Raman spectrum online monitoring in aspirin synthesis process

Article abstract of DOI:10.14233/ajchem.2014.15447

In this work, the Raman spectra were obtained directly in the synthesis process of aspirin as prediction set. The concentration sample according to the amount of material changes before and after reaction and it's extracted Raman information made up training set and a quantitative analysis model of the salicylic acid-acetic anhydride-aspirin mixed system was established by applying partial least squares (PLS). The values of correlation coefficient squared between measured and regression (R2), root mean square error of prediction set (RMSEP), mean deviation (MD) indicate that the influence of the prediction precision by the baseline drift can be effectively reduced in the method of spectrum pretreatment via mormalize + savitzky-go1ay smoothing first derivative. After the prediction set was used in this analysis mode, the mass per cent between salicylic acid and aspirin of online monitoring was 4.59 %, while the off-line Raman analysis and high performance liquid chromatography analysis results were 5.66 and 2.75 %. The relative deviations were -1.07 and 1.84 %. This method provides effective data support for synthesis of aspirin Raman spectroscopy online monitoring.

Full text of DOI:10.14233/ajchem.2014.15447

Asian Journal of Chemistry; Vol. 26, No. 2 (2014), 467-471
http://dx.doi.org/10.14233/ajchem.2014.15447
Raman Spectrum Online Monitoring in Aspirin Synthesis Process
*
HUI SU, KUO SUN, ZHIXIANG YAO , PEIXIAN HUANG and LIU LIU
Department of Biology and Chemical Engineering, Guangxi University of Science and Technology, Liuzhou 545006, Guangxi P.R. China
*
Corresponding author: E-mail: hailfellow512@163.com
Received: 7 March 2013; Accepted: 31 July 2013;
Published online: 15 January 2014;
AJC-14568
In this work, the Raman spectra were obtained directly in the synthesis process of aspirin as prediction set. The concentration sample
according to the amount of material changes before and after reaction and it’s extracted Raman information made up training set and a
quantitative analysis model of the salicylic acid-acetic anhydride-aspirin mixed system was established by applying partial least squares
2
(
PLS). The values of correlation coefficient squared between measured and regression (R ), root mean square error of prediction set
RMSEP), mean deviation (MD) indicate that the influence of the prediction precision by the baseline drift can be effectively reduced in
(
the method of spectrum pretreatment via mormalize + savitzky-go1ay smoothing first derivative. After the prediction set was used in this
analysis mode, the mass per cent between salicylic acid and aspirin of online monitoring was 4.59 %, while the off-line Raman analysis
and high performance liquid chromatography analysis results were 5.66 and 2.75 %. The relative deviations were -1.07 and 1.84 %. This
method provides effective data support for synthesis of aspirin Raman spectroscopy online monitoring.
Keywords: Pretreatment, Partial least squares, Online monitoring, Raman spectroscopy, Aspirin synthesis.
1
component attribute matrix is considered . Thomas and
2
INTRODUCTION
13
Haaland compared classical least squares (CLS), PCR with
partial least squares, they draw the conclusion that partial least
squares was the optimal method for multicomponent prediction.
However, during the application of Raman spectroscopy,
fluorescence of organic compounds in the samples, sometimes
several orders of magnitude more intense than the weak Raman
Increasing demands on process safety, product quality and
cost reduction are the main reasons for the growing interest in
the online monitoring whose features are real-time, non-
destructive, rapid analysis. The traditional analysis methods
(
titration, high performance liquid chromatography (HPLC))
cannot achieve requirements for online monitoring. Raman
spectroscopy analysis has advantages whose properties are
non-destructive monitoring, high sensitivity, short detection
time etc. and Raman spectrum information on reaction medium
can be rapidly obtained with a fiber optic probe. Simultaneous
determination of multicomponent can be performed in this
14-16
scatter and interfere with the Raman signals . So in order to
eliminate fluorescence influence, to remove a portion of the
17
noise and increase the signal to noise ratio , pretreatment
should be adopted in normal conditions before regression. The
normal pretreatment methods include normalization, smoothing,
18-26
derivative, etc.
.
1-4
method . Due to the presence of multiple characteristic peaks
in the Raman spectrum data of measured components and the
characteristic information of measured components were over-
This work chose sulphamic acid catalytic synthesis of
aspirin system as a study object and adopted the Raman
spectrum analysis method based on the partial least squares
algorithm via spectrum pretreatment to test and verify online
5
lapping , it is difficult to separate characteristic peaks to
determine the component concentration of the synthesis process.
In order to overcome the influence of multiple correlation
caused by overlapping information, multivariate calibration
27
monitoring of three components in aspirin synthesis process .
Theory
6
-8
techniques need to be used in Raman spectrum analysis . To
determine the content of known object component that is white
Partial least squares: The spectra are collected as rows
in a matrix X, with n rows and k columns. Each row represents
a spectrum and each column a single wavelength, the corres-
ponding, concentration of major components are placed as
rows in a matrix y.
9
analytical system , both principal component regression (PCR)
and partial least squares (PLS) can extract component variable
1
0,11
characteristic information , when the latter decompose
spectral matrix, it makes the region of the spectrum with the
measured remarkable attribute greater weight and the measured
Partial least squares is a regression method with a matrix
X and a vector y (or matrix). Partial least squares calculates

468 Su et al.
Asian J. Chem.
components that both capture the variation in X and correlates
with the variation in y. The loading weight vector w , where a
n
 f(−m)  1 − m
(−m) a   ε 
a
…
0
1





  
n

denotes partial least squares component number, shows the
spectral features in X that correlate with y, calculated as in
eqn. 1. The score vector, which can be interpreted, is calculated
as in eqn. 2. A scalar coefficient is then calculated (eqn. 4)
which in turn together with constitutes an equation for y, where
f is a residual with non modeled variation in y (eqn. 4).
f(−m +1) 1 − m +1 … (−m +1)  a1  ε2 
=
+



  

M
f(m)
M
1
M
m
M
…
M
m
M
M
εk
(9)








 





n
  
  
an
BA = g
(10)
(11)
(12)
(13)
T
A = (B B) B g
−1
T
X ′a −1ya−1
w =
a
(1)
(2)
T
−1
T
Xa−1ya−1
g
ˆ =
BA = B(B B) B g = Bg
T
K = B(B B)
−1
T
ta = Xa−1wa
B
(
14)
i = 2m +1
here, B is a matrix whose rows provide the terms of the
polynomial f(x) which take the coefficients in a and g is same
t ′a y
a−1
ca =
(3)
(4)
t t
a
a
as f(x). gˆ is the filtering value. K is the filtering coefficient
y = c t + c t +…+ c t + f
A A
1
1
2
2
matrix, which is only relevant to both filtering window and
polynomial order n.
The number of partial least squares components to be used
is often selected with cross validation and/or independent test
sample sets. For partial least squares, the sign of the loading
vector is determined by the direction of the change in y.
The predictive performance was assessed on the basis of
the root mean squared error of calibration and prediction,
calculated by:
Derivative: Spectrum of the first order and second order
derivative are commonly used in pretreatment methods of
spectral analysis to correct baseline drift and extract spectral
characteristic value. Compared with the original spectrum
signal, derivative spectrum can effectively eliminate the inter-
26
ference of baseline drift and background . But the derivation
may reduce the signal to noise ratio. Thus original spectra
need to be pretreated in the method of smoothing before deriva-
tion. Savitzky-Golay convolution first order derivative is
relatively commonly used as eqn. 15
n
2
(
y − y )
i i
∑
i=1
(5)
RMSE =
n
∂
f
Normalization: Normalization methods normally include
area normalization, maximum normalization and vector
normalization. Vector normalization is commonly used as a
Raman spectral data processing method to correct the spectral
changes caused by the small optical path distinction, which is
following equation:
=
a1
(15)
∂x
EXPERIMENTAL
Salicylic acid (AR), acetylsalicylic acid (AR), acetic
anhydride (AR), acetic acid (AR), amino sulfuric acid (AR)
were used as received. Laser Aperture (Laser-785, Danger,
America). Raman Spectrometer (Scientific-grade QE65000,
Ocean optics, America). Fiber Optic Probe (BAC100-785-
OEM, Ocean optics, America). LC (1200 SL, Agilent,
America). Electronic balance (ALC-210.4, Sartorius, German).
Quartz cuvettes(1cm).Three-necked round-bottomed flask
xi
xi,norm
=
k
2
(
6)
xi
∑
i=1
where xi,norm is the transformed Raman intensity for wavenumber
, k wavenumbers is the spectrum. Eqn. 6 shall be repeated
for all k wavenumbers in the spectrum.
x
i
(
100 mL). Water bath, etc.
Raman platform setting:A laser with the wavelength of
Smoothing: Smoothing can effectively eliminate the high
frequency component of the spectrum while retaining low
frequency components, can effectively improve signal to noise
ratio. The weakness in this method was that useful high
frequency signal data may therefore suffer losses. In this
connection Gorry proposed the convolution least squares
smoothing method based on Gram polynomial recursive
7
85 mm was used as the excitation light source and Raman
information was obtained through the optical fiber probe, the
setting of ocean optics spectra suite was that the x axis on
workstation menu was selected Raman shift; selected integral
-1
time was 1/s to obtain the Raman spectrum of 0-2000 cm
spectral range.
Preparation of training sample set: Aspirin synthetic
route are shown in Fig. 1.
2
1-24
nature . We define the order n polynomial model function
f(x) as eqn. 7.
O
O
n
2
f(x) = a + a x + a x +…ak−1x
k−1
j
=
a jx
0
1
x
∑
(7)
(8)
OH
OH
Sulphamic acid
j=0
(CH
3
CO)
2
O
3
CH COOH
OH
OCOCH
3
f(x ) = g + ε
i
i
i
Fig. 1. Aspirin synthetic route

Vol. 26, No. 2 (2014)
Raman Spectrum Online Monitoring in Aspirin Synthesis Process 469
As Fig. 1 shows, suppose that salicylic acid completely
reacted at the end of the reaction, before and after the reaction,
the mole ratio of salicylic acid, acetic anhydride, aspirin are
n: 2 n: 0 and 0: n: n. 41 standard samples were prepared by
mixing appropriate amounts of salicylic acid, acetic anhydride,
aspirin and acetic acid. According to mole number range of
salicylic acid, acetic anhydride, aspirin, respectively was 5-0,
eliminate the spectral differences from the baseline shifts, first
derivative spectra were adopted as pretreatment model. Raman
spectrum data of both Training set and Prediction set were
preprocessed via normalize + Savitzky-Go1ay smoothing first
derivative (Fig. 3).
8
6
4
2
000
000
000
000
0
10-5 and 0-5 mmol, the three materials were weighed to 10 mL
15000
(a)
(b)
volumetric flasks by decreasing and increasing every 0.125
mmol and acetic acid was used as solvent to 10 mL constant
volume.
1
0000
5
000
The samples were, respectively transferred to 1 cm optical
path quartz cuvettes to collect and save its Raman spectrum
information as the training spectroscopy set.
0
0
0
0500 1000 1500 2000
00 0500 1000 1500 2000
Preparation of prediction sample set:According to the
–3
–4
×
10
×10
2
7
3
2
1
0
1
2
15
literature , we precisely weighed acetic anhydride 41 g,
salicylic acid 27.7 g, sulphamic acid 0.5 g. When the tempe-
rature rise to about 81 ºC using a water bath, acetic anhydride,
salicylic acid and sulphamic acid was transferred to 100 mL
three-necked round-bottomed flask one after another, the
reaction was lasting for 18 min by magnetic stirring at 81 ºC.
Raman spectrum was saved every minute as prediction sample
through scanning reaction system. The online monitoring
device is shown in Fig. 2.
(c)
(d)
1
0
5
0
-
-
-5
00 0500 1000 1500 2000
–1
Raman shift (cm )
0
0
0500 1000 1500 2000
)
–1
Raman shift (cm
Fig. 3. Raman spectrum of before and after pretreatment. (a) Raman
spectrum of training set; (b) Raman spectrum of prediction set; (c)
Raman spectrum of training set via pretreatment; (d) Raman
spectrum of prediction set via pretreatment
Fiber
Laser aperture
Fig. 3(a-b) show that there are baseline drift in the Raman
spectrum of both training set and prediction set, especially in
prediction set, degree of baseline drift is larger as Fig. 3(b)
shown. Fig. 3(a)(c) display the Raman feature of acetic acid
background in training set exist sharp and well-defined
Raman spectrometer
Fiber optic probe
Cable
-1
Optucs spectrasuite
peaks at 600 and 900-800 cm and observed Raman peaks of
Water bath
both acetic acid and measured components (salicylic acid,
-1
acetic anhydride and aspirin) exit in 200-50 cm . There are
Fig. 2. Online monitoring device
Raman features of measured components clearly shown in
Fig. 3(b-d).
Preparation of off-line sample: Because of short reac-
tion period of this reaction process, off-line sample can not be
prepared per minute. We sampled at equilibrium of reaction
approximately 10 min after sulphamic acid was added. 1 g
reaction sample was taken to a 10 mL volumetric flask using
acetic acid to volume and mixed. Its Raman information as
the off-line spectroscopy set was extracted and saved three
times, the relative content of salicylic acid and aspirin was
determined byAgilent 1200 HPLC system (AgilentTechnologies,
USA) equipped with a vacuum degasser, a quaternary pump,
an autosampler, a thermostatic column compartment, a diode
array detector (DAD). Separation was performed on anAgilent
Zorbax SB-C18 column (250 mm × 4.6 mm, 5 µm), using
methanol 0.5 % acetic acid (30:70, v/v) as mobile phase. The
detection wavelength was set at 278 nm. The flow rate of the
mobile phase was 1 mL/min.
Form Fig. 3 it can be deduced that the influence of
the baseline drift can be reduced and the Raman spectrum
differences caused by chemical component enhanced via
normalize + Savitzky-Go1ay smoothing first derivative pre-
treatment.
Establishment of the training set: Ten samples used as
test set were selected every 4 counts from 3rd to the 39th in 41
samples of the training set, others were used as calibration
set. Principal factors that main spectrum variables caused by
changes of sample in training set, they are acetylsalicylic acid,
salicylic acid, acetic acid and acetic anhydride. Selected
number of partial least squares factors was 4.
Evaluation of the training set: Component concentration
and Raman spectrum of all wavelength ranges were regressed
by both direct partial least squares regression and partial least
squares regression via normalize + Savitzky-Go1ay smoothing
RESULTS AND DISCUSSION
(polynomial = 2, points = 5) first derivative. In order to assess
regression effect, some parameters of analysis were made for
2
the partial least squares model, they were values of R ,
29,30
RMSECV, RMSEP and MD . Regression effects were shown
Spectrum pretreatment: It was investigated that deriva-
tives can reduce peak overlap and eliminate constant and
28
linear baseline drift . The principle is that inflection points
in close peaks become turning points in the derivatives. To
in the Fig. 4.

470 Su et al.
Asian J. Chem.
TABLE-1
EVALUATION PARAMETERS COMPARISON OF TRAINING MODEL
2
R
-1
RMSEP (mg mL )
-1
MD (mg mL )
Component
Un-pretreatment
0.9910
Pretreatment
0.9970
Un-pretreatment
Pretreatment
1.1228
Un-pretreatment
-0.6960
Pretreatment
-0.1904
Salicylic acid
Acetic anhydride
Aspirin
2.1032
2.1377
2.8227
0.9850
0.9960
1.0966
-1.0750
0.6271
0.9900
0.9970
1.5266
0.8574
0.1762
7
6
5
4
3
2
1
0
0
0
0
0
0
0
0
70
60
50
40
30
20
10
0
Related evaluation parameters of training model are shown
in Table-1.
Table-1 resumes the effect of partial least squares regres-
(
a1)
(a2)
sion via normalize + Savitzky-Go1ay smoothing (polynomial
= 2, points = 5) first derivative pretreatment in the training set
are optimistic compared with unpretreatment. Fig. 4 (Table-1)
deduced that the influence of the prediction precision by the
baseline drift can be effectively reduced in this pretreatment
method.
0
.
10 20 30 40 50 60 70
Y Measured 1
0. 10 20 30 40 50 60 70
Y Measured 1
1
00
100
Application of analysis model
(b1)
(b2)
9
8
7
6
5
0
0
0
0
0
90
80
70
60
50
Results of off-line determination: The off-line spectro-
scopy set was applied in the calibration set, after the form of
each component concentration in calibration set was translated
into the mass percent by the principle of mass conservation.
The results are shown in the Table-2.
5
0
60
70 80
Y Measured 2
90
100
50
60
70 80
Y Measured 2
90
100
TABLE-2
RESULTS OF OFF-LINE RAMAN ANALYSIS
9
8
7
6
5
4
3
2
1
0
90
80
70
60
50
40
30
20
10
0
0
0
0
0
0
0
0
0
0
(c1)
(c2)
Salicylic acid
Acetic anhydride
(%, m/m)
Aspirin
(%, m/m)
48.9898
48.9614
48.9785
48.9766
No.
(
%, m/m)
2.7624
2.7841
2.7710
2.7725
1
2
3
31.9216
31.9376
31.9279
31.9291
Mean value
The offline Raman analysis result of percentage between
salicylic acid and aspirin was 5.66 % and it was 2.75 % in the
method of HPLC.
0
.
20
40
Y Measured 3
60
80
0.
20
40
Y Measured 3
60
80
Fig. 4. Evaluation of training model. (a1) Salicylic acid correlogram by direct
partial least squares regression; (b1) Acetic anhydride correlogram
by direct partial least squares regression; (c1) Aspirin correlogram
by direct partial least squares regression; (a2) Salicylic acid
correlogram by partial least squares regression via spectrum pre-
treatment; (b2)Acetic anhydride correlogram by partial least squares
regression via Spectrum pretreatment; (c2) Aspirin correlogram by
partial least squares regression via Spectrum pretreatment
Application in online monitoring: Due to concentration
range of components between calibration set and reaction
solution is inconsistent, the form of each component concen-
tration need to translated into the mass percent by principle of
mass conservation. Mass percent of each component in
synthetic process were obtained after spectral data of prediction
set was applied in the calibration set. The results are shown in
Table-3.
In Fig. 4, (a1) (b1) (c1) show correlation between measured
and regressed by direct partial least squares regression and
there are most of residual levers of the sample point evenly
distributing near the fit line but the 6th sample point. From
Fig. 3a the conclusion can be drawn that the large residual
lever of the 6th sample point was caused by its Raman spec-
trum obviously drifting in the training set. While (a2) (b2)
The data in Table-3 multiplied by the total mass of reactant,
the mass of each component every minute in synthetic process
were obtained and they were shown in the Fig. 5.
45
Salicylic acid
Acetic anhydride
Aspirin
40
3
3
2
2
1
5
0
5
0
5
(
c2) show correlation between measured and regressed via
normalize + Savitzky-Go1ay smoothing (polynomial = 2,
points = 5) first derivative. All residual levers of the sample
point evenly distribute near the fit line, no obvious high lever
exist.
10
5
Fig. 4 indicates above that the influence of the prediction
precision by the baseline drift can be effectively reduced via
normalize + Savitzky-Go1ay smoothing (polynomial = 2,
points = 5) first derivative pretreatment.
0
0
1
2
3
4
5
6
7
8
9
Time (min)
10 11 12 13 14 15 16 17 18
Fig. 5. On-line results of each component

Vol. 26, No. 2 (2014)
Raman Spectrum Online Monitoring in Aspirin Synthesis Process 471
TABLE-3
MASS PER CENT OF PREDICTION SET
Time
(min)
Salicylic acid
Acetic anhydride
(%, m/m)
Aspirin
(%, m/m)
Time
(min)
Salicylic acid
(%, m/m)
Acetic anhydride
(%, m/m)
Aspirin
(%, m/m)
(%, m/m)
27.1054
18.9049
12.7266
9.0669
5.8254
4.1926
3.1053
2.4921
2.3484
1
49.9122
43.8504
39.2836
36.5784
34.1823
32.9755
32.1719
31.7187
31.6126
17.2378
27.9351
35.9943
40.7682
44.9967
47.1265
48.5448
49.3447
49.5320
10
11
12
13
14
15
16
17
18
2.2768
2.7394
2.7902
2.8844
3.3036
3.2374
3.4971
3.9492
4.0881
31.5598
31.9019
31.9395
32.0091
32.3191
32.2702
32.4622
32.7966
32.8992
49.6253
49.0217
48.9554
48.8326
48.2857
48.3720
48.0332
47.4433
47.2622
2
3
4
5
6
7
8
9
6
7
.
.
J.B. Cooper, Chemom. Intell. Lab. Syst., 46, 231 (1999).
F. Estienne, D.L. Massart, N. Zanier-Szydlowski and P. Marteau, Anal.
Chim. Acta, 424, 185 (2000).
Fig. 5 indicates that, in this synthetic process, 10 min after
initial reaction 0.5g sulphamic acid was added, the reaction
reached equilibrium. Then the online result of percentage
between salicylic acid and aspirin was 4.59 %. Compared with
results of offline Raman analysis and HPLC, the relative
deviations are -1.07 and 1.84 %.
8
9
.
.
H. Witjes, M. van den Brink, W.J. Melssen and L.M.C. Buydens,
Chemom. Intell. Lab. Syst., 52, 105 (2000).
Y. Liang, Multivariate Calibration Methods for White Analytical Systems,
In: Q. Chen, editor. White, Grey and Black Multicomponent Systems
and Their Chemometric Algorithms. Changsha, China: Hunan Publis-
hing House of Science and Technology, pp. 29-36 (1996).
Conclusion
1
1
0. F. Estienne and D.L. Massart, Anal. Chim. Acta, 450, 123 (2001).
1. J.B. Cooper, K.L. Wise, W.T. Welch, M.B. Sumner, B.K. Wilt and R.R.
Bledsoe, Appl. Spectrosc., 51, 1613 (1997).
This work used Raman spectroscopy characteristics of
salicylic acid, acetic anhydride and aspirin in the synthesis
system of aspirin under sulphamic acid catalytic, based on a
study in the changes of the material in the reaction process for
model optimization, to establish a rapid simultaneous quanti-
tative analysis model for salicylic acid, acetic anhydride and
aspirin. In this analysis model, by using partial least squares
regression via Normalize + Savitzky-Go1ay smoothing (poly-
nomial = 2, points = 5) first derivative pretreatment, selected
number of partial least squares factors was 4. Online determi-
nation of the three components in aspirin synthesis process
can be achieved. This analytical method can be used in process
monitoring of other synthetic system and offer base data for a
dynamic optimization control.
1
1
1
2. P. Geladi and B.R. Kowalski, Anal. Chim. Acta, 185, 1 (1986).
3. E.V. Thomas and D.M. Haaland, Anal. Chem., 62, 109l (1990).
4. C.L. Stanford, B.A. Mantooth and B.T. Jones, J. Chem. Educ., 78,
1
221 (2001).
15. Y. Hu, T. Jiang, A. Shen, W. Li, X. Wang and J. Hu, Chemom. Intell.
Lab. Syst., 85, 94 (2007).
1
6. R.L. McCreery, In ed.: J.D. Winefordne, Magnitude of Raman Scattering,
Raman Spectroscopy for Chemical Analysis, Wiley-Interscience, New
York, USA, pp. 25-26 (2000).
17. O.E. De Noord, Chemom. Intell. Lab. Syst., 23, 65 (1994).
1
8. Y.G. Shi, X.L. Feng and Z.C. Li, In ed.: G.Y. Zhang, Chemical Pattern
Recognition, Chemometrics.Beijing, China: China Petrochemical Press,
pp. 141-143 (2002).
1
2
2
9. K. Kachrimanis, D.E. Braun and U.J. Griesser, J. Pharm. Biomed. Anal.,
43, 407 (2007).
0. W. Wu, Q. Guo, D. Jouan-Rimbaud and D.L. Massart, Chemom. Intell.
Lab. Syst., 45, 39 (1999).
1. A. Savitzky and M.J.E. Golay, Anal. Chem., 36, 1627 (1964).
ACKNOWLEDGEMENTS
The authors acknowledged the foundation support from
Science and Technology Planning Project of Guangxi (129825-
22. H.H. Madden, Anal. Chem., 50, 1383 (1978).
2
2
2
2
3. J. Steinier, Y. Termonia and J. Deltour, Anal. Chem., 44, 1906 (1972).
4. P.A. Gorry, Anal. Chem., 62, 570 (1990).
20 and 1355010-15).
5. C.D. Brown and P.D. Wentzell, Chemometrics, 13, 133 (1999).
6. L.S. Mendes, F.C.C. Oliveira, P.A.Z. Suarez and J.C. Rubim, Anal.
Chim. Acta, 493, 219 (2003).
REFERENCES
2
2
7. S. Yang, J. Kunming Teachers College, 29, 108 (2007).
8. W.J. McCarthy, TQ Analyst User's Guide, Thermo Nicolet Corp., Madison,
WI, p. 155 (2003).
1
.
C. Wang, T.J. Vickers, J.B. Schlenoff and C.K. Mann, Appl. Spectrosc.,
6, 1729 (1992).
4
2
3
4
5
.
.
.
.
C. Wang, T.J. Vickers and C.K. Mann, Appl. Spectrosc., 47, 928 (1993).
J.B. Cooper, K.L. Wise and B.J. Jensen, Anal. Chem., 69, 1973 (1997).
N. Everall and B. King, Macromol. Symp., 141, 103 (1999).
2
3
9. S. Macho and M.S. Larrechi, TrAC-Trends Anal. Chem., 21, 799 (2002).
0. J.M. Andrade, S. Garrigues, M. de la Guardia, M. Gómez-Carracedo
and D. Prada, Anal. Chim. Acta, 482, 115 (2003).
G.V.L.N. Murty and T.R. Seshadri, Proc. Math. Sci., 19, 17 (1944).