Journal of Critical Reviews

ISSN- 2394-5125 Vol 7, Issue 9, 2020

DIFFERENT ARTIFICIAL OLFACTION TECHNIQUES - ELECTRONIC NOSE

Shanthi K G, Keerthi B, Manimegalai S,Nisha A,Niviya B R
Department of Electronics and Communication Engineering,RMK College of Engineering and Technology,TamilNadu,India
Corresponding Author: [email protected]

Received: 24.03.2020 Revised: 13.04.2020 Accepted: 21.05.2020

Abstract
Artificial Olfaction system is developed to target the functionality of human olfactory system to serve people who are insensate to
smell and also as a safety alarming system in hazardous environment. This paper elucidates the various methods of implementation
of olfactory system beginning with traditional and extending to the latest modern techniques. There are different techniques like
Metal Oxide Semiconductors(MOS) based sensors, optical methods, acoustics and application-specific sensors of various ranges,
which further leads to extension of using advanced techniques like Artificial Neural Network, Recurrent Neural Network and
Convolutional Neural Network. Usage of several neural networks gives us a minimal sampling time and also they do not require pre-
processing which are being implemented in traditional networks such as Linear Discriminant Analysis (LDA),Back Propagation
Neural Network(BPNN), Support Vector Machine (SVM). Further, Colloidal Quantum Dots (CQD) gas sensors are used for their
reduced size and low power consumption in comparison to traditional gas sensors.

Keywords: Olfaction system, Sensors, Odour, Artificial neural networks, Machine learning, Colloidal Quantum Dots.

© 2020 by Advance Scientific Research. This is an open-access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/)
DOI: http://dx.doi.org/10.31838/jcr.07.09.107

INTRODUCTION Primitive method for detection of specific gas

Olfaction is generally a natural process of responding to According to Winson web article on sensors [2] ,different
chemical stimuli that results in the sense of odour. Olfaction is sensors were used for specific gases like Carbon Di-Oxide
a process whereby humans directly interact with the real Sensors, Hydrogen sulphide and methane. The most common
world. People with inability to do this olfaction process conventions for such sensors are Infrared (IR) gas sensors and
naturally are called anosmic. This affects the person’s normal the chemical gas sensors .In monitoring indoor air quality , the
living. In order to aid people with insensitivity to smell, measurement of CO2 is important, in the form of a
artificial olfaction system came into existence and plays a capnographic device. CO2 in gaseous environment is observed
significant role. Naturally the enzymes that the receptor by the characteristic absorption and the key components are
detects are further responsible for the classification of smell, generally infrared source, wavelength filter, and an infrared
whereas in artificial olfaction system, the abilities are detector. The most often used sensors for measuring CO2 is
restricted and the detection is purely based on the hardware Nondispersive infrared (NDIR) sensors that have
and software parts, and generally adhere to specific odour characteristics with good sensitivities of 20–50ppm. The
detection. This paper focuses on various methods that are Chemical CO2 gas sensors are generally based on polymer or
available to serve the purpose. The methods vary based on the hetero polysiloxane with major merit of consuming very less
form of input that the designed system takes to determine the energy. As the size is reduced, it fits into microelectronic-
type of odour. Few methods vary in how the data is being based systems. Most sensors are fully calibrated before
processed and few are based on the techniques used to analyse shipping is done. Along the time, the null point of the sensor
the obtained samples. With the use of advanced machine needs to be calibrated to maintain sensor’s long term stability.
learning techniques, the system can be trained using various To monitor the quality of air, CO2 sensors can be used. H2S
sample sets to match the performance of natural olfaction Sensor: This sensor is mostly under constant development
system. No matter how far the technological advancements because of the hazardous and corrosive nature of Hydrogen
are, there is still probability of error which sometimes is sulphide air.
negligible, and few of them can be eliminated by regular This method identifies hydrogen sulphide in the hydrogen feed
maintenance of system, and by training it with new data sets. stream of fuel cells for the prevention of catalyst formation
Few of the applications of artificial olfaction system include a and also used in personal protective equipments in order to
safety system for manual scavengers, chemical industries and alert the presence of hydrogen sulphide gas . The maximum
in Mining. sensitivity is 1000ppm.

ARTIFICIAL OLFACTION TECHNIQUES Gas chromatography in the detection of Hydrogen(H2

Various sensor modules with specific functionality )and CH4
Correlation using separate sensors Fan Gao et al., describes that, an electronic nose is used for the
Chandler Kemp et al., elucidated a gas leakage technique that detection of H2 and CH4 in breath samples which contains
mainly deals with Carbon Monoxide (CO), Hydrogen Sulphide alveolar air from lungs [3]. Therefore using a single sample of
(H2S) and Methane (CH4) sensor. CO Sensor Module has digital the alveolar air, e-nose will measure the concentration of H2
input signals being high and low [1]. There is no gas detected if and Methane. The main principle is gas chromatography, a
the input is low and if the input signal is High, gas has been technique used to separate and analyze the
detected and LCD displays Carbon monoxide Gas Leakage and samples(Fig.1).MOS sensor is used in this method for the
LED glows. Based on input of the switch, corresponding LED’s detection of every component after they are pushed into the
will glow with three different levels low, medium, high. chromatographic column which consists of a three-way valve ,
Beyond threshold (35ppm) workers are alerted. The flow valve, pump and flow valve .The gas concentration is
operations of hydrogen sulphide sensor module and methane displayed within 50 seconds. The characteristic peak overlaps
sensor module are similar to CO sensor module. However the when the flow rate of H2 and CH4 is high and when the flow
threshold is 40ppm for H2S and 750ppm for Methane. rate is low they are separated .With greater reliability than
other test, this method serves its purpose.

Journal of critical reviews 539

 DIFFERENT ARTIFICIAL OLFACTION TECHNIQUES - ELECTRONIC NOSE

Figure:1 Gas chromatographic detection

Development of an Electronic Nose by detection of odor peaks in varied conditions. The odor is
Leonardi et al., considers that the E-Nose is developed to detected with respect to the change in resistance by analyzing
monitor all the possible environmental odors [4] and focuses the sample airflow with and without neutral air. The collected
on measuring the performance of the E-Nose by checking its samples are given to the E-nose during the training period in
ability of correct classification of odours. Laboratory tests order to recognize it correctly during the period of testing
check the ability to distinguish between the odors. Their (Fig.2). The K Nearest Neighbour algorithms are used in
concentration values are measured and use field tests to check classification of odors and one e-nose differs from the other by
effectiveness, capability, and stability to continuously monitor having features like threshold, less sensitivity to noise, etc.

Figure 2: E-nose schematic diagram

Composite Feature Extraction Based on Image Covariance calculated from two sets of primitive variables instead of two
Matrix: primitive variables. Every set of primitive variables is termed
Chunghoon Kim et al.,elucidates pattern recognition where in as “composite vector”. The Primitive variables are predefined
data is commonly stored in vectors and its elements known as in a data sample as window. Hence this work focuses more on
primitive variables[5]. The Primitive feature extraction vectors for analysis.
methods such as Principal Component Analysis (PCA), Big Data
Analysis (BDA), or Linear Photodiode Array (LDA) use the Toxic gas monitoring based on Arduino
covariance of the primitive variables. The features are F. Tsow et al.,proposed a system that deals with a safety
extracted by solving the particular objective function in every application designed to detect harmful gases and serve as an
method that is defined using different types of covariance electronic nose in the toxic environment using various sensors
matrices. However, dealing with high-dimensional data, a huge using the ATmega328p microcontroller [6]. The toxicity level
number of possible combinations is expected to be computed of the gases is detected by comparing it with its threshold
for obtainment of the covariance matrix. As there are very high value and indicates the users by sending SMS to the mobile via
correlations between neighbouring primitive variables, all GSM module connected with Arduino . The threshold values
combinations usage has become redundant. Above all, it is are based on real time analysis results. This system is designed
likely to encounter Sequence to Sequence Similarity (SSS) to alert when the level of toxicity exceeds the threshold.
problem in the process of eigenvalue analysis of the covariance
matrix. The feature extraction methods are based on the
covariance matrix of image, wherein the covariance is

Journal of critical reviews 540

 DIFFERENT ARTIFICIAL OLFACTION TECHNIQUES - ELECTRONIC NOSE

The Swipe Card Model of Odorant Recognition connected to a valve regulating the sample airflow directed to
Jennifer C. Brookes et al.,emphasizes on the theory of odor the sensor chamber. During the process of “cleaning”, or the
recognition[7]. The Swipe Card model is designed based on reference phase, “neutral air” flows over the present sensors.
Turin’s theory which considers the molecular vibrations Therefore based on the air flow, detection is done.
caused by the tunneling of inelastic electrons. This becomes
vital for the activation process of olfactory receptors which Vapour Classification Using Composite Features
depends on the amount of energy absorbed by the Sang-Il Choi et al.,deciphered that the Volatile Organic
molecule.Both shape of the molecule and additional Compounds (VOC) measurement data used consists of 8
information such as vibrations are considered in this model for different classes like acetone, benzene,toluene, heptane,
odor recognition. The main advantage of this model is its cyclohexane, methanol, propanol and ethanol[9]. There are
predictability. 160 samples in data set, i.e., N = 160. Each sample was
measured through 16 channels over 2K-time points, which can
Concept of fluxing neutral air for continuous monitoring be viewed as a 16 × 2K matrix. In order to make composite
using “older” EOS 835, and the “new” EOS 507. vectors more effective, the matrix was transformed into a
Licinia Dentoni et al., depicted that the EOS 835 was first 32K-dimensional(D) vector using a lexicographic ordering
designed for the use in laboratory, and an initiative for the operator. A composite vector is constructed by grouping
development of a specific approach to be used for continuous adjacent elements and moving as much as the step size of the
monitoring of environmental odours[8]. Those two were shift. This is how the VOC is measured .
equipped with six MOS sensors, that responds to the presence
of odorous compounds in the analyzed air by changing their Detection using MAX30100 Oximetry Sensor
corresponding resistance with respect to specific reference Augustine Onubeze explained the oximetry sensor and the
conditions. These are obtained by fluxing neutral air (which is positives in using it [10]. MAX30100 is a complete oxymetric
clean, non-odorous air, obtained by filtration using activated system that has an inbuilt memory buffer with optics
carbon or other generation devices of neutral air). The optimized electronics, high Signal to Noise Ratio (SNR), using
variation of resistance towards reference produces a response low pass filters (LPF) and analog signal processing with
curve, and the significant features are extracted for further comparatively reduced noise (Fig.3).Two LEDs are included in
numerical elaborations and classification. The older EOS 835 this sensor, an infrared and red LED, which is considered as a
has two air inlets: one is connected to a filter with active visual source and further processing is carried using the
carbon and silica gel for the realization of the reference analog signals.
“neutral air”, whereas the second (“sample air” line) is

Figure 3: System using MAX30100 Oximetric sensor

Flammable gas and smoke sensor use of MQ6 Gas sensor. The flammable gas concentration is
Bo Cheng et al., [11] annotates that the flammable gas and the 300 to 10,000ppm and operates at -20 to 50 degrees Celcius
smoke sensor are used to measure the congregation of and low power consumption.
combustible gas in and around as analog. Figure 4 portrays the

Journal of critical reviews 541

 DIFFERENT ARTIFICIAL OLFACTION TECHNIQUES - ELECTRONIC NOSE

Figure 4:System using MQ6 sensor

E-nose using CQDs gas sensors Convolutional Neural Network (CNN) based Electronic
Long Li et al., gave an overview of data processing, quantum nose
dot material, and analysis that leads to the achievement of Pei-Feng Qi et al., unraveled a technique based on CNN to
small size and low power consumption [12]. The traditional reduce the sampling time of the electronic noses and also to
gas sensors were larger in size and have high power simplify the processing procedure of data[13]. The traditional
consumption. This led to the use of Colloidal quantum dots methods like Linear Discriminant Analysis, Back Propagation,
(CQDs) gas sensors. CQDs are three-dimensional (3D) Support Vector Machine, and neural networks required signal
semiconductor materials that have been used in the detection pre-processing and feature generation. There is no need of
of Nitrogen dioxide(N02), Hydrogen(H2), and Hydrogen signal pre-processing when using CNN as shown in Fig 5. The
Sulphide(H2S). This device is composed of array of sensors CNN structure consists of layers for high-level reasoning. The
such as In2O3, TGS 2602, FSP-SnO2, quantum dots, WO3 results are presented by the output layer. Thus the CNN usage
quantum dots and NiO quantum dots. minimizes the sampling time from 364 seconds to 15 seconds
and also two convolution layers with two pooling layers
Nose using Advanced Neural Network Techniques including a fully connected layer and an output layer. The layer
of convolution is CNN’s core building block. To reduce the
number of computations in the network, pooling layers is used
and thereby higher accuracy is obtained.

Figure 5: CNN Structure

Urban Monitoring using E-nose based on CNN phenol and diphenylamine. The affected areas were
Andres Gongora et al, disentangled a probable E-nose that was immediately evaluated with a portable electronic nose where
established to validate the chemical spillage that occured in the data was gathered with deep learning to detect smells
Spain [14]. The residents of the southern part of Spain noticed similar to phenol. Never minding, this method had its
an odour similar to gasoline (petrol) or plastic in 2007 and limitations since it could not determine the exact
finally found it to be the combinations of chemicals substances concentration of gases.
in the town's drainage system. There were abnormal levels of

Journal of critical reviews 542

 DIFFERENT ARTIFICIAL OLFACTION TECHNIQUES - ELECTRONIC NOSE

Multilayer Perceptron Neural Network (MLPNN) for a Neural Networks (NN). [4-4-1] MLPNN was used, which
Portable Electronic Nose refers to four input layers, four hidden layers and an output
Chih-Heng Pan et al., focused on implementing power-efficient layer with several nonlinear activation functions. The sensor
MLPNN using an analog VLSI circuit to implement a classifier sends data in the form of normalised voltage between 0.85V to
in a portable Electronic-nose[15]. Algorithms that are most 0.95V range as shown in Figure 6. When sensor reacts with the
odor molecules its resistance decreases and the change in
resistance corresponds to the activity of the sensor. The
trained and tested system provides an accuracy of 91.7% with
efficient in recognizing and identifying the odor patterns are 0.553mW power consumption.

Figure 6: MLPNN based electronic nose

CONCLUSION 6. F. Tsow, E. Forzani, R. Wang, R. Tsui, S. Mastroianni, C.

Progressing from the traditional olfaction techniques that Knobbe, et al. A fixed and wireless sensor design for real-
made use of gas chromatography to making use of advanced time monitoring of toxic environmental volatile organic
artificial neural network and machine learning techniques, the compounds. IEEE Sensors Journal 2013; 9: 1734–40.
sampling time and accuracy were greatly improved. This 7. Brookes J, Horsfield A, Stoneham AM. The Swipe Card
further increased the effectiveness of the system. The artificial Model of Odorant Recognition. Sensors 2012; 12:15709-
olfaction system is indeed evolved over time, providing us 49.
with its ability to monitor continuously by detecting odour 8. Licinia Dentoni, Laura Capelli, Selena Sironi, Renato Del
peaks in various conditions and proved to be beneficial across Rosso, Sonia Zanetti, Matteo Della Torre. Development of
various fields. The inference is that these techniques are not an Electronic Nose for Environmental Odour Monitoring.
just the replacement for those who have inability to detect and Sensors (Basel) 2012;12: 14363-81.
identify smell, but also can be used in various fields where the 9. Sang-Il Choi, Gu-Min Jeong, Chunghoon Kim.
environment is a threat to human subjects .The application of Classification of Odorants in the Vapor Phase Using
these techniques are innumerable with each method being Composite Features for a Portable E-Nose System.
uniquely designed and can be transformed based on the Sensors (Basel) 2012;12: 16182-93.
requirement of different purposes. 10. Augustine Onubeze. Developing a Wireless Heart Rate
Monitor with MAX30100 and nRF51822. 2016.
REFERENCES 11. B. Cheng, S. Zhao, S. Wang, J. Chen. Lightweight Mashup
1. Chandler E. Kemp, Arvind P. Ravikumar, Adam R. Brandt. Middleware for Coal Mine Safety Monitoring and Control
Comparing Natural Gas Leakage Detection Technologies Automation. IEEE Transactions on Automation Science
Using an Open-Source “Virtual Gas Field” Simulator. and Engineering 2017;14: 1245-55.
Environmental science and technology 12. Long Li, Hua Yuo Li, Zhixiang Hu, Jingyao Liu, Licheng
2016; 4: 4546-53. Zhou ,Zhilai Tian et al. Construction and Simple
2. Winson. Web article on different gas sensors. https:// Application of a Primary Electronic Nose System.
www. winson_ sensor.com.2019. International Symposium on Olfaction and Electronic
3. Fan Gao, Xi Zhang, Xusheng Zhang, Min Wang, Ping Wang. Nose 2019; 1-4.
Virtual electronic nose with diagnosis model for the 13. Pei-Feng Qi,Qing-Hao Meng, Ming Zeng. A CNN-based
detection of Hydrogen and Methane in breath from simplified data processing method for electronic noses.
gastrointestinal bacteria. International Symposium on ISOCS/IEEE International Symposium on Olfaction and
Olfaction and Electronic Nose 2017; 1-3. Electronic Nose (ISOEN) 2017; 1-3.
4. Leonardi S G, Cannistraro M, Patti E, Aloisio D, Donato N, 14. Andres Gongora, David Chaves, Javier
Neri G et al. Development of Electronic-Nose Gonzalez. Urban Monitoring of Unpleasant Odors with a
Technologies for Biomedical Applications. Sensors and Handheld Electronic Nose. International Symposium on
Microsystems 2013; 475-78. Olfaction and Electronic Nose 2019.
5. Chunghoon Kim, Chong-Ho Choi. A discriminant analysis 15. Chih-Heng Pan, Hung-Yi Hsieh, Kea-Tiong Tang. An
using composite features for classification problems. Analog Multilayer Perceptron Neural Network for a
Pattern Recognition 2007; 40: 451-60. Portable Electronic Nose;13:193-207.

Journal of critical reviews 543