USOO6137104A

United States Patent (19) 11 Patent Number: 6,137,104

Webb et al. (45) Date of Patent: Oct. 24, 2000

54 FAST AUTOMATED SPECTRAL FITTING Article by Karstang et al., entitled "Multivariate Prediction
METHOD and Background Correction. Using Local Modeling and
Derivative
erivative SSpectroScopy, published in Anal. Chem.
Ch In
75 Inventors: Christopher Webb, Los Altos, Calif.; 1991, in vol. 63, pp. 767-772. (Month Unknown).
Philip V. Wilson, Mount Waverley, Article by Sasaki et al., entitled "Estimation of component
Australia spectral curves from unknown mixture spectra,” published
73 Assignee: Varian, Inc., Palo Alto, Calif. RCS 1984, in vol. 23, No. 12, pp. 1955-1959.
Article by Tahboub at al., entitled "Evaluation of Multi
21 Appl. No.: 09/158,648 wavelength First- and Second-Derivative Spectra for the
22 Filed: Sep. 22, 1998 Quantitation of Mixtures of Polynuclear Aromatic Hydro
carbons,” published in Anal. Chem. in 1985, in vol. 57, pp.
Related U.S. Application Data 38-41. (Month Unknown).
60 Provisional application No. 60/089,148, Jun. 12, 1998. Article by Bauer et al., entitled "Selectivity and error
(51) Int. Cl. .................................................. G01J 3/50 estimates in multivariate calibration: application to Sequen
52) 250/226; 250/208.1; 356/346 tial ICP-OES,” published in Spectrochimica Acta, in 1991,
58 Field of Search ................................. 250,226, 2081, in vol. 46B, No. 8, pp. 1185–1196. (Month Unknown).
250/576, 461.2, 339.07; 356/346, 300, 349, Article by Thomas et al., entitled "Comparison of Multi
326; 126/665 variate Calibration Methods for Quantitative Spectral
Analysis,” published in Anal. Chem. in 1990, in vol. 62, pp.
56) References Cited 1091–1099. (Month Unknown).
U.S. PATENT DOCUMENTS
Primary Examiner-Que T. Le
5,308.982 5/1994 Ivaldi et al. ...... ... 250/339.01 Attorney, Agent, or Firm-Edward H. Berkowitz
5,596,407 1/1997 Zander et al. .......................... 356/328
5,784,162 7/1998 Cabib et al. ............................ 356/346 57 ABSTRACT
FOREIGN PATENT DOCUMENTS A spectral interval excited by ICP-ES containing expected
O 502 495 9/1992 European Pat. Off.. lines is modeled from wavelength calibration information
O 560 006 9/1993 European Pat. Off.. and from Spectra of model components. A Sample is then
excited and fit to the data with concurrent observation of
OTHER PUBLICATIONS wavelength calibration and a linear least Squares fit obtained
Article by van Veen et al., entitled “Kalman filtering of data with analysis of residuals to determine presence of unex
from Overlapping lines in inductively coupled plasma pected Spectral features which, if present, provide additional
atomic emission spectrometry, published in Spectro model components for an iterative fitting.
chimica Acta in 1990, in vol. 45B, No. 3, pp. 313–328.
(Month Unknown). 8 Claims, 6 Drawing Sheets

Obtain Spectra for Known -25 Obtain Spectra for Unknown

Species Species -50B

Determine Wavelengths of Argon Determine Wavelengths of Argon

wavelength calibration lines 20 wavelength calibration lines - 50A

Refer to relative displacement

of argon wavelength
Calibration lines in model
and analysis -> Shift model
Obtain 1st Gauss by fitting to 80A appropriately -55
each side of peak in turn. Use
parameters derived from
narrower side -30
Perform linear least squares fit H

80s
-35 ANAYSS
Examine residuals for more Y COMPLETE->
peaks. Fit another gauss to H Apply chi-squared USEWEIGHTS
largest. test-sls fit AND STANDARD
close enough? CONCENTRATIONS
CDRMN
UNKNOWN CONCS.

Does last gauss

exceedsoe
pre-set fraction Determine position of largest 70
of all previous? peak in residuals

ls it displaced Add new model component

Mask the from peak of interest
peak in by more than some
consisting of prime model
Store model consisting of all residuals preset multiplier shifted to position of the
gauss parameters, residuals, of sigma? peak in residuals
and argon wavelengths is 45 /
8O 75 85
U.S. Patent Oct. 24, 2000 Sheet 1 of 6 6,137,104

Obtain Spectra for Known

Species

Determine Wavelengths of Argon

wavelength calibration lines

Obtain 1st Gauss by fitting to

each side of peak in turn. Use
parameters derived from
narrower Side

Examine residuals for more

peaks. Fit another gauss to
largest.

Does last gauSS

exceed some
pre-set fraction
of all previous?

Store model consisting of all

gauSS parameters, residuals,
and argon wavelengths

FIG. 1A
U.S. Patent Oct. 24, 2000 Sheet 2 of 6 6,137,104

Obtain Spectra for Unknown

Species -5OB

Determine Wavelengths of Argon

wavelength calibration lines - 50A

Refer to relative displacement FIG. 1B

of argon wavelength
Calibration lines in model
and analysis -> Shift model
appropriately

ANALYSIS
COMPLETE->
Apply chi-squared USE WEIGHTS
test -> is fit AND STANDARD
close enough? CONCENTRATIONS
TO DETERMINE
UNKNOWN CONCS.

Determine position of largest

peak in residuals

ls it displaced Add new model component

Mask the YES from peak of interest consisting of prime model
peak in by more than some shifted to position of the
residuals preset multiplier
of sigma? peak in residuals
8O
75 85
U.S. Patent Oct. 24, 2000 Sheet 3 of 6 6,137,104

1p–6ux1n0–eoq

tsun que/eubs
U.S. Patent Oct. 24, 2000 Sheet 4 of 6 6,137,104

–puno6x1–0eq
|GZ0O9#[(?e6uxl?nedm)
G

tsun que/eu6
0 09
U.S. Patent Oct. 24, 2000 Sheet 5 of 6 6,137,104

G|ZO9
G0 9– 0
ts un que/euf IS
U.S. Patent Oct. 24, 2000 Sheet 6 of 6 6,137,104

sun que/eu6S
0 09
 6,137,104
1 2
FAST AUTOMATED SPECTRAL FITTING SUMMARY OF THE INVENTION
METHOD The invention expresses an expected Spectral interval in
an analytic form and by linear least Squares techniques,
This application claims benefit of Provisional Appl. extracting the coefficients of concentration for the elements
60/089,148 filed Jun. 12, 1998. 5
represented by lines excited during ICP-AES. Plasma gas
FIELD OF THE INVENTION and Standards yield data from which wavelength calibration,
concentration calibration and line shapes are obtained for
The invention relates to the practice of quantitative evalu analytical representation of the Spectral interval of interest.
ation of the components of Spectral distributions and par Wavelength shifts, in the nature of translation of the cali
ticularly to the automated analysis of Spectra. bration are monitored at each Sample data acquisition. Any
shift detected in the calibration causes a corresponding shift
BACKGROUND OF THE INVENTION in model components employed to realize the analytical
representation. Unexpected Spectral features in proximity to
The evaluation of spectral distributions has been highly peaks of interest are detected from the residuals of the fitting
developed over many decades in conjunction with the Vari 15
procedure and the presence thereof leads to an iteration
ous spectroscopies which have evolved. In the context of wherein a model component (peak) is shifted to form an
analytic instruments, the positions of Spectral lines to be additional model component for which the fitting operation
observed is Substantially known a priori and it is the relative is repeated. Such features which are located Sufficiently far
concentration of the elements present which is to be deter from peaks of interest are masked from further analysis.
mined. Positions of expected lines may be perturbed slightly
by instrumental effects or by the Superposition of proximate BRIEF DESCRIPTION OF THE DRAWINGS
lines. At high resolution, line Spectra exhibit lineshapes
which may be studied by fitting techniques to reveal the FIG. 1A shows in schematic form the steps of the model
quantitative presence of partially overlapping lines. building phase of the invention.
In the case of certain Specific modern Spectroscopic 25 FIG. 1B shows in schematic form the steps of the analysis
instruments the data is photo-electronically accumulated at phase of the invention.
discrete positions (pixels) with the result that a narrow peak FIG. 2 is a 2 component fit, without drift, for a B-Fe test
may be defined by only a few datums. Given the discrete Sample.
nature of the photoreceptive pixels, finite resolution and FIG. 3 is a 2 component fit, with actual 0.5 pixel drift, for
dispersion of the instrument and possible instability of the a B-Fe test sample
instrument to drift, peak position and drift may be deter FIG. 4 is a 1 component fit to an Fe sample with 0.5 pixel
minable as fractional pixel quantities. The Spectral evalua drift.
tion technique of the present invention accommodates this FIG. 5 is a 1 component fit to a Fe:B (1000:1) sample with
combination of circumstances.
35 0.5 pixel drift.
In the particular case where atomic emission Spectra are
excited from an inductively coupled plasma, the nature of DETAILED DESCRIPTION OF THE
ICP excitation is such that the density of spectral lines is INVENTION
quite high compared to more Sedate excitation. Thus, inter The functional Structure of the invention may be seen as
ferences are frequently encountered. 40 comprising two phases: building the (expected) spectral
In the prior art, spectral decomposition from ICP-AES has model, and analyzing the observed spectrum.
been discussed by Van Veen et al., Spectrochimica Acta, The model building phase, Schematically illustrated in
v.45B, pp.313 (1990). The method reported there is based FIG. 1A, has the goal of expressing the expected Spectrum
upon a Kalman filter technique with the consequence that as an analytic representation, e.g., as a Sum of gaussians
this prior art procedure is iterative and further requires 45 together with Slowly varying residuals. The invention is not
multiple data acquisitions to examine wavelength shifts. limited to use of a gaussian form (one may employ lorent
This particular work was accomplished with a Scanning type Zian or other appropriate forms), but reference will be made
instrument arranged to yield a far greater density of points to gaussian functions, both as a preferred and a generic
for peak definition than would ordinarily be available in a choice. Model building should be regarded as inclusive of
broad range instrument with discrete detection elements for 50 the Straightforward referencing of Standard wavelengths to
the Simultaneous acquisition of data over a wide spectral the instrument through observation 20 of characteristic
range. Quite apart from the nature of the particular spectro Spectral lines of the plasma gas including the Sample.
Scopic instrument, the Spectral treatment as described, is Typically argon is employed for the purpose of Supporting
very intensive in computation and time required therefore. the plasma and references to argon should be interpreted to
Another approach to spectral decomposition of ICP-AES 55 include any alternate choice for the plasma gas. Selected
data is described in U.S. Pat. No. 5,308.982 to Ivaldi, et al. calibration lines are smoothed during this step 20 (for
This procedure employs a first (and preferably Second) example using a well known procedure Such as the Savitsky
derivative of the Spectrum in process to construct a model Golay Smoothing filter, -see A. Savitsky and M. J. E. Golay,
Spectrum which conforms to the observed spectrum by Analyt. Chem., v. 8, p. 1627, 1964 and J. Steiner, Y. Temonia
adjustment of appropriate parameters, thereby yielding con 60 and J. Deltour, ibid., v.6, p.1906, 1972) and then the three
centration information. Such methods are practiced at the highest amplitude points of a calibration peak are fit to a
price of a significantly diminished Signal to noise parameter. quadratic to extract the peak maximum coordinate of wave
In the present work, ICP-AES is carried out in an instru length. The combination of these two procedures was found
ment which is described in U.S. Pat. No. 5,596,407, incor to achieve excellent accuracy with overall computational
porated by reference herein. This instrument includes an 65 Simplicity. Expressed in pixels of a discrete focal plane
electronic focal plane detector characterized by Sequences of detector, it has been confirmed that this method yields results
discrete pixels arranged in continuous linear arrayS. which agree (typically within much less than 0.1 pixel) with
 6,137,104
3 4
computationally more intense non-linear fitting procedures, used to re-define the wavelength-pixel calibration and the
It should also be noted that this combination of steps, the corrected value is employed with the Stored wavelength
Savitsky-Golay filter effectively fits a polynomial (using 3 dependent spectral representation. The method of perform
points) to the data points. The Subsequent quadradic fitting ing the shift of each model components to re-calculate the
is therefore well matched to this step. Use of a greater model from its component function (gaussian) and residues.
number of points would be largely redundant. It was con The gaussian(or other function) is simply re-computed as
firmed using this procedure for a group of peaks over a wide f"" (x+A)=f'(x) for a shift of A. The residuals are shifted
range of wavelengths that shifts with respect to an argon by the method of linear interpolation whereby the new
calibration line at 419.832 nm could be corrected to better residual function is is evaluated at each required point
than 0.1 pixels. assuming a linear variation of the residual function in the
During the model building phase, one separately cali interval comprising the two adjacent measured (pixel) val
brates the intensity Scale against Spectral lines from Standard CS.
Sources of known concentration by introduction of known The concentration coefficients are extracted in the usual
Standard(s) for spectral acquisition and recording 25. The manner 60 of linear least Squares analysis. Briefly, one
procedure 30 for fitting an observed set of points, constitut 15 represents a measured spectral interval y(x) over a wave
ing a peak, consists of fitting observed points on each length interval X as
(roughly) one half of the peak separately to the Selected
(preferably, gaussian) shape. That is, the data comprising a
peak is treated as comprising two Statistically similar por
tions. Various ways of making this division are acceptable.
The portions may be chosen to exhibit roughly equal area for
example, or a maximum point (if there is one) may be treated
as common to the two portions for fitting purposes. From the where the XS represent basis or model functions and are
Separate fits, there is extracted corresponding width and peak Selected functions of the variable X. These basis functions
position parameters. The peak position corresponding to the 25
may include Such choices as gaussians to represent peaks
narrower width is Selected to represent the model Spectral and polynomial or other appropriate form to represent
component for the respective Spectral feature. This permits background or like continuum. The residuals e(x) express
a Systematic treatment for peaks which may exhibit an the departure of the observed data from the model. One
asymmetry for any of a variety of reasons and minimizes the seeks to obtain the coefficients a. The well known proce
number of gaussians ultimately required to form the model. dure consists of forming the NXM matrix A where N is the
It is advantageous to construct a model including two number of values of the variable X for which measurements
closely overlapping peaks from the same element. An are obtained and M is the number of basis functions
example is the case of thallium near 190 nm. In Such case, employed. The elements of A are the model functions
the larger peak is matched first and the residual are analyzed evaluated at the wavelength shifted pixel locations
40 to determine the presence of additional Spectral features. 35 A=X(x) Eq. 2
The residuals containing the Second peak data are fit Sub The criteria of least Squares is employed to extract the
Sequently. Asymmetry may result from Some instrumental elements of the array, e.g., minimization of the quantity.
influence or from a Standard which contains multiple peaks
in extremely close proximity. The Selected shape (typically, W
Eq. 3
gaussian) together with the residuals now represent the
(possibly asymmetric) model function associated with the
particular peak wavelength.
40
X. t S. ak X: c}
Analysis of residuals 40 concludes with the determination
that the residuals meet Some predetermined criteria for Minimization of the above expression is recognized as
goodness of fit. At the model building phase, iteration is 45 yielding residuals of magnitude Small compared to the
terminated when the fitting of a Secondary or tertiary gauS modeled spectra. In the present invention, the distribution of
sian provides a component having less than Some Selected these residuals is analyzed for Structure of Significance, for
fraction (for example 0.1 in area) compared to the primary example, forming a Statistically significant peak. The dis
gaussian. placement of any Such resulting peak in the residuals is
The Second phase of the procedure, as Schematically 50 considered. If a disturbance in the residuals is sufficiently far
shown in FIG. 1B, is the analysis where each spectral from the peak(s) of interest within the spectral window, this
interval or window, containing spectral features of interest region may be masked out from the analysis. That is, the
excited from an uncontrolled Sample, is observed (together measured data in this region may be excluded from the least
with observation 50 of wavelength calibration lines). Stored Squares fit. Should Such residual peak be Sufficiently proxi
calibration data from the proceeding model building phase 55 mate the peak(s) of original interest, an additional model
are compared with currently observed calibration line(s) in function is assumed for the position of the residual peak and
order to monitor potential instrumental drift at step 55. Such the analysis in either instance is repeated. Distant and
drift may arise, for example, from local heating of the proximate are judged on the Scale of units of Standard
instrument with consequent mechanical distortion or deviation, O. The determination of Significance of residuals
expansion/contraction affecting optical paths. Differences in 60 is determined using chi-Square analysis of the residuals as
position of calibration wavelength(s) on the focal plane of indicated at Step 65. The details of chi-square analysis are
the instrument define a displacement which redefines the well known and may be found in many Standard works, for
positions of the model functions. The Spectral interval is example, W. H. Press, S. A. Teukolsky, W. T. Vetterling, B.
then represented by model functions and background rep P. Flannery, “Numerical Recipes in C, p. 659-661, Cam
resentation (if any) which are explicitly corrected to express 65 bridge University Press, 2" ed. (1992). Following the usual
the drift by a shift in the (fractional) pixel location and the prescription, a goodness of fit parameter is given by mini
concentration coefficients. Observed wavelength drift is mization of the quantity
 6,137,104
6
an interference on that argon line ordinarily used for the
wavelength reference in the Spectral interval of observation,
or Some unusual Source of wavelength drift might occur
which is not uniform across the focal plane. To cope with
this situation a more robust but more complicated embodi
ment is appropriate.
A group of argon lines is observed and the Scatter in the
The value so determined must be compared with a refer indicated shifts monitored. That is, each of these lines is
ence value X of to determine the acceptability of the fit. The compared to its corresponding previous pixel value to estab
quantities O. are values of a weighting parameter which is lish a corresponding shift and the average and Standard
preferably given as the product of two variables which may deviation of these shifts. If the scatter (standard deviation)
be termed RSD and SIG. RSD is preferably obtained by exceeds Some amount, Say 0.1 pixels, we employ an alter
defining a number N of contiguous points forming a native method for deriving the value of the shift to be
window. Each set of Ns contiguous points within the applied to spectral modeling and analysis.
15
window, N>>N, is examined to find the minimum relative
Standard deviation of Such set (for example, each 5 contigu (a) Evaluate the group of argon lines for outliers, and
ous points in a 21 point window interval). RSD, for all reject the one, or Some Small number which are furthest
values of i, is then Set equal to this local minimum value, or removed from the average indicated shift and employ
an appropriate value, Such as 0.02, whichever is larger. those remaining if their Scatter is Sufficiently Small to
These choices are dependent upon the noise present in the establish an adopted value for the shift.
data. If one expects to obtain about 1% accuracy for the Sum Another alternative is to
of 4 pixels, then 2% for individual pixels should suffice.
The quantity SIG is determined Separately for each i and (b) Switch immediately to a “bootstrapping” methodol
is taken as the electronically offset corrected maximum ogy. The technique employed is to Search the focal
Signal level in proximity to the peak of interest, or 25 plane region adjacent to the line of interest. Additional
alternatively, it is taken as the actual signal level at X, information identifying the principal elements present
whichever is larger. Thus in the presence of a significant in the Sample is required. This information is either (i)
interference peak, the criteria of best fit will be governed by Solicited from the user or (ii) derived from some other
the peak of interest rather than the interference peak. intelligent software (the nature of which is outside the
The evaluated quantity X must be compared to a refer Scope of the present work).
ence value. It is common to refer to Standard tabulations, for Next, for the focal plane region adjacent to the peak of
example, W. Mendenhall, Introduction to Probability and interest, each line there detected, is correlated with a com
Statistics, Duxbury Press, or to Numerical Recipes in C, prehensive wavelength database, for example, a wavelength
op.cit. It is also common to estimate the reference value by database containing over 30,000 lines observable with exci
multiplication of a value near unity with the number of 35
tation via inductively coupled plasma. Filters are applied to
degrees of freedom present in the data. In the present work, the database to search only those lines in a small “cell”
the number of degrees of freedom is taken as equal to the Surrounding the line of interest and which belong only to
number of (unmasked) data points less the number of model those elements which have been determined by either (i) or
components (including points used to fit any background). It
is preferred, where the number of model components, apart 40 (ii) to be present. If a peak is located in the region of interest
from background, is one, to use 3 times the number of and which lies within Some tolerance (for example, 8
degrees of freedom as the multiplier. This relaxation of the picometers or a value which is allowed to increase linearly
fitting criteria has been found to yield acceptable fit param with wavelength) of a line in the database, the two are
eters; otherwise, the multiplier is taken as 1.5. The test of the asSociated and an observation error determined. Such drift
residuals against a reference (preferably chi-square derived) 45 measures are obtained for preferably a large number of lines
may then terminate the procedure (65). Properties of the (which may be done conveniently in a line-rich
Standard Samples employed yield absolute measurements in environment). An average drift value is determined from all
accord with known methods. the lines for which associations are possible. Some chance
During the analysis phase, the result of Statistically Sig asSociations can occur which are not correct, but with proper
nificant residuals leads to an iterative Set of StepS as outlined 50 choice of tolerance and good calibration Starting point, this
above for the model building phase. A Single peak in the number will be relatively small and contribute randomly to
residuals is located (70) and its proximity to the major peak the average which is formed. Given a Sufficiently large
is ascertained (75). If the residual peak is sufficiently far number of lines contributing to the average, an accurate
from the peak of interest, there is no “interference” and these measure of drift is obtained.
residuals may be masked out (80) from further analysis. If 55 Illustrative examples of the invention in use are shown in
an interference condition is determined, then the (nearest) the following tables and FIGS. 3 through 6.
model peak is shifted by the indicated wavelength interval to Boron/Iron lines near 249 mm form a well known
match the residual peak position and treated as an additional interference, and data were obtained for evaluation with the
model component to represent the unexpected interference invention. Astrong memory effect is exhibited by boron; this
(85). The least squares fit procedure is then re-entered with 60 was reduced by changing the transfer tubing between the
the new (additional) model component included in the fitting Spray chamber and torch to Teflon lined tygon. It at first
constraints. appeared that the “boron’ peak Still accompanied the pure
Although the method of determining the required wave Fe Solution even when reduced to negligible proportions for
length offset using an argon line is found adequate for the the blank. It was verified from wavelength databases that
present invention in most circumstances, alternative 65 another Fe peak is expected with near the observed intensity
embodiments are possible, and may be desirable in Some nominally only about 1 pm away from the B peak, the
cases. For example, a particular line-rich Sample may cause Somewhat crowded Spectral region is given in Table 1.
 6,137,104
7 8
FIG. 4 shows the same sample, but now with the 0.5 pixel
TABLE 1. shift between models and data (ref. Q2). Models of both B
and Fe were Supplied during the fit and there was a relative
B/Fe lines in the spectral region of interest. shift between models and data of ~0.5 pixels in this case.
el. WL/nm
The fit is slightly less accurate when the models have been
shifted by the 0.5 pixel amount, but the derived concentra
Fe 249.782 tion is identical (see Table 1). Note that if the 0.5 pixel shift
B 249.773 is ignored, then a grOSS error in derived concentration occurs
Fe(weak) 249.772 (>30%).
The one component model-given the near exact overlap
The presence of the Second Fe peak is not a problem for of the second weaker Fe peak-yields results better than
a 2-component fit where the Fe response is modeled but does expected. It may be the case that the overlap is not as exact
represent a Severe test for the unexpected interference pro as Suggested by Standard wavelength databases, Since for the
tocol. 0 and 0.1 lug/ml B Solutions a Second unexpected interfer
Two sets of data-referenced by “P” and “Q' were 15
ence was detected at a slightly larger Separation than the 1
obtained for various pure and mixed B/Fesolutions. The two pm mentioned. The Second peak was not inferred for higher
sets differed in that a simulated wavelength drift was concentrations, however, and this is reflected in the higher B
induced between them by a deliberate adjustment of instru concentrations determined (see Table 3).
ment alignment. Further the difference was chosen to be The 1-component case with Zero B concentration i.e. only
about 0.5 pixels, i.e. as difficult as possible for compensation Fe present, is shown in FIG. 5, where a spectral fit results for
by the calibrating Ar peak. B and Fe models were created the case reference Q1 with 0 tug/ml B concentration. Only
from set P, and data from both sets P & Q processed with the the B model is Supplied during the fit and there was a relative
Same models. The results are Summarized in the tables 2 &
3 for 2-component and 1-component fits respectively. shift between models and data of -0.5 pixels in this case.
This is a particularly exacting test and the Spectrum is
25
accurately interpreted as containing essentially no B. The
TABLE 2 automatic procedures built into the peak fitting procedure
2-component model determine whether the fit is satisfactory at each stage (where
the largest unexpected residual peak has been fitted). The
Fe conc./ppm B conc.fppm
nominal nominal
B conc./ppm
Present invention
0.5 pixel
shift ref.
procedure then either adds a new component based upon a
shifted model component or a spectral region may be
1OO O.1 O.10 O P2 masked from the fit when Such spectral region is too far
1OO O.2 O.21 O P2 removed from the peak of interest to be of Significance. This
1OO 0.5 O.53 O P2
1OO O -0.01 yes O2
test would defeat the prior art procedures where the models
1OO O.1 O.10 yes O2 are automatically shifted according to the data and any
1OO O.2 O.21 yes O2 35
unexpected interference is essentially fatal.
1OO 0.5 O.49 yes O2 FIG. 5 shows the case similar to that in FIG. 4 but in this
instance the sample includes 0.1 lug/ml B. Only the B model
Results of the fitting method are obtained in table 2 as is Supplied during the fit and there was a relative shift
described in text and shown in FIGS. 2 and 3. Reference P2 between models and data of -0.5 pixels.
refers to data Set P processed using 2 model components. A 40
Although the present invention has been described with
2-component fit refers to a fitting procedure where data reference to particular means, materials and embodiments,
obtained from both pure B and Fe solutions are used to from the foregoing descriptions, one skilled in the art can
model the mixed Solution data. ascertain the essential characteristics of the present inven
tion and various changes and modifications may be made to
TABLE 3 45 adapt the various uses and characteristics thereof without
departing from the Spirit and Scope of the present invention
1-component model as set out in the claims which follow.
Fe conc.fppm B conc./ppm B conc./ppm 0.5 pixel What is claimed is:
nominal nominal Present Invention shift ref. 1. A method for treatment of spectra observed on the focal
1OO O O.OO yes O1 50
plane of an optical spectrometer wherein spectra are excited
1OO O.1 O.11 yes O1
by inductively coupled plasma processes, comprising the
1OO O.2 O.24 yes O1 Steps of
1OO 0.5 0.52 yes O1 (a) forming a representation for expected features in a
predetermined spectral interval,
Results of fitting as described in text and shown in FIGS. 55 (b) locating at least one first Selected wavelength corre
5 and 6. A 1-component fit refers to a fitting procedure where sponding to a position on Said focal plane, whereby a
only the B model is provided and the present inventive wavelength reference is established over Said predeter
procedure detects and compensates for Fe as an “unexpected mined spectral interval on Said focal plane,
interference'. (c) observing at least one predetermined spectral feature
Accuracy in most cases was better than 3% for the mixed 60 having a known concentration, whereby a Scale of
Solutions with the 2-component model. Precision was typi concentration is established,
cally-1%. FIG. 3 shows the actual spectral fit in the case P2 (d) representing each said spectral feature by the combi
characterized by 0.2 lig/ml B concentration, no 0.5 pixel nation of a Selected mathematical form dependent upon
shift between models and data and an Fe model employed wavelength and the residuals realized by comparison of
for the fit. Models of both B and Fe are supplied during the 65 Said form and Said observed spectral feature whereby a
fit and there is no relative shift between models and data in model Spectral component is obtained,
this case. (e) analyzing a Sample,
 6,137,104
10
(f) recording the spectra of Said Sample at least within said 3. The method of claim 2 wherein either of said portions
Spectral window and Simultaneously there with observ includes the residual maximum of Said peak.
ing again Said at least one first Selected wavelength, 4. The method of claim 1 wherein said step of locating
(g) determining any shift in location on said focal plane comprises applying a Smoothing filter to data in Said spectral
for said first Selected wavelength as obtained at Step (b) interval and performing a quadratic fit to the three most
compared to step (f) and incorporating any said shift in prominent datums whereby said wavelength is located.
each said model Spectral component by transforming 5. The method of claim 4 wherein said step of locating
Said mathematical form and Said residuals accordingly, further comprises Selecting wavelengths corresponding to
(h) performing a linear least Squares fit of Said recorded prominent Spectral features attributable to a plasma gas of
Spectra to Said model Spectral components whereby the Said inductively coupled plasma.
intensity of spectral features corresponding to Said 6. The method of claim 5 further comprising establishing
model components is established together with the the integrity of Said Selected first wavelength, comprising
distribution of residuals resulting therefrom, locating additional Said wavelength references attributable
(i) analyzing said residuals and determining whether said 15 to Said plasma gas and Subsequently observing each Said
fit meets predetermined criteria for terminating Said wavelength reference spectral features and deriving the
treatment, and in default thereof, corresponding apparent shift in the position thereof, and
(j) inspecting said distribution of residuals to determine determining whether the consistency among Said derived
whether a significant residual maximum is located shifts indicates the presence of an interference affecting at
therein and in response thereto, least one said first and additional wavelength reference
locating the wavelength corresponding to Said residual Spectral feature.
maximum and defining another model component 7. The method of claim 6 wherein the aggregate of Said
located at the wavelength of Said residual maximum derived shifts is treated to obtain the expected Statistical
and repeating steps (h) and (i), or, variance in the aggregate of Said derived shifts, disregarding
where no significant residual maximum is located, 25 any one of Said aggregate deviating from the average thereof
displaying Said intensity and terminating the treat by more than Said expected Statistical variance,
ment. re-determining the average value of Said derived shift and
2. The method of claim 1 wherein said step of represent adopting said re-determined value for said step (g).
ing each said spectral feature comprises treating each Spec 8. The method of claim 1 wherein said step of transform
tral feature forming a peak as comprising two portions of ing comprises shifting Said mathematical form by Said shift
Similar Statistical Significance, Separately fitting the portions determined at Step (g) and linearly interpolating the residuals
to a Selected function, and Selecting that fitting correspond as determined at Step (d) by said shift.
ing to the more narrow of the two fits as an initial repre
Sentation of Said peak. k k k k k