Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Machine Translated by Google

Microchemical Journal 68 Ž2001


. 11 1

Simultaneous determination of cobalt and nickel.


Comparison of prediction ability of PCR and PLS using
original, first and second derivative spectra

J. Ghasemi, A. Niazi
Chemistry Department, Razi Uniersity, Kermanshah, Iran

Received June 17, 2000; received in revised form 18 August 2000; accepted August 21, 2000

Abstract

Two multivariate calibration methods, partial least squaresŽ.PLS and principal component regressionŽPCR., were
applied to the simultaneous determination of the divalent ions of nickel and cobalt based on the formation of their
complexes with 4-Ž 2-pyridylazo. resorcinol
Ž. PAR . The original, first and second derivative absorption spectra of the
binary mixtures were used to perform the optimization of the calibration matrices by the PLS and PCR methods. The
results obtained by the application of the different chemometric approaches are discussed and compared. The
SavitzkyGolay convolution method is used for calculating the first- and second-derivative absorption spectra of
binary mixtures. Calibration matrices were 15540 and 15560 ng ml1 for cobalt and nickel, respectively. 2001
Elsevier Science BV All rights reserved.

Keywords: Partial least-squares; Principal component regression; Cobalt; Nickel; Determination

1. Introduction 4 , atomic absorption spectrometry 5 , etc., have


been used for the immediate determination of
Cobalt and nickel appear together in many realities these ions in different samples. Among the most
and synthetic samples. Several techniques such as widely used analytical methods are those based
,
X-ray fluorescence 1 atomic fluorescence spec- on the ultraviolet UV Ž .visible spectrophotome-try
trometry 2 , polarography 3 , chromatography techniques, due to both the resulting experi-mental
rapidity and simplicity and the wide application.
However, the simultaneous determination
Corresponding author. of these ions by the use of traditional spectropho-
E-mail address: ghasemi@razi.ac.irŽ J.
. Ghasemi .
tometry techniques is difficult because, generally,

0026-265X01$ - see front matter 2001 Elsevier Science BV All rights reserved.
PII: S 0 0 2 6 - 2 6 5 X 0
Ž 0 0. 0 1 5 9 - 4
Machine Translated by Google

2 ( 1 11
J. Ghasemi, A. Niazi Microchemical Journal 68 2001 )

the absorption spectra overlap in a bright region 1.00 cm quartz cells. A Metrohm 692 pH-meter
and the superimposed curves are not suitable for furnished with a combined glass-saturated calomel
quantitative evaluation. electrode was used for pH measurements. The
Under computer-controlled instrumentation, pH-meter was calibrated with at least two buffers
derivative techniques and multivariate calibration solutions at pH 3.00 and 9.00.
methods are playing a very important role in the
multicomponent analysis of mixtures by ultravio-let 2.2. Computer hardware and software
UV Ž .visible molecular absorption spec-trophotometry
6 8 . The advantage of multicom-ponent analysis All absorbance spectra were digitized and
using multivariate calibration is stored at wavelengths from 380 to 620 nm in steps
the speed of the method of determination for of 1 nm and then transferred inŽ ASCII format .
components of interest in a mixture, as a to a Pentium 200 MHz computer for subsequent
separation step can be avoided. manipulation by PCR and PLS programs.
The application of quantitative chemometrics, The data treatment was performed with MATLAB
particularly least classical squares CLS ŽŽ.. , inverse for windows math
Ž. works, version 4.2 . PCR and
least squares ILS , principal component regres- Ž . Ž. PLS programs Žfor calibration-prediction and ex-
sion PCR and , partial least squares PLS to , perimental design .were applied to a program
multivariate chemical data is becoming more written in MATLAB according to the algorithm
widely owing to the availability of digitized described by Martens and Naes 9 .
spectroscopic data and commercial software for
laboratory computers. 2.3. Reagents
The theory and application of PCR and PLS in
spectrometry has been discussed by several
workers 9 12 . Also several multicomponent de- All the chemicals used were of analytical-
reagent grade, subboiling, distilled water were used
termination of inorganic substances based on the
throughout.
application of these methods to spectrophotometer- Stock solutions of cobaltŽ.II and nickel Ž.
II were
ric data have been reported 13,14 . Recently, the
prepared from their commercial salts nitrate
Ž or
combination of derivative techniques with multi-
sulfate. and standardized titrimetrically. Stan-
variate calibration methods has been proposed,
dards of working solution were made by appropri-
and the convenience of such an approach has
ate dilution daily as required. A stock PAR solu-
evaluated by several workers with contradic- 5
.
tion 4.010
Ž M was prepared by dissolving reagent.
tory results 15,16 . . by sodium
A buffer solution pH 8.2 was Ž prepared
In this work reports the simultaneous spec-
borate.
trophotometric determination of cobalt and nickel
with 4- Ž2-pyridylazo. resorcinol
Ž. PAR using PCR
2.4. Procedure
and PLS algorithms is reported. The results ob-
tained by the two methods are compared and
discussed. discussed.
Known amounts of the standard solutions were
placed in a 10-ml volumetric flask and completed
to the final volume with deionized water final pH Ž
8.2. . The final concentration of these solutions
2. Experimental
varied between 15540 and 15560 ng ml1 for
cobalt and nickel, respectively.
2.1. Apparatus

Electronic absorption measurements were car-ried 3. Results and discussion


out on a CECIL 9000 spectrophotometer slit Ž
width 0.2 nm and scan rate 500 nmmin using . Fig. 1 shows the original, first and second
Machine Translated by Google

J. Ghasemi, A. Niazi Microchemical Journal 68 2001 1 )(11 3

termining cobalt IIŽ. Ž.


and nickel II in mixtures, the
optimal working conditions were studied un-der
the conditions previously established for each
metal ion. A sodium borate buffer solution of pH
8.2 was selected. The effect of PAR concentration
was also investigated, a reagent concentration of
4.0105 M was chosen because it ensures a
sufficient reagent excess. In order to choose the
optimal pH value at which the minimum overlap
occurs, influence of the pH of the medium on the
absorption spectra of metal ions were studied
over the pH range 7.09.0.
The absorption depends critically on the order
of addition of reactants, the most suitable order
of addition was metal ion solutions, PAR and
buffer. buffer. Individual calibration curves were
con-structured with several points Ž Fig. 2, . as
absorbance vs. metal ion concentration in the range of
15540, 15560 ng ml1 for cobalt and nickel,
respectively. Linear regression results, line equa-
tions and R2 are also shown on the Fig. 2.

3.1. Calibration and prediction data sets

Multivariate calibration methods are suitable


for the analysis of large number of samples. How-
ever, they are not advised for the determination
of large numbers of analyzes because of the com-
plexity of the calibration matrix. Moreover, the
preparation and analysis of the standards belong
to the calibration set are the most expensive step
in the multivariate calibration procedure.
Multivariate calibration methods such as PLS
require a suitable experimental design of the
standards belonging to the calibration set in order
to provide good predictions. A synthetic set of 43
Fig. 1. I: Absorbance; II: first; and III: second derivative solutions of mixtures of Co2 and Ni2 were
Ž. Ž.
spectra of a cobalt a nickel , and binary mixture com- Ž .
b .
prepared Tables
Ž 1 and 2 . From the series, 33
plexes c with PAR.
solutions were chosen for the calibration set and
the other 10 were used as prediction solutions.
derivative absorption spectra in aqueous solution This selection was performed on the basis of their
of the individual metal complexes and a mixture distributions on the first and second principal
of them at pH 8.2. The SavitzkyGolay convolu-tion components. Prediction solutions were chosen ac-
method is used for calculating the first- and cording to their situation on the graph, inside the
second absorption derivative spectra 17 . With the range of calibration solutions ie liesŽ in the calibration
aim of investigating the possibility of de- range ..
Machine Translated by Google

4 J. Ghasemi, A. Niazi Microchemical Journal 68 2001( 1 ) 11

Fig. 2. Analytical curve for univariate determination of cobalt and nickel complexes.

The spectral region between 380 and 580 nm, eral methods to obtain the number of significant
which implies working with 201 experimental factors. factors. At first according to Malinowski
points Ž. Ž per
thespectrum
spectra are
as digi-. algo-rithm ,18 a significant factor analysis applied
tized every 1-nm interval was selected for using the Malinowski indicator function INDŽ . ,
analysis. In this spectral region, the maximum spectrum
REV, Eigenvalue and Real Error Indicator. The
information is shown. The obtained spectra were obtained results are shown in Fig. 3. As can be
just mean centered. seen from Fig. 3 at least a significant factor of 3 is
needed to reconstruct the absorbance data
3.2. Number of significant factors matrix. Another function to confirm the number of
significant factors latent
Ž variables in. the PCR
In order to model the system without and PLS algorithm, is a cross-validation method,
overfitting the data concentration, we employ the sev-leaving out one sample at a time 19 . Given the
Machine Translated by Google

J. Ghasemi, A. Niazi Microchemical Journal 68 2001 1 ) 11 ( 5

Table 1
Concentration data for the different mixtures used in the calibration set for the determination of cobalt and nickel

Mixture Cobalt Nickel Mixture Cobalt Nickel


first first first first

ng ml ng ml ng ml ng ml

m1 550 538 m18 25 175


m2 536 538 m19 390 452
m3 17 75 m20 263 402
m4 320 410 m21 240 250
m5 481 510 m22 115 132
m6 217 104 m23 456 502
m7 246 285 m24 257 357
m8 450 487 m25 65 40
m9 230 210 m26 498 532
m10 362 426 m27 435 462
m11 506 537 m28 250 325
m12 115 165 m29 25 175
m13 221 130 m30 285 402
m14 355 417 m31 225 160
m15 171 103 m32 374 435
m16 516 537 m33 106 74
m17 490 518

set of 33 calibration spectra, the PCR and PLS PLS models are as follows:
calibration on 32 calibration spectra were per- No. PRESS
Calibration Metal Type of spectra method
formed, and using this calibration, the concentra- ion factor 10000
2
tions of the compounds, in each sample were PLS Co Original 12 0.0206 First derivative 12 0.0464
2
PLS Co Second derivative 13 0.0701 Original 11
compared with the known concentrations of the 2
PLS Co 3.2329 First derivative 13 0.0640 Second
compounds in this reference sample and the pre-. 2
PCR Co derivative 11 0.7644 Original 11 0.0144 First
diction error sum of squares PRESSŽ was calculated. PCR Co
2
derivative 12 0.0047 Second derivative 13
The PRESS was calculated in the same PCR Co
2
0.0701 Original 13 0.1736 First derivative 12
2
manner each time a new factor was added to the PLS Ni 0.0236 Second derivative 13 0.3202
2
PLS or PCR models. PLS Ni
2
PLS Ni
One reasonable choice for the optimal number of 2
PCR Ni
factors would be that number which yield PCR Ni
2
2
the minimum PRESS. However, using the number PCR Ni
of factors h . that
Ž yields a minimum PRESS

usually leads to some overfitting. A better crite-rion Print Fig. 4, the cumulative PRESS obtained by
for selecting the optimal number of factors optimizing the calibration matrix of the original,
involves the comparison of PRESS from models first- and second derivative of absorption spectra
with fewer than h factors. The F-statistics was with these methods, are shown. By inspection of
used to make the significance determination. Hol- PRESS plots of different cases, using the PLS on
land and Thomas 20 empirically determined that an original spectra gives fewer significant factors than
F-ratio probability of 0.75 is a good choice. PLS on derivative and PCR on all types of spectral
We selected as the optimum the number of factor- data. data.

tors for the PRESS value the F-ratio probability


of which drops below 0.75. The maximum number 3.3. Statistical parameters
of factors used to calculate the optimum PRESS
was selected as 15 and the optimal number of The root mean squares difference RMSD Ž is an .
factors obtained by the application of PCR and indication of the average error in the analysis
Machine Translated by Google

6
Table 2

Mixture Co

P1
P2
P3
J.
Added and found results of the synthetic mixtures of cobalt and nickel by PLS and PCR methods using original, first and second derivative spectral data

Added

440.5
250.0
430.0
Ni

410.5
75.0
395.0
2
Found

Original data
PCR

Co

434.1
211.7
459.5
2
Ni

432.1
101.1
331.6
2
PLS

Co

438.8
242.9
435.5
2
Ni

411.7
79.7
394.1
2
First derivative data

PCR

Co

435.7
222.2
446.9
2
Ni

387.2
2

83.8
387.3
PLS

Co

436.6
223.6
439.1
2
Ni

409.3
97.8
391.2
2
Second derivative data

PCR

Co

436.9
220.3
454.5
2
Ni

396.6
99.4
349.7
2
PLS

Co

426.2
240.1
437.4
2
Ni

406.8
77.3
369.4
2

P4 305.0 380.0 335.9 340.0 304.0 380.8 356.4 344.2 331.9 371.4 342.6 351.3 312.4 390.8
P5
P6
P7
P8
P9
P10
Ghas
462.5
450.0
537.5
125.0
210.0
375.0
307.5
207.5
312.5
175.0
275.0
200.0
483.8
424.5
541.9
158.9
209.6
359.0
328.4
226.7
317.7
116.8
284.0
206.3
459.0
448.1
539.0
129.1
212.5
379.0
304.9
204.7
315.6
170.5
273.1
199.5
455.9
439.7
548.0
127.0
244.3
351.1
330.0
216.6
306.2
157.9
251.8
223.5
462.0
442.4
538.9
140.4
226.3
376.2
299.1
216.8
311.2
150.9
242.2
202.4
479.0
435.1
497.5
162.4
174.4
342.6
314.5
226.4
328.9
135.9
318.5
225.4
478.2
440.0
539.2
135.8
211.1
366.4
305.5
216.0
325.4
162.5
278.5
200.8

1
Machine Translated by Google

( 1 ) 11
J. Ghasemi, A. Niazi Microchemical Journal 68 2001 7

Fig. 3. Significant factor analysis results of calibration matrix of cobalt and nickel mixtures.

for each component, is PCR and PLS, are summarized in Table 3. Ac-
cording to PRESS and statistical values, PLS on

(
n
original spectral data has given the best results
RMSD 1n x Italy Ž ˆi xi .2
with respect to the other mentioned data manipu-
i1 lation methods.
2
and the square of the correlation coefficient R .,Ž
which is an indication of the quality of fit of all 3.4. Application to alloy matrix and synthetic
the data to a line, is mixtures

n n
2
R 2Italy Italy iŽ. Ž. ˆx x xx i
2 Four synthetic mixtures of the two components
i1 i1 that their compositions have been prepared ac-
cording to some known alloys 21 . The proposed
where x isi the true concentration of the analyte in PLS and PCR methods were applied to the origi-
the sample i, ˆx represents
i the estimated con- nal, first and second derivative spectra, allowing the
centration of the analyte in the sample i, x is the resolution of their components. The composition
mean of the true concentration in the prediction of the binary mixtures assayed and PCR and PLS
set, and n is the total number of samples used in predictions are shown in Table 4.
the prediction sets, that were calculated. Print Fig. 5a,b show the recovery results obtained
The statistical parameters found for the original, by the application of the PLS and PCR methods
first and second derivative spectral data using on the three types of spectral data are repre-
Machine Translated by Google

8 (
J. Ghasemi, A. Niazi Microchemical Journal 68 2001 1 ) 11

sent. Satisfactory recovery values are obtained


in most of the binary samples analyzed. Recovery
values for Co2 and Ni2 by using PLS and PCR
and three types of spectral data are as follows:
Calibra- Type of spectral llltion Range of recovery %Ž .
data method
Co 2 Ni 2

PLS Original 97.2103.2 97.4106.3


PLS First derivative 89.4112.3 86.2130.4
PLS Second derivative 96.0108.6 92.9104.1
PCR Original 84.7127.1 66.7134.8
PCR First derivative 88.9116.9 90.2111.8
PCR Second derivative 83.0129.9 77.7132.5

There is a good agreement between obtained


results using PLS on original spectral data, and
known values indicate the successful applicability
of proposed method for simultaneous determina-
tion of cobalt and nickel in complex real samples.

4. Conclusion

The cobalt and nickel mixture is a complex


system due to the high spectral overlap
observed between the absorption spectra for these
components. For overcoming the drawbacks of
spectral interferences, PLS and PCR multivariate
calibration approaches are applied and compared.
Analysis of the results for the two-component
system revealed difference in the accuracy of
prediction between the two procedures for origi-nal,
first- and second derivative data. The best
recovery values are obtained by the application of
Fig. 4. Cumulative PRESS as a function of number of factors. PLS model for original absorbance data. The
I: original; II: first; and III: second derivative spectra. good agreement clearly demonstrates the utility

Table 3
Statistical parameters of the PLS and PCR methods using of original, first and second derivative spectral prediction data sets at
optimal significant factors

Type of Component PLS PCR


spectral data RMSD R2 RMSD R2

Original Co 2 3.7293 0.9956 24.2107 0.9944


2
Ni 2.7074 0.9989 33.3348 0.9310
2
First derivative Co 14.4321 0.9522 23.9480 0.9773
2
Ni 15.5911 0.9316 19.9322 0.8526
2
Second derivative Co 9.7632 0.9833 29.5283 0.9514
2
Ni 10.9671 0.9762 31.4410 0.7673
Machine Translated by Google

Table 4

Added

Udimet 700
J.
Actual composition and calculated concentration of cobalt and nickel in synthetic mixtures of some alloys

Co

37.2
2
Ni

106
2
Found

Original data
PCR

Co

22.2
2
Ni

81.6
2
PLS

Co

21.2
2
Ni

91.0
2
First derivative data

PCR

Co

34.4
2
Ni
2

24.5
PLS

Co

15.7
2
Ni

128.5
2
Second derivative data

PCR

Co

30.2
2
Ni

105.7
2
PLS

Co

50.9
2
Ni

112.4
2

Waspaloy
Niomonic 115
Rene 41

Ghas
aSnell and Ettre 21
27
26.4
22

.
115
120.1
110.1
20.7
71.9
23.2
124.3
111.1
120.4
22.7
22.7
28.0
113.8
120.7
111.9
37.2
27.8
27.5
75.4
100.6
121.9
30.8
18.0
23.1
136.5
146.8
103.5
49.3
63.7
30.6
101.9
123.6
124.5
13.3
24.6
25.9
108.4
116.1
116.1

1
9
Machine Translated by Google

ten J. Ghasemi, A. Niazi Microchemical Journal 68 (20011 ) 11

Fig. 5. Diagrammatic representation of the percentage recovery found in the analysis of 10 mixtures of cobalt and nickel by PLS aŽ .
and PCR Ž
b .methods, using I: original; II: first; and III: second derivative spectral data.

of this procedure for the simultaneous determina- 4 N. Xic, C. Huang and HU Fu, 1990, sepu 8 2 , 114, Ž .
chem abstr. 113Ž1990 094170
tion of the cobalt and nickel in synthetic and alloy .

samples. 5 Z. Schen, F. Nie, Y. Chen, Fensi Huaxane November 19, 1991.


Ž .Ž
1272.
6 TC O'Haver, GL Green, Anal. Chem. 48 1976 312. Ž Ž .
References 7 TC O'Haver, Anal. Chem. 51 1979 91A. .
8 P. Geladi, BR Kowalski, Anal. Bird. Acta 185 1986Ž 1. .
9 H. Martens, T. Naes, Multivariate Calibration, John
Ž ..Ž
1 OL Lan, SY Ho, Anal. Bird. Acta 280 2 1993 269. Ž .Ž . Wiley, New York, 1991.
2 V. Rigin, Anal. Bird. Acta 283 2 1993 895. Ž .Ž . 10 DM Haaland, EV Thomas, Anal. Chem. 60 1988 Ž .
3 Z. Shen, Z. Wang, Fensi Huaxue 21 11 1993 1313. 1193.
Machine Translated by Google

J. Ghasemi, A. Niazi Microchemical Journal 68 (20011 ) 11 11

11 KR Beebe, BR Kowalski, Anal. Chem. 59 1987 Ž . 17 A. Savitzky, MJE Golay, Anal. Chem. 36 August 1964 Ž.Ž .
1007A. 1627.
12 S. Wold, P. Geladi, K. Esbensen, J. Ochman, J. Chemom. 18 ER Malinowski, Factor Analysis In Chemistry, 2nd ed.
41 Ž1987
. 1. John Wiley, New York, 1991.
13 LG Hargis, JA Howell, Anal. Chem. 64 1992Ž 66R.. 19 MJ Stone, Stat. Soc. 36 1974
Ž . 111.

14 JA Howell, LG Hargis, Anal. Chem. 66 1994 Ž445R.. 20 DM Haaland, EV Thomas, Anal. Chem. 63 1989 Ž .
15 P. Maclaurin, PJ Warsfold, M. Crane, P. Norman, 1193.
Anal. Proc. 29 1992
Ž .65. 21 FD Snell, LS Ettre, Encyclopedia of Industrial Chemi-
16 R. Jones, TJ Coomber, JP Mccormich, FA Fell, BJ cal Analysis, Vol. 15, Wiley Interscience, New York,
Clark, Anal. Proc. 25 Ž1988.381. 1972.

You might also like