You are on page 1of 10

Spectrochimica Acta Part B 61 (2006) 200 – 209

www.elsevier.com/locate/sab

A semi-quantitative standard-less analysis method for laser-induced


breakdown spectroscopy
Pavel Yaroshchyk a,c , Doug Body b , Richard J.S. Morrison c , Bruce L. Chadwick b,⁎
a
Cooperative Research Centre for Clean Power From Lignite, 8/677 Springvale Rd., Mulgrave, Victoria 3170, Australia
b
Laser Analysis Technologies, PO BOX 879, Bayswater, Victoria 3153, Australia
c
School of Chemistry, Monash University, Victoria 3800, Australia
Received 26 September 2005; accepted 18 January 2006
Available online 28 February 2006

Abstract

We report on recently developed analytical software to model laser-induced breakdown spectroscopy emission spectra and predict sample
composition using a proposed calibration-free algorithm. The model uses a database of atomic emission lines to create a theoretical emission
spectrum for selected elements using defined plasma parameters. The resulting theoretical spectrum is fitted to experimental data obtained from a
laser-induced breakdown spectroscopy instrument comprising of four compact spectrometers that image the plasma emission. Elemental
concentrations are obtained by comparing observed and predicted spectra while varying the plasma temperature and relative elemental
concentrations. The use of the model for analysis of major elements in bauxites, brass and mineral samples as well as the analysis of laboratory air
is demonstrated. For the majority of elements investigated agreement within 25% is achieved between estimated and certified values.
© 2006 Elsevier B.V. All rights reserved.

Keywords: Calibration-free laser-induced breakdown spectroscopy (CF-LIBS); Laser-induced breakdown spectroscopy (LIBS); Quantitative analysis

1. Introduction intensities at these wavelengths can then be used to identify


the sample's composition both qualitatively and quantitatively,
Laser-induced breakdown spectroscopy (LIBS) is an assuming the plasma composition is representative of the target.
emerging analytical technique with a wide range of potential A substantial body of work has been published concerning both
applications. The method involves use of a high-power laser LIBS fundamentals and the application of LIBS to material
beam focused onto the surface of a sample for analysis. If the analysis; recent reviews can be found elsewhere [10–15].
laser power exceeds the threshold value, an optical breakdown In LIBS, a conventional quantitative analysis procedure
occurs. The breakdown threshold depends on the ablation starts with determination of a calibration curve of emission line
wavelength and pulse duration as well as on the analyzed intensity versus elemental concentration, using a series of
media. For instance, for nanosecond laser ablation the calibration standards or certified reference materials. The
threshold could vary from 1.9 × 108 Wcm− 2 for copper [1] up composition of an unknown sample is then found once the
to 1010–1011 Wcm− 2 for aqueous solutions [2,3]. During the emission intensity of the analytical lines has been measured.
optical breakdown a highly ionized gas, i.e. a plasma, is formed. A number of manuscripts dedicated to development of
Electron densities of Nd:YAG laser-induced plasmas are alternative LIBS analysis procedures have been published in
typically in the range of 1016–1020 electrons/cm3 [3–9]. recent years. Among these are reports of the use of chemometric
Heavy particles in LIBS plasmas are mainly represented by methods for the quantitative analysis of noble metals [16,17].
neutral species and both singly charged and doubly charged ions Chemometric techniques such as principal component regres-
that emit radiation at characteristic wavelengths. Spectral sion analysis (PCR) and partial least squares regression analysis
(PLSR) have also been successfully employed for analysis of
spectroscopic and other data [18–20]. Their main benefit to
⁎ Corresponding author. LIBS could be in the indirect analysis of “difficult elements”,
E-mail address: bruce.chadwick@laseranalysis.com (B.L. Chadwick). for example sulfur and chlorine or for inferred measurements of
0584-8547/$ - see front matter © 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.sab.2006.01.004
P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209 201

sample bulk properties (e.g. in coal analysis, determination of theoretical emission spectrum of selected elements that is
ash content, calorific value, etc) by correlating these parameters compared to the experimental spectrum obtained by analyzing
to the concentrations of key elemental species. the plasma emission using four separate compact spectrometers
Several manuscripts that describe the development of covering the spectral region 185–950 nm. We assume that the
standard-less analysis methods have also been published. The chemical elements present in the plasma are in their neutral
calibration-free LIBS (CF-LIBS) procedure described by Ciucci and singly ionized forms. Computations proceed until the
et al. in 1999 [21] demonstrated that within a certain accuracy it temperature that provides the best least squares fit of cal-
is in principle possible to find the composition of a laser- culated to observed intensities is found. Depending on the
induced plasma using computational methods and applying no number and character of selected elements, the calculations
calibration standards. Their method has a number of general take 30 s–10 min on a 3 GHz P4 PC with 1024 MB of RAM.
assumptions including: the plasma composition represents Improvements in the computation speed are expected as the
exactly the composition of the sample; the plasma is in local coding improves and faster PCs become available.
thermal equilibrium (LTE) during the period of observation; and The aim of this research is to create fully automatic software
the plasma is optically thin. Both the plasma temperature and that requires few input parameters, performs the analysis in a
the electron density, which are found from a Boltzmann plot are reasonable timeframe, and is easy to use by the operator. In this
used in the calculations. The CF-LIBS method has later been manuscript we describe the methodology used in this software
extended to the case of optically thick plasmas when non-linear and demonstrate its application to the analysis of a laboratory air
self-absorption effects were taken into account using a recursive sample, to the major components of both brass and bauxite
correction algorithm. A significant improvement in the analysis samples, as well as to some other minerals including andesite,
has been achieved via this approach [22]. The CF-LIBS method dacite, and skarns.
demonstrated good analytical results, however there are several
disadvantages associated with it. In the method, each spectral 2. Theoretical
line used to create a Boltzmann plot is treated equally, making
the selection of the lines critical to the method's accuracy. Line Classical theory gives the following equation for the spectral
interference, a high noise level, and spectral line broadening can line integral intensity (W m− 2) [22,23,26]
all have a strong influence on the model's performance. To Z
8phc Nk gi  
avoid this, a level of user interaction is required that would I ¼a 3 1−e−kðmÞl dm: ð1Þ
make it difficult to create a fully automatic application that deals k Ni gk
with any unknown sample matrix.
Here α is a constant characteristic of the instrument, λ is the
Another approach has been used by Gornushkin et al. in their
transition wavelength (m), h is Planck's constant (J s), Nk, Ni,
modeling of laser-induced plasmas [23–25]. In their later work
are number densities (m− 3) and gk, gi are statistical weights
they solve the inverse task, i.e. they predict the composition of
with subscripts referring to levels k (upper) and i (lower), k(ν)
the samples from the experimental spectra [5]. Their model
is the absorption coefficient (m− 1), and l is the absorption
incorporates the plasma temperature and number densities for
path-length (m).
selected species as input parameters and performs a search for
The frequency dependence of the spectral absorption
the best fit between simulated and experimental spectra. The
coefficient k(ν) [22,23,26] is expressed by
best fit is obtained using a Monte Carlo approach for
Z
a þl e−t dt
2
minimisation of multivariate functions. They obtained good
agreement between experimental and theoretical data, and linear k ð mÞ ¼ k0 ð2Þ
p −l a2 þ ðx−tÞ2
correlations between measured and predicted elemental con-
centrations. However, the method is time-consuming, with where a, k0, and x are defined as follows
several hours needed to perform the full computation restricting
the model's application. ðDmN þ DmL Þ pffiffiffiffiffiffiffi DmL pffiffiffiffiffiffiffi
a¼ ln2c ln2 ð3Þ
Compared with the approach used by Ciucci et al. [21] and DmD DmD
Bulajic et al. [22], our model does not require line identification,
e2 Ni f abs
more intense peaks of the same element are given automatically k0 ¼ 2p3=2 ð4Þ
greater weight and limited spectral line interference is me c b
acceptable. This makes it possible to create a fully automatic
pDmD
analytical software application designed to work with any type b ¼ pffiffiffiffiffiffiffi ð5Þ
of sample matrix. Our method is also significantly less time- ln2
consuming when compared to the model reported by Gornush- 2ðm−v0 Þ pffiffiffiffiffiffiffi
kin et al. [5]. x¼ ln2 : ð6Þ
DmD
Our model is also based on the local thermal equilibrium
(LTE) assumption and takes plasma temperature and plasma Here e is the elementary charge (C), me is the electron mass
electron density into account as well as the characteristics of (kg), fabs is the transition oscillator strength in absorption
atomic emission transitions as main input parameters. The (dimensionless), ΔνN, ΔνD, and ΔνL are the natural, Doppler,
calculation uses an atomic emission lines database to create a and Lorentzian line widths (Hz) respectively, and ν0 is the
202 P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209

frequency of the of the line center (Hz). ΔνL and ΔνD can be compilations [31,32]. The atomic partition function U(T) is
expressed as defined as
   X
2p 1 1 1 U ðT Þ ¼ gj e−Ej =kT ð11Þ
DmL ¼ 2pRT þ 2r ð7Þ j
pkT m M
where the sum is taken over all bound levels. Unlike the
 
m0 8pkT 1=2 pffiffiffiffiffiffiffi transition parameters, information about partition functions
DmD ¼ ln2 ð8Þ for atoms and ions is spread throughout the literature.
c m
Several polynomial approximations for calculation of
where p is the pressure of perturbing species (Pa), R is the partition functions exist, however these are given for limited
universal gas constant (J mol− 1K− 1), k is the Boltzmann temperature intervals (up to 16,000 K) [33,34]. The new
constant (J K− 1), T is the absolute temperature (K), m and M are approach of calculation of partition functions has been
the masses (kg) of the emitting atom and the perturber, and σ is described by Halenka [35–37]. In a hot plasma, bound levels
the collision cross-section (m2). of very high excitation energies move to the continuum due to
interactions with surrounding ions and free electrons. Thus, the
ionisation energy decreases with change in plasma temperature
3. The model
and electron density, the so-called lowering of the ionisation
energy effect, δE. Halenka computed the partition functions
3.1. Spectral line intensity
taking into account all energy levels predicted by quantum
mechanics, including also levels lying above the so-called
Using Eqs. (1)–(6) we obtain the following formula for the
normal ionisation energy. Unfortunately these partition func-
peak intensity profile I, (J m− 2)
tions were tabulated for a limited number of elements only.
" pffiffiffiffiffiffiffi #!
8phc −ðEk Ei Þ ln2 k2 gk Aki N −Ei In the work reported here, partition functions have been
I ¼ a 3 e kT 1−exp − pffiffiffi e kT K ða; xÞ l computed using NIST energy level data [38] from the defining
k 4p p DmD U ðT Þ
Eq. (11). A comparison of the values calculated for Fe I and Fe
ð9Þ II, with values reported by Halenka [36] and by Irwin [33] are
shown in Fig. 1. Irwin's U(T) values are generally lower than
where Aki is the atomic transition probability (s− 1), U(T) is a
those reported by Halenka. Partition functions evaluated for Fe I
partition function of an atom or ion under consideration
using (11) are lower than reported by Halenka but higher than
(dimensionless), N is the total number density (m− 3), Ek and
those of Irwin, while our data for Fe II are lower than given by
Ei are energy of the states k and i (J), and K(a, x) is the Voigt
both authors. In general, however, there is good agreement over
function that is defined by
the temperature range of 1000–10,000 K, when the ionisation
Z energy lowering effect is negligible. At higher temperatures the
kðmÞ a þl e−t dt
2

K ða; xÞ ¼ ¼ ð10Þ agreement is reasonable with Halenka's values reported for


k0 p −l a2 þ ðx−tÞ2
δE = 1 eV.
Several numerical methods for computation of the Voigt
integral (10) are known [27–30]. We used the algorithm 3.2. Spectral line shape
proposed by Humliček [30].
To store values of Aki, g, Ek, Ei, and λ, a database has From Eq. (8) it is easily demonstrated that the typical width of
been created and populated using major spectroscopic Doppler-broadened lines at laser-induced plasma temperatures

Fig. 1. A comparison of partition functions used in this work with values reported in [33] and [36].
P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209 203

is of the order of several picometers. It is clear that the pressure in the plasma. In this work, we find the values of
observed Gaussian line shapes are mainly attributable to the Lorentzian components statistically using the “Golden Section
spectrometer resolution. Therefore for our application ΔνD in Search in One Dimension” algorithm [39]. It is our general
Eq. (9) can be replaced with ΔνI (instrumental line width). assumption that when the fit error is minimized over the whole
The instrumental line width depends on such parameters as the spectral range, the correct values for Lorentzian components are
width of the spectrometer entrance slit, dispersion of the found.
spectrometer, optical aberrations, etc., and does not depend on To accelerate the computations, only a selected number of
the analyte. In order to reduce the number of input parameters the most intensive emission lines (typically 10–20) of the
this value is found from the spectra assuming that the full- lighter elements (H to Al) have been considered to be
width at half-maximum (FWHM) of the narrowest line in each potentially Stark-broadened. The rest of the peaks have been
spectrometer is to the first approximation equal to ΔλI assumed to show instrumental broadening only, and therefore
(between 0.12 and 0.22 nm). to be characterized by a Gaussian line profile. Fig. 2(a–c)
Lorentz broadening is difficult to estimate theoretically as it shows sections of the experimental spectrum of a bauxite
has contributions from natural, resonance, Van der Waals, and sample obtained, and the corresponding model output for
Stark broadening. In order to calculate it from Eq. (7), we need comparison. The Stark-broadened H, He, and Al lines as
to know (a) collisional cross-sections for different atoms, which well as non-Stark-broadened Ca, Ti and Fe, are in evidence
are tabulated only for the most important transitions, and (b) the here.

Fig. 2. Examples of experimental (gray) and simulated spectra (black) for an Australian bauxite sample; in (a, b) line widths are attributable to spectrometer resolution
limitations (Gaussian profile); in (c) line widths are attributable to both spectrometer resolution limitations and Stark broadening (Voigt profile).
204 P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209

3.3. Plasma temperature and the total number density

Plasma temperature is the most important parameter in our


model, and both spectral line intensities and shapes strongly
depend on it. In the works of Gornushkin [5], Ciucci [21], and
Bulajic [22] a Boltzmann plot has been used to estimate the
plasma temperature. Such procedure is conventionally used to
determine the temperature by plotting Ek values (eV) on the
horizontal axis and ln(Iλ / gk · Aki) on the vertical axis. Here Iλ is
the integrated intensity of the corresponding atomic emission
line. The temperature is then estimated from the slope of the
corresponding trendline. In our method the Boltzmann plot is
not used in any way and the “Golden Section Search in One
Dimension” routine [39] is employed to find the optimum
temperature, i.e. the temperature value that provides the best
least squares fit between experimental and theoretical spectra.
The value N = 1017 cm− 3 has been used as an initial number
density value for all species in all measurements in this work.
The same value has been used by Gornushkin et al. in [5]. In
fact, as long as the self-absorption effect is negligible, i.e. there
is a linear relationship between the emitted and registered
radiation, the range of 1015–5 × 1018 cm− 3 could be used
equally as well and produces no difference in the analysis
results. It has been observed that increasing N to 1020 cm− 3
leads to a reduction in the optimized plasma temperature value. Fig. 3. A schematic diagram of the algorithm of the software used in this work.
Further increases in the value of N cause the appearance of self-
absorption that is characterized by the appearance of spectral
lines with “flat-topped” Voigt profiles. form an overdetermined system for the entire spectrum that
has the form
3.4. Intensity calibration 0 1
0 1 eðk Þ
I1 ðk1 Þ I2 ðk1 Þ N In ðk1 Þ 0 1 B 1 C
There are several factors that contribute to non-linearity of B I1 ðk2 Þ I2 ðk2 Þ N In ðk2 Þ C
x1 B eðk2 Þ C
the spectrometer's response. These are grating efficiency, B C B x2 C B C
B Cd B C ¼ B C ð13Þ
B C@ N A B C
detector response, coating response, and the optical fiber @ N A B C
transmission. Each of the four spectrometers used in this work xn @ N A
I1 ðkp Þ I2 ðkp Þ N In ðkp Þ
has been intensity-calibrated prior to the investigations. A eðkp Þ
calibrated deuterium light source has been used to correct the
intensity in the interval 190–400 nm, while a tungsten–halogen where p is the number of pixels, p ≫ n. The solution vector xi
light source was used in the 400–950 nm region. is found using the singular value decomposition (SVD) algo-
rithm [39]. To accelerate the computations, the intensities Ii
3.5. The algorithm are calculated using Gaussian peak shapes for most of the
species. Voigt profiles are used to calculate peak shapes for
A schematic diagram of the algorithm is shown in Fig. 3. the light elements, which are more affected by Stark
Once experimental data are loaded, the intensity calibration is broadening.
performed. If the values for ΔλI are not supplied, the software A subroutine has been implemented that searches for the
estimates them as described above. After selecting the list of Lorentzian component of the Voigt profile line width. This
elements present in the plasma (analyte and atmosphere), the search also allows correcting for spectrometer aberrations that
method executes automatically. could noticeably affect the shape and the width of the lines on
Eq. (9) is applied to construct the profiles of all lines of the edges of the spectra, as in Fig. 2a. Here, the Al I 308.216
selected species within a corresponding wavelength interval and 309.271 nm doublet is distorted and Stark-broadened, and
using the database, which form spectra (Ii). Each pixel e(λk) in the lines are wider compared to the Ti multiplet located in the
the experimental spectra yields a linear equation of the form center of the spectrum. The initial system of equations is
updated after each run giving an improvement to the least
x1 I1 ðkk Þ þ x2 I2 ðkk Þ þ N þ xn In ðkk Þ ¼ eðkk Þ ð12Þ squares fit. As the Lorentzian components are found, the outer
loop continues from the beginning with the next value for the
where n is the number of species and x1…xn are coefficients plasma temperature provided by the Golden Section Search
that are proportional to the species number density. Eq. (12) algorithm [39]. The calculation then continues until the
P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209 205

plasma temperature that provides the best least squares fit is 4.2. Samples
found.
Concentrations Ci of the analyzed species could be found A number of different sample matrices have been analyzed
from during this study. These include:
1 X X
xi ¼ Ci ¼ 1 ð14Þ 1. A set of Australian bauxites including a British Chemical
Fexp i i Standard bauxite (395).
2. A set of Brammer brass standards (314, 655, 360, 482, 675,
where Fexp is a constant experimental parameter that takes
863 and 630).
into account the optical efficiency of the collection system
3. A set of pressed pellets made of Tasmanian mineral samples
as well the plasma density and volume [21]. The summa-
including andesites, dacites, and skarns.
tion is performed over all species excluding the atmospheric
4. Laboratory air.
elements.
5. Analytical results
4. Experimental
5.1. Analysis of the bauxite samples
4.1. Experimental setup
Spectra have been collected in a He atmosphere using a laser
A commercially available LIBS instrument 1000 HR (Laser energy of 150 mJ/pulse at an acquisition gate delay of 1 μs. The
Analysis Technologies, Australia) has been used in our helium flow to the sample chamber was set to approximately 3 l/
experiments [40]. It is equipped with four separate gated min. The use of buffer gas ensured that any observed oxygen
Czerny–Turner spectrometers providing spectral coverage from emission delivers from the analyzed sample and not from the
190 to 950 nm, and a Q-switched Nd:YAG laser. The ambient atmosphere. The instrument has a translation stage that
fundamental output in the range of 90–150 mJ per pulse has moves the sample during each run to ensure that a fresh surface
been used in the present study. 0.5–1 μs gate delays have been is available for each laser shot. Each full spectrum consisted of
used in the reported investigations. four single spectra acquired by four spectrometers. An average
A NIST-traceable deuterium tungsten–halogen calibration run consisted of 50 laser shots and 5 replicate measurements
light source (Mikropack) has been used to perform intensity were taken for each sample.
calibration for each of the four spectrometers in the Spectrolaser A comparison of a standard-less analysis of Al, Fe, Ti, and Si
instrument. in the set of bauxites against the ICP-OES values is shown in

Fig. 4. A comparison of the ICP-OES measured composition of a series of bauxite samples (ICP-OES, white bars) with calibration-free LIBS analysis results (CF-
LIBS, black bars), and with matrix-corrected results using kcorr = 0.13 (CF-LIBS-corr, shaded bars). The sixth sample is a British Chemical Standard bauxite (395).
Error bars represent ± standard deviation.
206 P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209

Fig. 5. Correlation between certified concentrations of Al, Fe, Ti, and Si in bauxite samples (horizontal axis) against calibration-free analysis results (vertical axis).

Figs. 4 and 5). The balance is oxygen with traces of Ca, Mg, Na, for most of the analyzed samples. In contrast, our values for Ti
and K. Concentrations reported here were obtained using an highly overestimate the true values across the whole concen-
average plasma temperature, bTN = 15 100 ± 500 K. Fig. 4 tration range, which could possibly be explained by the
suggests that agreement for Al, Fe, and Si is within 25–30% uncertainties of Aki values.

Fig. 6. A comparison of the certified composition of a series of brass samples (Cert, white bars) with calibration-free LIBS analysis results (CF-LIBS, black bars).
P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209 207

has been found to be within 10–20% for most of the samples


analyzed.
Additional results are also presented in the lower part of Fig.
6 pertaining to some of the minor elements in three of the brass
standards investigated. Agreement with certified values is still
generally quite reasonable. Where correlations are less
satisfactory, this can likely be explained by spectral interfer-
ences and uncertainty of the intensity calibration, which will
have a greater influence for weak emission from minor
components.

5.3. Analysis of the laboratory air


Fig. 7. Results of the CF-LIBS quantitative analysis of laboratory air.
As carried out in the work of Ciucci et al. [21], analysis of an
atmospheric air sample is a convenient way to validate the
The trends reported for the bauxite samples in Fig. 5 clearly method. A laser energy of 150 mJ/pulse has been used in the
demonstrate linearity between known and predicted values, experiment with an acquisition delay of 0.5 μs. Each spectrum
with R2 in the range of 0.836–0.973 for all four elements. has been an average of 50 laser shots, 5 replicas were taken for
Results for analytes whose CF-LIBS versus ICP-OES plots the analysis. Fig. 7 exhibits the results of analysis obtained at
have a slope markedly different to one can be improved by the bTN = 6700 ± 100 K. The composition of dry air is ap-
determining a single correction coefficient that effectively proximately 75.5% of nitrogen, approximately 23% of oxygen
serves as a “one-point calibration”. In the case of Ti, as shown in and up to 1.2% of argon (which has not been observed in the
Fig. 4 a correction factor (fitting parameter) kcorr = 0.13 greatly spectra). In a typical indoor atmosphere, H that comes from
improves the agreement between the LIBS results and the H2O and C from CO2 have fractions of approximately 0.1% and
certified ones. 0.01% respectively.

5.2. Analysis of brass samples 5.4. Minerals

For the brass investigations experimental data have been Silicate minerals including andesites, dacites and skarns have
obtained using a laser energy of 90 mJ/pulse and a gate of 1 μs. been analyzed. Andesites and dacites are igneous rocks made up
No buffer gas has been used in this case. A lower laser energy of andesine, pyroxene and biotite and quartz, plagioclase, quartz
was necessary in these experiments to avoid saturation of the and pyroxene, respectively. Skarns are usually dominated by
CCD detectors. Again, each total spectrum consisted of four garnet and pyroxene. Analysis of a small set of these pressed
single spectra acquired by four spectrometers. Each run was an mineral samples has been conducted using a laser energy of
average of 50 laser shots and 3 replicates were taken for each 120 mJ/pulse and a 1 μs gate delay. A He flow of 3 l/min has
sample. Results presented in Fig. 6 have been obtained using been used in the experiment to ensure no air remains in the
average value for the plasma temperature of bTN = 11 000 sample chamber and all observed oxygen emission results
± 1000 K. The agreement for the major components Cu and Zn from the sample itself. Fig. 8 demonstrates the comparison of

Fig. 8. A comparison of the XRF-determined composition of a series of mineral samples (andesites, dacites, and skarns) (XRF, white bars) with calibration-free LIBS
analysis results (CF-LIBS, black bars).
208 P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209

Fig. 9. Correlation between XRF-determined concentrations of Si, Fe, Al, Na and Ti in mineral samples (horizontal axis) against calibration-free analysis results
(vertical axis).

CF-LIBS data obtained at bTN = 14,900 ± 1500 K with XRF Acknowledgements


analysis of the major components. Not analyzed include
minor components (Ca, Cu, Mg, and Ti, b4% total). The The authors gratefully acknowledge financial and other
balance is O. The agreement for Si and Al is within 10–15%, support received from the Cooperative Research Centre for
while for Fe, Na, and K it is within 25–50%. Again, we Clean Power From Lignite, which is established and supported
consider these results to be quite promising as all trends of the under the Australian Government's Cooperative Research
CF-LIBS vs. XRF analysis are linear (Fig. 9) indicating that a Centres Program.
simple linear correction factor can be used to further optimize
the method. References

6. Conclusions and future work [1] Zh. Chen, A. Bogaerts, Laser ablation of Cu and plume expansion into 1
atm ambient gas, J. Phys., B 97 (2005) 063305.
The aim of this work has been to develop a fully automatic [2] J. Noack, A. Vogel, Laser-induced plasma formation in water at
nanosecond to femtosecond time scales: calculation of thresholds,
method for standard-less LIBS analysis. A calibration-free
absorption coefficients, and energy density, IEEE J. Quantum Electron.
algorithm has been developed and implemented. In contrast to 35 (1999) 1156–1167.
previous reports, peak identification is not a necessary [3] P. Yaroshchyk, R.J.S. Morrison, D. Body, B.L. Chadwick, Theoretical
requirement and the Boltzmann plot is not used in any modelling of optimal focusing conditions using laser-induced breakdown
calculations. The simultaneous analysis of the UV-IR spectral spectroscopy in liquid jets, Appl. Spectrosc. 58 (2004) 1353–1359.
[4] M. Sabsabi, P. Cielo, Quantitative analysis of aluminum alloys by laser-
region allows rapid multi-element analysis.
induced breakdown spectroscopy and plasma characterization, Appl.
Preliminary experimental verification has been conducted Spectrosc. 49 (1995) 499–507.
using three sets of samples with quite complicated sample [5] I.B. Gornushkin, A.Ya. Kazakov, N. Omenetto, B.W. Smith, J.D.
matrices in addiction to an atmospheric air. Promising Winefordner, Experimental verification of a radiative model of laser
agreement has been demonstrated for most of the elements in induced plasma expansion into vacuum, Spectrochim. Acta Part B 60
(2005) 215–230.
these experiments. At most a linear correction factor may be
[6] B. Le Drogoff, J. Margot, M. Chaker, M. Sabsabi, O. Barthélemy, T.
used to further improve the results. W. Johnston, S. Laville, F. Vidal, Y.v. Kaenel, Temporal characteriza-
The possibility of using selected number of “trusted” strong tion of femtosecond laser pulses induced plasma for spectrochemical
lines of each element in contrast to the current method where all analysis of aluminium alloys, Spectrochim. Acta Part B 56 (2001)
peaks are taken into account could be considered. 987–1002.
[7] F. Colao, V. Lazic, R. Fantoni, S. Pershin, A comparison of single and
In the longer term the model will be tested using a wider
double pulse laser-induced breakdown spectroscopy of aluminum samples,
range of sample matrices including ores, coal, and alloys, and Spectrochim. Acta Part B 57 (2002) 1167–1179.
the method will be extended so that self-absorption effects in the [8] B. Charfi, M.A. Harith, Panoramic laser-induced breakdown spectrometry
plasma are fully taken into account. of water, Spectrochim. Acta Part B 57 (2002) 1141–1153.
P. Yaroshchyk et al. / Spectrochimica Acta Part B 61 (2006) 200–209 209

[9] M. Ying, Y. Xia, Y. Sun, Q. Lu, M. Zhao, X. Liu, Study of the plasma [23] I.B. Gornushkin, J.M. Anzano, L.A. King, B.W. Smith, N. Omenetto, J.D.
produced from laser ablation of a KTP crystal, Appl. Surf. Sci. 207 (2003) Winefordner, Curve of growth methodology applied to laser-induced
227–235. breakdown spectroscopy, Spectrochim. Acta Part B 54 (1999) 491–503.
[10] D.A. Rusak, B.C. Castle, B.W. Smith, J.D. Winefordner, Fundamentals [24] I.B. Gornushkin, L.A. King, B.W. Smith, N. Omenetto, J.D. Winefordner,
and applications of laser-induced breakdown spectroscopy, Crit. Rev. Line broadening mechanisms in the low-pressure laser-induced plasma,
Anal. Chem. 27 (1997) 257–290. Spectrochim. Acta Part B 54 (1999) 1207–1217.
[11] I. Ahmed, B.J. Goddard, An overview of laser induced breakdown [25] I.B. Gornushkin, A.Ya. Kazakov, N. Omenetto, B.W. Smith, J.D.
spectroscopy, J. Fiz. Malays. 14 (1993) 43–54. Winefordner, Radiation dynamics of post-breakdown laser induced
[12] L.J. Radziemski, Review of selected analytical applications of laser plasma, Spectrochim. Acta Part B 59 (2004) 401–418.
plasmas and laser ablation, Microchem. J. 50 (1994) 218–234. [26] R. Mavrodineanu, H. Boiteux, Flame Spectroscopy, John Wiley & Sons,
[13] L.J. Radziemsky, From laser to LIBS, the path of technology development, Inc., NewYork, 1965.
Spectrochim. Acta Part B 57 (2002) 1109–1113. [27] W.J. Thompson, Numerous neat algorithms for the Voigt profile function,
[14] K. Song, Y.-I. Lee, J. Sneddon, Applications of laser induced breakdown Comput. Phys. 7 (1993) 627–631.
spectrometry, Appl. Spectrosc. Rev. 32 (1997) 183–235. [28] M. Kuntz, A new implementation of the Humlicek algorithm for the
[15] E. Tognoni, V. Palleschi, M. Corsi, G. Cristoforetti, Quantitative micro- calculation of the Voigt profile function, J. Quant. Spectrosc. Radiat.
analysis by laser-induced breakdown spectroscopy: a review of the Transfer 57 (1997) 819–824.
experimental approaches, Spectrochim. Acta Part B 57 (2002) 1115–1130. [29] F. Schreier, The Voigt and complex error function: a comparison of
[16] J. Amador-Hernandez, L.E. Garæia-Ayuso, J.M. Fernandez-Romero, M.D. computational methods, J. Quant. Spectrosc. Radiat. Transfer 48 (1992)
L.d. Castro, Partial least squares regression for problem solving in precious 743–762.
metal analysis by laser induced breakdown spectrometry, J. Anal. At. [30] J. Humlièek, Optimized computation of the Voigt and complex probability
Spectrom. 15 (2000) 587–593. functions, J. Quant. Spectrosc. Radiat. Transfer 27 (1982) 427–444.
[17] L.E. Garćia-Ayuso, J. Amador-Hernandez, J.M. Fernandez-Romero, M.D. [31] http://physics.nist.gov/PhysRefData/ASD/lines_form.html.
L.d. Castro, Characterization of jewellery products by laser-induced [32] http://cfa-www.harvard.edu/amdata/ampdata/kurucz23/sekur.html.
breakdown spectroscopy, Anal. Chim. Acta 457 (2–2) (2002) 247–256. [33] A.W. Irwin, Polynomial partition function approximations of 344 atomic
[18] P. Geladi, Chemometrics in spectroscopy. Part 1. Classical chemometrics, and molecular species, Astrophys. J., Suppl. Ser. 45 (1981) 621–633.
Spectrochim. Acta Part B 58 (2003) 767–782. [34] L. De Galan, R. Smith, J.D. Winefordner, The electronic partition
[19] R. DiFoggio, Guidelines for applying chemometrics to spectra: feasibility functions of atoms and ions between 1500 and 7000 K, Spectrochim. Acta
and error propagation, Appl. Spectrosc. 54 (2000) 94A–113A. Part B 23 (1968) 521–525.
[20] P.A. Aguilera, A. Garrido Frenich, H. Castro, J.L.M. Vidal, PLS and PCR [35] http://draco.uni.opole.pl/~halenka/apf_hal.htm.
methods in the assessment of coastal water quality, Environ. Monit. [36] J. Halenka, B. Grabowski, Atomic partition functions for iron, Astron.
Assess. 62 (2000) 193–204. Astrophys., Suppl. Ser. 57 (1984) 43–49.
[21] A. Ciucci, M. Corsi, V. Palleschi, S. Rastelli, A. Salvetti, E. Tognoni, New [37] J. Halenka, J. Madej, K. Langer, A. Mamok, Tables of the partition
procedure for quantitative elemental analysis by laser-induced plasma functions for nickel, Ni I–Ni X, Acta Astron. 53 (2001) 347–353.
spectroscopy, Appl. Spectrosc. 53 (1999) 960–964. [38] http://physics.nist.gov/PhysRefData/ASD/levels_form.html.
[22] D. Bulajic, M. Corsi, G. Cristoforetti, S. Legnalioni, V. Palleschi, A. [39] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical
Salvetti, E. Tognoni, A procedure for correcting self-absorption in Recipes in C++, Cambridge University Press, Cambridge, 2002.
calibration-free laser-induced breakdown spectroscopy, Spectrochim. [40] http://www.laseranalysis.com.
Acta Part B 57 (2002) 339–353.

You might also like