Figure A.1: Two real branches of the Lambert W function [Chung, 2005]
236
Appendix B
Actuator Properties and Specifications
B.1
Introduction
The following provides key actuator properties and data.
B.2
Magnetic Data
Linear relative permeabilities for the back iron leg, µirnlg and root portions, µirnrt and
armature, µarm , are estimated from the induction curves for the respective materials as
plotted in Figures B.1 and B.2. Since the linear permeability of steel is approximately
1000 times greater than air, the steel material properties are expected to be of little
significance in the inductance calculation at large air gaps (or prior to the onset of
saturation). Upon material saturation, the steel permeability is the same as free space,
µ0 or air, and hence the reluctance of the steel flux path lengths become as significant
as the air gaps. Saturation is expected to occur at relatively large excitation levels
and small air gaps, as it is at those operating points where the magnetic field intensity
will be greatest.
Table B.1 lists pertinent results of LPM fitting and manufacturer material data for
both opener and closer magnets. The armature and core are made from 0.3 mm thick
sheets of silicon steel alloy. The material is provided by Vacuumschmelze Hanau and
237
238
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
2.5
Flux Density [T]
2
Armature
Backiron (root)
Backiron (legs)
Air (free space)
1.5
1
0.5
0
0
0.2
0.4
0.6
0.8
1
1.2
Applied Field Intensity [A/m]
1.4
1.6
1.8
2
5
x 10
Figure B.1: Induction curves of the various materials used in the hinged actuator
magnetic path (large field)
their respective trade names are VX17.1 and ORSI H100-30.
B.3
Mechanical Data
The parameters for the spring mass and gas force models are derived through measured and system identification techniques. The following sections provide a brief
summary of the techniques used to measure key parameters. All relevant parameters
are listed in Table B.2.
B.3.1
Torsion Bar Force Measurement
As part of the actuator performance evaluation, the torsion bar force as a function
of armature position was measured. Figure B.3 illustrates the measured torsion bar
response.
239
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
2
1.8
1.6
Flux Density [T]
1.4
1.2
Armature
Backiron (root)
Backiron (legs)
Air (free space)
1
0.8
0.6
0.4
0.2
0
0
1000
2000
3000
4000
5000
Applied Field Intensity [A/m]
6000
7000
8000
Figure B.2: Induction curves of the various materials used in the hinged actuator
magnetic path (small field)
B.3.2
Testbench System Identification
A greybox model of the mechanical system was created with MATLAB’s System Identification Toolbox [Ljung, 1987, Ljung, 2004] to identify the effective moving mass,
spring stiffness, viscous damping and spring pre-load parameters. The actuator is
closed then released and the unforced position and velocity response data are recorded.
Three sets of data are collected for both a ‘cold’ and ‘warm’ operating condition. In
the ‘cold’ case, the free response is recorded after being unused for 24 hrs. In the
‘warm’ case, the actuator is run for approximately 3000 cycles immediately before
recording the response. For each operating case, two data sets are used to generate
predictive error method models. Upon which, the model performance is contrasted
with the third data set as shown in Figures B.4 and B.5. The proposed mechanical
model appears to be a valid with a minimum agreement of approximately 97% among
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
240
Table B.1: Magnetic and Electric Lumped Model Parameters
Parameter
ψ [Wb]
β [m/A]
κ [m]
α [A−1 ]
Resistivity [Ωm]
Armature
Core
Coil (copper)
R [Ω]
ℓm [m]
Nf c [turns]
Nec [turns]
Opener
0.1354
7.6801 × 10−5
4.1042 × 10−3
8.3263 × 10−3
Closer
0.1184
5.7479 × 10−5
4.0700 × 10−3
3.5562 × 10−3
5.0 × 10−3
2.0 × 10−3
3.0 × 10−8
0.51
19.4 × 10−3
10
50
5.0 × 10−3
2.0 × 10−3
3.0 × 10−8
0.46
20.8 × 10−3
10
50
the various data sets when contrasted with validation data. It is observed that the
parameter that varied most significantly was friction with an increase of 5.9% from
the cold to the warm operating conditions.
241
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
600
Measured Torsioin Bar Force [N]
500
400
300
200
100
0
0
1
2
3
4
5
Valve Position [mm]
6
7
8
Figure B.3: Torsion Bar Force Exerted on Valve Vs. Valve Position
242
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
−3
Displacement [m]
4
x 10
2
0
−2
−4
4
Velocity [m/s]
Measured Set 1
PEM Set 2 Fit: 98.22%
PEM Set 3 Fit: 97.96%
2
0.005
0.01
0.015
0.02
0.025
Measured Set 1
PEM Set 2 Fit: 97.04%
PEM Set 3 Fit: 96.99%
Average Parameter
Values:
m = 116.16 g
k = 117.01 N/mm
b = 5.10 Ns/m
x = 0.18 mm
0
−2
off
0
0.005
0.01
0.015
0.02
0.025
Time [s]
Figure B.4: Measured and predicted response during ‘cold’ operation
243
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
−3
Displacement [m]
4
x 10
2
0
−2
−4
0
0.005
0.01
0.015
Measured Output
PEM Set 2 Fit: 96.47%
PEM Set 3 Fit: 98.21%
0.02
0.025
Velocity [m/s]
4
2
Average Parameter
Values
m = 116.01 g
k = 117.01 N/mm
b = 3.39 Ns/m
xoff = 0.20 mm
0
−2
−4
0
Measured Output
PEM Set 2 Fit: 96.57%
PEM Set 3 Fit: 97.17%
0.005
0.01
0.015
0.02
0.025
Time [s]
Figure B.5: Measured and predicted response after approximately 3000 cycles
APPENDIX B. ACTUATOR PROPERTIES AND SPECIFICATIONS
Table B.2: Mechanical Lumped Model Parameters
Parameter
m [kg]
mv [kg]
Io [kg m2 ]
b [Ns/m]
bv [Ns/m]
b̂ [Nsm/rad]
k [N/m]
kv [N/m] [Ωm]
k̂ [Nm/rad]
Fv [N]
S [m]
ℓv [m]
Av [m2 ]
Cd [-]
Cgf [-]
c1 [-]
c2 [1/m]
c3 [1/m2 ]
Opener
Closer
0.116
0.037
6.190x 10−5
3.390
1.700
0.010
117.010x 103
40.000x 103
112.000
23.402
8.00x 10−3
38.0x 10−3
5.309 x 10−4
0.9
0.8
0.997
31.875
-11437.500
-
244
Appendix C
Program and Data File Summary
C.1
Introduction
The following lists all model, command, program and data files that were used and
generated over the course of this work.
245
APPENDIX C. PROGRAM AND DATA FILE SUMMARY
C.2
ANSYS Static Command and Result Files
Table C.1: ANSYS static command and result files
File Description
2D armature geometry used by ANSYS macro files
2D opener geometry used by mopnrmac.tex
2D opener geometry used by mclsrmac.tex
Pro/E file used to create 2D geometry files
ANSYS master command file (opener)
ANSYS master command file (closer)
ANSYS macro command file (opener)
ANSYS macro command file (closer)
Output torque result file (opener)
Output force in x-direction result file (opener)
Output force in y-direction result file (opener)
Output flux result file (opener)
Output torque result file (closer)
Output force in x-direction result file (closer)
Output force in y-direction result file (closer)
Output flux result file (closer)
MATLAB file used to process ANSYS result files
MATLAB output ANSYS results file (opener)
MATLAB output ANSYS results file (closer)
MATLAB LPM parameter fitting script (opener)
MATLAB LPM parameter fitting script (closer)
MATLAB Model script called by nlinfit.m
File Name
armature.igs
obkirn.igs
cbkirn.igs
ansyssection.prt
mopnr.tex
mclsr.tex
mopnrmac.tex
mclsrmac.tex
torque1op.res
forcex1op.res
forcey1op.res
flxopen.res
torque1cl.res
forcex1op.res
forcey1op.res
flxclose.res
linpost.m
opentab.m
clostab.m
openerparam.m
closerparam.m
nlinforfit.m
246
247
APPENDIX C. PROGRAM AND DATA FILE SUMMARY
C.3
ANSYS Transient Command and Result Files
Table C.2: ANSYS transient command and result files
File Description
ANSYS master command file (opener)
ANSYS macro command file (opener)
ANSYS transient result file (opener)∗
File Name
mopnrtrans.tex
mopnrtransmac.tex
transopnr X42VXXXmm.rst
∗ - transopnr 42VXXXmm represents the air gap in millimeters ×102 for the corresponding experimental input.
C.4
Simulink Models and Parameter Files
Table C.3: Simulink lumped parameter model files
File Description
Simulink parameter file
for all models
Simulink full dynamic model
Simulink simplified landing
control model
Simulink hybrid and LPM
transient magnetic model
Exhaust valve opening
simulation results files∗
Exhaust valve opening
simulation results files∗∗
File Name
param.m
fullsim.mdl
simpland.mdl
transientcomp simple.mdl
fdfwdflxFBXXbar rwTB X.mat
fdfwdflxFBXXbar rwEXXXXrpm X.mat
∗ - fdfwdflxFBXXbar rwTB X represents the EVO pressure and case number.
∗∗ - fdfwdflxFBXXbar rwEXXXXrpm X.mat represents the EVO pressure, simulated
engine speed and case number.
APPENDIX C. PROGRAM AND DATA FILE SUMMARY
C.5
248
Trajectory Optimization Files
Table C.4: MATLAB files used for generating landing control reference trajectories
File Description
Main script (opener)
Main script (closer)
Open loop current output function (opener)
Open loop current output function (opener)
C.6
File Name
trajgen.m
trajgen cl.m
nc.m
nc cl.m
Material Testing Machine Experimental Program and Data Files
In all files, the following signals are recorded: crosshead position (0.5”/V), small load
cell (51lb/V), large load cell (40lb/V), commanded opener coil current, actual opener
coil current (0.2A/V), opener coil voltage (0.2V/V), opener sensing coil signal (integrated) and H-bridge TTL switch signals. ∗ - s 25 XX XXX represents a 25 ms step
Table C.5: Material testing machine experimental and raw fata files
File Description
ControlDesk experiment file
ControlDesk interface layout file
dSPACE make file (to be compiled)
Variable mapping file
Compiled dSpace Executable File
Most recent parameter settings file
Control program Files
Steady state raw data files ∗
Transient voltage input raw
data files ∗ ∗
Sinusoidal current input raw
data files ∗ ∗ ∗
Armature motion raw data files †
File Name
. . . \MTScurrent\MTSDataAcquisition.cdx
. . . \MTScurrent\test.lay
. . . \MTScurrent\fr.mk
. . . \MTScurrent\fr.trc
. . . \MTScurrent\fr.ppc
. . . \MTScurrent\mtspar2.par
. . . \MTScurrent\src\∗.c
. . . \step\s 25 XX XXX.mat
. . . \impulse\i XX XXX XXX.mat
. . . \sinusoid\f XX XXX XXX.mat
. . . \sinusoid\m XXX XX XXX XX a.mat
input at a current level of XX amps and XXX indicating air gap in mm ×102 .
APPENDIX C. PROGRAM AND DATA FILE SUMMARY
249
∗∗ - i XX XXX XXX represents the switched voltage level, input duration in ms ×102
and air gap in mm ×102 .
∗ ∗ ∗ - f XX XXX XXX represents the switched voltage level, current input frequency
in kHz ×102 , mean current level amps and air gap in mm ×102 .
†- m XXX XX XXX XX a represents the armature amplitude in mm ×102 , displacement frequency in Hz, minimum air gap in mm ×102 and the current input in amps.
A suffix of ‘ a’ indicates the armature is moving away from the pole face during
recording, otherwise it is moving towards it.
C.7
Testbench Experiment and Data Files
In all files, the following signals are recorded: laser position (1.00mm/V), pressure
transducer (9psi/V), reconstructed position, commanded opener feedforward coil current, estimated position, estimated velocity, estimated gas force, actual coil currents
(0.2A/V), sensing coil signals (integrated), analog integrator reset signal, control
state, CPU task execution time and H-bridge TTL switch signals. Older data files
also contain recorded coil voltages (0.2V/V). Caution should be used in interpreting
such signals as they may be erroneous due to aliasing since both the sampling and
PWM carrier frequencies are at a rate of 50KHz. As a result, the commanded switch
signals are better indicators of coil voltage as they are control outputs, not sampled
inputs. ∗ - XXXX represents the emulated engine rpm, X indicates the EVO pressure
in bar and XX indicates the data set number.
∗∗ - XbarXbar indicates the respective minimum and maximum EVO pressure and
XXXX is the emulated rpm with X indicating the data set number.
250
APPENDIX C. PROGRAM AND DATA FILE SUMMARY
Table C.6: Testbench experimental and raw data files
File Description
ControlDesk experiment file
ControlDesk interface layout files:
dSPACE make file (to be compiled)
Variable mapping file
Compiled dSpace Executable File
Most recent parameter settings file
Control program Files
Landing control with open-loop
feedforward files ∗
Nominal data files with
disturbance observer ∗
Switched pressure data files ∗∗
C.8
File Name
. . . \dSPACECode\fluxrev5\ValveControl3a.cdx
. . . \dSPACECode\fluxrev5\plotters.lay
. . . \dSPACECode\fluxrev5\control.lay
. . . \dSPACECode\fluxrev5\anschwingen.lay
. . . \dSPACECode\fluxrev5\fr.mk
. . . \dSPACECode\fluxrev5\fr.trc
. . . \dSPACECode\fluxrev5\fr.ppc
. . . \dSPACECode\fluxrev5\bk1 hand flt4.par
. . . \dSPACECode\fluxrev5\src\∗.c
flatnessXXXXrpmXbar XX.mat
flxflataff XXXXrpm Xbar XX.mat
dp XbarXbarXXXXrpm X.mat
Primary Testbench Control Program Files
The following lists the primary C program files that contain the key feedforward,
estimation, position reconstruction and landing control algorithms.
Table C.7: Primary testbench control program files
File Description
Task manager
Control state manager (state machine)
Data acquisition, gas pressure controller
and flux-based reconstruction
FEA data for flux-based position measurement
Feedforward and landing controllers
State and disturbance estimation
File Name
tasks.c
control.c
hw inter.c
fluxdat.h
ctrl traj.c
filter.c
251
APPENDIX C. PROGRAM AND DATA FILE SUMMARY
C.9
Primary Analysis and Postprocessing Files
The following MATLAB files are various scripts used to help automate the data
process, plotting and analysis of the material testing machine and testbench results.
Table C.8: Files and scripts used to compile and analyze experimental results
File Description
Analog integrator drift fitting and plots
Material testing machine static data
processing and plotting
Multiple cycle data segmenter and
statistical analysis
Simple landing control simulation results plotter
Gas force model comparison and plotting
Flux-based position sensor sensitivity analysis
System ID main script
System ID greybox model file
System ID results plotter
File Name
drftcalc.m
stplt.m
flatVaffmultipressplot.m
simplandplot.m
press comp plt.m
fluxsens.m
mechid.m
sprngms.m
sysidplot.m
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement