Citation
Plasma-particle interactions for the quantitative analysis of individual aerosol particles using laser-induced breakdown spectroscopy

Material Information

Title:
Plasma-particle interactions for the quantitative analysis of individual aerosol particles using laser-induced breakdown spectroscopy
Creator:
Carranza, Jorge E., 1964-
Place of Publication:
Gainesville FL
Publisher:
University of Florida
Publication Date:
Language:
English
Physical Description:
xiii, 152 leaves : ill. ; 29 cm.

Subjects

Subjects / Keywords:
Aerosols ( jstor )
Average linear density ( jstor )
Diameters ( jstor )
Emission spectra ( jstor )
Lasers ( jstor )
Particle density ( jstor )
Particle emission ( jstor )
Particle mass ( jstor )
Plasma volume ( jstor )
Plasmas ( jstor )
Dissertations, Academic -- Mechanical Engineering -- UF ( lcsh )
Mechanical Engineering thesis, Ph. D ( lcsh )
City of Gainesville ( local )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 2002.
Bibliography:
Includes bibliographical references (leaves 147-151).
General Note:
Printout.
General Note:
Vita.
Statement of Responsibility:
by Jorge E. Carranza.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Jorge E. Carranza. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
029186840 ( ALEPH )
50860659 ( OCLC )

Downloads

This item has the following downloads:


Full Text










PLASMA-PARTICLE INTERACTIONS FOR THE QUANTITATIVE ANALYSIS OF INDIVIDUAL AEROSOL PARTICLES USING LASER-INDUCED BREAKDOWN SPECTROSCOPY











By

JORGE E CARRANZA


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


2002














ACKNOWLEDGMENTS

I would like to thank Dr. David Hahn for being the best teacher that I had in a

very long time, for his constant and invaluable advice, guidance, and patience during the development of my research, and for instilling in me a better understanding of science. What is more, I would like to express my sincere gratitude to him for his confidence shown in me three years ago when we met and I was struggling with the English language. I would also like to thank the members of my graduate committee for their help throughout this process.

My thanks also go to my lab mates for their helpful talks and company during experiments, and for their support during class periods. Thanks especially to Brian Fisher and Greg Yoder for their great deal of help during the experiments of ambient air monitoring in the summer of 2000. In addition, I would like to express my appreciation to my friends in Gainesville for sharing enjoyable times, and very special thanks to my friend Diana Serrano for our constant lunch talks, and for her constant help and support.

Finally, I want to acknowledge to the Department of Mechanical Engineering at the University of Florida for their financial support and for giving me the chance to be a teaching assistant.















TABLE OF CONTENTS


ACKNOW LEDGM ENTS .................................................................................................. ii

LIST OF TABLES ........................................................................................................ vi

LIST OF FIGURES .......................................................................................................... vii

ABSTRACT ...................................................................................................................... xii

CHAPTERS

1 INTRODUCTION ........................................................................................................... 1

1.1 Basic Principle of LIBS ........................................................................................ 3
1.2 Laser-Induced Breakdown Initiation and Plasma Formation .............................. 4
1.3 Fundam ental Studies ............................................................................................. 6
1.4 LIBS as an Analytical Technique ........................................................................ 9
1.5 Conditional Data Analysis for LIBS ................................................................. 14
1.6 Single-Shot LIBS-Based Aerosol Analysis ...................................................... 16
1.7 Vaporization of a Single Aerosol Particle Using LIBS ..................................... 18
1.8 M otivation and Objectives for Present Study .................................................... 20

2 EXPERIMENTAL FACILITIES AND FUNDAMENTAL PLASMA
CHARACTERISTICS ............................................................................................. 22

2.1 Aerosol Generation System ............................................................................... 23
2.2 Nebulizer Characterization ............................................................................... 26
2.2.1 M ass Flow-Rate Calibration .................................................................... 27
2.2.2 Lifetim e of Generated Droplets ............................................................... 28
2.2.3 Aerosol Particle Characterization ........................................................... 30
2.2.4 Nebulizer Droplet Characterization ......................................................... 34
2.3 LIBS Instrum entation ......................................................................................... 37
2.4 LIBS Spectral Analysis ...................................................................................... 39
2.5 LIBS Calibration ............................................................................................... 43
2.6 Laser Induced Breakdown Threshold ............................................................... 48
2.7 Global Energy Balance of the Optical Breakdown ........................................... 51
2.8 Plasm a Temperature ........................................................................................... 52
2.8.1 Spectral W indow Calibration ................................................................. 57
2.8.2 Plasm a Temperature Decay .................................................................... 26
2.9 Sum m ary ................................................................................................................ 59









3 FEASIBILITY OF AEROSOL DETECTION IN AMBIENT AIR ........................ 61

3.1 Ambient Aerosol Sampling System ................................................................. 62
3.2 LIBS Data Analysis .......................................................................................... 64
3.2.1 Collection of Data ................................................................................... 64
3.2.2 Processing of Data .................................................................................... 65
3.3 Results and Discussion ..................................................................................... 68
3.3.1 Mass Concentration ................................................................................. 70
3.3.2 Particle Analysis ...................................................................................... 77
3.3.3 Particle Size Distribution ........................................................................ 79
3.4 Sum m ary ................................................................................................................ 83


4 SAMPLING STATISTICS FOR SINGLE SHOT ANALYSIS ............................... 85

4.1 Experimental and Data Processing Considerations .......................................... 85
4.2 Energetic State of the Laser-Induced Plasma ................................................... 87
4.3 Time Delay Optimization with Pulse Energy .................................................... 89
4.4 Single-Shot Plasma Emission Fluctuations ........................................................ 91
4.5 Single-Shot Plasma Temperature ...................................................................... 94

5 ASSESSMENT OF THE UPPER PARTICLE SIZE LIMIT OF VAPORIZATION
USING THE LIBS TECHNIQUE .............................................................................. 100

5.1 Experimental and Data Processing Methodology ................................................ 100
5.2 Vaporization of a Collection of Nanoparticles .................................................... 102
5.3 Vaporization of a Single Silica Particle ............................................................... 104
5.4 Laser-Particle or Plasma-Particle Vaporization ................................................... 108
5.5 Particle Vaporization Driven by Laser-Induced Plasma ...................................... 112
5.6 Precision and Accuracy of Particle Sizing Using the LIBS Technique ............... 116

6 CHARACTERISTIC PLASMA VOLUMES FOR ANALYSIS OF AEROSOL
PARTICLES USING LIBS ......................................................................................... 119

6.1 Experimental Methodology ................................................................................. 121
6.2 Statistical Sample Volume ................................................................................... 124
6.3 Mass-Based Sample Volume ............................................................................... 128
6.4 Physical Plasma Volume ...................................................................................... 130
6.5 Discussion of the Characteristic Plasma Volumes ............................................... 133

7 CONCLUSIONS .......................................................................................................... 137

8 RECOMMENDATIONS FOR FUTURE WORK ...................................................... 140

APPENDICES








A ZERO ORDER LOGNORMAL DISTRIBUTION .................................................... 142

B QUARTZ TUNGSTEN HALOGEN LAMP .............................................................. 144

C SAVITZKY-GOLAY ALGORITHM ......................................................................... 145

R EFER EN C ES ................................................................................................................ 147

BIOGRAPHICAL SKETCH ........................................................................................... 152














LIST OF TABLES


Table page

2.1. Aerosol Generator Specifications ........................................................................ 25

2.2. Standard Aqueous Solutions ................................................................................. 26

2.3. Microsphere Silica Particle Suspensions ............................................................ 26

2.4. Summary of Generated Aerosol Particle Characteristics ...................................... 33

2.5. Laser and Optics Specifications .......................................................................... 39

2.6. Fe II Atomic Emission Lines for Plasma Temperature Calculations .................. 53

3.1. Parameters for Aerosol Particle Detection ........................................................... 65

3.2. Equations for Calibration of Mass Concentration .............................................. 66

3.3. Summary of Analyte Mass Concentration .......................................................... 76

3.4. Summary of Particle Size ...................................................................................... 82

6.1. Actual Silica Particle Sample Rates ......................................................................... 128

6.2. Summary of the Characteristic Plasma Volumes for 315 mJ per Laser Pulse ......... 133















LIST OF FIGURES


Figure pag.e

2.1 Schematic of the aerosol generation system ........................................................ 24

2.2 Nebulizer calibration curve of deionized water. Each data point is the average of
a minimum of 3 runs (error bar = � one standard deviation). 27

2.3 Light scattering responses from the testing chamber due to the excitation of a
laser pulse of 30 mJ under a co-flow of 60 1pm of purified air ............................. 30

2.4 Aerosol particles generated by a nebulization rate of 0.09 ml/min into a co-flow
of 42 1pm of air: a) For 2500 ag/ml of Iron, and b) For 10000 ag/ml of titanium
(TEM micrograph of 50K magnification) ............................................................ 31

2.5 Iron-based particle size distribution at airborne mass concentration of 4425 jg/m3
corresponding to an aqueous solution of 2500 jg/ml .......................................... 32

2.6 Titanium-based particle size distribution at airborne mass concentration of 17700
jig/m3 corresponding to an aqueous solution of 10000 ag/ml ............................. 32

2.7 Aspect ratio for the titanium-based crystal particles at airborne mass
concentration of 17700 jg/m3. 33

2.8 Nebulizer droplet size distribution based on the combination of iron and titanium
solution at 10000-jig/ml aqueous concentrations ................................................. 35

2.9 Zero-order lognormal distribution with modal diameter 340 nm at three
dimensionless widths a0 that describes the droplet size distribution created by a
m edical nebulizer ................................................................................................. 37

2.10 Experimental setup of a LIBS apparatus ............................................................ 38

2.11 LIBS spectrum for an iron mass concentration of 4425 jg/m3 in purified air.
Ensemble average of 1200 laser pulses (8 jIs delay time with 5 js integration
tim e) ......................................................................................................................... 40

2.12 Effect of time delay in atomic emission and continuum emission using an
integration time of 2 js at each designated delay time. Atomic emission line C I
247.8 nm .................................................................................................................. 43








2.13 Peak-to-base ratio and signal-to-noise ratio of the atomic emission line at ko,
where AX0 is the width of the peak, rms is the root-mean square of the noise at
the off-peak baseline, Bi is the average continuum intensity .............................. 45

2.14 Calibration curve for iron 256.26-nm emission peak signal using 8-ps delay and
5-gs time width. Error bars represent � one standard deviation (R2 = 0.996) ........ 46

2.15 Time delay optimization at fixed 2-ps time width for the maximization of the
LIBS signal P/B using the carbon emission line C I at 247.8 nm and at pulse
energy of 247 m J ................................................................................................. 47

2.16 Calibration curve for magnesium 280.27-nm emission peak signal using a 40-is
delay time and 40-gs time width. Error bars represent � one standard deviation
(R 2 = 0.998) ........................................................................................................ 48

2.17 Percentage of laser pulses producing breakdown as a function of laser pulse
energy for various sample stream s ...................................................................... 49

2.18 Transmitted energy thru the laser-induced plasma in air as a function of laser
pulse energy. Error bars represent � one standard deviation ............................. 51

2.19 Correlation wavelength versus pixel number of the iCCD array for the 270-nm
spectral w indow ....................................................................................................... 55

2.20 Reference lamp irradiance and collected irradiance by the LIBS system in the
spectral w indow 270 nm ...................................................................................... 56

2.21 Correction factor to provide relative intensities line-by-line in the spectral
w indow 270 nm .................................................................................................... 56

2.22 Representative 1200-shot average spectrum in air obtained using a laser pulse of
300 mJ at 5-ms delay time and 5-ms integration time after intensity correction ..... 57

2.23 Boltzmann plot using Fe II emission lines for energy pulse of 300 mJ at a delay
time of 10 jis an d width time of 5 ps ................................................................. 58

2.24 Plasma temperature decay in air and nitrogen atmosphere using pulse energy of
300 mJ an integration time of 5 gs. Error bar = + one standard deviation ...... 59

3.1 Schematic of the ambient air sampling system ...................................................... 62

3.2 Transport efficiency of ambient air particles from the inlet to the LIIBS sample
chamber as a function of the particle diameter .................................................... 64

3.3 Aluminum and magnesium calibration curves using the emission lines A 11394.4
nm and M g 1279.55 nm ........................................................................................ 66








3.4 Two single-shot LIBS spectra. The lower spectrum represents a single MgO
particle. The upper spectrum that has been vertically shifted for clarity
represents a single Mg-Si-containing particle ..................................................... 69

3.5 Two single-shot LIBS spectra. The lower spectrum represents a single A1O
particle. The upper spectrum that has been vertically shifted for clarity
represents a Ca-Al-containing particle ................................................................. 70

3.6 The lower Na-LIBS spectrum is the 9600-laser-shot average, and the upper NaLIBS spectrum is the 30-identified, Na-shot average from the 9600 laser shots.
Both spectra have the same intensity scale but have been vertically shifted for
clarity ......................................................................................................................... 7 1

3.7 Enhancing of the Mg-LIBS signal using the conditional analysis. The 14-Mg hits
are a subset of the 9600 laser shots. Both spectra have the same intensity scale
but have been vertically shifted for clarity .......................................................... 72

3.8 Mass concentration of magnesium as a function of time. Each data point
represents the average LIBS-based concentration over a two-hour sampling
p eriod ......................................................................................................................... 72

3.9 Mass concentrations of calcium and sodium as a function of time. Each data
point represents the average LIBS-based concentration over a two-hour sampling
period ......................................................................................................................... 75

3.10 Sampling frequency over a period of two hours. Each data point corresponds to
the frequency of hits for 1200-shot laser sequence ............................................. 77

3.11 Histogram of calculated diameters for magnesium-containing particles for the
Fourth of July holiday period. The particles were modeled as magnesium oxide
(M gO ) ...................................................................................................................... 79

3.12 Histogram of calculated diameters for calcium-containing particles for the
Fourth of July holiday period. The particles were modeled as calcium carbonate
(C aC O 3) .................................................................................................................... 80

3.13 Histogram of calculated diameters for calcium-containing particles for the nonholiday period. The particles were modeled as calcium carbonate (CaCO3) ......... 81

3.14 Histogram of calculated diameters for sodium-containing particles for the
Fourth of July holiday period. The particles were modeled as sodium chloride
(N aC I) ...................................................................................................................... 82

4.1 Percentage of incident pulse energy absorbed by the laser-induced plasma as a
function of laser pulse energy. Error bars represent � one standard deviation ........ 88

4.2 Time delay optimization at fixed 2-gs time width for the maximization of the P/B
using the Carbon emission line C I at 247.8 nm ................................................... 89








4.3 Optimal delay times for the spectral carbon emission line (P/B) at 247.8 nm as a
function of laser pulse energy. The solid line is a second-order curve fit ........... 90

4.4 Carbon peak emission (P), continuum emission (B), carbon peak-to-base (P/B),
and carbon signal-to-noise ratio (SNR) as a function of laser pulse energy. Each
data point represents the calculated value of a single 100-shot average spectrum ... 91

4.5 Laser shot-to-shot variability of the peak emission (open circles), continuum
emission (solid squares), and P/B (solid diamonds) for (a) 200 mJ laser pulse
energy, and (b) 344 mJ laser pulse energy. Data correspond to the optimal time
delay at different pulse energies .......................................................................... 93

4.6 Precision of the LIBS signal, expressed as P/B ratio and relative standard
deviations (RSD) of P/B and SNR as a function of laser pulse energy. Each data point represents the average of 100 single-shot calculations. Error bars represent
� one standard deviation ...................................................................................... 94

4.7 Effect of the smoothing algorithm on the Fe II atomic emission lines selected
(indicated by arrows) corresponding to a single-shot spectrum. The original
spectrum was shifted for clarity .......................................................................... 95

4.8 Plasma temperature and RSD of the plasma temperature as a function of laser
pulse energy. Each data point is the average single-shot temperature
calculations, and is based on iron II emission. Error bars represent � one
standard deviation .................................................................................................. 97

5.1 Peak-to-base (P/B) ratio of the 288.16-nm silicon emission line as a function of
silicon mass concentration for a well-disperse aerosol stream of silicon-based
nanoparticles generated by nebulization of aqueous silicon standards. (R2=0.999) 102

5.2 Single-shot spectra corresponding to a single 2.1-jim-diameter silica microsphere
as collected and following application of the Savitzky-Golay smoothing
algorithm. The smoothed spectrum has been shifted vertically for clarity ............. 104

5.3 Ensemble-averaged spectra corresponding to individually detected monodisperse
silica microspheres with diameters of 1.0, 1.5, 2.1, and 2.5 jtm .............................. 106

5.4 P/B at Si 1288.16 rm for ensemble-averaged spectra of individually detected
monodisperse silica microspheres as a function of the cube of the silica particle
diameter. The continuous line is a linear fit of the first three data points ............... 107

5.5 Gaussian temporal profile corresponding to a pulse energy of 160 mJ given in a
total time of 2.8 times the full-width half maximum (V,) ......................................... 110

5.6 Time and energy for the laser- and plasma-particle interaction in pulse energy of
320 m J ...................................................................................................................... 111








5.7 Silica particle diameter distributions for a nominal diameter of 1.02 pm. Sample
of 47 particles ........................................................................................................... 117

5.8 Silica particle diameter distributions for a nominal diameter of 1.50 pm. Sample
of 126 particles ......................................................................................................... 117

5.9 Silica particle diameter distributions for a nominal diameter of 2.08 Pm. Sample
of 574 particles ......................................................................................................... 118

6.1 The natural logarithm of the transmission as a function of the product of the
extinction cross-section and dilution factor for the 2.1 -pm-silica particle
suspension. The error bars represent � 1 standard deviation .................................. 125

6.2 Independence of the threshold value in the conditional analysis for determination
of the hit rate. Silica particles of 2.1 pLm in diameter and at a 45-cm"3 number
density were used. DIW equals purified deionized water only ............................... 126

6.3 The natural logarithm of one minus the experimental silica particle-sampling rate
as a function of the silica particle number density. The error bars represent � 1
standard deviation ..................................................................................................... 127

6.4 Silicon calibration curve for a low mass concentration range used in the
quantification of the mass-average sample volume. The error bars represent � 1
standard deviation ..................................................................................................... 129

6.5 The LIBS-based equivalent mass concentration of the 1.0, 1.5, and 2.1 pm silica
particles as a function of the silicon mass contained in the silica particles. The
error bars represent � 1 standard deviation .............................................................. 130

6.6 Temporal scale for the transmission measurements thru the plasma created by the
Nd:YAG 1064-nm laser, and probed by the Nd:YAG 532-nm laser ....................... 131

6.7 Cross-section of the physical plasma measured at the end of the 1064-nm plasmagenerating laser pulse. The boundary represents an optical thickness of 0.1, and
the error bars represent � 1 standard deviation. The dashed line is an elliptic
profile fit to the experimental plasma volume and aspect ratio ............................... 132













Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

PLASMA-PARTICLE INTERACTIONS FOR THE QUANTITATIVE ANALYSIS OF INDIVIDUAL AEROSOL PARTICLES USING LASER-INDUCED BREAKDOWN SPECTROSCOPY

By

Jorge E. Carranza

August 2002

Chair: David W. Hahn

Department: Mechanical Engineering

In response to the need for discrete characterization of ambient air fine particles due to the direct relation of particle size and particle composition to human health effects, laser-induced breakdown spectroscopy (LIBS) has been studied in this dissertation to support its development as a real-time aerosol analysis technique capable of measuring particle size and particle composition.

In Chapter 1, a literature review and the principles of LIBS are presented in the context of aerosol analysis. The complete experimental facilities are described in Chapter 2. An aerosol generation system was developed to produce well-dispersed aerosolized nanoparticles that served as the calibration source for the LIBS system. The LIBS system consists of a 1064-nm Nd:YAG pulse laser, supporting optics, and an intensified charge-coupled device for plasma emission quantification. The developed LIBS system was successfully deployed as discussed in Chapter 3 for ambient air monitoring (specifically aluminum, magnesium, calcium, and sodium). Mass








concentrations were recorded on the order of low parts per trillion and minimum particle sizes about 200 nm. In Chapter 4, issues of plasma homogeneity and signal fluctuations on a shot-to-shot basis were addressed to elucidate optimal laser pulse energy for single shot analysis, thereby identifying a characteristic state of the plasma where signal fluctuations are minimized. The implicit assumption of complete particle vaporization is investigated in detail in Chapter 5, with the determination that silica particles up to 2.1gm diameters are completely vaporized due to plasma-particle interactions. Finally, the characteristic plasma volumes related to the scheme of particle sizing are determined and analyzed in the context of the analysis of single aerosol particle detection.

The outcome of this research yielded an enhanced understanding of single aerosol analysis with the LIBS technique. Important conclusions are that LIBS measurement should be made at the plasma saturation condition, particle sizing should be limited to about 2 gxm or less, particle vaporization is driven by plasma-particle interactions, and that real-time ambient air monitoring is feasible with LIBS on time periods as short as 4 minutes.













CHAPTER 1
INTRODUCTION

Aerosols are a mixture of either solid or liquid particles suspended in a gas phase medium. Aerosols of considerable interest in the atmosphere are those with the particle size ranging from 1 nm to 100 jWn in diameter, and with particle number concentration ranging from a few particles per cm3 in clean environments to 108 particles per cm3 in highly polluted regions (Noble et al. 2000). Particles in the low size range (i.e., submicrometer to several micrometers in diameter) are of the most concern in atmospheric sciences, process and control monitoring, and especially regarding human health. The potential impact of the small particles on human health led the United States Environmental Protection Agency (US EPA 1996) to conclude that particulate matter was associated with increased morbidity and mortality. In fact, a recent study (Peters et al. 2001) reported that the smallest particles could penetrate deep into the respiratory system causing adverse health effects (such as the increasing incidence of heart attacks) in as little as 2 hours.

In response to these health risk factors, research agendas have targeted a number of program areas, including measures of outdoor particulate matter (PM) and associated human exposures, characterization of emissions sources, air quality model development and testing, and assessment of hazardous PM components (U.S. National Research Council 1999). Characterization of the behavior and properties of airborne particles is difficult because environmental aerosols comprise a diverse and constantly fluctuating system. Primary particles (emitted into the atmosphere) and secondary particles (formed








in the atmosphere) may undergo growth, evaporation, or chemical reactions, and their complex interactions are reflected mainly in the particle size and composition distribution (Hinds 1999). Many questions concerning the nature of these processes remain unanswered. The need for ambient air PM data, specifically size and composition measurements of individual particles, is common to all of these research areas; however, a bulk analysis of the particles is of limited use. Therefore, an individual particle analysis, specifically size and composition measurements, must be carried out to facilitate the understanding of not only these dynamic processes (i.e., transport and formation of toxic particles) but also their implication in the quality of life of the human beings (i.e., global warming, ozone depletion, and respiratory disorders). This need for PM data combined with the development of advanced aerosol analysis techniques have served as a motivation for the development of on-line or real-time aerosol particle analyzers.

Several techniques have been used to analyze individual airborne particles in real time. Most of these techniques are based on mass spectrometry methods such as rapid single-particle mass spectrometry (Johnston and Wexler 1995, and Ge et al. 1998), ultrasensitive particle analysis system (Reents et al. 1995 and 2000), and aerosol time-offlight mass spectrometry (Prather et al. 1994). However, recently the feasibility of the first atomic spectroscopy approach for the analysis of single particles based on a technique called laser-induced breakdown spectroscopy (LIBS) has been shown (Hahn 1998, and Hahn and Lunden 2000). The LIBS technique is the basis of this research, and it will be discussed in detail below. All the above techniques are able to determine particle size and composition. However, the mass spectrometry techniques are limited to the use of auxiliary techniques for sizing, typically light scattering or time-of-flight. In








contrast, the LIBS technique makes use of a direct mass measurement to resolve the determination of particle size with simultaneous multi-elements mass composition. In addition, LIBS is suitable to be miniaturized while functioning as a real-time, in situ or remote sensing monitor (Davies et al. 1995 and 1996). One more advantage of LIBS over mass spectrometry methods is its speed and experimental simplicity, as the LIBS instrumentation does not require the high vacuum instrumentation needed for mass spectrometry. All these characteristics of LIBS make it a promising technique for the analysis of single aerosol particles.

1.1 Basic Principle of LIBS

Laser-induced breakdown spectroscopy (LIBS), also known as laser-induced

plasma spectroscopy (LIPS), is an atomic emission spectroscopy technique that dates to the 1960s. When a pulsed laser beam is focused onto a small spot, the temperature of the locally heated region increases rapidly to vaporize the material, and to dissociate the molecules, thereby forming an optically induced plasma. The plasma will be formed if the laser power density exceeds the breakdown threshold value for the medium, as described in detail below. The laser-induced plasma is used as both the sample volume and the excitation source, dissociating all molecules and fine particulates within the highly energetic micro-plasma. The resulting plasma emission can be resolved both spectrally and temporally to yield spectra containing the atomic emission lines corresponding to the atoms present in the plasma volume, including atoms originating from aerosols initially enveloped by the plasma. The aggregate energy levels of each element of the periodic table are unique in the emission spectrum. Therefore, the spectral location of the atomic emission lines in the spectrum can be used to identify elements,








and their emission intensities (height, width, or integrated peak) can be used to provide a measure of the amount of mass of the constituent elements in the sample.

1.2 Laser-Induced Breakdown Initiation and Plasma Formation

The laser-induced plasma can be defined as a near-totally-ionized gas induced by laser irradiance sufficient to generate an electric field at the focal spot that exceeds the dielectric strength of the medium (Weyl 1989). This event is qualitatively marked by a glow in the focal region and followed by a distinct shock wave, while quantitatively distinguished by the absorption of the laser beam due to ionization. Two mechanisms involving the generation of ions, avalanche ionization and multi-photon ionization, are summarized here after Weyl (1989). In the avalanche ionization mechanism, the laser radiation is absorbed by free electrons that later collide with neutral atoms or molecules to release new electrons. The releasing of a new electron occurs for the free electron that gained sufficient energy and it is described by following relation

e- +M -) 2e- +M+. (1.1)

This process leads to a cascade phenomenon that increases the electron

concentration exponentially with time. In the second mechanism, the multi-photon ionization mechanism does not require the presence of free electrons as the avalanche ionization does. The multi-photon ionization involves the direct absorption of photons by atoms or molecules to produce ionization. This process is described by

M + n(hv)-+ e- + M. (1.2)

Both mechanisms of ionization require high laser irradiances, usually over 108 W/cm2, and occur simultaneously. In general, the breakdown initiation is started by multi-photon ionization and followed by cascade ionization, however, the literature






5

indicates that avalanche ionization is characteristic for larger irradiation wavelengths (i.e., Nd:YAG laser 1064 nm), and photon-ionization is associated with shorter irradiation wavelengths (i.e., excimer laser 193 nm) (Biswas et al. 1988, Ho et al. 1997, and Martin et al. 1999).

The evolution of the plasma after the breakdown initiation stage depends on many factors such as irradiance, ambient gas composition, and laser wavelength; a short description of the three major types of laser absorption wave is condensed here after Lee et al. (2000). At low plasma irradiation, laser-supported combustion waves are produced mainly via radiative transfer from the hot plasma to the cool high-pressure gas created in the shock wave. The plasma radiation is primarily in the extreme ultraviolet and it is generated by photo-recombination of electrons and ions into the ground-state atom. At intermediate irradiance, the laser-supported detonation wave model governs the expansion of the plasma. During the propagation of the shock wave, the shocked gas is heated enough to begin absorbing the laser radiation without requiring additional heating from the plasma; as a consequence, the laser absorption zone follows directly behind the shock wave and moves at the same velocity. At high irradiance, the laser-supported radiation model rules the absorption of the laser radiation. The plasma irradiation is so intense that, prior to the arrival of the shock wave, the ambient gas is heated to temperatures where the laser absorption begins.

In the first nanoseconds, after the deposition of the pulse energy of the laser, the plasma expands rapidly emitting electromagnetic radiation primarily in the ultravioletvisible region. Several hundreds of nanoseconds later, the plasma slows down and the radiation in the ultraviolet band decreases faster than in the visible band. The plasma








starts to decay by radiative transfer, quenching, and electron-ion recombination processes, and as a consequence, isolated emission lines and structured bands appear in the spectrum. These processes usually occur within tens of microseconds after the breakdown initiation. The characteristic lifetime of the plasma depends on the transient nature of electron densities inside the plasma. The sequential spectral emission of ionized, neutral, and molecular species generally occurs between 0.5-20 jis, 2-100 JIs, and after 50 is, respectively.

1.3 Fundamental Studies

The laser-induced plasma is an important part in the LIBS technique and many works have been dedicated for its characterization. Due to diverse variations in the experimental measurements, the literature reports breakdown threshold irradiance in clean air from 90-1600 GW/cm2 for 1064 nm lasers, 10-ns pulse width, and from 30-380 GW/cm2 for 266-nm lasers (Smith et al. 2001). Simeonsson and Miziolek (1994) studied the effect of the laser wavelength on the breakdown threshold in ambient air. They reported the lowest value of 9.7 GW/cm2 at 193 nm, the values of 270 GW/Cm2 at 355 rum and 200 GW/cm2 at 1064 inm. It was also found that due to the presence of particles in the focal region of the laser beam, the breakdown threshold could be reduced by two orders of magnitude (Leoncioni and Pettingill 1977, and Lushnikov and Negin 1993). The complex nature of aerosol particles and their interaction with the laser beam make it difficult to predict the breakdown threshold. While Reilly et al. (1977) found a decreasing breakdown threshold for water droplets from 0.5 to 15 Inm using a 10.6 gm CO2 laser, Pogodaev and Rozhdestvenskii (1979) showed that the breakdown threshold of water droplets increased with increasing particle size using a pulsed ruby laser.








Radziemski and Cremers (1989) reported that the breakdown threshold was a function of the ratio of particle diameter to laser wavelength. They also suggested a dependency of the breakdown threshold on the total number of particles and size distribution, and on the focal volume of the laser beam (as a function of laser irradiance). An important consideration is that any particle smaller than 10 jim is completely vaporized and incorporated into the plasma (Smith et al. 2001). It is noted that although 10-jKm particle sizes are often reported as an upper limit for complete dissociation, no detailed measurements addressing this issue have been reported; a literature survey regarding this matter is given in Section 1.7.

The laser-induced plasma is a non-equilibrium system from a global scale, but a possible solution to obtain analytical information from the plasma is to consider the plasma as in local thermodynamic equilibrium (LTE). At LTE the plasma is modeled stepwise in depth with the assumption that at any depth there are thermodynamic equilibrium conditions, and at other depths the conditions are different but still in thermodynamic equilibrium. At thermodynamic equilibrium the plasma emitting radiation is directly coupled to the physical conditions of all the matter in the plasma; specifically, the continuous radiation is described by the Planck function, the atomic occupation numbers are specified by the Boltzmann and the Saha equations, and the atomic transitions are determined by the Einstein relations. In 1983, Radziemski et al. (1983) showed that the laser-induced plasma behaves as in local thermodynamic equilibrium (LTE) after about 1 is from the onset of breakdown. They concluded in obtaining LTE by observing the agreement between the results of a theoretical radiationhydrodynamic model and the experimental measurements of the intensities ratios of








C(II)/C(I), N(Il)/N(I), and Be(II)/Be(I) emission lines. They also observed that after certain pulse energy, the excess of energy increases mainly the plasma volume instead of increasing plasma temperature (i.e., 20% increase of temperature over 5-fold pulse energy, 60 mJ to 300 mJ). Capitelli et al. (2000) discussed the LTE assumption under theoretical considerations, and concluded that laser-induced plasmas seemed to satisfy such a condition, but they also noted that phenomena such as low ionization, low excitation temperature, or non-equilibrium recombination could occur in the same temporal scale of LIBS measurements, thus implying the violation of LTE. Yalcin et al. (1999) measured electron density and temperature for the plasma formed by a Nd:YAG laser (100 mJ, 1064 rum, 10 ns pulse). Based on the agreement between the ionization temperature and the excitation temperature, they reported that LTE was reached at delay times of 0.35 pts.

Yalcin and coworkers (Yalcin et al. 1996, and 1999) also investigated the

influence of ambient gas conditions on the laser-induced plasma. They observed that by changing the ambient gas, particulate level, humidity, and laser energy, the electron density and plasma temperature (what characterize the plasma under LTE) varied little. They also confirmed that once sufficient laser energy was available to produce the plasma, additional energy produced a larger plasma volume with the same thermodynamic conditions, similar to the conclusions of Radziemski et al. (1983). Chen et al. (2000) used an Nd:YAG laser (1064 rum, 6.5 ns pulse) to study the effect of the laser energy on the plasma formation in air. They found that 50% of the incident energy was absorbed when the incident pulse energy was just above the breakdown threshold, and this percent of absorption increased up to a maximum value in which the plasma is








apparently saturated. They confirmed also that after the saturation of the plasma, the excess of deposited energy induced an expansion of the plasma, and observed that the higher the incident energy, the farther the initial plasma moved away from the focal spot. Overall, the nature of the laser-induced plasma has been well studied, leading the way to LIBS as an analytical technique.

1.4 LIBS as an Analytical Technique

The technique of laser-induced plasma spectroscopy has been applied to the

analysis of solids, liquids, and gases, and a number of literature reviews are available that cover a wide range of LIBS-based analyses (Smith et al. 2001, Darke and Tyson 1993, Radziemski 1994, Rusak et al. 1997, Schechter 1997, and Song et al. 1997). Relevant to environmental aerosols, several papers have focused specifically on LIBS as a method for the detection and determination of overall elemental mass concentrations, as a technique for continuous emission monitoring, and as a potential method for particle sizing. It has also been investigated as a remote sensing technique.

1.4.1 Elemental Detection and Overall Mass Concentration

The applicability of LIBS in the elemental detection and overall mass

concentration of aerosols has been shown in the large volume of publications during the last decades. In 1981, Radziemski and Loree (1981) introduced time resolution for LIBS to overcome the predominately continuum emission at early time and to optimize the analyte atomic emission lines. Using this approach they were able to detect chlorine and phosphorus-based aerosols in air down to concentrations of 60 ppm and 15 ppm, respectively. Later, Radziemski and coworkers succeeded in the direct sampling of beryllium in air with a limit of detection of 0.6 ng/g (0.7 gig/m3), which is comparable to the detection limit of 0.2 ng/g reported by the well-known technique inductively-couple








plasma atomic emission spectrometry (ICP-AES), and lower than the maximum permissible exposure of 25 ig/rm3 in 8 h given by the Occupational Safety and Health Administration (OSHA). The detection of beryllium was done as a cloud of beryllium particles (estimated diameter less than 10 Jim) generated by laser ablation and as beryllium-chloride aerosol produced by a nebulizer/heat-chamber system.

Essien et al. (1988) reported the detection and quantification of the metals

cadmium, lead, and zinc in aerosol form in air using a detection system comprised by a monochromator, photomultiplier tube and boxcar averager. The aerosol samples were generated in the sub-micron range with a wide variation in diameters by a nebulizerheating chamber arrangement. The limit of detection for cadmium, lead, and zinc were

0.019 (22), 0.2 (252), and 0.24 Jg/g (288 Jg/m3), respectively. The permissible exposure limit reported by OSHA are 5, 50, 5000 jig/m3 for cadmium, lead, and zinc respectively. These results showed the promising capability of LIBS for detection of toxic metals.

Rusak et al. (1997) summarized the versatility of LIBS for the detection of

different elements as aerosols in air. Mass concentration for elements such as lithium, sodium, magnesium, aluminum, silicon, potassium, calcium, manganese, and gallium were reported to have detection limits below 10 ppm. For other elements such as chlorine, strontium, indium, and mercury, the detection limits were in the range of 10-100 ppm, and for elements such as sulfur and arsenic the mass concentration limits were just above 100 ppm. These results were obtained under the criterion of time delay optimization of the appropriate atomic emission line. Other research and review papers focusing on the study of elements and compounds including phosphine (PH3), arsine








(AsH3), and fluorine (CF3H) are reported in the literature (Sneddon and Lee 1999, Peng et al. 1995, and Singh et al. 1997).

1.4.2 Continuous Emission Monitoring

The necessity of real-time and in situ measurements of elemental concentrations makes LIBS a strong candidate for industrial process monitoring. There are very few techniques that can meet these requirements and therefore work as continuous emissions monitoring (CEM) systems. Techniques such as non-disperse infrared and differential optical absorption spectroscopy are mostly restricted for gas phase species, while other techniques such as laser-desorption/mass spectrometry do not perform as in situ and/or non-intrusive measurements. The feasibility of LIBS as a CEM system has been demonstrated by Zhang et al. (1999) who performed LIBS measurements at the EPA rotary Kiln Incinerator Simulator for toxic metals in near real time. The multi-metal CEM test was conducted to target antimony, arsenic, beryllium, cadmium, chromium, lead, and mercury at low (15 gtg/m3), medium (60 jLg/m3), and high (600 tg/m3) mass concentrations. Efforts to monitor antimony, arsenic, and mercury failed because of the target concentration below the LIBS detection limit, but for the remaining metals LIBS successfully monitored them in real time. Because the LIBS system used only a single detection system, each metal was explicitly monitored from one fourth to one half of the reference-method sampling time with a time response of 17 s; EPA establishes RM29 as the reference method for multi-metals analysis. They reported several problems such as interference lines and damage of laser optics that avoided the detection of metals at different target concentrations, and diminished the accuracy of the LIBS measurements. At high target concentrations the LIBS accuracy with respect to that of RM29 for all-four








metals varied from 49% to 77%. At medium concentrations lead could not be detected and the relative accuracy ranged 38%-86% for the other metals. At low concentrations only beryllium and chromium were identified, with accuracy within 31% to 272%.

On-line analysis of chromium-based aerosols using LIBS was reported by Neuhauser et al. (1999) in their efforts for the development of a CEM in industrial effluents. The actual performance test was run at a German electroplating facility regulated to 1 mg/m3 for chromium. They compared the LIBS approach with a conventional filter-based reference analysis (off-line approach) performed simultaneously by an independent laboratory during a 30-min measurement period. The results showed a direct correlation between the on-line LIBS analysis and the filter-based analysis, but the LIBS measurements were overestimated by a factor of two. They attributed the disagreement due to reassembling of the LIBS system at a different sampling point. They also pointed out the difference of both methods. While the filter-based approach had an integrative character (over 30 min), the LIBS measurements were a quasi-continuous and non-integrative method (15 measurements over 30 min with 20-s duration per measurement). In addition, they reported that the dynamic response of the LIBS system ranged from 50 to 100 s. Overall, they showed the potential of LIBS for on-line monitoring of heavy metals in industrial effluents.

In another study of CEM technology, Hahn et al. (1997) investigated the

application and process conditions of LIIBS to metal emissions monitoring in waste combustion systems. Using sequences of 1000 shots, individual shots corresponding to the presence of chromium, manganese, and iron (hits) were collected on-line, in situ at a natural gas-fired, pilot-scale waste processing unit. All spectra for a given analyte were








grouped to calculate an equivalent concentration based on an a priori laboratory calibration scheme. They determined the actual mass concentration for a given analyte by multiplying the corresponding equivalent concentration by the hit frequency for each analyte (i.e., the percent of shots that registered the presence of the analyte emission). The results were compared to those obtained by the RM29, in accordance with EPA standards. For the determination of manganese, LIBS and RM29 gave the same concentration of 3.3 .Lg/m3, while for chromium and iron the LIBS technique yielded results that were consistently 21% and 36% lower than those obtained by RM29 in the range of 2-3 gtg/m3 and 40-140 gtg/m3, respectively. In general, they concluded that finetuning the overall instrument to specific conditions might lead to enhanced sensitivity and performance.

1.4.3 Single Particle Analysis

An extension of the LIBS technique for real-time sizing and elemental analysis of single particles was reported recently by Hahn (1998) and Hahn and Lunden (2000). This novel approach would be the first pure spectroscopy technique in development for the analysis (elemental mass and size) of single aerosol particles. They based their analysis on the point-sampling nature of LIBS and the discrete presence of the particles in a gaseous medium to detect single particles. A two-part calibration scheme (of known mass concentration and known discrete particle mass) was used to calculate a characteristic plasma volume, which is needed for particle sizing. Their results showed that the technique was both robust and highly sensitive during initial laboratory and field measurements. Measurements of calcium- and magnesium-based aerosols were performed and compared to independent measurements using a laser aerosol








spectrometer, with the resulting size distributions in agreement within a 25% maximum deviation. The understanding and potential development of this approach is the focus of this dissertation. In a recent application, which is explained in detail in the following chapters, aerosols in ambient air containing Al, Ca, Mg, and Na were monitored for a period spanning the Fourth of July holiday (Carranza et al. 2001). Measured mass concentrations ranged from 1.7 ppt to 1.7 ppb, and the measured aerosol diameters varied from 200 nm to 2 gim, thereby demonstrating the feasibility of LIBS for analysis of particulate matter in ambient air.

In addition to the previous cited applications, it is worthwhile to mention LIBS as an atmospheric remote sensing technique. Davies et al. (1995 and 1996) coupled a fiber optic system to a conventional LIBS system to deliver the incident laser pulse to the zone of testing and to transmit back to a spectrometer the optical radiation emitted by the plasma. The system was tested for distances up to 100 m between the remote location and the apparatus. The system was capable of detecting chromium, cooper, manganese, molybdenum, silicon, and vanadium concentrations down to 200 ppm.

1.5 Conditional Data Analysis for LIBS

The discrete nature of anthropogenic or natural aerosols combined with the essentially point-volume sampling method of the LIBS technique make the detection of particulates possible. For instance, at low particulate conditions, the traditional ensemble averaging of LIBS spectra may reduce considerably the signal-to-noise ratios of the targeted element contained in the particles. Knowing of this peculiarity for LIBS based aerosol measurements, Hahn et al. (1998 and 2000) developed a conditional data analysis algorithm for LIBS-based detection of individual aerosol particles that rejected spectral








data based on the absence of atomic emission corresponding to the targeted analytes. Spectra were classified as particle hits if the ratio of the atomic emission intensity of the analyte line to the intensity of the adjacent featureless spectral region (i.e. continuum) exceeded a threshold value. The algorithm needs at least two emission lines of the analyte, one for triggering, and the other lines for analysis. The threshold value was typically set to obtain from 0.1% to 0.3% false hits in order to retrieve spectra with analyte emission line intensities comparable to the root mean square (rms) values of the spectral signal shot-noise. Using the conditional data analysis approach, single-shot LIBS spectra were identified and analyzed corresponding to aerosol mass concentration levels from low parts-per-billion to parts-per-trillion levels.

Analogously, Schechter (1995) used a novel rejection algorithm to identify singleshot spectra and to screen anomalous events out such as those that did not contain elemental emission lines, and those events related to laser fluctuations or screening particle effects. His algorithm did not address the discrete nature of aerosols and its low particle density case; rather it was more of a data filtering approach to improve the analyte signal. Nevertheless, the criterion of the algorithm to identify the presence of the targeted element was related to the sum of the intensities of all expected lines. If the total intensity did not reach a threshold value, the spectrum would be rejected. As a result of this approach, Schechter demonstrated that the ensemble of the filtered spectra for Znbased aerosols improved the detection limit by a factor of three compared to direct ensemble averaging. A similar approach was used by Cheng (2000) and Martin and Cheng (2000) to measure trace metals of fine particles. The screening data algorithm was based on the (i) presence of multiple analyte emission lines, (ii) specified maximum








analyte emission peak fluctuations, and (iii) specified maximum full-width, halfmaximum analyte line widths.

The use of a suitable conditional analysis algorithm can yield significant increases in signal-to-noise ratios, but more importantly, opens the door to the analysis of singleshot LIBS spectra that correspond to individual aerosol particles. Such analysis is the basis of this research.

1.6 Single-Shot LIBS-Based Aerosol Analysis While Schechter (1995), Cheng (2000), and Martin and Cheng (2000) used

single-shot measurements to improve the signal-to-noise of trace aerosol composition, Hahn et al. (1997, 1998, and 2000) used it mainly to analyze single aerosol particles. Schechter and the others based their analysis on spectral filtering, which was used to obtain a representative ensemble-averaged spectrum. These filtered spectra were subsequently used to make a calibration curve of the targeted element (peak-to-base ratio versus analyte concentration). Xu et al. (1997) also applied single-shot measurements to analyze absolute particulate materials. They processed emission signals (absolute peak) from filtered single-breakdown events to obtain absolute concentrations, basing their method on the assumption that the same fluctuation pattern observed in the spectral peaks is present in the baseline as well. Consequently, a plot of the peak intensity against the average baseline intensity should provide a straight line with a slope related to the concentration of the analyte.

Hahn and coworkers (1998 and 2000) developed a two-part calibration scheme to determine the size of aerosol particles. The idea was to correlate an equivalent analyte mass concentration to the analyte mass contained in the plasma volume when a single particle was sampled by the plasma volume, designated a particle hit. The sequence of








the algorithm was to calculate first the plasma volume and then the particle size. First, the LIBS signal (ratio of atomic emission peak area to emission continuum intensity near to the peak) for the average of thousands of laser pulses was correlated to a known average analyte mass concentration. Second, the LIBS signal was calculated from the ensemble-averaged spectrum corresponding to a monodisperse particle stream of known size, mass, and composition in which the average particle concentration was adjusted to promote single-particle detection. Third, an equivalent mass concentration (also expressed as the ratio of single particle mass to plasma volume) was determined for the known particle mass to calculate the plasma volume. This last statement can be expressed by equating the actual mass concentration given by LIBS-based analysis and by aerosol mechanics

ActualMass.Concentration = X . F = 5.i N, (1.3) where Xis the equivalent mass concentration of all particle hits, F is the sample frequency or the percentage of laser pulses that sample an aerosol particle, R is the average particle mass, and N is the number density of particles (particles per unit volume). Assuming a Poisson sampling mode and p as the average number of particles per plasma volume, the sample frequency can be written as the probability of collecting one or more particles by

F = 1 - exp(-p), (1.4)

where p is readily expressed by the characteristic plasma volume V and the particle number density N noting that p = VpN. For the case of low particle sampling rate (i.e.,/p <<1), Equation 1.4 may be expanded using a Taylor series to become
F = Il-(I-,u + 2 -.








which neglecting higher order terms yields

F =,u = VP *N, (1.5)

Substituting Equation 1.5 in Equation 1.3 and solving for X yields the relation between the equivalent mass concentration, single particle mass, and the characteristic plasma volume:

X = (single particle mass) / VP. (1.6)

Finally, knowing the chemical state of the analyte, the equivalent spherical

diameter of the particle in the plasma volume is calculated using the following relation = [6XVp 11/3
D L- - , (1.7) [rp fJ

where p is the bulk particle density andf is the analyte mass fraction with respect to the bulk particle.

Therefore, the particle sizing technique follows from the steps outlined above. For an unknown particle distribution the equivalent mass concentration is calculated for an identified, single-particle hit using a traditional LIBS calibration curve. The absolute mass of the particle is then calculated using the equivalent mass concentration and the characteristic plasma volume. Last, by knowing or assuming the composition and density of the particle, the equivalent diameter is calculated.

1.7 Vaporization of a Single Aerosol Particle Using LIBS

Notwithstanding the significant body of research seen in previous sections

regarding laser-induced breakdown spectroscopy for aerosol analysis, to date no research has systematically addressed the fundamental assumption inherent in all quantitative LIBS measurements, namely the assumption of complete breakdown and vaporization of








all analyte species that comprise the aerosol particles of interest. Then, the important question remains as to what is the largest particle size for complete dissociation and vaporization of individual particles suspended in a gas stream?

A survey of the literature regarding the largest particles that can be completely vaporized for quantitative LIBS-based analysis reveals no systematic study designed to specifically address this issue. Cremers and Radziemski (1985) explored LIBS for the detection of beryllium particles deposited on filters corresponding to particle diameters of 50 nm, an ensemble collection of particles ranging from 0.5 to 5 pm, and for nominally 15-jim sized particles. They used a cylindrical lens to focus a laser beam directly on the surface of the filters, thereby producing a long plasma volume that engulfed the deposited beryllium particles, and subsequently collected the spectral emission. They reported a different analyte response, manifest as different calibration curve slopes, for these three different particle size classes, and concluded based on their experimental observations that incomplete particle vaporization occurred for particles with diameters greater than about 15 pWm. Other studies explored direct LIBS-based analysis of beryllium aerosols using particles less than 10 pWn in diameter (Radziemski et al. 1983, and Essien et al. 1988), and noted in the latter study that such a particle size is consistent with complete particle vaporization.

Several contemporary research papers have cited a 1 0-p r upper size limit for complete particle vaporization (Ottesen et al. 1989, Yalcin et al. 1996, and, Hahn et al. 1997), some referring to the extensive work of Radziemski and Cremers, and some not citing any particular reference source. Based on the overall body of LIBS research, it appears that what was presented by Radziemski and coworker as a useful guideline for








quantitative LIBS analysis of aerosols, namely the 1 0-tm upper size limit, has in essence become the accepted upper size limit for complete vaporization of aerosol particles within laser-induced plasmas. However, following extensive literature reviewing, no research has been reported that specifically addresses the issue of complete vaporization of individual airborne particles using laser-induced breakdown spectroscopy. This issue is addressed in Chapter 5 of this dissertation.

1.8 Motivation and Objectives for Present Study

The need for the development of new real-time techniques for the analysis of single aerosol particles challenges LIBS capabilities shown in the last decades. In contrast to the traditional ensemble averaging of global mass concentration measurements, single-shot LIBS analyses opens the door to the sizing and mass quantification of single particles. No studies have been reported to date to answer new questions regarding important issues such as shot-to-shot plasma fluctuations, experimental precision, and plasma-particle interactions. Therefore, this research directly supports the longer-term goal of establishing the LIBS technique as a quantitative diagnostic tool for the simultaneous measurement of particle size and particle composition, with a focus on ambient air fine particulate matter. The purpose is to gain an understanding of the effects of plasma-particle interactions on the optimization and precision of LIBS-based single-shot analysis of gaseous and aerosol systems. The following technical objectives are addressed by this research.

1. Investigate the degree of laser-induced plasma homogeneity, and the correlation

between plasma homogeneity and LIBS signal response to discrete aerosol

particles.






21

2. Investigate the optimal laser pulse energy for enhanced signal response and signal

robustness.

3. Investigate the relations between laser-induced plasma volume and particle

sampling volume.

4. Investigate the upper limit of single particle vaporization in the laser-induced

plasma.













CHAPTER 2
EXPERIMENTAL FACILITIES AND FUNDAMENTAL PLASMA CHARACTERISTICS

As the LIBS technique is further developed and refined as an analytical tool for aerosol analysis, well-controlled experimental conditions are required. The use of nebulizers for sample introduction is a standard mode for inductively coupled plasma atomic emission spectrometry (ICP-AES) and inductively coupled plasma mass spectrometry (ICP-MS). Nebulizers generally work by converting bulk analyte solutions into an aerosol spray for transport to and injection into the plasma. The aerosol characteristics of a given nebulizer design are closely related to the overall precision of the analytical system, with resulting limitations attributed to coarse aerosol size and low analyte transport efficiencies. Ideal nebulizer characteristics include 100% analyte transport efficiency, the generation of fine aerosol droplets, and operation without clogging.

Another desirable feature of an aerosol generation system is the ability to

introduce aqueous particle suspensions into a gaseous sample stream. For instance, introduction of particles from suspension into a gaseous stream is necessary for singleparticle diagnostics as developed for the LIIBS technique. Therefore, this section describes such an aerosol generation system along with the LIBS instrumentation required for plasma excitation and light collection, and the characteristics of the plasma produced with the coupling systems to assess the performance as a calibration source for the development of laser-based diagnostic techniques based on LIBS.








2.1 Aerosol Generation System
The aerosol generation system is shown schematically in Figure 2.1. The primary components are the pneumatic-type medical nebulizer (Hudson model 1720), the conical mixing/drying section, and the sample analysis section. A flow of compressed, dry nitrogen is used to drive the nebulizer flow, and the generated-aerosol droplets are delivered directly into the mixing/drying section. The nebulizer nitrogen flow is filtered through a HEPA filter cartridge, and regulated with a laminar-flow element flow controller. The full-scale flow is 10 liters per minute (1pm) with accuracy of� 0.1 1pm. The nebulizer flow is directed through the center hole of a stainless steel, porous plate 125-mm in diameter. O-rings seal the porous plate along the perimeter and at the center hole. A gaseous co-flow is introduced below the porous plate. The function of the gaseous co-flow is to provide bulk flow through the aerosol generator, both facilitating the drying of aerosol droplets and transporting the resulting solid particulates to the sample analysis section. Two parallel thermal-type mass flow controllers, each one with 30-lpm capacity and � 0.3-lpm accuracy, are used to meter the co-flow gas.

The mixing/drying section consists of a 175-mm length of micropolished stainless steel. The inner diameter of the mixing/drying vessel is 125 mm at the porous plate, and the vessel tapers to an inner diameter of 25 mm over a linear distance of 125 mm. A final 50-mm long, 25-mm constant diameter section connects the mixing/drying section to the sample analysis chamber. The sample analysis chamber is a standard stainless steel 70mm (2.76") 6-way vacuum cross. The cross is mounted directly to the mixing/drying section and set to provide optical access to the generated aerosol stream. Three of the horizontal flanges are fitted with optical quality windows, and the fourth horizontal









flange is fitted with a 50-mm diameter, 75-mm focal length UV grade fused silica lens. The focal point of the lens is centered on the central vertical axis of the mixing/drying section.




[ Exhaust vent


Two mass-flow
controllers, 0-30 1pm


Coarse filter (Desiccant dryer) j HEPA filter


Six-way cross testing chamber


Laser beam


Generated Aerosol


Nebulizer


Mass-flow controller,
0-10 1pm


N2 cylinder


Figure 2.1 Schematic of the aerosol generation system.


A detailed description of all the components is given in the Table 2.1. The last

four items listed in this table correspond to the subsystem for ambient air monitoring, and it was used in the firework particle monitoring experiments, which are developed in the next chapter.








Table 2.1 Aerosol Generator Specifications

Unit Description Manufacturer

Co-flow controller Model 8270, 0-30 1pm Matheson Gas Products
(two units) nitrogen @ 25�C
Co-flow meter Transducer model 8272- Matheson Gas Products
(two units) 0434
Nebulizer flow meter and Model MC12PS, 0-10 1pm Alicat Science Inc.
controller air @ 250C
Two pressure gauges 0-300 Psi Ashcroft

Two HEPA filters Product # 12144 capsule Gelman Laboratory

Coarse filter Part # X03-02-000, with Wilkerson Corporation dryer's silica gel,

Nebulizer Pneumatic medical type Hudson #1724

Air compressor Model PK-5060V, 5 HP Puma and 125 Psi @ 15.7 CFM,

489 PE tubing Flexible tubing, 1/8xl/4xl/16 Fisher Scientific and 1/4x3/8xl/16

Testing chamber Stainless-steel six-way Huntington Labs cross, I.D. 35 mm

Drying section Micro-polished Stainless- Custom design and steel, I.D125x L175x I.D25 fabrication Vacuum pump Model SR-0015-VP Thomas Compressors & Vacuum Pumps
Flow meter Tube # GJ502, 3-48 lpm air Gilmont Instruments

Stainless Tubing O.D. 5/8" ASTM A-269- MSC Industrial Supply Co.
94A/ A-213-94B
Remove PM larger than 10- Rupprecht and Patashnick P ln, 1 m3/h Co., Inc.








2.2 Nebulizer Characterization The nebulizer may be characterized by the consumption rate of nebulized liquid as a function of gas flow rate into the nebulizer, as well as by the resulting size distribution of aerosol droplets. A high degree of precision in the nebulizer mass flow rate is desirable for the current use for sample introduction into the aerosol generation apparatus and subsequent LIBS calibration. A summary of the specification for all the standard solutions and suspensions used in this research is presented in the Table 2.2 and Table 2.3, respectively.

Table 2.2 Standard Aqueous Solutions

Solutions Description Manufacturer DI water Deionized ultra filtered Fisher Scientific water, W2-20

Industrial Grade Nitrogen 99.7% N2, less than 32 ppm Prax Air of H20
Calcium, Iron, Aluminum, 10000 pg/mI ICP Standard Spex, Inc. Sodium, and Magnesium in a matrix of 5% HNO3 Titanium 10000 jig/ml ICP standard Spex, Inc.
in water


Table 2.3 Microsphere Silica Particle Suspensions

Particle Diameter, Standard Particle Density, Pm Deviation Particles/cm3
1.02 NA 9.149109 1.5 NA 3.122 109 2.08 NA 1.17 109 2.52 NA 6.4 109 3.0 NA 3.79 109 4.5 NA 1.12 le 5.13 NA 7.58 108








2.2.1 Mass Flow-Rate Calibration

The mass flow rate of the nebulizer was calibrated using distilled and deionized water over a range of nebulizer gas flow rates from 4.5 to 6.0 1pm of dry nitrogen. A gravimetric approach was used for flow periods ranging from 10 to 30 minutes for each gas flow rate. Liquid mass lost from the nebulizer varied from 689 to 3933 mg over the test matrix, while the analytical balance had an accuracy of 0.5 mg. The nebulizer calibration curve is presented in Figure 2.2 for the nebulization of deionized water. As observed in the figure, the nebulizer output is highly linear over this reported range of gas flows.


0.14
C E
E 0.12 S0.1


0.08


S0.06 y =-0.15119 + 0.048246x R= 0.99965
4 4.5 5 5.5 6 6.5 Nominal Gas Flow Rate, Llmin

Figure 2.2 Nebulizer calibration curve of deionized water. Each data point is the average of a minimum of 3 runs (error bar = - one
standard deviation).


It was noted that the nebulizer did not produce an aerosol output for nitrogen gas flow rates below a minimum value. This minimum flow rate was approximately 4 lpm. It was also observed that as the gas flow rate increased to values significantly larger than








6 1pm, the high precision characteristic of lower flow rate diminished as seen the Figure

2.2 data. Once the regions of linear operation were determined, experiments were limited to the linear regime, with a usual nominal gas flow rate of 5 1pm.

2.2.2 Lifetime of Generated Droplets

The nebulizer produces a precise mass flow rate of fine droplets, which are introduced into the gaseous co-flow within the mixing/drying section. When the generated droplets mix with the co-flow gas stream, the droplets begin to evaporate via mass diffusion of water to the surrounding gas (Turns 1999). Diffiusion-controlled evaporation is an appropriate model for the current range of nebulizer droplets (minimum diameter of -0.1 pm) and corresponding range of Knudsen numbers (Kn < 0.001) (Davis et al. 1980). The time t0 for a droplet to completely evaporate, assuming a binary mixture of water vapor in air, is given by the expression to = Do2 1K, where D0 is the original droplet diameter and K is the evaporation constant.

For the present exercise, the evaporation constant was calculated as 1.7x 1 09 m2/s by using a binary diffusion coefficient of 2.2x10-5 m2/s, and a water mass fraction of

0.0 16 at the droplet interface (equilibrium vapor pressure) and 0.008 for the gas stream (all nebulized water has evaporated). Using these parameters, drying times for initial water droplets of 250, 500, and 2000 nm in diameter were calculated as 0.04, 0.15, and 2.3 ms, respectively. These initial droplet diameters are consistent with those produced by the current nebulizer, as discussed below. For a nominal total gas flow rate of the aerosol generator of 50 1pm (nebulizer flow plus co-flow), a single droplet with drying time on the order of 10 ms has an equivalent drying distance of less than 1 mm. This is a very short distance compared with the 175-mm-mixing/drying-section length, which has








a total residence time of approximately 790 ms. The nebulizer, however, produces a concentrated aerosol mist of droplets, hence phenomena such as saturation or nearsaturation effects over the droplet-to-droplet length scales should be considered. Accordingly, for a two-order of magnitude decrease in the evaporation constant resulting from multiple droplet effects, sufficient residence time is still provided to dry a 2-riM droplet diameter (230 ms).

Laser light scattering measurements were performed to assess the evaporation of the nebulizer droplets prior to the optical sampling chamber. The laser beam of a Qswitched frequency-doubled Nd:YAG laser (X = 532 nm), at 30-mJ-pulse energy was focused at the center of the sample chamber using a 25-mm diameter, 250-mm focal length lens. Scattered light was collected at 900 using a telescope with unity magnification, and recorded with a 200-ps rise time phototube detector. Specifically, laser light scattering measurements were recorded in a pure gaseous co-flow stream of air, and in flow streams corresponding to the nebulization of pure deionized water and aqueous solutions of iron. The scatter laser-pulse profiles are shown in Figure 2.3, and correspond to 100-shot ensemble average.

For the case of nebulized deionized water, any water droplet entering the sample chamber would contribute to the measured light scattering signal. The scattering signal corresponding to the baseline case of the purified co-flow gas only is equal to the sum of Rayleigh scattering from the air molecules and stray light, including reflections from within the optical sample chamber. The light scattering signal was found to increase by only 10 % for the nebulization of deionized water and by more than 300 % with the nebulization of 10,000 ptg/ml of dissolved iron with respect to the baseline case of pure








air in both cases. The increase in scattering with the nebulization of deionized water only is attributed to either increased stray light due to light scattering within the mixing/drying section or to the penetration of a small number of very large droplets that did not sufficiently evaporate. The 3-fold increase in scattering with the nebulization of the iron solution is attributed to the formation of a large number of iron-based particulates formed by the evaporation of the nebulized water droplets.

14 . ' ' . . . . .

12
a i Fe@ 10000 p.g/ml 10
8 1 Fe@3333 pg/ml

I DI water
4 I
SPurified
2 air

0
0 5 10 15 20 25 30 Time, ns

Figure 2.3 Light scattering responses from the testing chamber due to the excitation of a laser pulse of 30 mJ under a co-flow of 60
1pm of purified air.


2.2.3 Aerosol Particle Characterization

The size of aerosol particles was obtained corresponding to the nebulization of either iron or titanium aqueous solutions ranging between 2500 and 10000 Lg/ml. Aerosol particles were collected directly on carbon coated, copper TEM grids, which were placed directly into the center of the aerosol stream at the exit plane of the 6-way sample cross. Aerosol particles were collected for 20 minutes for each sample condition and the samples were subsequently analyzed using transmission electron microscopy








(TEM) and X-ray diffraction. Aerosol size measurements were made directly from TEM micrographs using a light box and micrometer. TEM micrographs for generated-aerosol particles are presented in Figure 2.4.

' " A











Figure 2.4 Aerosol particles generated by a nebulization rate of 0.09 mi/mmn into a co-flow of 42 1pm of air: a) For 2500 g~g/ml of Iron, and b)
For 10000 g~g/ml of titanium (TEM micrograph of 50K magnification).


As can be seen from Figure 2.4, while the nebulization of iron solution produced primarily spheroidal-shaped particles, the nebulization of titanium solutions produced highly crystalline particles with a linear, rectangular geometry. In the case of iron solution, the reported particle diameter represents the average of the major and minor diameter of the spheroid particles. In the case of the titanium-based particles, they were modeled as rectangular rods with a square cross-sectional area. Width and length were recorded directly from the TEM images, and volume-equivalent spherical particle diameters were then calculated from the relation d = (6LW2/g)"3'1. The following Figure 2.5 and Figure 2.6 show the equivalent particle size distribution obtained from aqueous solutions of iron, and titanium and formed at the testing chamber. Both iron- and titanium-based particles showed well-defined distributions, with a skewing toward larger sizes, thus given a good representation of the generated particles. The well-dispersed









aspect ratios (length over width) observed from Figure 2.7 also suggest that the titaniumbased crystals are formed by non-preferential growth along all facets.


Fe at 4425 jig/m3 air as FeO Particles with
Mean Diameter = 13.2 nm











30 40 50 60 70 80
Diameter, nm


Figure 2.5 Iron-based particle size distribution at airborne mass concentration of 4425 Ptg/m3 corresponding to an aqueous solution of 2500 ptg/ml.


1 1/11


I '. I . .


35 30 25 .,20 15 10

5


17700 Axg/m3 iO Particles with n Diameter =73.5 nm











200 250 300


Ti at as Ti Meat


Diameter, nm

Figure 2.6 Titanium-based particle size distribution at airborne mass concentration of 17700 [Lg/M3 corresponding to an aqueous solution of 10000 pg/ml.












0
10 0 TiO Particles
0 at 17700 1g/m3
8 00 00
II 00
0
0 0
2 000 o
Qo. 4 0��_ 0


-,0 0 000
0
0 . . . . I ... m I .. . . I .. I . . . .
0 50 100 150 200 250 300
Equivalent Spherical Diameter, nm

Figure 2.7 Aspect ratio for the titanium-based crystal particles at airborne mass concentration of 17700 jig/m3.


Table 2.4 Summary of Generated Aerosol Particle Characteristics

Aqueous solution 2500 10000 2500 10000
concentration pg/ml Fe pg/mI Fe jig/ml Ti jig/ml Ti

Nebulization rate, ml/min 0.09 0.09 0.09 0.09

Co-flow, 1pm air 42 42 42 42

Mass concentration, jig/m3 4425 17700 4425 17700

Mean diameter, nm 13.2 68 74 73.5

Standard deviation, nm 7 29 52 58

Particle concentration, cm3 64.5 107 1.9 107 0.4 107 1.8 107

Chemical state FeO FeO TiO TiO








The analysis of the X-ray diffraction micrographs allowed the determination of the lattice d-spacing of the collected particles. They were compared with reference library values to determine the chemical state of the particles, resulting in the conclusion of primarily FeO for the iron-based particles and TiO for the titanium-based particles. It was noted that no d-spacings were observed to be consistent with the formation of hydrated or nitrated species. A summary of the aerosol particle characteristics is reported in Table 2.4.

2.2.4 Nebulizer Droplet Characterization

The TEM analysis and X-ray diffraction data were used to infer information about the size characteristics of the nominal water droplets produced by the nebulizer. Specifically, the initial diameter of the liquid droplet that subsequently produced each solid particulate can be calculated from a conservation of mass relationship based on the analyte species; namely the mass of the analyte species (e.g., titanium or iron) in the dry particle must equal the mass of analyte in the original liquid droplet. This relation may be expressed by the identity

)r D' S= ICD 3 pf, (2.1) 60:


where Do is the initial liquid droplet diameter, S is the solution concentration of the aqueous analyte (J.tg analyte/ml), D, is the particle diameter of the measured solid aerosol, p is the bulk density of the solid aerosol, andf is the mass fraction of analyte in the solid aerosol particle. For the FeO particles, the density is 5.7 g/cm3 and the iron mass fraction is 0.78. For the TiO particles, the density is 4.9 g/cm3 and the titanium mass fraction is 0.75. These values were utilized to convert the respective aerosol particle size distributions recorded for the titanium and iron-based solid particles into








primary water droplet diameters produced by the nebulizer. The iron-based and titaniumbased particle distributions both yielded similar nebulizer droplet size distributions, namely mean droplet diameters of 520 and 528 nm, respectively. This consistency is significant, in that the droplet size distribution created by the nebulizer is not influenced appreciably by the nature of the nebulized aqueous analyte solution. A composite distribution of the nebulizer droplet sizes based on the combination of TiO- and FeObased particle measurements is presented in Figure 2.8, as calculated using Equation 2.1 as described above. The mean droplet diameter is 524 nm, and standard deviation and modal diameters equal to 398 and 340 rnn, respectively.

100 #
Mean Diameter = 524 nm
St Deviation = 398 nm
80 Modal Diameter = 340 nm

S60
0
40 201

0
0.0 0.5 1.0 1.5 2.0 Nebulizer Droplet Diameter, grm Figure 2.8 Nebulizer droplet size distribution based on the combination of iron and titanium solution at 10000-jRg/ml aqueous
concentrations.


The droplet size distribution is well described using a zero-order lognormal distribution -ZOLD (Espenscheid et al. 1964) of the form e [_(lad2dM)2
p(d) p(o/2) exp ,(d 1 (2.2) aod. L 2 02








which is an equivalent expression for the lognormal distribution (see Appendix A). The lognormal distribution is widely known for the description of aerosols; however, the ZOLD is conveniently parameterized by a modal diameter (d4) and by a dimensionless width (ao) instead of the geometric mean and the geometric standard deviation needed by the lognormal distribution. The ZOLD that describes the present droplet size distribution uses a modal diameter dm = 340 un, and a dimensionless measure of width cr. = 0.45. This a. was obtained as the best value that fits the main characteristics of the droplet size. The mean (dmea,,) and standard deviation (r) of the droplet size distribution based on ZOLD are given by the relations

dm, =d,. exp(o- a), (2.3) o = dM [exp(4ao) - exp(3' )1 /2, (2.4) which yields a value din,,,,= 460 rn and a corresponding standard deviation of a= 218 un. Alternatives values for a. can be calculated based on the mean and standard deviation of the measured particle distribution (see Figure 2.8). For example, using the mean droplet diameter of 524 nm and the modal diameter of 340 rm, the Equation 2.3 would report a value of 0.537 for ao, and in the case of using the standard deviation and modal of 398 nm and 340 un respectively, the Equation 2.4 can be solved for a ao of

0.609. The agreement of the distributions using these dimensionless widths (cro) with the droplet size histogram is clearly seen in Figure 2.9. Then, Equation 2.2 may subsequently be used to estimate the size distribution of generated aerosol particles based on the analyte concentration of the aqueous solution and an a priori estimate of the density and mass fraction of the solid particulates.








100


80

=0.609
60
0.537

S40- c= 0.45

20 -,

0 0 0.5 1 1.5 2 Droplet Diameter, im

Figure 2.9 Zero-order lognormal distribution with modal diameter 340 nm at three dimensionless widths a0 that describes the droplet
size distribution created by a medical nebulizer.


2.3 LIBS Instrumentation

A schematic of the experimental LIBS setup is shown in Figure 2.10, and a

detailed description of the components is cited in Table 2.5. The excitation source is a 1064-nm Q-switched Nd:YAG laser with a nominal pulse width of 10 ns, maximum pulse energy of 400 mJ, and operating with a 5 Hz pulse repetition rate. The laser pulseto-pulse energy stability is specified as 4% rms. The output laser beam of 7 mm in diameter is expanded to an effective beam diameter of 12 mm using a telescope and an aperture (not shown schematically). Then, the expanded beam is passed through a 50mm diameter pierced mirror, and focused into the sample chamber with a 75-mm-focal length lens (UV grade quality) of 50 mm in diameter to create the plasma. The plasma emission is collected along the incident beam in a backward direction (1800) by the same primary focusing lens, separated by a pierced mirror, and launched into a fiber optic bundle using a matched 75-mm-focal length lens. The fiber optic cable is coupled to a









0.275-m spectrometer, with an optical dispersion of 0.035 nm/pixel for the 2400groove/mm grating. The dispersed emission is then recorded using a time-gated, intensified charge-coupled device (iCCD) detector array.


UV-enhanced 1
Dierced m


Six-way crossPlasma
(plasma chamber) volume 50mm




Beam dump

Sapphire Piano-convex lens window UV grade quartz
1064-nm AR (both sides) (f= 75mm)


Piano-convex lens broadband UV-AR coating (f= 75 mm)


broadband irror
Laser Power
Supplier


12mm 7mm


1.7x telescope


Nd-YAG laser


Figure 2.10 Experimental setup of a LIBS apparatus.


The 14-bit A/D converter of the iCCD readout gives a factor of 50 dynamic range for a nominal baseline signal level of 325 counts for a given on-chip binning factor. In addition, the spectral stripe is uniform over 100 pixel rows of the iCCD array. By varying the software-controlled on-chip binning of rows by multiples of 5, and additional factor of 20 in signal gain is achieved. Therefore, a factor of 1000 is realized for the overall dynamic range of the current LIBS detector system, which enables the size analysis of aerosols over a factor of 10 with respect to particle size (e.g., mass a D3).


UV-Quartz window








Table 2.5 Laser and Optics Specifications


Unit Description Manufacturer

Nd:YAG laser 1064 nm, 10 ns pulse, CFR Big Sky Laser 400 mJ Technologies, Inc. iCCD Pulse Generator Model PG-200, Princeton Instruments, Inc.
programmable pulse

iCCD Detector Controller Model ST-135 controller Princeton Instruments, Inc.

Intensified charge-couple
iCCD Camera device, model 1024-MLDS- Princeton Instruments, Inc.
El

ICCD Chiller Refrigerator recirculator, Neslab model CFT-25

Spectrometer 275 0.275 meter spectrometer, Action Research 2400 groves/mm Corporation Action Research
Spectrometer Controller Scan controller, model 275 coRerai Corporation


Optical Telescope xl.7, two UV-grade 1064- CVI Laser Corporation nm AR lenses

Elliptical Pierced Mirror (2 UV-grade AR enhanced CVI Laser Corporation
inches)

Focusing Lens 75-mm focal length, 50-mm CVI Laser Corporation dia, UV-grade, 1064 nm AR

Fiber Optic 1 -m fiber optic bundle, LG- Action Research 455-020-1 Corporation


2.4 LIBS Spectral Analysis

A representative spectrum, corresponding to the ensemble average of 1200 laser pulses, is shown in Figure 2.11 for iron at 4425 gg/m3 in purified air. As observed in the figure, a typical spectrum is characterized by continuum emission and by atomic emission lines. In a laser-induced plasma, continuum emission is a result of two main








mechanisms: a Bremsstrahlung process (free-free transition) and a recombination process (free-bound transition). The Bremsstrahlung process occurs when an electron passes close to a positive ion and due to quantum mechanic effects its trajectory changes. The acceleration of the electrons in this way causes it to radiate electromagnetic energy -this radiation is called Bremsstrahlung that in German means breaking radiation. All electrons go from a free-high energetic state to another free-low energetic state, and the differences in energy correspond to radiate energy. In a different mechanism, the recombination process or the free-bound transition takes place when free electrons, which are in a continuous energetic state in the plasma, return to the nucleus of the atoms. The free electrons are captured by the electric field of positive ions and release electromagnetic energy corresponding to the difference of energy between the continuous state and the level of energy that the electron occupies in the nucleus.

2000 . I . . I . . I . I .I I.
E - FeII 259.94 nm 1500 C q: E .: C
C14 CDC C N E cIlt E M LL..






240 245 250 255 260 265 Wavelength, nm

Figure 2.11 LIBS spectrum for an iron mass concentration of 4425
/3in purified air. Ensemble average of 1200 laser pulses (8 s

delay time with 5 lxs integration time).








In contrast to the continuum emission that is a continuous spectral

electromagnetic radiation, the atomic emission is a type of bound-bound transition in which an excited bound electron emits a discreet quantum of energy (usually due to spontaneous emission) corresponding to its de-excitation to a lower energy level in the atom. The quantum nature of the electron in the atom makes the transitions governed by selection rules based on the quantum numbers of the initial and final state. The initial and final state of the electron can be described by wave functions, and the probability that the transition occurs increases as these wave functions overlap. These atomic transitions (also called atomic emission lines) are characteristic for each element of the periodic table and superimposes to the continuum emission as observed in Figure 2.11. A formal definition of the continuous emission (Wsr- .Hz .cm-3) is presented in the following Equation 2.5 (Griem 1997)


1'4 = 2 5a 3Z4 .('&!i'312 E a 3N N' *[.1eXn( AE, +h v) (2.5)
B 343-,rr kT) H~e 0j kTJ and .. 2zE E - 'exp k +E exp[ k where a is the fine-structure constant, z is the charge of the ion, EH is the ionization energy of hydrogen, k is the Boltzmann constant, T is the thermodynamic temperature, a is the Bohr radius, N is the electron density, N is the ion density, AEz is the reduction in ionization energy, h is the Planck constant, vis the photon frequency, gf is the Maxwellaverage free-free Gaunt factor, g,, is the bound-free Gaunt factor, and n and n. are the principal quantum number and the maximum quantum number.








The Einstein-Boltzmann relation (Wsfr .cm"3) expressed by the Equation 2.6 determines the intensity of a spectral emission line (Lochte-Holtgreven 1995)


i' = I Akigk h. N.exp(_ Ek ) ,'(2.6) 41r U A Ukr

where Aki is the transition probability for the emission line, gk is the statistical weight of the excited state, U is the partition fimction, N is the atom density, 2 is the ki transition wavelength, and Ek is the upper-level excitation energy.

In practice, the electromagnetic radiation from the plasma is collected after a

period of time in which the atomic emission lines of the analyte of interest appear (due to the differing rates of decay between continuous and atomic emission); this time is called the time delay or gate delay. The collected light is averaged for a period of time called integration time or just time width. The evolution of the atomic emission and the continuum emission with time delay is shown in Figure 2.12 for the case of carbon I at 247.8 nm in the spectral window 240-260 nm. It is clearly visible from the graph when the spectra are compared that the continuum emission and the atomic emission have different rates of decay, indicating in some way that an optimal relation may exit to enhance the analyte atomic emission. This behavior is explained in detail below for carbon. The designation used to typify an atomic transition is related to the state of ionization, specifically I describes a neutral atomic transition that expresses a boundbound transition when the atom has all its electrons, an II corresponds to the first ionization transition that is a bound-bound transition when the atom misses an electron. Figure 2.11 shows a significant number of iron atomic emission lines present in the spectrum (Fe II) at 238.2, 239.56, 240.49, 256.26, 258.59, 259.94, 260.7, 261.18, and 263.1 nm, and the atomic emission line carbon I at 247.8 nm.








7000 , , , .

6000 5000

4000

3000 4-ps time delay ,...
2000
....-.......... lO-p~s time delay _ .1000
- 2-pstime delay

240 245 250 255 260
Wavelength, nm

Figure 2.12 Effect of time delay in atomic emission and continuum emission using an integration time of 2 s at each
designated delay time. Atomic emission line C 1 247.8 nm.


2.5 LIBS Calibration
A significant use of the described aerosol generation system in section 2.2 is the development of calibration curves for laser-based diagnostic techniques, such as laserinduced breakdown spectroscopy in the present study. Knowledge of the nebulization rate, the analyte solution concentration, and the co-flow rate enable calculation of the resulting analyte mass concentration produced in the LIBS sample chamber. The analyte concentration C (mass of analyte per volume of gas) is given by the equation NR . SS (2.7) Qo-flow + Qb

where the nebulization rate (NR) is given by the correlation in Figure 2.2, SS is the analyte solution concentration (pg/ml) in the nebulizer, Qc,-loW is the flow rate of the coflow gas, and Qeb is the nebulization gas flow rate. A nominal analyte solution concentration of 2500 jtg/ml yields a mass concentration of 4425 gg-of-analyte/m3 of gas








when utilizing a co-flow of 42 1pm and a nebulization gas flow of 5 lpm. In addition to the mass concentration, the number density of solid aerosol particles may be estimated based on the nebulizer droplet distribution reported above and the liquid nebulization rate. For a nebulization rate of 0.09 ml/min and a co-flow of 42 lpm, the aerosol particle number density in the sample chamber is in the order of 107 particles/cm3. Such a high number density of nanometer-size particles (i.e., 10 to 100 nm) provides a uniform aerosol stream that is well suited for the ensemble averaging of multiple LIBS spectra at a constant overall mass concentration.

As a calibration source for LIBS it is important to determine the atomic emission intensity for a known analyte mass concentration. To ensure a true average analyte response, the analyte must be well dispersed and spatially averaged throughout the plasma for many laser shots. The current number density of 107 particles/cm3 corresponds to an average of more than 104 particles per plasma volume (plasma volume

-1.7 10-3 cm3 as reported in Chapter 6). Hence a true average analyte response is determined for LIBS calibration by ensemble averaging.

As an example, LIBS spectra were recorded for generated iron particles

corresponding to a concentration range from 1734 to 5780 lag/m3, and in addition to the nebulization of pure deionized water. From Figure 2.11, the 256.26-nm emission line was chosen and utilized for calibration. Two parameters are generally used in LIBS analysis, the peak-to-base ratio (P/B) and signal-to-noise ratio (SNR). A formal definition of these parameters is given by Equations 2.8, 2.9, 2.10, 2.11 and 2.6, and illustrated by Figure 2.13.

LIBS Signal = P/B = Integrated Peak Emission Intensity (2.8) Base Emision Intensity









Integrated Peak = P= i1 - d-(Bo .AAo) (2.9)


Base = Bo = Average(Bft, Bright) (2.10)


where Bi I - JI . d PA

SNR= (2.11) nns.A&A4

In the above relations, Im and Ia are the intensity of the continuum emission and atomic emission line defined by the equations 2.5 and 2.6 respectively, Xo is the wavelength of the atomic emission line, A2O is the width of the atomic emission line, Bi is the average intensity of the continuum emission at the left and right of the atomic emission over a AX region, and rms is the root-mean square of the shot noise at the off-peak baseline.

10000 . * . . . .


8000
Atomic emission line
or Peak
=6000




2000


0245 X 250 255 Wavelength, nm

Figure 2.13 Peak-to-base ratio and signal-to-noise ratio of the atomic emission line at 4o, where AXo is the width of the peak, rms is the root-mean square of the noise at the off-peak baseline, Bi is
the average continuum intensity.








The LIBS signal P/B is defined as the integrated peak intensity normalized by the continuum baseline intensity about the peak. The baseline intensity was estimated by interpolating the off-peak intensity values on each side of the analyte emission line. Similarly, the signal-to-noise ratio is defined as the average peak intensity normalized by the root-mean square of the noise at the continuum close to the position of the emission line. For the iron the calibration curve is presented in Figure 2.14, which is the relation P/B versus known iron concentration in the gaseous stream corresponding to the 256.26nm emission line. The P/B used in the calibration curve was maximized for an optimal time delay of 8 ps with an integration time of 5 pts. The resulting linear correlation based on a least-square fit output a 0.996 regression coefficient (R2). A similar linear calibration curve was produced for the 259.94-nm iron emission line, yielding a 0.991 regression coefficient.

10
....... ........... . =-0.13745 + 0.0015267x FR= 0.9962

8I i

C6
a.
C 4

2

0,
0 1000 2000 3000 4000 5000 6000
Fe Mass Concentration, pig/m3

Figure 2.14 Calibration curve for iron 256.26-nm emission peak signal using 8-pts delay and 5-ts time width. Error bars represent
� one standard deviation (R2 = 0.996).






47

To illustrate the process of P/B maximization, the carbon emission line C 1247.8 nm was used. The time-resolved process requires choosing a time delay that maximizes the signal P/B for fixed integration time and pulse energy. Figure 2.15 shows the variation of the P/B as a function of delay time using an integration time of 2 gis at pulse energy of 247 mJ. As it was noticed before, the atomic and the continuum emission have different rate of decay that are used advantageously by the definition of the LIBS signal. The optimization of the P/B favorably uses the maximization of the ratio of the emitted energy by the processes related to bound-bound transition and free-free and free-bound transition. This process of optimization was followed to build the calibration curve of all analytes used in this research. Figure 2.16 shows another calibration curve for an analyte of interest. Such calibration curve reaffirms the well-suited system developed for the analysis of aerosols using LIBS.

30 I I I


25

d 20
a.

15

C,,,
10 Delay Time


50
05 10 15 20 25
Time Delay, gts

Figure 2.15 Time delay optimization at fixed 2-ts time width for the maximization of the LIBS signal P/B using the carbon emission
line C I at 247.8 nm and at pulse energy of 247 mJ.











200-

150
a.
~100170
I)
50


0 1000 2000 3000 4000 5000 Mg Mass Concentration, Ag/m3

Figure 2.16 Calibration curve for magnesium 280.27-nm emission peak signal using a 40-g.s delay time and 40-ps time width. Error
bars represent � one standard deviation (R2 = 0.998).


2.6 Laser Induced Breakdown Threshold The focus of a pulsed laser beam onto a small spot may increase the temperature rapidly of a localized region to vaporize the material, and to dissociate the molecules, thereby forming an optically induced plasma. The Nd:YAG laser used to create the plasma for LIBS experiments of this research delivers a well-collimated Gaussian beam with a divergence less than 0.0045 rad, a pulse-to-pulse stability less than 4% rms, and an energy drift of less than 10%. The plasma will be initiated and formed if the laser power density exceeds the threshold value for the medium, which is the power density required to hold electrons in their energy levels and avoid cascade ionization. Therefore, pulse-topulse stability and matrix effects are significant at laser pulse energies near the breakdown threshold, noting the extreme case of non-breakdown.

For the determination of laser-induced breakdown thresholds, the experimental set-up described in Section 2.3 was used, and a conditional data analysis was applied to






49

calculate the percentage of laser shots generating a plasma, as based on the presence of

optical emission at a delay time of 1 pgs with respect to the incident laser pulse. A

breakdown is said to occur when the detected signal was 15% higher than the dark signal (no breakdown). Experiments were performed for different gas stream conditions, including pure air, pure nitrogen, air with the introduction of deionized water at a mole fraction of 0.003, air with the introduction of titanium-based particles at a mass concentration of 17 ppm (mass basis), and air with the introduction of sodium-based particles at a mass concentration of 17 ppm. The mean particle diameters were on the order of 25 nm with aerosol number densities on the order of 5x108 cm"3, and generated as described in Section 2.2. Figure 2.17 shows the result of these experiments.
| | I II I I I I I I I I * '* I~
100 ..................... ........ ...
* in Air
80 N in N2
o in Diwater
>1,
o + in Ti particles X in Na particles
~60

40 - ,
+ Breakdown
c 2 0 1P
UI
0 , , -= 1 , = , I , , I ! ,. ,* .. . . I
0 50 100 150 200 250 Pulse Energy, mJ

Figure 2.17 Percentage of laser pulses producing breakdown as a
function of laser pulse energy for various sample streams.


The statistical breakdown threshold is defined in the present study as the required

energy to produce a breakdown frequency of 50% (percent of laser pulses yielding breakdown). As observed in Figure 2.17, the threshold energy ranged from 155 mJ to








166 mJ over all gas stream conditions. The pure gas streams are characterized by an abrupt breakdown threshold, changing from 10% to 90% breakdown frequency in a change from 160 to 175 mJ per pulse, respectively. In contrast, the addition of solid particulates to the gas stream lowered the laser pulse energy required for breakdown, producing breakdown frequencies of about 10% at 70 mJ per pulse. The addition of water only mirrored the results of the solid particles, which is presumed due to the presence of submicron-sized water droplets acting as aerosols. The difference in the breakdown phenomena at low pulse energies between the purely gaseous and particleladen streams is related to the generation of seed electrons during plasma initiation. Multiphoton ionization requires a well-defined photon flux for breakdown of nitrogen or oxygen, hence the well-defined profiles in Figure 2.17. However, the presence of solid particles provides surface sites for multiphoton ionization, which though less photonintensive than gas-phase breakdown, is a much less predictable process due to the complicated photon-particle surface interactions.

For an estimated laser beam diameter of 100 gm at the focal point (measured at laser pulse that yields no breakdown), the 50% breakdown frequency corresponds to average power density of about 200 GW/cm2. For all experimental conditions in the present study, a breakdown frequency of 100% was realized at pulse energy of 190 mJ. At this pulse energy, the incident photon flux is sufficiently high to uncouple the breakdown process from local matrix effects (e.g., aerosol loading), resulting in a repeatable, hence robust, plasma process. All subsequent experiments were performed using pulse energies above 190 mJ.








2.7 Global Energy Balance of the Optical Breakdown

Not all the energy of a laser pulse is used to produce and sustain the optically

induced plasma. Using a Gaussian-temporal pulse-laser profile, it can be understood that the optical breakdown initiation would not start until the power density of the leading edge (of the laser pulse) reaches the required threshold of the medium. Therefore, before reaching this point, an amount of energy passes by the focal region without being absorbed. In addition, after the breakdown initiation and later development, the remaining laser pulse is partially absorbed by free electrons of the plasma. The amount of absorbed energy depends on the population of free electrons, which also depend on other factors such as laser wavelength, pulse duration, and size of the focal region. Figure 2.18 shows the transmitted energy through the optical breakdown in air as a function of the incident laser pulse energy.
4.00 I
400 ,1 _ 1 .... I .... .. 1 .... I .... . ..!' ,.
350 - No Breakdown : 100% Breakdown
*-4 i300
E 250 0 200
0
i 10 : u:





0 50 100 150 200 250 300 350 400 Energy IN, mJ

Figure 2.18 Transmitted energy thru the laser-induced plasma in air as a function of laser pulse energy. Error bars represent � one
standard deviation.








The average energy per pulse was recorded using a volume absorbing calorimeter (AC5001 model calibrated at 1064 rim), both before and after the laser-induced plasma. No reflected or scattered energy was detected from the testing chamber. As can be seen from Figure 2.18, the incident laser pulse energy is transmitted 100% in the no breakdown region since the laser pulse energy does not reach the power density threshold of the medium (e.g., purified ambient air). At 100% breakdown frequency, the transmitted energy thru the formed plasma has a minimum of about 100 mJ in the range of laser pulse of 230-250 mJ. At higher pulse energies, the transmitted energy increases slower than the supplied pulse energy indicating that even though more energy is going thru the plasma, it is possible that the plasma is absorbing more energy than that of the range 230-250 mJ.

2.8 Plasma Temperature
It has been stated in the Section 1.3 that laser-induced plasma (LIP) is in local

thermodynamic equilibrium (LTE) at least 1 lis after the onset of the laser pulse. In LIP, the Boltzmann and Saha relations describe the atomic populations at different energy levels and ionization states, while the Maxwell distribution governs the electron speed distribution (dominant particles in the kinetic energy of the plasma). The temperature parameterizes the Boltzmann, Saha, and Maxwell relations adopting the name of excitation, ionic, and electron temperature respectively. At LTE, all these temperatures would have the same value corresponding to the thermodynamic equilibrium temperature of the species. In the present research the term plasma temperature is used to engulf the meaning of these temperatures and is calculated using the Boltzmann plot that is based on the Einstein-Boltzmann relation (Equation 2.6), rewritten as follows









I)-L1 = - I ~'+ cols (2.12)
tA-g) T( k

The Boltzmann plot graphs basically Elk in the abscise and In(A1lAg) in the

ordinate to have the inverse of the slope of a linear fit as the temperature; noting the atom density and partition function are constant for transition at the same ionization state of a particular species. The number of emission lines should be at least three, corresponding to the same ionization level, with well-spread upper excitation energy. The intensity of the emission line need not be in absolute value, but must be relative line-to-line. To assess the plasma temperature, 10 iron atomic emission lines in a single spectral window centered at 270 nm were used to build Boltzmann plots. The selected spectral lines were all Fe II emission lines with relatively broad spread of upper energy states as is shown in Table 2.6.

Table 2.6 Fe II Atomic Emission Lines for Plasma Temperature Calculations

Wavelength A, 10e s- g Upper energy level E,
X, Urn cm j

258.5876 0.81 8 38660.043 7.69 10"9 259.9396 2.20 10 38458.981 7.65 101 9 261.1874 1.10 8 38660.043 7.69 10"'9 261.7618 0.44 6 38858,958 7.73 10"' 271.4413 0.55 6 44784.761 8.91 10"' 272.7539 0.85 4 45044.168 8.96 10'9 273.9548 1.90 8 44446.878 8.84 10.' 275.5737 2.10 10 44232.512 8.80 10"'9
277.93 0.76 8 62322.431 1.24 10-1"
278.3691 0.70 10 62083.108 1.23 10"'8
Source: http:/physics.nist.gov








2.8.1 Spectral Window Calibration

The use of Boltzmann plots to determine the plasma temperature requires the precise identification of the Fe II atomic emission lines in the selected 270-nm spectral window. The spectral window centered at 270 nm has a width of about 35 rum. The collected light is recorded by an iCCD detector array with 1024 pixel to resolve the spectral radiation and 256 binned rows to quantify the collected light. First it is needed to establish accurately the relation between pixel number and wavelength, and second a pixel-by-pixel intensity response is measured over a specified number of binned rows to assess the net intensity (accounting by losses through lenses, fiber optic, spectrometer, and intensifier).

For the wavelength calibration, a set of spectrally resolved emission lines was selected. Nebulization of aqueous solution of lead, titanium, and aluminum were performed to obtain aerosol concentrations of 5700 Jtg/m3, and LIBS spectra were collected at this condition. Lead at 261.417, 266.315,280.199, and 283.305 nm (Pb I), titanium at 284.194 nm (Ti 11), and aluminum at 256.798, 257.509, and 266.038 nm (Al I), and 281.618 rm (Al II) were visually identified in the spectral window at different delay times varying from 5 to 20 us. The pixel corresponding to the maximum intensity of the emission line was assigned to the wavelength of the corresponding emission line. The wavelength calibration curve for the 270-nm spectral window is presented in Figure 2.19. The wavelength response is highly linear over the pixels of the detector array with an adjusted correlation coefficient 2 = 0.99998.

For the pixel-by-pixel intensity response, 100-binned rows were selected to

integrate the light during the plasma temperature measurements. The spectral intensity








response in the 270-nm spectral window, considering all losses in signal, was correlated to a reference signal given by a calibrated quartz tungsten halogen lamp. The quartz halogen tungsten filament lamp simulates a blackbody radiation over a range 250-2400 nm; details of this lamp can be seen in Appendix B.


290 .
285 Y = 250.68 + 0.035858x R2= 0.99998

280
E
c 275 o 270 265
260 255
250
0 200 400 600 800 1000 Pixel Number

Figure 2.19 Correlation wavelength versus pixel number of the
iCCD array for the 270-nm spectral window.


The pixel-by-pixel intensity calibration procedure is as follows. The quartz

tungsten halogen lamp was located in the testing chamber of the LIBS experimental setup. The lamp was turned on and spectra were collected after a delay time of about 1 minute with an integration time of 100 ps. Figure 2.20 shows the collected spectrum by the LIBS system, and the actual spectral irradiance of the lamp given by the manufacturer. The spectral irradiance of the lamp has been scaled by a factor of 10000 to obtain an appropriated correction factor (scaled lamp irradiance over collected irradiance) that is also plotted in the Figure 2.21. The correction factor adjusts the collected spectra









being used to build the Boltzmann plots, thereby providing a relative measure of intensities for all atomic emission lines.

10000


8000 Lamp


6000
p

S4000

Collected
2000 Spectrum

0 1I ,J I ,,, .I .... I11 11 11 . ... I I
250 255 260 265 270 275 280 285
Wavelength, nm

Figure 2.20 Reference lamp irradiance and collected irradiance by
the LIBS system in the spectral window 270 nm.


nI


I I ~ ~ ~ I I I I


. . . . I . . . . I


I I I I


255 260 265 270 275 280 285
Wavelength, nm


Figure 2.21 Correction factor to provide relative intensities lineby-line in the spectral window 270 nm.


A
250









2.8.2 Plasma Temperature Decay

The evolution of the laser-induced plasma can be quantified by the decay of its

temperature. The calculation of the plasma temperature requires the construction of

Boltzmann plots. To calculate the plasma temperature, LIBS spectra corresponding to

iron-based aerosols with mass concentration of 5700 gg/m3 were used. Spectra

corresponding to 1200-shot average were collected at 5-, 10-, 15-, and 20-ps delay time

with an integration of time of 5 gs for a laser pulse of 300 mJ. The temperature decay of

the plasma was monitored in air and in nitrogen atmospheres. Figure 2.22 shows a

typical spectrum collected in air after its intensity having been corrected by the factor

discussed in the previous section. The integrated emission lines (peaks) are obtained

according to the equation 2.9 and used as IA in the Equation 2.12.

6000 ,.
259.94 nm Fe II emission lines
1200-shot average spectrum
5000 for 300 mJlpulse
261.19 nm 275.57 nm 4000 /
E [ 261.76 nm
23000 273.95 nm E a) CV N 272.75 nm 0! ri
V.- CO
C4 C4
271.44 nm
1000

0
255 260 265 270 275 280
Wavelength, nm

Figure 2.22 Representative 1200-shot average spectrum in air obtained using a laser pulse of 300 mJ at 5-ms delay time and 5-ms
integration time after intensity correction.








A Boltzmann plot corresponding to a collected spectrum at a delay time of 10 jts with an integration time of 5 lis is displayed in Figure 2.23. The major component of uncertainty in the temperature calculated by this method comes from the uncertainty of the values of excitation energies. Boltzmann plots could report uncertainties in the temperature as high as 20 % (Griem 1997), however, the advantage of using widespread upper excitation energies leads to a better accuracy in the temperature determination. A plasma temperature of 11500 K was calculated from the plot corresponding to a laserinduced plasma created in ambient air. Finally, the evolution of the plasma temperature in air and nitrogen atmosphere is presented in Figure 2.24. The plasma temperature profile is basically indistinguishable between air and nitrogen. The error bars showed the precision of the calculated values and correspond to one standard deviation.
"3 . . . . I= 1 . .I . . . . I . . . I '
y 0.69005 - 8.693e-05x IF- 0.98549
-4




-J


-7


50000 60000 70000 80000 90000
E/k, K

Figure 2.23 Boltzmann plot using Fe II emission lines for energy
pulse of 300 mJ at a delay time of 10 ps and width time of 5 pts.








14000 * . * . * . .

13000 [ I Nitrogen EE
I12000i
200

1 1000
E
10000
E
9000

8000

7000 I . a . . . I . . . . I . . . . I .
0 5 10 15 20 25 30 Delay Time,lis

Figure 2.24 Plasma temperature decay in air and nitrogen atmosphere using pulse energy of 300 mJ an integration time of 5
pts. Error bar = + one standard deviation.


2.9 Summary

In conclusion, an aerosol generation system was implemented that enables the production of precise mass flow streams of well-characterized, submicron-sized aerosol particles. As a calibration source for laser-induced breakdown spectroscopy, linear calibration curves were produced for iron and other elements of interest over mass concentrations ranging from 0 to 17700 p.g/m3. The aerosol generation system is suitable to the production of multi-species aerosols via the nebulization of multi-species aqueous solutions. The gaseous co-flow provides additional flexibility to the aerosol generation system by enabling variations in the co-flow gas independent of the nebulizer operation. The laser-induced plasma produced with the present system is characterized by a breakdown frequency of 100 % at pulse energies greater than 190 mJ, the energy needed to obtain repeatable plasma every laser shot. In addition, the minimum laser pulse energy transmitted thru the plasma is about 225 mJ per pulse. At typical L1IBS time scales, the





60

plasma is still extremely hot (about 11000 K). The measured temperature profiles give an insight as to the plasma energetic state after the onset of the laser pulse, which is a useful reference in the following chapters. Overall, the system is well suited for the development, assessment, and calibration of in situ laser-based diagnostic techniques such as laser-induced breakdown spectroscopy. In the following chapter, the present system is tested to monitor ambient air in real time, especially for the sizing of single particles.













CHAPTER 3
FEASIBILITY OF AEROSOL DETECTION IN AMBIENT AIR

The implemented LIBS system described in the previous chapter was tested for the detection of aerosols in ambient air, including quantitative mass concentration measurements and size/composition measurements of individual aerosol particles in the present chapter. Ambient air was monitored on the University of Florida campus between Monday, June 28, 2000 and Friday, July 7, 2000, and again between Tuesday, August 1, 2000 and Friday, August 5, 2000. These sample windows were selected to overlap with the Fourth of July holiday period, a holiday associated with the use of fireworks. Specifically, there were at least 5 municipal fireworks displays within a 50mile radius of the University of Florida between July 3 and July 4 (Tuesday). LIBS data were collected typically twice each day, generally in the morning hours and in the afternoon/evening hours, for four elements of interest: aluminum, calcium, magnesium, and sodium. For each sampling period, spectral data were collected for approximately two hours. During the course of experiments, daily high temperatures ranged from 28 to 340C, overnight low temperatures ranged from 18 to 230C, and daily wind speeds ranged from 0 to 13 mph, with winds primarily to the north direction. Rain was periodic (maximum of 2.4 cm on June 29) during the nearly three weeks of sampling; however, no data were recorded during periods of rain. In the following sections, the aerosol collection system, the methodology of analysis, and the finding of this experiment are presented.








3.1 Ambient Aerosol Sampling System An ambient air sampling system was designed to bring a well-controlled flow of ambient air to the prototype LIBS instrument. A US EPA standard PM10 inlet was utilized at the air inlet of the sample line. A PM 10 inlet removes all particles greater than 10 microns in diameter via inertial impaction, while also ensuring uniform sampling of air. The air was drawn into the PM10 inlet at a volumetric flow rate of 16.7 liters per minute (-1 m3/hr), and subsequently transferred to the LIBS sample chamber using 12mm inner diameter stainless steel tubing. The sampling inlet was located on the roof of a single story building (about 4.6 m high) adjacent to the second story laboratory housing the LIBS system. The inlet was positioned approximately 1 m from the edge of the roof and approximately 3 m from the laboratory. The inlet was positioned to be as free as possible from any obstructions; however, no detailed analysis of building interactions with ambient air sampling was included for this study. A schematic of the sampling system is shown in Figure 3.1.


Figure 3.1 Schematic of the ambient air sampling system.


Vacuum Pump


- 8.2 m of 5/8"O.D. stainless steel tubing
- Four 900 elbows, 15 cm radius
- Five 350 elbows, 20 cm radius








A stainless steel transfer line of about 8.2-m long was connected directly to the inlet of the sample chamber, where the airflow subsequently entered the 6-way stainlesssteel vacuum cross. A vacuum pump was connected to the outlet of the sample chamber, and was used to maintain the overall sampling rate. The pressure in the sample chamber was maintained about 1 torr below atmospheric pressure. An aerosol deposition program (McFarland 1996) was used to calculate the particle transfer efficiency throughout the entire sampling system. The transport efficiency from the sampling inlet to the LIBS sample chamber as a function of the particle size is presented in Figure 3.2. The efficiencies ranged from 92 to 99 percent for particulate diameters between 0.1 and 2.5 microns. Essentially all particle diameters reported in this study ranged between 0.2 and

1.6 microns, with corresponding transport efficiency between 96 and 99 percent. In consideration of these calculations, no corrections were made to particle size distributions or particle mass concentrations based on the aerosol transport efficiency.

100 . . . I ' " ' ' I ' ' ' I . . ' '


98



Ui
CL 0 C
~92

90 0 , * , * , , , , � . . . I .. . a . . . . a a
0 0.5 1 1.5 2 2.5 Particle Diameter, Im

Figure 3.2 Transport efficiency of ambient air particles from the inlet to the LIBS sample chamber as a function of the particle
diameter.








3.2 LIBS Data Analysis

3.2.1 Collection of Data

Four elemental species were targeted for the ambient air monitoring namely

aluminum, magnesium, calcium, and sodium. The corresponding atomic emission lines utilized were the 394.40 and 396.15-nm Al I lines, the 279.55, 280.27, and 285.21-nm Mg I and II lines, the 393.37 and 396.85-nm Ca H lines, and the 589.00 and 589.59-nm Na I doublet. Three different spectral windows were utilized to monitor these species. They were centered at 275, 408, and 590 rm, with each having a spectral bandwidth of approximately 36, 32, and 25 nm, respectively. Aluminum and magnesium were selected because they are used as energetic fuels in many fireworks, while sodium and calcium were selected primarily as controls due to their general prevalence in atmospheric particulates.

Ambient air mass concentration measurements of the targeted elements were recorded using a conditional data analysis scheme. During each nominally two-hour sampling period, spectra were recorded using 1200-shot pulse sequences, which correspond to a 4-minute sample for the 5-Hz laser repetition rate. For each spectral window, between 8 and 14 1200-shot sequences were recorded. Spectra that contained significant emission intensity for a target analyte species were identified as particle hits. For each LIBS spectrum, the emission intensity about the expected analyte peak was compared to the emission intensity of an adjacent, featureless continuum spectral region. The expected analyte peak was triggered on the target emission line, which was centered averaged over 9 pixels of the iCCD detector array, and the featureless continuum intensity was the average of continuum intensities called base 1 and base 2, which were averaged over 15 pixels of the iCCD detector array. The spectrum was classified as a








particle hit if the ratio of emission intensities exceeded a threshold value, typically 75 to 150% above the nominal ratio corresponding to the absence of any analyte emission line intensity. These threshold values set the false hit rate to no more than one in a 1200-shot sequence. The designed conditional data analysis scheme was performed in real-time for each 1200-shot sequence, and the identified spectra for all particle hits were stored along with the ensemble-average of all laser pulses. Parameters used in the process of particle detection are listed in Table 3.1.

Table 3.1 Parameters for Aerosol Particle Detection


Analyte Aluminum, Magnesium, Calcium, Sodium, Al Mg Ca Na Emission Al 1-394.40 Mg 1-279.55 Ca 11-393.37 Na 1-589.0 (target), Al I- (target), Mg I (target), Ca II- (target), Na ILines, nm 396.15 (control) 280.27, Mg II- 396.85 (control) 589.59 (control) 285.21 (control)

Spectral 408 nm, -32 nm 275 nm, -36 nm 408 ram, -32 nm 590 nm, -25 nm Window bandwidth bandwidth bandwidth bandwidth Delay/Width 40 pis/ 60 lis 40 gs/ 40 gs 40 ps/ 60 gs 30 ps/150 gs Base I & 2 400 & 416 nm 270 & 284 nm 400 & 416 nm 584 & 594 nm LaserPulse 315mJ 315mJ 315mJ 315mJ
Energy

Hit 150% above no 85% above no 100% above no 75% above no Threshold hit case value hit case value hit case value hit case value


3.2.2 Processing of Data

After each sampling period, the stored spectra of all identified particle hits for a given analyte species were ensemble-averaged and an equivalent mass concentration was calculated using calibration curves. Typical calibration curves used in this experiment








are presented in Figure 3.3 and a summary of the correlations of mass concentration versus peak-to-base is listed in Table 3.2.

2000 Aluminum


.2
.:1000

0


0

0 20 40 60 80 100 120 140 160 LIBS Signal P/B, a.u.

Figure 3.3 Aluminum and magnesium calibration curves using the
emission lines A 11394.4 nm and Mg 1279.55 nm.


Table 3.2 Equations for Calibration of Mass Concentration

Analyte Mass Concentration, tg/m3

Target 279.55 rum C 0.0447(P/B)2 + 4.1575(P/B) - 0.2765 Mg Control 280.27 nm C = 0.0684(P/B)2 + 10.813(P/B) + 8.0581

Control 285.21 rum C = 0.0599(P/B)2 + 1.614(P/B) + 0.1284 Al Target 394.4 nm C 0.01837(P/B)2 + 10.6383(P/B) - 17.0215 Control 396.15 nm C = 0.00392(P/B)2 + 5.69876(P/B) - 0.6707 Ca Target 393.37 nm C = 0.0044(P/B)2 + 0.21186(P/B) + 1.2966 Control 396.85 rum C = 0.00028(P/B)2 + 0.45545(P/B) + 0.50404 Na Target 589.0 nm C = 0.0000563(P/B)2 + 0.37805(P/B) - 1.5396
Control 589.59 rum C = 0.0000910(P/B) + 0.7528(P/B) -0.8473








The equivalent mass concentration corresponds to the analyte signal strength as

calculated using the subset of data identified with the conditional sampling algorithm. As such, the actual mass concentration of the ambient air sample is equal to the product of the equivalent mass concentration and the frequency of identified hits, as expressed by the Equation 1.3 and described in an earlier publication (Hahn et al. 1997). The frequency of hits is the number of spectra identified with the target emission line divided by the total number of laser pulses. It is noted that as the frequency of hits approaches 100%, the conditional analysis scheme converges to the traditional ensemble-averaging approach.

A short description of the quantitative analysis of LIBS data for particle size and composition measurements is given in the Section 1.6 of this dissertation. The equivalent mass concentration is calculated using a typical calibration curve relating the analyte LIBS signal to a known analyte mass concentration. All calibration curves used in this study were generated with well-controlled mass flow streams as discussed in the previous chapter. The LIBS signal is also defined by Equation 2.8. As explained above, the equivalent mass concentration is equal to the single particle mass divided by the plasma volume. Using this relationship, the equivalent spherical diameter of a single particle may be calculated by the Equation 1.7, which is repeated here D= k-- 1 , (1.7) L rp f

where the characteristic plasma volume Vp is estimated to be 2.38 x 10-3 cm3 as it is shown later in Chapter 6.

The algorithm discussed above enables the identification of spectra corresponding to individual particles, and the calculation of overall mass concentrations and the









mass/size of individual particles. Two additional controls were added to the data reduction schemes. For mass concentration measurements, a total of two or more emission lines were utilized for each species. Typically, the most intense line (target emission line) was used to trigger the conditional data analysis routine. Then one or two additional lines (control emission lines) were used to calculate the equivalent mass concentration based on the ensemble-averaged spectrum of the identified particle hits. Because the spectral noise is random on a given shot, a high noise spike on the analyte trigger wavelength would not correspond to high noise spikes on the alternative control analysis wavelength or wavelengths. For single-shot analysis, comparing the analyte signal from two different atomic emission lines screens false signals or other spectral irregularities. A spectrum is rejected from subsequent single-shot size analysis if both atomic emission lines (target and control lines) do not yield the same analyte mass to within a factor of two. In other words, spectra are retained for size analysis only if the ratio of analyte masses calculated from two different emission lines is between 0.5 and 2.

3.3 Results and Discussion

An advantage of LIBS-based analysis of individual aerosol particles is the ability to determine the composition of constituent elements within a given particle. Several unique particle types were identified, as shown in the following two figures. Figure 3.4 contains two single-shot spectra. One spectrum is characterized by the presence of intense magnesium emission lines only, and the other one features both magnesium and silicon (288.16-nm Si I) emission lines. As discussed below, while the Mg-only spectrum is attributed to a single MgO particle, the second spectrum is recognized as a type of magnesium-silicate particle (i.e., composed MgSiO3 or an agglomerate of MgO with a silicate particle). Such magnesium-silicate spectra were rare in comparison to the









magnesium oxide spectra, with the latter accounting for greater than 95% of the recorded magnesium-containing spectra. Figure 3.5 contains two spectra corresponding to aluminum-containing particles, namely one spectrum containing only aluminum emission lines and a second spectrum characterized by the presence of both aluminum and calcium emission lines. Spectra characterized by the presence of both aluminum and calcium emission were less than 1% of the spectra containing calcium emission lines, and were about 10% of the spectra containing aluminum emission lines. Particle size data are presented in detail below, however, the four spectra presented in Figures 3.4 and 3.5 correspond to particle diameters in the range of 0.4 to 1.0 pn. The above data demonstrate the overall sensitivity of the LLBS technique for single particle analysis, as observed by the relative strength of the various atomic emission lines.

3000 oil Mg
2500

2000 Mg
.1500- Si
S 100
50




0
276 278 280 282 284 286 288 290
Wavelength, nm

Figure 3.4 Two single-shot LIBS spectra. The lower spectrum represents a single MgO particle. The upper spectrum that has been vertically shifted for clarity represents a single Mg-Sicontaining particle.










8000 Ca
Ca

6000


4000 A)
C


2000


392 393 394 395 396 397 398 399 400
Wavelength, nm

Figure 3.5 Two single-shot LIBS spectra. The lower spectrum represents a single AlO particle. The upper spectrum that has been vertically shifted for clarity represents a Ca-Al-containing particle.


3.3.1 Mass Concentration

A novel feature of laser-induced breakdown spectroscopy is the ability to reject null spectral data while utilizing the infrequent but signal-rich spectra corresponding to discrete aerosol particles. The enhanced signal-to-noise ratio (SNR) resulting from the conditional analysis-based LIBS monitoring is illustrated in Figure 3.6 for sodium-based particle sampling. The figure presents the spectrum corresponding to the ensemble average of 9600 laser pulses, along with the corresponding spectrum based on the ensemble average of 30 identified particle hits only. The corresponding sodium-based particle sampling frequency is 30/9600 or about 0.31% of laser pulses. The increase in analyte SNR is very significant, with the 9600-shot ensemble-average essentially equal to a non-detectable sodium emission signal. The equivalent mass concentration of the sodium hits spectrum is about 139 parts per billion (ppb) (mass sodium/mass air), while the overall sodium concentration is 139*(0.31/100) or about 0.43 ppb. The nominal









improvement in SNR is about two orders of magnitude for this sodium sampling frequency, which is apparent when comparing spectra in the Figure 3.6.

1400 1 I,* I I

1200 Na

1000

w 800 Na

c 600 Avg. Na-based Hits
C
400 Avg. All Shots

200
0 I I , , I , , , I , , I , , , I 586 588 590 592 594
Wavelength, nm

Figure 3.6 The lower Na-LIBS spectrum is the 9600-laser-shot average, and the upper Na-LIBS spectrum is the 30-identified, Nashot average from the 9600 laser shots. Both spectra have the
same intensity scale but have been vertically shifted for clarity.


Magnesium-containing emission spectra revealed similar enhancement in

magnesium emission line intensity with the use of conditional data analysis, as it is also seen in Figure 3.7. Specifically, Figure 3.7 reports the 14-hit averaged spectrum corresponding to an overall magnesium mass concentration of 108 ppt. As illustrated above, the use of a conditional data analysis routine yields spectra with excellent signalto-noise ratios for the observed ambient air particle sampling rates. The spectral data were subsequently used to calculate overall mass concentration measurements of aluminum, calcium, magnesium, and sodium throughout the sample period. The mass concentration of magnesium recorded in ambient air is presented in Figure 3.8 as a function of time.









1000 800


200


274 276 278 280 282 284 286 288 290
Wavelength, nm

Figure 3.7 Enhancing of the Mg-LIBS signal using the conditional analysis. The 14-Mg hits are a subset of the 9600 laser shots. Both spectra have the same intensity scale but have been vertically shifted for clarity.


140

120

5. 100
0.
C
.o 80

C
� 60
C
0
(o 40


6 6/2


.8


7/3 7/9 7/15 7/21 7/26 8/1


Date (Month/Day) - 2000

Figure 3.8 Mass concentration of magnesium as a function of time. Each data point represents the average LIBS-based concentration over a two-hour sampling period.


., , , ,,ri , , i ,, , , , , ,,,r, , . Holidays
-0


" O



- 0



- � o _ _ . . ,-, ; ., S, ; 1 .,,,r; ; , . , . ,,h -








The data presented in Figure 3.8 reveal a significant rise in magnesium during the Fourth of July holiday period, considered from July 1 to July 7. The magnesium mass concentration ranged from 0 to 108 parts per trillion (ppt) on a mass basis. It is noted that

1 part per trillion on a mass basis is equivalent to 1.16 ng/m3 of ambient air. A mass concentration of zero corresponds to a sampling frequency of zero or an equivalent mass concentration (based on hit spectrum) of less than 10 ppb, the lower detection limit established for magnesium spectra corresponding to the particle hits. The magnesiumbased particle sampling rates were typically 0.1% or less; hence the actual analyte signals corresponding to the Figure 3.8 data were approximately 3 orders of magnitude larger than the overall mass concentration values. The average magnesium concentration during the holiday period was 36.7 ppt, while the average concentrations of magnesium before and after the holiday period were zero and 3.5 ppt, respectively. The nearly 50fold increase in magnesium during the Fourth of July holiday period in comparison to the average of pre- and post-holiday concentrations is strongly suggestive that the source of increased magnesium is derived from the discharge of fireworks in the troposphere. The mass concentration data recorded for aluminum during the same sampling period revealed trends similar to the magnesium data. However, the overall aluminum-sampling hit rates were significantly lower than the magnesium sample rate. As such, the aluminum data contained a larger percentage of zero concentration measurement, essentially non-detects. The average aluminum concentration was 45.8 ppt during the holiday period, compared to an average mass concentration of 6.9 ppt for the non-holiday period.








The current findings of increased magnesium and aluminum are consistent with

results reported during a similar study that utilized laser desorption/mass spectrometry for the analysis of ambient-air particles (Liu et al. 1997). The study reported significant increases in the levels of ambient air magnesium and other elements attributed to fireworks in the days following the Fourth of July holiday at the University of California in Riverside. Although the total masses of magnesium and aluminum released by fireworks are diluted with a significant volume of ambient air, the current study and the work of Liu et al. suggest that pyrotechnic-derived particulates persist in the troposphere with a time scale on the order of days.

For the analytes calcium and sodium, Figure 3.9 shows the analyte mass

concentration response during the testing period. In contrast, the mass concentrations of calcium and sodium do not correlate as strongly as the magnesium and aluminum data do for the Fourth of July holiday period. Their concentrations are nearly double during the holiday period. The average calcium and sodium concentrations during the holiday period were 0.21 and 0.65 parts per billion (ppb) by mass respectively, while the pre- and post-holiday average mass concentrations were 0.14 and 0.25 ppb for calcium and sodium respectively. It is also difficult to determine the contribution of fireworksderived calcium and sodium to the recorded mass concentrations since the presence of these species in ambient air under normal conditions and their relatively large daily level fluctuations are well known (Laj et al. 1997, and Lee et al. 1999). In addition, for example, the standard deviations of the measured sodium mass concentration data are equal to 0.43 and 0.33 ppb, corresponding to the holiday and non-holiday sampling periods, respectively. The difference between holiday and non-holiday sodium








concentrations (about 0.4 ppb) is within the standard deviations for these two sample sets, making it difficult to conclude any statistical significance to the fluctuations in sodium and calcium over the Fourth of July holiday. Therefore, this increased concentration during the holiday period could be due to a normal fluctuation of the species.

2.5 ..... i i... . I ..... is.. ..i. .
Holiday 0 Ca, ppb 2 - o Na, ppb
0.
C.
.o 1.5
e
JET

0.5
] 08
0 -.

6/28 7/3 7/9 7/15 7/21 7/26 8/1 8/'7 Date (Month/Day) - 2000

Figure 3.9 Mass concentrations of calcium and sodium as a function of time. Each data point represents the average LlIBSbased concentration over a two-hour sampling period.


A summary of the mass concentrations for the analyte studies is reported in Table 3.3. No independent sampling analysis techniques were used to corroborate the reported mass concentration. However, in an early study, independent extractive sampling in accordance with US EPA Method 29 standard yielded excellent agreement for chromium and manganese concentration at 2-3 ppb mass concentrations (Haim et al. 1997). In aggregate, the current mass concentration data demonstrate the ability of the LlIBS technique to measure element-specific concentrations of particulate matter at very low overall mass concentration levels.








Table 3.3 Summary of Analyte Mass Concentration


Analyte Magnesium Aluminum Calcium Sodium
Detection Limit
(tectrum) 10 ppb 7.5 ppb 2.3 ppb 3.2 ppb
(on hit spectrum)

Holiday Mass 36.7 ppt 45.8 ppt 206 ppt 650 ppt Concentration* � 34 ppt � 126 ppt � 239 ppt � 425 ppt Non-Holiday Mass 2.8 ppt 6.9 ppt 140 ppt 254 ppt Concentration* � 5.8 ppt � 22.3 ppt � 127 ppt � 333 ppt

Maximum 108 ppt, 420 ppt, 755 ppt, 1.7 ppb,
concentration July 7 - PM July 7 - PM July 5 - AM July 2 - PM
* Average values � one standard deviation



While the mass concentration data presented above are based on approximately 2hour sampling period, it is worthwhile to examine the data with finer temporal resolution. The sampling frequencies of recorded sodium- and calcium-based particle hits are presented in Figure 3.10. The almost 2-hour sampling period monitored individual 1200shot laser sequences using each sequence a time of 4 minutes. The 4-min sampling frequencies ranged within a factor of two and three with respect to the average sampling rate for sodium-based particles and calcium-based particles respectively. Specifically, in the case of sodium-based particles the frequency varied from 0.25 to 0.92 % when the average during this 2-hour period was 0.49 %, while in the case of calcium-based particles the frequency ranged from 0.1 to 2.4% when the average was 0.7%. The time resolved sampling data presented in Figure 3.10 are typical of the relatively steady nature of particulate matter observed on the time scale of minutes to several hours.









2.5
2 1.5
1
0.5
0 10:3
3 2.5
2 1.5
1
0.5
0


13:20 11:06:40 11:40:00 12:13:20 12:4


-- Calcium August 1

* Avg. - 0.7 %
--------------------------.7.--------II I 1511 Ii------------I


13:20:00 13:53:20 14:26:40 15:00:00 15:33:20 Time

Figure 3.10 Sampling frequency over a period of two hours. Each data point corresponds to the frequency of hits for 1200-shot laser
sequence.


3.3.2 Particle Analysis

In addition to the evaluation of mass concentrations, the spectra corresponding to individual hits were analyzed for subsequent particle size. As outlined in detail previously, the elemental mass is directly calculated from the product of the equivalent analyte mass concentration of a single-shot spectrum and the characteristic plasma volume, 2.38xl 0-3 cm3. As observed in Equation 1.7, the calculation of a mass-based particle diameter requires the specification of parameters based on the particle type, namely the bulk particle density and the elemental mass fraction of the measured analyte species.

For the present study, the magnesium-based particles were modeled as

magnesium oxide, MgO, with a density of 3.58 g/cm3 and a magnesium mass fraction of

0.60. MgO was utilized as a model based on several observations. First, the perceived


Sodium July 4


Avg. - 0.5 %
S. . . .- - - - - - - - - - - - I . . I . . I . . . I . . . I








particle source of magnesium was primarily via combustion-generation during the discharge of fireworks. Second, nearly all of the magnesium particles were characterized by the absence of any recorded silicon or iron atomic emission lines. Third, modeling the particles as sea salt-based particles (Na/Mg is about 7.8 as reported by Mclnnes et al. 1999) was discounted for the magnesium-based particles due to the mass values of magnesium and sodium recorded. Specifically, the average mass of particulate magnesium was 232 fg, while the average mass of particulate sodium was 266 fg, with a corresponding Na/Mg ratio of only 1.15. Although the largest recorded sodium particle mass is consistent with the nearly 8 to 1 Na/Mg ratio, the recorded sampling frequency of the largest sodium particles is inconsistent with the recorded sampling frequency of magnesium particles.

The calcium particles were modeled as calcium carbonate, CaCO3, with a density of 2.71 g/cm3 and a calcium mass fraction of 0.40. Calcium in ambient air particulates is mostly composed of calcium carbonate or gypsum (Laj et al. 1997), although CaSO4 particulates from anthropogenic sources may be significant in number (Hoornaert et al. 1996). The sodium particles were modeled as two different particle types, namely sodium chloride, with a density of 2.17 g/cm3 and a sodium mass fraction of 0.39, for particles with a corresponding mass-based diameter less than 1.6 gm, and sodium nitrate for particles greater than 1.6 pm. Such a categorization is consistent with recent measurements reported of nitrate-containing particles using an aerosol time-of-flight spectrometer (Liu et al. 2000). Specifically, sodium nitrate particles were limited to the course size mode between 1.6 and 3.5 pm.








3.3.3 Particle Size Distribution

The histogram of calculated diameters for magnesium-contained particles

recorded during the Fourth of July holiday period is presented in Figure 3.11. This figure displays a particle distribution with mean diameter, standard deviation and modal diameter of 586, 96, and 515 nm, respectively. The size distribution exhibits skewness toward larger particle sizes, as expected for general aerosol populations. The mean particle size of magnesium-based particles recorded after the holiday period was 418 nm, with a standard deviation of 159 rn. The approximately 170 nm difference in mean diameters, about a 29% decrease, between the sizes of these two magnesium-based particle populations is not statistically significant, but nonetheless may be indicative of two different sources of magnesium particles, namely fireworks derived and perhaps marine-derived.


41-MgO Particles Mean = 586 nm St Dev = 196 nm Modal =515 nm









0 200 400 600 800 1000 1200 1d


400


Particle Diameter, nm

Figure 3.11 Histogram of calculated diameters for magnesiumcontaining particles for the Fourth of July holiday period. The particles were modeled as magnesium oxide (MgO).






80

In contrast, the calcium and sodium particle size distributions of the non-holiday period revealed essentially no variations with respect to the holiday period. The size distribution for calcium-contained particle (as CaCO3) is presented in Figure 3.12 and 3.13 for the holiday and non-holiday period, respectively. For the holiday period, the mean particle diameter is 685 nm with a standard deviation of 382 nm and a modal diameter of approximately 427 nm. A long tail is extended toward larger diameters, reaching a maximum diameter of 2.4 Km (not showed in the graph) with an equivalent mass of about 175 times that of the modal particle diameter. In the case of non-holiday particle distribution, the mean, the standard deviation, and the modal were basically the same with a variation within 5 % of the particle values recorded during the holiday period.


155-CaCO3 Particles Mean = 685 nm St Dev =381.7 nm Modal = 427 nm 15
C
0
10


5


00 500 1000 1500 2000
Particle Diameter, nm

Figure 3.12 Histogram of calculated diameters for calciumcontaining particles for the Fourth of July holiday period. The particles were modeled as calcium carbonate (CaCO3).






81

20
108-CaCO3 Particles
Mean = 705.7 nm
15 St Dev = 362.6 nm Modal = 460 nm


0 10


5


0 500 1000 1500 2000 Particle Diameter, nm

Figure 3.13 Histogram of calculated diameters for calciumcontaining particles for the non-holiday period. The particles were
modeled as calcium carbonate (CaCO3).


The histogram of sodium particles as NaCl is presented in Figure 3.14 for the holiday period. The mean and modal particle diameters are approximately 844 m and 698 nm respectively, with a standard deviation equal to 274 rum. As with the calcium results, the non-holiday associated sodium particles were consistent with the size distributions recorded during the holiday period. Specifically, the mean and modal diameter changed within 5 % and the standard deviation varied about 15 % respect to the particle values corresponding to the holiday period. As discussed above, sodium-based particles larger than 1.6 jim were modeled as sodium nitrate. The sodium nitrate particles accounted for less than 5 % of the sodium-based particle hits recorded, and were characterized by an equivalent size that ranged between 1.6 and 4.1 jim. A summary of the finding in this section is listed in Table 3.4.









339-NaCI Particles
Mean = 844.2 nm St Dev = 273.9 nm Modal = 698 nm


0 500 1000 1500 2000 2500
Particle Diameter, nm

Figure 3.14 Histogram of calculated diameters for sodiumcontaining particles for the Fourth of July holiday period. The
particles were modeled as sodium chloride (NaCl).


Table 3.4 Summary of Particle Size

Analyte Magnesium Calcium Sodium

Particle Model MgO CaCO3 NaCl Holiday Period: 41 155 339
Number of Particles 586 m �196 nm 685 nm �382 nim 844 nm � 274 nm
Mean � 1 StDev
Modal 515 nm 427 nm 698 nm Non-Holiday Period: 17 108 206 NumberofParticles 418 nm � 159 rm 706 nm � 363 nm 822 nm � 229 nm
Mean � 1 StDev
Modal 460 nm 720 nm Minimum Diameter 219 nm 196 rim 412 nm


It is also useful to discuss the absolute detectable analyte masses and the relative sensitivity of the LIBS technique for the species analyzed. The smallest diameters reported in Table 3.4 represent the minimum detected particles with mass of magnesium,








calcium, and sodium of 12, 4, and 31 femtograms, respectively. Because the probability is low that the actual ambient air size distributions shown in Figure 3.11 to 3.14 naturally ended at these points, these minimum detected sizes are considered representative of the analyte mass detection limits. It is noted that particle sizes on the order of tens of nanometers may exist for select particle types (i.e., metallurgical fumes and combustion derived). At present these nanoparticles were not detected using LIBS as implemented here. An additional parameter that should be noted is the rate of acceptance of the singleshot spectra. As detailed above, the criteria for analysis of a single-shot spectrum was that the calculated analyte mass was consistent (within a factor of two) for two atomic emission lines. The retention rate for the spectra of sodium-based particles was 90%, and the retention rate was 63% and 50% for the spectra of calcium and magnesium-based particles, respectively. These rates reflect a number of factors, including the overall LIBS sensitivity to each analyte emission line, the signal-to-noise ratio in the various spectral regions, and the specific nature of the emission line pairs (e.g., neutral line to ionic line as with magnesium).

3.4 Summary

The LIBS technique was successfully used to monitor ambient air particulate containing a number of species for a six-week sampling period spanning the Fourth of July 2000 holiday period. The implemented conditional analysis functioned to increase the sensitivity in the determination of overall mass concentrations up to tens of parts per trillion, and to identify spectra corresponding to individual aerosol particles with sizes of hundreds of nanometers. Changes in mass concentration for metallic species associated with the discharge of fireworks of the Fourth of July holiday period, such as magnesium and aluminum, were significant and increased by about one order of magnitude with








respect to the non-holiday period. In contrast, recorded ambient air concentrations of sodium and calcium revealed no significant correlation. Analysis of single particles yielded composition-based aerosol size distribution with diameters from 200 nm to 4 jin. The absolute mass detection limits for single particle analysis was in the order of tens of femtograms for magnesium-, calcium-, and sodium-based particles. Overall, LIBS-based analysis of ambient air aerosols is a promising technique for the challenging issues associated with real-time collection and analysis of ambient air particulate matter data. However, it is still needed to understand issues regarding to plasma-particle interactions for the development of LIBS as a deploying technique. The next chapter addresses the issue of single-shot variations in atomic emission spectra in consideration with plasma properties.













CHAPTER 4
SAMPLING STATISTICS FOR SINGLE SHOT ANALYSIS

Most of the research addressing LIBS sensitivity and precision is based on

traditional ensemble averaging of spectra (including ensemble of filtered spectra) as a means to overcome the extensive spectral fluctuations observed on a laser shot-to-shot basis. However, it is not readily apparent as the applicability of these studies to singleshot LIBS analysis. New questions regarding shot-to-shot variability and precision are raised with the specific application to single-shot aerosol analysis. This section is focused on the precision of LIBS-based single-shot analysis of gaseous and aerosol systems due to different laser pulse energies. A statistical analysis of fluctuations of single-shot spectral data in both atomic emission and plasma continuum emission were investigated together for a homogenous gaseous flow. The ubiquitous nature of carbon dioxide in ambient air provided the analyte source for this homogeneous gas flow. In addition, shot-to-shot fluctuations in plasma temperature are reported based on iron atomic emission in an aerosol-seeded flow, which were analyzed at optimal temporal delay for each pulse energy value.

4.1 Experimental and Data Processing Considerations

Single-shot experiments were conducted using the 247.86-nm carbon I atomic emission line. The source of carbon was carbon dioxide in the purified ambient air stream, which provided a homogeneous carbon source (-100 ppm) at the molecular level. Such uniform dispersion of the analyte species on a shot-to-shot level is not readily achieved by the introduction of aerosol species using the nebulizer. All experimental








data sets were recorded in time intervals no longer than two hours to minimize fluctuations of CO2 in the compressed air stream. A signal integration time of 5 pis was used for most experiments and the delay time optimization with pulse energies is presented in the next section.

Single-shot plasma temperature measurements were calculated from Boltzmann plots using 10 iron atomic emission lines in a single spectral window centered at 270 nm. The process of plasma temperature calculation was already described in Section 2.8, and the Fe II emission lines were listed in Table 2.4. During the processing of single-shot spectra for temperature measurements, all spectra were first smoothed using the SavitzkyGolay algorithm (Savitzky and Golay 1964, and Madden 1978) and then, individual spectra were selected according to a filtering algorithm. The filtering algorithm is based on the ratio of the integrated emission lines that defines the following parameters:

- A as the ratio of the peak at 259.9 nm to the peak at 258.58 nm,

- B as the ratio of the peak at 272.75 nm to the peak at 271.44 nm,

- C as the ratio of the peak at 277.93 nm to the peak at 271.44 rm, and

- D as the ratio of the peak at 278.36nm to the peak at 271.44nm.

Based on examination of ensemble-averaged spectra, the above parameters must satisfy the following criteria: i) the peak at 258.58 rm larger than 10000 counts, ii) A between 2.5 and 5, iii) B between 0.8 and 1.8, and iv) C and D larger than 0.1 and their ratio DIC between 1/3 and 3. Nominal values for these parameters based on 1200-shot ensemble average spectra varied as follow: peak at 258.58 rm from 17000 to 20000, A from 3.5 to 3.8, B from 1.09 to 1.12, C from 0.12 to 0.14, D from 0.15 to 0.19, and D/C from 1.2 to 1.4.








4.2 Energetic State of the Laser-Induced Plasma

The primary goal of the present study is to assess the precision obtained during single-shot spectral analysis of LIBS data, noting any variability of precision with laser pulse energy. However, keeping in context the pulse-to-pulse variability related to the breakdown process discussed in Section 2.6 (100% breakdown frequency about 190 mJ per laser pulse), the transmitted energy thru the plasma (or absorbed energy) is significant and is important at laser pulse energies above 100% breakdown frequency. Therefore, a more careful analysis of the absorbed energy by the plasma was performed first.

Figure 2.17 is re-plotted in a more convenient form in Figure 4.1 that shows the energy deposited into the plasma, expressed as percentage of incident pulse energy, for purified air as a function of the pulse energy. At pulse energy of 192 mJ, which ensures 100% breakdown frequency, the plasma absorbs 48% of the incident pulse energy. Above this pulse energy, the percentage of absorbed energy increases steadily to a maximum value of approximately 60% for pulse energies higher than about 250-260 mJ. This apparent saturation effect was observed first by Radziemski et al. (1983), who reported a transmission of 5% (95% absorption) at 300 mJ, and later by Chen et al. (2000) who obtained an absorption of 90% for incident energies from 45mJ to 80mJ (13.4 mJ energy threshold, f/5,and 6.5-ns pulse width). Radziemski et al. explained that after sufficient energy is deposited in the plasma to ionize nearly all of the matter, the additional energy tends to expand the size of the plasma rather than increasing the temperature or electron density. The current experiments suggest a saturation pulse energy of about 255 mJ, with a corresponding energy density of 0.18 kJ/cm3 (based on a plasma volume of 1.44x10-3 cm3 as calculated in Chapter 6). This value is one order of magnitude lower than the energy density of 5 U/cm3 reported by Chen et al. (2000).




Full Text

PAGE 1

PLASMA-PARTICLE INTERACTIONS FOR THE QUANTITATIVE ANALYSIS OF INDIVIDUAL AEROSOL PARTICLES USING LASER-INDUCED BREAKDOWN SPECTROSCOPY By JORGE E CARRANZA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2002

PAGE 2

ACKNOWLEDGMENTS I would like to thank Dr. David Hahn for being the best teacher that I had in a very long time, for his constant and invaluable advice, guidance, and patience during the development of my research, and for instilling in me a better understanding of science. What is more, I would like to express my sincere gratitude to him for his confidence shown in me three years ago when we met and I was struggling with the English language. I would also like to thank the members of my graduate committee for their help throughout this process. My thanks also go to my lab mates for their helpful talks and company during experiments, and for their support during class periods. Thanks especially to Brian Fisher and Greg Yoder for their great deal of help during the experiments of ambient air monitoring in the summer of 2000. In addition, I would like to express my appreciation to my friends in Gainesville for sharing enjoyable times, and very special thanks to my friend Diana Serrano for our constant lunch talks, and for her constant help and support. Finally, I want to acknowledge to the Department of Mechanical Engineering at the University of Florida for their financial support and for giving me the chance to be a teaching assistant. 11

PAGE 3

TABLE OF CONTENTS Eige ACKNOWLEDGMENTS ii LIST OF TABLES vi LIST OF FIGURES vii ABSTRACT xii CHAPTERS 1 INTRODUCTION 1 1 . 1 Basic Principle of LIBS 3 1.2 Laser-Induced Breakdown Initiation and Plasma Formation 4 1.3 Fundamental Studies 6 1.4 LIBS as an Analytical Technique 9 1 .5 Conditional Data Analysis for LIBS 14 1.6 Single-Shot LIBS-Based Aerosol Analysis 16 1 .7 Vaporization of a Single Aerosol Particle Using LEBS 18 1.8 Motivation and Objectives for Present Study 20 2 EXPERIMENTAL FACILITIES AND FUNDAMENTAL PLASMA CHARACTERISTICS 22 2.1 Aerosol Generation System 23 2.2 Nebulizer Characterization 26 2.2.1 Mass Flow-Rate Calibration 27 2.2.2 Lifetime of Generated Droplets 28 2.2.3 Aerosol Particle Characterization 30 2.2.4 Nebulizer Droplet Characterization 34 2.3 LIBS Instrumentation 37 2.4 LIBS Spectral Analysis 39 2.5 LIBS Calibration 43 2.6 Laser Induced Breakdown Threshold 48 2.7 Global Energy Balance of the Optical Breakdown 51 2.8 Plasma Temperature 52 2.8.1 Spectral Window Calibration 57 2.8.2 Plasma Temperature Decay 26 2.9 Summary 59 iii

PAGE 4

3 FEASIBILITY OF AEROSOL DETECTION IN AMBIENT AIR 61 3.1 Ambient Aerosol Sampling System 62 3.2 LIBS Data Analysis 64 3.2.1 Collection of Data 64 3.2.2 Processing of Data 65 3.3 Results and Discussion 68 3.3.1 Mass Concentration 70 3.3.2 Particle Analysis 77 3.3.3 Particle Size Distribution 79 3.4 Summary 83 4 SAMPLING STATISTICS FOR SINGLE SHOT ANALYSIS 85 4.1 Experimental and Data Processing Considerations 85 4.2 Energetic State of the Laser-Induced Plasma 87 4.3 Time Delay Optimization with Pulse Energy 89 4.4 Single-Shot Plasma Emission Fluctuations 91 4.5 Single-Shot Plasma Temperature 94 5 ASSESSMENT OF THE UPPER PARTICLE SIZE LIMIT OF VAPORIZATION USING THE LIBS TECHNIQUE 100 5.1 Experimental and Data Processing Methodology 100 5.2 Vaporization of a Collection of Nanoparticles 102 5.3 Vaporization of a Single Silica Particle 104 5.4 Laser-Particle or Plasma-Particle Vaporization 108 5.5 Particle Vaporization Driven by Laser-Induced Plasma 112 5.6 Precision and Accuracy of Particle Sizing Using the LIBS Technique 116 6 CHARACTERISTIC PLASMA VOLUMES FOR ANALYSIS OF AEROSOL PARTICLES USING LIBS 119 6.1 Experimental Methodology 121 6.2 Statistical Sample Volume 124 6.3 Mass-Based Sample Volume 128 6.4 Physical Plasma Volume 130 6.5 Discussion of the Characteristic Plasma Volumes 133 7 CONCLUSIONS 137 8 RECOMMENDATIONS FOR FUTURE WORK 140 APPENDICES IV

PAGE 5

A ZERO ORDER LOGNORMAL DISTRIBUTION 142 B QUARTZ TUNGSTEN HALOGEN LAMP 144 C SAVITZKY-GOLAY ALGORITHM 145 REFERENCES 147 BIOGRAPHICAL SKETCH 152 V

PAGE 6

LIST OF TABLES Table page 2. 1 . Aerosol Generator Specifications 25 2.2. Standard Aqueous Solutions 26 2.3. Microsphere Silica Particle Suspensions 26 2.4. Summary of Generated Aerosol Particle Characteristics 33 2.5. Laser and Optics Specifications 39 2.6. Fe II Atomic Emission Lines for Plasma Temperature Calculations 53 3.1. Parameters for Aerosol Particle Detection 65 3.2. Equations for Calibration of Mass Concentration 66 3.3. Summary of Analyte Mass Concentration 76 3.4. Summary of Particle Size 82 6. 1 . Actual Silica Particle Sample Rates 128 6.2. Summary of the Characteristic Plasma Volumes for 315 mJ per Laser Pulse 133 VI

PAGE 7

LIST OF FIGURES Figure page 2. 1 Schematic of the aerosol generation system 24 2.2 Nebulizer calibration curve of deionized water. Each data point is the average of a minimum of 3 runs (error bar = ± one standard deviation). 27 2.3 Light scattering responses from the testing chamber due to the excitation of a laser pulse of 30 mJ under a co-flow of 60 1pm of purified air 30 2.4 Aerosol particles generated by a nebulization rate of 0.09 ml/min into a co-flow of 42 1pm of air; a) For 2500 pg/ml of Iron, and b) For 10000 pg/ml of titanium {TEM micrograph of 50K magnification) 31 2.5 Iron-based particle size distribution at airborne mass concentration of 4425 pg/m^ corresponding to an aqueous solution of 2500 pg/ml 32 2.6 Titanium-based particle size distribution at airborne mass concentration of 17700 pg/m^ corresponding to an aqueous solution of 10000 pg/ml 32 2.7 Aspect ratio for the titanium-based crystal particles at airborne mass concentration of 17700 pg/m^. 33 2.8 Nebulizer droplet size distribution based on the combination of iron and titanium solution at 10000-pg/ml aqueous concentrations 35 2.9 Zero-order lognormal distribution with modal diameter 340 nm at three dimensionless widths ao that describes the droplet size distribution created by a medical nebulizer 37 2.10 Experimental setup of a LIBS apparatus 38 2.1 1 LIBS spectrum for an iron mass concentration of 4425 pg/m in purified air. Ensemble average of 1200 laser pulses (8 ps delay time with 5 ps integration time) 40 2.12 Effect of time delay in atomic emission and continuum emission using an integration time of 2 ps at each designated delay time. Atomic emission line C I 247.8 nm 43 Vll

PAGE 8

2.13 Peak-to-base ratio and signal-to-noise ratio of the atomic emission line at Xo, where AXo is the width of the peak, rms is the root-mean square of the noise at the off-peak baseline, 5,is the average continuum intensity 45 2.14 Calibration curve for iron 256.26-nm emission peak signal using 8-ps delay and 5-ps time width. Error bars represent ± one standard deviation (R^ = 0.996) 46 2.15 Time delay optimization at fixed 2-ps time width for the maximization of the LIBS signal P/B using the carbon emission line C I at 247.8 nm and at pulse energy of 247 mJ 47 2.16 Calibration curve for magnesium 280.27-nm emission peak signal using a 40-ps delay time and 40-ps time width. Error bars represent ± one standard deviation 0.998) 48 2.17 Percentage of laser pulses producing breakdown as a function of laser pulse energy for various sample streams 49 2.18 Transmitted energy thru the laser-induced plasma in air as a function of laser pulse energy. Error bars represent ± one standard deviation 51 2.19 Correlation wavelength versus pixel number of the iCCD array for the 270-nm spectral window 55 2.20 Reference lamp irradiance and collected irradiance by the LIBS system in the spectral window 270 nm 56 2.21 Correction factor to provide relative intensities line-by-line in the spectral window 270 nm 56 2.22 Representative 1200-shot average spectrum in air obtained using a laser pulse of 300 mJ at 5 -ms delay time and 5 -ms integration time after intensity correction 57 2.23 Boltzmann plot using Fe II emission lines for energy pulse of 300 mJ at a delay time of 10 ps and width time of 5 ps 58 2.24 Plasma temperature decay in air and nitrogen atmosphere using pulse energy of 300 mJ an integration time of 5 ps. Error bar = ± one standard deviation 59 3.1 Schematic of the ambient air sampling system 62 3.2 Transport efficiency of ambient air particles from the inlet to the LIBS sample chamber as a function of the particle diameter 64 3.3 Aluminum and magnesium calibration curves using the emission lines A1 1 394.4 nm and Mg I 279.55 nm 66 viii

PAGE 9

3.4 Two single-shot LIBS spectra. The lower spectrum represents a single MgO particle. The upper spectrum that has been vertically shifted for clarity represents a single Mg-Si-containing particle 69 3.5 Two single-shot LIBS spectra. The lower spectrum represents a single AlO particle. The upper spectrum that has been vertically shifted for clarity represents a Ca-Al-containing particle 70 3.6 The lower Na-LIBS spectrum is the 9600-laser-shot average, and the upper NaLIBS spectrum is the 30-identified, Na-shot average fi'om the 9600 laser shots. Both spectra have the same intensity scale but have been vertically shifted for clarity 71 3.7 Enhancing of the Mg-LIBS signal using the conditional analysis. The 14-Mg hits are a subset of the 9600 laser shots. Both spectra have the same intensity scale but have been vertically shifted for clarity 72 3.8 Mass concentration of magnesium as a function of time. Each data point represents the average LIBS-based concentration over a two-hour sampling period 72 3.9 Mass concentrations of calcium and sodium as a function of time. Each data point represents the average LIBS-based concentration over a two-hour sampling period 75 3.10 Sampling frequency over a period of two hours. Each data point corresponds to the fi-equency of hits for 1200-shot laser sequence 77 3.11 Histogram of calculated diameters for magnesium-containing particles for the Fourth of July holiday period. The particles were modeled as magnesium oxide (MgO) 79 3.12 Histogram of calculated diameters for calcium-containing particles for the Fourth of July holiday period. The particles were modeled as calcium carbonate (CaCOs) 80 3.13 Histogram of calculated diameters for calcium-containing particles for the nonholiday period. The particles were modeled as calcimn carbonate (CaCOa) 81 3.14 Histogram of calculated diameters for sodium-containing particles for the Fourth of July holiday period. The particles were modeled as sodium chloride (NaCl) 82 4. 1 Percentage of incident pulse energy absorbed by the laser-induced plasma as a function of laser pulse energy. Error bars represent ± one standard deviation 88 4.2 Time delay optimization at fixed 2-ps time width for the maximization of the P/B using the Carbon emission line C I at 247.8 nm 89 IX

PAGE 10

4.3 Optimal delay times for the spectral carbon emission line (P/B) at 247.8 nm as a function of laser pulse energy. The solid line is a second-order curve fit 90 4.4 Carbon peak emission (P), continuum emission (B), carbon peak-to-base (P/B), and carbon signal-to-noise ratio (SNR) as a function of laser pulse energy. Each data point represents the calculated value of a single 100-shot average spectrum ... 91 4.5 Laser shot-to-shot variability of the peak emission (open circles), continuum emission (solid squares), and P/B (solid diamonds) for (a) 200 mJ laser pulse energy, and (b) 344 mJ laser pulse energy. Data correspond to the optimal time delay at different pulse energies 93 4.6 Precision of the LIBS signal, expressed as P/B ratio and relative standard deviations (RSD) of P/B and SNR as a function of laser pulse energy. Each data point represents the average of 100 single-shot calculations. Error bars represent ± one standard deviation 94 4.7 Effect of the smoothing algorithm on the Fe II atomic emission lines selected (indicated by arrows) corresponding to a single-shot spectrum. The original spectrum was shifted for clarity 95 4.8 Plasma temperature and RSD of the plasma temperature as a function of laser pulse energy. Each data point is the average single-shot temperature calculations, and is based on iron II emission. Error bars represent ± one standard deviation 97 5.1 Peak-to-base (P/B) ratio of the 288.16-nm silicon emission line as a function of silicon mass concentration for a well-disperse aerosol stream of silicon-based nanoparticles generated by nebulization of aqueous silicon standards. (R^=0.999) 102 5.2 Single-shot spectra corresponding to a single 2.1-pm-diameter silica microsphere as collected and following application of the Savitzky-Golay smoothing algorithm. The smoothed spectrum has been shifted vertically for clarity 104 5.3 Ensemble-averaged spectra corresponding to individually detected monodisperse silica microspheres with diameters of 1.0, 1.5, 2.1, and 2.5 pm 106 5.4 P/B at Si 1 288.16 nm for ensemble-averaged spectra of individually detected monodisperse silica microspheres as a function of the cube of the silica particle diameter. The continuous line is a linear fit of the first three data points 107 5.5 Gaussian temporal profile corresponding to a pulse energy of 160 mJ given in a total time of 2.8 times the fullwidth half maximum (Tq) 110 5.6 Time and energy for the laserand plasma-particle interaction in pulse energy of 320 mJ Ill X

PAGE 11

5.7 Silica particle diameter distributions for a nominal diameter of 1.02 pm. Sample of 47 particles 117 5.8 Silica particle diameter distributions for a nominal diameter of 1.50 pm. Sample of 126 particles 117 5.9 Silica particle diameter distributions for a nominal diameter of 2.08 pm. Sample of 574 particles 118 6. 1 The natural logarithm of the transmission as a function of the product of the extinction cross-section and dilution factor for the 2.1-pm-silica particle suspension. The error bars represent ± 1 standard deviation 125 6.2 Independence of the threshold value in the conditional analysis for determination of the hit rate. Silica particles of 2.1 pm in diameter and at a 45-cm'^ number density were used. DfW equals purified deionized water only 126 6.3 The natural logarithm of one minus the experimental silica particle-sampling rate as a function of the silica particle number density. The error bars represent ± 1 standard deviation 127 6.4 Silicon calibration curve for a low mass concentration range used in the quantification of the mass-average sample volume. The error bars represent ± 1 standard deviation 129 6.5 The LIBS-based equivalent mass concentration of the 1.0, 1.5, and 2.1 pm silica particles as a function of the silicon mass contained in the silica particles. The error bars represent ± 1 standard deviation 130 6.6 Temporal scale for the transmission measurements thru the plasma created by the Nd:YAG 1064-nm laser, and probed by the Nd:YAG 532-nm laser 131 6.7 Cross-section of the physical plasma measured at the end of the 1064-nm plasmagenerating laser pulse. The boundary represents an optical thickness of 0.1, and the error bars represent + 1 standard deviation. The dashed line is an elliptic profile fit to the experimental plasma volume and aspect ratio 132 XI

PAGE 12

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PLASMA-PARTICLE INTERACTIONS FOR THE QUANTITATIVE ANALYSIS OF INDIVIDUAL AEROSOL PARTICLES USING LASER-INDUCED BREAKDOWN SPECTROSCOPY By Jorge E. Carranza August 2002 Chair: David W. Hahn Department: Mechanical Engineering In response to the need for discrete characterization of ambient air fine particles due to the direct relation of particle size and particle composition to human health effects, laser-induced breakdown spectroscopy (LIBS) has been studied in this dissertation to support its development as a real-time aerosol analysis technique capable of measuring particle size and particle composition. In Chapter 1, a literature review and the principles of LIBS are presented in the context of aerosol analysis. The complete experimental facilities are described in Chapter 2. An aerosol generation system was developed to produce well-dispersed aerosolized nanoparticles that served as the calibration source for the LIBS system. The LIBS system consists of a 1064-nm Nd:YAG pulse laser, supporting optics, and an intensified charge-coupled device for plasma emission quantification. The developed LIBS system was successfully deployed as discussed in Chapter 3 for ambient air monitoring (specifically aluminum, magnesium, calcium, and sodium). Mass Xll

PAGE 13

concentrations were recorded on the order of low parts per trillion and minimum particle sizes about 200 nm. In Chapter 4, issues of plasma homogeneity and signal fluctuations on a shot-to-shot basis were addressed to elucidate optimal laser pulse energy for single shot analysis, thereby identifying a characteristic state of the plasma where signal fluctuations are minimized. The implicit assumption of complete particle vaporization is investigated in detail in Chapter 5, with the determination that silica particles up to 2.1pm diameters are completely vaporized due to plasma-particle interactions. Finally, the characteristic plasma volumes related to the scheme of particle sizing are determined and analyzed in the context of the analysis of single aerosol particle detection. The outcome of this research yielded an enhanced understanding of single aerosol analysis with the LIBS technique. Important conclusions are that LIBS measurement should be made at the plasma saturation condition, particle sizing should be limited to about 2 pm or less, particle vaporization is driven by plasma-particle interactions, and that real-time ambient air monitoring is feasible with LIBS on time periods as short as 4 minutes. Xlll

PAGE 14

CHAPTER 1 INTRODUCTION Aerosols are a mixture of either solid or liquid particles suspended in a gas phase medium. Aerosols of considerable interest in the atmosphere are those with the particle size ranging from 1 nm to 100 pm in diameter, and with particle number concentration ranging from a few particles per cm^ in clean environments to 10* particles per cm^ in highly polluted regions (Noble et al. 2000). Particles in the low size range (i.e., submicrometer to several micrometers in diameter) are of the most concern in atmospheric sciences, process and control monitoring, and especially regarding human health. The potential impact of the small particles on human health led the United States Environmental Protection Agency (US EPA 1996) to conclude that particulate matter was associated with increased morbidity and mortality. In fact, a recent study (Peters et al. 2001) reported that the smallest particles could penetrate deep into the respiratory system causing adverse health effects (such as the increasing incidence of heart attacks) in as little as 2 hours. In response to these health risk factors, research agendas have targeted a number of program areas, including measures of outdoor particulate matter (PM) and associated human exposures, characterization of emissions sources, air quality model development and testing, and assessment of hazardous PM components (U.S. National Research Council 1999). Characterization of the behavior and properties of airborne particles is difficult because environmental aerosols comprise a diverse and constantly fluctuating system. Primary particles (emitted into the atmosphere) and secondary particles (formed 1

PAGE 15

2 in the atmosphere) may undergo growth, evaporation, or chemical reactions, and their complex interactions are reflected mainly in the particle size and composition distribution (Hinds 1999). Many questions concerning the nature of these processes remain unanswered. The need for ambient air PM data, specifically size and composition measurements of individual particles, is common to all of these research areas; however, a bulk analysis of the particles is of limited use. Therefore, an individual particle analysis, specifically size and composition measurements, must be carried out to facilitate the understanding of not only these dynamic processes (i.e., transport and formation of toxic particles) but also their implication in the quality of life of the human beings (i.e., global warming, ozone depletion, and respiratory disorders). This need for PM data combined with the development of advanced aerosol analysis techniques have served as a motivation for the development of on-line or real-time aerosol particle analyzers. Several techniques have been used to analyze individual airborne particles in real time. Most of these techniques are based on mass spectrometry methods such as rapid single-particle mass spectrometry (Johnston and Wexler 1995, and Ge et al. 1998), ultrasensitive particle analysis system (Reents et al. 1995 and 2000), and aerosol time-offlight mass spectrometry (Prather et al. 1994). However, recently the feasibility of the first atomic spectroscopy approach for the analysis of single particles based on a technique called laser-induced breakdown spectroscopy (LIBS) has been shown (Hahn 1998, and Hahn and Lunden 2000). The LIBS technique is the basis of this research, and it will be discussed in detail below. All the above techniques are able to determine particle size and composition. However, the mass spectrometry techniques are limited to the use of auxiliary techniques for sizing, typically light scattering or time-of-flight. In

PAGE 16

3 contrast, the LIBS technique makes use of a direct mass measurement to resolve the determination of particle size with simultaneous multi-elements mass composition. In addition, LIBS is suitable to be miniaturized while functioning as a real-time, in situ or remote sensing monitor (Davies et al. 1995 and 1996). One more advantage of LIBS over mass spectrometry methods is its speed and experimental simplicity, as the LIBS instrumentation does not require the high vacuum instrumentation needed for mass spectrometry. All these characteristics of LIBS make it a promising teehnique for the analysis of single aerosol particles. 1.1 Basic Principle of LIBS Laser-induced breakdown spectroscopy (LIBS), also known as laser-induced plasma speetroscopy (LIPS), is an atomic emission speetroscopy teehnique that dates to the 1960s. When a pulsed laser beam is focused onto a small spot, the temperature of the loeally heated region increases rapidly to vaporize the material, and to dissociate the molecules, thereby forming an optically induced plasma. The plasma will be formed if the laser power density exeeeds the breakdown threshold value for the medium, as described in detail below. The laser-induced plasma is used as both the sample volume and the excitation source, dissociating all molecules and fine particulates within the highly energetic micro-plasma. The resulting plasma emission can be resolved both spectrally and temporally to yield spectra containing the atomic emission lines corresponding to the atoms present in the plasma volume, including atoms originating from aerosols initially enveloped by the plasma. The aggregate energy levels of each element of the periodic table are unique in the emission spectrum. Therefore, the spectral location of the atomic emission lines in the speetrum can be used to identify elements.

PAGE 17

4 and their emission intensities (height, width, or integrated peak) can be used to provide a measure of the amount of mass of the constituent elements in the sample. 1.2 Laser-Induced Breakdown Initiation and Plasma Formation The laser-induced plasma can be defined as a near-totally-ionized gas induced by laser irradiance sufficient to generate an electric field at the focal spot that exceeds the dielectric strength of the medium (Weyl 1989). This event is qualitatively marked by a glow in the focal region and followed by a distinct shock wave, while quantitatively distinguished by the absorption of the laser beam due to ionization. Two mechanisms involving the generation of ions, avalanche ionization and multi-photon ionization, are summarized here after Weyl (1989). In the avalanche ionization mechanism, the laser radiation is absorbed by free electrons that later collide with neutral atoms or molecules to release new electrons. The releasing of a new electron occurs for the free electron that gained sufficient energy and it is described by following relation e~ +M ^2e+M\ (1.1) This process leads to a cascade phenomenon that increases the electron concentration exponentially with time. In the second mechanism, the multi-photon ionization mechanism does not require the presence of free electrons as the avalanche ionization does. The multi-photon ionization involves the direct absorption of photons by atoms or molecules to produce ionization. This process is described by M + n{hv)^ e~ + M* . (1.2) g Both mechanisms of ionization require high laser irradiances, usually over 10 W/cm^, and occur simultaneously. In general, the breakdown initiation is started by multi-photon ionization and followed by cascade ionization, however, the literature

PAGE 18

5 indicates that avalanche ionization is characteristic for larger irradiation wavelengths (i.e., Nd;YAG laser 1064 nm), and photon-ionization is associated with shorter irradiation wavelengths (i.e., excimer laser 193 nm) (Biswas et al. 1988, Ho et al. 1997, and Martin et al. 1999). The evolution of the plasma after the breakdown initiation stage depends on many factors such as irradiance, ambient gas composition, and laser wavelength; a short description of the three major types of laser absorption wave is condensed here after Lee et al. (2000). At low plasma irradiation, laser-supported combustion waves are produced mainly via radiative transfer from the hot plasma to the cool high-pressure gas created in the shock wave. The plasma radiation is primarily in the extreme ultraviolet and it is generated by photo-recombination of electrons and ions into the ground-state atom. At intermediate irradiance, the laser-supported detonation wave model governs the expansion of the plasma. During the propagation of the shock wave, the shocked gas is heated enough to begin absorbing the laser radiation without requiring additional heating from the plasma; as a consequence, the laser absorption zone follows directly behind the shock wave and moves at the same velocity. At high irradiance, the laser-supported radiation model rules the absorption of the laser radiation. The plasma irradiation is so intense that, prior to the arrival of the shock wave, the ambient gas is heated to temperatures where the laser absorption begins. In the first nanoseconds, after the deposition of the pulse energy of the laser, the plasma expands rapidly emitting electromagnetic radiation primarily in the ultravioletvisible region. Several hundreds of nanoseconds later, the plasma slows down and the radiation in the ultraviolet band decreases faster than in the visible band. The plasma

PAGE 19

6 starts to decay by radiative transfer, quenching, and electron-ion recombination processes, and as a consequence, isolated emission lines and structured bands appear in the spectrum. These processes usually occur within tens of microseconds after the breakdown initiation. The characteristic lifetime of the plasma depends on the transient nature of electron densities inside the plasma. The sequential spectral emission of ionized, neutral, and molecular species generally occurs between 0.5-20 ps, 2-100 ps, and after 50 ps, respectively. 1.3 Fundamental Studies The laser-induced plasma is an important part in the LIBS technique and many works have been dedicated for its characterization. Due to diverse variations in the experimental measurements, the literature reports breakdown threshold irradiance in clean air from 90-1600 GW/cm^ for 1064 nm lasers, 10-ns pulse width, and from 30-380 GW/cm^ for 266-nm lasers (Smith et al. 2001). Simeonsson and Miziolek (1994) studied the effect of the laser wavelength on the breakdown threshold in ambient air. They reported the lowest value of 9.7 GW/cm^ at 193 nm, the values of 270 GW/cm^ at 355 nm and 200 GW/cm^ at 1064 nm. It was also found that due to the presence of particles in the focal region of the laser beam, the breakdown threshold could be reduced by two orders of magnitude (Leoncioni and Pettingill 1977, and Lushnikov and Negin 1993). The complex nature of aerosol particles and their interaction with the laser beam make it difficult to predict the breakdown threshold. While Reilly et al. (1977) found a decreasing breakdown threshold for water droplets from 0.5 to 15 pm using a 10.6 pm CO 2 laser, Pogodaev and Rozhdestvenskii (1979) showed that the breakdown threshold of water droplets increased with increasing particle size using a pulsed ruby laser.

PAGE 20

7 Radziemski and Cremers (1989) reported that the breakdown threshold was a function of the ratio of particle diameter to laser wavelength. They also suggested a dependency of the breakdown threshold on the total number of particles and size distribution, and on the focal volmne of the laser beam (as a function of laser irradiance). An important consideration is that any particle smaller than 10 pm is completely vaporized and incorporated into the plasma (Smith et al. 2001). It is noted that although 10-pm particle sizes are often reported as an upper limit for complete dissociation, no detailed measurements addressing this issue have been reported; a literature survey regarding this matter is given in Section 1.7. The laser-induced plasma is a non-equilibrium system from a global scale, but a possible solution to obtain anal)'tical information from the plasma is to consider the plasma as in local thermodynamic equilibrium (LTE). At LTE the plasma is modeled stepwise in depth with the assumption that at any depth there are thermodynamic equilibrium conditions, and at other depths the conditions are different but still in thermodynamic equilibrium. At thermodynamic equilibrium the plasma emitting radiation is directly coupled to the physical conditions of all the matter in the plasma; specifically, the continuous radiation is described by the Planck function, the atomic occupation numbers are specified by the Boltzmann and the Saha equations, and the atomic transitions are determined by the Einstein relations. In 1983, Radziemski et al. (1983) showed that the laser-induced plasma behaves as in local thermodynamic equilibrium (LTE) after about 1 ps from the onset of breakdown. They concluded in obtaining LTE by observing the agreement between the results of a theoretical radiationhydrodynamic model and the experimental measurements of the intensities ratios of

PAGE 21

8 C(II)/C(I), N(II)/N(I), and Be(II)/Be(I) emission lines. They also observed that after certain pulse energy, the excess of energy increases mainly the plasma volume instead of increasing plasma temperature (i.e., 20% increase of temperature over 5-fold pulse energy, 60 mJ to 300 mJ). Capitelli et al. (2000) discussed the LTE assumption under theoretical considerations, and concluded that laser-induced plasmas seemed to satisfy such a condition, but they also noted that phenomena such as low ionization, low excitation temperature, or non-equilibrium recombination could occur in the same temporal scale of LIBS measurements, thus implying the violation of LTE. Yalcin et al. (1999) measured electron density and temperature for the plasma formed by a Nd:YAG laser (100 mJ, 1064 run, 10 ns pulse). Based on the agreement between the ionization temperature and the excitation temperature, they reported that LTE was reached at delay times of 0.35 ps. Yalcin and coworkers (Yalcin et al. 1996, and 1999) also investigated the influence of ambient gas conditions on the laser-induced plasma. They observed that by changing the ambient gas, particulate level, humidity, and laser energy, the electron density and plasma temperature (what characterize the plasma under LTE) varied little. They also confirmed that once sufficient laser energy was available to produce the plasma, additional energy produced a larger plasma volume with the same thermodynamic conditions, similar to the conclusions of Radziemski et al. (1983). Chen et al. (2000) used an Nd:YAG laser (1064 nm, 6.5 ns pulse) to study the effect of the laser energy on the plasma formation in air. They found that 50% of the incident energy was absorbed when the incident pulse energy was just above the breakdown threshold, and this percent of absorption increased up to a maximum value in which the plasma is

PAGE 22

9 apparently saturated. They confirmed also that after the saturation of the plasma, the excess of deposited energy induced an expansion of the plasma, and observed that the higher the incident energy, the farther the initial plasma moved away from the focal spot. Overall, the nature of the laser-induced plasma has been well studied, leading the way to LIBS as an analytical technique. 1.4 LIBS as an Analytical Technique The technique of laser-induced plasma spectroscopy has been applied to the analysis of solids, liquids, and gases, and a number of literature reviews are available that cover a wide range of LIBS-based analyses (Smith et al. 2001, Darke and Tyson 1993, Radziemski 1994, Rusak et al. 1997, Schechter 1997, and Song et al. 1997). Relevant to environmental aerosols, several papers have focused specifically on LIBS as a method for the detection and determination of overall elemental mass concentrations, as a technique for continuous emission monitoring, and as a potential method for particle sizing. It has also been investigated as a remote sensing technique. 1.4.1 Elemental Detection and Overall Mass Concentration The applicability of LIBS in the elemental detection and overall mass concentration of aerosols has been shown in the large volume of publications during the last decades. In 1981, Radziemski and Loree (1981) introduced time resolution for LIBS to overcome the predominately continuum emission at early time and to optimize the analyte atomic emission lines. Using this approach they were able to detect chlorine and phosphorus-based aerosols in air down to concentrations of 60 ppm and 1 5 ppm, respectively. Later, Radziemski and coworkers succeeded in the direct sampling of beryllium in air with a limit of detection of 0.6 ng/g (0.7 pg/m^), which is comparable to the detection limit of 0.2 ng/g reported by the well-known technique inductively-couple

PAGE 23

10 plasma atomic emission spectrometry (ICP-AES), and lower than the maximum permissible exposure of 25 pg/m^ in 8 h given by the Occupational Safety and Health Administration (OSHA). The detection of beryllium was done as a cloud of beryllium particles (estimated diameter less than 1 0 pm) generated by laser ablation and as beryllium-chloride aerosol produced by a nebulizer/heat-chamber system. Essien et al. (1988) reported the detection and quantification of the metals cadmium, lead, and zinc in aerosol form in air using a detection system comprised by a monochromator, photomultiplier tube and boxcar averager. The aerosol samples were generated in the sub-micron range with a wide variation in diameters by a nebulizerheating chamber arrangement. The limit of detection for cadmium, lead, and zinc were 0.019 (22), 0.2 (252), and 0.24 pg/g (288 pg/m^), respectively. The permissible exposure limit reported by OSHA are 5, 50, 5000 pg/m^ for cadmium, lead, and zinc respectively. These results showed the promising capability of LIBS for detection of toxic metals. Rusak et al. (1997) summarized the versatility of LIBS for the detection of different elements as aerosols in air. Mass concentration for elements such as lithium, sodium, magnesium, aluminum, silicon, potassium, calcium, manganese, and gallium were reported to have detection limits below 10 ppm. For other elements such as chlorine, strontium, indium, and mercury, the detection limits were in the range of 10-100 ppm, and for elements such as sulfur and arsenic the mass concentration limits were just above 100 ppm. These results were obtained under the criterion of time delay optimization of the appropriate atomic emission line. Other research and review papers focusing on the study of elements and compounds including phosphine (PH 3 ), arsine

PAGE 24

11 (AsHs), and fluorine (CF 3 H) are reported in the literature (Sneddon and Lee 1999, Peng et al. 1995, and Singh et al. 1997). 1.4.2 Continuous Emission Monitoring The necessity of real-time and in situ measurements of elemental concentrations makes LIBS a strong candidate for industrial process monitoring. There are very few techniques that can meet these requirements and therefore work as continuous emissions monitoring (CEM) systems. Techniques such as non-disperse infrared and differential optical absorption spectroscopy are mostly restricted for gas phase species, while other techniques such as laser-desorption/mass spectrometry do not perform as in situ and/or non-intrusive measurements. The feasibility of LIBS as a CEM system has been demonstrated by Zhang et al. ( 1 999) who performed LIBS measurements at the EPA rotary Kiln Incinerator Simulator for toxic metals in near real time. The multi-metal CEM test was conducted to target antimony, arsenic, beryllium, cadmium, chromium, lead, and mercury at low (15 pg/m^), medium (60 pg/m^), and high (600 pg/m^) mass concentrations. Efforts to monitor antimony, arsenic, and mercury failed because of the target concentration below the LBS detection limit, but for the remaining metals LBS successfully monitored them in real time. Because the LBS system used only a single detection system, each metal was explicitly monitored from one fourth to one half of the reference-method sampling time with a time response of 17 s; EPA establishes RM29 as the reference method for multi-metals analysis. They reported several problems such as interference lines and damage of laser optics that avoided the detection of metals at different target concentrations, and diminished the accuracy of the LBS measurements. At high target concentrations the LBS accuracy with respect to that of RM29 for all-four

PAGE 25

12 metals varied from 49% to 77%. At medium coneentrations lead could not be detected and the relative accuracy ranged 38%-86% for the other metals. At low concentrations only beryllium and chromium were identified, with accuracy within 31% to 272%. On-line analysis of chromium-based aerosols using LIBS was reported by Neuhauser et al. (1999) in their efforts for the development of a CEM in industrial effluents. The actual performance test was run at a German electroplating facility regulated to 1 mg/m^ for chromium. They compared the LIBS approach with a conventional filter-based reference analysis (off-line approach) performed simultaneously by an independent laboratory during a 30-min measurement period. The results showed a direct correlation between the on-line LIBS analysis and the filter-based analysis, but the LIBS measurements were overestimated by a factor of two. They attributed the disagreement due to reassembling of the LIBS system at a different sampling point. They also pointed out the difference of both methods. While the filter-based approach had an integrative character (over 30 min), the LIBS measurements were a quasi-continuous and non-integrative method (15 measurements over 30 min with 20-s duration per measurement). In addition, they reported that the dynamic response of the LIBS system ranged from 50 to 100 s. Overall, they showed the potential of LIBS for on-line monitoring of heavy metals in industrial effluents. In another study of CEM technology, Hahn et al. (1997) investigated the application and process conditions of LIBS to metal emissions monitoring in waste combustion systems. Using sequences of 1000 shots, individual shots corresponding to the presence of chromium, manganese, and iron (hits) were collected on-line, in situ at a natural gas-fired, pilot-scale waste processing unit. All spectra for a given analyte were

PAGE 26

13 grouped to calculate an equivalent concentration based on an a priori laboratory calibration scheme. They determined the actual mass concentration for a given analyte by multiplying the corresponding equivalent concentration by the hit frequency for each analyte (i.e., the percent of shots that registered the presence of the analyte emission). The results were compared to those obtained by the RM29, in accordance with EPA standards. For the determination of manganese, LIBS and RM29 gave the same concentration of 3.3 pg/m^, while for chromium and iron the LIBS technique yielded results that were consistently 21% and 36% lower than those obtained by RM29 in the range of 2-3 pg/m^ and 40-140 pg/m^, respectively. In general, they concluded that finetuning the overall instrument to specific conditions might lead to enhanced sensitivity and performance. 1.4.3 Single Particle Analysis An extension of the LIBS technique for real-time sizing and elemental analysis of single particles was reported recently by Hahn (1998) and Hahn and Lunden (2000). This novel approach would be the first pure spectroscopy technique in development for the analysis (elemental mass and size) of single aerosol particles. They based their analysis on the point-sampling nature of LIBS and the discrete presence of the particles in a gaseous medium to detect single particles. A two-part calibration scheme (of known mass concentration and known discrete particle mass) was used to calculate a characteristic plasma volume, which is needed for particle sizing. Their results showed that the technique was both robust and highly sensitive during initial laboratory and field measurements. Measurements of calciumand magnesium-based aerosols were performed and compared to independent measurements using a laser aerosol

PAGE 27

14 spectrometer, with the resulting size distributions in agreement within a 25% maximum deviation. The understanding and potential development of this approach is the focus of this dissertation. In a recent application, which is explained in detail in the following chapters, aerosols in ambient air containing Al, Ca, Mg, and Na were monitored for a period spanning the Fourth of July holiday (Carranza et al. 2001). Measured mass concentrations ranged from 1.7 ppt to 1.7 ppb, and the measured aerosol diameters varied from 200 nm to 2 pm, thereby demonstrating the feasibility of LIBS for analysis of particulate matter in ambient air. In addition to the previous cited applications, it is worthwhile to mention LIBS as an atmospheric remote sensing technique. Davies et al. (1995 and 1996) coupled a fiber optic system to a conventional LIBS system to deliver the incident laser pulse to the zone of testing and to transmit back to a spectrometer the optical radiation emitted by the plasma. The system was tested for distances up to 100 m between the remote location and the apparatus. The system was capable of detecting chromium, cooper, manganese, molybdenum, silicon, and vanadium concentrations down to 200 ppm. 1.5 Conditional Data Analysis for LIBS The discrete nature of anthropogenic or natural aerosols combined with the essentially point-volume sampling method of the LIBS technique make the detection of particulates possible. For instance, at low particulate conditions, the traditional ensemble averaging of LIBS spectra may reduce considerably the signal-to-noise ratios of the targeted element contained in the particles. Knowing of this peculiarity for LIBS based aerosol measurements, Hahn et al. (1998 and 2000) developed a conditional data analysis algorithm for LIBS-based detection of individual aerosol particles that rejected spectral

PAGE 28

15 data based on the absence of atomic emission corresponding to the targeted analytes. Spectra were classified as particle hits if the ratio of the atomic emission intensity of the analyte line to the intensity of the adjacent featureless spectral region (i.e. continuum) exceeded a threshold value. The algorithm needs at least two emission lines of the analyte, one for triggering, and the other lines for analysis. The threshold value was typically set to obtain from 0.1% to 0.3% false hits in order to retrieve spectra with analyte emission line intensities comparable to the root mean square (rms) values of the spectral signal shot-noise. Using the conditional data analysis approach, single-shot LIBS spectra were identified and analyzed corresponding to aerosol mass concentration levels from low parts-per-billion to parts-per-trillion levels. Analogously, Schechter (1995) used a novel rejection algorithm to identify singleshot spectra and to screen anomalous events out such as those that did not contain elemental emission lines, and those events related to laser fluctuations or screening particle effects. His algorithm did not address the discrete nature of aerosols and its low particle density case; rather it was more of a data filtering approach to improve the analyte signal. Nevertheless, the criterion of the algorithm to identify the presence of the targeted element was related to the sum of the intensities of all expected lines. If the total intensity did not reach a threshold value, the spectrum would be rejected. As a result of this approach, Schechter demonstrated that the ensemble of the filtered spectra for Znbased aerosols improved the detection limit by a factor of three compared to direct ensemble averaging. A similar approach was used by Cheng (2000) and Martin and Cheng (2000) to measure trace metals of fine particles. The screening data algorithm was based on the (i) presence of multiple analyte emission lines, (ii) specified maximum

PAGE 29

16 analyte emission peak fluctuations, and (iii) specified maximum full-width, halfmaximum analyte line widths. The use of a suitable conditional analysis algorithm can yield significant increases in signal-to-noise ratios, but more importantly, opens the door to the analysis of singleshot LIBS spectra that correspond to individual aerosol particles. Such analysis is the basis of this research. 1.6 Single-Shot LIBS-Based Aerosol Analysis While Schechter (1995), Cheng (2000), and Martin and Cheng (2000) used single-shot measurements to improve the signal-to-noise of trace aerosol composition, Hahn et al. (1997, 1998, and 2000) used it mainly to analyze single aerosol particles. Schechter and the others based their analysis on spectral filtering, which was used to obtain a representative ensemble-averaged spectrum. These filtered spectra were subsequently used to make a calibration curve of the targeted element (peak-to-base ratio versus analyte concentration). Xu et al. (1997) also applied single-shot measurements to analj^e absolute particulate materials. They processed emission signals (absolute peak) from filtered single-breakdown events to obtain absolute concentrations, basing their method on the assumption that the same fluctuation pattern observed in the spectral peaks is present in the baseline as well. Consequently, a plot of the peak intensity against the average baseline intensity should provide a straight line with a slope related to the concentration of the analyte. Hahn and coworkers (1998 and 2000) developed a two-part calibration scheme to determine the size of aerosol particles. The idea was to correlate an equivalent analyte mass concentration to the analyte mass contained in the plasma volume when a single particle was sampled by the plasma volume, designated a particle hit. The sequence of

PAGE 30

17 the algorithm was to calculate first the plasma volume and then the particle size. First, the LIBS signal (ratio of atomic emission peak area to emission continuum intensity near to the peak) for the average of thousands of laser pulses was correlated to a known average analyte mass concentration. Second, the LIBS signal was calculated from the ensemble-averaged spectrum corresponding to a monodisperse particle stream of known size, mass, and composition in which the average particle concentration was adjusted to promote single-particle detection. Third, an equivalent mass concentration (also expressed as the ratio of single particle mass to plasma volume) was determined for the known particle mass to calculate the plasma volume. This last statement can be expressed by equating the actual mass concentration given by LIBS-based analysis and by aerosol mechanics Actual. Mass. Concentration = X • F = m • N , (1.3) where X is the equivalent mass concentration of all particle hits, F is the sample frequency or the percentage of laser pulses that sample an aerosol particle, m is the average particle mass, and N is the number density of particles (particles per unit volume). Assuming a Poisson sampling mode and ju as the average number of particles per plasma volume, the sample frequency can be written as the probability of collecting one or more particles by F = 1 exp(-//) , ( 1 . 4 ) where ju is readily expressed by the characteristic plasma volume Vp and the particle number density N noting that /j = VpM . For the case of low particle sampling rate (i.e., /u «1), Equation 1.4 may be expanded using a Taylor series to become

PAGE 31

which neglecting higher order terms yields F = M = Vp-N , (1.5) 18 Substituting Equation 1.5 in Equation 1.3 and solving for X yields the relation between the equivalent mass concentration, single particle mass, and the characteristic plasma volume: X = (single particle mass) / V . ( 1 . 6 ) Finally, knowing the chemical state of the analyte, the equivalent spherical diameter of the particle in the plasma volume is calculated using the following relation D = 6XV^ npf -|l/3 (1.7) where p is the bulk particle density and f is the analyte mass fraction with respect to the bulk particle. Therefore, the particle sizing technique follows from the steps outlined above. For an unknown particle distribution the equivalent mass eoncentration is calculated for an identified, single-particle hit using a traditional LIBS calibration curve. The absolute mass of the particle is then calculated using the equivalent mass eoncentration and the characteristic plasma volume. Last, by knowing or assuming the eomposition and density of the particle, the equivalent diameter is calculated. 1.7 Vaporization of a Single Aerosol Particle Using LIBS Notwithstanding the significant body of research seen in previous sections regarding laser-induced breakdown spectroscopy for aerosol analysis, to date no research has systematically addressed the fundamental assumption inherent in all quantitative LIBS measurements, namely the assumption of complete breakdown and vaporization of

PAGE 32

19 all analyte species that comprise the aerosol particles of interest. Then, the important question remains as to what is the largest particle size for complete dissociation and vaporization of individual particles suspended in a gas stream? A survey of the literature regarding the largest particles that can be completely vaporized for quantitative LIBS-based analysis reveals no systematic study designed to specifically address this issue. Cremers and Radziemski (1985) explored LfflS for the detection of beryllium particles deposited on filters corresponding to particle diameters of 50 ran, an ensemble collection of particles ranging from 0.5 to 5 pm, and for nominally 15-pm sized particles. They used a cylindrical lens to focus a laser beam directly on the surface of the filters, thereby producing a long plasma volume that engulfed the deposited beryllium particles, and subsequently collected the spectral emission. They reported a different analyte response, manifest as different calibration curve slopes, for these three different particle size classes, and concluded based on their experimental observations that incomplete particle vaporization occurred for particles with diameters greater than about 15 pm. Other studies explored direct LIBS-based analysis of beryllium aerosols using particles less than 10 pm in diameter (Radziemski et al. 1983, and Essien et al. 1988), and noted in the latter study that such a particle size is consistent with complete particle vaporization. Several contemporary research papers have cited a 10-pm upper size limit for complete particle vaporization (Ottesen et al. 1989, Yalcin et al. 1996, and, Hahn et al. 1997), some referring to the extensive work of Radziemski and Cremers, and some not citing any particular reference source. Based on the overall body of LIBS research, it appears that what was presented by Radziemski and coworker as a useful guideline for

PAGE 33

20 quantitative LIBS analysis of aerosols, namely the 10-|am upper size limit, has in essence become the accepted upper size limit for complete vaporization of aerosol particles within laser-induced plasmas. However, following extensive literature reviewing, no research has been reported that specifically addresses the issue of complete vaporization of individual airborne particles using laser-induced breakdown spectroscopy. This issue is addressed in Chapter 5 of this dissertation. 1.8 Motivation and Objectives for Present Study The need for the development of new real-time techniques for the analysis of single aerosol particles challenges LIBS capabilities shown in the last decades. In contrast to the traditional ensemble averaging of global mass concentration measurements, single-shot LIBS analyses opens the door to the sizing and mass quantification of single particles. No studies have been reported to date to answer new questions regarding important issues such as shot-to-shot plasma fluctuations, experimental precision, and plasma-particle interactions. Therefore, this research directly supports the longer-term goal of establishing the LIBS technique as a quantitative diagnostic tool for the simultaneous measurement of particle size and particle composition, with a focus on ambient air fine particulate matter. The purpose is to gain an understanding of the effects of plasma-particle interactions on the optimization and precision of LEBS-based single-shot analysis of gaseous and aerosol systems. The following technical objectives are addressed by this research. 1. Investigate the degree of laser-induced plasma homogeneity, and the correlation between plasma homogeneity and LIBS signal response to discrete aerosol particles.

PAGE 34

21 2. Investigate the optimal laser pulse energy for enhaneed signal response and signal robustness. 3. Investigate the relations between laser-induced plasma volume and particle sampling voliune. 4. Investigate the upper limit of single particle vaporization in the laser-induced plasma.

PAGE 35

CHAPTER 2 EXPERIMENTAL FACILITIES AND FUNDAMENTAL PLASMA CHARACTERISTICS As the LIBS technique is further developed and refined as an analytical tool for aerosol analysis, well-controlled experimental conditions are required. The use of nebulizers for sample introduction is a standard mode for inductively coupled plasma atomic emission spectrometry (ICP-AES) and inductively coupled plasma mass spectrometry (ICP-MS). Nebulizers generally work by converting bulk analyte solutions into an aerosol spray for transport to and injection into the plasma. The aerosol characteristics of a given nebulizer design are closely related to the overall precision of the analytical system, with resulting limitations attributed to coarse aerosol size and low analyte transport efficiencies. Ideal nebulizer characteristics include 100% analyte transport efficiency, the generation of fine aerosol droplets, and operation without clogging. Another desirable feature of an aerosol generation system is the ability to introduce aqueous particle suspensions into a gaseous sample stream. For instance, introduction of particles from suspension into a gaseous stream is necessary for singleparticle diagnostics as developed for the LIBS technique. Therefore, this section describes such an aerosol generation system along with the LIBS instrumentation required for plasma excitation and light collection, and the characteristics of the plasma produced with the coupling systems to assess the performance as a calibration source for the development of laser-based diagnostic techniques based on LIBS. 22

PAGE 36

23 2.1 Aerosol Generation System The aerosol generation system is shown schematically in Figure 2.1. The primary components are the pneumatic-type medical nebulizer (Hudson model 1720), the conical mixing/drying section, and the sample analysis section. A flow of compressed, dry nitrogen is used to drive the nebulizer flow, and the generated-aerosol droplets are delivered directly into the mixing/drying section. The nebulizer nitrogen flow is filtered through a HEPA filter cartridge, and regulated with a laminar-flow element flow controller. The full-scale flow is 10 liters per minute (1pm) with accuracy of ± 0.1 1pm. The nebulizer flow is directed through the center hole of a stainless steel, porous plate 125-mm in diameter. 0-rings seal the porous plate along the perimeter and at the center hole. A gaseous co-flow is introduced below the porous plate. The fimction of the gaseous co-flow is to provide bulk flow through the aerosol generator, both facilitating the drying of aerosol droplets and transporting the resulting solid particulates to the sample analysis section. Two parallel thermal-type mass flow controllers, each one with 30-lpm capacity and ± 0.3-lpm accuracy, are used to meter the co-flow gas. The mixing/drying section consists of a 175-mm length of micropolished stainless steel. The inner diameter of the mixing/drying vessel is 125 mm at the porous plate, and the vessel tapers to an inner diameter of 25 mm over a linear distance of 125 mm. A final 50-mm long, 25-mm constant diameter section connects the mixing/drying section to the sample analysis chamber. The sample analysis chamber is a standard stainless steel 70mm (2.76”) 6-way vacuum cross. The cross is mounted directly to the mixing/drying section and set to provide optical access to the generated aerosol stream. Three of the horizontal flanges are fitted with optical quality windows, and the fourth horizontal

PAGE 37

24 flange is fitted with a 50-mm diameter, 75-mm focal length UV grade fused silica lens. The focal point of the lens is centered on the central vertical axis of the mixing/drying section. Exhaust vent Figure 2.1 Schematic of the aerosol generation system. A detailed description of all the components is given in the Table 2. 1 . The last four items listed in this table correspond to the subsystem for ambient air monitoring, and it was used in the firework particle monitoring experiments, which are developed in the next chapter.

PAGE 38

25 Table 2.1 Aerosol Generator Specifications Unit Description Manufacturer Co-flow controller (two units) Model 8270, 0-30 1pm nitrogen @ 25°C Matheson Gas Products Co-flow meter (two units) Transducer model 82720434 Matheson Gas Products Nebulizer flow meter and controller Model MC12PS, 0-10 1pm air @ 25°C Alicat Science Inc. Two pressure gauges 0-300 Psi Ashcroft Two HEPA filters Product #12144 capsule Gelman Laboratory Coarse filter Part # X03-02-000, with dryer’s silica gel, Wilkerson Corporation Nebulizer Pneumatic medical type #1724 Hudson Air compressor Model PK-5060V, 5 HP and 125 Psi @ 15.7 CFM, Puma 489 PE tubing Flexible tubing, 1/8x1/4x1/16 and 1/4x3/8x1/16 Fisher Scientific Testing chamber Stainless-steel six-way cross, I.D. 35 mm Huntington Labs Drying section Micro-polished Stainlesssteel, I.D125X L175x I.D25 Custom design and fabrication Vacuum pump Model SR-0015-VP Thomas Compressors & Vacuum Pumps Flow meter Tube # GJ502, 3-48 1pm air Gilmont Instruments Stainless Tubing O.D. 5/8” ASTM A-26994A/ A-213-94B MSC Industrial Supply Co. PM 10 sampler Remove PM larger than 10pm, 1 m^/h 1 Rupprecht and Patashnick Co., Inc.

PAGE 39

26 2.2 Nebulizer Characterization The nebulizer may be characterized by the consumption rate of nebulized liquid as a function of gas flow rate into the nebulizer, as well as by the resulting size distribution of aerosol droplets. A high degree of precision in the nebulizer mass flow rate is desirable for the current use for sample introduction into the aerosol generation apparatus and subsequent LIBS calibration. A summary of the specification for all the standard solutions and suspensions used in this research is presented in the Table 2.2 and Table 2.3, respectively. Table 2.2 Standard Aqueous Solutions Solutions Description Manufacturer DI water Deionized ultra filtered water, W2-20 Fisher Scientific Industrial Grade Nitrogen 99.7% N 2 , less than 32 ppm ofH 20 Prax Air Calcium, Iron, Aluminum, Sodium, and Magnesium 1 0000 pg/ml ICP Standard in a matrix of 5% HNO 3 Spex, Inc. Titanium 10000 pg/ml ICP standard in water Spex, Inc. Table 2.3 Microsphere Silica Particle Suspensions Particle Diameter, pm Standard Deviation Particle Density, Particles/cm^ 1.02 NA 9.149 10^ 1.5 NA 3.122 10^ 2.08 NA 1.17 10^ 2.52 NA 6.4 10^ 3.0 NA 3.79 10 Â’ 4.5 NA 1.12 10^ 5.13 NA 7.58 10^

PAGE 40

27 2.2.1 Mass Flow-Rate Calibration The mass flow rate of the nebulizer was calibrated using distilled and deionized water over a range of nebulizer gas flow rates from 4.5 to 6.0 1pm of dry nitrogen. A gravimetric approach was used for flow periods ranging from 10 to 30 minutes for each gas flow rate. Liquid mass lost from the nebulizer varied from 689 to 3933 mg over the test matrix, while the analytical balance had an accuracy of 0.5 mg. The nebulizer calibration curve is presented in Figure 2.2 for the nebulization of deionized water. As observed in the figure, the nebulizer output is highly linear over this reported range of gas flows. Figure 2.2 Nebulizer calibration curve of deionized water. Each data point is the average of a minimum of 3 runs (error bar = ± one standard deviation). It was noted that the nebulizer did not produce an aerosol output for nitrogen gas flow rates below a minimum value. This minimum flow rate was approximately 4 1pm. It was also observed that as the gas flow rate increased to values significantly larger than

PAGE 41

28 6 1pm, the high precision characteristic of lower flow rate diminished as seen the Figure 2.2 data. Once the regions of linear operation were determined, experiments were limited to the linear regime, with a usual nominal gas flow rate of 5 1pm. 2.2.2 Lifetime of Generated Droplets The nebulizer produces a precise mass flow rate of fine droplets, which are introduced into the gaseous co-flow within the mixing/drying section. When the generated droplets mix with the co-flow gas stream, the droplets begin to evaporate via mass diffusion of water to the surrounding gas (Turns 1999). Diffusion-controlled evaporation is an appropriate model for the current range of nebulizer droplets (minimum diameter of ~0.1 pm) and corresponding range of Knudsen numbers (Kn < 0.001) (Davis et al. 1980). The time to for a droplet to completely evaporate, assuming a binary mixture of water vapor in air, is given by the expression to = Dq^ /K, where Do is the original droplet diameter and K is the evaporation constant. For the present exercise, the evaporation constant was calculated as 1.7x10'^ m^/s by using a binary diffusion coefficient of 2.2x10'^ m^/s, and a water mass fraction of 0.016 at the droplet interface (equilibrium vapor pressure) and 0.008 for the gas stream (all nebulized water has evaporated). Using these parameters, drying times for initial water droplets of 250, 500, and 2000 nm in diameter were calculated as 0.04, 0.15, and 2.3 ms, respectively. These initial droplet diameters are consistent with those produced by the current nebulizer, as discussed below. For a nominal total gas flow rate of the aerosol generator of 50 1pm (nebulizer flow plus co-flow), a single droplet with drying time on the order of 10 ms has an equivalent drying distance of less than 1 mm. This is a very short distance compared with the 175-mm-mixing/drying-section length, which has

PAGE 42

29 a total residence time of approximately 790 ms. The nebulizer, however, produces a concentrated aerosol mist of droplets, hence phenomena such as saturation or nearsaturation effects over the droplet-to-droplet length scales should be considered. Accordingly, for a two-order of magnitude decrease in the evaporation constant resulting from multiple droplet effects, sufficient residence time is still provided to dry a 2-|o,m droplet diameter (230 ms). Laser light scattering measurements were performed to assess the evaporation of the nebulizer droplets prior to the optical sampling chamber. The laser beam of a Qswitched frequency-doubled Nd:YAG laser (X = 532 nm), at 30-mJ-pulse energy was focused at the center of the sample chamber using a 25-mm diameter, 250-mm focal length lens. Scattered light was collected at 90° using a telescope with unity magnification, and recorded with a 200-ps rise time phototube detector. Specifically, laser light scattering measurements were recorded in a pure gaseous co-flow stream of air, and in flow streams corresponding to the nebulization of pure deionized water and aqueous solutions of iron. The scatter laser-pulse profiles are shown in Figure 2.3, and correspond to 100-shot ensemble average. For the case of nebulized deionized water, any water droplet entering the sample chamber would contribute to the measured light scattering signal. The scattering signal corresponding to the baseline case of the purified co-flow gas only is equal to the sum of Rayleigh scattering from the air molecules and stray light, including reflections from within the optical sample chamber. The light scattering signal was found to increase by only 10 % for the nebulization of deionized water and by more than 300 % with the nebulization of 10,000 pg/ml of dissolved iron with respect to the baseline case of pure

PAGE 43

30 air in both cases. The increase in scattering with the nebulization of deionized water only is attributed to either increased stray light due to light scattering within the mixing/drying section or to the penetration of a small number of very large droplets that did not sufficiently evaporate. The 3-fold increase in scattering with the nebulization of the iron solution is attributed to the formation of a large number of iron-based particulates formed by the evaporation of the nebulized water droplets. Figure 2.3 Light scattering responses from the testing chamber due to the excitation of a laser pulse of 30 mJ under a co-flow of 60 1pm of purified air. 2.2.3 Aerosol Particle Characterization The size of aerosol particles was obtained corresponding to the nebulization of either iron or titanium aqueous solutions ranging between 2500 and 10000 pg/ml. Aerosol particles were collected directly on carbon coated, copper TEM grids, which were placed directly into the center of the aerosol stream at the exit plane of the 6-way sample cross. Aerosol particles were collected for 20 minutes for each sample condition and the samples were subsequently analyzed using transmission electron microscopy

PAGE 44

31 (TEM) and X-ray diffraction. Aerosol size measurements were made directly from TEM micrographs using a light box and micrometer. TEM micrographs for generated-aerosol particles are presented in Figure 2.4. Figure 2.4 Aerosol particles generated by a nebulization rate of 0.09 ml/min into a co-flow of 42 1pm of air: a) For 2500 pg/ml of Iron, and b) For 10000 pg/ml of titanium (TEM micrograph of 50K magnification). As can be seen from Figure 2.4, while the nebulization of iron solution produced primarily spheroidal-shaped particles, the nebulization of titanium solutions produced highly crystalline particles with a linear, rectangular geometry. In the case of iron solution, the reported particle diameter represents the average of the major and minor diameter of the spheroid particles. In the case of the titanium-based particles, they were modeled as rectangular rods with a square cross-sectional area. Width and length were recorded directly from the TEM images, and volume-equivalent spherical particle diameters were then calculated from the relation d = (6 LW^/k)'^\ The following Figure 2.5 and Figure 2.6 show the equivalent particle size distribution obtained from aqueous solutions of iron, and titanium and formed at the testing chamber. Both ironand titanium-based particles showed well-defined distributions, with a skewing toward larger sizes, thus given a good representation of the generated particles. The well-dispersed

PAGE 45

32 aspect ratios (length over width) observed from Figure 2.7 also suggest that the titaniumbased crystals are formed by non-preferential growth along all facets. Figure 2.5 fron-based particle size distribution at airborne mass concentration of 4425 pg/m^ corresponding to an aqueous solution of 2500 pg/ml. Diameter, nm Figure 2.6 Titanium-based particle size distribution at airborne mass concentration of 17700 pg/m^ corresponding to an aqueous solution of 10000 pg/ml.

PAGE 46

33 Figure 2.7 Aspect ratio for the titanium-based crystal particles at airborne mass concentration of 17700 pg/m^. Table 2.4 Summary of Generated Aerosol Particle Characteristics Aqueous solution concentration 2500 pg/ml Fe 10000 pg/ml Fe 2500 pg/ml Ti 10000 pg/ml Ti Nebulization rate, ml/min 0.09 0.09 0.09 0.09 Co-flow, 1pm air 42 42 42 42 Mass concentration, pg/m^ 4425 17700 4425 17700 Mean diameter, nm 13.2 68 74 73.5 Standard deviation, nm 7 29 52 58 Particle concentration, cmÂ’^ 64.5 10Â’ 1.9 lOÂ’ 0.4 lOÂ’ 1.8 lOÂ’ Chemical state FeO FeO TiO TiO

PAGE 47

34 The analysis of the X-ray diffraetion micrographs allowed the determination of the lattice d-spacing of the collected particles. They were compared with reference library values to determine the chemical state of the particles, resulting in the conclusion of primarily FeO for the iron-based particles and TiO for the titanium-based particles. It was noted that no d-spacings were observed to be consistent with the formation of hydrated or nitrated species. A summary of the aerosol particle characteristics is reported in Table 2.4. 2.2.4 Nebulizer Droplet Characterization The TEM analysis and X-ray diffraction data were used to infer information about the size characteristics of the nominal water droplets produced by the nebulizer. Specifically, the initial diameter of the liquid droplet that subsequently produced each solid particulate can be calculated from a conservation of mass relationship based on the analyte species; namely the mass of the analyte species (e.g., titanium or iron) in the dry particle must equal the mass of analyte in the original liquid droplet. This relation may be expressed by the identity ^DlS = ^Dlpf, (2.1) where Dg is the initial liquid droplet diameter, S is the solution concentration of the aqueous analyte (pg analyte/ml). Dp is the particle diameter of the measured solid aerosol, p is the bulk density of the solid aerosol, and /is the mass fraction of analyte in the solid aerosol particle. For the FeO particles, the density is 5.7 g/cm^ and the iron mass fraction is 0.78. For the TiO particles, the density is 4.9 g/cm^ and the titanium mass fraction is 0.75. These values were utilized to convert the respective aerosol particle size distributions recorded for the titanium and iron-based solid particles into

PAGE 48

35 primary water droplet diameters produeed by the nebulizer. The iron-based and titaniumbased particle distributions both yielded similar nebulizer droplet size distributions, namely mean droplet diameters of 520 and 528 nm, respectively. This consistency is significant, in that the droplet size distribution created by the nebulizer is not influenced appreciably by the nature of the nebulized aqueous analyte solution. A composite distribution of the nebulizer droplet sizes based on the combination of TiOand FeObased particle measurements is presented in Figure 2.8, as calculated using Equation 2.1 as described above. The mean droplet diameter is 524 nm, and standard deviation and modal diameters equal to 398 and 340 nm, respectively. The droplet size distribution is well described using a zero-order lognormal distribution -ZOLD (Espenscheid et al. 1964) of the form 100 80 c o 20 0 0.0 0.5 1.0 1.5 2.0 Nebulizer Droplet Diameter, pm Figure 2.8 Nebulizer droplet size distribution based on the combination of iron and titanium solution at 10000-pg/ml aqueous concentrations. ( 2 . 2 )

PAGE 49

36 which is an equivalent expression for the lognormal distribution (see Appendix A). The lognormal distribution is widely known for the description of aerosols; however, the ZOLD is conveniently parameterized by a modal diameter {d„^ and by a dimensionless width (cto) instead of the geometric mean and the geometric standard deviation needed by the lognormal distribution. The ZOLD that describes the present droplet size distribution uses a modal diameter = 340 nm, and a dimensionless measure of width CTq = 0.45. This Go was obtained as the best value that fits the main eharacteristics of the droplet size. The mean {dmean) and standard deviation (ct) of the droplet size distribution based on ZOLD are given by the relations -expCfo-'), (2.3) a = d„ [exp(4o-„' ) exp(3o-„' , (2.4) which fields a value dmean= 460 nm and a corresponding standard deviation of
PAGE 50

37 Figure 2.9 Zero-order lognormal distribution with modal diameter 340 nm at three dimensionless widths ctq that describes the droplet size distribution created by a medical nebulizer. 2.3 LIBS Instrumentation A schematic of the experimental LESS setup is shown in Figure 2.10, and a detailed description of the components is cited in Table 2.5. The excitation source is a 1064-nm Q-switched Nd:YAG laser with a nominal pulse width of 10 ns, maximum pulse energy of 400 mJ, and operating with a 5 Hz pulse repetition rate. The laser pulseto-pulse energy stability is specified as 4% rms. The output laser beam of 7 mm in diameter is expanded to an effective beam diameter of 12 mm using a telescope and an aperture (not shown schematically). Then, the expanded beam is passed through a 50mm diameter pierced mirror, and focused into the sample chamber with a 75-mm-focal length lens (UV grade quality) of 50 mm in diameter to create the plasma. The plasma emission is collected along the incident beam in a backward direction (180°) by the same primary focusing lens, separated by a pierced mirror, and launched into a fiber optic bundle using a matched 75-mm-focal length lens. The fiber optic cable is coupled to a

PAGE 51

0.275-m spectrometer, with an optical dispersion of 0.035 nm/pixel for the 2400groove/mm grating. The dispersed emission is then recorded using a time-gated, intensified charge-coupled device (iCCD) detector array. 38 Figure 2.10 Experimental setup of a LIBS apparatus. The 14-bit A/D converter of the iCCD readout gives a factor of 50 dynamic range for a nominal baseline signal level of 325 counts for a given on-chip binning factor. In addition, the spectral stripe is uniform over 100 pixel rows of the iCCD array. By varying the software-controlled on-chip binning of rows by multiples of 5, and additional factor of 20 in signal gain is achieved. Therefore, a factor of 1000 is realized for the overall dynamic range of the current LIBS detector system, which enables the size analysis of aerosols over a factor of 10 with respect to particle size (e.g., mass a D^).

PAGE 52

39 Table 2.5 Laser and Optics Specifications Unit Description Manufacturer Nd:YAG laser 1064 nm, 10 ns pulse, CFR 400 mJ Big Sky Laser Technologies, Inc. iCCD Pulse Generator Model PG-200, programmable pulse Princeton Instruments, Inc. iCCD Detector Controller Model ST135 controller Princeton Instruments, Inc. iCCD Camera Intensified charge-couple device, model 1024-MLDSE1 Princeton Instruments, Inc. ICCD Chiller Refngerator recirculator, model CFT-25 Neslab Spectrometer 275 0.275 meter spectrometer, 2400 groves/mm Action Research Corporation Spectrometer Controller Scan controller, model 275 Action Research Corporation Optical Telescope xl.7, two UV-grade 1064nm AR lenses CVI Laser Corporation Elliptical Pierced Mirror (2 inches) UV-grade AR enhanced CVI Laser Corporation Focusing Lens 75 -mm focal length, 50-mm dia, UV-grade, 1064 nm AR CVI Laser Corporation Fiber Optic 1-m fiber optic bundle, LG455-0201 Action Research Corporation 2.4 LIBS Spectral Analysis A representative spectrum, corresponding to the ensemble average of 1200 laser pulses, is shown in Figure 2.1 1 for iron at 4425 pg/m^ in pimfied air. As observed in the figure, a typical spectrum is characterized by continuum emission and by atomic emission lines. In a laser-induced plasma, continuum emission is a result of two main

PAGE 53

40 mechanisms: a Bremsstrahlung process (free-free transition) and a recombination process (free-bound transition). The Bremsstrahlung process occurs when an electron passes close to a positive ion and due to quantum mechanic effects its trajectory changes. The acceleration of the electrons in this way causes it to radiate electromagnetic energy —this radiation is called Bremsstrahlung that in German means breaking radiation. All electrons go from a free-high energetic state to another free-low energetic state, and the differences in energy correspond to radiate energy. In a different mechanism, the recombination process or the free-bound transition takes place when free electrons, which are in a continuous energetic state in the plasma, return to the nucleus of the atoms. The free electrons are captured by the electric field of positive ions and release electromagnetic energy corresponding to the difference of energy between the continuous state and the level of energy that the electron occupies in the nucleus. Figure 2.11 LIBS spectrum for an iron mass concentration of 4425 pg/m in purified air. Ensemble average of 1200 laser pulses (8 ps delay time with 5 ps integration time).

PAGE 54

41 In contrast to the continuum emission that is a continuous spectral electromagnetic radiation, the atomic emission is a type of bound-bound transition in which an excited bound electron emits a discreet quantum of energy (usually due to spontaneous emission) corresponding to its de-excitation to a lower energy level in the atom. The quantum nature of the electron in the atom makes the transitions governed by selection rules based on the quantum numbers of the initial and final state. The initial and final state of the electron can be described by wave functions, and the probability that the transition occurs increases as these wave functions overlap. These atomic transitions (also called atomic emission lines) are characteristic for each element of the periodic table and superimposes to the continuum emission as observed in Figure 2.1 1. A formal definition of the continuous emission (Wsr’*.Hz''.cm‘^) is presented in the following Equation 2.5 (Griem 1997) 2^aV (EA 3 • \kTj 3/2 •[•••]• exp ^E^+hv' kT . (2.5) kT and [•••] = 2 ^ g/-exp n max kT ( z^EjA^ kT where a is the fine-structure constant, z is the charge of the ion. Eh is the ionization energy of hydrogen, k is the Boltzmann constant, T is the thermodynamic temperature, is the Bohr radius, Ng is the electron density, Nz is the ion density, AEz is the reduction in ionization energy, h is the Planck constant, vis the photon frequency, gf\s the Maxwellaverage free-free Gaunt factor, g„ is the bound-free Gaimt factor, and n and Umax are the principal quantum number and the maximum quantum number.

PAGE 55

42 The Einstein-Boltzmann relation (Wsr‘.cm'^) expressed by the Equation 2.6 determines the intensity of a spectral emission line (Lochte-Holtgreven 1995) T _ 1 AiSk h .. ( eA 1 1 — i, A'^-exp 4^ U A ^ V ( 2 . 6 ) where Ak,is the transition probability for the emission line, gk is the statistical weight of the excited state, 1/ is the partition function, is the atom density, A is the /a transition wavelength, and Ek is the upper-level excitation energy. In practice, the electromagnetic radiation from the plasma is collected after a period of time in which the atomic emission lines of the analyte of interest appear (due to the differing rates of decay between continuous and atomic emission); this time is called the time delay or gate delay. The collected light is averaged for a period of time called integration time or just time width. The evolution of the atomic emission and the continuum emission with time delay is shown in Figure 2.12 for the case of carbon I at 247.8 nm in the spectral window 240-260 nm. It is clearly visible from the graph when the spectra are compared that the continuum emission and the atomic emission have different rates of decay, indicating in some way that an optimal relation may exit to enhance the analyte atomic emission. This behavior is explained in detail below for carbon. The designation used to typify an atomic transition is related to the state of ionization, specifically I describes a neutral atomic transition that expresses a boundbound transition when the atom has all its electrons, an II corresponds to the first ionization transition that is a bound-bound transition when the atom misses an electron. Figure 2.1 1 shows a significant number of iron atomic emission lines present in the spectrum (Fe II) at 238.2, 239.56, 240.49, 256.26, 258.59, 259.94, 260.7, 261.18, and 263.1 nm, and the atomic emission line carbon I at 247.8 nm.

PAGE 56

43 Figure 2.12 Effect of time delay in atomic emission and continuum emission using an integration time of 2 |as at each designated delay time. Atomic emission line C 1 247.8 nm. 2.5 LIBS Calibration A significant use of the described aerosol generation system in section 2.2 is the development of calibration curves for laser-based diagnostic techniques, such as laserinduced breakdown spectroscopy in the present study. Knowledge of the nebulization rate, the analyte solution concentration, and the co-flow rate enable calculation of the resulting analyte mass concentration produced in the LIBS sample chamber. The analyte concentration C (mass of analyte per volume of gas) is given by the equation C = NRSS Qco-Jlow Qneb (2.7) where the nebulization rate {NR) is given by the correlation in Figure 2.2, SS is the analyte solution concentration (pg/ml) in the nebulizer, Qco-flow '^^ the flow rate of the coflow gas, and Q^b is the nebulization gas flow rate. A nominal analyte solution concentration of 2500 pg/ml yields a mass concentration of 4425 pg-of-analyte/m^ of gas

PAGE 57

44 when utilizing a co-flow of 42 1pm and a nebulization gas flow of 5 1pm. In addition to the mass concentration, the number density of solid aerosol particles may be estimated based on the nebulizer droplet distribution reported above and the liquid nebulization rate. For a nebulization rate of 0.09 ml/min and a co-flow of 42 1pm, the aerosol particle number density in the sample chamber is in the order of lOÂ’ particles/cm^. Such a high number density of nanometer-size particles (i.e., 10 to 100 nm) provides a uniform aerosol stream that is well suited for the ensemble averaging of multiple LIBS spectra at a constant overall mass concentration. As a calibration source for LIBS it is important to determine the atomic emission intensity for a known analyte mass concentration. To ensure a true average analyte response, the analyte must be well dispersed and spatially averaged throughout the plasma for many laser shots. The current number density of 10^ particles/cm^ corresponds to an average of more than 10^ particles per plasma volume (plasma volume ~ 1.7 10 cm as reported in Chapter 6). Hence a true average analyte response is determined for LIBS calibration by ensemble averaging. As an example, LIBS spectra were recorded for generated iron particles corresponding to a concentration range from 1734 to 5780 pg/m^ and in addition to the nebulization of pure deionized water. From Figure 2.1 1, the 256.26-nm emission line was chosen and utilized for calibration. Two parameters are generally used in LIBS analysis, the peak-to-base ratio (P/B) and signal-to-noise ratio (SNR). A formal definition of these parameters is given by Equations 2.8, 2.9, 2.10, 2.1 1 and 2.6, and illustrated by Figure 2.13. r me P / p Integrated Peak Emission Intensity Base Emision Intensity

PAGE 58

45 Integrated Peak = P= jl^dX{b^ AX ^ ) (2.9) Base = B,= Average{B,^^,B^^,,) (2.10) where B.=~\l^^.dX SNR ^ rms • AX^ (2.11) In the above relations, Iba and Ix are the intensity of the continuum emission and atomic emission line defined by the equations 2.5 and 2.6 respectively, Xq is the wavelength of the atomic emission line, AXq is the width of the atomic emission line, 5, is the average intensity of the continuum emission at the left and right of the atomic emission over a AX region, and rms is the root-mean square of the shot noise at the off-peak baseline. Figure 2.13 Peak-to-base ratio and signal-to-noise ratio of the atomic emission line at Xo, where AXo is the width of the peak, rms is the root-mean square of the noise at the off-peak baseline, 5, is the average continuum intensity.

PAGE 59

46 The LIBS signal P/B is defined as the integrated peak intensity normalized by the continuum baseline intensity about the peak. The baseline intensity was estimated by interpolating the off-peak intensity values on each side of the analyte emission line. Similarly, the signal-to-noise ratio is defined as the average peak intensity normalized by the root-mean square of the noise at the continuum close to the position of the emission line. For the iron the calibration curve is presented in Figure 2.14, which is the relation P/B versus known iron concentration in the gaseous stream corresponding to the 256.26nm emission line. The P/B used in the calibration curve was maximized for an optimal time delay of 8 ps with an integration time of 5 ps. The resulting linear correlation based on a least-square fit output a 0.996 regression coefficient (R^). A similar linear calibration curve was produced for the 259.94-nm iron emission line, yielding a 0.991 regression coefficient. Figure 2.14 Calibration curve for iron 256.26-nm emission peak signal using 8-ps delay and 5-ps time width. Error bars represent ± one standard deviation (R^ = 0.996).

PAGE 60

47 To illustrate the process of P/B maximization, the carbon emission line C 1 247.8 nm was used. The time-resolved process requires choosing a time delay that maximizes the signal P/B for fixed integration time and pulse energy. Figure 2.15 shows the variation of the P/B as a function of delay time using an integration time of 2 ps at pulse energy of 247 mJ. As it was noticed before, the atomic and the continuum emission have different rate of decay that are used advantageously by the definition of the LIBS signal. The optimization of the P/B favorably uses the maximization of the ratio of the emitted energy by the processes related to bound-bound transition and free-free and free-bound transition. This process of optimization was followed to build the calibration curve of all analytes used in this research. Figure 2.16 shows another calibration curve for an analyte of interest. Such calibration curve reaffirms the well-suited system developed for the analysis of aerosols using LIBS. Figure 2.15 Time delay optimization at fixed 2-ps time width for the maximization of the LIBS signal P/B using the carbon emission line C I at 247.8 nm and at pulse energy of 247 mJ.

PAGE 61

48 3 d OQ £ ro c D) w 250 "T I ' I I I I I I I I — I — I — I — I — I — I — I — I — I — I — 1 — I — r y = 4.0329 + 0.054645X f^= 0.9979 A 200 150 100 50 4. _L _L _L 1000 2000 3000 4000 5000 Mg Mass Concentration, ng/m^ Figure 2.16 Calibration curve for magnesium 280.27-nm emission peak signal using a 40-ps delay time and 40-ps time width. Error bars represent ± one standard deviation (R^ = 0.998). 2.6 Laser Induced Breakdown Threshold The focus of a pulsed laser beam onto a small spot may increase the temperature rapidly of a localized region to vaporize the material, and to dissociate the molecules, thereby forming an optically induced plasma. The Nd;YAG laser used to create the plasma for LIBS experiments of this research delivers a well-collimated Gaussian beam with a divergence less than 0.0045 rad, a pulse-to-pulse stability less than 4% rms, and an energy drift of less than 10%. The plasma will be initiated and formed if the laser power density exceeds the threshold value for the medium, which is the power density required to hold electrons in their energy levels and avoid cascade ionization. Therefore, pulse-topulse stability and matrix effects are significant at laser pulse energies near the breakdown threshold, noting the extreme case of non-breakdown. For the determination of laser-induced breakdown thresholds, the experimental set-up described in Section 2.3 was used, and a conditional data analysis was applied to

PAGE 62

49 calculate the percentage of laser shots generating a plasma, as based on the presence of optical emission at a delay time of 1 ps with respect to the incident laser pulse. A breakdown is said to occur when the detected signal was 15% higher than the dark signal (no breakdown). Experiments were performed for different gas stream conditions, including pure air, pure nitrogen, air with the introduction of deionized water at a mole fraction of 0.003, air with the introduction of titanium-based particles at a mass concentration of 17 ppm (mass basis), and air with the introduction of sodium-based particles at a mass concentration of 1 7 ppm. The mean particle diameters were on the order of 25 nm with aerosol number densities on the order of 5x10* cm'^, and generated as described in Section 2.2. Figure 2.17 shows the result of these experiments. >, o c (U 3 a0) c 5 o a re a? CO OU 100 1 1 1 1 1 1 1 1 — 1 — j — I — 1 — I — 1 — 1 — n" • in Air 80 in N2 in Dlwater -1_ + in Ti particles X in Na particles 60 X 40 X + 20 X _L ±1 1 100 %) 50 100 150 Pulse Energy, mJ Breakdown 200 250 Figure 2.17 Percentage of laser pulses producing breakdown as a function of laser pulse energy for various sample streams. The statistical breakdown threshold is defined in the present study as the required energy to produce a breakdown frequency of 50% (percent of laser pulses yielding breakdown). As observed in Figure 2.17, the threshold energy ranged from 155 mJ to

PAGE 63

50 166 mJ over all gas stream conditions. The pure gas streams are characterized hy an abrupt breakdown threshold, changing from 1 0% to 90% breakdown frequency in a change from 160 to 175 mJ per pulse, respectively. In contrast, the addition of solid particulates to the gas stream lowered the laser pulse energy required for breakdown, producing breakdown frequencies of about 10% at 70 mJ per pulse. The addition of water only mirrored the results of the solid particles, which is presumed due to the presence of submicron-sized water droplets acting as aerosols. The difference in the breakdown phenomena at low pulse energies between the purely gaseous and particleladen streams is related to the generation of seed electrons during plasma initiation. Multiphoton ionization requires a well-defined photon flux for breakdown of nitrogen or oxygen, hence the well-defined profiles in Figure 2.17. However, the presence of solid particles provides surface sites for multiphoton ionization, which though less photonintensive than gas-phase breakdown, is a much less predictable process due to the complicated photon-particle surfaee interactions. For an estimated laser beam diameter of 1 00 pm at the focal point (measured at laser pulse that yields no breakdown), the 50% breakdown frequency corresponds to average power density of about 200 GW/cm^. For all experimental conditions in the present study, a breakdown frequency of 100% was realized at pulse energy of 190 mJ. At this pulse energy, the incident photon flux is sufficiently high to uncouple the breakdown process from local matrix effects (e.g., aerosol loading), resulting in a repeatable, hence robust, plasma process. All subsequent experiments were performed using pulse energies above 190 mJ.

PAGE 64

51 2.7 Global Energy Balance of the Optical Breakdown Not all the energy of a laser pulse is used to produce and sustain the optically induced plasma. Using a Gaussian-temporal pulse-laser profile, it can be understood that the optical breakdown initiation would not start until the power density of the leading edge (of the laser pulse) reaches the required threshold of the medium. Therefore, before reaching this point, an amount of energy passes by the focal region without being absorbed. In addition, after the breakdown initiation and later development, the remaining laser pulse is partially absorbed by free electrons of the plasma. The amount of absorbed energy depends on the population of free electrons, which also depend on other factors such as laser wavelength, pulse duration, and size of the focal region. Figure 2.18 shows the transmitted energy through the optical breakdown in air as a function of the incident laser pulse energy. Figure 2.18 Transmitted energy thru the laser-induced plasma in air as a function of laser pulse energy. Error bars represent ± one standard deviation.

PAGE 65

52 The average energy per pulse was recorded using a volume absorbing calorimeter (AC5001 model calibrated at 1064 nm), both before and after the laser-induced plasma. No reflected or scattered energy was detected from the testing chamber. As can be seen from Figure 2.18, the incident laser pulse energy is transmitted 100% in the no breakdown region since the laser pulse energy does not reach the power density threshold of the medium (e.g., purified ambient air). At 100% breakdown frequency, the transmitted energy thru the formed plasma has a minimum of about 100 mJ in the range of laser pulse of 230-250 mJ. At higher pulse energies, the transmitted energy increases slower than the supplied pulse energy indicating that even though more energy is going thru the plasma, it is possible that the plasma is absorbing more energy than that of the range 230-250 mJ. 2.8 Plasma Temperature It has been stated in the Section 1.3 that laser-induced plasma (LIP) is in local thermodynamic equilibrium (LTE) at least 1 ps after the onset of the laser pulse. In LIP, the Boltzmann and Saha relations describe the atomic populations at different energy levels and ionization states, while the Maxwell distribution governs the electron speed distribution (dominant particles in the kinetic energy of the plasma). The temperature parameterizes the Boltzmann, Saha, and Maxwell relations adopting the name of excitation, ionic, and electron temperature respectively. At LTE, all these temperatures would have the same value corresponding to the thermodynamic equilibrium temperature of the species. In the present research the term plasma temperature is used to engulf the meaning of these temperatures and is calculated using the Boltzmann plot that is based on the Einstein-Boltzmann relation (Equation 2.6), rewritten as follows

PAGE 66

53 In I 1 — — + const T\k) ( 2 . 12 ) The Boltzmann plot graphs basically E/k in the abscise and ln(M;\/Ag) in the ordinate to have the inverse of the slope of a linear fit as the temperature; noting the atom density and partition function are constant for transition at the same ionization state of a particular species. The number of emission lines should be at least three, corresponding to the same ionization level, with well-spread upper excitation energy. The intensity of the emission line need not be in absolute value, but must be relative line-to-line. To assess the plasma temperature, 10 iron atomic emission lines in a single spectral window centered at 270 nm were used to build Boltzmann plots. The selected spectral lines were all Fe II emission lines with relatively broad spread of upper energy states as is shown in Table 2.6. Table 2.6 Fe II Atomic Emission Lines for Plasma Temperature Calculations Wavelength X, nm A, 10* s’ g Upper energy level E, cm"’ J 258.5876 0.81 8 38660.043 7.69 10-’® 259.9396 2.20 10 38458.981 7.65 lO'’® 261.1874 1.10 8 38660.043 7.69 lO’’® 261.7618 0.44 6 38858.958 7.7310'’® 271.4413 0.55 6 44784.761 8.91 10'’® 272.7539 0.85 4 45044.168 8.96 10'’® 273.9548 1.90 8 44446.878 8.84 10'’® 275.5737 2.10 10 44232.512 8.80 10'’® 277.93 0.76 8 62322.431 1.24 10'’® 278.3691 0.70 10 62083.108 1.23 10'’® Source: http:/physics.nist.gov

PAGE 67

54 2.8.1 Spectral Window Calibration The use of Boltzmann plots to determine the plasma temperature requires the precise identification of the Fe II atomic emission lines in the selected 270-nm spectral window. The spectral window centered at 270 nm has a width of about 35 nm. The collected light is recorded by an iCCD detector array with 1024 pixel to resolve the spectral radiation and 256 binned rows to quantify the collected light. First it is needed to establish accurately the relation between pixel number and wavelength, and second a pixel-by-pixel intensity response is measured over a specified number of binned rows to assess the net intensity (accounting by losses through lenses, fiber optic, spectrometer, and intensifier). For the wavelength calibration, a set of spectrally resolved emission lines was selected. Nebulization of aqueous solution of lead, titanium, and aluminum were performed to obtain aerosol concentrations of 5700 pg/m^, and LIBS spectra were collected at this condition. Lead at 261.417, 266.315, 280.199, and 283.305 nm (Pb I), titanium at 284.194 nm (Ti II), and aluminum at 256.798, 257.509, and 266.038 nm (A1 I), and 281.618 nm (A1 II) were visually identified in the spectral window at different delay times varying from 5 to 20 ps. The pixel corresponding to the maximum intensity of the emission line was assigned to the wavelength of the corresponding emission line. The wavelength calibration curve for the 270-nm spectral window is presented in Figure 2.19. The wavelength response is highly linear over the pixels of the detector array with an adjusted correlation coefficient r^ = 0.99998. For the pixel-by-pixel intensity response, 100-binned rows were selected to integrate the light diuing the plasma temperature measurements. The spectral intensity

PAGE 68

55 response in the 270-nm spectral window, considering all losses in signal, was correlated to a reference signal given by a calibrated quartz tungsten halogen lamp. The quartz halogen tungsten filament lamp simulates a blackbody radiation over a range 250-2400 nm; details of this lamp can be seen in Appendix B. Figure 2.19 Correlation wavelength versus pixel number of the iCCD array for the 270-nm spectral window. The pixel-by-pixel intensity calibration procedure is as follows. The quartz tungsten halogen lamp was located in the testing chamber of the LEBS experimental setup. The lamp was turned on and spectra were collected after a delay time of about 1 minute with an integration time of 100 ps. Figure 2.20 shows the collected spectrum by the LIBS system, and the actual spectral irradiance of the lamp given by the manufacturer. The spectral irradiance of the lamp has been scaled by a factor of 10000 to obtain an appropriated correction factor (scaled lamp irradiance over collected irradiance) that is also plotted in the Figure 2.21. The correction factor adjusts the collected spectra

PAGE 69

56 being used to build the Boltzmann plots, thereby providing a relative measure of intensities for all atomic emission lines. Figure 2.20 Reference lamp irradiance and collected irradiance by the LIBS system in the spectral window 270 nm. Figure 2.21 Correction factor to provide relative intensities lineby-line in the spectral window 270 nm.

PAGE 70

57 2.8.2 Plasma Temperature Decay The evolution of the laser-induced plasma can be quantified by the decay of its temperature. The calculation of the plasma temperature requires the construction of Boltzmann plots. To calculate the plasma temperature, LIBS spectra corresponding to iron-based aerosols with mass concentration of 5700 pg/m^ were used. Spectra corresponding to 1200-shot average were collected at 5-, 10-, 15-, and 20-ps delay time with an integration of time of 5 ps for a laser pulse of 300 mJ. The temperature decay of the plasma was monitored in air and in nitrogen atmospheres. Figure 2.22 shows a typical spectrum collected in air after its intensity having been corrected by the factor discussed in the previous section. The integrated emission lines (peaks) are obtained according to the equation 2.9 and used as h in the Equation 2.12. Wavelength, nm Figure 2.22 Representative 1200-shot average spectrum in air obtained using a laser pulse of 300 mJ at 5-ms delay time and 5-ms integration time after intensity correction.

PAGE 71

58 A Boltzmann plot corresponding to a collected spectrum at a delay time of 10 ps with an integration time of 5 ps is displayed in Figure 2.23. The major component of imcertainty in the temperature calculated by this method comes from the uncertainty of the values of excitation energies. Boltzmann plots could report uncertainties in the temperature as high as 20 % (Griem 1997), however, the advantage of using widespread upper excitation energies leads to a better accuracy in the temperature determination. A plasma temperature of 11500 K was calculated from the plot corresponding to a laserinduced plasma created in ambient air. Finally, the evolution of the plasma temperature in air and nitrogen atmosphere is presented in Figure 2.24. The plasma temperature profile is basically indistinguishable between air and nitrogen. The error bars showed the precision of the calculated values and correspond to one standard deviation. Figure 2.23 Boltzmann plot using Fe II emission lines for energy pulse of 300 mJ at a delay time of 10 ps and width time of 5 ps.

PAGE 72

59 14000 13000 £ 12000 3 S 11000 Q. E ® 10000 n i 9000 (0 Q. 8000 7000 0 5 10 15 20 25 30 Delay Time.^s Figure 2.24 Plasma temperature decay in air and nitrogen atmosphere using pulse energy of 300 mJ an integration time of 5 ps. Error bar = ± one standard deviation. 2.9 Summary In conclusion, an aerosol generation system was implemented that enables the production of precise mass flow streams of well-characterized, submicron-sized aerosol particles. As a calibration source for laser-induced breakdown spectroscopy, linear calibration curves were produced for iron and other elements of interest over mass concentrations ranging from 0 to 17700 pg/m^. The aerosol generation system is suitable to the production of multi-species aerosols via the nebulization of multi-species aqueous solutions. The gaseous co-flow provides additional flexibility to the aerosol generation system by enabling variations in the co-flow gas independent of the nebulizer operation. The laser-induced plasma produced with the present system is characterized by a breakdown frequency of 100 % at pulse energies greater than 190 mJ, the energy needed to obtain repeatable plasma every laser shot. In addition, the minimum laser pulse energy transmitted thru the plasma is about 225 mJ per pulse. At typical LIBS time scales, the 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 ' ' ' 1 ' ' ' ' 1 ' ' I 1 t • Nitrogen Air i • i 1j i_ 1 1 ill 1 1 1 1 till 1—1 L 1 i ~

PAGE 73

60 plasma is still extremely hot (about 11000 K). The measured temperature profiles give an insight as to the plasma energetic state after the onset of the laser pulse, which is a useful reference in the following chapters. Overall, the system is well suited for the development, assessment, and calibration of in situ laser-based diagnostic techniques such as laser-induced breakdown spectroscopy. In the following chapter, the present system is tested to monitor ambient air in real time, especially for the sizing of single particles.

PAGE 74

CHAPTER 3 FEASIBILITY OF AEROSOL DETECTION IN AMBIENT AIR The implemented LIBS system described in the previous chapter was tested for the detection of aerosols in ambient air, including quantitative mass concentration measurements and size/composition measurements of individual aerosol particles in the present chapter. Ambient air was monitored on the University of Florida campus between Monday, June 28, 2000 and Friday, July 7, 2000, and again between Tuesday, August 1, 2000 and Friday, August 5, 2000. These sample windows were selected to overlap with the Fourth of July holiday period, a holiday associated with the use of fireworks. Specifically, there were at least 5 mimicipal fireworks displays within a 50mile radius of the University of Florida between July 3 and July 4 (Tuesday). LIBS data were collected typically twice each day, generally in the morning hours and in the aftemoon/evening hours, for four elements of interest: aluminum, calcium, magnesium, and sodium. For each sampling period, spectral data were collected for approximately two hours. During the course of experiments, daily high temperatures ranged from 28 to 34 C, overnight low temperatures ranged fi'om 18 to 23°C, and daily wind speeds ranged from 0 to 1 3 mph, with winds primarily to the north direction. Rain was periodic (maximum of 2.4 cm on June 29) during the nearly three weeks of sampling; however, no data were recorded during periods of rain. In the following sections, the aerosol collection system, the methodology of analysis, and the finding of this experiment are presented. 61

PAGE 75

62 3.1 Ambient Aerosol Sampling System An ambient air sampling system was designed to bring a well-controlled flow of ambient air to the prototype LIBS instrument. A US EPA standard PM 10 inlet was utilized at the air inlet of the sample line. A PM 10 inlet removes all particles greater than 10 microns in diameter via inertial impaction, while also ensuring uniform sampling of air. The air was drawn into the PMIO inlet at a volumetric flow rate of 16.7 liters per minute (~1 m^/hr), and subsequently transferred to the LIBS sample chamber using 12mm inner diameter stainless steel tubing. The sampling inlet was located on the roof of a single story building (about 4.6 m high) adjacent to the second story laboratory housing the LIBS system. The inlet was positioned approximately 1 m from the edge of the roof and approximately 3 m from the laboratory. The inlet was positioned to be as free as possible from any obstructions; however, no detailed analysis of building interactions with ambient air sampling was included for this study. A schematic of the sampling system is shown in Figure 3.1. Figure 3.1 Schematic of the ambient air sampling system.

PAGE 76

63 A stainless steel transfer line of about 8.2-m long was connected directly to the inlet of the sample chamber, where the airflow subsequently entered the 6-way stainlesssteel vacuum cross. A vacuum pump was connected to the outlet of the sample chamber, and was used to maintain the overall sampling rate. The pressure in the sample chamber was maintained about 1 torr below atmospheric pressure. An aerosol deposition program (McFarland 1996) was used to calculate the particle transfer efficiency throughout the entire sampling system. The transport efficiency from the sampling inlet to the LIBS sample chamber as a fimction of the particle size is presented in Figure 3.2. The efficiencies ranged from 92 to 99 percent for particulate diameters between 0.1 and 2.5 microns. Essentially all particle diameters reported in this study ranged between 0.2 and 1.6 microns, with corresponding transport efficiency between 96 and 99 percent. In consideration of these calculations, no corrections were made to particle size distributions or particle mass concentrations based on the aerosol transport efficiency. Figure 3.2 Transport efficiency of ambient air particles from the inlet to the LIBS sample chamber as a fimction of the particle diameter.

PAGE 77

64 3.2 LIBS Data Analysis 3.2.1 Collection of Data Four elemental species were targeted for the ambient air monitoring namely aluminum, magnesium, calcium, and sodium. The corresponding atomic emission lines utilized were the 394.40 and 396.15-nm A1 1 lines, the 279.55, 280.27, and 285.2 1-nm Mg I and II lines, the 393.37 and 396.85-nm Ca II lines, and the 589.00 and 589.59-nm Na I doublet. Three different spectral windows were utilized to monitor these species. They were centered at 275, 408, and 590 nm, with each having a spectral bandwidth of approximately 36, 32, and 25 nm, respectively. Aluminum and magnesium were selected because they are used as energetic fuels in many fireworks, while sodium and calcium were selected primarily as controls due to their general prevalence in atmospheric particulates. Ambient air mass concentration measurements of the targeted elements were recorded using a conditional data analysis scheme. During each nominally two-hour sampling period, spectra were recorded using 1200-shot pulse sequences, which correspond to a 4-minute sample for the 5-Hz laser repetition rate. For each spectral window, between 8 and 14 1200-shot sequences were recorded. Spectra that contained significant emission intensity for a target analyte species were identified as particle hits. For each LIBS spectrum, the emission intensity about the expected analyte peak was compared to the emission intensity of an adjacent, featureless continuum spectral region. The expected analyte peak was triggered on the target emission line, which was centered averaged over 9 pixels of the iCCD detector array, and the featureless continuum intensity was the average of continuum intensities called base 1 and base 2, which were averaged over 15 pixels of the iCCD detector array. The spectrum was classified as a

PAGE 78

65 particle hit if the ratio of emission intensities exceeded a threshold value, typically 75 to 150% above the nominal ratio corresponding to the absence of any analyte emission line intensity. These threshold values set the false hit rate to no more than one in a 1200-shot sequence. The designed conditional data analysis scheme was performed in real-time for each 1200-shot sequence, and the identified spectra for all particle hits were stored along with the ensemble-average of all laser pulses. Parameters used in the process of particle detection are listed in Table 3.1. Table 3.1 Parameters for Aerosol Particle Detection Analyte Aluminum, A1 Magnesium, Mg Calcium, Ca Sodium, Na Emission Lines, nm A1 1-394.40 (target), A1 1396.15 (control) Mg 1-279.55 (target). Mg I280.27, Mg II285.21 (control) Ca 11-393.37 (target), Ca II396.85 (control) Na 1-589.0 (target), Na I589.59 (control) Spectral Window 408 nm, ~32 nm bandwidth 275 nm, ~36 nm bandwidth 408 nm, ~32 nm bandwidth 590 nm, ~25 nm bandwidth DelayAVidth 40 ps/ 60 ps 40 ps/ 40 ps 40 ps/ 60 ps 30 ps/ 150 ps Base 1 & 2 400 & 416 nm 270 & 284 nm 400 & 416 nm 584 & 594 nm Laser Pulse Energy 315 mJ 315 mJ 315 mJ 315mJ Hit Threshold 1 50% above no hit case value 85% above no hit case value 100% above no hit case value 75% above no hit case value 3.2.2 Processing of Data After each sampling period, the stored spectra of all identified particle hits for a given analyte species were ensemble-averaged and an equivalent mass concentration was calculated using calibration curves. Typical calibration curves used in this experiment

PAGE 79

66 are presented in Figure 3.3 and a summary of the correlations of mass concentration versus peak-to-base is listed in Table 3.2. Figure 3.3 Aluminum and magnesium calibration curves using the emission lines A1 1 394.4 nm and Mg 1 279.55 nm. Table 3.2 Equations for Calibration of Mass Concentration Analyte Mass Concentration, pg/m^ Mg Target 279.55 nm C = 0MA1(P/Bf + A.\515(P/B) 0.2765 Control 280.27 nm C = 0.0684rP/5/ + \0.U3(P/B) + 8.0581 Control 285.21 nm C = 0.0599fP/S/ + \.6\A(P/B) + 0.1284 A1 Target 394.4 nm C = 0.01837fP/5/ + \0.6Z%7>(P/B) 17.0215 Control 396.15 nm C = 0.00392fP/8/ + 5.69%16(P/B) 0.6707 Ca Target 393.37 nm C = 0.00044fP/SÂ’/ + 02\\%6(P/B) + 1.2966 Control 396.85 nm C = 0.00028fP/fi/ + 0.A55A5(P/B) + 0.50404 Na Target 589.0 nm C = 0.0000563fP/ff/ + 0.37805(P/Bj 1.5396 Control 589.59 nm C = 0.00009 lO^P/S/ + Q.151Z(P/B) -0.8473

PAGE 80

67 The equivalent mass concentration corresponds to the analyte signal strength as calculated using the subset of data identified with the conditional sampling algorithm. As such, the actual mass concentration of the ambient air sample is equal to the product of the equivalent mass concentration and the frequency of identified hits, as expressed by the Equation 1.3 and described in an earlier publication (Hahn et al. 1997). The frequency of hits is the number of spectra identified with the target emission line divided by the total number of laser pulses. It is noted that as the fiÂ’equency of hits approaches 100%, the conditional analysis scheme converges to the traditional ensemble-averaging approach. A short description of the quantitative analysis of LIBS data for particle size and composition measurements is given in the Section 1.6 of this dissertation. The equivalent mass concentration is calculated using a typical calibration curve relating the analyte LIBS signal to a known analyte mass concentration. All calibration curves used in this study were generated with well-controlled mass flow streams as discussed in the previous chapter. The LIBS signal is also defined by Equation 2.8. As explained above, the equivalent mass concentration is equal to the single particle mass divided by the plasma volume. Using this relationship, the equivalent spherical diameter of a single particle may be calculated by the Equation 1.7, which is repeated here D = 6xF ^Pf -il/3 (1.7) where the characteristic plasma volume Vp is estimated to be 2.38 x 10'^ cm^ as it is shown later in Chapter 6. The algorithm discussed above enables the identification of spectra corresponding to individual particles, and the calculation of overall mass concentrations and the

PAGE 81

68 mass/size of individual particles. Two additional controls were added to the data reduction schemes. For mass concentration measurements, a total of two or more emission lines were utilized for each species. Typically, the most intense line (target emission line) was used to trigger the conditional data analysis routine. Then one or two additional lines (control emission lines) were used to calculate the equivalent mass concentration based on the ensemble-averaged spectrum of the identified particle hits. Because the spectral noise is random on a given shot, a high noise spike on the analyte trigger wavelength would not correspond to high noise spikes on the alternative control analysis wavelength or wavelengths. For single-shot analysis, comparing the analyte signal from two different atomic emission lines screens false signals or other spectral irregularities. A spectrum is rejected from subsequent single-shot size analysis if both atomic emission lines (target and control lines) do not yield the same analyte mass to within a factor of two. In other words, spectra are retained for size analysis only if the ratio of analyte masses calculated from two different emission lines is between 0.5 and 2. 3.3 Results and Discussion An advantage of LEBS-based analysis of individual aerosol particles is the ability to determine the composition of constituent elements within a given particle. Several unique particle types were identified, as shown in the following two figures. Figure 3.4 contains two single-shot spectra. One spectrum is characterized by the presence of intense magnesium emission lines only, and the other one features both magnesium and silicon (288.16-nm Si I) emission lines. As discussed below, while the Mg-only spectrum is attributed to a single MgO particle, the second spectrum is recognized as a type of magnesium-silicate particle (i.e., composed MgSiOs or an agglomerate of MgO with a silicate particle). Such magnesium-silicate spectra were rare in comparison to the

PAGE 82

69 magnesium oxide spectra, with the latter accounting for greater than 95% of the recorded magnesium-containing spectra. Figure 3.5 contains two spectra corresponding to aluminum-containing particles, namely one spectrum containing only aluminum emission lines and a second spectrum characterized by the presence of both aluminum and calcium emission lines. Spectra characterized by the presence of both aluminum and calcium emission were less than 1% of the spectra containing calcium emission lines, and were about 10% of the spectra containing aluminum emission lines. Particle size data are presented in detail below, however, the four spectra presented in Figures 3.4 and 3.5 correspond to particle diameters in the range of 0.4 to 1 .0 pm. The above data demonstrate the overall sensitivity of the LIBS technique for single particle analysis, as observed by the relative strength of the various atomic emission lines. 3000 2500 2000 3 nj ^ 1500 (0 c ^ 1000 500 Mg : Mg : U Si ^ 276 278 280 282 284 286 288 290 Wavelength, nm Figure 3.4 Two single-shot LIBS spectra. The lower spectrum represents a single MgO particle. The upper spectrum that has been vertically shifted for clarity represents a single Mg-Sicontaining particle.

PAGE 83

70 Wavelength, nm Figure 3.5 Two single-shot LIBS spectra. The lower spectrum represents a single AlO particle. The upper spectrum that has been vertically shifted for clarity represents a Ca-Al-containing particle. 3.3.1 Mass Concentration A novel feature of laser-induced breakdown spectroscopy is the ability to reject null spectral data while utilizing the infi'equent but signal-rich spectra corresponding to discrete aerosol particles. The enhanced signal-to-noise ratio (SNR) resulting from the conditional analysis-based LIBS monitoring is illustrated in Figure 3.6 for sodium-based particle sampling. The figure presents the spectrum corresponding to the ensemble average of 9600 laser pulses, along with the corresponding spectrum based on the ensemble average of 30 identified particle hits only. The corresponding sodium-based particle sampling frequency is 30/9600 or about 0.31% of laser pulses. The increase in analyte SNR is very significant, with the 9600-shot ensemble-average essentially equal to a non-detectable sodium emission signal. The equivalent mass concentration of the sodium hits spectrum is about 139 parts per billion (ppb) (mass sodium/mass air), while the overall sodium concentration is 139*(0.31/100) or about 0.43 ppb. The nominal

PAGE 84

71 improvement in SNR is about two orders of magnitude for this sodium sampling frequency, which is apparent when comparing spectra in the Figure 3.6. Figure 3.6 The lower Na-LIBS spectrum is the 9600-laser-shot average, and the upper Na-LIBS spectrum is the 30-identified, Nashot average from the 9600 laser shots. Both spectra have the same intensity seale but have been vertically shifted for clarity. Magnesium-containing emission spectra revealed similar enhancement in magnesium emission line intensity with the use of conditional data analysis, as it is also seen in Figure 3.7. Specifically, Figure 3.7 reports the 14-hit averaged spectrum corresponding to an overall magnesium mass concentration of 108 ppt. As illustrated above, the use of a conditional data analysis routine yields speetra with excellent signalto-noise ratios for the observed ambient air particle sampling rates. The spectral data were subsequently used to calculate overall mass concentration measurements of aluminum, calcium, magnesium, and sodium throughout the sample period. The mass concentration of magnesium recorded in ambient air is presented in Figure 3.8 as a function of time.

PAGE 85

72 Wavelength, nm Figure 3.7 Enhancing of the Mg-LIBS signal using the conditional analysis. The 14-Mg hits are a subset of the 9600 laser shots. Both spectra have the same intensity scale but have been vertically shifted for clarity. Date (Month/Day) 2000 Figure 3.8 Mass concentration of magnesium as a function of time. Each data point represents the average LIBS-based concentration over a two-hour sampling period.

PAGE 86

73 The data presented in Figure 3.8 reveal a significant rise in magnesium during the Fourth of July holiday period, considered fi-om July 1 to July 7. The magnesium mass concentration ranged from 0 to 108 parts per trillion (ppt) on a mass basis. It is noted that 1 part per tnllion on a mass basis is equivalent to 1.16 ng/m of ambient air. A mass concentration of zero corresponds to a sampling frequency of zero or an equivalent mass concentration (based on hit spectrum) of less than 10 ppb, the lower detection limit established for magnesium spectra corresponding to the particle hits. The magnesiumbased particle sampling rates were typically 0.1% or less; hence the actual analyte signals corresponding to the Figure 3.8 data were approximately 3 orders of magnitude larger than the overall mass concentration values. The average magnesium concentration during the holiday period was 36.7 ppt, while the average concentrations of magnesium before and after the holiday period were zero and 3.5 ppt, respectively. The nearly 50fold increase in magnesium during the Foiuth of July holiday period in comparison to the average of preand post-holiday concentrations is strongly suggestive that the source of increased magnesium is derived from the discharge of fireworks in the troposphere. The mass concentration data recorded for aluminum during the same sampling period revealed trends similar to the magnesium data. However, the overall aluminum-sampling hit rates were significantly lower than the magnesium sample rate. As such, the aluminum data contained a larger percentage of zero concentration measurement, essentially non-detects. The average aluminum concentration was 45.8 ppt during the holiday period, compared to an average mass concentration of 6.9 ppt for the non-holiday period.

PAGE 87

74 The current findings of increased magnesium and aluminum are consistent with results reported during a similar study that utilized laser desorption/mass spectrometry for the analysis of ambient-air particles (Liu et al. 1997). The study reported significant increases in the levels of ambient air magnesium and other elements attributed to fireworks in the days following the Fourth of July holiday at the University of California in Riverside. Although the total masses of magnesium and aluminum released by fireworks are diluted with a significant volume of ambient air, the current study and the work of Liu et al. suggest that pyrotechnic-derived particulates persist in the troposphere with a time scale on the order of days. For the analytes calcium and sodium, Figure 3.9 shows the analyte mass concentration response during the testing period. In contrast, the mass concentrations of calcium and sodium do not correlate as strongly as the magnesium and aluminum data do for the Fourth of July holiday period. Their concentrations are nearly double during the holiday period. The average calcium and sodium concentrations during the holiday period were 0.21 and 0.65 parts per billion (ppb) by mass respectively, while the preand post-holiday average mass concentrations were 0.14 and 0.25 ppb for calcium and sodium respectively. It is also difficult to determine the contribution of fireworksderived calcium and sodium to the recorded mass concentrations since the presence of these species in ambient air imder normal conditions and their relatively large daily level fluctuations are well known (Laj et al. 1997, and Lee et al. 1999). In addition, for example, the standard deviations of the measured sodium mass concentration data are equal to 0.43 and 0.33 ppb, corresponding to the holiday and non-holiday sampling periods, respectively. The difference between holiday and non-holiday sodium

PAGE 88

75 concentrations (about 0.4 ppb) is within the standard deviations for these two sample sets, making it difficult to conclude any statistical significance to the fluctuations in sodium and calcium over the Fourth of July holiday. Therefore, this increased concentration during the holiday period could be due to a normal fluctuation of the species. 2.5 .Q Q. Q. C o 9 1.5 CD i_ c 0) ^ 1 o O M (/) B 0.5 Ill ' ' 1 " " _ Holiday Period 1 [f r Df n a n* ^ r ^ » » 1 . 1 . T I t I I I I I T T « D _L I I I I I I I I Ca, ppb Na, ppb !• 6/28 7/3 7/9 7/15 7/21 7/26 Date (Month/Day) 2000 8/1 8/7 Figure 3.9 Mass concentrations of calcium and sodium as a function of time. Each data point represents the average LIBSbased concentration over a two-hour sampling period. A summary of the mass concentrations for the analyte studies is reported in Table 3.3. No independent sampling analysis techniques were used to corroborate the reported mass concentration. However, in an early study, independent extractive sampling in accordance with US EPA Method 29 standard yielded excellent agreement for chromium and manganese concentration at 2-3 ppb mass concentrations (Hahn et al. 1997). In aggregate, the current mass concentration data demonstrate the ability of the LIBS technique to measure element-specific concentrations of particulate matter at very low overall mass concentration levels.

PAGE 89

76 Table 3.3 Summary of Analyte Mass Concentration Analyte Magnesium Aluminum Calcium Sodium Detection Limit (on hit spectrum) lOppb 7.5 ppb 2.3 ppb 3.2 ppb Holiday Mass Concentration* 36.7 ppt + 34 ppt 45.8 ppt ±126 ppt 206 ppt ± 239 ppt 650 ppt ± 425 ppt Non-Holiday Mass Concentration* 2.8 ppt ± 5.8 ppt 6.9 ppt ± 22.3 ppt 140 ppt ±127 ppt 254 ppt ± 333 ppt Maximum concentration 108 ppt, July 7 PM 420 ppt, July 7 PM 755 ppt, July 5 AM 1.7 ppb, July 2 PM * Average values ± one standard deviation While the mass concentration data presented above are based on approximately 2hour sampling period, it is worthwhile to examine the data with finer temporal resolution. The sampling frequencies of recorded sodiumand calcium-based particle hits are presented in Figure 3.10. The almost 2-hour sampling period monitored individual 1200shot laser sequences using each sequence a time of 4 minutes. The 4-min sampling frequencies ranged within a factor of two and three with respect to the average sampling rate for sodium-based particles and calcium-based particles respectively. Specifically, in the case of sodium-based particles the frequency varied from 0.25 to 0.92 % when the average during this 2-hour period was 0.49 %, while in the case of calcium-based particles the frequency ranged from 0.1 to 2.4% when the average was 0.7%. The time resolved sampling data presented in Figure 3.10 are typical of the relatively steady nature of particulate matter observed on the time scale of minutes to several hours.

PAGE 90

77 Time Figure 3.10 Sampling frequency over a period of two hours. Each data point corresponds to the frequency of hits for 1200-shot laser sequence. 3.3.2 Particle Analysis In addition to the evaluation of mass concentrations, the spectra corresponding to individual hits were analyzed for subsequent particle size. As outlined in detail previously, the elemental mass is directly calculated from the product of the equivalent analyte mass concentration of a single-shot spectrum and the characteristic plasma volume, 2.38x10'^ cm^. As observed in Equation 1.7, the calculation of a mass-based particle diameter requires the specification of parameters based on the particle type, namely the bulk particle density and the elemental mass fraction of the measured analyte species. For the present study, the magnesium-based particles were modeled as magnesium oxide, MgO, with a density of 3.58 g/cm^ and a magnesium mass fraction of 0.60. MgO was utilized as a model based on several observations. First, the perceived

PAGE 91

78 particle source of magnesium was primarily via combustion-generation during the discharge of fireworks. Second, nearly all of the magnesium particles were characterized by the absence of any recorded silicon or iron atomic emission lines. Third, modeling the particles as sea salt-based particles (Na/Mg is about 7.8 as reported by Mclnnes et al. 1999) was discounted for the magnesium-based particles due to the mass values of magnesium and sodium recorded. Specifically, the average mass of particulate magnesium was 232 fg, while the average mass of particulate sodium was 266 fg, with a corresponding Na/Mg ratio of only 1.15. Although the largest recorded sodium particle mass is consistent with the nearly 8 to 1 Na/Mg ratio, the recorded sampling frequency of the largest sodium particles is inconsistent with the recorded sampling frequency of magnesium particles. The calcium particles were modeled as calcium carbonate, CaCOs, with a density of 2.71 g/cm^ and a calcium mass fraction of 0.40. Calcium in ambient air particulates is mostly composed of calcium carbonate or gypsum (Laj et al. 1997), although CaS 04 particulates from anthropogenic sources may be significant in number (Hoomaert et al. 1996). The sodium particles were modeled as two different particle types, namely sodium chloride, with a density of 2.17 g/cm^ and a sodium mass fraction of 0.39, for particles with a corresponding mass-based diameter less than 1 .6 pm, and sodium nitrate for particles greater than 1 .6 pm. Such a categorization is consistent with recent measurements reported of nitrate-containing particles using an aerosol time-of-flight spectrometer (Liu et al. 2000). Specifically, sodium nitrate particles were limited to the course size mode between 1.6 and 3.5 pm.

PAGE 92

79 3.3.3 Particle Size Distribution The histogram of calculated diameters for magnesium-contained particles recorded during the Fourth of July holiday period is presented in Figure 3.1 1. This figure displays a particle distrihution with mean diameter, standard deviation and modal diameter of 586, 96, and 515 run, respectively. The size distrihution exhibits skewness toward larger particle sizes, as expected for general aerosol populations. The mean particle size of magnesium-based particles recorded after the holiday period was 418 nm, with a standard deviation of 159 nm. The approximately 170 nm difference in mean diameters, about a 29% decrease, between the sizes of these two magnesium-based particle populations is not statistically significant, but nonetheless may be indicative of two different sources of magnesium particles, namely fireworks derived and perhaps marine-derived. Particle Diameter, nm Figure 3.11 Histogram of calculated diameters for magnesiumcontaining particles for the Fourth of July holiday period. The particles were modeled as magnesium oxide (MgO).

PAGE 93

80 In contrast, the calcium and sodium particle size distributions of the non-holiday period revealed essentially no variations with respect to the holiday period. The size distribution for calcium-contained particle (as CaCOa) is presented in Figure 3.12 and 3.13 for the holiday and non-holiday period, respectively. For the holiday period, the mean particle diameter is 685 nm with a standard deviation of 382 nm and a modal diameter of approximately 427 nm. A long tail is extended toward larger diameters, reaching a maximum diameter of 2.4 pm (not showed in the graph) with an equivalent mass of about 175 times that of the modal particle diameter. In the case of non-holiday particle distribution, the mean, the standard deviation, and the modal were basically the same with a variation within 5 % of the particle values recorded during the holiday period. Particle Diameter, nm Figure 3.12 Histogram of calculated diameters for calciumcontaining particles for the Fourth of July holiday period. The particles were modeled as calcium carbonate (CaCOa).

PAGE 94

81 20 — I — I — I — I — I — I — I — I — I — I — I — I — I — I — I — I — I — I — r Particle Diameter, nm Figure 3.13 Histogram of calculated diameters for calciumcontaining particles for the non-holiday period. The particles were modeled as calcium carbonate (CaCOs). The histogram of sodium particles as NaCl is presented in Figure 3.14 for the holiday period. The mean and modal particle diameters are approximately 844 nm and 698 nm respectively, with a standard deviation equal to 274 nm. As with the calcium results, the non-holiday associated sodium particles were consistent with the size distributions recorded during the holiday period. Specifically, the mean and modal diameter changed within 5 % and the standard deviation varied about 15 % respect to the particle values corresponding to the holiday period. As discussed above, sodium-based particles larger than 1 .6 pm were modeled as sodium nitrate. The sodium nitrate particles accounted for less than 5 % of the sodium-based particle hits recorded, and were characterized by an equivalent size that ranged between 1.6 and 4.1 pm. A summary of the finding in this section is listed in Table 3.4.

PAGE 95

82 Particle Diameter, nm Figure 3.14 Histogram of calculated diameters for sodiumcontaining particles for the Fourth of July holiday period. The particles were modeled as sodium chloride (NaCl). Table 3.4 Summary of Particle Size Analyte Magnesium Calcium Sodium Particle Model MgO CaC03 NaCl Holiday Period: Number of Particles Mean ± IStDev Modal 41 586 nm +196 nm 515 nm 155 685 nm ±382 nm 427 nm 339 844 nm ± 274 nm 698 nm Non-Holiday Period: Number of Particles Mean ± IStDev Modal 17 418 nm ± 159 nm 108 706 nm ± 363 nm 460 nm 206 822 nm ± 229 nm 720 nm Minimum Diameter 219 nm 196 nm 412 nm It is also useful to discuss the absolute detectable analyte masses and the relative sensitivity of the LIBS technique for the species analyzed. The smallest diameters reported in Table 3.4 represent the minimum detected particles with mass of magnesium,

PAGE 96

83 calcium, and sodium of 12, 4, and 31 femtograms, respectively. Because the probability is low that the actual ambient air size distributions shown in Figure 3.11 to 3.14 naturally ended at these points, these minimum detected sizes are considered representative of the analyte mass detection limits. It is noted that particle sizes on the order of tens of nanometers may exist for select particle types (i.e., metallurgical fumes and combustion derived). At present these nanoparticles were not detected using LIBS as implemented here. An additional parameter that should be noted is the rate of acceptance of the singleshot spectra. As detailed above, the criteria for analysis of a single-shot spectrum was that the calculated analyte mass was consistent (within a factor of two) for two atomic emission lines. The retention rate for the spectra of sodium-based particles was 90%, and the retention rate was 63% and 50% for the spectra of calcium and magnesium-based particles, respectively. These rates reflect a number of factors, including the overall LIBS sensitivity to each analyte emission line, the signal-to-noise ratio in the various spectral regions, and the specific nature of the emission line pairs (e.g., neutral line to ionic line as with magnesium). 3.4 Summary The LIBS technique was successfully used to monitor ambient air particulate containing a number of species for a six-week sampling period spanning the Fourth of July 2000 holiday period. The implemented conditional analysis functioned to increase the sensitivity in the determination of overall mass concentrations up to tens of parts per trillion, and to identify spectra corresponding to individual aerosol particles with sizes of hundreds of nanometers. Changes in mass concentration for metallic species associated with the discharge of fireworks of the Fourth of July holiday period, such as magnesium and aluminum, were significant and increased by about one order of magnitude with

PAGE 97

84 respect to the non-holiday period. In contrast, recorded ambient air concentrations of sodium and calcium revealed no significant correlation. Analysis of single particles yielded composition-based aerosol size distribution with diameters from 200 nm to 4 pm. The absolute mass detection limits for single particle analysis was in the order of tens of femtograms for magnesium-, calcium-, and sodium-based particles. Overall, LIBS-based analysis of ambient air aerosols is a promising technique for the challenging issues associated with real-time collection and analysis of ambient air particulate matter data. However, it is still needed to understand issues regarding to plasma-particle interactions for the development of LESS as a deploying technique. The next chapter addresses the issue of single-shot variations in atomic emission spectra in consideration with plasma properties.

PAGE 98

CHAPTER 4 SAMPLING STATISTICS FOR SINGLE SHOT ANALYSIS Most of the research addressing LIBS sensitivity and precision is based on traditional ensemble averaging of spectra (including ensemble of filtered spectra) as a means to overcome the extensive spectral fluctuations observed on a laser shot-to-shot basis. However, it is not readily apparent as the applicability of these studies to singleshot LIBS analysis. New questions regarding shot-to-shot variability and precision are raised with the specific application to single-shot aerosol analysis. This section is focused on the precision of LIBS-based single-shot analysis of gaseous and aerosol systems due to different laser pulse energies. A statistical analysis of fluctuations of single-shot spectral data in both atomic emission and plasma continuum emission were investigated together for a homogenous gaseous flow. The ubiquitous nature of carbon dioxide in ambient air provided the analyte source for this homogeneous gas flow. In addition, shot-to-shot fluctuations in plasma temperature are reported based on iron atomic emission in an aerosol-seeded flow, which were analyzed at optimal temporal delay for each pulse energy value. 4.1 Experimental and Data Processing Considerations Single-shot experiments were conducted using the 247.86-nm carbon I atomic emission line. The source of carbon was carbon dioxide in the purified ambient air stream, which provided a homogeneous carbon source (~100 ppm) at the molecular level. Such uniform dispersion of the analyte species on a shot-to-shot level is not readily achieved by the introduction of aerosol species using the nebulizer. All experimental 85

PAGE 99

86 data sets were recorded in time intervals no longer than two hours to minimize fluctuations of CO 2 in the compressed air stream. A signal integration time of 5 ps was used for most experiments and the delay time optimization with pulse energies is presented in the next section. Single-shot plasma temperature measurements were calculated from Boltzmann plots using 10 iron atomic emission lines in a single spectral window centered at 270 nm. The process of plasma temperature calculation was already described in Section 2.8, and the Fe II emission lines were listed in Table 2.4. During the processing of single-shot spectra for temperature measurements, all spectra were first smoothed using the SavitzkyGolay algorithm (Savitzky and Golay 1964, and Madden 1978) and then, individual spectra were selected according to a filtering algorithm. The filtering algorithm is based on the ratio of the integrated emission lines that defines the following parameters: A as the ratio of the peak at 259.9 nm to the peak at 258.58 nm, B as the ratio of the peak at 272.75 nm to the peak at 271 .44 nm, C as the ratio of the peak at 277.93 nm to the peak at 271 .44 nm, and D as the ratio of the peak at 278.36nm to the peak at 271 .44nm. Based on examination of ensemble-averaged spectra, the above parameters must satisfy the following criteria: i) the peak at 258.58 nm larger than 10000 counts, ii) A between 2.5 and 5, iii) B between 0.8 and 1.8, and iv) C and D larger than 0.1 and their ratio D/C between 1/3 and 3. Nominal values for these parameters based on 1200-shot ensemble average spectra varied as follow: peak at 258.58 nm from 17000 to 20000, A from 3.5 to 3.8, B from 1.09 to 1.12, C from 0.12 to 0.14, D from 0.15 to 0.19, and D/C from 1.2 to 1.4.

PAGE 100

87 4.2 Energetic State of the Laser-Induced Plasma The primary goal of the present study is to assess the precision obtained during single-shot spectral analysis of LEBS data, noting any variability of precision with laser pulse energy. However, keeping in context the pulse-to-pulse variability related to the breakdown process discussed in Section 2.6 (100% breakdown frequency about 190 mJ per laser pulse), the transmitted energy thru the plasma (or absorbed energy) is significant and is important at laser pulse energies above 100% breakdown frequency. Therefore, a more careful analysis of the absorbed energy by the plasma was performed first. Figure 2.17 is re-plotted in a more convenient form in Figure 4.1 that shows the energy deposited into the plasma, expressed as percentage of incident pulse energy, for purified air as a function of the pulse energy. At pulse energy of 192 mJ, which ensures 100% breakdown frequency, the plasma absorbs 48% of the incident pulse energy. Above this pulse energy, the percentage of absorbed energy increases steadily to a maximum value of approximately 60% for pulse energies higher than about 250-260 mJ. This apparent saturation effect was observed first by Radziemski et al. (1983), who reported a transmission of 5% (95% absorption) at 300 mJ, and later by Chen et al. (2000) who obtained an absorption of 90% for incident energies from 45mJ to 80mJ (13.4 mJ energy threshold, fr5,and 6.5-ns pulse width). Radziemski et al. explained that after sufficient energy is deposited in the plasma to ionize nearly all of the matter, the additional energy tends to expand the size of the plasma rather than increasing the temperature or electron density. The current experiments suggest a saturation pulse energy of about 255 mJ, with a corresponding energy density of 0.18 kJ/cm^ (based on a plasma volume of 1.44x10'^ cm^ as calculated in Chapter 6). This value is one order of magnitude lower than the energy density of 5 kJ/cm^ reported by Chen et al. (2000).

PAGE 101

88 Figure 4.1 Percentage of incident pulse energy absorbed by the laser-induced plasma as a function of laser pulse energy. Error bars represent ± one standard deviation. The pulse energy deposited into the plasma should be considered in the context of two additional factors, namely spherical aberration and optical thickness. Parigger et al. (1997) conducted a theoretical study and concluded that spherical aberration due to a single plano-convex lens with monochromatic light can induce a large spot size with multiple breakdown sites along the optical axis, thereby increasing the transmission of energy and reducing overall absorption. Borghese et al. (1998) determined an average optical thickness of 0.8 in the plasma during the deposition of the second half of the incident pulse energy. Using this value and the moving breakdown model developed by Docchio et al. (1988), an estimated value of 66% for the absorption of pulse energy was calculated, in very good agreement with the current experimental value of 60%. The correlations between plasma characteristics and laser pulse energies, noting that saturation corresponds to about 255 mJ, form the basis for the analysis of the following sections.

PAGE 102

89 4.3 Time Delay Optimization with Pulse Energy It is well known that the atomic emission signal strength of a given analyte species in a laser-induced plasma is a function of both laser pulse energy and of the temporal signal integration. To provide a valid comparison of carbon emission at different pulse energies, the carbon emission signal was optimized temporally for different pulse energies. The process of delay time optimization for fixed pulse energy was explained in Section 2.5. Figure 4.2 shows the result of this optimization process for only three pulse energies, 192 mJ, 247 mJ, and 335 mJ using the carbon emission line C I 247.86 nm. A similar trend is observed in all three plots, namely a parabolic-like profile with a clear maximum. Such behavior has been widely reported, and is the result of differing rates of decay between the continuum and atomic emission processes. Figure 4.2 Time delay optimization at fixed 2-ps time width for the maximization of the P/B using the Carbon emission line C I at 247.8 nm.

PAGE 103

90 Similar plots to that of Figure 4.2 were construeted for each of seven laser pulse energies used in this study. The data, delay times corresponding to the maximum LIBS signal P/B, were used to construct a plot of the optimal delay time as a function of laser pulse energy. This plot is presented in Figure 4.3, and a second-order fit (y = -0.802549 + 0.119246x 1.66*10'‘*x^, = 0.99811) of the data enabled selection of the optimal delay time for each pulse energy for carbon analysis for all subsequent experiments. As discussed in detail below, the plasma temperature was found to be essentially constant (1 1200 K) at each optimal point, which is consistent with the plasma physics, whereby the plasma temperature dictates the continuum emission and the populations of the various atomic transitions when the plasma behaves as in local thermodynamics equilibrium, hence the P/B ratio. All subsequent experiments were performed at laser pulse energies over 190 mJ to ensure complete breakdown for all laser shots, as it was found and discussed in Section 2.5. Pulse Energy, mJ Figure 4.3 Optimal delay times for the spectral carbon emission line (P/B) at 247.8 nm as a function of laser pulse energy. The solid line is a second-order curve fit.

PAGE 104

91 4.4 Single-Shot Plasma Emission Fluctuations Corresponding to the respective optimal delay times, the integrated carbon peak (P) and the continuum base intensity (B), along with the corresponding peak-to-base (P/B) and signal-to-noise (SNR) ratios as a function of pulse energy are plotted in Figure 4.4. All data points represent the calculated value of a single 100-shot average spectrum. As observed in the figure, all four parameters increase with increasing pulse energy. When the pulse energy was increased from 200 to 344 mJ, the carbon emission peak and continuum intensity increased by 150% and 100%, respectively, while the carbon P/B and SNR increased by 20% and 60%, respectively. In addition, the rate of increase with pulse energy was not constant, but rather the parameters are characterized by a more marked increase up to pulse energies of about 250 to 275 mJ, which coincides with the initiation of the saturation effect identified in Section 4.2. 5 CD lU 3 Q. CO z 5 O Q) C Figure 4.4 Carbon peak emission (P), continuum emission (B), carbon peak-to-base (P/B), and carbon signal-to-noise ratio (SNR) as a function of laser pulse energy. Each data point represents the calculated value of a single 100-shot average spectrum.

PAGE 105

92 While the P/B and SNR ratios based on ensemble averages are widely used for optimization with LIBS-based analysis, the extension of the LIBS technique for singleshot analysis needs consideration of shot-to-shot fluctuations to enhance both accuracy and precision. From the ensemble averaging approach, the use of high pulse energy is rewarded by a decent increment of the P/B and the SNR; however, the same approach is disappointing by losing information regarding the shot-to-shot behavior at any level of pulse energy. To explore such an effect, the shot-to-shot variability over 100 laser pulses of the carbon emission peak, continuum intensity, and carbon P/B is examined in Figure 4.5 for pulse energies of (a) 200 mJ and (b) 344 mJ. Several observations are made with respect to the two data sets observed in Figure 4.5. Careful inspection reveals a similar fluctuation pattern between the carbon atomic emission and continuiun emission intensities. This scaling is observed in the P/B ratios, which vary by less than a factor of two between the maximum and minimum values in Figure 4.5 (a), while the atomic emission and continuum values vary by more than a factor of 6 and 4, respectively. Of greater significance, the fluctuations in the carbon emission peak, continuum emission, and P/B ratios are decreased with an increase in pulse energy. This is readily observed by comparing Figures 4.5 (a) and (b). The relative standard deviation (RSD) for the atomic emission and continuum emission intensities are 29% and 21%, respectively, at 200 mJ, and decrease to 1 1% and 9.5%, respectively, at 344 mJ. Similarly, the RSD of the peak-to-base ratio decreased from a value of 11% at 200 mJ to 6% at 344 mJ. In addition to elucidating the effects of pulse energy on shot-toshot precision, as discussed in detail below, these results demonstrate the usefulness of processing LIBS data based on the normalized peak-to-base ratio. While the atomic

PAGE 106

93 emission peak undergoes intense fluctuations, the normalized P/B ratio exhibits a significantly less variability, which is desirable for single-shot analysis. I I 9 E TJ o3 B> c Figure 4.5 Laser shot-to-shot variability of the peak emission (open circles), continuum emission (solid squares), and P/B (solid diamonds) for (a) 200 mJ laser pulse energy, and (b) 344 mJ laser pulse energy. Data correspond to the optimal time delay at different pulse energies. In Figure 4.6 the effect of laser pulse energy on the precision of the LEBS signal, as measured by both the peak-to-base ratio (P/B) and the signal-to-noise ratio (SNR), is illustrated. The relative standard deviation (RSD) of 100 single-shot measurements of

PAGE 107

94 both the carbon P/B and SNR values decreases with increasing laser pulse energy. Moreover, the rate of decrease from 200 to about 255 mJ was significantly higher than for pulse energies in excess of 255 mJ. In fact, the RSD of the peak-to-base values is essentially constant above 250 mJ. As presented above, a pulse energy of about 255 mJ corresponds to the saturation value for absorption of pulse energy by the plasma. The results suggest a nahiral breakpoint at the satmation energy, resulting in a more repeatable, hence more precise, LIBS analyte signal for these laser pulse energies. As such, a greater robustness for quantitative single-shot LIBS measurements is realized after the plasma achieves a saturation condition. Figure 4.6 Precision of the LIBS signal, expressed as P/B ratio and relative standard deviations (RSD) of P/B and SNR as a fimction of laser pulse energy. Each data point represents the average of 100 single-shot calculations. Error bars represent ± one standard deviation. 4.5 Single-Shot Plasma Temperature To gain insight into the processes involved in the behavior discussed above, a series of single-shot plasma temperatures (based on Fe II emission) were calculated from

PAGE 108

95 individual spectra for varying laser pulse energies. Due to the high degree of signal noise associated with single-shot spectra, individual spectra were first smoothed by the Savitzky-Golay algorithm and then sorted using a filtering algorithm to reject irregular spectra. The Savitzky-Golay algorithm is a type of weighted-linear smoothing technique in which the noise is minimized by considering the real intensity at a specific location depending on the intensities before and after that location. The selected algorithm used seven discrete values to calculate the new value and it was applied ten times; details of the implementation of this algorithm are given in Appendix C. The application of this algorithm yielded a change in the integrated Fe-II emission lines of no more than 2%, and more importantly increased the sensitivity to detect the smaller emission lines (e.g., Fe II at 277.93 nm and 278.37 nm). Figure 4.6 presents the effect of the smoothing algorithm on an originally noise spectrum. Figure 4.7 Effect of the smoothing algorithm on the Fe II atomic emission lines selected (indicated by arrows) corresponding to a single-shot spectrum. The original spectrum was shifted for clarity.

PAGE 109

96 In addition to the application of the smoothing algorithm, a filtering algorithm was implemented to screen out irregular spectra. Irregular spectra may arise from different sources, such as weak ionization of the plasma, a small amount of targeted analyte in the plasma volume, and the non-optimal collection of plasma emission. It is noted that iron was added in the form of an iron-based aerosol, as described in Chapter 2, hence the homogeneity of iron atoms is greatly reduced as compared to the gaseous carbon source. The weak ionization of the plasma can result fi'om the absorption of laser energy by larger particles prior to reaching the focal spot; this induces a pre-breakdown that disrupts a normal energy-matter interaction at the nominal breakdown point. Because the analyte is derived from discrete iron particles, some plasma volumes may be characterized by weak analyte emission due to statistical fluctuations in the spatial distribution of iron particles. In the limiting case, no detectable quantity of the analyte may be present in a particular plasma volume. Using nominal values based on 1200-shot ensemble averages, a filtering algorithm was developed as described above. The filtering algorithm resulted in the rejection of about 60 to 70% of the single-shot spectra. Nonetheless, the average plasma temperatures based on single-shot measurements were in excellent agreement (within 10%) with plasma temperatures calculated directly from 1200-shot ensemble-averaged spectra, although a slight decrease in agreement was noted with increasing laser pulse energy. The mean temperature and the corresponding relative standard deviation based on 100 single-shot temperature measurements are presented in Figure 4.8 as a function of the laser pulse energy. As can be observed, the plasma temperature shows no significant change with increased laser pulse energy. Specifically, the highest and lowest plasma

PAGE 110

97 temperatures were 1 1,500 K (1 1% RSD) and 10,800 K (1 1% RSD), with no statistical difference between these values. The average plasma temperature was 1 1,200 K for the seven laser pulse energies investigated. It is noted that the optimal delay time with respect to the carbon emission P/B ratio was used for each of the laser pulse energies. This supports the discussions above regarding Figure 4.2, namely that the envelope of optimal analyte emission lines for varying signal delays and laser pulse energies corresponds to the condition of constant (optimal) plasma temperature for a given analyte. 16000 14000 ^ 12000 2 1 10000 E 8000 (0 6000 E g 4000 2000 0 200 225 250 275 300 325 350 Pulse Energy, mJ Figure 4.8 Plasma temperature and RSD of the plasma temperature as a function of laser pulse energy. Each data point is the average single-shot temperature calculations, and is based on iron II emission. Error bars represent ± one standard deviation. I I I I I I I I I I I I I I I T I i I ' I I I I I I I I I I Plasma Temperature T40 I I 30 7 } W D 20 ^ RSD 10 A final observation regarding Figure 4.8 is that the variability of the plasma temperature, based on the relative standard deviations, was found to be approximately constant as a function of laser pulse energy, with an average value of 12.8%. This stands in contrast with the behavior observed with the carbon emission parameters, which

PAGE 111

98 showed a marked decrease in variability with the onset of plasma saturation (see Figure 4.5), as based on absorption of incident laser energy. This difference between shot-toshot variations in atomic emission and plasma temperature offers insight into the overall precision of single-shot measurements. The temperature measurements make use of a Boltzmann plot, which depends on the relative intensities of the integrated iron atomic emission peaks. In contrast, quantitative analyte measurements (as defined in this dissertation) utilize atomic emission peak-to-base ratios, which depend on both the atomic emission and continuum emission signals. The actual continuum and atomic emission signals correspond to a spatial integration over almost the entire plasma (as is discussed in Chapter 6), as coupled to the spectrometer/detector system using appropriate collection optics. The plasma itself is not completely homogeneous, and necessarily has property gradients due to boundary effects and its transient nature. Continuum emission and analyte atomic emission processes are non-linear processes, as observed in Figure 4.2; hence it is fully expected that these two processes will differ relative to each other (i.e., peak-to-base ratios) throughout the plasma. In addition, theoretical description of the line-to-continuum ratio demonstrates the non-linear nature of this parameter with regard to plasma temperature, as discussed in a recent paper (Gomushkin et al. 1999). Accordingly, spatial variations in the position of the plasma will change the overall coupling of the plasma emission into the collection optics, thereby changing the effective spatial integration. Such changes are expected to be manifest in the resulting peak-to-base ratios, hence as the shot-to-shot variations observed in the data. In contrast, atomic emission intensities of a specific analyte are expected to be more robust with respect to spatial integration simply because the

PAGE 112

99 coupling of two non-linear processes is avoided. Moreover, variations in absolute signal level are inconsequential for temperature measurements due to the logarithmic nature of the Boltzmaim plot. While no quantitative measurements of plasma spatial stability were performed in the present study, it was readily observed that the plasma exhibited more variation in axial position at near-threshold pulse energies. It is further noted that the relative standard deviations of the shot-to-shot carbon P/B ratios and the iron-based plasma temperature values are 6 % and 12.8%, respectively, in the saturation regime. Hence these values may be interpreted as the inherent precision of the current LIBS system, encompassing intensifier and detector shot-noise as well as laser pulse energy stability. The additional variability (i.e., greater RSD) realized at sub-saturation pulse energies most likely represents additional spatial variability, manifest through different plasma emission spatial integration as discussed above. In concert, all experiments lead to the suggestion that the reported decrease in plasma variability, hence enhanced precision, obtained in the plasma saturation regime, is due to a decrease in shot-to-shot spatial variability of the plasma. It is therefore concluded that single-shot LIBS based measurements should be made with sufficient laser pulse energy to achieve saturation with respect to absorbed pulse energy, as well as made with a suitable collection geometry (e.g., backscatter mode) to minimize spatial variability.

PAGE 113

CHAPTER 5 ASSESSMENT OF THE UPPER PARTICLE SIZE LIMIT OF VAPORIZATION USING THE LIBS TECHNIQUE Laser-induced breakdown spectroscopy (LIBS) has been widely used as an analytical technique for the quantitative analysis and monitoring of a wide range of aerosol particles, as surveyed in Chapter 1. Specifically, the feasibility for detection and quantification of single ambient aerosol particles was explored in Chapter 3. Results of this study set most of the calculated particle diameters in the range from 200 nm to 2 pm and a small percent of particle size above 2 pm (reaching diameters up to 4 pm). However, in spite of the significant body of research regarding LIBS for aerosol analysis, to date no research has systematically addressed the fundamental assumption inherent in all quantitative LIBS measurements, namely the assumption of complete breakdown and vaporization of all analyte species that comprise the aerosol particles of interest. The important question remains as to what is the upper particle size limit of complete dissociation and vaporization of individual particles suspended in a gas stream? 5.1 Experimental and Data Processing Methodology Aerosols from silicon aqueous solutions (ICP-grade standard) and monodisperse silica particle suspensions (spherical Si 02 particles with diameters ranging from 1.0 pm to 5.1 pm) in deionized water were generated using the experimental facilities described in Chapter 2. Using a laser pulse energy of 320 mJ, a LIBS calibration curve was made from nebulized aqueous silicon solutions. Single-shot conditional analysis was performed from nebulized silica particle suspensions to record spectra corresponding to 100

PAGE 114

101 single silica microspheres. For the case of the monodisperse silica particle suspensions, the particle concentrations in solution (see Table 2.3) were adjusted using ultra-purified water dilution to produce aerosol densities of about 45 particles/cm^ in the LIBS sample chamber, thereby promoting a low sample rate and single particle detection. LIBS-based analysis for all experiments used the neutral silicon atomic emission line at 288.16 nm to define the processed LIBS signal, referred to as the peak-to-base (P/B) ratio. The continuum intensity was interpolated using the adjacent, featureless continuum emission intensity on both sides of the silicon emission line. For temporal signal integration, the 288.16-nm Si I P/B was optimized using a 35-ps time delay with respect to the incident laser pulse and an integration time of 5 ps. Single-shot conditional analysis was used to identify and analyze individual spectra corresponding the various sized monodisperse silica particles. The conditional data analysis routine was reported previously (Hahn and Lunden 2000, and Hahn et al. 1997) and summarized in Section 1.5 of this dissertation. This routine entails the identification of individual LIBS spectra corresponding the presence of discrete particles within a given plasma volume based on the targeted analyte atomic emission signal exceeding a predetermined threshold value. For the current study, the threshold value for the 288.16-nm silicon emission peak was set to obtain an average of 3 false hits (i.e., hits recorded for the nebulization of purified water) for every 1000 laser shots for nearly all experiments. To assess the effect of the threshold value, additional experiments were performed for analysis of the 1.0-pm silica particles using a more relaxed threshold value such that about 30 false hits were recorded for each 1000 shots. For all spectral data processing, individual spectra were smoothed using the Savitzky-Golay algorithm (as

PAGE 115

102 implemented in Appendix C) and subsequently processed using the 288.16-nm silicon emission line profile as discussed below. 5.2 Vaporization of a Collection of Nanoparticles The nebulization of the aqueous silicon solutions produces well-dispersed nanoparticles in the LIBS aerosol sample chamber following droplet evaporation. As characterized in Chapter 2, the aerosol generator produces a particle number density on the order of 10 cm' and particle mean diameters varying from 50 to 1 10 nm (based on Si 02 particles) for aqueous silicon concentrations ranging from 1000 to 10000 pg/ml, respectively. At these conditions, the first set of experiments was designed to quantify the response of the LIBS signal at 288.16-nm silicon line for a range of known silicon mass concentrations coming from a collection of nanoparticles. Figure 5.1 shows this signal quantification, which in other words is the resulting silicon calibration curve. Figure 5.1 Peak-to-base (P/B) ratio of the 288.16-nm silicon emission line as a function of silicon mass concentration for a well-disperse aerosol stream of silicon-based nanoparticles generated by nebulization of aqueous silicon standards. (rM.999).

PAGE 116

103 It is noted that for the plasma volume realized in the present study (measured in Chapter 6 as an approximately 1 .4 mm equivalent spherical diameter), there were on the order of lO'* silicon nanoparticles per plasma volume. As such, the resulting data represent the idealized LIBS-based analyte response corresponding to a highly homogenous aerosol gas stream. The silicon calibration curve is characterized by a linear correlation (R^ ~ 0.999) between the 288.16-nm silicon emission signal (P/B) and the known silicon mass concentration over the range from 0 to 17300 pg/m^. Each data point represents the average of 6 individual 1000-shot ensemble-averaged spectra, with error bars equal to one standard deviation. The high degree of linearity over the full range of silicon concentrations confirms the absence of self-absorption for concentrations as high as 17300 pg/m^, or about 14.7 parts per million on a mass basis. This upper limit was the highest concentration investigated (corresponding to the most concentrated standard silicon aqueous solution used for this study) and does not imply the onset of selfabsorption above this value. A final observation with respect to the linearity of the calibration curve shown in Figure 5.1 states that thousands of Si02 nanoparticles of up to 1 10 nm in average diameter and with individual masses in the order of femtograms (10Â’'^ g), which are formed at the highest aqueous mass concentration investigated, are completely vaporized by the interaction with laser beam and/or plasma volume. In the case of the maximum mass concentration investigated (17300 pg/m^ at the sample chamber), an average global silicon mass of 250 ng is vaporized within the plasma volume (1.44x10' cm ). If hypothetically all this vaporized mass were gathered in one single particle, the equivalent Si02 particle diameter would be 8 pm. As a consequence, the calibration curve suggests

PAGE 117

104 that a virtual Si 02 particle of 8 pm in diameter is a candidate for complete vaporization and linear analyte response when a laser pulse energy of 320 mJ is used and if the total silicon mass is the only limiting factor. 5.3 Vaporization of a Single Silica Particle To investigate the upper size limit for complete particle vaporization, single-shot spectra corresponding to individual silica particles were collected and quantitatively analyzed for a range of the monodisperse silica particle diameters. It is noted that singleshot LIBS spectra display considerable spectral noise as compared to ensemble-averaged spectra; hence it is desirable to improve the signal-to-noise ratio for quantitative analysis of spectra corresponding to the relatively small silica microspheres. 4000 3 3000 n s (0 g 2000 C 1000 0 280 285 290 295 300 Wavelength, nm Figure 5.2 Single-shot spectra corresponding to a single 2.1-pmdiameter silica microsphere as collected and following application of the Savitzky-Golay smoothing algorithm. The smoothed spectrum has been shifted vertically for clarity. As it was used before for single-shot spectra in plasma temperature calculations, the Savitzky-Golay smoothing algorithm was used again to suppress the level of random

PAGE 118

105 shot-noise for each spectrum identified as corresponding to a silicon particle “hit” using the conditional analysis routine. This algorithm was demonstrated to have a negligible effect on the LIBS analyte signal (about 1% change in the silicon emission line P/B ratio) as applied utilizing a filter template width greater than the line-width of the targeted atomic emission peak. An example of the Savitzky-Golay smoothing algorithm effect is shown in Figure 5.2, which presents a single-shot spectrum corresponding to a 2.1micron silica microsphere, along with the same single-shot spectrum following application of the smoothing algorithm. After all single-shot spectra were smoothed, a spectral filtering algorithm was used to screen out false hits and any other anomalous spectra resulting from the discrete sampling of aerosol particles. As noted above, the conditional analysis threshold value was set to allow the detection of a small number of false hits, that is spectra that contain no actual silicon emission. This was accomplished by setting the threshold in a purified air stream with the nebulization of only ultrapurified water. The presence of a small number of false hits ensures that the single-shot detection threshold is sufficiently low to enable identification of actual analyte hits at signal levels approaching the upper single-shot noise limit realized for a given spectral location. The false hits were subsequently filtered out using the following approach. The filtering algorithm is based on the similarity of the silicon emission line profile for a given single-shot spectrum as compared to emission line profile corresponding to the ensemble-average of thousands of individual spectra recorded for the aqueous silicon standard solutions as discussed above. This filtering algorithm resulted in the rejection of approximately 60% of all identified particle “hits,” consistent for all silica microsphere particle sizes, for the trigger threshold

PAGE 119

106 corresponding to an average of three false hits (i.e., no silicon present) per 1000 shots. When the relaxed conditional analysis threshold criteria was used, namely 30 false hits per 1000 shots, the data rejection rate scaled exactly with the tenfold increase in false particles hits, thereby demonstrating the relative independence of the final data set on the exact threshold value used for hit detection; additional comments and experiments regarding the conditional analysis threshold are provided in Chapter 6. Overall, the spectral data processing results were in excellent agreement with the expected outcome based on the estimated silica particle loadings, plasma volume sampling rates, and conditional sampling threshold values. Figure 5.3 Ensemble-averaged spectra corresponding to individually detected monodisperse silica microspheres with diameters of 1.0, 1.5, 2.1, and 2.5 pm. Single-shot spectra were recorded for each of the monodisperse silica particle suspensions, namely for particle diameters equal to 1.0, 1.5, 2.1, 2.5, 3.0, 4.5, and 5.1 pm. Using the spectral data processing schemes outlined above, all single-shot spectra were processed and a final ensemble-averaged spectrum was calculated for each silica

PAGE 120

107 particle size comprised of the individual single-shot spectra accepted after all data filtering. Representative ensemble-averaged spectra for the 1.0, 1.5, 2.1, and 2.5-pm diameter silica microspheres are shown in Figure 5.3. Lastly, the P/B of the 288.16-nm silicon line was calculated for each final ensemble-averaged spectrum. In Figure 5.4, the corresponding P/B for the different silica particle sizes are plotted as a fimction of the cube of the particle diameter, noting that the cubed diameter is proportional to the particle mass. The most significant result of the Figure 5.4 data is the clearly linear relation between the silicon P/B ratio as a fimction of diameter-cubed for the smallest three silica particle diameters (1.0, 1.5, and 2.1 pm), and the abrupt deviation from this linear trend for the particle diameters larger than 2.1 pm. Figure 5.4 P/B at Si 1 288.16 nm for ensemble-averaged spectra of individually detected monodisperse silica microspheres as a fimction of the cube of the silica particle diameter. The continuous line is a linear fit of the first three data points. For complete silica particle dissociation and vaporization, the resulting analyte signal (i.e., P/B) should scale as the analyte mass contained in the particle, hence as the

PAGE 121

108 diameter cubed. Accordingly, the break point in the Figure 5.4 data represents the upper size limit in which an aerosolized silica particle is completely vaporized in the laserinduced plasma. The limiting particle size of 2.1 pm is significantly below the fi'equently used 10-pm-particle diameter limit, and below the 8-pm virtual particle diameter estimated fiÂ’om the calibration curve. This limiting 2.1-pm-particle diameter must be taken into consideration for LIBS-based analysis of aerosol systems. 5.4 Laser-Particle or Plasma-Particle Vaporization In view of the well-defined upper size limit observed for the current study of silica particles, it is useful to discuss two fundamental phenomena that may play key roles in the dissociation and vaporization of the single aerosol particle, namely laserparticle interactions and plasma-particle interactions. The first step in understanding the possible mechanisms responsible for an upper size particle limit is determining the relative importance of particle vaporization as a result of direct interaction with the incident laser pulse (i.e., laser-particle interaction) as compared to particle vaporization within the laser-induced plasma subsequent to the optical breakdown initiation (i.e., plasma-particle interaction). Careful examination of the silica particle sampling rates was performed to further elucidate the contributions of these two effects. The details of these findings are presented in the next chapter. For the present experiments, the laser beam was focused to a spot size measured as approximately 100 pm in diameter. The resulting laser-induced plasma had a volume corresponding to an equivalent spherical diameter of about 1400 pm, as directly measured using transmission of a 532-nm probe beam synchronized to the tail end of the plasma-generating laser pulse. For sampling of discrete aerosol particles, the probability

PAGE 122

109 of a particle being in a specified sample volume (either the spatial region defined by the laser beam or the spatial region defined by the entire plasma volume) can be modeled by a Poisson probability distribution based on the respective sample volume and the aerosol number density, see Equation 1.5. For the silica microsphere suspensions, LIBS-based particle sampling rates were predicted based on either the laser beam sample volume or the plasma sample volume. The predicted particle sample rates based on the plasma sample volume were in excellent agreement when compared to the actual experimental silica particle sampling rates (6.5% and 3%, respectively). In contrast, the predicted sample rates based on the more limiting laser beam sample volume under predicted the observed sample rate by two orders of magnitude (about 0.05%). Based on this assessment, particle vaporization via direct laser-particle interaction can only account for about 1% of the sampled silica particles actually measured; hence the current experiments support the conclusion that plasma-particle interactions drive the dissociation and vaporization of individual aerosol particles. Additional insight to the relevancy of laseror plasma-particle interaction may be gained by visualizing the interaction within the time scale of the laser pulse. Assume the laser pulse follows a Gaussian temporal profile with a full-width half maximum (tq) of 10 ns (e.g., Nd:YAG laser used in this research) and that the pulse energy lasts a total time of 28 ns (i.e., time width of 2.8ro to obtain 99.9% of the total area under the Gaussian profile). At 50% breakdown frequency, nominal pulse energy of 160 mJ was determined in purified ambient air. This energy is represented by the area under the Gaussian profile having the time as abscissa and the instant power as ordinate. Chen et al. (2000) reported based on absorption of about 50% of the pulse energy at plasma energy threshold that the

PAGE 123

plasma is mainly the result of the second half of the laser pulse. Then, the threshold irradiance that triggers the breakdown initiation corresponds to the maximum instant 110 power achieved during the laser pulse. This maximum instant power is 15 MW for a nominal threshold energy pulse of 160 mJ, and it is illustrated in Figure 5.5. Figure 5.5 Gaussian temporal profile corresponding to a pulse energy of 160 mJ given in a total time of 2.8 times the fiill-width half maximum (To). Consider now the presence of a particle in the laser region interaction for a pulse energy of 320 mJ. Having the same medium (ambient air) and the same focal spot (100 pm in diameter), the breakdown threshold irradiance would be the same, and as a result the same instant power of 15 MW would be required to produce the initial optical breakdown and subsequence plasma formation. Figure 5.6 shows the corresponding laser pulse of 320 mJ with the time for breakdown initiation. Considering the definition of the pulse energy, the breakdown initiation appears 9 ns after the initiation of the laser pulse. During this time (i.e., 32% of interaction time with respect to the total laser-pulse time) the phenomenon is purely laser-particle interaction, the pulse delivers about 12% of its

PAGE 124

Ill energy, and the average irradiance is 54 GW/cm^. It is noted that after the breakdown initiation, the process of vaporization lays in the domain of the formed plasma that lasts beyond the laser pulse duration for several microseconds. Figure 5.6 Time and energy for the laserand plasma-particle interaction in pulse energy of 320 mJ. In view of global energy requirements for a silica particle of 2 pm in diameter, the calculated energy to completely vaporize it is about 5x10'^ J, the laser energy available for the particle before the optical breakdown is 1.5x10'^ J, and the calculated absorbed energy by the particle is in the order of 10Â’^ J. It is seen that there is enough laser energy to vaporize the particle, but due to optical properties of the particle there is not enough energy to completely vaporize. Smith (1977) developed a model to calculate the required laser irradiance to vaporize a particle. Using this model irradiances in the order of lO" W/cm are estimated to vaporize silica particles in the range of 0.5-5 pm; irradiances that are an order of magnitude larger than the calculated average irradiance for the period before the breakdown initiation. If for this 320 mJ/pulse the breakdown initiation shifts

PAGE 125

112 up to the maximum instant power (e.g., 30 MW), the average irradiance of this first half of the pulse would reach the order of 10*' required to vaporize the silica particles. Overall, it appears to be a time constraint problem for pure laser-particle interaction. In summary, particle optical properties and time constraint for the pure laserparticle interaction support the previous sampling-based discussion that the vaporization of the particle via laser interaction plays a minor role as compared with the plasmaparticle interaction when particles are sampled by the LIBS technique. 5.5 Particle Vaporization Driven by Laser-Induced Plasma With the concept of plasma-particle interaction as the driving mechanism for particle vaporization, it is useful to examine the vaporized particle mass as compared to the plasma parameters and in consideration of the overall range of linear analyte response as reported above (see Figures 5.1 and 5.4). Spectral analysis of the LIBS signal corresponding to the upper size limit for complete vaporization (2.1 -pm Si02 particle) produced a silicon P/B signal correlating to an equivalent concentration of 1925 pg/m^, as based on the calibration curve shown in Figure 5.1. In contrast, the 2.5-pm sized silica particles, the smallest size examined that is characterized by incomplete vaporization, yielded an equivalent concentration of 21 16 pg/m based on the silicon emission signal. This value is significantly less than the equivalent concentration of ~3340 pg/m^ expected for a linear analyte response corresponding to complete vaporization of a 2.5-pm silica particle. It is noted that the calibration curve based on the high aerosol number density of silicon nanoparticles (~10‘' particle per plasma) yielded a linear silicon analyte response to concentrations of at least 17300 pg/m^. A silicon mass concentration of 17300 pg/m^ corresponds to more than 5

PAGE 126

113 times the silicon mass within a given plasma volume as compared to the mass that would be produced by the complete vaporization of a single 2.5-micron sized silica particle. This behavior is somewhat surprising, and provides strong evidence that the limiting factor for complete particle vaporization is not simply the total heat capacity of the plasma nor the total analyte mass contained within the plasma volume. In fact, it is estimated that complete dissociation and vaporization of a single 2.5-micron silica particle would consume on the order of 0.0005% of the total plasma energy. These results suggest that the plasma-particle vaporization process is controlled by parameters other than global energy conservation within the plasma, and that the consideration of process rates must be important. To better understand the plasma-solid particle interaction, it is useful to further explore the temporal and spatial development of the laser-induced plasma. The laser beam is focused in a region of gas to produce the initial breakdown, which subsequently expands as a plasma to engulf a region much larger than the original volume defined by the actual laser beam. During this process, the plasma starts to fill the focal volume with highly energetic electrons (~60 eV) and ions (Isaac et al. 1998), and to emit UV radiation to the surrounding region. For the extremely high laser beam irradiance characteristic of LIBS, the laser-supported radiation wave provides a means for the propagation of the plasma along the laser beam in the backward direction; however there is no similar photon flux in the radial direction to explain the plasma propagation (i.e., ionization). Alternatively, fast electrons leaving the plasma core can produce the plasma propagation by collision with the atoms/molecules in the surrounding gas (Isaac et al. 1998). Regardless of the exact mechanisms, by the end of the incident laser pulse, the boundary

PAGE 127

114 of the plasma has expanded sufficiently to engulf any nearby aerosol particle. Consequently, the solid particle is bombarded by highly energetic electrons and ions, which are primarily responsible for the transfer of energy (Chen and He 1986). This process subsequently leads to the heating, fusion, vaporization, and ionization of the solid particle. The energy transfer may last up to tens of nanoseconds after the end of the laser pulse, eventually, however, the electrons are unable to deliver the ionization energy required for bond breaking (maximum electron density is reached) and the process then gives way to recombination. Radziemski (1994) reports that in the temporal history of a typical laser-induced plasma, the ion emission lines (electron-ion recombination) may be present on a time scale as early as tens of nanoseconds. It was not found any work related to the interaction of laser-induced plasmas and solid particles; however, the closest case to the present study is given by Smith et al. (1977) who reported the ablation rate of stationary solid polystyrene microspheres in a steady hot plasma (e.g., theta-pinch plasma with electron density about 10*® cm'^ and electron temperature lower then 120 eV). They studied the reduction in size of particles in the range of 100-300 pm in diameters and obtained reductions of up to 8% in their diameters after 2 ps of plasma-particle interaction. Extrapolating their results for similar conditions of electron number density (~10** cm'^) and electron temperature (~3 eV average during the first hundreds of nanoseconds) found in a laser-induced plasma, it is possible to determine that a 12-pm-diameter polystyrene particle could be completely vaporized (i.e., 100% reduction in diameter) in approximately 2 ps. Moreover, considering the scaling rule (that define the ablation rate in the plasma) for the lifetime of the solid particle, x oc r®^^, a 2-pm-diameter particle could be predicted to be vaporized in

PAGE 128

115 ~0.1 |is. Once again, the interaction time between a solid particle and energetic plasma seems to constrain the vaporization process and to be within the first hundred nanoseconds of the lifetime of the laser-induced plasma. Several other factors are proposed that may affect the process of particle vaporization, as manifest in the observed upper size limit for complete vaporization. Due to its transient nature, the plasma itself is a non-homogeneous system in which high temperature gradients are present; hence thermophoretic forces could propel the solid particle out the plasma or to cooler regions, subsequently limiting the vaporization process prior to complete vaporization. Another contributing factor toward incomplete vaporization may be linked to the physical vaporization processes. During the first few nanoseconds of plasma expansion, one side of the particle is subjected to a greater radiant flux and may consequently undergo a faster rate of vaporization. As a result of differing rates of surface evaporation or phase explosion, significant momentum may be transferred to the remaining solid particle, which like thermophoresis may displace the particle to less energetic plasma regions. Finally, vaporization may be limited by the rate of energy transfer (during the first nanoseconds of plasma-particle interaction) from the plasma to the particle. When the heating time is shorter than a characteristic time of mechanical relaxation, the particle does not have time to expand and heating occurs at nearly constant volume. This constant volume heating induces a pressure buildup that can fi'acture and break particles into many smaller pieces (Zhigilei and Garrison 1998), which are readily vaporized within the plasma volume. Such a process favors smaller sized particles characterized by greater surface-to-volume ratios. In summary, most likely the particle vaporization starts within the time scale of the laser pulse and ends

PAGE 129

116 within tens of nanoseconds later, before recombination processes become of significant component of the laser-induced plasma evolution. While factors such as thermophoretic forces and vapor expulsion may influence the vaporization dynamics, additional experimental work and plasma modeling are needed to further determine the exact processes that govern particle vaporization. 5.6 Precision and Accuracy of Particle Sizing Using the LIBS Technique The particle size quantification performed in Chapter Three reported diameters mainly in the range 0.2-2 pm; however, no estimation of the precision and accuracy of these values was stated. The precision and accuracy of the LIBS technique for particle sizing can be evaluated using the single-particle spectra of silica particles collected during the experiment of complete particle vaporization reported above. Knowing the particle properties (e.g., silica bulk density and silicon mass fraction), the characteristic plasma volume (e.g., 2.38 10'^ cm^ as calculated in the next chapter), and the equivalent mass concentration for each single-particle spectrum, the process of sizing or assigning an equivalent diameter to each hit spectra corresponding to a nominal particle size follows using the Equation 1.7. Figure 5.7, 5.8, and 5.9 show the calculated single-shot particle sizes for nominal particle diameters of 1.02, 1.50, and 2.08 pm respectively.

PAGE 130

117 Figure 5.7 Silica particle diameter distributions for a nominal diameter of 1.02 pm. Sample of 47 particles. Figure 5.8 Silica particle diameter distributions for a nominal diameter of 1 .50 pm. Sample of 126 particles.

PAGE 131

118 Figure 5.9 Silica particle diameter distributions for a nominal diameter of 2.08 pm. Sample of 574 particles. The particle distributions in the above figures display a high symmetry with respect to their mean values except in the case of the Figure 5.7, which probably is due to the low number of collected particles and the lower detection limit slightly below 1 pm. In addition, the accuracy of the calculated diameters increases with the increasing nominal diameter (e.g., increase of the LIBS signal P/B) and shows deviations of 13%, 9%, and 1% for the particles of 1.02, 1.50, and 2.08 pm in diameters. With respect to the precision, the value is steadier for all three diameters with a minimum relative standard deviation of 13% for the 1.5-pm particle diameter and a maximum relative standard deviation of 18% for the 2.08-pm particle diameter. These values of accuracy and precision can characterize the LIBS technique for the sizing of particles. Therefore, combining these values, an average uncertainty of less than 20% is determined for each calculated particle diameter.

PAGE 132

CHAPTER 6 CHARACTERISTIC PLASMA VOLUMES FOR ANALYSIS OF AEROSOL PARTICLES USING LIBS The use a single-shot conditional analysis for LIBS has been applied in this research as an approach to detect and quantify the size of individual aerosol particles. As it was stated earlier in this dissertation, the single-shot conditional analysis takes advantage of the discrete nature of aerosol particles and the point sampling volume realized with LIBS to collect and analyze individual aerosol particles and to improve the sensitivity of the technique itself This methodology leads to an analysis approach that is rooted in a probabilistic model of collecting single aerosol particles in consideration of a characteristic plasma volume or plasma size. The size, characteristics, and physical meaning of the sampling plasma (i.e., plasma sample volume) are the focus of the present chapter. Three distinct characteristic plasma volumes are experimentally measured in the context of the analysis of single aerosol particles using laser-induced breakdown spectroscopy. A more detailed explanation of this analysis can be foimd in the works of Hahn et al. (1997, 2000) and in Section 1.6 of this dissertation, and only a short clarification of the dependence of the LIBS sample volume is given here. When a given laser-induced plasma samples an aerosol particle, the event may be defined as a particle hit. The overall hit rate realized with single particle analysis may be modeled using a Poisson sampling probability, in which the probability of collecting at least a single particle in a given plasma volume is expressed by the Equation 1 .4, and rewritten as F 119

PAGE 133

120 =l-exp(-NV), where F is the sampling frequency (i.e., the number of particle hits divided by the total number of laser shots), N is the aerosol number density (number of particles/volume) and V is the plasma sample volume. It is noted that the sampling frequency may be approximated as F i^NV for small average number of particles per plasma volume. Rearranging the above relation, the next expression may be obtained for what is defined as the statistical plasma sample volume Vs, -Ln(l-F) = VsN. (6.1) The statistical sample volume is interpreted as the effective plasma sampling volume corresponding to a given particle sampling frequency of F realized at an aerosol number density N. To improve the analyte signal-to-noise ratio for low aerosol loadings, the actual aerosol mass concentration C (total aerosol analyte mass per volume of gas) can be determined using LEBS-based conditional data analysis as C = XF, where X is the equivalent aerosol mass concentration that is obtained by processing the ensembleaveraged LIBS spectrum corresponding to the aerosol particle hits only, and F is aerosol sampling rate as described above. Note that as the sample rate approaches 100%, the conditional analysis scheme converges to traditional ensemble averaging of all LIBS spectra. In addition to the above relations, the actual aerosol mass concentration may be defined using traditional aerosol mechanics as C = m N, where m is the mean aerosol analyte mass. Equating this last relation with the previous one, and making the approximation F »NV yields the expression XNV = mN, which may be solved to yield the important LIBS relation X= m/V„. ( 6 . 2 )

PAGE 134

121 This relation defines the mass-based sample volume Vm and states that the LIBSbased analyte concentration of a single spectrum corresponding to a single particle (or of an ensemble-average of single particle hits) is related to the actual analyte mass contained in a characteristic (i.e., mass-based) plasma volume. In other words, the mass-based sample volume can be defined as the required plasma volume to yield a LIBS-based mass concentration corresponding to the actual analyte mass of an individually sampled aerosol particle. In this chapter, both the statistical sample volume and the mass-based sample volume are evaluated along with a measured physical plasma volume (the third characteristic volume) based on plasma optical density. These three characteristic plasma volumes are discussed in concert to elucidate fundamental plasma-particle interactions, as well as to further the implementation of LIBS as an analytic technique suitable for analysis of aerosol systems, notably individual aerosol particles. 6.1 Experimental Methodology Three complementary experiments were performed to assess the various LIBS plasma volumes. In the first set of experiments, monodisperse silica particle aqueous suspensions were used to create aerosol streams with a known aerosol number density, enabling evaluation of the statistical sample volume. The second set of experiments was designed to calculate the LEBS-based equivalent mass concentration for various sized silica particles in order to evaluate the mass-based plasma volume. Finally, spatially and temporally resolved transmission measurements were recorded to provide a physical measure of the laser-induced plasma volume. The experimental facilities for LIBS and aerosol generation are the same that those were described in Chapter 2. In addition, spherical Si02 particles with

PAGE 135

122 monodisperse diameters ranging from 1.0, 1.5 and 2.1 pm (particles corresponding to complete vaporization) were introduced into the co-flow stream through nebulization to produce aerosol particle number densities ranging from about 5 to 90 particles/cm^ in the LIBS sample chamber. All LIBS-based parameters such as selected atomic emission line, laser pulse energy, time delay, and integration time were the same as those for the experiments in Section 5.1. A standard calibration curve from 0 to 3030 pg/m^ was made to accurately evaluate the signal corresponding to the smaller silica particle diameters. Ensemble averaging (minimum of 3000 individual spectra at each known aqueous silicon concentration) was used to construct a linear calibration response of the 288.16-nm silicon P/B signal as a function of known silicon mass concentration in the LIBS sample chamber. Single-shot conditional analysis was performed identically to that described in the previous chapter with the difference of choosing a higher false hit rate to assess the independence of the particle hit rate with respect to the threshold value. For the current study, the threshold value for the 288.16-nm silicon emission peak was set to obtain an average of 15 false hits for every 1000 laser shots for nearly all experiments. For all spectral data processing, individual spectra were smoothed using the Savitzky-Golay algorithm and subsequently processed using the P/B of the 288.16-nm silicon emission line profile in the same way that was done in the previous chapter. To determine the exact number density of the silica microspheres in suspension prior to nebulization, transmission measurements were recorded for various diluted samples of the prepared stock suspension (see Table 2.2 for original particle densities). A

PAGE 136

123 25-mW He:Ne laser operating at 632.8 nm was used to record the transmission through a standard 1-cm pathlength scattering cell. Transmission was recorded for various dilution factors with respect to the original silica particle suspension. The dilutions ranged from a factor of 1/3 to 1/6, and were selected to provide an optical density sufficiently small (i.e., NCexiL < 1) to ensure the absence of any multiple scattering effects. For the physical plasma volume measurements, a probe Nd:YAG laser (532-nm, 6.4-ns pulse width) was synchronized to the plasma-creating laser. A time delay of zero between the two lasers was defined when the temporal peaks of both laser pulse profiles were coincident in time. The probe laser was then set to a time delay of 20 ns with respect to the plasma-creating laser (Nd:YAG laser with 315 mJ per pulse), which placed the probe laser at the end of the plasma-creating laser pulse. Both the flashlamp and the Q-switch of the probe laser were externally triggered using a digital delay generator synchronized to the plasma-creating laser, resulting in less than 1 ns of jitter between the two pulses. The probe laser power was recorded with a volume-integrating laser pulse calorimeter, using a 1 -minute average at a 5-Hz repetition rate. A two-axis precision stage was assembled and used to translate the 532-nm probe beam horizontally (xcoordinate) and vertically (y-coordinate) through the plasma at 90 degrees with respect to the plasma-creating laser pulse direction. A 200-mm plano-convex lens was mounted to the stage to focus the probe beam to a 200-pm spot within the laser-induced plasma volume. The probe laser beam was translated vertically through the plasma volume at each horizontal position, using 100-pm steps in the y-direction and 200-pm steps in the x-direction. The measurements were reported four times for each vertical traverse through the plasma.

PAGE 137

124 6.2 Statistical Sample Volume The determination of the statistical sample volume requires an accurate and independent calculation of the silica aerosol number density in the LIBS sample chamber. This is readily calculated from the gas co-flow rates and the rate at which liquid is nebulized, provided the number density of the silica particles in the liquid suspension is known. The suspension number density (N) was determined from laser transmission measurements making use of the Beer-Lambert law I/I„=cxp(-C^,NL), (6.3) where / and Iq are the irradiances (or laser power) transmitted and incident on particle suspension contained in the scattering cell, Cext is the single particle extinction crosssection, and L is the optical path length (1 cm) through the scattering cell. Using complete Mie light scattering theory, an extinction cross-section of 8.80x10'^ cm^ was calculated for the 2.1 -pm silica particles (refractive index = 1.46) in purified water (refractive index = 1.33). Then as discussed above, the original suspension was diluted to ensure single-scattering effects, and the number density in Equation 6.3 may be modified as the original number density times a dilution factor fj. Rearranging the Equation 6.3, the natural log of the transmission is linearly proportional to the extinction cross-section and the partiele number density in suspension as seen in Equation 6.4 jLn(I/I^) = N-f,C^, . (6.4) A plot of the natural log of the transmission as a function of the product Cext fd is presented in Figure 6.1, with the slope equal to the number density of the original silica particle suspension.

PAGE 138

125 Figure 6.1 The natural logarithm of the transmission as a function of the product of the extinction cross-section and dilution factor for the 2.1-pm-silica particle suspension. The error bars represent ± 1 standard deviation. The transmission measurements yielded a particle number density of 1.87x10^ cm' for the stock ofa2.1-pm silica suspension. The highly linear relation (R =0.9997) observed in Figure 6.1 suggests no multi-scattering effects and a high degree of precision in the measured particle concentration. It is noted that this value was in excellent agreement with the result obtained based on serial dilution of the initial suspension number density provided by the manufacturer. The particle concentration was subsequently scaled to reflect dilution for all additional experiments, enabling the generation of a range of known silica aerosol concentrations in the LIBS sample chamber To assess the effect of the threshold value on the particle hit rate, experiments with a restrictive threshold value, such as about 2 false hits in 1000 shots, were performed in addition to the selected 15 false hits for each 1000 shots. Figure 6.2 shows the results of this experiment.

PAGE 139

126 DIW DIW+Si02 DIW DIW-(-Si02 Low Low High High Figure 6.2 Independence of the threshold value in the conditional analysis for determination of the hit rate. Silica particles of 2.1 pm in diameter and at a 45-cm'^ number density were used. DIW equals purified deionized water only. As seen from Figure 6.2, an average of 15.8 hits per 1000 shots with a standard deviation of 3.34 was recorded with purified water, and an average of 45.3 hits with a standard deviation of 4.74 was recorded for the silica particle seeded aerosol stream. The difference yielded a silica particle-sampling rate of 29.5 hits per 1000 shots with a standard deviation of 5.8. Using the same experimental conditions hut with the more restrictive sample threshold, experiments yielded an average of 1.8 hits per 1000 shots with a standard deviation of 0.42 recorded with purified water, and an average of 32.5 hits with a standard deviation of 3.6 recorded for the silica particle seeded aerosol stream, for an overall sampling rate of 30.7 hits per 1000 shots with a standard deviation of 3.6. Comparison of the two sampling rates (29.5 ± 5.8 and 30.7 ± 3.6) reveals no statistical difference between the actual silica particle sample rates when using the two different conditional analysis threshold values.

PAGE 140

127 For the determination of the statistical sample volume, the silica particle LIBSbased sampling hit rate was determined for various silica particle concentrations using the 2.1 -pm silica particles. As noted above, the conditional analysis threshold value was set to allow the detection of about 15 false hits in 1000 shots. The actual hit rate was obtained by subtraction of the average number of false hits from the total number of hits recorded in a 1000-shot sequence. The LIBS-based silica particle sampling rate, expressed as the -/n(l-F), is presented in Figure 6.3 as a function of the actual silica particle number density in the LIBS sample chamber. The error bars correspond to the standard deviation based on multiple 1000-shot experiments. Additionally, Table 6.1 reports the actual hit rate and the standard deviation for each particle number density. Figure 6.3 The natural logarithm of one minus the experimental silica particle-sampling rate as a function of the silica particle number density. The error bars represent ± 1 standard deviation. A plot of the -//j(l-F) as a function of A is characterized by a slope equal to the statistical plasma sample volume Vs in consideration of Poisson sampling statistics and as it is expressed by Equation 6.1. A linear least squares fit (R^ = 0.931) yielded a statistical

PAGE 141

128 plasma volume of 1.17x10Â’^ cm^, which corresponds to an equivalent spherical diameter of 1.31 mm. It is noted with regard to the Figure 6.3 data that the observed sample frequency does not exactly follow an ideal Poisson probability distribution in that the observed rates tend to increase beyond those predicted at the upper and lower extremes. This deviation from the ideal linear trend could be due to aerosol-laser beam interactions before the laser focal region, an effect that may induce a particle breakdown sufficient to excite the analyte and trigger a conditional analysis hit. Such a breakdown would be decoupled from the actual laser-induced plasma volume, thereby inflating the sampling rate based on Poisson statistics as defined solely by the plasma volume. Table 6,1 Actual Silica Particle Sample Rates N at Sample Chamber, cm'^ Actual Sample Rate, F Standard Deviation ofF 7.48 0.0081 0.0051 14.96 0.0111 0.0073 44.88 0.0297 0.0051 59.84 0.0512 0.0123 89.77 0.1022 0.0139 6.3 Mass-Based Sample Volume To calculate the mass-based LIBS plasma volume, single-shot spectra of individual silica particles were collected and quantitatively analyzed for monodisperse silica particle diameters of 1.0, 1.5, and 2.1 pm. These particle sizes were all of a size characterized by complete particle vaporization, as determined in the previous chapter. The process schemes of detection of single particles, smoothing of single spectra due to spectral noise (using the Savitzky-Golay algorithm), and filtering (e.g., screen out false

PAGE 142

129 hit and anomalous events) were implemented in the same way as the schemes applied in Chapter 5 for determination of the upper particle size limit of vaporization. Using the spectral data processing schemes outlined previously, all single-shot spectra were processed, and a final ensemble-averaged spectrum was calculated for each silica particle size comprised of the individual single-shot spectra accepted after all data filtering. Finally, the P/B of the 288.16-nm silicon line was calculated for each final ensembleaveraged spectrum, which was used with the silicon calibration curve to calculate the equivalent mass concentration (pg/m^) corresponding to each silica particle size. Figure 6.4 shows the linear correlation (i.e., nanoparticles calibration curve) between the silicon 288.16-nm P/B signal and the corresponding low range silicon mass concentration. Figure 6.4 Silicon calibration curve for a low mass concentration range used in the quantification of the mass-average sample volume. The error bars represent ± 1 standard deviation. Figure 6.5 presents the measured equivalent mass concentration as a function of the silicon mass contained in each single silica particle for the particle diameters of 1.0, 1 .5, and 2. 1 pm. The silicon mass is calculated based on a silica (Si02) particle density

PAGE 143

130 of 1.96 g/cm^ and a silicon (Si) mass fraction of 0.467. The highly linear behavior (R^ = 0.996) observed in Figure 6.5 enables a direct calculation of the mass-based plasma volume using the measured slope and the relation X= rri/Vm, which yields a value of 2.38x10'^ cm^ with a corresponding equivalent spherical diameter of 1.66 mm. The volume is larger than the statistical sample volume measured in the preceding section, a characteristic that will be discussed later in this chapter. Figure 6.5 The LIBS-based equivalent mass concentration of the 1.0, 1.5, and 2.1 pm silica particles as a function of the silicon mass contained in the silica particles. The error bars represent ± 1 standard deviation. 6.4 Physical Plasma Volume The final experiment provides a physical measurement of the plasma volume induced by the 1064-nm Nd:YAG laser pulse. As described above, the plasma was formed in ambient air and probed using a 532-nm Nd:YAG pulse laser to determine the boundaries of the plasma at a 20-ns delay with respect to the 1064-nm plasma-initiating pulse. Figure 6.6 shows the temporal scale in which the transmission measurements were

PAGE 144

131 performed. Both laser pulses show high symmetry for delivering the total energy of the pulse and resemble typical Gaussian profiles. A slight tail appears at the end of the pulse, however, this asymmetry does not account for more than 8% of the total pulse energy. The probe laser (with 6.4 ns full-width half maximum) releases the total pulse energy (99.9%) in about 1 8 ns, so the transmission measurements were averaged over this period and centered at 20 ns after the peak of the plasma creating laser pulse. Figure 6.6 Temporal scale for the transmission measurements thru the plasma created by the Nd:YAG 1064-nm laser, and probed by the Nd:YAG 532-nm laser. For a given point within the plasma, the transmission of the probe laser was calculated as the ratio of the probe pulse power through the laser-induced plasma to the probe laser power in the absence of the laser-induced plasma. The power meter was well separated from the plasma and it was determined that the laser-induced plasma emission contributed a negligible signal to the probe pulse power meter. The plasma boundary was defined as the point at which an optical thickness of 0.1 was recorded. The optical thickness is ideally defined as the weighted sum of the product of each absorbing species

PAGE 145

132 cross-section, absorbing species niunber density, and optical pathlength. The optical thickness is readily calculated as the opposite sign of the natural log of the transmission, as defined by the BeerLambert law. An optical thickness of 0.1 corresponds to the population of free electrons of the plasma absorbing approximately 10% of the probe laser energy. The plot of measured optical thickness versus distance for each traverse was used to define the plasma boundary by interpolating the resulting profile using a order polynomial to define the exact point corresponding to an optical thickness of 0. 1 . A plot of the plasma volume cross-sectional profile is shown in Figure 6.7. Also included in the figure is an ideal elliptic profile based on the measured plasma volume and plasma aspect ratio. 10.5 10.0 I 9.5 >-“ 9.0 8.5 18.0 18.5 19.0 19.5 20.0 20.5 21.0 X, mm Figure 6.7. Cross-section of the physical plasma measured at the end of the 1064-nm plasma-generating laser pulse. The boundary represents an optical thickness of 0.1, and the error bars represent ± 1 standard deviation. The dashed line is an elliptic profile fit to the experimental plasma volume and aspect ratio. 7T — I — rn — I I I I t — |“T — I — I — T“| — I — I — n — j — i i i — r-j — i — i — i — rz Volume = 1.44 10'* cm® _ ‘111 I I I I I I I I I I I I I I I I 1 I 1 I I I 1—LJ

PAGE 146

133 A high degree of symmetry is observed in the Figure 6.7 data, with the overall shape corresponding well to a prolate spheroid (with semi-axis of 2.25 mm and 1.15 mm). This type of shape has been reported before for laser-induced plasmas (Borghese et al. 1998, and Chen et al. 2000) and it is characteristic for high-energy pulses. A numerical integration using 360 degree rotation of the measured profile (average at upper and lower cross sections) yielded a calculated plasma volume of 1.44x10' cm , with an equivalent spherical diameter of 1 .4 mm. 6.5 Discussion of the Characteristic Plasma Volumes In this study, three different characteristic plasma volumes were determined using three different physical measurement schemes. The resulting plasma volumes have been defined as a statistical sample volume, a mass-based volume, and a physical plasma volume. The experimental results are summarized in following Table 6.2, and while it is noted that all three plasma volumes agree to within a factor of 2, the actual differences should be interpreted in the context of the relevant plasma processes. Table 6.2 Summary of the Characteristic Plasma Volumes for 315 mJ per Laser Pulse Characteristic Plasma Volume, mm^ Diameter, mm Statistical 1.17 1.31 Physical 1.44 1.40 Mass-based 2.38 1.66 To understand the relations between these different characteristic plasma volumes, it is useful to consider the interaction of a single aerosol particle with the plasma as defined by the measured optical thickness of 0. 1 . The plasma itself must have

PAGE 147

134 a significant boundary region due to its highly transient nature and its formation within a continuum gaseous medium. With this in mind, the physical plasma volume is somewhat arbitrary, in that a larger plasma volume could be defined if experimental precision was sufficient to measure an optical thickness less than 0. 1 . However, while the physical plasma volume is defined in the current study based on optical thickness, it is recognized that as the plasma state weakens (e.g., reduction in free electron density) on the outer boundaries, sufficient energy may not be available to completely vaporize a given aerosol particle in this regime. With the additional need for complete particle vaporization with LIBS-based aerosol analysis, consideration of the physical plasma volume gives way to the statistical plasma volume. The statistical sampling volume (Vs = f/N) represents the effective volume consistent with the LIBS system to detect the presence of an individual particle following vaporization and subsequent analyte species atomic emission. There was no a priori basis in the present study to expect complete convergence of the physical and statistical plasma volumes. Nonetheless, the current data suggest that sufficient vaporization for subsequent analyte detection is confined within a region somewhat smaller than the region defined by an optical thickness of 0.1. For the current study, defining the physical plasma volume to a region based on an optical thickness of 0.15 yields convergence between the physical and statistical sample volumes. The important conclusion is not the degree of exactness between these two characteristic plasma volumes, but rather that the plasma volume is not ideally homogeneous with regard to plasma-particle interactions. Additional plasma transmission measurements, including Abel inversion and plasma modeling, may provide further insight into the local plasma state conditions necessary for particle sampling and quantitative analysis.

PAGE 148

135 In contrast to particle sampling considerations, once the particle is vaporized and ionized, a different set of processes are relevant to LIBS-based analysis. If it is expected that the analyte mass released from the aerosol particle will subsequently diffuse throughout the plasma volume with a relatively short time-scale. Hence it follows that the analyte atomie emission and plasma continuum emission light that is collected and reported as the signal peak-to-base ratio reflects a plasma volume characteristie of the entire emitting plasma at the relevant signal integration delay time, typically microseconds to tens of mieroseconds following plasma initiation. The mass-based plasma volume refleets this LIBS signal corresponding to the partiele analyte mass contained in an equivalent emitting plasma volume. In view of the present data, it is reasonable to eonclude that the emitting plasma volume extends beyond the plasma size defined by the limit necessary for initial particle vaporization or an optical thickness of 0.1. Because the equivalent mass eoneentration is linearly proportional to the peak-tobase ratio (by the calibration curve), the calculated mass-based plasma volume (i.e., emitting volume) is independent of the size of the particle, provided eomplete particle vaporization, as demonstrated by the linear response observed in Figure 6.5. In summary, the analysis of individual aerosol particles using laser-induced plasma spectroseopy has been discussed in this chapter in the context of the characteristic plasma volume. While the statistical sample volume refleets the probabilistic nature of capturing single particles using the laser-induced plasma, the statistical volume also reflects a fundamental threshold necessary for significant vaporization of the sampled particle. In contrast, both the physical and mass-based plasma volumes are eharacterized by the absorption or emission properties of the plasma state, respectively, and appear to

PAGE 149

136 be less restrictive than considerations of plasma-induced particle vaporization. While all three characteristic plasma volumes were found to be of similar value in the present study, each is coupled to specific plasma processes, and when considered together provide additional insight into the plasma processes relevant to LIBS-based aerosol analysis.

PAGE 150

CHAPTER 7 CONCLUSIONS In response to the need for discrete characterization of aerosol particles, notably ambient air fine particles due to their direct relation to human health, laser-induced breakdown spectroscopy (LIBS) has been the focus in this dissertation for development as a real-time technique for the analysis of single aerosol particles. Key issues addressed include implementation and demonstration of LIBS as a real-time monitor for aerosol particles, as well as the fundamental interactions between laser-induced plasma and individual aerosol particles. The achievements and conclusions of this research are summarized as follows. 1 . The aerosol generation system was fully characterized and determined to produce well-dispersed nanoparticles about 70 nm in mean diameter with number densities in the order of lOÂ’ particles per cm^. The aerosol system excellently coupled to the LIBS system for purposes of calibration, including multi-species detection and quantification, and for introduction of particle suspensions. 2. The LIBS system as implemented was able to create a repeatable plasma (100% breakdown frequency) for pulse energies greater than 190 mJ, and a saturated plasma (60% absorption) for pulse energies larger than 255 mJ. 3. A conditional analysis technique was successfully used with LIBS to monitor ambient particulate matter with mass concentrations on the order of parts per trillion, and with minimum particle diameters approaching 200 nm. LIBS was successful for the realtime detection of transient changes in metallic-based aerosols in the ambient air due to 137

PAGE 151

138 the discharge of fireworks. The time response for ambient air monitoring was assessed and it was foimd that the variability of 4-min sampling frequencies in 2-hour sampling period ranged at most by a factor of three with respect to the 2-hour average sampling rate. 4. The average uncertainty with the LIBS system as implemented for single particle sizing was determined to be less than 20%. While the accuracy for sizing was observed to increase with increasing LIBS analyte signals, the precision remained consistent for any level of LIBS signal. 5. Single-shot LIBS based measurements should be performed with sufficient laser pulse energy to achieve the condition of saturation with respect to absorbed laser energy by the plasma (i.e., above ~ 255 mJ/pulse in the present study). Above plasma saturation, the shot-to-shot spatial variability of the plasma is minimized, thereby producing the most robust and precise plasma for single-shot analysis. 6. Average single-shot plasma temperature measurements and measurement precision were determined to be constant and independent of the laser pulse energy at the optimal time delay (i.e., maximum P/B). 7. For quantitative analysis of aerosol particles, complete vaporization of the entire analyte mass should be achieved. In this study it was determined that the largest single silica particle for complete vaporization at airborne condition was 2.1 pm in diameter when using a pulse energy of 320 mJ and the current LIBS parameters. This upper limit may change for different species and different optical configurations, but it is significantly less than the often-cited limit oflO pm.

PAGE 152

139 8. Based on silica particle sampling rates and the analyte response, it is concluded that the major mechanisms of particle vaporization in the use of the LIBS technique are due to plasma-particle interactions. Furthermore, the present analysis suggests that complete vaporization of a particle is not limited by the heat capacity of the plasma and appears to be constrained to the first tens of nanoseconds of plasma lifetime. 9. Three distinct plasma volumes were defined as characteristic of the LIBS technique, namely a sampling-based volume, a mass-based volume, and a physical plasma volume. These characteristic plasma volumes were determined to have sizes within a factor of two, and are governed by the nature of the plasma absorption, growth and emission processes.

PAGE 153

CHAPTER 8 RECOMMENDATIONS FOR FUTURE WORK Additional work is needed to further develop LIBS as a technique for single particle analysis. The following issues are recommended for future work. 1 . The effect of optics configuration, such as f/number and laser beam diameter, on the production of a more energetic plasma (e.g., plasmas with more than 60% laser pulse absorption), and on the variation of the spatial position of the plasma (at constant pulse energy) as a means to improve the accuracy and precision of the LIBS signal. 2. The rate of acceptance of single-shot spectra by using multiple emission lines (more than two) in the discrimination of an actual particle hit. Present spectral retention rate is about 60% using two emission lines with mass concentrations within a factor of two, as implemented in this study. 3. The effects of laser pulse fluctuations on the variability of the LIBS signal to elucidate the origin of its precision. Actual P/B precision is about 6% (RSD) in the present study and could be due to laser pulse-to-pulse fluctuations, intensifier and detector shot noise, or intrinsic laser-matter interactions. 4. Study the upper limit of vaporization for different types of particles and with different LIBS parameters to quantify the importance of the physical (e.g., heat capacity and dissociation energy) and optical properties of the particles in the mechanisms of plasma-particle vaporization with the LIBS technique. 5. Develop a model and simulate numerically the mechanism of particle vaporization due to the laser-induced plasma to identify the most important parameters. 140

PAGE 154

such as time, electron and/or ion number density, plasma temperature, thermophoretic force, and others. 141 6. Resolve experimentally the diffusion time of a vaporized single particle analyte species in the plasma to clarify the representativeness of the collected signal corresponding to the mass of the particle. 7. Expand the Poisson model for LIBS-based aerosol particle sampling rate to include features such as particle size and aerosol screening effects.

PAGE 155

APPENDIX A ZERO ORDER LOGNORMAL DISTRIBUTION Let the lognormal distribution of the Equation al defined by the geometric mean Xg and the geometric standard deviation
PAGE 156

143 AogW = 4^ • •exp [lnx-In(x„exp(o-g)f lal Expanding, simplifying, and regrouping, the equation a8 is obtained Aog(^) = AogW = -cr. 42z -a „ exp [lnx-lnx„ -al^ 2al AogW •exp<^exp (inx-lnx^y (in X In x„ y / \ + ln V fn y \^m J •exp AogW = exp(-io-J (lnx-lnx„)' -exp yllTT a X„ [ J (a8) The equation a8 is the rewritten lognormal distribution, and comparing this equation with the equation a5 that define the ZOLD, it conclude that if and Therefore, the zero-order lognormal distribution is a new form of writing the and = cr mean mean lognormal distribution with the appropriate values. This similarity has been noted before by Honig (1965), Watterson (1971), Pyun (1973), Yan (1974), and Ross (1978).

PAGE 157

APPENDIX B QUARTZ TUNGSTEN HALOGEN LAMP Manufacturer: Oriel Instruments Model Number: 63350 or 63351 Serial Number: 7-1121 Material: one 1000-watt, quartz halogen, tungsten filament Spectral range: 250-2400 nm Lamp current: 8.20 Amps Operating room temperature: 26.5 °C Calibration curve: r X ,-5 r. D E F G H) Irradiance { — — ) = A • exp(/4 + B/ A)Ch — + ^ + + m nm y AAA A A J where A is given in nm and A = 44.5712 5 = -4.63923 -10^ C = 9.0937210'' D = 4.13307 £ = 2.07519-10’ £ = -1.47164-10* G = 3.87410-10“* // = -3.80406 -10'' 144

PAGE 158

APPENDIX C SAVITZKY-GOLAY ALGORITHM The selected algorithm for plasma temperature spectra uses 7 contiguous discrete values to calculate the new centered value of the set. The smoothing equation is T/ = ^(2 x ,._3 + 3jc,_ 2 + 6x,._, + lx, + 2x,^,), where yt is the updated value (new value) corresponding to the previous value x,(old value) at the position i, and Xi. 3 , Xi. 2 , x,./, x,+/, x,+ 2 , x ,+5 are the previous values corresponding to 1,2, and 3 position before and after the position of x,. This sequence was applied 10 times to the entire spectrum to obtain the final smoothed spectrum Original Data A,_I A +3 A2 >1 Aj— A4— As^^ A6— ^ A 7 — 1/21 Ag A 9 • B 3 B 4 B5B6B6Bg^^ B9 > 1 +3 +6 +3 -2 1/21 C4 C5 C6 C7 Cg C9 • • Smoothed Data i K, 1 K 2 K 3 K 4 Ks K6 K1021 K 1022 A 1023 B 1023 Cl 023 K 1023 |Aio 24 I 1 B 1024 1 1 C 1024 I 1 Kio 24 I Original First Pass Second Pass Tenth Pass Data Smoothing Smoothing Smoothing 145

PAGE 159

146 For smoothing LIBS spectra corresponding to single silica particles, the selected Savitzky-Golay algorithm was slightly different to the previously described. The algorithm uses 5 contiguous discrete values to calculate the new centered value of the set. The smoothing equation is T/ =^(-3^,-2 +12x,_, +\lx, +12x,^i +3 x,^2) The definition of the terms was the same as stated before and this sequence was applied 3 times to the entire spectrum to obtain the final smoothed spectrum. The reason of change was the minimum effect of the algorithm in a real emission line, which had a line width of 12 pixels. The idea in select the width of the algorithm (number of discrete values or pixels depending for smoothing) is to be less or equal to the halfwidth full maximum of the emission line. Original Smoothed Data Data Ai • Di • • • D2 A +12 Aa > B3 • D3 A4-^^ 1/35 B4 C4 D4 Bs Cs Ds A6 — — > BeCe De +12 At Be> C7 • +17 1/35 „ Ag Bg> — Cg • A9 B9C9 # • • . • D1021 • • • • • • D1022 A 1023 B1023 Cl 023 D1023 |Aio 24 I 1 B1024 1 1 C1024 1 1 D1024 1 Original First Pass Second Pass Third Pass Data Smoothing Smoothing Smoothing

PAGE 160

REFERENCES Biswas A., Latifi H., Radziemski L.J., and Armstrong R.L, Applied Optic, 27 (1988) 2386. Borghese A. and Merola S.S., Applied Optics, 37 (1998) 3977-3983. Capitelli M., Capitelli F., and Eletskii A., Spectrochimica Acta Part B, 55 (2000) 559574. Carranza J.E., Fisher B.T., Yoder G.D., and Hahn D.W., Spectrochimica Acta Part B, 56 (2001)851-864. Chen X. and He P., Plasma Chemistry and Plasma Processing, 6 (1986) 313-333. Chen Y.L., Lewis J.W.L., and Parigger C., Journal of Quantitative Spectroscopy & Radiative Transfer, 67 (2000) 91-103. Cheng M-D., Fuel Processing Technology, 65-65 (2000) 219-229. Cremers D.A. and Radziemski L.J., Applied Spectroscopy, 39 (1985) 57-63. Darke S.A. and Tyson J.F., Journal of Analytical Atomic Spectrometry, 8 (1993) 145209. Davies C.M., Telle H.H., Montgomery D.J., and Corbett R.E., Spectrochimica Acta Part B, 50(1995) 1059-1075. Davies C.M., Telle H.H., and Williams A.W., Fresenius Journal of Analytical Chemistry, 355 (1996) 895-899. Davis E.J., Ravindran P., Ray A.K., Chemical Engineering Commun, 5 (1980) 251-268. Docchio F., Regondi P., Capon M.R.C., and Mellerio J., Applied Optics, 27 (1988) 36613668. Espenscheid W.F., Kerker M., and Matijevic E., The Journal of Physical Chemistry, 68 (1964) 3093-3097 Essien M., Radziemski L.J., and Sneddon J., Journal of Analytical Atomic Spectrometry, 3 (1988) 985-988. Ge Z., Wexler A.S., and Johnston M.V., Environmental Science Technology, 32 (1998) 3218-3223. 147

PAGE 161

148 Gomushkin LB., Smith B.W., Potts G.E., Omenetto N., and Winefordner J.D., Analytical Chemistry, 71 (1999) 5447-5449. Griem H.R., Principles of Plasma Spectroscopy, Cambridge University Press, UK 1997, Chapter 5. Hahn D.W., Applied Physics Letters, 72 (1998) 2960-2962. Hahn D.W., Flower W.L., and Hencken K.R., Applied Spectroscopy, 51 (1997) 18361844. Hahn D.W. and Lunden M.M., Aerosol Science and Technology, 33 (2000) 30-48. Hinds W.C., Aerosol Technology: Properties, Behavior, and Measurement of Airborne Particles. 2nd. Ed. John Wiley & Sons, NY 1999. Ho W.F., Ng C.W., and Cheung N.H., Applied Spectroscopy, 51 (1997) 87. Honig E.P., Journal of Physical Chemistry, 69 (1965) 4418. Hoomaert S., Van Malderen H., and Van Grieken R., Environmental Science and Technology, 30 (1996) 1515-1520. Isaac R.C., Gopinath P., Varier G.K., Nampoori V.P.N., and Vallabhan C.P.G., Applied Physics Letters, 73 (1998) 163-165. Johnston M.V., and Wexler A.S., Analytical Chemistry, 67 (1995) 721 A-726A. Laj P., Ghermandi G., Cecchi R., Maggi V., Riontino C., Hong S.M., Candelone J.P., and Boutron C., Journal of Geophysics Res. -Oceans, 102 (1997) 26615-26623. Lee D.S., Dollard G.J., Derwent R.G., and Pepler S., Water Air Soil Pollution, 113 (1999) 175-202. Lee Y-I, Song K., and Sneddon J., Laser-Induced Breakdown Spectroscopy. Nova Science Publishers Inc., New York 2000, chapter 3. Leoncioni D.E., and Pettingill L.C., Journal of Applied Physics, 48 (1977) 1848-1851. Liu D.Y., Prather K.A., and Hering S.V., Aerosol Science and Technology, 33 (2000) 7186 . Liu D.Y., Rutherford D., Kinsey M., and Prather K.A., Analytical Chemistry, 69 (1997) 1808-1814. Lochte-Holtgreven W., Evaluation of Plasma Parameters. In Plasma Diagnostics edited by W. Lochte-Holtgreven (1995), American Institute of Physics, Chapter 3. Lushnikov A. A. and Negin A.E., Journal of Aerosol Science, 24 (1993) 707-770.

PAGE 162

Madden H.H., Analytical Chemistry, 50 (1978) 1383-1386. Martin M.Z. and Cheng M-D., Applied Spectroscopy, 54 (2000) 1279-1285. 149 Martin M.Z., Cheng M-D, and Martin R.C., Aerosol Science and Technology, 31 (1999) 409-421. McFarland A., Deposition Version 4.0, Texas A&M University, College Station, Texas (1996). Mclnnes L.M., Covert D.S., Quinn P.K., Germani M.S., Journal of Geophysics Res. Atmos., 99 (1999) 8257-8268. Neuhauser R.E., Panne U., Niessner R., and Wilbring P., Fresenius Journal of Analytical Chemistry, 364 (1999) 720-726. Noble C.A. and Prather K.A., Mass Spectrometry Review, 19 (2000) 248-274. Ottesen D.K., Wang J.C.F., Radziemski L.J., Applied Spectroscopy, 43 (1989) 967-976. Parigger C., Tang Y., Plemmons D.H., and Lewis J.W.L., Applied Optics, 36 (1997) 8214-8221. Peng L.W., Flower W.L., Hencken K.R., Johnsen H.A., Renzi R.F., and French N.B., Process and Control Quality, 7 (1995) 39-49. Peters A., Dickery D.W., Muller J.E., and Mittleman M.A., Circulation, 103 (2001) 28102815. Pogodaev V.A. and Rozhdestvenskii A.E., Soviet Technology Physics Letters, 5 (1979) 257-261. Prather K.A., Nordmeyer T., and Salt K., Analytical Chemistry, 66 (1994) 1403-1407. Pyun C.W., Journal of Macromolecular Science and Chemistry, A7 (1973) 1721-1725. Radziemski L.J., Microchemical Journal, 50 (1994) 218-234. Radziemski L.J., and Cremers D.A., Spectrochemical Analysis Using Laser Plasma Excitation. In "Laser-Induced Plasma and Applications ” edited by L. J. Radziemski and D.A. Cremers (1989), Marcel Dekker, New York, Chapter 7. Radziemski L.J. and Loree T.R., Plasma Chemistry and Plasma Processing, 1 (1981) 281293. Radziemski L.J., Loree T.R., Cremers D.A., and Hoffman N.M., Analytical Chemistry, 55 (1983) 1246-1252.

PAGE 163

150 Reents W.D., Downey S.W., Emerson A.B., Mujsce A.M., Muller A.J., Siconolfi D J., Sinclair J.D., and Swanson A.G., Aerosol Science Technology, 23 (1995) 263-270. Reents W.D. and Ge Z.Z., Aerosol Science Technology, 33 (2000) 30-48. Reilly J., Singh P., and Weyl G., Multiple Pulse Propagation through Atmospheric Dust at 10.6 um, AIAA 10*Â’Â’ Fluid and Plasma Dynamics Conference, Albuquerque, NM, June 27-28, paper 77-697, 1977. Ross W.D., Journal of Colloid and Interface Science, 67 (1978) 181-182. Rusak D.A., Castle B.C., Smith B.W., and Winefordner J.D., Critical Reviews in Analytical Chemistry, 27 (1997) 257-290. Savitzky A. and Golay M.J.E., Analytical Chemistry, 36 (1964) 1627-1639. Schechter I., Analytical Science & Technology, 8 (1995) 779-786. Schechter I., Reviews in Analytical Chemistry, 16 (1997) 173-298. Simeonsson J.D. and Miziolek A.W., Applied Physics B, 59 (1994) 1-9. Singh J.P., Yueh F.Y., Zhang H.S., and Cook R.L., Process and Control Quality, 10 (1997) 247-258. Smith B.W., Hahn D.W., Gibb E., Gomushkin I., and Winefordner J.D., KONA Powder and Particle N*Â’ 19 (2001) 25-33. Smith D.L., Kristiansen M., and Haler M.O., Journal of Applied Physics, 48 (1977) 45214527. Sneddon J. and Lee Y-I, Analytical Letters, 32 (1999) 2143-2162. Song K., Lee Y.I., and Sneddon J., Applied Spectroscopy Reviews, 32 (1997) 183-235. Turns S.R., An Introduction to Combustion, 2"^ ed. (1999). McGraw-Hill, New York, chapter 4. U.S. Environmental Protection Agency, Federal Register 61, 77, (1996) 17357-17358. U.S. National Research Council, Research Priorities for Airborne Particulate Matter: IL Evaluating Research Progress and Updating the Portfolio (1999). Watterson J.G., Journal of Macromolecular Science and Chemistry, A5 (1971) 10071009. Weyl G.M. Physics of Laser-Induced Breakdown: An Update, in Laser-Induced Plasmas and Applications, edited by L.J. Radziemski and D.A. Cremers (1989), Marcel Dekker, New York, Chapter 1.

PAGE 164

151 Xu L., Bulatov V., Gridin V.V., and Schechter L, Analytical Chemistry, 69 (1997) 21032108. Yalcin S., Crosley D.R., Smith G.P., and Paris G.W., Hazardous Waste & Hazardous Materials, 13 (1996)51-61. Yalcin S., Crosley D.R., Smith G.P., and Paris G.W., Applied Physics B, 68 (1999) 121130. Yan J.P., Journal of Colloid and Interface Science, 49 (1974) 152-153. Zhang H., Yueh P.Y., and Singh J.P., Applied Optics, 38 (1999) 1459-1466. Zhigilei, L.V., Garrison, B.J., Applied Surface Science, 127-129 (1998) 142-150.

PAGE 165

BIOGRAPHICAL SKETCH Jorge Carranza was bom in Pern in 1964. He is the third of Pablo Carranza and Nelida QuipuzcoÂ’s four children. He spent part of his life in the small city named Trujillo. In 1989, he achieved the degree of Bachelor in Mechanical Engineering from the University of Tmjillo and worked for the university during the following years. In 1995, he obtained the unique opportunity of pursuing a masterÂ’s degree at the University of Puerto Rico. He moved to the small island of the Caribbean Sea, Puerto Rico, where after two intensive years of schoolwork reached his goal. He graduated with the degree of Master in Science in 1998, and due to his outstanding work, the University of Puerto Rico hired him during the following year. One year later, he moved to Gainesville with the intention of pursuing the next step of his career, to obtain a Ph.D. from the University of Florida. He studied English intensively for six months in order to satisfy all requirements of the University of Florida for an international student, and by then he met Dr. David Hahn who would be later his adviser and mentor. Jorge Carranza started in the University of Florida in August 1999 and graduated in August 2002. He spent enjoyable time in Gainesville where he met many great friends. 152

PAGE 166

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. David W. Hahn, Chairman Assistant Professor of Mechanical Engineering I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Jacob N. Chung Professor of Mechanical Engineering 1 certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. f William E. Lear, Jr. ( ' Associate Professor of Mechanical Engineering I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. ChangWW^i (j Assistant Professor of Environmental Engineering Science I certify that 1 have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. 2nuomin Zhang Associate Professor of Mechanical Engineering

PAGE 167

This dissertation was submitted to the Graduate Faculty of the College of Engineering and to the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. August 2002 Pramod P. Khargonekar* Dean, College of Engineering Winfred M. Phillips Dean, Graduate School