Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components
A multistage recognition approach is advanced for poorly classified clusters. According to this approach, if a sample is related to a cluster that is common for several samples, then further object recognition (within that cluster) is possible. Such two-stage recognition procedure is based, at ea...
Gespeichert in:
| Datum: | 2008 |
|---|---|
| 1. Verfasser: | |
| Format: | Artikel |
| Sprache: | English |
| Veröffentlicht: |
Інститут фізики напівпровідників імені В.Є. Лашкарьова НАН України
2008
|
| Schriftenreihe: | Semiconductor Physics Quantum Electronics & Optoelectronics |
| Online Zugang: | http://dspace.nbuv.gov.ua/handle/123456789/119056 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Zitieren: | Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components / I.V. Kruglenko // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2008. — Т. 11, № 3. — С. 240-244. — Бібліогр.: 14 назв. — англ. |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-119056 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-1190562017-06-04T03:03:19Z Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components Kruglenko, I.V. A multistage recognition approach is advanced for poorly classified clusters. According to this approach, if a sample is related to a cluster that is common for several samples, then further object recognition (within that cluster) is possible. Such two-stage recognition procedure is based, at each stage, on the fuzzy logic concept and enables one to perform practically complete recognition of all the samples under consideration. 2008 Article Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components / I.V. Kruglenko // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2008. — Т. 11, № 3. — С. 240-244. — Бібліогр.: 14 назв. — англ. 1560-8034 PACS http://dspace.nbuv.gov.ua/handle/123456789/119056 en Semiconductor Physics Quantum Electronics & Optoelectronics Інститут фізики напівпровідників імені В.Є. Лашкарьова НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| description |
A multistage recognition approach is advanced for poorly classified clusters.
According to this approach, if a sample is related to a cluster that is common for several
samples, then further object recognition (within that cluster) is possible. Such two-stage
recognition procedure is based, at each stage, on the fuzzy logic concept and enables one
to perform practically complete recognition of all the samples under consideration. |
| format |
Article |
| author |
Kruglenko, I.V. |
| spellingShingle |
Kruglenko, I.V. Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components Semiconductor Physics Quantum Electronics & Optoelectronics |
| author_facet |
Kruglenko, I.V. |
| author_sort |
Kruglenko, I.V. |
| title |
Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components |
| title_short |
Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components |
| title_full |
Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components |
| title_fullStr |
Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components |
| title_full_unstemmed |
Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components |
| title_sort |
improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components |
| publisher |
Інститут фізики напівпровідників імені В.Є. Лашкарьова НАН України |
| publishDate |
2008 |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/119056 |
| citation_txt |
Improvement of sensor response reproducibility and multistage recognition approach for samples with dominant components / I.V. Kruglenko // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2008. — Т. 11, № 3. — С. 240-244. — Бібліогр.: 14 назв. — англ. |
| series |
Semiconductor Physics Quantum Electronics & Optoelectronics |
| work_keys_str_mv |
AT kruglenkoiv improvementofsensorresponsereproducibilityandmultistagerecognitionapproachforsampleswithdominantcomponents |
| first_indexed |
2025-07-08T15:09:31Z |
| last_indexed |
2025-07-08T15:09:31Z |
| _version_ |
1837091911312080896 |
| fulltext |
Semiconductor Physics, Quantum Electronics & Optoelectronics, 2008. V. 11, N 3. P. 240-244.
© 2008, V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
240
PACS
Improvement of sensor response reproducibility and multistage
recognition approach for samples with dominant components
I.V. Kruglenko
V. Lashkaryov Institute of Semiconductor Physics, NAS Ukraine
41 Prospect Nauky, Kyiv 03028, Ukraine
Tel.: (380-44) 525-56-26; Fax: (380-44) 525-83-42; e-mail: kruglenko@yahoo.com
Abstract. A multistage recognition approach is advanced for poorly classified clusters.
According to this approach, if a sample is related to a cluster that is common for several
samples, then further object recognition (within that cluster) is possible. Such two-stage
recognition procedure is based, at each stage, on the fuzzy logic concept and enables one
to perform practically complete recognition of all the samples under consideration.
Keywords: “Electronic Nose”, dominant components, multistage recognition procedure.
Manuscript received 00.00.00; accepted for publication 00.00.00; published online 00.00.08.
1. Introduction
The smart sensor systems are important facilities applied
when dealing with provision of efficiency and
environmental safety of industrial production, as well as
improvement of life quality [1].One of the problems
occurring in sensor devices is temporal variation of
sensitivity and selectivity profile of sensitive elements.
Such drifts affect image recognition by limiting
recognition ability of the system used, thus requiring
additional gauging. In some cases, however, it is
possible to avoid sensor surface modification in the
course of measurements by using a specific procedure of
surface treatment. Indeed, just the sensitive layers
determine almost all limitations of “Electronic Nose”
stability. Therefore, stability of sensor operation during
its service life involves stability and reproducibility of
the sensitive layer parameters [2].
A possibility to choose specific unique selectivity
profiles out of various molecular organic materials
makes them promising for application as sensitive
coatings in multichannel analyzers. The organic
molecules can interact with a wide range of various
analytes. This means also prevalent effect of dominant
components in complex mixtures, such as drinks [3, 4],
perfumes [5] and pharmaceuticals [6, 7], on sensor
response. Two different types of sensor−gas mixture
interaction may form the basis of the above process,
namely, (i) gas molecules interact only with the receptor
centers at the film surface or in the layer bulk; (ii)
sorption occurs in the lattice voids and is not related
directly to the receptor−analyte interaction; these two
processes lead to variation of the layer structure.
Obviously, the above two processes can result in
long-term relaxation. Indeed, the investigations with
surface plasmon resonance and ellipsometry techniques
[8] (that were confirmed with the results of atomic force
microscopy studies) showed that both topography and
thickness of calixarenes sensitive layers varied due to
adsorption of organic molecules, such as toluene,
chloroform, and ethanol [9]. These variations depended
on the gas nature and time of exposure.
To illustrate, the thickness of thin C[4]A layers
increased by 12% at the initial stage of sorption (Fig. 1).
After exposure to chloroform vapor for over 30 min., the
layer thickness increase was 6÷8% the initial thickness
value. The above process seems to be due to film
structure variation because of solid structure relaxation.
The atomic force microscopy studies showed that
topography of thin (≈100 nm) calixarenes layers changed
(see Fig. 2) after their surface was exposed to saturated
vapors of solvents [10]. Thus, in some cases, interaction
of thin organic layers with solvent vapors leads to
undesired processes (from the viewpoint of stability of
the gas sensor parameters). It is obvious that the
properties of such layers in what concerns their
interaction with different gaseous analytes will vary.
2. Experimental procedure
A fully automated quartz crystal microbalance set
designed at ISP, Kyiv (8 channels, 10 MHz AT-cut
crystals) with a measurement interval of 1 s was used in
our experiments [7]. The measuring procedure involved
the following stages: argon circulation; brandy
vapor−argon mixture circulation (bubbling, 12 ml of
Semiconductor Physics, Quantum Electronics & Optoelectronics, 2008. V. 11, N 3. P. 240-244.
© 2008, V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
241
0 10 20 30 40 50 60
0.06
0.08
0.10
0.12
Th
ic
kn
es
s r
el
at
iv
e
ch
an
ge
, ∆
d/
d
Exposure time, min
Fig. 1. Variation of thickness of deposited C[4]A layers with
time of exposure to chloroform vapor (layer thickness of
60−70 nm).
probe, rate 50 ml/min, and temperature of 36±0.3 °С);
argon circulation; ethanol−argon mixture circulation, if
any. Thin films of C[n]A (tret-butylcalyx[n]arene, n = 4,
6, 8), tetracene, pentacene, phtalocyanine (H2Pc),
dibenzotetraazaannulene (H2TAA) and tetramethyl-
dibenzotetraazaannulene (H2TMTAA) were prepared by
thermal sputtering onto the metal electrodes of QCM
plates [11]. Thin films (about 100 nm) were prepared by
thermal sputtering in vacuum (VUP-5М, pressure of
5×10-4 Pa, temperature of 297 ± 2 К, deposition rate of
0.1 nm/min).
Five different samples were studied, namely, those
of ethanol and four brandy sorts (State Standard
ГОСТ13741) produced by Public Corporation APF
“Tavriya” (Nova Kakhovka, Ukraine): “Borysphen”,
“Georgiyivs’kyi”, “Oleksandriys’kyi” and “Tavriya”
(from here on B, G, O, and T, respectively).
Restoration of sensitive layers using ethanol vapor.
If several (three-four) experiments with vapors of drinks
(e.g. brandy) having volatile organic components were
performed in succession, then irreproducibility of the
results of measurements due to surface contamination
was observed It is obvious that image recognition in
such situation is impossible. Indeed, a comparative
analysis showed that direct measurements of brandy
samples without additional treatment did not give
unambiguous solution to the problem of chemical image
recognition.
Bearing in mind that the results obtained could be
due to incomplete surface cleaning as well as surface
modification, we tested the following restoration
procedure for sensitive coating parameters. Between the
successive experiments with drink samples, the sensors
were cleaned by exposing to gas-vapor mixture of
saturated ethanol vapor and argon during 5 min. It was
supposed that this procedure could ensure not only
removal of organic molecules from the sensitive layer
but restoration of the surface initial state as well. We
also made check measurements without sensor surface
treatment with vapor.
a
2.0µm
b
Fig. 2. Topography of calixarene C[4]A surfaces before (left)
and after (right) treatment in chloroform vapor.
An analysis of chemical images obtained
demonstrated complete coincidence of the images for
three successive experiments (Fig. 3). One can see that a
simple procedure of sensor array treatment with ethanol
vapor can improve substantially system restoration thus
ensuring formation of stable chemical images [12].
“Electronic Nose” is intended, first of all, for
recognition of various multicomponent chemical
systems. A necessary condition for this is improvement
of discriminating ability of sensor arrays. However,
possible ways for sensor array optimization in what
concerns required selectivity of sensitive layers and
parameterization of sensor response still remain not
understood adequately. Another important problem is
that of “extra” sensors with small information content.
The reason is that such “extra” sensors introduce
“additional information noise”. Indeed, due to diversity
and complexity of interactions between the
multicomponent chemical systems and sensor elements,
the data obtained for such systems are multivariate, so
one cannot say in advance how important a given
property is for identification of a certain analyte (or class
of analytes). In this case, the problem of recognition
from multivariate data gets first priority. In other words,
it becomes necessary to optimize grouping from the
viewpoint of the best correspondence of the results
obtained to the final aim of recognition.
Semiconductor Physics, Quantum Electronics & Optoelectronics, 2008. V. 11, N 3. P. 240-244.
© 2008, V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
242
0
45135
80
225
270
315H2TMTAA
H2TAA
H2Pc
C[8]A
C[6]A
C[4]A
pentacene
tetracene
first, second, third experiments
H2TAA
H2Pc
C[8]A
C[6]A
C[4]A
pentacene
tetracene
H2TMTAA
first, second, third experiments
Fig. 3. Radial diagrams for three measurements of brandy
samples without (a) and with (b) additional treatment of sensor
surface.
3. Results and discussion
Quantitative analysis of recognizing ability of sensor
array. To determine the possibilities for drinks
recognition, we performed quantitative analysis of
identification ability of sensor array. The cluster analysis
methods that apply the fuzzy logic concept [13] were
used as criterion. The input matrix involved the data of
three experiments for each sample (for brandies and
ethanol).
The key element of any optimization procedure is
existence of recognition measure. It enables one to judge
indirectly appropriateness of different sensors in a set
and single out the most informative part of
multidimensional response surface. The Rousseeuw
version of cluster analysis makes it possible to perform
such estimation using the so-called silhouette width s(i)
as parameter [13]. Being a discrimination degree, this
parameter characterizes factually measurement
association with a certain cluster. If the s(i) values are
close to unity, then the data are arranged compactly in a
cluster, especially if all the s(i) values are comparable to
each other for all the cluster elements. Small values of
s(i) indicate that the corresponding measurement lies
between the compact cluster regions. This means that,
remaining within the framework of the fuzzy logic
concept, one cannot conclude to what cluster the
corresponding measurement belongs. Negative value of
s(i) indicates that, most probably, the corresponding
measurement was erroneously associated with this
cluster, i.e., it cannot be correctly associated with any of
clusters formed from the data set under consideration.
Minimization of the corresponding objective function for
a certain combination of variables and observations
makes it possible to calculate s(i) for each observation,
as well as its mean value S(i) for the whole data set.
The iV-parameter (the normalized area below the
kinetic curve) was calculated for various time intervals.
A comparative analysis led to the conclusion that in this
case application of that parameter enables one to
improve considerably the degree of recognition for
different types of samples [14]. Just such kind of sensor
response parameterization was applied in our
quantitative analysis.
The iV-parameters for complete set of sensors were
used in different intervals of time: 0.7÷1.2, 1.5÷5.2, and
5.3÷7.8 min. In the 0.7÷1.2 min. interval, no sample was
recognized. (S(i) = 0.45). In the case of 1.5÷5.2 min.
time interval, ethanol was recognized (S(i) = 0.33), while
brandy G and ethanol were recognized in the 5.3÷7.8
min. time interval (S(i) = 0.37). For further analysis, the
iV-parameter at sampling times from 5.3 up to 7.8 min.
was chosen; in this case, the biggest S(i) values (S(i) =
0.37) were observed.
It is known that cluster analysis does not make it
possible to choose optimal set of criteria (types of
sensors) for solving the problem of drinks recognition.
Therefore, the recognition ability was estimated for
different combinations (sets) of sensors. An analysis of
the results of recognition for normalized responses for
five different samples (B, G, O, T, and ethanol) and
different combinations (sets) of three sensors were made
with iV-parameter for 5.3÷7.8 min. time interval. This
enabled us to estimate the effect of sensors on
discrimination ability. It was shown that the best
recognition occurs by sensors with pentacene, C[6]A,
and H2TAA coatings; two samples (brandy G and
ethanol) are recognized in practically all cases. The
sensor set used cannot recognize directly the brandies:
B, O, and T. Thus, for the samples with a dominant
component, recognition ability is lower than that for
other samples under consideration.
It is of interest to compare statistics of different
sensors appearance in those three-sensor sets (formed of
six combinations of C[4]A (2), C[6]А (3), pentacene (1),
H2Pc (2), H2TMTAA (2), tetracene (2), C[8]А (1), and
Semiconductor Physics, Quantum Electronics & Optoelectronics, 2008. V. 11, N 3. P. 240-244.
© 2008, V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
243
H2TAA (4)) that give true recognition. In this case, one
should note essential role of a sensor with H2TAA
coating (almost hydrophobic surface). The reason for
this is that such drinks as brandy have heavy fractions
with big organic molecules.
The sensor set under consideration could recognize
directly and correctly only two objects of five. However,
it is possible to develop a multistage approach. According
to it, if a sample was assigned to a cluster that is common
for several drinks, then further object recognition (within
the cluster) is possible. Bearing this in mind, we analyzed
the three unrecognized samples (brandies B, O, and T)
using different three-sensor sets (Table 1.). In this case,
one can clusterize directly the brandy B, and it is possible
to separate it from two other brandy samples.
When studying statistics of appearance of different
sensors in the three-sensor sets giving true recognition of
three unclusterized samples, five combinations of C[4]A
(2), C[6]А (1), pentacene (4), H2Pc (0), H2TMTAA (5),
tetracene (1), C[8]А (2), and H2TAA (0) were considered.
The optimal set involves pentacene, C[6]А, and
H2TMTAA. Thus, only two unrecognized drinks
(brandies O and T) remain after two stages of recognition.
Table 1. Chemical image recognition of three brandy
samples (“Borysphen” - B, “Oleksandriys’kyi” – O and
“Tavriya” – T) with three-sensor sets.
Sample
recognition
Type of sensor
B O T S(i)
tetracene pentacene H2TMTAA + − − 0.4
pentacene C[4]A H2TMTAA + − − 0.42
pentacene C[6]A H2TMTAA + − − 0.5
pentacene C[8]A H2TMTAA + − − 0.38
C[4]A C[8]A H2TMTAA + − − 0.31
Table 2. Chemical image recognition of two brandy
samples (“Oleksandriys’kyi” – O and “Tavriya” – T)
with three-sensor sets.
Sample
recognition
Type of sensor
O T S(i)
tetracene C[4]A H2TAA + + 0.33
tetracene C[6]A H2TAA + + 0.36
C[6]A H2Pc H2TAA + + 0.25
C[6]A H2Pc H2TMTAA + + 0.25
H2Pc H2TAA H2TMTAA + + 0.25
The results of analysis of the two unrecognized
brandy samples (O and T) made using different three-
sensor sets (the iV-parameters for 5.3÷7.8 min. time
interval) are presented in Table 2. One can see that the
best recognition is achieved using sensors with tetracene,
C[6]A, and H2TAA coatings. In this case, one should
consider five combinations of C[4]A (1), C[6]А (3),
pentacene (0), H2Pc (3), H2TMTAA (2), tetracene (2),
C[8]А (0), and H2TAA (3) when studying statistics of
different sensors appearance in the three-sensor sets
giving true recognition of two unclusterized samples.
Thus, the hierarchical clusterization procedure that
is based, at each stage, on the fuzzy logic concept makes
it possible, in the final analysis, to achieve complete
recognition of all the samples under consideration.
8. Conclusion
Application of sensor cleaning procedure is proposed
that not only ensures removal of organic molecules from
the sensitive layer but makes it possible to restore the
initial state of sensor surface as well. It is shown that
treatment of sensor array with ethanol vapor can
improve considerably system state restoration. This
ensures formation of stable chemical images in the case
of samples with dominant components. It is shown also
that a two-stage clusterization procedure that is based, at
each stage, on the fuzzy logic concept enables one to
make complete recognition of all the samples
considered.
Acknowledgements
The author is grateful to Dr. B. Snopok for helpful
discussions.
References
1. K.J. Albert, N.S. Lewis, C.L. Schauer,
G.A. Sotzing, S.E. Stitzel, T.P. Vaid, D.R. Walt,
Cross-reactive chemical sensor arrays // Chem.
Rev. 100, p. 2595 (2000).
2. B.A. Snopok, I.V. Kruglenko, Multisensor systems
for chemical analysis: state-of-the-art in Electronic
Nose technology and new trends in machine
olfaction // Thin Solid Films 418(1), p. 21-41
(2002).
3. N. Paulsson, E. Larsson, F. Winquist, Extraction
and selection of parameters for evaluation of breath
alcohol measurement with an electronic nose //
Sens. Actuators A: Phys. 84(3), p. 187-197 (2000).
4. L. Lvova, R. Paolesse, C. Di Natali, A. D’Amico,
Detection of alcohols in beverages: An application
of porphyrin-based Electronic Tongue // Sens.
Actuators B: Chem. 118(1), p. 439-447 (2006).
5. T. Nakamoto, Y. Nakahira, H. Hiramatsu,
T. Moriizumi, Odor recorder using active odor
sensing system // Sens. Actuators B: Chem. 76(1),
p. 465-469 (2001).
6. Y. Liu, X. Yu, R. Zhao, D.-H. Shangguan, Z. Bo,
G. Liu, Quartz crystal biosensor for real-time
monitoring of molecular recognition between
protein and small molecular medicinal agents //
Biosensors & Bioelectronics 19(1), p. 9-19 (2003).
7. F. Jalali, E. Arkan, G. Bahrami, Preparation of a
gabapentin potentiometric sensor and its
application to pharmaceutical analysis // Sens.
Actuators B: Chem., 127(1), p. 304-309 (2007).
8. R.M.A. Azzam, N.M. Bashara, Ellipsometry and
Polarized Light, North-Holland, Amsterdam, 1980.
Semiconductor Physics, Quantum Electronics & Optoelectronics, 2008. V. 11, N 3. P. 240-244.
© 2008, V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
244
9. J. Simon, J.-J. Andre, Molecular Semiconductors,
Springer-Verlag, Berlin (1985).
10. Y.M. Shirshov, B.A. Snopok, O.V. Rengevych,
V.I. Kalchenko, A. Coleman, Relaxation of
nanostructured molecular materials under the
influence of solvent vapors, in Frontiers of
Multifunctional Nanosystems, Eds. E. Buzaneva, P.
Scharff, p. 347-368, Kluwer Academic Publishers
(2002).
11. I.V. Kruglenko, B.A. Snopok, Y.M. Shirshov,
E.F. Venger, Digital aroma technology for
chemical sensing: temporal chemical images of
complex mixtures // SQO 3(4), p. 529-541 (2000).
12. I.V. Kruglenko, B.A. Snopok, Sensor renewal
using displacement reactions, in 2nd Workshop of
the Second Network on Artificial Olfactory
Sensing (NOSEII), New concepts for chemical
sensing, Linköping University, Sweden, May 18-
21, 2003.
13. L. Kaufman, P.J. Rousseeuw, Finding Groups in
Data: An Introduction to Cluster Analysis, Wiley,
New York (1990).
P.J. Rousseeuw, Silhouettes: A graphical aid to the
interpretation and validation of cluster analysis // J.
Comput. Appl. Math. 20, p. 53-65 (1987).
|