Physics and Applications of A Plasma Deflagration Accelerator

PHYSICS AND APPLICATIONS OF A PLASMA DEFLAGRATION
ACCELERATOR
A DISSERTATION
SUBMITTED TO THE DEPARTMENT OF MECHANICAL ENGINEERING
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Keith T. K. Loebner
June 2017
iv
Abstract
Pulsed, gas-fed plasma accelerators are known to operate in a mode that is characterized by ultra-
high velocity, collimated plasma jets known as a plasma deflagration. Despite a degree of con-
tinuous scientific attention over several decades, both the basic underlying physics and the complex
dynamics that contribute to the formation and evolution of these jets remains poorly understood.
The high energy densities and inherently transient nature of these pulsed systems make for a challeng-
ing diagnostic environment, and offer an array of both physical questions and potential applications
that are the focus of this dissertation.
This work concerns the experimental investigation of several aspects of a pulsed plasma acceler-
ator operated in the deflagration mode. Particular focus is given to physics governing the formation
and subsequent behavior of the collimated outflowing jet, whereas prior studies have primarily en-
gaged with the acceleration process in the inter-electrode region of the plasma gun.
Three main diagnostics were implemented and are discussed, each with the aim of characterizing
a specific feature of the plasma accelerator. First, an immersed probe was used to simultaneously
measure multiple magnetohydrodynamic (MHD) and thermodynamic state variables at a single
point in the plasma flow. This enabled empirical validation of an MHD Rankine-Hugoniot model for
plasma deflagrations and detonations, as well as the establishment of baseline plasma parameters
such as density and temperature in the far-field of the jet.
Turning to less-invasive diagnostic techniques, spectroscopic methods were then applied to quan-
tify the plasma density distribution in the near-field of the accelerator plume. A fast-framing ICCD
camera coupled to a high-resolution spectrometer enabled spatially and spectrally resolved imaging
of a transaxial slice of the jet. The images showed the Stark-broadened lineshape of the H- hydro-
gen emission as a function of vertical position in the slice, and from this image the radial density
profile was tomographically reconstructed. Scaling laws for the peak density and radial distribution
were established, and these scalings demonstrate that a magnetically compressed core forms within
the deflagration-produced jet.
The third diagnostic, schlieren imaging, was aimed towards understanding the interface between
the axial plasma flow and the radially compressed, magnetically confined region, as well as the
fluctuations and dynamics at that interface. A laser schlieren imaging apparatus was developed,
v
and combined with a CMOS camera capable of collecting 256 consecutive frames at a 10 MHz
framerate. This system enabled the time-resolved imaging of the plasma density structure in a large
viewing area. Using this tool, coherent structures within the jet and quasi-periodic self-modulation
of the pinching core were directly observed for the first time.
Finally, various applications of the deflagration-produced plasma jets were considered. These
included: space propulsion, plasma-material interactions, and laboratory astrophysics. All three
of the developed diagnostics were used, alone and in combination, to characterize the deflagration
accelerator as applied to these areas. This work represents a comprehensive investigation of the
detailed physics of a complex and practically useful plasma accelerator, and improves our theoretical
understanding of its operation and performance.
vi
Acknowledgements
Having grown up a mere short bike ride from the Stanford campus, my conception of the University
for the better part of my adolescence was as the home of the local shopping mall and the occasional
site of a bar or bat mitzvah reception. It was not until I began to apply to colleges on both the
West and East Coasts that I began to think of Stanford as a place that I could one day attend
not as a shopper or partygoer, but as a student (though I have found that the three are far from
mutually exclusive amongst the Stanford student body). Even after I made the decision to attend
the Massachusetts Institute of Technology for my undergraduate education, I knew that if at all
possible I would seek to return to these temperate climes to pursue a graduate degree.
When I was so fortunate as to be accepted into the M.S./Ph.D. program in Mechanical Engi-
neering at Stanford, I was able to realize the broad strokes of this vision; however, I had no idea
then how truly transformative the experience would be. Over the six years of my academic affilia-
tion with Stanford, I have met countless brilliant and engaging people who have shaped my life in
uncountable and unfailingly positive ways. A separate dissertation would be required to enumerate
each of these individuals and their contributions to my success; however, I will make an attempt,
though admittedly incomplete, to acknowledge and thank some of them.
First, I would like to thank my adviser, Professor Mark Cappelli. As far as it is within my ability
to discern, Mark is a truly unique breed of academic adviser. In fact, academic adviser is an
unduly limiting misnomer when it comes to Mark; he does not confine his mentorship merely to the
academic arena, though his support and guidance in that area is without parallel. His dedication to
his craft as a scientist is at least matched by his commitment and concern for each of his students,
and it is exceedingly unlikely that my experience at Stanford would have been a tenth as rich and
rewarding without his role in it.
I would also like to extend special appreciation to the remainder of my reading committee,
Professor Christopher Edwards and Dr. Flavio Poehlmann-Martins. Your feedback and knowledge
of the material has vastly improved the final product of this dissertation. Special thanks are owed
as well to Professor Ronald Hanson and Professor Per Enge for agreeing to serve on and chair,
respectively, my thesis committee. At times, as the sole representative of the Stanford Plasma
Physics Laboratory in several of Professor Hansons classes, I felt as though I were an honorary
vii
member of the Hanson Group. Professor Enge was the first faculty member that I spoke to regarding
attending Stanford; it is especially fitting, then, that you were able to chair the committee that helped
to mark the end of my tenure here.
I was fortunate to receive generous financial assistance throughout my graduate career, including
three years of Government support under and awarded by the U.S. Department of Defense, Air
Force Office of Scientific Research, National Defense Science and Engineering Graduate (NDSEG)
Fellowship, under 32 CFR 168a. Additional financial support for myself and the work described in
this dissertation was provided by the U.S. Department of Energy, Office of Science, and the National
Nuclear Security Administration Stewardship Science Academic Program.
I was also privileged to have a couple of extracurricular experiences outside the confines of the
university during my time at Stanford: Thank you to Dr. Jay Salmonson of Lawrence Livermore
National Laboratory, who supervised me as I experienced a taste of the life of a government scientist
and investigated capsule implosion physics on the National Ignition Facility during a summer in the
High Energy Density Physics Internship Program. Thanks as well to Professor Christopher Brophy
of the Naval Postgraduate School, who made sure I didnt get lost or thrown in the brig as I spent
a summer studying high pressure pulsed corona discharges as a Naval Research Enterprise Intern.
A deep thanks to Dr. Victor Miller, whose knowledge of Schlieren is without parallel (excepting,
of course, for the great Gary Settles, hallowed be his name) and whose 90s cover band I will someday
see perform live, I promise. Without your help, it is unlikely that I would have gotten to this point
having done anywhere near the same quality of work, such as it is. The jury is still out on whether
I should thank you or curse you for getting me hooked on cycling; Ill let you know in thirty years,
when the long term effects either have or havent taken hold. Thanks as well to the gentlemen of
Hadland Imaging, and in particular to Vince Morton, who provided the Shimadzu HPV-X2 camera
on loan to SPPL and thus enabled the capture of some truly breathtaking high speed plasma imagery.
A hearty thanks to all the members of the SPPL, without whom I would have been toiling away
in obscurity AND a thorough lack of entertainment and camaraderie, instead of merely in obscurity.
Ben, your willingness to ask for forgiveness instead of permission led to rapid progress in the early
days that might not have otherwise been achievable (nor, perhaps, advisable, but were alive so alls
well that ends well); thank you for moving us forward. Tom, your assistance in the performance of
nearly all the experiments described herein was invaluable, as were your deft numerical analysis skills
and your critical (and occasionally even constructive) feedback of our results; in all seriousness, very
little of this work would have been possible without you (but without me...who knows). I wish you
the good luck of a long life and a bare minimum of electrical fires as you inherit the gun experiment;
as the rest of the lab knows, you may need it. Chris, I hope you enjoy your career at the Lab; maybe
one of these days Ill join you, if theyll have me. In the meantime, see if you can get that whole
fusion thing worked out; Ill work on the office playlist. To the rest of the lab, both old and new:
David, Bob, Taemin, Nic, Andrea, Fabio, the rest of Italy that Im certain will eventually become
viii
lab members if they arent already... thanks for all the laughs, and remember to keep pretending to
clean the lab occasionally in order to extract more of that free pizza from Mark.
To my MIT friends who left the Least Coast to slum it in Palo Alto with me, thanks for keeping
it properly nerdy (not like these poser nerds at Stanford that know the rules of games like football
and the one with the hoops). Malcolm, Benson, Rubin, and Flynn, I know I havent converted all
of you to PA4Lyfers yet, but one way or another well live at most a block away from each other
forever...even if that block is really, really long.
Thanks to the outstanding team at Schox: Jeff, Kate, Steph, Ivan, Diana, Tom, Brian, and Sam;
your patience and support as I completed this journey and prepared for the next has meant a great
deal to me.
I was also fortunate to live within minutes of my parents, Pam and Ben, as well as my sister,
Sarah, who even snagged a job at Stanford just so we could get lunch together on campus (at least,
I assume thats why she wanted the job). I was even lucky enough to see other members of my local
and nearly-local family (and my local nearly-family) every so often between Thanksgiving, Passover,
and Cardinal football games: Jon, Michelle, David, Steve, Gary, Barbara, Alex, Mark, and Marla,
thanks for being there for me. Theres nothing quite like family to make a place feel like home, so
thank you all for making that a reality.
And finally, words are entirely inadequate to express the love and gratitude I have for my fiancee
Ellen, who will soon become my wife. She has been my stalwart partner for all (less a couple lonely
months) of what has been a long and often bumpy road to reach this goal, and throughout has
possessed the unique ability to seamlessly transition between celebrating my few scientific successes
and providing a welcome respite from pondering my (all too many) failures. You kept me (more or
less) sane with your love and support, and I cannot wait for the life that we will continue to build
together; I love you.
ix
Contents
Abstract v
Acknowledgements vii
1 Introduction 1
1.1 Plasma Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Plasma Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Range of Plasma Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Plasma Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Review of Electromagnetic Plasma Guns . . . . . . . . . . . . . . . . . . . . . 5
1.3 Plasma Deflagration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3.1 The Plasma Analogy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 The Rankine-Hugoniot Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.3 Applicability of the MHD Rankine-Hugoniot Model . . . . . . . . . . . . . . 11
1.4 Dissertation Organization and Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Stanford Plasma Gun Facility 15

2.1 Facility Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Vacuum System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Pulsed Power Driver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Gas Injection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Custom Valve Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.1 Test Apparatus and Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5.2 Valve Characterization Results . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Coaxial Plasma Gun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6.1 SPG Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6.2 Prior Work on the SPG Facility . . . . . . . . . . . . . . . . . . . . . . . . . . 28
x
3 Exploring the Far Field 32
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.1 Phenomenological Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.1.2 Derivation of the MHD Rankine-Hugoniot Relations . . . . . . . . . . . . . . 36
3.1.3 SPG Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1.4 Probe Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.1.5 Spatial Scanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.1 Probe Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.2 Spatiotemporal Contours . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3.1 Model Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4 Exploring the Near Field 55

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Line Broadening Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Stark Broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.2 Plasma Density Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.3 Doppler Broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.4 Temperature Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3 Methods: Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.1 Test Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.2 Imaging Spectrometer Apparatus . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.3 Spectral & Spatial Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4 Methods: Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4.1 Emissivity Inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.4.2 Fitting the Emissivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.5.1 Radial Density Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.5.2 Current Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.6.1 Radial Compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.6.2 Constant Drift Velocity Model . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.6.3 Model Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
xi
5 Dynamics of the Discharge 79
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 Plasma as a Refractive Medium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.3 Schlieren Photography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3.1 A Basic Schlieren System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3.2 Principles of Schlieren . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.4 Challenges of Imaging Schlieren in Plasmas . . . . . . . . . . . . . . . . . . . . . . . 84
5.5 Methods: Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.5.1 Light Source and Chromatic Filtering . . . . . . . . . . . . . . . . . . . . . . 85
5.5.2 Beam Formation and Spatial Filtering . . . . . . . . . . . . . . . . . . . . . . 85
5.5.3 Cutoff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.5.4 Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.6 Methods: Image Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6.1 Background Subtraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.6.2 Diffraction Artifact Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.6.3 Wavelet Denoising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.6.4 Intensity Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.7 Methods: Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.7.1 Spatial Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.7.2 Radiometric Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.8 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.8.1 Distinguishing the Schlieren Effect . . . . . . . . . . . . . . . . . . . . . . . . 99
5.8.2 Shadowgraph Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.8.3 Evidence of a Dual Flow Structure . . . . . . . . . . . . . . . . . . . . . . . . 103
5.9 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.9.1 What We Can Learn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.9.2 Long Timescale Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.9.3 Emission Cinematography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.9.4 Schlieren Cinematography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
5.9.5 Tomographic Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.10 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.10.1 Z pinch stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.10.2 Calculating the Alfven Time . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.10.3 Applying the Stability Criterion to the Observed Pinch . . . . . . . . . . . . 123
5.11 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
xii
6 Applications 127
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.2 Plasma-Material Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.2.1 Plume Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2.2 Energy and Heat Flux Parameters . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2.3 Witness Plate Target Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
6.2.4 PMI with the SPG Facility . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.3 Space Propulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.3.1 Facility Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.3.2 Polarity Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.3.3 Geometry & Gas-Injection Comparison . . . . . . . . . . . . . . . . . . . . . 138
6.3.4 Space Propulsion with the SPG . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.4 Laboratory Astrophysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.4.1 Jet Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.4.2 Pinch Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.4.3 Astrophysical Scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.4.4 Laboratory Astrophysics with the SPG . . . . . . . . . . . . . . . . . . . . . 150
6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7 Conclusion and Future Work 151
A Experimental Facility Procedures 155

A.1 Chamber Pumpdown Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
A.2 Gun Firing Procedure and Safety Checklist . . . . . . . . . . . . . . . . . . . . . . . 159
B Valve Engineering Drawings 163
C Imaging Spectrometer Addendum 173

C.1 Calibration Image & Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
C.2 H- Broadening Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
C.3 Spectrometer Analysis Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
D Schlieren Data & Processing 213

D.1 Schlieren Utilities Library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
xiii
List of Tables
3.1 Parameters used to compute detonation and deflagration velocities. . . . . . . . . . . 52
6.1 Scales of typical astrophysical phenomena and related experiments . . . . . . . . . . 149

6.2 Comparison of critical length and temporal scales for the laser produced jets [1, 2]
and those from the device considered in this work. . . . . . . . . . . . . . . . . . . . 149
xiv
List of Figures
1.1 The parameter space of plasmas in nature . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 Electromagnetic plasma gun schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Illustration of a deflagration versus detonation in a cylindrical tube . . . . . . . . . . 8
1.4 (a) Schematic depiction of a magnetohydrodynamic deflagration. (b) Schematic de-
piction of a magnetohydrodynamic detonation. . . . . . . . . . . . . . . . . . . . . . 9
1.5 Rankine-Hugoniot curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Time integrated image of the gun firing . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1 Experimental Facility Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2 Photograph of the straight glass tube . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Photograph of the multi-port glass tube . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Valve cross section (closed) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.5 Valve cross section (open) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Valve testing equipment setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7 Pressure rise rate v. plenum pressure over range of restrictor positions . . . . . . . . 24
2.8 Mass bit v. restrictor position at several plenum pressures . . . . . . . . . . . . . . . 24
2.9 Schematic cutaway of the Stanford Plasma Gun . . . . . . . . . . . . . . . . . . . . . 26
2.10 Photograph of the Stanford Plasma Gun . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.11 Acceleration process within the electrodes . . . . . . . . . . . . . . . . . . . . . . . . 27
2.12 Rogowski coil array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.13 Installed Rogowski coil array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.14 Radial current distribution obtained from the Rogowski coil array . . . . . . . . . . 30
2.15 ICCD image sequence of the discharge . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1 ICCD images of the consecutive deflagration and detonation events . . . . . . . . . . 35

3.2 Rankine-Hugoniot curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
xv
3.3 (a) Schematic depiction of a deflagration, with relevant state variables labeled. Note
that the upstream gas in this case is the cold gas injected by the valve, assumed to be
instantaneously ionized. The arrows are indicative of the flow velocity relative to the
wave region. (b) Schematic depiction of a detonation, with relevant state variables
labeled. Note that in this case, the upstream gas is the residual gas left behind by
the initial deflagration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Schematic of the Quadruple Langmuir Probe (QLP) experiment . . . . . . . . . . . 42
3.5 Photographs of the QLP experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6 Example set of QLP current traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.7 Computed density vs. time at two axial positions . . . . . . . . . . . . . . . . . . . . 47
3.8 Computed temperature vs. time at one axial position . . . . . . . . . . . . . . . . . 48
3.9 Spatiotemporal contour of plasma density . . . . . . . . . . . . . . . . . . . . . . . . 49
3.10 Measured vs. predicted velocity of the deflagration front . . . . . . . . . . . . . . . . 50
3.11 Measured vs. predicted velocity of the detonation front . . . . . . . . . . . . . . . . 51
3.12 Deflagration process diagram, part two . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.13 The revised parameter space of the SPG subsequent to the far-field study . . . . . . 53
4.1 Time integrated image of SPG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4.2 Illustration of the Stark effect on atomic energy levels . . . . . . . . . . . . . . . . . 58
4.3 Stark broadening correlation to plasma density . . . . . . . . . . . . . . . . . . . . . 59
4.4 Illustration of the Doppler effect on spectral line shape . . . . . . . . . . . . . . . . . 60
4.5 Imaging spectrometer setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.6 Example raw spectral image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.7 Diagram of chord integration differential area . . . . . . . . . . . . . . . . . . . . . . 65
4.8 Half-plane image of the H- line vs. vertical position . . . . . . . . . . . . . . . . . . 66
4.9 Inverted image of the H- line vs. radial position . . . . . . . . . . . . . . . . . . . . 67
4.10 Comparison of Lorentzian and Gaussian lineshapes . . . . . . . . . . . . . . . . . . . 68
4.11 Lineshapes at various combinations of plasma density and temperature . . . . . . . . 68
4.12 Voigt fit of example radial slice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.13 Radial profiles computed from the Voigt fits of the inverted spectral data . . . . . . 71
4.14 Peak plasma density versus average current scaling . . . . . . . . . . . . . . . . . . . 72
4.15 Line density versus average current scaling . . . . . . . . . . . . . . . . . . . . . . . . 73
4.16 Schematic of a pinch occurring in the SPG . . . . . . . . . . . . . . . . . . . . . . . . 74
4.17 Pinch model computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.18 Updated illustration of the stages of the deflagration . . . . . . . . . . . . . . . . . . 77
5.1 Example schlieren image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

5.2 A basic schlieren system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
xvi
5.3 schlieren experimental apparatus in this work . . . . . . . . . . . . . . . . . . . . . . 85
5.4 Illustration of both methods for fabrication of sooted slide ND cutoffs from glass slides 87
5.5 Photograph of the Shimadzu HPV-X2 . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6 Results of each step of the image processing sequence, as applied to an example frame 90
5.7 Example of a ring artifact in a schlieren image . . . . . . . . . . . . . . . . . . . . . 92
5.8 Graphical depiction of the streak-removal algorithm . . . . . . . . . . . . . . . . . . 93
5.9 Spatial calibration grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.10 Radiometric calibration image and resultant remapping curves . . . . . . . . . . . . 97
5.11 Proof of schlieren gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.12 Schlieren gradient in air . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.13 Shadowgraph effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.14 Returning to the dual flow picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.15 Discharge current vs. time, with pinch duration indicated . . . . . . . . . . . . . . . 106
5.16 Comparison of various neutral density filters . . . . . . . . . . . . . . . . . . . . . . . 107
5.17 Analysis of optical emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.18 Coherent structure evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.19 Diameter and Intensity Periodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.20 Diagram of deflection pattern resulting from an axisymmetric refractive index . . . . 114
5.21 Average radiometrically calibrated image . . . . . . . . . . . . . . . . . . . . . . . . 115
5.22 Symmetrized radiometrically calibrated image . . . . . . . . . . . . . . . . . . . . . . 116
5.23 Abel inverted time-averaged image . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.24 Selected profiles of the Abel inverted image . . . . . . . . . . . . . . . . . . . . . . . 117
5.25 Abel inverted time-averaged image with density quantification . . . . . . . . . . . . . 118
5.26 Illustration of the sausage instability . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.27 Growth rates of m = 0 mode during deflagration . . . . . . . . . . . . . . . . . . . . 124
5.28 Growth rates of m = 0 mode, closely spaced between 20-30% current coupling . . . . 124
5.29 Growth rates at higher pinch density . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6.1 SEM images of observed material damage on Si wafers . . . . . . . . . . . . . . . . . 131

6.2 Compilation of optical micrographs of damage to Cu tokens . . . . . . . . . . . . . . 133
6.3 Photographs of SBT and mSPG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.4 Plasma density spatiotemporal contours . . . . . . . . . . . . . . . . . . . . . . . . . 137
6.5 Fast framing ICCD images of the SBT firing, normal polarity . . . . . . . . . . . . . 138
6.6 Fast framing ICCD images of the SBT firing, reverse polarity . . . . . . . . . . . . . 138
6.7 Spatiotemporal density contour of mSPG, reverse polarity . . . . . . . . . . . . . . . 139
6.8 Fast framing images of the mSPG firing, reverse polarity . . . . . . . . . . . . . . . . 140
6.9 Sequence of emission images of the plasma jet leading edge . . . . . . . . . . . . . . 142
6.10 Leading edge velocity plot of the plasma jet . . . . . . . . . . . . . . . . . . . . . . . 143
xvii
6.11 Schlieren images of the jet at 5 kV and 9 kV charging voltages . . . . . . . . . . . . 144
6.12 Current vs. time at 5 kV and 9 kV charging voltages . . . . . . . . . . . . . . . . . . 146
6.13 Pinch properties vs. radius . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
C.1 Spectrometer calibration image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

C.2 Current calibration plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
xviii
Chapter 1
Introduction
The goal of this dissertation is to further understand the physical mechanisms governing the oper-
ation of a particular class of electromagnetic plasma accelerator which, when operated in a specific
mode, is known as a plasma deflagration accelerator. One unique feature of the plasma deflagra-
tion accelerator is the high-velocity, directed plasma jet that it produces; this work also explores
several practical applications of these plasma outflows.
The remainder of this chapter provides the necessary information required to understand the
context of this work, as well as the motivation for carrying out these investigations. A brief descrip-
tion and background is provided for each of the substantive titular terms in turn (i.e., plasma,
deflagration, and accelerator), including a brief review of the body of research into the class of
plasma accelerators to which the plasma deflagration accelerator belongs. This is followed by an
outline of the scope and organization of the remaining chapters, as well as an enumeration of the
objectives of this work.
1.1 Plasma Definition

While plasma is the most abundant form of matter in the universe, it is not commonly encountered in
the corner of the cosmos most frequented by humanity. Thus, any proper dissertation with plasma
in the title should seek first to define what a plasma is before exploring what a plasma does.
Though there are many ways to approach this definition, the most intuitive manner in which
to think of a plasma is as the terminal state of matter after successive additions of energy. In the
lowest energy state of matter, it exists as a solid; the atoms are content to lie in close proximity to
one another due to a lack of bulk thermal motion. As energy is added to this population of atoms,
they soon become a liquid, followed by a gas. Finally, when enough energy is added to the system, a
significant fraction of the electrons that were previously bound to the nuclei of the atoms are freed,
1
2 CHAPTER 1. INTRODUCTION
and the matter is now a collection of neutral atoms and charged particles (i.e., ions and electrons).
We call this collection of both charged particles and neutrals a plasma.
1.1.1 Plasma Criteria
It is important to note that the above description is merely a thought experiment; in fact, achieving
the plasma state is rarely so simple as merely adding heat. Thus, a more rigorous definition is
typically used to define a plasma, which includes meeting the following three criteria: collective
behavior, quasineutrality, and the ratio of plasma oscillation frequency to the collision frequency
between charged and neutral particles being greater than unity.
Collective behavior is truly the defining feature of a plasma from the perspective of its practical
implications: the particles of a plasma do not behave as a simple gas, caroming off of one another
as if comprised of hard spheres. Rather, the particles of a plasma are charged, which means that
as they move in relation to one another, local space-charge can develop which gives rise to electric
fields. Furthermore, the motion of a charged particle is itself a current, and currents give rise to
magnetic fields. Both electric and magnetic fields affect the motion of charged particles in space;
thus, the motion of the charged plasma particles at one region of the plasma affect the behavior of
charged particles elsewhere in the plasma. This is the essence of collective behavior, and is in large
part the reason for the complexity and wonder of plasma phenomena.
Quasineutrality is a term of art in plasma physics, and it means that while a plasma may develop
local concentrations of charge due to internal or external potentials, these charges are shielded out
such that they are apparent neither in the bulk plasma nor external to the plasma. Thus, the plasma
is neutral enough, such that we may consider the ion density (ni ) to be approximately equal to
the electron density (ne ), and regard the plasma as having a common density, n, called the plasma
density.
Whether a gas meets the criteria of quasineutrality and collective behavior is often determined
based on the relative scale lengths of the shielding distance of these local charges and the length
scale of the system; the former is called the Debye length D , which is defined as:
1/2
0 kB Te
D , (1.1)
ne2
where 0 is the vacuum permittivity, kB is Boltzmanns constant, Te is the electron temperature, n

is the plasma density, and e is the elementary charge. Thus, one criterion for a plasma is:
D L (1.2)
1.1. PLASMA DEFINITION 3
where L is the characteristic length scale of the system. We may also define a related parameter:
the Debye sphere, which is a sphere of radius D , and the number of particles therein, ND :
4
ND = n 3D . (1.3)
3
The second criterion of a plasma, in order for the first to be valid (i.e., for the plasma to be
sufficiently dense such that there are enough charged particles to meaningfully shield out local
charge concentrations), is that the Debye sphere have sufficient particles within it, i.e.:
ND 1. (1.4)
Finally, as stated above, the ratio between the characteristic frequency of the oscillations of the
plasma, , and the collision frequency of charged particles within the plasma with neutrals, , should
be greater than unity. Though a gas may be ionized, if the collision frequency of charged particles
with neutrals is significantly higher than the frequency with which they respond to electromagnetic
fields, then the dynamics of the system will be governed by standard hydrodynamic behaviors (which
are mediated by collisions). Thus, the third and final plasma criterion is:

> 1. (1.5)

Matter that exhibits these three criteria is, with few exceptions, a plasma.
1.1.2 Range of Plasma Conditions

Though they may seem restrictive, the plasma criteria are met by plasmas that span a wide range
of conditions. Even reducing the state space to merely electron temperature and electron density
(neglecting the fact that a plasma may have various temperatures defined for different species, such
as separate ion and electron temperatures), Fig. 1.1 shows that plasmas exist over six orders of
magnitude in temperature and fifteen orders of magnitude in number density.
Our horizon is not quite so broad; the red oval in Fig. 1.1 indicates the general plasma regime that
we expect the plasma deflagration accelerator to operate within. However, even this narrower view
spans several orders of magnitude in temperature and density, and the uncertainty in the precise
values of these parameters provides one key motivation for the diagnostics performed in this work.
The detailed findings that drastically narrow the parameter space covered by the plasma deflagration
accelerator are contained in Chapters 3 and 4.
Figure 1.1: Various plasmas plotted against their associated electron temperature and number den-
sity ranges, from Ref. [3]. The red oval indicates the general parametric region in which many pulsed
plasmas, such as the discharge under study herein, reside.
1.2 Plasma Acceleration
In most practical applications, plasmas are accelerated in one of two primary ways: electrostatically
and electromagnetically. In the former case, charged particles (typically ions) are accelerated across
a potential drop, and if the purpose of the acceleration is to create thrust (e.g., for a spacecraft),
the ion beam is neutralized by a separate electron beam to avoid charging of the spacecraft. The
force on the ions is from the electric component of the Lorentz force, given by
F~L = q E,
~ (1.6)
~ is the electric field vector. Examples of accelerators in this

where q is the particle charge and E
class include cusped-field and Hall-effect accelerators (also known as Hall thrusters), about which a
tremendous body of historic and active research exists [46].
In an electromagnetic accelerator, a combination of electric and magnetic forces accelerate

both the ions and electrons of a quasineutral plasma. An electromagnetic accelerator can rely
on externally-applied or internally-generated magnetic fields to contribute to the acceleration; the
latter is typically called a self-field accelerator. In either case, the force on the plasma includes
1.2. PLASMA ACCELERATION 5
the magnetic term of the Lorentz force, and thus experiences the total force
F~L = q(E
~ + ~v B)
~ (1.7)
~ is the magnetic flux density vector. Such self-field accelerators

where ~v is the particle velocity and B
are typically referred to generically as plasma guns.
1.2.1 Review of Electromagnetic Plasma Guns
The most common geometric configuration of such plasma guns is coaxial, wherein the accelerator
electrodes are configured as an outer electrode (e.g., the anode) and an inner electrode (e.g., the
cathode); however, other geometries are sometimes used (e.g., plane geometries [7]). The Stanford
Plasma Gun studied in this work is of the coaxial type. As illustrated in Fig. 1.2, the acceleration
process in electromagnetic plasma guns proceeds as follows:
1. An external power source, such as a high-energy capacitor bank, applies a potential between the
electrodes. In some variations, a process gas resides in the electrode volume prior to switching
power across the electrodes; this is known as the pre-fill mode. Less commonly, the power is
applied to the electrodes held in vacuum (i.e., with a negligible background gas density) and
the process gas is puffed into the electrode volume; this is fittingly known as the gas-puff
mode. The Stanford Plasma Gun is operated in the gas-puff mode, the consequences of
which will be described in greater detail below.
2. Whether in situ or injected, the process gas breaks down in the inter-electrode region nearly
instantaneously, forming a plasma armature that conducts current radially between the outer
and inner electrode.
3. The current collected by and flowing through the central electrode generates an azimuthal
magnetic field concentrically about the electrode axis.
gas puff
capacitor
bank
Figure 1.2: Schematic of an electromagnetic plasma gun

4. The plasma armature is accelerated axially under the j B component of the Lorentz force.
Note that the sign of the charge on a plasma particle cancels with the sign of the current and
the induced magnetic field, and the accelerating force always points downstream towards the
exit plane of the accelerator.
While seemingly straightforward in principle, the above process can occur in several distinct
ways. The most common operational mode is termed the snowplow mode, in which the current
conducted radially through the plasma is confined into a narrow sheet (e.g., 1 cm thick) that is
accelerated downstream by the Lorentz force, sweeping up encountered neutral gas and accelerating
it (hence the term snowplow). A related mode, termed the mass slug mode, is a special case
of the snowplow mode in which no mass is accumulated by the current sheet as it travels axially
downstream. Ref [8] reviews models describing these plasma gun modes in detail; as they are not
the subject of this work, we will not review them here. The important shared trait of both the
snowplow and mass slug descriptions of plasma gun operation is that they involve a hydromagnetic
propagating shock, and predict relatively high plasma temperatures and electrode erosion (due to
high current densities within the sheet).
In 1958, Marshall observed a third operating mode in a coaxial plasma gun, one that had unex-
pected properties that were unlike the behavior predicted by the snowplow or mass slug models. The
ejected ions had a relatively low temperature of less than 100 eV, and the gun exhibited markedly
lower electrode erosion (indicative of low current densities at the electrode surfaces) [9]. Though
Marshall did not endeavor to explain these discrepancies, Mather performed a detailed study several
years later in which he identified two distinct operating regimes of a similar plasma gun. He could
control the operating regimes by adjusting the relative timing between gas injection and voltage
application. When the voltage was applied to the electrodes shortly after gas injection (' 140 s),
Mather observed significantly higher particle energies in the emergent jet than in the case where
voltage was applied a longer period after gas injection (' 380 s) [10]. The implication of this
observation is that the operating mode of the plasma gun is highly dependent upon the neutral gas
distribution within the electrode volume prior to breakdown: in the first mode, the neutral gas is
concentrated at the breach end of the gun(i.e., as in the gas-puff condition); in the second mode,
the neutral gas has distributed itself along the full axial length of the gun (i.e., approximating a
pre-fill condition).
The first attempt to provide a theoretical explanation for the two modes observed by Marshall and
Mather was formulated by Cheng in 1970. His plasma gun configuration differed in two important
ways from the Marshall gun: first, he eschewed the high voltage switch and instead triggered the
discharge via the gas injection itself. Second, he injected the gas through the base of the electrode,
whereas the Marshall gun configuration included injecting the gas partway down the gun. There
are three consequences of these differences: first, because the gas density monotonically decreases
downstream of the gas inlet, the potential difference required to ionize the gas will always be lowest
1.3. PLASMA DEFLAGRATION 7
upstream of the ionization front where the density is greater (according to Paschens Law applied
to a vacuum gap [11]). This leads to a stationary ionization wavefront in the laboratory frame, as
new process gas is continuously ionized as long as there is a sufficiently high potential drop across
the electrodes. Second, the accelerating plasma can freely expand into vacuum, preventing a shock
from forming as accelerated plasma impacts slower neutrals downstream. Third, the magnetic flux
generated by the return current suffuses the breach region as ionization occurs, which enables the
plasma to be magnetized immediately upon ionization without the need for macroscopic internal
currents that could lead to shock formation [12].
These two modes of acceleration in plasma guns have distinct phenomenological characteristics:
the snowplow (and/or mass slug) mode includes a shock that propagates into a neutral gas, adding
energy to the flow as it propagates and causing a pressure and density rise across the wave. The
unknown mode observed by Marshall, Mather, and Cheng includes an ionization wave as well, but
in contrast, the pressure and density appear to drop across the wave as a cooler, more diffuse plasma
is accelerated downstream of the wave. These characteristics are strikingly similar to two classes
of quasi-one dimensional combustion waves, known as detonations and deflagrations, which are
discussed in the next section.
1.3 Plasma Deflagration

Having firmly established the meaning of plasma and accelerator, we now consider the meaning
of the term deflagration. While the usage of plasma deflagration to describe the mechanics of a
plasma gun was first coined by Cheng in 1970 [13], a deflagration is a significantly more venerable
concept that was first identified in the realm of chemical combustion. The word comes from the Latin
deflagrare, which means to burn down, and is the name given to a class of combustion wave (i.e.,
a propagating combustion reaction zone or flame). In fact, there are two types of combustion waves
that propagate through premixed gases (i.e., combustible mixtures in which the fuel and oxidizer
are homogeneously distributed), and they are termed detonations and deflagrations.
The classic thought experiment used to describe these two classes of combustion wave proceeds
thusly: one imagines a cylindrical tube, closed at one end, and filled with a premixed combustible
gas as shown in Fig. 1.3a. If the mixture is ignited at the closed end, as shown in Fig. 1.3b,
the combusting gas causes a rapid local pressure rise that drives a strong shock wave through the
unburned gas, which is followed by a combustion zone in which the chemical combustion reactions
progress to completion. The flame front and the burned gases move in the same direction (though
at different speeds), towards the open end of the tube. This process is called a detonation, and
real-world examples of materials that can undergo such detonations include most explosives.
However, if the mixture is ignited at the open end, as shown in Fig. 1.3c, the combustion products
can freely expand away from the combustion zone. In a similar manner as in the detonation case, the
(a) A cylindrical tube (b) The tube lit at the (c) The tube lit at the
filled with an unburned, closed end, forming a open end, forming a de-
premixed gas. detonation wave. flagration wave.
Figure 1.3: A premixed combustible system.
burned gasses expand rapidly out of the tube, but in the deflagration configuration the flame front
moves into the unburned gas and toward the closed end (i.e., in the opposite direction). Real-world
examples of deflagration-like processes include both the mundane, such as a propane burner, and
the exciting, such as a chemical rocket engine. The latter of these examples highlights an important
difference between detonations and deflagrations: a deflagration converts a larger fraction of the
chemical energy of the reactants into directed kinetic energy, in large part because of the absence
of the shock wave formed in the detonation case. As will be discussed in more detail in subsequent
chapters, the efficient conversion of input energy into directed kinetic energy is a key advantage of
the deflagration mode and lies at the core of several applications of a plasma deflagration.
a)
p
Upstream Downstream
J
B
b)
p
Downstream Expansion Upstream
Sheet
Zone
B
Figure 1.4: (a) Schematic depiction of a magnetohydrodynamic deflagration. (b) Schematic depic-
tion of a magnetohydrodynamic detonation.
1.3.1 The Plasma Analogy
One can imagine a similar thought experiment, in which energy is added to a plasma in a relatively
narrow region approximating the idealized flame front of the combustion case. However, the plasma
case has two important differences. First, the local pressure rise is due to contributions from both
particle effects (i.e., hydrodynamic forces) as well as electromagnetic forces; in most cases, the elec-
tromagnetic forces dominate. Second, energy is added to the plasma by an external electrical energy
source instead of by releasing the internal chemical energy of a combustible mixture. Additionally,
there is no stark distinction between reactants and products in the plasma case; in this idealized
description, the plasma on both sides of the wavefront is treated as a conducting fluid.
Fig. 3.3 shows cutaway views through the cylinder of our thought experiment, adapted to the
plasma case; alternatively, one can imagine the illustrations as a cutaway through a radius of the
coaxial volume of a plasma accelerator. In the gas combustion case, the key state variables that are
conserved across the wavefront are the kinetic pressure and the mass density of the gas; the plasma
case is identical, with the exception that the total pressure p includes a magnetic pressure term,
i.e.:
B2
p = p + (1.8)
20
In the electromagnetic accelerator, the energy addition to the flow is primarily in the form of stored
magnetic energy that is expanded into directed kinetic energy. Thus, from at least a qualitative
perspective, the attractiveness of the deflagration and detonation descriptions is apparent.
1.3.2 The Rankine-Hugoniot Theory
The set of allowable solutions for the mass, momentum, and energy conservation equations solved
across a shock in the flow (e.g., a wavefront) that includes energy addition (e.g., a deflagration
or a detonation) are defined by what is known as the Rankine-Hugoniot relations. The name
originates from the fact that the curve combines Rankines seminal shock jump equations, developed
in 1870 [14] for non-reacting compressible flows, with the addition by Hugoniot nineteen years later
of the treatment of the entropy near and across the jump region [15]. If the conservation equations
are extended to include magnetohydrodynamic flow variables, such as the magnetic flux density
associated with the conducting fluid, the theory is referred to as the magnetohydrodynamic (MHD)
Rankine-Hugoniot relations.
The general form of the Rankine-Hugoniot curve is shown in Fig. 3.2, and succinctly illustrates the
two allowable solution branches as well as several other salient features. A more detailed discussion
of the Rankine-Hugoniot curve and relations (primarily as they pertain to combustion theory) is
contained in Ref. [16]; a more detailed analysis of the conservation equations in the MHD context
is presented in Chapter 3. The origin of the curve, labeled as such in Fig. 3.2, represents the initial
state of the flow (i.e., p1 and 1 ). The families of lines emanating from the origin, shown as dotted
red or green lines in Fig. 3.2, are referred to as Rayleigh lines and are derived from combining the
mass and momentum conservation equations relating the initial and final states. The Hugoniot curve
itself, the black line in the figure, represents the locus of points corresponding to the intersection
of both families of curves with the set of allowed final states given by conserving energy across the
wave and applying the appropriate entropy considerations.
The distance of the curve away from the origin is proportional to the energy added to the flow
between one side of the wave and the other. The tangency points of the Rayleigh lines to the
Hugoniot curve are referred to as Chapman-Jouget (C-J) points, and there are two such points for
any given Hugoniot curve (one corresponding to the upper branch, and one to the lower branch).
The slope of the Rayleigh lines that intersect the C-J points correspond to the minimum allowed
detonation velocity and the maximum possible deflagration velocity. There are no solutions for the
velocity of the end-state of the gas that surpass these respective minimum and maximum velocities;
in other words, the downstream (i.e., burned) gas velocity will be subsonic in the detonation case
and supersonic in the deflagration case.
detonation waves
deflagration waves
strong detonation
upper C-J point
weak detonation
P
prohibited
weak deflagration
origin strong deflagration
lower C-J point
1/
Figure 1.5: The Rankine-Hugoniot curve, illustrating the beginning and end states of a flow tran-
sitioning from the origin (i.e., the initial state) to the final state in one of the two allowed branches.
The strong and weak designations have more clear relevance to the combustion waves they were
originally developed to describe than to the plasma phenomena to which we draw analogy; as such,
we do not consider the implications of strong and weak combustion waves in the plasma case. The
prohibited region, also labeled in the figure, is not a permitted end state because the pressure and
specific volume (the inverse of the density) cannot both increase and still satisfy the continuity
relation.
1.3.3 Applicability of the MHD Rankine-Hugoniot Model

Though some analysis of the applicability of the plasma analogy to the detonation and deflagration
phenomena is conducted in this work, an important point must be made about the limitations of
such analyses. On the one hand, the Rankine-Hugoniot relationships are merely a restatement of
conservation equations that present two sets of solutions, wherein for one set of solutions the gas is
accelerated as it passes through the energy-adding wave, and in the other set of solutions the gas
is decelerated. There are no statements that limit the energy addition to be of a particular type,
which would seem to indicate that purely electromagnetic energy addition should pose no inherent
complication. However, certain assumptions are implicit in the Rankine-Hugoniot formulation that
present certain difficulties in adapting it to electromagnetic plasma acceleration.
At the core of the Rankine-Hugoniot model is the assumption that a Hugoniot curve exists; in
other words, there is an assumption that for a given quantity of energy coupled into a flow with
a given initial state, two families of solutions exist for the downstream state (i.e., corresponding
to the detonation and deflagration waves). While in the combustion case, such an assumption is
testable and empirically well supported, the same cannot strictly be said of the plasma case. The
mechanisms of energy addition and entropy generation to and by the plasma flow are complex
and difficult to directly measure, and thus we cannot state with any semblance of certainty that
the assumptions of constant energy addition and smooth entropy generation near the wave are
obeyed by the plasma accelerator. Even the exact quantity of energy that is coupled into either
the plasma detonation or plasma deflagration from the external capacitor bank is difficult to
measure precisely. Furthermore, the behavior of the plasma produced by the Stanford Plasma Gun is
unsteady and the result of coupled collisional, magnetohydrodynamic, and even radiative processes
that cannot be easily isolated in their effects. In short, an experiment cannot be practically devised
(at least, not within the scope of this work) that quantitatively demonstrates that, given a fixed
energy addition, two sets of wave solutions are observed that are consistent with properties predicted
by a Rankine-Hugoniot curve computed using magnetohydrodynamic state variables.
However, despite the above caveats, we do not herein concern ourselves primarily with the strict
quantitative applicability of every detail of the MHD Rankine-Hugoniot model. In most areas of
scientific investigation, and in particular one as complex and multidisciplinary as plasma physics,
models that can provide even limited utility, or that shed light on one area of the behavior of a
system while leaving another unexplained, are welcomed for what use they do possess. Likewise,
where such models fail, they should rightly be discarded in favor of more accurate descriptions. We
apply this philosophy here, and recognize that while the MHD Rankine-Hugoniot model has both
advantages and drawbacks when applied to real systems such as the Stanford Plasma Gun, it is most
useful as an analogy for the general behavioral characteristics rather than treated dogmatically in
its every aspect. Instead, we rely on the direct measurements of the parameters and behaviors of
the system, collected over the course of producing this work, to enhance our understanding of the
Stanford Plasma Gun and related devices of its class.
1.4 Dissertation Organization and Scope

The experimental inquiries laid out in this dissertation were fortunate to benefit from previous work
and existing facilities developed by previous Stanford Plasma Physics Laboratory (SPPL) students
1.4. DISSERTATION ORGANIZATION AND SCOPE 13
Electrode Volume
Near Field
Far Field
Figure 1.6: Time integrated of the accelerator during firing, with three regions labeled: the electrode
volume, the near-field, and the far-field.
and collaborators. The Stanford Plasma Gun, the capacitor bank, and the power transmission
system were constructed by Flavio Poehlmann-Martins [12], who also conducted extensive studies
on the current distribution within the gun during firing using an array of Rogowski coils [17]. The
originator of the deflagration accelerator concept, D.Y. Cheng, provided counsel and equipment
to earlier generations of SPPL students as well.
The scope of this work can best be defined as building off of the investigations, conducted
both at Stanford and elsewhere, directed towards that which occurs within the coaxial electrode
volume. Cheng himself performed time-resolved measurements of the magnetic field distribution
inside the gun, and found that they were consistent with a diffuse current distribution [18, 19].
Woodall and Len performed similar measurements, while also varying the delay between gas injection
and the applied voltage, and were able to observe the transition between the diffuse mode they
themselves termed the deflagration mode and the propagating current sheet they referred to as
a snowplow, but which may also be thought of as a detonation [2023]. Black followed this work
with a miniaturized, fast-response magnetic probe array placed inside a coaxial gun operating in the
deflagration mode, designed to obtain a two-dimensional, time-resolved profile of the time-varying
magnetic field. He found that the current distribution was substantially evenly distributed along the
electrodes during the majority of the first half period of the oscillating discharge current, but that
current began to concentrate in the vicinity of the exit plane of the gun towards the end of the current
pulse [24]. Others have, over time, performed similar magnetic probe measurements on coaxial
guns, and largely confirmed the preexisting qualitative understanding of the guns behavior [2527].
At Stanford, Poehlmann conducted experiments to directly measure the current distribution (i.e.,
instead of inferring the current distribution from magnetic flux probes) using an array of Rogowski
coils, as will be described further in section 2.6.2.
Each of these studies focus on the plasma behavior in the electrode volume, as shown in Fig. 4.1.
However, little attention has been given to diagnosing the plasma state and dynamics in either
the near- or far-field regions, also shown in Fig. 4.1, of the deflagration-produced jet. The new
investigations contained herein focus, in detail, upon the parameters and dynamics of the plasma
outside the electrodes, in these regions.
Chapter 1 has provided the conceptual underpinnings for important terminology, reviewed the
operating principles of this class of device, and outlined the underlying MHD Rankine-Hugoniot
model that suggests the existence of plasma deflagrations. Chapter 2 briefly reviews the existing SPG
facility, summarizes the specific findings from prior studies conducted at Stanford, and details the
upgrades to the facility that were implemented to enable the diagnostic work herein. In particular, a
custom-built, high-mass-bit puff valve that was specifically designed for driving the SPG experiment
is discussed and its operation characterized.
Chapter 3 describes an immersed probe study which examines the temporal evolution of four
plasma state variables (plasma density, electron temperature, plasma potential, and ion thermal
Mach number) within the jet in the far-field, well downstream of the exit plane of the SPG. With this
data, in combination with the neutral gas injection characterization data collected while evaluating
the puff valve, empirical support for the MHD Rankine-Hugoniot model is discussed.
Chapter 4 discusses a spectroscopic technique, wherein plasma emission was used to quantify
the radial plasma density distribution just downstream of the exit plane, in the near-field, where
immersed probes are unsuitable. A parametric study was performed using this diagnostic at a range
of operating conditions, and scaling laws are presented for the peak density and core radius that
demonstrate a current driven radial compression occurs in the core of the jet.
Chapter 5 details a newly developed ultra high speed schlieren imaging system, and explores the
fascinating and previously unseen jet dynamics it revealed. Various applications of these high energy
jets are explored in Chapter 6, employing the diagnostics developed in Chapters 3, 4, and 5. Finally,
Chapter 7 summarizes the conclusions reached as a result of this work, and provides recommended
directions for future research. In various chapters, this dissertation includes material from previously
published work contained in Refs. [2843].
Chapter 2
Stanford Plasma Gun Facility
This chapter describes the Stanford Plasma Gun Facility, which includes the plasma deflagration
accelerator itself as well as the associated systems used for its proper and safe operation. In the
following sections, the details of four principal subsystems of the overall facility are outlined: 1)
the vacuum system, 2) the pulsed power driver, 3) the gas injection, and 4) the coaxial plasma gun
itself, including a summary of the known features of its operation based on prior work conducted at
Stanford.
2.1 Facility Overview

A schematic of the overall facility is shown in Fig. 2.1. The SPG facility was developed for the
express purpose of studying the behavior of the discharge in various operating regimes, rather
than for characterizing the suitability of the gun for any particular application. As such, physical
accessibility was prioritized in the facility design: the SPG can be easily removed and maintained
while the majority of the system remains in vacuum. Care was also taken to design the facility for
safe operation at high-energy conditions: the SPG is electrically isolated from the main chamber
by several feet of glass insulator, and the power supply and control system is far-removed from the
capacitor banks.
15
16 CHAPTER 2. STANFORD PLASMA GUN FACILITY
1 Large Vacuum Tank

2 Main Door 23
3 Large Gate Valve
4 Large Turbo Gate Valve
19 20
5 Gun Section Gate Valve
6 Glass Tube
22
7 6-way Cross
8 Large Turbopump
9 Small Turbopump 21 10
10 Polycold
11 Cryo Compressor 1
6 17
12 Cryo Compressor 2 18
13 Poly Shroud 1 15
14 Poly Shroud 2 7 27
15 Cryopump 1 9 14
16 Cryopump 2 5 11
17 HV Power Supply and Controller
18 Charging Circuitry Cabinet 2
25
19 Capacitor Banks 1 & 2 26
20 Capacitor Banks 3 & 4 8 4 3
21 Stanford Plasma Gun
12
22 Puff Valve
23 Hydrogen Cylinder
1
24 Roughing Pump Outflow Line
25 Ionization Pressure Gauge 13
26 Thermocouple Gauge 1
27 Thermocouple Gauge 2 24 16
Figure 2.1: Labeled diagram of the experimental facility.
2.2 Vacuum System

The vacuum chamber consists of three major sections, each of which can be controllably isolated from
other portions of the vacuum and pumping systems: the large vacuum tank, the gun section, and
the cryogenic pump bay. The large vacuum tank comprises a non-magnetic stainless steel chamber
approximately 3.3 m in length and 1.5 m in diameter, and during most experiments acts as a high
volume dump tank into which the SPG exhausts. The main tank is connected to the cryogenic
pump bay, but can be separated therefrom, using the large pneumatic gate valve, for any reason
(e.g., to avoid damaging the cryopumps at such a time as the main tank must be rapidly vented to
atmosphere and the cryopumps have not yet returned to ambient temperature). The main tank is
Figure 2.2: Photograph of the straight glass tube, by F. Poehlmann.

2.2. VACUUM SYSTEM 17
Figure 2.3: Photograph of the six-port, axially-reversible tube designed for optical investigations.
likewise connected to the gun section at one eight-inch conflat (CF) port and isolatable via a small
pneumatic gate valve, which enables the gun section to be vented and adjustments or modifications
to be performed on the SPG system without breaking vacuum on the main tank.
The gun section includes all the vacuum components between the gun itself and the small gate
valve: the glass tube, which mates directly to a mating surface of the gun (e.g., a CF flange or a
face-sealing gasket); a tee-section, which mates to the glass tube; and, a six-way cross comprising
several CF ports of various sizes to enable the mounting of vacuum instrumentation.
Two different glass tubes were used over the course of the several studies outlined in this work.
The first, shown with the plasma gun installed in Fig. 2.2, was a six foot long Pyrex tube having an
inner diameter of six inches, which was used for the probe study outlined in Chapter 3, as well as the
space propulsion study described in Section 6.3. The second tube, shown in Fig. 2.3, was specially
designed to allow high quality optical through-access to facilitate various optical diagnostics; it
includes six ports arranged as shown, and is axially reversible to maximize the axial range to which
optical through-access is afforded. Both tubes required custom manufacturing, which was provided
for the straight tube by Adams & Chittenden Scientific Glass and for the eight-way tube by Allen
Scientific Glass.
The cryogenic pump bay houses a pair of two-stage cryogenic pumping systems (CVI-TM1200)
that provide the final base vacuum pressure of approximately 2 107 Torr. The vacuum across
all three portions of the vacuum system is provided by a combination of four different pumping
systems, including the cryogenic pumps. A 2,500 cfm Kinney mechanical roughing pump can be
directly connected to the main tank as well as to the gun section, and can reduce the system
pressure to approximately 1.7 101 Torr in 30 minutes. A large turbomolecular pump (Pfeiffer
TPH 520), backed by the roughing pump, is also connected to the main tank and can be used to
accelerate cryogenic pumping process as well as to increase the operation time of the system before
the cryopanels saturate. The turbopump can also be used to bring the main tank to a lower base
pressure before reconnecting the cryogenic pump bay, if the main tank has been isolated and vented.
A small turbomolecular pump (Pfeiffer TPH 380), also backed by the roughing pump, is connected
to the gun section and can be used similarly.
The pressure of the system is monitored by two standard thermocouple gauges, one connected
at the gun section and the other at the cryogenic pump bay, while the system is at rough vacuum
(i.e., vacuum achievable without the cryopumps). At high vacuum, a standard ionization pressure
gauge is used, and it is connected to the main tank proximal the junction between the main tank
and the gun section.
Detailed procedures for operating the vacuum system are contained in Appendix A.
2.3 Pulsed Power Driver

The pulsed accelerator is driven by a modular capacitor bank consisting of sixteen Sangamd Energy
Discharge Capacitors. Each capacitor has a capacitance of 14 F, a series inductance of 15 nH,
and a rated maximum voltage of 17 kV. The capacitors are connected to the gun via RG-8 coaxial
cables, each of which is 1.75 m long. Care must be taken to avoid large discrepancies in total line
inductance from one transmission line to the next, to ensure balanced transmission to the load (i.e.,
the plasma gun); thus, each transmission line is substantially the same length. The capacitor bank
is configured modularly, and is divided into four individual banks of four capacitors each. Each bank
can be hardwired to either connect to the charging circuitry and the SPG or to remain disconnected
and therefore unused. All of the experiments conducted in this work were performed using a single
bank (i.e., four capacitors), operated between 2-9 kV.
The capacitor banks are charged using a purpose-built capacitor charging power supply (TDK
Lambda 500A), which can be isolated from both the ground plane and the high-voltage side of the
capacitor using two high-voltage relays. After the capacitor bank is charged, the power supply is
isolated from the system to prevent potential damage to the power supply during firing.
By reconfiguring the number of capacitors connected to the system, as well as the line inductances
between the capacitor banks and the gun, the time response of the system (e.g., the underdamped
RLC oscillation period) can be changed. However, since the capacitor configuration was kept con-
stant for the duration of the various experiments performed as part of this work, the time response
was also substantially unchanged. When the gun is fired in this configuration, the drive current
oscillation has a period of 20 s.
Detailed procedures for operating the pulsed power driver are contained in Appendix A. To
initiate the discharge (e.g., fire the gun) after the capacitor bank is charged, gas is injected into the
inter-electrode volume; in other words, the SPG is operated in the gas puff mode described briefly
in Chapter 1. The details of the gas injection process and the associated equipment developed for
these studies are described in the next two sections1 .
1 The majority of the material discussed in these sections is reproduced from Ref. [37], written by K. Loebner et
al.
2.4. GAS INJECTION 19
2.4 Gas Injection

As discussed in Ch. 1, a pulsed coaxial plasma accelerator requires the process gas to be introduced
rapidly into an otherwise-evacuated electrode region in order to properly access the deflagration
mode. Typically, the necessary high pressure rise-rates for these devices are achieved by means of
so-called puff valves that inject a transient, high density gas slug into the inter-electrode region
to initiate the discharge[44]. Increased rise-rates are generally achieved by increasing the upstream
pressure within the plenum of the puff valve. However, in a valve of fixed plenum size, increasing the
upstream pressure likewise increases the overall mass of gas that is delivered to the inter-electrode
region. In some applications, particularly those where total mass utilization is an important param-
eter, failing to mitigate this effect results in wasted process gas. The utilization of the gas by the
driving circuit of the plasma source is dictated by the pulse energy and discharge current, which dic-
tates the recruitment of charge carriers from the injected neutral gas during the ionization process.
For a fixed driving current waveform, there is a particular mass flow profile that can be sustained
without wasting undue amounts of injected gas. While the fast rise rate afforded by a particular high
plenum pressure may be necessary to access the deflagration mode, a shorter characteristic decay
time of the discharge current may require a lower overall mass flow. Matching these conditions is
difficult, if not impossible, without a means for varying the mass bit independently from the rise
rate, or sharpness, of the pressure front of the puffed gas.
Some attempts have been made to adjust the plenum size and/or duty cycle time of the valve
according to the needs of an individual experiment, but the existing methods for doing so come with
attendant difficulties. One previously implemented solution is to modify the valve driver circuit
to produce different valve opening times, but this process is highly nonlinear due to the coupling
between the electromagnetic drive mechanism and the gas dynamic resistance to the valve opening,
leading to intensive calibration requirements at each desired operating point[44]. Other designs
incorporate an adjustable plenum size, but require disassembly of the valve in order to change the
mechanism and thus fail to allow real time adjustment and calibration while the valve is in use[45].
Overcoming these limitations typically requires an excessively complex (and thus expensive), high
part-count mechanism. In this work, we describe the design and operation of a fast rise-rate puff
valve with a dynamically and continuously variable mass bit, wherein the effective actuation time of
the valve can be mechanically adjusted while the valve is in use. This feature allows rapid tuning of
the mass flow characteristics to match a given plasma driver circuit, and does so without a complex
mechanism or high part-count.
2.5 Custom Valve Design

As shown in Fig. 2.4, the valve consists primarily of a plenum, driver coil, aluminum hammer ring,
and a poppet, all contained within a housing. When the coil is energized with a pulsed current, the
Restrictor Flange Gas Inlet
Movable
Restrictor
Tail Cap
Hammer
Return Spring
Poppet
Guide Poppet
Body O-ring
Poppet
Return Spring Grooves
Hammer
Coil
Coil
Section
Coil Bobbin
Head Cap O-ring Grooves
1 in. Gas Outlet
Figure 2.4: A cross section view of the valve assembly in the closed position, with all principal
elements labeled.
hammer is diamagnetically accelerated upwards by the induced eddy currents within the aluminum
hammer. When the moving hammer strikes the outer rim of the poppet, it impulsively drives the
poppet off of an O-ring seal, opening the valve and releasing a puff of gas. The valve in the open
position, with the hammer driving the poppet against the movable restrictor, is shown in Fig. 2.5.
A return spring for both the hammer and the poppet, respectively, resets the valve to its initial
state after actuation, enabling operation of the valve in any orientation. This overcomes a drawback
present in other designs that employ the gravity-assisted pressure differential as the only means of
resetting the valve[46], which prohibits operating the valve in inverted or horizontal positions. The
concentricity of the poppet travel is maintained by a 3D-printed guide that mechanically locates the
poppet in the center of the driver coil and over the O-ring seal, while permitting gas flow through
and around the guide. The travel distance, or throw of the poppet during actuation is determined
by the position of the restrictor, which is adjustable while the valve is pressurized. This adjustment
is achieved by means of rotating the externally-accessible, threaded stem of the restrictor either
2.5. CUSTOM VALVE DESIGN 21
Gas Inlet
Hammer
Return Spring
Poppet
Body
Poppet
Return Spring
Hammer
1 in. Gas Outlet
Figure 2.5: A cross section view of the valve assembly in the open position, with only the elements
that move during actuation labeled.
clockwise or counterclockwise, which linearly positions the restrictor either closer to or farther away
from, respectively, the base of the interior cavity of the poppet. This mechanically limits the travel
distance of the poppet when the valve is actuated. Pressure integrity of the plenum during restrictor
positioning is maintained by a dynamically-sealing O-ring located around the stem of the restrictor,
that is itself captured by a recessed groove in the valve tail-cap and the removable restrictor flange.
Additional O-ring grooves at the planar interface of each valve section prevent static leakage, as
shown in Fig. 2.4.
The valve body is constructed of a clear thermoplastic, which allows visual confirmation of proper
valve operation during use. The head and tail caps are made of aluminum, as is the restrictor,
restrictor flange, and the hammer ring. The mass of the hammer ring was selected to match the
mass of the poppet, itself made of machinable Nylon, in order to maximize efficient momentum
transfer between the hammer and poppet during impact. The driver coil was wound from 18 AWG
copper magnet wire on a custom-fabricated Delrin bobbin, and was designed to have an inductance
of 100 H, a value that was verified after winding the wire onto the bobbin using an LCR meter.
The poppet guide was 3D-printed in an Acrylic-like plastic resin on a 3D Systems ProJet 3500 HD.
The design of the valve is such that the rate of pressure rise of the gas front expelled from the
valve is determined by the plenum pressure, while the overall mass bit is determined by a combina-
tion of the plenum pressure and the throw of the poppet, controlled by the restrictor position. The
construction materials and the nature of the design were selected to enable low part count, straight-
forward fabrication, and maximum ease of use with a variety of operating conditions. Detailed
engineering drawings of the key components of the valve can be found in Appendix B.
2.5.1 Test Apparatus and Procedure

In order to verify the reliable tunability of both the pressure rise rate and mass bit, an extensive
characterization study of the valve properties was conducted. The experimental apparatus, as shown
in Fig. 2.6, consisted of an array of piezoelectric pressure transducers (PZTs) as well as a Baratron
pressure gauge positioned within a small vacuum chamber. The specific PZTs were PCB Piezotronics
ICP pressure sensors, Model No. 113B26, and were driven by an eight channel power supply /
signal conditioner also manufactured by PCB Piezotronics, Model No. 482A20. With this signal
conditioner, the rise time is 1.0 s with a sensitivity of 1.45 mV/kPa. The rise rate of the
pressure in the vicinity of the valve output was measured directly by the PZTs, while the mass bit
was calculated based on the equilibrium pressure in the chamber after a single actuation of the valve.
The chamber volume was measured to be 17.07 L by filling the chamber with a known mass of gas
using a calibrated mass-flow controller. With the chamber having a known volume, the mass bit for
a given firing of the valve was estimated assuming the gas to be ideal and at an equilibrium ambient
temperature of 293 K upon reaching its equilibrium pressure after valve actuation.
The plenum pressure was varied between 10-40 psig in increments of 5 psi and the restrictor
position was varied over ten increments from 0.02-1.33 cm away from the base of the interior cavity
of the poppet. Each combination of plenum pressure and restrictor position was tested three times
to establish an estimate of the repeatability of the valve output from firing to firing. The valve
was driven by a capacitor bank with a capacitance of 1350 F, charged to 900 V for all trials. The
chamber into which the valve was fired was evacuated before each firing to a background pressure
on the order of 103 Torr; the specific pressure was recorded precisely for each individual test and
used to calculate the P. A gate valve was used to isolate the chamber during each firing in order
to allow the pressure to rise to an equilibrium value. The process gas used for all trials was N2 .
2.5. CUSTOM VALVE DESIGN 23
Capacitor Bank Triggered Baratron

Spark-Gap Pressure
Switch Gauge
Piezoelectric Pressure
Transducers
Pumps
Pu Valve Probe Holders

Gate
Valve
Oscilloscope
Signal
Conditioner
Figure 2.6: Schematic of the valve and driver circuit installed on the test chamber, along with the
data acquisition equipment.
2.5.2 Valve Characterization Results
The results of the valve operating characteristics are summarized in Figs. 2.8 and 2.7, which show the
mass bits and rise rates, respectively, of the gas puffs under various combinations of plenum pressure
and restrictor position. Fig. 2.8 in particular shows that the mass bit can be varied substantially
linearly by adjusting the restrictor position, with the mass bits for the shown plenum pressures
converging as the overall poppet throw decreases. At lower pressures, the pressure effects that work
against the opening mechanism are weaker, leading to less mechanical jitter and thus closer overall
agreement with the linear trend.
The resulting mass bit is consistent with estimated mass contained inside the valve plenum at
the tested pressures, implying that, at a maximum, only the gas contained within the plenum is
released. At increasingly limiting restrictor positions, only a fraction of the plenum gas is measured
to have been emitted by the valve. The measured mass bit is proportional to the overall time that the
Figure 2.7: Rise rate as a function of plenum pressure, each point representing an average over all
10 tested restrictor positions.
300
Plenum: 10 psi, Nitrogen
250 Plenum: 20 psi, Nitrogen
200
Mass Bit [mg]
150
100
50
0
0.0 0.2 0.4 0.6 0.8 1.0 1.2
Distance of Poppet Stop From Fully Open Position [cm]
Figure 2.8: Mass bit as a function of restrictor position, along with linear fits, for plenum pressures
from 10-25 psig.
2.6. COAXIAL PLASMA GUN 25
valve is in the open position, and the results are consistent with the movable restrictor mechanism
successfully varying the opening time of the valve during actuation in a linear fashion.
Fig. 2.7 depicts the steady increase in pressure rise-rate obtainable with increasing plenum pres-
sure. Each data point in Fig. 2.7 represents the measured rise-rate over the entire range of all ten
restrictor positions at that plenum pressure. The presented error is based on the residual of the
measured rise-rates against the average value of all ten operating points for the specified plenum
pressure, wherein each operating point was tested three times, as specified above.
The mass bit and rise rate data collected in the course of characterizing the valve here are used
throughout this dissertation where knowledge of the inlet conditions, pre breakdown, are required.
This is of particular importance in Chapter 3, where conditions across the deflagration wave (i.e.,
both upstream/inlet and downstream conditions) are needed to assess the applicability of the MHD
Rankine-Hugoniot model.
2.6 Coaxial Plasma Gun

An illustrative schematic of the coaxial plasma gun itself is shown in Fig. 2.9, along with a photograph
of the gun attached to the straight glass tube of the gun section of the vacuum chamber in Fig. 2.10.
The gun itself consists of a set of stainless-steel rod anodes, arrayed in a circular fashion around
a central copper cathode with 6.5 mm gaps between each consecutive electrode. The diameters of
the stainless-steel rods and the copper cathode are each 5 mm. There is a transparent plastic sleeve
around the rod anode array which extends from the breach end up to the last 3 cm of the electrodes.
This sleeve serves to confine the gas while enabling optical access to the inter-electrode region. A
stainless steel retaining ring captures the ends of the stainless steel rods and holds the plastic sleeve
against the breech end of the gun.
Note that the overall mechanism of operation is the same whether the anode is a solid cylindrical
shell or composed of rods, but for the fact that the interior of the gun volume is visible through
the rods (and it is for this reason the rods were employed). However, the inductance per unit
length of the coaxial electrode arrangement does differ between the rod-anode and solid-anode
configurations, which can impact the electrodynamics of the coupled capacitor bank (i.e., energy
source), accelerator, and plasma armature. We use only the rod-anode configuration throughout the
experiments described in this dissertation, with the exception of Section 6.3 in which experiments
conducted on a miniature plasma deflagration accelerator that included a solid anode are described.
gas puff
Figure 2.9: Schematic cutaway of the Stanford Plasma Gun, showing gas injection through the base
of the cathode at the breach end.
Figure 2.10: Photograph of the Stanford Plasma Gun through a window of the six-port glass tube.
2.6.1 SPG Operation

From a practical standpoint, and ignoring for the time being the detailed physics, the SPG is
operated as follows: first, the capacitor banks, which are directly connected to the gun electrodes
via transmission lines external to the chamber, are charged using the capacitor-charging power
supply. Second, the valve is activated and process gas is puffed into the breech end, which initiates
breakdown as shown in Fig. 2.11. The ionized gas is accelerated downstream by the Lorentz force, and
the ionization front remains substantially stationary in the laboratory frame as additional inflowing
gas is ionized at the breech.
This conceptual picture was experimentally investigated in detail by Poehlmann using several
methods [12, 17, 4749]. The following section briefly summarizes the results of one such study, in
order to establish what was known of the SPG facility prior to the experimental investigations of
Chapters 3, 4, and 5.
gas puff
Figure 2.11: Schematic depiction of the acceleration process in the inter-electrode region.
2.6.2 Prior Work on the SPG Facility

In order to directly measure the axial distribution of the radially flowing plasma current, Poehlmann
et al. devised an array of Rogowski coils that was configured to be inserted into the electrode
volume during operation. A Rogowski coil is a type of fast current transformer consisting of a
toroidal solenoid, wound about a central conductor. A transient current passing through the central
conductor induces a voltage in the toroidal solenoid, which can be measured at an oscilloscope to
infer the current passing through the central conductor as a function of time. Using an array of such
coils, as shown in Fig. 2.12, the current flowing through the central electrode at each axial point can
be determined as a function of time. By taking the difference between the currents flowing at each
successive coil of the array, the radial current that passes from the plasma to the electrode between
each pair of coils can be determined. This gives the axial distribution of the radially flowing current
at each time.
Fig. 2.13 shows the array installed into the SPG. While occupying such a significant fraction of
the electrode volume can be expected to have an appreciable impact on the device operation, it was
assumed that the fundamental operation remains qualitatively similar, at least insofar as the current
distribution dynamics.
Fig. 2.14 depicts a plot of the current distribution within the SPG as a function of elapsed time
since discharge initiation, reproduced from Ref. [12]. Note that at early times (e.g., 3, 5, and 6
s), the current is localized at the upstream end of the electrode volume. At later times (e.g., 7, 8,
and 10 s), the current distribution is biased slightly toward the downstream end, but appreciable
current densities continue to be present at each axial position along the accelerator axis. This is
consistent with a deflagration-type discharge, as described in Ch. 1.
In addition to the Rogowski coils, a fast-framing ICCD camera (Cordin 220) was used to take a
sequence of frames of the discharge as it evolved. An example image sequence is shown in Fig. 2.15.
It is plain from these images that the underdamped RLC oscillation of the drive current produces two
discharge events of drastically different character. The first is diffuse and produces a high velocity,
collimated jet. The second generates a confined luminous region, assumed to correspond to current
conduction, that appears to propagate downstream.
The behaviors measured by the Rogowski coil array and observed in the ICCD images are strongly
indicative of the solutions to the MHD Rankine-Hugoniot relations. However, in order to evaluate
a set of conservation equations across a discontinuity, a direct measurement of the conserved quan-
tities must be made on both sides of the discontinuity (or, alternatively, an accurate estimate of
the conserved quantity must be made). Though the upstream neutral gas parameters were well
characterized in the process of designing and testing the puff-valve, there are no such measurements
of the plasma conditions downstream of the hypothesized plasma deflagration. It is the need for
these data that motivates the experiments conducted in the next chapter.
Figure 2.12: Rogowski coil array used in Ref [12]. Photograph by F. Poehlmann.
Figure 2.13: Rogowski coil array used in Ref [12], installed in the inter-electrode volume. Photograph
by F. Poehlmann.
! %&
'()*+,- './/-0) 1234

#" " %& G>(0E-
#! %& ) @ " A7
30 #! CHAPTER 2. STANFORD PLASMA GUN FACILITY
#" %&
" $! %&
) @ 9 A7
! $" %&
! " #! #"
56&- 1 74
) @ < A7
BDC =(,6(> './/-0)7
#$
; 7 < 7
#! " 7 : 7
=(,6(> './/-0) 1234
9 7 #! 7
: ) @ : A7
$ ) @ #! A7
! " #! #" $! $"

3%%->-/()+/ 3?67 1%&4
!
Figure 2.14: Radial current distribution as a function of elapsed time since the onset of breakdown, " #! #" $!
obtained from the Rogowski coil array. Figure by F. Poehlmann.
FIG. 7. !Color online" Cathode current at different axial locations !after digital low-pass filtering, " = 1 # 107 s" !a"
currents !b" and fast framing camera images of plasma discharge inside coaxial plasma gun !c" for operation at 1 k
!112 !F" and 50 nH transmission line inductance.
The measurements show that the current was concen- process when the anode vo
trated within the first 5 cm of the gas entrance during the first the coils can facilitate brea
4 !s of the pulse. During the next 5 !s the current spread causing a portion of the cu
along the cathode, and by 7 !s into the discharge, the cur- at downstream locations
rent was distributed almost uniformly along the first 15 cm down limit. This sudden ch
of the cathode. This measurement of the current distribution transient in the measured
is consistent with the optical emission detected by the fast Alternatively, the sudden
framing camera images presented in Fig. 7!c". The frame rate be caused by a local rec
and exposure time of the shown image sequence were 2 !s distribution. However, the
2 s 20 s
6 s 22 s
12 s 24 s
14 s 26 s
16 s 30
28 s
18 s 34
30 s
Figure 2.15: Sequence of fast-framing images of the discharge, showing two distinct discharge modes
sequentially in time.
Chapter 3
Exploring the Far Field: Scanning

Quadruple Langmuir Probe
3.1 Introduction
As discussed in the previous chapter, the bulk of the work performed to characterize the Stanford
Plasma Gun by Poehlmann et al. was directed toward understanding the mechanics of the discharge
process within the electrode volume. However, no significant data was collected on the plasma
parameters within the plasma jet itself. This region is of interest for at least two reasons. First,
every practical application of the plasma deflagration accelerator concept is at least partially based
upon the properties of the plasma within the jet, because this is the region of the discharge that
either interacts with physical objects to generate conditions of interest or the properties of which
dictate the efficacy of the system as a propulsive device. Second, having characterized the upstream
gas parameters as a result of assessing the operation of the valve (see Section 2.5.2), a subset of the
conservation equations underlying the MHD Rankine-Hugoniot relations can be evaluated using the
downstream plasma parameters without a detailed mechanistic understanding of the intermediate
processes (i.e., the plasma deflagration and/or detonation waves).
The majority of the material contained in this chapter is reproduced from Refs. [34, 36], with
minor modifications for clarity. As such, certain introductory and explanatory portions may be
duplicative of material contained in previous chapters, but have been retained for completeness.
In this chapter, we show that current-driven ionization waves exhibit analogous behavior to
that of classical combustion-driven shocks described by the Rankine-Hugoniot relations. We collect
concurrent, time-resolved experimental data capturing the MHD field variables both upstream and
downstream of current-driven ionization waves using an immersed multiple-probe technique. Having
thus determined the jump conditions across the wave, we are able to categorize it as either a plasma
32
3.1. INTRODUCTION 33
detonation or a plasma deflagration. We use this information to compare our data against conditions
predicted by the Rankine-Hugoniot relations for MHD shock waves [13, 50], appropriately modified
to take into account various unique aspects of the discharge. We not only observe the two distinct
wave types predicted by the theory, but also obtain excellent quantitative agreement among the
measurable flow parameters for both types of wave. These results provide a rigorous basis for a
straightforward, reduced-order theoretical tool that can be used to analyze these complex, three
dimensional, multi scale, and non-equilibrium pulsed plasma systems. This work also cements the
marriage between magnetohydrodynamic phenomena and the classical Rankine-Hugoniot model for
combustion waves, which each appear in a multitude of physical systems but can now be unified
beneath one coherent theoretical description.
Current-driven magnetohydrodynamic (MHD) ionization waves and acceleration mechanisms are

a key feature of many natural and artificial plasma systems. In such systems, an external energy
source drives current through a plasma and in doing so generates electromagnetic field gradients that
influence the acceleration, compression, and heating dynamics of the system. These processes provide
pathways for the coupling of magnetic, thermal, and kinetic energy modes in magnetized plasmas,
strongly impacting their behavior. The magnetic pressure gradients produced as a consequence of
such waves have been shown to have an essential role in the generation of astrophysical jets [51
54]. These waves also underpin the formation and acceleration of spheromak and compact toroid
plasmas [5560], as well as the recent advances in plasma-jet driven magnetized target fusion [61, 62].
Plasma jets produced as a consequence of MHD ionization wave dynamics can generate strong shear
flows, which have themselves been used in novel Z pinch stabilization schemes [6365] and plasma-
material interaction studies [38, 66, 67]. Due to increasing interest in these plasma phenomena,
there is a critical need for a reliable model that accurately predicts the behavior of plasmas formed
and sustained by MHD ionization waves.
Often, fruitful progress in developing such a theoretical description can be made by importing
ideas and theories from separate, but related, fields of physics. In combustion science, similar dy-
namic energy transformations as those discussed above also occur in combustible gases, and take
the form of combustion waves, driven by the latent chemical energy of the flow. When a steady,
one-dimensional (or quasi steady, quasi one-dimensional) description is appropriate, these waves
fall into one of two categories: deflagrations and detonations. Each of these two categories corre-
sponds to different sets of qualitative and quantitative characteristics, of both the wave behavior
and the properties of the processed gas. Furthermore, they each correspond to a different branch
of allowable solutions to the relevant conservation laws, applied across the idealized discontinuity
of the combustion wave; collectively, these allowable solutions are described by what is known as
the Rankine-Hugoniot curve, which is simply a graphical representation of the thermodynamic pa-
rameter space that satisfies the Rankine-Hugoniot relations. Though this theory was developed
for one-dimensional, steady combustion waves, circumstantial evidence has been steadily collected
34 CHAPTER 3. EXPLORING THE FAR FIELD
over the course of several successive investigations [12, 13, 21, 24, 62, 65, 6870] that indicates a
similar dually branched, quasi-one-dimensional model may appropriately describe non-chemically-
reacting flows with electromagnetic energy addition; that is, current-driven MHD ionization waves
may exhibit Rankine-Hugoniot behavior. However, a coherent set of experiments that simultaneously
measures sufficient flow field data across such waves that permits direct, quantitative comparison
with the MHD Rankine-Hugoniot model has not been conducted. We convincingly demonstrate
herein that the Rankine-Hugoniot relations, modified to include additional state variables appropri-
ate for magnetohydrodynamic flows, provide a practical and reliable model for quasi-one-dimensional
current-driven ionization waves that is supported by direct measurement of the change in state vari-
ables across the waves. The experiments executed and described in this chapter were designed with
the need for these direct measurements in mind.
3.1.1 Phenomenological Motivation

A typical pair of discharge events is shown in Fig. 3.1 as a series of ICCD images [12, 36]. A first
ionization wave forms at the breech end of the accelerator and rapidly broadens to fill the coaxial
electrode volume (region I), continuing to accelerate plasma in the axial direction downstream once
it has fully developed (region II). As the current decays and crosses zero, the first wave dissipates
and a second ionization wave forms at the breech (region III). This second wave (i.e., a current
sheet) propagates into the gas left behind by the preceding ionization wave and is expelled from the
accelerator (region IV).
Two distinct wave modes are clearly visible in Fig. 3.1, which we hypothesize to correspond to the
detonation and deflagration solutions of the Rankine-Hugoniot jump conditions. The detonations
correspond to when the pressure and density increase across the wave. These waves consist of a
shock front followed by an expansion wave, and typically manifest as relatively sharp discontinuities
in the flow. Different characteristics are observed in the deflagration solutions, wherein the pressure
and density decrease across the wave. In contrast, deflagration waves are broader, propagate slowly,
and produce higher accelerated gas velocities due to the lack of a decelerating normal shock. As
shown in Fig. 3.1, the consecutive waves appear to be a deflagration followed by a detonation: the
first wave (regions I & II) is broad and approximately stationary in the laboratory frame while
accelerating the plasma in the form a jet in the downstream direction. The second wave (regions III
& IV) is a narrow current sheet, and propagates away from the breech into the upstream residual
gas left by the deflagration process before being expelled.
2 s 20 s
I III
6 s 22 s
12 s 24 s
14 s 26 s
II IV
16 s 30
28 s
18 s 34
30 s
Figure 3.1: ICCD images of consecutive current-driven ionization waves. I. Establishment of the
plasma deflagration in the coaxial accelerator volume, broadening towards exit plane. II. Acceler-
ation and expulsion of the plasma through the quasi-stationary deflagration. III. Formation of the
detonation at the breech of the accelerator as the deflagration dissipates. IV. Propagation of the
detonation wave along the accelerator axis. The dark silhouetted lines are the anode rods, as shown
in Fig. 3.4. Time stamps are correlated with the contour in Fig. 3.9.
3.1.2 Derivation of the MHD Rankine-Hugoniot Relations
The general form of the Rankine-Hugoniot curve is shown in Fig. 3.2a, and succinctly illustrates the
two allowable solution branches as well as several other salient features. A more detailed discussion
of the Rankine-Hugoniot curve and relations (primarily as they pertain to combustion theory) is
contained in Ref. [16]; for the present work, only the fundamental features require discussion. The
origin of the curve, labeled as such in Fig. 3.2a, represents the initial state of the flow (i.e., p1 and 1 ).
The families of lines emanating from the origin, shown as dotted red or green lines in Fig. 3.2a, are
referred to as Rayleigh lines and are derived from combining the mass and momentum conservation
equations relating the initial and final states. The Hugoniot curve itself, the black line in the figure,
represents the locus of points corresponding to the intersection of both families of curves with the
set of allowed final states given by conserving energy across the wave. The distance of the curve
away from the origin is proportional to the energy added to the flow between one side of the wave
and the other. The tangency points of the Rayleigh lines to the Hugoniot curve are referred to as
Chapman-Jouget (C-J) points, and there are two such points for any given Hugoniot curve (one
corresponding to the detonation branch, and one to the deflagration branch).
In the physical device used in this study (described fully in Ch. 2), two consecutive ionization
waves are produced. This leads to successive transitions between states, such as that shown in
Fig. 3.2b. The end state of the first event is roughly the origin for the second event, hence the two
different Hugoniot curves (labeled 1 and 2 in the figure). We measure the beginning and end states
of both waves, and the theory predicts the wave and plasma velocities (which are each related to the
slope of the corresponding Rayleigh line). By also independently measuring the plasma velocity, we
are able to obtain a check on the practical descriptive power of the quasi one-dimensional model.
Another important (but less mathematically rigorous) test can be applied by observing the
qualitative features of the discharge, and comparing them to the expected qualitative features of
detonations and/or deflagrations that have long been associated with their respective appearances
in combustion phenomena. The qualitative difference in structure between the two types of wave
is shown in Fig. 3.3; a combustion deflagration has a broader, diffuse reaction zone, analogous to
a broad and diffuse region of injected current density in the plasma deflagration case. A chemical
detonation has a thin shock wave followed by an expansion region, analogous to a current sheet
moving through, sweeping up, and accelerating the plasma as in plasma detonation.
We now develop the expressions from MHD Rankine-Hugoniot relations that were tested by
direct measurement, and which are arranged in terms of the measurable input and output variables
(i.e., the mass density, temperature, total pressure, and velocity of the flow). For a quantitative
comparison, we require expressions relating the jump conditions across each type of wave in terms of
the measured quantities (i.e., density, temperature, total pressure, and velocity). We have chosen the
fixed-frame accelerated plasma velocity as the output variable for comparison, so our expressions
relate the downstream plasma velocity directly to the other state variables in both cases.
a)
detonation waves
deflagration waves
strong detonation
upper C-J point
weak detonation
P
prohibited
weak deflagration
origin strong deflagration
lower C-J point
1/
b)
1 2
origin
Det.
Def.
1/
Figure 3.2: (a) The Rankine-Hugoniot curve, illustrating the beginning and end states of a flow tran-
sitioning from the origin (i.e., the initial state) to the final state in one of the two allowed branches.
The Chapman-Jouguet (C-J) points are tangency points of allowed Rayleigh lines to the Hugoniot
curve (i.e., the lines corresponding to the minimum allowed detonation velocity and the maximum
possible deflagration velocity). (b) The state-transition diagram of a process such as that which
occurs in the Stanford Plasma Gun, in which a deflagration leads to an initial state corresponding
to a subsequent detonation. The intersection of the Rayleigh lines with the Hugoniot curve(s) are
illustrative only, and are not meant to imply that the successive waves are C-J deflagrations and/or
detonations.
a)
LB Wave Frame: uu 0 -ud
pu* pd*
Upstream Downstream
u J d
Bu Bd
External Frame: V u 0 Vw= Vu- uu VdLB

b)
UB Wave Frame: ud 0 -uu
p2*
pu*
Downstream Expansion Upstream
Sheet
Zone u
B2 Bu
External Frame: VdUB Vw= uu- Vu > 0 Vu
Figure 3.3: (a) Schematic depiction of a deflagration, with relevant state variables labeled. Note
that the upstream gas in this case is the cold gas injected by the valve, assumed to be instantaneously
ionized. The arrows are indicative of the flow velocity relative to the wave region. (b) Schematic
depiction of a detonation, with relevant state variables labeled. Note that in this case, the upstream
gas is the residual gas left behind by the initial deflagration.
In all expressions, the subscript u should be understood to refer to the upstream conditions, the
subscript d to refer to the downstream conditions, and the subscript 2 to refer to the conditions in
region 2 of the upper branch wave structure. Velocities defined as V correspond to the laboratory
(i.e., externally fixed) frame, whereas velocities defined with a u correspond to the reference frame
fixed to the wave (i.e., moving at the same velocity as the wave).
In the case of the deflagration solutions, we first combine the continuity equation,
u uu = d ud , (3.1)
with the momentum equation (incorporating the magnetic pressure term into the total pressure, p ),
pu + u u2u = pd + d u2d , (3.2)

to obtain the following relation:
pd pu

2
(u uu ) = . (3.3)
1/u 1/d
In terms of the upstream velocity uu directly, this becomes

s
1 pd pu
uu = . (3.4)
u 1/u 1/d
For the plasma deflagration wave, we desire the velocity of the accelerated gas in the laboratory
frame, Vddef . In combustion, deflagration waves are empirically observed to move at a slow and
constant velocity in the laboratory frame; we make the same assumption here. Also, in our system,
the upstream gas velocity is negligible compared to the accelerated plasma velocity, such that we
can make the following approximation:
u
Vddef = Vd Vu Vd = ud uu = uu uu . (3.5)
d
Combining the above expression with Eqn. 3.4, we readily obtain

s
1 1
Vddef (pd pu ), (3.6)
u d
the downstream velocity of the plasma accelerated by the deflagration wave.
For the plasma detonation waves, the analysis proceeds in two steps. First, the plasma prop-
erties in region 2 are calculated from the parameters in the upstream region (i.e., the conditions
in the downstream region of the preceding deflagration event) via the hydromagnetic shock jump
conditions. Then, the downstream plasma conditions are calculated using a model for the expan-
sion region similar to that developed for the deflagration solutions. The hydromagnetic shock first
compresses the upstream gas and raises the pressure to p2 , which is dominated by the magnetic
component, giving
2 2
p2 B22 1 Mu 1
= = +1 , (3.7)
pu 20 pu 1
where here we have used the adiabatic shock jump equation based on the magnetosonic Mach
number Mu . In our system, B2 is a measured parameter derived from the current, and thus Mu can
be calculated and then used to obtain the jump in the density, , across the shock as well. This is
given by
+1 2

2 2 Mu
= . (3.8)
u 1 + 1
2 Mu
2
The pressure and density in region 2 are now known, and the second step is to calculate the down-
stream gas velocity as though the expansion zone is a lower branch wave, with upstream conditions
given by p2 and 2 . Since the expansion zone must be moving at the same speed as the hydromag-
netic shock, the velocity of the upstream gas with respect to the deflagration wave (u2 ) is already
determined from the computed Mu .
The downstream velocity, ud , can only be determined as a function of the strength of the hydro-
magnetic shock at the front of the detonation wave, since the degree to which the shock decelerates
the flow determines the degree to which it can be re-accelerated by the following expansion wave
towards Md = 1. The strength parameter, alpha (), is thus defined to express ud as a fraction of
the magnetosonic speed, i.e. s
pd
ud = . (3.9)
d
Given this free parameter, we employ the continuity relation to obtain
(2 u2 )2
pd = , (3.10)
2 d
and combine the above equation with the form in Eqn. 3.4 to directly obtain d in terms of the flow
conditions in region 2, i.e.:
2
(2 u2 )2 +1
2
d = . (3.11)
22 u22 + p2
If is assumed to be known, then Eqns. 3.10 and 3.11 can be substituted into Eqn. 3.9 to
obtain the downstream plasma speed resulting from acceleration by the plasma detonation wave.
The upper limit of is = 1, but we can compute the minimum value by noting that the weakest
possible expansion wave following the shock front corresponds to the case pd = p2 . This allows pd
to be eliminated from Eqn. 3.9 and the expression to be solved for min , given by
r
2
min = u2 . (3.12)
p2
The shock is observed to move significantly faster than the gas ahead of it, so the velocity of
the accelerated plasma downstream of the wave in the laboratory frame can be determined by
approximating Vddet uu ud . We can thus combine Eqns. 3.9, 3.10, and 3.11 into a single
expression for the downstream velocity of the plasma detonation wave, given by
2

p2 pu +
2 +1
u
p
2 u p
2
Vddet = r . (3.13)
(p2 pu ) u 1 u2
Eqns. 3.6 and 3.13 are the expressions necessary to test the theoretical framework, provided the
requisite data are collected regarding the initial and final states of the plasma being processed by
each wave.
3.1.3 SPG Operating Conditions

The process gas used in these experiments was molecular nitrogen gas, puffed into the breech end
of the accelerator (the left side in Fig. 3.4a) to initiate the discharge. A single 56 F capacitor bank
(i.e., four of the capacitors described in Ch. 2) was used to drive the discharge, initially charged
to 3 kV, such that the current and voltage waveforms are decaying sinusoids (an example current
trace is shown in Fig. 3.6) that lead to a secondary breakdown event corresponding to the second
half-period of oscillation.
Gas injection is performed using the custom built and designed fast-puff valve, previously de-
scribed in detail in Chapter 2. The short rise-time valve injects a high density puff of neutral gas into
the breach end of the accelerator, which initiates breakdown nearly instantaneously. The incoming
neutral density is known from the neutral gas inflow characterization data (i.e., the mass bit and
pressure rise time data at the gas inlet of the gun volume). It is from this information that the
initial plasma conditions immediately following breakdown are estimated.
Electrode Bias Network

18V 3V
Arrangement 18V
1
2 I4 I2 I3
4 Measured Currents
3
Vacuum Chamber
3 kV
Linear Axis Stage
56 F QLP
Accelerator
Inflow Breech Exit Plane
Figure 3.4: Schematic of the quadruple Langmuir probe (QLP) setup, and the bias network asso-
ciated with the individual probes.
a) Probe 1 Noise Probe
Accelerator Exit Plane Probe 4

Probe 2
Probe 3
Probe Lead Shield Linear-Axis Stage
b) translation direction
noise probe
QLP
Figure 3.5: (a) Photograph of the experimental test section, including a closeup view of the QLP
and associated noise probe. (b) Time-integrated photograph of the SPG during firing (encompassing
both events) and depicting the QLP at a particular axial position.
3.1.4 Probe Theory
The downstream plasma state is measured using a quadruple Langmuir probe (QLP) in current-
saturation mode [71], which provides the plasma density and temperature at high spatial and tem-
poral resolution. The probe consists of four independent electrodes, biased relative to one another
as shown in Fig. 3.4a, and supported by an insulating ceramic substrate. Electrodes 1,2, and 4
are oriented parallel to the flow direction, while electrode 3 is perpendicular to the flow direction.
The plasma currents collected by electrodes 2, 3, and 4 are measured via wideband Pearson current
monitors, and the current collected by electrode 1 is calculated via a current balance between the
four electrodes. Conversion of the measured currents and bias voltages to the plasma state variables
is accomplished by the solution of a nonlinear system of equations derived from kinetic theory [72],
and developed for this particular probe configuration to provide plasma temperature Te , plasma
density ne , plasma potential , and ion thermal Mach number Si [71].
In order to calculate the four independent plasma state variables, the measured current and
associated bias voltages must be coupled with a theory describing the current collected by the probe
surface. Similar to other electrostatic plasma probes, the collected probe current, Ip , is composed
of contributions due to both electrons and ions and can be written as
Ip = Ie Ii , (3.14)
where Ie and Ii correspond to electron and ion current, respectively. For any electrode exposed to
the plasma flow, the collected current will correspond to electron flux if the probe potential is less
than the plasma potential (probe plasma ) according to the expression:

e
Ike = Aprobe Je0 exp (plasma probe ) , (3.15)
kTe
which is a function of the probe area both parallel and/or perpendicular to the plasma flow direction,
Aprobe , and the thermal diffusion of electrons. In Eqn. 3.15, Ike is the electron current parallel to the
flow direction, Je0 is the initial electron current density, e is the electron charge, k is the Boltzmann
constant, and Te is the electron temperature. Current due to ion collection, however, is a more
complex function of the operating regime of the probe. More specifically, the ratio of the local
probe radius to the Debye length, rp /D , and the ratio of ion to electron temperature, Ti /Zi Te ,
governs which theory should be applied. If 5 rp /D 100 and Te /Zi Ti 1, where Zi is the ionic
charge, Ref. 73 developed a relationship for ion current collection as a function of empirical fitting
parameters and based upon experimental measurements [74],

e
Iki = Aprobe Ji0 + (plasma probe ) , (3.16)
kT
0.75
2.9 Ti
= + 0.07 0.34, (3.17)
ln(rp /D ) + 2.3 Zi Te
" 3 #
Ti rp
= 1.5 + 0.85 + 0.135 ln . (3.18)
Zi Te D
An analytical expression for the ion current collected by the perpendicular probe in particular was
obtained by assuming a negligible sheath thickness ds [75], i.e., ds /rp 1, such that
1/2
kTi 2
exp Si2

Ii = A ne e
2me

XS j
i 3
j+ , (3.19)
j=0
j! 2
where Si is the ion thermal Mach number, A is the cross section of the probe perpendicular to the
plasma flow, ne is the electron density, and me is the electron mass.
Coupling these expressions with the assumptions of quasineutrality (ne = n) and thermal equi-
librium (Ti = Te = T ) between the electron and ion plasma components, a set of four coupled
nonlinear equations for the desired parameters can be developed, giving:

ep1 ep1
I1 = Ak Je0 exp Ak Ji0 + , (3.20)
kT kT

e(p1 + 12 )
I2 = Ak Je0 exp
kT

e(p1 + 12 )
Ak Ji0 + , (3.21)
kT

e(p1 + 13 )
I3 = Ak Je0 exp
kT

e(p1 + 13 )
Ak Ji0 + , (3.22)
kT

e(p1 + 14 )
I4 = A Je0 exp
kT
1/2
Sij

kT 2 2
X 3
A ne e exp Si j+ , (3.23)
2me j=0
j! 2
where ij represents the potential difference between probe i and probe j as applied by the bias
network, with subscript p representing the plasma potential. These equations are solved at each
recorded data point in time to obtain a time-series of the derived plasma parameters in a single shot.
In the event that rp /D > 100, the ion collection theory used in Eqn. 3.16 is invalid. In this case, a
thin sheath is assumed and the ion current follows a Bohm expression [76],
r
kT 1
Iki = Ak ne e exp . (3.24)
mi 2
Incorporation of this thin sheath assumption leads to the following revised set of current balance
equations, valid for a thin sheath rp /D > 100:
r
ep1 kT 1
I1 = Ak Je0 exp Ak ne e exp (3.25)
kT mi 2

e(p1 + 12 )
I2 = Ak Je0 exp
kT
r
kT 1
Ak ne e exp (3.26)
mi 2

e(p1 + 13 )
I3 = Ak Je0 exp
kT
r
kT 1
Ak ne e exp (3.27)
mi 2

e(p1 + 14 )
I4 = A Je0 exp
kT
1/2
Sij

kT 2 2
X 3
A ne e exp Si j+ . (3.28)
2me j=0
j! 2
The systems described by Eqns. 3.20-3.23 and Eqns. 3.25-3.28 are solved using a standard
Newton-Raphson algorithm with a centered trust region. The initial guess of each state variable
provided to the solver are obtained by solving the algebraically-reduced system, which provides a
first order solution of the derived plasma parameters. In order to ensure the correct ion collection
model is used for each data point, the system of equations corresponding to the thin sheath assump-
tion are solved first, and the resulting Debye length is computed. If rp /D is such that the theory
developed in Ref. [74] is valid, the state variables are recalculated using the equations corresponding
to the empirical fit.
The chief assumption regarding the kinetic theory of the ion and electron fluxes to the probe was
that the electron and ion populations were substantially thermalized with one another (i.e., Te = Ti ).
The first basis for this assumption is the measurement of the ion thermal Mach number (Si ) by the
probe itself. Though the measurement was not highly sensitive to changes during the sequence of
wave events (i.e., the structure is not as clearly visible), comparison of the Mach number to the
measured plasma velocity allows the ion temperature to be approximated. For a representative
measurement, Mi ' 2.5 3.0, which gives an ion temperature of 10 eV for the first event, which is
of the same order as the peak observed electron temperature. The second basis for this assumption
is that the energy relaxation time, based on the Coulomb collision frequency between ions and
electrons, is of the same order as the plasma residence time in the experiment. Furthermore, the
additional probe parameters are only minimally sensitive to the Te /Ti ratio and even variation from
0.1-1.0 does not significantly alter the results.
3.1.5 Spatial Scanning

The experimental procedure for collecting the QLP data was as follows: with the probe at a given
axial position of the linear stage (shown in Fig. 3.4a), the accelerator was fired and the probe currents
were measured for the duration of the drive current ringdown time (including both the primary and
secondary breakdown events). The QLP was then traversed along the accelerator axis in increments
of 5 mm, and the process was then repeated in order to sample axial locations between 50-215 mm
from the accelerator exit plane. A repeatability study was also conducted, and the magnitude of the
measured currents was found to vary by less than 1% from shot to shot at a single axial position. The
spatial response of the probe was dictated by the included volume of the multi-electrode assembly,
which was approximately 0.125 cm3 , based on a 5 mm linear scale length. The 5 mm axial increments
described in the main text are the result of this minimum spatial resolution.
The temporal response was obtained by measuring an applied current driven through the probe
wires, and the bandwidth was found to be unaffected by any stray capacitance of the probe assembly
and was instead limited by the bandwidth of the Pearson Current Monitors used to measure the
currents (100 MHz).
3.2 Results
3.2.1 Probe Solutions

The axial-scan technique described in Sec. 3.1.5 enabled the acquisition of representative spatiotem-
poral contours of the plasma parameters along the accelerator axis, downstream of the exit plane,
for the consecutive ionization waves. In order to compare our results to the theory, it remains to
combine the experimentally determined state variables on the righthand side of each downstream
velocity equation (Eqns. 3.6 and 3.13, respectively) in order to obtain the predicted downstream ve-
locity for each type of wave. The state variables are obtained as follows: For the deflagration wave,
the downstream pressure is calculated assuming the plasma equation of state to be pd = ne kB Te ,
consistent with observations by Woodall et al. that the downstream plasma has very little entrained
radial current and magnetic flux [21]. For the detonation wave, the time at which the current sheet
is formed is determined using the discharge current trace (i.e., at the peak voltage magnitude of the
3.2. RESULTS 47
4 I1 40
ID
I2
Discharge Current [kA]

2 20
Probe Current [A]
0 0
2 20
I4
4 I3 40
5 10 15 20 25 30 35 40 45
Time [ s]
Figure 3.6: An example time series of all three measured probe currents, the fourth probe current
calculated via current balance, and the discharge current (alternative y-axis).
4.0 10
21
50 mm
3.5 165 mm
3.0
2.5
[1/m3 ]
2.0
ne
1.5
1.0
0.5
0.0
5 10 15 20 25 30 35 40 45
Time [s]
Figure 3.7: Calculated plasma density time series for two example probe locations (50 mm and 165
mm from the exit), showing the clear time-shift of the density front of the LB wave and the peak of
the UB wave.
12
10
8
[eV]
6
Te
0
5 10 15 20 25 30 35 40 45
Time [s]
Figure 3.8: Calculated plasma temperature time series for the 50 mm probe location.
negative oscillation), and the upstream conditions are taken to correspond to the probe data col-
lected at that time. The pressure in region 2 (as shown in Fig. 3.3 is again assumed to be dominated
by the magnetic component, obtained via the measured discharge current. Given p2 , the hydro-
magnetic Mach number of the deflagration wave can be calculated, allowing 2 to be determined
from the adiabatic hydromagnetic shock jump equations. To incorporate the measured downstream
densities into the model, we renormalize the density to a Gaussian integral over the radial profile
with the measured axial value as the peak, consistent with the observed radial fall-off in bulk plasma
emission ( n2e ).
3.2.2 Spatiotemporal Contours

The velocity of the accelerated plasma is determined from the slope of the leading edge of the
spatiotemporal plasma density contours, providing the key state variable to test against the pre-
dicted downstream plasma velocities. Other important information necessary to fully describe the
plasma state, but not collected directly by the QLP, include: the upstream conditions for the first
ionization wave (i.e., the injected gas conditions) and the magnetic pressure profile as a function
of time and space. The former was measured at the injection location using piezoelectric pressure
transducers [37], and the upstream plasma density was thus obtained assuming full ionization. The
latter was calculated by averaging the theoretical azimuthal field (based on the measured discharge
current) over the radius of the plasma accelerator to obtain a spatially averaged field strength as a
function of time (see Eqn. 3.29).
3.3. DISCUSSION 49
Figure 3.9: Contours of free electron density in z-t space, shown as function of distance from the
accelerator exit plane in the axial direction and time from initial breakdown following primary gas
injection. Regions of space and time in which the QLP equation system did not converge (i.e. no
currents were being collected from the plasma) are represented by the black area in the lower portion
of the figure.
The two distinct events are also evident in the z-t plasma density contours derived from the
QLP data, shown in Fig. 3.9. The dotted white lines indicate the leading edges of the contours
corresponding to the first and second ionization waves, which were used to determine the accelerated
plasma velocities in each case. The plasma plume resulting from the first wave is broad in time,
and expands as it moves axially downstream. The second wave is narrower in time, and the plume
moves downstream at a slower velocity. These characteristics are consistent with the two solution
branches of the Rankine-Hugoniot relations, and strongly suggest they govern the dynamics of the
system.
3.3 Discussion
3.3.1 Model Comparison
The numerical values used to validate the theoretical model are shown in Table 3.1. For the detona-
tion, these values were used in conjunction with variations in between min < < 1 to establish a
range of downstream plasma velocities for comparison, based on Eqn. 3.13. These parameters were
obtained directly from measurements made of the upstream and/or downstream plasma states, with
Deflagration
250 Fit (39.28 km/s)
Ideal Theory (42.64 km/s)
Distance from exit plane, mm
200
150
100
50
11 12 13 14 15 16
Time from initial breakdown, s
Figure 3.10: Comparison of the measured downstream velocity (based on the density front) and
the calculated downstream velocity for the lower branch solution (i.e., the first event). The data
values correspond to a specific contour of the leading edge of the first measured plasma expulsion,
indicated by the lower dotted white line in Fig. 3.9.
the exception of the B field value which is calculated from a spatial and temporal average of the
measured driving current, derived from
Z2 Zr2
1 0 I(t)
B= dr dt (3.29)
(2 1 )(r2 r1 ) 2r
1 r 1
where 2 1 is the characteristic lifetime of the ionization wave and r2 r1 is the radius of the
coaxial plasma accelerator.
The comparisons between the predicted and measured downstream velocities for both waves are
shown in Figs. 3.10 and 3.11. The total pressure and density both decrease across the first wave (see
Table 3.1), indicating that this event is indeed a deflagration-like solution. The second wave exhibits
the opposite behavior, acting as a hydromagnetic piston that ionizes and compresses the upstream
gas. The predicted accelerated plasma velocity for the deflagration case is within 8% of the measured
value, while the measured velocity of the downstream plasma accelerated by the detonation wave
falls within the range of predicted velocities, as indicated by the shaded region of Fig. 3.11. The
measured velocity corresponds to a strength parameter of ' 0.74.
3.3. DISCUSSION 51
Figure 3.11: Comparison of the measured and calculated downstream velocity for the upper branch
solution (i.e., the second event). The range of theoretically allowable downstream velocities are
shown for min < < 1. The data values correspond to a specific contour of the second measured
plasma expulsion, indicated by the upper dotted white line in Fig. 3.9.
This remarkable quantitative agreement between the complex experimental system and the one-
dimensional, steady-state MHD Rankine-Hugoniot model has significant implications for the contin-
ued study of these wave phenomena. The evident subordination of higher-dimensional and unsteady
effects to the straightforward relationships captured by the model indicate that the dominant accel-
eration mechanism, particularly for the deflagration wave, is a highly efficient expansion of magnetic
pressure into directed kinetic energy. It is also of note that a detonation wave always forms sub-
sequent to the deflagration wave, instead of vice-versa. The Rankine-Hugoniot framework predicts
that this should occur; the first wave expands into vacuum downstream, and will thus experience a
density and pressure drop across the wave, rendering it a deflagration solution. The second wave, by
contrast, necessarily must propagate into the conditions left behind by the deflagration wave, which
leads a compressive shock to form, thus rendering it a detonation solution.
This evidence of branching phenomena in consecutive current-driven ionization waves is strong
support for consistency between these systems and the MHD Rankine-Hugoniot model. Such con-
sistency is not typically expected for such complex systems, and while fully capturing the three
dimensional, multi-scale and non-equilibrium mechanisms underlying these processes may require
involved numerical simulations, we have shown that good agreement can be obtained from a straight-
forward theoretical description.
Table 3.1: This table contains the measured values used to calculate the downstream gas velocity
for both the detonation and deflagration solutions. Note that the subscripts refer to the parameters
corresponding to each wave (see Fig. 3.3).
Ahead of the wave Behind the wave

u = 4.6 kg m3 d = 7.3 105 kg m3
Bu = 0.58 T Bd ' 0 T
deflagration
p0u = 0.53 kPa p0d = 3.4 kPa
pu = 136.1 kPa pd = 3.4 kPa
u = 4.1 105 kg m3 2 = 2.2 104 kg m3
Bu ' 0 T B2 = 0.33 T
detonation
p0u = 0.9 kPa p02 ' 0 kPa
pu = 0.9 kPa p2 = 42.7 kPa
Figure 3.12: The plasma deflagration acceleration process, updated to reflect the features investi-
gated in Ch. 3.
3.4. SUMMARY 53
Figure 3.13: Revisited plasma parameter space from Fig. 1.1, wherein the solid red oval indicates
the parametric region mapped out using the QLP.
3.4 Summary
In this chapter, we have demonstrated that the SPG produces a jet velocity that is in agreement
with what we would expect to be produced by a magnetic pressure driven wave that corresponds to
a deflagration-like solution of the MHD Rankine-Hugoniot relations. Likewise, the parameters of the
secondary breakdown are consistent with a plasma detonation (or, alternatively, a snowplow). We
have extended our understanding of the SPG acceleration process to include the region downstream
of the exit plane, as reflected in Fig. 3.12. We have also refined our knowledge of the parameter
space accessible by way of the expelled plasma jet, at least in the far-field region; an approximation
of this region in Te -ne space is shown in Fig. 3.13.
However, our understanding even of this region is fundamentally incomplete: it is plain from the
QLP data that the plasma parameters are likely to vary significantly along the axial direction as the
plasma jet expands into vacuum, and both the time-integrated and time-resolved imagery indicates
that the region in the vicinity of the exit plane of the SPG is of great interest. In correcting these
deficiencies, however, we are limited by the QLP diagnostic, which has three critical disadvantages
if our aim is to satisfy the objectives of this dissertation as set out initially in Ch. 1. First, the
QLP cannot probe the region near the exit plane, due to the risk of arcing between the central
electrode of the SPG and the probe assembly (and the likelihood of surpassing the usable plasma
parameter ranges associated with the QLP). Second, the useful parameters extracted from the QLP
study relied upon an axial scan consisting of dozens of independent trials, assembled together into
a spatiotemporal contour; a scaling study relying upon this methodology is thus clearly infeasible.
Finally, the QLP perturbs the plasma state as an inherent consequence of its nature as an immersed
probe; this forces us to rely upon the models and assumptions outlined above regarding its interaction
with the plasma, which places an upper bound on the accuracy of the probe.
In the next chapter, we present a study designed around a more advanced diagnostic tool that is
designed to address the aforementioned challenges, in order to explore the near-field of the expelled
plasma jet and further enhance our understanding of the SPG system.
Chapter 4
Exploring the Near Field: Imaging

Spectroscopy
4.1 Introduction
The QLP study described in the previous chapter provided a wealth of data about the plasma
parameters of the deflagration plume, and enabled a quantitative comparison of the quasi-one-
dimensional MHD Rankine-Hugoniot model to the results obtained directly from the SPG. However,
a number of questions remain that cannot be readily answered with an immersed probe, due to
limitations of such tools.
First, the energy density of the plasma deflagration reaches its extreme in the vicinity of the
exit plane, and the QLP cannot be located there without the risk of arcing and other damage
to the probe, the gun, and other systems. This is true even at the comparatively low charging
voltage (3 kV) operating condition considered in Chapter 3, and would naturally increase at higher
energy operating conditions. Unfortunately, for the purposes of applying the high energy jet to
various related fields (e.g., plasma material interactions), it is precisely this region where the plasma
properties, and in particular the density and temperature, are desired. Thus an immersed probe is
wholly unsuitable for deeper study of these features.
Second, a probe that is placed in the plasma flow inherently disturbs the local plasma environment
and thus obscures, to an extent that can only be corrected through the various assumptions discussed
in Ch. 3, the true state of the unperturbed plasma. This is an inescapable fact about any plasma
diagnostic that must be directly placed in the plasma environment, and is the principal reason for
the popularity of optical diagnostics. One class of optical diagnostics of plasmas is based on the
spectrum of light that they emit; in other words, plasma spectroscopy.
55
56 CHAPTER 4. EXPLORING THE NEAR FIELD
Figure 4.1: Time-integrated image (5 second exposure) of the exit plane of the Stanford Plasma
Gun during firing. The dotted oval indicates the approximate region from which light is collected
by the imaging spectrometer.
The majority of the material contained in this chapter is reproduced from Refs. [33, 41], with
minor modifications for clarity. As such, certain introductory and explanatory portions may be
duplicative of material contained in previous chapters, but have been retained for completeness.
The spectroscopic diagnostic we discuss in this chapter focuses on lineshape analysis. In particu-
lar, we measure the Stark broadened profile of a spectral line of hydrogen emitted by the plasma, as
well as the Doppler broadened lineshape of an impurity line spectrally adjacent to the hydrogen line.
The Stark broadening is related to the plasma density through an empirically derived scaling law,
and the Doppler broadened lineshape allows the estimation of the bulk plasma temperature. An
imaging spectrometer was used to gather the data in order to collect spatially-resolved spectroscopic
data across a trans-axial slice of the plasma jet, immediately downstream of the exit plane where
an immersed probe would be unusable. We collected data from the region indicated in Fig. 4.1.
This path-integrated spectroscopic data is mathematically transformable into a radial profile of the
average plasma density distribution over the first 10 s of the discharge, which encompasses the
deflagration in its entirety.
By collecting this data at a range of operating conditions, we measured the scaling of the peak
plasma density and the line density of the jet scales in terms of the discharge current. The scaling
of these parameters with current suggests a radial compression within the expelled jet that occurs
several centimeters downstream of the exit plane of the SPG. Based on these observations, we
assumed that the plasma could be considered to be in radial quasi-equilibrium and applied a suitable
equilibrium relation corresponding to a simple pinch model. We then used this model to predict the
4.2. LINE BROADENING MECHANISMS 57
plasma temperature based on the measured density profiles, and checked the predicted temperatures
against the estimated temperatures based on the Doppler-broadened impurity lines. Finally, we
modified the simple model discussed in Chapter 3 with a loss term based on our findings, and
reevaluate its applicability in light of the added term.
4.2 Line Broadening Mechanisms

The visible spectrum of light emitted by plasmas is a rich source of data that can be used to
diagnose various plasma properties without perturbing the plasma. In particular, using a suitably
sensitive spectrometer, the shape of individual lines can be measured, and plasma parameters that
directly influence the lineshape thus deduced. In this study, we use two broadening mechanisms to
determine the plasma density and temperature, respectively. We use the Stark-broadening of the
H- hydrogen line to determine the plasma density, and Doppler-broadening of an adjacent impurity
line to determine the temperature of the impurity species within the plasma.
4.2.1 Stark Broadening

An external electric field applied to an atomic or molecular emitter causes what is known as the
Stark effect, in which spectral lines of the atom or molecule are shifted or split. There are both
first- and second-order effects, which are linear and quadratic in terms of the field, respectively. The
Stark effect is the electrostatic analog of the magnetically-driven Zeeman effect, and can be used
similarly to measure the strength of the external field applied to the emitter.
The mechanism of the Stark effect can be understood by imagining a negatively charged electron
orbiting a positive nucleus. This electron has an energy state associated with its orbital. In the
presence of an applied electric field, the electron and nucleus will experience opposing forces, which
in turn affects the energy potential between the electron and the nucleus that dictates the associated
energy level. This will shift the energy (and thus wavelength) of the spectral line; in the case of
degenerate states, it may also cause the line to split into two distinct levels/lines.
Even in the absence of an applied electric field, the Stark effect causes spectral lines of atomic
species in a plasma (or otherwise surrounded by charged particles) to broaden from their theoretical
minimum width in a neutral gas, . This is because the atom (i.e., emitter) is influenced by microscopic
electric fields resulting from free electrons in the plasma, even though the macroscopic electric field
is negligible (that is, in the bulk plasma beyond the sheath). The electron energy level is tugged
relative to its unperturbed state by these microscopic fields, leading to perturbations in the energy
level, as shown in Fig. 4.2.
These small stochastic perturbations in the energy levels of the many emitters in the plasma
result in a broadened spectral line. The degree to which the line is broadened can be correlated with
the plasma density (i.e., the free electron density in the plasma). This is typically done empirically,
E=0 E0
n=4 n=4
n=3 n=3
n=2 n=2

Figure 4.2: Schematic illustration of a Stark-broadened transition (right) versus a theoretical
infinitesimal-width transition in the absence of Stark-broadening (left).
as the correlation depends strongly on the emitting species, and can also depend to varying degrees
upon other plasma parameters (e.g., electron temperature, ion temperature, etc.). Hydrogen is
the most closely studied species in relation to Stark broadening, due to its ubiquity in plasmas of
interest and strong correlation between line width and electron density with weak dependence on
other plasma parameters.
4.2.2 Plasma Density Correlation

When excited hydrogen atoms present in the plasma emit radiation, they do so according to well-
characterized series of spectral lines (and corresponding energy levels). One of these named series,
the Balmer series, comprises visible lines at 410 nm, 434 nm, 486 nm, and 656 nm resulting from
electron transitions to the n = 2 energy level. These lines are referred to as H-, H-, H-, and H-,
respectively. In these experiments, both the H- and H- lines were bright enough to see on the
imaging spectrometer, but at the range of conditions tested only H- had a signal of high enough
quality (i.e., the line shape was discernible at lower energy conditions with relatively low emitted
intensity).
The H- line is well correlated to the plasma density, and is only weakly sensitive to temperature.
The correlation relied upon for this work was investigated for various hydrogen lines in Refs. [77, 78],
from which Fig. 4.3 is reproduced. This piecewise function results in the following correlation
4.2. LINE BROADENING MECHANISMS 59
Figure 4.3: The Stark-broadening correlation used in this work, reproduced from Ref. [77].
between the full-width at half-maximum (FWHM), in nanometers, of the Lorentzian line shape and
the plasma density, ne :
1
0.09014
log10 (FWHM + 3.292)
ne = log10 FWHM 3.292
0.0272
log10 FWHM + 21.12708
log10 ne = log10 FWHM < 3.292 (4.1)
0.99262
By fitting a line shape to the broadened, spectrally-resolved emission intensity data, we obtain the
FWHM used to compute the plasma density from Eqn. 4.1.
4.2.3 Doppler Broadening

The well-known Doppler shift occurs in a sound wave when the source is moving towards or away
from the observer; as the source moves away from the observer, the apparent frequency of the sound
decreases, and vice versa. In a cosmological context, the same phenomenon occurs with light, and
objects in the universe that are moving away from observers on Earth display a spectrum that is
red-shifted relative to its stationary wavelength (whereas an object moving towards the observer will
emit a blue-shifted spectrum).
The same phenomenon occurs for an emitter within the plasma, such as a hydrogen atom, as it is
jostled around by other plasma particles as shown in Fig. 4.4. The emitters have some distribution of
velocities that is related to the plasma temperature (and therefore the temperature of the emitting
species, which may not be in thermal equilibrium with the bulk plasma). The instantaneous velocity
T=0 T0

Figure 4.4: Schematic illustration of a Doppler-broadened transition (right) versus a theoretical
infinitesimal-width transition in the absence of Doppler-broadening (left).
of each emitting atom will have some component toward or away from the observer (in this case, the
spectrometer), and will thus emit light that is slightly blue- or red-shifted, respectively. This has
the net effect of broadening the spectral line according to the distribution of velocities, and thus also
according to the temperature of the emitting species. Thus, we can measure the Doppler-broadened
width of the spectral line shape and calculate the temperature of the species, based solely on the
energy of the transition (i.e., the frequency/wavelength of the line), the mass of the emitting species,
and the width of the line.
4.2.4 Temperature Computation
Instead of using an empirical correlation as in the case of the plasma density, the temperature was
directly calculated according to the relation:
2
mc2

T = , (4.2)
20 2kB ln 2
where is the full-width at half maximum of the spectral line intensity distribution, 0 is the
center wavelength, m is the ion mass, c is the speed of light in vacuum, and kB is the Boltzmann
constant.
4.3. METHODS: DATA COLLECTION 61
For reasons to be discussed further in Section 4.4.2, we do not rely on Doppler broadening of
the H- line to compute the temperature, despite the fact that in general a spectral line experiences
both Stark and Doppler broadening that can, in theory, be deconvolved to give the plasma density
and temperature. Instead, the temperature was estimated by measuring the Doppler broadening of
the impurity lines that were visible on the captured images of the H- line. If we assume that the
impurity species are in local thermal equilibrium (LTE) with the ions of the bulk plasma (as they
would be if they are localized in a high density, highly collisional region of the plasma), then this
provides an estimate of the bulk plasma temperature. In Section 4.6.3, we assess the validity of that
assumption under the tested experimental conditions.
4.3 Methods: Data Collection

In this section, we review the details of the imaging spectrometer apparatus used in this study, and
in particular the steps taken to execute the calculations described above across a range of machine
operating conditions. The objective of the imaging spectroscopy experiment is to quantitatively
measure the plasma density in the hot, dense region that was inaccessible by the probe study. The
apparatus as described in this section is shown in Fig. 4.5.
4.3.1 Test Conditions

The plasma density was not sufficiently high in the region surrounding the visibly bright core to
measurably broaden the H- line beyond the background line width. In addition, the signal was not
strong enough (i.e., the optical emission intensity was insufficiently high) to capture a snapshot on a
timescale significantly shorter than the deflagration timescale, and thus we captured the integrated
emission over the entire deflagration event (10 s).
In order to understand how the measured quantities scale with the operating condition of the
accelerator, we imaged the emitted spectrum around the H- line (1.5 nm) at a range of charging
voltages. We operated the accelerator between 3 kV and 8 kV, so as to overlap with the probe study
conducted at 3 kV for comparison purposes. These voltages correspond to peak discharge currents
between 32.5 kA and 72.5 kA. Ten trials were performed at each operating condition, and the results
were averaged.1
4.3.2 Imaging Spectrometer Apparatus

This section describes the imaging spectrometer apparatus. As shown in Fig. 4.5, a single lens (10 cm
diameter, f /2) imaged the plasma emission onto the entrance slit of the spectrometer, matching the
f -number of the 0.75 meter SPEX 750M spectrometer using an aperture stop. A 1200 groove/mm
1 Note that there was a systematic current measurement error during the execution of these experiments, and a
subsequent campaign was conducted for the express purpose of correcting this error. For more details, see Appendix C.
Vacuum
Chamber
PIMAX
ICCD
Window
Aperture Stop
Anode Focusing Lens Entrance Slit

Cathode Spectrometer
Gas Injection
Figure 4.5: Schematic of the experimental setup.
grating, blazed at 5 100 , dispersed the light; the entrance slit was set to maximum height to avoid
cutting off any spatial information in the vertical direction. The dispersed light was then imaged
onto a Princeton Instruments PI-MAX intensified CCD camera, which could be gated down to a
minimum gate width of 2 ns. The camera was triggered simultaneously with breakdown of the gas
using the current monitor, and was gated to a 10 s exposure so as to capture emission exclusively
from the deflagration portion of the discharge.
An example of the image captured by the imaging spectrometer for each trial is shown in Fig. 4.6.
The H- line is centered in the frame, with two impurity lines adjacent at slightly longer wavelengths.
Based on a line survey of the major elements in the system (Cu, C, Fe, H) these lines were identified
as Fe I and Fe II at 656.92 nm and 657.4 nm, respectively[79]. The Fe I line at 656.92 nm was used
in lieu of the Fe II line at 657.14 nm, as a more consistent signal was obtained from the former across
the full range of conditions.
Fe I
H-
Fe II

Figure 4.6: Example image of chord-integrated intensity as a function of wavelength and vertical
height [I(, y)], collected by the PI-MAX camera, including the impurity lines (Fe I and Fe II) and
H- line.
4.3.3 Spectral & Spatial Calibration
In order to interpret the raw image data obtained by the imaging spectrometer, the image must
be calibrated such that the relative spectral and spatial distances between pixels are known. This
was performed by placing a mercury lamp in the vacuum chamber in the region to be imaged, and
selecting a doublet in the vicinity of the H-. The structure of the lamp was arranged such that
two light-emitting regions separated by a known distance were visible in the viewable region of
the imaging spectrometer, which enabled spatial calibration to be performed in situ. The process
followed for spectral and spatial calibration of the imaging spectrometer is outlined in detail in
Appendix C. The obtained calibrations were 0.125 A/pixel and 0.16 mm/pixel, respectively.
4.4 Methods: Data Analysis

In this section, we discuss the process of computing the radial plasma density distribution from the
raw spectrometer images. This includes symmetrization of the H- line shape, Abel inversion of the
intensity to obtain a radial emissivity profile, and fitting of each radial location with an appropriate
line shape profile.
4.4.1 Emissivity Inversion
As shown in Fig. 4.7, the image quantifies the emission intensity as a function of wavelength (in
the x-direction) and as a function of vertical position (in the y-direction) within the trans-axial
slice. At each y-position, the emission signal is path-integrated, comprising the cumulative intensity
contribution of each emitter lying along the chord projected onto that position before being dispersed
by the spectrometer. In order to determine the radial profiles of the H- line emission from the
chordwise integrated emission intensity, we utilize the inverse Abel transform, given in analytical
form by:
Z R
1 dI(y) dy
(r) = p , (4.3)
r dy y r2
2
where I(y) is the experimentally measured, chord-integrated intensity, R is the radius at which the
measured intensity reaches the background level, and (r) is the calculated radial emissivity. To
compute Eqn. 4.3, we use the Nestor-Olsen method, summarized by
N 1
2 X
k (r) = I(yn )Bk,n , (4.4)
y
nk
where the integers k and n are the position indices of the radial and vertical intensities, and y is
the distance between adjacent experimental data points. Calculation of the weights Bk,n is given in
Ref. [80], according to:
Bk,n = Ak,k for n = k

Bk,n = Ak,n1 Ak,n for n k + 1 (4.5)
where
[(n + 1)2 k 2 ]1/2 [n2 k 2 ]1/2
Ak,n = . (4.6)
2n + 1
The coded implementation of this algorithm, along with the code for ingesting the raw image data
and producing the processed and inverted image, is contained in Appendix C. The Abel transform
is highly sensitive to the symmetry of the measured integral data. However, as is evident from
the example data, the raw image of the H- line is not particularly symmetrical. Accordingly,
4.4. METHODS: DATA ANALYSIS 65
Figure 4.7: Rendering of a trans-axial slice, with the superposition of a differential area that
represents the projected area of a chordwise segment of the plasma onto the camera sensor. Figure
by T. Underwood.
we identified the centroid of the H- line, and considered only the half-plane of the image from
the calculated center as shown in Fig. 4.8. Each vertical slice of the half-plane image was inverted
according to Eqn. 4.4, and then fit with a Voigt profile as discussed in the next section. The inverted
emissivity profile (, r), computed from the half-plane image, is shown in Fig. 4.9.
Note that the intensity of the inverted image is low at the top of the image, which is located
at r = 0. This is a consequence of both the assumptions implicit in the Abel transform, and the
mechanism by which the plasma emission is generated. First, the Abel transform assumes that the
point of zero emission is at r = R; that is, the edge should represent the edge of the plasma
profile. However, the pixel values at the edge are finite (due to the noisy background signal atop
which the emission signal appears); this causes a nontrivial emission to be computed at the edge by
the inversion integral. If the signal does not increase proportionally in intensity towards the center
(r = 0) line, as in this case, then the data does not numerically account for the increasing path
length factored into the inversion integral with decreasing r.
Figure 4.8: Example halfplane image of the chord-integrated emission intensity as a function of
wavelength (x-axis) and vertical height (y-axis).
Second, this behavior is not unexpected when we consider the physical source of the emission
signal. In a fully ionized hydrogen plasma, the emitting species are not the ions (which are protons
and thus do not emit) but are the neutral hydrogen atoms that are either entrained in or otherwise
present at the structured plasma flow. However, if the plasma density increases radially towards
the center of the slice (as is qualitatively apparent), then the fraction of neutral emitters will fall off
radially as more of the neutrals are ionized. This results in the phenomenon of burnout, wherein
there are no longer a meaningful number of neutral emitters present after a certain radial shell is
surpassed. We see evidence of this in the inverted emissivity profile, and it will become relevant in
subsequent sections as well as we endeavor to determine the plasma density as a function of radius.
To convert the inverted emissivity profile into a radial profile of the plasma density, a line shape
must be fit to the intensity at each r position, which is discussed in the next section.
Figure 4.9: Abel-inverted emissivity profile vs. radius and wavelength [(r, )], based on the example
chord-integrated intensity map in Fig. 4.8.
4.4.2 Fitting the Emissivity
At each radial location, an inverted spectrum has now been computed that contains both Fe lines and
the H- line. To obtain the useful plasma parameters from these lineshapes, the spectrum must be
fit to a curve that approximates the shape of each line. There are two types of broadening a line can
experience due to a host of possible mechanisms: broadening that results in a Lorentzian lineshape,
which is the result of collisional processes (or pseudo-collisional processes such as Coulomb collisions)
that are homogeneous across velocity classes, and broadening that results in a Gaussian lineshape,
which is the result of collisional processes that are inhomogenous across velocity classes. Examples
of both lineshapes having the same FWHM are shown in Fig. 4.10, reproduced from Ref. [81].
In the SPG plasma, the two primary contributions to the broadened shape of the emission line
are the Stark broadening, which leads to a Lorentzian line profile, and Doppler broadening, which
leads to a Gaussian line profile. A spectral line subject to both types of broadening exhibits a Voigt
profile, which is simply the convolution of a Lorentzian and Gaussian profile, given by:
V (x; , ) = G(x; ) L(x; ) (4.7)

Atoms 2014, 2
Figure 6. Spectral line shapes of the Lyman- (a) and - (b) line in a t
Figure 4.10: Comparison of the Lorentzian lineshape and Gaussian lineshape having the same
created by interacting protons. Comparisons of the results obtained by
full-width at half-maximum (FWHM). Reproduced from Ref. [81].
codes: SimU (yellow), BinGo (black) and DM-simulation (purple).

18 -3
300 40
n=10 cm , T=1 eV
18 -3
n=10 cm , T=10 eV
19 -3
250 n=10 cm , T=1 eV
ytisnetni dezilamroN
ytisnetni dezilamroN
19 -3 30
n=10 cm , T=10 eV
(a) (b)
200
150 20
100
10
50
0 0
-2
-1.0 -0.5 0.0 0.5 -0.1
1.0x10
Energy (eV) En
Figure 4.11: Lineshapes for H- for various combinations of plasma density and temperature.
Figure 7. Spectral line shapes of the Lyman- (a) and - (b) line in a t
created by interacting protons at n = 1019 cm3 and T = 1 eV. Compa
obtained by the different codes.
300
14
ER-simulation
Figure 4.12: Triple Voigt fit to the inverted spectrum at two different radial locations of an example
data point.
where G(x; ) is the (centered) Gaussian profile given by
1 2 2
G(x; ) = ex /(2 ) (4.8)
2
and L(x; ) is the (centered Lorentzian profile given by

L(x; ) = . (4.9)
(x2 + 2 )
The shape factors and determine the degree of Gaussian and Lorentzian aspects, respectively,
that are evident in the line shape. Examples of Voigt profiles with different combinations of and
(and thus different combinations of ne and Te ) are shown in Fig. 4.11. The greater the Lorentzian
component relative to the Gaussian, the more pronounced the wings of the distribution become.
At each radial position (i.e., each horizontal slice of Fig. 4.9), a Voigt profile was fit to the under-
lying data. An example fit at two different radii is shown in Fig. 4.12. The Lorentzian component
of the Voigt fit is due to the Stark broadening, from which the plasma density was calculated using
the empirical correlation for the Stark broadening of hydrogen introduced in Section 4.2.2. The
Gaussian component of the Voigt fit converged inconsistently and was negligible at most r-values,
and so it was not used to determine the temperature. Instead, the Fe I line was used to determine
the Doppler broadening, for the opposite reason: the Lorentzian component of the Voigt fit to the
Fe I line was negligible in all tested cases. The code used to fit the triple Voigt profile and produce
the radial density profiles discussed in the next section is contained in Appendix C.
4.5 Results
The goal of implementing this diagnostic was to quantify the density distribution in the bright,
focused core of the plasma jet that is clearly visible in time-integrated images but inhospitable
to probe studies. In this section, we present the results obtained from the imaging spectrometer
diagnostic that were directed to that goal.
4.5.1 Radial Density Profiles
An example radial density profile calculated from the Abel-inverted emission spectra is shown in
Fig. 4.13. As discussed above, due to the effect of optical burnout within the jet core a complete
radial profile could not be readily constructed. In this region, a Voigt fit to the inverted emissivity
resulted in either no convergence to a suitable profile or exhibited an amplitude error greater than
a predetermined threshold.
However, this region is the portion of the plasma that exhibits the most extreme properties,
and thus we attempt to estimate a density profile based on a fit of two different curves to the
variance-weighted raw data.The variance of each plasma density data point was computed from the
least-squares fit of a Voigt lineshape to the spectra, the routine for which outputs a number of
parameters including the standard error of the fit (see App. C). If the error of the fit exceeded a
predetermined threshold percentage error over the fit value, the data point was discarded; the vast
majority of such data points fell in the burnout region, where the signal to noise ratio was extremely
poor. We fit both an assumed Gaussian profile and a modified exponential profile2 to the data in
the region outside the burnout zone, in order to estimate the profile within the burnout zone. It is
obvious that this approach increases the uncertainty in the exact value of the peak density; however,
we accept this uncertainty as unavoidable. The second important parameter, the line density, defined
as Z R
Ne = 2 ne r dr (4.10)
0
2 The reason the exponential profile is a modified exponential profile is that the portion of the profile that spans
the burnout zone is actually a line having a slope equal to the slope between the last two data points fit to the
exponential, to avoid the profile blowing up at r = 0.
4.5. RESULTS 71
Figure 4.13: Example result of the computed radial density profiles, including a comparison of
the Gaussian and modified exponential curve fitting methods used to estimate the density profile
in the burnout region. The green circles are data points computed from the numerical fit shown
in Fig. 4.3, and the black and blue curves are the Gaussian and modified exponential fits to the
profile, respectively. The red region represents the burnout region within which no line shape was
discernible.
and having the units of inverse meters, is not as significantly impacted by our choice of assumed den-
sity profile because it involves integrating over the entire radial profile (for most of which measured
data points exist).
Due to shot-to-shot variability of the SPG, an average of 3 trials was conducted at each operating
point. The resulting images were averaged together to produce a composite spectral-spatial image
corresponding to the operating point, which was then fit according to the procedure outlined above.
The average temperature was computed for each trial and compared to the predicted temperature
using the model described in Sec. 4.6.2; if the two differed by greater than an order of magnitude,
the data point was discarded when computing the scaling relations computed and shown in the
following sections.
Figure 4.14: Scaling of the peak number density with average discharge current during the first
half-period of the current oscillation, with proportionality of the form ne I .
4.5.2 Current Scaling

The empirical scaling law governing the peak density as a function of the peak discharge current (I)
is shown in Fig. 4.14. For a power law of the form I , we obtain a fit parameter of = 1.641 1.170.
From this fit, it is clear that the scatter in the peak density values is too large to draw any firm
conclusions. However, neglecting the high fit error, the scaling of the peak density with the current
is indicative of a positive relationship with magnetic pressure (which scales as I 2 assuming that
B I).
The scaling of the line density is shown in Fig. 4.15. In contrast to the peak density, the error
of the power law fit is within a reasonable range, and the computed exponent is = 0.984 0.159.
This nearly linear scaling with the average discharge current, coupled with the scaling of the peak
density, implies a radial compression driven by magnetic pressure. The following section discusses
the underlying mechanism of this compression, and summarizes the derivation and application of a
simple model to compare against the experimentally obtained results.
4.6. DISCUSSION 73
Figure 4.15: Scaling of the line densities with peak current, with proportionality of the form
Ne I .
4.6 Discussion
This section discusses the results of the scaling study, and the implications of the results on the
physical mechanism governing the apparent radial compression of the plasma in the core of the jet.
4.6.1 Radial Compression
The dependence of the peak density and the core radius on the drive current indicates that a
radial compression occurs along the centerline of the accelerator. The fact that such a compression
occurs generally in coaxial guns operated in the pre-fill mode is well known; the demonstration of
such a compression in a deflagration accelerator has previously only been presumed. The physical
mechanism of such a radial compression, or pinching, is shown schematically in Fig. 4.16.
As previously discussed in Chapters 1, 2, and 3, the primary axial acceleration mechanism is the
~ ~ force; in this case, it is this same force that generates the radial compression. As the plasma is
J B
accelerated downstream, the current passing through the plasma must still return to ground through
~ point in the negative z
the gun electrodes. As shown in Fig. 4.16, the current streamlines (i.e., J)
direction, and the self-generated B-field remains azimuthal. The Lorentz force now points radially
Figure 4.16: Schematic illustration of the mechanism of pinch formation along the centerline in
the near-field of the SPG exit plane. The dotted rectangle indicates the approximate region imaged
using the spectroscopic imaging diagnostic.
in the vicinity of the jet core, which causes the radial compression. This geometric configuration
that results in self-induced plasma compression is known as a Z-pinch [82, 83].
The simple geometry of a Z-pinch is well suited to an analytical treatment. The next section
describes a method for obtaining an analytical solution for the plasma parameters within the core,
assuming that it exists as a quasi-equilibrated pinch.
4.6.2 Constant Drift Velocity Model

In order to show a consistent relationship between the measured plasma temperature and plasma
density distribution, we employ an equilibrium Z-pinch model that captures the radial compression
of the plasma at the exit plane. The constant drift velocity (CDV) equilibrium Z-pinch model has
been shown to correspond well to the parameters of a high energy shear-stabilized Z-pinch [84]; we
now summarize the analytical development of this model.
The CDV assumption constrains the current density to be proportional to the number density
(i.e., j = Cne ); in other words, the drift velocity of the electrons through the pinch, which carry
the axial current, is constant irrespective of radius. Given this assumption, the azimuthal B-field is
determined from Amperes law: Z r
0
B (r) = j(r0 )r0 dr0 , (4.11)
r 0
where the proportionality constant C is determined from the number density profile and the mea-
sured discharge current (assumed to be the axial current in the focus), i.e.
I
C = R 2 R a . (4.12)
0 0
ne (r)rdrd
The radial momentum equation for the plasma column, eliminating extraneous terms, simply
balances the magnetic pressure (which applies a force radially inwards and is proportional to the
square of the current) against the kinetic pressure (which is related to the plasma temperature and
density through an equation of state). Thus, in the absence of any applied external fields, equilibrium
4.6. DISCUSSION 75
in a radially compressed plasma column is given by
dp B (r) d [rB (r)]

= , (4.13)
dr 0 r dr
which can be solved numerically to yield p(r). Assuming an ideal gas equation of state, the pressure
profile p(r) can be combined with the known (i.e., measured) density profile ne (r) to yield the
temperature profile, i.e.
p(r)
T (r) = . (4.14)
(1 + 1/Z)ne k
Here, Z is the ionization state of the bulk plasma ions (assumed to be Z = 1 for hydrogen).
Thus, for each data point at which a density profile was computed and a discharge current
was measured, the above model can be used to compute the magnetic field, kinetic pressure, and
temperature profiles. If the model is found to be valid for the SPG, this finding can be exceptionally
useful in cases wherein the density profile is easily measured but these secondary parameters are less
easily found. In the next section, we assess the applicability of the CVD pinch model.
4.6.3 Model Comparison
Fig. 4.17 shows four radial profiles, three of which (i.e., the kinetic pressure, azimuthal magnetic
field, and plasma temperature) are computed from the fourth (i.e., the plasma density) based on
Eqns. 4.11, 4.12, 4.13, and 4.14. Note that the kinetic pressure distribution is similar in shape to
the density distribution, and that the magnetic field strength peaks at a radial location distal the
centerline.
The measured average temperature is printed on the temperature plot of Fig. 4.17, and was
computed by spatially averaging the inverted emissivity profile, fitting a Voigt profile to the Fe I
line, and computing the temperature from the Gaussian component of the Voigt fit according to
Eqn. 4.14. The predicted average temperature was computed by averaging the temperature profile
over the radius. As shown in the figure, the measured and predicted temperature for the case shown
are within 3 eV. For all tested conditions in which a suitable density profile could be successfully fit
to the broadening data, the measured and predicted spatially averaged temperatures were within one
order of magnitude, which is adequate agreement to assert that the compressing core is reasonably
approximated by a quasi-equilibrated radial pinch.
Another method of comparison can be obtained from the so-called Bennett pinch relations [83],
and further assuming that the plasma is compressed adiabatically. The Bennett pinch relations
dictate, for an adiabatic compression, that the plasma density, discharge current, and line density
are related as: 2
I 1
ne 1 . (4.15)
Ne1
Figure 4.17: Example computation of the radial profiles of the azimuthal magnetic field strength,
kinetic pressure, and plasma temperature based on the measured plasma density profile combined
with Eqns. 4.11, 4.12, 4.13, and 4.14, with the measured temperatures based on the Doppler broad-
ened Fe I emission line. Note that the predicted and measured average temperatures are within 3
eV.
Given the scalings computed and shown in Figs. 4.14 and 4.15, the above equation allows us to
compute an effective of
' 1.4 0.3. (4.16)
For an ideal monatomic hydrogen plasma, we would expect:
ideal = 1.667. (4.17)
Given the error in the scaling fits and the pursuant error in the calculation of , this calculation
alone does not demonstrate unequivocal for the model. However, when combined with the compari-
son between the measured and predicted plasma temperatures, as well as a geometric configuration
of the SPG that is known to produce such pinches in other systems, the evidence strongly indicates
that a radially equilibrated Z-pinch forms within the core of the jet.
4.7. SUMMARY 77
Figure 4.18
4.7 Summary
In this chapter, we discussed measurements of the conditions at the exit plane of the accelerator, and
demonstrated that they are consistent with the hallmark radial magnetic compression of a Z-pinch.
We have developed empirical scaling models for both the peak number density and line density in
the focus region downstream of the SPG exit plane. The peak number density scales roughly with
the magnetic pressure, and the line density scales roughly linearly with the current. Based on these
observations of the plasma parameters at increasing drive currents, we infer that during a single
firing of the accelerator, the rising drive current pulse leads to radial plasma compression in the
focal region, consistent with a Z-pinch.
Using the CDV Z-pinch model, we predicted the plasma temperature, azimuthal magnetic field
strength, and kinetic pressure in the compression region according to the measured plasma density
profiles at various operating conditions. We also derived an estimate of the bulk plasma temperature
from the Doppler broadened impurity lines, and compared the predicted and measured plasma
temperatures and found them to be consistent. Using the well-known Bennett pinch relations, we
combined the scaling laws to compute an effective ratio of specific heats, , of the plasma in the pinch
and found that it was consistent with an ideal monatomic hydrogen plasma. Taken in combination,
both the scaling analysis and temperature modeling and measurement comparisons provide strong
evidence of the formation of a radially-equilibrated Z-pinch within the deflagration-produced plasma
jet.
This finding adds to the physical picture of the discharge developed in Chapter 3; accordingly,
we revisit the figure illustrating the stages of the plasma deflagration, but with the addition of
the Z-pinch suggested by the spectroscopic data, in Fig. 4.18. At this stage in our study of the
deflagration-produced jet, we have identified two distinct aspects of the flow. In Chapter 3, we
obtained time-resolved state variable data for the downstream plasma conditions well beyond the
radial compression zone. In this chapter, we traded temporal resolution for the ability to quantify
the radially-resolved plasma density profile in the hottest, most dense region of the plasma, and in
so doing verified the presence of a Z-pinch in the core of the jet.
However, just as the QLP was insufficient in several areas and the imaging spectrometer was
developed to rectify those deficiencies, the imaging spectrometer itself has several limitations. First,
the measurement of the plasma density is indirect, relying upon neutral emitters in the vicinity
of the plasma to generate the emission signal. Second, the diagnostic is time-averaged, and as
such provides only a general understanding of what occurs during the transient plasma acceleration
process. Third, while the necessary data could be obtained in a single firing of the gun, all the
emission signal is gathered from a single axial position that must be painstakingly aligned. In short,
understanding the axial variation in the plasma compression characteristics would be prohibitively
difficult using the imaging spectrometer.
The above limitations must be addressed to examine the dynamics of the pinch formation and
the interactions between the pinch and the deflagration-produced jet. It is this topic that is the
principal focus of the following chapter, in which a time-resolved optical diagnostic that directly
measures the plasma density is developed.
Chapter 5
Dynamics of the Discharge:

Emission and Schlieren
Cinematography
5.1 Introduction
The characterization studies thus far have focused on two aspects of the deflagration. First, the
probe study described in Chapter 3 collected time-resolved data at a single low-energy operating
condition, comparatively far from the exit plane. Second, the imaging spectrometer study described
in Chapter 4 collected time-integrated plasma density data at a single axial position immediately
downstream of the exit plane, at various operating conditions. The combined results of these studies
suggest that two predominant flow phenomena coexist in the initial deflagration phase of the dis-
charge: a high velocity, diffuse jet, which itself surrounds a magnetically compressed plasma focus
attached to the central electrode and extending approximately one gun diameter downstream.
Having studied first one piece, then the other, we have developed an understanding of what
the properties of each are in isolation. However, determining precisely what occurs at the interface
between the external shear flow and the radially compressing core is of great interest, particularly
the ways in which the dynamics of the pinching core may affect the behavior of the jet downstream.
The diagnostics previously employed, however, cannot shed light on these dynamical features in the
hot, dense environment of the plasma focus, and so we turned to other diagnostic methods. We
once again relied upon optical techniques to avoid disrupting the plasma behavior with an immersed
probe, but in this case we achieved time resolution by capturing sequences of images at an ultra
high frame rate. A small portion of the collected imagery is of the raw emission, which provides a
qualitative record of the dynamics of the deflagration process. The majority of the imagery, however,
79
80 CHAPTER 5. DYNAMICS OF THE DISCHARGE
was of a light beam that was passed through the plasma, instead of light emitted by the plasma. The
intrinsic properties of the plasma govern how light propagates through it, and thus we were able to
infer those properties from the characteristics of the imaged light beam. In effect, this comprises a
measurement of the refractive index of the plasma.
This chapter details the principles, practical considerations, and results of a study of the plasma
refractive index and dynamics using high-speed schlieren and emission imaging. While the emission
experiment was simply a matter of pointing the camera at the viewing area of the chamber and se-
lecting a suitable neutral density filter, the schlieren imaging required substantially greater effort to
implement and refine, and thus a majority of the exposition is devoted to this topic. First, we intro-
duce schlieren imaging in the context of typical neutral gas flows, and then the special considerations
involved in applying the technique to plasma diagnostics. Next, we discuss the apparatus devised
for this study in particular, and the methodology associated with data collection and processing.
Finally, we discuss a series of experiments and corresponding results obtained using the schlieren
imaging apparatus, and complete the picture of the plasma jet dynamics introduced in Chapters 3
and 4.
5.2 Plasma as a Refractive Medium
As discussed in Chapter 1, a major source of complexity with respect to plasmas as compared

to neutral gases is their inherently electromagnetic character. Though the interactions between a
plasma and electromagnetic fields, both externally applied and self-induced, often make the plasma
scientists job a more difficult one, they can also be leveraged to measure properties of the plasma.
As an electromagnetic wave passes through a plasma, its propagation is altered by these properties;
namely, the refractive index of the plasma (and gradients thereof) changes the relative phase and
propagation direction of the electromagnetic wave. Thus, by measuring either or both of these
effects, the refractive index of the plasma can be determined.
We require an expression for the refractive index of a plasma in terms of the properties of interest
(namely, the plasma density). As in a metal, the electrons of a plasma are free to roam within the
material, and the dielectric constant (i.e., the relative permittivity) is given by
p2
=1 , (5.1)
2
where is the angular frequency of the electromagnetic wave interacting with the plasma and p is
the plasma frequency, given in turn by
s
ne e2
p = , (5.2)
0 me
5.3. SCHLIEREN PHOTOGRAPHY 81
where ne is the plasma density, e is the fundamental charge of the electron, 0 is the vacuum
permittivity, and me is the mass of the electron.

The refractive index n = of the plasma is thus given by
ne e2
n2 = 1 . (5.3)
0 m e 2
This is a simplified picture of how electromagnetic waves behave as they interact with a plasma;
it assumes negligible magnetic field effects, and that the frequency of the wave is much greater
than the plasma frequency (i.e., p ). For a more detailed treatment of the plasma refractive
index in all manner of field configurations, as well as a comprehensive description of the zoo of
electromagnetic waves supported by plasmas in general, the reader is referred to the seminal work
by Stix, as well as those of Chen, Dendy, and others [8588]. Nevertheless, within the scope of
this work, the previous description is adequate. Most importantly, this framework provides a direct
link between the parameter of interest, the plasma density ne , and the measurable property, the
refractive index n.
Having recognized such a link, we turn now to the principles involved in the measurement of the
refractive index as a function of space and time using high-speed schlieren photography, in order to
resolve the evolving structure of the plasma jet.
5.3 Schlieren Photography

Schlieren photography is a venerable technique for the imaging of refractive index gradients in
transparent media. A refractive index gradient or local inhomogeneity is known as a Schliere,
from the German word meaning streak, and the plural (likewise from the German) is therefore
Schlieren. Schlieren imaging or photography is the class of techniques that convert these invisible
inhomogeneities into observable images, such as that shown in Fig. 5.1. The method was refined by
August Toepler in 1864 [89] for the study of supersonic flows, and has enjoyed widespread use in
the decades hence for imaging the structure of fluid flow around objects. Virtually all principles of
schlieren and shadowgraph imaging discussed in this work originate from the immensely descriptive,
detailed, and practically useful book by Settles [90]; enough material is excerpted here for the reader
to follow and understand the application of these principles to the current study.1
It is assumed that the reader has no familiarity whatsoever with a practical schlieren system
and how schlieren images are formed, and an overview is thus given here; those familiar with the
principles of schlieren imaging of typical gas flows may safely skip to Section 5.4 for a discussion of
concerns specific to schlieren imaging of plasmas.
1 For a much deeper discussion, as well as a thoroughly enjoyable reading experience, the reader is enthusiastically
referred to the original.

Figure 5.1: Schlieren photography of an airplane model with straight wings, at Mach 1.2. Photo-
graph by NASA.
5.3.1 A Basic Schlieren System

In its most basic form, a functional schlieren apparatus requires only: a light source, a collimating
optic, a focusing optic, a cutoff, and a screen (or light sensor) on which to project the schlieren image.
In fact, even the most complex configurations can typically be reduced to these essential elements.
An example linear arrangement is shown in Fig. 5.2, consisting of a light source, a collimating lens,
a focusing lens, a knife-edge cutoff, and a screen on which to project images of the schlieren object.
5.3. SCHLIEREN PHOTOGRAPHY 83
screen focusing lens collimating lens
schlieren object point light source

knife edge
Figure 5.2: A basic schlieren system with a point light sourcesource.
5.3.2 Principles of Schlieren
The schlieren concept is straightforward: the cutoff is placed axially at the focus of the beam,
and moved into the focal spot of the beam orthogonally to the optical axis. If the cutoff is properly
positioned, the image will darken uniformly and in proportion to the degree of cutoff; this is because,
in the geometric optic approximation, each bundle of rays at the focus comprises all of the image
information. Thus, if the cutoff is positioned at 50% cutoff (i.e., halfway into the focal spot) then half
the rays passing through the image are extinguished at the cutoff and fail to reach the screen. When
a schlieren object is placed along the optical path between the source and the cutoff, rays passing
through the object are refracted, altering their trajectory as shown in Fig. 5.2. The refraction angle
of the ray will depend upon the magnitude of the refractive index gradient, and rays which are
refracted away from the cutoff will appear as a region of greater relative brightness on the screen,
whereas rays which are refracted towards the cutoff and extinguished will appear as the opposite.
Thus, the orientation of the cutoff will impose a gradient in the schlieren image. The resulting
images are rich in detail and often quite beautiful, and can shed light on complex flow physics of
even simple systems [91]. Cutoffs are typically either a horizontally or vertically oriented knife-edge,
resulting in either a vertical or horizontal gradient in the image, respectively. Other, more exotic
types of cutoffs, such as grids, pinholes, and discs, are occasionally used for various purposes. In
this study, we utilized a slight variation on the knife-edge theme: a neutral-density (ND) cutoff, as
will be discussed below.
It is important to note a technique, related to schlieren imaging, that is similar in principle and
in general simpler to implement. That technique is known as shadowgraph imaging; when one sees
the shadows cast by rising thermal plumes on a hot and sunny day, one is observing shadowgraph
imaging in the wild. A shadowgraph system is fundamentally the same as that for schlieren, with
the exception that there is no cutoff. Local regions of darkness and/or brightness are due only to
bunching of the rays passing through refractive index gradients; thus, a focusing optic is not even
required. The shadowgraph method is less sensitive than its schlieren counterpart, and was not
particularly relied upon in this work. An example result from the schlieren diagnostic operated
without a cutoff, so as to produce an unadulterated shadowgraph signal, is shown in Section 5.8.2.
5.4 Challenges of Imaging Schlieren in Plasmas

The fact that a plasma possesses a refractive index that is proportional to the plasma density makes
schlieren imaging, and the related technique of shadowgraph imaging, a natural diagnostic choice
for qualitatively imaging the plasma structure. However, plasmas present a number of challenges
that can thwart the typical schlieren techniques described above. In conducting this study, we have
identified, and overcome, many of these unique challenges.
The first and foremost complication is that plasmas tend to be quite luminous, particularly
pulsed plasmas of appreciable energy densities such as those that are the subject of this work. In
fact, Z-pinches, dense plasma focii, and other devices similar to the SPG have been used as light
sources in lithography applications; in short, they are exceptionally bright [9296]. A self-luminous
schlieren object competes with the schlieren image, and will most assuredly obscure the schlieren
signal if great care is not taken to eliminate as much self-emission as possible [97]. The natural
solution to this problem is to obtain a backlight source with an optical power that dwarfs that of the
plasma, and furthermore to concentrate that optical power in a narrow optical range (in contrast
to the plasma emission, which is relatively broad in frequency). A seemingly simple answer readily
comes to mind: a laser. However, though a laser light source solves the issue of overcoming the
optical power emitted by the plasma, it introduces additional complications related to its coherence
that must be overcome in order to be implemented, as discussed in greater detail in Sections 5.5 and
5.6.
Second, pulsed plasmas are, by their nature, short-lived; this is not a problem for schlieren specif-
ically insomuch as for time-resolved imaging in general. The timescale of the discharge dynamics we
investigate here is sub-microsecond, while the duration is tens of microseconds; this leads to stringent
requirements on the time response and time duration of the light source, the camera, or both. Time
resolution can be obtained using a pulsed light source, but in that case, the evolution of the plasma
structure cannot be ascertained in a single firing of the gun. Time resolution can, conversely, be
obtained using fast-framing cameras; however, the number of frames that can be collected using an
intensified CCD camera typically ranges between 1-16 frames [98103], from which an understanding
of the full discharge dynamics is difficult to obtain.
5.5 Methods: Data Collection

In this section, we review the details of the schlieren apparatus used in this study, and in particular
the steps taken to overcome the challenges described above. The apparatus as described in this
section is shown in Fig. 5.3.
+
Earth ground
collimating mirror
cathode rod
anode rods biconcave
CW diode
gas inlet lens
laser
Pyrex
vacuum
chamber
capacitor bank
to dump
tank
line filter
...
ND cutoff
first second Shimadzu HPV-X2
focusing mirror baffle baffle CMOS camera
Figure 5.3: Schematic of the schlieren apparatus developed for this work.
5.5.1 Light Source and Chromatic Filtering

For backlight illumination, we employed a continuous-wave (CW) tunable diode laser at a nominal
wavelength of 636 nm. The diode was placed in a thermo-electrically cooled driver module, and
operated at its maximum continuous operating current. At the maximum current, the diode delivers
a nominal output optical power of 250 mW. An optical bandpass filter centered at 635 2 nm with
a FWHM of 10 nm was placed before the camera lens, and the laser diode wavelength was thermally
tuned (using the driver module) to maximize the background signal intensity on the camera sensor.
The absolute wavelength of the diode was not directly measured, but based on the temperature vs.
wavelength curve provided by the manufacturer, the nominal wavelength used during the imaging
experiments was determined to be 636 nm.
5.5.2 Beam Formation and Spatial Filtering

We used a conventional Z-type schlieren configuration consisting of two 6 diameter mirrors, each
with a focal length of 23.75 inches. The output beam of the diode was first collimated, then sub-
sequently passed through an adjustable focal length f -mount 50 mm lens, and then focused onto
an adjustable aperture located at the focal point of the first mirror. The adjustable aperture acts
as an approximate point source, which is collimated by the first mirror. The collimated beam was
passed through the test section to the focusing mirror. As the beam is focused, it is passed through
multiple successive apertures (or baffles) of decreasing diameter before reaching the cutoff, in order
to filter portions of the beam that were improperly focused or suffer from other spatial aberrations.
The beam then passes through the cutoff, which is described in the next section. The final optic
prior to the camera lens is the optical bandpass filter previously described.
5.5.3 Cutoff
To ameliorate the coherence issue caused by the use of the diode laser, a neutral density (ND) cutoff
is used instead of the traditional knife edge. The ND cutoff prevents the on-off behavior that results
from the diffraction effects, at the expense of some schlieren sensitivity, while retaining the gradient
character of the final image. We attempted to incorporate several commercial ND filters as the
cutoff, but obtained the highest quality results with a sooted glass slide. Such a cutoff is typically
fabricated by applying a piece of transparent tape to a clear glass slide, holding the slide above a lit
flame to allow a measure of soot to accrue, and then removing the tape. This traditional method of
creating the cutoff is shown in Fig. 5.4a.
The ease of producing these cutoffs somewhat mitigates the imprecision of this method; cutoffs
can be repeatedly manufactured and tested until a suitable one is found. However, in performing
this process many times, a particular source of defective cutoffs was identified: the adhesive of the
tape tends to melt when exposed to the flame for too long (e.g., when attempting to achieve a high
optical density) and at the cutoff itself, the adhesive residue (as shown in Fig. 5.4b) can exacerbate
the diffraction effects.
To combat this, we developed a new technique, wherein we firmly affixed a second glass slide to
the first and laterally offset it by approximately 50% of the width, as shown in Fig. 5.4c, to act as a
mask in lieu of the tape. This method is similarly straightforward and simple, but lacks the problem
of the melting adhesive. When properly applied, it results in a cleaner cutoff edge, as shown in
Fig. 5.4d.
The cutoff was first roughly aligned at the horizontal focus of the beam, and fine adjustments
(including to the degree of cutoff) were then made using a three-axis micrometer stage. The target
schlieren sensitivity was obtained by attempting to visually align the beam spot at approximately
50% cutoff, and making fine adjustments with the micrometer stage until the signal from a reference
schlieren object (e.g., the thermal plume from a human hand, the expelled jet of a compressed air
canister, etc.) was maximally visible.
Figure 5.4: a) The traditional cutoff manufacture method, in which tape is applied to the slide
before sooting with a flame, and then removed to create the cutoff edge. b) Optical microscope
image of the border between the sooted and unsooted region of a cutoff with poor performance (a
bad cutoff). The edge definition is uneven and inconsistent, and residue from the tape adhesive is
clearly visible. The scale bar represents a distance of 100 m. c) The improved cutoff manufacture
method, developed to prevent adhesive residue from accruing along the cutoff edge, in which a second
glass slide is used as the mask. d) Optical microscope image of the border between the sooted and
unsooted region of a cutoff with good performance (a good cutoff). The definition of the edge is
clear and consistent, and no adhesive residue is visible. The scale bar represents a distance of 100
m.
5.5.4 Imaging
In conducting this study, we were fortunate to have the temporary use of an exceptional camera that
met the stringent imaging requirements of experiment. From the fine gentlemen at Hadland Imaging,
we obtained the Shimadzu HPV-X2 high speed camera. The system provided a 10 MHz maximum
frame rate, along with the ability to record 256 consecutive frames. This equated to a snapshot
every 100 ns (with a 50 ns integration time) for 25.6 s, neatly matching the time span and time
resolution requirements of the deflagration and its dynamics. In fact, the time span was sufficient to
record both the initial deflagration and the subsequent detonation, and the distinguishing features
of each.
5.6. METHODS: IMAGE PROCESSING 89
Figure 5.5: Photograph of the Shimadzu HPV-X2 camera aligned in the schlieren system. The
baffling and spatial filters (black foamcore and business cards with custom apertures cut into them)
are also visible.
5.6 Methods: Image Processing

Digital image processing techniques can be essential to transforming raw schlieren image data,
particularly when collected under challenging experimental conditions, into the classic, recognizable,
and often beautiful portraits of fluid motion with which most readers are familiar. In this section,
we systematically discuss the steps taken to process the captured images in order to produce the
final results, from which a clear picture of the plasma dynamics is then gleaned. An unprocessed
(raw) frame from an example sequence is shown in Fig. 5.6a, which is transformed into the processed
schlieren image of Fig. 5.6 d and finally the quantified deflection angle mapping of Fig. 5.6e.
Figure 5.6: a) Raw image frame. b) Frame after background subtraction. c) Frame after diffraction
artifact (streak) removal. d) Frame after wavelet denoising. e) Frame after intensity remapping to
be proportional to deflection angle.
5.6.1 Background Subtraction

The choice of a laser source and the efforts to conserve as much optical power as possible in the
probe beam led to a highly non-uniform, fixed-pattern background. In addition, the CMOS sensor
background intensity is not completely consistent between sequences, so a tare image cannot be
easily constructed by capturing an image sequence of the background and subtracting it from an
image sequence captured even minutes or seconds later during a firing of the gun. However, using
the rolling buffer feature of the camera system, we can pre-trigger the camera such that a selectable
number of frames are recorded before the time point of the trigger. Thus, we can ensure that 5-10
frames at the start of the sequence are taken prior to the appearance of a schlieren signal, and are
therefore suitable tare images.
In some sequences, we found that the first image was excessively over- or under-exposed. To
avoid incorporating these skewed images into the background subtraction step, we averaged the first
five frames after the first frame (i.e., frames two through six) to construct the background image,
then subtracted the background image from the entire sequence. An example frame after background
subtraction is shown in Fig. 5.6b, and even with no further processing it is clear that the background
subtraction step greatly enhances the visibility of the signal.
5.6.2 Diffraction Artifact Removal

Though background subtraction is sufficient to obtain a decent qualitative picture of the schlieren
signal, features that are not fixed pattern are not removed by way of background subtraction,
because they appear in the image only when there is a schlieren signal, and continue to vary in
intensity throughout the sequence. A prime example of these types of features are the vertical
streaks that appear in front of the signal, such as those in Fig. 5.6b. These result from the diffraction
patterns produced by light scattering off of the soot particles at the cutoff. Other related diffraction
patterns appear in some image sequences, such as concentric rings that change in size and intensity
as the image sequence progresses (see, for example, the indicated portion of Fig. 5.7). In short, in
order to avoid viewing the image sequences as if through a dirty window, these artifacts must be
removed.
A key feature of this method relies on the fact that we have 256 frames in each sequence, and
thus have a time series of the intensity of each pixel; the image data can thus be imagined as a three
dimensional data cube (in x, y, and t). Artifacts in the x y plane (i.e., the conventional image
view) are complicated morphologically. Though they typically comprise vertical streaks, they are
not completely vertical, and since they differ from frame to frame morphological filtering would need
to be performed on each frame individually without a prior estimate of the artifact characteristics.
However, by their nature the artifacts are persistent in time, at least somewhat, on a pixel by pixel
basis. These manifest as intensity-modulated features that are exactly vertical in the x t plane,
and are thus significantly more amenable to morphological filtering.
ring artifacts
Figure 5.7: Example of a ring-like artifact that persists in the image sequence after background
subtraction. The streaks are also clearly visible.
The process for removing these artifacts is shown step by step in Fig. 5.8. First, a slice in the
x t plane is considered. The temporally-persistent features are clearly visible as vertical streaks.
Second, the discrete Fourier transform in two dimensions is computed; the result is shown in the
figure. As is always the case upon transformation between the spatial and wavenumber (Fourier)
domains in two dimensions, features that are vertically aligned appear on the horizontal axis of the
transformed image (and vice versa). These can be selectively filtered, as shown in step three of
Fig. 5.8. The filtering mask is blurred with a Gaussian kernel in order to reduce ringing artifacts in
the reconstituted image, post-filtration. The image, with the vertical streaks substantially removed,
is regenerated from the filtered Fourier domain representation using an inverse discrete Fourier
transform. Finally, the filtered x t plane image is properly replaced in the three dimensional data
cube. This process is performed on each x t slice, and the entire image sequence is thus filtered.
The results of this streak filtering algorithm are shown in the difference between Fig. 5.6b and c.
The filtering process can be halted after this step while maintaining the overall high quality of
the resulting image sequences. However, a final filtering step was employed to smooth and reduce
the appearance of sensor noise in the final images, as discussed below.
3 2
Figure 5.8: Depiction of the process for removal of temporally persistent diffraction artifacts in an
image sequence. 1) A slice is made in the xt plane at a position in y. 2) A 2-D Fourier transform is
applied to the slice. 3) A 2-D filter is applied to the transformed slice. 4) The 2-D inverse transform
is applied to the filtered transformed image. 5) The slice at the original y position is now filtered.
5.6.3 Wavelet Denoising

As a final filtering step, we employed a wavelet thresholding denoising method. A wavelet decompo-
sition of a signal (in any number of dimensions, though here we speak exclusively of two dimensional
signals) is similar in concept to a Fourier decomposition, in that it is simply a different representa-
tion of the signal in terms of a different basis. However, while a Fourier decomposition is made up
of sine and cosine functions, a wavelet decomposition can be made up of a host of different mutu-
ally orthogonal basis functions, at various spatial scales. The full scope of wavelet decomposition
is outside that of this work, and we recommend one of several excellent references for the theory
and applications of generalized wavelet transforms for a comprehensive review [104, 105]. However,
the essentially important feature is that large wavelet coefficients are dominated by signal informa-
tion, whereas small wavelet coefficients carry a preponderance of noise information. Furthermore,
the wavelet coefficients retain localized signal information, as opposed to a transform such as the
Fourier transform in which each component in the Fourier domain represents image components
over the entire spatial domain.
Thus, a wavelet thresholding approach to denoising simply includes determining a magnitude
threshold, below which the wavelet coefficients are set equal to zero. A related step, referred to
as soft thresholding, applies a linear scaling to wavelet coefficients above the threshold to soften
the transition across the threshold value. We determined the threshold empirically, and used a
biorthogonal wavelet for the decomposition, which resulted in the denoising apparent in the difference
between Fig. 5.6c and d. Comparing the two images, it is evident that the signal features and edges
are retained while the flatness of the background is greatly improved (a sign of successful denoising).
5.6.4 Intensity Normalization

The next step, the result of which is shown in Fig. 5.6e, is neither strictly essential nor filtering per
se. However, transforming the schlieren signal into a radiometrically-symmetrized image allows the
signal to be examined without the characteristic imposed schlieren gradient, which can be useful.
In addition, if the image is transformed such that the pixel intensity is equal (or proportional) to
the absolute deflection angle of the ray passing through the schlieren object at that point, then the
refractive index (or shape of the refractive index contour) can be determined from tomographic meth-
ods. The calibration necessary to perform this transformation is discussed further in Section 5.7.2,
and the results in Section 5.9.5.
5.7 Methods: Calibration

Schlieren is typically a qualitative diagnostic; it is in the intuitively-understood depictions of fluid
flows and the structures that develop within them that schlieren truly shines. However, even in cases
5.7. METHODS: CALIBRATION 95
where the refractive index itself is not directly measured, various aspects of the discharge dynamics
can be determined if we know the absolute and relative spatial dimensions of the flow. Furthermore,
there are methods of obtaining a direct measurement of the refractive index from a schlieren image,
by calibrating the schlieren system using an object that generates a known schlieren signal. In this
section, we describe the calibration steps taken to extract quantified spatiotemporal and plasma
density information from the schlieren data.
5.7.1 Spatial Calibration

The spatial calibration method was straightforward. We attached a transparent sheet to one of the
windows of the test section, such that the beam passes through it before reaching the camera. On
the transparent sheet is printed a regular grid of dots, as shown in Fig. 5.9.
In cases where the optical pathway of the beam is not orthogonal (or at least nearly so) to
the plane of the imaging surface (e.g., the camera or a screen), such an array can be used to
algorithmically remap a skewed image to a flat plane. For example, if the camera is viewing a
scene at an oblique angle, a regularly spaced grid can remap the projection such that the image
appears as if captured from an orthogonal direction. However, in our case, the camera is already
aligned sufficiently normal to the optical axis of the beam, and the grid can be used simply to
determine the number of pixels per millimeter. The computed spacing for the calibration shown in
Fig. 5.9 is 4.212 pixels/mm. Any time the position or zoom level of the camera was adjusted, a new
spatial calibration image was taken so that it could be properly associated with any image sequences
subsequently captured.
Spatial calibration permits the extraction of semi-quantitative data from the schlieren images,
such as the size of observed plasma structures, the rate at which they expand and/or advect, and
similar parameters, all in real spatial units that can be incorporated into dimensional analyses
and scaling laws. However, in order to reduce the schlieren images in such a way as to obtain
true quantification (i.e., of the quantitative relationship between the schlieren signal itself and the
refractive index of the plasma), further calibration of the apparatus is required.
9 mm
Figure 5.9: Image of the spatially uniform grid used to obtain relative and absolute spatial calibra-
tion of the schlieren images. The center-to-center spacing of the dots on the grid is 3 mm.
5.7.2 Radiometric Calibration

While qualitative assessment of the refraction angle induced upon a ray by the plasma is apparent
merely by inspecting the image, a true quantitative understanding can be obtained as well. The
method employed here is among the class of standard photometric methods, in which a standard
refractive object (i.e., a standard Schliere containing a known refraction benchmark or refraction
profile) is placed in the field of view and used to quantify the functional relationship between
refraction angle and pixel intensity. While many standard schlieren exist and can be used, we
chose a weak convex lens (f = 10 m) to produce a known distribution of refraction angles in the
calibration image.
As shown in Fig. 5.10 a, collimated rays which pass through the weak lens (the standard Schliere)
generate a distribution of intensities that are proportional to the refraction angle experienced by the
ray. For a ray hitting an arbitrary radial position of the lens, this angle is given by:
r
= tan (5.4)
f
5.7. METHODS: CALIBRATION 97
a) slice b)
extent of lens
Intensity [A.U.]
pixels
background
level
pixels pixels
c) = intensity data d) maximum
= full exponential fit = pseudo exponential fit
= pseudo exponential fit
Intensity [A.U.]
Intensity [A.U.]
background
minimum
pixels pixels
Figure 5.10: a) An image of the calibration object in the image plane, showing a slice through the
diameter of the calibration object. b) The intensity along the slice through the calibration object as
a function of position. c) A zoomed-in portion of the intensity variation, focusing on the portion of
the curve that corresponds to the calibration object. The data, an exponential fit to the data, and a
pseudo exponential fit to the maximum, minimum, and background intensity values are shown. d)
A pseudo exponential fit to an average image from a non-calibrated image sequence.
assuming that the deflection angle is small (which, in almost all instances of a schlieren-type deflec-
tion, it is). Since the deflection of a ray is known, the corresponding pixel intensity at the camera
sensor can be mapped directly to the deflection corresponding to its position in the calibration im-
age, to obtain a curve of deflection angle vs. pixel intensity. This curve is then used to tranform,
pixel by pixel, a schlieren image collected during the experiment into a calibrated image in which
the intensity of each pixel is proportional to the deflection angle experienced by the ray terminating
on that pixel.
Fig. 5.10b shows the curve that results from taking a slice through the image of the standard
Schliere. There is a background pixel level that corresponds to zero deflection, and in the region
containing the weak lens itself (i.e., the standard schliere) there is a distribution of intensities,
ranging from somewhat darker than the background to significantly brighter. The reason for this
asymmetry about the background level is due to the fact that the cutoff is not an absolute cutoff,
but rather a region of higher relative optical density. Therefore, the maximum extinction level for
a ray that is deflected into the cutoff is not as stark as it is in the case of an opaque cutoff, such as
a knife edge.
We isolate the region of the curve corresponding to the standard Schliere, and perform two fits to
the data as shown in Fig. 5.10c. The first fit is a non-linear least-squares fit of an exponential function
to the data (full exponential fit) and the second is a non-linear least-squares fit of an exponential
function to the set of three points corresponding to the minimum, maximum, and background levels
of the curve. As is apparent from the figure, a fit based on the full curve and one based on the three
points previously mentioned are nearly identical. This is a useful finding, because in cases where we
lack a calibration image corresponding to a particular image sequence, we can estimate a calibration
curve by constructing a pseudo-exponential fit of the minimum, maximum, and background values
from the sequence. Such an estimated curve is shown in Fig. 5.10 d.
Unfortunately, a combination of factors led to this calibration method falling short of ideal
when put into practice. The ideal situation is that the exact deflection angle can be determined
for each pixel of each frame The best way to utilize the standard schlieren method is to place the
standard schliere in an unused region of the test image, and do in situ calibration for each captured
image. However, this requires that the image undergo minimal post processing, and in particular
that the background be generally quite uniform for this to be effective[106]. In our study, this
was plainly not the case, as described above in Section 5.6. The calibration images were captured
in separate sequences, prior to the sequences that were eventually calibrated; this can introduce
additional uncertainty and error. Furthermore, not every calibration image was properly exposed,
and due to the limited amount of time available to use the camera, it was not determined until
after the camera was returned that some of the data was unusable. However, the above issues
notwithstanding, the remapping of the schlieren signal to be merely proportional to the deflection
angle was straightforward, and could be done using the estimated calibration curve method as
described. This permitted the determination of the shape of the refractive index contours, and thus
the structure of the plasma density, even if the absolute density value was not attainable. Given that
the previously performed quantitative measurements of the plasma density detailed in Chapter 4
can be combined with the shape function, a wholistic view of the evolution of the plasma density
contours as a function of space and time can be attained.
5.8 Validation
Before describing the physically meaningful results, it is worthwhile to discuss the means by which
we demonstrated that the diagnostic was performing as expected; in other words, the steps that were
taken to validate the diagnostic as such. This includes verifying that the signal we measured was in
5.8. VALIDATION 99
fact a schlieren signal (i.e., proportional to the refractivity gradient and not due to self-emission or
other sources), so that we could be certain that we were in fact sensitive to the plasma density.
5.8.1 Distinguishing the Schlieren Effect

There are many examples of plasma schlieren in the literature that purport to show classical schlieren
images of plasma flows [107]. However, almost without exception, the nature of the experiment is
such that it is unclear whether the image presented is indeed of plasma schlieren (i.e., plasma density
inhomogeneities) because either no significant schlieren gradient is visible [7, 24, 108110], or it is an
atmospheric plasma that generates thermal gradients that contribute the majority of the schlieren
signal [111, 112].
As described in Section 5.2, a plasma has a refractive index that is proportional to the free
electron density (i.e., the plasma density ne ), and it is this that we seek to measure with schlieren
imaging. However, a purported schlieren image of the plasma density can be adulterated by several
factors, including an (opposing) contribution to the refractive index due to neutral gas, as well as
self-emission that can wash out the underlying schlieren signal. In this work, we endeavored to show
that we are, in fact, observing a true schlieren effect.
The refractive index of the plasma is, in fact, slightly different than the expression given in
Eqn. 5.3, and should include the contribution from the neutral gas:
ne
n = 1 + k (5.5)
2nc
where k is the proportionality coefficient between the density and the refractive index, empirically
determined for various gases as a function of temperature, and known as the Gladstone-Dale coeffi-
cient, and where nc is the critical density, given by
2 me 0
nc = . (5.6)
e2
However, the positive contribution to the refractive index typically constitutes no more than one
part in ten thousand to the refractive index, whereas the negative contribution from the free electron
density is on the order of several parts in one thousand or greater. Thus, we need not consider the
contribution from the neutral gas, in particular because the plasma is highly ionized in the region
we are imaging (i.e., in the pinching region). In many cases in the literature, particularly those in
which atmospheric plasma jets are imaged using schlieren techniques, it is in fact the refractive index
gradient induced in the surrounding neutral gas that constitute the schlieren. This is typically the
result of thermally-generated gradients that are created by the plasma jet, and not electron density
gradients of the plasmas themselves.
The best evidence of a properly configured schlieren imaging system is that the characteristic
schlieren gradient is present in the image; if the signal is all (or mostly) emission from a luminous
Gradient
Figure 5.11: Average of 100 frames, depicting the vertical schlieren gradient indicative of a properly
configured schlieren apparatus.
source, or if it is a shadowgraph signal, the gradient will not be readily apparent. So-called plasma
schlieren is often one or both of these types of signal, instead of true schlieren. In contrast, we
captured a clear schlieren gradient in our images. In Fig. 5.11, the vertical gradient (due to the
horizontal ND cutoff) is clearly visible; this figure is an average over one hundred frames, which is
performed only to emphasize the schlieren gradient (though it is present, clearly, throughout the
sequence).
We can also verify the schlieren effect on a non-plasma schlieren object, taking advantage of the
same apparatus. To quickly and easily test this, we fired a compressed air jet in the test section;
the schlieren image is shown in Fig. 5.12. Notice the clear schlieren gradient in the image, wherein
the top half of the image trends towards dark and the bottom half trends towards light. The fine
structure of the turbulent jet is also visible.
5.8. VALIDATION 101
Gradient Direction
Figure 5.12: A single frame from a sequence captured of a compressed air canister being fired into
air.
5.8.2 Shadowgraph Comparison
Shadowgraph effects, which are related to schlieren effects as previously described, can appear even
in a schlieren image. The characteristics of a shadowgraph image are similar to those of a schlieren
image, with the key difference being that any inhomogeneity in the refractive index will create a
symmetric shadow at the image plane. In other words, since there is no cutoff there will be no
imposed schlieren gradient perpendicular to the orientation of the cutoff. The shadowgraph effect
is proportional to the first derivative of the refractive index gradient (i.e., the second derivative of
the refractive index as a function of space) instead of to the refractive index gradient itself; thus, in
general shadowgraph is less sensitive than schlieren [90]. Shadowgraph can be superior to schlieren
in resolving certain types of flow features that are especially sharp, such as shock waves, but a
properly configured schlieren system will resolve these as well.
In order to assess the magnitude of the shadowgraph effect in our system, we captured a sequence
with the cutoff removed such that any deflectometry signal would be the result of shadowgraph,
instead of schlieren. The resulting image is shown in Fig. 5.13, which is an average of the first 100
frames of the sequence (so as to include only the quasi-stationary pinch) and shows the characteristic
symmetric signal of shadowgraph. Note, also, that the signal is relatively weak, in particular when
compared to the schlieren signal of Fig. 5.11, and resolves primarily the edges of the strongest
symmetric signal
Figure 5.13: Shadowgraph signal obtained by removing the schlieren cutoff, with the symmetric
signal clearly visible.
parts of the pinch as opposed to the continuum of densities and full axial extent of the pinch visible
in the schlieren image.
5.8. VALIDATION 103
5.8.3 Evidence of a Dual Flow Structure

Over the course of this study, we were able to add significant supporting evidence to the dual-flow
picture of the discharge, established and suggested in previous Chapters 3 and 4. In doing so, we
also serendipitously showed that the self-emission mitigation techniques we devised and implemented
were successful at isolating the schlieren effects from the plasma luminosity. Fig. 5.14 shows evidence
of both of these aspects.
Figs. 5.14 a-e are similar to the dual flow Figs. 3.12 and 4.18 of the previous chapters, but we
have combined the outer, diffuse flow measured via quadruple probe in Ch. 3 with the radially
compressed core identified in Ch. 4. Using the cinematographic schlieren diagnostic, we are able
to observe the dynamics of the entire process in the near-field region of the accelerator after the
exit plane. We can capture the formation of the pinch (Fig. 5.14c), the peak-current, maximum
flow velocity driving-phase of the flowing pinch (Fig. 5.14d), and the decreasing-current, quiescent
phase of the pinch as the discharge diminishes (Fig. 5.14e). Fig. 5.14f depicts an illustration of the
predicted velocity profile that would be generated by the presumptive dual-flow configuration, as
the axial current and self-generated azimuthal magnetic field of the pinch would combine to retard
the axial flow in the z-direction within the pinch. This is only an assumed profile, however, as we
did not directly measure the bulk flow velocity itself in a spatially-resolved manner.
While testing the schlieren system, we inadvertantly removed one of the two baffles being used
to block stray light before it reached the cutoff, resulting in the image shown in Fig. 5.14g. This
image is taken at the same time point as the image in Fig. 5.14h, the only difference being the lack
of one baffle and thus the increase in self-emission from the bulk plasma. The surrounding, diffuse
flow is of too low a density to create a significant schlieren signal, being below the limit of schlieren
sensitivity for our system, but we are convinced of its presence by the washed-out schlieren image
of Fig. 5.14g. We can also plainly observe how essential the blocking of stray light is to the ability
to observe the structure of the high density, magnetically-compressed plasma.
a)
gas puff
f)
b)
g)
c)
first baffle
d) only
first & second h)

baffle
e)
Figure 5.14: Schematic depiction of the discharge process (a-e), as well as the assumed velocity
profile through the jet and jet core (f), an image obtained using a single baffle showing the background
emission (g), and an image obtained at the same time point using two baffles showing significantly
reduced background emission.
5.9. RESULTS 105
5.9 Results
The goal of implementing this diagnostic was to directly observe the dynamical processes at work
in the focusing, pinching core of the deflagration jet. In this section, we present the results obtained
from the schlieren diagnostic that were directed to that goal. In the limited time that the camera
was available, additional studies were performed that involved placing various objects into the flow
and observing the resulting effects; these studies are discussed in more detail in Chapter 6. After
presenting the results of this set of experiments, we discuss and analyze the results and the new
physics they reveal about the jet dynamics in Section 5.10.
At a general, high level, our picture of the discharge progression is fairly complete, on the
timescale of the overall deflagration event. However, this system is inherently transient, and through
the use of high speed cinematography (of both the self-emission and the schlieren signal) we can
begin to understand the dynamics of the complex system on a timescale much smaller than that
of the complete deflagration. In the following sections, we present a selection of the results and
analysis obtained from the emission and schlieren studies. In many cases, references will be made
to a sequence of images; a selection of image sequences is contained in the Supplementary Material
to this work. A static sequence of images presented in such a format as this written work can lack
a certain power of explanation; the interested reader is encouraged to seek out the supplementary
image sequence associated with the image data presented here, and to compile a video sequence
using the analysis and processing software for which the source code is given in Appendix D.
5.9.1 What We Can Learn
Through time-resolved imaging, we are able to observe and measure several features of the flow.
In a qualitative sense, we can distinguish between the character of the first and second breakdown
events (i.e., the deflagration and detonation waves) and verify that they are, in fact, different in
kind rather than merely in degree. Schlieren imaging is particularly well suited to providing an
intuitively-grasped, qualitative understanding of flow fields and their structure and evolution, and
we take advantage of this aspect as well.
Spatially calibrated schlieren images allow us to also measure aspects of the pinch structure itself:
its lifetime, assembly rate, diameter (as a function of axial distance), density, and fluctuations of
the spatial distribution of the pinch in general. Since the pinch is formed via compression by the
self-generated azimuthal B-field, information about the structure of the pinch as a function of time
and space also provides indirect data regarding the magnetic field strength and topology, particularly
when coupled with knowledge of the total discharge current as a function of time.
duration (~8 s)
rundown (~2 s)
Figure 5.15: Discharge current vs. time, with the duration of the rundown and pinch phases
indicated. The width of the blue rectangles at the regions between phases is indicative of the
uncertainty in the times at which these phases begin and end.
5.9.2 Long Timescale Dynamics

Through both emission and schlieren imaging, certain temporal features of the discharge on the
timescale of the entire deflagration event (i.e., > 1 s) become apparent. First, and most obviously,
there is a gap of 2 s between breakdown (i.e., when the trigger is received from the current
monitor) and when any signal is observed in the viewable area. The uncertainty in this delay is
largely due to the fact that some variable-amplitude noise riding on the signal from the current
monitor is often associated with breakdown, and thus the signal can reach the trigger level of the
scope earlier on some shots. This delay results from the acceleration of the plasma along the electrode
length, and corresponds well with the measured velocity of the leading edge (downstream of the exit
plane) of the plasma jet of ' 100 km/s.
The pinch begins to coalesce at approximately 2-3 s after breakdown, and persists until approx-
imately 10 s after breakdown (in the schlieren image sequence). This corresponds to the duration
indicated in Fig. 5.15, showing the current trace for the 9 kV charging voltage operating condition.
5.9.3 Emission Cinematography

Before considering the schlieren data, it is instructive to see the advantages and limitations of pure
emission imaging, which was also conducted with the 10 MHz frame-rate camera. The signal received
by the camera is comprised only of light emitted by the plasma, and this intensity I is related to the
5.9. RESULTS 107
a)
b)
c)
Figure 5.16: A sequence of images, captured at the same relative time during the discharge process
using different neutral density filters: a) ND2, b) ND3, and c) ND4.
plasma density approximately as I n2e assuming that the plasma is optically thin and, of course,
that it emits from regions of high plasma density.
However, since the goal is to study the plasma structure (and therefore plasma density) di-
rectly and in detail, emission has some serious disadvantages. Plasma emission can come from
several sources, the most important (in the optical regime) being atomic spectra and thermal
Bremsstrahlung. In the former case, the desire to observe the structure of the plasma at its most
dense is in inherent opposition to collecting an emission signal, for the simple reason that a fully
ionized hydrogen plasma consists of protons and electrons which do not emit an atomic spectra.
In the latter case, there can be significant broad-spectrum background Bremsstrahlung radiation
(resulting from free/free charged particle interactions) at high enough temperatures, where the total
intensity from an optically thin hydrogen plasma is given by:
Z
I n2e T 1/2 d` (5.7)
along the collection line-of-sight `. However, it is impractical to distinguish between these two types
of emission, and only one (Bremsstrahlung) is indicative of electron density and thus the structure
of the plasma itself.
Second, the exact relationship between the emission signal and the plasma density presents an
issue with quantification. An exact relationship between the plasma density and the emission signal
intensity cannot be easily determined without complex modeling, and would require the use of
certain other assumptions (such as the optical thickness of the plasma). Clearly, any quantification
of the plasma density based solely on emission would have high uncertainty, directly related to the
accuracy of the underlying radiative model.
Third, the quadratic proportionality between the presumed plasma density and the intensity of
the emission signal limits the useful dynamic range of the imaging sensor. That is, if the density
varies by a factor of two, the intensity will vary by a factor of four. Since the CMOS sensor used in
this work has a linear response, it is impossible to capture the full range of plasma densities present
throughout the discharge (spatially and temporally) in emission. This is evident from the images in
Fig. 5.16, where the signal in the vicinity of the cathode, where the plasma density and temperature
are high, is saturated relative to the downstream signal.
We captured three different sequences of the gun firing at the same operating condition, with
neutral density filters of successively increasing blocking ratio (ND2, ND3, and ND4 filters), as
shown in Fig. 5.16. This sequence of filters represents three orders of magnitude of neutral density,
illustrating the difficulty in capturing the plasma density structure in a single firing using emission.
Quasi-Periodicity
One advantage of the emission data is that it requires no digital post-processing, and thus there
is no noise injected into the image data due to image filtering steps. The light also only passes
5.9. RESULTS 109
a)
b)
c) d)
Figure 5.17: a) image from the sequence of emission images, wherein the vertical red line represents
the column of pixels shown for each time point in b), wherein the vertical red line represents the
column of pixels for which the intensity values are shown in c), wherein the red dots represent the
peaks between which the instantaneous frequency is shown for each successive peak in d).
through a single fused silica window and a neutral density filter before reaching the camera, and
thus diffraction effects due to the schlieren cutoff are not present. This provides a clear picture
across all three dimensions of the data (two spatial dimensions and time) that we can examine to
observe trends in the emission intensity. Though the intensity is related in an undetermined way to
the plasma density, it is instructive to examine nonetheless.
For example, one type of slicing of the data that is well suited to the emission sequences is shown
in Fig. 5.17. Fig. 5.17a shows a single frame in the sequence, and the vertical red line located 6
cm downstream is a set of y-pixels which, when swept in time over the 100 frames containing the
deflagration, result in the swept profile in Fig. 5.17b. The horizontal line in Fig. 5.17b is a set of
t-pixels, which can be seen plotted against time in Fig. 5.17c. This is simply tracking the intensity
of a single pixel in time, at a point in the image that captures coherent fluctuations in the intensity.
These fluctuations are visible as vertical striations in Fig. 5.17 and peaks in Fig. 5.17c. Considering
each sequential pair of intensity peaks, we can compute the frequency at which the peaks appear
as a function of time; in other words, we estimate the changing frequency based on the changing
effective period between peaks. The result of this calculation is shown in Fig. 5.17d, and we can see
that the frequency appears to increase and decrease along with the discharge current (which drives
and thus aligns with the intensity signal in time).
We return to this observed quasi-periodicity in Section 5.10.3, but it is clear that to understand
the plasma structure in the vicinity of the cathode (the saturated region of Fig. 5.17a), we cannot
rely on emission. Instead, we turn to schlieren imaging for this purpose.
5.9.4 Schlieren Cinematography
The primary advantage of schlieren imaging over that of the raw optical emission is that the signal
is directly proportional to the electron density. In addition, we are not limited by the dynamic range
of the sensor, so we do not wash out the signal in the dense (and thus strongly emitting) parts of
the discharge as we do in the case of emission. We can therefore observe the structure of the plasma
in the vicinity of the electrode, where the current densities, magnetic fields, and plasma densities
are highest.
As the operating voltage and current of the gun are increased, the schlieren signal improves
(i.e., the signal over background increases) due to higher current and plasma densities. Therefore,
we consider images from the sequence corresponding to an operating voltage of 9 kV in Fig. 5.18.
Fig. 5.18a shows a representative sequence of frames between 3.4-4.1 s, in 100 ns intervals. This
sequence depicts the emergence, growth, and advection of coherent structures just prior to and at
the peak discharge current. Fig. 5.18b shows the manner in which the axial position z and effective
size are calculated, from the image data, for each structure in each image in the sequence.
Amplitude Growth Rate
By tracking the growth in amplitude of these structures over 5-7 frames, corresponding to 400-600
ns, we can measure the effective growth rate of the structure (e.g., d/dt). The results of this process
are shown, for the five structures that were discernible in the full sequence, in Fig. 5.18c. These
structures were formed sequentially, and thus they did not all overlap directly in time; the markers
in the figure are aligned in time, however, in order to indicate similarities between the time evolution
of the structure size. As shown in the figure, the growth rate is linear and consistent among all five
structures for the first 200 ns, before diverging somewhat over the latter 300 ns. We note that
the variation in growth rate at later times does not appear to be strongly correlated to the order
of the structures; that is, structures formed later in time do not have a consistently higher or lower
growth rate than those formed earlier in time.
5.9. RESULTS 111
a)
b)
c)
non-linear
growth
linear growth
d)
Figure 5.18: a) A selection from a sequence of frames. b) An annotated snapshot depicting the axial
position and maximum radial extent of the coherent structures along the pinch. c) A plot of radial
extent versus time for a series of structures emitted from the cathode. d) A plot of axial position
versus time for the same series of structures.
Effective Advection Velocity
We also track the effective velocity of the structures as they are advected downstream in the plasma
flow. Fig. 5.18b shows how the axial displacement z of each structure is determined from the
schlieren images. Aligning the displacement vs. time points in a similar manner as described above,
we observe that the effective velocity is consistent between each of the observed structures and does
not change meaningfully over the 400 ns that each structure remains in the field of view. The
effective velocity of these structures is '80 km/s, which is 75% of the velocity of the plasma front as
measured by tracking the front position of the emission. The velocity of the bulk flow may change
slightly between the initial front and the time at which these structures are observed (2-4 s), but
the data suggests that the structures are perturbations to the background flow that are advected
uniformly slower than the bulk flow velocity. The implication of this observation is that the Taylor
hypothesis, which states that the velocity of convected structures in the flow are equal to the bulk
flow velocity, is likely invalid for the plasma jet.
Quasi-Periodicity Revisited
We also perform a similar analysis as that described in Section 5.9.3, in which we track the evolution
of a selected vertical column of pixels over the 100 frames of the deflagration. We consider two axial
locations, 1 and 25 pinch diameters downstream from the center electrode, respectively, as shown
in Fig. 5.19a. Figs. 5.19b and c show the resulting slices in the y-t plane for these axial positions.
At z/d = 1, the schlieren signal is starkly differentiable from the background and the diameter of
the pinch can thus be easily determined; the resultant time-varying diameter of the pinch is shown
in Fig. 5.19d. At z/d = 25, the boundary of the pinch is less easily discernible, but the intensity of
the schlieren signal is clearly modulated. Averaging the upper half of the slice so as to avoid dark
schlieren signal canceling out bright schlieren signal, we obtain the intensity vs. time curve shown
in Fig. 5.19e. Some amount of quasi-periodicity is observed in each of Figs. 5.19d and e; that is, in
the pinch diameter and the intensity of the schlieren signal.
At first glance, this effective frequency of multiple pinch parameters, at least as calculated from
the 9 kV shot analyzed here, does not appear to agree with that calculated from the emission data
taken at nearly the same operating condition. The observed frequency is lower than that of of the
8.5 kV shot. There are several plausible explanations for this observed phenomenon. First, the
periodicity may be an artifact of the processing and analysis that must be applied to the schlieren
data and to which the emission data is not subjected (this is particularly true of the periodicity in the
diameter, which is difficult to systematically calculate from the underlying data). Second, the lower
frequency modulation may be the result of a different phenomenon than that which is observed
in emission, with a different characteristic frequency; there may be many competing dynamical
phenomena that contribute to modulation of both the emission and schlieren signals at various
time-varying frequencies, some of which are more easily observed in emission and others in schlieren.
5.9. RESULTS 113
a)
b) c)
d) e)
Figure 5.19: a) Illustration of the sequence of frames, showing the y t slices. b) The first y t
slice, located at z/d = 1. c) The second y t slice, located at z/d = 25, showing a region of interest.
d) A plot of diameter, as computed from the schlieren image, versus time. e) A plot of average
intensity versus time in the region of interest.
Finally, because the maximum frame-rate of the camera is 10 MHz, a modulation at a frequency
of 1.3 MHz could be the result of aliasing. In other words, if structures are shed and advected
downstream at 11.3 MHz, higher than the effective sampling rate, it would be apparent as a low
frequency aliased component. We also note that the phase velocity of the modulation is not apparent
from the time series, and could in fact be negative, suggesting that high frequency perturbations
of the diameter and/or schlieren signal that move downstream could manifest as low frequency
perturbations that appear to propagate backwards in the upstream direction. This would be a case
of the classic wagon-wheel effect, a well-known temporal aliasing problem.
5.9.5 Tomographic Reconstruction
Since the schlieren signal is the result of rays being deflected through a specific plasma distribu-
tion (and therefore refractive index distribution), certain tomographic reconstruction techniques are
available to us if we permit some assumptions to be made about the distribution. Namely, for an
axisymmetric refractive index profile, integrated deflection data from a single perspective (such as
that collected here) is sufficient to recover said refractive index profile via the inverse Abel transform.
The calculation is in the same class of inversion as that performed in Chapter 4 with respect to the
Stark broadening data, but the integrated variable is the deflection angle instead of the emissivity.
An axisymmetric refractive index distribution will result in a deflection pattern similar to that
shown in Fig. 5.20, where the total deflection angle is the result of the integrated refractive index
gradient across the plasma column in the transaxial direction (the propagation direction of the ray).
For such a distribution, the Abel transform between the refractive index at a given radial location
and the deflection angle at a given vertical position is given by[113, 114]:
Figure 5.20: Illustration of the deflection pattern in a collimated beam that results from an axisym-
metric refractive index.
5.9. RESULTS 115
pixels
pixels
Figure 5.21: A time-averaged frame of the radiometrically calibrated image sequence captured at
the 9 kV operating condition.
Z
1 (y)
n(r) = 1 p dy, (5.8)
y2 r2
r
where (y) is the deflection angle of the beam at the radial location r. Since we do not actually
measure a continuous distribution of deflection angles but rather a discrete set (i.e., corresponding
to a set of N pixel values), the discrete transform is given by:
N Z 1
1 X j + (j+1 j )`
ni = n(ri ) = 1 p d`, (5.9)
j=i (j + `)2 i2
0
where ri is the radial distance from the centerline (i.e., ir, where r is the pixel dimension) and
j is the deflection angle at rj . The integral in Eqn. 5.9 can be performed analytically to yield:
N
X
ni = 1 Dij j , (5.10)
j=0
pixels
pixels
Figure 5.22: A symmetrized image generated from the average image in Fig. 5.21.
where the coefficients Dij are given by
Dij = 0 j < i,
1
= Ai,j j = i,

1
= (Ai,j Bi,j1 ) j > i, (5.11)

with p
p p j + 1 + (j + 1)2 i2
Ai,j = (j + 1)2 i2 j 2 i2 (j + 1) ln p , (5.12)
j + j 2 i2
and p
p p j+1+ (j + 1)2 i2
Bi,j = (j + 1)2 i2 j 2 i2 j ln p . (5.13)
j + j 2 i2
Thus, given an image where each pixel value corresponds to the deflection angle (in radians), we
can compute the refractive index as a function of radius and axial position (assuming axisymmetry).
Even if the pixel value is only proportional to the deflection angle, we can compute the shape of the
axisymmetric refractive index distribution.
5.9. RESULTS 117
pixels
pixels
Figure 5.23: The Abel-inverted refractive index distribution computed from the symmetrized image
in Fig. 5.22. The heat-map units are arbitrary.
Figure 5.24: Selected profiles (of an arbitrary quantity proportional to density) at various axial
positions through the Abel-inverted image.
0 frame 1e25
1.35
1.20
50
1.05
0.90
100
0.75
150 0.60
0.45
200 0.30
0.15
0 50 100 150 200
Figure 5.25: An Abel-inverted, time-averaged image with the density quantified.
First, we consider an average of the first 100 frames of the deflagration, remapped from schlieren
data to deflectometric data as described in Section 5.7.2. This average is shown in Fig. 5.21. Note
that there is a relatively clear dark region along the centerline; this is consistent with the deflection
pattern one would expect for rays passing through the central chord of an axisymmetric distribution
(i.e., no deflection for rays passing through the center), as shown in Fig. 5.20.
The image is, of course, not perfectly symmetric due to various aspects of the schlieren data
and calibration process. As we found in Chapter 4, the success of the Abel transform relies on
symmetry (or, in other words, is only valid for a truly symmetric, axisymmetric distribution). We
must therefore symmetrize the data before applying the inversion. One approach is to average the
upper and lower halves of the image together, to compose an average profile. However, this approach
unnecessarily compounds the complication of having averaged two regions (i.e., the bright schlieren
signal and the dark schlieren signals) with different, nonlinear mappings between signal intensity
and deflection angle. Instead, we simply locate the axial centerline (as the dark central region
previously identified) and symmetrize the image about that line by considering only the upper half
image; the result is shown in Fig. 5.22.
We consider each vertical slice as independent from each other vertical slice, and compute the
inverse Abel transform for each slice consecutively. The inverted profile of the time-averaged,
radiometrically-calibrated image in Fig. 5.22 is shown in Fig. 5.23. The spatial dimensions are
listed in pixels; only a 125 250 pixel area is computed, and then reflected to create the full image.
Note that artifacts of the inversion process appear along the center line and in the vicinity of r = 0,
5.10. DISCUSSION 119
due to the singularity at that point, as a high intensity (i.e., high density) region; these are an
unavoidable consequence of the computation and should be ignored.
A different way of viewing the inverted data, as density profiles, is shown in Fig. 5.24. Slices
through the time-averaged pinch at normalized axial slices corresponding to z/d = 0, 2, 4, 6, and 8
show that the pinch is approximately Gaussian several diameters downstream of the cathode, but
appears to have a hollowed region in the center in the vicinity of the cathode and 2-3 diameters
downstream. There are two physically-motivated explanations for this observation. The first is that
at high frequencies, current flowing through a conductor (or, as in this case, through a conductive
plasma column) localizes at the outer surface of the conductor instead of uniformly through the cross
section. This is known as the skin-effect and is commonly observed in wire pinches [115119]. The
second is that from a fluid-dynamics perspective, a small stagnation zone should form at the tip of
the central electrode as the plasma is accelerated along the centerline, and approach the free stream
conditions as the flow axially develops.
If we consider the intensity remapping in a fully quantitative way, wherein the pixel value is equal
to the integrated deflection angle instead of merely proportional to it, we can compute the absolute
plasma density via the steps outlined above. This method of effectively guessing that the calibration
curve associated with one shot is identical to that associated with the data from a second shot is, in
short, highly suspect. At the very least, a large degree of uncertainty should rightly be associated
with any such computation. Nevertheless, the result of this calculation is shown in Fig. 5.25. If we
neglect the density values associated with the centerline, which are artifacts of the inversion process
as described above, the densities remain approximately one order of magnitude higher than that
which is predicted via the scaling determined in Chapter 4. The reasonable values range from
3 1024 to 1 1025 m3 , which is higher than densities reported in similar experiments [120122].
For this reason, we put greater faith in the calculations that assume only proportionality between
the remapped intensity and the deflection angle, and the resultant density shape functions, than the
absolute quantification results.
5.10 Discussion
The current understanding of continuous-flow Z-pinches that form in a gas-puff plasma focus is
that a stable plasma column is formed, attached to the cathode, and lasts significantly longer than
stagnant Z-pinches and dense plasma focuses typically persist before instabilities cause them to
break up. What this study has revealed is that this picture is incomplete, and fails to account for
the inherently dynamical nature of even this stabilized pinch.
The above results demonstrate an underlying periodicity of the discharge that is at a much higher
frequency than the drive current oscillation (i.e., 100 kHz); a quasi-periodic (i.e., periodic with
a changing period) emergence of regular, coherent plasma structures (often referred to as plasma
blobs or plasmoids) is visible in both the self-emission and schlieren image sequences. Using these
time-resolved imaging techniques, we were able to identify the frequency of these modulations of the
plasma density, as well as other dynamical features of the structures themselves. With additional
physical analysis, we can form hypotheses regarding the origins of these plasma structures and add
still further to the dynamical picture of the deflagration jet.
5.10.1 Z pinch stability

As discussed in Chapter 4, the Z pinch that forms in a plasma focus closely resembles the geometry of
a truly cylindrical Z pinch, particularly in the region immediately downstream of the cathode. The
rate at which the pinch diameter increases with axial distance, due to diverging current streamlines
(and thus decreasing current density), is small compared to the characteristic diameter of the pinch.
In other words,
D
1, (5.14)
D
and therefore we will consider the plasma column to be roughly cylindrical for the purposes of
dynamical analysis.
Here, we give a brief discussion of the relevant portions of linear and ideal MHD stability theory
as applied to cylindrical Z pinches. An excellent review/overview of this material is contained in a
monograph concerning the physics of Z pinches, from their initial development to the modern era,
written by Liberman et al. [123]; we reproduce the relevant analysis here.
The radial stability is governed by the same equation as that used in Chapter 4 (Eqn. 4.13), that
is:
dP B d
+ (rB ) = 0 (5.15)
dr 4r dr
where both P and B are functions of radius, i.e. P = P (r) and B(r) = (0, B (r), 0). The lack of
cylindrical symmetry (which we are not considering) leads to an additional dependence of B with
z, but for a given axial location Eqn. 5.15 holds given the assumption of Eqn. 5.14.
This can be reformulated as an eigenvalue problem, which we can Fourier analyze by assuming
perturbations to the equilibrium vary with the form:
(r, t) = (r) exp(t + im + ikz). (5.16)
The eigenvalues of this solution (2 ) are real; if the eigenfrequencies are negative (2 = 2 ) the
eigenmode is a wave oscillation, whereas positive values (2 > 0) correspond to unstable eigenmodes
(i.e., they grow exponentially). The real part of is the growth rate, m is the azimuthal wavenumber,
and k is the axial wavenumber.
We focus here on the m = 0 sausage mode, best illustrated by the diagram in Fig. 5.26. The
mechanism for the growth of such an instability mode is as follows: a perturbation in the pinch
SMALLER
B
LARGER
B
B
J
Figure 5.26: Illustration of the sausage (m = 0) instability.
radius causes the local current density to increase, and the azimuthal field (generated by the axial
current density j) to increase as well. This increasing field further compresses the neck formed at
the region of locally higher field, due to the magnetic pressure increase, causing a runaway process
in which the perturbation grows exponentially.
The growth rate of the sausage instability can be determined analytically, for certain geometric
cases and distributions of current density in the plasma column. For example, when the current
distribution is strongly influenced by the skin effect and therefore localized in a thin layer at the
boundary of a uniform column, Eqn. 5.15 reduces to the Bessel equation and provides the following
transcendental dispersion relation:
2 Im () m2 Km (kR)
=1+ , (5.17)
Im () kRKm (kr)
where = (k 2 R2 + 22 /)1/2 , Im (x) and Km (x) are modified Bessel functions.[124, 125]. However,
even this analytical expression is adapted to a Z pinch with a free boundary, which is inconsistent with
the picture we have developed thus far. In the case of the deflagration jet, it is more appropriately
described as a diffuse Z pinch that lacks a clear free plasma boundary.
A diffuse Z pinch with either a uniform or Gaussian density profile exhibits only m = 0 and
m = 1 instability modes, of which the m = 0 mode has the fastest growth rate [123]. As found by
Refs. [126, 127], the growth rate saturates, for k , in a Gaussian pure Z pinch at the rate
1 VA
= = (5.18)
A R
which is the inverse of the Alfven time over the characteristic pinch radius R. Our time-resolved
imaging data allows us to assess whether the observed pinch dynamics are related to the quasi-
periodic excitation of this sausage instability, as detailed in the following subsection.
5.10.2 Calculating the Alfven Time
Possessed of the requisite diagnostic data, we can compute (with some uncertainty) the Alfven
time at the narrowest point of the observed pinch, and thus the expected growth rate of an m = 0
instability mode in our pinch. We can then compare this to the observed periodicity of the generated
structures, and determine whether they are likely the result of the repetitively-saturated sausage
instability.
We measure the total discharge current, at the capacitor bank; some fraction cpinch of this total
current must pass through the pinch plasma. That is,
Ipinch = cpinch ID . (5.19)
Given the total current flowing through the pinch, we can model the current density distribution
in the pinch and thus determine the azimuthal B-field. As in Chapter 4, we assume a constant drift
velocity model in the pinch such that
j = Cn (5.20)
where
Ipinch
C= 2
, (5.21)
R RR
n(r)r drd
0 0
with R being the radius of the plasma column. Amperes law, as before, gives us the azimuthal
B-field distribution from the combination of Eqns. 5.20 and 5.21:
Zr
0
B (r) = j() d. (5.22)
r
0
Taking the average azimuthal field in the pinch, Bavg , as
ZR
1
Bavg = B (r) dr, (5.23)
R
0
and an estimation of the average density, navg , as
ZR
1
navg = n(r) dr, (5.24)
R
0
we can calculate the characteristic Alfven speed:
Bavg
VA = , (5.25)
0 navg mi
where mi is the ion mass (having neglected the mass contribution of the electrons). In these studies,
the ion mass corresponds to the proton mass because hydrogen was used as the working gas. The
growth rate and Alfven time can be computed as a function of these variables using Eqn. 5.18. Each
of these variables is a function of time during the discharge, and thus the maximum growth rate
(and growth time) of the sausage instability will also vary during the discharge.
5.10.3 Applying the Stability Criterion to the Observed Pinch
Let us consider the operating point with the maximum current, which corresponds to the 9 kV
charging voltage shot with the results shown in Figs. 5.18 and 5.19. The damped sinusoidal total
discharge current, with the first half period corresponding to the deflagration event, has a maximum
current of 90 kA and a half period of 10 s.
A full account of the time-varying stability criterion would be forced to consider the fact that the
density distribution, as well as the discharge current and coupling constant between the discharge
current and the pinch current, changes as a function of time. However, to first order, the dominant
feature that governs the Alfven time across the pinch is the magnetic field and, by extension,
the current flowing through it. The average pinch radius changes slightly, but for the purpose of
this analysis, we will consider the average equilibrium radius at the narrowest part of the pinch
throughout the discharge, which is '0.35 cm as determined above in Section 5.9.4 and the scaling
in Ch. 4, and consider the sausage instability as arising from perturbations to that radius.
Likewise, the plasma density changes over the course of the discharge (and is related to the
instantaneous current being driven through the pinch), but here we will consider the background
equilibrium density to be of the order 1023 m3 , consistent with the scaling established in Ch. 4.
Figure 5.27: Growth rates of the m = 0 mode during the deflagration.
Figure 5.28: Growth rates of m = 0 mode, closely spaced between 20-30% current coupling.
Figure 5.29: Growth rates of the m = 0 mode assuming a higher pinch density.
This results in the following expression for the growth rate:
[Hz] = 1.97 107 s A kg1 Bavg [T]

(5.26)
It should be noted, at this point, how sensitive the analysis is to each of these parameters,
and that the parameter about which we have the least information is the pinch current coupling
parameter, Cpinch . We can plot the growth rate, as a function of time during the discharge half-
period, for various values of Cpinch . This family of curves is shown in Fig. 5.27, and curves closely
spaced between Cpinch = 0.2 and 0.3 are shown in Fig. 5.28. The relationship between the average
field and the pinch current (assuming constant background values) is given by
Bavg = 2.776 105 Ipinch 1.0 103 , (5.27)
where the current and field are in SI units. This linear relationship was computed from the equilib-
rium calculation of Ch. 4.
To illustrate the sensitivity of the growth rate to other parameters about which we have more
certainty, note the distribution of curves in Fig. 5.29. In this case, the density has been increased
by a factor of five, and thus the growth rate for a fixed coupling parameter changes by a factor of
five as well. However, as discussed in Section 5.9.4, the frequency at which the plasma column is
modulated does not appear to change rapidly during the deflagration, and in the case of the 9 kV
shot is within a few MHz of 10 MHz throughout. This pattern fits more closely with the measured
density of ' 1 1023 m3 and a coupling factor of '0.25 than with a higher density and widely-
varying growth rate corresponding to a higher coupling factor. Given the diffuse radial current that
is conducted to the cathode along the length of the coaxial geometry during the entirety of the
deflagration, a fractional coupling of 20-30% of the total discharge current to the pinch is a more
reasonable conclusion from the available data.
Note that the times in Figs. 5.27, 5.28, and 5.29 are relative to the record length, and to obtain
the time relative to the breakdown time 3.2 s should be subtracted from the time on the x-axis.
5.11 Summary
This chapter discussed the high speed schlieren and emission imaging apparatus and its application
to studying time-resolved discharge dynamics in the deflagration accelerator. This tool offers an
unprecedented ability to understand the complex time-varying behavior and evolution of the plasma
structure in the near-field of the deflagration-produced jet. With respect to the schlieren, we devel-
oped several key methods that advance laser plasma schlieren beyond the current state of the art, in
particular: a new method of fabricating high performance neutral density cutoffs, and a new method
of processing the schlieren images to remove temporally-persistent artifacts. We also endeavored to
calibrate the schlieren apparatus so that semi-quantitative and fully quantitative descriptions of the
absolute plasma density and plasma structure could be extracted from the schlieren data. This
valuable density field data was analyzed in multiple ways in order to provide physical explanations
for the observed dynamic behavior of the deflagration.
First, we considered the emission data. We took advantage of the low image noise to quanti-
tatively analyze the quasi-periodic modulation of the axially accelerated plasma; we observed that
the deflagration appears as a guttering flame. Second, we considered the schlieren data. The
ability to see the plasma structure in the most dense, hot part of the pinching core revealed the
emergence and evolution of coherent, magnetically-confined structures within the jet. We calculated
their growth rates and advection velocities, and compared the observed periodicity of the schlieren
signal with that of the emission signal. Finally, we employed a tomographic technique to obtain
2-1/2 dimensional maps of the plasma density of the pinch.
Briefly, we developed a model of the pinch stability to the radial m = 0 mode, or sausage
instability, and applied it to the experimental data. We determined that a time-varying, current-
driven sausage mode repetitively saturating and re-exciting at the cathode attachment point is a
plausible explanation for the observed self-modulation of the pinch diameter. However, further data
collection at a higher time resolution would be necessary to determine whether the self-modulation
occurs throughout the discharge at all operating conditions, and if this instability is in fact the
dominant driver of the discharge dynamics.
Chapter 6
Applications
6.1 Introduction
One of the principal goals of understanding how the plasma deflagration accelerator works is to put
the high energy density directed plasma jet to practical use. In this chapter, we present several
self-contained studies in which the SPG facility was used to investigate a particular application.
First, we examine the suitability of the deflagration-produced jet for generating fusion-relevant
conditions at a plasma-material interface [31, 38]. Second, we use a variant of the SPG facility de-
signed for low-power operation to study the practical implementation of the deflagration accelerator
as a space propulsion mechanism [32]. Finally, we evaluate the capability of the SPG facility to serve
as an experimental platform for laboratory astrophysics [43].
6.2 Plasma-Material Interactions

As international experimental fusion reactor facilities move closer towards operational status, atten-
tion is turning to the next generation of reactor designs and materials for practical power plants.
One of the key engineering challenges in this development is the identification and understanding of
the processes occurring at the interface between the plasma exhaust and the reactor first wall, as it
is these phenomena that drive the suitability and lifetime of plasma facing components (PFCs) in
the reactor, as well as whether the reactor is able to achieve the necessary confinement conditions.
It is known that the bulk of the transient loading experienced by PFCs is the result of edge
localized modes (ELMs) and disruption events. As it is difficult to isolate these transient loads in
an operating experimental reactor, other techniques have been devised for studying the interaction
between plasmas at ELM-like conditions and materials often found in PFCs. These include the use
of plasma guns, laser irradiation, and charged particle beams to produce high heat and/or particle
fluxes directed onto a target substrate [69, 128131]. The latter two techniques can provide heat flux
127
128 CHAPTER 6. APPLICATIONS
parameters on the order of those expected in ELMs, but may not capture the complex effects that
result from a quasi neutral plasma acting as the source of the energy flux. The former technique is the
platform afforded by the SPG Facility. By coupling detailed knowledge of the plasma environment
both in the jet and the vicinity of the target material, the potential exists for a significantly deeper
understanding of the plasma material interactions (PMI) and plasma edge physics that plays a
critical role in fusion engineering.
In this work, we report on the suitability of the SPG experiment for use as an analog plasma
source designed to replicate ELM-like conditions and behavior. Using the QLP data outlined in
Chapter 3, we obtained the energy flux and heat flux parameter of the jet as a benchmark to
compare against known ELM behavior [66, 132]. Off-normal heat fluxes are considered (for current
and next-step machines) to be those that are > 20 MW m2 , with streaming plasma densities of
< 1022 m3 (to avoid shocks and vapor shielding at the plasma-material interface)[66]; our goal is
to achieve parameters within this range.
We then performed a set of target studies using two different witness plate materials. The
first set of target data was collected from a series of individual shots at the same charging energy,
using polycrystalline silicon wafers. The results of these shots were analyzed via scanning electron
microscopy (SEM). We then performed a two-parameter target study using copper witness plates,
in which we vary the peak and integrated fluxes independently in order to determine which drives
the degree and character of the resulting material damage. The copper witness plate samples are
analyzed using a high powered optical microscope with <1 m resolution. From the combination
of these two studies, we show that the observed damage appears to be strongly tied in type and
severity to the peak fluxes, and that the morphology of the damage is closely linked to the initial
surface characteristics and material of the target sample.
6.2.1 Plume Characterization

In the context of this study, the primary goal of the plume characterization is to obtain a mea-
surement of the bulk heat and particle fluxes experienced by the target surfaces. These measured
parameters enable the observed effects on material surfaces to be closely coupled to the plasma
conditions, instead of relying on estimates based upon the input energy to the accelerator. To this
end, a QLP was used to obtain the plume parameters necessary to calculate a total energy flux and
heat flux parameter to use as a benchmark for comparison against other plasma gun target studies
in the literature, well summarized in Ref. [66].
6.2.2 Energy and Heat Flux Parameters

Using only the density and temperature contours of Chapter 3, and estimating the bulk plasma
velocity via the slope of the leading edge of the plasma density contour, we calculate the relevant
bulk energy flux and heat flux parameter for the jet. This data was collected only for the 3 kV
6.2. PLASMA-MATERIAL INTERACTIONS 129
charging voltage, corresponding to 252 J/pulse, due to the fact that at higher energies it was not
technically feasible to collect probe data this close to the exit plane of the accelerator as a result of
arcing. Scaling of the plasma parameters with machine operating conditions is detailed in Chapter 4,
where we found that the key state variables (i.e., density and temperature) scale approximately
with the square of the peak current (and, thus, with the magnetic pressure) while the velocity
scales approximately linearly with charging voltage (and, thus, with peak current). This allows for
estimation of the expected conditions at a material target for a range of machine operating points.
The energy flux (E ) is assumed to consist entirely of internal energy (i.e., the kinetic temperature)
of the flowing plasma and the kinetic energy carried by the plasma jet. This is calculated as
1
E = cv Te ne V + mp V 2 ne V (6.1)
2
where cv is the per-particle heat capacity of hydrogen at constant volume (assumed to be 23 kB ), Te

is the plasma temperature (where we have assumed that Te = Ti as per the QLP theory), mp is
the proton mass, ne is the plasma density and V is the bulk jet velocity. Using the nominal values
from the probe data of Te ' 10 eV, ne ' 1021 m3 , and V ' 40 km/s (obtained via the slope of the
leading edge of the spatiotemporal contour), we obtain an energy flux of
E ' 150 MW m2
For a nominal pulse with of 10 s (obtained from the width of the plasma pulse in the contours),
the heat flux parameter (H ) is thus
H = 0.58 MJ m2 s1/2 .
This is somewhat lower than the heat flux parameter typically used to replicate ELMs in systems
developed by others[129]. However, this apparent shortcoming is mitigated for two reasons: first,
this heat flux parameter is based on measured data in the plasma plume, so it is not immediately
apparent that this is in fact a significantly lower number than that actually realized by systems based
on optical heating, given the lack of actual heat flux data in those cases. Second, we expect the
heat flux parameter to scale approximately linearly with input energy, and this data was collected
at a comparatively low input energy (250 J/pulse). The heat flux parameter is therefore likely to
be in the range of 4.1 MJ m2 s1/2 at the highest pulse energy tested in this study. Furthermore,
the overall energy flux is higher than that generally associated with ELMs, and thus multiple firings
should be able to achieve a higher effective heat flux parameter (by increasing the effective period)
while still maintaining the minimum energy flux to replicate ELM-like damage in material targets.
We also note that the plasma temperatures measured in the plume are much lower than what
would be expected in a large tokamak (a pedestal electron temperature of 2-4 keV is thought
necessary for optimal ITER performance [133]). This limitation is difficult to overcome in relatively
small plasma sources such as the SPG, and efforts to increase the plasma temperature using exter-
nal sources (such as pulsed bias fields) often comes at the expense of directed kinetic energy and
velocity (see, for example, Ref. [67]), reducing the overall particle and energy fluxes. In addition, the
authors recognize that the near field magnetic topology of the SPG is similar to a continuous-flow
Z-pinch, and thus substantially different from that in the vicinity of a tokamak divertor under typical
configurations [41, 134]. However, plasma-material interaction dynamics that are governed by the
total energy and heat fluxes, more so than the ion or electron temperatures themselves and/or the
particular magnetic field configuration, are well within the experimental reach of systems such as
the SPG.
6.2.3 Witness Plate Target Studies

In an effort to determine the type and degree of damage caused by the accelerated plasma jets, a
series of experiments using witness plates was performed as a means of investigating the bulk effects
of the jet on the sample. The first set of targets used were polycrystalline silicon wafers, placed
normal to the flow direction along the central axis of the accelerator. The second set of targets used
were unpolished, mirror-finish pure copper tokens, also placed normal to the flow direction along
the central axis of the accelerator. In each case, the target was located 10 cm from the exit plane
of the accelerator, and held in a sample holder mounted to the linear axis stage; the stage was kept
at the same position for all trials. The stage was isolated from any external electronics, and the
samples were electrically isolated from the stage, such that the targets were allowed to float relative
to the plasma potential and ground.
The objective of this study was twofold: first, to establish a baseline characterization of whether
the facility can produce observable damage on a target substrate, and second, to preliminarily
investigate whether the aggregate energy delivered to the target or the peak energy flux drive the
scale and qualitative aspects of the observed damage.
Results: Silicon Targets
The first set of shots was performed using polycrystalline silicon wafers. A series of representative
SEM images, depicting typical damage morphologies and observed features, is shown in Fig. 6.1.
There is clear evidence of large voids that could be produced by macro-particles entrained by
the plasma jet, but there are no obvious indications of surface melting. The depth of the macro-
particle bore into the sample is beyond what is resolvable using the SEM, as shown in Fig. 6.1e.
A single shot also produced small island-like features, visible in Fig. 6.1c, which we interpret to be
either redeposited silicon material from elsewhere on the wafer or deposited ablated material from
the accelerator itself. The stress ripples and surface cracking shown in Fig. 6.1b and 6.1d show
that significant mechanical stress is being generated in the surface layers of the target. These results
(a) particle damage (b) stress ripples
(c) material deposition (d) surface cracking
(e) particle milling (f) impact cavities
Figure 6.1: SEM images of observed material damage on Si wafers after exposure to a single 6 kV
shot. Subfigures (a)-(e) correspond to the same magnification (scale bar visible), subfigure (f) is a
ten times greater magnification of the image from (e).
indicate that the plasma environment produced by the SPG is capable of producing material damage
in crystalline targets that can be observed and related back to the plasma source; however, since
metallic targets are more representative of actual PFCs in a reactor and have substantially different
bulk properties than Si, a more detailed study using copper targets was also conducted. Though
copper itself is unlikely to be a plasma facing material, the relationship between observable damage
morphologies (e.g., melting) and energy flux is well-characterized for copper, making it a suitable
material choice without a priori knowledge of what the effects of the plasma might be on a more
robust target material (e.g., tungsten).
Results: Copper Targets
The optical micrographs of the series of targets irradiated by the SPG facility are shown in Fig. 6.2.
Since the copper targets were not additionally polished beyond the stock mirror finish prior to
exposure, the initial surface characteristics displayed a fairly large degree of initial roughness on the
micro scale (shown in Fig. 6.2a) due to residual machining marks. However, this led to what is one
of the more interesting results of the target study: the damage from the impinging plasma jet was
observed to localize preferentially along the linear grooves in the material. Examples of this are
clearly visible in Figs. 6.2f, 6.2g, and 6.2l. The latter micrograph shows the boundary between the
exposed and unexposed portions of the witness plate, where the transition from unexposed linear
grooves to linearly-structured damage zones is clearly visible.
The additional objective of the target study was to determine whether total energy flux, i.e.
over multiple low-energy shots, would achieve the same types of damage generated by a single high-
energy-flux shot. As is clearly visible from the micrographs, equivalent total energy fluxes do not
produce the same results. For example, the SPG fired ten times at 2.0 kV (Fig. 6.2j) corresponds
to the same total energy flux as the SPG fired at 6.32 kV a single time (Fig 6.2f). This comparison
assumes that the actual energy flux in the jet correlates to the initially stored energy in the capacitor
bank, but it is evident that even if there is a moderately non-linear scaling of the jet parameters with
energy, that the damage sustained is much more strongly dependent on the peak fluxes experienced
by the target.
(a) Unexposed (b) 2.00 kV, 1 shot (c) 3.00 kV, 1 shot
(d) 4.00 kV, 1 shot (e) 5.29 kV, 1 shot (f) 6.32 kV, 1 shot
(g) 8.00 kV, 1 shot (h) 2.00 kV, 4 shots (i) 2.00 kV, 7 shots
(j) 2.00 kV, 10 shots (k) 8.00 kV, 10 shots (l) 8.00 kV, 1 shot
(m) 8.0 kV, 10 shots (n) 8.00 kV, 10 shots (o) 8.00 kV, 10 shots
Figure 6.2: Compilation of optical micrographs of damage to Cu tokens, corresponding to all the
tested conditions in the witness plate study. Subfigures (a)-(l) correspond to the same magnfication
(scale bar visible), whereas subfigures (m)-(o) are images taken at ten times greater magnification.
6.2.4 PMI with the SPG Facility

The above results and analysis indicates that the SPG facility is suitable for generating ELM-
like loads for testing candidate PFC materials and configurations. The tested configurations were
confined to normal-incidence impacts of the plasma jet onto the target surface, but the facility is
easily adaptable to oblique incidence angle configurations, grouped targets with fixed separation
(as in the ITER divertor cassettes), and other such configurations. Though the overall fluence and
energy density remains somewhat below that expected for single ELM events in the regimes tested,
the peak fluxes are sufficiently large at high energy (> 1 kJ/pulse) that the shortcoming in total
energy deposition can be mitigated over the course of multiple firings. Furthermore, the facility is
capable of operating at maximum energies of 12 kJ/pulse, i.e. one order of magnitude higher input
energy , and thus the predicted energy densities for ELM events are at least theoretically attainable
over the course of a single shot.
The data collected via the QLP enabled the calculation of the direct plasma parameters neces-
sary to determine the bulk energy flux of the jet, and the target studies indicated that achieving
the requisite damage threshold is a strong function of the peak energy flux. The target studies
also provided a promising avenue for further work, studying the specific mechanism responsible for
damage localization at initial sites of surface irregularity. One explanation is the reduced ability of
the substrate to conduct heat away from the increased surface area in the vicinity of the groove, as
well as increased field electron emission from the surface in the vicinity of the sharp groove edges
driving higher ion fluxes into the surface. This interesting result reinforces the fact that the proper
evaluation of potential PFCs under fusion conditions requires the use of high energy plasma sources
such as the SPG facility, in order to capture the complex interaction of the plasma and material
interface. In future work, we will expand the selection of PFC targets to include materials of greater
relevance to fusion systems, such as pure tungsten, tungsten alloys, and carbon-containing materials.
6.3 Space Propulsion

Pulsed plasma thrusters are an important class of electric propulsion devices, providing a number of
advantages over other ion accelerators. The deflagration mode has several advantages over the more
frequently seen detonation or snow-plow mode, such as increased electrostatic to kinetic energy
conversion efficiency and reduced electrode erosion. In spite of this, a practical implementation as
a small-to-medium-sized electric propulsion system has not yet been realized. In this section, we
investigate a thruster designed specifically with low average power and high specific impulse in mind,
as these attributes are required for such a system.
The QLP was used to obtain time-resolved measurements of the plasma properties in the plume of
two different thrusters. We also use a fast-framing ICCD camera to make qualitative observations of
the discharge characteristics at multiple times during a single firing of the thruster. We compare and
6.3. SPACE PROPULSION 135
contrast the behavior of the thrusters operated under various conditions, and indicate the apparent
optimal operating mode in order to achieve maximal directed plume kinetic energy.
6.3.1 Facility Modifications

Vacuum Chamber
All experiments were carried out in the 6 internal diameter Pyrex tube with Conflat flanges at either
end, coupled to a 6 Conflat T-section, which is in turn mated to a six-way cross before finally linking
to the large chamber. All tests reported here were performed at a nominal background pressure of
107 Torr.
Power System
The power system relied upon for this study was unchanged, in spite of the fact that the SPG
capacitor banks are plainly unsuitable for a space propulsion system. In a practical implementation
of a complete system, a power processing unit of similar energy density and output characteristics
could be obtained in the requisite form factor.
Thrusters
Two thrusters were used in these experiments, driven by the same power system. The first thruster is
an integrated unit designed as a stand-alone thruster, containing a microfluidic pulsed solenoid valve
as a gas injector and a solid coaxial electrode configuration. The second thruster is of a modular
design, enabling modification of the electrode geometry. This second thruster is based on the larger
Stanford Plasma Gun, and gas injection is performed by the same custom puff-valve.
Both thrusters were sealed to 6 Conflat flanges so that they could be mounted directly to the
vacuum chamber, and each thruster was made to pass through its corresponding flange such that
the electrodes were accessible from outside the vacuum.
The Stanford-Busek Thruster is shown in Fig. 6.3a. The electrode assembly consists of a solid
outer cylindrical conductor made of stainless steel, with a copper center electrode. The outer elec-
trode is 5 cm in outer diameter, while the outer diameter of the center electrode is 0.75 cm. The
copper center electrode fastens to a copper backplate, through which the injected gas flows via a set
of gas-distribution inlets (shown in Fig. 6.3a). The electrodes are held at high potential, separated
by a sealed insulator, and breakdown is initiated by the gas puff entering the interelectrode region.
The Miniature Stanford Plasma Gun (mSPG or M-SPG) is designed to mimic the bulk geometry
of the SBT, but includes the advanced custom-built puff valve to increase the density of the injected
gas as well as a rod-electrode configuration for the outer electrode. The rod anode configuration
(shown in Fig. 6.3b) both alters the inductance per unit length of the thruster, and also allows
(a) Photograph of the SBT, not mounted to the CF (b) Photograph of the M-SPG thruster, mounted to
flange. the CF flange.
Figure 6.3
residual gas and unaccelerated neutrals to diffuse radially away from the center electrode after the
plasma is ignited.
Method: Plume Characterization
The QLP was used to obtain simultaneous time-resolved measurements of the plasma density, tem-
perature, potential, and ion Mach number at a single spatial point along the acceleration axis of
the thruster plume. As in Chapter 3, the QLP was mounted to a linear axis stage, and moving the
probe along the axis over the course of multiple firings enabled the compilation of a spatiotemporal
contour of each of the plasma state variables.
Fast-Framing Imaging
In order to characterize the general behavior of each thruster configuration, the Cordin 220-UV
ICCD camera was used to obtain several frames of the optical emission produced by the plasma
during each discharge event. The operation mode of the thruster, i.e. detonation/snow plow or
deflagration, was determined based in part on the observed evolution of the accelerated plasma.
6.3.2 Polarity Comparison

The first set of experiments was performed on the SBT device, operated at a constant energy of 44 J
per pulse on a single capacitor (corresponding to a charging voltage of 2.5 kV). A full QLP contour
was collected for each polarity condition. In the first case, the contour of which is shown in Fig. 6.4a,
the center electrode of the SBT was positively charged and the conventional current direction (ion
current) was driven radially outward. The contour in Fig. 6.4b shows the second case, in which the
(a) Plasma density contour of the SBT operated in (b) Plasma density contour of the SBT operated in
outer-cathode, positive-polarity mode. inner-cathode, negative-polarity mode.
Figure 6.4
center electrode of the SBT was negatively charged and the ion current was driven radially inwards.
The white regions in both figures are noise due to the fact that the QLP system of equations does
not solve here, since the first real signal due to plasma contact with the probe is represented by
the lower edge of the contour. The difference in the time from initial breakdown between the two
test cases is due to different trigger levels on the oscilloscope, which was triggered by the discharge
current.
From a purely magnetohydrodynamic perspective, the direction of the current should not alter
the acceleration process; the vector of the Lorentz force remains in the axial direction. However,
it is clear from the results that the behavior of the thruster is substantially different between each
case. The multiple expelled plasma plumes are visible as striations in the contours, and the velocity
of the plume can be estimated from the slope of the leading edge of the contour. In the positive-
polarity case, there is not a significant drop in plasma density along the axis between the first two
acceleration events. Contrasted with the negative-polarity mode, where a clear separation is seen in
time between the first two events, this indicates that the plumes are localized more strongly in time
when the center electrode is the cathode. One explanation of this phenomenon is that the ions are
accelerated radially inwards as well as axially in the negative-polarity case, leading to a collimated
jet on the axis and an overall less-diffuse accelerated plasma plume.
The ICCD images, shown in Figs. 6.5 and 6.6, show clearly that in the positive-polarity case,
the thruster is operating in a detonation or snow-plow mode. This is indicated by what appears
from the images to be a plasma sheet propagating axially away from the thruster exit. The thruster
exit is not visible in the images in either case. In the negative-polarity case, for the same ICCD
gain settings and frame timing, we observe a radically different behavior even at the same charging
energy. Instead of a propagating sheet, there is a much brighter region concentrated nearer to the
(a) 5 s (b) 6 s (c) 7 s (d) 8 s
Figure 6.5: Fast framing images of a representative firing of the SBT in positive-polarity mode.
(a) 5 s (b) 6 s (c) 7 s (d) 8 s
Figure 6.6: Fast framing images of a representative firing of the SBT in negative-polarity mode.
central axis. There may be a less-strongly emitting plasma that is further downstream than this
region but is invisible on the CCD array due to its significantly lower emission strength.
The most probable explanation for this behavior is that the ion current is driven radially inwards
in the negative-polarity case, such that the strongly-emitting excited ion species are concentrated
about the center electrode. Though we do calculate a higher plume velocity in the negative-polarity
case (12 km/s vs 10 km/s), the difference is within the error of our method of calculating the
overall plume velocity on the basis of the contour composed of an ensemble of shots. From a
propulsion perspective, however, it is clear that the overall directed kinetic energy is higher in the
negative-polarity case based on the observations of the plume itself. The reduction of the diffuse
character of the accelerating plume in the tube section is an indication that in the space environment,
plume divergence would be significantly less for a thruster employing the polarity shown in Fig. 6.6.
6.3.3 Geometry & Gas-Injection Comparison

In order to compare the effects of electrode geometry and gas injection methods on the plume
characteristics, the M-SPG thruster was employed (shown in Fig. 6.3b). The rod-anode configuration
allows the gas to expand radially as well as axially while inside the electrode volume, which should
decrease the upstream residual gas density during repeated pulses. The gas-injection method in
this thruster is a custom-built puff valve designed expressly for the introduction of high density gas
Figure 6.7: Plasma density space-time contour of the M-SPG, operated at 2.5 kV charging voltage
(44 J/pulse) in the negative-polarity mode.
puffs into the inter-electrode region. As previously discussed, high density gas puffs, as well as low
upstream density, are believed to be essential to accessing the deflagration mode.
The QLP contour of the plasma density for the M-SPG thruster is shown in Fig. 6.7. The thruster
was operated in the negative-polarity mode, at the same energy of 44 J/pulse. The probe behavior
during this test was somewhat erratic, as evidenced by the streaks of high calculated plasma density
in the contour; these are not physically meaningful, and are the result of probe currents that did not
return to zero in the observed time window and thus the equations did not converge to a reasonable
value. However, the leading edge of the plume is still visible and a plume time-of-flight velocity can
be calculated. For the M-SPG, we calculate a velocity of 37 km/s, which is approximately a factor
of four higher than both the negative and positive-polarity operating modes of the SBT.
A series of ICCD images of the M-SPG in operation are shown in Fig. 6.8. As we can see in the
photographs, the rod-anode configuration allows a substantial amount of the processes gas to escape
around the periphery. However, there is a localized jet that emerges from the gun as well. Based on
the QLP data, this axial jet has a significantly higher velocity than the on-axis plume produced by
both sets of SBT experiments.
(a) 5 s (b) 6 s (c) 7 s (d) 8 s
Figure 6.8: Fast framing images of a representative firing of the M-SPG in negative-polarity mode.
6.3.4 Space Propulsion with the SPG

In the study of this application of the SPG facility, we have measured the plume velocity of the SBT
system in both a positive-polarity and negative-polarity mode, and found the plume parameters
and velocity along the axis to be similar but with vastly different observable characteristics. The
expelled plasma of the former mode behaves as a radially-expanding, quasi-spherical sheet followed
by a diffuse plasma that fills the test section. The latter mode, by contrast, behaves as a substantially
more collimated, brighter plasma plume that expands axially downstream, indicating that the thrust
efficiency in the negative-polarity mode should be greater.
The M-SPG thruster was used to test whether sharpness of the injected gas puff plays a role in
the observed plume conditions, as it employs a specialized valve designed to inject such a gas puff
instead of the commercial microfluidic valve incorporated into the SBT. We found that significantly
higher axial velocities were observed using the M-SPG thruster, while keeping all other operating
condition variables the same. In future work, we will replace the rod-anode configuration with an
enclosed outer electrode as in the SBT design, in order to de-couple the impact of the custom valve
and anode design on the plume characteristics.
Based on these preliminary results, we see great potential in a gas-fed pulsed plasma deflagration
thruster employing a high-density gas puff valve. The high plume velocity, if it can be extended
to the remainder of the injected gas by containing it within the electrode volume to be ionized
and accelerated, would enable a high Isp and low average power pulsed electric propulsion system,
capable of providing orbit-transfer or station-keeping capacity to a host of small and medium-size
spacecraft.
6.4 Laboratory Astrophysics

Astrophysical plasma jets are ubiquitous throughout the universe, occurring in environments such
as planetary nebulae, active galactic nuclei, and young stellar objects. Many of the jets arising
from these sources achieve high velocities, v 100 300 km/s, while still maintaining remarkable
6.4. LABORATORY ASTROPHYSICS 141
collimation over vastly different timescales [135, 136]. For example, planetary nebulae have been
observed to produce periodic jets with characteristic timescales around 1000 years where as Herbig-
Haro objects yield outflows lasting up to 105 years. In terms of spatial evolution, the majority of
jets are fractions of parsecs in length and feature large length-to-width ratios indicating they remain
narrow as the jets propagate over vast spatial scales [137]. Even with the ubiquitous nature and
contemporary interest in plasma jets, there is still little known about the formation dynamics, role
of instabilities, and the nature of interactions with their respective backgrounds.
Many of the remaining unknowns in astrophysical systems are driven not only by the complexity
of the environments but also the vast disparity in scales involved. When also considering the enor-
mous distances such objects are from Earth, it comes with little surprise that many astrophysical
data sets are limited in their spatial, temporal, and even spectroscopic resolution. Laboratory ex-
periments overcome much of these shortcomings by offering a repeatable, high resolution platform
that can be used to complement astrophysical observations and numerical simulations. Although
the benefits of superior repeatability and access are apparent for laboratory experiments, it is still a
challenge to ensure that a similarity is maintained to relevant astrophysical systems. Without such
similarity, there is no guarantee that predictions made regarding the behavior of laboratory experi-
ments will hold when scaled to larger systems. The most direct way of ensuring this similarity is by
achieving the exact astrophysical conditions in the laboratory, something employed in the study of
equations of states in planetary interiors [138]. For situations where this is not possible, similarity
is still maintained when quantities of interest (such as pressure, density, space, time, etc.) can be
mapped between the systems via multiplicative constants.
With the advent of high energy devices (lasers, fast z-pinches), it has become possible over recent
years to produce and study hypersonic jets in the laboratory setting. Using such facilities as the
OMEGA laser, recent work [139] has produced and studied the interaction of jets with ambient media
leading to the formation of bow shocks. Other research [140] has utilized fast z-pinches in the form of
a conical array of fine exploding wires to produce radiatively cooled hypersonic jets. Although scaled
jets with impressive velocities and densities can be produced with such devices, there is significant
research [141143] that points to the jets magnetic field, associated electric current, and lifetime as
important metrics when trying to emulate the structure and dynamics of astrophysical flows.
In this study we combine the data detailed in Chapters 4 and 5 into relevant dimensionless groups
and discuss how these numbers relate to astrophysical systems. The plasma density, velocity, and the
resulting pinch structure are specific properties which, along with calculations of both the plasma
temperature and magnetic field, were determined to be critical parameters in deciding whether or
not there may be relevant similarity to astrophysical flows.
8 mm
a) t = 1.6 s b) t = 1.8 s c) t = 2.0 s
8 mm
d) t = 2.2 s e) t = 2.4 s f) t = 2.6 s
Figure 6.9: Selected time resolved images of the early formation stages of the plasma deflagration
event for a capacitor charging voltage of 9 kV. These images are a view of broadband plasma self-
emission and are used to to estimate jet velocity in Fig. 6.10
6.4.1 Jet Velocity

The plasma deflagration velocity was estimated using a time of flight method where the leading
edge of the jet was tracked as a function of time. To achieve this, the Shimadzu HPV-X2 CCD
camera was employed and setup in the configuration detailed in Chapter 5. Both the laser backlight
and gradient filters required for Schlieren imaging were removed to allow an unobstructed view of
broadband plasma self-emission. Unlike other studies where a questionable assumption of exper-
imental repeatability must be assumed due to device limitations, this camera allowed a complete
time-resolved video of each individual trial.To illustrate the higher end of the velocities achievable
with the device, a capacitor charging voltage of 9 kV was selected for this component of the study.
Of the 100 images of the jet, a small subset were identified at early times in the deflagration
event where the interface between the plasma and vacuum was clearly visible. A total of six of
these images are detailed in Fig. 6.9. It is clear from these emission images that many of the
structural features of astrophysical flows, namely, high-velocity and collimation are produced from
the accelerator. To estimate velocity from these emission snapshots, an edge tracking algorithm was
used to identify the position within each image of the plasma-vacuum interface. Using the spatial
calibration determined for the plasma self-emission optical system, 0.36 mm/pixel, a mapping of
position as a function of time was found and is shown in Fig. 6.10. A resulting jet velocity of
V = 109 1 km/s was found via the slope of the resulting line. Reported uncertainties in this
number were due to both spread in the identified plasma-vacuum edge and structural variations
observed within the jet from frame to frame.
110
11
Leading Edge Data
10 Fit
100
V = 109 1 km/s
Jet Leading Edge Position [cm]
90
9
80
8
70
7
60
6
50
5
40
4
30
3
0
1.6 100
1.7 200
1.8 300
1.9 400
2.0 500
2.1 600
2.2
Time [s]
Figure 6.10: Plot of plasma jet leading edge as a function of time for a capacitor charging voltage
of 9 kV. A corresponding velocity of V = 109 1 km/s was found via the slope of the resulting fit.
6.4.2 Pinch Characteristics
Although optical emission itself provides detailed information about plasma dynamics, the fact that
it depends on n2 makes it less attractive for resolving fine spatial structures. Schlieren imaging was
used instead for uncovering the dynamics and structure of the pinch region because its signal is
directly proportional to the gradient in density. Although schlieren and other refractometry based
methods have been used to visualize flow fields of countless experiments, great care was taken in the
design of our optical setup. When attempting to image dense plasmas in particular, it is mandatory
that the source of collimated light completely dominate any self-emission over the timescales relevant
to flows of interest. A 637 nm, 250 mW diode laser was used to satisfy this constraint and ensure a
signal to noise ratio, SNR, of 100 or more was achieved even during periods of the greatest optical
emission.
y y
x = 0.11 x = 0.16
5 mm 5 mm
a) 5 kV b) 9 kV
Figure 6.11: Schlieren images of the plasma jet 11 s after initial gas breakdown for a) 5 kV and b)
9 kV capacitor charging conditions. Both the pinch radius and jet collimation were observed to be
strong functions of charging voltage.
The z-type schlieren imaging apparatus of Chapter 5 was used to capture the spatial structure
and length scales of the pinch and jet, which is essential to evaluating the relevant dimensionless
parameters.
The spatial structure of the deflagration jet was investigated for two different capacitor charging
conditions, namely, 5 kV and 9 kV. Unlike the time-average density measurements, which can also
theoretically be used to obtain spatial properties of the jet, a unique advantage of the schlieren
configuration in this study is the ability for the diagnostic to capture transient structures and
instabilities of the plasma. Here, however, the schlieren diagnostic was used only to identify gradient
length scales within the jet and specific estimates of its total length and radius. Fig. 6.11 illustrates
still images of the plasma jet 11 s after breakdown once the essential features of the pinch have
been established. Using the spatial calibration of the system, 0.23 mm/pixel, a minimum pinch
radius of 5.3 mm and 3.6 mm was found for the 5 kV and 9 kV trials respectively while a jet
length of more than 56 mm was found for both. It was also found that both the pinch radius
and associated jet collimation were strong functions of the charging voltage of the capacitors and
thus device current. As the induced magnetic field is a critical parameter in determining the pinch
properties, the conclusion that the pinch becomes narrower as the current flow (and thus magnetic
field/pressure) increases is consistent with expectations.
The same equilibrium pinch model as that used in Chapters 4 and 5 was used to determine the
remaining critical features of the jet, namely magnetic field and temperature. Prior work [84] has
been performed that concludes that the equilibrium profile predicted for the classic imploding Z-
pinch, and the profile of a shear-stabilized pinch such as that formed in the deflagration accelerator,
are the same. Such a model used in conjunction with the Stark-broadened density and pinch current
profiles allows estimates of spatially dependent parameters based on empirically measured values.
The constant drift velocity model was applied to fully close the equilibrium pinch model, as
detailed in Chapter 4. The evaluation of Eqs. 4.12-4.14 requires explicit knowledge of both n(r) and
Ip . As with the jet velocity measurement, the peak jet parameters attainable over our operating
range are of particular interest. Thus calculations were carried out for capacitor charging voltages
of 5 kV and 9 kV. The density profile required for Eq. 4.12 was obtained from the Stark-broadened
spatial profiles (computed as detailed in Ch. 4) associated with the respective charging voltages. The
device current was measured by placing a wide-band current transformer around the transmission
lines connecting the anodes and capacitors. A representative trace of the current waveform measured
in this manner is detailed in Fig. 6.12.
Although the device current is convenient to measure, the Ip that is referenced in Eq. 4.12 is
the amount of current flowing through the pinch. Past studies [144] have investigated the spatial
and temporal distribution of current within the deflagration accelerator and concluded that both
effects are important to consider. It was found that virtually all of the current for much of the
first positive half-period of the underdamped LRC oscillation is collected by the cathode within the
accelerator volume and used to induce a B-field to accelerate the plasma. By the time the plasma
has reached the end of the accelerator volume, the vast majority of the current flows directly through
the concentrated area occupied by the pinch. Thus as with the estimates of characteristic length
scales, a time of 11 s after initial gas breakdown was used to obtain Ip , as marked in Fig. 6.12.
The resulting pinch properties calculated using the equilibrium model are detailed in Fig. 6.13.
These properties and specifically of note, the pressure profile, indicate the magnitude and structure
required to maintain a stable pinch. The spatial scale shown in Fig. 6.13 is a reflection of the
measured density profiles which feature characteristic radii comparable to the time-resolved gradient
length scales measured with the schlieren diagnostic. Calculations made using the model are also
consistent in both structure, trend, and value with surface mounted magnetic probe and Thomson
scattering measurements for B and T respectively for a similar device in Refs. [84, 122].
100
100
Voltage: 5 kV
80
80 Voltage: 9 kV
Time = 11 s
60
60
40
40
20
Current [kA]
20
00
20
-20
40
-40
60
-60
80
-80
00 5 10
10 15
15 20
20 25
25 30
30 35
35 40
40
Time [s]
Figure 6.12: Plot of the measured current being fed into the device from discharging capacitors at
both 5 kV and 9 kV charging voltages as a function of time. The waveform of the capacitor discharge
resembles a underdamped LRC oscillator. The current going through the pinch, Ip , was taken at
t = 11 s.
a) 2.0
2.0 V = 5 kV
V = 9 kV
1.5
1.5
Magnetic Field [T]
1.0
1.0
0.5
0.5
0.0
0.0
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
Radius [cm]
b) 55
44
Pressure [MPa]
33
22
11
00
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
Radius [cm]
c) 70
70
60
60
50
50
Temperature [eV]
40
40
30
30
20
20
10
10
00
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
Radius [cm]
Figure 6.13: Plots of the pinch properties as a function of radius derived from the equilibrium model:
a) magnetic field, b) pressure, and c) temperature. Note the pressure here is the equilibrium pinch
pressure and not the ram pressure of the jet.
6.4.3 Astrophysical Scaling
It is important to establish both device parameters and to calculate relevant dimensionless numbers
for laboratory astrophysics experiments. The experimental measurements presented in this work
were used to quantify the plasma jet properties. Namely, Stark broadening was used to measure
the density, time of flight was used to measure velocity, Schlieren imaging was used to measure
gradient length scales and the pinch model was used to obtain both the plasmas magnetic field
and temperature. From these properties, a number of other quantities can be calculated that are
relevant to magnetohydrodynamic flows.
A number of studies have investigated the scaling and similarity properties of the governing
plasma fluid equations by establishing dimensionless groups. Some research such as Ref. [147] include
additional physics to account for optical depth effects of the plasma. As there are so many different
similarity variables for virtually any astrophysical body, there are no laboratory experiments that can
claim to recreate all essential physics. Thus in many cases, the suitability of a laboratory experiment
is instead determined by its ability to recreate specific physics of interest. Three dimensionless groups
that are universally cited as important metrics for establishing the similarity between astrophysical
and laboratory systems are the Reynolds number, Re, magnetic Reynolds number, Rem , and Euler
number, Eu,
VL
Re = , (6.2)

Rem = V L0 , (6.3)
r

Eu = V . (6.4)
p
Within these expressions, V refers to the jet velocity, L is the characteristic length scale in the flow,
is the plasma kinematic viscosity and is the conductivity. Expressions for both and are
given in Refs. [148, 149] as,

6 A[T (eV)]5/2
3.3 10 ,

[m2 s1 ] = Min Z 4 (kg m3 ) (6.5)

2T (eV)

ZB(T)
Z
[ m] = 1.96k = 1.039 104 3/2 . (6.6)
Te (eV)
Within the equation for , two different formulas are presented where in one the effect of a magnetic
field is neglected and the second it is taken into account. As detailed in Ref. [148], the minimum
of the two is taken as the plasma viscosity. Remaining terms in Eqs. 6.5-6.6 include, the Coulomb
logarithm ( 10), the ion atomic mass number, A, and magnetic field entanglement parameter, .
Table 6.1: Characteristic scales and dimensionless parameters for a variety of astrophysical envi-
ronments and laboratory scale experiments [145, 146]. The properties specified for the deflagration
accelerator were the maximum measured and L was taken as the pinch radius.
Environment L [m] V [m/s] n [m3 ] T [eV] B [T] Re Rem Eu

Warm ISM 3 1016 105 106 1 109 107 1019 10
Dense Cloud 3 1016 5 103 109 102 108 7 1013 6 1014 6
Stellar Atmosphere 107 105 1021 1 102 5 1012 4 109 10
AGN Disk 2 1011 3 106 1018 103 102 7 107 6 1013 10
XRB Disk 104 3 105 3 1027 103 102 109 4 1011 1
Omega Hohlraum 104 105 1028 102 102 3 103 5 101 1
NIF Hohlraum 3 104 2 105 1028 3 102 102 103 2 103 1
Z Experiment 103 105 1028 102 102 3 104 5 102 1
Short Pulse Laser 105 106 1030 103 104 103 103 4
Deflagration Accel. 4 103 105 1023 6 101 2 10 2 102 1
Table 6.2: Comparison of critical length and temporal scales for the laser produced jets [1, 2] and
those from the device considered in this work.
Quantity Laser Deflagration

Jet Length [m] 5 103 6 102
Jet Radius [m] 1.5 104 4 103
Time Scale [s] 2 109 105
In Eq. 6.6, it was further assumed that the current flow in the pinch is normal to the self-generated
magnetic field.
The measured characteristic parameters for the deflagration accelerator along with a variety of
other experimental platforms and astrophysical flows is detailed in Table 6.1. Compared to many
of the laser experiments, our device offers a larger characteristic length scale, similar velocity/
temperature, but lower plasma density and magnetic field. However, the deflagration jet offers
significant advantages in lifetime (over 1000 times longer than those summarized in Refs. [1, 2]) and
both radial and axial extent, as shown in Table 6.2. These features allow the accelerator to offer a
unique ability to study dynamics and instabilities with longer timescales.
In terms of dimensionless numbers, the scale between experiments and astrophysical flows is
large. In many cases for validation of numerical codes and testing of physics, it is sufficient to ensure
the regime of the governing equations is maintained between systems. As nearly all astrophysical
flows of interest, and all of those shown in Table 6.1, feature both large Re and Rem , an important
determination is if a given experiment can maintain the ideal magnetohydrodynamic constraints,
Re 1 and Rem 1. In terms of the magnetic Reynolds number, our experiment maintains
the requirement that Rem 1 implying that flux lines of the magnetic field are advected by the
resulting flow. The other constraint is not strictly met as Re 10 implying that viscous effects
might be important, however it is still an order of magnitude above unity. The applicability of the
experiment for simulating specific physics not considered directly in this paper can be answered by
taking the experimental parameters in Table 6.1 and evaluating the appropriate dimensionless group
of interest.
6.4.4 Laboratory Astrophysics with the SPG

This study has presented the exhaustive experimental characterization of the SPG in the context
of astrophysical flows. The deflagration accelerator boasts enormous advantages of inexpensive
upkeep, high repeatability, and nearly unprecedented diagnostic access compared to conventional
laser facilities. Beyond that, the device removes practical issues such as target fabrication while
incorporating essential physics in that the produced jet is completely driven by internal current flow
and resulting induced magnetic fields. With the measured similarity parameters and ability to study
time-resolved phenomena including instability dynamics and plasma interactions, this experimental
platform is well suited to investigate contemporary astrophysical problems.
6.5 Summary
In this chapter, we presented detailed studies in a number of application areas in which the SPG
facility can be employed. We relied, in various combinations, on the diagnostics previously developed
and described in Chapters 3, 4, and 5. First, we outlined a study of plasma-material interactions
using the SPG as a high energy directed plasma source, and found that the heat flux parameter
ranges corresponding to ELM events can be suitably replicated at a plasma-material interface.
Second, we used two miniaturized forms of the SPG designed to operate as low-power gas-fed plasma
deflagration accelerators to study the feasibility of the deflagration accelerator concept in the area of
space propulsion. Finally, we analyzed the imaging spectrometer and schlieren data in the context
of various non-dimensional scaling parameters to demonstrate the usefulness of the SPG facility as
a platform for laboratory astrophysics.
Chapter 7
Conclusion and Future Work
This dissertation has detailed the experimental investigation of the Stanford Plasma Gun, a plasma
deflagration accelerator, using several interrelated diagnostics and methods. Over the course of the
three primary testing campaigns described herein, we developed our understanding of the physical
mechanisms governing the behavior and dynamics of the deflagration-produced jet in increasingly
complex ways.
First, we used an immersed probe to establish a baseline measurement of the plasma parameters
in the jet. We also expanded the probe study to include aggregated time-of-flight measurements
that allowed us to determine the velocities of the apparent plasma detonation- and deflagration-like
events within the gun. By combining the plasma parameters upstream of the discharge region,
the plasma parameters downstream of the discharge region, and the parameters of the flow (i.e.,
velocities) both upstream and downstream of the discharge region, we were able to quantitatively
evaluate both the mass and momentum conservation equations across the ionization and acceleration
waves. Without a detailed understanding of the precise mechanisms of energy addition to the flow,
we nevertheless demonstrated that Rayleigh lines corresponding to both a plasma detonation, in
which the total pressure and density rose across the wavefront, and a plasma deflagration, in which
the total pressure and density fell across the wavefront as the downstream plasma was efficiently
accelerated downstream. While this study does not, in itself, unequivocally prove that the MHD
Rankine-Hugoniot model is the primary theoretical description of the operation of the SPG and
accelerators of its class, it does provide an important addition to the existing body of work that
points in this direction.
However, in spite of this step forward, our understanding of the SPG itself was missing an
important category of information. It was plain from the qualitative imagery of the discharge and
other supporting data that the interesting physics occurs in the vicinity of the exit plane of the
gun (i.e., the near-field), where the plasma conditions clearly reach their extremes. The QLP was
unsuitable for studying this region in several ways: first, the probe was not designed to operate
151
152 CHAPTER 7. CONCLUSION AND FUTURE WORK
in the high density and temperature regime of the near-field. Second, the inherent perturbation
to the plasma state induced by the probe is undesirable from the perspective of accuracy. Third,
obtaining the full complement of information from the QLP required the probe to be axially scanned,
which made studies of how the plasma parameters scaled with the operating conditions of the gun
practically infeasible.
In the second phase of this work, we developed a spectroscopic diagnostic that enabled us to
collect a snapshot of the radially-resolved plasma parameters during a single firing of the gun in
order to address these issues with the immersed probe and to focus our investigation toward the
region of interest. The imaging spectrometer enabled us to uncover the scaling laws for both the
peak number density and line density in the focus region downstream of the SPG exit plane. Our
findings indicated that the peak number density and the line density scaled in a manner consistent
with a radial compression of the plasma core with increasing drive current. We then related this
feature of the SPG to the geometry and mechanic of the classical Z-pinch discharge, and employed a
few common and empirically-validated assumptions to develop an equilibrium model of such pinches
that related an expanded set of plasma parameters to the measured plasma density profile.
Using the model thus developed, we predicted the radial profiles of the plasma temperature,
azimuthal magnetic field strength, and kinetic pressure based on the measured plasma density profiles
at each operating condition. We checked the model predictions against the plasma temperature
estimated from the Doppler broadening of iron impurity ions, and found that the model predictions
were broadly consistent. As a further check, we computed an effective specific heat ratio of the plasma
by combining the empirical scaling laws with the scaling relations developed for an adiabatically
compressed Z-pinch, and found the results to be consistent with an ideal monatomic hydrogen
plasma. Thus, we found that both the scaling and modeling analyses were mutually consistent and
demonstrated the likely formation within the emergent jet of a Z-pinch in radial magnetohydrostatic
equilibrium.
The imaging spectrometer was not without its own drawbacks, however. First, though the
emission measurement is non-perturbative, it relies on light emitted by neutral hydrogen within the
highly ionized plasma instead of a direct measurement of the plasma density. Second, the minimum
required signal to noise ratio to perform suitable lineshape analysis dictated that light be collected
for the entire duration of the deflagration. Thus, time resolution and an understanding of the
dynamics of the discharge were outside the scope of this diagnostic tool. Third, spatial resolution
was achievable only in the radial direction; collecting data from various axial positions was infeasible
given the alignment and calibration difficulties associated with a single axial position.
The final diagnostic was designed to retain the advantages of the imaging spectrometer while
addressing the aforementioned disadvantages. We chose to develop a high speed schlieren and emis-
sion imaging apparatus in order to satisfy these criteria. At the highest frame-rate ever achieved for
153
a time period including an entire drive current cycle, we obtained cinematic schlieren and conven-
tional imagery of the deflagration and detonation events. In the schlieren case, we developed new
experimental methods and techniques to advance the state of the art of laser plasma schlieren, as
well as novel image processing techniques to overcome several image artifact types that commonly
appear in laser schlieren images. We also calibrated the schlieren apparatus to enable extraction of
the plasma density structure from the schlieren imagery.
Cinematic imaging of the plasma density structure revealed the formation and dynamic evolution
of coherent, magnetically-confined structures within the jet, and enabled the calculation of their
growth and advection rates from the image sequences. Next, we applied a similar inversion technique
as that used for the imaging spectrometer data to generate rotationally-symmetric three dimensional
maps of the plasma density of the pinching core of the plasma jet. We also applied a simple stability
model to the asymmetric pinch, which suggested that a time-varying sausage mode excited at the
tip of the cathode could explain the observed quasi-periodic self-modulation of the pinch diameter.
Finally, we briefly reviewed studies of several applications of the plasma deflagration accelerator.
The SPG facility was employed for the study of: plasma material interactions, particularly the
replication of the damage induced by ELM events in tokamak fusion reactors; space propulsion,
particularly low power, gas-fed pulsed plasma thrusters that access the efficient acceleration qualities
of the deflagration mode; and laboratory astrophysics, particularly the suitability of the SPG for
replicating the scaled parameters of various stellar outflows and other astrophysical phenomena.
However, though significant forward progress was made towards understanding the physics of the
SPG and related devices, several open questions remain that would provide fruitful avenues for future
research. First, the MHD Rankine-Hugoniot model could be explicitly evaluated by making direct
measurements of the energy within the plasma flow on both sides of the hypothesized deflagration
and/or detonation waves. Such measurements would enable all three of the relevant conservation
relations (i.e., mass, momentum, and energy) that form the basis of the Rankine-Hugoniot curve
to be quantitatively applied to the SPG system, which would in turn permit the existence (or
nonexistence) of an actual Hugoniot curve to be determined.
Second, though it has drawbacks as a plasma density diagnostic, the imaging spectrometer (or
a similar spectrographic instrument) could be repurposed as a Doppler-shift velocimeter in order to
measure the velocity profile of the jet (which is an as-yet-unknown jet parameter). By adjusting the
viewing direction of the spectrometer from perpendicular to the jet flow direction to an oblique angle,
a component of the jet velocity will be aligned along the optical axis of the imaging spectrometer.
The light emitted by the entrained (or recombined) hydrogen atoms will be blueshifted due to the
relative velocity between the emitter and the spectrometer, and the velocity can be computed from
the frequency shift of the optical signal. Such a spectroscopic diagnostic as this would enable a
direct measurement to be made of the velocity profile through the plasma jet, and for the shear flow
between the high-density plasma core and the surrounding plasma flow to be explicitly observed.
154 CHAPTER 7. CONCLUSION AND FUTURE WORK
Third, several steps could be taken to improve the schlieren diagnostic to build of the work already
performed. While the schlieren signal is technically quantifiable (and such quantification was made,
to the extent possible) it is not ideally suited for absolute quantification of the plasma density.
However, the schlieren apparatus could be converted into a shearing interferometer whose output
intensity is directly proportional to the plasma density, instead of related to the plasma density
according to the non-linear relationship defined by the integrated deflection angle in combination
with the schlieren cutoff. The remainder of the apparatus (i.e., the light source, the camera, etc.)
could be preserved or upgraded as necessary to facilitate the conversion, but there is no constitutive
barrier to performing an interferometric measurement with the existing equipment.
Beyond these concrete improvements upon the specific investigations and diagnostics imple-
mented in this dissertation, other potential areas of interest surrounding the underlying operation of
the SPG include: the symmetry and dynamics of the discharge as it forms at early times (i.e., imme-
diately post-breakdown), the magnetic field structure of the plasma jet between the exit plane and
several gun diameters downstream, and the sensitivity of the gun operation to various drive-current
waveforms (e.g., a square wave instead of a decaying sinusoid). With regard to potential future
applications of the SPG facility, further investigation into the dynamics of the plasma-material in-
terface and the basic flow physics of the interaction between the high velocity magnetized jet and
various immersed objects are both areas of broad interest in the plasma physics community.
Appendix A
Experimental Facility Procedures
A.1 Chamber Pumpdown Procedure

1. Vent chamber with side valve make sure red isolation valve above the chamber is shut
There will be a loud hissing noise as the chamber returns to atmospheric pressure; this
is normal.
2. Check gas feed systems:
Is the correct gas attached to the feed valve?
Are all valves shut, and all connections tightened?
Is the feed valve in the closed position?
3. Check gun subsystems:
IMPORTANT: Ensure grounding loops are in place on ALL capacitors, and touch
grounding rod to ALL metal surfaces, BEFORE touching any part of the gun systems.
Is the gate valve between the gun and chamber closed ?
CLOSED: check with last user before opening, as the gun section may be depres-
surized.
To open, plug in the black power cord attached to the gate valve.
OPEN: proceed with pump down.
Is the gun installed onto the glass tube, and are all connections tight?
If gun is not installed, replace copper gasket and reinstall. Gaskets are in the supply
closet.
155
156 APPENDIX A. EXPERIMENTAL FACILITY PROCEDURES
4. Vent the main line:

IMPORTANT: Is the switch to the small chamber in the other lab (located in the
southeast / back left corner) CLOSED? If not, ASK before venting the line.
Open the large red isolation valve above the chamber.
5. Turn on the roughing pump (black), located outside the back doors of the lab:
(a) Hit the stop button to ensure that the pump is off.
(b) Open the black oil pipe (adjustable wrench required) on the right to check oil level. If
you can see oil then proceed, if NOT then replace using Inland 77 oil until oil is visible
in the pipe.
(c) Check inverted oil container on the left. To refill, hold inverted inside a container of oil
and manually hold the restricting rod to the side so oil (slowly) flows in.
(d) Open the green air intake valve.
(e) Open the brown gate valve between the pump and the main line 5 turns (enables easier
pump start by reducing initial pump load).
(f) Turn on the breakers (inside the circuit box, beneath the tape labeled Black) for the
pump and blowers.
(g) Hit the start button.
High pitch squeaking or squealing = bad.
If the sound does not disappear in a few seconds/minutes or smoke becomes visible,
there is too much tension in the belts and immediately STOP by hitting the stop
button.
(h) Go inside and close the venting valve to the left of the chamber (with the main chamber
door closed).
(i) Close the green air intake valve on the roughing pump outside.
(j) Open the brown gate valve on the roughing pump all the way.
(k) If necessary, can open the left lower red twist valve to drain oil momentarily, but must
be kept closed.
6. Prepare the Polycold:
(a) Turn on the Polycold cooling water.

Open highest red valve to the right of the chamber when facing the main chamber
door (use ladder to reach).
Check that the float meter is floating (i.e. water is flowing).
A.1. CHAMBER PUMPDOWN PROCEDURE 157
Check that the pressure gauges on either side of the filter read close to equal. If the
pressure drop is too high, filter must be replaced.
(b) Check that all switches on the Polycold control box (located in southeast / back left
corner) are on standby.
(c) Turn on the large AC Power switch on the bottom-right of the control panel.
(d) Hit the on switch in the outlined box under System Control.
IMPORTANT: If the pressure in the chamber is below 30 mTorr (3.0 102 T on

readout): CLOSE THE RED ISOLATION VALVE! Do not suck roughing pump oil
into the chamber!
(e) Let the temperature get down to 150 C before starting.
(f) Set the local switch in the box marked PFC Circuit 1 to cool.
Crackling in the black insulated pipe is normal.

Pressure and temperature will both rise (pressure indicator is the gauge readout in
the lower center of the control panel).
(g) When discharge pressure gauge reaches 25 bar, manually set switch to standby.
A significant amount of cooling occurs in the 24-25 bar range; do not set to standby
too early or more cycles will be needed.
Watch to ensure that the pressure and temperature begin to decrease again.
(h) Wait for the temperature to return to 150 C again.
IMPORTANT: If the discharge pressure in the Polycold goes over 26 bar, TURN IT
OFF, let the pressure come back to 10 bar, turn it on again, and ensure the pressure does
not rise over 26 bar!
7. IF the pressure increases in the 100 mTorr (1.0 101 T on readout) range before the Polycold
can run a second cycle (the next step): open the red isolation valve 4 turns and WATCH the
pressure drop again.
8. Polycold cycling:
(a) Once the temperature has returned to 150 C, et the local switch in the box marked
PFC Circuit 1 to cool.
Pressure and temperature will both rise again.
(b) When discharge pressure gauge reaches 25 bar, manually set switch to standby.
Watch to ensure that the pressure and temperature begin to decrease again.
IMPORTANT: If the pressure in the chamber is below 30 mTorr (3.0 102 T on

readout): CLOSE THE RED ISOLATION VALVE! Do not suck roughing pump oil
into the chamber!
(c) Continue to cycle the Polycold until the pressure stays under 25 bar by itself, and then
leave on cool for the rest of the session.
9. Turn on the emissive pressure readout when pressure is below 20 mTorr by pressing the Fil-
ament button on the chamber pressure readout.
10. Turn on the cryo-pump:
(a) Start when the temperature is significantly below freezing ( 30 C).

(b) Turn on the cryo-pump cooling water by opening the green valve (on the pipe on the
ground near cryo controls)
Check the midsize float to the right and the pressure drop across the filters.
(c) Press the on button on the cryo-pump control box on the floor to power the pump on.
The screeching squeak is normal for the first 5 minutes, and should stop after that.
11. Leave the cryo-pump on, and the pressure should reach (4.0 8.0) 107 Torr after a few
hours.
The pressure may drop in steps, rather than continuously, as different components of air
are frozen out of the interior atmosphere at discrete temperatures.
12. Turning off the system:
(a) Turn the cryo-pump power off, leaving the cooling water on (do not shut green valve
yet).
(b) Set the Polycold to standby, then turn large AC Power switch to off.
(c) Turn off the emissive pressure sensor (press and hold the Filament button on the
pressure readout).
(d) Turn off the cryo-pump cooling water (shut the green valve)
(e) Turn off the Polycold cooling water (using the ladder)
(f) Close the brown gate valve on the roughing pump (outside)
(g) Hit the stop button on the roughing pump.
(h) Switch the breakers for the Black pumps Blowers and Breakers to off.
A.2. GUN FIRING PROCEDURE AND SAFETY CHECKLIST 159
A.2 Gun Firing Procedure and Safety Checklist

I. Preparation:
1. Verify that all capacitors are grounded.

2. Unplug sensitive lab equipment and move it to a safe location.
3. Connect grounding cable connection to external ground.
4. Check grounding rod (chicken stick) connection to external ground.
5. Check continuity to ground.
6. Establish safety perimeter, hanging high voltage in use signs as needed at lab entrances.
7. Put on safety goggles.
8. Check that all non-essential equipment is unplugged from wall power.
9. Manually test high voltage relays by pushing down on the armature and ensure they slide
easily.
10. Connect gas supply to valve.
11. Inspect all connections (including grounding):
cables coming out of charging cabinet
electronics on the optical table
capacitor banks
power supply and power supply controller
puff valve power cable
etc.
12. Ensure that the gun is properly seated and installed on vacuum chamber.
13. Ensure necessary probes are emplaced:
high voltage probe on capacitor bank(s)
current monitor on capacitor bank(s)
oscilloscope probes for any additional experiment(s)
etc.
14. Check air gaps between HV surfaces and grounded surfaces.
15. Check that gate valve at end of gun section is plugged in and open.
16. Test puff valve:
i. Turn on the power supply for the valve charging relays.
ii. Check that both the HV relay and the dump relay function properly (listen for actua-
tion).
iii. Turn on valve capacitor charging power supply.

iv. Set output voltage to 900 V.
v. Set output current to maximum (turn the knob all the way clockwise).
vi. Charge the valve by connecting the HV output to the valve capacitors using the switch-
controlled relay.
vii. Once current output goes to zero and the output voltage reaches 900 V, disconnect the
valve power supply.
viii. Monitor the chamber pressure (may require an additional pair of eyes).
ix. Switch on the valve spark-gap controller and depress the trigger button.
x. Check that the pressure momentarily rises upon puffing the valve, and listen for valve
actuation.
xi. Check that the pressure returns to baseline. If not, the valve may be leaking or a failure
may have occurred.
xii. Hold the momentary switch to connect the dump resistor to the valve capacitor bank
for 15 seconds to drain excess charge.
xiii. Leave the valve capacitors uncharged during remaining preparation.
17. Ensure power supply is turned off.
18. Remove shorting wires from capacitors on the bank(s) being used (central lead first).
19. Inspect lineman gloves for holes or defects.
20. Put on lineman gloves.
21. Ensure HV power supply controller is in Inhibit ON and HV OFF modes, and the knob
is turned fully counterclockwise.
22. Ensure relay switches are in correct positions:
power supply HV and GND relays are open

dump resistor relay is closed
23. Plug in the power cables for the relays and HV power supply.
24. Ensure that HV power supply is reading the lowest voltage on the display (0.3-0.6 kV).
25. Test relay switches by actuating the relay switches and listening for actuation.
26. Assuming successful completion of all preceding steps, the system is ready to fire.
II. Firing:
1. Charge puff valve by connecting the valve power supply output to the valve capacitor using
the switch.
A.2. GUN FIRING PROCEDURE AND SAFETY CHECKLIST 161
2. Once the valve capacitor is charged, disconnect the valve power supply using the toggle
switch.
3. Place the HV power supply GND relay into the closed position.
4. Place the HV power supply HV relay into the closed position.
5. Place the dump resistor relay into the open position.
6. Ensure the HV power supply controller is in Inhibit OFF and HV ON modes.
7. SLOWLY turn the knob on the power supply controller clockwise, allowing the displayed
voltage to rise as the knob is turned. The display reads approximately the true output
voltage of the power supply, NOT the set point; if the knob is turned too quickly, the
power supply can inadvertently be set to an output voltage that may cause sparkover on
the capacitors.
8. Once the display reads the desired charging voltage:
i. Place the HV power supply HV relay into the open position.
ii. Place the HV power supply GND relay into the open position.
iii. Ensure the HV power supply controller is in the Inhibit ON and HV OFF modes.
iv. The display voltage should droop relatively quickly; but the capacitor voltage is still
at the set voltage.
9. Turn the valve spark gap switch controller on using the toggle switch; the green light should
illuminate.
10. Loudly state your intention to fire the gun (e.g., by yelling Firing!).
11. Depress the white momentary button to puff the valve; the gun should fire and a flash should
be visible. If no such flash occurs, or some other malfunction is apparent, go to section III.
12. Turn off the valve spark gap switch controller.
13. Place the dump resistor relay into the closed position.
14. Sift through the wreckage and see whos still alive. If the population of the lab has not
decreased, repeat until science is complete.
III. In case of malfunction:
1. Assume the system is in an energized state unless proven otherwise.

2. Ensure the dump resistor relay is in the closed state.
3. Ensure the HV power supply is in the HV OFF and Inhibit ON modes, and that the
knob is turned all the way counter clockwise.
4. If the cause of the malfunction is unknown, move to section IV and power the system down
in order to begin troubleshooting.
IV. Power down:
1. Assume the system is in an energized state unless proven otherwise.

2. Ensure the dump resistor relay is in the closed state.
3. Ensure the HV power supply is in the HV OFF and Inhibit ON modes, and that the
knob is turned all the way counter clockwise.
4. Disconnecting the valve power supply from the capacitor (if it is connected) using the toggle
switch.
5. Hold the dump resistor momentary switch for 30 seconds to bleed any excess charge off the
valve capacitor.
6. Turn the voltage and current knobs on the valve power supply full counter clockwise and
turn off the valve power supply.
7. Unplug HV power supply and switching relays.
8. Ensure all power relays are in the closed position.
9. Ensure voltage on capacitor bank(s) is zero (e.g., using HV probe attached to an oscillo-
scope).
10. Ensure the grounding rod is connected to ground.
11. Using the grounding rod, discharge all metallic surfaces in the vicinity of the gun:
capacitors (including the cart and housing)
the gun flanges, both ground and HV
all exposed nodes of any transmission lines
etc.
12. Connect shorting wires across the capacitor leads of the bank(s) which were in use (ground
lead first).
13. Safely store the lineman gloves, and thank your lucky stars you made it.
Appendix B
Valve Engineering Drawings
163
8 7 6 5 4 3 2 1
APPENDIX B. VALVE ENGINEERING DRAWINGS
D D
0.0938
C C
1.480
2.188
B B
UNLESS OTHERWISE SPECIFIED: DATE
DIMENSIONS ARE IN INCHES
7/10/14
TOLERANCES:
ANGULAR: 1 degree TITLE:
ONE DECIMAL PLACE 0.1
A
Sliding Hammer Ring
TWO DECIMAL PLACE 0.01
THREE DECIMAL PLACE 0.005
MATERIAL
Aluminum Author: REV
DO NOT SCALE DRAWING
BREAK ALL EDGES
SCALE: 1:1 WEIGHT: SHEET 1 OF 1
8 7 6 5 4 3 2 1
164
8 7 6 5 4 3 2 1
0.170 THRU ALL

D 0.281 0.138 D
1.00 0.331 X 90, Near Side
0.25
C C
00
1.
B B
0.25 THRU
0.4375
DIMENSIONS ARE IN INCHES 6/12/14

TOLERANCES:
square hole pattern ANGULAR: 1 degree TITLE:
A TWO DECIMAL PLACE 0.01
Poppet Seal Flange
MATERIAL
BREAK ALL EDGES

8 7 6 5 4 3 2 1
165
8 7 6 5 4 3 2 1
0.1875
D D
1.50
Threaded with 1/4-20

please cut groove
B
0.750
suitable for rotating

3.50
0.250
part with a flathead
screwdriver
0.50
C C
0.75
DETAIL B
SCALE 2 : 1
0.486
0.0878
0.029
B B
6/12/14
TOLERANCES:
Adjustable Poppet Stop
A
Scale 5:1
MATERIAL
BREAK ALL EDGES
8 7 6 5 4 3 2 1
166
8 7 6 5 4 3 2 1
A
D 0.625 D
0.109
0.984
.20 R0
R0 .2
0 0.0625
1.109
0.75
C C
A
SECTION A-A
0.750
B B
Only break edges when necessary;
see assembly drawing for intended
fit.
1.73

6/12/14
TOLERANCES:
Poppet
MATERIAL
Nylon Author: REV
DO NOT SCALE DRAWING 1
BREAK ALL EDGES
8 7 6 5 4 3 2 1
167
8 7 6 5 4 3 2 1
D D
Poppet guide; already Coil bobbin; already made
made
C
C Poppet return spring C
Hammer return spring
C
B B
6/12/14
TOLERANCES:
SECTION C-C THREE DECIMAL PLACE 0.005
Valve Assembly Cutaway
SCALE 2 : 1
MATERIAL
Multiple Author: REV
BREAK ALL EDGES
8 7 6 5 4 3 2 1
168
8 7 6 5 4 3 2 1
3.50
D D
2.19
E
1.25
E F
SECTION E-E
C C
O-ring groove for 037 elastomer, one on

each side of the part.
0.052
6x 0.25 THRU ALL

0.0865
DETAIL F
B B
1.5
SCALE 5 : 1
2.19
15
0
Bolt Pattern
Radius

6/12/14
TOLERANCES:
THREE DECIMAL PLACE 0.005 Valve Body
MATERIAL
Acrylic Author: REV
BREAK ALL EDGES

8 7 6 5 4 3 2 1
169
8 7 6 5 4 3 2 1
2.250
D D
1.480
C
0.656
0.75
C
SECTION C-C
3.50
C C
1.51
50
B B
6x 0.25 THRU ALL
0.25 diameter thru to interior
w/ center 0.33 from top surface UNLESS OTHERWISE SPECIFIED: DATE
6/12/14
TOLERANCES:
A
Valve Coil Section
MATERIAL
Acrylic Author: REV
BREAK ALL EDGES
8 7 6 5 4 3 2 1
170
8 7 6 5 4 3 2 1
6x 0.266 THRU ALL

0.44 0.25
0.49 X 90, Near Side
D D
0.49
1.5150
0.8125 bolt pattern
radius
3.49 external mate
bolt pattern radius
C +0.002 C
0.076 - 0.002 6x 0.070 0.25
B Parker O-ring tap for #2-56
132 C inner diameter
+0.002
+0.002
0.551 0.000
0.122 - 0.002
0.000
0.451
B B
+0.002
1.981 -0.002
Parker O-ring
0.086 - 0.002
015
outer diameter
B +0.002
0.054 - 0.002
DETAIL C
SECTION B-B SCALE 2 : 1

7/23/14
TOLERANCES:
Valve Head Cap
MATERIAL
BREAK ALL EDGES

8 7 6 5 4 3 2 1
171
8 7 6 5 4 3 2 1
dimensions on detail
B are for an O-ring
A acting as a gasket
B
D D
0.297
0.058
0.75
A SECTION A-A
3.50 no thread above
this line
DETAIL B
SCALE 5 : 1
C 0.107 0.25 C
tap for #6-32
6x 0.201 THRU ALL
tap for 1/4-20
1.515 0.201 THRU ALL
0.32 X 90, Near Side
tap for 1/4-20
B
0.44 THRU ALL B
pipe tap for 1/4 NPT 0.4375
square bolt pattern
6/12/14
TOLERANCES:
A
Valve Tail Cap
MATERIAL
BREAK ALL EDGES
8 7 6 5 4 3 2 1
172
Appendix C
Imaging Spectrometer Addendum
This appendix contains additional information regarding the imaging spectrometer experiment in-
cluding: the image used for spectral and spatial calibration, the complete set of imaging spectrometer
data points collected to produce the scalings outlined in Ch. 4, and the analysis scripts used to gen-
erate the data.
C.1 Calibration Image & Procedure

We determined the wavelength calibration factor (in A/pixel) and height calibration factor (in
mm/pixel) by placing a mercury (Hg) spectral lamp in the target region (i.e., the region circled in
Fig. 4.1). The mercury spectrum has a number of lines across the visible spectrum, and by tuning
the spectrometer on two closely spaced Hg lines we determined the wavelength calibration factor.
We selected lines that were between H- and H-, and made the assumption that the dispersion
efficiency (i.e., the dispersion of incoming light in A/pixel) was the same across the measuring
range of the spectrometer. Because the lamp was configured as a bent tube, two portions of the
mercury lamp discharge tube that are separated by a known distance (the separation distance
between portions of the tube) were used to calibrate vertical spatial resolution. The instrument
broadening was determined using a hydrogen spectral lamp, and used to establish a baseline to
subtract from broadening due to the plasma density alone.
The calibration image of the Hg spectral lamp is shown in Fig. C.1. The vertical separation of
the two spots on an individual spectral line is due to the vertical separation of the two emissive
portions of the lamp tube, which are separated by 4.95 mm. The horizontal distance is determined
from the vacuum wavelengths of the adjacent mercury lines, located at 434.7 nm and 435.8 nm,
respectively. From these values, along with the pixel counts in the horizontal and vertical directions,
we obtain the calibrated per-pixel values of 0.125 A/pixel and 0.16 mm/pixel, respectively.
173
174 APPENDIX C. IMAGING SPECTROMETER ADDENDUM
87 pixels
31 pixels
4.95 mm
435.8 nm 434.7 nm
Figure C.1: Calibration image of the mercury spectral lamp, which provided both the spatial and
spectral resolution (0.16 mm/pixel and 0.125 A/pixel).
7500
data
7000 linear fit
6500
Peak Current [A]
6000
5500
5000
4500
4000
3500
3000
2000 3000 4000 5000 6000 7000 8000
Charging Voltage [V]
Figure C.2: Plot of peak measured current agains the measured charging voltage, used to correct
the current measurements collected during the experiments performed in Ch. 4.
C.2. H- BROADENING DATA 175
Additionally, the current probe used for the experiments conducted in Ch. 4 was found to be
mis-calibrated, after the experiments had been performed, by a factor of approximately 10. How-
ever, the voltage probe that measured the charging voltage associated with each firing of the gun
was properly calibrated. It was also found that the time response of the defective probe was not
impaired, but merely the amplitude. Therefore, the current traces were simply scaled according to
the correlation shown in Fig. C.2 in order to obtain the correct current values for the scaling analyses
performed in Ch. 4.
C.2 H- Broadening Data

Each data point includes the raw spectroscopic image, Abel inverted image, a selection of spectral
slices and accompanying Voigt fits, and the plasma density profiles obtained from the inverted data.
The source data is representative of the mean intensity over three successive trials.
Note that in some cases, the sets of three trials were performed over the course of multiple
campaigns; therefore, for certain charging voltages (i.e., operating points), multiple data points
exist and are arranged below.
Analysis for 3kV, 3 trial avg

0 0
5 5
Radial position [mm]

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
Wavelength [nm] Wavelength [nm]
1.0
intensity Gaussian
lineshape at r = 6.4 mm Modified Exponential
intensity
Plasma density, ne [m3 ]
0.8
10 23
Normalized Intensity
lineshape at r = 12.8 mm
0.6
10 22
0.4
10 21
0.2
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14
Wavelength [nm] Radial position [mm]

0 0
5 5
Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
lineshape at r = 6.4 mm 10 24 Modified Exponential
intensity
0.8
10 23
0.6
0.4 10 22
0.2 10 21
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14 16
Analysis for 5-5kV, 3 trial avg

0 0
5 5

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity 10 23

0.8
0.6
10 22
0.4
10 21
0.2
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5
Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity
0.8
10 23
0.6
10 22
0.4
10 21
0.2
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0 10 25
intensity Gaussian
intensity
0.8 10 24
0.6 10 23
0.4 10 22
0.2 10 21
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5
Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity
0.8
0.6 10 22
0.4
10 21
0.2
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity

0.8 10 23
0.6
10 22
0.4
10 21
0.2
0.0 10 20 0
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5
Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0 10 24
intensity Gaussian
intensity
0.8
lineshape at r = 12.8 mm 10 23
0.6
10 22
0.4
10 21
0.2
0.0 10 20 0
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity
0.8
0.6 10 22
0.4
10 21
0.2
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5
Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity 10 24 Gaussian
intensity
0.8
10 23
0.6
10 22
0.4
10 21
0.2
0.0 10 20 0
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity

0.8
10 23
0.6
0.4 10 22
0.2 10 21
0.0 10 20 0
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14

0 0
5 5
Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity
0.8
10 23
0.6
10 22
0.4
0.2 10 21
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14 16

0 0
5 5

Y position [mm]
10 10
15 15
20 20
655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8
1.0
intensity Gaussian
intensity
0.8
10 23
0.6
10 22
0.4
10 21
0.2
0.0 10 20
655.2 655.4 655.6 655.8 656.0 656.2 656.4 656.6 656.8 657.0 2 4 6 8 10 12 14 16
C.3 Spectrometer Analysis Libraries
1 # ********************************************************
2 # ABEL INVERSION CODE
3 # ********************************************************
4 # Revision History:
5 # 1. Version 1.0 (7/21/2015): Initial Implementation
6 # ******************* LOAD PACKAGES ********************
7 from matplotlib.pylab import *
8 import matplotlib.pyplot as plt
9
10 import matplotlib.patches as patches

11 from mpl toolkits.axes grid1.inset locator import inset axes, zoomed inset axes
12 from mpl toolkits.axes grid1.anchored artists import AnchoredSizeBar
13 from matplotlib.colors import LogNorm
14 import numpy as np
15 from scipy.optimize import curve fit
16 from scipy import interpolate, constants, integrate
17
18
19 from scipy.ndimage import interpolation

C.3. SPECTROMETER ANALYSIS LIBRARIES 183
20 from scipy.signal import savgol filter

21
22 from lmfit.models import VoigtModel, PolynomialModel, GaussianModel

23 import Equilibrium Code j func as EQ code
24
25 import os
26 import skimage
27 from skimage import io
28
29
30 # Define Gaussian Fitting Function

31 def Gauss(x,sigma,a,b,d,x0):
32 return d*x + b + a*exp((xx0)**(2.)/(2*sigma**(2.)))
33
34 def Baseline(x,a,b):
35 return a*x + b
36
37 # list of density data files

38 exp dirs = {"1016 Line BroadeningB" : "Op" , "1030 Line BroadeningB" : "Op"}
39 # other: {"Gun Experiment 8 28 2015" : "kV", "Gun Experiment 9 4 2015B" : "kV"}
40 exp dir = exp dirs.keys()[0]
41 parent dir = os.getcwd()+"/" + exp dir+"/"
42
43 m pixel = 0.04/250.0 # meters per pixel (from lamp cal)

44
45 # initialize data structures for variables to store

46 # gaussian fit
47 gaus hwhms = []
48 gaus line densities = []
49 gaus peak temps = []
50 gaus mean temps = []
51 gaus peak densities = []
52
53 # exponential fit w/offset

54 exp offset hwhms = []
55 exp offset line densities = []
56 exp offset peak temps = []
57 exp offset mean temps = []
58 exp offset peak densities = []
59
60 # exponential fit w/offset

61 exp hwhms = []
62 exp line densities = []
63 exp peak temps = []
64 exp mean temps = []
65 exp peak densities = []
66
67 #integrated values
68 set avg dens = []
69 set avg temp = []
70
71 for dirName,subDirList,fileList in os.walk(parent dir):

72
73 if exp dirs[exp dir] in dirName:

74 print "Starting " + dirName.split('/')[2] + ", " + dirName.split('/')[1] ...
+ " ..."
75 sys.stdout.flush()
76
77 images = []
78
79 for fname in fileList:

80 if 'kV' in fname and "pdf" not in fname:
81 temp = skimage.img as float(io.imread(dirName + "/" + fname))
82 param list = fname.split(' ')
83 trial str = param list[1]
84 op pt = param list[2]
85 images.append((temp, trial str, op pt))
86
87 image = np.zeros like(images[0][0])

88
89 for image package in images:

90 image = image+image package[0]
91
92 image = image / (1.0*len(images))

93
94 for k in np.arange(1):
95 image = image
96
97 start strip = 410 #490

98 stop strip = 550
99
100 raw image = image[:,start strip:stop strip]

101
102 theta = 0.9

103 raw image = interpolation.rotate(raw image,theta)
104 y center avg = np.sum(raw image, axis=1)
105 y center = np.argmax(y center avg)
106 x center avg = np.sum(raw image, axis=0)
107 x center = np.argmax(x center avg)
108
109
110 Intensity Matrix = raw image

111 top stop = np.array([])
112 n=0
113 while top stop.size == 0:

114 top stop = np.where(Intensity Matrix[:,n] > 0)[0]
115 n = n+1
116 Intensity Matrix = Intensity Matrix[top stop[0]:,:] # remove rotation ...
cells from top
117 right stop = np.array([])
118 n=0
119 while right stop.size == 0:
120 right stop = np.where(Intensity Matrix[n,:] > 0)[0]
121 n = n+1
122 Intensity Matrix = Intensity Matrix[:,:right stop[1]]
123 left stop = np.array([])
124 n=1
125 while left stop.size == 0:
126 left stop = np.where(Intensity Matrix[n,:] > 0)[0]
127 n = n+1
128 Intensity Matrix = Intensity Matrix[:,left stop[0]:]
129 bot stop = np.array([])
130 n=1
131 while bot stop.size == 0:
132 bot stop = np.where(Intensity Matrix[:,n] > 0)[0]
133 n = n+1
134 Intensity Matrix = Intensity Matrix[:bot stop[1],:]
135
136 ## get the midline point and line center guesses if not in folder
137 if 'Line params avg image.txt' not in fileList:
138 fig1 = plt.figure()
139 plt.imshow(Intensity Matrix)
140 plt.title("Click the center of the H alpha line (on the right)")
141 #sys.stdout.flush()
142 pts = fig1.ginput(1,timeout=0)
143 midpoint = int(np.round(pts[0][1]))
144 top half Int = np.flipud(Intensity Matrix[:midpoint,:])
145
146 plt.close(fig1)
147

149 plt.plot(np.sum(top half Int, axis=0))
150 plt.title("Click the peaks of the three lines, from left to right")
151 #sys.stdout.flush()
152 pts = fig2.ginput(3, timeout=0)
153 feII center guess = int(np.round(pts[0][0]))
154 feI center guess = int(np.round(pts[1][0]))
155 Halpha center guess = int(np.round(pts[2][0]))
156
157 plt.close(fig2)
158

160 plt.imshow(Intensity Matrix)
161 plt.title("Click the upper, then lower bounds of the central jet ...
region")
162 pts = fig3.ginput(2, timeout=0)
163 top = int(np.round(pts[0][1]))
164 bottom = int(np.round(pts[1][1]))
165
166 plt.close(fig3)
167
168 np.savetxt(dirName + "/"+'Line params avg image.txt', \

169 np.array([midpoint, feII center guess, feI center guess, \
170 Halpha center guess, top, bottom]))
171 else:
172 pts = np.loadtxt(dirName + "/" + 'Line params avg image.txt', ...
dtype=np.int16)
173 midpoint = pts[0]
174 feII center guess = pts[1]
175 feI center guess = pts[2]
176 Halpha center guess = pts[3]
177 top = pts[4]
178 bottom = pts[5]
179
180 top half Int = np.flipud(Intensity Matrix[top:midpoint,:]) # grab top ...

half only
181 bot half Int = Intensity Matrix[midpoint:12,:]
182
183 integrate segment = Intensity Matrix[top:bottom,:]

184 integ spectrum = np.sum(integrate segment, axis=0)
185 bgnd segment = Intensity Matrix[:20,:]
186 bgnd spectrum = np.sum(bgnd segment, axis = 0)
187
188 # estimate the bg density

189
190 line sum = bgnd spectrum

191 y = 1.*line sum
192 x = np.arange(line sum.shape[0])*1.
193
194 ix1 = index of(x, 12)

197
198 pol mod = PolynomialModel(2,prefix='pol ')

199 pars = pol mod.guess(y[:ix1], x=x[:ix1])
200
201 voigt1 = GaussianModel(prefix='feII ')

202 pars.update( voigt1.make params())
203
204 voigt2 = GaussianModel(prefix='feI ')

206
207 voigt3 = VoigtModel(prefix='Halpha ')

209
210 voigt1.guess(y[ix1:ix2],x=x[ix1:ix2])
211 pars['feII center'].set(feII center guess, min=feII center guess7, ...
max=feII center guess+7)
212 voigt2.guess(y[ix2:ix3], x=x[ix2:ix3])
213 pars['feI center'].set(feI center guess, min=feI center guess7, ...
max=feI center guess+7)
214 voigt3.guess(y[ix3:], x=x[ix3:])
215 pars['Halpha gamma'].set(value=0.7, vary=True, expr='')
216 pars['Halpha center'].set(Halpha center guess, ...
min=Halpha center guess10, max=Halpha center guess+10)
217
218 mod = voigt1 + voigt2 + voigt3 + pol mod #exp mod

219
220 out = mod.fit(y, pars, x=x)

221
222 x fine = np.linspace(x[0],x[1],1000)

223 shift y fine = mod.eval(params = out.params, x=x fine) ...
np.min(mod.eval(params = out.params, x=x fine))
224 normal y fine = shift y fine / np.max(shift y fine)
225
226 integ Lw = out.values['Halpha gamma']*2.*nm pixel

227 integ Lw err = out.params['Halpha gamma'].stderr*2.*nm pixel
228 bg density = Densify(integ Lw) # Lorentzian FWHM in nm
229 bg density err = np.min([np.abs(integ density Densify(integ Lw + ...
integ Lw err)),np.abs(integ density Densify(integ Lw ...
integ Lw err))])
230
231 integ Gw = 2.*out.values['feI sigma']*np.sqrt(2.*np.log(2.))*nm pixel

232 integ Gw err = ...
2.*out.params['feI sigma'].stderr*np.sqrt(2.*np.log(2.)) * nm pixel
233 bg temp = (integ Gw * constants.c / (2.*FeI nom *1.e9))**2. * 9.27e26 ...
/ (2.*np.log(2.))
234 bg temp err = np.abs(integ temp ((integ Gw+integ Gw err) * ...
constants.c / (2.*FeI nom *1.e9))**2. * 9.27e26 / (2.*np.log(2.)))
235
236 def filter raw image(image=top half Int):

237 filt image = np.zeros like(image)
238 (rows,cols) = image.shape
239
240 for col in np.arange(cols):

241 filt col = savgol filter(image[:,col],25,3)

242 filt image[:,col] = filt col
243
244 return filt image

245
246 # Integrate the intensity, and fit the lineshapes

247 def nestor olsen(image):
248 # this algorithm sucks
249 num r = image.shape[0]
250 num s = image.shape[1]
251 inv frame = np.zeros like(image)
252
253 k,n = np.ogrid[0:num r,0:num r]

254
255 Akn = (np.sqrt(n**2 (k1)) np.sqrt((n 1)**2 (k 1)**2)) / ...

(2*n 1)
256
257 Bkn = np.zeros like(Akn)

258
259 for k in np.arange(num r):

260 for n in np.arange(num r):
261 if k == n:
262 Bkn[k,n] = 1. * Akn[k,n]
263 elif n k + 1:
264 Bkn[k,n] = Akn[k,n1] Akn[k,n]

265
266 for i in np.arange(num s):

267 Q i = image[:,i]
268 for k in np.arange(num r):
269 summand = 0.0
270 for n in np.arange(k,num r1):
271 summand = summand + Bkn[k,n]* Q i[n]
272 f k = summand
273 inv frame[k,i] = 2./np.pi * f k
274
275 return inv frame #np.flipud(inv frame)

276
277 def manual abel(frame):

278 # this algorithm actually works
279 num r = frame.shape[0]
280 num z = frame.shape[1]
281 inv frame = np.zeros like(frame)
282
283 i,j = np.ogrid[0:num r,0:num r]

284
285 Aij = ((j+1.0)**2. i**2.)**0.5 (j**2. i**2.)**0.5 (j + 1.) ...

* \
286 np.log((j + 1. + ((j + 1.)**2. i**2.)**0.5) / (j + (j**2.0 ...

i**2.)**0.5))
287
288 Bij = ((j+1.0)**2. i**2.)**0.5 (j**2. i**2.)**0.5 j * \

289 np.log((j + 1. + ((j + 1.)**2. i**2.)**0.5) / (j + (j**2.0 ...
i**2.)**0.5))
290
291 Dij = np.zeros like(Aij)

292
293 for i in np.arange(num r):

294 for j in np.arange(num r):
295 if j < i:
296 Dij[i,j] = 0.0
297 elif j == i:
298 Dij[i,j] = 1./np.pi * Aij[i,j]
299 else:
300 Dij[i,j] = 1./np.pi * (Aij[i,j] Bij[i,j1])
301
302 for i in np.arange(num z):

303 eps j = frame[:,i]
304 summand = Dij* eps j
305 i = np.sum(summand,axis=1)
306 inv frame[:,i] = i

307
308 return 1.*inv frame

309
310 filt image = filter raw image(top half Int)

311 inverted intensity = nestor olsen(filt image)[1:2,:]
312 #inverted intensity = manual abel(top half Int)[1:,:]
313
314 # only really need to fit profiles to half the extent of the invert
315 # ed profile, line shapes reach steady state quickly
316 extent = inverted intensity.shape[0]
317
318 def index of(arrval, value):

319 "return index of array *at or below* value "
320 if value < min(arrval): return 0
321 return max(np.where(arrvalvalue)[0])
322
323 temps = np.zeros(extent)

324 densities = np.zeros(extent)
325 dens err = np.zeros(extent)
326
327 ### initialize the main figure

328 mainfig = plt.figure(figsize=(14,12))
329 mainfig.suptitle("Analysis for " + image package[2] + ", " \
330 + "3 trial avg", size=22)
331
332 slice ax = mainfig.add subplot(223)

333 slice ax.set xlabel(r"Wavelength [nm]", size = 18)
334 slice ax.set ylabel(r"Normalized Intensity", size = 18)
335
336
337 for i in np.arange(extent):

338 line sum = inverted intensity[i,:]
339 y = 1.*line sum
341

345

348
349 voigt1 = VoigtModel(prefix='feII ')

351
352 voigt2 = VoigtModel(prefix='feI ')

354

357
359 pars['feII gamma'].set(value=0.7, vary=True, expr='')
360 pars['feII center'].set(feII center guess, ...
min=feII center guess7, max=feII center guess+7)
362 pars['feI gamma'].set(value=0.7, vary=True, expr='')
367

369

371
372 # calculate nm/pixel

373 FeI nom = 656.92
374 Halpha nom = 656.28
375
376 nm pixel = 0.011666851820915406 #np.abs((FeI nom ...

Halpha nom)/(out.values['feI center'] ...
out.values['Halpha center']))
377 offset = Halpha nom out.values['Halpha center']*nm pixel # ...
anchors to center of Halpha, which is at 656.28 nm nominal
378
379 nms = offset + x*nm pixel

380 nms fine = np.linspace(nms[0],nms[1],1000)
382
383 mask = np.all([(nms fine > 656.0), (nms fine < 656.205)], axis=0)
384
385
386 # plot the triple voigt fit to the integrated lineshape

387
388 if (i%40 == 0 and i != 0):# and i 60):

389 normal y = (y np.min(y))/(np.max(y np.min(y)))
390 shift y fine = mod.eval(params = out.params, x=x fine) np.min(y)
392 slice ax.plot(nms, normal y, marker = '.', linestyle='', ...
markersize=8, label = "intensity")
393
394 fitline, = slice ax.plot(nms fine, normal y fine, ...

linestyle='', linewidth=2, label = 'lineshape at r = ' + ...
str(i*m pixel *1000.0) + " mm")
395
396 #inset ax.plot(nms fine[mask], normal y fine[mask], linewidth ...

= 2, color=fitline.get color())
397
398 Lw = out.values['Halpha gamma']*2.*nm pixel # Lorentzian FWHM in nm

399 Lw err = out.params['Halpha gamma'].stderr*2.*nm pixel # 1sigma ...
error in nm
400 Gw = 2.*out.values['Halpha sigma']*np.sqrt(2.*np.log(2.)) \
401 * nm pixel # Gaussian FWHM in nm
402
403 a param = out.values['Halpha gamma']/(np.sqrt(2) \

404 * out.values['Halpha sigma'])
405
406 if a param < 10.0:

407 print "Gettin' Gaussian at r = ", i* m pixel *1000.0, "a = ", ...
a param
408
409
410
411 def Densify(FWHM):

412 if (log10(FWHM)+3.292 > 0):
413 return ((log10(FWHM)+3.292)/0.0272)**(1./0.09014)

414 print "if"
415 else:
416 return 10**(((log10(FWHM)+21.12708)/0.99262))
417
418 densities[i] = Densify(Lw)

419 dens err[i] = np.abs(Densify(Lw+Lw err) densities[i])
420
421 temps[i] = (2.*out.values['feI sigma']*np.sqrt(2.*np.log(2.))* \

422 nm pixel * constants.c / (2.* FeI nom*1.e9))**2. * \
423 9.27e26 / (2.*np.log(2.))
424
425
426 #### analyze the integrated spectrum that was fit #####
427
428 line sum = np.sum(inverted intensity, axis=0)#integ spectrum

429 y = 1.*line sum
431

435
436

439
440 voigt1 = GaussianModel(prefix='feII ') #VoigtModel(prefix='feII ')

442
443 voigt2 = GaussianModel(prefix='feI ')#VoigtModel(prefix='feI ')

445

448
450
451 pars['feII center'].set(feII center guess, min=feII center guess7, \

452 max=feII center guess+7)
454


459

461

463

465 shift y fine = mod.eval(params = out.params, x=x fine) ...
np.min(mod.eval(params = out.params, x=x fine))
467
468 integ Lw = out.values['Halpha gamma']*2.*nm pixel

469 integ Lw err = out.params['Halpha gamma'].stderr*2.*nm pixel
470 integ density = Densify(integ Lw) # Lorentzian FWHM in nm
471 integ density err = np.min([np.abs(integ density Densify(integ Lw + ...
integ Lw err)),np.abs(integ density Densify(integ Lw ...
integ Lw err))])
472
473 integ Gw = 2.*out.values['feI sigma']*np.sqrt(2.*np.log(2.))*nm pixel

474 integ Gw err = ...
2.*out.params['feI sigma'].stderr*np.sqrt(2.*np.log(2.)) * nm pixel
475 integ temp = (integ Gw * constants.c / (2.*FeI nom*1.e9))**2. * ...
9.27e26 / (2.*np.log(2.))
476 integ temp err = np.abs(integ temp ((integ Gw+integ Gw err) * ...
constants.c / (2.*FeI nom *1.e9))**2. * 9.27e26 / (2.*np.log(2.)))
477
478 # fit the calculated density profile

479 # remove the artifacts
480 rr = (np.arange(0,densities.shape[0]) \
481 *m pixel *1000.0)[np.isnan(densities)] # radii in mm
482 dens err = dens err[np.isnan(densities)]
483 densities = densities[np.isnan(densities)]
484 rr = rr[dens err > 0.0]
485 densities = densities[dens err > 0.0]
486 dens err = dens err[dens err > 0.0]
487
488 # remove nonphysical tiny densities

489 mask good = np.where(densities > 1.e20)
490 rr good = rr[mask good]
491 densities good = densities[mask good]
492 dens err good = dens err[mask good]
493
494 good fits = np.where(densities good > dens err good)

495 rr good fit = rr good[good fits]
496 densities good fit = densities good[good fits]
497 dens err good fit = dens err good[good fits]
498
499 if 'Burnout Data.txt' not in fileList:

501 plt.errorbar(rr good, densities good, yerr = dens err good, marker ...
= 'o',linestyle='none')
502 plt.title(r"Click: R$ {burnout}$, R$ {extent}$")#, and Threshold ...
$n e$")
503 plt.yscale("log", nonposy = 'clip')
504
505 pts = fig1.ginput(2,timeout=0)

506 rburn = pts[0][0]
507 rextent = pts[1][0]
508
509 iburnt = np.argmin(np.abs(rr good rburn))

510 iextent = np.argmin(np.abs(rr good rextent))
511
512
513 plt.close(fig1)
514
515 np.savetxt(dirName + "/"+'Burnout Data.txt', \

516 np.array([iburnt, iextent]))#, ithresh]))
517 else:
518 pts = np.loadtxt(dirName + "/" + 'Burnout Data.txt', dtype=np.int16)
519 iburnt = pts[0]
520 iextent = pts[1]
521
522
523
524 # define functions for fitting the radial profile

525 def half gaus(x,a,sigma, offset):
526 return a*exp((x)**2/(2*sigma**2)) + offset
527
528 def exp decay offset(x, c, alpha, offset):

529 return c*(x)**(alpha) + offset
530
531 def line extrap(x, a, b):

532 return a*x + b
533
534 def rsquared decay(x, C, offset):

535 return C*x**(2.) + offset
536
537 def exp decay offset wrapper(x, rburnout, popt, fitflag):

538 # extrapolates a line from the edge of burnout region
539 iburnt = np.argmin(np.abs(x rburnout))
540 soln = np.zeros like(x)
541
542 if fitflag == "successful":

543 soln[iburnt:] = exp decay offset(x[iburnt:], *popt)

544 else:
545 soln[iburnt:] = rsquared decay(x[iburnt:], *popt)
546
547 a = (soln[iburnt+1] soln[iburnt])/(x[iburnt+1] x[iburnt])

548 b = soln[iburnt] a*x[iburnt]
549
550 soln[:iburnt] = line extrap(x[:iburnt], a, b)

551
552 return soln

553
554 def exp decay(x, c, alpha):

555 return c*(x)**(alpha)
556
557 rr new = np.linspace(0.01,rr good[iextent],100) # fewer samples for EQ ...

calc, increases by 10x
558
559
560 ### fit the densities to each profile

561 #Gaussian
562 p0 = [densities good[iburnt],rr good[iburnt],densities good[iextent]]
563 popt,pcov = curve fit(half gaus,rr good[iburnt:iextent], \
564 densities good[iburnt:iextent],p0=p0, maxfev = 5500, \
565 sigma=dens err good[iburnt:iextent])
566
567 density gaus bg = popt[2]

568 density gaus = half gaus(rr new, *popt)
569 density gaus err = np.sqrt(pcov[2,2])
570
571 hwhm gaus = popt[1]*np.sqrt(2.*np.log(2.))

572 hwhm gaus err = np.sqrt(pcov[1,1])*np.sqrt(2.*np.log(2.))
573
574 # spline fit to densities

575 tck = interpolate.splrep(rr good[iburnt:iextent], n)
576
577 # exp decay w/offset + linear fit

578 fitflag = "successful"
579 p0 = [densities good[iburnt],1.0 ,np.min(densities good)]
580 popt,pcov = curve fit(exp decay offset,rr good[iburnt:iextent], \
583
584 if popt[2] < 0.0 or (popt[0] < 0.0 and popt[1] < 0.0):
585 fitflag = "failed"
586 print "Exponential decay failed, switching to rsquared decay"
588 p0 = [densities good[iburnt],np.min(densities good)]
589 popt,pcov = curve fit(rsquared decay,rr good[iburnt:iextent], \

592
593 if popt[0] < 0.0:

594 print "rsquared decay failed, switching to noweight fit"
596 fitflag = "successful"
597 p0 = [densities good[iburnt],1.0 ,np.min(densities good)]
598 popt,pcov = ...
curve fit(exp decay offset,rr good[iburnt:iextent], \
599 densities good[iburnt:iextent],p0=p0, maxfev = 5500)
600
601 if popt[2] < 0.0 or (popt[0] < 0.0 and popt[1] < 0.0):
602 print "Good lord man, noweight fit failed. Switching to ...
noweight rsquared"
604 fitflag = "failed"
605 p0 = [densities good[iburnt],np.min(densities good)]
606 popt,pcov = ...
curve fit(rsquared decay,rr good[iburnt:iextent], \
607 densities good[iburnt:iextent],p0=p0, maxfev = 5500)
608
609 density exp offset bg = popt[1]

610 density exp offset = exp decay offset wrapper(rr new, rr good[iburnt], ...
popt, fitflag)
611 density exp offset err = np.sqrt(pcov[1,1])
612
613 if fitflag == "successful":

614 density exp offset percenterr = ...
np.sqrt(pcov[0,0])/exp decay offset(rr new, *popt)[1]
615 else:
616 density exp offset percenterr = ...
np.sqrt(pcov[0,0])/rsquared decay(rr new, *popt)[1]
617
618 upper hwhm = rr new[np.where(density exp offset > ...

0.5*(np.max(density exp offset)*(1.+density exp offset percenterr) ...
+ density exp offset err))][1]
619 lower hwhm = rr new[np.where(density exp offset > ...
0.5*(np.max(density exp offset)*(1.density exp offset percenterr) ...
density exp offset err))][1]
620
621 hwhm exp offset = np.mean([upper hwhm, lower hwhm])

622 hwhm exp offset err = np.abs(upper hwhm lower hwhm)/2.0
623
624 # exp decay w/o offset

625 p0 = [densities good[iburnt],1.0]
626 popt,pcov = curve fit(exp decay,rr good[iburnt:iextent], \

629
630 density exp = exp decay(rr new, *popt)

631 density exp bg = density exp[1]
632 density exp err = np.sqrt(pcov[0,0])
633
634 upper hwhm = rr new[np.where(density exp > 0.5*(np.max(density exp) + ...

density exp err))][1]
635 lower hwhm = rr new[np.where(density exp > 0.5*(np.max(density exp) ...
density exp err))][1]
636
637 hwhm exp = np.mean([upper hwhm, lower hwhm])

638 hwhm exp err = np.abs(upper hwhm lower hwhm)/2.0
639
640
641 # calculate the equilibrium profile, once for each fit

642 op str = image package[2]
643 if '' in op str:
644 op = np.float(op str[0] + '.' + op str[2])
645 else:
646 op = np.float(op str[0])
647 I = op*9.67858159e3 # kV to peak current, from calibration
648 c factor = 1.0 # estimate of fractional current in the pinch
649 I = I*0.637*c factor # avg current in sinusoidal period
650
651 n gaus, r gaus, p gaus, T gaus, B gaus = ...

EQ code.Equilibrium Profile(density gausdensity gaus bg,rr new/1000.,I)
652 # perturb by density error
653 pert n, pert r, pert p, pert T, pert B = \
654 EQ code.Equilibrium Profile(density gausdensity gaus bg \
655 +density gaus err,rr new/1000.,I)
656
657 T gaus = T gaus/11604.5221 # convert to eV

658 pert T = pert T/11604.5221
659 r gaus = r gaus *1000.0
660
661 r temps = r gaus[:np.argmin(np.abs(r gaus 10.0))] #weird shit ...

happens w/ temperature
662 T gaus = T gaus[:np.argmin(np.abs(r gaus 10.0))]
663 pert T = pert T[:np.argmin(np.abs(r gaus 10.0))]
664
665 line density gaus = ...

2*np.pi*integrate.simps(n gaus * r gaus *1000.,r gaus *1000.)
666 line density gaus err = np.abs(line density gaus ...
2*np.pi*integrate.simps(pert n * r gaus *1000.,r gaus *1000.))
667 peak temp gaus = np.max(T gaus)

668 peak temp gaus err = np.abs(peak temp gaus np.max(pert T))
669 mean temp gaus = np.mean(T gaus)
670 mean temp gaus err = np.abs(mean temp gaus np.mean(pert T))
671
672
673 modelfig = plt.figure(figsize=(14,12))

674 modelfig.suptitle("EQ calculation for " + image package[2] + ", " \
675 + "3 trial avg, Gaussian Fit", size=22)
676
677 n ax = modelfig.add subplot(221)

678 n ax.plot(r gaus, n gaus, 'b', linewidth=2)
679 n ax.set xlim(r gaus[0],r gaus[1])
680 n ax.set ylim(0.0, np.max(n gaus)*1.05)
681 n ax.set xlabel(r"Radial position [mm]", size = 18)
682 n ax.set ylabel(r"Density Profile [m${3}$]", size = 18)
683
684 p ax = modelfig.add subplot(222)

685 p ax.plot(r gaus, p gaus *1.e3, 'b', linewidth=2)
686 p ax.set xlim(r gaus[0],r gaus[1])
687 p ax.set xlabel(r"Radial position [mm]", size = 18)
688 p ax.set ylabel(r"Pressure Profile [kPa]", size = 18)
689
690 T ax = modelfig.add subplot(223)

691 T ax.plot(r temps, T gaus, 'b', linewidth=2)
692 T ax.set xlim(r temps[0],r temps[1])
693 T ax.set ylim(0.0,np.min([np.max(T gaus[np.isnan(T gaus)]), 1000.]))
694 T ax.set xlabel(r"Radial position [mm]", size = 18)
695 T ax.set ylabel(r"Temperature Profile [eV]", size = 18)
696 # write comparison temp
697 T ax.text(0.02, 0.3, 'Predicted Temp (Avg) =' + ...
"{:10.2f}".format(np.mean(T gaus[np.isnan(T gaus)])) + " eV", \
698 verticalalignment='bottom', horizontalalignment='left', \
699 transform=T ax.transAxes, \
700 fontsize=15)
701 T ax.text(0.02, 0.2, 'Measured Temp (Avg) =' + ...
"{:10.2f}".format(integ temp) + " eV", \
704 fontsize=15)
705 #T ax.cla()
706
707 B ax = modelfig.add subplot(224)

708 B ax.plot(r gaus, B gaus, 'b', linewidth=2)
709 B ax.set xlim(r gaus[0],r gaus[1])
710 B ax.set xlabel(r"Radial position [mm]", size = 18)
711 B ax.set ylabel(r"B$ {\theta}$ Profile [T]", size = 18)
712
713 plt.tight layout()

714 plt.subplots adjust(top=0.9)
715
716 ### for the exponential fit w/offset

717 n exp offset, r exp offset, p exp offset, T exp offset, B exp offset = ...
EQ code.Equilibrium Profile(density exp offset ...
density exp offset bg,rr new/1000.,I)
718 pert n, pert r, pert p, pert T, pert B = ...
EQ code.Equilibrium Profile(density exp offset ...
density exp offset bg+density exp offset err, rr new/1000.,I)
719
720 T exp offset = T exp offset/11604.5221 # convert to eV

721 pert T = pert T/11604.5221
722 r exp offset = r exp offset *1000.0
723
724 #weird shit happens w/ temperature

725 T exp offset = T exp offset[:np.argmin(np.abs(r gaus 10.0))]
727
728 line density exp offset = \

729 2*np.pi*integrate.simps(n exp offset * r exp offset \
730 *1000.,r exp offset *1000.)
731 line density exp offset err = np.abs(line density exp offset \
732 2*np.pi*integrate.simps(pert n * r exp offset *1000., \
733 r exp offset *1000.))
734 peak temp exp offset = np.max(T exp offset)
735 peak temp exp offset err = np.abs(peak temp exp offset np.max(pert T))
736 mean temp exp offset = np.mean(T exp offset)
737 mean temp exp offset err = np.abs(mean temp exp offset np.mean(pert T))
738
739 modelfig2 = plt.figure(figsize=(14,12))

740 modelfig2.suptitle("EQ calculation for " + image package[2] + ", " \
741 + "3 trial avg, Offset Exponential", size=22)
742
743 n ax = modelfig2.add subplot(221)

744 n ax.plot(r exp offset, n exp offset, 'b', linewidth=2)
745 n ax.set xlim(r exp offset[0],r exp offset[1])
746 n ax.set ylim(0.0, np.max(n exp offset)*1.05)
749
750 p ax = modelfig2.add subplot(222)

751 p ax.plot(r exp offset, p exp offset *1.e3, 'b', linewidth=2)
752 p ax.set xlim(r exp offset[0],r exp offset[1])
755
756 T ax = modelfig2.add subplot(223)

757 T ax.plot(r temps, T exp offset, 'b', linewidth=2)
759 T ax.set ylim(0.0,np.min([np.max(T exp offset[np.isnan(T exp offset)]), ...
1000.]))
"{:10.2f}".format(np.mean(T exp offset[np.isnan(T exp offset)])) ...
+ " eV", \
766 fontsize=15)
770 fontsize=15)
771 #T ax.cla()
772
773 B ax = modelfig2.add subplot(224)

774 B ax.plot(r exp offset, B exp offset, 'b', linewidth=2)
775 B ax.set xlim(r exp offset[0],r exp offset[1])
778

781
782 ### ####

783 ### exponential fit without offset ###
784 ### ####
785 n exp, r exp, p exp, T exp, B exp = ...
EQ code.Equilibrium Profile(density exp ...
density exp bg,rr new/1000.,I)
786 pert n, pert r, pert p, pert T, pert B = \
787 EQ code.Equilibrium Profile(density expdensity exp bg+ \
788 density exp err, rr new/1000.,I)
789
790 T exp = T exp/11604.5221 # convert to eV

791 pert T = pert T/11604.5221
792 r exp = r exp *1000.0
793
794 #weird shit happens w/ temperature

795 T exp = T exp[:np.argmin(np.abs(r gaus 10.0))]

797
798 line density exp = 2*np.pi*integrate.simps(n exp * r exp *1000.,r exp *1000.)
799 line density exp err = np.abs(line density exp ...
2*np.pi*integrate.simps(pert n * r exp *1000.,r exp *1000.))
800 peak temp exp = np.max(T exp)
801 peak temp exp err = np.abs(peak temp exp np.max(pert T))
802 mean temp exp = np.mean(T exp)
803 mean temp exp err = np.abs(mean temp exp np.mean(pert T))
804
805 modelfig3 = plt.figure(figsize=(14,12))

806 modelfig3.suptitle("EQ calculation for " + image package[2] + ", " \
807 + "3 trial avg, Pure Exponential", size=22)
808
809 n ax = modelfig3.add subplot(221)

810 n ax.plot(r exp, n exp, 'b', linewidth=2)
811 n ax.set xlim(r exp[0],r exp[1])
812 n ax.set ylim(0.0, np.max(n exp)*1.05)
815
816 p ax = modelfig3.add subplot(222)

817 p ax.plot(r exp, p exp *1.e3, 'b', linewidth=2)
818 p ax.set xlim(r exp[0],r exp[1])
821
822 T ax = modelfig3.add subplot(223)

823 T ax.plot(r temps, T exp, 'b', linewidth=2)
825 T ax.set ylim(0.0,np.min([np.max(T exp[np.isnan(T exp)]), 1000.]))
"{:10.2f}".format(np.mean(T exp[np.isnan(T exp)])) + " eV", \
832 fontsize=15)
836 fontsize=15)
837 #T ax.cla()
838
839 B ax = modelfig3.add subplot(224)

840 B ax.plot(r exp, B exp, 'b', linewidth=2)

841 B ax.set xlim(r exp[0],r exp[1])
844

847
848
849 # finish the main analysis plot

850 slice ax.legend(loc='best')
851 leg = slice ax.get legend()
852 ltext = leg.get texts()
853 plt.setp(ltext, fontsize='small')
854
855 rad profile ax = mainfig.add subplot(224)

856 rad profile ax.set yscale("log", nonposy='clip')
857 #rad profile ax.set xscale("log", nonposx='clip')
858 rad profile ax.scatter(rr good[iburnt:], densities good[iburnt:],
859 marker='o', color='g', s=28)#, label='calculated densities')#,
860 #yerr=[dens err[ipeak coarse:]*0.2,dens err[ipeak coarse:]])
861
862 # plot each fit

863 rad profile ax.errorbar(rr new, density gaus, yerr = ...
[np.zeros like(rr new),density gaus err*np.ones like(rr new)],
864 color = 'black', linewidth = 2, label='Gaussian')#'k', linewidth ...
= 2, label = "Gaussian fit", )
865 rad profile ax.errorbar(rr new, density exp offset, yerr = ...
[np.zeros like(rr new),density exp offset err *np.ones like(rr new)],
866 color = 'blue', linewidth = 2, label='Modified Exponential')
867 #rad profile ax.errorbar(rr new, density exp, yerr = ...
[np.zeros like(rr new),density exp err *np.ones like(rr new)],
868 #color = 'green', linewidth = 2, label='Pure Exponential')
869 rad profile ax.set xlabel(r"Radial position [mm]", size = 18)
870 rad profile ax.set ylabel(r"Plasma density, $n e$ [m${3}$]", size = 18)
871 rad profile ax.set xlim([rr[0],rr good[iextent]])
872 #imax dens = np.argmax(densities+dens err)
873 max gaus = np.max([np.max(density gaus), ...
np.max(density exp offset)])#, np.max(density exp)])
874 rad profile ax.set ylim([1.e20,5.*max gaus])
875 rad profile ax.add artist(patches.Rectangle((rr[0],1.e20),
876 rr good[iburnt1] rr[0], rad profile ax.get ylim()[1] ...
rad profile ax.get ylim()[0],
877 alpha = 0.25, facecolor = 'red'))
878 #rad profile ax.ticklabel format(style='sci', axis='y', scilimits=(0,0))
879 #rad profile ax.yaxis.major.formatter. useMathText = True
880 rad profile ax.legend(loc='best')
881 leg = rad profile ax.get legend()

882 ltext = leg.get texts()
883 plt.setp(ltext, fontsize = 'small')
884
885 ypos min, ypos max = 0.0, top half Int.shape[0]*m pixel *1000.0
886 raw image ax = mainfig.add subplot(221)
887 raw image ax.imshow(top half Int, extent = ...
[nms[0],nms[1],ypos max,ypos min], aspect = 'auto')
888 raw image ax.set xlabel(r"Wavelength [nm]", size = 18)
889 raw image ax.set ylabel(r"Y position [mm]", size = 18)
890
891 rpos min, rpos max = 0.0, inverted intensity.shape[0]*m pixel *1000.0
892 inv image ax = mainfig.add subplot(222)
893 inv image ax.imshow(inverted intensity, extent = ...
[nms[0],nms[1],rpos max,rpos min], aspect = 'auto', cmap='jet')
894 inv image ax.set xlabel(r"Wavelength [nm]", size = 18)
895 inv image ax.set ylabel(r"Radial position [mm]", size = 18)
896

899
900 plt.show()
901
902 #os.system('exit')
903 #exit
904
905 mainfig.savefig(dirName + "/" + "Averaged " # + image package[1] + ...

" " \
906 + image package[2] + " analysis.pdf")
907
908 modelfig.savefig(dirName + "/" + "Averaged " # + image package[1] + ...

" " \
909 + image package[2] + " EQcalc GaussFit.pdf")
910
911 modelfig2.savefig(dirName + "/" + "Averaged " # + image package[1] + ...

" " \
912 + image package[2] + " EQcalc OffsetExpFit.pdf")
913
914 modelfig3.savefig(dirName + "/" + "Averaged " # + image package[1] + ...

" " \
915 + image package[2] + " EQcalc PureExpFit.pdf")
916
917 plt.close('all')
918
919 ### LOAD THE DATA STRUCTURES WITH COMPUTED VALUES ###
920
921 gaus hwhms.append((I, hwhm gaus, hwhm gaus err))

922 exp offset hwhms.append((I, hwhm exp offset, hwhm exp offset err))
923 exp hwhms.append((I, hwhm exp, hwhm exp offset err))
924
925 gaus peak densities.append((I, np.max(density gaus density gaus bg), ...
density gaus err))
926 exp offset peak densities.append((I, ...
np.max(density exp offsetdensity exp offset bg), ...
density exp offset err))
927 exp peak densities.append((I, np.max(density expdensity exp bg), ...
density exp err))
928
929 gaus line densities.append((I, line density gaus, line density gaus err))
930 exp offset line densities.append((I, line density exp offset, ...
line density exp offset err))
931 exp line densities.append((I, line density exp, line density exp err))
932
933 gaus peak temps.append((I, peak temp gaus, peak temp gaus err))
934 exp offset peak temps.append((I, peak temp exp offset, ...
peak temp exp offset err))
935 exp peak temps.append((I, peak temp exp, peak temp exp err))
936
937 gaus mean temps.append((I, mean temp gaus, mean temp gaus err))
938 exp offset mean temps.append((I, mean temp exp offset, ...
mean temp exp offset err))
939 exp mean temps.append((I, mean temp exp, mean temp exp err))
940
941 set avg dens.append((I, integ density, integ density err))

942 set avg temp.append((I, integ temp, integ temp err))
943
944 print "Finished " + dirName.split('/')[2] + ", " + dirName.split('/')[1]

946
947 # save the recorded values

948 # gaussian profiles
949 np.savetxt(parent dir + "/Analyses/hwhmsgaus.txt", np.array(gaus hwhms))
950 np.savetxt(parent dir + "/Analyses/line densitiesgaus.txt", ...
np.array(gaus line densities))
951 np.savetxt(parent dir + "/Analyses/peak tempsgaus.txt", np.array(gaus peak temps))
952 np.savetxt(parent dir + "/Analyses/mean tempsgaus.txt", np.array(gaus mean temps))
953 np.savetxt(parent dir + "/Analyses/peak densitiesgaus.txt", ...
np.array(gaus peak densities))
954
955 # pure exponential fits

956 np.savetxt(parent dir + "/Analyses/hwhmsexppure.txt", np.array(exp hwhms))
957 np.savetxt(parent dir + "/Analyses/line densitiesexppure.txt", ...
np.array(exp line densities))
958 np.savetxt(parent dir + "/Analyses/peak tempsexppure.txt", np.array(exp peak temps))
959 np.savetxt(parent dir + "/Analyses/mean tempsexppure.txt", np.array(exp mean temps))

960 np.savetxt(parent dir + "/Analyses/peak densitiesexppure.txt", ...
np.array(exp peak densities))
961
962 # offset exponential fits

963 np.savetxt(parent dir + "/Analyses/hwhmsexpoff.txt", np.array(exp offset hwhms))
964 np.savetxt(parent dir + "/Analyses/line densitiesexpoff.txt", ...
np.array(exp offset line densities))
965 np.savetxt(parent dir + "/Analyses/peak tempsexpoff.txt", ...
np.array(exp offset peak temps))
966 np.savetxt(parent dir + "/Analyses/mean tempsexpoff.txt", ...
np.array(exp offset mean temps))
967 np.savetxt(parent dir + "/Analyses/peak densitiesexpoff.txt", ...
np.array(exp offset peak densities))
968
969 # integrated fits

970 np.savetxt(parent dir + "/Analyses/avg tempsinteg.txt", np.array(set avg temp))
971 np.savetxt(parent dir + "/Analyses/avg densitiesinteg.txt", np.array(set avg dens))
1 #! usr/bin/env python
2
3 ### Halpha processing script ###

4
5 # import necessary libraries

6 from matplotlib.pylab import *
8
9 import matplotlib.patches as patches

10 from mpl toolkits.axes grid1.inset locator import inset axes, zoomed inset axes
11 from mpl toolkits.axes grid1.anchored artists import AnchoredSizeBar
15 from scipy import interpolate, constants, integrate
16
17
18 from scipy.ndimage import interpolation

19 from scipy.signal import savgol filter
20 from scipy.misc import imread
21
22 from lmfit.models import VoigtModel, PolynomialModel, GaussianModel

23 import Equilibrium Code j func as EQ code
24
25 import os
26 import skimage
28
29 # data format: N trials rows, 3 columns arranged (avg current, param, std err)
30
31 # list of density data files

32 data dir = "ManualSort 112916"
33 main dir = os.getcwd()+"/" + data dir
34
35 params = {}
36
37 # functions to fit Bennet pinch

38 #/(N**(1./(gamma1.))))
39
40 def PowerLaw(xdata, C, alpha):

41 return C*xdata**(alpha)
42
43 def hwhm func(I, C, gamma):

44 return C*(I**(1./(gamma 1.)))
45
46 fileList = []
47 for (dirpath, dirnames, filenames) in os.walk(main dir):

48 fileList.extend(filenames)
49 break
50
51 for fname in fileList:

52 if "txt" in fname:
53 param name = fname.split('.')[0]
54 params[param name] = np.loadtxt(main dir + "/" + fname)
55
56 modelfig = plt.figure(figsize=(14,12))
57 modelfig.suptitle("Scaling Results", size=22)
58
59 #### TEMPERATURES ####

60 # offset exponential #
61 data = params['peak tempsexpoff']
62
63 currs = data[:,0]
64 vals = data[:,1]
65 errs = data[:,2]
66
67 curr fine = np.linspace(np.min(currs), np.max(currs), 1000)

68
69 beta guess = 1.
70 C guess = np.mean(vals)
71
72 p0 = [C guess,beta guess]
73
74 popt,pcov = curve fit(PowerLaw, currs, vals ,p0 = p0, maxfev = 5500, sigma=errs)
75
76 beta = popt[1]
77 beta err = np.sqrt(pcov[1,1])
78
79 fitline = PowerLaw(curr fine, *popt)

80 T ax = modelfig.add subplot(223)
81 T ax.errorbar(currs/1000.,vals,yerr=errs, marker = 'o', linestyle = 'none', ...
label=(r"$\beta$ = %.3f $\pm$ %.3f" % (beta,beta err)))
82 T ax.plot(curr fine/1000., fitline, 'k.') #" % (beta)))#
83 beta top = beta+beta err
84 beta bot = betabeta err
85 mean err = np.mean(np.abs(PowerLaw(currs, *popt) vals))
86 C top = fitline[0]/(curr fine[0]**beta top)
87 C bot = fitline[0]/(curr fine[0]**beta bot)
88 top err = PowerLaw(curr fine, C top, beta top) + np.sqrt(mean err)
89 bot err = PowerLaw(curr fine, C bot, beta bot) np.sqrt(mean err)
90 T ax.fill between(curr fine/1000., bot err, top err, alpha = 0.1,
91 edgecolor = 'none', facecolor = 'blue', linewidth=4,
92 linestyle='', antialiased=True)
93 T ax.set title(r"Peak Temperature, $\propto$ $I\beta$", size = 18)

94 T ax.set xlabel(r"Average Current [kA]", size = 14)
95 T ax.set ylabel(r"T [eV]", size = 14)
96 T ax.set xlim([0.9*np.min(currs/1000.),1.1*np.max(currs/1000.)])
97 T ax.set ylim([0.0,1.5*np.max(vals)])
98 T ax.legend(loc='best')
99
100 logTExp = np.array([np.log10(11604.5221*vals), np.log10(11604.5221*errs)])

101
102 # gaussian #
103
104 data = params['peak tempsgaus']

105

107 vals = data[:,1]
108 errs = data[:,2]
109
110 curr fine = np.linspace(np.min(currs), np.max(currs), 1000)

111
112 logTGaus = np.array([np.log10(11604.5221*vals), np.log10(11604.5221*errs)])

113
114 #### LINE DENSITIES ####

115 # offset exponential
116 data = params['line densitiesexpoff']
117

121
122 n ax = modelfig.add subplot(221)

123 alpha guess = 1.0
125
126 p0 = [C guess, alpha guess]

127
128 popt, pcov = curve fit(PowerLaw, currs, vals, p0 = p0, maxfev = 5500, sigma = errs)
129
130 alpha = popt[1]

131 alpha err = np.sqrt(pcov[1,1])
132

134
135 n ax.set yscale("log", nonposy='clip')

136 n ax.errorbar(currs/1000.,vals,yerr=errs, marker = 'o', linestyle = 'none', ...
label=(r"$\alpha$ = %.3f $\pm$ %.3f" % (alpha,alpha err)))
137 n ax.plot(curr fine/1000., fitline, 'k')
138 alpha top = alpha+alpha err
139 alpha bot = alphaalpha err

141 C top = fitline[0]/(curr fine[0]**alpha top)
142 C bot = fitline[0]/(curr fine[0]**alpha bot)
143 top err = PowerLaw(curr fine, C top, alpha top) + 0.1*mean err
144 bot err = PowerLaw(curr fine, C bot, alpha bot) 0.1*mean err
145 n ax.fill between(curr fine/1000., bot err, top err, alpha = 0.1,
148 n ax.set title(r"Line Density, $\propto$ $I\alpha$", size = 18)
149 n ax.set xlabel(r"Average Current [kA]", size = 14)
150 n ax.set ylabel(r"N$ e$ [m${1}$]", size = 14)
151 n ax.set xlim([0.9*np.min(currs/1000.),1.1*np.max(currs/1000.)])
152 n ax.set ylim([0.1* np.min(vals),10.*np.max(vals)])
153 n ax.legend(loc='best')
154
155
156
157
158 #### PEAK DENSITIES #####

159 data = params['peak densitiesexpoff']
160

164
165 k guess = 1.
167
168 p0 = [C guess,k guess]

169
170 popt,pcov = curve fit(PowerLaw, currs, vals ,p0 = p0, maxfev = 5500, sigma=errs)
171
172 k = popt[1]
173 k err = np.sqrt(pcov[1,1])
174

176
177 p ax = modelfig.add subplot(222)

178 p ax.set yscale("log", nonposy='clip')
179 p ax.errorbar(currs/1000.,vals,yerr=errs, marker = 'o', linestyle = 'none', ...
label=(r"$\kappa$ = %.3f $\pm$ %.3f" % (k,k err)))
180 p ax.plot(curr fine/1000., fitline, 'k')
181 k top = k+k err
182 k bot = np.max([0.0,kk err])
184 C top = fitline[0]/(curr fine[0]** k top)
185 C bot = fitline[0]/(curr fine[0]** k bot)

186 top err = PowerLaw(curr fine, C top, k top) + 0.1*mean err
187 bot err = PowerLaw(curr fine, C bot, k bot) 0.1*mean err
188 p ax.fill between(curr fine/1000., bot err, top err, alpha = 0.1,
191 p ax.set title("Peak Density, $\propto$ $I\kappa$", size = 18)
192 p ax.set xlabel(r"Average Current [kA]", size = 14)
193 p ax.set ylabel(r"n$ e$ [m${3}$]", size = 14)
194 p ax.set xlim([0.9*np.min(currs/1000.),1.1*np.max(currs/1000.)])
195 p ax.set ylim([0.1* np.min(vals),10.*np.max(vals)])
196 p ax.legend(loc='best')
197
198 logNeExp = np.array([np.log10(vals*1.e6),np.log10(errs*1.e6)])

199
200 # gaussian
201 data = params['peak densitiesgaus']
202

206
207 logNeGaus = np.array([np.log10(vals*1.e6),np.log10(errs*1.e6)])

208
209 gamma = (2. alpha) / k + 1.

210 gamma err = np.abs((2. (alpha + alpha err)) / (k + k err) + 1 gamma)
211
212
213 print "Gamma from n, N = ", gamma, " +/ ",gamma err

214
215 scale ax = modelfig.add subplot(224)

216
217 # add the parameter ovals

218 # zpinch
219 scale ax.text(16.89, 5.32, "Z Pinch", verticalalignment = 'bottom',
220 horizontalalignment = 'left', fontsize = 18, bbox=dict(fill = True, ...
linewidth=2, edgecolor='red', facecolor='white'))
221 scale ax.add artist(patches.Ellipse(xy=(16.74,5.85), width = 1.1,
222 height = 1.18, alpha = 0.25, facecolor = 'blue', edgecolor = 'red'))
223
224 #farfield
225 scale ax.text(15.1,5.5, "(farfield)", verticalalignment = 'bottom',
226 horizontalalignment = 'left', fontsize = 14)
228 height = 0.398, alpha = 0.25, facecolor = np.random.rand(3)))
229
230 # theta pinch

231 scale ax.text(17.43, 6.79, "Theta Pinch", verticalalignment = 'bottom',

235
236 # open traps

237 scale ax.text(15.09, 6.4, "Open Traps", verticalalignment = 'bottom',
241
242 # excimer laser

243 scale ax.text(15.1, 4.85, "Excimer Laser", verticalalignment = 'bottom',
247
248 # spark
249 scale ax.text(15.67, 4.58, "Spark", verticalalignment = 'bottom',
253
254 # cathode spot

255 scale ax.text(17.02, 4.65, "Cathode Spot", verticalalignment = 'bottom',
259
260 # cathode torch

261 scale ax.text(18.0, 5.2, "Cathode Torch", verticalalignment = 'bottom',
265
266 logNeAvg = np.array([np.log10(params['avg densitiesinteg'][:,1]*1.e6), \

267 np.log10(params['avg densitiesinteg'][:,2]*1.e6)])
268 logTAvg = np.array([np.log10(params['avg tempsinteg'][:,1]*11604.5221), \
269 np.log10(params['avg tempsinteg'][:,2]*11604.5221)])
270 scale ax.errorbar(logNeExp[0], logTExp[0], yerr = 0.05*logTExp[1],
271 xerr = 0.01*logNeExp[1], linestyle='none', marker = 'o',
272 color = 'green', markersize = 12, label='Scaled')
273 scale ax.errorbar(logNeAvg[0], logTAvg[0], yerr = 0.05*logTAvg[1],
274 xerr = 0.01*logNeAvg[1], linestyle='none', marker = 'o',
275 color = 'red', markersize = 12, label = 'Avg')
276
277 scale ax.set xlim([15.,19.])

278 scale ax.set ylim([4.5,7.])

279 scale ax.set xlabel(r"log$ {10}$ n$ e$ [cm${3}$]", size = 14)
280 scale ax.set ylabel(r"log$ {10}$ T$ e$ [K]", size = 14)
281 scale ax.legend(loc=5)#(18.0, 6.2))
282
283 plt.tight layout(h pad = 2.0, w pad=2.0)

285
286 modelfig.savefig(main dir + "/Scaling AnalysisExp.pdf")

287
288 plt.show()
Appendix D
Schlieren Data & Processing
This appendix is intended to provide the necessary computational tools to analyze the raw image
sequences available in the Supplementary Files hosted alongside the digital copy of this work.
D.1 Schlieren Utilities Library
1 ### schlieren utilities.py defines a number of functions for manipulating

2 # schlieren images, and is particularly designed for working with image
3 # sequences generated by the Shimadzu HPVX2 camera (i.e., stacks of 256
4 # frames)
5 import sys
6
8
9 import scipy
10 from scipy import fftpack
12 from scipy.optimize import fsolve
13
14 import skimage
16 from skimage import exposure
17 from skimage.restoration import denoise tv chambolle
18 from skimage.filters.rank import equalize, mean bilateral, median
19 from skimage.filters import gaussian
20 from skimage.morphology import disk, rectangle, dilation, watershed
21
22 import pywt
23
213
214 APPENDIX D. SCHLIEREN DATA & PROCESSING
24 import abel
25
26 from mayavi import mlab

27 mlab.options.backend = 'envisage'
28

30 from matplotlib import animation
32 plt.rcParams['animation.ffmpeg path'] = '/opt/local/bin/ffmpeg'
33
34
35 def get stack(path, trial, num = 256, Nstart=1, Nstop=256):

36 """ Returns a (M x N) x num array of raw images, where M and N are determi
37 ned by the shape of the images in the folder. 'num' is 256 by default,
38 and Nstart and Nstop (the first and last desired frame, respectively)
39 can also be specified.
40 """
41 stack = []
42
43 for i in np.arange(Nstart,Nstop+1):
44 image = skimage.img as float(io.imread(path+trial+"/"+trial+" " + \
45 str(i).zfill(3)+".tiff"))
46 stack.append(image)
47
48 return np.asarray(stack)
49
50 def plot frame(img, title="frame",flip=False, cmap='gray'):

51 """ Plots an individual image and displays it. If flip==True, the image will
52 be flipped horizontally.
53 """
54 plt.figure()
55 plt.title(title)
56 if flip:
57 plt.imshow(np.fliplr(img), cmap=cmap)
58 else:
59 plt.imshow(img,cmap=cmap)
60 plt.show()
61
62 def extract diams(stack):

63 """ Using manual user checks, iterate over a series of frames and extract
64 the pinch diameter from a Schlieren image."""
65 diams = []
66
67 def pick x loc():

68 print "Click on the axial location of desired diameter measurement."
70 plt.figure()
D.1. SCHLIEREN UTILITIES LIBRARY 215
71 plt.imshow(stack[0], cmap='gray')
72 xloc = int(np.round(plt.ginput(1,timeout=0)[0][0]))
73 plt.axvline(x=xloc)
74 plt.show()
75 return xloc
76
77 def calc diam(xloc, frame):

78 fig, (ax slice, ax frame) = plt.subplots(2, 1, gridspec kw = ...
{'height ratios':[1, 3]})
79 frame slice = frame[75:185,xloc]
80 maxloc = np.argmax(frame slice)
81 minloc = np.argmin(frame slice)
82
83 ax slice.plot(frame slice)
84 ax slice.plot(maxloc,frame slice[maxloc],'r+',linestyle="none")
85 ax slice.plot(minloc,frame slice[minloc],'r+',linestyle="none")
86
87 ax frame.imshow(frame[75:185,:], cmap='gray')
88 ax frame.scatter(xloc,maxloc, color='r', s=14)
89 ax frame.scatter(xloc,minloc, color='r', s=14)
90
91
92 plt.show()
93
94 uinput = raw input("Is the guess correct? (y/n):")

95
96 if uinput == 'y':
97 pass
98 else:
99 print "Click on the two points defining the diameter."
101 pts = fig.ginput(2,timeout=0)
102 maxloc = int(np.round(pts[0][1]))
103 minloc = int(np.round(pts[1][1]))
104
105 ax slice.cla()
106 ax slice.plot(frame slice)
107 ax slice.plot(maxloc, frame slice[maxloc], 'r+', markersize=22)
108 ax slice.plot(minloc, frame slice[minloc], 'r+', markersize=22)
109
110 ax frame.cla()
111 ax frame.imshow(frame[75:185,:],cmap='gray')
112 #ax frame.axvline(x = xloc)
113 ax frame.scatter(xloc,maxloc, color='r', s=14)
114 ax frame.scatter(xloc,minloc, color='r', s=14)
115 plt.show()
116
117 plt.close(fig)
118
119 return np.abs(maxloc minloc)

120
121
122 while(True):
123 xloc = pick x loc()
124
125 uinput = raw input("You selected x=" + str(xloc) + ". Is that okay? (y/n):")
126

128 print "Great! Starting sequence..."
129 plt.close()
130 break
131 elif uinput == 'n':
132 print "Ok, try again!"
133 else:
134 "Sorry, that's not an input I understand. Try again!"
135
136 for frame in stack:

137 diams.append(calc diam(xloc, frame))
138
139 return diams

140
141 def track feature(stack, find center=True, yloc=0):

142 """ takes a series of frames in a stack, and returns a list of points that
143 track a specific feature based on manual user input to each frame."""
144
145 pts = []
146
147 def pick y loc():

148 print "Click on the vertical location of the centerline."
150 plt.figure()
151 plt.imshow(stack[0], cmap='gray')
152 yloc = int(np.round(plt.ginput(1,timeout=0)[0][1]))
153 plt.axhline(y=yloc)
154 plt.show()
155 return yloc
156
157 def get user pt(frame, pts, yloc=0):

158 fig = plt.figure()
159 plt.imshow(frame, cmap='gray')
160 plt.axhline(y=yloc)
161 for pt in pts:
162 plt.scatter(int(np.round(pt[0])),int(np.round(pt[1])),s=20,color='red',marker='.')
163
164 uinput = fig.ginput(1,timeout=0)[0]

165
166 plt.show()
167
168 return uinput

169
170
171 if find center:

172 while(True):
173 yloc = pick y loc()
174
175 uinput = raw input("You selected y=" + str(yloc) + ". Is that okay? ...
(y/n):")
176

178 print "Great! Starting sequence..."
179 plt.close()
180 break
181 elif uinput == 'n':
182 print "Ok, try again!"
183 else:
184 "Sorry, that's not an input I understand. Try again!"
185
186
187 for frame in stack:

188 pts.append(get user pt(frame, pts, yloc=yloc))
189
190 pts array = np.asarray(pts)

191 for i in np.arange(len(pts)):
192 pts array[i,1] = np.abs(pts array[i,1]yloc)
193
194 return pts array

195
196
197 def eps from intensity(frame, cal curve, idr0):

198 """ returns a frame with the intensity values remapped to refraction angles,
199 based upon a calibration lens of R = 25.4 mm and f = 10 m.
200 """
201 m per pixel = 0.0127/np.shape(cal curve[idr0:])[0] # meters per pixel
202
203 def pixel to eps(pixel val):

204
205 return 0.1 * (np.argmin(np.abs(cal curve pixel val)) idr0) * m per pixel
206
207 frame to eps = np.vectorize(pixel to eps)

208
209 return frame to eps(frame)

210
211
212 def bgnd correct(stack, Nframes = 4):

213 """ Returns a backgroundcorrected stack of frames, by subtracting an
214 average of Nframes from the beginning of the stack from each image in
215 the stack.
216 """
217 N in stack = stack.shape[0]
218 bgnd stack = np.zeros like(stack)
219
220 bgnd frame = np.zeros like(stack[0])

221
222 for i in np.arange(1,Nframes+1):

223 frame = stack[i]
224 float frame = frame
225 bgnd frame = bgnd frame + float frame
226
227 bgnd frame = bgnd frame/float(Nframes)

228
229 origin val = np.mean((stack[0] bgnd frame)[200:,350:])

230
231 for i in np.arange(N in stack):

232 frame = stack[i] bgnd frame
233
234 bgnd stack[i] = frame

235
236 return bgnd stack

237
238 def lineByline FFT(frame, cutoff=15, band=20):

239 """ Applies a linebyline lowpass filter to the image in the horizontal
240 direction to remove vertical streaks.
241 """
242 filt frame = np.zeros like(frame)
243 num lines = frame.shape[0]
244
245 if cutoff + band > frame.shape[1]:

246 band = frame.shape[1]cutoff
247
248 for i in np.arange(num lines):

249 line FFT = np.fft.rfft(frame[i,:])
250 line FFT[cutoff:] = scipy.signal.medfilt(np.real(line FFT[cutoff:]), ...
kernel size=(line FFT[cutoff:].shape[0]/2)).astype("complex128")
251 filt line = np.fft.irfft(line FFT)
252 filt frame[i,:] = filt line
253
254
255 return filt frame

256
257 def lineByline med(frame, kernel size=3):

258 """ Applies a linebyline median filter to the image in the horizontal
259 direction to remove vertical streaks.
260 """
261 filt frame = np.zeros like(frame)
262 num lines = frame.shape[0]
263
264 for i in np.arange(num lines):

265 filt line = scipy.signal.medfilt(frame[i,:], kernel size=kernel size)
266
267 filt frame[i,:] = filt line

268

270
271 def bilateral filt(frame, offset):

272 """ Returns bilateral filtered image of frame
273 """
274
275 filt frame = mean bilateral(frame + offset,rectangle(1,10),s0=30,s1=30)

276

278
279 def tv filt(frame, offset):

280 """ Returns total variance filtered image of frame
281 """
282 filt frame = denoise tv chambolle(frame, weight=0.015, multichannel=False)
283

285
286 def twoD FFT(frame, orientation = 'xt'):

287 """ perform 2D FFT filtering operation on a single frame. Designed to be
288 used on frame with vertical streaks (e.g., in the xt plane)
289 """
290
291 frame fft = fftpack.fftshift(fftpack.fft2(frame))

292
293 y dim, x dim = frame fft.shape

294 y0 = y dim/2
295 x0 = x dim/2
296 #y0,x0 = frame fft.shape[0]/2 , frame fft.shape[1]/2
297
298 mask = np.ones like(np.real(frame fft))

299
300
301 # preferentially mask vertical stripes

302 if orientation == 'xt':
303 mask[y024:y0+25,:x03] = 0.0

304 mask[y024:y0+25,x0+4:] = 0.0
305
306 # add low pass filter

307 if orientation == 'xy':
308 # create mask indices
309 y,x = np.ogrid[y0:y dimy0, x0:x dimx0]
310
311 lpmask = x**2.0 + y**2.0 (x0/2.5)**2.0

312 mask[lpmask] = 0.0
313
314 # blur the mask

315
316 mask = scipy.ndimage.filters.gaussian filter(mask, 5, order=0, output=None, ...

mode='reflect', cval=0.0, truncate=4.0)
317
318 #plot frame(mask)

319
320 frame fft = frame fft*mask

321
322 return np.real(fftpack.ifft2(fftpack.ifftshift(frame fft)))

323
324 def wvlt denoise(frame,threshold):

325 """ Returns a denoised frame using thresholding of wavelet coefficients.
326 """
327 wavelet = pywt.Wavelet('rbio5.5')
328 WC = pywt.wavedec2(frame,wavelet)
329 NWC = map(lambda x: pywt.threshold(x,threshold, mode='soft'), WC)
330
331 return pywt.waverec2( NWC, wavelet)

332
333 def boost contrast(frame):

334 """ Applies histogram equalization to a single frame to improve contrast.
335 """
336
337 return skimage.exposure.equalize adapthist(frame,clip limit=0.5)

338
339 def crop stack(stack, xlim, ylim):

340 """ Returns a cropped stack of frames, cropped according to the limits in
341 xlim and ylim, which should be passed as tuples.
342 """
343
344 num frames = stack.shape[0]

345 frame template = np.zeros like(stack[0][ylim[0]:ylim[1],xlim[0]:xlim[1]])
346 cropped stack = ...
np.zeros((num frames,frame template.shape[0],frame template.shape[1]))
347
348 for i in np.arange(num frames):

349 cropped stack[i] = stack[i][ylim[0]:ylim[1],xlim[0]:xlim[1]]
350
351 return cropped stack

352
353 def filt stack(stack, filter type = ["FFT"]):

354 """ Returns a filtered stack of frames, based on the filter type(s) in the
355 list filter type.
356 """
357
358 for filt in filter type:

359 print "Filtering stack with: " + filt
360

362 fltd stack = np.zeros like(stack)
363
364 offset = np.abs(np.min(stack))

365

367 if "FFT" in filter type:
368 fltd stack[i] = lineByline FFT(stack[i], cutoff=8) # 8 is good
369 elif "2DFF Txt" in filter type:
370 fltd stack[i] = twoD FFT(stack[i])
371 elif "2DFF Txy" in filter type:
372 fltd stack[i] = twoD FFT(stack[i], orientation = 'xy')
373 elif "bilat" in filter type:
374 fltd stack[i] = bilateral filt(stack[i], offset)
375 elif "med" in filter type:
376 fltd stack[i] = lineByline med(stack[i], kernel size=21)
377 elif "wvlt" in filter type:
378 fltd stack[i] = wvlt denoise(stack[i], 0.025)
379 elif "TV" in filter type:
380 fltd stack[i] = tv filt(stack[i], offset)
381 else:
382 print "#nofilter"
383 fltd stack[i] = stack[i]
384
385 return fltd stack

386
387 def get cal curve2(par, cal name, cal bgnd name):
388
389
390 cal stack = get stack(par,cal name)

391 cal bgnd = get stack(par,cal bgnd name)
392 cal stack = crop stack(cal stack, xlim, ylim)
393 cal stack = twist xt(cal stack)
394 filts = ["2DFF Txt"]
395 cal stack = filt stack(cal stack, filter type = filts)

396 cal stack = twist xt(cal stack)
397 filts = ["wvlt"]
398 cal stack = filt stack(cal stack, filter type = filts)
399 avg cal = np.mean(cal stack, axis=0)
400
401 return avg cal

402
403 def get cal curve(cal roi frame):

404 """ Returns a callable calibration function that maps pixel intensity values
405 to refraction angle values, after receiving a backgroundcorrected ROI
406 of a calibration image.
407 """
408
409 cal line = np.median(cal roi frame, axis=1)

410
411 def logistic(x, A, K, B, nu, Q, C):

412 y = A + (K A) / ((C + Q*np.exp(B*x))**(1./nu))
413 return y
414
415 xdata = np.arange(cal line.shape[0])

416 ydata = cal line
417
418 popt, pcov = curve fit(logistic, xdata, ydata, p0=[1.0, 0.5, 10.0, 1.0, 30.0, ...
1.0])
419 print popt
420
421 x = np.linspace(0, cal line.shape[0], 100)

422 y = logistic(x, *popt)
423
424 plt.plot(xdata, ydata, 'o', label='data')

425 plt.plot(x,y, label='fit')
426 plt.show()
427
428 def twist xt(stack):

429 """ Returns the frame stack, transposed such that each frame is now an xt
430 slice of the 3D image cube. Also can rotate the cube back to the origin
431 al orientation.
432 """
433 return np.transpose(stack, (1, 0, 2))
434
435 def xt filt stack(stack, filter type = ["med"]):

436 """ Returns a filtered stack of frames, based on the filter type(s) in the
437 list filter type, with the filter applied to the xt plane of the image
438 cube.
439 """
440
441 for filt in filter type:

442 print "Filtering stack with: " + filt
443
444 twst stack = np.transpose(stack, (1, 0, 2))

445 num frames = twst stack.shape[0]
446 fltd stack = np.zeros like(twst stack)
447

449 if "FFT" in filter type:
450 fltd stack[i] = lineByline FFT(twst stack[i], cutoff=21)
451 elif "bilat" in filter type:
452 fltd stack[i] = bilateral filt(twst stack[i])
453 elif "med" in filter type:
454 fltd stack[i] = lineByline med(twst stack[i], kernel size=21)
455 else:
456 print "#nofilter"
457 fltd stack[i] = stack[i]
458
459 return np.transpose(fltd stack, (1, 0, 2))

460
461 def bgnd scale(stack):

462 """ Returns a background corrected stack of frames, using the mean of the
463 upper righthand corner of the frame as the scaling anchor for the center
464 of the distribution, relative to the minimum from among the stack of
465 frames.
466 """
467

469 corr stack = np.zeros like(stack)
470
471 # calculate minimum bgnd intensity

472 min lvl = [1.0,0]
473

475 bgnd lvl = np.mean(stack[i][:10,200:])
476 if bgnd lvl < min lvl[0]:
477 min lvl[0] = bgnd lvl
478 min lvl[1] = i
479

481 bgnd lvl = np.mean(stack[i][:10,200:])
482 offset = bgnd lvl min lvl[0]
483 if offset > 0.0:
484 corr stack[i] = stack[i] offset
485 else:
486 corr stack[i] = stack[i] + offset
487
488 return corr stack

489
490 def intensity (stack):

491 """ Returns a stack of frames of length N1, each frame the intensity
492 difference between the Nth and N+1 frame.
493 """
494

496 stack = np.zeros((num frames1,stack[0].shape[0], stack[0].shape[1]))
497
498 for i in np.arange(num frames1):

499 stack[i] = stack[i+1]stack[i]
500
501 return stack

502
503 def remap stack(stack):

504 """ Returns a stack with values remapped to account for asymmetry between
505 negative and positive ranges of the Schlieren signal.[DEPRECATED]
506 """
507
508 maxval = np.max(stack)

509 minval = np.min(stack)
510 newstack = np.copy(stack)
511
512 for frame in newstack:

513 posmask = frame > 0.0
514 negmask = frame 0.0
515
516 frame[posmask] = frame[posmask]/maxval

517 frame[negmask] = frame[negmask]/minval
518
519 return newstack

520
521 def symmetrize stack(stack):

522
523 numframes = stack.shape[0]

524 symm stack = np.zeros like(stack)
525
526 for i in np.arange(numframes):

527 quads = abel.tools.symmetry.get image quadrants(stack[i], symmetry axis=1, ...
symmetrize method=u'fourier')
528 symm stack[i] = ...
abel.tools.symmetry.put image quadrants(quads,symm stack[0].shape, ...
symmetry axis=1)
529
530 return symm stack

531
532
533
534 def mean fluct(stack,Nstart,Nstop):

535 """ Returns a stack of frames of length endstart, each frame the intensity
536 of the fluctuation from the mean intensity over frames start:end.
537 """
538
539 # compute avg

540 mean frame = np.mean(stack[Nstart:Nstop+1],axis=0)
541
542 fluct stack = stack mean frame

543
544 return fluct stack

545
546 def plot sequence(stack, frames, fname, ftype=".pdf"):

547 """ Receives a list of frames in 'frames', then plots the sequence and saves
548 the plotted sequence to a file.
549 """
550 ncols,nrows = frames.shape
551 fig, axs = plt.subplots(nrows=nrows,ncols=ncols)
552 cmap = 'gray'
553 minval = np.min(stack[frames.flatten()])
554 maxval = np.max(stack[frames.flatten()])
555
556 for i in np.arange(ncols):

557 for j in np.arange(nrows):
558 cropframe = stack[frames[i,j],75:175,:]
559 y,x = cropframe.shape
560
561 axs[j,i].set frame on(False)

562 axs[j,i].get xaxis().set visible(False)
563 axs[j,i].get yaxis().set visible(False)
564
565 axs[j,i].imshow(np.fliplr(stack[frames[i,j],75:175,:]), cmap=cmap, ...

clim=(minval,maxval))
566 t = frames[i,j]*0.1
567 axs[j,i].text(x/10, y/6, r"t = " + str(t) + r" $\mu$s", fontsize=8, ...
color='white')
568
569 fig.subplots adjust(left=0.2, bottom=0, right=0.85, top=1, wspace=0.02, ...

hspace=0.01)
570
571 fig.savefig(fname + ftype, bbox inches = 'tight', pad inches=0.0, dpi=300)

572
573
574
575 def make video(stack, fname, Nstart=1, Nstop=256, cmap='gray'):

576 """ Converts a stack of images to a video and saves it to fname.

577 """
578
579 num frames = (NstopNstart) + 1

580
581 fig = plt.figure()

582 ax = plt.axes() # for subimage size
583 fig.subplots adjust(left=0, bottom=0, right=1, top=1, wspace=None, hspace=None)
584 ax.set frame on(False)
585 ax.get xaxis().set visible(False)
586 ax.get yaxis().set visible(False)
587 temp frame = stack[Nstart1]
588 im=plt.imshow(np.fliplr(temp frame),cmap = cmap, ...
clim=(np.min(stack),np.max(stack)))#, clim=(0.05,1.0)) #norm = ...
LogNorm(vmin=0.075, vmax=1.0))#norm = LogNorm(vmin = 0.1, vmax = 0.9)) ...
#clim=(np.min(stack),np.max(stack))
589 y,x = temp frame.shape
590 time text = plt.text(x/10, y/5, r"t = " + str((Nstart*0.1)) + r" $\mu$s", ...
fontsize=24, color='white')
591
592 def init():

593 im.set data(np.fliplr(temp frame))
594 time text.set text(r"frame = 000")
595 return [im,time text]
596
597 def animate(i):

598 a=im.get array()
599 frame = stack[Nstart+i1]
600
601 a = np.fliplr(frame) #filt frame

602 im.set array(a)
603
604 t = (Nstart+i)*0.1
605 time text.set text(r"t = " + str(t) + r" $\mu$s")
606 return [im, time text]
607
608 anim = animation.FuncAnimation(fig, animate, init func=init, \

609 frames=num frames, interval=1, blit=True)
610
611 FFwriter = animation.FFMpegWriter(fps=num frames/20, bitrate=4000)

612 anim.save(fname+'.mp4',writer = FFwriter, extra args=['vcodec', 'libx264'])
613
614 def save stack(stack, dest):

615

618 scipy.misc.imsave(dest + 'frame' + str(i).zfill(3)+".tiff", stack[i])
619
620
621 if name == " main ":

622
623 # manual measurement: pixel/mm = 4.212

624
625 # example script:

626 # parent directory
627 par = "2016 04 20/TIFF16/" #"2016 04 19/TIFF16/" #"2016 04 20/TIFF16/" ...
#"2016 4 13/TIFF16/"# "2016 04 13/TIFF16/"#
628
629 # shot directories

630 shotname = "85kV emission nd4"
631 calname = "knife cal"
632 calbgnd = "knife bgnd"
633
634 raw stack = get stack(par,shotname)

635
636
637 corr stack = bgnd correct(raw stack)

638
639 xlim = (75,325)#(5,319)

640 ylim = (0,251)#(50,210)
641
642 crp stack = crop stack(corr stack, xlim, ylim)

643
644
645 twstd stack = twist xt(crp stack)

646
647 filts = ["2DFF Txt"]

648 fltd stack = filt stack(twstd stack, filter type = filts)
649 #
650 fltd stack = twist xt(fltd stack)
651
652
653
654 filts = ["wvlt"]

655 fltd stack = filt stack(fltd stack, filter type = filts)
656
657
658 knots = np.array([0.17457974, 0.00054841, 0.46632542]) # min, bgnd, max ...

from pinch region in fltd stack[20:100,105:155,205:]
659 knots corr = knots + np.abs(np.min(knots))
660 rr = [0.0,40.0,80.0]
661
662 def func1(x, a, b, c):

663 return a*np.exp(b*x + c) a*np.exp(c)
664
665 popt, pcov = curve fit(func1, rr, knots corr, maxfev=5000, p0 = (0.02924622, ...
0.02467059, 1.26308551))
666
667 def func2(x, a, b):

668 return a*x**2 + b*x
669
670 def intensity to radius(intensity):

671 """ converts the signal intensity to radius of the calibration lens.
672 Further manipulation is required to convert the radius to an actual
673 deflection angle, but the result is proportional thereto."""
674 def equations(p):
675 x = p
676
677 return func1(x, *popt) + np.min(knots) intensity

678
679 x = fsolve(equations, (40.0))

680
681 return np.abs(40.0 x)

682
683 thresh stack = np.copy(fltd stack)

684 thresh stack[thresh stack > knots[2]] = knots[2]
685 thresh stack[thresh stack < knots[0]] = knots[0]
686
687 vec intensity to radius = np.vectorize(intensity to radius)

688 print "remapping intensities..."
689 thresh stack = vec intensity to radius(thresh stack[15:120])
690
691 def make UH symm(stack):

692

695

np.vstack([frame[:132,:],np.flipud(frame[:132,:])])[7:257,:]
699

701
702 def UH to radial(stack):

703

705 radial stack = np.zeros like(stack[:,:132,:])
706

708 radial stack[i] = np.flipud(np.fliplr(stack[i,:132,:]))
709
710 return radial stack

711
712 def make LH symm(stack):

713

716

np.pad(np.vstack([np.flipud(frame[133:,:]),frame[133:,:]]),((8,8), ...
(0,0)),mode='constant', constant values = 0.0)
720

722
723 def invert stack(stack):

724
725 newstack = np.copy(stack)

726 newstack = make UH symm(newstack)
727 invert stack = np.zeros like(stack)
730 invert stack[i] = ...
abel.transform.Transform(np.transpose(newstack[i])).transform
731
732 return invert stack

733
734 def manual abel(frame):

735
736 num r = frame.shape[0]

737 num z = frame.shape[1]
738 inv frame = np.zeros like(frame)
739
740 i,j = np.ogrid[0:num r,0:num r]

741
742 Aij = ((j+1.0)**2. i**2.)**0.5 (j**2. i**2.)**0.5 (j + 1.) * \

743 np.log((j + 1. + ((j + 1.)**2. i**2.)**0.5) / (j + (j**2.0 ...
i**2.)**0.5))
744
745 Bij = ((j+1.0)**2. i**2.)**0.5 (j**2. i**2.)**0.5 j * \

746 np.log((j + 1. + ((j + 1.)**2. i**2.)**0.5) / (j + (j**2.0 ...
i**2.)**0.5))
747
748 Dij = np.zeros like(Aij)

749
750 for i in np.arange(num r):

751 for j in np.arange(num r):

752 if j < i:
753 Dij[i,j] = 0.0
754 elif j == i:
755 Dij[i,j] = 1./np.pi * Aij[i,j]
756 else:
757 Dij[i,j] = 1./np.pi * (Aij[i,j] Bij[i,j1])
758
759 for i in np.arange(num z):

760 eps j = frame[:,i]
761 summand = Dij* eps j
762 i = np.sum(summand,axis=1)
763 inv frame[:,i] = i
764
765 return np.vstack((np.flipud(inv frame[1:,:]),inv frame[1:,:]))[7:257,:]

766
767 def abel stack(stack):

768
769 rad stack = UH to radial(stack)

770 inv stack = np.zeros like(stack)
772

774 inv stack[i] = manual abel(rad stack[i])
775
776 return inv stack

777
778 def dens from n(stack):

779
780 from scipy import constants

782 dense stack = np.zeros like(stack)
783

785 dense stack[i] = (constants.epsilon 0 * constants.m e) * (2.0 * np.pi ...
* constants.c / 636.e9)**2 * (1. stack[i]) / (constants.e)**2.0
786
787 return dense stack

788
789 def get 2dspline(frame):

790 spl = scipy.interpolate.RectBivariateSpline(np.arange(frame.shape[0]), ...
np.arange(frame.shape[1]), frame)
791
792 return spl

793
794 def get scalar field(frame):

795
796 r = frame.shape[0]
797 z = frame.shape[1]
798 x, y, z = np.mgrid[40.0:40.0:160j, 40.0:40.0:160j, 0:200:500j]
799
800 r = x/(np.cos(np.arctan(y/x)))
801 z = z
802
803 spl = get 2dspline(frame)

804
805 return spl.ev(np.abs(r),z)#[:,:,:]

806
807 def get all scalar fields(stack):

808

810 volumes = np.ndarray((numframes, 160, 160, 500))
811

813 rad frame = np.flipud(stack[i,:125,:])
814 volumes[i] = 1.0 get scalar field(rad frame)
815
816 return volumes

817
818 def plot 3D pinch(volume, view, roll, n, stackmin=None, stackptp=None):

819
820 s = volume
821 if stackmin == None:
822 stackmin = s.min()
823
824 if stackptp == None:

825 stackptp = s.ptp()
826
827 src = mlab.pipeline.scalar field(s)

828 for i in np.linspace(0.1,1.0, num=5):
829 #mlab.pipeline.iso surface(src, contours=[stackmin+i*stackptp, ], ...
opacity=i)
830 mlab.pipeline.iso surface(src, contours=[s.max()i*stackptp, ], ...
opacity=1.0 i)
831
832 mlab.view(*view)
833 mlab.roll(roll)
834 mlab.savefig("stacks/volume frames/frame" + str(n).zfill(3) + ".png")
835 mlab.show()
836 mlab.clf()
837
838
839 make video(fltd stack, "videos/" +shotname+" "+filts[0], Nstart=15, Nstop=135) ...
#, cmap='plasma')
840
841 plt.show()
Bibliography
[1] V. T. Tikhonchuk, P. Nicolai, X. Ribeyre, C. Stenz, G. Schurtz, A. Kasperczuk, T. Pisarczyk,

L. Juha, E. Krousky, K. Masek, M. Pfeifer, K. Rohlena, J. Skala, J. Ullschmied, M. Kalal,
D. Klir, J. Kravarik, P. Kubes, and P. Pisarczyk, Plasma Physics and Controlled Fusion 50,
124056 (2008).
[2] D. R. Farley, K. G. Estabrook, S. G. Glendinning, S. H. Glenzer, B. A. Remington, K. Shige-

mori, J. M. Stone, R. J. Wallace, G. B. Zimmerman, and J. A. Harte, Phys. Rev. Lett. 83,
1982 (1999).
[3] B. Smirnov, Physics of ionized gases (2008).
[4] G. Janes, J. Dotson, and T. Wilson, ELECTROSTATIC ACCELERATION OF NEU-

TRAL PLASMASMOMENTUM TRANSFER THROUGH MAGNETIC FIELDS , Tech.
Rep. (1962).
[5] N. MacDonald, K. Loebner, and M. Cappelli, in APS Gaseous Electronics Conference (2011).
[6] C. V. Young, A. Lucca Fabris, and M. A. Cappelli, Applied Physics Letters 106, 044102
(2015).
[7] T. E. Markusic, Current sheet canting in pulsed electromagnetic accelerators, Ph.D. thesis,
Princeton University (2002).
[8] R. G. Jahn, Physics of electric propulsion (Courier Dover Publications, 1987).
[9] J. Marshall, Physics of Fluids (US) (1960).
[10] J. W. Mather, Physics of Fluids 7, S28 (1964).
[11] F. Paschen, Annalen der Physik 273, 69 (1889).
[12] F. Poehlmann, Dept. of Mechanical Engineering, Stanford University, . . . , Ph.D. thesis, Stan-
ford University (2010).
[13] D. Y. Cheng, Nuclear Fusion 10, 305 (1970).
233
234 BIBLIOGRAPHY
[14] W. J. M. Rankine, Philosophical Transactions of the Royal Society of London 160, 277 (1870).
[15] H. Hugoniot, J. Ecole Polytechnique (1889).
[16] K. Kenneth, John Willey & Sons, New York (1986).
[17] F. R. Poehlmann, M. A. Cappelli, and G. B. Rieker, Physics of plasmas 17, 123508 (2010).
[18] D. Y. Cheng, AIAA Journal 9, 1681 (1971).
[19] D. Y. Cheng and C. N. Chang, Deflagration plasma thruster , Tech. Rep. (NASA, 1984).
[20] L. Len, The snowplow and deflagration modes of operation in coaxial plasma guns, Ph.D. thesis,
University of New Mexico (1984).
[21] D. M. Woodall and L. K. Len, Journal of Applied Physics 57, 961 (1985).
[22] D. Woodall and L. Len, Journal of applied physics (1985).
[23] J. H. Degnan, W. L. Baker, S. W. R. Warren, D. W. Price, M. P. Snell, R. J. Richter-Sand,

and P. J. Turchi, Journal of Applied Physics 61, 2763 (1987).
[24] D. Black, R. Mayo, and R. Caress, Physics of Plasmas 4, 2820 (1997).
[25] S. Zhi-Gang, L. Cheng-Hui, L. Chun-Hian, and W. Cheng, Journal of Physics D: Applied

Physics 28, 2 (1995).
[26] H. Bruzzone and J. Martnez, Plasma Sources Science and Technology 10, 471 (2001).
[27] V. I. Tereshin, A. N. Bandura, O. V. Byrka, V. V. Chebotarev, I. E. Garkusha, I. Landman,

V. A. Makhlaj, I. M. Neklyudov, D. G. Solyakov, and A. V. Tsarenko, Plasma Physics and
Controlled Fusion 49, A231 (2007).
[28] K. Loebner, F. Poehlmann, and M. Cappelli, in Bulletin of the American Physical Society,
Vol. 57 (American Physical Society, 2012).
[29] K. Loebner, B. Wang, and M. Cappelli, in Bulletin of the American Physical Society, Vol. 59
(American Physical Society, 2014).
[30] K. T. K. Loebner, B. C. Wang, F. R. Poehlmann, Y. Watanabe, and M. A. Cappelli, IEEE

Transactions on Plasma Science 42, 2500 (2014).
[31] K. T. K. Loebner, T. C. Underwood, B. C. Wang, and M. A. Cappelli, in Proceedings of the

2015 IEEE 26th Symposium on Fusion Engineering (SOFE) (IEEE, Austin, TX, 2015).
BIBLIOGRAPHY 235
[32] K. T. K. Loebner, T. C. Underwood, A. L. Fabris, M. A. Cappelli, and J. J. Szabo, in Joint

Conference of the 30th International Symposium on Space Technology and Science, 34th In-
ternational Electric Propulsion Conference and 6th Nano-satellite symposium (Hyogo-Kobe,
Japan, 2015).
[33] K. Loebner, T. Underwood, T. Mouratidis, and M. Cappelli, in Bulletin of the American

Physical Society, Vol. 60 (American Physical Society, 2015).
[34] T. Underwood, K. Loebner, and M. Cappelli, in Bulletin of the American Physical Society,
Vol. 60 (American Physical Society, 2015).
[35] K. T. K. Loebner, T. C. Underwood, A. Lucca-Fabris, M. A. Cappelli, and J. J. Szabo, in 34th

International Electric Propulsion Conference (Hyogo-Kobe, Japan, 2015).
[36] K. T. K. Loebner, T. C. Underwood, and M. A. Cappelli, Physical Review Letters 115, 175001
(2015).
[37] K. T. K. Loebner, T. C. Underwood, and M. A. Cappelli, Review of Scientific Instruments 86,

063503 (2015).
[38] K. T. K. Loebner, T. C. Underwood, B. C. Wang, and M. A. Cappelli, IEEE Transactions on

Plasma Science , 1 (2016).
[39] K. Loebner, T. Underwood, and M. Cappelli, in APS Division of Plasma Physics Meeting
2016 (2016) p. abstract #NO8.012.
[40] K. Loebner, V. Miller, T. Underwood, and M. Cappelli, in 69th Annual Meeting of the APS
Division of Fluid Dynamics - Gallery of Fluid Motion (American Physical Society, 2016).
[41] K. T. K. Loebner, T. C. Underwood, T. Mouratidis, and M. A. Cappelli, Applied Physics

Letters 108, 094104 (2016).
[42] T. Underwood, K. Loebner, and M. Cappelli, in APS Division of Plasma Physics Meeting
2016 (2016) p. abstract #PP10.032.
[43] T. C. Underwood, K. T. Loebner, and M. A. Cappelli, High Energy Density Physics 23, 73
(2017).
[44] J. Kriesel, R. Prohaska, and A. Fisher, Review of scientific instruments 62, 2372 (1991).
[45] C. J. Keyser, M. Dembinski, and P. K. John, Review of Scientific Instruments 51, 425 (1980).
[46] M. M. Milanese, J. O. Pouzo, O. D. Cortazar, and R. L. Moroso, Review of Scientific Instru-

ments 77, 036106 (2006).
236 BIBLIOGRAPHY
[47] F. Poehlmann, N. Gascon, and M. Cappelli, AIAA Joint Propulsion . . . , 1 (2007).
[48] F. Poehlmann and N. Gascon, in AIAA Joint Propulsion . . . , July (2006) pp. 16.
[49] F. Poehlmann, N. Gascon, C. Thomas, and M. Cappelli, in 29th International Electric Propul-
sion Conference (2005) pp. 111.
[50] G. W. Sutton and A. Sherman, Engineering Magnetohydrodynamics (McGraw-Hill, 1965) p.

548.
[51] J. Contopoulos, The Astrophysical Journal 450, 616 (1995).
[52] D. L. Meier, S. Koide, and Y. Uchida, Science 291, 84 (2001).
[53] P. M. Bellan, S. You, and S. C. Hsu, Astrophysics and Space Science 298, 203 (2005).
[54] J. S. T. Ng and R. J. Noble, Physical Review Letters 96, 115006 (2006).
[55] P. B. Parks, Physical Review Letters 61, 1364 (1988).
[56] J. H. Hammer, C. W. Hartman, J. L. Eddleman, and H. S. McLean, Physical Review Letters

61, 2843 (1988).
[57] R. Raman, F. Martin, B. Quirion, M. St-Onge, J. L. Lachambre, D. Michaud, B. Sawatzky,

J. Thomas, A. Hirose, D. Hwang, N. Richard, C. Cote, G. Abel, D. Pinsonneault, J. L.
Gauvreau, B. Stansfield, R. Decoste, A. Cote, W. Zuzak, and C. Boucher, Physical Review
Letters 73, 3101 (1994).
[58] R. Raman, F. Martin, E. Haddad, M. St-Onge, G. Abel, C. Cote, N. Richard, N. Blanchard,

H. Mai, B. Quirion, J.-L. LaChambre, J.-L. Gauvreau, G. Pacher, R. DeCoste, P. Gierszewski,
D. Hwang, A. Hirose, S. Savoie, B.-J. LeBlanc, H. McLean, C. Xiao, B. Stansfield, A. Cote,
D. Michaud, and M. Chartre, Nuclear Fusion 37, 967 (1997).
[59] H. S. McLean, D. Q. Hwang, R. D. Horton, R. W. Evans, S. D. Terry, J. C. Thomas, and

R. Raman, Fusion Science and Technology 33, 252 (1998).
[60] D. Kumar and P. M. Bellan, Physical Review Letters 103, 105003 (2009).
[61] J. T. Cassibry, R. J. Cortez, S. C. Hsu, and F. D. Witherspoon, Physics of Plasmas 16, 112707
(2009).
[62] J. Cassibry, Y. Thio, and S. Wu, Physics of Plasmas 13, 053101 (2006).
[63] U. Shumlak and C. W. Hartman, Physical Review Letters 75, 3285 (1995).
[64] U. Shumlak, B. A. Nelson, R. P. Golingo, S. L. Jackson, E. A. Crawford, and D. J. Den Hartog,

Physics of Plasmas 10, 1683 (2003).
BIBLIOGRAPHY 237
[65] R. P. Golingo, U. Shumlak, and B. a. Nelson, Physics of Plasmas 12, 062505 (2005).
[66] G. Federici, C. Skinner, J. Brooks, J. Coad, C. Grisolia, A. Haasz, A. Hassanein, V. Philipps,

C. Pitcher, J. Roth, W. Wampler, and D. Whyte, Nuclear Fusion 41, 1967 (2001).
[67] S. Jung, M. Christenson, D. Curreli, C. Bryniarski, D. Andruczyk, and D. Ruzic, Fusion

Engineering and Design 89, 2822 (2014).
[68] R. Wallace, Theoretical, computational and experimental analysis of the deflagration plasma
accelerator and plasma beam characteristics, Ph.D. thesis, Virginia Polytechnic Institute and
State University (1991).
[69] J. T. Bradley, IEEE Transactions on Plasma Science 27, 1105 (1999).
[70] H. Sitaraman and L. Raja, AIAA Journal , 1 (2014).
[71] N. Gatsonis and L. Byrne, IEEE Transactions on Plasma Science 32 (2004).
[72] S. Chen and T. Sekiguchi, Journal of Applied Physics 36 (1965).
[73] E. Peterson and L. Talbot, AIAA Journal 8, 2215 (1970).
[74] J. Laframboise, in Rarefied Gas Dynamics, Volume 2, Vol. 1 (1965) p. 22.
[75] B. H. Johnson and D. Murphree, AIAA Journal 7, 2028 (1969).
[76] S.-L. Chen and T. Sekiguchi, Journal of Applied Physics 36, 2363 (1965).
[77] P. V. Storm, Thesis (PhD). STANFORD UNIVERSITY, Source DAI-B 58/02, p. 939, Aug
1997, 125 pages. (1997).
[78] H. Ehrich and D. E. Kelleher, Physical Review A 21, 319 (1980).
[79] A. Kramida, Y. Ralchenko, J. Reader, and NIST Atomic Spectra Database Team, NIST
Atomic Spectra Database (version 5.2), [Online], (2014).
[80] O. H. Nestor and H. N. Olsen, SIAM Review 2, 200 (1960).
[81] R. K. Hanson, R. M. Spearrin, and C. S. Goldenstein, Spectroscopy and Optical Diagnostics

for Gases (Springer International Publishing, Cham, 2016).
[82] J. Pollock and S. Barraclough, (1905).
[83] O. Buneman, in Plasma Physics, edited by J. Drummond (McGraw-Hill, Inc., New York, NY,
1961) p. 202.
[84] S. L. Jackson, Dissertation Abstracts International, Ph.D. thesis, University of Washington

(2006).
238 BIBLIOGRAPHY
[85] T. H. Stix, The Theory of Plasma Waves, New York: McGraw-Hill, 1962 (1962).
[86] F. F. Chen, Plasma Physics (1984).
[87] R. O. Dendy, Plasma dynamics (Clarendon Press, 1990).
[88] J. Bittencourt, Fundamentals of plasma physics (2013).
[89] A. Toepler, Annalen der Physik (1866).
[90] G. Settles, Schlieren and Shadowgraph Techniques: Visualizing Phenomenon in Transparent

Media, 2nd ed. (Springer, 2006).
[91] V. A. Miller, M. Tilghman, and R. K. Hanson, Physics of Fluids 27, 091105 (2015).
[92] P. Lee, X. Feng, G. Zhang, and M. Liu, Plasma Sources Science (1997).
[93] E. Bogolyubov, V. Bochkov, and V. Veretennikov, Physica (1998).
[94] Y. Kato and S. Be, Applied physics letters (1986).
[95] S. Lee, P. Lee, G. Zhang, and X. Feng, on Plasma Science (1998).
[96] Y. Kato, I. Ochiai, and Y. Watanabe, Journal of Vacuum (1988).
[97] E. T. Jens, V. A. Miller, and B. J. Cantwell, Experiments in Fluids 57, 39 (2016).
[98] A. Raga, H. Sobral, and M. Villagran-Muniz, Monthly Notices of (2001).
[99] R. Morrow and J. Lowke, Journal of Physics D: Applied Physics (1997).
[100] E. Robert, V. Sarron, T. Darny, and D. Ries, Plasma Sources (2014).
[101] C. Ohl, O. Lindau, and W. Lauterborn, Physical Review Letters (1998).
[102] N. Babaeva and G. Naidis, Journal of Physics D: Applied Physics (1996).
[103] B. Sands and B. Ganguly, Transactions on Plasma . . . (2008).
[104] J. Weaver, Y. Xu, and D. Healy, Magnetic Resonance in (1991).
[105] Y. Xu, J. Weaver, D. Healy, and J. Lu, IEEE transactions on image (1994).
[106] M. J. Hargather and G. S. Settles, Optics and Lasers in Engineering 50, 8 (2012).
[107] R. Lovberg, (1963).
[108] D. Kalantar and D. Hammer, Physical review letters (1993).
[109] G. Decker, R. Deutsch, and W. Kies, Plasma physics and (1985).

BIBLIOGRAPHY 239
[110] E. Choueiri, T. Markusic, J. Berkery, and J. Cooley, Physics and Dynamics of Current Sheets
in Pulsed Plasma Thrusters, Tech. Rep. (Princeton University, Princeton, 2002).
[111] J. Bradley, J. Oh, O. Olabanji, and C. Hale, on Plasma Science (2011).
[112] M. Boselli, V. Colombo, E. Ghedini, and M. Gherardi, Plasma Chemistry and (2014).
[113] R. Rubinstein and P. S. Greenberg, Applied Optics 33, 1141 (1994).
[114] A. K. Agrawal, B. W. Albers, and D. W. Griffin, Applied optics 38, 3394 (1999).
[115] H. Bloomberg and M. Lampe, Journal of Applied (1980).
[116] M. Haines, Proceedings of the Physical Society (1959).
[117] C. Deeney, T. Nash, R. Spielman, and J. Seaman, Physical Review E (1997).
[118] M. Haines, IEEE transactions on plasma science (1998).
[119] D. Ryutov, M. Derzon, and M. Matzen, Reviews of Modern Physics (2000).
[120] U. Shumlak and C. W. Hartman, Phys. Rev. Lett. 75, 3285 (1995).
[121] U. Shumlak, B. Nelson, and B. Balick, High Energy Density Laboratory Astrophysics (Springer
Netherlands, 2007) pp. 4145.
[122] U. Shumlak, R. P. Golingo, B. A. Nelson, and D. J. Den Hartog, Phys. Rev. Lett. 87, 205005
(2001).
[123] M. Liberman, J. D. Groot, A. Toor, and R. Spielman, Physics of high-density Z-pinch plasmas
(1999).
[124] R. J. Tayler, C. S, D. J. W. E, L. R, H. R, K. M. M, Schwarzschild, L. R. E, L. Rayleigh,

T. R. J, and T. G. I, Proceedings of the Physical Society. Section B 70, 31 (1957).
[125] V. D. Shafranov, Plasma Physics and the Problem of Controlled Thermonuclear Reactions,
vol. ii ed. (1959).
[126] F. Felber, Physics of Fluids (1958-1988) (1982).
[127] A. Budko, F. Felber, and A. Kleev, Physics (1989-1993) (1989).
[128] G. Federici, A. Zhitlukhin, N. Arkhipov, R. Giniyatulin, N. Klimov, I. Landman, V. Pod-

kovyrov, V. Safronov, A. Loarte, and M. Merola, Journal of Nuclear Materials 337-339, 684
(2005).
[129] I. E. Garkusha, N. I. Arkhipov, N. S. Klimov, V. A. Makhlaj, V. M. Safronov, I. Landman,

and V. I. Tereshin, Physica Scripta T138, 014054 (2009).
240 BIBLIOGRAPHY
[130] S. Kajita, N. Ohno, S. Takamura, W. Sakaguchi, and D. Nishijima, Applied Physics Letters
91, 261501 (2007).
[131] T. Hirai, J. Linke, P. Sundelin, M. Rubel, W. Kuhnlein, E. Wessel, J. P. Coad, C. P. Lungu,

G. F. Matthews, L. Pedrick, and G. Piazza, Physica Scripta T128, 166 (2007).
[132] A. W. Leonard, Physics of Plasmas 21, 090501 (2014).
[133] A. Leonard, A. Herrmann, K. Itami, J. Lingertat, A. Loarte, T. Osborne, W. Suttrop, , t. I.

D. M. Database Expert Group, and t. I. D. Physics Expert Group, Journal of Nuclear Materials
266-269, 109 (1999).
[134] A. Prinn and B. Ricketts, Journal of Physics D: Applied Physics 2026 (1972).
[135] B. Balick and A. Frank, Annual Review of Astronomy and Astrophysics 40, 439 (2002).
[136] B. Reipurth and and John Bally, Annual Review of Astronomy and Astrophysics 39, 403
(2001).
[137] K. J. Borkowski, J. M. Blondin, and J. P. Harrington, The Astrophysical Journal Letters 482,
L97 (1997).
[138] R. Jeanloz, P. M. Celliers, G. W. Collins, J. H. Eggert, K. K. M. Lee, R. S. McWilliams,

S. Brygoo, and P. Loubeyre, Proceedings of the National Academy of Sciences 104, 9172
(2007).
[139] J. M. Foster, B. H. Wilde, P. A. Rosen, R. J. R. Williams, B. E. Blue, R. F. Coker, R. P. Drake,

A. Frank, P. A. Keiter, A. M. Khokhlov, J. P. Knauer, and T. S. Perry, The Astrophysical
Journal Letters 634, L77 (2005).
[140] S. V. Lebedev, J. P. Chittenden, F. N. Beg, S. N. Bland, A. Ciardi, D. Ampleford, S. Hughes,

M. G. Haines, A. Frank, E. G. Blackman, and T. Gardiner, The Astrophysical Journal 564,
113 (2002).
[141] H. Baty and R. Keppens, The Astrophysical Journal 580, 800 (2002).
[142] S. Appl, T. Lery, and H. Baty, Astronomy and Astrophysics 355, 818 (2000).
[143] T. Lery, H. Baty, and S. Appl, Astronomy and Astrophysics 355, 1201 (2000).
[144] F. R. Poehlmann, M. A. Cappelli, and G. B. Rieker, Physics of Plasmas 17 (2010),

http://dx.doi.org/10.1063/1.3526603.
[145] P. Babington, Jets from Young Stars IV, 1st ed., 793 (Springer-Verlag Berlin Heidelberg, 2010).
BIBLIOGRAPHY 241
[146] J. I. Castor, High Energy Density Laboratory Astrophysics, edited by S. V. Lebedev (Springer
Netherlands, Dordrecht, 2007) pp. 207211.
[147] J. E. Cross, B. Reville, and G. Gregori, The Astrophysical Journal 795, 59 (2014).
[148] D. Ryutov, R. P. Drake, J. Kane, E. Liang, B. A. Remington, and W. M. Wood-Vasey, The

Astrophysical Journal 518, 821 (1999).
[149] M. Mitchner and C. Kruger, Partially Ionized Gases (John Wiley and Sons, 1992).

Physics and Applications of A Plasma Deflagration Accelerator

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Physics and Applications of A Plasma Deflagration Accelerator

Transféré par

Droits d'auteur :

Formats disponibles

PHYSICS AND APPLICATIONS OF A PLASMA DEFLAGRATION

2 Stanford Plasma Gun Facility 15

4 Exploring the Near Field 55

7 Conclusion and Future Work 151

A Experimental Facility Procedures 155

B Valve Engineering Drawings 163

C Imaging Spectrometer Addendum 173

D Schlieren Data & Processing 213

3.1 Parameters used to compute detonation and deflagration velocities. . . . . . . . . . . 52

6.1 Scales of typical astrophysical phenomena and related experiments . . . . . . . . . . 149

1.1 The parameter space of plasmas in nature . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1 Experimental Facility Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1 ICCD images of the consecutive deflagration and detonation events . . . . . . . . . . 35

4.1 Time integrated image of SPG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5.1 Example schlieren image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

6.1 SEM images of observed material damage on Si wafers . . . . . . . . . . . . . . . . . 131

C.1 Spectrometer calibration image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

1.1 Plasma Definition

1.1.1 Plasma Criteria

where 0 is the vacuum permittivity, kB is Boltzmanns constant, Te is the electron temperature, n

1.1.2 Range of Plasma Conditions

1.2 Plasma Acceleration

~ is the electric field vector. Examples of accelerators in this

In an electromagnetic accelerator, a combination of electric and magnetic forces accelerate

~ is the magnetic flux density vector. Such self-field accelerators

1.2.1 Review of Electromagnetic Plasma Guns

Figure 1.2: Schematic of an electromagnetic plasma gun

1.3 Plasma Deflagration

Figure 1.3: A premixed combustible system.

1.3.1 The Plasma Analogy

1.3.2 The Rankine-Hugoniot Theory

upper C-J point

origin strong deflagration

lower C-J point

1.3.3 Applicability of the MHD Rankine-Hugoniot Model

1.4 Dissertation Organization and Scope

Stanford Plasma Gun Facility

2.1 Facility Overview

1 Large Vacuum Tank

Figure 2.1: Labeled diagram of the experimental facility.

2.2 Vacuum System

Figure 2.2: Photograph of the straight glass tube, by F. Poehlmann.

2.3 Pulsed Power Driver

2.4 Gas Injection

2.5 Custom Valve Design

Restrictor Flange Gas Inlet

1 in. Gas Outlet

2.5.1 Test Apparatus and Procedure

Capacitor Bank Triggered Baratron

Pu Valve Probe Holders

2.5.2 Valve Characterization Results

2.6 Coaxial Plasma Gun

2.6.1 SPG Operation

2.6.2 Prior Work on the SPG Facility

'()*+,- './/-0) 1234

! " #! #" $! $"

Exploring the Far Field: Scanning

Current-driven magnetohydrodynamic (MHD) ionization waves and acceleration mechanisms are

3.1.1 Phenomenological Motivation

3.1.2 Derivation of the MHD Rankine-Hugoniot Relations

upper C-J point

origin strong deflagration

lower C-J point