Prezentace aplikace PowerPoint

1
Active Compensation in RF-driven Plasmas by Means
of Selected Evolutionary Algorithms a
Comparative Study
Ivan Zelinkahttp//www.ft.utb.cz/people/zelinkaE
mail zelinka_at_ft.utb.czTomas Bata University in
ZlinFaculty of TechnologyInstitut of
Information TechnologiesMostni 5139Zlin 760
01Czech Republic
2
Structure of the Lecture I
3
Plasmas Status Quo I
Plasmas are conductive assemblies of charged
particles, neutrals and fields that exhibit
collective effects. Further, plasmas carry
electrical currents and generate magnetic fields.
Plasmas are the most common form of matter,
comprising more than 99 of the visible universe.

• Plasmas are radically multiscale in two senses
• most plasma systems involve electrodynamics
  coupling across micro-, meso- and macroscale and
• plasma systems occur over most of the physically
  possible ranges in space, energy and density
  scales. The figure here illustrates where many
  plasma systems occur in terms of typical density
  and temperature conditions.

4
Plasmas Status Quo II
Major topical areas of plasma science and technology Major topical areas of plasma science and technology
Plasma Equilibria, dynamic and static Wave and Beam Interactions in Plasmas
Wave and Beam Interactions in Plasmas Numerical Plasmas and Simulations
Plasma Sources Plasma Theory
Plasma-based Devices Plasma Diagnostics
Plasma Sheath Industrial Plasmas
Alan Watts of Environmental Surface Technologies
in Atlanta, Georgia has suggested the following
grid for organizing industrial plasmas with
reference to the major "revolutions" in
technology
Revolution Technologies
Industrial Engines, Metallurgy
Chemical Waste handling, Catalysts
Electrical Transformers, Switches
Nuclear Reactors, Isotopes
Electronic Electronics, Semiconductors
Optical Lighting Sources, Lasers
5
Plasmas Status Quo III
Benefits at Home. High efficiency lighting
manufacturing of semiconductors for home
computers, TVs and electronics flat-panel
displays and surface treatment of synthetic
cloth for dye adhesion.
Plasmas in Transportation. Plasma spraying of
surface coatings for temperature and wear
resistance, treatment of engine exhaust
compounds, and ion thrusters for space flight.
Plasma Thrusters for Spacecraft - test of
electrostatic ion thruster in large vacuum
chamber (NASA)
Plasma spraying of high-temperature resistance
surface coatings for a diesel engine turbocharger
housing
6
Plasmas Status Quo IV
Modification of Aerodynamic Drag. A flat panel
with a layer of one-atmosphere plasma undergoing
wind tunnel testing. This technology may lead to
improvements in aircraft flight range and landing
on short runways. (University of Tennessee)
Microwave generated plasma around a catalyst for
removal of NOx and CO from engine exhausts
Plasma Lighting. The most prevalent man-made
plasmas on our planet are the plasmas in lamps.
There are primarily two types of plasma-based
light sources, fluorescent lamps and
high-intensity arc lamps. Fluorescent lamps find
widespread use in homes, industry and commercial
settings.
Inside every fluorescent lamp there lurks a
plasma. It is the plasma that converts electrical
power to a form that causes the lamp's phosphor
coating to produce the light we see. The phosphor
is the white coating on the lamp wall. A
fluorescent lamp is shown here with part of the
phosphor coating removed to reveal the blue
plasma glow inside.
7
Plasmas Status Quo V
New one-atmosphere plasma systems make possible
new methods for surface cleaning and
sterilization for food, medical, and other
applications. Whereas standard heat sterilization
is time consuming and irradiation can damage
materials, this new plasma technology has been
shown to kill bacteria on various surfaces in
seconds to minutes. In addition to destroying
bacteria, such plasma systems also destroy
viruses, fungi and spores. These systems also
provide an environmentally benign method for
pre-treating surfaces. One-atmosphere plasma
systems are now becoming available for various
industrial applications. The photo shows
laboratory testing of non-thermal amospheric
pressure plasmas for the inactivation (or
destruction) of microorganisms.
8
Impact of Plasmas on Technology

• Products manufactured using plasmas impact our
  daily lives on
• Computer chips and integrated circuits
• Computer hard drives
• Electronics
• Machine tools
• Medical implants and prosthetics
• Audio and video tapes
• Aircraft and automobile engine parts
• Printing on plastic food containers
• Energy-efficient window coatings
• High-efficiency window coatings
• Safe drinking water
• Voice and data communications components
• Anti-scratch and anti-glare coatings eyeglasses
  and other optics

• Plasma technologies are important in industries
  with annual world markets approaching 200
  billion
• Waste processing
• Coatings and films
• Electronics
• Computer chips and integrated circuits
• Advanced materials (e.g., ceramics)
• High-efficiency lighting

9
Motivation and Aims
Radio frequency inductively-coupled plasma source
for plasma processing

• Use of evolutionary algorithms to deduce
  fourteen Fourier terms in a radio-frequency (RF)
  waveform in plasma reactor.
• Previous experiment
• Dyson, A., Bryant, P., Allen, J. E. Multiple
  harmonic compensation of Langmuir probes in RF
  discharges, Meas. Sci. Technol. 11(2000), pp
  554-559
• L Nolle, A Goodyear, A A Hopgood, P D Picton, N
  StJ Braithwaite, Automated Control of an Actively
  Compensated Langmuir Probe System Using Simulated
  Annealing
• Extension of a previous study as an comparative
  study
• SA, DE in K.V. Price, R.Storn, Lampinen J., DE
  Global Optimiser for Scientists and Engineers,
  Springer-Verlag
• SA,DE, SOMA in journal is in searching process

10
Introduction
Langmuir probes are important electrostatic
diagnostics for RF-driven gas discharge plasmas.
These RF plasmas are inherently non-linear, and
many harmonics of the fundamental are generated
in the plasma. RF components across the probe
sheath distort the measurements made by the
probes. To improve the accuracy of the
measurements, these RF components must be
removed. This has been achieved during this
research by active compensation, i.e. by applying
an RF signal to the probe tip. Not only amplitude
and phase of the applied signal have to match
that of the exciting RF, also its waveform has to
match that of the harmonics generated in the
plasma.   The active compensation system uses
seven harmonics to generate the required
waveform. Therefore, fourteen heavily interacting
parameters (seven amplitudes and seven phases)
need to be tuned before measurements can be
taken. Because of the magnitude of the resulting
search space, it is virtually impossible to test
all possible solutions within an acceptable time.
An automated control system employing EAs has
been developed for online tuning of the waveform.
This control system has been shown to find better
solutions in less time than skilled human
operators do. The results are also more
reproducible and hence more reliable.
Radio-frequency (RF) driven discharge plasmas
are widely used in the material processing
industry. Plasmas are partially ionized gases,
which are not in a thermal equilibrium with their
surroundings. They are used, for example, for
etching, deposition and surface treatment in the
semiconductor industry. In order to achieve best
results, i.e. quality, it is essential for users
of such plasmas to have tight control over the
plasma and hence they need appropriate diagnostic
tools. Better diagnostics lead to better control
of the plasma and hence to better quality of the
products.
11
Schematics of a RF driven plasma system

• Problem domain low temperature plasma systems
• Radio-frequency driven plasmas
• RF-powered plasmas by an external power source,
  usually operating on 13.56 MHz (industrial use)
• The main application of RF-powered plasmas is to
  produce a flux of energetic ions, which can be
  applied continuously to a large area of work
  piece, e.g. for etching or deposition.

13.56 MHz
Langmuir probe
12
Langmuir probes

• Developed in 1924 by Langmuir, are one of the
  oldest probes used to obtain information about
  low-pressure plasma properties. They are metallic
  electrodes, which are inserted into a plasma. By
  applying a positive or negative DC potential to
  the probe, either an ion or an electron current
  can be drawn from the plasma, returning via a
  large conducting surface such as the walls of the
  vacuum vessel or an electrode. This current is
  used to analyze the plasma properties, e.g. for
  the determination of the energy of electrons,
  electron particle density, etc.
• The region of space-charge (the sheath) that
  forms around a probe immersed in a plasma has a
  highly non-linear electrical characteristic. As a
  result, harmonic components of potential across
  this layer give rise to serious distortion of the
  probes signal. In RF-generated plasmas this is a
  major issue as the excitation process necessarily
  leads to the space potential in the plasma having
  RF components. As a consequence of this fact a
  serious distortion of the probes signal can be
  observed. It is caused by harmonic components of
  potential across this layer. In order to achieve
  accurate measures, this harmonic component has to
  be eliminated.

13
Problem Complexity and Active Compensation in
RF-driven Plasmas and Automated Control System
Where n number of points in search
space b resolution per channel in bits p number
of parameters to be optimized

• Resolution of 12 bits
• Dimensionality of the search space was 14
  (Dyson, A., Bryant, P., Allen, J. E. reported in
  Multiple harmonic compensation of Langmuir
  probes in RF discharges, Meas. Sci. Technol.
  11(2000), pp 554-559 only 3 harmonics)
• Search space consisted of n ? 3.7 x 1050 search
  points
• Mapping out the entire search space would take
  approximately 1041 years i.e. 1032 x longer that
  our universe exist
• 240s -gt 10-47s

14
Software Experiment Equipment and Requirements on
XWOS System

• Before the xwos (xwindow waveform optimization
  system) control software was developed, the
  following requirements were identified
• The optimization should take place within
  reasonable time,
• The search results (fitness) over time should be
  plotted on-line on screen in order to allow a
  judgement of the quality of the result,
• The operator should be able to select values for
  the EAs parameters,
• The operator should have the opportunity to set
  any of the fourteen parameters manually,
• The operator should have the opportunity to
  fine-tune the settings found by the automated
  system,
• The DC bias (fitness parameter) had to be
  monitored.
• The control software was developed in C on a
  500 MHz Pentium III PC running the Linux 2.2
  operating system. The graphical user interface
  was coded using X-Windows and OSF/Motif.

Correlation analysis window
History of one evolution of the best and average
individual
DC Bias
7 amplitudes
7 phases
15
Hardware Experiment Equipment

• All experiments were carried out at the Oxford
  Research Unit, The Open University, UK. Figure
  shows the experiment setup. Apart from the
  control system described above, a digital
  oscilloscope was used to measure the actual
  waveforms found by the three optimization
  algorithms.
• The control software was running on a PC under
  the Linux operating system. The algorithms used
  for this experiments were written in C and
  integrated in the existing Langmuir probe control
  software. The plasma system used was a standard
  GEC cell.

16
Optimization Algorithms Used

• Simulated Annealing (SA)
• Van Ginneken, L. P. P. P., Otten, R. H. J. M.
  The Annealing Algorithm (Kluwer International
  Series in Engineering and Computer Science,72),
  Kluwer Academic Publishers, 1989
• Differential Evolution (DE)
• Price K. An Introduction to Differential
  Evolution, in New Ideas in Optimization, D.
  Corne, M. Dorigo and F. Glover, Eds., s. 79108,
  McGraw-Hill, London, UK, 1999.
• Self-Organizing Migrating Algorithm (SOMA)
• Zelinka Ivan , SOMA Self Organizing Migrating
  Algorithm,chapter 7, 33 p. in B.V. Babu, G.
  Onwubolu (eds), New Optimization Techniques in
  Engineering, Springer-Verlag

17
SOMA Main Idea

• The main idea on which SOMA is based is
  competetive-cooperative behavior of the
  intelligent beings who are together solving given
  task. Examples can be observed arround the world
• Ants
• Bees
• Termites
• Wolves
• People
• Gold miners of 19th century
• Battle strategies
• Bacause of used philosophy, terminology used with
  this algorithm a little bitt differ from standard
  terminology used with classics EAs.

At http//www.ft.utb.cz/people/zelinka/soma/ are
available source codes, test functions, and
more...
18
SOMA Terminology and Recommended Parameters
19
SOMA Principles

• Parameter definition - Migrations, MinDiv,
  PopSize, PathLength, Step, PRT, Specimen and
  Dimension of the problem.
• Start of SOMA - population generating
• Run of SOMAprecisely

(1)
(2)
(3)
(4)
(5)
20
SOMA Principles
21
SOMA Basic Versions

• Versions
• AllToOne
• AllToOneRandomly
• AllToAll
• AllToAllAdaptive

22
SOMA Ability to Avoid Local Minimas
SOMAs ability to avoid local minimas - during
migration loops is created false function -
polyhedron and individuals move along to edges of
this polyhedron
23
SOMA Constraints Handling

• Handling of boundary constraints
• Boundary position setting
• Reset of wrong parameter
• Spiral movement on N1 dimensional sphere
• Random replacement
• Handling of integer variables
• Rounding in the population
• Rounding in the cost function argument input
• Handling of discrete variables
• Integer index use
• Handling of constraints given to the fitness
• Penalty

24
SOMA Problem Complexity

• Objective function -
• unimodal multimodal
• Linear nonlinear
• None-fractal type (but because everything in the
  real world has constrains, fractal type functions
  can also be optimized)
• Defined at real, integer or discrete argument
  spaces
• Constrained, multiobjective
• Needle-in-haystack problems
• NP problems
• Degree of parameter interactions low high,
  separable non-separable
• Type of variables continuous discrete /
  integer / mixed
• Number of variables low high
• Search space small large, finite infinite,
  continuous non-continuous

25
SOMA Selected Tests I
26
SOMA Selected Tests II
EggHolder
StretchedSine
StretchedSine
27
SOMA Tests Functions
Sphere model, 1st De Jong's function
Rosenbrock's saddle
3rd De Jong's function
4th De Jong's function
Stretched V sine wave function (Ackley)
Rastrigin's function
Schwefel's function
Griewangk's function
Test function (Ackley)
Test function - egg holder
Ackley's function
Rana's function
Cosine wave function (Masters)
Pathological function
Michalewicz's function
28
SOMA Selected Problems
Chemical reactor optimization and control
Chemical reactor structural stability
Mechanical engineering examples
Analytic programming
Predictive model estimation
Fuzzy controller setting
Antena
Inverse Fractal Problem
29
Previous Experiments

• SA had shown better floating potential than
  human operator
• SA had shown smaller diversity in floating
  potential and time

For following experiments were parameters set so
that used EA showed the best performance as much
as possible
30
Experiment Setting SA, DE
Plasma parameters used for the experiments
Parameter settings for the optimization
algorithms used in experiments
31
Results I SA, DE
DE
All data were carefully collected and used to
draw a flow of all histories so that average,
minimal and maximal values can be easily observed.
SA
32
Results II SA, DE
DE
Efficiency of used algorithms can be also judge
according to correctness and reproducibility of
reached results based on statistical point of
view
SA
33
Results III SA, DE
DE
Results were used to restore waveforms observed
on osciloscope. Here are depicted average values,
minimal and maximal values reached during all
experiments.
SA
34
Results VI SA, DE
Results were used to create an algorithm
efficiency chart to show efficiency of both
algorithms. They shows minimal, maximal and
average values reached during the active
compensation of RF-driven plasmas.
35
Experiment Setting SA, DE and SOMA
Plasma parameters used for the experiments
Parameter settings for the optimization
algorithms used in experiments
36
Results I SA, DE and SOMA
DE
All data were carefully collected and used to
draw a flow of all histories so that average,
minimal and maximal values can be easily observed.
SOMA
SA
37
Results II SA, DE and SOMA
DE
All data were carefully collected and used to
draw a flow of all histories so that average,
minimal and maximal values can be easily observed.
SOMA
SA
38
Results III SA, DE and SOMA
DE
Results were used to restore waveforms observed
on osciloscope. Here are depicted average,
minimal and maximal values reached during all
experiments.
SOMA
SA
39
Results III a) SA, DE and SOMA
DE
Here are all waveforms in one just for
demonstration. Average, minimal and maximal
values reached during all experiments cannot be
observed here.
SOMA
SA
40
Results VI SA, DE and SOMA
Results were used to create four charts four
different view on algorithm efficiency
SA, SOMA, DE
41
Conclusion
Ability to be used all three algorithms can be
used for active compensation in RF-driven
plasmas. However, based on results it is clear
that SOMA and DE has greater potential for this
task.   Preciseness and reproducibility one of
the crucial points in science is reproducibility,
i.e. the ability to achieve the same results for
two identical experiments. In practical
applications like this one, a high degree of
reproducibility is needed. From figures it is
visible, that SOMA and DE has a greater
reproducibility than SA. They are is also more
precise than SA.   Speed the speed of the
optimization process was not determined by the
computer power available, but by the time
constants of the analogue equipment, e.g.
harmonic box. Therefore, all three algorithms
have shown similar speed performance in this
specific application.   Diversity is tightly
connected with preciseness and reproducibility.
From this point of view SOMA and DE performed
almost three times better than SA. If one
remembers that plasmas are highly nonlinear
dynamical systems with complicated behavior, then
the results produced by SOMA and DE are very
sufficient. Algorithms efficiency from figures
it is clearly visible that the best results were
obtained by SOMA algorithm, second place took DE
and third SA. While results given by SA are
significantly the worst one, in the case of SOMA
and DE should be mentioned that difference
between them was wery small almost negligible.
This small difference shows, that both algorithms
are highly usable for dealing with systems kind
of blackbox which plasma reactor in fact is.
Dynamical position of global extreme global
extreme (thus whole cost function landscape) was
not static in time. During above described
experiments which took almost 12 hours of
noninterrupted works (for 5 days ?), plasma in
reactor changed its behaviour. This change was
linear dependent. Based on experiences with SOMA
and DE, it can be stated that both algorithms has
follows global extreme (or founded suboptimal
solution) position quite well.
42
Acknowledgements

• This work was partly funded by the
• Ministry of Education of the Czech Republic,
  under grant reference MSM 26500014,
• Grant Agency of the Czech Republic under grand
  references GACR 102/03/0070 and GACR 102/02/0204.
  
• The authors whish to express their thanks to
• Lars Nolle School of Computing and Technology,
  The Nottingham Trent University, Burton
  Street, Nottingham, NG1 4BU, UK
• A.A. Hopgood
• N.St.J. Braithwaite Oxford Research Unit, The
  Open University
• Alec Goodyear
• Jafar Al-Kuzee
• for assistance with the plasma equipment.