Skip to main content

Full text of "Laser-induced plasma spectroscopy in the analysis of phosphate mining samples and archaeological materials"

See other formats


LASER-INDUCED PLASMA SPECTROSCOPY 

IN THE ANALYSIS OF PHOSPHATE MINING 

SAMPLES AND ARCHAEOLOGICAL MATERIALS 



MARK ANTHONY VILLORIA 



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 

UNrVERSrTY OF FLORIDA 



2002 



This dissertation is sincerely dedicated to my parents, 
Florecito Bacho Villoria and Delynn Aya-ay Villoria, 
who have instructed me over many years the value 
of an education and have equipped me with the 
determination to obtain one. 



ACKNOWLEDGMENTS 

I would like to extend sincere gratitude to Dr. Jim Winefordner for his generous 
support during my graduate career at the University of Florida. His environment of 
enormous instrumental capabilities combined with a diverse, cooperative team has 
encouraged much intellectual and experimental freedom. Also, special appreciation is 
extended to Dr. Benjamin Smith, whose sound, patient guidance and encouraging support 
have given much direction to my research. 

I am also grateful to all past and present members of the Winefordner group for 
their wisdom, support, and friendship. I especially thank Dr. Igor Gornushkin for his 
resourcefulness and vast contributions to these investigations. I appreciate the many 
invaluable visiting scientists to the group, namely Dr. Jesus Anzano, Dr. Chris Stevenson, 
and Dr. Ga'bor Gal'bacs. Other researchers receiving special recognition include Dr. 
David Powell, Dr. Rolf Hummel, Lee Pearson, and Michael Stora. Apart from the 
laboratory, many people have contributed to the success of my graduate career. Jeanne 
Karably, Steve Miles, Larry Hartley, Joe Shalosky, Joseph Carusone, and Matt Glover 
are all appreciated for their friendships, fine workmanship, and priceless advice. 

I sincerely would like to extend gratitude to those professors who have planted 
deep roots not only in chemistry but also in my life prior to graduate school. Dr. K. C. 
Nainan of Stone Mountain High School introduced me to the concepts of chemistry and 
challenged me through national competition during my teenage years. On the 
undergraduate level, Dr. Larry McRae and Dr. Barbara Mixon allowed me the freedom to 

iii 



explore many intellectual pursuits and encouraged me to continue my explorations in 
graduate school. 

My family has always been a source of encouragement. I appreciate the liberty 
they have given me in pursuing my goals. I am deeply indebted to my parents for their 
continual love and sacrifice over the years. I also am grateful for the many relationships I 
have formed in Gator Christian Life, people with whom I have shared my life during my 
graduate career. I am extremely appreciative of my wonderful fiancee, Oleah Hodge, for 
her encouragement and invaluable advice. Jesus Christ, my Lord and Savior, is the 
foundation of my faith, without Whom none of this would have ever been possible. 
Finally, I acknowledge the financial support of IMC-Agrico, which has funded a portion 
of this research. 



IV 



TABLE OF CONTENTS 

page 

ACKNOWLEDGMENTS iii 

LIST OF TABLES vii 

LIST OF FIGURES viii 

ABSTRACT xi 

CHAPTERS 

1 INTENT AND SCOPE OF DISSERTATION 1 

2 INTRODUCTION TO LASER-INDUCED PLASMA SPECTROSCOPY 3 

Basic Principles 4 

Design Considerations 8 

Applications 1 1 

3 CHEMOMETRIC APPROACH TO LIPS 21 

Linear Correlation 22 

Rank Correlation 23 

Principal Component Analysis 25 

PCA of Phosphate Mining Samples 26 

Comparison of Chemometric Methods 28 

4 RAPID FIELD IDENTIFICATION OF PHOSPHATE MINING SAMPLES 46 

Phosphate Mining 46 

Background 48 

Preliminary studies 49 

Correlation Studies with the Benchtop Instrument 50 

Experimental Setup and Methodology 50 

Results 52 

Effect of Sample Position 53 

Continuous Correlation on a Moving Sample 54 

Fiber-Optic Probe 54 



Experimental Setup and Methodology 54 

Results 55 

Remote LIPS 55 

Experimental Setup and Methodology 55 

Results 56 

The Portable LIPS Probe 57 

Experimental Setup and Methodology 57 

Wet vs. Dry sampling 58 

Real Sample Analysis 58 

IMC-Agrico Site Visit 58 

Results 59 

Conclusions 62 

5 LIPS FOR CHARACTERIZATION OF ARCHAEOLOGICAL MATERIALS 92 

Introduction 92 

Experimental 94 

Instrumentation 94 

Micro-LIPS system 94 

Mini-LIPS system 95 

Samples 96 

LIPS Libraries 96 

Software 96 

Results and Discussion 97 

Summary and Conclusions 99 

6 ANALYTICAL MATRIX EFFECTS IN GEOLOGICAL MATERIALS 104 

Introduction 104 

Experimental 105 

Apparatus 105 

Sample Preparation 106 

Results and Discussion 107 

Conclusion 109 

7 CONCLUSIONS AND FUTURE WORK 1 16 

APPENDIX 

OPERATIONAL INSTRUCTIONS FOR THE PORTABLE LIPS PROBE 1 18 

REFERENCES 127 

BIOGRAPHICAL SKETCH 130 



VI 



LIST OF TABLES 

Table Page 

2-1 Early laser systems 13 

2-2 Comparison of lasers utilized in LIPS systems 18 

2-3 Advantages and disadvantages of laser-induced plasma spectroscopy 20 

3-1 Principal components and the amount of variance each includes 43 

3-2 Comparison of chemometric methods 45 

4-1 Average spectral line intensity ratios 65 

4-2 Correlation coefficients for overburden, matrix, and bedrock samples 68 

4-3 Identification of materials by chemical and LIPS analysis 70 

4-4 Criteria for classification by chemical analysis 71 

4-5 Identification using a fiber optic LIPS probe 78 

4-6 Correlation coefficients for overburden, matrix, and bedrock untreated samples 87 

4-7 Chemical and LIPS analysis results from FMC-Agrico samples 91 

5-1 Description of pottery samples 101 

5-2 Calculated probabilities in ceramic archaeological samples using the mini-LIPS 

system from 230-315 nm 102 

5-3 Calculated probabilities in ceramic archaeological samples using the micro-LIPS 

system from 180-315 nm 103 

6-1 Determination of iron in ores 112 

6-2 Determination of aluminum in particles of AI2O3 of different sizes 113 

6-3 Simultaneous determination of AI2O3 and SiC»2 in ore standards 115 



Vll 



LIST OF FIGURES 

Figure Page 

2-1 Laser plasma interaction and absorption 14 

2-2 Interaction of the plasma with the ambient atmosphere 15 

2-3 Evolution of a laser-induced plasma 16 

2-4 Typical LIPS set-up 17 

2-5 Temporal development of a series of lead emission lines in a LIPS plasma 19 

3-1 A plot of the intensities of a probe sample spectrum against the intensities of a 

spectrum in a spectral library 30 

3-2 A plot of the ranks of a probe sample spectrum against the ranks of a spectrum in a 

spectral library 31 

3-3 Laser-induced plasma spectral averages of three classes of phosphate mining 

samples: bedrock, matrix, and overburden 32 

3-4 Scree plot of log-eigenvalues of bedrock principal components 33 

3-5 Scores of the first two principal components for bedrock spectra 34 

3-6 Laser-induced plasma spectra of 30 bedrock samples 35 

3-7 Scree plot of log-eigenvalues of matrix principal components 36 

3-8 Scores of the first two principal components for matrix spectra 37 

3-9 Laser-induced plasma spectra of 30 matrix samples 38 

3-10 Scree plot of log-eigenvalues of overburden principal components 39 

3-1 1 Scores of the first two principal components for overburden spectra 40 

3-12 Laser-induced plasma spectra of 30 overburden samples 41 



vm 



3-13 Scree plot of log-eigenvalues of principal components of three classes of phosphate 

mining samples 42 

3-14 Linear discriminant analysis of bedrock, matrix, and overburden spectra 44 

4-1 LIPS spectra of overburden, matrix, and bedrock 64 

4-2 Schematic of the LIPS benchtop experimental system 66 

4-3 LIPS emission spectrum of matrix sample 67 

4-4 Correlation coefficients of matrix sample 69 

4-5 Correlation coefficients as a function of distance from focal length at maximum laser 

power 72 

4-6 Correlation coefficients as a function of distance from focal length at lower laser 

power 73 

4-7 Plot of correlation coefficients vs. distance using motorized sample translation 74 

4-8 Fiber-optic probe system 75 

4-9 Fiber optic LIPS probe 76 

4-10 Trigger circuit for Big Sky laser 77 

4-11 Experimental apparatus for remote LIPS 79 

4-12 Remote LIPS spectra 80 

4-13 Line intensity ratio as a function of distance in remote LIPS analysis 81 

4-14 Experimental setup of the field LIPS probe 82 

4-15 The field LIPS probe 83 

4- 1 6 Spectra of wet and dry bedrock samples 84 

4-17 Spectra of wet and dry matrix samples 85 

4- 1 8 Spectra of wet and dry overburden samples 86 

4-19 Bedrock correlation coefficients 88 

4-20 Matrix correlation coefficients 89 

4-21 Overburden correlation coefficients 90 



IX 



5-1 LIP spectra from archaeological ceramic samples 100 

6-1 Calibration curve of iron 1 10 

6-2 Fe & Al compounds prepared as pellets or powders 1 1 1 

6-3 Simultaneous determination of AI2O3 and Si02 in ore standards 114 

A-l View on the main menu window before any acquisition or processing of data 125 

A-2 Resulting screen after the data were collected and processed 126 












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 

LASER-INDUCED PLASMA SPECTROSCOPY 

IN THE ANALYSIS OF PHOSPHATE MINING 

SAMPLES AND ARCHAEOLOGICAL MATERIALS 

By 

Mark Anthony Villoria 
May 2002 

Chairman: Professor James D. Winefordner 
Major Department: Chemistry 

Laser-induced plasma spectroscopy (LIPS) is a versatile technique used in many 
academic and industrial settings. The LIPS technique is a well-established method for 
the rapid elemental analysis of various materials with little or no sample preparation. A 
laser pulse of sufficiently high power is tightly focused onto a sample surface. A hot, 
intense plasma is formed as the surface is heated by the laser and as material is ablated. 
The emitted radiation is spectrally resolved and the emitting species in the laser-induced 
plasma are identified. The elemental composition of the sample is then determined by its 
unique spectral wavelengths and line intensities. 

Identification of materials is achieved by using these spectral "fingerprints" that 
are unique to each sample. The focus of this research is to use these fingerprints in a 
variety of applications. Spectral data were analyzed by linear correlation, nonparametric 



XI 



rank correlation, and principal component analysis. The feasibility of using these 
chemometric techniques in LIPS was compared. 

Several studies involved the development and evaluation of various instrumental 
configurations, with the goal of optimizing a configuration for use in the phosphate 
industry. This research will discuss the development of field instruments that help to 
minimize contamination of matrix material (phosphate ore) by overburden or bedrock 
material through rapid field identification. Research has demonstrated the feasibility of 
accurately identifying material in its untreated, natural state with no sample preparation. 
The application of LIPS involves acquiring spectra of several selected samples, 
developing a library from these spectra, and using a correlation technique to match 
unknown spectra with well-characterized library spectra. Software was developed to 
rapidly carry out the correlation procedure and display material identification. 

Furthermore, LIPS was used to study the archaeological significance of certain 
ceramics from the first century BC. Finally, the analytical matrix effects commonly 
found in LIP spectra were investigated. 



xii 



CHAPTER 1 
INTENT AND SCOPE OF DISSERTATION 

It is the author's impression that the direction of graduate research does not 
always follow the path for which it was initially intended. Such is the case with the 
graduate career of the author. The original focus of a couple of years of research was the 
investigation of silicon surfaces in a new field known as Desorption/Ionization on Silicon 
Time-of-Flight Mass Spectrometry. However, the direction of research changed when 
financial support of LMC-Agrico became available. Thus, this dissertation represents 
only the research conducted in laser-induced plasma spectroscopy (LIPS) and does not 
include research involving mass spectrometry. 

Chapter 2 provides a brief overview of the LIPS technique, including an 
explanation of the theory, instrumental design, and applications. Some of the theories 
associated with statistical analyses, such as linear and rank correlation and principal 
component analysis, are explained in Chapter 3. Chapter 4 presents background 
information about the phosphate mining process relevant to the industrially driven 
research and includes the development and evaluation of different LIPS configurations 
for phosphate analysis. The following two chapters report on collaborative projects with 
Jesus Anzano, a visiting professor from the University of Zaragoza in Spain. Chapter 5 
reports the investigations on the analysis of archaeological materials by LIPS. This 
comprises experiments dealing with both qualitative and quantitative analyses of ancient 
pottery samples. Chapter 6 presents studies on the matrix effects in LIPS analyses. 



Finally, conclusions are presented in Chapter 7. An operational manual for the LIPS field 
instrument is included in the Appendix. 



CHAPTER 2 
INTRODUCTION TO LASER-INDUCED PLASMA SPECTROSCOPY 

Laser-induced plasma spectroscopy (LIPS) has enjoyed recent popularity as an 
analytical technique. Many research groups have realized the potential of LIPS, resulting 
in an ever-increasing number of publications. The reason for this success is perhaps the 
wide applicability of the technique as a method for elemental analysis. In the thirty years 
or so since its conception, thorough research has been conducted in academic, industrial, 
and government laboratories in an effort to comprehend its effectiveness in a variety of 
applications. 

The development of LIPS started in the early 1960's when Brech and Cross first 
demonstrated the possibility of using lasers as excitation sources in atomic emission 
spectroscopy [1]. By using a pulsed ruby laser and a pair of electrodes, they recorded a 
spectrum of elemental components in a microplasma following laser ablation. Soon 
afterwards, Runge et al. used a giant pulsed ruby laser to produce spectra from the 
coincident vaporization and excitation of metals and nonmetals [2]. Experiments showed 
that a breakdown of air occurred when laser radiation was brought tightly into focus. The 
purpose of these early experiments and others was to determine the mechanisms that led 
to this breakdown and to study the influence of various parameters (e.g., wavelength, 
focal diameter, pressure, pulse length, material) on breakdown thresholds [3]. 

Early instruments employed ruby or Nd:glass laser systems, a microscope for 
positioning and focusing the laser beam, graphite electrodes for additional excitation, and 
large spectrometers with photographic plates as emission detectors [4]. A summary of 



these systems is given in Table 2-1. These designs, however, suffered from poor 
precision due to laser instability and an inadequate spectral detection system. 
Improvements were made in the substitution of photographic plates with 
photomultipliers, and later charge-coupled devices (CCDs) and charge injection devices 
(CIDs) [4]. The evolutionary development of the LIPS apparatus incorporates such 
advances in optical detection and in laser technology. Due to the achievements of 
conventional atomic spectroscopy techniques such as atomic absorption spectroscopy and 
inductively coupled plasma optical emission spectroscopy, it was not until the 1980's that 
LIPS began to find widespread use in chemical analysis [5]. Presently, with the advances 
in laser technology, LIPS has become reliable, versatile, and readily available at a low 
cost. The present design complements the advantages of almost no sample preparation, 
the analysis of all states of matter, and its capability of simultaneous multi-element 
detection. 

Basic Principles 

In LIPS, a form of atomic emission spectroscopy, a high-power density laser 
pulse is focused onto a target material. A small portion of the sample is vaporized, and a 
hot plasma develops. The emitted radiation from the short duration plasma is collected 
by a spectrometer and analyzed for elemental composition. 

Ahmad and Goddard indicate five steps in the LIPS process: heating, melting, 
vaporization, excitation, and ionization [6]. As the laser pulse strikes the surface of a 
solid sample, an initial heating causes the temperature of the surface to rise to thousands 
of degrees. The surface responds by melting, and a small portion is vaporized into an 
ionized gas. In order to form this ionized gas, the metal from the surface is oxidized by 



the loss of electrons. There are two main mechanisms in which the ionized metal may be 
formed [3] : 

(eqn. 1) e' + M— »2e" + M + Electron ionization 

(eqn. 2) M + mhv -» M + + e Multiphoton ionization . 

In electron ionization (eqn. 1), electrons absorb laser radiation, and then collide 
with the neutral metal to ionize the solid which forms a gas. The electrons must acquire 
an energy greater than the band gap of the solid (or the ionization energy of the gas). 
Consequently, the electron concentration increases exponentially with time throughout 
the lifetime of the laser pulse, leading to a cascade breakdown. 

Multiphoton ionization (eqn. 2) is characterized by the absorption of multiple 
photons by an atom or molecule. When the energy of these photons is sufficient to eject 
an electron from the valence to the conduction band of the solid, a metal ion and an 
electron result. This electron, in turn, can react with neutrals to again ionize the metal 
(eqn. 1). 

Thus, electrons that absorb the photons undergo many collisions with themselves 
as well as with surrounding atoms. The energy absorbed by the electrons is distributed 
and passed on to the metal lattice. Once in the lattice, the energy is instantaneously 
converted to heat, causing a rapid rise in the surface temperature of the material, 
ultimately resulting in vaporization. For evaporation to occur, the energy deposited in the 
layer of molten metal must be greater than the latent heat of vaporization of the target, L v 
(J/g) [5]. Thus, no evaporation occurs below the minimum absorbed irradiance, F m j„ 
(typically 10 12 W/m 2 for a Q-switched laser). The two are related in the following 
equation: 



(eqn. 3) F min = p U a /, 



V4* -Vi 
e 



where p is the mass density of the target (g/mm 3 ), a is the thermal diffusivity (mm 2 /s), 
and t e the duration in seconds of the laser pulse. 

Equation 3 shows the dependence of the vaporization threshold on the duration of 
the laser pulse. Vaporization does not result from high power short pulses, whereas 
longer, lower power pulses produce deep holes in the target. Accordingly, an equation 
has been derived which relates the time of vaporization, t v (s), to the irradiance of the 



laser, F(W/cm 2 ): 



(eqn. 4) t v = 



kK P C(T v -T ) 2 



AF 2 



where K is the thermal conductivity (J/s cm K), p is mass density (g/cm ), C is heat 
capacity per unit mass (J/g K), and T v and T„ are the vaporization and initial temperatures 
in Kelvin, respectively. This equation is used to estimate the surface temperature, depth 
vaporized, or the time taken to reach the boiling temperature when irradiance is close to 
the threshold value [5]. 

Once the vaporization is established at the sample surface, gas dynamic processes 
govern the behavior of the plasma. These processes are based on the two assumptions 
that the laser beam is always spatially and temporally uniform within its extent and 
duration, and that the molten material ejected is negligible compared with atomic ejection 
[6]. 

As the plasma is formed, the vapor pressure increases, thus affecting the laser 
absorption. The weakly ionized plasma is partially transparent to the laser beam, 
allowing direct heating of the target surface to continue. The inverse Bremsstrahlung 
process heats primarily the electrons, which consequently increase the plasma 
temperature and electron density [3]. At high laser powers, the number density of 



7 

electrons increases to a point (critical electron density), which makes the plasma opaque, 
preventing the laser from reaching the underlying target. (If the laser radiation has a 
wavelength greater than the plasma wavelength, X p , light is reflected and the plasma 
shrinks instead of expands, as it is no longer absorbing the light. The plasma wavelength 
is defined by X p ~ 3.35 x 10 7 (n e )' v \ where n e is the electron density in the partially 
ionized layer.) Heating of the surface continues only indirectly by thermal conduction 
from the plasma. Figure 2-1 shows the absorption processes that occur as the surface is 
ablated. 

Laser heating of the plasma continues as the plasma plume grows toward the laser 
[7]. The plasma thus rapidly heats and expands. Therefore, the laser is no longer heating 
the surface, only heating the already vaporized sample. As the plasma expands, its 
volume increases, reducing the density of the plasma. This decrease in density allows the 
laser radiation to once again penetrate the surface of the sample, causing vaporization. 
The cycle of vaporization, heating, and expansion is repeated throughout the laser pulse 
[5]. 

Expansion of the plasma allows interaction with the surrounding atmosphere. 
This atmosphere may be simply the ambient gas that existed before the laser pulse, or in 
the case of a vacuum, it may be the neutral gas species resulting from the escape of fast 
ions [3]. The ablated material, in the form of particles, free electrons, atoms, and ionized 
atoms, expands at a velocity much faster than the speed of sound and forms a shock wave 
in the surrounding atmosphere. Behind the shock wave is a region of the shock-heated 
ambient gas followed by the expanding plasma [3]. Figure 2-2 illustrates the interactions 
between the plasma and the atmosphere. 



8 

After several microseconds, collisions with ambient gas slow down the plasma 
plume, and the shock wave detaches from the plasma front. Plasma temperatures in the 
range of 10 4 to 10 5 K and electron densities on the order of 10 15 to 10 19 cm 3 have been 
measured [8]. The plasma then decays through radiative quenching and electron-ion 
recombination processes that lead to formation of high-density neutral species in the post- 
plasma plume. Decay ends with the formation of clusters and with the thermal and 
concentration diffusion of species into the surrounding gas. The emitted radiation 
(integrated over the first tens of microseconds) is spectrally resolved, and the emitting 
species in the laser-induced plasma are identified and quantified by their unique spectral 
wavelengths and line intensities. 

The LIPS plasma is similar to those plasmas used in the conventional atomic 
emission methods such as electrode spark, arc, and inductively coupled plasma. These 
techniques require the sample to be placed into a plasma for excitation and emission. The 
LIPS plasma, however, is formed from the sample and therefore contains the desired 
metal. Figure 2-3 provides a visual summary of the stepwise evolution of the plasma, as 
well as an approximate time scale for each event. 

Design Considerations 

Many LIPS designs have been proposed and used successfully in a variety of 
applications. Some even incorporate portable capabilities for field analysis, as discussed 
in Chapter 4. Each of these systems consists of two major parts, one for production of 
the plasma and another for analysis of the radiant emission from the plasma. Plasma 
production usually incorporates a pulsed laser, a radiation delivery system, the target, and 
a movable stage for the sample. The emitted radiation is analyzed using collection optics, 






a dispersion system, a detector, and a computer for control, data acquisition, and analysis. 
A simplified LIPS system is shown in Figure 2-4. 

In order to generate the plasma, the laser must generate pulses of sufficient power. 
Such suitable lasers include solid state lasers, gas lasers (C0 2 and excimer), and Nd:YAG 
pumped- and flashlamp pumped-dye lasers [3]. Table 2-2 presents a general comparison 
of these three types of lasers. 

Solid state lasers, such as Nd:YAG lasers, are commonly used in LIPS analyses 
because of their good output reproducibility, compactness, and high irradiance [5]. The 
fundamental wavelength of 1064 nm is the most widely used, although frequency 
doubled (532 nm), tripled (355 nm), and quadrupled (266 nm) wavelengths have also 
been successful. A Nd:YAG laser is a solid state laser composed of Nd + ions in an 
yttrium-aluminum-garnet host [9]. A Q-switch mode is responsible for the laser pulse. 
In this mode, the cavity of the laser is only switched on after the population inversion has 
been allowed to grow greater than the threshold value. By switching on the cavity past 
the threshold, the laser emits a very robust pulse [10]. This laser provides a reliable pulse 
(-10 ns) with an energy between 10 - 100 mJ. It has a small beam divergence allowing 
for efficient focusing and a choice of operation wavelengths [11]. 

To produce a sufficiently high power density, the laser is focused to a spot size of 
about 1 mm 2 by means of a spherical lens [12]. An XYZ translation stage moves the 
target. The stage allows movement of the sample between laser pulses, allowing fresh 
sample surface to be exposed for each pulse. The emitted radiation from the plasma is 
collected and dispersed by the spectrometer and quantified by the detector. A black box 
set-up of a typical LIPS apparatus can be seen in Figure 2-4. 



10 

As the laser pulse is emitted from the laser cavity, two processes occur. The 
plasma is formed and an array detector detects the laser pulse. The plasma shown in 
Figure 2-2 consists of three primary regions: the high temperature core, the lower 
temperature middle, and the expanding shock wave discussed above. Under atmospheric 
conditions, the total volume occupied by the plasma is about 3 mm' and its lifetime is 
approximately 50 microseconds. The photodiode, connected to the photodiode arrary 
(PDA), triggers the detector to begin recording specta. 

Figure 2-5 shows the decay of the lead emission spectra observed upon analysis 
of a lead containing sample. In this case, over a 14 (is interval, emission signals rise to a 
maximum and then begin to decay. Thus, the use of gated detection allows optimization 
of the signal to noise ratio. Usually, light emitted only 5 - 20 |is after plasma formation 
is detected [13]. The ideal time delay is crucial as a discrimination between the 
background produced by the Bremsstrahlung continuum and the emission lines of the 
sample must be made [14]. 

Light is collected and directed to the detection system using a fiber optic cable. In 
some cases, a lens is used to focus the emitted light onto the end of the fiber optic [10]. 
By using a fiber optic to collect light, sensitivity of detection to spark position is reduced 
because of the large acceptance angle of the fiber [15]. In addition, the fiber optic allows 
the detection system to be positioned remotely from the plasma emission. 

The fiber optic transfers the collected emission to narrow pass filters, a 
monochromator, or a spectrograph. The composition of the sample and the application 
determine which device is most appropriate for analysis. The spectrally resolved light is 
then detected using either a photomultiplier tube (PMT) or an array detector. The PMT is 
used to monitor a particular emission line, while the array detector allows for a 



11 

continuous recording of the spectrum [10]. Finally, a computer records the line 
intensities for further analysis. 

Applications 

There are several attractive features inherent to LIPS. A summary of these, as 
well as disadvantages, is given in Table 2-3. One of the most attractive characteristics is 
its capability in the remote sensing and process monitoring areas, possible since only 
optical access to the sample is required. Another attractive feature is its simultaneous 
multi-element capability with minimal, if any, sample preparation. There is usually no 
sample preparation, which eliminates the need for tedious and time-consuming sample 
digestion and preparation procedures. This increases throughput, since the analyte signal 
is not reduced by dissolution or contaminated with chemical reagents. This advantage is 
extremely important in the analysis of phosphate samples (Chapter 4). However, some 
severe problems, such as variable mass ablation, must be overcome before the technique 
can reach its full potential. Other problems to consider include sample heterogeneity, 
particle size, and the effect of moisture on samples. 

LIPS has applications in numerous fields, including biology, the environment, 
geology, metallurgy, and nuclear industry, to name just a few. New applications are 
continually being discovered, and a comprehensive review on all applications is virtually 
impossible. Here, a few examples are briefly discussed to survey the wide applicability 
of the technique. 

The detection of almost 20 elements present in biological fluids such as serum 
and blood was shown to be possible by Loree [16]. Reported levels were as low as 50 
ng/ml. Radziemski et al. have used LIPS for the direct detection of dangerous elements 
(chlorine, fluorine, and beryllium) in the atmosphere [17]. Other environmental 



12 

applications include those of the Cremers group and Ciucci et al. The former group 
evaluated their instrument for the analysis of metal in soils, paints, and particles collected 
on filters [12], while the latter used the detection of dangerous soil pollutants to estimate 
the sensitivity of the technique [14]. 

In metallurgy, the characterization of impurities and the inhomogeneity of 
nominally pure metal, such as in corroded materials, is extremely important [5]. In-situ 
analysis of metals in their operation area, such as at a nuclear reactor, is also an important 
measurement. One group used fiber optics for remote measurements of elemental traces 
in the hostile environment of nuclear reactor buildings [11, 18]. 

Obvious chemical applications include surface analysis and depth profiling. 
Surface analysis is achieved by altering the position of the target as described above. 
Since the technique is minimally destructive, LIPS has been used in the pigment 
identification of painted artworks [19]. In depth profiling, successive laser pulses are 
shot at a stationary target, and the spectrum is recorded as material is ablated [5]. 



13 



Table 2-1 . Early laser systems. 



Manufacturer 


Model No. 


Laser type 
(wavelength) 


Year of 
manufacture 


Thermo Jarrell Ash 
(Franklin, MA) 


Mark I 


Ruby (694.3 nm) 


1963 


Markn 


Nd:glass(1060nm) 


1966 


LOMO (Leningrad, 
Russia) 


MSL2 


Nd:glass(1060nm) 


1967 


Shimadzu (Tokyo, 
Japan) 




Ruby (694.3 nm) 


1967 


CISE Segrate 
Milano (Milan, 
Italy) 


CISE I 


Ruby (694.3 nm) 


1966 


Ford Motor Co. 
(Dearborn, MI) 






1963 


VEB Carl Zeiss 
(Jena, Germany) 


LMA 


Nd:glass(1060nm) 
or Ruby (694.3 nm) 


1964 


LMAI 


Nd:glass(1060nm) 


1965 


Optical Technology, 
Inc. 


Model 120 


Nd:glass(1060nm) 


1965 


Model 190 


Nd:glass(1060nm) 


1968 









14 



• Thomson scattering 

Optical probing • Faraday rotation 

• Interferometry 




Hard x-rays 
Soft x-rays 

Harmonics (ncoo) 
Raman scattered light 

Laser coo 



Ablation 
surface 



Critical n c /4 Expanding 

surface sheath and 

(n c ) plasma blowoff 



Figure 2-1. Laser plasma interaction and absorption [3]. 



15 




Hot, high-pressure 
strongly absorbing 
vapor plasma 



Temperature 
profile 



Conduction 



Shock wave 



Radiation 



Ambient 
Atmosphere 



Figure 2-2. Interaction of the plasma with the ambient atmosphere [3]. 









16 



Laser strikes 
the surface 



t = Ons 



Instant increase in 
surface temperature 



:.v-;.;-v.v.;-v.v 

t~l ns 



Vaporization starts, 
heat dissipates slowly 



t~l ns 



Underlying layers 
reach critical T, p 

►- 




t~lns 



Surface explodes, 
breakdown occurs, 
plasma forms 




t ~5ns 



Plasma expands, 
becomes opaque, 
heats environment 




t~10ns 



Plasma continues 
expanding, shock 
wave forms 




Plasma cools 
down, becomes 
unstable, decays 



t - 1000 ns 




t ~ 10000 ns 



Strong electron-ion 
recombination, 
neutrals form 



Plasma products 
diffuse into 
environment 



t - 20000 ns 



Figure 2-3. Evolution of a laser-induced plasma. 






17 



Pulsed 














Display 


laser 




Time 
control 




■■■1 • • 






< 






c 


r ' 


i 


T arises 


iH IfTffm 


i\ 


S r 








llllllllll w 




plasma 


Spectrograph ££L 



Figure 2-4. Typical LIPS set-up [3]. 



18 





Table 2-2. Comparison of lasers utilized in LIPS systems. 


Type 


Wavelength 


Pulse width 
(nsec) 


Pulse energy 
(J) 


Comments 


Nd:YAG 


1.06 
0.53 


7-12 


0.3 to 1.0 


Compact, low 
maintenance, reasonably 
priced, glass optics 


C0 2 


10.6 


1 to 300 |isec 


0.5 to 500 


Simple design, inert gas, 
special IR optics required 
for laser pulses 


Excimer 


0.194 to 0.351 


10 to 30 


0.25 


Rep rates to 250 Hz, UV 
wavelengths, toxic gases, 
quartz optics required for 
laser pulses 


Source: [3" 











19 



Pb 
357.27 nm 






Pbl 
368.35 nm 



Pbl 
373.99 nm 




355 



t— -i 1 1 1 r 

365 370 



Wavelength (nm) 



Figure 2-5. Temporal development of a series of lead emission lines in a LIPS plasma. 



20 



Table 2-3. Advantages and disadvantages of laser-induced plasma spectroscopy. 



Advantages 



Disadvantages 



1 . Minimal or no sample preparation 

2. All states of matter can be analyzed, as 
well as both conductive and nonconductive 
samples 

3. Very small amounts of material are 
vaporized 

4. Easy analysis of refractory materials 
such as ceramics 

5. Microanalysis is possible with spatial 
resolving powers of 1 - 10 ^.m 

6. Capability of remote analysis in harsh 
environments 

7. Atomization and excitation are in one 
step 

8. Capable of simultaneous multi-element 
analysis 



1 . Variation in the mass ablated caused by 
changes in the bulk matrix 

2. Difficulty in obtaining matrix matched 
standards 

3. Detection limits higher than standard 
solution techniques (e.g. ICP - OES) 

4. Poor precision, typically 5 - 10% 

5. Standard emission disadvantages, such 
as spectral interferences and self-absorption 

6. Possibility of optic damage from high 
energy density lasers 

7. Complexity of the operation of the 
system 






CHAPTER 3 
CHEMOMETRIC APPROACH TO LIPS 

Experimental data in this dissertation were analyzed using three different 
statistical methods: linear correlation, nonparametric rank correlation, and principal 
component analysis. Linear and rank correlations were each used extensively in the 
identification of numerous samples because of the familiarity of these techniques with 
laser-induced plasma spectra [20]. LIPS is often used to identify the elemental 
composition of materials by the presence of one or more spectral lines in the sample's 
emission spectrum. Although a detailed chemical composition could be obtained, the 
goal of this work was to use a material's spectrum as a "fingerprint" for instant 
identification of that material. This fingerprint is a LIP spectrum that is compared to 
many other spectra in a spectral library. Each material has its own unique fingerprint, 
thus giving a unique positive identification of that material. 

The primary objective of these statistical methods is to identify a compound 
belonging to a known class of compounds stored in a certain spectral library. A probe 
spectrum can be sequentially superimposed with the library spectra and the difference or 
similarity will immediately show up even without the use of a computer. In many cases, 
however, visual identification is not obvious, especially in transition regions. The spectra 
may look close to identical. Thus, powerful statistical methods are required in order to 
reliably identify such materials. 

Goals for the qualitative identification of the compounds investigated included the 
rapid identification of material without extensive, laborious calculation. The 

21 



22 

incorporation of the statistical method into customized software was critical. This was 
especially important for those samples used in the phosphate industry, as the software 
was integrated into a unique instrument adapted for rapid field analysis (as discussed in 
Chapter 4). 

The correlation methods of choice, then, were linear correlation and 
nonparametric rank correlation, for these were most suitable for compound identification. 
The choice of correlation method is determined by the particular experimental 
arrangement, the type of data obtained, and time requirements [20]. In this research, 
linear and nonparametric rank correlation were completely adequate for spectral 
identification of each sample spectrum, which consists of 2048 data points. The potential 
of using another statistical method, principal component analysis, with LIPS data was 
also briefly investigated. 

With the various instrumental configurations described in Chapter 4, spectra of 
several selected samples were acquired and used to compile the LIPS spectral libraries. 
A correlation technique was used to match unknown spectra with well-characterized 
library spectra. Software was developed to perform the correlation algorithm, rapid data 
processing, and simple spectrometer operation. 

Linear Correlation 

Linear correlation (also known as Pearson product moment correlation, or more 
simply, Pearson's correlation) measures the association between variables. The 
correlation between two variables reflects the degree to which the variables are related. 
The goal of a simple linear correlation is to determine whether a change in one of the 
independent variables is associated linearly with a change in the other independent 
variable. The linear correlation coefficient r is calculated as: 



23 

r = [ i(x,-x)(y,-y) ]/[ E(x,-x) 2 i(y,-}0 2 ] 1/2 

where X is the mean of X,'s, and Y is the mean of K,'s. A value of r = 0.0 indicates no 
linear relationship between the two variables and that these two variables are 
uncorrelated. However, a value of +1.0 signifies a perfect positive linear relationship 
between variables. This complete positive correlation occurs when the data points lie on 
a perfect straight line with the positive slope; high values on the x axis (wavelength) are 
associated with high values on the y axis (spectral intensity). 

As an example, in the phosphate rock samples, spectral libraries were compiled 
from the various types of material that could be examined. The spectral intensities of a 
new sample spectrum were plotted against the spectral intensities of those spectra in the 
library (Figure 3-1). The algorithm determines the closest match by using the above 
equation, and the new sample is identified as the material in the library that has the 
highest correlation coefficient. Practically, this is the spectrum that the new sample most 
closely resembles. 

Rank Correlation 

The linear correlation coefficient r does not take into account the individual 
distributions of x and y. Hence, the linear correlation coefficient r is not the most 
accurate statistic for deciding whether an observed correlation is statistically significant. 
This is especially important in this research, since fluctuating single-shot spectra (x's 
with different distributions) are often compared with stable library spectra (y's with 
similar distributions) [20]. Nonparametric rank correlation (also known as Spearman's 
rank correlation) is likely to be more robust since the numbers are drawn from a perfectly 
known distribution function. 



24 

Nonparametric rank correlation can be applied to compare two independent 
random variables. Unlike the linear correlation, nonparametric rank correlation works on 
ranked data, rather than directly on the data itself. It is with ranking (or relative) 
measurements, as opposed to linear measurements, that the nonparametric rank 
correlation method is used to analyze data. Similar to the linear correlation coefficient r, 
the Spearman's r coefficient indicates agreement. A value of r near one indicates good 
agreement; a value near zero implies poor agreement. Because of its dealings with 
ranked data, nonparametric rank correlation does not make any assumptions about the 
distribution of the underlying data [21]. 

The nonparametric rank correlation method assigns a rank to each observation in 
each group separately. Each value in a spectrum is replaced with the value of its rank, an 
integer between 1 and 2,048 in accordance with its magnitude [20]. Thus, the most 
intense pixel in a spectrum is assigned the number 2,048, since there are 2,048 data points 
(or pixels) in the spectrometer. The resulting list of numbers is drawn from a perfectly 
known distribution function, namely, uniformly from the integers between 1 and 2,048. 
Each integer in the distribution function occurs precisely once. For this research, the 
ranks of the probe sample spectrum were plotted against the ranks of a spectrum stored in 
a spectral library (Figure 3-2). The equation for nonparametric rank correlation is the 
same as that of linear correlation, with the exception that the values of the x's and y's are 
replaced by their corresponding ranks R's and S's: 

r = H.(Ri- R) (S, - 5) ] / [ Z (/?,- - R) 2 I (5, - S) 2 ] m . 



25 



Principal Component Analysis 

Principal component analysis (PCA) is a statistical method used to break down a 
set of data into its most basic variations. PCA is widely used in a variety of disciplines, 
some of which include signal processing, statistics, and neural computing. In some 
application areas, PCA is also known as the Karhunen-Loeve transform, the Hotteling 
transform, or eigenanalysis. The PCA algorithm can be applied to sets of spectroscopic 
data from plasma spectra, since the spectroscopic data consist of lists of measurements 
made on a collection of objects. 

There exists a number of objectives of principal component analysis. The first is 
to reduce the dimensionality of data. Reduction of dimensionality is practical if the new 
axes account for approximately 75% or more of the variance in a data set. Another goal 
is to determine linear combinations of variables. Because eigenvectors are reduced to a 
centered point, linear combinations to relate points (spectra) to each other may be found. 
Next, PCA allows the visualization of multidimensional, or multivariate, data. The 
variance explained by a pair of axes defining a plane can be viewed on a planar plot. 
Finally, PCA permits the identification of groups of spectra or of outliers. Visual 
inspection of a planar plot indicates objects that are grouped together, thus indicating that 
they belong to the same type of compound or result from the same process. Anomalous 
objects may also be detected, in which case they may be excluded from analysis because 
of the perturbation that they introduce or that analysis may require repetition. 

Spectroscopic data could be classified as multivariate data. The information from 
a set of spectra could be organized such that each datum in the data set is identified with a 
point. Of a sample set of about 20 spectra, for example, one spectrum can be reduced to 



26 

a single point on a graph of much fewer (2 or 3) principal components. This reduction in 
the dimensionality of data aids material classification. 

In the spectroscopic analysis of real samples, a spectrum may be defined not only 
by the elemental composition of the sample, but also by the effect of a number of 
variables. Constituents within the sample may interact; detection, including noise, may 
vary among instruments; environmental conditions may affect the baseline; samples may 
be prepared or handled differently. Despite these variations, however, there must always 
be a finite number of independent variations occurring in the spectral data. It is likely 
that the largest variations in the spectral set would be the changes in the spectrum due to 
the different concentrations of the constituents of the mixtures. 

PC A breaks apart the spectral data into the most common spectral variations 
(factors, eigenvectors, loadings) and the corresponding scaling coefficients (scores). In 
any set of spectra for this research, the data is 2,048 dimensional, since there exist 2,048 
pixels in each spectrum. Because of the extremely large dimensionality of this data set, it 
is beneficial to have this dimensionality reduced to principal components to observe 
groupings in the data. The detailed algorithm for determining principal components is 
quite extensive and can be found in the literature [22, 23]. 
PCA of Phosphate Mining Samples 

Ninety spectra were taken of three categories of phosphate material: 30 each of 
bedrock, matrix, and overburden. Figure 3-3 displays the laser-induced plasma spectral 
averages of these three classes of phosphate mining samples. Because the differences in 
these samples lay in the various concentrations of its constituents, the main variation in 
these spectra is in the intensity of certain peaks, or, more accurately, the ratio of the 
intensities of distinctive peaks relative to others. The intensities of each of these spectra 



27 

were compiled in a database; the final worksheet contained a matrix of 90 x 2,048 cells 
(number of spectra x pixels). The principal component analysis was conducted using 
customized programs written in MATLAB. 

A log-eigenvalue plot (Figure 3-4) can be generated to illustrate the number of 
principal components that may be used. Each of the points of the resulting "scree" plot 
indicates one spectrum. This scree plot shows a number of data points that lie along an 
imaginary line. The points that deviate most from this line are considered to be 
significant principal components. Here, about 4 to 8 principal components may be 
considered significant. 

Figure 3-5 is a plot of scores of the first two principal components plotted against 
one another for 30 bedrock samples. These scores represent the maximum variation of 
spectra. As shown, points 2 and 3 are situated at positions farther from the rest of the 
points. An observation of the superimposed spectra of all 30 bedrock samples (Figure 3- 
6) reveals spectra 2 and 3 to contain more broad peaks than that of the remaining spectra. 
Thus, PCA can be used to show slight variations in spectra while still retaining important 
spectral information. 

Principal component analyses were likewise performed on matrix and overburden 
spectral data. The resulting data from the similar analyses are shown in the scree plots, 
plots of scores, and superimposed spectra (Figures 3-7 to 3-12). 

The final scree plot from all 90 spectra is presented in Figure 3-13. The plot, 
along with a corresponding table of eigenvalues (Table 3-1), shows the reduction in 
dimensionality to 7 principal components to account for 99.15% of the cumulative 
variance. 



28 

Discriminant function analysis is commonly used to determine which variables 
discriminate best between two or more groups. The basic idea underlying discriminant 
analysis is to determine whether groups differ with regard to the mean of a feature 
variable. This variable is then used to predict group membership. If discriminant 
function analysis is effective for a set of data, the classification of correct and incorrect 
estimates will yield a high percentage correct. Linear discriminant analysis was used in 
the classification of the phosphate spectra. Further computational analysis in MATLAB 
allows a two-dimensional linear discriminant function plot of all three groups of spectra 
(Figure 3-14). The plot shows the distinct regions for the three groups. As expected, the 
bedrock spectra are the most distinct, separated on the plot very noticeably. Overburden 
and matrix are more similar, but yet separated very well. 

The goals of PCA have been accomplished with the analysis of spectra from 
bedrock, matrix, and overburden samples. The dimensionality of the data was reduced 
from a 90 x 2,048 data set to a two-dimensional data plot with 90 points. A set of 
eigenvectors, calculated from the original calibration data, serve as scaling factors in a 
linear combination of the included spectra. Finally, the resulting two-dimensional plot 
allows the visualization of classes of spectra. 

Comparison of Chemometric Methods 

Table 3-2 lists the results of linear correlation, rank correlation, and principal 
component analysis. Principal component analysis yields results that are considerably 
more accurate than that of correlation analysis. The percent error for each of these 
analyses was 16.7%, 17.8%, and 1.11% for linear correlation, rank correlation, and 
principal component analysis, respectively. 



29 

Due to the ranking process in nonparametric rank correlation, the relative 
magnitudes of high intensity peaks are reduced, whereas the magnitudes of low intensity 
peaks are enhanced. This sensitivity to background noise is one major disadvantage of 
nonparametric rank correlation. Another drawback is that it is slower than linear 
correlation, since the signal intensities (pixels) must first be rearranged (ranked) prior to 
the correlation. The reduction of speed, however, is only on the order of tens of seconds, 
at most, for analyses containing fewer than a hundred spectra. 

Nevertheless, linear correlation was primarily used to conduct the research in this 
dissertation. Principal component analysis, while a more precise algorithm for the 
classification of materials, requires hours of data post-treatment. In order to accumulate 
the data needed for PCA, it was first necessary to combine all the spectral data into a 
worksheet for import into the MATLAB program. Customized programs then allowed 
the handling of the data that concluded in the visualization of the data from a two- 
dimensional discriminant functions plot. While the hours of data post-treatment may be 
reduced to minutes with specific, customized computer programs, the handling of 
massive amounts of spectroscopic data would remain an enormous computational effort. 
The linear and rank correlations, however, are simple, robust, and not as mathematically 
challenging. For the purposes of the research included here, the accuracy of these 
correlation methods was entirely satisfactory. 



30 



a 

1 c 

w h 

V y 

*3 v 

e en 
B 



1400 

1200 

1000 

800 

600 

400 

200 





Linear Correlation 



-♦♦- 




^^ 



200 



400 



600 



800 



Intensities Library Spectrum 



Figure 3- 1 . A plot of the intensities of a probe sample spectrum against the intensities of 

a spectrum in a spectral library. 






31 



E 2000 



Rank Correlation 




500 1000 1500 

Rank Library spectrum 



2000 



Figure 3-2. A plot of the ranks of a probe sample spectrum against the ranks of a 

spectrum in a spectral library. 






250 



260 



32 



Emission Spectra - Class Averages 




270 



280 



290 



Wavelength, ram 



300 



310 



Figure 3-3. Laser-induced plasma spectral averages of three classes of phosphate mining 

samples: bedrock, matrix, and overburden. 



33 






10 



10 



10 



™io 7 

o 



10" 



10" 



1(f 



Log-Eigenvalue Plot 



4-8 significant PCs 



* 



** 



* ■*• 



*********- 



10 15 20 

PC number 



25 30 



Figure 3-4. Scree plot of log-eigenvalues of bedrock principal components. 



34 



3000 



Bedrock Data Scores Plot 

l 




-WHO 



Score* fix PC #1 



Figure 3-5. Scores of the first two principal components for bedrock spectra. 






35 



4000 



3500 



3000 



2500 



2000 



E 

m 



1500 



1000 



500 



Bedrock Data - All Spectra 




Variable Index 



Figure 3-6. Laser-induced plasma spectra of 30 bedrock samples. X-axis is pixel 
number. Anomalies (#2 and #3) in spectra are easily visible. 






36 



1 r -*- 



10 J 



Log-Eigenvalue Plot 



10 



1 10' 



10' 



10 J , 



10 






4 PCs significant 



*# 



** 



** 



* * 



'*#■ 



10 15 20 

PC number 



25 30 



Figure 3-7. Scree plot of log-eigenvalues of matrix principal components. 



37 



Matrix Data Scores Plot 




-1 o i 

Scores lor PC #1 



Figure 3-8. Scores of the first two principal components for matrix spectra. 



38 



Matrix Data - All Spectra 




200 



400 600 



800 1000 1200 

Variable Index 



1400 



Figure 3-9. Laser-induced plasma spectra of 30 matrix samples. X-axis is pixel number. 
Greater inhomogeneity exists in matrix spectra. 






39 



10' 



10 



Log-Eigenvalue Plot 



> 

LU 
8* 



10 



10 l 



10" 



10 



4-8 significant PCs 



*** 



* * * 



****** 



* * * : 



10 15 20 25 30 

PC number 



Figure 3-10. Scree plot of log-eigenvalues of overburden principal components. 









40 



Overburden Data Scores Plot 



5000 






;"«» i 



■10000 

1 



-05 



0.5 



— - 1 


■■ i 


— 


■ 


















• \ 


*2 

• 
















• • 


• 








• • 


#i 






\ • 










\ #26 










- 








.... 






! 


i.. - 





1.5 



Scores for PC #1 



»10 



Figure 3-11. Scores of the first two principal components for overburden spectra. 



41 



Overburden Data - All Spectra 




1000 12O0 

Variable Index 



Figure 3-12. Laser-induced plasma spectra of 30 overburden samples. X-axis is pixel 

number. 






42 



Log-Eigenvalue Plot 




10 20 30 40 50 60 70 80 

PC number 

Figure 3-13. Scree plot of log-eigenvalues of principal components of three classes of 

phosphate mining samples. 









43 



Table 3- 



1 . Principal components and the amount of 


variance each i 


Principal 
Component 


Eigenvalues 


Variance 


Cumulative 
Variance 


1 


2.04 x 10 1U 


59.36% 


59.36% 


2 


1.07X10 1U 


31.14% 


90.50% 


3 


1.39 x 10 y 


4.04% 


94.54% 


4 


8.98 x 10 s 


2.61% 


97.15% 


5 


4.05 x 10 s 


1.18% 


98.33% 


6 


1.71 x 10* 


0.50% 


98.83% 


7 


1.12 x 10" 


0.33% 


99.15% 


8 


5.73 x 10' 


0.17% 


99.32% 


9 


4.77 x 10' 


0.14% 


99.46% 


10 


4.42 x 10' 


0.13% 


99.59% 


11 


2.96 x 10 7 


0.09% 


99.67% 


12 


2.23 x 10 7 


0.06% 


99.74% 


13 


1.59 x 10 7 


0.05% 


99.78% 


14 


1.43 x 10' 


0.04% 


99.83% 


15 


1.20 x 10 7 


0.03% 


99.86% 


16 


7.05 x 10" 


0.02% 


99.88% 


17 


6.03 x 10 b 


0.02% 


99.90% 


18 


4.64 x 10 b 


0.01% 


99.91% 


19 


4.59 x 10 b 


0.01% 


99.93% 


20 


3.35 x 10 b 


0.01% 


99.94% 



includes. 



44 



8000 



6000 



4000 



2000 



g -2000 

u. 
O 

_i 

-4000 



•6000 



BOOO 



-10000 





J. 



J_ 



95% probability e%>ses shown 
' 1 



• Bedrock 
■ Matrix 

♦ Overburden 



~r 



-8000 -6000 -4000 -2000 



2000 4000 6000 8000 1000C 
LDF#1 



Figure 3-14. Linear discriminant analysis of bedrock, matrix, and overburden spectra. 






45 



Table 3-2. Comparison of chemometric methods. 
Linear correlation: 16.7% error 



Actual 


Predicted 


Bedrock 


Matrix 


Overburden 


Bedrock 


30 








Matrix 


13 


16 


1 


Overburden 





1 


29 



Rank correlation: 17.8% error 



Actual 


Predicted 


Bedrock 


Matrix 


Overburden 


Bedrock 


30 








Matrix 


13 


16 


1 


Overburden 





1 


29 



Principa 


component analysis: 1.11% error 


Actual 


Predicted 


Bedrock 


Matrix 


Overburden 


Bedrock 


30 








Matrix 


13 


16 


1 


Overburden 





1 


29 



CHAPTER 4 
RAPID FIELD IDENTIFICATION OF PHOSPHATE MINING SAMPLES 

The objective of this research is to develop field instruments that will help to 
minimize contamination of matrix material (phosphate ore) by overburden or bedrock 
material through rapid field identification. This project has demonstrated the feasibility 
of accurately identifying overburden, matrix, and bedrock material in their untreated, 
natural state with no sample preparation. The application of laser-induced plasma 
spectroscopy as described in the previous chapter involves acquiring spectra of several 
selected samples, developing a library from these spectra, and using a correlation 
technique to match unknown spectra with well-characterized library spectra. Software 
was developed to rapidly carry out the correlation procedure and display material 
identification. 

In the development of field instruments for industrial applications, several 
experimental avenues have been explored. Identification of overburden, matrix, and 
bedrock samples was achieved by using four different configurations: a prototype 
benchtop instrument, a hand-held fiber-optic probe, the telescopic probe and finally a 
field LIPS probe. A review of these results is presented. Finally, a field portable 
instrument was designed, constructed, optimized and delivered to EVIC-Agrico. 

Phosphate Mining 

Mining in Florida, after tourism and agriculture, is the third largest industry in the 
state. The mining industry and its associated industries such as the processing and 
shipping of minerals provide numerous employment opportunities and a wide variety of 

46 



47 

products. About 90 percent of the rock mined is used in the production of agricultural 
fertilizers which are sold internationally. About 5 percent is used for livestock feed 
supplements, and the remainder is used for common items such as soft drinks, toothpaste, 
light bulbs, vitamins, and shaving cream. 

The role of Florida in the phosphate industry is paramount, as it is the supplier of 
75 percent of the United States' fertilizer and other phosphate needs and 25 percent of the 
world demand for phosphate products. Phosphate production has been an important 
component of the Florida economy for the past 100 years. 

The production of phosphate is a two step process: (1) the mining of the raw 
phosphate, including cleaning and separating out impurities and (2) the processing of the 
phosphate to make it suitable for commercial use. The phosphate matrix, a mixture of 
pebbles, sand, and clay, is typically found an average of 25 feet below the surface. 
Overburden, the material above the matrix, is removed by large draglines. The draglines 
then deposit the matrix in containment wells, where high-pressure water guns liquefy 
material into a slurry for pipeline transport to the processing plant. 

Phosphorus exists in nature with calcium, magnesium, and other elements in 
phosphate rock. After phosphate rock is ground to a fine and uniform size, it is reacted 
with sulfuric acid, releasing phosphorus as phosphoric acid, a soluble, readily available 
form that can be utilized by growing plants. 

Ca 5 (P0 4 ) 3 OH(j) + 5H 2 S0 4 (aq) -* 3H 3 P0 4 (atf) +5CaS0 4 (5) + H 2 O(0 

The phosphoric acid is concentrated, then reacted with ammonia, a source of 
nitrogen. The two most common products produced are diammonium phophate (DAP) 
and monoammonium phosphate (MAP). Other common products include fertilizers 



48 

materials such as granular triple superphosphate and urea and animal feed ingredients 
including calcium phosphate products. 

The phosphate industry is unique in that Florida law requires that all land that is 
mined be reclaimed; every acre mined must be reshaped. The industry is very sensitive 
to environmental needs. Nearly every phosphate plant relies on the self-generation of its 
electricity needs, driven by the waste heat from its facilities. In addition, factories re-use 
over 98 percent of the water used in mining and processing. These issues address cost 
efficiency as well as environmental concerns. 

Background 

In the mechanical removal of apatite ore from the earth, it is necessary to 
distinguish the transitional interface between the undesirable material that lies directly 
above (overburden) and beneath (bedrock) the matrix layer of ore. This is traditionally 
done by preliminary visual examination of core samples by trained geologists, limited 
chemical analysis of field samples, and visual observations of the exposed mine pit by the 
drag line operator. The locations of these interfaces are generally not very well known 
due to the limited number of core samples which can be economically obtained and the 
uncertainties in the identification of the core sample composition. Therefore, the dragline 
operator must rely on approximate or extrapolated data for the depth of the overburden 
and matrix. This rough estimation often leads to reduced efficiency in the recovery of 
raw material or contamination of valuable ore with undesirable compounds such as 
excessive MgO from the bedrock. It would be extremely useful to have a rapid, reliable 
field measurement technique to accurately identify overburden, matrix, and bedrock 
material in core samples. This would provide far more accurate preliminary 
identification of the topography in the mining operation. Improvements would come 



49 

from using a larger number of sampling locations with improved depth resolution. In 
addition, it would be very useful for the drag line operator to have available real time 
measurements of the content of each load of material or of the spatially-resolved 
composition of the exposed surface of the mine pit. The goal of this research is to 
develop and evaluate an approach to solve these measurement problems using laser- 
induced plasma spectroscopy and thereby enhance the efficiency with which apatite ore is 
removed from the earth. 

Preliminary studies 
Figure 4-1 shows typical LIP spectra of overburden, matrix and bedrock samples 
in a spectral window from 240 - 340 nm. In this range, spectral lines for Si, Fe, Al and 
Mg can be easily observed. Several obvious compositional differences are clear. 
Overburden tends to have higher Si concentration, matrix tends to be relatively low in Fe 
and bedrock is particularly high in Mg levels. In a preliminary study, measurement 
indices were devised which related the ratio of these spectral lines to the individual 
materials. Using a measurement approach first suggested by Regis Stana, the feasibility 
of identifying natural soils as overburden, matrix or bedrock was tested. Unknown 
(blind) samples of these materials were provided by EVIC-Agrico. Six samples labeled 
Al, A2, A3, Bl, B2 and B3 were received, sealed in plastic bags in a five-gallon shipping 
bucket. From each sample, about 5 g of material was removed with a small scoop and 
pressed loosely into a small sample dish, 2.5 cm in diameter and 4 mm deep. Scraping 
with a microscope slide roughly leveled the surface. This single dish of soil constituted 
the analytical sample for each bag of material. The samples were measured wet, as taken 
from the original bags. 



50 

Laser-induced plasma spectra were acquired for each of the six analytical samples 
using the compact LIPS instrument developed previously [20]. A laser pulse energy of 
50 mJ was focused with a 10 cm focal length lens resulting in a spot size at the sample 
surface of about 0.5 mm. For each sample, 11 or 12 runs were made, each consisting of 
10 laser samplings at random points on the sample surface. Therefore, a total of 1 10 or 
120 laser shots were averaged for each sample. The laser was operated at a repetition 
rate of 1 Hz. Spectra were captured through a fiber optic link to a compact Ocean Optics 
spectrometer. Customized software identified the spectral lines, located the background, 
and calculated the net line intensities for each laser shot. 

Table 4-1 shows the normalized, averaged results for the six samples. Four 
different indices, Si/P, Si/Mg, Si/Al and Si/Fe were evaluated. The average relative 
uncertainties were 50%, 33%, 33% and 30%, respectively. The Si/P ratio did not vary 
consistently and, having the poorest precision was rejected as an indicator. The Si/Al 
ratio did not vary significantly among the 6 samples tested. The Si/Mg and Si/Fe ratios 
both showed statistically significant differences between the 6 samples with the Si/Mg 
ratio proving to be the most reliable indicator. 

Based upon these preliminary results, a full evaluation of the method proceeded 
and the correlation data analysis approach was developed to use the entire content of the 
measured spectra rather than any particular pair of spectral lines. 

Correlation Studies with the Benchtop Instrument 
Experimental Setup and Methodology 

The preliminary work was repeated to confirm the ability to reliably identify soils 
as overburden, matrix, or bedrock. Approximately 8 g of unknown samples provided by 



51 

IMC-Agrico were loosely pressed into a sample dish, 3.0 cm in diameter and 8 mm deep. 
A spatula was used to roughly level the surface. 

Laser-induced plasma spectra were obtained from each of six samples using the 
configuration depicted in Figure 4-2. A laser pulse (Big Sky Laser Technologies, Inc., 
1064 nm) at a repetition rate of 1 Hz and pulse energy of 50 mJ was aligned through a 
pierced mirror. The laser was then focused with a 15 cm focal length lens resulting in a 
spot size at the sample surface of about 0.5 mm. Light emitted at the sample surface was 
collected by the pierced mirror and focused by a lens (12 cm focal length) through a 
neutral density filter and onto a fiber optic cable linked to an Ocean Optics spectrometer. 

For each sample, 10 runs were made, each consisting of 10 laser samplings at 
random points on the sample surface, resulting in 100 laser shots for each sample. The 
resulting spectra were averaged to form a library for each sample. Single shot spectra 
were obtained for random samples and compared against the libraries for identification. 
Spectra were obtained in the 250-330 nm spectral window. Figure 4-3 shows a typical 
spectrum resulting from the average of 10 laser shots. 

Several samples, obtained from IMC-Agrico, were analyzed using the customized 
software. These same samples were also sent back to IMC-Agrico for identification by 
chemical analysis. The phosphorus (P2O5) and magnesium (MgO) content of each of the 
samples was obtained by wet digestion. The identification by LIPS was correlated 
against these analytical results obtained by IMC-Agrico. 

Because it is expected to have variations in surface topography with core 
sampling, the effect of sample height (position relative to the laser focus) on the 
identification of the soils was studied. To examine this effect, spectra were acquired for 



52 

each sample at various positions above and below the focal plane of the focusing lens. 
The laser pulse energy was 50 mJ, the repetition rate of the laser was 2 Hz, and 10 laser 
shots were averaged for each sample. A library was made for each specific position from 
the focal plane (e.g., +/-1.0 cm from the optimum focus). Each library contained the data 
for each of the three layers. 

Core sampling movement was simulated by the mechanical translation of a 
sample tray packed with overburden, matrix, and bedrock. The motorized translation 
stage moves at approximately 0.25 cm/s over a distance of about 18 cm. Overburden, 
matrix, and bedrock material were tightly packed with distinct transitions into a 5 cm x 
30 cm sample tray placed on top of the translation stage. First, a correlation library was 
made for overburden by collecting and averaging about 30 spectra of overburden sample. 
Then, the translation stage and the correlation algorithm in the software were 
simultaneously initiated. Finally, a graph of the correlation coefficient vs. laser shot 
number (distance) was plotted as each spectrum was collected. 
Results 

The averaged correlation coefficients from the ten libraries are displayed in Table 
4-2. As shown previously, the software easily distinguishes among overburden, matrix, 
and bedrock. As an example, the correlation coefficients for a matrix sample are 
graphically depicted in Figure 4-4. Correlation coefficients for both matrix samples are 
high, while those of other samples are significantly lower. In addition, excellent 
precision (standard deviation <0.06) was obtained. From the single shot spectra of 
random samples, we observed a high degree of confidence in classifying the soils as 
overburden, matrix, or bedrock. This data analysis approach could even distinguish 
between different samples of material within the three sample categories. 



53 

The results from chemical analyses are shown in Table 4-3. The criteria used for 
classifying the samples as overburden, matrix, or bedrock are shown in Table 4-4. 
Samples with values in between those given values in Table 4-4 would likely be obtained 
from the interface between two layers. The LIPS determination of the samples correlates 
quite well with the results obtained from chemical analysis. There are occasions when 
there may be a discrepancy between the two methods of identification (e.g. Sample 4). 
Particles of the matrix may have been embedded within the bedrock sample, giving a 
false identification. This error, however, would be minimized or perhaps eliminated by 
averaging several laser probings of each sample. 
Effect of Sample Position 

As expected, the manipulation of the sample height below the focusing lens 
showed that the highest spectral quality is observed at the focal length of the lens. Figure 
4-5 shows the effect of sample distance on the correlation coefficients using the 
maximum laser pulse energy (50 mJ). In this figure, a spectrum taken at the lens focal 
length (distance = cm) was used for the correlation library and the matrix spectrum is 
the source spectrum in each library. It is important to note that, even though the 
correlation coefficient was poorer out of the plane of laser focus, the sample was always 
identified as matrix, regardless of the sample position. Another significant observation is 
the difference in correlation coefficients at each specific position. Within each layer, the 
correlation coefficients follow the same trend: matrix, bedrock, overburden. A similar 
study at lower laser energy showed that the reliability of identification degrades for 
material closer than the lens focal length, as can be seen in Figure 4-6. 



54 

Continuous Correlation on a Moving Sample 

Results from motorized sample translation are shown in Figure 4-7. Three 
distinct regions of correlation coefficients are apparent on the graph. The large variations 
in the correlation coefficients for the middle region (matrix) are likely due to 
inhomogeneity in the sample. Nevertheless, the resulting plot still indicates when the 
transitions between layers occurred and correctly identifies the 3 materials in all cases. 
Another alternative is to perform the correlation against the last spectrum observed, thus 
detecting changes in the sample composition. Another method is to correlate against 
each of the three libraries, which would improve the precision within any one region. 
These alternative algorithms may be developed and evaluated for their effectiveness and 
ease of identification of layers. 

Fiber-Optic Probe 
Experimental Setup and Methodology 

The portability of this technology in the field is an obvious advantage. This 
avenue was explored with the design of a miniaturized LIPS probe, shown in Figure 4-8. 
The miniature probe measured 2 inches in diameter and was easily held in one hand 
(Figure 4-9). A small trigger button on the probe is connected to a customized laser 
trigger circuit, shown in Figure 4-10, to initiate the laser pulse. As the probe is pressed 
against the sample, the operator depresses the trigger button and a spectrum is 
immediately obtained. The laser (1064 nm) was coupled through an optical fiber to the 
probe head. The resulting power at the output of the optical fiber was -40 mJ. The fiber 
output was focused on the sample with a W diameter lens, precisely located at the 
optimal focal distance. Emitted light from the plasma was collected by another optical 
fiber positioned at an angle with respect to the laser beam. The optical fiber (400 urn) 



55 

guided the light into the spectrometer (Ocean Optics, Inc.), which was interfaced to a 

laptop computer. 

Results 

The results from the fiber optic LIPS probe are shown in Table 4-5. Two hundred 
laser shots were used on each of the samples of known origin. Every laser shot was used 
in the classification of each sample. In a field setting, the poor spectra would not be 
used; instead, those spectra would be discarded and another spectrum taken. A threshold 
value could easily be set in the software, prompting the operator to repeat a measurement 
if the overall spectrum intensity was too low. At maximum power, an identification 
accuracy of 87% was achieved when all spectra were evaluated. This improves to 95% 
when the poor spectra are discarded. 

Although this system performed adequately in the laboratory, the coupling 
efficiency of the laser through the optical fiber was difficult to maintain and the amount 
of energy delivered to the sample was about 10X lower than the system without the fiber 
optic link. This resulted in a much weaker plasma, requiring better sample presentation 
(surface uniformity, moisture content). It was therefore concluded that the field 
instrument design would not use a fiber optic laser link. 

Remote LIPS 
Experimental Setup and Methodology 

The possibility of remote analysis was investigated by use of a telescopic focusing 
system and standard hardware. The experimental arrangement, shown in Figure 4-11, 
included the same 50 mJ Nd:YAG laser (Big Sky Laser Technologies, Inc.) which was 
used in the other experiments, a telescopic focusing system, a large diameter collection 
lens, and the fiber optic mini spectrometer (Ocean Optics, Inc.). The laser spark was 



56 

induced on a solid target placed at a distance of 6 m from the laser operating at its 
fundamental wavelength of 1064 nm at a maximum repetition rate of 20 Hz. The laser 
light was focused on the target by a telescopic beam compressor with an adjustable focal 
length. Emission from the spark was collected by a large quartz lens (10 cm diameter, 20 
cm focal length) at a small angle with respect to the laser beam. The lens focused the 
plasma emission light on the face of an optical fiber (600 urn) and the fiber guided the 
light into the spectrometer. The two channel spectrometer alternatively covered the 
spectral ranges of 230-310 nm and 200-850 nm with resolutions of 0.5 nm and 1 nm, 
respectively. 
Results 

Typical spectra obtained with the remote LIPS system are shown in Figure 4-12. 
The spectra from the low-resolution channel (200-850 nm spectral range), Figure 4- 12(b), 
are twice as intense as the spectrum from the high-resolution channel (230-310 nm 
spectral range), Figure 4-12 (a), due to the higher spectrometer sensitivity in the wide 
spectral range. Nevertheless, strong lines of elements could clearly be seen and resolved 
in both channels. Figure 4-13 shows a dramatic decrease in the line intensity ratio as the 
target distance was increased. 

Overall, it has been demonstrated that a compact and reliable moderate power 
laser (the 50 mJ Big Sky) together with a simple and inexpensive detection system (the 
Ocean Optics spectrometer) can efficiently be used for remote detection of elements in 
solid samples within an operating distance of -10 ft. The small size of the setup 
components is an additional advantage that allows the engineering of a compact setup for 
field applications. 



57 

The Portable LIPS Probe 
Experimental Setup and Methodology 

A schematic and picture of the portable LIPS probe are shown in Figures 4-14 and 
4-15. Including the handle, the unit is 3 feet tall, allowing for convenient measurements 
at ground level by a standing operator. The operator grasps the handle of the unit at waist 
level and initiates the laser pulse by the trigger button on the handle. The trigger button, 
on/off switch, and safety interlock switch, are connected in series to initiate the laser 
trigger. The safety interlock switch is a precautionary device, allowing the laser to 
trigger only if the unit is pressed against a firm surface. This trigger circuitry is then 
connected by a 20 ft coaxial cable to the laser trigger box (Figure 4-10) on the back of the 
laser ICE (integrated cooler and electronics). The trigger circuit serves as the external 
trigger for the initiation of each laser pulse. 

The customized unit is constructed within an aluminum frame. Laser light 
emitted from the laser head is focused by a lens at a fixed distance at the bottom of the 
probe. Light emitted from the plasma is collected by a pierced mirror and focused into an 
optical fiber. The 15 inch optical fiber is attached to a USB2000 Ocean Optics 
spectrometer. The spectrometer and notebook computer are linked by a short USB cable. 

The software was customized for this application in Visual Basic 6.0 to allow for 
rapid identification of materials as overburden, matrix, or bedrock. "Training" of the 
software is done by constructing a spectral library of known materials. The spectra of 
unidentified samples are then correlated against the reference library. With each laser 
pulse, the software identifies the material as overburden, matrix, or bedrock, based on the 
magnitude of the correlation coefficient. A lightweight, Sony VAIO notebook computer 
was attached to the portable instrument for data acquisition and display. 



58 

Wet vs. Dry sampling 

The effect of moisture on the spectroscopic behavior of real samples was 
investigated. The moisture content of the samples was determined by weight loss after 
drying. Spectroscopic data were taken on both wet and dry samples. The spectra of wet 
vs. dry samples were compared and the accuracy of identification was evaluated. 
Real Sample Analysis 

Several spectra were taken of overburden, matrix, and bedrock samples provided 
by IMC-Agrico. Samples were studied as received and had varying, unknown moisture 
content. Three samples of each material were used. 

Ten spectral libraries were made from these nine samples, each library consisting 
of the data points from each spectrum of overburden, matrix, or bedrock. Correlation 
coefficients were obtained for each sample with respect to other samples within the same 
library. These correlation coefficients from the 10 libraries were averaged and the 
standard deviation was calculated. 
IMC-Agrico Site Visit 

At the New Wales facility, a number of samples were taken directly from the 
mining facility and analyzed by LIPS. The portable LIPS probe was first used on a 
variety of samples on hand. Samples were placed in a 13 x 9 inch pan for the analysis by 
the probe. Here, an identification of each of the 1 1 samples was determined after 20 
single shots of the probe. The following day, the probe was taken outdoors for analysis 
directly on mounds of material that a nearby dragline had recently unearthed. A 
generator placed in the bed of the truck supplied power. Finally, a number of samples 
were obtained from various mine pits and were sent to the University of Florida for LIPS 
analysis. 



59 

Results 

Samples obtained directly from IMC-Agrico are typically moist. Moisture 
content as received is at most 20%. Average moisture content for matrix and bedrock 
samples is 16% and for overburden less than 2%. Correspondingly, samples of 
overburden are less affected by moisture since the typical real sample does not contain as 
much moisture as that of matrix and bedrock samples. The intensity of the signal from 
overburden is reduced by approximately 15-20%, while the reduction in signal intensity 
of the more moist samples of matrix and bedrock is typically 30-60% (Figures 4-16 to 4- 
18). 

The reduction in signal intensity when dealing with real samples, however, does 
not influence the software's ability to accurately identify the samples. Each spectrum is 
rich with spectral information, provided a good laser shot is taken. The training set in the 
library accounts for the moisture in the samples, resulting in accurate identification with 
each laser trial. Thus, a comprehensive library will yield more accurate identification. 
Because of the distinct differences among the three types of samples, the effect of 
moisture has little effect on the accuracy of identification. 

The averaged correlation coefficients from real sample analysis are displayed in 
Table 4-6. Again, as has been shown previously in preliminary results and in the 
benchtop design, the software easily distinguishes among overburden (01, 02, 03), 
matrix (Ml, M2, M3), and bedrock (Bl, B2, B3) with this LIPS probe design. The 
correlation coefficients for the samples are graphically displayed in Figures 4-19 to 4-21. 
Correlation coefficients of each layer are high within each group, indicating a high 
accuracy in classifying the soils. The average standard deviation is 0.108, indicative of a 
high precision from these unprepared samples. There are some occasions in which the 



60 

correlation coefficients of a sample do not correlate well with others of the same 
classification (Ml and B2). Practically, this inaccuracy would easily be resolved as 
successive determinations are expected to be made on samples of the same classification. 
It is observed, however, that this instrument and software provide a very reliable 
identification of material as overburden, matrix, or bedrock. 

At the New Wales facility, the results from the first day of analysis were positive. 
Of the 11 samples, 8 were identified correctly according to the chemical analysis. 
Positive identification was established as the material identified in majority relative to the 
other two possibilities. Typically, this was achieved by about 16 of the 20 laser shots 
identifying a particular material. 

Results from the outdoor analysis near the mine pit were similar. Issues regarding 
the application of the instrument to the field became apparent. The most obvious was the 
visibility of the computer monitor in bright sunlight. Reflection of the sun on the 
computer monitor made the operation of the instrument cumbersome and difficult. In 
addition, the sampling of materials was also laborious. The coolant and power cables for 
the unit allowed only a 20 ft measuring range, which was found to be inconveniently 
short. Thus, visible mounds of material just several yards away needed to be brought to 
the unit for analysis. Despite these rather simple problems, the field performance of the 
probe was encouraging overall, noting the ruggedness of the instrument and the accuracy 
of the identification. 

The results from the analysis of the samples that were sent to the University of 
Florida for LIPS analysis are included in Table 4-7. The LIPS method was accurate with 
9 of the 15 samples, assuming the chemical analysis is accurate. However, an 



61 

interpretation of the chemical analyses provides insight into the errant samples. For 
example, lab analysis of sample 1 indicates a high content of bone phosphate lime (BPL) 
and a marginal (but high) MgO. Officially, the geology department cuts off the ore body 
at MgO of 0.8%. In the chemical classification algorithm used, 2% MgO was used as a 
cutoff because the tendency is to blend away the higher MgO product. At the particular 
location where this sample was mined (Hopewell), the MgO increases more gradually so 
that the ore contact is not sharply defined. Thus, the sample is marginally bedrock, but is 
likely in the lower contact transition region. Sample 1 could be classified as either matrix 
or bedrock. It was classified as matrix by the chemical analysis and bedrock by the LIPS 
analysis. 

Sample 5 was classified as overburden by chemical analysis and as matrix by 
LIPS analysis. A visual analysis of the material indicates that the sample is clearly 
overburden. This was likely a case in which the spectral library used is not as 
comprehensive as possible. If more representative spectra were included in the library, 
perhaps this sample would have been identified correctly. 

Sample 10 was taken from the center of the ore body in what was believed to be a 
barren strip of clay. Chemical analysis reveals it to be overburden, while LIPS analysis 
identifies it as matrix. The sample does have lower BPL levels than desirable (9.78%), 
but the marginal MgO content indicates that the sample is probably adequate for mining. 
Sample 13 behaves similarly. The MgO level is low, while the BPL content (9.06%) is 
slightly lower than the cutoff value desired (10%). These are good examples of samples 
in which the analyses using LIPS might be more reliable than those obtained from 
conventional chemical analyses. 



62 

Finally, with these interpretations in mind, the LIPS algorithm appears to give an 
accurate, or at least reasonable, identification with 13 of the 15 samples that were sent to 
the University for analysis. The identification of the 2 incorrect samples may be due, in 
large part, to the exclusiveness of the present spectral library. 

Conclusions 

One immediate application of the remote probe could be the analysis of raw ores 
at industrial beneficiary sites. No sample preparation is required and a quick 
identification of the nature of a material can be obtained. 

Although the fiber optic probe was functional, it was difficult to obtain sufficient 
laser irradiance at the sample, after re-collimating and re-focusing the output of the fiber 
optic link. With the irradiance obtained, it was difficult to obtain a reproducible, 
energetic plasma on some soils, especially if the sample was wet. The study was 
therefore completed using a system that incorporated the laser head within the 
measurement probe. 

The benchtop laboratory instrument was used as a prototype for preliminary study 
before the development of the field LIPS probe. Similar studies were performed on both 
instruments to assess the accuracy of sample identification with real samples. The field 
LIPS probe is engineered for robust use on site and provides a rapid identification of 
material using the software developed. 

The studies presented here have given encouraging results. Specifically, the soils 
were easily identified as overburden, matrix, or bedrock using a single shot spectrum. In 
summary, we have 

• shown that overburden, matrix, and bedrock can be easily distinguished; 



63 

demonstrated remote identification at a range of 10 m; 

developed a field LIPS probe for single shot material identification; and 

developed software that displays correlation results with each laser pulse. 









64 



800 
700 
<00 

.{? mo 
S too 

300 
200 
100 



- Overburden 



« I* U, 



_Ju^ 




240 



2300 -, 



2000 



1300 



* 



2(0 2S0 300 

Ware length fttm) 



320 



Matrix 



1000 



300 



- 



240 



2<0 290 300 

Wwelength fcan) 



320 



3300 -. 
3000 
2300 
2000 
1300 
1000 
300 - 
- 



Bedrock 



240 



2<0 280 300 

Wavelength frm) 



320 



340 




340 




340 



Figure 4-1 . LIPS spectra of overburden, matrix, and bedrock. 



65 



Table 4- 1 . Average spectral 1 


me intensity ratios. 


Sample 


Si/P 


Si/Mg 


Si/Al 


Si/Fe 


Al 


169 ±91 


11 ±6 


16 ±5.5 


10.4 ±6.4 


A2 


48 ±30 


2.4 1 0.4 


15 ±4.3 


3.3 ±0.9 


A3 


39 ±22 


0.33 ± 0.07 


17 ±7.7 


2.2 ±0.7 


Bl 


89 ±42 


17.3 ±8 


10.4 ±3.3 


12.8 ±3.7 


B2 


1 18 ±55 


2.0 ±0.4 


16.7 ±4 


2.2 ± 0.4 


B3 


118 ±40 


0.53 ±0.2 


27.3 ±9 


1.86 ±0.3 



66 



Laptop 
Computer 




' ' ' — r- 



Spectrometer 



Opttcal 
Fiber 



n 



V 



Las*r 



Lent 



\ 



Pierced 
Mirror 



Lens 



o 



Laser 
Power 

Snpph 



Sample 




Adjustable 
Stage 



Figure 4-2. Schematic of the LIPS benchtop experimental system. 






67 



1600- 
1400- 
1200- 
1000- 
S 800-1 

e 

1 600 

c 

400H 
200 

OH 

-200 



240 



- 1 — 
260 



Matrix spectrum 

Mg 




280 300 320 

Wavelength (nm) 



- 1 

340 



Figure 4-3. LIPS emission spectrum of matrix sample. 



68 



Table 4-2. Correlation coefficients for overburden, matrix, 


and bedrock 


samples. 


Sample 


0' burden 1 


O' burden 2 


Matrix 1 


Matrix 2 


Bedrock 1 


Bedrock 2 


O' burden 1 


1 


0.9568 


0.5756 


0.7026 


0.5056 


0.4424 


0' burden 2 


0.9568 


1 


0.5617 


0.6599 


0.4503 


0.3753 


Matrix 1 


0.5756 


0.5617 


1 


0.9289 


0.7639 


0.7106 


Matrix 2 


0.7026 


0.6599 


0.9289 


1 


0.8184 


0.7674 


Bedrock 1 


0.5056 


0.4503 


0.7639 


0.8184 


1 


0.9785 


Bedrock 2 


0.4424 


0.3753 


0.7106 


0.7674 


0.9785 


1 



69 







Matrix correlation coefficients 




m 1.1 - 














c 








.2 , 








o 1 - 








£ „„ 








« 0.9 - 








o 








o 








c 0.8 - 










o 






^r 




S 0.7 ■ 












a> 












£ 0.6 - 


* 








o 








O 0.5, 










i 1 1 1 


1 




01 


02 Ml M2 Bl 
Sample category 


B2 



Figure 4-4. Correlation coefficients of matrix sample. 



70 



Table 4-3. 


Identification of materials by chemical and LIPS analysis. 


Sample 


%P 2 5 


%MgO 


Chemical ID 


LIPS ID 


1 


7.31 


6.02 


Bedrock 


Bedrock 


2 


3.03 


6.72 


Bedrock 


Bedrock 


3 


3.62 


1.67 


Matrix/Bedrock 


Bedrock 


4 


1.12 


15.03 


Bedrock 


Matrix 


5 


4.80 


3.81 


Bedrock/Matrix 


Bedrock 


6 


ND 


0.22 


Overburden 


Overburden 


7 


16.19 


0.21 


Matrix 


Matrix 


8 


2.95 


6.14 


Bedrock 


Bedrock 






71 



Table 4-4. Criteria for classification by chemical analysis 



Overburden 



Matrix 



Bedrock 



P 2 5 



Negligible 



>3% 



Variable 



MgO 



Negligible 



<1.5% 



>5% 



Anything in between is a mixture. 



1.2 



72 




■Matrix 
- Cburden 
Bedrock 



-1.5 -1 -0.5 0.5 1 1.5 
Distance from focal length (cm) 



2.5 



Figure 4-5. Correlation coefficients as a function of distance from focal length at 

maximum laser power. 









73 



1.2 



Si 

1 0.8 

I 
e 

B 0.6 



1 0.4 

E 

0.2 




*W 



■Matrix 

- Cburden 

Bedrock 



-1.5 -1 -0.5 0.5 1 1.5 2 
Distance from focal length (cm) 



2.5 



Figure 4-6. Correlation coefficients as a function of distance from focal length at lower 

laser power. 



74 





1.0- 










0.9- 


.«* * + ** 

• 


m 

• * 
• 




X 

1 


0.8- 
0.7- 


Overburden 


• 

• • 


-. 




0.6- 




• 
• • 
• 


• \% 


00 


0.5- 




* Matrix 


• 


CO 








• 




0.4- 

A *% 






Bedrock 




0.3 - 
0.2- 
0.1 - 














0.0- 










i ■ i ' i ' i > i > i 



50 100 150 

Distance (mm) 



200 



250 



Figure 4-7. Plot of correlation coefficients vs. distance using motorized sample 

translation. 



75 



Trigger wire 




Laser 
power 
supply 



Coaxial 



Laser 
i umbilicals 



Laser head 




Coupler 



lJLl 



Probe 



Spectrometer 



Laptop computer 



Figure 4-8. Fiber-optic probe system. 



76 



To spectrometer 



Trigger button 




Handle 



From laser head 



To trigger circuit 



Figure 4-9. Fiber optic LIPS probe. 



77 



+9V 



+W 



52.3 kD 



.iokn 



Con. 5 



Con. 3 



Con. 1 
+9V< (§) ♦ 

-9V^-®— 



Con. 3 



lkQ 

WW 




lkQ 

vVW 



1,4,8,15 



No connection 



7, 9, 10, 14 



3,11,12,13,16 



Con. 3 



+9V 



For external lamp trigger input into Big Sky Laser 
+5VDC, 100ms, into 50 Q 



+9V 



.49.9 kQ 



2N3904 



tv 



-® 

Con. 4 




Figure 4-10. Trigger circuit for Big Sky laser. 



78 



Table 4-5. Identification using a fiber optic LIPS probe 





Overburden 


Matrix 


Bedrock 


Correct 


174 


159 


188 


Incorrect 


4 


10 


11 


Poor spectra 

(Corr. Coeff. < 0.5) 


22 


31 


1 






79 



Target 



Laptop 
Computer 






Ocean Optics 
Spectrometer 




Figure 4-11. Experimental apparatus for remote LIPS. 



80 



(a) Spectra obtained with LIP-spectrometer 

at the distance of 6 m using low- and 
high resolution spectrometer channels 



J 


I 

T™ — — 


Mg 

L 

-| ' 1 ■ 1 ■ 1 ■ r ■ 



200 300 



soo eoo too 



Wavaiangln. nm 

a) Low resolution channel {200-850 nm. 
600 mm ' graling. 25 nm slit) 




Wavalangth. nm 

b) High resolution channel {230-310 nm. 
3600 mm ' grating. 25 »m slit) 



(b) Spectra obtained with LIP-spectrometer at 

the distance of 6 m 



Phosphate Rock #17 
(from IMC Agrico) 



'jJiA^ 



30O 400 500 600 TOO SOO 

Wavalengm. DID 




Figure 4-12. Remote LIPS spectra. 



81 



3.0- 



2.5- 

E 
c 

i 

E 
c 

co 

to 

CM 

2 1.5 



2.0 



| 1.0 



0.5 



0.0- 




1 ■ I ■ I ' I ' I ■ I ■ I ■ I ■ I ■ I ■ I 

2 4 6 8 10 12 14 16 18 20 22 



Distance, ft 

Figure 4-13. Line intensity ratio as a function of distance in remote LIPS analysis. 



82 










p 

&0 

•c 

H 



o 



Oh« 



<d .« 



o o 

£ .9 w 

PQ O £ iS 



T3 
cd 





d 

■1—1 



u 

JO 

2 

& 
•4-» 

o 
a. 

c 

s 

•c 

0) 
D. 

w 



=3 



c 





o 


CI 


O 


J3 


V 

B 
o 


a. 
o 
u 

x> 


c 
o 


| 

S 
> 


o 


0) 


tii 


■»-> 


o 






O 


o 

q=l 


B 
I 








<* 




r 


l-H 

3\ 







c 



-►*- 



en 



83 




Figure 4-15. The field LIPS probe. 



84 



4500 
4000 
3500 
3000 

2500 

& 

•| 2000 

a 

S 1500- 

1000- 
500- 


-500 




— 1 i | i 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1— 

200 220 240 260 280 300 320 340 360 380 



Wavelength (nm) 



Figure 4-16. Spectra of wet and dry bedrock samples. 






85 



3400- 

3200- 

3000- 

2800- 

2600- 

2400- 

2200- 

2000- 

- 1800- 

■| 1600- 

| 1400- 

>B 1200- 

1000- 

800- 

600- 

400- 

200- 

0- 

-200- 




— i — i — | — i — r- 

200 220 240 



- 1 — ■ — r~ 

260 280 



—J— 
300 



320 



- 1 ' 1 ' 1 - 

340 360 380 



Wavelength (nm) 



Figure 4-17. Spectra of wet and dry matrix samples. 



86 



■ 
3 



1800 -, 
1700- 
1600- 
1500- 
1400- 
1300- 
1200- 
1100- 
1000- 

900- 

800- 

700- 

600 

500 

400 

300 

200 

100 



-100 




& Dry 



Wet 



— | — i — | — i — | — i — | — i — | — i — | — i — | — i — | — i — | — i — |— 

200 220 240 260 280 300 320 340 360 380 



Wavelength (nm) 



Figure 4-18. Spectra of wet and dry overburden samples. 



87 



Table 4-6. 


Correlation coefficients for overburden, 

samples. 


matrix, and bedrock untreated 




01 


02 


03 


M1 


M2 


M3 


B1 


B2 


B3 


01 


1 


0.9009 


0.9010 


0.6916 


0.5490 


0.6382 


0.6624 


0.8346 


0.7453 


02 


0.9009 


1 


0.9167 


0.7237 


0.6580 


0.6996 


0.6189 


0.8216 


0.7366 


03 


0.9010 


0.9167 


1 


0.7651 


0.6311 


0.6348 


0.6644 


0.8391 


0.7534 


M1 


0.6916 


0.7237 


0.7651 


1 


0.7548 


0.7284 


0.7360 


0.7603 


0.7543 


M2 


0.5490 


0.6580 


0.6311 


0.7548 


1 


0.7846 


0.6057 


0.5563 


0.5994 


M3 


0.6382 


0.6996 


0.6348 


0.7284 


0.7846 


1 


0.6506 


0.6539 


0.6874 


B1 


0.6624 


0.6189 


0.6644 


0.7360 


0.6057 


0.6506 


1 


0.7718 


0.8524 


B2 


0.8346 


0.8216 


0.8391 


0.7603 


0.5563 


0.6539 


0.7718 


1 


0.8387 


B3 


0.7453 


0.7366 


0.7534 


0.7543 


0.5994 


0.6874 


0.8524 


0.8387 


1 



88 




1 


1.1 


I 




<J 


1 


s 




1 


0.9 


• 1 




a 


0.8 


I 




C5 


0.7 



M1 M2 M3 
Sample 



B1 B2 B3 







Bedrock 2 Correlation Coefficients 




1 1.1 - 

1 

1 0.9 - 
| 0.8 - 
1 0.7 - 
! 0.6- 


1- 


-*-k. 


■- 












U 5 - 














01 


02 03 M1 M2 M3 B1 B2 B3 






Sample 






Figure 4-19. Bedrock correlation coefficients. 






89 








Matrix 2 Correlation Coefficients 






B 1.1 - 

1 1 " 
I 0.9 - 

c 0.8 - 

o 

1 0.7 




























t 0.6- 

e 

U OK - 




























01 


02 03 M1 M2 M3 B1 
Sample 


B2 


B3 




Figure 4-20. Matrix correlation coefficients. 



90 



I 1 

g 0.9 

a 0.8 

o 



E °- 6 H 

I 



0.5 



Overburden 1 Correlation Coefficients 



s 




01 02 03 M1 M2 M3 B1 B2 B3 
Sample 








Overburden 3 Correlation Coefficients 






1 1.1 - 

1 1 - 

g 0.9 - 

■ 0.8 - 

o 

1 0.7 

E 0.6- 

e 

y 05- 


< 


f-*-A 


tsj 




> — 


-Y 


A> 


-i 











1 02 03 M1 M2 M3 B1 
Sample 


B2 


B3 



Figure 4-21. Overburden correlation coefficients. 



91 








Table 4-7. 1 


Chemical and LIPS 


analysis 


results from MC-Agrico samples. 




1 


BPL 


Fe 2 3 


A1 2 3 


Insol. 


CaO 


MgO 


Chemical 
Analysis ID 


LIPS ID 




1 


21.25 


2.47 


1.93 


61.9 


13.95 


1.6 


Matrix 


Bedrock 


X 


2 


4.31 


0.22 


2.46 


89.45 


0.47 


0.06 


Overburden 


Overburden 


ok 


3 


29.29 


0.74 


2.34 


56.49 


18.3 


0.39 


Matrix 


Matrix 


ok 


4 


14.38 


1.8 


0.68 


8.18 


30.44 


14.59 


Bedrock 


Bedrock 


ok 


5 


0.26 


0.2 


0.08 


95.05 


0.21 


0.03 


Overburden 


Matrix 


X 


6 


43.57 


0.41 


3.8 


41.49 


24.77 


0.08 


Matrix 


Overburden 


X 


7 


7.18 


0.68 


0.59 


26.16 


24.11 


12.38 


Bedrock 


Bedrock 


ok 


8 


21.65 


0.47 


4.74 


67.58 


5.11 


0.21 


Matrix 


Matrix 


ok 


9 


19.71 


0.77 


1.3 


69.27 


12.58 


0.38 


Matrix 


Matrix 


ok 


10 


9.78 


1.96 


3.28 


73.78 


6.49 


0.82 


Overburden 


Matrix 


X 


11 


4.07 


0.53 


0.66 


15.61 


26.02 


16.08 


Bedrock 


Bedrock 


ok 


12 


16.14 


0.59 


1.54 


74.84 


9.73 


0.22 


Matrix 


Matrix 


ok 


13 


9.06 


0.43 


1.72 


82.82 


3.55 


0.03 


Overburden 


Matrix 


X 


14 


0.69 


0.31 


1.37 


89.68 


0.19 


0.02 


Overburden 


Overburden 


ok 


15 


12.01 


0.95 


1.38 


75.56 


8.68 


0.69 


Matrix 


Bedrock 


X 



CHAPTER 5 
LIPS FOR CHARACTERIZATION OF ARCHAEOLOGICAL MATERIALS 

Introduction 

In recent years, there has been increasing use of modern laser technology in 

archaeology, and the results are particularly promising. One of the principal tasks of 

excavation teams is the establishment of a meaningful classification of the many 

recovered ceramic materials. A number of techniques, involving visual examination as 

well as mineralogical, chemical, structural, and elemental analysis, has been used in 

characterizing and subsequently classifying ceramic materials [24]. Micromorphological 

comparison of pottery groups and soil materials has indicated that two main factors 

influence the raw material procurement strategies of the potters of these settlements: the 

proximity of the soil material to the potter's settlement and the suitability of the soil 

material for pottery manufacture. Wieder and Adan-Bayewitz studied different pottery 

groups from the Roman period [25]. Several techniques have been proposed as tools for 

archaeological analysis. Clark and Gibbs examined the pigments of a number of pieces 

by Raman microscopy using an Olympus BH-2 microscope with a lOOx objective 

coupled to a DILOR X-Y triple grating spectrometer with a photodiode-array detector 

[26]. The light sources were either a Coherent 170 argon ion laser or a krypton ion laser. 

An extensive evaluation of common pottery from Roman Galilee and Golan was carried 

out, employing neutron activation analysis to determine the primary classification of 

archaeological materials [27]. Characterization of archaeological material was performed 

using Mossbauer spectroscopy [28], and non-destructive x-ray fluorescence (XRF) has 

92 



93 

been used by several workers for the analysis of archaeological materials [29, 30]. 
Thermogravimetry (TG) and differential thermal analysis (DTA) were used by Papargyris 
and Cooke for ceramic powders [31]. X-ray diffraction (XRD) and scanning electron 
microscopy (SEM) analyses have been used to obtain fractionated particle sizes of 
dolomite [32]. 

LIPS has been used to identify metallic elements in vitrified glass [33] and 
pigments used in painting [34]. Analysis of pigments in paintings is of major 
significance in art conservation, as it can lead to detailed characterization of materials as 
discussed in Chapter 2. Thus, pigment analysis is important for dating and authentication 
as well as for possible conservation of restoration of the artwork. Burgio et al. used two 
laser-based analytical techniques — LIPS and Raman microscopy [35]. 

The LIPS technique has been successfully applied as an on-line diagnostic 
technique in the laser cleaning process of polluted limestone from historic buildings of 
the Spanish Renaissance [36]. The classification of pottery by laboratory analysis 
depends upon matching the composition of the pottery. The criteria for selecting 
reference pottery depend upon the archaeological problems and will not be discussed 
here. The analyzed materials have homogeneous compositions and are not grossly 
different in composition. The geographical domain for the present study was confined to 
Spain. The key Spanish city of Zaragoza was an important center for pottery 
manufacture during the Roman period. Over 2,000 years of history have left Zaragoza 
with a series of cultural strata that have gradually formed the personality and 
physiognomy of a large, modern city which, at the same time, has its roots firmly planted 
in its past. 






94 

In previous papers [20, 37], it has been shown that simple statistical correlation 
methods, such as linear and rank correlations, can be successfully applied for 
identification of solid and particulate materials using laser-induced plasma spectroscopy. 
A compact LIP spectrometer with microscopic sample imaging was used for rapid 
identification of plastics [38]. Emission spectra were collected with a compact, dual 
channel, fiber optic spectrometer and monitored either in a 230-310 nm or a 200-800 nm 
spectral window. Linear and nonparametric rank correlation methods were used for 
identification of plastic samples which had very similar compositions. A nearly 100% 
reliable identification was achieved. In another study, identification of particulate 
materials, such as iron ores and iron oxides, also yielded nearly 100 % accuracy [38]. 
The success of the correlation approach is based on the use of thousands of data points 
(pixels) representing the sample spectrum in a relatively large spectral window. 

In this chapter, the application of linear and nonparametric rank correlation for the 
identification of various pottery types is demonstrated. 

Experimental 
Instrumentation 
Micro-LIPS system 

A microscopic LIP spectrometer, developed in the laboratory, has been previously 
described [20]. Briefly, it consists of a compact Nd:YAG laser (model MK-367) which is 
attached to a modified microscope. The laser output is aligned with the microscope 
optical system using a dichroic mirror placed inside the microscope enclosure. The 
mirror reflects 99% of the laser radiation at 1064 nm and transmits visible light coming 
from the illuminated sample. A magnified image of the sample is monitored with a TV 
camera mounted on the top of the microscope and connected to a monitor. Microscope 



95 

magnification is varied with a set of objectives placed in a rotatable mount. For the 
present work, a lOx objective (total magnification 500x) is used with a working distance 
of 5 mm between the objective tip and the sample surface. With the lOx working 
objective, the laser is focused to a ~ 20 urn diameter spot on the sample surface providing 
an irradiance of ~ 10 12 W/cm 2 . The laser and the spectrometer are synchronized by a 
trigger pulse from a home-made compact pulse generator working either in a single pulse 
mode or at a 0.5 Hz repetition rate. 
Mini-LIPS system 

The mini-LIPS system consists of a compact Nd:YAG laser. The mirror reflects 
99% of the laser radiation at 1064 nm and transmits visible light coming from the 
illuminated sample. The laser and the spectrometer were synchronized by a trigger pulse 
from a custom-built compact pulse generator working either in a single pulse mode or at 
a 0.5 Hz repetition rate. 

In both cases the radiation from the laser spark was collected with a bifurcated 
optical fiber connected to a dual-channel Ocean Optics mini-spectrometer (SD2000). 
The spectrometer had the following characteristics: channel one — 230-310 nm spectral 
range, 3600 mm" 1 holographic grating, 25 urn slit, 0.16 nm spectral resolution, 2048 pixel 
linear CCD array; channel two — 200-850 nm spectral range, 600 mm" 1 grating blazed at 
400 nm, 25 pm slit, 1.3 nm spectral resolution, 2048 pixel linear CCD array. The 
spectrometer was driven from a laptop computer via a DAQCard-700 interface (National 
Instruments). Data acquisition and analysis were performed using software (Visual Basic 
6.0) written in-house. 



96 

Samples 

The first six samples examined were generously obtained form Department of 
Ancient Sciences of the University of Zaragoza and the majority were collected by 
excavations at Zaragoza, Spain. The city of Zaragoza, founded by Augustus Caesar, is 
very important for archaeology as it contains the remains of a Roman city established two 
thousand years ago. Descriptions of the sample fragments are given in Table 5-1. 
Samples 1-6 are from Spain and samples 7 and 8 are ceramics from Italy. 
LIPS Libraries 

LIPS libraries were compiled for identification of the ceramic materials. The 
library spectra were obtained by inducing the laser spark on 10 random surface spots and 
averaging the resulting 10 emission spectra. All libraries were stored in a computer and 
used on a day-to-day basis without being renewed. 
Software 

A program for correlation analysis was developed using Visual Basic 6.0 and the 
LabView drivers supplied with the Ocean Optics spectrometer. A detailed description of 
the program can be found elsewhere [20]. In brief, the software offered the following 
options to an operator: (i) choice of appropriate experimental parameters (integration 
time, trigger type, spectrometer channel, number of spectra to average, library file); (ii) 
pre-treatment of the spectrum (reduction in the spectral range, correction for continuous 
plasma background); and (iii) selection of a correlation method (linear or rank). The 
computer calculated all mutual correlation coefficients between the current spectrum and 
all library spectra. The correlation plot corresponding to the maximum correlation 
coefficient was displayed along with the statistical parameters (correlation coefficients, 
errors, probabilities) and the name of the compound which is identified with the highest 



97 

correlation probability. If desired, the correlation can be repeated with the use of another 
correlation method. The output data are saved and stored in a computer. 

Results and Discussion 

Small amounts of materials (micrograms) like ceramics are almost entirely 
atomized when exposed to laser radiation sufficient intense for breakdown. The 
application of LIPS for identification of materials is especially good because there is no 
loss of information in the plasma. As shown in these results, the large amount of 
spectroscopic data (2,048 pixels), used in the correlation procedure, allowed original 
information about the sample nature to be obtained. 

The correlation methodology was first applied to library ceramics SI through S6 
using the micro-LIPS system. Ten shot-averaged spectra from these samples are shown 
in Figure 5-1. The most prominent feature in all the spectra is an unresolved group of N 
II lines near 500 nm due to atmospheric nitrogen. A group of O II lines also appear in the 
region 350-450 nm and the O I triplet at 777.2-777.5 nm is also visible. Other features 
include a strong carbon line at 247.86 nm and the H a line at 656.28 nm. Visual 
examination of Figure 5-1 does not reveal any obvious differences in the spectra. 
However, one would only expect subtle variations in the ratios of line intensities, due to 
the varying stoichiometry of the various ceramics and also small variations in the nature 
of the background spectra due to differences in the laser material interaction. These 
subtle, yet consistent, differences in the spectra can be reliably discerned using a simple 
linear or rank correlation analysis [20]. 

Eight materials were chosen as indicated in Table 5-1. The results of 
identification using 10 shot-averaged spectra are shown in Tables 5-2 and 5-3. Both 
correlations show very high (>90%) probabilities of correct identification, except for a 



98 

few cases. There is some potential for improvement in the robustness of this technique. 
Additional software options can be applied in order to increase the probability of correct 
identification. For example, spectra pre-treatment options can be used which include 
choice of spectral range. Increasing the spectral range may be useful when a large 
portion of the recorded spectrum contains no information, or only spectral features which 
are related to the sample composition. Elimination of the uninformative portion of the 
spectrum may improve correlation results. 

In our case, the central portion of the spectrum (containing the strong group of N 
II lines) was entirely due to the surrounding air and not from the pottery. Therefore, it 
was logical to exclude this portion from the correlation analysis. The probability of 
identification was somewhat improved in the case of linear correlation and rank 
correlation (see Table 5-3). 

Obviously, any spectral information relating to the O and N content of the 
ceramics was lost in our measurements because the spectra were taken in air. Although 
the reliability of identification might be significantly improved if the samples were 
studied in an inert environment, this was not done because of the interest in the 
development of an approach that would be useful for rapid identification of ceramics in a 
normal environment. 

The other options for improvement of the robustness are instrumental. As shown 
previously [20], higher spectrometer resolution results in an increased probability of 
correct identification. Similarly, a wider spectral range is useful if it is not obtained at the 
expense of resolution. Ideally, high resolution and wide spectral range, combined 



99 

together (as in an echelle spectrometer) would provide the best possible performance for 
the proposed correlation routine, but at a significant increase in cost. 

Summary and Conclusions 

Pottery groups manufactured in ancient settlements of early Roman (late first 
century BC) were studied by LIPS. The goal of this work was the instant identification 
of ceramics by LIPS. LIP spectra from ceramics in a 200-800 nm spectral window were 
compared with reference spectral libraries stored in a computer. The libraries consisted 
of representative spectra from different groups of ceramic samples. The plasma emission 
spectra of eight archaeological samples were studied. Simple statistical correlation 
methods including linear and rank correlations were used. The probabilities of correct 
identification ranged from 0.8 to 1 with values close to unity for most of the samples 
studied. The present study aids in the characterization and identification of different 
types of materials used in pottery by means of LIPS, which is essentially non-destructive 
to the ceramic. 

A compact laser-induced plasma spectrometer has been used for rapid 
classification of different groups of old ceramics by using statistical correlation analysis. 
A customized software package was implemented which combined both data acquisition 
and data processing functions. Linear and non-parametric (rank) correlations were 
applied for classification of spectral data with approximately the same results. The 
robustness of the technique was demonstrated by achieving 90-99% reliable identification 
of almost all analyzed ceramics. The technique has excellent potential for on-line, real- 
time analysis of archaeological materials. 



100 



Sample 1 




Sample 2 



IXII) 



Mg 



r jJJJI^UuX Ua _j^ 



100 2O0 300 400 500 GOO 700 BOO 900 

Wavelength (nidi 



— I • 1 ' l • l • l 



100 200 300 400 500 600 700 SOO 900 

Wavelength (nm) 



Sample 3 



M* 



y 




100 200 300 400 500 



700 SOO 900 



Mg 


1) 


Sample 4 




<)( 




C 


N01) 


O(l) 


00 200 


300 i 


OO 500 «X 


700 WO 900 



Wavelength (nm) 



Wavelength (nm) 







Sample 5 




O(ll) 






J 


wvJL 


N(H> 


O(I) 

*» lu 



Sample 6 




O(I) 



100 2003004005006X700800900 

Wavelength (nm) 



Wavelength (nm) 



Sample 7 4 



Mf 




Sample 8 




100 200 300 400 SOO 600 

Wavelength (nm) 



700 600 900 



T • ' T 



100 200 300 400 5O0 600 700 BOO 900 

Wavelengtli (nm) 



Figure 5-1. LIP spectra from archaeological ceramic samples. 






101 



Table 5-1 . Description of pottery samples- 



Sample 



Material 



Origin 



Description 



SI 



Labitolosa. Terra Sigilata 
Hispanica 



Villaroya de la 

Sierra 

(Zaragoza), 

Spain 



Concave fragment, orange with a 

design length, 4.0 cm; width 4.0 

cm. 



S2 



Labitolosa. Terra Sigilata 
Hispanica 



Tricio (La Rioja), 
Spain 



Orange, curved fragment; length, 
4.5 cm; width, 2.0 cm. 



S3 



Labitolosa. Typical oxidant 
ceramic 



Zaragoza, Spain 



Orange, curved fragment: length, 
5.5 cm; width, 4.0 cm. 



S4 



Labitolosa. Typical reducing 
ceramic 



Zaragoza, Spain 



Black, curved fragment: length, 
5.0 cm; width, 3.5 cm. 



S5 



Lumpy ceramic 



Zaragoza, Spain 



Orange, curved fragment; length, 
6.0 cm; width, 4.5 cm. 



S6 



Typical ceramic 



Botorrita 

(Zaragoza), 

Spain 



White, curved fragment; length, 
4.5 cm; width, 4.0 cm. 



S7 



Recent pottery 



Italy 



Red, curved fragment; length, 6.5 
cm; width, 5.5 cm. 



S8 



Recent pottery 



Italy 



Red, curved fragment; length, 8.0 
cm; width, 6.5 cm. 



102 





Table 5-2. Calculated probabilities in ceramic archaeological samples 

LIPS system from 230-315 nm. 


using the mini- 


Linear correlation 
(Rank correlation) 


Samph 


;s SI 


S2 


S3 


S4 


S5 


S6 


S7 


S8 


SI 



(0) 


1 
(0.0730) 


0.5069 
(0.2011) 


0.9961 
(0.9983) 


0.9434 
(0.9526) 


1 

(1) 


0.9155 
(0.9976) 


0.9033 
(0.9992) 


S2 


0.9467 
(0.0924) 



(0) 


0.4654 
(0.4859) 


1 

(1) 


0.9490 
(0.9839) 


1 
(1) 


1 

(1) 


1 
(1) 


S3 


1 
(0.9453) 


1 
(0.9933) 



(0) 


1 
(0.9991) 


1 
(0.5088) 


1 

(1) 


1 
(1) 


1 

(1) 


S4 


0.9769 
(0.5454) 


0.9953 
(0.2963) 


0.9329 
(0.5088) 



(0) 


0.9999 
(0.9998) 


1 
(1) 


0.9985 
(0.9986) 


0.9854 
(0.9897) 


S5 


0.9910 
(0.7704) 


1 
(0.9857) 


1 

(0.8340) 


1 
(0.9991) 



(0) 


0.0970 
(0.8886) 


0.9887 
(1) 


0.9701 
(1) 


S6 


1 
(1) 


1 

(1) 


1 

(1) 


1 
(1) 


1 

(1) 



(0) 


1 

(1) 


1 
(1) 


S7 


0.9985 
(0.9563) 


0.9994 
(0.9986) 


0.9999 
(0.9097) 


1 
(0.9416) 


0.9105 
(0.9081) 


1 

(1) 



(0) 


0.9951 
(0.2565) 


S8 


0.9947 
(0.9987) 


1 
(0.3542) 


0.8393 
(0.9711) 


0.9816 
(0.8993) 


1 
(0.9768) 


1 

(1) 


1 

(0.9140) 



(0) 



103 





Table 5-3. Calculated probabilities in ceramic archaeological samples 

LIPS system from 180-315 nm. 


using the micro- 


Linear correlation 
(Rank correlation) 


Samplt 


is SI 


S2 


S3 


S4 


S5 


S6 


S7 


S8 


SI 




(0) 


1 
(1) 


1 
(1) 


1 
(1) 


1 
(1) 


1 
(1) 


1 
(1) 


1 
(1) 


S2 


0.9999 
(0.9988) 




(0) 


1 

(1) 


1 

(1) 


1 
(1) 


1 
(1) 


1 

(1) 


0.9812 
(0.9999) 


S3 


1 

(1) 


1 

(1) 




(0) 


1 

(1) 


1 

(1) 


1 
(1) 


I 
(1) 


1 
(1) 


S4 


1 
(1) 


1 

(1) 


1 

(1) 



(0) 


1 
(1) 


1 
(1) 


1 
(1) 


1 

(1) 


S5 


1 
(1) 


1 

(0.9999) 


1 

(1) 


1 

(1) 




(0) 


1 
(1) 


1 
(1) 


0.9999 
(0.9991) 


S6 


1 
(1) 


0.9989 
(1) 


1 

(1) 


1 

(1) 


0.9977 
(0.9985) 




(0) 


0.9999 
(0.9996) 


0.9790 
(0.9984) 


S7 


1 

(1) 


1 
(1) 


1 

(1) 


1 

(1) 


1 
(1) 


1 

(1) 




(0) 


1 
(1) 


S8 


1 
(1) 


0.9989 
(0.9963) 


1 

(1) 


1 

(1) 


1 
(0.9999) 


0.9999 
(0.9998) 


1 
(1) 




(0) 



CHAPTER 6 
ANALYTICAL MATRIX EFFECTS IN GEOLOGICAL MATERIALS 

A microscopic laser-induced plasma spectrometer was used to evaluate the 
analytical matrix effect commonly observed in the analysis of geological materials. 
Samples were analyzed in either the powder or pressed pellet forms. Calibration curves 
of a number of iron and aluminum compounds show a linear relationship between the 
elemental concentration and peak intensity. A direct determination of elemental content 
can thus be made from extrapolation on these calibration curves. To investigate matrix 
effects, synthetic model samples were prepared from various iron and aluminum 
compounds spiked with Si0 2 and CaC0 3 . The addition of these matrices had a 
pronounced analytical effect on those compounds prepared as pressed pellets. However, 
results indicated the absence of matrix effects when the samples were presented to the 
laser pulse as loose powders on adhesive tape. Results are comparable to certified values, 
indicating the reliability of this sampling approach for accurate analysis, provided the 
sample particle diameters are greater than a 100 u.m. Finally, the simultaneous analysis 
of two different elements was demonstrated using powders on tape. 

Introduction 

Laser-induced plasma spectroscopy is now a well-established technique for the 
elemental analysis of solid compounds. Although LIPS is advantageous in that it allows 
direct, rapid analysis with little or no sample preparation and is virtually non-destructive 
[39], LIPS signals emitted from the same element often depend on the matrix in which it 
is embedded [40]. 

104 



105 

Several groups have investigated analytical matrix effects in laser ablation. The 
dependence of matrix effects on plasma conditions was studied by measuring the ionic-to- 
atomic spectral line intensities of zinc and manganese following laser ablation into a dry 
inductively-coupled plasma (ICP) [41]. Others have also studied the effects of the matrix in 
laser ablation ICP mass spectrometry (LA-ICP-MS) and LA-ICP atomic emission 
spectroscopy (LA-ICP-AES) using geological materials prepared as pressed pellets [42, 43]. 
Chaleard et al. mathematically corrected for matrix effects by normalizing emission signals 
against the quantity of material vaporized and the plasma excitation temperature in LA 
optical emission spectrometry (LA-OES) [44]. Ciucci et al. incorporated a mathematical 
algorithm to overcome the analytical matrix effects [45]. Without the use of calibration 
curves, their method allowed precise and accurate determination of the elemental 
composition of materials using LIPS. 

In this study, the analytical matrix effects of powdered samples using LIPS are 
compared to those observed for the same samples prepared as pressed pellets. The 
elemental composition of ores is determined by extrapolation from calibration graphs made 
from compounds with known elemental concentrations. This chapter also discusses the 
simultaneous quantitative analysis of the major constituents in ore powders using LIPS. To 
maximize the analytical performance of LIPS, a better understanding of these matrix effects 
is necessary. 

Experimental 
Apparatus 

A compact LIP-spectrometer with microscopic sample imaging was used for 

quantitative spectrochemical analysis [20]. Spectra were collected with a compact dual 

channel fiber optic spectrometer and monitored either in a 230-310 nm or a 200-800 nm 



106 

spectral window. The instrumental configuration is the same as the micro-LIPS system 
described in Chapter 5. 
Sample Preparation 

To observe the effect of sample presentation in our system, samples having 
known concentrations of elements were prepared as either pressed pellets or powders. 
Each mixture was ground and homogenized for 30 minutes to ensure particle uniformity. 
For pressed samples, a portion of each mixture was transferred to an aluminum dish (32 
mm diam., 7 mm deep) and compacted into a pellet with a force of 5000 kPa for 5 
minutes. The powdered samples were placed on double-sided tape (3M) on a microscope 
slide. The powders were scattered over the tape surface ensuring that single particles, 
separated from each other by a distance of several particle diameters, could be distinctly 
seen. 

Iron compounds [Fe(N0 3 )3-9H 2 0, FeCl 2 4H 2 0, Fe 3 4 , Fe] in the pressed and pellet 
forms were studied. Similarly, several aluminum compounds were analyzed. With the 
signal intensities of particular atomic lines (274.6 nm for Fe and 282.2 nm for Al) and the 
varied concentrations of iron and aluminum in these compounds, calibration graphs were 
prepared. 

The matrices used were Si0 2 and CaC0 3 to represent the two important 
geochemical classes of silicates and carbonates. The various compounds of aluminum 
and iron were spiked with these matrices (at 70% matrix) to prepare calibration graphs. 
Matrix effects were investigated by examining the change in slope resulting from the 
addition of the matrix. 



107 

Other determinations were made to determine the accuracy of the technique, the 
effect of particle size, and the capability to simultaneously determine elemental content 
of different species in either the pressed or powdered form. 

Results and Discussion 

LIPS was used for several compounds of iron and aluminum using both powders 
and pellets. Calibration curves show a linear relationship between the concentration of 
the element and the peak intensity (Figure 6-1). The spectral lines used for compounds of 
iron and aluminum were 274 nm and 282 nm, respectively. These were chosen from 
among others based upon the more extensive self-absorption seen in other spectral lines. 

The addition of the matrices Si0 2 and CaC0 3 to the same iron and aluminum 
compounds indicated the preferential use of powders. Similar slopes observed for 
powdered iron samples indicate the absence of any matrix effect for iron (Figure 6-2). 
However, when iron compounds were prepared as pellets, the slope of the graph 
decreased by over 40% in the presence of calcium carbonate. Similarly, no matrix effect 
was observed for powdered aluminum samples, whereas the presence of both matrices in 
pellets of aluminum samples resulted in significantly different slopes (as much as 45%). 

The quantitative accuracy of the technique was determined by determining the 
iron content in powdered samples. Powdered samples were used because of a minimal 
observed matrix effect. The results in Table 6-1 are comparable (% relative errors of less 
than 10%) to certified values, indicating the reliability of the use of powders for accurate 
analysis. 

Study of the influence of particle diameter on the emission signal intensity is 
important because of the application of LIPS to geological samples. By extrapolation of 
the calibration curve made from aluminum compounds, the aluminum content of AI 2 3 



108 

particles of varying size was determined and compared to the certified value. The results 
from the analysis (Table 6-2) indicate a strong relationship between particle size and 
signal intensity. As shown, LIPS analysis was only reliable for sample particles with 
diameters greater than about 100 p,m. 

Elemental determination was also extended to the analysis of NIST SRM 
(National Institute of Standards and Technology - Standard Reference Material) ores. 
The amounts of A1 2 3 and Si0 2 in certified ore samples were determined simultaneously. 
Two calibration curves were constructed from analyses of known concentrations of ore 
standards 691 and 693 diluted with graphite (Figure 6-3). Elemental concentrations were 
obtained by extrapolation of these calibration curves. For example, NIST ore 691 ( 1 .22% 
A1 2 3 , 3.70% Si0 2 ) was diluted with graphite at various percentages: 10% 691 with 90% 
graphite, 30% 691 with 70% graphite, 50% 691 with 50% graphite, 70% 691 with 30% 
graphite, and 100% 691. Each of these concentrations could be stoichiometrically 
converted to the A1 2 3 and Si0 2 concentrations, each under 4%, as shown in Figure 6-3. 
These samples were then prepared as either pellets or powders. This same approach was 
used in the calibration with NIST ore 693, resulting in the four graphs of Figure 6-3. 

The results, shown in Table 6-3, indicate reasonable accuracy with samples 
prepared as powders, indicating the absence of any significant matrix effect. However, 
poorer results were obtained with samples prepared as pellets. The proposed method was 
applied to the simultaneous determination of A1 2 3 and Si0 2 in ore standards. A 5.6% error 
for AI2O3 and 0.3% error for Si0 2 were achieved using powders. The small percent error 
observed with the use of powders relative to that obtained from the use of pellets indicates 
the preferential utilization of powders for this methodology. Better precision was also 






109 

achieved using powders; the RSD obtained using powders is approximately 5%, while that 
obtained with pellets is 8%. 

Conclusion 
An accurate determination of powdered geological materials was achieved by 
using a compact laser-induced plasma spectrometer with microscopic sample imaging. 
Matrix effects were evaluated by examining calibration curves constructed from analyses 
of iron and aluminum compounds. These compounds were presented to the laser beam in 
either the pressed pellet or powder form. Accurate concentrations were determined from 
only those samples prepared as powders, indicating the absence of matrix effects that are 
typically observed. Experimental results agreed well with certified values, provided the 
sample particle size was greater than 100 ^.m. Finally, the technique was able to 
simultaneously determine the concentrations of two different elements with reasonable 
accuracy. The advantageous powder sample preparation has potential for on-line, real- 
time analysis of raw materials for mining and chemical processing industries. 



110 



1800 -, 



o 



1 




Fe(100%Fe) 



Concentration (%) 



Figure 6-1. Calibration curve of iron. 



Ill 




T" 
§ 



o i: 

m c 

■ 

U 

c 
o 
o 




(sjunoo) /Cjbibjuj 



tsjunoo) Ajisu3}U| 



1/1 
C 

<u 

o 

a. 

s- 

o 

en 
J 

"5 
ex 

•o 




rS 






ill 

u. wo 




p8 



I 
c 

8 

c 
o 
O 
o 



-r. 

-o 

c 

3 
O 

Q. 

s 

o 
o 



Hi 

■ 

U 

E 

E 



■a 



(sjunoo) Ajisuajui 



(sjunoo) \jisu3jui 



12 



Table 6-1. Determination of iron in ores (wavelength: 274.60 nm) 



°amnlc % fron 




LIPS value T Certified value 


w ciiui 


690 


70.69 66.85 


5.7 


691 


82.17 90.80 


9.5 


692 


64.72 59.58 


8.6 


693 


64.86 65.11 


0.4 



T Each value is an average of five shots. 



113 



Table 6-2. Determination of aluminum in particles of A1 2 3 of different sizes 

(wavelength: 282.20 nm). 



Particle size 
(um) 


%A1 
LIPS value 1 


% error 


15 


37.13 


38.6 


30 


47.87 


20.9 


44 


56.20 


7.1 


117 


59.33 


1.9 


208 


59.42 


1.8 


617 


60.92 


0.7 


T Certified value is 60.5% Al; 



each value is an average of five shots. 



114 



a 


-a 


i 


— 


E 


T3 


B 


C 


M 


B 


B 

O 




u 


© 




u 




a 




C 




Xjisuaju] 



jdisasjui 



115 





Table 6-3. Simultaneous determination of AI2O3 


and Si0 2 


in ore standards. 










% AI2O3 










% Si0 2 






Calibration 

with 691 


Certified 
value 


Powder 


% error 


Pellet 


%error 


Certified 
value 


Powder 


%error 


Pellet 


%error 


27f 


0.82 


0.98 


19.5 


1.25 


52.4 


4.17 


3.99 


4.3 


3.18 


23.7 


690 


0.18 


0.24 


33.3 


1.31 


627.8 


3.71 


3.64 


1.9 


1.98 


46.7 


693 


1.02 


1.24 


21.6 


1.14 


11.8 


3.87 


3.88 


0.3 


1.81 


53.2 




Calibration 

with 693 


Certified 
value 


Powder 


%error 


Pellet 


%error 


Certified 
value 


Powder 


%error 


Pellet 


%error 


27f 


0.82 


1.05 


28.0 


1.23 


50.0 


4.17 


4.57 


9.6 


6.90 


65.5 



690 0.18 



0.17 



5.6 



1.55 761.1 



3.71 



3.76 



1.3 



4.59 



691 



1.22 



1.15 



5.7 



23.7 



1.84 50.8 3.70 4.10 10.8 14.35 287.8 



CHAPTER 7 
CONCLUSIONS AND FUTURE WORK 

The experiments performed in this research have demonstrated the versatility of 
laser-induced plasma spectroscopy. Identification of materials was achieved by using the 
spectral "fingerprints" that are unique to each sample. LIP spectra are amenable to many 
types of data analysis techniques, only a few of which have been discussed in this 
dissertation. Several studies involved the development and evaluation of various 
instrumental configurations, with the goal of optimizing a configuration for use in the 
phosphate industry. Furthermore, LIPS was used to study the archaeological significance 
of certain ceramics from the first century BC. Finally, the analytical matrix effects 
commonly found in LIP spectra were investigated. 

The correlation techniques may still be improved. The software currently under 
development incorporates additional parameters, such as selection of a spectral window 
rather using the entire spectrum for analysis. The uninformative portions of spectra 
(background noise, unrelated peaks, tails) may be eliminated during spectra pre-treatment 
to increase the probability of correct identification. The primary concern in the 
chemometrics realm is the enhancement of the linear and rank correlation techniques, 
since these are the fastest and most robust of algorithms. 

Other future endeavors could include the incorporation of the principal 
component analysis into real-time analysis. Currently, PCA is only used in data post- 
treatment. It would be advantageous to combine the accuracy of PCA with a simple one- 
shot LIPS analysis. This computer undertaking would achieve a highly reliable LIPS 

116 



117 

system, although its speed would be limited by the computer's processor and the program 
which is written. Also, additional chemometric methods may be explored, including 
nonparametric methods such as K-nearest neighbor and neural networks. 

In the development of a LIPS instrument for the phosphate industry, it is clear that 
the design of the field instrument can be improved. For single shot measurements such as 
these, a much smaller (and potentially less expensive) laser could be used, although a 
commercial source for the ideal laser for this application does not presently exist. In 
principle, the entire field instrument could be self-contained, self-powered and no larger 
than the size of the present probe head. The display electronics could also be simplified 
and made more compact, perhaps with the use of a palmtop computer. To improve upon 
remote analyses at distances of 10-20 m, the engineering of a field version would be the 
next practical step. The main concern with the use of such an instrument in an industrial 
environment is eye safety. 

In order to pursue the use of LIPS as a reliable tool in archaeological analysis, 
several experimental avenues may be explored. A number of samples from known time 
periods may be analyzed in order to grasp the potential of the technique in this field. The 
virtual non-destructiveness of LIPS is an advantage in the analysis of precious or ancient 
artifacts. Finally, further investigations of analytical matrix effects are necessary. The 
analytical signals from a multitude of compounds must be scrupulously examined. 

In conclusion, the advantages of LIPS complement its flexibility and usefulness in 
a variety of applications. The future of LIPS in these particular applications is promising. 
Because of the broad applicability of the technique, the full potential of LIPS may never 
be realized. 



APPENDIX 
OPERATIONAL INSTRUCTIONS FOR THE PORTABLE LIPS PROBE 

Laser Operation 

Turn on the laser power supply by turning the key clockwise on the laser ICE 
(Integrated Cooler and Electronics). Flip the toggle switch on the probe to the "On" 
position. Press the "Run" button on the laser ICE when ready to acquire spectra. Ensure 
the PRF (Pulse Repetition Frequency) is set to "0." Press the triangular down key to 
change PRF to if necessary. The laser will fire and a spectrum will be taken when the 
trigger button on the handle is depressed and the probe is set on a sample surface. 

Software General Description 

A program for soil layer identification is written in Visual Basic 6.0. The 
software combines both data acquisition and data processing functions. It allows reliable 
classification of three groups of soil samples, overburden, matrix, and bedrock, by using 
statistical correlation analysis. Linear and nonparametric rank correlations can be applied 
for classification of spectral data with approximately the same results. 

The software offers the following options to an operator: (i) choice of number of 
laser shots, (ii) choice of auto or external trigger, (iii) adjustment of discrimination level, 
(iv) adjustment of graph scale, and (v) options regarding the visual display. Reference 
spectral data are stored in a library on the computer's hard drive. As soon as the laser is 
triggered, a new spectrum is obtained and compared to the spectra in the reference 
library. The software immediately identifies the ablated material as either overburden, 
matrix, or bedrock. 



118 



119 

Software Operational Instruction 

1. Turn on the Sony VAIO notebook computer by pressing the power button in the upper 
right region of the keyboard. The computer will take a few minutes before Windows is 
ready. 

2. In Windows, use the pointing device in the middle of the keyboard to highlight the 
"LIPS Probe" icon. Select by clicking the left button on the bottom of the keyboard. The 
operational window will be displayed (Figure A-l). The program is controlled by the 
selection of options in the pull down menus (see below). 

3. Select "Spectrometer" -» "Trigger" -» "External" 

This allows the software to acquire spectra from the manual use of the LIPS 
probe. Ensure that the check mark is to the left of "External." This step is only necessary 
if the spectrometer was operating in the automatic mode in the previous session. 

4. Set the Total number of laser shots to the number desired in the window "Number of 
laser shots" (Figure A-l). 

5. Select "Correlation" — » "Select correlation library" 

An "Open" dialog box appears. Select EVIC-Agrico Library (or the library of 
choice for correlation with a current spectrum) by clicking on it. 

(To create or add to a library, see instructions on "Creating a Library or Adding to an 
Existing Library" below.) 

6. Press "GO" and depress the trigger button on the handle of the probe. (Ensure that the 
laser ICE is in the "run" mode.) The software will identify the layer by highlighting the 
name of the layer in red and including a "!" by its identification. The resulting screen is 
shown in Figure A-2. A correlation summary, a list of correlation coefficients relative to 
each sample in the library, and a correlation plot are all displayed along with the newly 



120 

acquired spectrum. A summary of identification results can be obtained by clicking on 
"Correlation Statistics." The library spectrum is superimposed on the working spectrum. 
The colors of the two spectra can be set independently via menu option "View" and its 
submenus (see below). The correlation plot is depicted in the top right corner of the 
spectrum graph and shows graphically quality of correlation. If the correlation plot is 
close to a 45-degree straight line (as the one shown in Figure A-2), the correlation is 
good; if it is represented by a set of widely scattered dots, the correlation is poor. 

Software Available Options 
Changing Spectrometer Coefficients 

Select "Spectrometer" -> "Coefficients" 

These settings should only be changed if the spectrometer inside the probe is 
changed. The coefficients are included on the wavelength calibration data sheet provided 
by Ocean Optics. 
Adjusting Discrimination Level 

Select "Spectrometer" — ► "Discrimination Level" 

Enter a value between 1 and 10. A lower number indicates more sensitivity in the 
acquisition of spectra. However, while a higher number will prevent one from having to 
repeat measurements, there is a decreased chance of accurate identification with higher 
values. 
Creating a Library or Adding to an Existing Library 

Select "Library" -» "Create Library" 

Select "OK", then "GO" 

Depress the trigger button on the probe handle to acquire a spectrum. After a new 
spectrum is acquired, the software prompts to a "Save As" dialog box. To create a new 



121 

library, enter a name for the new library in the "File Name" line and select "Save." To 
add to an existing library, select the library to which the new spectrum would be added. 
Enter the layer identification as "Overburden," "Matrix," or "Bedrock" in the dialog box 
prompting for the name of the new library member and select "OK." 
Adjusting Graph Scale 

Select "View" -* "Scale" 

Check "Autoscale" for the software to determine the graph scale automatically. 

Check "Define Scale" to enter the graph scales for the x and y axes manually. 
Enter the desired values in each box and press "OK." 

Check "Full Scale" to use the maximum (default) scale for spectral intensity. 
Changing Visual Display 

The operator has the capability to adjust the color of the display according to 
preference. Select "View," and then select the item that is desired to be changed. The 
color of the background, foreground, current spectrum, and the library spectrum may be 
adjusted. 

The operator may also display a grid overlaid on the graph. Select "View" — *■ 
"Grid." Vertical or horizontal lines (or both) may be displayed to more closely examine 
spectral lines. These lines may be solid, dashed, or dotted, and the color may of the lines 
may be adjusted as well. 
Choosing Spectra Acquisition 

The data may be acquired from the spectrometer or from data previously 
acquired. Choose "Spectra acquisition" and select the appropriate option. Before 
acquiring data with the spectrometer, it is beneficial to run some previously collected 
spectra (from a correlation library, for example) and to check how the software works. 



122 

Background noise 

The operator may choose to acquire raw spectra or subtract the background from 
a spectrum. Choose "Options" and select the desired procedure. For successful 
correlation, however, the option "Raw spectra" must always be checked. 
Correlation Analysis 

A correlation library may be selected for comparison to the current spectra. The 
operator has the choice of selecting either a linear correlation or a rank correlation under 
the "Correlation" drop-down window. As a rule, correlation results for linear and rank 
correlations are the same, though linear correlation is processed much faster. However, 
rank correlation is believed to provide more trustful results. Therefore, in cases of 
doubtful classification, it is helpful to collect identification statistics using both linear and 
rank correlations. 
Saving Correlation Results 

The correlation results are saved automatically to the file "c:\CorrResults.txt" 
after each session of data acquisition/processing. This file will be rewritten upon the next 
session. Therefore, to save the results of each session, it is necessary to rename the file 
"c:\CorrResults.txt" and place it in the desired location. The corresponding prompt 
appears in the end of each acquisition/processing session. 



123 

Pull-down menus 

1. File 

Save 
Save As 
Print 
Exit 

2. Spectrometer 

Coefficients 
Trigger 

Auto 

External 
Discrimination Level 

3. Library 

Create Library 
Convert existing library 

From raw to bkg subtracted 

From one spectrometer to another 
From OOIBASE 

Convert single file 

Convert incremental sequence of files 

4. View 

Scale 

Autoscale 

Define Scale 

Full Scale 
Back Color 
Fore Color 
Graph Color 
Library Graph Color 
Grid 

Background color 
Cut-off limits 

Show (check) 

Color 
Peaks 

Show peaks 

Color 

5. Spectra acquisition 

From spectrometer 
Use stored data 

6. Options 

Raw spectra 

Auto bkg 

Forced bkg (Routines 2-64) 

Subtract bkg 

7. Correlation 

Select correlation library 
Linear correlation 
Rank correlation 



124 

Shut-down Procedure 

1 . Press "EXIT" in the software to exit the program. 

2. Click on the "Start" menu in the bottom left corner of the screen and select "Shut 
Down." Ensure "Shut Down" is highlighted and select "OK." After several seconds, the 
computer will automatically turn off. Fold the computer until it snaps to lock. 

3. Flip the toggle switch on the probe to the "OFF" position. 

4. Ensure the Laser ICE is in the "STOP" mode. The light next to "STOP" should be lit. 

5. Turn the key on the laser ICE counter-clockwise to turn off. 

Troubleshooting 

1. Avoid pressing keys in a random fashion; the software is not fully protected against 
unpredictable actions. Know exactly what you are doing at each step. 

2. If an error is generated, the software will terminate automatically. Simply restart the 
software. 

3. If the software continually indicates that a poor spectrum was obtained and the 
measurement needs to be repeated, adjust to a higher value for the discrimination level. 

4. In all other cases, please contact Ben Smith or Igor Gornushkin in the Department of 
Chemistry at the University of Florida. 



125 



Matrix 

Overburden 

Bedrock 



Number of laser shots [l 
Current run number 

go I 



JiBili 



Library spectrum 
Current spectrum 



EXIT 



Wavelength, ran 



Figure A-l . View on the main menu window before any acquisition or processing of 

data. 






126 



in x 



! Matrix 1 

Overburden 0.9491 
Bedrock 0.9723 



Number of laser shots l 
Current run number 



GO 



Source library 

C\IMC Software\Lib\1MC-Agrico Library/ 
Library spectrum of 
..Matrix 



Correlation library. 
C\IMCSaltwore\Lib\IMC-Agrico Ubrary.t 



Linear correlation 
Max Com. Coefl. M 
Most Likely Compound Is 
... Matrix 



Bedrock 09723 
Bedrock 9206 

rix 1 
Mafrix 09967 
Overburden 00845 
Overburden 00376 
Overburden 07542 
Overburden 0.936 
Overburden 0.9491 
Matnx 09228 
Matrix 09568 
Matrix 06616 



1 



d 



EXIT 





1b41 H 


! ""3 






Library spectrum 


.••' ! 




3200- 

?oon- 

2400 
2000- 


j i ... 8421: 











a r 








/i 








«-«-il«- •»» ISIl 






1200- 
800 




































i * i 


i ■ i ' r- — > 1 





Wavelength, nm 



Figure A-2. Resulting screen after the data were collected and processed. 









REFERENCES 

1. Brech, F. and L. Cross, Appl. Spectrosc. 16, 5, 1962. 

2. Runge, E. F., R. W. Minck and F. R. Bryan, Spectrochim. Acta, 20, 733, 1964. 

3. Cremers, D. A. and L. J. Radziemski, Eds., Laser-Induced Plasmas and Applications, 
Marcel Dekker, Inc.: New York, 1989. 

4. Thiem, T. L., S. Dementev, J. Melville, H. S. Melville, A. F. Kotyuk and M. 
Ulanovkiy, Amer. Lab. 20C, Nov. 1995. 

5. Sneddon, J. Thiem, T. L. Lee, Lasers in Analytical Atomic Spectroscopy, VCH 
Publishers, Inc.: New York, 1997. 

6. Ahmad, I. and B. J. Goddard, Jurnal Fizik Malaysia. 14, 43, 1993. 

7. Basov, N. G., V. A. Boiko, V. A. Gribbov, S. M. Zakharov, D. N. Krokhin and G. V. 
Sklizkov, Sov. Phys. JETP, 34, 81, 1972. 

8. Zapka, W. and A. C. Tarn, J. Opt. Soc. Am. 71, 1585, 1981. 

9. Noll, R., R. Sattman and V. Sturm, SPIE, 2248, 50, 1994. 

10. Ingle, J. D. and S. R. Crouch, Spectrochemical Analysis, Prentice Hall: Englewood 
Cliffs, New Jersey, 1988. 

11. Davies, C. M., H. H. Telle and A. W. Williams, Fres. J. Anal. Chem. 355, 895, 
1996. 

12. Yamamoto, K. Y., D. A. Cremers, M. J. Ferris and L. E. Foster, Appl. Spectrosc. 50, 
222, 1996. 

13. Cremers, D. A., M. J. Ferris, C. Y. Han, J. D. Blacic and D. R. Pettit, SPIE, 2385, 
28, 1995. 

14. Ciucci, A., V. Palleschi, S. Rastelli, R. Barbini, F. Colao, R. Fantonic, A. Palucci, S. 
Ribezzo and H. J. L. Van der Steen, Appl. Phys. 63B, 185, 1996. 

15. Cremers, D. A., Appl. Spectrosc. 41, 572, 1987. 

16. Loree, T. R., Laser Institute of America, ICALEO Conference, Nov. 1983. 



127 



128 



17. Radziemski, L. J., D. A. Cremers, and T. R. Loree, Spectrochim. Acta, 38B, 349, 
1983. 

18. Davies, C. M., H. H. Telle, D. J. Montgomery, and R. E. Corbett, Spectrochim. 
Acta, 50B, 1059, 1995. 

19. Anglos, D., S. Couris, and C. Fotakis, Appl. Spectrosc. 51, 1025, 1997. 

20. Gornushkin, I. B., B. W. Smith, H. Nasajpour, and J. D. Winefordner, Anal. Chem. 
71,5157,1999. 

21. Lehmann, E. L. and H. J. M. DAbrera, Nonparametrics: Statistical Methods Based 
on Ranks, rev. ed., Prentice-Hall: Englewood Cliffs, NJ, 1998. 

22. Anton, H., Elementary Linear Algebra. John Wiley & Sons: New York, 1994. 

23. Lay, D. C, Linear Algebra and Its Applications, Addison Wesley: New York, 2000. 

24. Wilson, M. J., Clay Mineralogy: Spectroscopy and Chemical Determinative 
Methods, Chapman & Hall: Oxford, U. K., 1994. 

25. Wieder, M. and D. Adan-Bayewitz, Catena, 35, 327, 1999. 

26. Clark, R. and P. J. Gibbs, J. Raman Spectrosc. 28, 99, 1997. 

27. Adan-Bayewitz, D. and I. Perlman, Archaeom. 27, 203, 1985. 

28. Adan-Bayewitz, D. and M. Wieder, J. Field Arch. 19, 189, 1992. 

29. Giauque, R. D.,X-Ray Spectrom. 23, 160, 1994. 

30. Giauque, R. D., F. Asaro, F. H. Stross, and T. R. Hester, X-Ray Spectrom. 22, 
44, 1993. 

31. Papargyris, A. D. and R. D. Cooke, Brit. Cer. Trans. 95, 107, 1996. 

32. Wang, D. and D. W. Anderson, Can. J. Soil Sci. 80, 251, 2000. 

33. Su, C. F., S. Feng, J. P. Singh, F-Y Yueh, J. T. Rigsby JH, D. L. Monts and R. L. 
Cook, Glass Technol. 41, 16, 2000. 

34. Anglos, D., S. Couris and C. Fotakis, Appl. Spectrosc. 51, 1025, 1997. 

35. Burgio, L., R. J. H. Clark, T. Stratoudaki, M. Doulgeridis and D. Anglos, Appl. 
Spectrosc. 54, 463, 2000. 



129 



36. Gobernado-Mitre, I., A. C. Prieto, V. Zafiropoulos, Y. Spetsidou and C. Fotakis, 
Appl. Spectrsoc. 51, 1 125, 1997. 

37. Gornushkin, I. B., A. Ruiz-Medina, J. M. Anzano, B. W. Smith and J. D. 
Winefordner, J. Anal. At. Spectrom. 15, 581, 2000. 

38. Anzano, J. M, I. B. Gornushkin, B. W. Smith and J. D. Winefordner, Polym. Eng. 
Sci. 40, 2423, 2000. 

39. Rusak, D. A., B. C. Castle, B. W. Smith and J. D. Winefordner, Crit. Rev. Anal. Chem. 
27,257,1997. 

40. Eppler, A. S., D. A. Cremers, D. D. Hickmott, M. J. Ferris and A. C. Kostelo, Appl. 
Spectrosc. 50, 1175, 1996. 

41 . Chan, G. C.-Y., W. T. Chang, X. Mao and R. E. Russo, Spectrochim. Acta, 55B, 
221,2000. 

42. Morrison, C. A., D. D. Lambert, R. J. S. Morrison, W. W. Ahlers, and I. A. Nicholls, 
Chem. Geol. 119, 13, 1995. 

43. Motelica-Heino, M., O. F. X. Donard and J. M. Mermet, J. Anal. At. Spectrom. 14, 
675, 1999. 

44. Chaleard, C, P. Mauchien, N. Andre, J. Uebing, J. L. Lacour and C. Geertsen, J. Anal. 
At. Spectrom. 12, 183, 1997. 

45. Ciucci, A., M. Corsi, V. Palleschi, S. Rastelli, A. Salvetti and E. Tognoni, Appl. 
Spectrosc. 53,960, 1999. 






BIOGRAPHICAL SKETCH 
The author was born August 21, 1975, in Atlanta, Georgia, and spent his 
childhood in Stone Mountain, a suburb of Atlanta. He is the youngest child in a family 
that includes his parents, Florie and Delynn, and his sister, Marie, and brother, Myles. 
He attended Stone Mountain High School graduating with honors in 1993. He attended 
Berry College in Rome, Georgia, and graduated cum laude in 1997 with a Bachelor of 
Science in chemistry and a minor in mathematics. He studied analytical chemistry under 
the direction of James D. Winefordner at the University of Florida to pursue his doctoral 
degree. 



130 



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. 




James D. Winefordner^Chairman 
raduate Research Professor of Chemistry 

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 H. Powell 
Scientist in Chemistry 



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. 



Willard W. Harrison 
Professor of Chemistry 



I certify that I have read this study and that in my opinion it conforms to 
acceptable standards of scholarly presentation and is fulh/ adequate, in scope and quality, 
as a dissertation for the degree of Doctor of Philosor 




D. Stewart 
Associate Professor of Chemistry 






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. 

t 




Jrfeph J. Delfino 

wfessor of Environmental Engineering 
Sciences 



This dissertation was submitted to the Graduate Faculty of the Department of 
Chemistry in the College of Liberal Arts and Sciences and to the Graduate School and 
was accepted as partial fulfillment of the requirements for the degree of Doctor of 
Philosophy. 



May 2002 



Dean, Graduate School 



LD 
1780 
20 £ 







\)15S 



UNIVERSITY OF FLORIDA 

3 1262 08555 3401