Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products
In the present work, the use of the cluster analysis method in the “fuzzy logic” concept for the optimization of the cross-selective sensor arrays (“electronic nose”, EN) is considered. This approach enables to purposefully form the sensor arrays with definite chemical functionality optimized for th...
Gespeichert in:
| Datum: | 2004 |
|---|---|
| Hauptverfasser: | , , , |
| Format: | Artikel |
| Sprache: | English |
| Veröffentlicht: |
Інститут фізики напівпровідників імені В.Є. Лашкарьова НАН України
2004
|
| Schriftenreihe: | Semiconductor Physics Quantum Electronics & Optoelectronics |
| Online Zugang: | http://dspace.nbuv.gov.ua/handle/123456789/118178 |
| 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: | Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products / I.V. Kruglenko, B.A. Snopok, Yu.M. Shirshov, F.J. Rowell // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2004. — Т. 7, № 2. — С. 207-216. — Бібліогр.: 49 назв. — англ. |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-118178 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-1181782017-05-30T03:02:53Z Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products Kruglenko, I.V. Snopok, B.A. Shirshov, Yu.M. Rowell, F.J. In the present work, the use of the cluster analysis method in the “fuzzy logic” concept for the optimization of the cross-selective sensor arrays (“electronic nose”, EN) is considered. This approach enables to purposefully form the sensor arrays with definite chemical functionality optimized for the solution of the specific applied problems. The criteria of the optimization of the sensor response, number and type of the sensor elements are considered with the goal to improve the classification of widely used pharmaceutical products. The optimization of EN array in the kinetic mode and selection of the most informative part of the sensor response enabled to reduce the analysis time, and also the number of sensors in array, to improve the discriminatory capability of the whole array. Being based on the analysis of response kinetic peculiarities, the physical mechanisms determining the peculiarities of adsorption-desorption processes at the interface have been considered. 2004 Article Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products / I.V. Kruglenko, B.A. Snopok, Yu.M. Shirshov, F.J. Rowell // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2004. — Т. 7, № 2. — С. 207-216. — Бібліогр.: 49 назв. — англ. 1560-8034 PACS: 07.07.Df http://dspace.nbuv.gov.ua/handle/123456789/118178 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 |
In the present work, the use of the cluster analysis method in the “fuzzy logic” concept for the optimization of the cross-selective sensor arrays (“electronic nose”, EN) is considered. This approach enables to purposefully form the sensor arrays with definite chemical functionality optimized for the solution of the specific applied problems. The criteria of the optimization of the sensor response, number and type of the sensor elements are considered with the goal to improve the classification of widely used pharmaceutical products. The optimization of EN array in the kinetic mode and selection of the most informative part of the sensor response enabled to reduce the analysis time, and also the number of sensors in array, to improve the discriminatory capability of the whole array. Being based on the analysis of response kinetic peculiarities, the physical mechanisms determining the peculiarities of adsorption-desorption processes at the interface have been considered. |
| format |
Article |
| author |
Kruglenko, I.V. Snopok, B.A. Shirshov, Yu.M. Rowell, F.J. |
| spellingShingle |
Kruglenko, I.V. Snopok, B.A. Shirshov, Yu.M. Rowell, F.J. Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products Semiconductor Physics Quantum Electronics & Optoelectronics |
| author_facet |
Kruglenko, I.V. Snopok, B.A. Shirshov, Yu.M. Rowell, F.J. |
| author_sort |
Kruglenko, I.V. |
| title |
Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products |
| title_short |
Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products |
| title_full |
Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products |
| title_fullStr |
Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products |
| title_full_unstemmed |
Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products |
| title_sort |
multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products |
| publisher |
Інститут фізики напівпровідників імені В.Є. Лашкарьова НАН України |
| publishDate |
2004 |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/118178 |
| citation_txt |
Multisensor systems for gas analysis: optimization of arrays for classification of pharmaceutical products / I.V. Kruglenko, B.A. Snopok, Yu.M. Shirshov, F.J. Rowell // Semiconductor Physics Quantum Electronics & Optoelectronics. — 2004. — Т. 7, № 2. — С. 207-216. — Бібліогр.: 49 назв. — англ. |
| series |
Semiconductor Physics Quantum Electronics & Optoelectronics |
| work_keys_str_mv |
AT kruglenkoiv multisensorsystemsforgasanalysisoptimizationofarraysforclassificationofpharmaceuticalproducts AT snopokba multisensorsystemsforgasanalysisoptimizationofarraysforclassificationofpharmaceuticalproducts AT shirshovyum multisensorsystemsforgasanalysisoptimizationofarraysforclassificationofpharmaceuticalproducts AT rowellfj multisensorsystemsforgasanalysisoptimizationofarraysforclassificationofpharmaceuticalproducts |
| first_indexed |
2025-07-08T13:31:16Z |
| last_indexed |
2025-07-08T13:31:16Z |
| _version_ |
1837085735973289984 |
| fulltext |
207© 2004, V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
Semiconductor Physics, Quantum Electronics & Optoelectronics. 2004. V. 7, N 2. P. 207-216.
PACS: 07.07.Df
Multisensor systems for gas analysis: optimization of arrays
for classification of pharmaceutical products
I.V. Kruglenko, B.A. Snopok, Yu.M. Shirshov, F.J. Rowell*
V. Lashkaryov Institute of Semiconductor Physics, National Academy of Sciences of Ukraine
41, prospect Nauky, 03028 Kyiv, Ukraine
*Centre for Pharmaceutical and Environmental Analysis, School of Science, University of Sunderland, UK
Abstract. In the present work, the use of the cluster analysis method in the �fuzzy logic�
concept for the optimization of the cross-selective sensor arrays (�electronic nose�, EN) is
considered. This approach enables to purposefully form the sensor arrays with definite chemi-
cal functionality optimized for the solution of the specific applied problems. The criteria of
the optimization of the sensor response, number and type of the sensor elements are consi-
dered with the goal to improve the classification of widely used pharmaceutical products. The
optimization of EN array in the kinetic mode and selection of the most informative part of the
sensor response enabled to reduce the analysis time, and also the number of sensors in array,
to improve the discriminatory capability of the whole array. Being based on the analysis of
response kinetic peculiarities, the physical mechanisms determining the peculiarities of ad-
sorption-desorption processes at the interface have been considered.
Keywords: electronic nose, identification, multisensor system, cluster analysis, fuzzy logic.
Paper received 17.03.04; accepted for publication 17.06.04.
1. Introduction
Development and investigation of the chemical and bio-
chemical sensors as a separate direction on the boundary
of physics, chemistry and biology is intensively devel-
oped in recent years [1�3]. The important place among
such systems is occupied by selective sensors aimed at the
single components of gas mixtures � sensors of oxygen,
hydrogen as well as other constituents of gas mixtures;
these systems use the concept of specific sensors built on
the principal �one gas � one sensor�. Despite wide spec-
tra of available sensors, they can�t provide the respective
description of the multicomponent mixtures that are char-
acteristic for the chemical environment of the natural
origin [4,5]. Besides, the creation of highly selective coat-
ings sensitive only to the specific molecules, is practi-
cally impossible, because the compounds presenting the
same chemical class have functional groups of the simi-
lar spatial structure with the close physical and chemical
properties.
This put forward as a task of paramaunt importance
to create multisensor systems for the analysis of the com-
position of the multicomponent chemical media (MCM),
which can be used for the ecological monitoring, control
of the technological processes, analysis of the products
quality and for the medical and biological investigations
[6�9]. Systems of that type are intensively developed in
many countries, with the aim of their use for the control
of the media, which are vitally important for human be-
ings. Thus, the medicine and pharmacy are one of the
largest markets for such systems, which require the devel-
opment of the new intelectual diagnostics. Indeed, even
B.C. the possibility of using the odors for the diagnostics
of different deseases was known. Thus, the development
of the intellectual gas matrixes and following develop-
ment of the �electronic nose� must provide more wide use
of the �scent� diagnostics in medicine. At the same time,
high effeciency (and, thus, a respective danger in the case
of the unproper use) of the modern pharmaceutical prod-
ucts supposes the necessity of the quality control of the
products and also their identification in the case of the
need, because they can cause latent and evident non-
wanted or adverse effects for patients. Among the tera-
togenic, that is such products that can cause adverse ef-
fects, are such medicine (drugs) as alkaloids, quinine,
tetracycline, sulfanilamides, etc. Thus, the express iden-
tification of the widely used pharmaceutical products is
practically very important.
The �Electronic nose� is designed, first of all, for the
identification of the various MCM. The necessary con-
dition for this is the ability of the multisensor array to
classify MCM: the increase of the discriminating ability
of the array is of primary importance. But the ways of the
optimization of the sensor array towards respective nec-
essary sensitivity of the sensitive layers and ways of the
parametrization of the sensor response still remain with-
208
SQO, 7(2), 2004
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
out answer. Besides, the problem of the �superfluous�
sensors with the small information capacity is very im-
portant due to �additional information noise� of such
sensors [9]. Indeed, in connection with the complex na-
ture of the interaction between MCM and sensor elements,
data on them have multidimensional and diverse charac-
ter, so before analysis it is unclear, how essential is this
or another property for identification of the certain analy-
te or class of analytes. At given conditions, in the first
place is the problem to classify the multidimensional data;
that is the necessity to optimize grouping from the view-
point of the best correspondence between obtained re-
sults and the final target of this classification. In contrast
to many statistical procedures, the methods of the cluster
analysis enable to determine the �most probable solution�,
when the identification problem is still in the descriptive
stage of investigation. This enables to develop the ap-
proach concerning the ways of the optimization of the
sensor array and to give it certain chemical functionality.
The objective of the present work was to consider ways
of optimization of the multisensor arrays for the gas
analysis with the aim to improve the quality of pharma-
ceutical product classification by using the methods of
the cluster analysis that are based on the concept of the
�fuzzy logic�.
2. Multisensor systems of gas analysis:
a general approach
The approach to the analysis of the multicomponent
chemical mixtures of the unknown composition can be
borrowed from the analysis of the principles, which are
habitual to the life support systems of the biological or-
ganisms, in particular, olfaction. Olfaction system of the
animals and human beings is the unique instrument for
the analysis of the chemical components in air [11�13].
This system is characterized by: (i) high rapidity of the
analysis (milliseconds); (ii) wide sensitivity range (up 9
orders in concentration); (iii) low detection limits; (iv)
memory possibility, possibility to distinguish complex
mixtures of compounds and (v) to identify fragrant com-
pounds in the complex mixtures; (vi) miniature sizes.
The idea of MCM detection consists in the simultane-
ous analysis of signals from the system of sensitive ele-
ments, each element has unique selectivity profile. That
is, MCM can be described not by the sum of its constitu-
ent components, but some model with the characteristic
for each MCM set of parameters � by a chemical image
(CI) [14, 15]. Instrumental system for the formation of CI
with the consequent identification using the methods of
pattern recognition are called as �electronic nose�, EN
[16, 17]. The basis of EN is the array of sensors with the
sensitive layers possessing different chemical functional-
ity. Such systems creates �image� in terms of the sensi-
tive element signal values in the form of a multidimen-
sional surface of responses. The approach enables to get
over the principal limitations of the specific sensors and
provide unlimited quantity of the possible �images� of
different components. Between the probe image in the
�chemical� base (concentration of the separate compo-
nents on axes) and its analog in the response space of the
sensor array one can observe the one-to-one correspond-
ence (Fig. 1). Because the dimensionality of the response
space is sufficiently smaller than the dimensionality of
the �chemical basis�, the sensors must be weakly selec-
tive and have a wide dynamical range of their signal
change. In this case, the commonly used sensitivity level
to the wide spectrum of the chemical compounds will be
provided, and the necessary number of the different com-
binations in the response space will be achieved [18, 19].
The first intellectual model of EN, which included
elements of pattern recognition, was developed in 1982
by Persaud and Dodd [20]. After that the different vesions
of EN, which were based on various physical converters,
were developed, which enabled to obtain simple, com-
pact, completely automated devices suitable to solve sev-
eral applied problems [6, 7, 16, 17]. But, as a whole, the
question concerning formation of the necessary for the
specific application, of the chemical functionallity of the
sensor array was open, due to, at first place, problem of
the provision of the uniqueness for the each separate im-
age of MCM of different origin.
3. Piezoelectric converters for EN systems
Among physical converters of various types used in EN
devices, piezoelectric converters are among most attrac-
tive, because these enable to develop approach, which
supposes separation of the chemical interaction processes
and its physical transformation into the respective sensor
response [21]. The dependence of the resonant frequency
f of an acoustic oscillator on the mass of substance ∆m
that is on its surface is used in a Quartz Crystal Micro-
balance (QCM) to register intermolecular interactions.
The linear dependence between the resonant frequency
shift f of QCM and m, which is supposed in many cases
[22], in reality is idealized representation obtained in
the conditions that the density and viscosity of the sensi-
tive layer do not differ from material of the converter
itself and do not change during its interaction with analyte.
But in such cases, when the hydrodynamic interaction
between the physical converter and environment can be
neglected (mainly in measurements in a gas phase), and
to consider a sensitive layer as solid, then the mass change
∆m with the change of the oscillation frequency ∆f of the
piezocrystal resonator should be described by the expres-
sion [23]:
⋅⋅+⋅⋅+∆−⋅⋅=∆ T
dT
df
f
P
dP
df
ff
f
dSm QQ
000
11ρ (1)
where S is the electrode area; ρQ and dQ � density and
thickness of the quartz plate with the surface layer; f0 -
natural oscillation frequency of the piezoelement. If it is
possible to neglect temperature and pressure changes
during the experiment (df/dP = 0, df/dT = 0), then the
equation (1) is transformed into the well-known Sauebrey
equation [24]:
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
209SQO, 7(2), 2004
SmCf Q /∆⋅−=∆ (2)
where CQ is the integral resonator sensitivity to the addi-
tional mass on the surface. That is, the crystal oscillation
frequency is linearly changed with the mass change on its
surface. For the ÀÒ-cut of the quartz crystal with main
frequency 10 MHz and 0.25 cm2 area, the converter sen-
sitivity is about 1 ng⋅Hz�1. Such ratio is valid only for the
small mass changes ∆m on the converters surface, which
is determined by sorption of the analyte by the sensitive
layer. This mass depends on the gas volume V, which
flows over the surface of the piezocrystal and can be writ-
ten as:
V = q ⋅ t = ∆M /C or ∆M = C ⋅ q ⋅ t (3)
where Ñ is the analyte concentration in the gas flow, ∆M �
analyte mass in the gas flow, q � gas flow velocity, and
t � time. If γ is the effective coefficient of analyte capture
by the sensitive layer (∆m = γ ⋅ ∆M), then in the region
far from saturation, the converter frequency shift will de-
pend on gas flow parameters in the following way (equa-
tion (2) and (3)):
∆f = (CQ / S) ⋅ (γ ⋅ C ⋅ q ⋅ t) (4)
Thus, the frequency shift of the crystal oscillations
depends on time, analyte concentration and carrier gas
velocity. If the analysis time will be fixed under the con-
stant velocity of the gas flow, then the change of the crys-
tal frequency is directly proportional to the analyte con-
centration in the gas flow. It is evident that the various
processes can lead to the violation of the conditions in
which the equation (4) was obtained. Among them � ad-
sorption of the mixture on the walls, fluctuations of the
temperature, pressure, carrier gas velocity, etc. The
standartization of the analysis conditions, choice of the
respective materials, and the increase of the wall tem-
perature are widely used methods to solve these prob-
lems. Besides, in these cases when the flowing measure-
ments regime is used, the contribution of these factors is
negligible.
4. Sensitive layers for EN
Piezocrystalline detector is the powerfull analytic de-
vice; this is clear from the proportion for the frequency
change to the analyte concentration (about 1 ng⋅Hz�1).
Taking into account that the piezosensor is non-specific
device, the detector reveals the frequency change due to
mass deposition of any substance onto the surface. Thus,
for the formation of the selectivity profile of the sensitive
element, it is necessary to choose the coatings that will be
sensitive to the respective analytes. In this case, the pro-
cedures of the formation of sensitive architectures with
the receptor centers on the surface of the physical con-
verters have a paramount importance because of the ne-
cessity of the molecula recognition principles realiza-
tion in the case, when receptor centers of various origin
and structure are fixed in the two-dimensional matrix at
the phase interface (homogenic-heterogenic processes).
Some information concerning the ways to solve this
problem is provided by odor systems developed on the
base of analysis of human beeings and animals scent pe-
culiarities. For example, in accordance with the Amoore
system, the �form� of the molecule determines its odor
[25]. This stereochemical concept has seven �primary�
odors, for each of them the sizes of the receptor �holes�
were calculated. Thus, the receptor centers that provide
several contacts of different nature with the respective
accomodation in the space of the pair receptor-analyte,
are capable to provide the necessary profile of the selec-
tive layer, in general. Taking into account the ability of
the organic materials concerning formation of various
types of intermolecular interactions and their diversity,
exactly the organic compounds are most attractive for
the formation of the sensor arrays of a new generation.
Additional information concerning the use of organic
materials in sensor technique can be found in [26�28].
5. Formation and recognition of CI
The above considered ways of the creation of intellectual
sensor systems assume the presence of interaction of
MCM target components with sensitive layers, the pecu-
liarities of which unambigously characterize the given
type of the product. But the response of the sensors also
depends on the procedure of the probe preparation and
analysis of the samples prehistory, irregularity of the gas
flow, etc., can lead to the unpredicted shape of the kinet-
ics curve at the initial stage [10, 29]. Besides, the part of
the signal that corresponds to the stationary level, also
doesn�t have additional information about MCM [30].
Indeed, the increase of the identification capability of
the �electronic nose� in the dynamic regime was shown
using the linear classificator � the principal component
analysis. Neuron networks (non-linear classificator) con-
firmed this result. As was supposed, for the different
classificators the best results on the quality of recogni-
tion are achieved when the initial part and stationary
state are excluded from the input data, that is the kinetics
information enables to provide the required level of the
0
2
4
6
8
10 0
2
4
6
8
10
2
4
6
8
10
0
2
4
6
8
10 0
2
4
6
8
10
Mainframe of sensor arrays
Chemical space
substance 3
su
b
st
a
n
c
e
1
su
bsta
nce 2
Response space
se
n
so
r
1
sensor 3
se
nso
r 2
2
4
6
8
10
Fig.1. General functioning procedure of the multisensor arrays.
210
SQO, 7(2), 2004
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
uniqueness of the selectivity profile as a single sensor
element as well as of the chemical image as a whole [29�
31]. At the same time, the developed approaches don�t
enable their effective use as an instrument to optimize the
sensor array: absence of the simple and understandable
classification standard and rigid requirements towards
quantity of the necessary set of data essentially limit pos-
sibilities of these methods as a useful experimental tools
in the given case. Let us consider the possible ways for
the solution of this problem.
Classification means here division of some set of
objects or observations into uniform, in certain sense,
groups � clusters, elements of which are similar, at the
same time between clusters the qualitative differences are
observed [32]. Thus, the objective of the cluster analysis
is the separation of the internal structure, in general case
multidimensional data, and attributing each of them from
the given set of objects to one of the class (sort). The
cluster analysis can be considered as one of the methods
of data compression, when to the big quantity of images
(for example, observations, which differ as a consequence
of the uncontrolled analysis conditions) is made corre-
sponded to the limited number of clusters, using some
function of mutual correspondence. The classical meth-
ods of the cluster analysis lead to separation of the data
set into clusters with clearly determined boundaries; this
means that irrespectively to the quality of the input data
they must be related to the certain class. In fact, this means
that the part of the data will be classified incorrectly, �
this is of course impermissible for a number of branches
such as medical diagnostics or the life support control at
big industrial enterprises with the continuous production
cycle. In order to prevent such inaccurate classification,
the methods based on the concept of �fuzzy logics� were
developed, where the concept of diffuse or fuzzy set is
used. In contrast to the classification according to the
prototype, �indistinctness� arises in the situation when
attributing to the certain category is assumed true only
to a certain degree. The source of such �typicality� can
be graduation concepts (a priori information about classes
� separation methods) or possibility to distinguish �more�
or �less� succesfull solutions by the analysis of the algo-
rithm peculiarities of the standard or function of the class
identity (hierarchy approaches). It is necessary to note
that just the separation methods are a powerful instru-
ment, when the task is in the optimization of the acquire-
ment procedure optimization of the multidimensional data
set with the aim of their following use in the procedures of
the image recognition.
The most important element of any optimization pro-
cedure is the availability of the classification degree;
which enables to evaluate expediency of the use of differ-
ent sensors in the array and enables to separate the most
informative part of the multidimensional response sur-
face. The variant of cluster analysis in the Rousseeuw
version enables to obtain such estimation, by using the
so-called s(i) (silhouette width) as a parameter [33, 34].
The minimization of the respective target function for
the certain combination of the variables and observations
enables to calculate s(i) both for a single observation and
for its average value S(i) in the whole data set. Because
s(i) is in fact the characteristic of the attribution to the
certain cluster, then the s(i) close to unity means that
data in the cluster are compactly placed, especially if all
s(i) values for all the cluster elements have comparable
values. Small s(i) values testify that the given observation
lies between regions of the compact clusters, � that is
remaining within the concept of the �fuzzy logic� it is
impossible to make a conclusion to which cluster the given
measurement belongs.Thus, despite lowering the classi-
fication ability of the methods based on the given con-
cept (because the quantity of the solutions on the attribu-
tion to the certain cluster is decreased), number of the
wrong classifications is practically reduced to zero. Nega-
tive s(i) value points that the given observation most prob-
ably is wrongly attributed to the given cluster � that is it
cannot be correctly placed to any of the clusters formed
from the considered observation.
Presented short review of the cluster analysis peculi-
arities based on the concept of �fuzzy� logic enables to
propose the method for the optimization of the multisensor
arrays for specific application, using the distribution of
s(i) as the criterion. Thus, the optimal combination of the
sensitive elements and procedures of the parametrization
of the sensor response must provide the formation of the
uniform clusters in the response space, which are charac-
terized by the high and close s(i) values both for single
observations within cluster, and the mean S(i) value for
all observations as a whole. We will consider the realiza-
tion of this approach for the classification of three types
of medical compounds.
6. Experiment
Conceptually the development of the EN systems sup-
poses the solution of the four main problems, namely: (i)
synthesis of the receptor centers with the specified chemi-
cal functionality; (ii) their integration with a physical
converter; (iii) complete integration of the separate com-
ponents of the system with taking into account the pecu-
liarities of the probes preparation and carrying out of the
measurements; (iv) parametrization of the response of the
sensor array, formation of the data bases of standard-
ized chemical images. In this case, the principle question
which is applied for all development stages of such sys-
tems is the search for the ways of the informativity in-
crease of the separate observation and the efficiency of
the conversion of the chemical information into the math-
ematical model.
Materials
Calixarenes, tert-bytil-calix[4]aren (Ñ[4]A), tert-bytil-
calix[6]aren (C[6]A) and tert-bytil-calix[8]aren (C[8]A)
were kindly provided by professor Kalchenko V.I. (Insti-
tute of Organic chemistry, NAS Ukraine, Kyiv). Linear
polyacenes: pentacen, tetracen, and phthalocyanine
(H2Pc) were kindly provided by professor Vertsimakha
Ya.I. (Institute of Physics, NAS Ukraine, Kyiv). Annu-
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
211SQO, 7(2), 2004
lenes: dibenzotetraazaannulene (H2TAA) and its methyl
substituted derivative (H2TÌÒAA) were kindly provided
by professor Lampeka Ya.D. (Institute of physical chem-
istry, NAS Ukraine, Kyiv). All compounds were used
without additional purification. The films of organic com-
pounds were obtained by thermal vacuum evaporation
(VUP-5M, pressure 5⋅10�4 Pa) on the metal electrodes of
PEC (from one side). The mean condensation rate was 10
nm/min; temperature of QCM was 297±2°Ê. The thick-
ness of the films during evaporation process was control-
led by the quartz resonator and was equal to 100 nm.
The following three types of the medicines were inves-
tigated: Excedrin®, Bristol-Myers SQUIBB Co. (USA);
Groseptol 480®, Polfa Grodzisk (PL); Ampicillini tri-
hydras-Darnitsa®, Darnitsa (UA). The experimental vol-
ume corresponded to the one pill crushed into powder.
The gas �image� of the pharmaceutical compounds,
where inorganic salts, cellulose and its derivatives are
usually used as fillers, is mainly determined by the active
substance (usually organic substance with the small vola-
tility), ingredients (such as benzol acid, mineral oils,
etc.), and also aromatizators. In the case of the absence
of the latter ones and sufficiently low volatility of the
active components, the gas �image� of the compound is
characterized by the low concentration of the target-ori-
ented substances in the gas phase and by the change of CI
with time due to the change of (both qualitative and quan-
titative) a gas phase composition over the solid probe.
Respectively, all measurements were carried out only on
the samples dispersed immediatly after unpacking before
the analysis.
Experimental setup
Multisensor QCM analyzer of the gas mixtures contain:
(i) temperature controlled measurement camera with the
sensor matrix of the flowing type; (ii) quartz generators
block; (iii) block of the frequency measurement and RS232
sequential interface constructed on the base of a specia-
lized microprocessor (AT89C2051); (iv) generator of the
gas mixtures; (v) system collection and processing of the
information on the base of personal computer [32]. The
system temperature was maintained at 37±0.3°Ñ level,
flowrate of the carier gas (argon) was about 180 ml/min.
The measurement procedure included the following
stages: gas circulation up to the frequency stabilization
of the sensors (±3 Hz); circulation of the vapor-gas mix-
Table 1. Classification ability of the triple sets for three types of the pharmaceutical products. Data set includes iV-parameters for
1.4�5.2 min.; IC and S(i) are presented only for the sets with adequate classification. Sets with partial classification are not shown
Sensor name Ampicillini® Excedrin® Groseptol 480® IC S(i)
tetracene pentacen C[4]A + + + 96.61 0.59
tetracene pentacen C[6]A + + + 96.43 0.6
tetracene C[4]A C[8]A + + + 89.62 0.41
tetracene C[4]A Pc + + + 91.38 0.53
tetracene C[4]A H2TMTAA + + + 84.45 0.48
tetracene C[6]A C[8]A + + + 91.23 0.41
pentacen C[4]A C[6]A + + + 97.66 0.69
pentacen C[4]A C[8]A + + + 89.23 0.56
pentacen C[4]A Pc + + + 96.58 0.7
pentacen C[4]A H2TAA + + + 87.49 0.66
pentacen C[4]A H2TMTAA + + + 92.62 0.52
pentacen C[6]A C[8]A + + + 90.37 0.55
pentacen C[6]A Pc + + + 97.01 0.73
pentacen C[6]A H2TMTAA + + + 92.98 0.44
C[4]A C[6]A Pc + + + 97.54 0.62
C[4]A C[6]A H2TAA + + + 98.75 0.47
C[4]A C[6]A H2TMTAA + + + 97.63 0.48
C[4]A Pc H2TAA + + + 87.06 0.55
C[4]A Pc H2TMTAA + + + 96.69 0.47
C[6]A Pc H2TAA + + + 90.52 0.61
C[6]A Pc H2TMTAA + + + 96.42 0.41
C[4]A H2TAA H2TMTAA + + + 81.82 0.43
....................
C[6]A C[8]A H2TMTAA � � � � �
C[8]A Pc H2TAA � � � � �
C[8]A Pc H2TMTAA � � � � �
Pc H2TAA H2TMTAA � � � � �
212
SQO, 7(2), 2004
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
ture; purging by carrier gas up to the restoration of the
initial frequency value of the QCM. The scheme of the
measurement part of the device is presented in Fig. 2.
Data processing
The responses of the sensors based on the piezoelectric
converters can be characterized by the value of the re-
sponse frequency (in the different moments of time), its
change rate, integral characteristics and their combina-
tions. Respectively, the formation of CI was carried out
using three parameters: frequency change rate of the sen-
sors Λ, saturation level ∆fs and iV-parameter. The Λ pa-
rameter was calculated at the initial part of the kinetic
curve and ∆fs was determined by the averaging of the
experimental values after reaching the stationary state
of the response; iV-parameter was calculated at different
times tth by numerical integrating of the experimental
dependencies with the following division of the value of
integral on tth values [17, 31]:
( ) ( )∫∆
−
=
2
1
11
21
1
,
t
t
dttf
tt
ttiV . (5)
Taking into account, that CI should not depend on
absolute values of the sensor responses, CI probes were
normalized by unity (that is the sum of the responses for
all sensors for each analysis was equal to unity). To proc-
ess of the experimental data, the programs Origin
(MicroCal. Software, Inc) were used. Statistical analysis
was carried out using S-PLUS 2000 (Math. Soft, Inc).
7. Results
Analysis of the sensor responses for these three types of
pharmaceutical compounds (Fig. 3) enabled to attract
attention to the following peculiarities in the kinetic be-
haviour of the curves: (i) for all the samples, the curves
are monotonic; (ii) the saturation levels are similar for
some sensors; (iii) response of the sensors respectively to
their values is differently changed with time for different
sensors; (iv) transition to the equilibrium state is achieved
after various times for different sensors. The qualitative
comparison of the sensor response with the aim to find
more stable and specific data parametrization for CI for-
mation, with the use of the signal growth rate at the ini-
tial part Λ and the saturation level ∆fs for CM formation
didn't enable to obtain the unique CI like to many other
cases [29-31]. As the use of these parameters didn't en-
able even to qualitatively identify these three samples,
the iV-parameter was calculated, because its use have
positive effect on the degree of the identification for dif-
ferent types of sensors in many cases.
As iV-parameter can be calculated for different time
intervals, initially it was necessary to determine the most
informative part of the kinetic curve, using the degree of
the clusterization as the criterion. Data matrix for clus-
ter analysis included iV-parameters for three samples for
the each type of drugs; variants of the paramerization
were considered for time intervals: 1�2; 1.3�2.6;
1.4�5.2; 1.1�6.8; 5�7 min. In Fig. 4 the dependencies
of the average value S(i) for the different time intervals
are presented. As can be seen from the figure at time dis-
cretion from 1.4 to 5.2 min., the highest S(i) values are
observed. Moreover, in this case single s(i) values within
classes have approximately equal values (Fig. 5). It is
interesting that the use of the wide interval of the input
data (1.1�6.8 min.) provide lower classification ability
in comparison with the best interval in the region of the
strongest signal change of sensors (Figs 3 and 5). Thus,
for the following analysis the interval time between 1.4
and 5.2 min of the sensor response was chosen.
The analysis of the kinetic curves in Fig. 3 shows that
some curves have similar character of relaxations (both
for the different sensors for the same sample and for one
sensor but three various samples). This testifies that not
only optimal time but also quantity and type of the sen-
sors can essentially influence the classification task. That
is why, it is important to ascertain how the discrimina-
tory ability depends on the type of the sensor coating.
Taking into account the statistical character of the task,
it is relatively hard to establish the tendency of the influ-
ence at a large number of variables (sensors). Therefore,
the number of sensors in the subset was reduced up to
three. It is understandable that the reduction of the
number of sensors effectively decreases the number of the
possible combinations in the response space. But from
the practical point of view, even for three sensors with the
respective selectivity profile and 10 grades of the signal,
the number of the possible combinations exceeds 104
(above 310). At the same time, for three pharmaceutical
Fig. 2. Scheme of the measurement part of the experimental
setup.
QCM
e-Nose
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
213SQO, 7(2), 2004
compounds under consideration, the decrease of the num-
ber of sensors from 8 up to 3 optimal results in the growth
of S(i) from 0.49 up to 0.73 (Table 1); and uniformity s(i)
inside clusters is increased � the clusters become more
compact (Fig. 5).
As is known, the cluster analysis can't enable at once
to find the optimal set of the sensors. So, the ability to
classify was estimated using various combinations of the
sensors (YES or NO for the right classification, see Table
1). Besides, for the different combinations of three sen-
sors (in all 54 sets) the discriminatory ability was esti-
mated by S(i) value in the case when the classification
was adequate (all YES). Thus, the analysis of the classi-
fication results for the different subsets provided possi-
bility to evaluate the influence of the sensors on the dis-
criminatory capability (Table 1). The highest discrimi-
natory ability was obtained for the set C[6]À, pentacene,
H2Pc. It is interesting to compare the statistics of the en-
try of the various sensors into triple subsets with the right
classification. From these 22 combinations C[4]A (13),
C[6]À (11), pentacene (10), H2Pc (9), H2TMTAA (7),
tetracen (6), C[8]À and H2TAA (4). At the same time, for
the triple sets with S(i) > 0.6 (6 combinations) we have:
C[4]A (4), C[6]A (4), H2Pc (4), pentacene (4) and H2TAA
(2): for three best (S(i) ≥ 0.69) � pentacene (3), H2Pc (2),
C[6]A (2) and C[4]A (2). It should be emphasized that for
the latter three ones the two-dimensional representation
preserves about 97% of information of the initial data sets
(IC, see Table 1). Thus, from the eight sensitive layers, the
sensors which are based on pentacene, C[4]A, C[6]À, H2Pc
and H2TAA coatings are essential for the high discrimi-
natory capability, when tetracene, C[8]À and H2TMTAA
have small informational capacity.
It is interesting to note that the different from the chemi-
cal point of view compounds were included into the best
sets. In respect to this, the question arises if it is possible
to establish the connection of the peculiarities of the mo-
lecular structure of these compounds, their behaviour
under their interaction with the analytes of different ori-
gin with their contribution to the discriminatory capabil-
ity of the multisensor array?
8. Discussion
The considered above testifies that it is the kinetic infor-
mation that enables to obtain the most unique CI for vari-
ous MCM; the presence of the experimental curves ena-
bles to compare them with the theoretical dependencies
within the frames of the certain model of adsorption with
the aim to clarify the mechanism of the process which
proceeds at the interface.
The approximation of the kinetic dependencies, pre-
sented in Fig. 3 by different analytical functions has
shown that all curves are adequately described by the
function of the following type:
0 2 4 6 8
�160
�120
�80
�40
0
Ampicillini trihydras
DarnitsaÒ
Df, Hz
0 2 4 6 8
�160
�120
�80
�40
0 Groseptol 480
Ò
0 2 4 6 8
�160
�120
�80
�40
0
te tra cen e
E x ced r in
p en ta cen e C [4 ]A C [6 ]A
C [8 ]A H 2 P c H 2 TA A H 2 T M TA A
t , m in
Ò
Fig. 3. Experimental dependencies of the frequency on time for
three pharmaceutical products.
1 2 3 4 5
0.0
0.1
0.2
0.3
0.4
0.5
n
S i( )
Fig. 4. Dependence of S(i) on iV-parameter for different time
intervals: 1�2 (1); 1.3�2.6 (2); 1.4�5.2 (3); 1.1�6.8 (4);
5�7 (5) min. For the cases (1) and (5) S(i) value is presented for
the comparison because the classification wasn't adequate.
214
SQO, 7(2), 2004
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
)))(exp(1()( β
τ
tftf s −−⋅∆=∆ (6)
where ∆fs � limiting value, τ � characteristic time, and
β � parameter, the value of which is within the [0,1] inter-
val. The average values for the parameters of the equa-
tion (6) for the considered samples are presented in Table
2. At β = 1, the equation (6) transforms into the classical
Langmuir dependence. It is interesting, that at t→0 (6)
transforms into the known kinetic Bangham-Bart equa-
tion [35]:
( ) BtAtf −⋅=∆ β (7)
where A, Â and β � constants, which depend on the nature
of the adsorbate-adsorbenpaart. The Bangham kinetics
is usually connected to the either non-uniformity of the
adsorbent surface or to the interaction in the adlayers.
The concept of the non-uniform surface, as well as
concept of the interaction are two basic classical ap-
proaches, which enable to explain the deviations from
the Langmuir theory [36]. The reasons of such devia-
tions are understandable: not only concentration limita-
tions are distorting the statements of the Langmuir model,
but the real surface itself is essentially different from the
ideal � it is the spatially organized systems in nano-mi-
cro scale. The peculiarities of the organization depend
on the molecular structure, fabrication procedure, used
substrate material, etc. The concept of the heterogeneous
surface supposes that the surface is characterized by some
distribution of the heats of adsorption. This can be a re-
sult of the presence of adsorption centers of different types
(in relation to the molecules of the specified type) and the
Fig. 5. The values of s(i) (left) and and classification of clusters (right) for 8 sensors (iV-parameter for 1.1�6.8 min.) � a; for 8 sensors
(iV-parameter for 1.4�5.2 min.) � b: for 3 sensors (C[6]À, pentacene, H2Pc; iV-parameter for 1.4�5.2 min.) � c. Ampicillini
trihydras-Darnitsa® (1, 2, 3), Excedrin® (4, 5, 6), Groseptol 480®,(7, 8, 9).
Component 1
0.0 0.2 0.4 0.6 0.8 1.0 �2
�
1
�
3
�
2
0
1
0
2
2
4
7
8
9
4
5
6
1
2
3 a
S i( )
C
o
m
p
o
n
e
n
t
2
0.0 0.2 0.4 0.6 0.8 1.0
�
1
�1�3
�
2
�2
1
1
0
0 2 3
7
8
9
4
5
6
1
2
3 b
S i( )
C
o
m
p
o
n
e
n
t
2
Component 1
0.0 0.2 0.4 0.6 0.8 1.0 �1
0
.5
0
.0
�
0
.5
�
1
.0
�2 1
1
.0
0 2
7
8
9
4
5
6
1
2
3
c
S i( )
C
o
m
p
o
n
e
n
t
2
Component 1
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
215SQO, 7(2), 2004
fluctuations of the adsorption heat as a consequence of
the influence, for example, of the surface structure, as
well. Thus, the peculiarities of the real surface, can be
taken into account by the respective choice of the distri-
bution type of the differential heat of adsorption. Taking
into account the relation between differential heat of ad-
sorption and the rate constant τ in the equation (6) [36], it
is possible to formulate the statement also in kinetic terms:
the respective choice of the distribution of the relaxation
times is the characteristic feature of the given surface.
The formulation of the problem in the terms of relaxa-
tion times enables to reveal the physical meaning of the
equation (6). Taking into account its coincidence with
the widely known KWW equation (Kohlrausch-
Williams-Watts, KWW, �stretched exponential� func-
tion), it is reasonable to suppose and general physical
mechanism which determines the origin of such depend-
encies [37-39]. It is known that KWW enables to ad-
equately describe the wide spectrum of physical phenom-
ena in various materials [40-42], the characteristic fea-
ture of which is the presence of the set of relaxation peri-
ods in the system. This law isn�t the empirical observa-
tion and has theoretical background [38, 43]: the relaxa-
tion of the system in the conditions of the stochastic fluc-
tuations has the form ~exp(�(t/τ)β). This conclusion
doesn't depend on the microscopic physical model of the
process, and is determined explicitly by the stochastic
nature of the relaxation process. It is interesting to note
that the analysis within the frames of the percolation
theory shows a direct link between KWW type of relaxa-
tion and the fractal nature of material [44]. For example,
the presence of the spatial fractal structures must result
in the appearance of KWW types of relaxation in them,
and vice versa, if the system is characterized by KWW
relaxation dependencies, this is indicative of the fractal
properties of the object itself.
The analysis of the parameters of the equation (6) cal-
culated and averaged for all the samples (Table 2) shows
the presence of the correlation between them: for C[8]A,
H2TMTAA and tetracene τ are minimal and β close to
1.0 (Langmuir model, �slow� responce), for C[4]À and
pentacene � β are maximal at β ≤ 0.5 (that is, the sensor
response is probably limited by the mass transfer proc-
esses, "quick" response); for C[6]A, H2Pc, and H2TAA �
τ average at 0.5 < β < 0.7. It is necessary to note that
obtained values of τ and β are characteristics of the sensi-
tive layer itself, not the analyte. This enable to suggest
that β parameter can be some characteristic of the sensi-
tive coating essential for the optimization of such cross-
selective arrays.
These results can be understood within the frame of
the phenomenological model of the adsorption-desorp-
tion on the surface [45]. Corresponding to this approach,
the adsorption-desorption processes can be characterized
as those controlled by the energy profile on the surface.
One of the main results of this method is the classification
of the systems on �quick� or �slow� adsorbents in de-
pendence on the relative values of the surface adsorption
and desorption barriers. �Slow� and �quick� systems
show very different surface dynamics (Table 2). Indeed,
in the case of the strong adsorbents, adsorption time of
particle near the surface is much smaller, then during
desorption � such systems influence equilibrium surface-
volume and cause the limitation of the mass transfer near/
on the surface. For �slow� adsorbents the desorption time
is much smaller then the particle adsorption time (the
Langmuir model), that is, the surface structure almost
doen't influence the dynamical behaviour of the particle
near/on the surface due to its �energetic� inertia. At the
same time, for the �quick� adsorbents the structure pecu-
liarities will determine the adsorption-desorption equi-
librium on the nanostructured surface due to influence
on the surface energy profile. It is understandable, the
means of the surface structurization on the micro- and
nano- scale in such a manner enable to control the selec-
tivity profile of these systems. The obtained result ena-
bles to consider the correlation between chemical func-
tionality of the sensitive layers of the EN systems and
their molecular structure. All molecules used in the given
work as components of sensitive layers consist of the aro-
matic fragments with the hydrophilic and hydrophobic
regions, which creates adsorption centers for the respec-
tive molecules. Besides, the presence of the nanovoids in
calixarenes can stimulate additional steric stabilization
of the analyte-receptor complex. Thus, the functional
possibilities for the diffeent types of the molecular struc-
tures are determined in the first place by the dominating
(from the viewpoint of ability to form intermolecular com-
plexes) peculiarities of their molecular structure: high
polarizability of pentacen and H2Pc [46]; the presence of
the intramolecular hydrogen bonds in Pc and H2TAA
[47]; by the presence of the nanovoids in calixarenes [48].
It is necessary to underline that the form of nanocavities
of calixarenes in the solid state is changing from the trun-
cated cone (C[4]A) through cylinder (C[6]A) up to irregu-
lar figures for (C[8]A); this is the result of the enlarge-
ment of the intramolecular hydrogen bond [49]. This
determines C[8]A structure disorder in the solid state,
which distorts the nanovoids and complicates interaction
of the possible analyte with the intramolecular polar part.
Thus, tetracen, H2TMÒAA and C[8]A (the latter ones
due to their spatial structure) have almost hydrophobic
surface under absence of the possibilities of the additional
Table 2. Average values of the parameters of the equation (6)
for three types of pharmaceutical products.
¹ Sensor Average β τ, min
name
1. tetracene 0.90±0.10 20�30 �Slow� adsorption
2. pentacene 0.50±0.01 1�10 �Quick� adsorption
3. C[4]A 0.35±0.01 1�10 �Quick� adsorption
4. C[6]A 0.55±0.05 10�20 �Mean� adsorption
5. C[8]A 0.90±0.10 19�21 �Slow� adsorption
6. H2Pc 0.65±0.10 10�17 �Mean� adsorption
7. H2TAA 0.60±0.10 14�20 �Mean� adsorption
8. H2TMTAA 0.95±0.05 20�50 �Slow� adsorption
216
SQO, 7(2), 2004
I.V. Kruglenko et al.: Multisensor systems for gas analysis: optimization of the array for �
intermolecular interactions. C[4]A among considered
calixarenes most fully maintain its structure in the solid
state, and, respectively, the ability to bound complemen-
tary analytes due to rigid structure of the nanovoid, which
essentially increase the lifetime of the analyte in the adsor-
bed state. Pentacen, in its turn, having high polarizabi-
lity, is capable to form Van-der-Waals complexes of diffe-
rent origin with wide spectrum of the various analytes.
H2Pñ, C[6]À and H2TAA, besides that are capable to
interact with different analytes, both hydrophobic and
hydrophilous, which enables to form multicontact com-
plexes with the different origin of the intermolecular
bonds.
9. Conclusions
The considered comparisons of the results of cluster analy-
sis and approximation of the kinetic dependencies ena-
bles to made a conclusion about the correlation between
β-parameter and the efficiency of the classification task:
optimally sensitive are such layers which are characteri-
zed by the values of β << 1 - this provides, on the one
hand, the sufficiently wide selectivity profile of the sensi-
tive layer, and, on the other hand, the possibility to con-
trol the selectivity profile of the sensitive element by us-
ing the surface structurization in the nano-micro-scale.
Acknowledgements
Authors are thankfull to the international science foun-
dation INTAS for the financial support of this work
(projects 00-00870 and 01-00257).
References
1. J.R. Stetter, W.R. Penrose // Journal of The Electrochemical
Society, 150, pp. S11-S16, (2003).
2. R. Brown, E. Zellers, Environmental Monitoring, in Sensors-
A Comprehensive Survey Grandke and W.H.Ko, eds., VCH
Publishers, Weinheim: FRG, (1989).
3. C.L. Honeybourne // Journal of Chemical Education, 77(3),
pp. 338-343 (2000).
4. F. Avila, D.E. Myers, Palmer C // J. Chemometrics, 5, pp.
455-465 (1991).
5. K.J. Albert, D.R. Walt, D.S. Gill, T.C. Pearce // Anal., 73,
pp. 2501-2508 (2001).
6. J.W. Gardner, P.N. Bartlett // Sensors and actuators B., 18,
pp. 211-220 (1994).
7. H.T. Nagle, R.G. Osuna, S.S. Schiffman // IEEE Spectrum,
Special issue on Electronic nose, pp. 22-38 (1998).
8. R. Lucklum, P. Hauptmann // Sensors and Actuators B, 70,
pp. 30-36 (2000).
9. I.V. Kruglenko, B.A. Snopok // Proceedings of The Ninth In-
ternational Symposium on Olfaction and Electronic Nose (Ed.
Arnaldo D'Amico and C.Di Natale), MMIII Aracne Editrice
S.r.l., pp.104-111 (2003).
10. M. Sharaf, D. Illman, B. Kowalski, Chemometrics, J. Willey
& Sons, (1986).
11. T.C. Pearce // BioSystems, 41, pp. 43 (1997).
12. A.K. Vidybida // Sur.Biophys, 29, 7B2 (2000).
13. K. Mori, H. Nagao, Y. Sasaki // Comput. Neural. Syst., 9,
P. 79R (1998).
14. W.Goepel // Sensors and Actuators B, 52, pp.125 (1998).
15. F. Davide, A. D'Amico // Sensors and actuators A, 32,
pp. 507-518 (1992).
16. J.W. Gardner, P.N. Bartlett, Electronic Noses. Principles and
Applications, Oxford, University Press, (1999).
17. B.A. Snopok, I.V. Kruglenko // Thin Solid Films, 418 (1), pp. 21-
41 (2002).
18. K. Faber, A. Lorber, B. Kowalski // Jornal of Chemometrics,
11, pp.419 (1997).
19. F.Davide, A. D'Amico // Sensors and Actuators A, 32,
pp. 507 (1992).
20. K.C. Persaud, J.Dodds // Nature, 299, pp. 352 (1982).
21. T. Tatsuma, Y. Watanabe, N. Oyama // Anal. Chem., 71,
pp. 3632-3636 (1999).
22. W.P. Carey, B.R. Kowalski // Anal. Chem., 58, pp. 3077-3084
(1986).
23. G.B. Altshuler, N.N. Efimov, V.G. Shakulin, Quartz genera-
tors., Ì.: Radio I svyaz�, (1984).
24. G. Sauebrey // Z.Phys. 155, pp.206-222 (1959).
25. J.E. Amoore, Stereochemical theory of Olfaction // Nature,
198, pp. 271-272 (1964).
26. F.L. Dickert, R. Sikorski // Materials Science and Engineer-
ing, C 10, pp. 39-46 (1999).
27. Zhong Cao1, Kazutaka Murayama, Katsuyuki Aoki //
Analytica Chimica Acta., 448, pp.47-59 (2001).
28. Di Natale C., Paolesse R., Macagnano A., Mantini A.,
MariP., D'Amico A // Sensors and Actuators B, 68, pp.319-
323 (2000).
29. T. Ekloev, P. Martensson, I. Lundstroem // Analytica Chimica
Acta, 381, pp. 221-232 (1999).
30. Kruglenko I.V., Snopok B.A., Shirshov Yu.M, Reznik A.M.,
Nowicki D.W., Dekhtyarenko A.K., Sensors for Environ-
mental Control (Ed.P.Siciliano), World Scientific Publishing
Co.Pte.Ltd., pp.239-243 (2003).
31. I.V. Kruglenko, B.A. Snopok, Y.M. Shirshov, E.F. Venger //
Semiconductor Physics, Quantum Electronics and Optoelect-
ronics, 3(4), pp. 529-541 (2000).
32. Kaufman, L. and Rousseeuw, P.J. Finding, Groups in Data:
An Introduction to Cluster Analysis, Wiley, New York.,
(1990).
33. P.J. Rousseeuw // J. Comput. Appl. Math., 20, pp. 53-65
(1987).
34. A. Struyf, M. Hubert, and P.J. Rousseeuw, // Computational
Statistics and Data Analysis, 26, pp. 17-37 (1997).
35. M. Breusse, L. Faure, B. Claudel, J. Veron // Progress in
Vacuum Microbalance Technique, 2, pp. 229 (1973).
36. F.F. Volkenshtein, Electronic processes on the surfaces of semi-
conductors at hemosorption, Ì:Nauka. (1987).
37. R. Metzler, J. Klafter // Physics Reports, 339, pp. 1-77 (2000).
38. I. Koponen // Journal of Non-Crystalline Solids, 189, pp. 154-
160 (1995).
39. K. Weron, A. Jurlewicz // J.Phys.A: Math. Gen., 26, pp. 395-
410 (1993).
40. B.J. West // Chemical Physics 284, pp. 45-57 (2002).
41. M. M. Ahmada, K. Yamadaa, T. Okudaa // Carbohydrate
Polymers, 53, pp. 289-296 (2003).
42. I. Avramov, I. Gutzow // Journal of Non-Crystalline Solids,
298, pp. 67-75 (2002).
43. O. Edholm, C. Blomberg // Chemical Physics, 252, pp. 221-
225 (2000).
44. I.Avramov, V.Tonchev // Journal of Non-Crystalline Solids,
194, pp. 122-128 (1996).
45. O.V.Bychuk, B.O'Shaughnessy // Journal of Colloid and In-
terface Science, 167, pp.193-203 (1994).
46. M. Pope, C.E. Swenberg, Electronic Processes in Organic
Crystals and Polymers (2nd Edition), Oxford University Press
(1999).
47. Andre Zh, Simon Zh.-Zh. Andre, Molecular semiconductors,
Ì.: �Mir� (1988).
48. Richard M. Crooks, Antonio J. Ricco // Accounts of Chemi-
cal Research, 31(5), pp. 219-225 (1998).
49. C.D.Gutsche. Calixarenes, Royal Society of Chemistry, Cam-
bridge (1989).
|