See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/287491219

Uncertainty estimation for SDAR-OES internal standard method

Conference Paper · January 2009

CITATIONS READS

0 85

11 authors, including:

Ion Pencea Mihai Tarcolea

Polytechnic University of Bucharest Polytechnic University of Bucharest
55 PUBLICATIONS   73 CITATIONS    53 PUBLICATIONS   64 CITATIONS   

SEE PROFILE SEE PROFILE

Catalin Sfat
Polytechnica University of Bucharest
32 PUBLICATIONS   17 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Improving the characteristics of the automotive components by changing the surfaces of the metal parts View project

ECO-NANOTEHNOLOGII DE DEPOLUAREA APELOR ȘI VALORIFICAREA DEȘEURILOR (ECO-NANO-ECO) View project

All content following this page was uploaded by Catalin Sfat on 20 December 2015.

The user has requested enhancement of the downloaded file.

 Uncertainty estimation for SDAR-OES internal
standard method
Ion Pencea, Assoc. Professor, Catalin Eugen Sfat, Lecturer,
Research Center BIOMAT Research Center BIOMAT
University „POLITEHNICA” of Bucharest University „POLITEHNICA” of Bucharest
Romania Romania
ipencea@yahoo.com sfateugen@yahoo.com

Iulian Virgil Calota, Eng.,

Mihai Tarcolea, Professor, Romanian Association for Accreditation – RENAR
Research Center BIOMAT Bucharest, Romania
University „POLITEHNICA” of Bucharest
Romania Sorin Parvu, PhD Candidate,
Research Center BIOMAT
University „POLITEHNICA” of Bucharest
Mariana Pencea, Teacher, Romania
”Viaceslav Harnaj” Lyceum, Bucharest,
Romania Cosmin Cotrut, PhD Candidate,
Research Center BIOMAT
University „POLITEHNICA” of Bucharest
Romania

Abstract: Spark discharge in argon optical emission spectrometry the single channel result of the analytical element. This is why
(SDAR-OES) is the most used technique for elemental analysis of often intensity ratios instead of raw intensities are used to
metallic alloys due to its highest efficiency/cost ratio. In this field calibrate the channels [3].
the SR EN 13005/2005 standard and other European documents as The relation between vaporization, excitation, intensity yield
EA 4/16- EA guidelines on the expression of uncertainty in
and measured variable are known only qualitatively, which is
quantitative testing, EURACHEM /CITAC Guide CG 4,
Quantifying Uncertainty in Analytical Measurement (second why spectrometers need to be calibrated. There is no
edition) 2000, etc. require that the uncertainty estimation to be done calibration curve theory which can be used for entire operating
on the fundamental basics as much as possible. In this sense, the practice. The most used method for OES spectrochemical
paper addresses mainly an advanced theoretical model for analysis is the internal standard method (ISM). The ISM
estimation of SDAR-OES internal standard method contribution to assumes a polynomial relationship between concentration ratio
the spectrochemical uncertainty budget. In this regard, the authors and the intensity ratio of the spectral lines emitted by the
have introduced new ways of estimation of the standard deviations alloying elements in the sample and the internal standard of
of the spectral line intensity ratios, mass concentration ratios and the form [1, 3]:
finally were given a route of computation of compound standard 2 3 4
uncertainties for dozed concentrations. Ci I I  I  I 
= a0 + a1 i + a2 i  + a3 i  + a4 i  (1)
Keywords: uncertainty, standard deviations, SDAR-OES, Cs Is  Is   Is   Is 
spectrochemical analysis, internal standard method

Introduction
Spark discharge in argon optical emission spectrometry
(SDAR-OES) is the most used technique for elemental
analysis of metallic alloys due to its highest efficiency ratio
[1-7]. The working principle of SDAR-OES analysis is the
measurement of intensity of characteristic spectral lines
emitted by atoms during sparking.
Generally, the intensity ratio of the analysis channel and a
suitable reference channel shows a better reproducibility than
where: Ci – mass concentration of the element “i”; Cs – mass
concentration of the internal standard which is, generally, the
matrix concentration, i.e. base concentration; Ii – intensity of
analytical line of the i-th alloying element; Is – intensity of the
analytical line of the chosen internal standard; ak – coefficients
determined by the least squares regression method, k=1…4.
Note 1: The a2, a3 and a4 coefficients are frequently
disregarded.
Spectrometrists usually adjust the spectrometer parameters
through linearization of Eq. (1), i.e. disregarding the a2, a3 and
a4 coefficients.
The ISM is applied with the hypothesis that all alloying
elements in the sample are dozed with the exception of trace
elements, whose concentration do not exceed 0.5(%) [1, 3]. In
this sense, the trace elements are disregarded and the sum of
the dozed elements is considered 100(%), respectively:
n
On the other hand, from the statistical point of view, the
independent variable for a least square regression of a set {xi,
yi}, i= 1, n , data should be the variable with the smaller
relative standard deviation (RSD). Thus, for a low alloyed
Certified Reference Material (CRM) the RSD’s of the yi
quotient are of the order of 1…2 % (see Table 1) while the
RSD of the xi, may be of the order of 2…10 % as is shown in
eq. (4)[1, 2, 4].

∑ C +C
i =1
i s = 100(%) (2)
RSDxi =
1 1
+ ⋅100(%) = 10−3 +10−5 ⋅100 > 3.0(%) (4)
IS IE
where: n – is the number of the alloying elements.
When the uncertainties of the concentrations a critical for the
The Eq. (2) is equivalent with: mechanical or chemical properties of a grade and should be
n
Ci 100 kept as low as possible then the RSD’s could be minimized by
∑C +1 =
Cs
(3) increasing IS and IE. Thus, for a modern spectrometer which
i =1 s provides
IS > 5·105 cps and IE >2·104 cps the RSDX is bring down to
about 1%. Nevertheless, when CE ≤0.1% the IE<103 and
In the case of routine analysis, on the abscissa is allocated to RSDXE >3%.
Ii Comparing the RSD’s for xi and yi (Table1) it results that
the ratios of intensities xi =
while the ordinate to
Is  Ii 
choosing   as independent variable is consistent when IS
C C  Is 
yi = i quotients (fig.1) [3]. Thus, the i quotients are is greater but, more important, when IE exceeds 104 cps.
Cs Cs
estimated by the yi which are calculated values on the base of
calibration method.
2. Concentration computation for Internal Standard
Method
Suppose that all yi have been estimated on the base of
functional dependence on xi e.g. on the base of calibration
curves then, the CS is easily calculated using the following
relation:

100
CS = n
(5)
1 + ∑ yi
i =1

Once the Cs is determined than, the concentrations of the

alloying elements can be easily calculated as:
n
Ci = Cs ⋅ yi = 100 yi (1+ ∑ y j ) = Ci ( y1,..., yi ,..., yn ) (6)
j =1

Fig. 1. An example of calibration curve with intensity ratios along x-axis and Cj
concentration ratios along y-axis [3]. where: yj – the estimate of on the base of its calibration
Cs
Fig.1 shows the intensity rations plotted along the x-axis and curve for element i.
the concentrations quotients along the y-axis. This might be
somehow backwards at first glance, as the concentration It is generally accepted that yj, j= 1, n , are independent
quotients would be as the “known” variable. But, for routine indirect variables and they are affected only by random errors
analyses the intensity ratios are known and the concentration [1, 2, 8].
quotients are to be determined.
Table 1. The main data of a CRM for low alloyed steel grade
Element C Si S P Mn Ni Cr Mo Cu Co Sn Al W Ti V Fe
C, % mass 0,257 0,498 0,131 0,074 0,533 0,38 0,348 0,242 0,162 0,031 0,124 0,09 0,14 0,246 0,042 96.7
U (95%) 0,005 0,01 0,005 0,003 0,011 0,007 0,005 0,006 0,004 0,002 0,006 0,004 0,012 0,006 0,002 0.2
RSDc (%) 0,97 1,00 1,91 2,03 1,03 0,92 0,72 1,24 1,23 3,23 2,42 2,22 4,29 1,22 2,38 0.22
2
C  n

3. Estimation of type B standard uncertainties of the

2
RSD =  S 
CS
 100 
∑S
j =1
2
yi (10)

concentrations for ISM

On the base of the above hypotheses and on the error
propagation law, the standard deviation of the concentration
If the first term in eq. (8) is replaced by RSDCS in eq. (10)
of the element i is expressed as: than the RSDCi get a more consistent expression,
n
respectively:
10 [(1+ ∑yj ) − yi ] ⋅ S
4 2 2
yi
104 yi2 n
  100− Ci  2  CS  2  
S 2
= ⋅ ∑S 2
+ j=1
2
Ci n yj n RSDCi = RSDCS + 
2
 −   * RSDy2i 
(1+ ∑yj )4 j=1
j≠1 (1+ ∑yj )4   100   100  
j=1 j=1
(11)
 n 2 
 ∑Syj  The eq. (10) reveals that the Achilles’ heel of ISM is the
2 yi 2 Syi 
j =1 2
j ≠1 summation of dispersions for each dozed concentration and
= Ci  n
+ (1− n ) ⋅ 2  (7) the critical value of internal standard RSD’s determines the
(1+ ∑yi )2 yi 
1+ ∑yi all other RSD’s as it is shown in Fig. 2. When RSD’s is 1%
 i=1 i=1  than RSD’s of the alloying elements are quite similar with
  calculated ones from the CRM’s Certificate but when RSDS
increase to 6% than the whole RSD’s increase beyond 6%.
Ci2  2 n 2 S2 
= 4 Cs ⋅ ∑Syj + (100 − Ci ) 2 
2 yi From the SR EN 13005:2005 point of view the SCS and SCi
10  j=1 yi  represent the standard uncertainties of type B that are
compound uncertainties [9-:11]
The eq. (7) can be expressed as:
1
 2 2 n 2 2 2
C   100 − Ci  S 
RSDCi (%) =  s  ⋅ ∑Sy2
yi
+ Elemental concentrations and their RSDs for a MBH MRC
 100  j=1 i  *100
 100  y 
2

 j ≠1  i
8,00 W
(8) Co
7,00 S P Sn Al V
C Si Mn Ni Cr Mo Cu Ti
where: RSDCi is the relative standard deviation per cent of 6,00
the i-th element concentration. W RSDc
5,00
2 C(%)
C (% ); R S D (% )

The expression of RSD Ci could be put in a more significant Co

4,00 RSD yi ; ys=1
Sn V
form if one takes into account the expression of the standard S P Al
RSDyi; ys=2
deviation of the inner standard that is: 3,00 C Si Mn Ni Mo Cu Ti
Cr RSD yi; ys=6
2,00

104
CS4 n 2 n

4 ∑
S = 2
Cs S = 4 ∑Syj 2
yj
1,00

 n
 j =1 10 j =1 (9)

 ∑ i
0,00

1 + y Elements
 i=1 
Fig. 2. CRM’s concentrations and their RSD’s depending on internal
standard RSD (%)
The relative standard deviation of the inner standard
concentration is:
4. Contribution estimation of the calibration process to x' i ,k = Ai y ik + Bi (17)
the spectrochemical uncertainty budget
According to LSRLM and to error propagation law the
The calibration practice consist in sparking of dozen of standard deviation of Ai is [13, 15]:
CRM’s for each calibration curve, i.e. obtaining the data
S O2i
(yik, xik), k = 1, N i for element i. S A2i = (18)
Vi 2
In this paper we will consider the case of linear relationship where:
between yi and xi for all dozed elements because this is the 2
Ni
case of spectrochemical routine analyses.
Applying the least square linear regression method (
Vi = ∑ y ' ik − y '
i ) (19)
k =1
(LSLRM) to the (yik, xik), k=1…Ni, data one obtains a
relationship of the form: y'ik - the “known” y ik of the k-th CRM used for
xi= Ai·yi +Bi (12)
where: Ai and Bi are well know regression coefficients [12 - calibration; yi' - the average value of the y 'ik , k =
-15]:
Ni
1, N i .
∑( x
k =1
ik − xi )( y ik − yi ) The derived expression of the standard deviation of a xmi
which was determined by interpolation using the calibration
Ai = Ni
(13) curve and the measured value yim is:
∑( y
2
ik − yi )  1 y im − y 'i ( )2

k =1
S 2m
xi = S ⋅ +
2
oi

(20)
where: Ni – the number of CRM’s used for calibration; y ik  N i Vi 2 
  Ci According eq. (20) and LSLR theory the derived expression
– the known quotient 
 of the k-th CRM; xik – the of the standard deviation of Bi is:
  Cs  1 2

2 
2
y'i 
measured intensity ratio associated to y ik ; yi - the S =S ⋅ +
Bi
 Ni
oi
Vi 2  (21)
average of the calibration “known” values y ik , i.e. :  
Ni
On the base of the above derived relations there were
1 deduced the expression of the standard deviation of the yim
yi =
Ni
∑y
k =1
ik (14)
as:
S oi2 2
xi - average of the xik measured values, i.e. S yi2 m = 2 2
Ai Vi
[ (
y i2 m + y im − y ' i )
2
+2 ~
yi
2
]+ N2SA oi
2
Ni i i
1
xi =
Ni
∑x
k =1
ik (15) (22)

As could be seen from Eq. (7), Eq. (9) and Eq. (11), the
calibration is among the most significant contributor to the
uncertainty budget of the ISM. In this sense, the estimation
of the standard deviation S yi of yi, i = 1, n is unavoidable.
The habit of spectrometrist to use the functional eq.
yi = a i xi + bi for routine analysis complicates the S y2i
estimation while the calibration is done backwards, e.g.
x' i ,k = Ai y ik + Bi . In this respect, the only way to estimate
S y2i is to consider the calibration curve residue S O2i as
measure of the dispersion of variable Xi [15], respectively:
Ni 2 a surprisingly results.
S O2i = ∑ (x
k =1
ik − x' ik ) / Ni − 2
Note 2. The quantity yim is identical with yi used in eq. (5)
and specifies that yim was estimated ones from measured
data.
The Eq. (22) shows that the standard deviation of measured
yim (yi) quotient depends on its value, but mainly on the
calibration parameters as the number of CRM used for
calibration Ni, on distribution of yi’ into analytical range
which is estimated by Vi and on the slope of regression line
(Ai).
As it is generally accepted the minimum value of S2yi
m
is reached when yi is closed to <yi’> [10, 15], but form eq.
(22) it results that S2yi is minimum for y m = y% i 2 that is
(16) From our derivation of the standard deviation S y m [see eq.
2
i

(22)] it results that it could be drawn down to an accepted

where: xik , k = 1, N i - measured values; x' i ,k - calculated value mainly by reducing the residue of LSLR procedure.
values LSLRM, i.e. This implies an accurate calibration with a large number
(Ni) of adequate CRM’s and, the last but not the least, using
a good “radiation measure machine” which is the
spectrometer. Thus, a precise and accurate measurement of
spectral line intensities depends on many factors as: sample
parameters (chemical homogeneity, microstructure,
cleanness, etc.), optic parameters (transmittance of the lens
and glass fiber; dispersion of diffraction lattice and other
aberrations etc.), electronic parameters (CCD’s efficiency,
electronic noise, etc.).
From the above results that the estimation of the RSDi
implies at least two steps:
1- calculation of RSDCS using Eq. (10);
2- calculation of RSDCi using Eq. (11);
The estimation of RSDCS and RSDCs imply to know the
values of Ni, Vi, S Oi for all dozed elements, i.e. i = 1, n .
The yim ≡ y i , i = 1, n values are not directly accessible to
the spectrometrist but they could be replaced with the
Ci
values that are values displayed by spectrometer.
Cs
Using an appropriate software, RSDCS and subsequently
RSDCs could be calculated but the date an ISM calibration
of a spectrometer is a matter of know-how that is protected.
Thus, the issue of the calibration contribution to the
uncertainty budget of SDAR-OES ISM remain, for a while,
to the theoretical level that is, anyhow, a big breakthrough.
Once having the validation of the correctness of eq. (22)
then will be quite easier to estimate RSDCi, i=1…n.
The authors will search for calibration data that could be
brought in public paper in the next couple of month and will
perform computations and simulations of the SDAR-OES
ISM contribution to the uncertainty budget of
spectrochemical analysis.
5. Conclusions
The paper addresses a very sensitive and important subject
concerning the contribution of the calibration to the
uncertainty budget of SDAR-OES internal standard method.
There were theoretically estimated the standard deviations
and relative standard deviations for all dozed elements.
The most two among the striking results of this issue are:
1. The relative standard deviation of an alloying
element i (RSD2Ci) consists of two terms, one which is the
same for all alloying elements and one that depends,
somehow in a complicate manner, on element i (see eq.
(10)). This fact shows that the RSD2Ci could not be
decreasing down to (CS/100)2·RSD2 Cs.
2. The standard deviation of the quotient yi, S2yi,
reaches its minimum value when yi is one half of the average
value of the quotient used to fit the calibration line, i.e.
yi = <yi’>/2.
The authors have gotten an adequate way to match the way
of spectrometer calibration to the way of spectrometer is
operated which consists The way of S2yi calculation using
{
the calibration on parameter set Ni, S 2 , V , i = 1, n .
oi i }
The paper specifies the way of computing RSD and
Cs
subsequently, RSDci but their application to a real case of
uncertainty estimation are hindered, for a while, because the
calibration data are classified for available commercial
spectrometers.
Anyhow, by the authors’ knowledge, it is for the first time
when was done a thoroughly theoretically estimation of the
SDAR-OES ISM contribution to the spectrochemical
uncertainty budget.
The results presented in the paper and postponed further
ones in this field will use to develop a more reliable
procedure for uncertainty estimation of spectrochemical
analyses that will comply better with the EA 04/16 and SR
EN 13005: 2005 requirements.
Until the submission of final form of this paper the author
will get at least a set of calibration data for applying the
above presented theoretical model for estimation of SDAR-
OES ISM contribution to the spectrochemical uncertainty
budget and to simulate the influence of different variable on
this uncertainty.
References
[1] K. Slickers, „The automatic emission spectroscopy”, ISBN
3980.3333-l-0 Brühl Universitäts – Drukerei, Germany, 1993
[2] I.Pencea, “Basics of arc or spark discharge optical emission
spectrochemical analysis”, Ed. Printech, Bucharest, 2007.
[3] ***, http://www.spectro.ro
[4] A. Truţia, I Ioniţă, Floriana Iova, G., “Optical, atomic and molecular
Spectrometry”, Ed. Univ. , Bucharest, 2003
[5] I. Ioniţă, „Spectral methods for medical analisys”, Bucharest, 2003.
[6] I. Lupu, F. Grigorescu, L. Lupu, “Instrumental analysis for
metallurgy and machine construction industry”, Ed. Tech.,
Bucharest, 1986.
[7] E.Tenescu, N.Vlahovici, A. Vancu, “Applied spectroscopy”,
Didactical and Pedagogical Ed., Bucharest, 1975.
[8] ***, “GUM-Guide to the Expression of uncertainty in measurement”
[9] ***, “EA 4/16- EA guidelines on the expression of uncertainty in
quantitative testing”
[10] ***, “EURACHEM / CITAC Guide CG 4, Quantifying Uncertainty
In Analytical Measurement” (second edition), 2000
[11] Tiron, M., „Statistical and informational processing of measurement
Data”, . Ed. Tech., Bucharest, 1976
[12] I. Voinea, A. Semenescu, M. Tarcolea, “Data processing and analysis
for medical engineering”, Ed. MatrixRom, Bucharest, 2002
[13] I. Voinea, A. Semenescu, M. Tarcolea, “Informatics for medical
Engineering”, Ed. MatrixRom, Bucharest, 2002.
[14] I. Ceausescu, “Experimental data processing for analytical
chemistry”, Ed. Tech., Bucharest, 1976.
View publication stats