IOP Conference Series: Materials Science and Engineering
OPEN ACCESS You may also like
- Operating temperature dependence of
A Smart Gas Sensor Insensitive to Humidity and sensitivity in Ag-TiO2 Schottky type gas
sensors
Temperature Variations S Rahbarpour, S Sajed, N Ghodsi et al.
- Light-Activated Chemoresistive and
Plasmonic-Resonant Optical Sensors for
To cite this article: Mohammadreza Hajmirzaheydarali and Vahid Ghafarinia 2011 IOP Conf. Ser.: NO2 and H2 Sensing Based on ZnO
Mater. Sci. Eng. 17 012047 Doped Nanoparticles
Valentina Paolucci
- Utilizing the response patterns of a
temperature modulated chemoresistive
gas sensor for gas diagnosis
View the article online for updates and enhancements. Amir Amini and Vahid Ghafarinia
This content was downloaded from IP address 110.54.182.73 on 11/12/2023 at 10:25
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
A Smart Gas Sensor Insensitive to Humidity and
Temperature Variations
Mohammadreza Hajmirzaheydarali1 and Vahid Ghafarinia1
1
Electronic Materials Laboratory, Electronic Engineering Department, K. N. Toosi
University of Technology, Tehran, Iran
Abstract The accuracy of the quantitative sensing of volatile organic compounds by
chemoresistive gas sensors suffers from the fluctuations in the background atmospheric
conditions. This is caused by the drift-like terms introduced in the responses by these
instabilities, which should be identified and compensated. Here, a mathematical model is
presented for a specific chemoresistive gas sensor, which facilitates these identification and
compensation processes. The resistive gas sensor was considered as a multi-input-single-
output system. Along with the steady state value of the measured sensor resistance, the
ambient humidity and temperature are the inputs to the system, while the concentration level
of the target gas is the output. The parameters of the model were calculated based on the
experimental database. The model was simulated by the utilization of an artificial neural
network. This was connected to the sensor and could deliver the correct contamination level
upon receiving the measured gas response, ambient humidity and temperature.
1. Introduction
Rapid growth of interest in artificial olfaction as a replacement for the human and animal odor sensing
panels has led the gas sensors and the electronic noses to play an important role in the selective gas
detection and odor analysis applications [1-3]. The main element in the structure of an electronic nose
is an array of non-specific gas sensors, the combined responses of which serve as a fingerprint to
identify certain chemicals in a gas mixture [4]. The sensors used are usually of chemoresistive type
[5] different in structural parameters [6, 7] such as the pallet thickness [8], filtered with different
membranes [9] or operated at different temperatures [10]. Various pattern recognition algorithms have
been utilized for the identification of an unknown target gas by comparing its response pattern with
the previously recorded responses [11, 12]. A successful chemical identification would depend on the
quality of the sensor array, the generality of the training database and the efficiency of the pattern
recognition algorithms utilized.
Fluctuations in the background atmospheric conditions can impose substantial errors in the
measurements and, hence, in the results of the subsequent mathematical analysis [13, 14]. It has been
shown that the responses of chemoresistive gas sensors, for instance, is affected by not only the nature
and concentration of the target gas, but is simultaneously affected by the temperature and humidity
changes in the surrounding atmosphere. Moreover, the traces of the interfering gases even at low
levels can affect the sensors baseline and, hence, nonlinearly modify the acquired responses. These
drift-like [13, 15] effects, categorized as measurement errors, cause multidimensional shifts in the
recorded response patterns and can hinder the intended gas classification process in the feature space
[10, 16]. In practice, such pattern aberrations can increase the misclassification probability of a
response pattern recorded for an unknown gas by blurring the between-class decision boundaries.
Published under licence by IOP Publishing Ltd 1
2
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
Different approaches have been adopted to deal with the measurement errors encountered with
sensor responses described above [17, 18]. The soft methods employed for the elimination of the
above described errors consider the ambient temperature and the ambient humidity as the additional
inputs to the sensory system. Different modeling techniques are, then, used to model the system based
on the input-output relationships recorded in training experiments. The model is then utilized for the
elimination of the drift components from the recorded responses [19].
Here, we are reporting the design and fabrication of a controlled atmosphere chamber which could
be utilized for the acquisition of the response data of a gas sensor to different target gases at various
predetermined environmental conditions. The results were piled up for the formation of a training
database, which was then utilized for compensation of the responses of the sensor for the ambient
condition fluctuations.
2. Experimental
The schematic diagram of the measurement system is presented in Fig.1. It consists of a three-walled
cubic controlled atmosphere chamber equipped with a humidity sensor, a temperature indicator and a
general chemoresistive gas sensor. The sensors are connected to a computer via an interface circuit.
The heating elements and the purging system, along with the cooling jacket installed between the
walls of the chamber, can affect the chamber temperature. Temperature control is achieved by a PID
power control unit connected to the heating elements and to the temperature sensor. The outermost
wall is made of a porous insulating material which thermally insulates the chamber from the ambient
atmosphere. The relative humidity in the chamber air is controlled by a flow of controlled mixture of
dry and humid air. The system can provide predetermined alcohol vapor concentrations in air at
known temperature, in the range of 0-50oC, and relative humidity levels, in the range of 2-95%, within
the volume of the innermost chamber. Two small fans mildly circulated the chamber air for
homogenization. It took ~20 min for the chamber to acquire a stable atmosphere with constant
humidity and temperature. The three parameters were constantly monitored during the measurements.
A predetermined amount of analytical grade methanol was injected into the chamber after
stabilization of the humidity level and temperature in the chamber. Another 15 min was allowed for
homogenization of methanol contaminant in the chamber's atmosphere.
The steady-state responses of the under test gas were recorded after digitization. The
corresponding chamber humidity and temperature were directly read and continuously recorded from
the respective standard sensors. The overall accuracy in the measurement of the chamber’s relative
humidity and temperature were +/- 3% and +/- 0.5oC, respectively. After the completion of each test,
the chamber was purged with the flow of dry air for 20 min. The block diagram of the measurement
system is presented in Fig.2. The results of the experiments were piled up to form a databank
containing the gas sensor’s steady-state resistances at the presence of various combinations of the
methanol contamination, chamber humidity and temperature.
The chemoresistive gas sensor used in these investigations is a commercially available tin dioxide
thick-film gas sensor (SP3-AQ2-01, FIS Inc., Japan) [20].
2
3
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
Air
Inlet
Figure1. The schematic diagram of the measurement system.
Temp.
Humid.
Gas
Gas
Power
Supply
Sens.
Sens.
Sens.
Cham.
For
All
Blocks
Control
RS232
A/D
Circuit
Interface
PC
Microcontroller
Program
Circuit
Figure 2. A block diagram describing different parts of the measurement system.
Methanol vapor at concentrations ranging from 100 to 2000 ppm was introduced to the chamber as
the target gas. Sensor responses to the methanol contaminated atmosphere at different combinations
of ambient humidity and temperature were investigated. The results are summarized in fig.3a and b
for 23 oC and 33 oC , respectively. Each diagram is related to different methanol concentrations and
demonstrates Rg variations with relative humidity at the stated temperature. The databank was
completed with sensor responses at the presence of six different methanol concentrations, each
measured at five different ambient temperatures ranging from 23 oC to 43 oC. In all cases the relative
humidity varied continuously in the span of 5% to 85%.
3. Modeling
The input-output relationship of a chemoresistive gas sensor is described through the power law [21,
22]:
(1)
3
4
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
In which R is the sensor resistance and C is the concentration of the target gas; s, m, and r are the
calibration parameters. This simple heuristic model can be tuned based on some experimental data;
the reverse model then can be utilized to estimate the gas concentration corresponding to the
measured value of the sensor resistance. While this single input single output model has performed
well in formulating the sensor behavior at well-controlled constant laboratory conditions, its
quantitative predictions based on the measurements carried out in the practical conditions can be
trusted only if the other influencing inputs remained unchanged since the last calibration.
Atmospheric conditions including the ambient humidity (H) and temperature (T) or traces of
undesired interfering gases can remarkably affect the sensing mechanism of the chemoresistive gas
sensor [18]. The variation of these atmospheric parameters invalidates the one-to-one relationship
between R and C. Fig.3 illustrates how the environmental parameters can change the sensor
resistance. At any fixed concentration of the target gas, the sensor resistance decreases by increasing
the ambient humidity and temperature. Hence, as depicted in Fig.4, there will be a span of sensor
resistance values related to any concentration of the target gas. The calibrated model, showed by solid
curve in Fig.4, is a good transfer function at normal atmospheric conditions. However, the ambient
humidity and temperature variations cause drift-like terms that alter the sensor resistance substantially
beyond the expected values and result in high estimation errors.
A more general model is obtained by including the effects of the additional inputs in the formation
of the model structure. Assuming the actual target gas concentration as a function of the sensor
resistance, ambient humidity and ambient temperature, we’ll have:
C=f(R, H, T) (2)
Figure 3. Variations of the sensor resistance with the relative humidity at the presence of the stated
methanol contamination levels in air at (a) 23 and (b) 33oC.
4
5
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
Figure 4. The best fit of the power function to the experimental data. Vertical bars represent the span
of variations caused by the drift-like terms.
There is no explicit mathematical formulation for f yet, but a data-based representation is possible
through universal approximators such as artificial neural networks [23, 24].
Following a series of numerical evaluations and based on the pruning method, a three layer
feedforward neural network of {3 2 1} structure was proved to be a good candidate for representing
the unknown function f. The transfer function of all 6 neurons was selected to be logarithmic sigmoid
defined as:
y=1/(1+e-x) (3)
where x is the input to the neuron and y is the output. Two third of the experimental data were
randomly selected as the training dataset. The Levenberg-Marquardt algorithm was repeatedly
performed on the training dataset to decrease the output error to the lowest possible level. The
remaining one third of the data, set aside as the test dataset, were used to evaluate the performance of
the trained neural network. The concentration values estimated by the trained neural network along
with their respective experimentally recorded values are presented in Fig.5. The results show that the
neural network has successfully modeled the sensor behavior within the trained range of the input
space.
For the online performing of the above described drift compensation process, a smart sensor module
was designed based on the above reported results. The block diagram of the designed sensory system
is demonstrated in Fig.6. The responses of the three sensors, i.e. temperature, humidity and gas
sensors are fed into the neural network software implemented on a suitable microcontroller after
signal manipulation and conversion to digital signals. The neural network program, then, predicts
the compensate gas concentration based on the previous training. The obtained target gas
concentration value, after linearization, is re-converted to analog voltage directly related to the actual
contamination level. The whole system acts as a smart gas sensor module that is insensitive to the
variation of the environmental parameters.
5
6
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
Figure 5. Comparison of the real (experimentally measured) target gas concentrations with those
predicted by the trained artificial neural network.
Gas
Humid.
Temp.
Sensor
Sensor
Sensor
Signal
Conditioning
Analog
to
Digital
ANN
on
DSP
Digital
to
Analog
Target
Gas
Concentration
Figure 6. The block diagram of the smart gas sensor module
4. Conclusion
The effect of ambient humidity and temperature fluctuations on the steady state responses of a
chemoresistive gas sensor was studied. The results of the experiments carried out by the developed
atmosphere controlled chamber revealed the dependency of the sensor output on the environmental
parameters. A set of experiments were conducted at the presence of predetermined methanol
concentration levels, the results of which indicated substantial deviations in the sensor response due
to the instabilities in the parameters of the background atmosphere. It was shown that a static SISO
model can not appropriately describe the input-output relationship of a chemoresistive gas sensor.
This was achieved by the formation of a database containing the information on the sensor's behavior
at different atmospheric conditions. This database was utilized for the training of an artificial neural
network which could relate the combined readouts from a humidity sensor, a gas sensor and a
thermometer to the actual target gas concentration. A comparison of the network-predicted target gas
concentrations with the actual concentration levels showed that the trained neural network can
considerably compensate the effects of the ambient humidity and temperature variations in the steady-
state responses of the gas sensor.
The results facilitated the design and development of a smart gas sensor module for the online
compensation of the drift-like terms. Beside the resistance of gas sensor that is proportional to the
concentration of the target gas, the module records background atmospheric conditions through the
incorporated humidity and temperature sensors. The three inputs are fed into the neural network that
6
7
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
is implemented on a microcontroller. The output of the module is a linear voltage that is only
dependent on the concentration of the target gas.
References
[1] D. D. Leec, D. S. Lee, Environmental gas sensors. IEEE Sensors Journal, 2001. 1: p. 214-224.
[2] S. Capone, A. Forleo, L. Francioso, R. Rella, P. Siciliano, J. Spadavecchia, D. S. Presicce, A. M.
Taurino, Solid state gas sensors state of the art and future activities. Journal of Optoelectronics
and Advanced Materials, 2003. 5: p. 1335-1348.
[3] F. Hossein-Babaei, M. Orvatinia, Gas diagnosis based on selective diffusion retardation in an air
filled capillary. Sensors and Actuators B, 2003. 96: p. 298-303.
[4] H. T. Nagle, S. S. Schiffman, R. Gutierrez-Osuna, The how and why of electronic noses. IEEE
Spectrum, 1998. 35: p. 22-21
[5] N. Yamazoe, G. Sakai, K. Shimanoe, Oxide semiconductor gas sensors. Catalysis Surveys from
Asia, 2003. 7: p. 63-75.
[6] F. Hossein-Babaei, M. Orvatinia, A novel approach to hydrogen sensing. IEEE Sensors Journal,
2004. 4: p. 802-806.
[7] F. Hossein-Babaei, M. Hemmati, M. Dehmobed, Gas diagnosis by a quantitative assessment of
the transient response of a capillary-attached gas sensor. Sensors and Actuators B, 2005. 107: p.
461-467.
[8] F. Hossein-Babaei, M. Orvatinia, Analysis of thickness dependence of the sensitivity in thin film
resistive gas sensors. Sensors and Actuators B, 2003. 89: p. 256-261.
[9] F. Hossein-Babaei, M. Orvatinia, Transient regime of gas diffusion-physisorption through a
microporous barrier. IEEE Sensors Journal, 2005. 5: p. 1004-1010.
[10] F. Hossein-Babaei, S. M. Hosseini-Golgoo, Analyzing the response of a thermally modulated gas
sensor using a linear system identification technique for gas diagnosis. IEEE Sensors Journal,
2008. 8: p. 1837-1847.
[11] R. Gutierrez-Osuna, Pattern analysis for machine olfaction A review. IEEE Sensors Journal,
2002. 2: p. 189-202.
[12] F. Hossein-Babaei, S. M. Hosseini-Golgoo, Amir Amini, Extracting discriminative information
from the Pade-Z-transformed responses of a temperature-modulated chemoresistive sensor for gas
recognition. Sensors and Actuators B, 2009. 142: p. 19–27.
[13] F. Hossein-Babaei, V. Ghafarinia, Compensation for the drift-like terms caused by environmental
fluctuations in the responses of chemoresistive gas sensors. Sensors and Actuators B, 2010. 143:
p. 641-648.
[14] G. Martinelli, M. C. Carotta, Thick-film gas sensors. Sensors and Actuators B, 1995. 23: p. 157-
161.
[15] A. Hierlemann, R. Gutierrez-Osuna, Higher-order chemical sensing. Chemical Reviews, 2008.
108: p. 563-613.
[16] M. Holmberg and T. Artursson, Drift compensation standards and calibration methods.
Handbook of machine olfaction. (Edited by T. C. Pearce, S. S. Schiffmann, H. T. Nagle & J. W.
Gardner), 2003: p. 325-346.
[17] S. Holmin, C. Krantz-Rulcker, I. Lundstriom, F. Winquist, Drift correction of electronic tongue
responses. Measurement Science and Technology, 2001. 12: p. 1348-1354.
[18] T. Artursson, T. Eklov, I. Lundstrom, P. Martensson, M. Sjostrom, M. Holmberg, Drift
correction for gas sensors using multivariate methods. Journal of Chemometrics, 2000. 14: p.
711-723.
[19] S. Hirobayashi, H. Kimura, T. Oyabu, Dynamic model to estimate the dependence of gas sensor
characteristics on temperature and humidity in environment. Sensors and Actuators B, 1999. 60:
p. 78-82.
[20] FIS Inc. Japan, http://www.fisinc.co.jp
[21] N. Yamazoe, K. Shimanoe, Theory of power laws for semiconductor gas sensors. Sensors and
Actuators B, 2008. 128: p. 566-573.
[22] S. W. Lee, P. P. Tsai, H. Chen, Comparison study of SnO2 thin and thick film gas sensors.
Sensors and Actuators B, 2001. 79: p. 163-170.
7
8
CAMAN IOP Publishing
IOP Conf. Series: Materials Science and Engineering 17 (2011) 012047 doi:10.1088/1757-899X/17/1/012047
[23] S. Chen, S. A. Billings, P. M. Grant, Non-linear system identification using neural networks.
International Journal of Control, 1990. 51: p. 1191-1214.
[24] M. O. Efe, O. Kaynak, A comparative study of neural network structures in identification of
nonlinear systems. Mechatronics, 1999. 9: p. 287-300.
