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

MISSING IMAGE

Material Information

Title:
Laser-induced plasma spectroscopy in the analysis of phosphate mining samples and archaeological materials
Physical Description:
Book
Creator:
Villoria, Mark Anthony, 1975-
Publication Date:

Record Information

Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 28839367
oclc - 50473398
System ID:
AA00020485:00001

Full Text










LASER-INDUCED PLASMA SPECTROSCOPY
IN THE ANALYSIS OF PHOSPHATE MINING
SAMPLES AND ARCHAEOLOGICAL MATERIALS









t "


IBy .

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

UNIVERSITY 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 Gomrnushkin 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








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.















TABLE OF CONTENTS

page

ACKN OW LED GM ENTS..................................................................................................iii

LIST OF TABLES ............................................................................................................ vii

LIST OF FIGU RES.......................................................................................................... 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 .................................................................................................................. 11

3 CHEMOMETRIC APPROACH TO LIPS .................................................................... 21

Linear Correlation ......................................................................................................... 22
Rank Correlation ........................................................................................................... 23
Principal Com ponent Analysis...................................................................................... 25
PCA of Phosphate M ining Sam ples ....................................................................... 26
Com prison of Chem om etric M ethods......................................................................... 28

4 RAPID FIELD IDENTIFICATION OF PHOSPHATE MINING SAMPLES ............. 46

Phosphate M ining.......................................................................................................... 46
Background ...................................................................................................................48
Prelim inary studies........................................................................................................ 49
Correlation Studies w ith the Benchtop Instrum ent ....................................................... 50
Experim ental Setup and M ethodology................................................................... 50
Results .................................................................................................................... 52
Effect of Sam ple Position....................................................................................... 53
Continuous Correlation on a M oving Sam ple........................................................ 54
Fiber-Optic Probe.......................................................................................................... 54








Experim ental Setup and M ethodology................................................................... 54
Results .................................................................................................................... 55
Rem ote LIPS ................................................................................................................. 55
Experim ental Setup and M ethodology................................................................... 55
Results .................................................................................................................... 56
The Portable LIPS Probe.......................................................................................... 57
Experim ental Setup and M ethodology................................................................... 57
W et vs. Dry sam pling.......................................................................................... ... 58
Real Sam ple Analysis ........................................................ ..................................... 58
IM C-Agrico Site Visit................................................... ......... ............................ 58
Results............................................................................................. ...................... 59
Conclusions...................................................................................................................62

5 LIPS FOR CHARACTERIZATION OF ARCHAEOLOGICAL MATERIALS.......... 92

Introduction................................................................................................................... 92
Experim ental................................................................................................................. 94
Instrum entation........................................................................................ ........ 94
M icro-LIPS system ......................................................................................... 94
M ini-LIPS system ..................................................................... ..... ........ 95
Sam ples ..................................................................................................................96
LIPS Libraries ........................................................................................................ 96
Software............................................................................................... ................... 96
Results and D iscussion............................... ........................................ .......... ................. 97
Sum m ary and Conclusions.................................................. ...................................... .... 99

6 ANALYTICAL MATRIX EFFECTS IN GEOLOGICAL MATERIALS .................. 104

Introduction................................................................................................................. 104
Experim mental ........................................................................................ ..................... 105
Apparatus..................................... ......................................................................... 105
Sam ple Preparation............................................................................................... 106
Results and Discussion ................................................................................. .............. 107
Conclusion...................................................................................... .............................109

7 CONCLU SION S AND FUTURE W ORK .................................................................. 116

APPENDIX

OPERATIONAL INSTRUCTIONS FOR THE PORTABLE LIPS PROBE................. 118

REFEREN CES................................................................................................................ 127

BIOGRAPHICAL SKETCH .......................................................................................... 130














LIST OF TABLES


Table Page

2-1 Early laser system s.......................................................................................................... 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 IMC-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 A1203 of different sizes .............................. 113

6-3 Simultaneous determination of A1203 and Si02 in ore standards................................... 115















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

3-11 Scores of the first two principal components for overburden spectra........................... 40

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








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

4-16 Spectra of wet and dry bedrock samples...................................................... 84

4-17 Spectra of wet and dry matrix samples....................................................... 85

4-18 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








5-1 LIP spectra from archaeological ceramic samples.......................................................... 100

6-1 Calibration curve of iron................................................................................................. 110

6-2 Fe & Al compounds prepared as pellets or powders....................................................... 111

6-3 Simultaneous determination of A1203 and SiO2 in ore standards................................... 114

A-i 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








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.













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 IMC-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

Jesis 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.






2


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,

(J/g) [5]. Thus, no evaporation occurs below the minimum absorbed irradiance, Fmin

(typically 1012 W/m2 for a Q-switched laser). The two are related in the following

equation:


Fmin = p Lv a te"


(eqn. 3)








where p is the mass density of the target (g/mm3), a is the thermal diffusivity (mm2/s),

and te 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, (s), to the irradiance of the

laser, F (W/cm2):

iffpC (Tv-T,,)2
(eqn. 4) tv = _______
4F2


where K is the thermal conductivity (J/s cm K), p is mass density (g/cm3), C is heat

capacity per unit mass (J/g K), and T, and To 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








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, Xp,, light is reflected and the plasma

shrinks instead of expands, as it is no longer absorbing the light. The plasma wavelength

is defined by Xp 3.35 x 107 (ne)y, where nle 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.








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 104 to 105 K and electron densities on the order of 1015 to 1019 cm3 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 (CO2 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 Nd3 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 mm2 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.








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 mm3 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 ps 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








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








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].










~~_____________Table 2-1. Early laser systems.________
Manufacturer Model No. Laser type Year of
______________________(wavelength) manufacture
Thermo Jarrell Ash Mark I Ruby (694.3 nm) 1963
(Franklin, MA) Mark II Nd:glass (1060 nm) 1966
L O M O (L en in grad oT iTi i i n \ ir "
LOMO (Leningrad, MSL 2 Nd:glass (1060 nm) 1967
Russia) ___________-__________________
Shimadzu (Tokyo, n / n n -
Shimadzu (Tokyo, Ruby (694.3 nm) 1967
Japan) _________
CISE Segrate
Milano (Milan, CISE I Ruby (694.3 nm) 1966
Italy)__________________________
Ford Motor Co. 1963
(Dearborn, MI)________________________
VEB Carl Zeiss LMA Nd:glass (1060 nm) 1964
(Jena, Germa) or Ruby (694.3 nm)
(Jena, Germany) LMA I Nd:glass (1060 nm) 1965
Optical Technology, Model 120 Nd:glass (1060 nm) 1965
Inc. Model 190 Nd:glass (1060 nm) 1968
Source: [4]











Thomson scattering
Optical probing Faraday rotation
Interferometry


-, ri- .-~


o e'sI
I


Critical
surface
(n,)


nc/4


Hard x-rays
Soft x-rays

Harmonics (nto)
Raman scattered light


Laser co


Expanding
sheath and
plasma blowoff


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


Thermal
conduction
/ \


/,


Ablation
surface













Hot, high-pressure
strongly absorbing
vapor plasma

Temperature
profile

Ambient
Atmosphere
Conduction

Shock wave


Radiation


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














Laser strikes
the surface








Vaporization starts,
heat dissipates slowly








Surface explodes,
breakdown occurs,
plasma forms


t =0ns


t-Ins


Instant increase m
surface temperature








Underlying layers
reach critical T, p








Plasma expands,
becomes opaque,
heats environment
Pm


t-5ns


Plasma continues
expanding shock
wave forms so


t 1000 ns


Strong electron-ion
recombination,
neutrals form


t 20000 ns


Plasma coo ls
down, becomes
unstable, decays
p


t- 10000 ns


Plasma products
diffuse into
environment
It,


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


t-lns


t-lns


t- 10ns
































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












Table 2-2. Comparison of lasers utilized in LIPS systems.
Type Wavelength Pulse width Pulse energy Comments
(Am) (nsec) (J)
1.06 Compact, low
Nd:YAG 0 7-12 0.3 to 1.0 maintenance, reasonably
0.53 priced, glass optics
Simple design, inert gas,
CO2 10.6 1 to 300 tsec 0.5 to 500 special IR optics required
for laser pulses
Rep rates to 250 Hz, UV
Excimer 0.194 to 0.351 10 to 30 0.25 wavelengths, toxic gases,
Excimer 0.194 to 0.351 10 to 30 0.25 quartz optics required for
I_________ laser pulses
Source: [3]








Pb I
368.35 nm
I


Pb
373.99 nm
I


355 360


365 370
Wavelength (nm)


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


375













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


Advantages


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 pgm
6. Capability of remote analysis in harsh
environments
7. Atomization and excitation are in one
step
8. Capable of simultaneous multi-element
analysis


Disadvantages


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 interference 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








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:








r = [ (X-iX)(Yi_-y) ]/[ I(XiX)2X(yi-Y)2 11/2

where X is the mean of Xi's, and Y is the mean of Y1'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.








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 = [ 7.(R,-R)(S,-S) ]/[ I(Ri-R)2 (S -S)2 ]1/2









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








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.

PCA 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








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.








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.








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.












Linear Correlation


1400
n 1200 -
S 1000 ----
S800 ~
C6 600-
400
S 2001 M 7
0 _-----9-----"-----------i
0 200 400 600 800
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.













Rank Correlation


2 2000-

S1500 -

' 1000
500


, o '0


0 500 1000 1500 2000
Rank Library spectrum


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


I I !










Emission Spectra Class Averages
4500- r - I T
SBedrock


40 -


3500
l


3000 -


2500






1500-


1000


~u0A
4,I *



0-
250 260. .
25O 26O


270 280 20
WaveWlength, nm


300 310 320


Figure 3-3. Laser-induced plasma spectral averages of three classes of phosphate mining
samples: bedrock, matrix, and overburden.


;N i
I' -














Log-Eigenvalue Plot


1010


109

108


107

106

105

i 4


r 4-8 significant PCs
**
[ ~**
*

t **,*


0 5 10 15 20 25 30
PC number

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

















Bedrock Data Scores Plot



3"T-


6040


4000
I "_- --- ~ -





*x 0'* *^






2000',
2


















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

I





'Y *2 y







-"OOi ..


3om 2 C 0 1 1 203



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



























3000-


Bedrock Data All Spectra
ff I ------T


2500





IJJ
2500D



icool

15001-







al i ................ J ._ i .. _.... ..... .... ..... .. .. .. .... .. ... .. A ..
10*1




200 400 600 800 1000 1200 140 16M 180o 2000
Varibe 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.









Log-Eigenvalue Plot


~1


4 PCs significani


w 107
51


-A- *


10 15 20 25 30
PC number


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


10

10

10


0 5







37





,0 Matrix Data Scores Plot




t5i







00



0
.. ..... -








0 C





05. 3 S o








15q




-1 b P #




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













40001



3500-



3000-



2500
a

1 2000

w

1500-







500


Matrix Data All Spectra
I I -- r - -


\ \ ~2


200 400 600 800 1000 1200 1400 o1600 1800 iU
Variable Index

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










Log-Eigenvalue Plot





4-8 significant PCs


I


.* ** * ^



5 10 15 20 25
PC number


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


10


108


10


104
0


30







40





Overburden Data Scores Plot


5000


5000 i


*1.S-1 -5 005 t1 5
Sres foW PC C 1 0

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





41

Overburden Data All Spectra __










,I i


D P I I *. i I I I
r) 200 ,M 800 ROD 1000 1200 1400 1600 1800 2000
Varale Index
Figure 3-12. Laser-induced plasma spectra of 30 overburden samples. X-axis is pixel
number.










Log-Eigenvalue Plot
1010





108
w



106

'' ".......................


0 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.










Table 3-1. Principal components and the amount of variance each includes.
Principal V a Cumulative
Comonent ,Eigenvalues Variance C l ei
Component __________Variance
1 2.04 x 10i'u 59.36% 59.36%
2 1.07 X 10' 31.14% 90.50%
3 1.39 x I09 4.04% 94.54%
4 8.98 x 10' 2.61% 97.15%
5 4.05 x 10" 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 I07 0.17% 99.32%
9 4.77 x 107 0.14% 99.46%
10 4.42 x 107 0.13% 99.59%
11 2.96 x 10' 0.09% 99.67%
12 2.23 x 107 0.06% 99.74%
13 1.59 x 107 0.05% 99.78%
14 1.43 x 107 0.04% 99.83%
15 1.20 x 107 0.03% 99.86%
16 7.05 x 10V 0.02% 99.88%
17 6.03 x 106 0.02% 99.90%
18 4.64 x l06 0.01% 99.91%
19 4.59 x 10 1 0.01% 99.93%
20 3.35 x 106 0.01% 99.94%









































-60OO Oh


-8000-
S Bedrock
95% probably e shon 0 Matix
SOverburden
10000 --_____ 1 -L I
8000 .6C000 -4000 2000 0 2000 4000 6000 8000 1000C
LDF#1


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










Table 3-2. Comparison of chemometric methods.

Linear correlation: 16.7% error
Actual I Predicted
_____Bedrock Matrix Overburden
Bedrock 30 0 0
Matrix 13 16 1
Overburden 0 1 29

Rank correlation: 17.8% error
Actual Predicted
_____Bedrock Matrix Overburden
Bedrock 30 0 0
Matrix 13 16 1
Overburden 0 1 29

Principal component analysis: 1.11% error
Actual Predicted
_____Bedrock Matrix Overburden
Bedrock 30 0 0
Matrix 13 16 1
Overburden 0 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 IMC-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








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.

Ca5(PO4)3OH(s) + 5H2SO4(aq) 3H3PO04(aq) +5CaSO4(s) + H20()

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








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








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 IMC-Agrico. Six samples labeled

Al, A2, A3, B 1, 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.








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 110 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








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 (P205) 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








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.








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 = 0 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.








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 V2" 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 gim)








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 o10X 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








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 pm) 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.








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.








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 11 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.








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








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








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.








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.


















Overburden


--

700


too -



200


10

300

2000


1300

1000


i~
300 -

07


Matrix


a4 Ix


280
W MlMh

Ii


300 320 3#'


Bedrock


2, 2, 280 300 320 349

FigureW4-1. LIPS spectra ofoverburden, matrix, and bedrock.

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


240 260 120 300 320 340


II


3300

300

2300

2000

1300












Table 4-1. Average spectral line intensity ratios.
Sample Si/P Si/Mg Si/Al Si/Fe
Al 16991 11 6 16 5.5 10.4 6.4
A2 48 30 2.4 0.4 15 4.3 3.3 0.9
A3 39 22 0.33 0.07 17 7.7 2.2 0.7
B1 8942 17.38 10.43.3 12.83.7
B2 118 55 2.00.4 16.74 2.20.4
B3 118 40 0.53 0.2 27.3 9 1.86 0.3





























Mrvw
\1iro


scKMtusr


AdJulrahk
Stbr


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


IAPtqP
CAMPuter







67











Matrix spectrum
1600- Mg

1400

1200

1000- Ca

- 800

MSi
Al
400- Fe
200 .


0

-200 E ..
240 260 280 300 320 340

Wavelength (nm)


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






68





Table 4-2. Correlation coefficients for overburden, matrix, and bedrock samples.
Sample O'burden 1 O'burden 2 Matrix 1 Matrix 2 Bedrock I Bedrock 2
O'burden 1 1 0.9568 0.5756 0.7026 0.5056 0.4424
O'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


















Matrix correlation coefficients


~-1.1


0

o
r 0.8 -
0
0.7
S0.6-
0
o 0.5
01 02 Ml M2 BI B2
Sample category

Figure 4-4. Correlation coefficients of matrix sample.






70





Table 4-3. Identification of materials by chemical and LIPS analysis.
Sample %P205 %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.
OvrbrdnP205 MgO
Overburden Negligible Negligible

Matrix >3% < 1.5 %
Bedrock Variable > 5%
Anything in between is a mixture.












1.2


I0.8
S0.8- Matrix

a 0.6 --O'burden
I -Bedrock
S0.4 ----___

0.2


-1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
Distance from focal length (cm)

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














1.2



0.8 U 7 Matrix
S0.6 -.S- O'burden
Bedrock
0.4

U 0.2

0
-1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

Distance from focal length (cm)

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













1.0- %o

0.9 -"
0.80
0 Overburden % *-

0.7 *. *
S0.7- "l *,

0.6

S0.5- Matrix *0'' '

S0.4 Bedrock

0.3
0.2 --


0.1 ,2: '..

0.0- I I I I I
0 50 100 150 200 250
Distance (mm)

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






























Probe Spectrometer Laptop computer


Figure 4-8. Fiber-optic probe system.






















To spectrometer









Trigger button


Handle







From laser head


i ----: To trigger circuit
-w ^.V"'. ' .j! **.
/ ...., ;... ... t :. -" .
k '*.
."


Figure 4-9. Fiber optic LIPS probe.











+9V


52.3 !-


10 k


Con. 3




Con. 1
+9V<---
-9V


1,4,8,15 =
"Cn 1


lka


Con. 4


7,9,10,14


3,11,12,13,16


-9V
For external lamp trigger input into Big Sky Laser
+5VDC, lOOms, into 50 Q


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 22 31 1
(Corr. Coeff.< 0.5)

















Laptop
Computer


Figure 4-11. Experimental apparatus for remote LIPS.















(a) Spectra obtained with LIP-spectrometer
at the distance of 6 m using low- and
high resolution spectrometer channels


V Mg








200 000 400 0O 00 0 O0 *00 00
200.




a) Low Meoluonk d0anm (200-850 nr.
600 m"n graig. 25 un Au)


2io *0o 20.
b) H140 osowJtm dwnIu (230-310 nm.
3600 rWmm' gU 25 mn mi)


(b) Spectra obtained with LIP-spectrometer at
the distance of 6 m


Phosphate Rock #17
(from IMC Agrico)








200 100 400 B; .O 70 8O0 00
WttftoHglk. urn



Wo". ca










*oo. j , .--1 t*. .._^-
z00 000 400 500 000 700 B00 004


Figure 4-12. Remote LIPS spectra.


Mg






IAi i


140 -







81





3.0-


2.5


I2.0-\
C\i



o 1.5-

0
4 1.0-.
E

0.5-


0.0


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.







Trigger


Handle


17 in


Spectrometer

19 in Fiber optic
19mi
Focusing lens
Reflective mirror
Safety interlock switch


BNC connections
On/Off switch
Umbilicals to power/
laser ICE (20 ft)

Laser head


Pierced mirror
Focusing lens


7 in


Figure 4-14. Experimental setup of the field LIPS probe.


3ft











































Figure 4-15. The field LIPS probe.




















4500-

4000-

3500-

3000-

2500-

" 2000-

1500-

1000-

500-


0-


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


Wet


I I I I I I I I I I I I 1 1
200 220 240 260 280 300 320 340 360 380

Wavelength (nm)



















3400
3200
3000
2800-
2600
2400-
2200
2000
1800-
"" 1600-
1400-
1200
1000
600
600: D ry

400- rf---- ---
200 -W"et ,
0
-200-
200 220 240 260 280 300 320 340 360 380

Wavelength (nm)


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






86












1800
1700
1600
1500
1400
1300
1200
1100-
1000
~900-

8~00-
700,
600
500-

24O0- D
200- Wet
100-
0-
-100
200 220 240 260 280 300 320 340 360 380
Wavelength (am)


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





87



Table 4-6. Correlation coefficients for overburden, matrix, and bedrock untreated
~~~______~________sam )les.
___ 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













Bedrock 1 Correlation Coefficients


Bedrock 3 Correlation Coefficients

g 1.1

0.9-
a 0.8
1 0.7.
0.6 -
06
S0.5 ....
01 02 03 M1 M2 M3 B1 B2 B3
Sample




Figure 4-19. Bedrock correlation coefficients.


1.1
"I 1

S 0.9
= 0.8
| 0.7
S0.6
S0.5


01 02 03 M1 M2 M3 81 B2 B3
Sample