Amperometric Electronic Tongue Applied to Coffee

Amperometric Electronic Tongue Applied to Coffee Classification
Alejandro Durán Carrillo de Albornoza, Albert Gutés Regidorb, Manel del Valle Zafrab*
a
ICTM, University of Havana, Zapata y G, 10400, Havana, Cuba
b
Sensors & Biosensors Group, Department of Chemistry, Autonomous University of Barcelona,
08193, Bellaterra, Barcelona, Spain
*Corresponding author. Tel: +34 935811017; Fax: +34 935812379
E-mail address: [email protected]
Abstract
An amperometric electronic tongue was developed and applied to discriminate liquid coffee samples
by chrono-amperometry technique. The instrument comprises a noble-metal and epoxy-graphite
multi-electrode system, the multi-channel potentiostat and also the Virtual Instrument, which were
developed in our laboratory. Principal component analysis was used to process the multivariate signal
from Bonka´s brand commercial coffee samples. Samples were obtained by three preparation
procedures: espresso, coffee pot and infusion. The system is capable to discriminate coffee samples
according to the preparation mode and also the variety.
Keywords: Coffee classification, Amperometric electronic tongue, Virtual Instrumentation, multichannel potentiostat.
potentiometric sensors have been developed for
water samples [9, 10], wines [9-11], tea [12],
juices [13, 14] and alcoholic beverages [15]
among others. On the other hand, the
application of amperometric e-ts in water
analysis for the industrial [16, 17] and domestic
[18], wines, milk [19, 20], juices [20], bacterial
analysis and tea [21, 22] are known.
The arrays of metal amperometric
electrodes [7, 23, 24] and graphite-epoxy [25]
are widely used in e-t, because of their
advantages and the variety of analytical
techniques employed (i.e. LAPV, SAPV [20],
etc).
Regarding
instrumentation,
microprocessors, embedded systems [26],
multifunction cards and computers together with
the well known virtual instrumentation [27-30] is
used for the development of e-ts [31].
With respect to the analysis of liquid coffee
samples, research using traditional laboratory
equipment as well as potentiometric and
amperometric e-ts combined with Artificial
Neural Networks and Principal Component
Analysis (PCA) [10, 12, 14] have been done.
The goal of this paper is to present the
application of an amperometric e-t using a
simple
chrono-amperometric
technique
combined with PCA for the classification of
coffee samples and employing an array of
electrodes of metal and epoxy-graphite and
suitable instrumentation, both custom developed
in our laboratory.
1 Introduction
Nowadays the multivariate analytical
systems known as electronic tongues (e-t),
which allow extracting significant qualitative and
quantitative information from liquid samples are
widely studied [1, 2].These systems are based
in arrays of non-selective sensors and pattern
recognition tools. The use of different types of
sensors (i.e. electrochemical, conducting
polymers, optical, etc.) is reported in the
literature [3-5] for the e-t, as well as a variety of
processing tools, like ANN, PCA and PLS [1, 4,
6].
The electronic tongues are being used in
different fields like clinical analysis, environment
studies, pharmaceutical, biotechnological and
food industries, among others, which has been
also reported [1-5, 7, 8]. Its use, particularly
when the instrumentation is developed ad hoc
and based on personal computers, is an
alternative compared with costly conventional
equipment for chemical analysis in the field, the
laboratory, and the industry.
Qualitative and quantitative analysis of
beverages using e-t (i.e. classification,
multidetermination, quality analysis, correlation
with expert panels, etc.) is one of the main
subjects of research and at the same time, an
area of greatest application of e-t. Regarding
this, many studies have been done where a
large range of sensors and analytical techniques
are employed [3-5] like potentiometry and
amperometry.
E-ts
with
electrochemical
173
2.3 Instrumentation. Hardware and software
2 Experimental
Our scheme, showed in Figure 2, is based
on a Pentium III PC and a commercial
multifunction
DAQ
card
from
National
Instruments PCI-6221 featuring 16-bit, 250 kS/s,
16 analog inputs, and two 16-bit analog outputs.
The card acquires the signals corresponding to
each channel, from the analog input lines, and
sets the working potential for each electrode by
means of the analog output channel. The
designed multichannel potentiostat (4 channels)
simultaneously controls the working potential
applied to the electrodes in a pseudoindependent way because the DAC output feeds
a voltage divider resistive network that
establishes the working potential respecting to
the reference electrode also connected to
ground. The configuration also uses an auxilliary
electrode which supply the working electrodes
current. The conditioning hardware (i.e.
potentiostat, currento to voltage converter,
voltage follower) is based on low power FETInput operational amplifier AD822 from Analog
Devices. The voltage signals were also low-pass
filtered by the UAF42, active filter from Burr
Brown, tunned to 2 Hz.
2.1 Reagent and samples
An experiment for classification of sort from
the same Nestle brand was carry out: the coffee
samples were prepared according to three
traditional methods: coffee pot, espresso and
infusion, from different sorts of Bonka (Nestle)
commercial brand: Puro Brasil, Mezcla 70:30,
Mezcla Intenso, Estilo Italiano Natural, Natural,
Puro Colombia, Filtro Natural, Descafeinado
Natural, Instantáneo Descafeinado and also a
commercial substitute prepared with barley.
Finally, nine samples were prepared from each
sort of coffee, with three replications for each
mode of preparation plus three samples of
instant coffee, for a grand total of 84.
For all the samples 15 g of ground coffee
were weighted, keeping a volume of Ribes
natural mineral water of 75 ml. For cleaning the
system a 0.1M KCl washing solution prepared in
the laboratory was employed. The attachments
(filter and coffee pot) were washed with bidistilled water from the laboratory, and the
measurements were carried out directly on the
coffee samples without using any reagent.
2.2 Sensor array and measuring cell
Two amperometric multi-electrodes built in
the laboratory were employed. The first one
includes three electrodes (Gold, Platinum and
graphite-epoxy) described and previously used
[32] and the second was similar but it comprised
4 Platinum electrodes. For manually injecting
the coffee samples and performing the
measurements, the system shown in Figure 1
was implemented. Two methacrylate cells were
also designed: the cell for mixing
and
measurement (1) with incorporated auxiliary
electrode and stirrer, where the working multielectrode is placed (2), and the cell (3) with the
Ag/AgCl Orion 900200 double junction reference
electrode (4). A teflon tube joins both cells and
the syringe (5) for establishing the flow of liquid.
2
5
1
Figure 2. Hardware scheme.
The Virtual Instrument (VI) which commands
the system was programmed using the
LabVIEW 7.0 TM NI-DAQmx functions. The VI
reads and displays the multi-electrode signals.
The resulting signal array is also pre-processed
in order to obtain the matrix of data patterns
which corresponds to each coffee sample.
Sampling rate is set to 10 samples per second
by software, while one voltage single value is
the mean of 100 acquired points.
Another VI was also implemented, which
automatically sorts the data, giving a response
pattern which represents the intensity profile vs
sample number with the information of the 7
electrodes in the following order: Platinum1,
Gold, Epoxy-Graphite, Platinum2 to Platinum5.
The final data matrix is of 84 rows and 4200
columns in the first experiment while in the
4
3
Figure 1. Flow system.
174
second is of 30 x 4200, with rows representing
the number of coffee samples and columns the
instant values of the intensities sampled.
It is interesting to note the similarity of the
response profiles.
The coffee samples were manually injected
with
a
syringe
coupled
to
the
mixing/measurement cell. The sample was
stirred with a constant speed. Two sequences of
measurements were carried out, one for each
multi-electrode. Each attachment was previously
washed with bi-distilled water. The electrodes
were polished with standard sandpaper
sandpaper as necessary. The experiments were
carried out at ambient temperature in the
laboratory. Figure 4 depicts an example of two
pattern profiles from Brazilian coffee and barley
substitute, theses patterns were arranged
serializing the responses from the 7 sensors
corresponding to each coffee sample.
2.4 Measurement procedure
The system was calibrated using homemade dummy cells against an AUTOLAB
PGSTAT 12 (Eco Chemie, The Nederlands)
multi-channel potentiostat with GEPS v 4.9
software in the laboratory. Also, tests were
carried out by standard additions with oxidable
compounds (ascorbic acid and K4[Fe(CN)6]).
Chrono-amperometry technique was used,
which allows to obtain the transient profiles of
intensity for each sample. The measurements
were carried out applying a DC potential and
recording the instant values of the intensity
during 60 seconds. Figure 3(A-B) depicts an
example of two sensor responses obtained by
applying the D.C. stimulus to the sensor arrays.
The figure 3A corresponds to Brazilian coffee
while figure 3B corresponds to the barley
substitute
Figure 4. An example of two patterns profile.
2.5 Data processing
Cyclic
Voltammetry
technique
was
previously applied to random samples of coffee
in order to determine the optimal values of the
working potentials, which also allows to
compress the data. A reasonable solution was
to establish the same electrode potential of 0.8
V for the Pt, Au, C multi-electrode, while for the
Platinum multi-electrode different potentials
were established from the voltage divider
resistive network (i.e. 0.25, 0.5, 0.75, 1.0V).
The PCA technique was used for the
analysis of the data considering that no previous
knowledge of the sample composition is
needed. This mathematical tool allows the
graphical projection of the data in the least
dimension that describes the largest variance.
From a visual interpretation of the observation
parameters (scores) the relation between the
data corresponding to a given experiment is
shown, which allows the classification of the
samples. SPSS 12 program (SPSS Inc., USA)
was used. A total of 4200 current values were
considered as variables in both experiments as
Figure 3A. Sensor responses corresponding to the Brazilian
coffee.
Figure 3B. Sensor responses corresponding to the barley
substitute.
175
it is shown in Figure 4. SigmaPlot (Jandel
Scientific, Germany) was employed in graphic
representations of the data.
barley were discriminated. Triplicates were
prepared using the same amount of coffee in all
cases, so a grand total of 18 samples (PCA
objects) were determined. 4200 variables were
obtained and standardised before PCA
modelling. 89.2% of the total data variance was
explained by the first two principal components.
Good discrimination is obtained for 4 samples
groups with a strong overlapping in two of the
coffee types; natural blend coffee and mixture
coffee.
3 Results and discussion. Principal
components analysis performance
Several PCA adjustments were performed
with the obtained data. The first one consisted in
coarse coffee types discrimination. For doing so,
two different coffees, Brazilian and Colombian,
were prepared by three different methods using
the same amount, 5 grams, of ground coffee,
i.e. infusion, espresso machine and coffee pot.
Each coffee was prepared by triplicate. Toasted
barley was prepared by the same procedure and
also by triplicate. A total of 27 samples (PCA
objects) were prepared and measured, resulting
in 4200 PCA variables. A simple standardization
was applied to data previous to PCA modelling.
The resulting PCA is showed in figure 5. A
clear Colombian coffee outlier was detected and
thus eliminated from PCA modelling. A total of
87.3% of variance was explained by the first two
principal components. As can be observed very
good classification was obtained, with clear
separation between the two coffees and
moreover to toasted barley.
Because of the good obtained results more
accurate discrimination was considered. In this
case we pretended to classify coffee varieties
using
the
three
previously
mentioned
preparation
methods
(infusion,
espresso
machine and coffee pot). In this case triplicates
of each variety were prepared and measured.
PCA modelling was performed for each
preparation method.
Figure 6. Coffee discrimination using infusion method.
Figure 7 shows the obtained results for the
espresso machine method. 5 types of coffees
and 1 type of toasted barley were prepared by
triplicate. Again, 5 grams of ground coffee were
used in the preparation. A total of 18 samples
were measured and again a total of 4200
variables were obtained. Simple standardization
was applied previous to measurement. A total of
83.8% variance was explained by the first two
principal components. Clear separation of the
six types of coffees was achieved by this
preparation method.
Figure 5. Example of coarse coffee type discrimination using
3 preparation methods.
Figure 6 shows the PCA model obtained by
the infusion preparation method. In this case a
total of 5 types of coffees and 1 type of toasted
Figure 7. Coffee discrimination from espresso method.
176
Finally, figure 8 shows the obtained PCA for
the last preparation method, using a
conventional coffee pot. Again 5 types of coffees
and 1 type of toasted barley were prepared by
triplicate and measured. In this case mixture and
instant coffees were changed for other two types
of coffees: Italian and filter coffees. This change
was made because mixture and instant coffees
were not suitable for being prepared in the
coffee pot used. The 4200 obtained variables
were standardised previous to PCA modelling.
85.5% of the total data variability was explained
by the first two principal components.
been developed in our laboratories. Automation
in data recording and pre-treatment was also
achieved by software development. A robust
system has been achieved as long as variability
in manual coffee samples preparation has not
affected the final discrimination capacity of the
system.
5. Acknowledgements
A. Durán wants to thank RED ALFA
BioSenIntg which supports the research stay
and the Sensors & Biosensors Group, from the
Autonomous University of Barcelona where this
work was carried out.
6. References
[1] Y. Vlasov, et al., Pure and Applied Chemistry, Vol. 77,
No. 11, pp. 1965-1983, 2005.
[2] K. Toko, Measurement Science and Technology, Vol. 9,
pp. 1919-1936, 1998.
[3] A. Legin, A. Rudnistskaya, Y. Vlasov, "Electronic
tongues: new analytical perspective for chemical sensors" in
Integrated Analytical Systems, (S. Alegret Ed.) Elsevier,
Amsterdam, pp. 437-486, 2003.
[4] A.K. Deisingh, D.C. Stone, M. Thompson, International
Journal of Food Science and Technology, Vol. 39, pp. 587604, 2004.
[5] F. Winquist, C. Krantz-Rülcker, I. Lundström, Sensors
Update, Vol 11, No. 1, pp. 279-306, 2002.
[6] V. Pravdová, M. Pravda, G.G. Guilbault, Analytical
Letters, Vol. 35, No. 15, pp. 2389-2419, 2002.
[7] C. Krantz-Rülcker, et al., Analytica Chimica Acta, Vol.
426, pp. 217-226, 2001.
[8] A. Gutés, F. Céspedes, M. del Valle, Analytica Chimica
Acta, Vol. 600, No. 1-2, pp. 90-96, 2007.
[9] A. Legin, et al., Electroanalysis, Vol. 11, No. 10-11, pp.
814-820, 1999.
[10] A. Legin, et al., International Journal of Food Science
and Technology, Vol. 37, pp. 375-385, 2002.
[11] A. Legin, et al., Analytica Chimica Acta, Vol. 484, pp.
33-44, 2003.
[12] L. Lvova, et al., Sensors and Actuators, B, Vol. 95, pp.
391-399, 2003.
[13] A. Rudnitskaya, et al., Analytical Sciences, Vol. 17, pp.
309-312, 2001.
[14] A. Legin, et al., Sensors and Actuators B, Vol. 44, pp.
291-296, 1997.
[15] A. Legin, et al., Analytica Chimica Acta, Vol. 534, pp.
129-135, 2005.
[16] M. Lindquist, P. Wide, Proc. of the 18th IEEE
Instrumentation and Measurement Technology Conference,
Budapest, Hungary, May 21-23, pp. 1320-1324, 2001.
[17] A. Gutés, et al., Sensors and Actuators B, Vol. 115, pp.
390-395, 2006.
[18] M. Lindquist, P. Wide, Sensor for Industry Conference,
New Orleans, Louisiana, USA, January 27-29, 2004.
[19] F. Winquist, et al., Measurement Science and
Technology, Vol. 9, pp. 1937-1946, 1998.
[20] F. Winquist, P. Wide, I. Lundström, Analytica Chimica
Acta, Vol. 357, pp. 21-31, 1997.
[21] M. Scampicchio, et al., Electroanalysis, Vol. 18, No. 17,
pp. 1643-1648, 2006.
[22] P. Ivarsson, P., et al., Sensors and Actuators B, Vol. 76,
No. 1-3, pp. 449-454, 2001.
[23] S. Holmin, et al., Electroanalysis, Vol. 14, No. 12, pp.
839-847, 2002.
[24] F. Winquist, et al., Analytica Chimica Acta, Vol. 471, pp.
159-172, 2002.
Figure 8. Coffee discrimination from coffee pot method.
As a general comment, it is possible to
observe a clear discrimination capacity of the
developed system by a simple Principal
Component Analysis. Two approaches of
discrimination
have
been
successfully
performed. In the first one we have clearly been
able to discriminate between two types of coffee
and one toasted barley brand independently of
the method of preparation. Explained variability
with the first two principal components was over
87%. Moreover in the second approach we have
been able to discriminate 5 types of coffees and
the same toasted barley brand using three
different methods of preparation. Explained
variability with just two principal components
was over 83% in all cases.
4 Conclusions
A new electronic tongue system has been
developed and used for a qualitative
discrimination between different types of coffees
and a toasted barley brand. Simplicity of the
system, using simple chrono-amperometry and
conventional noble metal electrodes, provided
successful results. All the components of the
system, including sensors, measuring cell, and
virtual instrumentation using LabView, have
177
[25] A. Gutés, et al., Electroanalysis, Vol. 18, No. 1, pp. 8288, 2006.
[26] E.M. Avdikos, M.I. Prodromidis, C.E. Efstathiou,
Sensors and Actuators B, Vol. 107, No. 1, pp. 372-378,
2005.
[27] F.J. Sáez de Viteri, D. Diamond, Analytical Proceedings
Including Analytical Communications, Vol. 31, pp. 229-232,
1994.
[28] A. Kraub, U. Weimar, W. Göpel, Trends in analytical
chemistry, Vol. 18, No. 5, pp. 312-318, 1999.
[29] A. Economou, et al., Analytica Chimica Acta, Vol. 467,
No. (1-2,3), pp. 179-188, 2002.
[30] M. Jensen, Journal of Chemical Education, Vol. 79, No.
3, pp. 345-348, 2002.
[31] A. Durán, et al., Sensors, Vol. 6, pp. 19-29, 2006.
178