Epfl TH3542

POUR L'OBTENTION DU GRADE DE DOCTEUR S SCIENCES
PAR
ingnieur physicien diplm EPF
de nationalit suisse et originaire de La Sagne (NE)
accepte sur proposition du jury:
Lausanne, EPFL
2006
Prof. R. Schaller, prsident du jury
Dr Ch. Hollenstein, directeur de thse
Prof. M. Rappaz, rapporteur
Dr G. Wlder, rapporteur
Prof. J. Winter, rapporteur
CHARACTERIZATION OF ELECTRICAL DISCHARGE
MACHINING PLASMAS
Antoine DESCOEUDRES
THSE N
O
3542 (2006)
COLE POLYTECHNIQUE FDRALE DE LAUSANNE
PRSENTE LE 9 JUIN 2006
LA FACULT SCIENCES DE BASE
Centre de Recherche en Physique des Plasmas
SECTION DE PHYSIQUE
La frousse, moi ! ... Jaime autant vous
dire, mille tonnerres ! que quand je le
rencontrerai, votre yti, a va faire des
tincelles !
le capitaine Haddock
Abstract
Electrical Discharge Machining (EDM) is a well-known machining technique since more
than fty years. Its principle is to use the eroding eect on the electrodes of successive
electric spark discharges created in a dielectric liquid. EDM is nowadays widely-used
in a large number of industrial areas. Nevertheless, few studies have been done on the
discharge itself and on the plasma created during this process. Further improvements of
EDM, especially for micro-machining, require a better control and understanding of the
discharge and of its interaction with the electrodes. In this work, the dierent phases
of the EDM process and the properties of the EDM plasma have been systematically
investigated with electrical measurements, with imaging and with time- and spatially-
resolved optical emission spectroscopy.
The pre-breakdown phase in water is characterized by the generation of numerous
small hydrogen bubbles, created by electrolysis. Since streamers propagate more easily in
a gaseous medium, these bubbles can facilitate the breakdown process. In oil, no bubbles
are observed. Therefore, the breakdown mechanism in oil could be rather enhanced by
particles present in the electrode gap. Fast pulses of current and light are simultaneously
measured during the pre-breakdown. These pulses are characteristic of the propagation of
streamers in the dielectric liquid. The pre-breakdown duration is not constant for given
discharge parameters, but distributed following a Weibull distribution. This shows that
the breakdown is of stochastic nature.
After the breakdown, the plasma develops very rapidly (< 50 ns) and then remains
stable. The plasma light is particularly intense during the rst 500 ns after the breakdown
and weaker during the rest of the discharge, depending on the current intensity. While the
gap distance is estimated to be around 10100 m, the discharge excites a broad volume
around the electrode gap, typically 200 m in diameter. This volume grows slightly during
the discharge. Vapor bubbles are generated in water and in oil by the heat released from
the plasma. At the end of the discharge, the plasma implodes and disappears quickly.
Light is still emitted after the discharge by incandescent metallic particles coming from the
erosion of the workpiece. Their temperature is measured around 2200 K, demonstrating
that they are still in a liquid state in the beginning of the post-discharge.
The spectroscopic analysis of the plasma light shows a strong H
and continuum
radiation, with many atomic metallic lines emitted by impurities coming from electrode
and workpiece materials. The EDM plasma is thus composed of species coming from
the cracking of the dielectric molecules (mainly hydrogen in the case of water and oil),
with contamination from the electrodes. The contamination is slightly higher in the
vicinity of each electrode, and the contamination from the workpiece increases during the
i
ii ABSTRACT
discharge probably due to vaporization. The electron temperature, measured from copper
line intensities with the two-line method, is found to be low. The temperature is around
0.7 eV ( 8100 K) in the whole plasma, slightly higher in the beginning of the discharge.
The electron density has been measured from Stark broadening and shift measurements
of the H
line. The density is extremely high, especially at the beginning of the discharge
(> 210
18
cm
3
during the rst microsecond). Then it decreases with time, remaining
nevertheless above 10
16
cm
3
after 50 s. During the whole discharge, the density is
slightly higher in the plasma center. The EDM plasma has such a high density because it
is formed from a liquid, and because it is constantly submitted to the pressure imposed by
the surrounding liquid. This extreme density produces spectra with strongly-broadened
spectral lines, especially the H
line, and with an important continuum. During the rst

microsecond when the density is at its maximum, spectral lines are so broadened that
they are all merged into a continuum.
The low temperature and the high density of the EDM plasma make it weakly non-
ideal. Its typical coupling parameter is indeed around 0.3, reaching 0.45 during the rst
microsecond. In this plasma, the Coulomb interactions between the charged particles
are thus of the same order as the mean thermal energy of the particles, which produces
coupling phenomena. Spectroscopic results conrm the non-ideality of the EDM plasma.
The strong broadening and shift of the H
line and its asymmetric shape and complex

structure, the absence of the H
line, and the merging of spectral lines are typical of non-

ideal plasmas. The EDM plasma has thus extreme physical properties, and the physics
involved is astonishingly complex.
Keywords: electrical discharge machining, EDM, plasma, spark, discharge, non-
ideal plasma, optical emission spectroscopy, spectroscopy, imaging, breakdown, dielectric,
liquid.
Version abrge
Llectro-rosion (ou EDM pour Electrical Discharge Machining) est une technique dusi-
nage bien connue depuis plus de cinquante ans. Son principe est dutiliser leet rosif
sur les lectrodes dtincelles lectriques successives cres dans un liquide dilectrique.
Llectro-rosion est aujourdhui trs utilise dans un grand nombre de secteurs industriels.
Nanmoins, peu dtudes ont t menes sur la dcharge elle-mme et sur le plasma cr
pendant ce processus. Les amliorations futures de llectro-rosion, en particulier pour
le micro-usinage, passent par un meilleur contrle et une meilleure comprhension de la
dcharge et de ses interactions avec les lectrodes. Dans ce travail, les direntes phases
du processus dlectro-rosion et les proprits du plasma ont t tudies de manire
systmatique, laide de mesures lectriques, dimagerie et de spectroscopie dmission
optique rsolue en temps et en espace.
La phase de pr-dcharge dans leau est caractrise par la gnration de nombreuses
petites bulles dhydrogne, cres par lectrolyse. Puisque les streamers se propagent plus
facilement dans un milieu gazeux, ces bulles peuvent faciliter le processus de claquage.
Dans lhuile, aucune bulle nest observe. Ainsi, le mcanisme de claquage dans lhuile
pourrait plutt tre facilit par des particules prsentes dans lespace inter-lectrodes.
Des impulsions rapides de courant et de lumire sont mesures simultanment durant la
pr-dcharge. Ces impulsions sont caractristiques de la propagation de streamers dans le
liquide dilectrique. La dure de la pr-dcharge nest pas constante pour des paramtres
de dcharge donns, mais elle est distribue selon une distribution de Weibull. Ceci montre
que le claquage est de nature stochastique.
Aprs le claquage, le plasma se dveloppe trs rapidement (< 50 ns) et reste ensuite
stable. La lumire du plasma est particulirement intense durant les premires 500 ns
suivant le claquage et plus faible pendant le reste de la dcharge, et dpend de lintensit du
courant. Alors que lespace inter-lectrodes est estim environ 10100 m, la dcharge
excite un large volume autour des lectrodes, dun diamtre typique de 200 m. Ce volume
crot lgrement pendant la dcharge. Des bulles de vapeur sont gnres aussi bien dans
leau que dans lhuile, d la chaleur libre par le plasma. A la n de la dcharge, le
plasma implose et disparat rapidement. De la lumire est encore mise aprs la dcharge
par des particules mtalliques incandescentes, provenant de lrosion de la pice. Leur
temprature a t mesure environ 2200 K, ce qui dmontre quelles sont toujours
ltat liquide au dbut de la post-dcharge.
Lanalyse spectroscopique de la lumire du plasma montre un forte radiation de la
ligne H
et une forte mission continue, avec la prsence de nombreuses lignes atomiques

mtalliques mises par des impurets provenant des matriaux de llectrode et de la
iii
iv VERSION ABREGEE
pice. Ainsi, le plasma dlectro-rosion est compos despces provenant de la dissociation
des molcules du dilectrique (principalement de lhydrogne dans le cas de leau et de
lhuile), avec une contamination des lectrodes. La contamination est lgrement plus
forte au voisinage de chaque lectrode, et la contamination venant de la pice augmente
au cours de la dcharge, probablement cause de son vaporation. La temprature lec-
tronique, mesure partir des intensits de lignes de cuivre avec la mthode dite two-line
method, est basse. La temprature est autour de 0.7 eV ( 8100 K) dans tout le plasma,
lgrement plus haute au dbut de la dcharge. La densit lectronique a t mesure
partir de llargissement et du dplacement par eet Stark de la ligne H
. La densit est
extrmement leve, particulirement au dbut de la dcharge (> 210
18
cm
3
durant la
premire microseconde). Elle dcrot ensuite avec le temps, restant nanmoins toujours
au-dessus de 10
16
cm
3
aprs 50 s. Pendant toute la dcharge, la densit est lgrement
plus leve au centre du plasma. Le plasma dlectro-rosion a une densit si leve car
il est form partir dun liquide, et parce quil est constamment soumis la pression
impose par le liquide environnant. Cette densit extrme produit des spectres avec
des lignes spectrales trs largies, particulirement la ligne H
, et avec une importante

radiation continue. Pendant la premire microseconde o la densit est son maximum,
les lignes spectrales sont tellement largies quelles fusionnent et ne forment quun continu.
La basse temprature et la haute densit du plasma dlectro-rosion le rendent faible-
ment non-idal. Son paramtre de couplage typique est en eet autour de 0.3, atteignant
0.45 pendant la premire microseconde. Dans ce plasma, les interactions coulombiennes
entre les particules charges sont ainsi du mme ordre que lnergie thermique moyenne
des particules, ce qui produit des phnomnes de couplage. Des rsultats de spectroscopie
conrment la non-idalit du plasma dlectro-rosion. Le fort largissement et dplace-
ment de la ligne H
ainsi que sa forme asymtrique et sa structure complexe, labsence

de la ligne H
, et la fusion des lignes spectrales sont en eet typiques des plasmas non-
idaux. Le plasma dlectro-rosion possde ainsi des proprits physiques extrmes, et
la physique y relative est tonnamment complexe.
Mots-cls: lectro-rosion, EDM, plasma, tincelle, dcharge, plasma non-idal, spec-
troscopie dmission optique, spectroscopie, imagerie, claquage, dilectrique, liquide.
Contents
Abstract i
Version abrge iii
1 Introduction 1
1.1 Electrical Discharge Machining (EDM) . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 State of the art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Purpose and structure of the work . . . . . . . . . . . . . . . . . . . . . . . 8
2 EDM plasmas : Background 11
2.1 Discharges in gases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 Spark, arc, glow & co. . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.2 Sparks and streamers . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.3 Electric arcs and cathode spots . . . . . . . . . . . . . . . . . . . . 14
2.2 Discharges in dielectric liquids . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Other similar plasmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Experimental setup and diagnostics 25
3.1 Electrical discharge machining device . . . . . . . . . . . . . . . . . . . . . 25
3.2 Electrical measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Discharge measurements . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.2 Pre-breakdown measurements . . . . . . . . . . . . . . . . . . . . . 28
3.3 Plasma light intensity measurements . . . . . . . . . . . . . . . . . . . . . 29
3.4 Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Optical emission spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5.1 Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5.2 Electron temperature measurement . . . . . . . . . . . . . . . . . . 33
3.5.3 Electron density measurement . . . . . . . . . . . . . . . . . . . . . 35
3.5.4 General experimental setup . . . . . . . . . . . . . . . . . . . . . . 37
3.5.5 Experimental setup for time-resolved spectroscopy . . . . . . . . . . 40
3.5.6 Experimental setup for spatially-resolved spectroscopy . . . . . . . 40
3.6 Experimental diculties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
v
vi CONTENTS
4 Pre-breakdown of EDM discharges 45
4.1 Bubbles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Pre-breakdown current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 Pre-breakdown duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5 Imaging diagnostics of EDM discharges 53
5.1 Imaging of the EDM process . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.1.1 Erosion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.1.2 Bubbles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.1.3 Short-circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1.4 Reproducibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 Plasma imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2.1 Evolution of the plasma light intensity . . . . . . . . . . . . . . . . 56
5.2.2 Typical plasma image . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2.3 Plasma evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.4 Eect of the discharge current . . . . . . . . . . . . . . . . . . . . . 59
5.2.5 H
emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.3 Beginning of the discharge : Fast imaging . . . . . . . . . . . . . . . . . . 61
5.4 End of the discharge and post-discharge . . . . . . . . . . . . . . . . . . . 62
6 Optical emission spectroscopy of EDM plasmas 67
6.1 Time-integrated emission spectra . . . . . . . . . . . . . . . . . . . . . . . 67
6.1.1 Typical spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.1.2 Eect of the dielectric . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.1.3 Eect of the electrode material . . . . . . . . . . . . . . . . . . . . 70
6.1.4 Eect of the electrode polarity . . . . . . . . . . . . . . . . . . . . . 72
6.1.5 Eect of the discharge current . . . . . . . . . . . . . . . . . . . . . 73
6.1.6 Eect of the discharge on-time . . . . . . . . . . . . . . . . . . . . . 74
6.1.7 First estimation of electron density and temperature . . . . . . . . 75
6.2 Time-resolved emission spectra . . . . . . . . . . . . . . . . . . . . . . . . 76
6.2.1 Merging of atomic lines . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.2.2 Evolution of the electron temperature . . . . . . . . . . . . . . . . . 76
6.2.3 Evolution of the electron density . . . . . . . . . . . . . . . . . . . 77
6.3 Spatially-resolved emission spectra . . . . . . . . . . . . . . . . . . . . . . 79
6.3.1 Asymmetry of the contamination . . . . . . . . . . . . . . . . . . . 79
6.3.2 Electron temperature prole . . . . . . . . . . . . . . . . . . . . . . 80
6.3.3 Electron density prole . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.4 Time- and spatially-resolved emission spectra . . . . . . . . . . . . . . . . 82
6.4.1 Evolution of the contamination . . . . . . . . . . . . . . . . . . . . 82
6.4.2 Evolution of the electron temperature proles . . . . . . . . . . . . 83
6.4.3 Evolution of the electron density proles . . . . . . . . . . . . . . . 84
CONTENTS vii
7 Non-ideality of EDM plasmas 87
7.1 Plasma coupling parameter . . . . . . . . . . . . . . . . . . . . . . . . . 87
7.2 Spectroscopic signs of the non-ideality . . . . . . . . . . . . . . . . . . . . 90
7.2.1 Broadening and shift of H
. . . . . . . . . . . . . . . . . . . . . . . 90
7.2.2 Asymmetry and structure of H
. . . . . . . . . . . . . . . . . . . . 90
7.2.3 Absence of H
and line merging . . . . . . . . . . . . . . . . . . . . 93

7.2.4 Inglis-Teller relation . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.3 Summary of the physical properties of EDM plasmas . . . . . . . . . . . . 95
8 Conclusions 101
A Stark broadening and shift of hydrogen spectral lines 105
A.1 The Stark eect on hydrogen levels . . . . . . . . . . . . . . . . . . . . . . 105
A.2 Consequences of the Stark eect on H
emission . . . . . . . . . . . . . . . 107
References 111
Acknowledgments 123
Curriculum Vitae 125
viii CONTENTS
Chapter 1
Introduction
1.1 Electrical Discharge Machining (EDM)
Electrical Discharge Machining (EDM) is a well-known machining technique since more
than fty years. Nowadays it is the most widely-used non-traditional machining process,
mainly to produce injection molds and dies, for mass production of very common objects.
It can also produce nished parts, such as cutting tools and items with complex shapes.
EDM is used in a large number of industrial areas: automotive industry, electronics,
domestic appliances, machines, packaging, telecommunications, watches, aeronautic, toys,
surgical instruments...
Figure 1.1: Examples of parts machined with EDM: high speed turbine and mold for the
screw thread of PET bottles, produced by die-sinking; die for manufacturing plastic cladding
and micro-parts, produced by wire-cutting (images c Agie-Charmilles [1, 2]).
1
2 CHAPTER 1. INTRODUCTION
The advantages of EDM over traditional methods such as milling or grinding are mul-
tiple. Any material that conducts electricity can be machined, whatever its hardness
(hardened steel, tungsten carbide, special alloys for aerospace applications, for example).
Furthermore, complex cutting geometry, sharp angles and internal corners can be pro-
duced. Final surface state with low rugosity (< 100 nm) and precise machining ( 1 m)
are other important advantages. Moreover, there is no mechanical stress on the machined
piece, no rotation of workpiece or tool is necessary, and the machines have a high au-
tonomy. On the other hand, the disadvantages are the relatively low material removal
rate (order of 100 mm
3
/minute), surface modication of the machined workpiece (white
layer and heat aected zone, typical depth 50 m), and limited size of workpiece and
tool, for example.
1.1.1 Principles
The principle of EDM is to use the eroding eect of controlled electric spark discharges
on the electrodes. It is thus a thermal erosion process. The sparks are created in a
dielectric liquid, generally water or oil, between the workpiece and an electrode, which can
be considered as the cutting tool. There is no mechanical contact between the electrodes
during the whole process. Since erosion is produced by electrical discharges, both electrode
and workpiece have to be electrically conductive. Thus, the machining process consists in
successively removing small volumes of workpiece material, molten or vaporized during a
discharge. The volume removed by a single spark is small, in the range of 10
6
10
4
mm
3
,
but this basic process is repeated typically 10000 times per second.
Figure 1.2 gives a simple explanation of the erosion process due to a single EDM dis-
charge. First, voltage is applied between the electrodes. This ignition voltage is typically
200 V. The breakdown of the dielectric is initiated by moving the electrode towards the
workpiece. This will increase the electric eld in the gap, until it reaches the necessary
value for breakdown. The location of breakdown is generally between the closest points
of the electrode and of the workpiece, but it will also depend on particles present in the
gap [3]. When the breakdown occurs, the voltage falls and a current rises abruptly. The
presence of a current is possible at this stage, because the dielectric has been ionized and
a plasma channel has been created between the electrodes.
The discharge current is then maintained, assuring a continuous bombardment of ions
and electrons on the electrodes. This will cause strong heating of the workpiece material
(but also of the electrode material), rapidly creating a small molten metal pool at the
surface. A small quantity of metal can even be directly vaporized due to the heating.
During the discharge, the plasma channel expands. Therefore, the radius of the molten
metal pool increases with time. The distance between the electrode and the workpiece
during a discharge is an important parameter. It is estimated to be around 10 to 100 m
(increasing gap with increasing discharge current).
At the end of the discharge, current and voltage are shut down. The plasma implodes
under the pressure imposed by the surrounding dielectric. Consequently, the molten metal
pool is strongly sucked up into the dielectric, leaving a small crater at the workpiece surface
(typically 1500 m in diameter, depending on the current).
1.1. ELECTRICAL DISCHARGE MACHINING (EDM) 3
v
o
l
t
a
g
e
c
u
r
r
e
n
t
time
time
(a) (b) (c) (d) (e)
(a) Pre-breakdown :
voltage applied
between the electrode
and the workpiece
(b) Breakdown :
dielectric breakdown,
creation of the
plasma channel
(c) Discharge :
heating, melting and
vaporizing of the
workpiece material
(d) End of the discharge :
plasma implosion,
removing of the
molten metal pool
(e) Post-discharge :
solidifying and flushing
of the eroded particles
by the dielectric
Figure 1.2: Principle of the EDM process.
The liquid dielectric plays a crucial role during the whole process: it cools down the
electrodes, it guarantees a high plasma pressure and therefore a high removing force on
the molten metal when the plasma collapses, it solidies the molten metal into small
spherical particles, and it also ushes away these particles. The post-discharge is in fact
a crucial stage, during which the electrode gap is cleaned of the removed particles for the
next discharge. If particles stay in the gap, the electrical conductivity of the dielectric
liquid increases, leading to a bad control of the process and poor machining quality. To
enhance the ushing of particles, the dielectric is generally owing through the gap. In
addition, the electrode movement can be pulsed, typically every second, performing a
large retreat movement. This pulsing movement also enhances the cleaning, on a larger
scale, by bringing fresh dielectric into the gap.
The material removal rate can be asymmetrically distributed between the electrode
(wear) and the workpiece (erosion). The asymmetry is mostly due to the dierent ma-
terials of the electrodes. But it also depends on the electrode polarity, on the duration
of the discharges and on the discharge current. Note that by convention, the polarity is
called positive when the electrode is polarized positively towards the workpiece, negative
otherwise. By carefully choosing the discharge parameters, 0.1% wear and 99.9% erosion
can be achieved.
Two main types of machining can be distinguished, as shown in gure 1.3. In die-
sinking EDM, the electrode is shaped and will produce its negative form into the work-
piece. The wear has to be very low, in order to keep the electrode original shape unmod-
ied during the whole machining process. The asymmetry in the material removal rate
is thus crucial for die-sinking. The electrode is generally in copper or graphite, and the
dielectric is oil. In wire-cutting EDM, the electrode is a continuously circulating metallic
wire (typical diameter 0.1 mm, generally in steel, brass or copper), which cuts the work-
piece along a programmed path. Deionized water is used as dielectric, directly injected
around the wire. The wire is capable of achieving very small cutting angles. The wear
is a lesser problem than in die-sinking, because eroded portions of the wire are continu-
ously replaced by new ones due to the wire circulation. Die-sinking EDM is mainly used
to produce injection molds, whereas the main applications of wire-cutting EDM are the
production of steel cutting dies and extrusion dies.
(a) die-sinking (b) wire-cutting
electrode
workpiece
cutting direction
wire circulation
Figure 1.3: Main types of EDM: die-sinking and wire-cutting.
The quality of the machining, i.e. precision and surface rugosity, is directly related
to the discharge parameters (current, voltage, discharge duration, polarity...), but also
on the dielectric cleanliness. Sparks with strong current produce deep craters: a high
removal rate is obtained but with a high surface rugosity. On the other hand, sparks with
low current will produce small craters: the surface rugosity is low but the removal rate
is also low. For a rough preliminary machining in die-sinking for example, high current,
long discharge duration and positive polarity will be chosen, typically 50 A and 1 ms. In
wire-cutting, current up to 1000 A can be used, but with shorter duration ( s). For
surface nishing and polishing, short discharges with low current and negative polarity
are used, typically 2 A and 5 s.
1.1.2 History
The historical roots of EDM date back to the discovery of electric discharges. Besides the
discharges produced by natural phenomena, namely lightning, the production of articial
discharges has been closely related to the development of electrical energy sources. First
investigations of electrostatic phenomena were performed with frictional machines, during
the rst half of the 18
th
century. After that, the rst sparks and pulsed arcs were produced
with Leyden jars, an early form of capacitor invented in Germany and in the Netherlands
around 1745 [4] (see gure 1.4 (a)). More powerful discharges were created by putting
several Leyden jars in parallel, creating thus a battery. Although scientists of this period
sensed that the nature of these articial discharges was the same as the nature of lightning,
the understanding of the observed phenomena was very incomplete.
Joseph Priestley (17331804), an English theologian and chemist, was the rst to
discover in 1766 erosion craters left by electric discharges on the cathode surface:
June the 13
th
, 1766. After discharging a battery, of about forty square
feet, with a smooth brass knob, I accidentally observed upon it a pretty large
circular spot, the center of which seemed to be supercially melted. (...) After
an interruption of melted places, there was an intrie and exact circle of shining
dots, consisting of places supercially melted, like those at the center. (see
gure 1.4 (b))
June the 14
th
, 1766. (...) Examining the spots with a microscope, both the
shining dots that formed the central spot, and those which formed the external
circle, appeared evidently to consist of cavities, resembling those on the moon,
as they appear through a telescope, the edges projecting shadows into them,
when they were held in the sun. [5]
Priestley also investigated the inuence of the electrode material and of the discharge
current on the craters size.
Figure 1.4: (a) Engraved plate sent by Alessandro Volta to Joseph Priestley, showing the spark
produced by short-circuit of a Leyden jar (1775) [6]; (b) sketches of erosion craters on cathode
surface, observed by Joseph Priestley in 1766 [5].
Whereas the discharges studied by Priestley were pulsed and oscillating (because cre-
ated by short-circuiting of Leyden jars), continuous discharges could only be produced
with battery of electrochemical cells, invented later by Alessandro Volta (17451827) in
1799. Developing very large voltaic batteries, the rst continuous carbon arc was produced
by Vasilii Petrov in St-Petersburg in 1802 [7]. Published in 1803 but only in Russian, his
discovery remained ignored and forgotten for over a century. The discovery of electric
arcs is thus often attributed to Humphry Davy (17781829). Unaware of Petrovs work,
he re-discovered independently carbon arcs around 1808, using the huge voltaic battery of
the Royal Institution of London (see gure 1.5 (a)). By separating two horizontal carbon
electrodes connected to the battery, Davy created a bright and stable discharge. The
shape of this discharge was arched, giving its name to the phenomenon.
Development of devices using electric arcs for lighting purposes followed quickly. Swiss
natural philosopher Auguste-Arthur de la Rive (18011873) proved in 1820 that arcs can
also burn in vacuum, by creating a discharge in an exhausted glass vessel. Figure 1.5 (b)
shows examples of these early carbon arc lamps.
Figure 1.5: (a) Public demonstration of the carbon arc discharge, probably by Humphry Davy
in the Royal Institution of London (early 19
th
century) [7]. The picture below shows the basement
lled with a huge battery, used to create the discharge; (b) early carbon arc lamps in air (left)
and in exhausted glass vessel (right), also known as Davys electric eggs or de la Rives electric
eggs [7].
With sophistication of electric sources and industrialization, Auguste de Meritens
(18341898) developed in 1881 in France a second major application using electric arcs.
He used the heat produced by an arc for joining lead plates, inventing the principle of arc
welding. Nowadays, electric arcs are also used for coating deposition, metal processing,
plasma spraying and as high power switches, for example [8].
The history of EDM itself begins in 1943, with the invention of its principle by Russian
scientists Boris and Natalya Lazarenko in Moscow. The Soviet government assigned them
to investigate the wear caused by sparking between tungsten electrical contacts, a problem
which was particularly critical for maintenance of automotive engines during the second
world war. Putting the electrodes in oil, they found that the sparks were more uniform
and predictable than in air. They had then the idea to reverse the phenomenon, and
to use controlled sparking as an erosion method [9]. Though they could not solve the
original wear problem, the Lazarenkos developed during the war the rst EDM machines,
which were very useful to erode hard metals such as tungsten or tungsten carbide. The
Lazarenko circuit remained the standard EDM generator for years.
In the 1950s, progress was made on understanding the erosion phenomenon [1012].
It is also during this period that industries produced the rst EDM machines. Swiss
industries were involved very early in this market, and still remain leaders nowadays.
Agie was founded in 1954, and les Ateliers des Charmilles produced their rst machine in
1955. Due to the poor quality of electronic components, the performances of the machines
were limited at this time.
2005
1955
Figure 1.6: 50 years of evolution in EDM machines: Eleroda D1 (1955) and
Robol 2050 TW (2005) from Charmilles (images c Charmilles Technologies [1]).
In the 1960s, the development of the semi-conductor industry permitted considerable
improvements in EDM machines. Die-sinking machines became reliable and produced
surfaces with controlled quality, whereas wire-cutting machines were still at their very
beginning.
With the introduction of numerical position control in the late 1960s and early 1970s,
the movements of electrodes became much more precise. This major improvement pushed
forward the performance of wire-cutting machines. Computer numerical controlled sys-
tems (CNC) improved further the performance of EDM in the mid 1970s.
During the following decades, eorts were principally made in generator design, process
automatization, servo-control and robotics. Applications in micro-machining became also
of interest during the 1980s [13]. It is also from this period that the world market of
EDM began to increase strongly, and that specic applied EDM research took over basic
EDM research [14]. Finally, new methods for EDM process control arose in the 1990s:
fuzzy control and neural networks.
1.1.3 State of the art
Fifty years after the rst industrial machine, EDM has made considerable progress. Recent
improvements in machining speed, accuracy and roughness have been achieved mostly
with improvements in robotics, automatization, process control, dielectric, ushing and
generator design [1518]. The other main research domains are the machining of non-
conductive materials such as ceramics [19, 20], micro-machining [2124], characterization
and improvement in the machined surface quality [25, 26], and modelling of the EDM
process [2729]. But so far, few studies have been done on the discharge itself, which lies
at the heart of the process.
However, in various elds, breakdown in dielectric liquids or solids have already been
studied: pre-breakdown and breakdown in dierent liquids [30, 31] and condensed mat-
ter [32], exploding wires in water [33], laser produced plasma in transformer oil [34], and
Teon capillary discharges [35] for example. These plasmas have similar properties to
those of the EDM plasma.
Though EDM keeps unmatched abilities such as the machining of hard materials and
complex geometries, this technique has to evolve constantly in order to stay competitive
and economically interesting in the modern tooling market against other traditional or
new machining techniques [17, 18, 36].
1.2 Purpose and structure of the work
Further improvements in EDM performances, especially for micro-machining, require a
better control and understanding of the discharge, and of its interaction with the elec-
trodes. A better comprehension of the sparking process will also reduce problems related
to its stochastic nature.
Until now, process optimization relied almost only on empirical methods and recipes.
It is necessary to go beyond this empirical optimization, and to use reliable numerical
models to predict important parameters, such as the material removal rate and wear for
example.
Although EDM is quite old, only a few theoretical and numerical studies on the EDM
plasma exist [3740], mainly due to the complex physics involved in this process. The
EDM process mixes indeed breakdown of liquids, plasma physics, heat transfer, radi-
ation, hydrodynamics, materials science, electrodynamics... These models still contain
several parameters which are empirically determined or articially introduced. Further-
more, experimental characterization of the EDM plasma is lacking. Some spectroscopic
measurements have already been made but remain very incomplete [41, 42]. As we will
see later (section 3.6), the EDM plasma is experimentally dicult to investigate. These
diculties are the main cause for the lack of experimental data, which are nevertheless
essential as inputs and for the validation of numerical models.
The purpose of this work is thus a systematic experimental investigation of the EDM
plasma, in order to measure its physical properties and to improve the understanding
of its complex basic physics. Besides the industrial aspect, the study of this plasma is
1.2. PURPOSE AND STRUCTURE OF THE WORK 9
also of fundamental interest. Other works on similar plasmas, notably measurements of
their density and temperature (see sections 2.2 and 2.3), showed that they can be classed
among non-ideal or strongly coupled plasmas. Such plasmas have very interesting physical
properties and are still not well known [4346]. Strongly coupled plasmas produced in
the lab are interesting also for astrophysical studies, because deep layers of giant planets
and superdense plasmas of the matter of white dwarves are of this kind [43].
This work is one part of a special EDM research project, initiated by the innovation
committee of the Technology-Oriented Program NANO 21 (project n
5768.2). This
project was a collaboration involving:
Charmilles Technologies, as the industrial partner (Dr G. Wlder);
the Centre de Recherches en Physique des Plasmas (CRPP) of the Ecole Polytech-
nique Fdrale de Lausanne (EPFL), for the study of the plasma (Dr Ch. Hollen-
stein);
the Laboratoire de Simulation des Matriaux (LSMX) of the EPFL, for a numerical
model of the temperature evolution in the workpiece surface (Prof. M. Rappaz);
the Dpartement de Physique de la Matire Condense (DPMC) of the Universit de
Genve, for the study of metallurgical processes occurring at the workpiece surface
(Prof. R. Flkiger);
the Electrochemistry group of the Bern Universitt, for the study of corrosion as-
pects of the EDM process (Prof. H. Siegenthaler).
The present manuscript is structured as follows: we will rst present in chapter 2 a
general background for the understanding of EDM plasmas, i.e. a brief summary of exist-
ing knowledge about similar phenomena and plasmas. The dierent experimental setups
and plasma diagnostics used in this work are described in chapter 3, and the experimental
results are presented and discussed in chapters 4 (results about pre-breakdown), 5 (imag-
ing results) and 6 (spectroscopy results). Chapter 7 treats the non-ideal character of EDM
plasmas, and also gives a summary of their measured physical properties. Finally, general
conclusions are given in chapter 8.
Chapter 2
EDM plasmas : Background
Although EDM discharges take place in a dielectric liquid, the rst section of this chapter
will deal with discharges in gases. They have been extensively studied [4749], and have
common features with discharges in liquids. After that, a brief review about the specic
characteristics of discharges in liquids will be given in the second section. Finally, other
plasmas similar to the EDM plasmas are presented in the last section.
2.1 Discharges in gases
2.1.1 Spark, arc, glow & co.
Depending on the gas pressure, the electrode gap and the electrode conguration, several
discharge regimes can be distinguished. They can be classied according to their current-
voltage characteristics, as shown in gure 2.1.
v
o
l
t
a
g
e
current
Townsend
discharge
corona
spark
arc
normal
glow
abnormal
glow
low-pressure
atmospheric pressure
Figure 2.1: Schematic current-voltage characteristics of the dierent types of discharges in
gases [47, 48].
11
12 CHAPTER 2. EDM PLASMAS : BACKGROUND
Four main types of steady or quasi-steady processes exist:
the Townsends dark discharge, characterized by a very weak current ( 10
8
A);
the glow discharge, widely used in many industrial processes, operating at low
current ( 10
2
A), fairly high voltage ( 1 kV) and low pressure ( mbar). The
glow plasma is weakly ionized and in a non-equilibrium state, and is visible as a
uniform glowing column. As in the Townsends discharge, electrons are emitted by
ion impacts on the cold cathode;
the corona discharge, also at low current ( 10
6
A) but at atmospheric pressure.
Corona discharges develop locally (typically around sharp ends of wires) in strongly
non-uniform electric eld;
the arc discharge, characterized by high current ( 100 A), low voltage ( 10 V)
and a bright light emission. The arc discharge diers from the glow discharge
in the electron emission mechanism. In arcs, electrons are emitted by thermionic
processes, due to the heating of the cathode. The plasma of high pressure arcs can
be considered to be in a state of thermodynamic equilibrium.
On the contrary to Townsend, glow, corona and arc discharges, the spark discharge
is not a steady process but a transient process, i.e. an unstable transition state of limited
lifetime towards a more stable regime (see gure 2.1). To be perfectly rigorous, it would
be more proper to say spark breakdown than spark discharge, since it is a transition
mechanism and not a state resulting from a transition [48]. However, we will keep (abu-
sively but for clarity) the term spark discharge or simply discharge when speaking about
the period during which the EDM plasma exists (part c in gure 1.2), and the term break-
down will refer to the transition from the no-plasma situation to the plasma situation
(part b in gure 1.2).
2.1.2 Sparks and streamers
Since the plasma created during the EDM process is precisely a spark, it is worthwhile
to describe this type of discharge in more detail. Note that lightning shows beautiful
examples of giant spark discharges.
The breakdown phenomenon leading to the creation of a spark is complex. The break-
down is too fast to be explained by repetitive electron avalanches through secondary cath-
ode emission, as in low pressure discharges. It consists rather of a very rapid growth of a
thin weakly-ionized channel called a streamer, from one electrode to the other.
A streamer is formed from an intensive primary electron avalanche, starting from the
cathode (see gure 2.2 (a)). A space charge eld is associated with this avalanche, due
to the polarization of charges inside it. This electric eld increases with the avalanche
propagation and growth. The avalanche has to reach a certain amplication before it
can create a streamer. As soon as the space charge eld is comparable or exceeding the
applied external eld, a weakly ionized region can be created due to this amplication of
the electric eld: the streamer is thus initiated.
2.1. DISCHARGES IN GASES 13
(a) electron avalanche (b) positive streamer (c) negative streamer
anode +
cathode -
anode +
cathode -
anode +
cathode -
+
+
+
+
+
+
+
+
+
+
s
t
r
e
a
m
e
r
secondary
avalanches
+
+
+
+
+ +
+
+ +
+ +
+
_
_
_
_
_
_
_
_ _
_
_ _
_
_
_
_ _
_
+
+
+
+
+
_
+
+
_
+
_ _
+
_
_
_
_
_
_
_
_
_
_
_
_
+
+
+
+
+
hn
p
r
o
p
a
g
a
t
i
o
n
t > t
2 1
t
1
+
+
+
_
_
_
t > t
2 1
t
1
t > t
2 1
t
1
+
+
+
__
_
+
+
+ +
+ +
_
_
_
+
_
_
_ _
+
+ +
_
_
+
_ _
+
+
+
_ _
space charge
field
Figure 2.2: Breakdown mechanisms leading to a spark discharge. Propagation of: (a) the
primary electron avalanche; (b) a positive streamer; (c) a negative streamer [49].
Once the streamer is initiated, it then grows and propagates, following a zigzagging
and branched path due to the random nature of the propagation mechanism. The speed
of this propagation is extremely high, reaching 10
6
m/s. The propagation can be directed
towards both the anode or the cathode, depending on the gap distance and voltage.
In moderate gaps and with moderate voltages, the avalanche-to-streamer
transition occurs only when the primary avalanche has crossed the gap and reached
the anode. The avalanche has not grown enough and the space charge eld is
not high enough to create a ionized region, before the avalanche has reached the
anode. Then, the streamer starts from the anode and grows towards the cathode.
This kind of streamer is called cathode-directed or positive. The streamer growth is
caused by secondary avalanches, created near the positive head of the streamer (see
gure 2.2 (b)). These secondary avalanches are initiated by electrons released by
photo-ionization. The electrons of the secondary avalanches are rapidly attracted
into the streamer, neutralizing the streamer positive head and leaving behind them
the positive ions of the secondary avalanches (ions move much slower than electrons).
These positive charges become the new head of the extended streamer. This is how
the positive streamer grows.
In large gaps and/or with strong gap voltages, the space charge eld of the pri-
mary avalanche can be suciently high to create the streamer even before reaching
the anode. Thus, the avalanche-to-streamer transition occurs in the gap. Then, the
streamer propagates towards both electrodes at the same time. If the avalanche-to-
streamer transition occurs while the avalanche has not yet gone far from the cathode,
the streamer grows mostly towards the anode. In this case, the streamer is called
anode-directed or negative. The growth mechanism towards the cathode remains
the same as described above, but the growth towards the anode is slightly dierent.
Here, the electrons of the primary avalanche form a negative head for the streamer.
These electrons rapidly neutralize the positive ions of secondary avalanches, also
initiated near the streamer head by photo-ionization and by moving electrons (see
gure 2.2 (c)). The electrons of the secondary avalanches then form the new head of
the extended streamer. Thus, for both positive and negative streamers, the streamer
is feeding on charges created ahead of its tip by secondary avalanches.
When the electrode gap is closed by a streamer, the breakdown phase is completed and
the discharge phase begins. The transition from a weakly-ionized channel (the streamer
bridging the gap) to a highly-ionized channel (the spark itself) is poorly understood. It is
probably caused by a back streamer, similar to the well-known return stroke in light-
ning discharges [49]. If we assume that a streamer is perfectly conducting, the head of a
positive streamer, for example, is at the same potential as the anode. When the streamer
head is approaching close to the cathode, all the potential fall is located over a very short
distance, the distance between the cathode and the streamer head. The electric eld is
so intense in this region that electrons are emitted in great number from the cathode and
from atoms near the cathode. Once the gap is closed by the streamer, these electrons,
multiplied at enormous intensity, are accelerated towards the anode in the initial streamer
channel, causing strong ionization. This ionization front is propagating backwards at
10
7
m/s. The formation of the true spark channel is thus probably caused by this
back streamer, which strongly increases the degree of ionization in the original streamer
channel.
The plasma composing the spark channel is highly ionized and conductive, capable
of sustaining a large current ( 10
4
A). The spark is accompanied by a cracking sound
(the thunder in the case of lightning), resulting from the shock wave created by the rapid
and localized heating of the gas surrounding the plasma channel. The channel radially
expands with time, because the surrounding gas is gradually ionized, by heat conduction
and by the shock wave. The temperature of a spark is typically around 1.8 eV (20000 K),
and the electron density around 10
17
cm
3
.
If the power source is capable of delivering the discharge current over a certain amount
of time, the spark will naturally transform into an arc, since the spark is a only a transient
process.
2.1.3 Electric arcs and cathode spots
Existing theories, measurements and models about vacuum and atmospheric arcs are
abundant [8, 5056]. It is still a widely studied subject, because cathode phenomena,
for example, remain poorly understood due to their extreme complexity. As in the EDM
process, it involves solid state, surface and plasma physics, electric and thermal processes.
Since arcs erode cathodes by leaving small craters on their surfaces, the knowledge of arc
phenomena can be useful to understand the EDM erosion process, although the EDM
plasma is not an arc but a spark. Furthermore, the plasma state in a high pressure arc
column is found to be relatively similar to that in a spark channel [49].
Cathode region
The cathode is a region of specic interest, because it has to produce the necessary electron
current for the arc to survive. A high current density is indeed one of the characteristic
feature of electric arcs. While electrons are simply falling from the plasma into the
anode because of its higher electrical potential, the cathode has to develop very ecient
emission mechanisms to extract electrons from the metal into the plasma. Electrons of
the conduction band in the metal need in fact some energy to overcome the energy gap
of the metalplasma interface, also called work function.
The transformation of a spark into an arc is accomplished by the creation of a hot
spot on the cathode surface, called cathode spot. This small spot ( 10 m in diameter)
has astonishing physical properties and is capable of supplying a great electron current.
Electrons are emitted from the cathode spot by:
thermionic emission (emission of the most energetic electrons from a heated metallic
surface);
eld electron emission (emission by tunnel eect due to the lowering of the external
potential caused by an electric eld at the surface);
thermionic eld emission, also called thermo-eld emission, which is a combination
of the two preceding processes. The electric eld enhances the thermionic emission
by the Schottky eect. This mechanism dominates by far in electric arc cathodes;
thermal runaway, a non-stationary explosive emission of electrons and explosive
evaporation. A signicative quantity of cathode material, typically a protrusion
at the surface, can be directly evaporated and transformed into a dense plasma.
This occurs when there is a suciently rapid heating of the cathode, i.e. when
the deposited power grows faster than heat conduction in the cathode can remove
it [57].
The discharge organizes itself in a such way as to create a strong electron emission
from the cathode by the mechanisms cited above, i.e. by raising its temperature and by
creating a strong electric eld at its surface. The voltage jump is mostly localized near
the electrodes, in the so-called anode and cathode layers (see gure 2.3). Consequently,
anode
layer
cathode
layer
arc column
x
V
C A
Figure 2.3: Schematic prole of electrical potential in an arc [49].
the electric eld is particularly high in these very thin regions. This type of potential
prole is due to the presence of space charged regions near the electrodes. The plasma in
the cathode layer, usually called cathode spot plasma because it is located in front of the
cathode spot, is a small and dense plasma. It remains poorly understood, but its electron
temperature is estimated around 5 eV (60000 K) and its density around 10
20
cm
3
. The
cathode spot plasma should thus be dense enough to present non-ideal eects [57].
A detailed structure of the cathode layer is given in gure 2.4. While the plasma
of the arc column is assumed to be in thermal equilibrium, the plasma located in the
cathode layer is characterized by deviations from equilibrium [53]. First, a deviation of
the electron temperature T
e
from the heavy particles temperature T
h
is present in a layer
of thermal relaxation. Then, we have a violation of the ionization equilibrium. In this
ionization layer, the production rate of ions is very high due to collisions with energetic
electrons strongly accelerated from the cathode. The ion ux leaving this layer towards the
cathode is thus much higher than the ux entering from the layer of thermal relaxation.
The dense cathode spot plasma is created in this layer. Finally, the quasineutrality is
violated in a very thin sheath near the cathode, where the ion density n
+
is higher than
the electron density n
e
. This is the space charged region creating the potential prole
shown in gure 2.3, and where almost all the voltage drop is located.
c
a
t
h
o
d
e
sheath ionization layer layer of thermal relaxation arc column
space charged
(n > n )
+ e
quasineutral quasineutral
collisionless
ionization > recombination ionization = recombination
~ 20 nm
(Debye length)
~ 10 m m ~ 100 m m
T > T
e h
Local
Thermal
Equilibrium
(LTE)
e
-
+
+
+
e
-
+
x
electric potential
+
+
+
Figure 2.4: Schematic structure of the cathode layer (not to scale). The dimensions of the
three layers are evaluated for an argon atmospheric pressure arc plasma at 10000 K and with a
charged particle density of 10
17
cm
3
[53].
By this plasma structure, the cathode spot is strongly heated by the ion bombardment
coming from the ionization layer through the collisionless sheath. This heating leads to
evaporation and melting, thermionic emission and thermal runaway in some cases. The
electron emission is further increased by the electric eld produced by the sheath, adding
eld emission and thermionic eld emission. In return, the electrons coming from the
cathode are crucial to create a high ionization rate in the ionization layer, producing
the ions that will heat the cathode in a self-consistently balanced regime. This coupled
ion/electron production is the sustaining mechanism of the arc discharge.
The terms involved in the energy balance at the cathode spot are multiple. The
energy is mainly brought by ion bombardment, but smaller contributions are also given
by Joule heating, atom and electron bombardment, recombination in the cathode, plasma
radiation (almost negligible [57]) and Thomson eect (also negligible). Energy is mostly
dissipated by electron emission and heat conduction in the electrode (slow process), but
also by evaporation, surface radiation and droplet emission. The relative importance of all
these energy sources and sinks depend strongly on the arc conditions (pressure, materials,
current, etc.). Heating mechanisms of electrodes in EDM sparks should be very similar
to those occurring in arcs as described here.
The temperature of a cathode spot is typically around 40005000 K [57], high above
the fusion temperatures of metals, which explains the erosive eect of arcs on cathodes.
Besides direct evaporation, ejection of cathode melted matter due to the plasma pressure
and explosive erosion by thermal runaway, another erosive mechanism can occur also dur-
ing the discharge. If energy dissipation dominates at the surface due to a high radiation
and electron emission (this is the case if the surface temperature is very high), the maxi-
mum temperature is located below the surface, in the cathode spot. This will lead to an
internal explosion, and consequently to ejection of solid or liquid matter into the plasma,
in the form of m droplets [57].
Anode region
At the anode, the situation is also complex, depending strongly on experimental conditions
with several dierent modes, and not completely understood. Generally, the current is
distributed over a larger surface than at the cathode. The current density is thus lower,
and no signicative erosion is visible. However, anode spots can develop under certain
conditions of current, pressure and anode geometry. In this case, matter can also be
evaporated and signicant erosion is observed. The plasma structure near the anode has
similarities with that near the cathode. A negative space charged region is located directly
near the anode surface (n
e
> n
+
), because ions are unable to cross the anode potential
barrier. The metal vapor coming from the anode can thus be ionized in the anode layer,
due to this space charged region. Energy is brought to the anode by the electron ux and
by recombination with ions in the anode; energy is dissipated through evaporation and
heat conduction. The temperature of the anode spot is slightly lower than that of the
cathode spot, still being around 3000 K [49].
Inter-electrode plasma
A large quantity of power is dissipated in the arc column by the Joule eect, heat conduc-
tion and radiation. On the axis of the column, the plasma temperature is typically around
5 eV (60000 K) and the electron density around 10
16
cm
3
, but they depend on the gas,
the pressure and the current. These values decrease radially, i.e. towards the edges of
the column. The plasma of the arc column is thus much less dense than the cathode spot
plasma. Due to frequent collisions and thus intensive energy exchange between particles,
the plasma of high-pressure arcs is in local thermal equilibrium. The degree of ionization
is also very high, close to 100%.
Arcs in EDM
EDM sparks can sometimes transform into arcs, especially if long discharge durations and
graphite electrodes are used. One hot spot is then created on each electrode surface, and
the following discharges will systematically take place between these two hot spots and
instantly turn into an arc. The result is catastrophic from the erosion point of view, leading
to a localized burn of the workpiece surface and to the destruction of the electrode shape.
This is particularly problematic when performing smoothing and polishing operations.
Hot spots have to be avoided as much as possible during the EDM process, because one
of the advantages of this technique is precisely to distribute the erosion spots over the
whole workpiece surface. The stochastic change in the localization of the successive spark
discharges is thus crucial in EDM.
The arcing phenomenon during EDM is well known, and is generally avoided by stop-
ping the discharge as soon as the voltage reaches a value below a xed threshold, typically
1520 V. A decrease in the discharge voltage is the sign of a transition into an arc, because
arcs burn with lower voltage than sparks (see gure 2.1). Other arc detection methods
exist, based on the measurement of the time lag between the voltage application and the
breakdown, on the measurement of the ignition voltage value, or on the measurement of
the voltage descending ank at the breakdown [58].
2.2 Discharges in dielectric liquids
The principal dierence between discharges in gases and discharges in liquids is the den-
sity of the medium in which the breakdown occurs. The higher density of liquids makes
them more dicult to break down, i.e. it requires a higher electric eld. As described
below, the breakdown mechanism is also slightly dierent, and the plasma properties are
strongly inuenced by the pressure imposed by the surrounding liquid.
Breakdown in dielectric liquids have been widely studied, especially for insulation
problems of electric transformers [32]. Published articles on breakdown phenomena in
liquids are very numerous, and the existing knowledge is summarized in a few review
articles [32, 5963]. A wide range of liquids have been investigated: water [64], salted
water solutions [65], oils with dierent aromatic constituents [30,66,67], silicon uids [59],
liquid argon and nitrogen [31, 68], benzene, toluene, carbon tetrachloride [69], and other
hydrocarbon liquids [7072] such as propane, pentane, cyclopentene, hexane, cyclohexane,
isooctane, decane, etc. The inuences of hydrostatic pressure, conductivity [65], viscosity
[59], additives [73] and particles [74, 75] on the breakdown mechanism have also been
studied.
As in high-pressure gases, streamers are involved in breakdown in liquids. Two types of
streamer exist: the positive streamer starting from the anode, and the negative streamer
starting from the cathode. The use of a point-to-plane geometry permits a separate obser-
2.2. DISCHARGES IN DIELECTRIC LIQUIDS 19
vation of the two types of streamer, depending on the polarity of the electrodes. Contrary
to the situation in gases, the structure and propagation speed of positive and negative
streamers are dierent. As shown in gure 2.5, the positive streamer is lamentary and
fast ( 110 km/s), and the negative streamer is bushy and slow ( 100 m/s). In
point-to-point geometry, the two types of streamer are emitted from both electrodes.
Figure 2.5: Shadowgraphs of negative and positive streamers in oil (taken from [30]).
The sequence of events leading to breakdown can be broken down as follows:
1. Initiation
The initiation of the breakdown mechanism diers from that in gases. Direct prop-
agation of electrons into the liquid and impact ionization are dicult, due to the
strong collisions between electrons and molecules occurring in this dense medium.
An avalanche of electrons in the liquid phase is thus unlikely, or at least only of short
range. It is generally accepted that the development of the primary avalanche, which
will create the streamer, takes place in a region of lower density, created beforehand
near the electrode. If the liquid pressure is not too high, this low density region is
a vapor bubble (or a vortex of hot liquid in very viscous liquids [59]).
However, this point is controversial. According to [70, 76], the positive streamer
would be a purely electronic process occurring in the bulk liquid, while negative
streamer would rst involve the formation of a bubble. This should explain the
dierent propagation speed of the two types of streamer. On the other hand, other
authors claim that the breakdown process cannot be initiated without the presence
of bubbles [77, 78]. The eect of the pressure on streamer initiation, for example,
is in good agreement with this second theory. Increasing the pressure inhibits in
fact streamer development and increases the breakdown voltage. This shows that
the phenomena involved in the breakdown mechanism occur in a gaseous medium.
Additional experimental evidence seems to demonstrate that the role of bubbles in
the breakdown triggering is indisputable [79].
Micro-bubbles can pre-exist in the liquid [78] or form electrically, i.e. from heat
released by small electron avalanches in the liquid or by eld emission near electrode
asperities [68, 80]. In salted solutions, ionic currents can also slightly enhance the
formation of bubbles [62].
2. Streamer formation
The propagation of the primary electron avalanche is strongly facilitated in the bub-
ble. Since electrons are continuously heating the liquid in the front of the avalanche,
and consequently lowering its density, the bubble is growing (time scale 10 ns).
New avalanches are formed and thus new bubbles grow ahead of the preceding ones.
The ionization of a channel in the liquid, i.e. the formation of the streamer, is
caused by this cycle of heating, density lowering and avalanche growth.
3. Streamer propagation
The streamer then grows and propagates according to the mechanism described
above, governed by a combination of electrostatic and hydrodynamic forces and
even instabilities [32, 59]. This mechanism is quite similar to that in gases, but
more complex. The structure of the streamer is systematically branched, more
than in gases. This reects the diculty for the ionization front to propagate in a
dense medium. It could also be due to other inhomogeneously distributed micro-
bubbles leading the streamer path, or due to electrostatic repulsion between adjacent
streamer branches [78]. Small electric currents (in the form of bursts of fast pulses),
localized weak light emission and shock waves are associated with the propagation
of streamers.
4. Gap completion and breakdown
When the streamer reaches the other electrode, a reverse ionizing front is observed
as in gases, starting from the reached electrode and going back towards the initial
electrode. The ionized channel thus thickens, establishing the spark or arc discharge.
With negative streamers, the return stroke can even be emitted before the streamer
has reached the anode [76]. Intense emission of light is recorded simultaneously with
the gap completion, as shown in gure 2.6.
This general description of the breakdown mechanism does not take into account the
eects of the numerous experimental parameters. The characteristics of the streamers
depend on the gap voltage and distance, on the electrode materials, geometry and surface
state, on the liquid pressure, viscosity, density, conductivity, temperature, composition,
molecular structure, purity, etc. The presence of particles, for example, strongly facilitates
the triggering of a breakdown. The contamination of the dielectric is thus as important as
its type for the initiation of a breakdown. The addition of so-called electron scavengers
can also increase the propagation speed of negative streamers by one order of magnitude.
This type of additive (e.g. SF
6
or C
2
H
5
Cl [30]) increases the electronic anity of the
dielectric molecules. Though streamers are of gaseous nature (cf. inuence of pressure,
shock waves, role of bubbles), this dependence on electron scavenging additives show that
electronic processes are also operating. The emission of light is another evidence of this
fact.
2.2. DISCHARGES IN DIELECTRIC LIQUIDS 21
Figure 2.6: Propagation of a positive streamer and breakdown in oil (time in microseconds,
gap 1.27 cm, 82 kV, taken from [30]).
Spectroscopic analysis of the light emitted by streamers in hydrocarbon liquids and oil
shows the presence of C
2
, C
3
and H
2
molecules, but atomic hydrogen as well [34, 70, 81].
In liquid nitrogen, CN molecules are observed, probably produced by recombination of
nitrogen atoms with carbon atoms coming from the electrodes [31]. These results indicate
that the uid molecules are decomposed inside the streamers. Temperatures and densities
in streamers have also been measured with spectroscopy. The rotational and vibrational
temperatures of N
2
molecules in streamers in liquid nitrogen are 500 and 4000 K respec-
tively [31]; the electron temperature of streamers in oil are about 500010000 K [34].
The electron density is very high, about 10
17
10
19
cm
3
for positive streamers, and
about 10
16
10
17
cm
3
for negative streamers [31, 34, 81]. As a general rule, streamers
and resulting plasmas in liquids have smaller dimensions and higher densities than those
in gases, due to the pressure imposed by the liquid.
To conclude about discharges in liquids, we mention that specic applications have
already been developed. Streamer discharges can produce chemical reactants in liquids,
which will be used, for example, to degrade organic pollutants by oxidation [82]. Another
application is the sterilization of microorganisms present in water, with the aid of chemical
species (e.g. H
2
O
2
) produced by electric discharges [83].
Furthermore, bubbles are of great interest since single-bubble sonoluminescence has
been discovered. A gas micro-bubble, trapped in a liquid and periodically driven by intense
acoustic waves, can emit light when it is violently collapsing. This phenomenon is called
sonoluminescence, and still remains puzzling [84]. The emitted light is in the UV range,
which implies a high energy density inside the bubble. Recent experiments seem to prove
the existence of a plasma inside a sonoluminescing bubble [85]. Its temperature would
be above 15000 K. Thus, these bubbles could possibly be used to produce thermonuclear
fusion.
2.3 Other similar plasmas
Besides sparks and high-pressure arcs in gases, other plasmas have similar properties to
those created by electric discharges in dielectric liquids and to EDM plasmas.
Exploding wires
Plasmas in liquids can not only be created by applying high voltage between two im-
mersed electrodes. They can also be formed by exploding wires. In such experiments, a
thin metal wire placed in water is rapidly vaporized due to the ow of a strong current
( 100 kA on s time scale). The plasmas produced in this way have extremely high elec-
tron densities (10
21
10
22
cm
3
) and pressure ( 10 kbar), and temperatures of a few eV
(1000030000 K) [33, 8688]. These high densities are caused by the strong connement
created by the liquid inertia. Such plasmas are so dense that their core is optically highly
opaque. Explosives have even been used to further increase the plasma connement, and
consequently increase the plasma pressure and density [89].
Capillary discharges
In this type of discharge, the conning medium is a small capillary made of glass or
another dielectric (Teon, polyethylene, alumina, BeO for example), typically with an
inner diameter of 1 mm. This tube is in contact with two electrodes applying a fast current
pulse ( 1 kA over 100 ns). The plasma is formed by ablation of the capillary walls. The
electron densities reached are also high (10
16
10
19
cm
3
), and the temperatures are of a
few eV [35, 9092].
Laser-produced plasmas
Plasmas of metals and alloys can be created by focusing energetic Nd:YAG laser pulses
on solid targets, placed in a vacuum chamber. Typical pulses have a power density of
10
10
W/cm
2
, a duration of 7 ns, a repetition frequency of 30 Hz, and can be focused on a
300 m diameter spot. These plasmas have an electron density of 10
16
10
18
cm
3
, and
a temperature of about 11.5 eV (1000015000 K) [9395].
2.3. OTHER SIMILAR PLASMAS 23
Shock waves
The compression by shock waves of gases or metal vapor is another way to produce a dense
plasma. The gas is generally contained in a tube heated by a resistor furnace. The shock
wave is generated with a system of chambers at dierent pressures, or with condensed
explosives for powerful compression [43, 96]. The gas is then irreversibly compressed
and heated. This method produces plasmas with spectacular pressure ( 100 kbar)
and densities ( 10
23
cm
3
), with temperatures around 10 eV (100000 K). The use of
underground explosions and even nuclear (!) explosives permitted Russian scientists to
obtain gigantic pressures, up to hundreds of Mbar [43].
Astrophysical plasmas
Electron densities of plasmas present in the universe cover an astonishingly broad range:
from 1 cm
3
for interstellar matter and solar wind for example, up to 10
30
cm
3
for
white dwarfs, which are the late evolution stage of stars having a mass comparable or
lower than that of the sun. Among the most dense plasmas, let us cite the plasmas of
the deep layers of giant planets (hydrogen plasma, 6 10
24
cm
3
, 1 eV for Jupiter), the
interior of the sun (hydrogen plasma, 6 10
25
cm
3
, 1.5 keV), and other exotic objects
such as the matter of brown dwarfs (hydrogen plasma, 10
28
cm
3
, 1 keV) and the matter
of white dwarfs (carbon plasma, 5 10
30
cm
3
, 10 keV) [43, 44]. Dense plasmas can also
be produced by spacecraft entering the bottom layers of giant planet atmospheres. The
atmospheric pressure on these planets is very high, due to their strong gravitational eld.
Thus, a dense plasma will be formed ahead of a travelling spacecraft by heating of the
atmosphere.
Micro-plasmas
In contrast to the plasmas described above, micro-plasmas do not resemble to EDM
plasmas by their densities, but by their typical gap distance. Plasmas with micro- and
submicrometric gaps have recently been produced in air and other gases, mainly due
to progress in scanning probe microscopy (SPM) piezoelectric gap controllers [97, 98],
and progress in fabrication of micro-devices by photolithography [99]. Typical electron
density is 10
12
10
15
cm
3
, and the electron temperature around 5 eV [99, 100]. Chips
creating micro-discharges can be used as ionizing sources for micro-sensors, gas analyzers
and mass spectrometers [101, 102], or as integrated plasma chemical vapor deposition
(PCVD) apparatus, for example [103]. Micro-plasmas are also studied for plasma display
panel (PDP) applications [104]. Recent progress and applications of micro-plasmas have
been reviewed in [105].
Moreover, micro-gaps are of fundamental interest and new physical questions are aris-
ing. Since the gap distance becomes of the same order as the sheath thicknesses and even
as the plasma Debye length, the discharge has perhaps to organize itself dierently from
the traditional macro-gap discharges.
Chapter 3
Experimental setup and diagnostics
The rst section of this chapter presents the EDM device used in this work, along with
the machining parameters. The various plasma diagnostics are then described in the
following sections. Finally, in the last section we address the specic diculties related
to the experimental study of EDM plasmas.
3.1 Electrical discharge machining device
Figure 3.1 presents dierent views of the machining equipment at the CRPP. We use a
small and versatile die-sinking EDM machine, equipped with a generator of the Roboform
type from Charmilles Technologies. The electrode is cylindrical, with a diameter of 3 or
5 mm. In order to better control the localization of the sparks, its tip is conical. The
servo-controlled movement of the electrode is only vertical. The workpiece used in our
experiments is generally a at cylinder, 5 cm in diameter. In order to ush the particles
contaminating the electrode gap during machining, the dielectric can be pumped, and
re-injected into the gap with a shower. No dielectric cleaning is performed during this
closed circuit circulation.
The dielectric, the electrode and the workpiece can easily be changed. We use:
deionized water (typical conductivity 1.5 S/cm), mineral oil (FluX Elf 2, oil for
EDM, viscosity 6) or liquid nitrogen as dielectric. With liquid nitrogen, the work-
piece is placed in a dewar to avoid boiling as much as possible;
electrodes in copper, tungsten, graphite and zinc;
workpieces in W300 steel (AISI Type H11).
The machining process is completely controlled by the generator. It supplies the dis-
charge voltage and current, regulates them, and controls the servomotor for the electrode
displacement. The generator uses principally the measurement of the gap voltage to
regulate and control the process.
25
26 CHAPTER 3. EXPERIMENTAL SETUP AND DIAGNOSTICS
(b)
(c)
(a)
EDM pulse
generator
electric motor
(for servo-controlled
electrode displacement)
pump
dielectric
shower
electrode
workpiece
fibre optics (for
optical diagnostics)
dielectric circuit
EDM machine
control
Figure 3.1: EDM device. (a) General view; (b) EDM machine; (c) electrodes.
Figure 3.2 shows the discharge parameters which can be set with the generator: igni-
tion voltage V , discharge current I, discharge on-time, o-time (pause between the end
of a discharge and the voltage rise for the next one), electrode polarity.
It is impossible to control the pre-breakdown duration, i.e. the time lag between the
voltage application and the breakdown, because it depends on the electrode gap and on
physico-chemical properties of the dielectric. Note that the pre-breakdown duration is also
called ignition delay time or breakdown time lag. The value of the voltage during the
discharge can not be set by the generator either. Its value depends on electrode materials,
but is typically around 2025 V. The values that we can choose with our generator for
V , I, on-time and o-time are given in table 3.1.
The machining mode schematically presented in gure 3.2 is called Isopulse, because
every discharge has the same on-time, independently of the pre-breakdown duration. This
mode is the standard machining mode used in this work. By adding a capacitor in parallel
to the gap, it is possible to use the generator in a capacitive mode, generally used for surface
nishing [106]. The sparks are produced by successive discharges of the capacitor. In this
mode, the polarity is always chosen negative. The discharge on-time and current are not
controlled, and thus can slightly vary from a discharge to the other, in contrary to the
Isopulse mode. Typically, with a 10 nF capacitor, the discharge duration is around 1.5 s
and the current around 6 A.
3.2. ELECTRICAL MEASUREMENTS 27
v
o
l
t
a
g
e
c
u
r
r
e
n
t on-time
(plasma)
off-time
time
pre-breakdown
(stochastic)
?
breakdown
I
V
time
Figure 3.2: Main adjustable discharge parameters in Isopulse mode: V , I, on-time and o-time.
Parameter possible values
V [V] 80, 120, 160, 200.
I [A] 0.5, 1, 1.5, 2, 3, 4, 6, 8, 12, 16, 24, 32, 48, 64.
on-time [s] 0.4, 0.8, 1.6, 3.2, 6.4, 12.8, 25, 50, 100, 200, 400, 800, 1600, 3200.
o-time [s] 0.8, 1.6, 3.2, 6.4, 12.8, 25, 50, 100, 200, 400, 800, 1600, 3200.
Table 3.1: Discharge parameters: possible values.
3.2 Electrical measurements
3.2.1 Discharge measurements
The most basic plasma diagnostics consist in measuring the evolution of voltage and
current during a discharge. The voltage is measured with a dierential probe (SI-9002 by
Sapphire Instruments, 25 MHz), connected in parallel to the electrode gap, as shown in
gure 3.3. We use two current probes depending on the application:
a fast current transformer (FCT from Bergoz Instrumentation, 1.6 GHz) for fast
measurements, typically for the breakdown study;
a DC 50MHz probe (AP015 from LeCroy), for a general characterization of the
discharges.
The probe is placed around the current cable, near to the upper electrode. Both voltage
and current probes are connected to a fast oscilloscope (WavePro 950 from LeCroy, 1 GHz).
workpiece
electrode
fibre optics
dielectric
spark
EDM pulse
generator
G
voltage
probe
current probe
V
to photomultiplier
(or spectrograph)
servo-controlled
displacement
Figure 3.3: Schematic drawing of the diagnostics experimental setup.
Figure 3.4 shows a comparison of measurements with the two current probes. Since
the FCT probe acts as a passive transformer, it dierentiates DC currents as shown in
gure 3.4 (a). On the other hand, this probe is well suited for fast measurements. One
can see in gure 3.4 (b) that the response of the DC probe is slower than that of the FCT
probe.
0
100
200
-2
0
2
0 50 100 150
time [ s] m
0
2
4
6
0 0.1 0.2 0.3 0.4
v
o
l
t
a
g
e

[
V
]
c
u
r
r
e
n
t

[
A
]
0
100
200
- 0.1
time [ s] m
FCT probe
DC probe
FCT probe
DC probe
12 ns
(a) complete Isopulse discharge (b) breakdown phase
Figure 3.4: Comparison of the two current probes. (a) Measurements during a complete Isopulse
discharge; (b) measurements at the breakdown.
3.2.2 Pre-breakdown measurements
For pre-breakdown study, the gap voltage is simply supplied by a standard DC source
rather than by the EDM generator, as shown in gure 3.5. Pre-breakdown currents are in
3.3. PLASMA LIGHT INTENSITY MEASUREMENTS 29
fact weak and fast signals (see section 4.2). The EDM generator produces regular electrical
interferences due to its current regulation system, even before a discharge. The intensity
of these interferences are of the same order of pre-breakdown current and strongly perturb
their measurement. Since the interesting point studied here is pre-breakdown phenomena,
a DC source is sucient to simply maintain a constant electrode gap voltage, without
producing any parasitic signal.
R
DC source
(0 - 1000 V)
manual
displacement
voltage
probe
current probe
V
bubbles
Figure 3.5: Experimental setup for pre-breakdown current measurements.
Without the EDM generator, the upper electrode has to be moved manually, with a
micro-screw system. An additional resistance (typically R = 12 k) is placed in series,
in order to avoid short-circuiting of the DC source when the electrodes come into con-
tact. Both electrodes used are pointed, because this geometry is found to enhance the
occurrence of pre-breakdown current and the bubble generation.
3.3 Plasma light intensity measurements
The temporal evolution of the light intensity emitted during a discharge is measured with
a fast photomultiplier (H6780 from Hamamatsu, spectral sensitivity 200 800 nm, rise
time 0.78 ns). The light is transmitted to the photomultiplier by a quartz bre located
near the plasma (see gure 3.3). Inside the photomultiplier, the photons are transformed
into electrons by a photocathode, and then multiplied by dynodes. The photomultiplier
delivers an output current, which is proportional to the incoming light intensity. To further
improve the sensitivity of the measurement, the output current is amplied again (with a
C6438 amplier from Hamamatsu, DC 50 MHz). Finally, the current is measured with
the oscilloscope.
This measurement is not an absolute intensity measurement, which would require a
specic calibration. This is not necessary, since the interest of this measurement is simply
to determine how the emitted light intensity evolves during the discharge.
3.4 Imaging
Due to the small size of the electrode gap, close images of the electrode region show de-
tails which are not visible to the naked eye. It gives useful information about the EDM
process, especially for its control: inuence of bubbles, short-circuit, localization of the
discharges, behavior of eroded particles, etc. Imaging of the plasma itself is also interest-
ing, particularly to see the evolution of the plasma size and shape, from the breakdown
to the afterglow. Analysis of the plasma development at the breakdown requires a fast
camera, because of the rapidity of the phenomenon (< 100 ns).
The most convenient way to acquire images of EDM plasmas is to use an endoscope.
This instrument, well-known for medical applications, is composed of thousands of in-
dividual optical bres, which make the sampling of the studied object image. The
exibility of the endoscope makes it very easy to position near the electrode gap.
Here we use an endoscope from Myriad Fiber Imaging, 1.5 mm in diameter, composed
of 30000 quartz bres. The endoscope is equipped with a small built-in quartz lens at
its tip, in order to have a magnication of the plasma region. To obtain a compact ge-
ometry, this lens is not a conventional one, but a gradient index lens. The light rays are
bent inside the lens, because it is made of a material with a gradually varying index of
refraction. The minimal working distance of the endoscope is 4 mm, the depth of eld
2.5 15 mm, the eld of view 30
in air and 20 25
in water. An example of an image

obtained with the endoscope is given in gure 3.6. The individual bres composing the
instrument can be seen on the zoomed image.
Figure 3.6: Image of a ruler graduated every 100 m, taken in air with the endoscope at a
distance of 4 mm.
Figure 3.7 shows the experimental arrangement for imaging. The endoscope is directly
immersed in the dielectric. Due to its short working distance, it can be placed a few
3.4. IMAGING 31
millimeters from the spark, assuring therefore a good image magnication and small light
absorption by the dielectric. When studying the EDM process in general, the endoscope
is placed a few centimeters from the electrode gap and slightly above, in order to have
a general view of the electrodes and their surrounding environment. To observe the
electrodes, an external lighting is necessary to illuminate them, because our endoscope is
not equipped with a source for illumination. We use two halogen lamps for this purpose.
For plasma imaging, the endoscope is placed much closer to the plasma (5 mm typically),
and the at workpiece is replaced by a pointed one. With this point-to-point geometry,
the plasma is thus localized as much as possible.
endoscope
(a) (b)
to camera
lens
Figure 3.7: Schematic drawing of the experimental setup for imaging. (a) Imaging of the EDM
process; (b) plasma imaging.
The endoscope is directly connected to a CCD camera (SensiCam Fast Shutter from
PCO Imaging, 12 bit, exposure time 100 ns 10 ms, 1280 768 pixels). The image
focusing is done by a lens with adjustable focal length, placed between the end of the
endoscope and the CCD camera.
For time-resolved imaging, i.e. acquisition of successive images of short exposure time
taken at dierent moments of the discharge, the camera can be gated. In this mode,
the exposure begins with an external trigger signal, produced by a pulse generator. The
delay between the breakdown (t = 0) and the beginning of the exposure is set by the pulse
generator. The t = 0 trigger is obtained by measuring the current rise or the voltage drop
which occur at the breakdown.
For fast imaging, we use another camera, an intensied CCD camera (ICCD PI-MAX
camera from Princeton Instruments, 16 bit, < 2 ns gating, 1024 1024 pixels). Since
short exposure time implies low measured intensities, light intensication is required.
This is performed in ICCD cameras by a micro-channel plate (MCP) placed just behind
a photocathode. The electrons created in the photocathode are multiplied in the MCP,
and then re-transformed into photons by a uorescent screen. The MCP is also used as a
shutter for gated exposure. Since the electrons in the MCP can be abruptly blocked by
reversing the electric eld, fast gating is achievable with ICCD cameras. For time-resolved
imaging with the PI-MAX camera, the delay from the breakdown can be set without a
pulse generator. The t = 0 trigger is directly sent to the camera controller, and the delay
is set by software.
3.5 Optical emission spectroscopy
Optical emission spectroscopy (OES) is a classical plasma diagnostic. Several impor-
tant plasma measurements can be performed with this non-intrusive method: chem-
ical composition, temperatures, densities, electric and magnetic elds present in the
plasma [107109].
3.5.1 Principles
OES consists in the spectral analysis of the light emitted in the visible region by the
plasma. The light is dispersed in a spectrograph, generally by a grating, and detected by
photodiodes, a photomultiplier or a CCD camera.
Atoms, ions and molecules emit light with a discrete spectrum when they pass from
an excited state to a state of lower energy, i.e. when an electron makes a transition from
an upper energy level E
i
to a lower energy level E
j
inside the atom, ion or molecule. The
wavelength of the spectral line emitted during this transition will be
=
hc
E
i
E
j
, (3.1)
where h is Plancks constant and c the speed of light. Since the atomic or molecular
quantum energy levels are dierent for every element or molecule, the discrete spectrum
emitted by these electronic transitions, called bound-bound transitions, is like a charac-
teristic signature of the emitting species. The plasma composition can thus be determined
with OES by identifying the dierent emitted spectral lines.
While the wavelength of a spectral line is only determined by the type of the emitting
species, the intensity of this line depends on the abundance of the emitter, i.e. on its
density, and also on the population of the excited state considered. This means that line
intensities are dependent on the plasma temperature.
The energy levels of the hydrogen atom can be exactly calculated with quantum me-
chanics. The main levels are the eigenvalues of the Hamiltonian describing the electron
of the hydrogen atom, and are given by
E
n
=
13.6 eV
n
2
, n = 1, 2, 3 ... (3.2)
The energy E = 0 corresponds to the ionization threshold. A special terminology is used
for the lines emitted by atomic hydrogen. They are given in gure 3.8.
A plasma can also emit light with a continuous spectrum. This occurs when free
electrons recombine in atoms or molecules: they pass from a state E > 0 in the energy
continuum (i.e. with no dened energy) to a discrete level inside an atom or a molecule.
These transitions are called free-bound transitions. Continuous light is also emitted when
free electrons lose energy by passing near an ion or an atom. This is the well-known
Bremsstrahlung, or free-free radiation.
3.5. OPTICAL EMISSION SPECTROSCOPY 33
Balmer series Paschen series Lyman series
- 13.6
- 3.39
- 1.51
- 0.85
- 0.54
0
n=1
n=2
n=3
n=4
n=5
eV
H
Figure 3.8: Main quantum levels and emission lines of the hydrogen atom.
3.5.2 Electron temperature measurement
To determine the dierent temperatures of the species present in a plasma, several spec-
troscopic methods exist [107, 108]. In practice, their respective applicability depends on
the type of spectrum emitted by the plasma studied: atomic lines, ionic lines, molecular
bands, continuum emission... Most of these methods require the assumption of thermal
equilibrium, or at least of local thermal equilibrium (LTE). They need a theoretical re-
lation between the populations of the various excited states and the temperature or the
pressure, for example. In case of LTE, this relation is simply given by equilibrium statis-
tical mechanics or thermodynamics. In case of non-equilibrium plasmas, it is much more
complicated to develop a theoretical model.
Here, for the electron temperature determination in EDM plasmas, we use the two-line
method. This method, which is a particular case of the Boltzmann plot method, is based
on measurements of the intensity ratio of two spectral lines, emitted by atoms or ions of
the same element [107, 108]. Since it is a relative and not an absolute intensity measure-
ment, the calibration of the absolute spectral sensitivity of the optical arrangement is not
necessary.
If we assume that the plasma is in a state of LTE (the validity of this assumption for
EDM plasmas will be discussed later, in 6.1.7), the distribution of atoms and ions in
the dierent excited states is described by the Boltzmann distribution. Thus, the density
n
i
of a certain type particle, atom or ion, in an excited state i is given by
n
i
=
n
0
g
i
Z
exp
_
E
i
k
B
T
e
_
, (3.3)
where n
0
is the total density of the considered particles in the plasma, g
i
the statistical
weight of the excited level and E
i
its energy, Z the partition function, k
B
the Boltzmann
constant and T
e
the electron temperature. In case of LTE, T
e
is also equal to the ion
temperature and to the gas temperature; T
e
could thus be replaced in (3.3) with a unique
plasma temperature T.
The emission coecient of the line emitted by the transition from the upper state i
to a lower state j is
=
1
4

hc
A
ij
n
i
, (3.4)
where is the line wavelength and A
ij
the probability for the transition from the state i
to the state j. Replacing n
i
by its expression given in (3.3), we obtain
=
hc
4

n
0
Z

g
i
A
ij
exp
_
E
i
k
B
T
e
_
.
Considering now two lines
1
and
2
emitted by the same species, i.e. by atoms or
even charged ions of the same element, the ratio of their intensities is given by
I
1
I
2
=

1
2
=
g
1
A
1

2
g
2
A
2
exp
_
E
2
E
1
k
B
T
e
_
,
where g
i
, A
ij
and E
i
for the line k (k = 1, 2) are noted g
k
, A
k
and E
k
respectively.
Extracting T
e
from the preceding equation, we nally obtain
T
e
=
E
2
E
1
k
B
_
ln
_
I
1
1
g
2
A
2
I
2
2
g
1
A
1
_
_
1
. (3.5)
The ratio I
1
/I
2
is experimentally determined from an emission spectrum. The electron
temperature can be thus calculated with equation (3.5) if E
k
,
k
, g
k
and A
k
are known
values for the two lines.
Practically, the choice of the two lines is of importance. Since the relative error on the
temperature is
T
e
T
e
=
k
B
T
e
|E
2
E
1
|

(I
1
/I
2
)
I
1
/I
2
according to (3.5), it is crucial to choose two lines with a dierence |E
2
E
1
| as large
as possible. Moreover, the chosen lines have to be intense and isolated from other lines,
in order to avoid overlapping and to minimize (I
1
/I
2
). It is also desirable that
1
and
2
are close to each other. In this case, the absorption by the dielectric and the
spectral sensitivity of the optical arrangement can be considered equal for the two lines.
No spectral correction is then required for the calculation of I
1
/I
2
. Finally, the lines have
to be non-resonant ones, to avoid self-absorption. It has to be emphasized that with the
two-line method, a minimum error of 10% on T
e
is expected [107].
As we will see in chapter 6, copper lines are emitted by EDM plasmas when using
a copper electrode. Three atomic copper lines full the conditions cited above. Their
parameters are given in table 3.2. Since |E
2
E
1
| must be large, we can use the pairs
(a)(b) and (a)(c) for the determination of T
e
with the two-line method.
Line [nm] upper state (i) lower state (j) E
i
[eV] E
j
[eV] g
i
g
j
A
ij
[10
8
s
1
]
(a) 510.554 4p
2
P
3/2
4s
2 2
D
5/2
3.817 1.389 4 6 0.02
(b) 515.324 4d
2
D
3/2
4p
2
P
1/2
6.191 3.786 4 2 0.6
(c) 521.820 4d
2
D
5/2
4p
2
P
3/2
6.192 3.817 6 4 0.75
Table 3.2: Line parameters of the Cu I lines used for electron temperature determination [110].
In practice, to calculate the line intensities from a spectrum, the background has to be
subtracted rst. Then the line shapes are tted with Lorentzian functions (see 3.5.3),
and eventually deconvolved if there are adjacent lines. The line intensity I is not simply
the maximum value of the peak, but the whole area under the Lorentzian t.
3.5.3 Electron density measurement
With the electron temperature T
e
, the electron density n
e
is another important physical
parameter characterizing a plasma, which can also be determined with optical emission
spectroscopy. Again, there are several methods based on measurements of line and con-
tinuum intensities, but also on measurements of line prole [107, 108, 111]. The electron
density has in fact a direct inuence on line shapes, while they are relatively insensitive
to electron and ion temperatures. Spectral lines are broadened and shifted from their
theoretical position, increasingly with the electron density. The inuence of the density
on line broadening can be qualitatively explained as follows: the collision frequency will
increase with density, and so the lifetime of atoms in excited states will decrease due to
collisions. By the Heisenberg principle, this will increase the uncertainty on the energy
levels, i.e. the uncertainty on the emitted wavelength. Spectral lines are thus more broad-
ened in a dense plasma.
There are several broadening mechanisms:
the natural broadening, due to the niteness of the lifetime of an atom in an excited
state;
the broadening due to collisions, such as:
the pressure broadening, caused by collisions between emitters and neutral
atoms or molecules;
the Stark broadening, caused by collisions between emitters and charged per-
turbers. The atomic quantum levels of emitters are modied by the Stark
eect, due to the electric eld created by the colliding ions or electrons;
the Doppler broadening, due to the thermal movements of the emitters along the
line of sight.
Stark broadened lines are experimentally valuable for the density determination, be-
cause the Stark broadening is particularly sensitive to the electron density. In a dense
plasma, and this is the case of EDM plasmas as we will see in chapter 6, this broadening
mechanism will dominate by far all the others [34, 111]. For more details on the Stark
eect and its consequences on line emission, see appendix A.
The Doppler broadening leads to a Gaussian line shape, while processes involving par-
ticle impacts, such as the Stark broadening, lead to a line shape close to a Lorentzian. If
both eects take place, the line prole will be a Voigt prole, i.e. a combination of the
two. Since Stark broadening prevails in EDM plasmas, our lines have a shape close to a
Lorentzian.
In our case, we use the H
line emitted by atomic hydrogen at 656.28 nm to deter-

mine the electron density. Along with H
, this line is a classical candidate for density

measurement, because the hydrogen atom is particularly sensitive to the Stark eect.
Furthermore, Stark broadening and shift of the hydrogen Balmer lines have been well
studied since many years, because this type of diagnostic is particularly important in
fusion experiments and astrophysics, for example. Thus, numerous theoretical and ex-
perimental works have been published about Stark broadening and shift of the H
line,
especially for electron density calculations from full width at half maximum (FWHM)
measurements [112115] and from shift measurements [113117].
Computation of spectral lineshapes is quite dicult, because collisions of both elec-
trons and ions with the emitter have to be taken into account as the broadening mecha-
nism. But in recent years progress has been made to include several dynamic eects in
calculations, especially ion dynamics. Here we used the theories of Gigosos and Cardenoso
on broadening [112] and of Oks on shift [115] to calculate the electron density.
Figures 3.9 (a) and (b) show respectively the simulation results for the dependence of
the FWHM and of the shift of H
on the electron density. Though the plasma temperature

has a weak inuence on the line prole, the simulations of FWHM are calculated for
dierent temperatures and for dierent reduced mass of the emitterperturbing particle
pair. Here, we use the calculation for a plasma of 10000 K (which is close to the actual
EDM plasma temperature, see chapter 6), and with = 1 (i.e. hydrogen atoms perturbed
by heavy particles). For the shift simulation, we use calculations made for plasmas with
temperatures from 13000 to 20000 K. EDM plasmas are cooler, but simulation results
for lower temperature are not available. The determination of n
e
is more precise with the
FWHM measurements, because the broadening of H
is experimentally easily measurable

for densities at which the shift is still very low. At 5 10
17
cm
3
for example, the FWHM
is 3 nm but the shift is only 1.7 .
By tting the simulation results of the gure 3.9, we obtain two equations giving the
electron density as a function of the H
FWHM and of the H
shift. Thus, to calculate

n
e
from our FWHM and shift measurements, we use the following relations:
n
e
= 8.8308 10
16
W
_
1.6005
(3.6)
10
14
10
16
10
18
0.01 0.1 1 10
FWHM of the H line [nm]
a
e
l
e
c
t
r
o
n

d
e
n
s
i
t
y

[
c
m
]
-
3
simulation
fit
10
18
210
18
3 10
18
4 10
18
5 10
18
5 10 15 20
shift [] of the H line
a
simulation
fit
e
l
e
c
t
r
o
n

d
e
n
s
i
t
y

[
c
m
]
-
3
(a) broadening (b) shift
Figure 3.9: Relation between the electron density and: (a) the full width at half maximum
(FWHM) of the H
line (simulation results for a 10000 K plasma with = 1, according to [112]);

(b) the shift of the H
line (simulation results for plasmas from 13000 to 20000 K, according to

Oks [115]).
with n
e
in cm
3
and the FWHM
W
in nm, and
n
e
= 10
17
(1.3591 + 2.0757
S
+ 0.0037
2
S
) (3.7)
with n
e
in cm
3
and the shift
S
in .
In practice, we have to take into account the broadening caused by the spectrograph.
This experimental broadening is measured with a laser, which theoretically emits a single
wavelength. The H
line is treated as the copper lines for T

e
determination: background
subtraction and Lorentzian t. Then, the line FWHM is obtained by subtracting the
experimental broadening from the FWHM of the Lorentzian t.
3.5.4 General experimental setup
The general experimental setup for OES is presented in gure 3.10. The emitted light is
collected by a quartz bre, 1 mm in diameter. The bre is immersed in the dielectric,
located a few millimeters from the spark to optimize the collected light intensity and
to reduce absorption from the dielectric. The light is then dispersed by a spectrograph
equipped with gratings, and detected by a CCD camera or photodiodes. Finally, the
signal is digitalized and treated with a computer.
We use dierent spectrographs and detectors depending on the application:
for time-integrated spectroscopy, a 0.275 m spectrograph (SpectraPro 275 from
Acton Research Company) with three gratings (150, 600 and 1200 grooves/mm),
spectrum
CCD
camera
fibre optics
spectrograph
gratings
Figure 3.10: Schematic drawing of the experimental setup for optical emission spectroscopy.
equipped with a CCD camera (TE/CCD-1024E from Princeton Instruments, 16 bit,
1024 256 pixels);
for time-resolved spectroscopy, the same spectrograph SP275 as above, equipped
with a detector composed of an array of 700 gated photodiodes (OMA III system
from EG&G PARC, 14 bit, 100 ns 10 ms gating);
for spatially-resolved spectroscopy, a 0.75 m imaging spectrograph (SpectraPro 750i
from Acton Research Company) with three gratings (150, 600 and 1800 grooves/mm),
equipped with a CCD camera (Spec-10:400B from Princeton Instruments, liquid N
2
cooled, 16 bit, 1340 400 pixels);
for time- and spatially-resolved spectroscopy, the same spectrograph SP750i as
above, equipped with the ICCD PI-MAX camera used for fast imaging (see sec-
tion 3.4).
The resolving power and the spectral range for each conguration are given in table 3.3.
spectrograph SP275 spectrograph SP750i
CCD camera photodiodes CCD camera ICCD camera
150 600 1200 150 600 1200 150 600 1800 150 600 1800
resolving power [nm] 1.25 0.35 0.17 2.8 0.8 0.25 0.5 0.12 0.03 0.33 0.09 0.02
spectral range [nm] 650 160 70 450 100 45 240 60 15 120 30 10
Table 3.3: Resolving power and spectral range for the dierent experimental congurations
(spectrograph / detector / grating).
The spectral sensitivity of the optical system (bre + spectrograph + detector) has
been measured with a reference light source, whose real emission spectrum is known.
The sensitivity is calculated by dividing the spectrum of the reference source measured
through the optical system by the real spectrum. Figure 3.11 shows the results for the
spectrograph SP275 with the CCD camera.
0
50
100
150
400 600 800
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
(a) real spectrum
0
200
400
400 600 800
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
(b) measured spectrum
0
0.2
0.4
0.6
0.8
1
300 400 500 600 700 800 900
wavelength [nm]
s
p
e
c
t
r
a
l

s
e
n
s
i
t
i
v
i
t
y
(c) spectral sensitivity ( ) (b) (a)
= /
Figure 3.11: Spectral calibration of the optical system composed of the quartz bre, the spectro-
graph SP275 and the CCD camera. (a) Real spectrum of the reference light source; (b) spectrum
measured through the optical system; (c) spectral sensitivity (normalized).
The spectral sensitivity is maximum between 600 and 750 nm, and falls o below
500 nm and above 900 nm, mainly due to the quantum eciency of the CCD camera. For
all spectrograph-detector congurations, the optical system is sensitive to light emitted
roughly between 300 and 1000 nm.
The spectral calibration was made under dierent conditions: without dielectric, with
clean oil and with oil contaminated by eroded particles. Though the light absorption
by the dielectric increases drastically in contaminated oil, the relative spectral sensitivity
remains approximately the same without oil, in clean or contaminated oil.
Though the plasma light is intense enough for imaging (even with short exposure
time < 1 s), it is not possible to perform emission spectroscopy of a single spark. For
imaging, the total plasma light is concentrated on a few pixels of the CCD chip. But for
spectroscopy, the light is dispersed in the spectrograph and spread over the whole chip.
The intensity of a single spark is then far too weak to be measured. The light has to be
accumulated over thousands of discharges.
3.5.5 Experimental setup for time-resolved spectroscopy
Time-resolved spectra are obtained with gated photodiodes. As for time-resolved imaging,
a pulse generator is used to generate the detector gate. The light is detected only during a
short period of the discharge, which is the time resolution of the measurement. Changing
the delay between the beginning of the discharge and the beginning of the light acquisition,
we obtain a succession of spectra. Each spectrum is obtained by accumulating the light
of the same part of thousands sparks, as shown in gure 3.12.
delay
exposure time
v
o
l
t
a
g
e
c
u
r
r
e
n
t
d
e
t
e
c
t
o
r

g
a
t
e
time
"t=0" trigger (V or I)
time
time
Figure 3.12: Timing diagram for time-resolved spectroscopy.
3.5.6 Experimental setup for spatially-resolved spectroscopy
Figure 3.13 shows the experimental setup for spatially-resolved optical emission spec-
troscopy.
In order to have a spatial sampling of the emitted light, the magnied plasma image
captured with the endoscope is projected onto an in-line array of 16 bres, arranged over
2 mm. In this way, each bre collects the light coming from a dierent zone of the emitting
region. With a typical endoscope-plasma distance of 5 mm, the achievable resolution is
20 m/bre.
The bre bundle brings the sampled light into the imaging spectrograph. The 16
dierent spectra are recorded simultaneously with the CCD camera. Due to the small
size of the light emitting region (see section 5.2), this arrangement is the easiest way
to perform the spatial sampling. In principle, it could be done directly without the
endoscope but with an array of bres directly located near the plasma, but in this case
the miniaturization of the bres in the bundle is then an issue.
3.6. EXPERIMENTAL DIFFICULTIES 41
endoscope
16 spectra
image
lens
bundle of 16 in-line fibres
imaging
spectrograph
with CCD camera
Figure 3.13: Experimental setup for spatially-resolved spectroscopy.
Since the light acquisition is made over thousands of discharges due to the weak emitted
intensity, it is necessary that the successive plasmas stay located, as far as possible, in the
same position with respect to the endoscope. Otherwise, one bre of the bundle would
sample light coming from dierent zones of the plasma, and the spatial sampling is no
longer valid. As for plasma imaging, we use a point-to-point electrode geometry to localize
the sparks as well as possible. Discharges with high current are also avoided, because they
produce large craters, and so large variations in the plasma localization already after a
few discharges.
For time- and spatially-resolved spectroscopy, the setup is the same as shown in g-
ure 3.13, but the CCD camera is replaced by the ICCD camera, which can be gated.
3.6 Experimental diculties
The experimental investigation of EDM plasmas is not straightforward. Since this work is
precisely an experimental characterization of these plasmas, it is important to emphasize
the practical problems encountered.
The main source of diculties comes from the small size of the plasma. With a typical
electrode gap of 10 100 m, moreover in a liquid environment, it is almost impossi-
ble to apply all the classical plasma diagnostics: electrostatic probes, mass spectrometry,
microwave diagnostics, actinometry... Intrusive and active methods either are not appli-
cable, or too dicult. Optical methods, such as emission spectroscopy and imaging, are
almost the only usable diagnostics.
Furthermore, even the optical diagnostics are dicult to implement in practice. A
spatial characterization of such a small plasma is challenging. In addition to the problems
related to the miniaturization of light acquisition instruments (optical bre, endoscope),
the weak intensity emitted by the plasma is also an issue. It is necessary to have sensitive
detectors and to make the measurements in a dark environment. Even so, measurements
of a single discharge remains extremely dicult.
The short duration of the discharges is also problematic. Since the timescale of the
discharges is 1 100 s and the timescale of breakdown phenomena is 10 100 ns typ-
ically, the measuring devices must have a fast response. Related to the rapidity of the
phenomena studied, electrical interference is another problem. In a typical discharge, the
current rises at breakdown from 0 to 12 A in 200 ns, for example. The electromagnetic
waves associated with this sudden current rise perturb all the electrical measurements,
especially weak signals such as photomultiplier signals. Shielding of the measuring instru-
ments becomes of importance.
Another major source of diculties is the poor reproducibility of the discharges, even
with the same experimental conditions (see 5.1.4). The localization of the successive
plasmas is constantly changing, due to the evolution during machining of the dielectric
cleanliness and of the electrodes surface and geometry. Even the plasma properties can
vary, also because of the intrinsic stochastic nature of the breakdown process. In such
conditions, one spark can be quite dierent from another. Since our spectroscopic mea-
surements have been done over thousands of discharges due to the weak intensity of a
single spark, this light accumulation can be viewed as a way to overcome the reproducibil-
ity problem. In other words, our spectroscopic results are describing an average plasma.
Variation in the plasma localization is particularly problematic for time- and spatially-
resolved spectroscopy (section 6.4). For such measurements, we need short exposure times
in order to have a ne temporal resolution. At the same time, the plasma has to be
localized and as stable as possible for the spatial sampling. For a sucient measured
intensity, a short exposure time requires a large number of recorded discharges (typically
100000 for 2 s exposure). But a large number of discharges leads to signicative erosion,
and thus to movements of the plasma. After a certain time of machining, the workpiece
surface is in fact no longer at and presents the beginning of a hole. Then the following
plasmas begin to move around the pointed electrode (they are located where the gap
is the smallest, i.e. between the electrode tip and the side of the newly machined hole).
Figure 3.14 shows an experimental arrangement which minimizes the plasma movements
when performing time- and spatially-resolved spectroscopy.
endoscope : fixed
slow manual
displacement
servo-controlled
displacement
to bundle of fibres,
spectrograph and camera
Figure 3.14: Experimental setup for time- and spatially-resolved spectroscopy.
3.6. EXPERIMENTAL DIFFICULTIES 43
The erosion is considerably reduced by using a workpiece with a large at surface,
which is manually and slowly moved horizontally. Even with a large number of sparks,
the erosion is distributed over the whole surface, which remains roughly at. The plasma
movement is thus minimized, permitting a good spatial sampling with a short exposure
time.
Chapter 4
Pre-breakdown of EDM discharges
The rst of the four chapters presenting experimental results is dedicated to the pre-
breakdown phase. The study of the mechanisms leading to breakdown is physically of
great interest, but it is also important from the industrial point of view. Since no material
removal is performed during this phase, it would be interesting to understand it and to
control its duration, in order to reduce this waste of time as much as possible.
The rst section of this chapter summarizes some observations made about the bubbles
created during the pre-breakdown. Current measurements are presented in the second
section, and nally a short analysis of the pre-breakdown duration is given in the last
section. These results are certainly not exhaustive, but they give a rst insight into
pre-breakdown phenomena.
4.1 Bubbles
Depending on the conditions, small bubbles can be generated at the electrodes as soon as
the gap voltage is applied. One can observe that:
bubbles are created only in water, and not in oil;
more bubbles are generated when the electric eld in the gap is increased. A few
bubbles are already visible with a gap of 15 mm and 80 V, and the emission is
strongly increased by diminishing the gap distance and/or increasing the gap voltage
(see 5.1.2 for images). Note that no bubbles are generated in oil even with a gap
of 100 m and 1 kV;
bubbles are created at the cathode;
the contamination of the water, and consequently its conductivity, has an eect.
For example, very few bubbles are generated in clean conditions (water conductivity
< 3 S/cm).
These observations suggest that the bubbles are produced by electrolysis of the water.
The ow of a small electric current in the water releases gaseous hydrogen and oxygen at
the electrode surfaces, according to the following reactions:
45
46 CHAPTER 4. PRE-BREAKDOWN OF EDM DISCHARGES
2 H
2
O O
2
+ 4 H
+
+ 4 e
(at the anode),

4 H
2
O + 4 e
2 H
2
+ 4 OH
(at the cathode).

The observed inuences of the electric eld and of the water conductivity on bubble
generation is a direct sign that bubbles are related to the ow of an electric current. Since
we observe only bubbles coming from the cathode, they are bubbles containing hydrogen.
The oxygen released at the anode is probably directly consumed for the oxidation of the
anode surface. One can in fact observe that the anode, in copper or in steel, is rapidly
oxidized after a few minutes.
To conrm that electrolysis is really the source of bubbles, a basic analysis of the
bubble gas was performed. Bubbles were collected in a test tube as shown in gure 4.1,
and the gas was tested with a detector sensitive to explosive gases (hydrogen, methane or
propane for example). The detector was quickly saturated already with a small volume
of gas, indicating that the bubbles are certainly made of hydrogen.
collected gas
Figure 4.1: Experimental setup for bubbles collection.
Note that the electric eld responsible for the electrolysis is much higher than the
value simply obtained by dividing the gap voltage by the gap distance. Since we use a
point-to-point geometry, the eld near the electrode tip is strongly enhanced by the point
eect. Figure 4.2 shows a calculation for a gap of 5 mm and 200 V, assuming electrode
tips with a curvature radius of 100 m. The eld at the tip can reach 2300 V/cm, nearly
six times higher than the plane electrode, uniform eld value of 400 V/cm.
4.2 Pre-breakdown current
Before the breakdown, which is characterized by a large voltage drop and current rise,
fast current pulses are sometimes measured. An example of such pre-breakdown current
is given in gure 4.3. The measured current pulses reach a few milliamperes and last
about 520 ns.
4.2. PRE-BREAKDOWN CURRENT 47
0 V
200 V
R = 100 m m
20 V
2
2
0

V
/
c
m
2
3
0
0

V
/
c
m
1 mm
9
0
0

V
/
c
m
Figure 4.2: Increase in the electric eld du to the point eect. Simulation result for a point-
to-point geometry, 100 m curvature radius, 5 mm gap, 200 V, in water: lines of equipotential
and electric eld strength (in grey scale).
0
100
200
0
1
2
-0.6 -0.4 -0.2 0
breakdown
time [ s] m
c
u
r
r
e
n
t

[
m
A
]
v
o
l
t
a
g
e

[
V
]
Figure 4.3: Typical pre-breakdown current (AC component) measured in 1.5 S/cm water (gap
voltage 180 V).
We observe this type of current not necessarily just before a breakdown as in gure 4.3,
but randomly as soon as the gap voltage is applied, provided that the electrode spacing
is small enough. The occurrence of these irregularly-spaced pulses increases when moving
the electrode towards the workpiece. Figure 4.4 shows, on a longer timescale, the mul-
tiplication of these current bursts when reducing the gap distance. A DC component is
measured (electrolysis current), because the water used is not a perfect dielectric due to
its contamination. The water lling the gap acts therefore as a resistance, whose value
is roughly proportional to the gap distance. This explains why we measure a decrease in
the gap voltage and an increase in the DC current, when the gap distance is reduced.
0
50
100
150
200
0
5
10
15
20
0 5 10
time [s]
15 20
v
o
l
t
a
g
e

[
V
]
c
u
r
r
e
n
t

[
m
A
]
DC component
AC component
Figure 4.4: Current bursts measured in water when the gap distance is slowly diminished (the
electrode is moved manually towards the workpiece).
As mentioned in section 2.2, this type of current signal is typically associated with
the propagation of streamers in the gap [31, 34, 63, 66, 72]. Each pulse is due to a sudden
growth of the ionized channel. A small current can be measured, because this propaga-
tion involves a movement of charged species. The electrode polarity should inuence the
shape of the pre-breakdown current signal, since positive and negative streamers propa-
gate slightly dierently. However, no clear eect can be observed in our measurements,
because our electrode geometry is symmetric (point-to-point). To clearly dierentiate
current signals of the two types of streamer, a strongly asymmetric electrode congura-
tion must be used (point-to-plane, for example [63]).
As for the bubble generation, oil and water behave dierently. No AC or DC pre-
breakdown current are measured in oil, for any condition. This remarkable point empha-
sizes the crucial role of gaseous bubbles for the propagation of streamers, as shown also in
other studies [7779]. In water, their initiation and propagation are strongly facilitated by
the pre-existing bubbles, created by electrolysis. On the other hand, the creation of a bub-
ble in oil is more dicult, because electrolysis does not occur. The initiation of a streamer
in oil requires a higher electric eld than that created with our source ( 0.110 MV/cm
is necessary, according to [30, 60, 63, 72]). This explains the absence of pre-breakdown
current in oil in our measurements.
4.2. PRE-BREAKDOWN CURRENT 49
Although the breakdown mechanisms in water and in oil should not be completely
dierent and should follow the general sequence described in section 2.2, they can nev-
ertheless be inuenced by dierent parameters. In water, the breakdown triggering is
certainly enhanced by the presence of bubbles. In oil, there are rather the contaminating
particles that could play this role, by forming a conductive bridge [3,75]. Although both
bubbles and particles certainly inuence the breakdown in water as in oil, the dominant
contribution is not the same in the two types of dielectric.
As expected, the conductivity of the water has an inuence on pre-breakdown current
[65]. Figure 4.5 presents measurements in clean water, in water contaminated with eroded
particles, and in a slightly salted solution.
200
100
0
200
100
0
200
100
0
v
o
l
t
a
g
e

[
V
]
time [ s] m
0 1 2 0 5 10 0 8 16
c
u
r
r
e
n
t

[
m
A
]
0
10
20
30
0
10
20
30
0
10
20
30
time [ s] m time [ s] m
DC component
DC component
DC
component
(a) clean water
1.5 S/cm m
(b) contaminated water
7 S/cm m
(c) salted water
20 S/cm m
Figure 4.5: Eect of the water conductivity on pre-breakdown current: (a) clean water; (b) wa-
ter contaminated with eroded particles; (c) salted water.
Increasing the conductivity naturally also increases the DC current. Therefore, the
bubble generation by electrolysis will be more ecient and pre-breakdown currents more
frequent. This is exactly what we observe in our measurements. In clean water, the DC
current is almost zero and current pulses are seldom measured (typically < 1 pulse every
10 s). Then, the average number of current pulses per time unit increases with the
conductivity.
The addition of external particles or other additives to the dielectric thus certainly
inuences and facilitates the breakdown mechanism. But for a stable machining process,
these additives should not increase the liquid conductivity too much. The liquid has to
keep some of its insulating properties. In the salted solution (c), short-circuits become
very frequent. A water conductivity higher than 20 S/cm will thus lead to an unstable
situation, prejudicial to the quality of the machining process. In real conditions of ma-
chining, the dielectric is always contaminated with particles, despite continuous ltering
and cleaning of the liquid. Figure 4.5 (b) thus reects a more realistic situation than (a)
and (c).
Other studies have shown that light is emitted during the propagation of streamers.
The light intensity follows the same evolution as the current [31,34,63,66,72]. An example
of light emission measured with a photomultiplier during the pre-breakdown is given in
gure 4.6.
100
120
140
0
4
8
0
0.1
0.2
0.3
-0.8 -0.6 -0.4 -0.2 0
time [ s] m
l
i
g
h
t

i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
c
u
r
r
e
n
t

[
m
A
]
v
o
l
t
a
g
e

[
V
]
Figure 4.6: Light emission associated with pre-breakdown current in water (gap voltage 140 V).
Whereas the light at the breakdown is intense (saturation of the detector), the emission
associated with streamers is very weak. Therefore, the gain of the photomultiplier has to
be set high, and this produces considerable electronic noise on the output signal. This
kind of measurement is therefore quite dicult. Nevertheless, one can see that the light
emission is clearly correlated with the current, as expected.
4.3 Pre-breakdown duration
Once the gap voltage is applied, it takes a certain amount of time until a breakdown
occurs. This duration is not constant from a discharge to another, even if they have the
4.3. PRE-BREAKDOWN DURATION 51
same parameters. Statistics of the pre-breakdown duration have been done for dierent
types of discharges. The mean duration diers according to the discharge parameters,
but the duration distribution is always of the same kind. Figure 4.7 shows histograms of
the pre-breakdown duration measured on two types of discharges.
0 10 20 30 40 50 60 70
0
100
200
300
400
500
n
u
m
b
e
r
pre-breakdown duration [ s] m
Weibull distribution fit
0
1000
2000
3000
0 100 200 300
pre-breakdown duration [ s] m
n
u
m
b
e
r
Weibull distribution fit
(a) (b)
50 150 250
mean = 11.3 s m mean = 46.9 s m
Figure 4.7: Histograms of the pre-breakdown duration. (a) Statistics on 8000 similar dis-
charges: 6 A, 8 s on-time, 8 s o-time, 200 V, in water; (b) statistics on 19000 similar
discharges: 6 A, 50 s on-time, 50 s o-time, +200 V, in water.
The distribution of pre-breakdown duration can be well tted with the Weibull distri-
bution [118]. The probability density function of this type of distribution is
P(x|a, b) = a b x
b1
exp(ax
b
) (x > 0) ,
where a and b are constant parameters. The general shape of this distribution is, starting
from zero, an increase to a maximum value, followed by an exponential decay. The Weibull
distribution is known for describing, among others, the repartition of the pre-breakdown
duration in various dielectric liquids [119, 120]. The breakdown process in EDM has thus
a strong stochastic nature, which clearly appears in these measurements. Although we
can optimize discharge parameters and experimental conditions to reduce the mean pre-
breakdown duration (addition of electron scavengers in the dielectric for example), we
will always have distributed values. The poor reproducibility of EDM discharges can
undoubtedly be related to this stochastic aspect of the breakdown.
As we will see in the next chapter, the stochastic nature of the breakdown has a
practical consequence for plasma imaging. Since it is impossible to predict precisely when
the next breakdown will occur, the camera opening for the exposure of a given discharge
can only be triggered with the beginning of this very discharge. We cannot anticipate the
beginning of the next discharge, as could be done with pulsed RF plasmas, for example.
Chapter 5
Imaging diagnostics of EDM discharges
The rst section of this chapter presents images taken from lms that were recorded
during the EDM machining process. Images of the plasma itself are analyzed in the
second section. Finally, the third and fourth sections are respectively dedicated to images
of the plasma at the beginning and at the end of the discharge.
5.1 Imaging of the EDM process
5.1.1 Erosion
Figure 5.1 shows a typical example of what can be observed during machining.
1 2 3 4
5 6 7 8
1 mm
electrode
workpiece
spark
bubbles
eroded cavity
Figure 5.1: Images taken during the machining process (1 ms exposure; Cu electrode, steel
workpiece, 8 A, 50 s on-time, + 200 V, water). Sparks and bubbles are clearly visible. The
duration of the whole sequence is approximatively 40 s.
53
54 CHAPTER 5. IMAGING DIAGNOSTICS OF EDM DISCHARGES
The workpiece surface, completely at on image 1 before machining, is slightly eroded
by the successive sparks. After a few tens of seconds of machining, i.e. after a few
hundreds of thousands of sparks, a small cavity is formed (image 8). Some of these sparks
are visible on images 2, 4, 5 and 7. The heat released by the discharges generates vapor
bubbles close to the electrode gap. They can be clearly seen on images 3, 6 and 7 for
example.
5.1.2 Bubbles
Dierent types of bubbles can be observed. As mentioned in section 4.1, hydrogen bubbles
are released during pre-breakdown in water by electrolysis. During machining, the heat
of the plasma is generating vapor bubbles, in water as in oil. Images of these dierent
kinds of bubbles are given in gure 5.2.
(a) H bubbles during
in water
2
pre-breakdown
(b) vapor bubbles during
erosion in water
(c) vapor bubbles during
erosion in oil
1 mm 1 mm
electrode
1 mm
electrode
Figure 5.2: Dierent types of bubbles generated during the EDM process: (a) small hydrogen
bubbles created by electrolysis during pre-breakdown in water (large and small gap, + 200 V);
(b) water vapor bubbles thermally induced during machining (8 A, 50 s); (c) oil vapor bubbles
thermally induced during machining (6 A, 6 s).
By zooming on the images of gure 5.2 (a), the diameter of the electrolysis bubbles can
be roughly estimated around 30 100 m. The generation of these hydrogen bubbles is
strongly enhanced with an increase in the electric eld. With a small gap (bottom image
of gure 5.2 (a)), one can in fact observe numerous bubbles ejected from the gap with
high velocity, leading even to turbulent phenomena (vortices).
5.1. IMAGING OF THE EDM PROCESS 55
On the other hand, the thermally-induced vapor bubbles are much larger. Their
diameter is 500 m, in water as in oil (gure 5.2 (b) and (c)). When machining in water,
both hydrogen and vapor bubbles are created. This can be seen on gure 5.1, image 6 for
example: two large vapor bubbles are visible, along with several small hydrogen bubbles
going up on both sides of the pointed electrode. In oil, only large vapor bubbles can be
observed.
5.1.3 Short-circuit
Imaging permits also the observation of undesirable phenomena, such as the short-circuit
of the electrodes. The phenomenon shown in gure 5.3 has been observed both in water
and in oil.
1 2 3 4
5 6 7 8
hot spot
(arc?)
vapor
bubble
new
bubble
electrode
workpiece
500 m m
Figure 5.3: Short-circuit, creation of a stable hot spot and continuous generation of large
growing vapor bubbles in oil. The duration of the whole sequence is approximatively 15 s
(images: 9 ms exposure).
A short-circuit occurs when the two electrodes come into contact, or if a conducting
debris bridges the gap. If the electrodes are not separated but kept in contact, the current
ow heats them by Joule eect. With a point-to-point geometry as in gure 5.3, a hot
spot is easily formed between the electrodes, because the current density becomes very
high in this region due to their small contact area. The heat released by this hot spot
generates a vapor bubble (image 2), just as the sparks do during machining. On image 3,
the inter-electrode distance has been slightly increased. The hot spot becomes bright and
is probably a small arc. The heat release is then more intense, leading to a growth of
the vapor bubble, which remains attached to the hot spot (images 3 and 4). Once the
bubble reaches a critical size, the buoyancy force is large enough to detach the bubble
from the hot spot (image 5). A new bubble is immediately formed, and grows just like
the preceding one (images 6, 7 and 8). This cycle of bubble generation is very stable and
will continue repeatedly, unless the electrodes are clearly separated. On images 7 and 8,
the inter-electrode distance is further increased. The arc then grows and becomes very
bright (image 8). The arc suddenly disappears as soon as the generator can not sustain
it anymore, due to a too large gap distance.
5.1.4 Reproducibility
To illustrate the problem of the reproducibility of the discharges (discussed in section 3.6),
gure 5.4 shows several images of discharges with the same parameters.
electrode
workpiece
spark
500 m m
Figure 5.4: Images of similar discharges, illustrating the problem of reproducibility (9 ms
exposure; 12 A, 50 s, oil). The electrode outlines are drawn.
Even with a point-to-point geometry, the plasma localization strongly changes from
one discharge to another. Furthermore, the size of the sparks, and consequently their
physical properties, also vary. Therefore, all our measurements were repeated several
times. Numerous plasma images were acquired, and the most representative ones are
presented in the following sections.
5.2 Plasma imaging
5.2.1 Evolution of the plasma light intensity
Before analyzing images of EDM plasmas, it is important to study the evolution of the
plasma light intensity during the discharge. Two examples of such measurements are
given in gure 5.5, along with voltage and current measurements.
5.2. PLASMA IMAGING 57
0
2
4
6
0 2 4 6 8
0
20
40
60
80
100
time [ s] m
c
u
r
r
e
n
t

[
A
]
l
i
g
h
t

i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
v
o
l
t
a
g
e

[
V
]
0
0.4
0.8
1.2
(a) short discharge (b) long discharge
0
4
8
0
100
200
0
0.8
1.6
0 20 40 60 80 100
time [ s] m
0.4
1.2
2
6
50
150
Figure 5.5: Evolution of the plasma light intensity, of the current and of the voltage during a
single discharge: (a) short discharge (4 A, 8 s, water); (b) long discharge (8 A, 100 s, water).
Two phases can be distinguished: the breakdown phase and the discharge phase. The
major part of the light is emitted during the breakdown phase, i.e. approximately during
the rst 500 ns after the breakdown (see gure 5.5 (a)). During the discharge itself,
the light emission is 10 to 150 times less intense. In gure 5.5 (b), the rst light peak
seems to last much longer than 500 ns. This is an artefact due to the saturation of the
photomultiplier. For this measurement, the gain of the detector has in fact been set higher
in order to better visualize the light emission during the discharge phase. The intensity
of the rst peak is thus in reality much higher and lasts less than 1 s.
The light emission during the discharge phase becomes much more intense when in-
creasing the discharge current. Furthermore, the oscillations observed on light intensity
measurements in gure 5.5 (b) are correlated to those on current (the current waveform
is caused by the regulation system of the generator). This shows that these two physical
values are unsurprisingly interdependent.
From the erosion point of view, the rst microsecond seems to be very dierent from
the rest of the discharge. The high intensity of the rst light peak suggests that a massive
ow of energy is brought to the workpiece during the breakdown phase. Due to the short
duration of this event (< 1 s), it is probable that only the workpiece surface is aected by
this primary energy ow, while the workpiece bulk remains unaected. A burst of energy
will in fact directly evaporate only the very supercial layers of the workpiece. The bulk is
not aected, because it can not absorb this high amount of energy by conduction on such
a short time scale. The bulk will be heated by the plasma during the discharge phase.
Like the workpiece, the electrode is probably also aected by this primary energy ow.
A small supercial erosion of the electrode during the breakdown phase could be consistent
with the well-known fact that wear mostly occurs at the beginning of the discharge.
The fast and supercial erosion process occurring at the breakdown could be similar
to the thermal runaway of protrusions on arc cathode surfaces (see 2.1.3). Protrusions
can be evaporated or even explode if the input power is higher than what the electrode can
evacuate by conduction. The typical time scale of fast evaporation processes is 1100 ns,
and below 1 ns for explosive processes [56]. Interest in using very short discharges could
thus arise in EDM, considering that such fast eroding mechanisms will produce surfaces
with very low roughness. A new type of machining could perhaps be developed, with
generators capable of producing powerful ultra-short discharges.
5.2.2 Typical plasma image
Contrary to the images of the section 5.1, plasma images are acquired without any external
lighting. The measured light comes only from the plasma itself. Figure 5.6 shows a typical
image of an EDM plasma. The electrodes have been outlined because they are not visible
without the external lighting.
200 m m
50 m m
(a) image
(b) contour plot
(c) vertical profile
-200 -100 0 100 200
distance from center [ m] m
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
Figure 5.6: (a) Typical plasma image (5 s exposure, 5 s delay after breakdown; 24 A, 100 s,
oil), the position of the electrodes is drawn; (b) contour plot of (a); (c) intensity prole of (a)
along the vertical axis.
The light emitting region is generally round or oval. As we will see later, its diameter
increases with the discharge current. The light mostly originates from a broader region
than the gap itself, i.e. the discharge excites a broad volume around the electrode gap.
In order to better analyze the images, it is useful to view them as contour and prole
plot, as shown in gure 5.6 (b) and (c). The dots in the center of gure 5.6 (b), and
the irregularities in the center of gure 5.6 (c), are not real but due to the individual
sampling by the endoscope bres. There are indeed small spaces between the bres in the
endoscope which do not bring the light to the camera. The brightest zone is the central
5.2. PLASMA IMAGING 59
region (electrode gap) and the light emission decreases rapidly with distance: at 40 m
from the center, the intensity is halved.
5.2.3 Plasma evolution
The evolution of the plasma size is another delicate point. Care has to be taken for
the image treatment and interpretation. First, the light emitting region (excited region)
could be broader than the plasma itself (ionized region), and the light could partly be
scattered from the liquid in the surrounding of the electrodes. Secondly, the images have
to be normalized in intensity. If not, the plasma seems to be bigger during the rst
microsecond, simply because the light intensity at this stage is much higher. Finally, we
cannot acquire more than one image during a single discharge, because the readout time
of the CCD camera is much longer than the discharge on-time. Thus, each image will
be acquired during a dierent discharge. Since the discharges are poorly reproducible,
the comparison between the images becomes dicult. However, gure 5.7 shows plasma
images at dierent stages of the discharge.
0 to 5 s m 5 to 10 s m 10 to 15 s m 20 to 25 s m 50 to 55 s m 90 to 95 s m
200 m m
Figure 5.7: Plasma images at dierent times during the discharge (5 s exposure, variable delay
after breakdown; 24 A, 100 s, water). The images are normalized in intensity. Each image is
obtained during a dierent discharge.
There is a slight growth of the emitting region, but its size remains fairly constant
during the whole discharge. Similar measurements have been reported elsewhere [27,121].
The emitting region was also found to grow with time. Even if the light emitting region
could be larger than the plasma itself, it is likely that these images reect an increase in
the plasma size during the discharge, as expected from simulations [40, 122].
Other results conrm this conclusion. Measurements of crater radius on the cathode
surface have been made for dierent discharge on-time [123]. The crater radius r
c
(t),
supposed to be close to the plasma radius r
p
(t), is found to increase with t
0.2
. In previous
models, r
p
(t) was assumed to grow with t
3/4
[40], or with tanh(t) [124]. We will not
propose any relation here. In our opinion, the reproducibility of our images is not good
enough to propose any numerical relation, which would be an over-interpretation of the
measurements.
5.2.4 Eect of the discharge current
As shown in gure 5.8, the diameter of the light emitting region increases with the dis-
charge current, being roughly from 150 to 400 m for a current increase from 3 to 48 A.
Thus, the plasma size most likely also increases with the discharge current. Again, this
can be correlated with the fact that discharges with high current produce larger craters
than discharges with low current.
3 A 6 A 12 A 24 A 48 A
200 m
185 m 210 m 270 m 275 m 410 m
Figure 5.8: Plasma images for dierent discharge currents and estimation of the emitting region
diameter (5 s exposure, 10 s delay after breakdown; 100 s, water). The images are normalized
in intensity.
5.2.5 H
emission
As we will see in chapter 6, the hydrogen H
line is the dominant contribution in EDM

plasma line emission. Intensity evolution and imaging of the H
emission has been mea-

sured by using a band-pass lter centered around 656 nm. The lter is placed between
the optical bre and the photomultiplier for intensity measurements, and between the
endoscope and the camera for imaging. Results are given in gure 5.9.
time [ s] m
c
u
r
r
e
n
t

[
A
]
l
i
g
h
t

i
n
t
e
n
s
i
t
y
[
a
r
b
.

u
n
i
t
]
v
o
l
t
a
g
e

[
V
]
(a) intensity evolution (b) image
0
100
200
0
5
10
15
0
0.5
1
1.5
0 20 40 60
100 m m
50 m m
-200 0 200
distance from center [ m] m
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
10 30 50
150
50
Figure 5.9: Emission of the H
line: (a) intensity evolution; (b) typical image (5 s exposure,

0 s delay), contour plot and vertical intensity prole. Discharge parameters: 12 A, 50 s, water.
5.3. BEGINNING OF THE DISCHARGE : FAST IMAGING 61
The intensity evolution of this line is similar to the evolution of the total emission,
given previously in gure 5.5. The major part of the light is emitted during the breakdown
phase. Similarly, the images obtained with the H
lter are very similar to those obtained

without it (see gure 5.6). The emitting region has the same shape, size and intensity
proles. We can thus conclude that the H
line is emitted by the whole plasma volume.

5.3 Beginning of the discharge : Fast imaging
Figure 5.10 shows fast imaging of the rst 300 ns of the discharge, obtained with the ICCD
camera. This time interval corresponds to the emission of the rst light peak and to the
rst fast current rise (see gure 5.5 (a)). This interval is thus important in the plasma
evolution: it is during the very beginning of the discharge that the plasma is forming and
that a high amount of energy is brought into it.
50 to 100 ns
100 to 150 ns
150 to 200 ns
200 to 250 ns
250 to 300 ns
time [ s] m
0 0.5 1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
exposure
V/200 [V]
I/8 [A]
Figure 5.10: Fast imaging of the discharge beginning along with current and voltage evolution
(50 ns exposure, variable delay; 6 A, water). The images are normalized in intensity. Each image
is obtained during a dierent discharge.
The exposure time is 50 ns for each image, with variable delays. There is always a 50 ns
intrinsic delay between the trigger signal (current rise at the breakdown) and the camera
opening, due to gating electronics and cables. Since we cannot anticipate the breakdown
for the camera triggering as previously mentioned in section 4.3, the rst image can only
be acquired 50 ns after the breakdown.
Between 100 and 200 ns after the breakdown, the current reaches its maximum. The
light intensity is also at its maximum. However, on the plasma images, no clear evolution is
visible from 50 to 300 ns: size and geometry remain constant. Images obtained with delays
longer than 300 ns are not shown here, but they are similar to those of the gure 5.10.
This shows that the plasma develops very rapidly (within 50 ns) after the breakdown,
and then remains quite stable.
5.4 End of the discharge and post-discharge
Since the major part of the workpiece material is removed at the end of the discharge
when the plasma collapses, the transition from the discharge to the post-discharge is
particularly interesting to study. This stage in the EDM process is certainly crucial for
the optimization of the material removal rate.
The evolution of the light intensity at the transition and during the post-discharge is
presented in gure 5.11.
0
150 250 350 450
time [ s] m
2
4
6
8
0
50
100
150
200
v
o
l
t
a
g
e

[
V
]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
50
0
1.4
0.7
c
u
r
r
e
n
t

[
A
]
post-discharge light
discharge light
Figure 5.11: Evolution of the light intensity during a post-discharge (discharge: 6 A, 50 s,
water).
As soon as the current is shut down, it drops rapidly to zero, and the light intensity is
simultaneously also dropping fast. But there is still a weak slowly-decaying light emission,
typically until 300 s after the end of the discharge. The corresponding images of the
transition are given in gure 5.12.
5.4. END OF THE DISCHARGE AND POST-DISCHARGE 63
time [ s] m
45 50
0
0.5
1
exposure
V/200 [V]
I/8 [A]
55
0
0.5
1
0
0.5
1
0
0.5
1
0
0.5
1
d
i
s
c
h
a
r
g
e

(
p
l
a
s
m
a
)
p
o
s
t
-
d
i
s
c
h
a
r
g
e
(
p
a
r
t
i
c
l
e
s
)
50 to 51 s m
51 to 52 s m
52 to 53 s m
53 to 54 s m
54 to 55 s m
Figure 5.12: Images of the transition between the discharge and the post-discharge, along with
current and voltage evolution (1 s exposure, variable delay; 12 A, 50 s, water). The images
are normalized in intensity. Each image is obtained during a dierent discharge.
As long as the discharge current is established, the plasma is visible as a bright spot.
When the current drops down, the disappearance of the plasma is almost immediate. The
plasma collapse is thus a very fast phenomenon.
Images of the post-discharge in gure 5.12 are less clear, due to the weak intensity of
the light emitted during this period. The objects which are emitting light are dicult to
identify. The quality of the images are much improved by setting the sensitivity of the
ICCD camera at its maximum and by increasing the exposure time. It is also easier to do
the measurements in oil, because the numerous bubbles present in water strongly perturb
the acquisition of the light. Some examples of better post-discharge images are shown in
gure 5.13.
It becomes clearer that the light emitted after the discharge is due to particles of heated
metal. These particles come from the molten metal pool created in the workpiece during
the discharge, and are then removed from the workpiece and ejected in the dielectric when
the discharge is shut down. We see their path and not only luminous dots, because they
are moving during the camera exposure. These explosions of particles were also clearly
200 m m
Figure 5.13: Incandescence of the removed particles after a discharge. Each image is taken
during a single post-discharge (100 s exposure; 12 A, 50 s, oil).
visible on gure 5.4. However, on these previous images, the exposure time was long
enough to record both the spark and the post-discharge. Therefore, the interpretation
of these luminous paths was not obvious. They could have been attributed to bubbles
for example, which would scatter the light of the spark. On the other hand, images of
gure 5.13 are time-resolved and the light is acquired only during post-discharges. The
interpretation is thus less ambiguous.
In order to conrm the hypothesis of the metal particles, optical emission spectroscopy
of the post-discharge light has been performed. An example of the spectrum is shown in
gure 5.14.
400 500 600 700 800
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
blackbody fit
2300 K
Figure 5.14: Optical emission spectrum of the post-discharge light, with a 2300 K blackbody
t. The light measurement is time integrated over thousands of post-discharges (50 s exposure,
150 g/mm grating; 16 A, 50 s, oil).
Spectroscopy shows that the post-discharge light emission is close to a blackbody
radiation, which conrms that the emitters are hot metal particles. The spectrum in
gure 5.14 is quite noisy, due to the low intensity of the light emission. Fitting several
post-discharge spectra with Plancks law, the temperature of the emitters is found to be
around 2200 K ( 100 K). Since the melting point of steel is 1700 K, the particles are
still in a liquid state in the very beginning of the post-discharge. The temperature found
5.4. END OF THE DISCHARGE AND POST-DISCHARGE 65
is consistent with metallurgical analyses of the modied layers of the workpiece [25, 125].
In these studies, the maximum temperature of the workpiece surface reached during the
EDM process has been estimated to be around 3000 K.
The particle size can be measured on the images. We found that the largest particles
have a diameter of 30 m, which is in good agreement with previous work on spark-
eroded particles [126, 127]. The particle speed can also be estimated from images. The
maximum speed is around 3 m/s. Since the particles move in the liquid, their trajectories
and speed can be aected if a liquid ow velocity is imposed.
Chapter 6
Optical emission spectroscopy of EDM
plasmas
The inuence of the discharge parameters on the emission spectra are analyzed in the rst
section, which presents measurements integrated over thousands of whole discharges. The
evolution of the spectra during the discharge duration is shown in the second section, with
time-resolved measurements. Spatially-resolved spectra are presented in the third section.
Finally, time- and spatially-resolved measurements are given in the fourth section.
6.1 Time-integrated emission spectra
6.1.1 Typical spectrum
An example of an optical spectrum emitted by the EDM plasma is shown in gure 6.1.
300 400 500 600 700 800 900 1000
H
a
C
Cu
C
2
Fe
Cr
Cu
Cu
Fe
Fe
Fe + Cr
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
Figure 6.1: Typical emission spectrum (Cu/steel, 12 A, 2 s, oil, 150 g/mm grating).
67
68 CHAPTER 6. OPTICAL EMISSION SPECTROSCOPY OF EDM PLASMAS
The dominant line is the Balmer H
line emitted by atomic hydrogen, which comes

from the cracking of the dielectric molecules. Some lines of atomic carbon and C
2
molecules (Swan system) are also visible. This indicates that the organic molecules of
oil are almost completely cracked by the discharge. As previously mentioned in 5.2.5,
imaging showed that the H
line is emitted by the whole plasma volume. This can be

understood considering the fact that the plasma is formed by cracking and ionizing the di-
electric molecules and atoms. Since the hydrogen source is the dielectric, hydrogen atoms
are present everywhere in the plasma. Although the H
line is dominant, the remaining

Balmer series lines such as H
(486.13 nm) or H
(434.05 nm) are not observed. This

remarkable point will be discussed later, in section 7.2.
The plasma is contaminated by impurities: several lines of atomic copper from the
electrode are present, along with many lines originating from atomic iron, chromium and
carbon of the removed material of the steel workpiece. These lines can be seen in more
detail in gure 6.2.
360 380 400 420 440
460 480 500 520 540 560 900 920 940 960
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
wavelength [nm]
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
C
C C
C
C
C
C
Cr
Fe
C
2
Fe
Fe
Fe
Fe
Fe
Fe
Cu
Cu
Cu
Cr
Fe
C
2
Fe
C
2
Cr
Fe
Fe
Cr
Cr
Fe
C
2
Fe Fe Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Fe
Figure 6.2: Typical emission spectrum, details (Cu/steel, 12 A, 100 s, oil, 600 g/mm grating).
The fact that all observed lines are atomic lines (along with a few neutral molecular
lines) and that no ionic lines are visible, is a rst indication that the plasma is cold. The
excitation of ions requires indeed higher energy than the excitation of atoms. Furthermore,
6.1. TIME-INTEGRATED EMISSION SPECTRA 69
atoms and ions of metallic elements are easily excited, more easily than those of non-
metallic elements. For example, the excited state energies of visible transitions of Fe atoms
and of Fe
+
ions are respectively around 3 and 11 eV (from the ground state of atoms and
ions respectively), but for C atoms and for C
+
ions they are respectively around 8 and
22 eV [128, 129]. Thus, the absence of ionic lines of iron, copper or chromium indicates
clearly that the plasma temperature is low. Since sparks in gases are known to be highly
ionized plasmas [49], EDM sparks should nevertheless also be fairly ionized.
The low excitation energy of metallic atoms explains the abundance and the relatively
high intensity of their lines in the EDM plasma emission (see for example the numerous
Fe lines in gure 6.2). It should be noted that this abundance does not mean that a
large amount of metal is present in the plasma. In electric arcs for example, the light
emission is dominated by metallic lines already with a contamination of a few percents of
metal. It is also well known that metallic impurities strongly cool down fusion plasmas
in tokamaks, because of their strong radiation. Thus, a quantitative estimation of the
metallic contamination in the EDM plasma from our spectra would be hazardous.
An intense broadband continuum radiation is also observed. The origin of this radia-
tion remains unclear. Blackbody ts do not match well with our measurements. Thus, a
continuous emission by heated electrodes or metal particles is unlikely. It is more probable
that this continuum is due to free-bound transitions, i.e. radiation emitted during recom-
bination processes. Free-free radiation is also another plausible source. Since molecules
generally emit a broadband spectrum, molecules or fragments of molecules from the di-
electric certainly also participate to the continuum [34, 63].
6.1.2 Eect of the dielectric
Figure 6.3 shows emission spectra obtained in dierent dielectrics.
400 600 800
(a) water (b) oil (c) liquid nitrogen
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
600 800 1000 500 700 900
wavelength [nm] wavelength [nm] wavelength [nm]
500 700 900 400 600 800 500 700 900
Cu
Cu
O
O
Cu
H
a
H
a
Cu
Cu
Cu
C
C
2
Cu
Cu
Cu
N
N
N
N
N
Figure 6.3: Eect of the dielectric on the emission spectrum (Cu/steel, 150 g/mm grating):
(a) water, 12 A, 2 s; (b) oil, 12 A, 2 s; (c) liquid nitrogen, 48 A, 2 s.
As mentioned previously, the discharge cracks the dielectric molecules, even the strong
triple bond of nitrogen N
2
molecules. We observe atomic hydrogen and oxygen lines from
water; atomic hydrogen, atomic carbon and C
2
lines from oil; atomic nitrogen lines from
liquid nitrogen. The dominant lines in gure 6.3 (c) are copper lines, because we have
more electrode wear than workpiece erosion under these experimental conditions in liquid
nitrogen.
The dielectric cleanliness has an eect on the emission spectrum, as shown in gure 6.4.
The continuum radiation is found to be more intense in clean dielectric than in dielectric
contaminated with eroded particles. Unsurprisingly, lines emitted by Fe, Cr and Cu
follow the opposite tendency. The high intensity of the continuum in clean dielectric is
not clearly understood. It is probably because the contribution of dielectric molecules
(continuous radiation) dominates the radiation of metallic elements in a clean dielectric.
400 600 800
(a) clean oil (b) contaminated oil
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
wavelength [nm] wavelength [nm]
500 700 900 1000 400 600 800 500 700 900 1000
Fe, Cr, Cu
Fe, Cr, Cu
Figure 6.4: Eect of the dielectric cleanliness on the emission spectrum (Cu/steel, 12 A, 50 s,
150 g/mm grating): (a) clean oil; (b) oil contaminated with eroded particles.
The addition of metallic powders to the dielectric has also been briey investigated
with spectroscopy. The aim of adding powders is to deposit new layers on the workpiece
surface and to improve its quality during the machining process. For example, the addition
of graphite nano-particles is found to improve the hardness of the white layer and to
reduce the number of cracks in it [125]. The addition of aluminium and tungsten powders
does not signicantly modify the emission spectrum. New Al and W atomic lines are
simply added. Thus, the plasma properties are certainly not drastically changed by the
introduction of powders. It is rather the breakdown process and the surface quality which
can be inuenced by the powders, depending on their size and composition.
6.1.3 Eect of the electrode material
Emission spectra obtained with an electrode in copper, in graphite and in tungsten are
shown in gure 6.5. The electrode material contaminates the plasma, and lines of re-
spectively atomic copper, of atomic carbon and C
2
, and of atomic tungsten are emitted.
With a graphite electrode, the Swan bands of C
2
and the C lines are much stronger than
in gure 6.1 for example, because the carbon contamination of the plasma due to the
graphite electrode is much greater than the carbon produced by the cracking of the oil.
The tungsten lines are very dicult to observe because they are located in the same
spectral region as many iron lines, which are more intense.
(a) copper (b) graphite (c) tungsten
400 600 800
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
wavelength [nm]
500 700 900 400 600 800
wavelength [nm]
500 700 900 400 600 800
wavelength [nm]
500 700 900
Cu
Cu
Cu
C
C
2
W
C
Figure 6.5: Eect of the electrode material on the emission spectrum (12 A, 5 s, oil, 150 g/mm
grating): (a) Cu/steel; (b) graphite/steel; (c) W/steel.
The use of a zinc electrode produces a spectrum quite dierent from the others, as
shown in gure 6.6. It should be noted that zinc is never used as electrode material in
die-sinking machines, but only as a coating layer of the wire in wire-cutting machines.
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
wavelength [nm]
300 400 500 600 700 800 900 1000
H
a
Zn
Zn
Zn
Zn
+
Zn
+
Zn
+
Zn
+
C
Figure 6.6: Emission spectrum with a zinc electrode (Zn/steel, 12 A, 12 s, 200 V, oil,
150 g/mm grating).
In addition to the H
line and C lines, both atomic Zn lines and ionic Zn

+
lines
are emitted, with high intensities. Zinc is the only material tested in this work that
produces ionic lines. The wavelengths of these lines, along with those of all other main
lines observed in EDM spectra, are given in table 6.1. The dierent behavior of zinc comes
from its physical properties. Excited states energies of Zn
+
transitions already exist from
8 eV [128]. Thus, Zn
+
ions are more easily excited than Fe
+
, Cu
+
or Cr
+
ions.
Moreover, this material has a low melting and boiling temperature (692 K and 1184 K
respectively), much lower than those of copper for example (1356 K and 2839 K respec-
tively). Zinc has also a vapor pressure 400 times higher than copper. Consequently,
the wear of a zinc electrode is high during the EDM process, and that explains why this
material is never chosen for real die-sinking EDM operations. A theoretical study based
on experimental data showed that the energy balance at the cathode is strongly related to
the electrode material properties [28]. It is demonstrated in this work that a zinc cathode
creates a higher density current and a lower ionic current fraction than other cathode
materials, due to the particular physical properties of zinc.
species wavelengths [nm]
H 656.3 (H
)
C 833.5, 906.2, 909.5, 911.2, 940.6, 962.1, 965.8
C
2
467.9, 468.5, 469.8, 471.5, 473.7, 512.9, 516.5, 563.6 (Swan system)
O 777.2, 777.4, 777.5, 844.6
N 746.8, 821.6, 868.0, 939.3, 1011.5
Cu 510.6, 515.3, 521.8, 578.2, 793.3, 809.3
W 400.9, 407.4, 426.9, 429.5
Fe 358.1, 382.0, 404.6, 438.4, 495.8, 516.7, 527.0, 561.6, 718.7, 751.1, 822.0
Cr 520.5, 520.6, 520.8, 529.8, 541.0, 723.6, 740.0
Zn 468.0, 472.2, 481.1, 636.2
Zn
+
491.2, 589.4, 602.1, 610.2, 747.9, 758.9, 773.3
Table 6.1: Wavelengths of the main lines emitted by EDM plasmas [128132].
6.1.4 Eect of the electrode polarity
The emission spectra also give qualitative information about the electrode wear and the
workpiece erosion. Figure 6.7 shows two spectra measured under the same experimental
conditions, except for the electrode polarity.
In gure 6.7 (a), the tungsten electrode is positively polarized with respect to the
workpiece. In this conguration, the wear of the electrode is very strong. In the other
conguration where the workpiece is positively polarized with respect to the electrode
(gure 6.7 (b)), the erosion is more eective and the electrode wears much less. The
polarity has a clear inuence on spectra. With negative electrode polarity, the iron lines
of the group around 520 nm are much more intense than with positive polarity. If we
consider the H
line as a reference of intensity (its intensity does not depend on erosion or

wear but originates only from the dielectric), the iron lines show clearly that the erosion is
more ecient with negative polarity. If we look at the region between 350 and 450 nm, it
(b) polarity - (a) polarity +
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
wavelength [nm]
300 400 500 600 700 800 900 1000
wavelength [nm]
300 400 500 600 700 800 900 1000
350 400 450
Fe
Fe
W
Fe
Fe H
a
H
a
350 400 450
Figure 6.7: Eect of the electrode polarity on the emission spectrum (W/steel, 1 A, 100 s,
water, 150 and 600 g/mm gratings): (a) positive polarity; (b) negative polarity.
is obvious that the wear is stronger with positive polarity: in (a) several tungsten and iron
lines are visible with comparable intensities, in (b) the tungsten lines are no longer visible
and the iron lines are more intense. Note that these conclusions are only qualitative, since
it is dicult to deduce the quantities of materials removed from these spectra.
6.1.5 Eect of the discharge current
Three emission spectra obtained with dierent discharge currents are presented in g-
ure 6.8.
400 600 800
(a) 6 A (b) 12 A (c) 48 A
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
500 700 900 400 600 800 500 700 900
wavelength [nm]
400 600 800 500 700 900
H
a
H
a
H
a
Fe, Cr, Cu
Fe, Cr, Cu
Fe, Cr, Cu
Cu
Cu
Cu
Figure 6.8: Eect of the discharge current on the emission spectrum (Cu/steel, 12 s, oil,
150 g/mm grating): (a) 6 A; (b) 12 A; (c) 48 A.
Increasing the discharge current strongly enhances the intensity of the emitted light.
But from the spectroscopic point of view, the increase in the current has little eect. It
only increases the intensities of the metallic lines. This can be understood by considering
that discharges with high current produce large craters and remove more material than
low current discharges.
6.1.6 Eect of the discharge on-time
The discharge on-time has a signicant inuence on the emission spectrum. Figure 6.9 (a)
and (b) are spectra for two dierent on-times of Isopulse discharges, where the current is
controlled and maintained constant during the discharge. To produce shorter discharges,
we use the generator in the capacitive mode (see section 3.1). Figure 6.9 (c) is the
spectrum of discharges produced in the capacitive mode with a 10 nF capacitor.
400 600 800
(a) 200 s m (b) 5 s m (c) ~ 1.6 s (capacitive mode) m
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
500 700 900 400 600 800 500 700 900
wavelength [nm]
400 600 800 500 700 900
O
H
a
H
a
H
a
O
O
O
O
Figure 6.9: Eect of the discharge on-time on the emission spectrum (Cu/steel, water,
150 g/mm grating): (a) 200 s, 12 A; (b) 5 s, 12 A; (c) discharges produced in the capac-
itive mode, 1.6 s, 6 A.
The broadening of the H
line increases drastically with decreasing on-time as shown in

gures 6.9 (a) and (b). The FWHM of the H
line reaches almost 10 nm in gure 6.9 (b).

Such an important broadening can only be due to dynamic Stark broadening, which
dominates other broadening mechanisms under these conditions. The H
line is also
shifted to higher wavelengths with decreasing on-time. The measured values of FWHM
and line shift are a sign of an extremely high electron density, in particular for brief
discharges and thus for the beginning of any discharge. Because the presented spectra
are time-integrated, the evolution of electron density can not be directly determined from
these FWHM measurements.
For discharges shorter than 2 s, the plasma is so dense that almost all the atomic lines
are completely merged into one broad continuum. We hardly distinguish on gure 6.9 (c)
the H
line, two oxygen lines and two copper lines. This continuum is of another nature
than the continuum observed on the previous spectra. It is created by a degeneracy
of atomic lines, due to the high plasma density. Extreme densities destroy the upper
energy levels of the atoms, and the free-bound radiation becomes more important than
line radiation [133, 134]. This eect will be discussed in more detail later, in section 7.2.
6.1.7 First estimation of electron density and temperature
With a H
FWHM of almost 10 nm for short discharges (gure 6.9 (b)), the order of
magnitude of the electron density is 10
18
cm
3
(equation (3.6)). This is a particularly
high value, 100 times higher than the density of vacuum arcs and 10 times higher
than the density of sparks in gases. This rst estimation is in good agreement with other
studies on discharges in liquids (see section 2.2), and is also conrmed by the spectrum
of gure 6.9 (c): this kind of line-merging phenomenon can only occur in extremely dense
plasmas. The pressure imposed by the liquid surrounding the plasma is the cause of its
high density.
With the two-line method applied to copper lines from time-integrated spectra, the
electron temperature is estimated around 0.7 eV 0.15 eV ( 8100 K 1750 K). The
errors in intensity measurements are principally due to other overlapping lines located
near the copper lines. The value found is a low electron temperature, but consistent
with previous studies on EDM plasmas [40, 41] and other similar plasmas [31, 34, 135].
It should be noted that this electron temperature estimation does not give information
on the temperature at the electrode surfaces, which would be a useful value to know for
simulations and practical applications. This temperature has to be measured dierently.
Results can be found in [123, 136].
The two-line method is based on the assumption of LTE. It is not an obvious as-
sumption, especially for non-ideal plasmas such as EDM plasmas (see chapter 7), where
coulombian interactions are comparable to thermal interactions. The evolution to the
equilibrium state by thermal interactions is thus perturbed by coulombian eects. How-
ever, due to the high plasma density, one can assume that the numerous collisions between
particles thermalize them rapidly.
A density criterion giving the validity of LTE can be derived theoretically [107, 137].
A plasma of temperature T is in complete LTE if its electron density n
e
is higher than
n
e
9.2 10
17
_
k
B
T
E
H
_
E
2
E
1
E
H
_
3
[cm
3
] , (6.1)
where E
H
is the ionization energy of the hydrogen atom, E
1
the ground state energy of
the atoms present in the plasma and E
2
the energy of their rst excited state. According
to this criterion, a hydrogen plasma at 0.7 eV is in LTE if
n
e
8.6 10
16
cm
3
. (6.2)
If we assume that the EDM plasma is mainly composed by hydrogen, its electron density
is suciently high for the assumption of LTE to be reasonably accepted. Therefore, even
if we have not measured it separately, the ion temperature can be estimated to be of the
same order as the electron temperature.
6.2 Time-resolved emission spectra
6.2.1 Merging of atomic lines
An example of time-resolved emission spectra taken during the rst 5 microseconds of the
discharge is given in gure 6.10.
Figure 6.10: Example of time-resolved spectra (Cu/steel, 16 A, water; time resolution 500 ns,
150 g/mm grating).
The intensities of the spectra decrease with time, because the light emission is particu-
larly intense during the rst microsecond of the discharge, as stated previously in 5.2.1.
No lines are clearly identiable in the very rst spectrum. The emission is almost contin-
uous. This eect of line merging has already been observed with very short discharges in
the preceding section (gure 6.9). The previous conclusions about the high plasma density
at the beginning of the discharge is thus conrmed by time-resolved measurements.
The merging can be more clearly seen in gure 6.11, which presents spectra of the rst
3 microseconds of the discharge measured with a better time resolution than in gure 6.10.
The spectral region investigated is centered around 520 nm, where the main Cu, Cr and
Fe lines are located. All the lines are completely merged during approximately the rst
400 ns. Their shapes then appear, strongly broadened between 400 ns and 1 s, much
less afterwards. This is evidence that the electron density is decreasing with time, as we
will see later in 6.2.3.
6.2.2 Evolution of the electron temperature
The evolution of T
e
during the discharge is measured by applying the two-line method
on each time-resolved spectrum. Note that it is impossible to measure T
e
during the rst
microsecond of the discharge with this method, because of the line merging. The electron
6.2. TIME-RESOLVED EMISSION SPECTRA 77
wavelength [nm]
t
i
m
e
[
s
]
m
0
1
2
3
460
480
500
520
540
560
580
0.4
0.6
0.8
1
i
n
t
e
n
s
i
t
y

(
n
o
r
m
a
l
i
z
e
d
)
Figure 6.11: Merging of the Fe, Cu and Cr atomic lines (Cu/steel, 12 A, water; time resolution
200 ns, 600 g/mm grating). The spectra are normalized to the intensity of the Cu line at
521.8 nm.
temperature is slightly higher during the rst 10 s than for the rest of the discharge,
being from 0.8 to 0.7 eV ( 9000 to 8000 K). Then it remains rather constant around
0.7 eV within the margin of error, even after 1 ms.
6.2.3 Evolution of the electron density
Figure 6.12 shows time-resolved emission spectra of the H
line during the rst 10 s of

the discharge.
640
650
660
670
680
0
5
10
656.28 nm
peak position
wavelength [nm]
t
i
m
e
[
s
]
m
0.2
0.4
0.6
0.8
1
i
n
t
e
n
s
i
t
y

(
n
o
r
m
.
)
Figure 6.12: Broadening and shift of the H
line (Cu/steel, 16 A, water; time resolution 1 s,

600 g/mm grating). The spectra are normalized to the intensity of the H
line.
The very rst spectra show extremely broadened and shifted H
lines. Thus, it is
quite dicult to nd a good t for the rst spectra, leading to non negligible errors in the
FWHM measurements. The black lines on gure 6.12 represent the theoretical position
of the H
line (656.28 nm) and the measured peak positions. These lines are drawn to
easily visualize the line shift. At the beginning of the discharge, we can measure FWHM
up to 10 nm and shift up to 10 .
The evolution of n
e
is calculated from these time-resolved H
spectra with equa-

tions (3.6) and (3.7). Results are given in gure 6.13.
1 10 100
broadening
shift
time [ s] m
e
l
e
c
t
r
o
n

d
e
n
s
i
t
y

[
c
m
]
-
3
10
18
10
17
Figure 6.13: Evolution of the electron density, deduced from time-resolved broadening and
shift measurements of the H
line (Cu/steel, 16 A, water).

In spite of some dierences, the FWHM and the shift measurements give comparable
density values and evolution. As expected, the electron density is extremely high during
the rst microsecond (above 10
18
cm
3
) and then decreases rapidly with time. The EDM
plasma is created from a liquid, which is a dense medium ( 310
22
water molecules/cm
3
,
for example). This is the main reason for the high plasma density. In the very beginning of
the discharge, the plasma has to overcome the extreme pressure imposed by the dielectric.
Then the plasma expands, which results in a decrease in its density. But during the whole
discharge, the density remains high (always above 10
16
cm
3
) due to the constant pressure
of the surrounding liquid. The density reduction with time is conrmed by other studies
on discharges in liquids [34].
The high electron density measured reinforces the assumption of free-bound and free-
free radiation for the explanation of the continuum given in 6.1.1. Since these two
types of radiation processes involve free electrons, the intensity of free-bound and free-
free radiation is in fact directly related to n
e
. Therefore, the continuum is expected to be
intense with such high values of n
e
. Moreover, the measured continuum decreases with
the discharge on-time (see gure 6.9), which is consistent with the decrease of n
e
with
time.
6.3. SPATIALLY-RESOLVED EMISSION SPECTRA 79
6.3 Spatially-resolved emission spectra
Figure 6.14 shows an example of spatially-resolved spectra along the vertical axis. The
intensities of the spectra located in the center are higher, because the plasma is brighter
in this region (see also gure 5.6).
450
500
550
600
650
700
5
10
15
# fibre
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
Figure 6.14: Example of spatially-resolved spectra, along the vertical axis (Cu/steel, 12 A,
4 s, water, 150 g/mm grating).
Note that the spatially-resolved axis is not calibrated in real distance, but simply
graduated with the numbers of the bres making the sampling. In order to make the
conversion into real distance, we have to know the spatial resolution obtained with the
bre bundle, i.e. the real distance between two points seen by two consecutive bres. This
spatial resolution depends on the optical arrangement, mostly on the plasmaendoscope
distance, which is not constant from one set of measurements to another. After each
set of measurements (the simultaneous acquisition of the 16 spectra), the electrodes have
to be changed due to their erosion, and the optical arrangement has to be re-aligned.
Thus, an exact calibration for each set of measurements is time-consuming and has not
been systematically done. The spatial resolution can nevertheless be estimated around
2030 m/bre (gap distance 100 m), for all the measurements presented here.
6.3.1 Asymmetry of the contamination
As previously said, the spectral region around 520 nm is interesting for a qualitative
characterization of the plasma contamination, because of the presence of three chromium
lines at 520.45, 520.6 and 520.84 nm, and of the copper line at 521.82 nm. With a copper
electrode and a stainless steel workpiece, the Cu line is emitted by particles coming
from the electrode and the Cr lines are emitted by particles coming from the workpiece.
Figure 6.15 shows spatially-resolved spectra along the vertical axis of this spectral region.
500
510
520
530
540
550
6
7
8
9
10
11
0
0.2
0.4
0.6
0.8
1
# fibre
wavelength [nm]
i
n
t
e
n
s
i
t
y

(
n
o
r
m
a
l
i
z
e
d
)
Cu line (=1)
Cr line
to steel
electrode
to copper
electrode
Figure 6.15: Vertical asymmetry of the plasma contamination (Cu/steel, 6 A, 100 s, water,
150 g/mm grating). The spectra are normalized to the intensity of the Cu line at 521.8 nm.
An asymmetry of the plasma contamination is clearly visible. Near the Cu electrode,
the Cu line is more intense than the Cr lines. On the other hand, the Cr lines are more
intense than the Cu line near the steel workpiece. Thus, even if Cu and Cr lines are present
everywhere in the plasma, we see that each electrode is contaminating the plasma mostly
in a region close to itself. Unsurprisingly, no asymmetry is visible along the horizontal
axis (measurement not shown here).
Depending on the discharge parameters, material transfer from one electrode to the
other can sometimes be observed. In this case, the intensity of lines emitted by both
electrode materials are quite constant across the vertical prole. With the discharge
conditions of gure 6.15, no material transfer is macroscopically visible on the electrodes
after several minutes of machining. This is qualitatively conrmed by the contamination
asymmetry measured by spatially-resolved spectroscopy.
6.3.2 Electron temperature prole
Spatially-resolved spectra of the three Cu lines used for T
e
calculations are shown in
gure 6.16, along with the electron temperature vertical prole calculated from them.
The central spectra have a better signal to noise ratio than the outer ones, because the
light intensity is higher in the center. The electron temperature is quite constant over
the whole prole. Measurements along the horizontal axis are not shown here but do not
show any dierence. The plasma has thus a homogeneous electron temperature around
0.7 eV, as previously measured.
6.3. SPATIALLY-RESOLVED EMISSION SPECTRA 81
505
510
515
520
525
6
8
10
12
0
0.2
0.4
0.6
0.8
1
wavelength [nm]
# fibre
# fibre
i
n
t
e
n
s
i
t
y

(
n
o
r
m
.
)
0
0.2
0.4
0.6
0.8
1
4 6 8 10 12
e
l
e
c
t
r
o
n

t
e
m
p
e
r
a
t
u
r
e

[
e
V
]
521.8 nm (=1)
515.3 nm
510.6 nm
Te 510-515
Te 510-521
Figure 6.16: Vertical prole of the electron temperature calculated from spatially-resolved Cu
lines intensities ratios (Cu/steel, 6 A, 100 s, water, 1800 g/mm grating). The spectra are
normalized to the intensity of the Cu line at 521.8 nm.
6.3.3 Electron density prole
Spatially-resolved H
spectra, along with the electron density vertical prole calculated

from them with equation (3.6), are shown in gure 6.17. The values found are in good
agreement with our previous results (gure 6.13), around 510
16
cm
3
for a 100 s dis-
charge. Despite large error bars, the electron density seems to be slightly higher in the
plasma center. The horizontal prole, not shown here, is very similar to the vertical one.
620
640
660
680
700
6
8
10
0
0.2
0.4
0.6
0.8
1
wavelength [nm]
# fibre
4 6 8 10 12
# fibre
i
n
t
e
n
s
i
t
y

(
n
o
r
m
.
)
e
l
e
c
t
r
o
n

d
e
n
s
i
t
y

[
c
m
]
-
3
10
16
10
17
Figure 6.17: Vertical prole of the electron density calculated from spatially-resolved H
broad-
ening measurements (Cu/steel, 6 A, 100 s, water, 600 g/mm grating). The spectra are normal-
ized to the intensity of the H
line.
6.4 Time- and spatially-resolved emission spectra
6.4.1 Evolution of the contamination
A temporal and spatial characterization of the contamination can be obtained by gating
the camera when acquiring spatially-resolved spectra of the Cr and Cu lines. In order
to present the contamination results in a convenient way, we have calculated, for each
spectrum, the lines intensities ratio R dened by
R
.
=
I(Cr)
I(Cu)
=
I(520.5 nm) +I(520.6 nm) +I(520.8 nm)
I(510.6 nm) +I(515.3 nm) +I(521.8 nm)
.
This value represents the relative importance of the contamination coming from the work-
piece (Cr lines) with respect to the contamination coming from the electrode (Cu lines).
The absolute value of R has no physical signicance and is not important, but its evolu-
tion in space and time is of interest for a description of the contamination. Figure 6.18
shows an example of a spectrum where these lines are located, along with the evolution
of R proles. The delay between the beginning of the discharge and the camera exposure
is indicated near each prole.
0.2
0.4
0.6
0.8
5 10 15
510 515 520
# fibre
wavelength [nm]
i
n
t
e
n
s
i
t
y
Cu Cr
Cu
Cu
R

=

r
a
t
i
o

C
r

/

C
u
0 s m
20 s m
30 s m
40 s m
4 s m
8 s m
steel copper
Figure 6.18: Evolution of the contamination distribution, deduced from time- and spatially-
resolved spectra of Cu and Cr lines (Cu/steel, 12 A, 50 s, water; time resolution 2 s).
The ratio R clearly increases during the discharge, from 0.35 to 0.7 in 40 s. It
indicates that the plasma contamination from the workpiece material increases with time,
probably by direct vaporization of the metal into the plasma. This vaporization is mostly
ecient after 10 s (R is roughly constant before). Again, it is impossible to quantify
the eciency of the vaporization, in terms of material removal rate. But even if some
amount of material is vaporized during the discharge, the largest part of the erosion occurs
certainly at the end of the discharge, with the implosion of the plasma.
The increase in R with time could also be partially due to a small decrease of the
contamination from the electrode during the discharge. It is known indeed that wear
mostly occurs at the beginning of the discharge.
6.4. TIME- AND SPATIALLY-RESOLVED EMISSION SPECTRA 83
Due to the asymmetry of the contamination observed in the preceding section, the
proles of R should be decreasing from the left to the right. The steel workpiece is indeed
located towards the left (towards bre #1) and the copper electrode towards the right
(towards bre #16). This tendency cannot be observed in gure 6.18. It is mainly due
to the large experimental errors (error bars are not shown for clarity). As stated in sec-
tion 3.6, time- and spatially-resolved measurements are particularly dicult. Contrary to
the increase in R with time, the asymmetry is probably an eect too subtle to be detected
during these measurements.
The eect of the electrode polarity on the contamination distribution and evolution is
shown in gure 6.19.
0.4
0.5
0.6
0.7
0.8
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
5 10 15
# fibre
R

=

r
a
t
i
o

C
r

/

C
u
steel copper
R

=

r
a
t
i
o

C
r

/

C
u
5 10 15
# fibre
steel copper
(a) polarity + (b) polarity -
8 s m
12 s m
28 s m
36 s m
20 s m
8 s m
16 s m
24 s m
44 s m
32 s m
Figure 6.19: Eect of the electrode polarity on the evolution of the contamination repartition,
deduced from time- and spatially-resolved spectra of Cu and Cr lines (Cu/steel, 6 A, 50 s,
water; time resolution 4 s): (a) positive polarity; (b) negative polarity.
As in gure 6.7, the situation where the wear is important (gure 6.19 (a)) can be
distinguished from the situation where the erosion is important (gure 6.19 (b)). The
ratio R is in fact lower with the positive polarity (0.50.7) than with the negative polarity
(0.82). R increases with time, for both polarities: it slightly increases from 0.5 to 0.65
in (a), but doubles from 0.9 to 1.8 in (b). The vaporization is thus more ecient in the
latter case. Again, it is dicult to observe the contamination asymmetry. Some of the
proles in gure 6.19 (b) nevertheless slightly decrease from left to right.
6.4.2 Evolution of the electron temperature proles
Spatially-resolved spectra of the Cu lines taken at three dierent times of the discharge are
given in gure 6.20. The decrease in the continuum intensity and in the line broadening
with time can be observed. It is remarkable that even the Cu lines show a signicant
broadening during several microseconds, due to the high plasma density. These lines are
in fact much less sensitive to Stark broadening than hydrogen lines. Broadening of Cu
lines is also observed in laser-produced plasmas with copper targets [94, 138].
(a) 0 to 2 s m (c) 48 to 50 s m (b) 4 to 6 s m
500
510
520
5
10
15
wavelength [nm]
# fibre
500
510
520
5
10
15
500
510
520
5
10
15
Figure 6.20: Time- and spatially-resolved spectra of the Cu lines used for electron temperature
calculation (Cu/steel, 12 A, 50 s, water; time resolution 2 s, 600 g/mm grating): (a) 0 to 2 s;
(b) 4 to 6 s; (c) 48 to 50 s.
Figure 6.21 shows the evolution of T
e
proles calculated from the previous spectra. As
previously mentioned, the electron temperature is slightly higher at the beginning, but
remains around 0.7 eV. The proles are roughly at within the margin of errors.
0.5
0.6
0.7
0.8
0.9
0 4 8 12 16
# fibre
e
l
e
c
t
r
o
n

t
e
m
p
e
r
a
t
u
r
e

[
e
V
]
8 s m
12 s m
16 s m
48 s m
28 s m
Figure 6.21: Evolution of the electron temperature proles (Cu/steel, 12 A, 50 s, water, time
resolution 2 s).
6.4.3 Evolution of the electron density proles
Time- and spatially-resolved H
spectra are shown in gure 6.22. The spectra are quite

noisy, due to the weak intensity emitted. Nevertheless, the H
broadening is easily ob-

servable.
The evolution of the electron density proles is presented in gure 6.23. These den-
sity measurements are consistent with time-resolved and spatially-resolved measurements
6.4. TIME- AND SPATIALLY-RESOLVED EMISSION SPECTRA 85
(a) 0 to 2 s m (c) 36 to 38 s m (b) 4 to 6 s m
640
650
660
5
10
15
# fibre
wavelength [nm]
640
650
660
5
10
15
640
650
660
5
10
15
Figure 6.22: Time- and spatially-resolved spectra of the H
line (Cu/steel, 12 A, 50 s, water;

time resolution 2 s, 600 g/mm grating): (a) 0 to 2 s; (b) 4 to 6 s; (c) 36 to 38 s.
previously presented (gures 6.13 and 6.17). The density diminishes during the discharge,
by almost two orders of magnitude in 50 s. The density is slightly higher in the plasma
center, and this tendency is veried during the whole discharge. This temporal and spatial
characterization of n
e
can be useful for the validation of EDM plasma simulations.
4 6 8 10 12 14
# fibre
10
18
10
17
10
16
e
l
e
c
t
r
o
n

d
e
n
s
i
t
y

[
c
m
]
-
3
0 s m
12 s m
4 s m
8 s m
16 s m
20 s m
48 s m
36 s m
Figure 6.23: Evolution of the electron density proles (Cu/steel, 12 A, 50 s, water, time
resolution 2 s).
Chapter 7
Non-ideality of EDM plasmas
In this chapter, we will show that EDM plasmas are weakly non-ideal. The denition of
a non-ideal plasma is given in the rst section. Spectroscopic evidence of the non-ideality
of EDM plasmas is discussed in the second section. Finally, the third section presents a
general summary of the EDM plasma physical properties, measured in this work.
7.1 Plasma coupling parameter
Natural plasmas and articially-produced plasmas are both very diverse. Generally, they
are classied according to their temperature and density, because the properties of a
plasma depend mostly on these two parameters.
In particular, the density n and temperature T dene whether a plasma is ideal or
not. In an ideal plasma, the particles are almost free, i.e. their movement is only weakly
inuenced by other particles. In this case, the mean kinetic energy of particles is much
higher than the mean Coulomb potential energy between them.
Knowing n and T for a given plasma, we can calculate its coupling parameter , which
quanties its degree of ideality. This parameter represents the ratio of the potential
energy of Coulomb interaction at the mean inter-particle distance divided by the mean
thermal energy of the charged particles in the plasma. It is dened by
.
=
Z
2
e
2
4
0
a k
B
T
, (7.1)
where Z is the charge of the particles (Z=1 for hydrogen ions, for example), e the ele-
mentary charge,
0
the vacuum permittivity, and a the mean inter-particle distance given
by
a =
_
3
4n
_
1/3
. (7.2)
Therefore, we see that

n
1/3
T
. (7.3)
87
88 CHAPTER 7. NON-IDEALITY OF EDM PLASMAS
The plasma is called:
ideal if 1;
weakly non-ideal if 1;
strongly coupled if > 1.
According to equation (7.3), we see that ideal plasmas are characterized by a low density
and a high temperature. Note that is linked with the Debye length
D
, dened by
D
.
=
_
0
k
B
T
e
2
n
, (7.4)
by the relation
3/2
1
n
3
D
.
Therefore, we see also that, in an ideal plasma, a very large number of particles is con-
tained in a Debye sphere. Nevertheless, the higher the number of particles in a Debye
sphere, the less the particles interact with each other.
On the other hand, non-ideal and strongly coupled plasmas are dense and/or cold. In
such plasmas, the inter-particle distance is short enough, and the particle velocity small
enough, to produce signicant electrostatic interactions between the charged particles.
For a given particle, the microelds created by the others have a signicant inuence on
it.
We have seen in the previous chapter that EDM plasmas have a typical electron tem-
perature of 0.7 eV and a typical electron density of 10
18
cm
3
. Thus, the typical coupling
parameter of EDM plasmas is around
0.33 .
If we consider that n
e
reaches 2.5 10
18
cm
3
in the very beginning of the discharge (see
gure 6.13), 0.45 during the rst microsecond. Thus, EDM discharges produce cold
and dense plasmas which are weakly non-ideal.
According to equation (7.4), the typical Debye length of EDM plasmas is around 6 nm.
The mean inter-particle distance a is also around 6 nm (equation (7.2)). Thus, a Debye
sphere is containing roughly only one particle. This is another illustration of the plasma
non-ideality.
Figure 7.1 shows the classication of the main types of natural and articial plasmas,
including EDM plasmas, in a log densitylog temperature diagram. Ideal, weakly non-
ideal and strongly coupled plasmas can easily be distinguished in such a log-log diagram,
because iso-contours are straight lines.
7.1. PLASMA COUPLING PARAMETER 89
log T [eV]
- 2 - 1 0 1 2 3 4 5
l
o
g

n

[
c
m
]
-
3
25
20
15
10
5
0
Earth
plasma sheet
Solar wind
(1 AU)
Earth
iono-
sphere
Flames
Glow
discharge
Solar
corona
Low
pressure
Alkali
metal
plasma
Fusion
experiments
Fusion
reactor
Focus
Z-pinches
Laser
plasma
Shock
tubes
High pressure
arcs
EDM
strongly-coupled
plasmas
ideal plasmas
w
e
a
k
l
y
n
o
n
-
i
d
e
a
l
p
l
a
s
m
a
s
Jupiter core
30
Sun core
white dwarf
brown dwarf
G
=
0
.
0
1
G
=
0
.
1
G
=
1
G
=
1
0
Interstellar gas
active
galactic
nucleus
6
log T [K]
2 3 4 5 6 7 8 9 10
Figure 7.1: Classication of plasmas according to their density and temperature (adapted
from [45, 139]).
One can see in gure 7.1 that the majority of natural and articial plasmas are ideal.
Besides extremely dense astrophysical objects, few non-ideal plasmas exist. From the
physical point of view, their extreme properties make them particularly interesting to
study. The physics of non-ideal plasmas is complex, diering from classical plasma
physics theories [4346]. Corrections have to be made because of deviations from the
binary collision model due to collective eects. For example, the screening mechanism
described by the Debye model has to be modied in non-ideal plasmas. Theories have
been developed quite recently, mainly due to the lack of experimental data before the
1980s. The creation of dense, and therefore non-ideal plasmas, is in fact dicult and
requires considerable experimental resources (see 2.3). In this context, EDM discharges
appear to be a very simple method to produce such plasmas. EDM plasmas are indeed
amongst the most dense articially-created plasmas. But unfortunately, these plasmas
are not the easiest to be investigated experimentally, as repeated several times in this
work.
7.2 Spectroscopic signs of the non-ideality
In addition, spectroscopic results indicate also that the EDM plasma is non-ideal.
7.2.1 Broadening and shift of H
Since non-ideal plasmas are dense, a strong Stark broadening and shift of spectral lines is
naturally a characteristic sign of non-ideality. The microelds deform the energy spectrum
of radiating atomic particles, i.e. they perturb the initial and nal states of radiative
transitions. The atomic levels are shifted and split, and this results in line broadening
and shift [43, 44].
This eect can be clearly observed on optical spectra emitted by EDM plasmas, espe-
cially on the H
line. This fact has already been discussed in detail in chapter 6. Similar
strong broadening and shift of hydrogen lines in dense plasmas have been reported in
numerous works [113, 115, 117, 135, 140144].
It should be noted that the calculation of spectral line shapes emitted by dense plasmas
has been recently much improved. Simulations in good agreement with experimental
data are obtained with models including several non-ideal eects, such as the eect of
the dynamic microeld created by the ions or the inuence of electron collisions [45].
Calculations are mostly done for lines of the hydrogen atom or hydrogen-like ions, because
of the simplicity of quantum calculations for these species.
7.2.2 Asymmetry and structure of H
Besides its broadening and shift, the H
line gives other useful information. Figure 7.2

shows a typical example of a high resolution spectrum of the H
line. If a lorentzian t
is applied, it is observed that the shape of the line is slightly asymmetric and deviates
from the t. This asymmetry is also due to interactions between ions and the emitter
(hydrogen atom in our case), and is another sign of plasma non-ideality [43, 114, 116].
We also see in gure 7.2 the complex structures around the H
line. The arrows mark

small bumps which are not due to radiation of other lines. These structures are created
by several phenomena, which occur in dense plasmas [145153]. Figure 7.3 gives a simple
explanation for the origin of these structures.
7.2. SPECTROSCOPIC SIGNS OF THE NON-IDEALITY 91
wavelength [nm]
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
n
i
t
]
650 655 660 665
Lorentzian fit
Figure 7.2: Complex structure of the H
line (C/steel, 24 A, 12 s, oil, 1200 g/mm grating).

w
0s
1s
2p
3p 4p
5s 6s 8p
(a) Stark components
w
(b) satellites
w
(c) combination of (a) and (b)
1
2
3
4
5
0
W
D /
Figure 7.3: Theoretical structure of the H
line in a non-ideal plasma [147]: (a) Stark com-

ponents of H
(see appendix A.2); (b) satellite lines due to plasma waves or turbulence; (c) re-
sulting H
structure. The frequency = 0 corresponds to the H
unperturbed position (
0
=
656.28 nm). The components on the other side of the line are symmetrically positioned.
First, the quasistatic ion microelds cause splitting of energy levels in hydrogen atoms
by the Stark eect. The H
emission is formed by several lines, due to transitions be-

tween these new levels (see appendix A for more details). These lines are represented in
gure 7.3 (a), where and denote the polarization of each component. Moreover, low
frequency electrostatic waves can also produce a substantial quasistatic eld, with the
same eect as the ion microeld.
Furthermore, if plasma waves or turbulence are present in the plasma, a high frequency
electric eld is superimposed on the quasistatic microeld. This dynamic eld, oscillating
with a frequency , produces satellite lines as shown in gure 7.3 (b). In the case of
Langmuir waves for example, the frequency of the dynamic eld is

pe
,
where
pe
is the electron plasma frequency dened by
pe
.
=
e
2
n
e
0
m
e
. (7.5)
For EDM plasmas,
pe
5 10
13
s
1
. The eld associated with Langmuir waves will
create satellite lines located at
k
=
0
k
pe
, k = 1, 2, 3, ...
where
0
is the unperturbed H
position, and where
pe
=

pe
2c

2
0
is the wavelength associated with the electron plasma frequency [148, 153].
The combination of the quasistatic and the oscillating eld produces a spectrum where
the Stark components and the satellites are superimposed, as shown in gure 7.3 (c).
Furthermore, if the frequency between the Stark components matches the frequency of
the dynamic eld (i.e. if / = ), resonant interactions occur. As a consequence,
depressions or dips will appear on the spectrum.
The addition of all these eects, typical in non-ideal plasmas, results in multiple mod-
ications to the H
prole, as seen in gure 7.2. The complexity of the emission spectrum

reects the complexity of the plasma itself: it is a cold and dense plasma, highly colli-
sional, with strong electric elds, turbulence and resonance phenomena.
In principle, the position of the structures in gure 7.2 should allow us to estimate the
value of the electric elds present in the plasma. But this estimation can be done only
if the structures are completely identied, which is extremely complex. The Stark com-
ponents have to be distinguished from the satellites, which can be produced by dierent
simultaneous waves or turbulence.
In our case, this identication is not possible. First, the structures of gure 7.2 are
barely observable, and it is also almost impossible to estimate which waves are present in
the EDM plasma. Turbulent phenomena are certainly more likely than classical plasma
waves. Nevertheless, the ion microeld can be estimated from the mean inter-particle
distance [111, 149]. With a density of 10
18
cm
3
, the microeld is around 10
6
V/cm.
7.2. SPECTROSCOPIC SIGNS OF THE NON-IDEALITY 93
7.2.3 Absence of H
and line merging

Another important eect of the plasma non-ideality is the transformation of the upper
members of a spectral series into a continuous spectrum [43, 44, 134]. This eect is due
to:
1. the lowering of the ionization threshold;
2. the merging of spectral lines near the series limit.
First, the ionization potential of atoms is lowered in non-ideal plasmas. Inter-particle
interactions in high density plasmas destroy in fact the upper atomic levels. More
precisely, the microelds which are superimposed on the Coulomb electric eld of an
atomic nucleus, transform the upper energy states of this atom into unbound states, as
shown in gure 7.4.
r
V(r)
(1) Coulomb potential
(2) microfield
unbound states
bound states
(1) + (2)
Figure 7.4: Lowering of the ionization threshold in non-ideal plasmas, due to the eect of
microelds (assumed to be uniform on the atomic scale) [44].
The shift in the ionization energy U
i
is given by
U
i
=
Z
2
e
2
4
0
D
, (7.6)
where
D
is the Debye length [44,46]. In our case, we found U
i
0.23 eV. This lowering
of the ionization threshold modies the optical spectrum of the plasma. Figure 7.5 explains
schematically the consequences in the case of an hydrogen plasma.
In an unperturbed atom, all the levels can be populated by electrons (gure 7.5 (a)).
Therefore, the dierent lines of all series can be emitted, in particular those of the Balmer
series: H
, H
, H
, etc. Continuous radiation by free-bound transitions is emitted by

electrons coming from the energy continuum, whose threshold is 13.6 eV above the ground
state n = 1 (see gure 3.8).
On the other hand, in a dense plasma, the ionization potential is lowered due to the
microelds. If the new threshold is located between levels n = 3 and n = 4 for example
n=2
n=3
n=4
n=5
H
a
H
b
H
g
ionization threshold
H
a
continuum
(a) unperturbed H atom (b) H atom in a dense plasma
continuum
radiation
continuum
radiation
n=2
n=3
ionization threshold
Figure 7.5: Absence of H
and increase in the continuum radiation due to the lowering of the

ionization threshold in a dense hydrogen plasma.
(gure 7.5 (b)), the only Balmer line which can still be emitted is the H
line, because
the upper levels of all other Balmer lines are now in the energy continuum. In addition
to the disappearance of these lines, another consequence of the lowering of the ionization
threshold is an increase in the continuum radiation. Free-bound and free-free transitions
become much more likely with a lowered threshold.
Secondly, some of the remaining spectral lines, whose upper states are yet below the
new ionization threshold, can still be erased. Although these lines are discrete transi-
tions, they appear as a continuum radiation not because their upper state is in the energy
continuum, but because of the Stark broadening. The upper lines of a series are close
to each other in the spectrum. If they are signicantly broadened so that their widths
exceed the distance between them, they will overlap with adjacent lines and merge into a
continuum. The continuum formed by this degeneracy of lines is the type of continuum
observed during the rst microsecond of the EDM discharge (see 6.1.6 and 6.2.1). The
Stark broadening of lines near the series limit has thus the same eect as the lowering
of the ionization threshold: it transforms the upper members of a spectral series into a
continuous spectrum. Therefore, this second eect can be considered as an apparent or
optical lowering of the ionization threshold [43, 44].
As showed in chapter 6, the spectra of EDM plasmas are characterized by the absence
of the H
line and other upper Balmer lines, by an strong continuum radiation and by
merging of almost all lines at the beginning of the discharge. All these characteristics can
thus be viewed as signs of the non-ideality of the EDM plasma.
7.2.4 Inglis-Teller relation
As previously said, upper spectral lines of a series can be merged if the plasma is dense,
due to their overlapping caused by Stark broadening. Since the number of these erased
7.3. SUMMARY OF THE PHYSICAL PROPERTIES OF EDM PLASMAS 95
lines depends on the plasma density, the last discernible line gives a rough estimation
of the density. For hydrogen Balmer lines, this estimation is given by the Inglis-Teller
relation:
log n = 23.26 7.5 log n
max
, (7.7)
where n is the density expressed in cm
3
and n
max
the quantum number of the upper
energy level of the last visible transition [43, 108, 154156]. This relation is mainly used in
astrophysics, to evaluate the electron density in the atmosphere of stars, for example. The
coecients of this relation are sometimes slightly dierent in literature. The uncertainty
factor of this relation is about 50% [108].
If we observe on EDM spectra the H
line (transition from level 3 to 2) but no H
line
(transition from level 4 to 2), then
3 n
max
< 4 ,
and the plasma density is at least 610
18
cm
3
according to relation (7.7). This rough
estimation is of the same order of magnitude as our previous density measurements and
conrms their validity.
Contrary to upper hydrogen Balmer lines which disappear because of their strong
broadening, metallic lines do not merge and can be observed in the EDM plasma spectrum,
as shown in chapter 6. The metallic lines are much less sensitive to Stark broadening, and
this explains why they are still observable. The Stark broadening of non-hydrogenic lines
is rather poorly documented, and few data are available [157]. Nevertheless, the FWHM of
a few atomic copper lines can be found. For example, in a plasma with an electron density
of 10
17
cm
3
and at 10000 K, the FWHM of the H
line is 10.8 , whereas the FWHM of

the Cu lines at 510.6, 515.3 and 521.8 nm are 0.43, 1.9 and 2.2 respectively [158]. This
shows that non-hydrogenic lines are not suciently broadened to merge and disappear
into a continuum, as upper hydrogen lines do.
7.3 Summary of the physical properties of EDM plas-
mas
According to the measurements presented in this work, EDM plasmas have the following
characteristics:
The plasma is composed of species coming from the cracking of the dielectric
molecules, mainly hydrogen, with contamination from the electrode materials.
The contamination is slightly higher in the vicinity of the electrodes, and increases
during the discharge.
The electron temperature T
e
is low, around 0.7 eV (8100 K), slightly higher at
the beginning of the discharge. The temperature is homogeneous in the plasma. The
assumption of LTE can be accepted due to the high plasma density, and therefore
the ion temperature T
i
should be similar to the electron temperature.
The electron density n
e
is extremely high, reaching 210
18
cm
3
during the rst
microsecond of the discharge, decreasing rapidly afterwards. The density is slightly
higher in the plasma center. The ion density n
i
should be roughly equal to the elec-
tron density, since multiple ions are unlikely with such a low electron temperature
and since quasineutrality should hold.
The plasma is very dense and cold, and therefore presents several non-ideal eects,
especially in the beginning of the discharge. The coupling parameter is around
0.33, and reaches 0.45 during the rst microsecond.
The Debye length
D
is around 6 nm, and the mean inter-particle distance a
is also around 6 nm. Therefore, a Debye sphere contains roughly only one particle.
Again, this is a sign of the plasma non-ideality.
The light emitted by the plasma is much more intense during the rst microsecond
than during the rest of the discharge.
Due to its high density, the EDM plasma could be not completely optically thin.
The opacity of a plasma increases in fact with the density. For example, ultra-
dense plasmas produced by exploding wires in water (n
e
= 2 10
21
cm
3
) are known
to be highly opaque and to radiate like a blackbody [33]. Consequently, the light
emitted by the very center of the EDM plasma at the beginning of the discharge
could be absorbed within the plasma and not be measured. The opacity can be a
problem for spectroscopic diagnostics, since the core of an opaque plasma cannot be
characterized with such diagnostics. Nevertheless, in our case, the plasma opacity
could be critical only during roughly the rst 500 ns of the discharge, when the
density reaches its maximum. But during this period, T
e
and n
e
can anyway not
be measured with the spectroscopic methods used in this work, because of the line
merging.
Energy is dissipated in the plasma by Joule heating, heat conduction, radiation and
mechanical work (plasma expansion). Energy is also used for evaporation, dissoci-
ation, ionization and excitation of the dielectric molecules and atoms. The energy
received by the workpiece is mainly brought by ion bombardment. Thermal run-
away of the very supercial layers is also probable in the beginning of the discharge,
due to the massive ow of energy at the breakdown.
For a plasma in LTE, the repartition of the densities of the dierent species is de-
scribed by the Saha equation. For a plasma containing electrons, one type of singly
charged ions and one type of atoms, the Saha equation is
n
e
n
i
n
a
=
Z
e
Z
i
Z
a
_
2m
e
k
B
T
h
2
_
3/2
exp
_
U
i
U
i
k
B
T
_
, (7.8)
where n
k
is the density of the species k, Z
k
its partition function, U
i
the ionization en-
ergy of the atom, and U
i
its correction due to non-ideality (equation (7.6)) [44,159].
In principle, this equation can be used for the calculation of the EDM plasma ion-
ization degree n
e
/n
a
. For a hydrogen plasma, Z
e
= 2, Z
i
= 1 and
Z
a
=
n
max
n=1
g
n
exp
_
E
n
k
B
T
_
=
n
max
n=1
2n
2
exp
_
U
i
(1 1/n
2
)
k
B
T
_
,
where U
i
= 13.6 eV and n
2
max
= U
i
/U
i
. In our case, we found Z
a
2. Assuming
in addition n
e
= n
i
= 10
18
cm
3
, we found n
e
/n
a
= 10
5
.
The ionization degree of the EDM plasma is certainly much higher, probably about
0.1 as for similar discharges [34]. High-pressure arcs have also similar densities
as EDM plasmas, and are known to be strongly ionized. The calculation gives
n
e
/n
a
= 10
5
, then we found n
a
= 10
23
cm
3
if we assume that n
e
= 10
18
cm
3
.
Such a high density of atoms is unlikely, since the density of molecules in the liquid
is around 10
22
cm
3
. This shows that the ionization degree calculated here is not
correct. The assumptions made are probably too strong. In particular, the Saha
equation as written in (7.8) does not take into account the complex composition of
the EDM plasma. But for a multi-component plasma, the Saha equation has no
analytical solution and becomes dicult to solve [159].
Assuming a temperature of 0.7 eV, a density of 210
18
cm
3
and that the plasma is
a fully ionized hydrogen plasma, the plasma pressure p at the beginning of the
discharge is
p = p
ideal
p = (n
e
+n
i
) k
B
T p ,
where p is a correction due to the plasma non-ideality [44, 46] given by
p =
e
2
24
0
D
(n
e
+n
i
) .
We found that p 4.2 bar. Since the density decreases during the discharge, the
pressure is also decreasing. The value found is lower than the pressure calculated
with the simulation [40] ( 10
3
bar). But the assumption of a fully ionized hy-
drogen plasma is obviously a simplication. In reality, the plasma composition is
more complex and the plasma is partially ionized (the H
line is emitted by neutral

hydrogen atoms). The pressure is thus higher than 4.2 bar, depending on the ion-
ization degree of the plasma. Nevertheless, the pressure is still much lower than the
value found with the simulation.
The estimated dimensions of the plasma are summarized in gure 7.6. The elec-
trode gap is around 100 m, and the plasma radius in the middle of the gap is also
around 100 200 m, growing with time. The contact surface between the plasma
and the electrodes can be estimated from measurements of crater diameter, and
is 10 m. Therefore, the current density j owing through the plasma, and
especially through the electrodeplasma interface, is extremely high, in the order
of 10
6
10
7
A/cm
2
.
anode
cathode
sheath (E MV/cm)
~

1
0
0
m
~ 10 m
~

1
0

n
m
j
M
A
/

c
m
2
sheath (E MV/cm)
p
l
a
s
m
a

c
o
l
u
m
n
~

1
0

n
m
E kV/cm
hydrogen plasma
n = 10 cm
e
18 -3
T = 0.7 eV
e
weakly non-ideal
p 10 bar
~ 100 - 200 m
Figure 7.6: Schematic image of the general physical properties of the EDM plasma.
Assuming a potential prole similar to those of electric arcs (see gure 2.3), most of
the voltage jump is located in thin sheaths near the electrodes. The sheath thickness
is in the order of the Debye length [53], 10 nm in our case. The electric eld
is thus very intense in these regions, about 10
6
V/cm. Consequently, electrons are
emitted from the cathode mainly by thermo-eld emission. The global eld in the
plasma column is lower, about 10
2
10
3
V/cm. On the other hand, the microeld
seen by the particles is also around 10
6
V/cm, due to the high plasma density.
In a non-ideal plasma, the movement of an electron is mainly inuenced by interac-
tions with ions. The mean free path of an electron
e
is thus given by
e
=
1
n
i
Q
ei
,
where Q
ei
is the cross section of thermal electronion scattering. This cross section
can be theoretically calculated for a strongly ionized non-ideal plasma [44]. The
result is
Q
ei
= 4
_
Ze
2
12
0
k
B
T
_
2
ln ,
where ln is the Coulomb logarithm. In the case of non-ideal plasma, the usual
Coulomb logarithm has to be replaced by
ln =
1
2
ln
_
1 +
_
3
D
_
2
_
, (7.9)
where
D
is a parameter similar to the coupling parameter dened by
D
.
=
Z
2
e
2
4
0
D
k
B
T
.
Note that
D

3/2
. For EDM plasmas,
D
0.33 and therefore ln 2.2 .
The cross section is thus Q
ei
1.3 10
13
cm
2
and the electron mean free path
e
80 nm.
Related to the electronion scattering cross section, the plasma electrical con-
ductivity can be estimated from the Spitzer conductivity, which is valid for fully
ionized plasmas where electronion collisions are dominant. The Spitzer conduc-
tivity is given by
=
E
2 (4
0
)
2
(2 k
B
T)
3/2
3/2
Ze
2
m
1/2
e
ln
, (7.10)
where
E
is a constant depending on Z, which takes account of electronelectron
collisions [44, 46, 160]. For singly charged ions (Z = 1), we have
E
= 0.582 and
thus
= 1.53 10
2
T
3/2
ln
[
1
m
1
] ,
with T in K. The Spitzer conductivity holds for ideal plasmas (ln 10), but can
be applied to weakly non-ideal plasmas if the Coulomb logarithm is replaced by
the modied Coulomb logarithm given in equation (7.9) [44]. The value found is
510
3
1
m
1
, much lower than the conductivity of ideal plasmas (10
6
1
m
1
).
Besides the discharge on-time, the other discharge parameters (current, dielectric,
electrode materials, polarity) have roughly no signicant inuence on the general
plasma physical properties enumerated here.
Chapter 8
Conclusions
Although Electrical Discharge Machining is known, used and studied since the 1950s, the
plasma created during this process remains puzzling. The important lack of knowledge
about this plasma is mainly due to the complexity of the phenomena occurring during
EDM, but also to the experimental diculties encountered for its characterization. The
EDM plasma is small, weakly luminous, of short duration, in a dicult environment, of
stochastic nature and poorly reproducible.
To our knowledge, this work is the rst systematic investigation of the EDM plasma
with various diagnostics. Considering the signicant experimental diculties related to it,
the rst success of this work is simply to have shown the applicability of plasma diagnos-
tics to the EDM discharges, particularly optical emission spectroscopy. The eorts made
for the application of the diagnostics have been well rewarded. The results obtained are
found to be extremely rich, and have permitted us to draw interesting conclusions about
the physical properties of this plasma, and also about the dierent phases of the EDM
process in general.
The pre-breakdown in water is characterized by the generation of numerous small
hydrogen bubbles at the cathode, created by electrolysis. These bubbles certainly enhance
the triggering of the breakdown, since the propagation of streamers is strongly facilitated
in a gaseous medium. On the other hand, no bubbles are observed during the pre-
breakdown in oil. In this dielectric, the breakdown process can rather be facilitated by
the presence of particles in the gap.
Fast current pulses have been measured during the pre-breakdown, which are associ-
ated with the propagation of streamers. The occurrence of these pulses increases with the
liquid conductivity. As expected, a weak light emission is also observable and is correlated
with the pre-breakdown current.
The duration of the pre-breakdown phase is not constant for given discharge param-
eters, but the values are distributed following a Weibull distribution. This shows that
the breakdown process is of stochastic nature, depending strongly on the dielectric and
electrode properties.
101
102 CHAPTER 8. CONCLUSIONS
After the breakdown, the plasma is established very rapidly, in less than 50 ns as
observed with fast imaging. The plasma light intensity is particularly high immediately
after the breakdown and approximately during the rst 500 ns. This rst phase is quite
dierent from the rest of the discharge, and the development of generators capable of
producing ultra-short discharges could be interesting for a very supercial machining.
During the rest of the discharge, the emitted light is weaker, and its intensity depends
mostly on the discharge current (more intense with higher current). Imaging shows that
the discharge excites a broad volume around the electrode gap, typically 200 m in diam-
eter, brighter in its center. This volume also clearly increases with the discharge current.
The plasma is found to grow slightly during the discharge, as expected according to crater
measurements reported elsewhere. Vapor bubbles are generated in water as in oil, by the
heat released from the plasma.
At the end of the discharge when the current is shut down, the plasma implodes and
disappears quickly. The ejection of eroded metallic particles out of the gap is observ-
able with imaging during the post-discharge. These particles emit light by incandescence,
which is conrmed by spectroscopic measurements. The spectrum of the post-discharge
light is indeed close to a blackbody spectrum at roughly 2200 K, demonstrating also that
the eroded particles are still in a liquid state immediately after the end of the discharge.
The spectroscopic analysis of the plasma light gives information about the plasma
composition. It is mostly composed by species coming from the cracking of the dielectric
molecules, mainly hydrogen in the case of water and oil. Since the plasma is formed from
the dielectric, the H
line, for example, is emitted by the whole plasma volume. The pres-
ence of numerous metallic spectral lines indicates that the electrodes are also contaminat-
ing the plasma. Time- and spatially-resolved measurements show that the contamination
is slightly higher in the vicinity of each electrode, and that the contamination coming
from the workpiece increases during the discharge, probably due to vaporization.
Workpiece erosion or electrode wear regime can also qualitatively be dierentiated with
spectroscopy. Changing the electrode polarity produces in fact very dierent spectra. An
increase in the discharge current has also an eect. Since discharges with high current
produce larger craters, the metallic spectral lines are more intense with a high discharge
current.
Almost all the spectral lines observed are atomic lines and not ionic lines, which
indicates that the plasma is cold. The electron temperature, determined with the two-
line method from copper line intensity measurements, is around 0.7 eV in the whole
plasma, slightly higher in the beginning of the discharge. This conrms that the EDM
plasma has a low temperature.
Discharges with short on-times produce spectra with strongly-broadened lines, es-
pecially the H
line, and with an important continuum radiation. Time-resolved spec-

troscopy conrms these observations. During the rst microsecond, spectral lines are so
broadened that they are all merged into a continuum. This eect is due to the high plasma
density in the very beginning of the discharge. Time- and spatially-resolved measurements
of the electron density, based on Stark broadening and shift measurements of the H
line,
show that the density is above 10
18
cm
3
in the beginning, and then decreases during
CONCLUSIONS 103
the discharge. But even after 50 s, the electron density remains high, above 10
16
cm
3
.
During the whole discharge, the density measured is slightly higher in the plasma center.
The EDM plasma has such a high density because it is formed from a liquid, which
is a dense medium, and because it is constantly submitted to the pressure imposed by
the surrounding liquid. The strong intensity of the continuum radiation can be related to
the high electron density. Free-bound and free-free radiations are indeed more likely in a
dense plasma. The high electron density also indicates that the plasma is certainly in LTE.
As previously mentioned, the EDM plasma is cold and dense. With a typical electron
temperature of 0.7 eV and a typical electron density of 10
18
cm
3
, its coupling parameter
is around 0.3, demonstrating that it is weakly non-ideal. Evaluations of the Debye
length, of the mean inter-particle distance and of the Coulomb logarithm, for example,
also show the non-ideality of the EDM plasma.
Furthermore, spectroscopic results conrm this conclusion. The broadening, the shift,
the asymmetrical shape and the structure of the H
line are characteristic signs of the

non-ideality of the plasma. The absence of upper Balmer lines such as the H
line, for
example, along with the merging of spectral lines in the beginning of the discharge are
other demonstrations of this fact.
The physical properties of this plasma are quite fascinating. In addition to an ex-
treme electron density, the electric elds present in the plasma and the current density
are particularly high, for example. It is also remarkable that the general plasma physical
properties are relatively insensitive to most of the discharge parameters, such as the dis-
charge current, the type of dielectric, the electrode materials and polarity. Whereas these
parameters certainly have a signicant inuence on the machining quality and eciency,
they roughly do not change the plasma densities, elds or temperatures.
Although this work has given many interesting and new results, further investigations
are necessary to have a complete understanding of the EDM plasma and process. For ex-
ample, we have only studied discharges used for standard erosion in die-sinking machines.
Plasma created during wire-cutting operations could have dierent properties, since very
high currents and short on-times are used (up to 1 kA during 2 s typically). Discharges
with low current (< 6 A) have neither been studied, because the plasma intensity with
such currents is too weak for our optical diagnostics. These discharges are used for pol-
ishing operations, where the workpiece is not signicantly eroded but just supercially
melted.
Furthermore, besides density and temperature measurements, optical emission spec-
troscopy gives mainly only qualitative results. Quantitative estimations of the plasma
contamination and of the workpiece vaporization would be useful for example, but can
unfortunately not be obtained with this diagnostic. The eective temperature at the
workpiece surface is also another important parameter which can not be measured with
emission spectroscopy.
It should be noted that the other diagnostics which can still be applied to the EDM
plasma are few. With miniaturized optical bres, active spectroscopic methods with
lasers could be possibly developed. Measurement of the electron density could thus be
104 CHAPTER 8. CONCLUSIONS
performed with interferometry, and information about bubbles and eroded particles could
be obtained with light scattering and absorption measurements. Acoustic diagnostics
could also be developed for the study of the formation, expansion and collapse of the
plasma and of bubbles [161].
The implementation of spectroscopic diagnostics on industrial machines could be con-
ceivable in the long term. Such diagnostics could be used to control the dielectric clean-
liness or to detect transitions into arcs, for example.
This rst investigation of the EDM plasma has emphasized its complexity and has
also opened new questions. It will be important in the future to correlate the properties
of the plasma with the performances and relevant characteristics of the EDM process,
in order to dene the perfect plasma for EDM. To have a complete comprehension of
the mechanisms of erosion and wear for example, we should understand precisely how
the energy is distributed between the plasma, the electrode and the workpiece, and how
it is transferred to them during the discharge. Until now, we can only compare with
similar situations, such as cathode spots in high-pressure arcs, for example. But we do
not have direct measurements of the heating and transfer mechanisms from the plasma
to the electrodes. A deep understanding of this could also explain in detail the origin of
the asymmetry between wear and erosion, which is a characteristic and crucial feature of
the EDM process. Moreover, with a better comprehension and control of the breakdown
phenomenon, more reproducible discharges could be obtained, which is also important
from the industrial point of view.
In conclusion, this work has given a rst insight into fundamental aspects of the EDM
plasma. In our opinion, we are convinced that such a study is necessary for further
improvements of the process, and that other investigations should be carried on. Like any
other fundamental study, it is clear that this work cannot answer now all the questions
related to the EDM process, and that no immediate industrial improvements and direct
applications can be obtained from it. Nevertheless, some of the results presented here have
already been implemented in a simulation program developed by Charmilles Technologies,
which calculates and optimizes the heat exchange during the discharge. This demonstrates
that this work is a rst promising step towards a better understanding of the EDM process,
and possibly towards future developments of this technique.
Appendix A
Stark broadening and shift of hydrogen
spectral lines
The Stark eect is the modication of the energy levels of an atom, a molecule or an ion,
due to the presence of an external electric eld. The Zeeman eect is the equivalent of
the Stark eect, but for the magnetic eld.
The Stark broadening is the broadening of spectral lines, caused by the modication
of the emitter energy levels by the Stark eect. In this case, the external electric eld is a
local micro-eld created by charged particles (ions or electrons) colliding with the emitter.
Though the Stark broadening aects every atom, we will present here some results for the
hydrogen atom only. Since we use the Stark broadening of the H
line for the electron

density determination, we will focus especially on the levels n = 2 and n = 3.
A.1 The Stark eect on hydrogen levels
The Hamiltonian H describing the electron of the hydrogen atom exposed to an electric
eld is composed of :
the unperturbed Hamiltonian
H
0
=
p
2
2m
e
e
2
4
0
R
,
where p = i
is the impulsion operator, R the position operator, m

e
and e the
mass and the charge of the electron. Note that if we take into account the spin-orbit
interaction and relativistic eects, the Hamiltonian becomes
H
0
=
p
2
2m
e
e
2
4
0
R
+
1
2m
2
e
c
2
e
2
4
0
R
3
L S + relativistic terms ,
where L is the angular momentum operator and S the spin operator [162]. This
Hamiltonian describes the ne structure of the hydrogen atom.
105
106 APPENDIX A. STARK BROADENING AND SHIFT OF HYDROGEN LINES
the Stark perturbing term W
S
due to the electric eld F
W
S
= e F R .
Thus we have :
H = H
0
+W
S
=
p
2
2m
e
e
2
4
0
R
e F R . (A.1)
Treating W
S
with the perturbation theory, we found that, to rst order, the unperturbed
energy level E
n
is split into (2n 1) new levels E
(1)
n
given by
E
(1)
n
= E
n
+n(n
2
n
1
) , (A.2)
with 0 n
1
, n
2
< n [163]. For example, the level n = 3 will be split into 5 levels : E
3
,
E
3
3 and E
3
6. The dierence in energy between the new levels are multiples of
, given by
=
6
2
0
m
e
e
F . (A.3)
We see that is linear with F. The eect of the electric eld on the levels given by (A.2)
is thus called the linear Stark eect.
Taking the ne structure into account, the equation (A.2) has to be modied as follows:
E
(1)
n
= E
n
+n(n
2
n
1
) +k m
l
m
s
, (A.4)
where m
l
is the magnetic quantum number, m
s
the spin quantum number and k a constant
(for example k = 3 10
5
eV for E
2
; 8.910
6
eV for E
3
3 ; 1.710
6
eV for E
3
and
E
3
+ 3).
The calculation of the energy levels (without the ne structure) at the second order
of the perturbation theory modies equation (A.2) by adding a term quadratic in F. We
obtain
E
(2)
n
= E
n
+
_
n(n
2
n
1
)
3a
0
e
2
_
F
. .
E
(1)
n
_
n
4
_
17n
2
3(n
2
n
1
)
2
9m
2
l
+ 19
_
a
3
0
16
_
F
2
,
where a
0
is the Bohr radius [163]. The energy shift given by this quadratic term is called
the quadratic Stark eect.
The linear Stark eect is almost only signicant for the hydrogen atom and hydrogen-
like ions, and with a strong electric eld (the eld is called weak when the Stark splitting
is small compared with the ne structure splitting, and strong otherwise). The quadratic
A.2. CONSEQUENCES OF THE STARK EFFECT ON H
EMISSION 107
eect is almost negligible for hydrogen. Other elements are sensitive to the quadratic
eect, but the eect is small especially for heavy elements. The hydrogen lines are thus
the most broadened ones in a multi-element spectrum.
A.2 Consequences of the Stark eect on H
emission
As previously mentioned, the H
line is emitted by the transition from the level n = 3 to

n = 2. Figure A.1 shows the splitting of these two levels by the Stark eect, according to
equation (A.2).
n=2
n=3
without electric field with electric field
E
3
E
2
E
3
E
2
E + 3
3
D
E + 6
3
D
E - 3
3
D
E - 6
3
D
E + 2
2
D
E - 2
2
D
H
a
n
H
a
emission
n=3
n=2
n
D/h
0 1 2 3 4 -1 -2 -3 -4
p s p
5 6 8 -8 -6 -5
p p
s s
Figure A.1: Splitting of the levels n = 2 and n = 3 of the hydrogen atom due to the Stark
eect, and its consequences on H
emission.
Due to the levels splitting, the possible transitions are multiplied. The H
emission is
not a single line anymore, but it is composed of several lines. Figure A.1 shows also the
frequencies emitted by these new transitions (the zero frequency reference is the transition
from E
3
to E
2
). The letters and denote the polarization of the emitted light. The
polarization corresponds to a transition with m
l
= 0, with the electric vector parallel
to the electric eld; the polarization corresponds to a transition with m
l
= 1, with
the electric vector perpendicular to the electric eld [163].
The frequency range covered by the H
emission is thus larger with an electric eld

than without it. This results in the broadening of the H
emission: this is the origin of the

Stark broadening. If the plasma density increases, the electric eld seen by the emitter
will also increase, because the charged perturbing particles will be closer. As a result,
will be larger according to equation (A.3). So we qualitatively see that the H
frequency
range, i.e. the Stark broadening of H
, increases with density.

Taking into account the ne structure of the hydrogen atom, the H
emission is com-
posed of several lines even without electric eld, because a splitting of the levels already
exist with F = 0. But the number of lines is also increased by the Stark eect, as shown
in gure A.2 according to equation (A.4).
n=2
n=3
no field
(fine structure)
weak field strong field
D
5/2
P
3/2
, D
3/2
S
1/2
, P
1/2
P
3/2
S
1/2
, P
1/2
n
0 1 2 3 4 -1 -2 -3 -4
p s p
5 6 8 -8 -6 -5
p p
s s
H
a
lines
(strong field)
E
3
E
2
E + 3
3
D
E + 6
3
D
E - 3
3
D
E - 6
3
D
E + 2
2
D
E - 2
2
D
Figure A.2: Fine structure of the levels n = 2 and n = 3 of the hydrogen atom, splitting due
to the Stark eect in weak and strong eld, and H
emission in strong eld.

In addition to the broadening, the H
emission is also shifted to higher wavelengths.

Equation (A.4) shows that the center of gravity of the H
emission (E
3
E
2
) is shifted
from its position without eld, due to the last term. However, this is not sucient to
explain the measured shifts of hydrogen lines. While we can give a simple explanation
for the Stark broadening, the origin of the Stark shift of spectral lines is much more
complicated. It involves several eects such as the quadratic Stark eect, interaction with
eld gradient and emitters-electrons collisions, whereas Stark broadening involves mainly
emitterion collisions [107].
A.2. CONSEQUENCES OF THE STARK EFFECT ON H
EMISSION 109
Although the origin of line shape modication by Stark eect can be simply understood
as presented here, the precise calculations of line broadening and shift require sophisticated
models [112117].
References
[1] http://www.charmilles.ch .
[2] http://www.agie-charmilles.com .
[3] B. M. Schumacher, After 60 years of EDM the discharge process remains still
disputed, J. Mater. Process. Technol., 149, 376381 (2004).
[4] A. Anders, Tracking down the origin of arc plasma science, I. Early pulsed and
oscillating discharges, IEEE Trans. Plasma Sci., 31(4), 10521059 (2003).
[5] J. Priestley, Experiments on the circular spots made on pieces of metal by large
electrical explosions, in The history and present state of electricity with original
experiments, vol. II, 3rd ed., London (1775).
[6] B. Dibner, Galvani - Volta, A controversy that led to the discovery of useful elec-
tricity, CT: Burndy Library, Norwalk (1952).
[7] A. Anders, Tracking down the origin of arc plasma science, II. Early continuous
discharges, IEEE Trans. Plasma Sci., 31(4), 10601069 (2003).
[8] Handbook of vacuum arc science and technology, fundamentals and applications,
edited by R. L. Boxman, D. M. Sanders and P. J. Martin, Noyes publications, Park
Ridge (1995).
[9] B. R. Lazarenko, About the inversion of metal erosion and methods to ght ravage
of electric contacts, WEI-Institut, Moscow (in Russian) (1943).
[10] L. H. Germer and F. E. Haworth, Erosion of electrical contacts on make, J. Appl.
Phys., 20(11), 10851109 (1949).
[11] J. D. Cobine and E. E. Burger, Analysis of electrode phenomena in the high-current
arc, J. Appl. Phys., 26(7), 895900 (1955).
[12] A. S. Zingerman, The eect of thermal conductivity upon the electrical erosion of
metals, Sov. Phys. Tech. Phys., 1, 19451958 (1956).
[13] T. Sato, T. Mizutani, and K. Kawata, Electro-discharge machine for microhole
drilling, National Technical Report, 31, 725733 (in Japanese) (1985).
111
112 REFERENCES
[14] D. F. Dauw and B. Van Coppenolle, On the evolution of EDM research, part 2:
From fundamental research to applied research, in Proceedings of the 11th Interna-
tional Symposium for Electro Machining (ISEM-11), 133142 (Lausanne, Switzer-
land, 1995).
[15] K. Kobayashi, The present and future developments of EDM and ECM, in Pro-
ceedings of the 11th International Symposium for Electro Machining (ISEM-11),
2947 (Lausanne, Switzerland, 1995).
[16] R. Lttgen, EDM die-sinking - state of the art, in Proceedings of the 12th Interna-
tional Symposium for Electro Machining (ISEM-12), 341350 (Aachen, Germany,
1998).
[17] K. H. Ho and S. T. Newman, State of the art electrical discharge machining
(EDM), Int. J. Mach. Tools and Manufacture, 43(13), 12871300 (2003).
[18] K. H. Ho, S. T. Newman, S. Rahimifard, and R. D. Allen, State of the art in wire
electrical discharge machining (WEDM), Int. J. Mach. Tools and Manufacture,
44(12-13), 12471259 (2004).
[19] B. Lauwers, J.-P. Kruth, W. Liu, W. Eeraerts, B. Schacht, and P. Bleys, Investi-
gation of material removal mechanisms in EDM of composite ceramic materials, J.
Mater. Process. Technol., 149, 347352 (2004).
[20] T. Tani, Y. Fukuzawa, N. Mohri, N. Saito, and M. Okada, Machining phenomena in
WEDM of insulating ceramics, J. Mater. Process. Technol., 149, 124128 (2004).
[21] H. Takezawa, H. Hamamatsu, N. Mohri, and N. Saito, Development of micro-EDM-
center with rapidly sharpened electrode, J. Mater. Process. Technol., 149, 112116
(2004).
[22] J. Fleischer, T. Masuzawa, J. Schmidt, and M. Knoll, New applications for micro-
EDM, J. Mater. Process. Technol., 149, 246249 (2004).
[23] I. Beltrami, C. Joseph, R. Clavel, J.-P. Bacher, and S. Bottinelli, Micro- and nano-
electric discharge machining, J. Mater. Process. Technol., 149, 263265 (2004).
[24] C. Joseph, Contribution laccroissement des performances du processus de micro-
EDM par lutilisation dun robot dynamique leve et de haute prcision, PhD
thesis, Ecole Polytechnique Fdrale de Lausanne, Switzerland (2005).
[25] G. Cusanelli, A. Hessler-Wyser, F. Bobard, R. Demellayer, R. Perez, and
R. Flkiger, Microstructure at submicron scale of the white layer produced by
EDM technique, J. Mater. Process. Technol., 149, 289295 (2004).
[26] F. Klocke, D. Lung, G. Antonoglou, and D. Thomaidis, The eects of powder
suspended dielectrics on the thermal inuenced zone by electrodischarge machining
with small discharge energies, J. Mater. Process. Technol., 149, 191197 (2004).
REFERENCES 113
[27] S. Das, M. Klotz, and F. Klocke, EDM simulation: nite element-based calculation
of deformation, microstructure and residual stresses, J. Mater. Process. Technol.,
142(2), 434451 (2003).
[28] R. Perez, H. Rojas, G. Wlder, and R. Flkiger, Theoretical modeling of energy
balance in electroerosion, J. Mater. Process. Technol., 149, 198203 (2004).
[29] H.-P. Schulze, R. Herms, H. Juhr, W. Schtzing, and G. Wollenberg, Comparison of
measured and simulated crater morphology for EDM, J. Mater. Process. Technol.,
149, 316322 (2004).
[30] J. C. Devins, S. J. Rzad, and R. J. Schwabe, Breakdown and prebreakdown phe-
nomena in liquids, J. Appl. Phys., 52(7), 45314545 (1981).
[31] P. E. Frayssines, N. Bonifaci, A. Denat, and O. Lesaint, Streamers in liquid nitro-
gen: characterization and spectroscopic determination of gaseous lament temper-
ature and electron density, J. Phys. D: Appl. Phys., 35(4), 369377 (2002).
[32] E. O. Forster, Progress in the understanding of electrical breakdown in condensed
matter, J. Phys. D: Appl. Phys., 23(12), 15061514 (1990).
[33] E. A. Martin, Experimental investigation of a high-energy density, high-pressure
arc plasma, J. Appl. Phys., 31(2), 255267 (1960).
[34] P. Barmann, S. Krll, and A. Sunesson, Spectroscopic measurements of streamer
laments in electric breakdown in a dielectric liquid, J. Phys. D: Appl. Phys.,
29(5), 11881196 (1996).
[35] S. V. Kukhlevsky, Cs. Ver, J. Kaiser, L. Kozma, L. Palladino, A. Reale, G. Tomas-
setti, F. Flora, and G. Giordano, Generation of pure, high-density metal-vapor
plasma by capillary discharge, Appl. Phys. Let., 74(19), 27792781 (1999).
[36] G. Wlder, New technologies and EDM, in Proceedings of the 12th International
Symposium for Electro Machining (ISEM-12), 331340 (Aachen, Germany, 1998).
[37] F. Van Dijck, Physico-mathematical analysis of the electro discharge machining pro-
cess, PhD thesis, Catholic University of Leuven, Belgium (1973).
[38] D. D. DiBitonto, P. T. Eubank, M. R. Patel, and M. A. Barrufet, Theoretical
models of the electrical discharge machining process. I. A simple cathode erosion
model, J. Appl. Phys., 66(9), 40954103 (1989).
[39] M. R. Patel, M. A. Barrufet, P. T. Eubank, and D. D. DiBitonto, Theoretical
models of the electrical discharge machining process. II. The anode erosion model,
J. Appl. Phys., 66(9), 41044111 (1989).
[40] P. T. Eubank, M. R. Patel, M. A. Barrufet, and B. Bozkurt, Theoretical models
of the electrical discharge machining process. III. The variable mass, cylindrical
plasma model, J. Appl. Phys., 73(11), 79007909 (1993).
114 REFERENCES
[41] K. Albinski, K. Musiol, A. Miernikiewicz, S. Labuz, and M. Malota, The tem-
perature of a plasma used in electrical discharge machining, Plasma Sources Sci.
Technol., 5(4), 736742 (1996).
[42] B. W. Pillans, M. H. Evensen, H. F. Taylor, P. T. Eubank, and L. Ma, Fiber optic
diagnostic techniques applied to electrical discharge machining sparks, J. Appl.
Phys., 91(4), 17801786 (2002).
[43] V. E. Fortov and I. T. Iakubov, Physics of nonideal plasma, Hemisphere, New York
(1990).
[44] Transport and optical properties of nonideal plasma, edited by G. A. Kobzev,
I. T. Iakubov and M. M. Popovich, Plenum Press, New York (1995).
[45] R. Redmer, Physical properties of dense, low-temperature plasmas, Phys. Rep.,
282(2-3), 35157 (1997).
[46] K. Gnther and R. Radtke, Electric properties of weakly nonideal plasmas,
Birkhuser Verlag, Basel (1984).
[47] J. M. Meek and J. D. Craggs, Electrical breakdown of gases, Clarendon Press, Oxford
(1953).
[48] L. B. Loeb, Electrical breakdown of gases with steady or direct current impulse
potentials, in Encyclopedia of Physics, volume XXII: Gas discharges II, edited by
S. Flgge, Springer Verlag, Berlin (1956).
[49] Y. P. Raizer, Gas discharge physics, Springer-Verlag, Berlin (1991).
[50] A. E. Guile and B. Jttner, Basic erosion processes of oxidized and clean metal
cathodes by electric arcs, IEEE Trans. Plasma Sci., 8(3), 259269 (1980).
[51] C. W. Kimblin, A review of arcing phenomena in vacuum and in the transition to
atmospheric pressure arcs, IEEE Trans. Plasma Sci., 10(4), 322330 (1982).
[52] V. I. Rakhovsky, State of the art of physical models of vacuum arc cathode spots,
IEEE Trans. Plasma Sci., 15(5), 481487 (1987).
[53] M. S. Benilov and A. Marotta, A model of the cathode region of atmospheric
pressure arcs, J. Phys. D: Appl. Phys., 28(9), 18691882 (1995).
[54] R. L. Boxman, S. Goldsmith, and A. Greenwood, Twenty-ve years of progress in
vacuum arc research and utilization, IEEE Trans. Plasma Sci., 25(6), 11741186
(1997).
[55] I. I. Beilis, State of the theory of vacuum arcs, IEEE Trans. Plasma Sci., 29(5),
657670 (2001).
[56] E. Hantzsche, Mysteries of the arc cathode spot: a retrospective glance, IEEE
Trans. Plasma Sci., 31(5), 799808 (2003).
REFERENCES 115
[57] B. Jttner, V. F. Puchkarev, E. Hantzsche, and I. Beilis, Cathode spots, in Hand-
book of vacuum arc science and technology, fundamentals and applications, edited
by R. L. Boxman, D. M. Sanders and P. J. Martin, Noyes publications, Park Ridge
(1995).
[58] A. W. Behrens, J. Ginzel, and F.-L. Bruhns, Threshold technology and its appli-
cation for gap status detection, J. Mater. Process. Technol., 149, 310315 (2004).
[59] P. K. Watson, W. G. Chadband, and M. Sadeghzadeh-Araghi, The role of elec-
trostatic and hydrodynamic forces in the negative-point breakdown of liquid di-
electrics, IEEE Trans. Electr. Ins., 26(4), 543559 (1991).
[60] W. G. Chadband, The electrical breakdown of insulating oil, Power Engineering
Journal, , 6167 (March 1992).
[61] R. Tobazon, Prebreakdown phenomena in dielectric liquids, IEEE Trans. Diel.
Electr. Ins., 1(6), 11321147 (1994).
[62] H. M. Jones and E. E. Kunhardt, Development of pulsed dielectric breakdown in
liquids, J. Phys. D: Appl. Phys., 28(1), 178188 (1995).
[63] A. Beroual, M. Zahn, A. Badent, K. Kist, A. J. Schwabe, H. Yamashita, K. Ya-
mazawa, M. Danikas, W. D. Chadband, and Y. Torshin, Propagation and structure
of streamers in liquid dielectrics, IEEE Electr. Insul. Mag., 14(2), 617 (1998).
[64] I. V. Lisitsyn, H. Nomiyama, S. Katsuki, and H. Akiyama, Thermal processes in a
streamer discharge in water, IEEE Trans. Diel. Electr. Ins., 6(3), 351356 (1999).
[65] H. M. Jones and E. E. Kunhardt, The inuence of pressure and conductivity on
the pulsed breakdown of water, IEEE Trans. Diel. Electr. Ins., 1(6), 10161025
(1994).
[66] O. Lesaint and G. Massala, Positive streamer propagation in large oil gaps, Ex-
perimental characterization of propagation modes, IEEE Trans. Diel. Electr. Ins.,
5(3), 360370 (1998).
[67] G. Massala and O. Lesaint, Positive streamer propagation in large oil gaps, Electri-
cal properties of streamers, IEEE Trans. Diel. Electr. Ins., 5(3), 371381 (1998).
[68] A. Denat, F. Jomni, F. Aitken, and N. Bonifaci, Thermally and electrically induced
bubbles in liquid argon and nitrogen, IEEE Trans. Diel. Electr. Ins., 9(1), 1722
(2002).
[69] H. Yamada, T. Sato, and T. Fujiwara, High-speed photography of prebreakdown
phenomena in dielectric liquids under highly non-uniform eld conditions, J. Phys.
D: Appl. Phys., 23(12), 17151722 (1990).
116 REFERENCES
[70] A. Beroual, C. Marteau, and R. Tobazon, Behavior of streamers in liquids under
step voltages in point-plane geometry, IEEE Trans. Electr. Ins., 23(6), 955959
(1988).
[71] H. Yamashita and H. Amano, Prebreakdown phenomena in hydrocarbon liquids,
IEEE Trans. Electr. Ins., 23(4), 739750 (1988).
[72] O. Lesaint and P. Gournay, Initiation and propagation thresholds of positive pre-
breakdown phenomena in hydrocarbon liquids, IEEE Trans. Diel. Electr. Insul.,
1(4), 702708 (1994).
[73] Y. Nakao, H. Itoh, S. Hoshino, Y. Sakai, and H. Tagashira, Eects of additives
on prebreakdown phenomena in n-hexane, IEEE Trans. Diel. Electr. Ins., 1(3),
383390 (1994).
[74] T. Zinoulis, I. Argirakis, D. W. Auckland, and W. G. Chadband, The role of
particles in discharge initiation in liquids, in IEEE 11th International Conference
on Conduction and Breakdown in Dielectric Liquids (ICDL 93), 421425 (Baden-
Dttwil, Switzerland, 1993).
[75] M. Hara, J. Suheiro, H. Maeda, and H. Nakashima, DC pre-breakdwon phenom-
ena and breakdown characteristics in the presence of conducting particles in liquid
nitrogen, IEEE Trans. Diel. Electr. Ins., 9(1), 2330 (2002).
[76] W. G. Chadband, From bubbles to breakdown, or vice-versa, in IEEE 11th In-
ternational Conference on Conduction and Breakdown in Dielectric Liquids (ICDL
93), 184193 (Baden-Dttwil, Switzerland, 1993).
[77] T. Aka-Ngnui and A. Beroual, Bubble dynamics and transition into streamers in
liquid dielectrics under a high divergent electric eld, J. Phys. D: Appl. Phys.,
34(9), 14081412 (2001).
[78] J. Qian, R. P. Joshi, J. Kolb, K. H. Schoenbach, J. Dickens, A. Neuber, M. Butcher,
M. Cevallos, H. Krompholz, E. Schamiloglu, and J. Gaudet, Microbubble-based
model analysis of liquid breakdown initiation by a submicrosecond pulse, J. Appl.
Phys., 97(11), 113304 (2005).
[79] S. M. Korobeinikov, A. V. Melekhov, and A. S. Besov, Breakdown initiation in
water with the aid of bubbles, High Temperature, 40(5), 652659 (2002).
[80] R. Kattan, A. Denat, and O. Lesaint, Generation, growth, and collapse of vapor
bubbles in hydrocarbon liquids under a high divergent electric eld, J. Appl. Phys.,
66(9), 40624066 (1989).
[81] A. Denat, N. Bonifaci, and M. Nur, Spectral analysis of the light emitted by stream-
ers in hydrocarbon liquids, IEEE Trans. Diel. Electr. Ins., 5(3), 382387 (1998).
[82] H. Akiyama, Streamer discharges in liquids and their applications, IEEE Trans.
Diel. Electr. Ins., 7(5), 646653 (2000).
REFERENCES 117
[83] M. Sato, T. Ohgiyama, and J. S. Clements, Formation of chemical species and
their eects on microorganisms using a pulsed high-voltage discharge in water,
IEEE Trans. Ind. Appl., 32(1), 106112 (1996).
[84] M. P. Brenner, S. Hilgenfeldt, and D. Lohse, Single-bubble sonoluminescence, Rev.
Mod. Phys., 74(2), 425484 (2002).
[85] D. J. Flannigan and K. S. Suslick, Plasma formation and temperature measurement
during single-bubble cavitation, Nature, 434(7029), 5255 (2005).
[86] J. W. Robinson, M. Ham, and A. N. Balaster, Ultraviolet radiation from electrical
discharges in water, J. Appl. Phys., 44(1), 7275 (1973).
[87] A. W. DeSilva and J. D. Katsouros, Electrical conductivity of dense copper and
aluminium plasmas, Phys. Rev. E, 57(5), 59455951 (1998).
[88] A. Grinenko, A. Sayapin, V. Tz. Gurovich, S. Emov, J. Felsteiner, and Ya. E.
Krasik, Underwater electrical explosion of a Cu wire, J. Appl. Phys., 97(2),
023303 (2005).
[89] J. W. Robinson, Measurements of plasma energy density and conductivity from 3
to 120 kbar, J. Appl. Phys., 38(1), 210216 (1967).
[90] S. V. Kukhlevsky, J. Kaiser, O. Samek, M. Liska, and J. Erostyak, Stark spec-
troscopy measurements of electron density of ablative discharges in Teon-(CF
2
)
n
capillaries, J. Phys. D: Appl. Phys., 33(9), 10901092 (2000).
[91] Y. Ping, I. Geltner, A. Morozov, and S. Suckewer, Interferometric measurements of
plasma density in microcapillaries and laser sparks, Phys. Plasmas, 9(11), 4756
4766 (2002).
[92] L. S. Caballero, H. Chuaqui, M. Favre, I. Mitchell, and E. Wyndham, Plasma jet
emission in fast-pulsed capillary discharges, J. Appl. Phys., 98(2), 023305 (2005).
[93] C. Colon, G. Hatem, E. Verdugo, P. Ruiz, and J. Campos, Measurement of the
Stark broadening and shift parameters for several ultraviolet lines of singly ionized
aluminium, J. Appl. Phys., 73(10), 47524758 (1993).
[94] M. A. Hafez, M. A. Khedr, F. F. Elaksher, and Y. E. Gamal, Characteristics of
Cu plasma produced by a laser interaction with a solid target, Plasma Sources Sci.
Technol., 12(2), 185198 (2003).
[95] Y. Zhang, L. Li, and G. Zhang, Spectroscopic measurements of plasma inside the
keyhole in deep penetration laser welding, J. Phys. D: Appl. Phys., 38(5), 703710
(2005).
[96] V. A. Alekseev and I. T. Iakubov, Non-ideal plasmas of metal vapours, Phys. Rep.,
96(1), 169 (1983).
118 REFERENCES
[97] K. Terashima, L. Howald, H. Haefke, and H.-J. Gntherodt, Development of
a mesoscale/nanoscale plasma generator, Thin Solid Films, 281-282, 634636
(1996).
[98] Y. Hirata, M. Fukushima, T. Sano, K. Ozaki, and T. Ohji, Micro-arc discharge
phenomena, Vacuum, 59(1), 142151 (2000).
[99] T. Ito, T. Izaki, and K. Terashima, Development of plasma chip, Surf. Coat.
Technol., 133-134, 497500 (2000).
[100] C. Penache, M. Miclea, A. Bruning-Demian, O. Hohn, S. Schlssler, T. Jahnke,
K. Niemax, and H. Schmidt-Bcking, Characterization of a high-pressure microdis-
charge using diode laser atomic absorption spectroscopy, Plasma Sources Sci. Tech-
nol., 11(4), 476483 (2002).
[101] M. Miclea, K. Kunze, J. Franzke, and K. Niemax, Plasmas for lab-on-the-chip
applications, Spectrochim. Acta B, 57(10), 15851592 (2002).
[102] R. Longwitz, Study of gas ionization in a glow discharge and development of a micro
gas ionizer for gas detection and analysis, PhD thesis, Ecole Polytechnique Fdrale
de Lausanne, Switzerland (2004).
[103] T. Ito, T. Izaki, and K. Terashima, Application of microscale plasma to material
processing, Thin Solid Films, 386(2), 300304 (2001).
[104] O. B. Postel and M. A. Cappelli, Parametric study of the vacuum ultraviolet emis-
sion and electrical characteristics of a He-Xe microdischarge, J. Appl. Phys., 89(9),
47194726 (2001).
[105] K. H. Becker, K. H. Schoenbach, and J. G. Eden, Microplasmas and applications,
J. Phys. D: Appl. Phys., 39(3), R55R70 (2006).
[106] G. Smon, Guide pratique dusinage par tincelage, Ateliers des Charmilles S.A.,
Genve (1977).
[107] H. R. Griem, Principles of Plasma Spectroscopy, University Press, Cambridge
(1997).
[108] R. Rompe and M. Steenbeck, Progress in plasmas and gas electronics, volume 1,
Akademie-Verlag, Berlin (1975).
[109] T. Fujimoto, Plasma Spectroscopy, Clarendon Press, Oxford (2004).
[110] K. Fu, M. Jogwich, M. Knebel, and K. Wiesemann, Atomic transition probabilities
and lifetimes for the Cu I system, At. Data Nucl. Data Tables, 61(1), 130 (1995).
[111] H. R. Griem, Spectral line broadening by plasmas, Academic, New York (1974).
REFERENCES 119
[112] M. A. Gigosos and V. Cardenoso, New plasma diagnosis tables of hydrogen Stark
broadening including ion dynamics, J. Phys. B: At. Mol. Opt. Phys., 29(20),
47954838 (1996).
[113] S. Bddeker, S. Gnter, A. Knies, L. Hitzschke, and H.-J. Kunze, Shift and width
of the H alpha line of hydrogen in dense plasmas, Phys. Rev. E, 47(4), 27852791
(1993).
[114] W. Olchawa, Computer simulations of hydrogen spectral line shapes in dense plas-
mas, J. Quant. Spectrosc. Radiat. Transfer, 74(4), 417429 (2002).
[115] S. A. Flih, E. Oks, and Y. Vitel, Comparison of the Stark widths and shifts of the
H-alpha line measured in a ash tube plasma with theoretical results, J. Phys. B:
At. Mol. Opt. Phys., 36(2), 283296 (2003).
[116] S. Gnter and A. Knies, Shifts and asymmetry parameters of hydrogen Balmer
lines in dense plasmas, Phys. Rev. E, 55(1), 907911 (1997).
[117] A. Escarguel, E. Oks, J. Richou, and D. Volodko, Highly nonlinear, sign-varying
shift of hydrogen spectral lines in dense plasmas, Phys. Rev. E, 62(2), 26672671
(2000).
[118] W. Weibull, A statistical distribution function of wide applicability, J. Appl.
Mech., 18, 293297 (1951).
[119] C. J. Frei, Weibull statistical analysis of dielectric breakdown in n-hexane, IEEE
Trans. on Electr. Insul., 20(2), 303307 (1985).
[120] C. J. Frei, L. He, and K. Mller, Implications from the inuence of electrohydrody-
namic motion on breakdown with respect to the discharge mechanism in dielectric
liquids, IEEE Trans. on Electr. Insul., 24(2), 169174 (1989).
[121] A. Karden, Der Entladekanal - Was passiert wirklich bei der Funkenerosion?, Bericht
Nr. 272-1/00, Laboratorium fr Werkzeugmaschinen und Betriebslehre, Rheinisch-
Westflischen Technischen Hochschule, Aachen (2001).
[122] E. Gidalevich and R. L. Boxman, Sub- and supersonic expansion of an arc channel
in liquid, J. Phys. D: Appl. Phys., 39(4), 652659 (2006).
[123] B. Revaz, G. Witz, and R. Flkiger, Properties of the plasma channel in liquid
discharges inferred from cathode local temperature measurements, J. Appl. Phys.,
98(11), 113305 (2005).
[124] P. C. Pandey and S. T. Jilani, Plasma channel growth and the resolidied layer in
EDM, Precis. Eng., 8(2), 104110 (1986).
[125] G. Cusanelli, PhD thesis, Universit de Genve, Switzerland (2004).
120 REFERENCES
[126] A. E. Berkowitz and J. L. Walter, Spark erosion: a method for producing rapidly
quenched ne powders, J. Mater. Res., 2(2), 277288 (1987).
[127] J. Carrey, H. B. Radousky, and A. E. Berkowitz, Spark-eroded particles: Inuence
of processing parameters, J. Appl. Phys., 95(3), 823829 (2004).
[128] C. E. Moore, A multiplet table of astrophysical interest, U.S. Department of Com-
merce, National Bureau of Standards, Washington (1959).
[129] PLASUS SpecLine Software, http://www.plasus.de .
[130] J. Reader, C. H. Corliss, W. L. Wiese, and G. A. Martin, Wavelengths and transition
probabilities for atoms and atomic ions, U.S. Department of Commerce, National
Bureau of Standards, Washington (1980).
[131] R. W. B. Pearse and A. G. Gaydon, The identication of molecular spectra, Wiley,
New York (1950).
[132] NIST Atomic Spectra Database, http://www.physics.nist.gov/PhysRefData/ASD .
[133] D. J. Heading, G. R. Bennett, J. S. Wark, and R. W. Lee, Novel plasma source for
dense plasma eects, Phys. Rev. Lett., 74(18), 36163619 (1995).
[134] Y. Vitel, M. El Bezzari, T. V. Gavrilova, M. Skowronek, L. G. Dyachkov, and
Y. K. Kurilenkov, On dense hydrogen and noble gases plasma spectra, Physica B,
228(1-2), 161165 (1996).
[135] A. Escarguel, B. Ferhat, A. Lesage, and J. Richou, A single laser spark in aqueous
medium, J. Quant. Spectrosc. Radiat. Transfer, 64(4), 353361 (2000).
[136] B. Revaz, J. Emery, G. Witz, R. Flkiger, R. Perez, J. Carron, and M. Rappaz,
Local temperature response to pulsed discharges in electronic discharge machining
(EDM) environment, IEEE Trans. Plasma Sci., 33(3), 10661071 (2005).
[137] H. R. Griem, Validity of local thermal equilibrium in plasma spectroscopy, Phys.
Rev., 131(3), 11701176 (1963).
[138] K. Song, H. Cha, J. Lee, and Y.-I. Lee, Investigation of the line-broadening mech-
anism for laser-induced copper plasma by time-resolved laser-induced breakdown
spectroscopy, Microchem. J., 63(1), 5360 (1999).
[139] J. D. Huba, NRL Plasma formulary, Naval Research Laboratory, Washington
(2004).
[140] M. M. Litvak and D. F. Edwards, Spectroscopic studies of laser-produced hydrogen
plasma, IEEE J. Quant. Electronics, 2(9), 486492 (1966).
[141] W. L. Wiese, D. E. Kelleher, and D. R. Paquette, Detailed study of the Stark
broadening of Balmer lines in a high-density plasma, Phys. Rev. A, 6(3), 1132
1153 (1972).
REFERENCES 121
[142] Y. Vitel, Experimental study of H
broadening and shift in dense argon plasmas,

J. Phys. B: At. Mol. Phys., 20(10), 23272337 (1987).
[143] C. Parigger, J. W. L. Lewis, and D. Plemmons, Electron number density and
temperature mesurement in a laser-induced hydrogen plasma, J. Quant. Spectrosc.
Radiat. Transfer, 53(3), 249255 (1995).
[144] Y. Vitel, T. V. Gavrilova, L. G. Dyachkov, and Y. K. Kurilenkov, Spectra of dense
pure hydrogen plasma in Balmer area, J. Quant. Spectrosc. Radiat. Transfer, 83(3-
4), 387405 (2004).
[145] A. Cohn, P. Bakshi, and G. Kalman, Linear Stark eect due to resonant interactions
of static and dynamic elds, Phys. Rev. Lett., 29(6), 324327 (1972).
[146] C. C. Gallagher and M. A. Levine, Balmer-line anomalies in a turbulent plasma,
Phys. Rev. Lett., 30(19), 897900 (1973).
[147] W. R. Rutgers and H. W. Kalfsbeek, Calculations and measurements of the dy-
namic Stark eect in hydrogen, Z. Naturforsch., 30(a), 739749 (1975).
[148] A. I. Zhuzhunashvili and E. A. Oks, Technique of optical polarization measurements
of plasma Langmuir turbulence spectrum, Sov. Phys. JETP, 46(6), 11221129
(1977).
[149] K. H. Finken, R. Buchwald, G. Bertschinger, and H.-J. Kunze, Investigation of the
H alpha line in dense plasmas, Phys. Rev. A, 21(1), 200206 (1980).
[150] V. P. Gavrilenko and E. A. Oks, Intra-Stark spectroscopy of Coulomb radiators
in a plasma with quasimonochromatic electric eld, Sov. J. Plasma Phys., 13(1),
2228 (1980).
[151] G. Kalman and P. Bakshi, Microeld eects: strongly coupled plasmas, J. Quant.
Spectrosc. Radiat. Transfer, 44(1), 19 (1990).
[152] P. Bakshi and G. Kalman, Hydrogenic spectroscopy for various unusual plasmas,
J. Quant. Spectrosc. Radiat. Transfer, 44(1), 93100 (1990).
[153] E. A. Oks, S. Bddeker, and H.-J. Kunze, Spectroscopy of atomic hydrogen in
dense plasmas in the presence of dynamic elds: Intra-Stark spectroscopy, Phys.
Rev. A, 44(12), 83388347 (1991).
[154] D. R. Inglis and E. Teller, Ionic depression of series limits in one-electron spectra,
Astrophys. J., 90, 439448 (1939).
[155] C.-R. Vidal, Determination of electron density from line merging, J. Quant. Spec-
trosc. Radiat. Transfer, 6(4), 461477 (1966).
[156] K. M. Roussel and R. F. OConnell, A comparison between Debye-Hckel screening
and the stark eect on the determination of the last observable spectral line from a
hydrogen-like plasma, Phys. Lett. A, 51(4), 244246 (1975).
122 REFERENCES
[157] N. Konjevic, Plasma broadening and shifting of non-hydrogenic spectral lines:
present status and applications, Phys. Rep., 316(6), 339401 (1999).
[158] N. Konjevic and W. L. Wiese, Experimental Stark widths and shifts for spectral
lines of neutral and ionized atoms, J. Phys. Chem. Ref. Data, 19(6), 13071385
(1990).
[159] M. R. Zaghloul, M. A. Bourham, and J. M. Doster, A simple formulation and
solution strategy of the Saha equation for ideal and nonideal plasmas, J. Phys. D:
Appl. Phys., 33(8), 977984 (2000).
[160] L. Spitzer and R. Hrm, Transport phenomena in a completely ionized gas, Phys.
Rev., 89(5), 977981 (1953).
[161] J. W. Mackersie, I. V. Timoshkin, and S. J. MacGregor, Generation of high-power
ultrasound by spark discharges in water, IEEE Trans. Plasma Sci., 33(5), 1715
1724 (2005).
[162] C. Cohen-Tannoudji, B. Diu, and F. Lalo, Mcanique quantique, Hermann, Paris
(1973).
[163] H. G. Kuhn, Atomic Spectra, Longmans, London (1962).
Acknowledgments
This last chapter is undoubtedly the most pleasant one to write. It would have been
impossible to complete this thesis without the help of many people. Since they all under-
stand my mother tongue, I wish to warmly thank them en franais dans le texte.
Mes remerciements vont naturellement tout dabord au Dr Christoph Hollenstein, di-
recteur du groupe des plasmas industriels du CRPP et directeur de ma thse, sans qui rien
naurait t possible et grce qui jai beaucoup appris. Travailler sous sa supervision,
et ainsi pouvoir proter de sa grande exprience et de son sens aigu de la physique, a
t un rel plaisir pendant ces cinq ans. Ses qualits scientiques mais aussi humaines
ont largement contribu au succs de mon travail de thse. Merci normment, Christoph !
Je remercie trs sincrement les personnes de Charmilles Technologies qui ont col-
labor ce projet, spcialement le Dr Georg Wlder qui ma suivi pendant tout ce tra-
vail et qui ma fait le plaisir dtre membre de mon jury. Un grand merci galement
Ren Demellayer et au Dr Roberto Perez, pour avoir partag leur grande connaissance
de llectro-rosion et pour leurs prcieux conseils. Merci vous tous pour votre soutien
technique et scientique, ainsi que pour votre enthousiasme face mon travail.
Je tiens aussi remercier les Drs Ivano Beltrami et Cdric Joseph ainsi que Marco
Boccadoro de Agie, pour les discussions fructueuses que jai eues avec eux et leur intrt
pour mon projet.
Je remercie sincrement MM. les professeurs Michel Rappaz et Jrg Winter qui ont
accept de faire partie de mon jury de thse, ainsi que M. le professeur Robert Schaller
qui la prsid.
Si le directeur dun groupe de recherches est essentiel, ses autres membres ne le sont
pas moins. Un trs trs trs grand MERCI au Dr Jean-Luc Dorier, qui a toujours su
couter et trouver des rponses mes questions tant scientiques que techniques. Merci
Jean-Luc de mavoir fait proter de tes impressionnantes connaissances, et merci de ne
jamais tre court de gentillesse et de bonne humeur ! Je remercie galement les Drs
Alan Howling et Laurent Sansonnens pour leur disponibilit et leurs conseils toujours
pertinents. Je suis particulirement reconnaissant envers Alan qui a relu et corrig ce
manuscrit, et qui a ainsi rendu le texte un peu plus oxfordien.
Un grand merci aux doctorants du groupe, Samantha Pavon, Lukas Derendinger,
Hannes Schmidt et Benjamin Strahm, mais aussi il Dottore Raaello Sobbia, pour avoir
contribu lexcellente ambiance de travail. Evidemment, un merci tout particulier
123
124 ACKNOWLEDGMENTS
Alban Sublet, ambassadeur de Haute-Savoie en Suisse, qui est devenu au l de ces annes
bien plus quun simple collgue de travail. Merci Alban, collgue physicien, joggeur,
fondeur, skieur, vttiste, randonneur, cinaste, cinphile, photographe, mlomane, et
jen oublie. Tes un jeune qui nen veut, moi je dis chapeau.
Je tiens galement remercier les anciens doctorants du groupe, tout spcialement
les Drs Thierry Delachaux et Malko Gindrat. Merci les gars pour les trs nombreux
moments de rire passs ensemble, sans vous ce nest plus pareil... Merci Thierry pour
les kilomtres courus en ta compagnie, en particulier ces 26.2 miles de sourance entre
Hopkinton et Boston un jour trop chaud davril. Et merci Malko, mon voisin de bureau
pendant plusieurs annes, de mavoir transmis un peu de ta connaissance inpuisable de
linformatique.
Mes remerciements vont bien sr galement aux dirents services techniques du
CRPP. Merci aux mcaniciens, lectriciens, lectroniciens, au service du vide et au service
informatique, pour les ptits boulots occasionnels sans lesquels on navance pas au labo.
Un trs grand merci au secrtariat pour toutes les tches administratives, avec mention
spciale pour Edith Grter. Merci de tout mon coeur Edith, pour tout et surtout pour
tout le reste !
Je tiens aussi saluer Alexandre Blais, de luniversit dOttawa certes, mais Qub-
cois avant tout. Merci Alex de ton passage en Suisse, et merci mille fois de mavoir fait
dcouvrir les richesses de ta belle province. Jai eu bin dsfonne avec to, mon go !
Enn, je remercie inniment tous les membres de ma famille, les personnes les plus
importantes de ma vie, pour leur soutien tout au long de ces annes. Merci la nouvelle
gnration, Arthur, Gaspard, Anna, Elise et Camille, pour tout le bonheur que vous
me transmettez. Merci mes indispensables soeurs, Vronique, Nomie et Sarah, mais
comment diable vivrais-je sans vous ?!? Merci vous davoir choisi Franoise, Antoine et
Stephan, je naurais pu faire meilleurs choix... Et un merci incommensurable Maman et
Papa, pour mavoir soutenu de toutes les manires possibles et mavoir donn la chance
de toujours pouvoir choisir et raliser ce que jai eu envie de faire. Vous mavez en plus
donn lamour de la Science, merci de mavoir fait ce si beau cadeau...
Antoine
Curriculum Vitae
Antoine DESCOEUDRES
N le 2 fvrier 1978 Lausanne (VD),
de nationalit suisse et originaire de La Sagne (NE).
FORMATION
2002 - 2006 Thse de doctorat
eectue dans le groupe des plasmas industriels,
Centre de Recherches en Physique des Plasmas (CRPP),
Ecole Polytechnique Fdrale de Lausanne (EPFL).
Assistant pour divers cours de Physique gnrale.
1996 - 2001 Diplme dingnieur physicien, EPFL
Travail de diplme eectu au CRPP.
1993 - 1996 Baccalaurat s lettres (mention latin-mathmatiques) et
Certicat de maturit - type A (latin-grec)
Gymnase cantonal du Bugnon, Lausanne.
PUBLICATIONS
A. Descoeudres, L. Sansonnens and C. Hollenstein, Attachment-induced ionization insta-
bility in electronegative capacitive RF discharges, Plasma Sources Sci. Technol., 12(2),
152-157 (2003).
A. Descoeudres, C. Hollenstein, R. Demellayer and G. Wlder, Optical emission spec-
troscopy of electrical discharge machining plasma, J. Phys. D: Appl. Phys., 37(6),
875-882 (2004).
A. Descoeudres, C. Hollenstein, R. Demellayer and G. Wlder, Optical emission spec-
troscopy of electrical discharge machining plasma, 14
th
International Symposium for Elec-
tromachining (ISEM XIV), April 2004, Edinburgh, Scotland.
J. Mater. Process. Technol., 149, 184-190 (2004).
A. Descoeudres, C. Hollenstein, G. Wlder and R. Perez, Time-resolved imaging and
spatially-resolved spectroscopy of electrical discharge machining plasma, J. Phys. D:
Appl. Phys., 38(22), 4066-4073 (2005).

Epfl TH3542

Uploaded by

Copyright:

Available Formats

Epfl TH3542

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Epfl TH3542

Uploaded by

Copyright:

Available Formats

POUR L'OBTENTION DU GRADE DE DOCTEUR S SCIENCES

line, and with an important continuum. During the rst

line and its asymmetric shape and complex

line, and the merging of spectral lines are typical of non-

et une forte mission continue, avec la prsence de nombreuses lignes atomiques

, et avec une importante

ainsi que sa forme asymtrique et sa structure complexe, labsence

and line merging . . . . . . . . . . . . . . . . . . . . 93

in water. An example of an image

line emitted by atomic hydrogen at 656.28 nm to deter-

, this line is a classical candidate for density

on the electron density. Though the plasma temperature

is experimentally easily measurable

FWHM and of the H

shift. Thus, to calculate

line (simulation results for a 10000 K plasma with = 1, according to [112]);

line (simulation results for plasmas from 13000 to 20000 K, according to

line is treated as the copper lines for T

(at the anode),

(at the cathode).

line is the dominant contribution in EDM

emission has been mea-

line: (a) intensity evolution; (b) typical image (5 s exposure,

lter are very similar to those obtained

line is emitted by the whole plasma volume.

line emitted by atomic hydrogen, which comes

line is emitted by the whole plasma volume. This can be

line is dominant, the remaining

(434.05 nm) are not observed. This

line and C lines, both atomic Zn lines and ionic Zn

line as a reference of intensity (its intensity does not depend on erosion or

line increases drastically with decreasing on-time as shown in

line reaches almost 10 nm in gure 6.9 (b).

line during the rst 10 s of

line (Cu/steel, 16 A, water; time resolution 1 s,

spectra with equa-

line (Cu/steel, 16 A, water).

spectra, along with the electron density vertical prole calculated

spectra are shown in gure 6.22. The spectra are quite

broadening is easily ob-

line (Cu/steel, 12 A, 50 s, water;

Besides its broadening and shift, the H

line gives other useful information. Figure 7.2

line. The arrows mark

line (C/steel, 24 A, 12 s, oil, 1200 g/mm grating).

line in a non-ideal plasma [147]: (a) Stark com-

structure. The frequency = 0 corresponds to the H

emission is formed by several lines, due to transitions be-

position, and where

prole, as seen in gure 7.2. The complexity of the emission spectrum

and line merging

, etc. Continuous radiation by free-bound transitions is emitted by

and increase in the continuum radiation due to the lowering of the

line (transition from level 3 to 2) but no H

line is 10.8 , whereas the FWHM of

line is emitted by neutral

line, and with an important continuum radiation. Time-resolved spec-

line are characteristic signs of the

line for the electron