MODELING AND CONTROL STRATEGIES FOR MULTIPROCESS ARC WELDING POWER SOURCES
by
JONATHON C. KELM
Submitted in partial fulfillment of the requirements
For the degree of Doctor of Philosophy
Thesis Adviser: Prof. Kenneth A. Loparo
Department of Electrical Engineering & Computer Science
CASE WESTERN RESERVE UNIVERSITY
January, 2020 Modeling and Control Strategies for Multiprocess Arc
Welding Power Sources
Case Western Reserve University Case School of Graduate Studies
We hereby approve the thesis1 of
JONATHON C. KELM
for the degree of
Doctor of Philosophy
Prof. Kenneth Loparo 11/15/2019
Committee Chair, Adviser Date Department of Electrical Engineering & Computer Science
Prof. Vira Changkong 11/15/2019
Committee Member Date Department of Electrical Engineering & Computer Science
Prof. Robert Gao 11/15/2019
Committee Member Date Department of Mechanical & Aerospace Engineering
Prof. Wei Lin 11/15/2019
Committee Member Date Department of Electrical Engineering & Computer Science
1We certify that written approval has been obtained for any proprietary material contained therein. Dedicated to my parents, who taught me everything that really matters. Table of Contents
List of Tables vii
List of Figures viii
List of Initialisms xiv
List of Symbols xviii
Acknowledgements xxii
Abstract xxiii
Chapter 1. Introduction1
Background and Motivation1
Literature Review 14
Goals of this Work 25
Contributions of this Work 26
Thesis Organization 26
Chapter 2. Modeling 28
Welding Cables, Workpiece, and Welding Fixture 28
Electrode and Contact Tip 33
The Welding Arc 35
Summary of Load Impedances 37
Piecewise Linear Model 38
Sources of Uncertainty 47
Chapter 3. Review of the Existing System 51
iv Overview of the Existing Control System 51
Waveform Generator Reference Tracking 54
Chapter 4. Proposed Control Strategy 60
Overview of Sliding Mode Control 61
Simplified Model 62
Current Tracking 63
Voltage Tracking 87
Power Tracking 93
Regulation Mode Switching 93
Output Limiting 93
Advantages of SMC for Welding Power Sources 97
Drawbacks of Sliding Mode Control 97
Chapter 5. Hardware Implementation 99
Abortive Efforts 99
Hybrid Analog/Digital Solution 101
Examples 111
Chapter 6. Results 115
Experimental Setup 115
Current Step Command 116
Current Exponential Command 124
Chapter 7. Summary, Conclusions, and Future Work 127
Summary and Conclusions 127
Recommendations for Further Work 128
v Appendix A. Simulation 130
General Structure 130
Integration Method 135
Simulation Configuration 138
Load Change Simulation 140
Simulation Parameters 141
Complete References 142
vi List of Tables
1.1 Timeline of welding power source technology 16
2.1 Ranges of Model Parameters 50
4.1 Current error dynamics for various reference signals 80
4.2 Voltage error dynamics for various reference signals 89
A.1 Simulation parameters 141
vii List of Figures
1.1 Illustration of arc welding terminology3
1.2 “Chopper” style welding power source structure6
1.3 “Inverter” style welding power source structure7
1.4 Shielded Metal Arc Welding (SMAW) setup8
1.5 Flux-Cored Arc Welding (FCAW) setup8
1.6 Gas Metal Arc Welding(GMAW) setup9
1.7 Gas Tungsten Arc Welding(GTAW) setup 10
2.1 Circuit diagram showing sources of impedance in the welding
circuit 29
2.2 Welding cable cross-section showing electrical and geometrical
properties 30
2.3 Circuit diagram for welding cable transmission line model 30
2.4 AWG 0000 welding cable resistance versus cable length 31
2.5 AWG 0000 welding cable inductance versus cable length 32
2.6 AWG 0000 welding cable capacitance versus cable length 33
2.7 AWG 0000 welding cable resonant frequency versus cable length 34
2.8 Wire electrode fed through contact tube as in Gas Metal Arc
Welding (GMAW) and FCAW 35
2.9 V-I characteristics for GMAW arcs of various lengths 37
2.10 Circuit diagram of power section and load 40
viii 2.11 Circuit diagram of power section and load when switch is closed 40
2.12 Circuit diagram of power section and load when switch is open 41
2.13 Relationship between switch position u and choke current iL(t) 43
2.14 Current filtering effect of the welding circuit impedance 44
2.15 Inductance vs. current for the output inductor of a typical
welding power source 47
3.1 Block diagram of the existing control system 53
3.2 Example of a simplified FSM for a GMAW welding program 54
3.3 Example of a constant reference signal 55
3.4 Example of a ramp reference signal 57
3.5 Example of an exponential reference signal 57
3.6 Example of a parabolic reference signal 58
3.7 Output regulation modes in a representative sample of welding
programs 59
4.1 Circuit diagram of simplified power source and load 63
4.2 Sample current error trajectories for u = 0 for a constant reference 65
4.3 Sample current error trajectories for u = 1 for a constant reference 66
4.4 Desired current trajectory z2 = kiz1 for a constant reference 67 − 4.5 Current error trajectories with u = 0, u = 1, and sliding line
σi = 0 68
4.6 Sliding mode existence for constant current, k < Rw + Ro + Ro 70 i Lw Lw L
ix Rw Ro Ro 4.7 Sliding mode existence for constant current, ki + + 71 ≈ Lw Lw L 4.8 Sliding mode existence for constant current, k > Rw + Ro + Ro 72 i Lw Lw L
4.9 Relationship between switching function σ, control u, and
hysteresis κ 74
4.10 Relationship between load resistance Rw and hysteresis κ for
Fsw = 20 kHz 75
4.11 Relationship between load inductance Lw and hysteresis κ for
Fsw = 20 kHz 76
4.12 Current step simulation, Fsw = 20 kHz 77
4.13 Current step simulation, Fsw = 80 kHz 78
4.14 Current standard deviation vs. switching frequency 79
4.15 Current exponential simulation 82
4.16 Current ramp simulation 84
4.17 Current parabola simulation 86
4.18 Voltage exponential simulation 91
4.19 Circuit diagram for the Open Circuit Voltage (OCV) case 92
4.20 Example operating region defined by current, voltage, and power
limits 95
4.21 Example of output limiting, 20 V regulation at a 100 A minimum
current 96
5.1 Original analog SMC control board 101
5.2 Hybrid analog/digital SMC PCB 103
x 5.3 Basic structure of the existing control hardware 104
5.4 Basic structure of the modified control hardware 104
5.5 Block diagram of the SMC hardware implementation 105
5.6 SMC algorithm flowchart 107
5.7 Band-limited derivative frequency responses for various values of
τ 109
5.8 State diagram for CPLD gate drive logic 110
5.9 Oscilloscope trace showing hysteresis modulation signals 111
5.10 Oscilloscope trace showing detail of hysteresis modulation signals 112
5.11 Oscilloscope trace of 200 A current regulation at 20 kHz 112
5.12 Oscilloscope trace of 200 A current regulation at 10 kHz 113
5.13 Oscilloscope trace of current ramp regulation 113
5.14 Oscilloscope trace of current exponential regulation 114
6.1 Experimental load bank 116
6.2 84 µH coil of welding cable 116
6.3 255 µH coil of welding cable 117
6.4 LCR meter used for measuring cable inductance 117
6.5 Current step, 10 A to 100 A, with varying load resistance (existing
controls) 119
6.6 Current step, 10 A to 100 A, with varying load resistance (SMC
simulation) 119
xi 6.7 Current step, 10 A to 100 A, with varying load resistance
(hardware implementation) 119
6.8 Current step, 20 A to 200 A, with varying load resistance (existing
controls) 120
6.9 Current step, 20 A to 200 A, with varying load resistance (SMC
simulation) 120
6.10 Current step, 30 A to 300 A, with varying load resistance (existing
controls) 121
6.11 Current step, 30 A to 300 A, with varying load resistance (SMC
simulation) 121
6.12 Current step, 10 A to 100 A, with varying load inductance
(existing controls) 122
6.13 Current step, 10 A to 100 A, with varying load inductance (SMC
simulation) 122
6.14 Current step, 20 A to 200 A, with varying load inductance
(existing controls) 123
6.15 Current step, 20 A to 200 A, with varying load inductance (SMC
simulation) 123
6.16 Current exponential, 20 A to 200 A, with varying load resistance
(existing controls) 125
6.17 Current exponential, 20 A to 200 A, with varying load resistance
(SMC simulation) 125
xii 6.18 Current exponential, 20 A to 200 A, with varying load inductance
(existing controls) 126
6.19 Current exponential, 20 A to 200 A, with varying load inductance
(SMC simulation) 126
A.1 Simulator “main loop” flowchart 132
A.2 Simulated load changes from R1 = 100 mΩ to R2 = 5 mΩ 140
xiii List of Initialisms
AC Alternating Current.
ADC Analog-to-Digital Converter.
ADRC Active Disturbance Rejection Control.
ANN Artificial Neural Network.
AWG American Wire Gauge.
AWPS Arc Welding Power Source.
AWS American Welding Society.
BLS Bureau of Labor Statistics.
CAG Carbon Arc Gouging.
CC Constant Current.
CPLD Complex Programmable Logic Device.
CTWD Contact Tip to Workpiece Distance.
DAC Digital-to-Analog Converter.
DC Direct Current.
DCEN Direct Current Electrode Negative.
DCEP Direct Current Electrode Positive.
DCM Discontinuous Conduction Mode.
DSP Digital Signal Processor.
xiv EMC Electromagnetic Compatibility. emf Electromotive Force.
EMI Electromagnetic Interference.
EPDM Ethylene Propylene Diene Monomer.
FCAW Flux-Cored Arc Welding.
FLC Fuzzy Logic Control.
FOC Fractional Order Control.
FPGA Field Programmable Gate Array.
FSM Finite State Machine.
GMAW Gas Metal Arc Welding.
GTAW Gas Tungsten Arc Welding.
GTAW-P Pulsed Gas Tungsten Arc Welding.
HM Hysteresis Modulation.
I/O Input/Output.
IEC International Electrotechnical Commission.
IGBT Insulated Gate Bipolar Transistor.
LAC Linear Average Control.
LCR Inductance/Capacitance/Resistance.
LTI Linear Time-Invariant.
xv MAWPS Multiprocess Arc Welding Power Source.
MOSFET Metal-Oxide Semiconductor Field-Effect Transistor.
MPC Model Predictive Control.
OCV Open Circuit Voltage.
PC Personal Computer.
PCB Printed Circuit Board.
PI Proportional-Integral.
PID Proportional-Integral-Derivative.
PWM Pulse Width Modulation.
QSMC Quasi-Sliding Mode Control.
RK4 Classic Runge-Kutta Method.
SCR Silicon Controlled Rectifier.
SISO Single Input Single Output.
SMAW Shielded Metal Arc Welding.
SMC Sliding Mode Control.
SMPS Switched-Mode Power Supply.
SPI Serial Peripheral Interface.
xvi USA United States of America.
ZCS Zero Current Switching.
ZVS Zero Voltage Switching.
xvii List of Symbols
Fsw switching frequency.
Ga electrical conductance of welding arc.
Lw inductance of welding circuit load impedance.
L output choke inductance.
P0 electric arc cooling power.
Re electrical resistance of welding electrode.
Ro resistance of welding power source output resistor.
Rt electrical resistance of electrode contact point.
Rw resistance of welding circuit load impedance.
Tsw switching period.
V0 steady-state arc electromotive force.
Vg power source supply (input or “bus”) voltage.
Zc electrical impedance of welding cables.
Zp electrical impedance of welding fixture.
... x third time derivative of x.
xviii x¨ second time derivative of x. x˙ time derivative of x.
`e length of electrode extension.
−1 inverse Laplace transform. L {·}
κ switching function hysteresis bound.
Laplace transform. L {·}
µ0 magnetic permeability of free space.
µins magnetic permeability of insulation material.
µ magnetic permeability.
ν1 voltage tracking error.
ν2 derivative of voltage tracking error.
φ unknown disturbance function.
ψ unknown disturbance function.
ρc electrical resistivity of cable.
σi sliding surface switching function for current control.
σp sliding surface switching function for power control.
xix σv sliding surface switching function for voltage control.
τa arc time constant.
ξh comparison function, upper.
ξl comparison function, lower.
cc electrical capacitance per unit length of cable.
dins insulation thickness.
dc distance between welding cables. d duty cycle.
iL inductor current.
iw electrical current in the welding circuit.
ki switching function parameter for current regulation.
kp switching function parameter for power regulation.
kv switching function parameter for voltage regulation.
lc electrical inductance per unit length of cable.
pw power in welding circuit; pw = vwiw.
ri current reference waveform.
xx rp power reference waveform.
rv voltage reference waveform. s Laplace transform operator variable (complex frequency).
vw voltage across the welding circuit.
z1 current tracking error.
z2 derivative of current tracking error.
xxi Acknowledgements
Everyone in my life has been incredibly supportive and understanding as I have tried
to balance working full-time in industry, pursuing graduate studies, and raising twin
boys, usually feeling like I’m not doing any one of those things very well. Thanks to
all of you.
This research was sponsored by The Lincoln Electric Company of Cleveland, Ohio,
and I would be remiss if I did not mention my gratitude for the knowledge and
experience I have gained while working there. I would like to give special thanks to
Tom Matthews, Joe Daniel, Ed Hillen, Todd Kooken, Lifeng Luo, Judah Henry, and
Nick Trinnes for their support and for the insights they provided into the welding and
electronics topics in this thesis.
I am grateful to my adviser, Ken Loparo, whose wide-ranging technical expertise was
invaluable. He patiently provided guidance and wisdom, and contributed insights at
several steps along the way. I would like to thank the other Systems & Control Engi-
neering faculty and staff members at Case Western Reserve University for everything
I have learned from them.
Finally, I thank my wife Ashley for dutifully accepting her role as a “Ph.D. widow”
for the last few years. Without her love and support this work would not have
been completed. The recent arrival of our sons, Carl and Robert, gave me renewed
motivation and they are, by far, the most important project I will ever pursue.
xxii Abstract
Modeling and Control Strategies for Multiprocess Arc Welding Power Sources
Abstract
by
JONATHON C. KELM
A modern Multiprocess Arc Welding Power Source(MAWPS) is a Switched-Mode
Power Supply(SMPS) that has been designed to produce waveforms used for multiple
arc welding processes, such as Shielded Metal Arc Welding(SMAW), Gas Metal
Arc Welding(GMAW), Gas Tungsten Arc Welding(GTAW), and Flux-Cored Arc
Welding(FCAW). MAWPS control is challenging for a number of reasons, including the complex dynamics of switching power converters, transient conditions encountered in the metal transfer process, wide variations in load impedance, a need for tracking complex reference waveforms, incomplete or inaccurate models of the welding process itself, the difficulty of addressing the needs of several welding processes using a single machine, an electrically harsh environment with high levels of electromagnetic noise, and health and safety concerns.
In this work, models of the equipment in a welding setup are developed that can
be used for analysis and control system design. The models are used to develop a
simulation environment and a new control strategy for a welding power source from
Lincoln Electric, using Sliding Mode Control (SMC). While SMC has been applied
xxiii to SMPS elsewhere in the literature, this work focuses on the particular needs of the welding power source and incorporates output current, voltage, and power reference
tracking, switching frequency control, and output constraints.
A hardware implementation of the SMC strategy is described, and its performance
is compared against the existing control system and computer simulations. While
some implementation details still need to be worked out, the SMC strategy is shown
to be feasible to implement and to provide significant improvements in the current, voltage, and power tracking performance. These improvements should have a direct
impact on the welding performance of the Multiprocess Arc Welding Power Source
(MAWPS).
xxiv 1
1 Introduction
1.1 Background and Motivation
High-quality arc welding is an important part of modern manufacturing, with welding-
related expenditures of at least $34.1 billion in the United States in the year 2000,
equivalent to $325 for every household in the country [1]. Welding technology is a key
factor in industries such as automotive manufacturing, oil and gas pipelines, building
construction, mining, and defense.
The central piece of equipment in an arc welding setup is the welding power source.
A modern MAWPS is a switching power converter that has been designed to produce waveforms used for various arc welding processes. MAWPS control is challenging for
a number of reasons, including the complex dynamics of switching power converters,
transient conditions encountered in the metal transfer process, wide variations in load
impedance, a need for tracking complex reference waveforms, incomplete or inaccurate
models of the welding process itself, the difficulty of addressing the needs of several welding processes using a single machine, an electrically harsh environment with high
levels of electromagnetic noise, and health and safety concerns.
Arc welding control has been studied in the past, but most previous studies have
been done from two perspectives: that of manufacturing or process engineering, where Introduction 2 the interest lies in integrating welding into a larger manufacturing system, and that of metallurgical and materials engineering, concerned with the physics of joining the materials to be welded. Most published research in welding control treats the power source as a “black box,” and few substantial studies have been published in the open literature regarding the control of the welding power source itself. Of the literature that does exist regarding power source control, the majority has not addressed the dynamics of the welding process. Most of the suggested control strategies have been tested only in simulation, or if in an actual power source then only in contrived situations. Not much of the research has demonstrated how the principles investigated affect the welding performance of the machines.
One reason for the scarcity of research in welding power source control is that, until recently, sufficiently fast and precise data acquisition equipment was not available for studying the electrical signals involved in arc welding. Another reason is that few researchers have had access to the internals of a power source in order to study and modify it to test improvements.
The work described here attempts to model the important phenomena in the welding process as they relate to the control of the power source, and suggests a control strategy for achieving better performance than traditional controls. The new strategy is implemented by modifying a standard welding power source from Lincoln
Electric. The performance of the new strategy is compared to the performance of the existing controls.
1.1.1 Arc Welding Overview
The goal of welding is to join pieces of material by obtaining continuity between multiple elements composing a structure. Two elements to be joined are placed Introduction 3
Weld Bead
Filler Material
Base Material
Weld Pool
Spatter Joint
Workpiece
Figure 1.1. Illustration of arc welding terminology
adjacent to one another to form a joint. The material comprising the pieces to be
joined is called the base material, and the structure being welded is referred to as
the workpiece. Heat from an energy source is applied to melt the base material and
possibly a separate filler material into a volume of liquid metal along the joint called
the weld pool. The cooled and solidified weld pool is called the weld bead. Figure 1.1
illustrates these terms. The quality of a weld depends upon many factors including
the purity of the material in the weld bead, the depth at which the bead penetrates
the base material, and the presence of any solidified droplets outside of the weld bead,
referred to as spatter.
The most common energy source used for welding is an electrical arc [2], and welding processes using this energy source are called arc welding processes. This
is in contrast to other types of welding processes such as resistance welding, laser Introduction 4 welding, or oxyacetylene welding. In arc welding, a metal electrode is connected to one terminal of a welding power source, and the workpiece is connected to the other terminal. A low voltage, high current electric arc is established between the end of the electrode and the workpiece. The arc is a gaseous conductor through which current
flows to complete the welding circuit. The heat from the arc melts the base material and the filler material (if present) to form the liquid metal weld pool. The power source monitors the current and voltage in the welding circuit, and manipulates them according to logic designed to promote metal transfer in a particular way. If the weld is made properly, the materials to be joined will have become one continuous, joined piece.
Arc Welding Power Sources. Arc welding can be performed using nearly any electrical power source, including motor generators, batteries, and simple step-down transformers connected to mains power [2]. The most advanced arc welding power sources are switching power converters built with solid state power electronics. These machines can be configured using digital computers, and can be easily integrated with other industrial equipment such as factory automation and data acquisition devices.
Because they are highly configurable and typically controlled by reprogrammable soft- ware, these machines can be used to implement multiple arc welding processes using a single power source. A machine with this capability is referred to as a Multiprocess
Arc Welding Power Source(MAWPS). The high switching frequency of these power sources (in the range of 20 kHz to 200 kHz) allows high bandwidth control, smaller magnetics, and smaller size and weight for the machines compared to older technolo- gies based on generators or Silicon Controlled Rectifiers (SCRs). The efficiency of traditional arc welding power sources is typically 75 % to 85 %, translating to quite Introduction 5 large losses at high outputs [3]. Switching converters can have much higher efficiency, exceeding 95 % [4].
In the welding literature, a distinction is made between static and dynamic char-
acteristics of power sources. Static characteristics reflect changes that occur on the
order of 250 ms to 500 ms (typically related to the length of the arc), while dynamic
characteristics reflect changes at speeds measured in milliseconds or faster [5]. Since
the control of switching converters is very rapid in comparison (on the order of mi-
croseconds), both the static and dynamic properties of the power source can be con-
trolled [6].
Arc welding may be performed using Direct Current (DC) in either positive or
negative polarity, or using Alternating Current (AC). The welding polarity and hence
the direction of current flow affects the distribution of heat in the weld, with certain
polarity configurations being more desirable for particular arc welding processes. Gas
Tungsten Arc Welding (GTAW), for example, is typically done using Direct Current
Electrode Negative (DCEN) or AC, while GMAW and SMAW are typically Direct
Current Electrode Positive (DCEP) but may also be performed using AC.
Modern welding power sources based on solid state switching power converters
can be roughly placed into two categories: choppers and inverters.
Figure 1.2 shows the structure of a “chopper” welding power source. In this
topology, isolation from the AC input power is provided by a low-frequency step-down
transformer directly connected to the input voltage, at 50 Hz or 60 Hz. This input voltage is rectified and filtered to provideDC power to a digitally-controlled switching
section that runs at a rate of 20 kHz to 200 kHz. This high-frequency “chopped” signal
is smoothed by an output filter section to produce the welding output. Introduction 6
AC Input Welding Power Output
step-down capacitor output rectifier switch transformer bank filter
Figure 1.2. “Chopper” style welding power source structure
There are two chief advantages to the chopper architecture. First is that there are
fewer parts than an inverter, since the switching action is implemented using a single
switch rather than a four-switch H-bridge. Second is that the switching is done on
the output of the transformer, rather than the input, meaning that it is less likely for
the control system to cause transformer flux imbalance.
The basic structure of an inverter-style power source is shown in Figure 1.3. AC
input voltage at 50 Hz or 60 Hz is rectified and filtered to produce aDC voltage that
provides power to a full-bridge. The full-bridge switching devices on the transformer
primary side are controlled to produce the desired output waveform. The output of
the full bridge is a high-voltage, high-frequency (20 kHz to 200 kHz) AC signal that
goes through a step-down transformer, then is further rectified by an output rectifier
and smoothed by an output filter section, to produce the final welding waveform.
An advantage of the inverter topology over the chopper is that the step-down trans-
former can be physically much smaller because it operates at a higher frequency. A
disadvantage is that the switching devices are on the primary side of the transformer,
meaning that the control algorithms that produce the output waveforms for welding
must ensure that transformer flux imbalance does not cause transformer saturation.
Shielded Metal Arc Welding. In SMAW, the arc is established between a flux-
coated metal rod or “stick” electrode that is consumed during welding to provide the
filler material. Figure 1.4 shows a typical SMAW setup. SMAW is almost always Introduction 7
AC Input Welding Power Output
input capacitor step-down output output H-bridge rectifier bank transformer rectifier filter
Figure 1.3. “Inverter” style welding power source structure
performed manually. The welder manipulates the weld pool by moving the electrode
along the joint using his or her hands, and adjusts the arc length by changing the
distance from the end of the electrode to the workpiece. Since the arc length can vary widely during a weld, maintaining the desired current level in SMAW typically
demands a Constant Current (CC) control characteristic from the power source. In
addition to constant current regulation, SMAW requires well-controlled response to
short circuits (avoiding explosive rises in current upon load changes) and may require
power regulation in certain portions of the waveform in order to maintain constant
energy input while the arc length changes.
Metal transfer in SMAW was studied by Pistorius and Liu, who documented
relationships between welding voltage frequency content and the size of the metal
droplets that were transferred [7]. Blasco et al. discussed the SMAW process and the
desirable characteristics of a power source used for SMAW, concluding that a full-
bridge converter provides a good balance of size, weight, and welding performance [8].
Flux-Cored Arc Welding. FCAW uses a consumable wire electrode that is con-
tinuously fed into the weld pool using a motorized wire feeder. The FCAW wire is
hollow and contains flux that, when heated, produces gas for shielding the welding
arc. Additional shielding gas may also be used. The equipment used in a typical
FCAW setup is shown in Figure 1.5 The wire is fed at a constant rate and the arc Introduction 8
Welding Torch
Power Source
Flux-Coated Electrode
E W
Workpiece
Figure 1.4. SMAW setup
Electrode Wire Feeder wire spool
Anode and electrode wire
Power Source
Gun Gas nozzle Contact tip E W (optional) Gas Supply Weld pool (optional)
Workpiece
Figure 1.5. FCAW setup length is controlled by the wire feeding rate and the Contact Tip to Workpiece Dis- tance (CTWD). Metal is normally transferred in a spray transfer with infrequent short circuits [9]. Thus for FCAW the power source mostly sees the impedance of the arc itself, with anomalous short circuit events that must be handled without explosive changes in the current level that can result in spatter. Introduction 9
Electrode Wire Feeder wire spool
Anode, electrode wire, and gas hose
Power Source
Gun Contact tip Gas nozzle E W Gas Supply Weld pool
Workpiece
Figure 1.6. Gas Metal Arc Welding(GMAW) setup
Gas Metal Arc Welding. Like FCAW, Gas Metal Arc Welding(GMAW) uses
a consumable wire electrode. GMAW differs from FCAW in that the wire is solid
and the shielding for the arc always comes from an externally-supplied shielding gas
(commonly argon, carbon dioxide, or a mixture of these) rather than a flux inside the wire. Figure 1.6 shows a GMAW setup. GMAW is quite common because it can be
done manually but is also easily automated. GMAW processes are typically classified
according to how the metal is transferred from the end of the wire. These trans-
fer modes depend on the average current level and include short circuiting transfer,
globular transfer, spray transfer, and pulsed transfer.
Gas Tungsten Arc Welding. Gas Tungsten Arc Welding(GTAW) was developed
in the 1930s and 1940s by the aircraft industry for welding of sheet metals such as
aluminum and magnesium [10]. The GTAW electrode is a nonconsumable tungsten
rod. A separate filler material may be fed into the arc to be melted and deposited
into the weld pool. Figure 1.7 shows a GTAW setup. The arc length is controlled by Introduction 10
Welding Torch
Power Source
Tungsten Electrode
E W Filler Material
Gas Supply Workpiece
Figure 1.7. Gas Tungsten Arc Welding(GTAW) setup
moving the welding torch toward or away from the workpiece, and GTAW typically
requires a power source with aCC characteristic. To avoid tungsten contamination
in the weld pool, it is important that the electrode is not melted or fused to the workpiece. Short circuits are not expected in GTAW, and when they occur the power
source should not react as it does in other processes; the power source may reduce
the current level or turn off to prevent the tungsten “sticking” to the workpiece. In
Pulsed Gas Tungsten Arc Welding (GTAW-P), the current may be moved at a high
frequency between a high peak and a low background. In this case low overshoot and
undershoot when performing a current step are important.
Other Requirements for Welding Power Sources. Aside from welding processes
themselves, additional demands are sometimes placed on MAWPS output regulation
such as touch sensing, implementing metal cutting processes, and operating while
connected to resistive load banks for testing and calibration.
Touch sensing is used by a robot connected to an MAWPS, where the electrode is
at the end of the robot’s end effector. The power source regulates a low Open Circuit
Voltage (OCV) as the robot moves the welding torch around its work area. If the Introduction 11
torch touches the workpiece, a small current begins to flow. The power source detects
this flow of current and informs the robot that the torch has touched the workpiece.
This can be used as a method of positioning the robotic arm when making welds.
Good OCV regulation and low-end current regulation are required to perform this
effectively.
Because welding and cutting of metals are frequently done together, it is common
for a MAWPS to implement Carbon Arc Gouging (CAG), a non-welding process used
for cutting metal. The heat from an arc melts the metal and a stream of air pushes
the molten metal out of the way. Gouging is typically done at very high output levels
and causes frequent transitions between open and closed circuits as the metal is cut
away.
For calibration and testing, a MAWPS is often connected to a load bank with
a fixed impedance. This allows a technician to calibrate the voltage and current
feedback by regulating a constant output into a known load and comparing the power
source readings with external meters.
1.1.2 The Need for Improved Controls
The Bureau of Labor Statistics (BLS) indicates that in 2010 there were 337,300 welding, cutting, soldering, or brazing jobs in the United States, and this is projected
to increase 15 % by 2020. This figure is slightly above the average growth projected
for all occupations (14 %), and is much higher than the 4 % average projected increase
in production occupations [11]. The American Welding Society (AWS) found that a large majority of welding industry personnel believe a lack of skilled welders is presently a problem and will continue to be a problem in the long-term [12]. Some
in the industry see increased automation as a solution, while others believe more Introduction 12
intelligent power sources that require less training to operate will be required. Both
situations could benefit from more advanced control systems. According to Norrish, welding control improvements are needed due to a paucity of skilled welders, a need for
continuous improvement in occupational health and safety, and competitive pressures
to improve productivity and reduce cost [13].
Advances in materials and metallurgy are bringing new alloys to industry, and the
present level of knowledge about how to weld these materials is often insufficient. New welding processes are being developed to take advantage of these materials, placing
new demands on power source control systems [12]. The AWS describes development
of materials as one of the key drivers of the future of the welding industry. These new
materials are developed to have increased strength, be lighter, and be more resistant
to corrosion. From [14]:
“Improvements in the quality and reliability of joints will help over-
come the image problem of the welded joint as the weakest link
in any structure... Processing research, artificial intelligence and
robotics, advanced materials, and other developments outside the
traditional scope of welding can all be applied to making dramatic
improvements in welds.”
Paul suggested that the goal of the power source is to reduce the diversity of
the welding process by making as many factors as possible robust across parameter variations [15]. Paul proposes the concept of a “goodness factor” for welding equip-
ment, which measures how appropriate the equipment is considering settling time of
reference tracking, reduction of losses in the welding cables, high power factor, and
high efficiency. Introduction 13
1.1.3 Challenges for Control
MAWPS control presents several challenges that makes its study interesting and pro- vides avenues for research in several fields of systems and control.
Switching Power Converter Dynamics. The dynamics of switching power con- verters is nonlinear and time-varying. While the electric circuits that make up power
converters can be simple, they can exhibit complex dynamics including bifurcations,
chaos, and period doubling [16].
Metal Transfer Transients. Because the purpose of welding is to transfer metal,
there is rarely a steady state achieved during welding. Instead, transient phenomena
are the norm. It is crucial to understand the way the power source control system
handles transients in metal transfer such as current pulses to promote molten metal
ball formation and transfer and short circuit events that occur when the electrode
contacts the workpiece.
Wide Load Impedance Variations. A welding power source is required to regulate voltage into an open circuit (in reality, a very high impedance load), and is also
required to regulate current into a very low impedance load (referred to as a short
circuit). Additionally, the electric arc has a nonlinear conductance that must be
properly handled by the power source control system.
Reference Tracking. In many welding processes, it is not sufficient to regulate to
a fixed setpoint such as a constant current or a constant voltage. Instead, periodic
events occur that require specific waveform shapes such as pulses, current ramps, etc.
to be tracked by the power source.
Incomplete and Inaccurate Models. There are many variables involved in the welding process, and it is challenging to combine the models of the power source, Introduction 14 welding circuit, metal transfer process, and weld pool dynamics in a way that produces
a detailed yet tractable model of the welding process. Simplifying assumptions are
necessary to make it feasible to analyze and simulate the welding process, and these
assumptions can limit the scope of applicability of the models.
Electrically Harsh Environments. Due to the high levels of electrical current
involved in arc welding, electrical noise is a primary concern in the design of welding
power sources. Both the hardware and the control system must take precautions to
ensure that electrical noise emitted from the power source is of an acceptable level,
and that interference from the welding process itself does not affect the operation of
the welding power source or surrounding equipment.
Health and Safety Concerns. The International Electrotechnical Commission
(IEC) publishes requirements for welding power sources in terms of health and safety.
Welding can be a dangerous process that can result in electrical shock or electrocu-
tion, burns, and inhalation of fumes. All of these things can be a factor in the design
of the power source and its controls.
1.2 Literature Review
1.2.1 Historical Overview
Metal-joining processes that could be considered welding were known as far back
as the Bronze Age, but the earliest known use of an electric arc for joining metals
occurred in the 19th century [2], shortly after the electric arc was discovered inde-
pendently by Sir Humphry Davy and Vasily Vladimirovich Petrov [17]. Carbon arc welding, a process by which metals are joined by melting them with an electric arc
between a carbon electrode and the workpiece, was created in 1881 by Bernardos and Introduction 15
Olszewski, and the first US patent for arc welding with bare metal electrodes was
awarded to Charles Coffin in 1890 [18]. In 1926, the use of helium gas as a shield
for the welding arc was described by Hobart and Devers. Shielded metal arc weld-
ing, commonly referred to as “stick welding,” became available in the 1930s, as did
Submerged Arc Welding(SAW).
In 1939 and 1943 A.M. Cassie and O. Mayr, respectively, published important
papers describing the electric arc, which would become the basis for many mathe-
matical models to follow [19]. Also around this time, welding became an important
part of manufacturing for the military efforts in World War II.
The 1950s saw many developments in Gas Metal Arc Welding(GMAW), including
a patent for spray transfer arc welding and the use of carbon dioxide as a shielding
gas. Plasma arc cutting, a process for precision cutting of metals using equipment
similar to welding, was first demonstrated publicly in 1956.
In 1960 the first semiconductor power device, the Silicon Controlled Rectifier
(SCR), became available. The SCR would become the basis for many industrial welding machines designed in the decades to follow. Pulsed GMAW was developed
in the 1960s, but would not become commonly used in industry until decades later.
“Synergic” GMAW was patented in 1964 by Manz. The first robotic welding instal-
lations also occurred during the 1960s.
In 1974 the US Maritime Commission supported Jim Thommes’ first-generation
prototype welding inverter, a technology that would dominate welding power source
technology beginning in the 1990s. The 1970s in general was an important decade for
the development of power electronics, and many of the analysis and control techniques
developed then are still in use today. Introduction 16
1881 - One of the earliest carbon arc welding machines invented by De • Meritens (France). 1885 - Carbon arc welding developed by Bernardos and Olszewski (Russia). • 1889-90 - First arc welding with bare metal electrodes (C.L. Coffin, USA). • 1926 - USA patents issued to M. Hobart and P.K. Devers for helium-shielding • in arc welding. 1930s - Shielded metal arc welding and submerged arc welding become avail- • able. 1941 - GTAW developed by Meredith (USA). • 1943 - Otto Mayr published an important paper describing a model of the • electric arc, focusing on the arcing phenomenon in electric relays. 1950 - First patent for spray transfer arc welding awarded to Muller, Gibson • and Anderson. 1956 - Plasma arc cutting first displayed publicly. • Late 1950s - CO2 used as a shielding gas for GMAW. • 1960 - The Silicon Controlled Rectifier(SCR), the first power semiconductor • device, becomes available [20]. 1960s - Development of pulsed GMAW. • 1961 - First robotic welding installation. • 1964 - Synergic GMAW patented by Manz. • 1974 - US Maritime Commission supports Jim Thommes’ first-generation • prototype welding inverter. 1990s - Solid-state inverter technology begins to dominate. • 2000s - Software-controlled machines become commonplace. • Table 1.1. Timeline of welding power source technology
1.2.2 Classical Approaches to Power Electronics Control
Classical approaches to switching power converter control are based on circuit av- eraging or state-space averaging of the switching action and linearization about an operating point [21, 22, 23]. Averaging techniques have been in widespread use since the 1970s. Using averaging methods, each possible combination of switching device position is considered as a separate circuit which leads to a set of differential equations for each configuration. These equations are “averaged” together using a duty cycle to transform the variable structure system into a continuous system. This Introduction 17
transformed system can then be linearized about an operating point to allow for Lin-
ear Average Control (LAC), normally using frequency-domain techniques, i.e. Bode
plots and gain and phase margins.
A 2003 paper by Guo, Hung, and Nelms discussed the use of root locus tech-
niques to design discrete-time controllers for switching converters using averaging
methods [24]. They were able to demonstrate good performance compared with controllers designed using frequency-domain techniques.
Sun and Grotstollen pointed out that averaging methods do not work well when a
Pulse Width Modulation (PWM) converter is operating in Discontinuous Conduction
Mode (DCM), and they cannot be applied to certain architectures such as resonant
or quasi-resonant converters [25].
It is well-known that classical approaches suffer from lack of robustness to input
power variations, output load variations, and parameter variations in the converter
components. Some researchers have addressed these issues by introducing nonlinear
extensions of classical control techniques. For example, Zhu investigated the applica-
tion of nonlinear Proportional-Integral-Derivative (PID) control, in which a nonlinear
combination of the proportional, integral, and derivative error is used to produce a
control signal to aDC-DC converter [ 26]. Compared to a traditional Proportional-
Integral (PI) controller, the nonlinear PID was found to be easier to tune and more
robust for a 1 kWDC-DC converter. Other attempts to address the robustness issues
of classical techniques include adaptive methods [27, 28] and gain scheduling [29].
Higuchi developed a multi-stage digital controller for a switching power converter that
estimated the load current and switched bumplessly between different controllers to
improve performance at regulating into an inductive load [30]. Introduction 18
1.2.3 Nonlinear Approaches to Power Electronics Control
Because of the difficulties with classical methods applied to power electronics, most
research in recent decades has focused on nonlinear forms of control such as those
using artificial intelligence (e.g. Fuzzy Logic Control(FLC), Artificial Neural Net- works (ANNs)) or variable structure techniques (e.g. Sliding Mode Control(SMC)).
Sometimes these approaches are combined to obtain desirable actions from several
types of controllers. Escobar et al. compared LAC to Fuzzy Logic Control (FLC) and various forms of SMC and found LAC to perform particularly poorly [31].
Fuzzy Logic Control(FLC). Fuzzy Logic Control(FLC) is an outgrowth of the
theory of fuzzy sets, formalized in 1965 by Zadeh [32]. Instead of describing systems
using numerical variables, fuzzy logic uses linguistic variables, i.e. variables whose values are sentences in a natural language. This is formalized mathematically using
the concept of a fuzzy set. Unlike crisp sets, elements are allowed to partially belong
to one or more fuzzy sets. The advantage of FLC is that it provides a mathemati-
cal framework for specifying human preferences for control system performance that would otherwise be difficult to express using analytical formulas [33]. An example
of a rule executed by a fuzzy controller would be “If setpoint error is positive and
large and the error change is positive and small, then the actuator output should
be negative and large.” These types of rules are most commonly used as high-level
supervisory controllers.
Advantages of FLC are that it does not require an accurate model of the system,
and the controller design relies only on intuitive descriptions of desired performance
instead of analytical calculations that may be difficult to perform. Although some
see this as an advantage, fuzzy logic was famously criticized by R.E. Kalman and Introduction 19
W. Kahan for lacking mathematical rigor [34]. A disadvantage of FLC is that it is difficult to implement the complex rules without using a microcontroller.
Some researchers have successfully applied FLC to power electronics, e.g. [35, 36].
Iskender and Kaarslan showed how an arc welding power source based on a two- switch single-phase inverter could be controlled using FLC or SMC with state-space averaging. In comparing the two methods, they found FLC to outperform SMC.
Their test, performed in simulation, assumed a linear, purely resistive load of 0.2 Ω and examined a step current change from 80 A to 110 A.
Golob, Koves, and Tovornik discussed the application of fuzzy logic to GMAW process control and weld process quality monitoring, but not to the control of the power source itself [37].
A 2002 paper by Viswanathan compared the performance of a “universal” fuzzy logic controller to a universalPI controller and a set of conventionalPI controllers tuned for specific operating points (output load and input line conditions) for a boost converter. The universal controllers had a single set of tuning parameters based on a nominal operating point. The fuzzy controller was found to perform better than the universalPI controller tuned for worse-case conditions, and about the same as the set ofPI controllers tuned for specific operating points [ 38].
Another 2002 paper by Balestrino discussed the application of fuzzy logic to thePI controller for a Cuk´ converter. FLC was used as the high-level supervisor to vary the coefficients of a conventionalPI controller, changing its action based on the difference between the actual and the desired output voltage [39].
Neural Networks. ANNs have been applied to inverter control. The idea is to train the network to create binary outputs for controlling the switching devices through a Introduction 20
nonlinear mapping of the input signals [40]. The chief advantage of neural networks in
control is that they are able to easily implement complex, nonlinear transfer functions.
However, they can be difficult to implement and require offline training before they
can be used [41].
Genetic Algorithms. Schutten and Torrey described the application of genetic al-
gorithms to switching power converter control, using the technique to optimize a
performance index [42].
Model-Predictive Control. Model Predictive Control (MPC) applied to switching power converters was documented by Geyer [43] and Kouro et al. [44].
Active Disturbance Rejection Control. Gao and Sun developed a control algo- rithm based on Active Disturbance Rejection Control (ADRC) which they applied to a 1 kWDC-DC converter and found to work well in rejecting disturbances and being easy to tune [45].
Fractional-Order Control. Fractional Order Control (FOC), based on the princi- ples of the fractional calculus, was proposed for the control of a buck converter by
Calder´on,Vinagre and Feliu [46]. In a related work, the same authors suggested the use of fractional order sliding surfaces when using SMC forDC-DC converter control [47].
1.2.4 Sliding Mode Control(SMC)
The general principle of SMC is to design a sliding manifold that causes the system’s state trajectory to move toward a desired operating point. SMC operates in two phases: a reaching phase and a sliding phase. In the reaching phase, the controller forces the trajectory from an arbitrary initial condition to the sliding manifold. Once the sliding manifold has been reached, the trajectory moves along it toward the desired Introduction 21
operating point. A reaching condition ensures that regardless of the initial condition,
the trajectory is directed to reach a vicinity of the sliding manifold. The sliding
manifold is subject to an existence condition which specifies that when the trajectory
is within a small distance of the sliding manifold, it will be “pulled” toward it at which
point the controller operates in the sliding phase. Once operating in the sliding phase,
a stability condition ensures that the trajectory will always move toward the desired
operating point. The controller designer has some freedom in designing the sliding
manifold provided these conditions are satisfied.
For a switching power converter, moving along the sliding manifold can be done
by opening and closing the switching devices at a high frequency (ideally infinitely
fast). Because of the practical impossibility of implementing infinitely fast switching
due to switching losses, Electromagnetic Interference (EMI) concerns and difficulty of
magnetics design, researchers have proposed alternatives such as hysteresis modula-
tion or quasi-sliding mode fixed frequency control which exhibit many of the benefits
of SMC with somewhat degraded performance from the ideal.
SMC grew out of the theory of variable structure systems, and began to be rec-
ognized in the 1960s and 1970s as described by Utkin [48]. The specific application
of SMC to power electronics was studied in the 1980s, e.g. [49, 50, 51, 52] and
in the 1990s, e.g. [53, 54, 55]. By the late 1990s it was apparent that the the-
ory of SMC had become well understood and that, in order to see more widespread
use, practicing engineers needed to be educated about how to integrate it into their work [56]. Today, SMC is beginning to be accepted as a prominent control strategy
for power electronics due to its inherently variable structure and its ability to pro-
duce controllers that are robust across a wide range of parameter variations as seen Introduction 22
in power electronics [57, 58, 59, 60, 61]. In 2008 Tan, Lai, and Tse pointed out
that SMC applied toDC-DC converters has been heavily researched but needs more
practical applications [62]. A 2012 book by the same authors presents the theoretical
and practical aspects of SMC applied to switching power converters [41].
In 1995, Nguyen published a paper where SMC was used to control a buck con- verter to track an arbitrary voltage reference signal with adaptive hysteresis [63].
Iskender and Karaarslan compared SMC with FLC for an arc welding power source
in a 2005 work [64].
In classical control methods using fixed-frequency PWM, PWM signal design is
usually performed as a separate step from controller design. That is, the dynamics
of the converter are not considered when designing the PWM switching patterns.
Sabanovic-Behlilovic et al., in a paper describing the application of SMC to three-
phase inverters and rectifiers, noted as an advantage over classical methods that
SMC allows the switching pattern and the controller design to be performed simulta-
neously [65]. SMC applied to a single-phase inverter was described by Jezernik and
Zadravec [66].
Because SMC inherently handles the variable structure of the switching power
converter, it does not suffer from the drawbacks of classical methods when converters
are operating in DCM. SMC tends to be easy to implement in either analog or digital
domains when compared to other advanced control techniques, and it is robust across
parameter variations.
A disadvantage of SMC is that the variable frequency it requires has practical
problems for implementations including increased switching losses at high frequencies
and complicating the design of transformers and output inductors. Some authors note Introduction 23
that the variable frequency can cause problems for Electromagnetic Compatibility
(EMC)[ 41]. However, Tanaka, Ninomiya and Harada proposed a random-switching
control method for a switching converter in order to reduce the power of the switch-
ing harmonics in the output frequency spectrum [67]. In their method, a random
perturbation was deliberately added to the switching instant of the converter, and
this was found to spread the power of the noise spectrum more evenly which may
actually be beneficial for EMC.
1.2.5 Reference Tracking
Most of the literature concerning switching power supply control assumes only that
a constant output voltage is of interest. One exception is the work done by Nguyen, where SMC was used to control a buck converter to track an arbitrary voltage refer-
ence signal with adaptive hysteresis [63]. This is one of the few resources describing
the tracking problem for switching power converters.
1.2.6 The Welding Power Source
Much of the published research regardingDC-DC converters, switching inverters and
switching rectifiers applies to the design of welding power sources. Some designs
specifically for welding have been discussed in the power electronics literature. Mecke,
Fischer, and Werther presented a design for a phase-shifted, Zero Voltage Switch-
ing (ZVS)/Zero Current Switching (ZCS) full-bridge inverter intended to minimize
switching losses in arc welding applications [68]. Aigner, Dierberger and Grafham
compared different topologies for Arc Welding Power Source (AWPS) use including
series resonant converters, quasi-resonant ZCS converters, and multi-resonant ZCS
and ZVS converters in terms of their desirability for arc welding applications [69]. Introduction 24
More recently a paper by Klumpner and Corbridge discussed a two-stage design where
the inverter bridge always operates at 50 % duty cycle and a prior stage regulates the
input voltage for this inverter [70]. Wang, Wang, and Xu described the design of a
double-loop control system for an AWPS with an outer voltage control loop providing
a reference for an inner current loop [71].
1.2.7 Power Source Dynamics and Their Effect on Welding
Several researchers have commented on the effect of the power source dynamics on metal transfer in GMAW. Richardson, Bucknall, and Stares explored the shape of a pulsed GMAW waveform having a trapezoidal current shape [72]. They found that two trapezoidal pulse waveforms having the same mean current can have different wire melting rates depending on the slew rate of the current, since this affects how much time is spent in the pulse peak and thus how much current contributes to I2R heating of the wire: faster current response leads to smaller liquid metal droplets. They also compared the behavior of the idealized trapezoidal pulse with the exponential shaped pulse that is more commonly seen in actual machines. Similar results had been found earlier by Yamamoto, Harada, and Yasuda [73]. Joseph et al. compared the behavior
of the pulse waveforms of four commercially available power sources, and found the
machines to produce different results with similar settings due to the differences in wave shapes [74]. Yarlagadda et al. described a method of detecting short circuits in
pulsed GMAW by combining digital signal processing of current and voltage signals with high-speed videos of metal transfer occurring [75].
Arc length control and metal transfer have been heavily studied for GMAW. Ex-
amples of more theoretical works include those by Modenesi and Nixon, Bingul, and
Thomsen. Modenesi and Nixon’s work focused on the instability in metal transfer Introduction 25
that occurs with abrupt changes in current and/or voltage during welding. Their work also considered the effects of the shielding gas on this instability [76]. Bingul
developed a model explaining that the metal transfer mode and the V-I characteristic
of the arc both play a role in the stability of arc length regulation for GMAW[ 77].
Thomsen developed a nonlinear arc length control system for GMAW using the feed-
back linearization method [78].
1.3 Goals of this Work
From the beginning, the goals of this work have been the following:
Understand the components of a welding setup, how they interact, and how • they can be modeled, focusing on how each component affects the control of
the welding power source.
Develop a simulation environment that can be used to explore the compo- • nents of the welding setup and to test different control strategies.
Develop a new power source control strategy that improves upon the ex- • isting controls in terms of reference tracking and handling of varying load
impedance.
Implement the new strategy in hardware by retrofitting an existing welding • power source, in order to prove that it is feasible and to test its performance
against the existing controls.
Compare the new control strategy to existing controls, in scenarios that • reflect events that occur during actual weld processes.
As will be explained throughout the thesis, some of these goals have been achieved
and, for others, further work is needed. Introduction 26
1.4 Contributions of this Work
This work contributes to the existing literature on SMC applied to switching power
converters by considering output current regulation in Section 4.3 and power regula-
tion in Section 4.5 in addition to output voltage regulation in Section 4.4. Addition-
ally, the focus is placed on a non-resistive load impedance, and on transient response
and reference tracking rather than only the stability of the closed-loop system as
commonly encountered in the literature. Section 4.7 describes a method of applying
constraints to the SMC strategy, forcing the system trajectory to stay within defined
limits. This type of constrained control has been briefly described in the general
case [79] but has not been applied to switching power converters.
Chapter 2 expands upon the models of welding equipment in the literature by
collecting models for all of the components of a welding setup into one place. The
simulation environment, described in detail in Appendix A, provides a way for en-
gineers to experiment with new control strategies and to easily test performance in
different scenarios.
The hardware implementation described in Chapter 5 gives a practical industrial
application of SMC using readily-available electronic components. This implementa-
tion can be used as a guideline for other researchers looking to implement SMC in
switching power converters.
1.5 Thesis Organization
Chapter 2 presents mathematical models for the various components of a welding
system, including the electric arc, welding fixture, cables, and the power source itself. Introduction 27
Chapter 3 reviews the performance of the existing control system, and discusses ways the performance could be improved. A proposed new control strategy based on Slid- ing Mode Control(SMC) is presented in Chapter 4. This strategy is implemented in hardware, and the implementation is described in Chapter 5. Experimental results comparing the existing system with the new control strategy are described in Chap- ter 6. The thesis concludes in Chapter 7 with a review of the work completed and suggestions for further work. 28
2 Modeling
In this chapter, models of the welding circuit and power source are derived from
basic principles. The welding circuit is first considered without regard for the power
source, accounting for all of the other impedances present in the circuit. The power
source is then modeled with the welding circuit as its load, using switching power
converter analysis techniques from power electronics.
The welding circuit comprises all of the components of a welding setup through which electrical current flows during welding. The circuit diagram in Figure 2.1 shows
the power section as a current source iw(t) supplying current to the welding circuit,
and illustrates the impedances in the circuit: the cables carrying current from the
power source to the arc, having an impedance Zc; the resistance Rt of the point at which the electrode comes into contact with the current in the cable (in the case of
consumable electrode processes); the resistance Re of the electrode itself; the welding
arc with conductance Ga; and Zp, the impedance of the workpiece and welding fixture.
2.1 Welding Cables, Workpiece, and Welding Fixture
Due to the high amperage required for arc welding, large cables are used between the
power source and workpiece; American Wire Gauge (AWG) 0000 or AWG 000 are Modeling 29
Zc Rt
cables contact point Re electrode
power source iw(t)
Ga arc
Zp
workpiece
Figure 2.1. Circuit diagram showing sources of impedance in the weld- ing circuit
common sizes. After the arc itself, the welding cables are typically the most significant
contributor to the impedance of the welding circuit. This impedance can vary widely
between setups. In some cases the power source is positioned near the workpiece with
short, straight cables, in which case the cable impedance has a negligible effect. In
other situations such as at construction sites, very long cables coiled around metal
structures may separate the power source from the workpiece. Cable impedance can
have a significant effect on the quality of welds as discussed by Peters and Allgood [80].
The cable is assumed to be two parallel conductors of length `c, each with radius
rc, separated by a distance dc as shown in Figure 2.2. The impedance can be analyzed
using a simplified form of Heaviside’s transmission line model, as shown in Figure 2.3.
The model consists of a resistance Rc, capacitance Cc, and inductance Lc [81].
The resistance Rc of the cables can be computed from the conductor resistivity
ρc, radius rc, and cable length `c:
`c Rc = ρc 2 . (2.1) πrc Modeling 30
µins µins
c
rc rc dins dins ρc ρc
dc
Figure 2.2. Welding cable cross-section showing electrical and geomet- rical properties
Rc Lc
Cc
Figure 2.3. Circuit diagram for welding cable transmission line model
Welding cables are typically stranded copper (ρ 1.75 10−8 Ω m) or stranded alu- ≈ × minum (ρ 2.83 10−8 Ω m) with a thick (0.25 mm to 1.75 mm) insulator made of a ≈ × synthetic rubber such as neoprene or EPDM[ 81, 82]. Figure 2.4 shows the relation-
ship between cable length and resistance for both copper and aluminum AWG 0000
cables. Modeling 31
Figure 2.4. AWG 0000 welding cable resistance versus cable length
The approximate inductance of the transmission line in microhenries is given by [83] dc 1 dc Lc = 0.004`c ln + . (2.2) rc 4 − `c Figure 2.5 plots the inductance versus cable length for AWG 0000 welding cables separated by various distances.
The approximate capacitance in farads of the welding cables is given by [81]
π`c Cc = . (2.3) ln(dc/rc) Modeling 32
Figure 2.5. AWG 0000 welding cable inductance versus cable length
In Figure 2.6, the capacitance versus cable length is plotted for AWG 0000 welding
cables separated by various distances.
The welding cables effectively introduce a harmonic oscillator into the welding
circuit, where the relationship between the voltage at the power source Vw(s) and the voltage at the end of the cables Va(s) is given by
1/LcCc Va(s) = Vw(s). (2.4) s2 + Rc s + 1 Cc LcCc This system has a resonant frequency given by
1 fn = . (2.5) 2π√LcCC Modeling 33
Figure 2.6. AWG 0000 welding cable capacitance versus cable length
Figure 2.7 shows how the resonant frequency decreases as the cable length increases,
becoming close to typical switching frequencies as the length approaches 200 m. This
implies an inverse relationship between cable length and control system bandwidth:
as the cable length increases, the bandwidth of the control system must decrease to
avoid amplifying the resonant frequency of the welding circuit.
2.2 Electrode and Contact Tip
There is some resistance associated with the consumable SMAW electrode or the
nonconsumable tungsten electrode in GTAW. In wire-fed processes such as GMAW Modeling 34
Figure 2.7. AWG 0000 welding cable resonant frequency versus cable length
and FCAW, the wire electrode is fed through a tubular conductor that is connected to the welding cable as shown in Figure 2.8. The point where the tubular conductor comes into contact with the electrode has an associated resistance. These consumable- electrode processes typically demand a consistent Contact Tip to Workpiece Distance
(CTWD), which describes the length of electrode wire that should extend beyond the contact tip during a weld. While this length of electrode is short its cross-sectional area is small, and due to the high amperage in the welding circuit its resistance is non- negligible. Computing the resistance of the contact tip and electrode is complicated Modeling 35
gas nozzle
contact tube
workpiece electrode
Figure 2.8. Wire electrode fed through contact tube as in GMAW and FCAW
by the fact that the resistivity of metals varies as a function of temperature [84]:
`e Re = ρe(T ) 2 , (2.6) 2πre where ρe(T ) is the resistivity of the electrode material as a function of temperature
T , `e is the length of the electrode that extends beyond the contact tip, and re is the
radius of the electrode.
2.3 The Welding Arc
The welding arc is a gaseous conductor, and the relationship between voltage and
current in an arc is nonlinear [2,5, 85 ]. At low current, the total potential of
the arc falls as current increases. At some point, the potential reaches a minimum
and then begins to rise with increasing current. After this point the relationship
is essentially ohmic. In general the potential increases as the gap between the arc Modeling 36
terminals increases, and in practice welders generally consider the arc voltage to be
proportional to arc length. Several researchers have created models of the arc leading
to a nonlinear relationship between voltage, current, and arc length. Figure 2.9 shows
the relationship between voltage and current for a GMAW arc at various arc lengths
as reported by Bingul [84].
There are many factors that affect the electrical characteristics of the arc, includ-
ing the shape and material makeup of the electrodes, the ambient temperature and
pressure, surrounding electromagnetic fields, gas composition, etc. [86]. Hence, the
published models of electric arcs make many simplifying assumptions.
The most well-known electric arc models are those of Mayr and Cassie. The
Mayr model gives an expression for the conductance of an arc Ga in terms of the arc voltage va and current ia and a steady-state “cooling power” dissipated by the arc to the environment: d ln G 1 v (t)i (t) a = a a 1 (2.7) d t τa P0 − where τa is the arc time constant and P0 is the cooling power dissipated by the arc
to the environment at steady-state.
The similar Cassie model gives an expression for the conductance of an arc in
terms of the voltage and a steady-state arc Electromotive Force (emf)
2 d ln Ga 1 v = 1 2 (2.8) d t τa − V0 where V0 is the steady-state arc emf.
Tseng et al. suggested combining the Cassie and Mayr models using a smooth,
monotonically decreasing function of the current such that the conductance is given Modeling 37
60 10 mm arc length 8 mm arc length 55 4 mm arc length
50
45
40 volts
35
30
25
20 150 200 250 300 350 amperes
Figure 2.9. V-I characteristics for GMAW arcs of various lengths
by i2 vi i2 i2 d G Ga = Gmin + 1 exp 2 2 + exp 2 τa (2.9) − −It V0 −It P0 − d t which tends to the Cassie model at low currents, and the Mayr model at high currents.
This model was experimentally demonstrated by Sawicki et al. to match welding
transformer arc characteristics in several scenarios [87].
2.4 Summary of Load Impedances
Given the descriptions of the various welding circuit impedances in the previous sec-
tions, it would seem intractable to develop a model that encompasses all of them.
Indeed, it will be shown that for the purposes of developing control strategies for
a MAWPS, it is sufficient to consider the stability and tracking performance across
ranges of impedances, or under specific situations that can be modeled using simpli-
fying assumptions. Modeling 38
2.5 Piecewise Linear Model
Classical systems and control theory provides a collection of tools for analyzing and
designing Single Input Single Output (SISO), linear systems that do not change over
time (i.e. Linear Time-Invariant (LTI) systems). In order to use these tools for power
electronics systems, which are nonlinear and time-varying, it is necessary to make
simplifying assumptions about the systems and develop an approximate linear, time-
invariant model. This model typically works well enough in practice when working with a simple converter with a single operating point, but becomes cumbersome and
inaccurate when a converter must operate across a wide range of input and output
power levels.
Usually, an averaging approach can be applied to switching power converters that
are controlled by PWM. The basic strategy is to develop a model of the circuit
equivalent to each possible switch position, then combine these into a single model
by a weighted average, with the PWM duty cycle giving the weight applied to each
switch position.
The drawback of the traditional averaging approach is that it results in only a
small-signal model and does not accurately describe the large-signal response of the
circuit. Furthermore, because the linearization is performed about a specific operating
point, the controllers developed using this technique are not very robust across ranges
of output loads, input power variations, and/or variations in the parameters of the
circuit elements.
The welding power source power section and load can be modeled using the circuit
diagram in Figure 2.10, which resembles a traditional buck DC-DC converter with a Modeling 39
resistive-inductive load. This is a variable-structure system containing two subsys-
tems, one of which is active at any moment depending on the position of the switch
S.
When S is closed (u = 1), the circuit in Figure 2.11 results. This circuit is
described by the set of linear ordinary differential equations
d iL 1 1 = vw(t) + vg(t) (2.10) d t −L L d vw 1 1 1 = iL(t) vw(t) iw(t) (2.11) d t C − RoC − C
d iw 1 Rw = vw(t) iw(t). (2.12) d t Lw − Lw
This system can be written in matrix form as
˙x(t) = Ax(t) + B1vg(t) (2.13) where iL(t) 0 1/L 0 1/L − x(t) = v (t) A = 1/C 1/R C 1/C B1 = 0 . (2.14) C o − − iw(t) 0 1/Lw Rw/Lw 0 − When the switch is open (u = 0), the circuit in Figure 2.12 results. This circuit is
described by
d iL 1 = vw(t) (2.15) d t −L d vw 1 1 1 = iL(t) vw(t) iw(t) d t C − RoC − C
d iw 1 Rw = vw(t) iw(t) d t Lw − Lw Modeling 40
L iL(t) iw(t) S
Lw
+ vg(t) D C Ro vw(t) −
Rw
Figure 2.10. Circuit diagram of power section and load
L iL(t) iw(t)
Lw
+ vg(t) C Ro vw(t) −
Rw
Figure 2.11. Circuit diagram of power section and load when switch is closed
which, in matrix form, can be written
˙x(t) = Ax(t) + B2vg(t) (2.16) where x(t) and A are as in (2.14) and 0 B2 = 0 . (2.17) 0 Modeling 41
iw(t)
iL(t) Lw
L C Ro vw(t)
Rw
Figure 2.12. Circuit diagram of power section and load when switch is open
2.5.1 Power Section Waveforms
Suppose the voltages vg(t) and vw(t) are constants, i.e. vg(t) = Vg and vw(t) = Vw.
This is a valid assumption when the time interval under consideration is short, or the converter is operating in steady-state. Then (2.10) gives
d iL Vg Vw = − (2.18) d t L which has solution
Vg Vw iL(t) = iL(t0) + − (t t0). (2.19) u=1 L −
This shows that when the switch is closed, the choke current iL(t) increases linearly with a ramp rate proportional to the voltage difference Vg Vw and inversely propor- − tional to the choke inductance L. A higher input voltage tends to increase the ramp
rate, while a higher output voltage or a higher inductance tends to decrease the ramp
rate. Assuming Vg > Vw is always true, the current will always increase when the
switch is closed. Modeling 42
When the switch is open, Vg disappears from (2.18) and
d i V L = w (2.20) d t − L
results. This has solution
Vw iL(t) = iL(t0) (t t0). (2.21) u=0 − L −
Thus the current decreases linearly at a rate proportional to Vw and inversely pro-
portional to L. Figure 2.13 shows the relationship between switch position and choke
current described by equations (2.19) and (2.21).
2.5.2 Welding Current and Voltage Waveforms
Consider the Laplace transform of equation (2.12),
1 Rw siw(s) = vw(s) iw(s). (2.22) Lw − Lw
Solving for iw(s) gives
1/Rw iw(s) = vw(s). (2.23) 1 + Lw s Rw
Equation (2.23) has the form of a first-order low-pass filter with time constant Lw/Rw
and gain 1/Rw. This shows that the welding current can be considered a scaled and
filtered version of the welding voltage. The frequency response of this filtering effect
is shown in Figure 2.14 for a fixed resistance Rw = 100 mΩ.
The filter time constant increases as Lw increases, or as Rw decreases. Notice that
vw(s) lim iw(s) = , s→0 Rw which shows that at steady-state the inductance has no effect and the current is
related to the voltage through the resistance Rw of the welding circuit, as expected
from Ohm’s law. Similarly, as Lw 0, the current iw(s) vw(s)/Rw. → → Modeling 43
iL
Vg Vw m1 = − L m = Vw 2 − L
iL(t0)
t
t0
u
1
0 t
t0 t0 + dTsw t0 + Tsw
Figure 2.13. Relationship between switch position u and choke current iL(t)
2.5.3 State-Space Average Model
The two subsystems described by (2.13) and (2.16) can be combined into a single
system using the state-space averaging technique, originally described by Middlebrook
and Cuk´ [21]. Suppose the switch S is closed during the interval t [0, t0]. During ∈ this time the system (2.13) is active. The solution of this system is given by
Z t At A(t−τ) x(t) = e x(0) + e B1vg(τ) d τ t [0, t0]. (2.24) 0 ∈ Modeling 44
Figure 2.14. Current filtering effect of the welding circuit impedance
At t = t0 the switch opens, and it is open during the interval t [t0,Tsw]. During ∈ this interval the system (2.16) is active, which has solution
A(t−t0) x(t) = e x(t0) t [t0,Tsw]. (2.25) ∈
Although the system changes structure at t = t0, its state x(t) is continuous. Using
this fact, we can write the value of the state at t = Tsw in terms of the state at t = 0:
Z t0 ATsw A(Tsw−τ) x(Tsw) = e x(0) + e B1vg(τ) d τ. (2.26) 0
By defining the duty cycle d(t) [0, 1] as the fraction of the switching period Tsw ∈ in which S is conducting and d0(t) = 1 d(t) as the fraction of the time the switch is − Modeling 45
not conducting, we can write