ble at ScienceDirect

Engineering Science and Technology, an International Journal 18 (2015) 229e234

brought to you by COREView metadata, citation and similar papers at core.ac.uk

provided by Elsevier - Publisher Connector

Contents lists availa

HOSTED BY

Engineering Science and Technology,an International Journal

journal homepage: http: / /www.elsevier .com/locate/ jestch

Full length article

Quantitative recognition of flammable and toxic gases withartificial neural network using metal oxide gas sensors inembedded platform

B. Mondal a, *, M.S. Meetei a, J. Das b, C. Roy Chaudhuri c, H. Saha d

a Department of Electronics and Communication Engineering, Tezpur University, Tezpur 784028, Indiab Department of Physics, Jadavpur University, Kolkata 700032, Indiac Department of Electronics and Telecommunication Engineering, Indian Institute of Engineering Science and Technology, Howrah, Indiad Centre of Excellence in Green Energy and Sensors Systems, Indian Institute of Engineering Science and Technology, Kolkata 711103, India

a r t i c l e i n f o

Article history:Received 15 November 2014Received in revised form19 December 2014Accepted 31 December 2014Available online 31 January 2015

Keywords:Neural networkReliable detectionSensor array

* Corresponding author. Tel.: þ91 03712 267007/8/E-mail address: biplobm@tezu.ernet.in (B. MondalPeer review under responsibility of Karabuk Univ

http://dx.doi.org/10.1016/j.jestch.2014.12.0102215-0986/© 2015 Karabuk University. Production anlicenses/by-nc-nd/4.0/).

a b s t r a c t

Artificial Neural Network (ANN) based pattern recognition technique is used for ensuring the reliableevaluation of responses from an array of Zinc Oxide (ZnO) based sensors comprising of pure ZnO nano-rods and composites of ZnOeSnO2. All the sensors were fabricated in the lab. The paper first reports thedevelopment of an artificial neural network based model for successfully recognizing different con-centration of hydrogen, methane and carbon mono-oxide. Feed forward back propagation neuralnetwork was used for the classification of the gases at critical concentrations. The optimized ANN al-gorithm is then embedded in the microcontroller based circuit and finally verified under lab conditions.© 2015 Karabuk University. Production and hosting by Elsevier B.V. This is an open access article under

the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Qualitative and quantitative detection of flammable and toxicgases like hydrogen (H2), methane (CH4) and carbon mono-oxide(CO) in coal mines, petroleum industries and gas storage plantsare of great importance for the safely of workers, health hazardsand to avoid risk of accidental explosion. H2 is a potential source ofenergy of future and is used in fuel cell [1], H2 engine cars, chemicalindustry etc. H2 is a colorless, odorless and highly explosive gashaving flammability in the range (4e75%) in air [2]. In coal minespresence of methane is always expected, which has caused mostloss of life than any other gas. Methane in the range (5e14%) in air isflammable [2]. CO is also a colorless and tasteless gas which isbelieved to be the most dangerous gas in mine. CO in the range of50 ppm is fatal and is also flammable and explosive [2] in higherrange of concentrations.

Oxide based semiconductor sensors present advantages likehigh sensitivity to low concentration of gases, less fabrication cost,long term stability, low power consumption etc. whichmakes them

9x5266.).ersity.

d hosting by Elsevier B.V. This is a

attractive material for the detection of toxic, hazardous andcombustible gases. A wide range of sensors based on semi-conductor metal oxides including ZnO, TiO2, SnO2, WO3, In2O3, etc.were used for the detection of gases like H2 [3], CH4 [4], LPG [5], CO[6] etc. Both ZnO and SnO2 are n-type wide band semiconductorswhich have attracted remarkable attention as highly sensitivesensor material. These material were expected to be superior ma-terial for sensing gases like H2 [7,8], CO [9,10], and CH4 [10,11]. Inspite of the high response, metal oxide semiconductor (MOx) basedsensor presents a serious drawback of poor selectivity, which hin-ders their commercial application. Therefore, MOx sensors fallshort when a precise detection of specific concentration of gaseouscomponent is required or when they are used to discriminate amixture of gas. Drift in the sensor output due to ageing, poor sta-bility or high sensitivity to humidity of the ambient environmentare also serious drawbacks of metal oxide based sensors. Both driftand poor selectivity make the sensor output unreliable [12].

Many authors have suggested various ways to overcome theproblem of cross-sensitivity to achieve selective sensor which mayinclude, a) addition of noble catalytic metal to promote the reactionto specific gas [13], b) operating the sensor in dynamicmode [14], c)surfacemodification [15] and d) use of multi-compositional sensingfilm [16,17]. However, extremely selective sensors are very

n open access article under the CC BY-NC-ND license (http://creativecommons.org/

Table 1Sensor type and operation temperature.

Sensor Sensing material Operating temperature (�C)

S1 ZnO 150S2 200S3 250

S4 ZnOeSnO2 150S5 200S6 250

B. Mondal et al. / Engineering Science and Technology, an International Journal 18 (2015) 229e234230

expensive and operating the sensor in dynamic mode makes themeasurement system complex. Although, the problem of cross-sensitivity can be addressed with any of the above mentionedways, the troubles of reliability due to the drift in sensor output stillpersist. A potentially more cost effective way to perform suchmeasurement is to use a sensor array coupled with a pattern clas-sification system [18e20]. The use of multiple sensors with anappropriately modeled pattern classifier is therefore expected toimprove the reliability in the recognition of specific concentrationof gaseous components. This will also improve the selectivity for aparticular gas when there is a mixture of gases.

Various pattern analysis methods like principle componentanalysis (PCA) [21], partial least square analysis (PLS) [22], andartificial neural network [23] etc. can be usedwith gas sensor arraysfor the classification and identification problem. Among thesemethods artificial neural network is reported to provide superiorperformance when quantification or multicomponent mixtureclassification is required [24]. Multilayer perceptron (MLP) andradial basis function (RBF) are the two commonly used neuralnetwork algorithms used for pattern recognition. However, theneuron connectivity in RBF is comparatively lower thanMLP. This isthe reason why the mean square error of an RBF network is lesscompared to an MLP one having the same number of neurons.Therefore the parameters for tuning the network are higher in MLP.MLP with nonlinear transfer function, like sigmoid function, allowthe network to learn any inputeoutput relationship adjusting theweights and biases in the network by a gradient descent of tech-nique known as back propagation of errors. Many authors havereported the successful utilization of ANN in many applicationsincluding pattern classification, identification, decision making etc.[25e28]. Multilayer perceptron neural network have been used byJoo et al. to identify breast nodule malignancy using sonographicimages [26]. Lv et al. reported the detection of low concentration offormaldehyde in binary gas mixtures using micro gas sensor arrayand a neural network [27]. Lee et al. reported about a combustiblegas recognition system including sensor array and neural networkbased pattern recognition [28].

Most of the commercially available gas recognition systems arePC based and hence are bulky. Embedded systems can be developedwith a suitable training function which can automatically read thesensor input from the sensor array and produce the neural networkoutput. This may overcome the drawback of PC based gas patternrecognizer, making them portable and convenient to the user. Inthis work, we report the development of a prototype toxic gasrecognizer using sensor array and multilayer neural network. Thesensors were based on pure ZnO nano-rods and ZnOeSnO2 com-posites which were fabricated in the lab. The sensors were exposedto H2, CH4 and CO in different concentration ranges. The datagenerated from the sensors were used to train a multilayer neuralnetwork with back propagation learning algorithm, which canreasonably produce output to unseen input data. A process flowchart for realization of the ANN in microcontroller is also formu-lated and verified by downloading the network in the microcon-troller. The use of an array of sensors instead of one or two alongwith a suitably trained neural network can increase the reliability ofthe measurement.

2. Sensor fabrication

The sensor used in the experiment consists of pure oxide of zincas well as composite oxide of zinc and tin, as the active gas sensingelement. For the fabrication of sensors separate sols were preparedfor pure ZnO and composite ZnOeSnO2 with polyvinyl (PVA) ascapping agent. The detail material synthesis and methodical char-acterization of the sensor materials were reported in our previous

communications [29,30]. In a typical procedure, sol for pure ZnO isprepared by mixing 0.6 g of zinc acetate di-hydrate[Zn(CH3COO)2$2H2O] with aqueous solution of PVA under con-stant stirring. The 1.0 M sodium hydroxide (NaOH) solutions wasthen mixed drop wise to the above solution until a pH valuereached 7.0. The solution was then stirred at 80 �C for 1 h. Sol forZnOeSnO2 sensor is prepared by mixing appropriate amount ofzinc acetate dehydrate (Zn(CH3COO)2$2H2O) and stannic chloride(SnCl4$5H2O) with PVA. 1.0 M NaOH solution was used as reducingagent until pH of the solution becomes 7.0. The final solution wasstirred at 110e120 �C for 1 h. Finally, the resulting precursors werespin-casted on the cleaned <100> oriented Si/SiO2 substrates fol-lowed by annealing at 550 �C for 30 min in a tube furnace.

3. Measurement techniques

A total of six numbers of sensors were used for the collection ofdata generated from the sensors when exposed to three differentgases namely hydrogen, methane and carbon monoxide. The sen-sors were exposed to each test gas in different concentrations (0.1,0.3, 0.5 and 1.0%) separately. The sensor types and their operatingconditions are given in Table 1. The detail arrangement of thesensor measurement set up and sensor structure were reportedelsewhere [29,30]. Typically, the sensors were placed inside aclosed test chamber with suitable arrangement for gas inlet andoutlet. Measured amount of test gas was fed to the test chamberusing Alicat® mass flow controllers (MFC). The test chamber wasfitted with heating arrangement that would maintain a uniformtemperature (with an accuracy of ±3 �C) inside the chamber. Thesensors were designed to operate in a resistive mode and Fig. 1shows a model of the sensor measuring circuit. PdeAg contactelectrodes were made on the top of the sensing layer by e-beamevaporation technique in order to provide electrical connection. Anelectrometer (Agilent U1253B) was used tomonitor the variation ofsensor resistance. The resistance variation of the sensor is con-verted into voltage output by the measurement circuit.

4. Design of ANN in matlab®

A feed forward multilayered neural network (Fig. 2) trained by agradient descent technique, also called back propagation of errors,was used to process the data arising from the sensors to investigatethe feasibility of discrimination of gases. The networkwas designedin the Neural Network toolbox in MATLAB® with six input neuronequal to the number of sensors used, two hidden layers with tenand twelve neurons in the hidden layer 1 and 2, respectively, andthirteen output nodes corresponding to a specific concentration ofa specific gas, as shown in Table 2. Sigmoid functionwas used as theactivation function for the hidden and the output neurons. Trainingof the network was performed using supervised back propagationlearning algorithmwhich was first initialized with randomweightsand biases. During the training process, weights and biases of thenetwork were iteratively adjusted in order to minimize the

Fig. 1. Schematic model of the circuit diagram.

Table 2Test gas and sensors designated output.

Test gas Concentration (%) Designated output

No gas 1000000000000

H2 0.1 01000000000000.3 00100000000000.5 00010000000001 0000100000000

CH4 0.1 00000100000000.3 00000010000000.5 00000001000001 0000000010000

CO 0.1 00000000010000.3 00000000001000.5 00000000000101 0000000000001

B. Mondal et al. / Engineering Science and Technology, an International Journal 18 (2015) 229e234 231

performance function of the network. The default performancefunction for feed-forward networks is mean square error (MSE),which is the average squared error between the network outputand the target output and is defined as [31],

MSE ¼ 1N

XN

i¼1

ðTi � YiÞ2 (1)

where T is the target and Y is the network output.In this study 260 data samples were used out of which 70% data

were used as training data, 20% as validation and 10% as testingdata. The performance of a feed forward MLP neural network issignificantly affected by the network parameters like number ofhidden layers, number of neurons in each hidden layer, trainingstyle and activation function used. The weight matrix for the neu-rons was initially chosen randomly and the training algorithm usedwas Resilent Backpropagation (RP).

4.1. Optimization of neural network structure

MLP with a single hidden layer and nonlinear transfer functionlike sigmoid function can learn any non-linear relationship. How-ever, it has been reported earlier by various authors that theaddition of a second hidden layer improves the efficiency of thenetwork to a great extent and also drastically reduces the totalnumber of hidden neurons [32e34]. A network with two hiddenlayers was chosen in the present case and the number of neurons ineach layer is varied in search of best performance. The networkgave the lowest MSE (0.005) when the number of neurons in hid-den layer 1 and layer 2 was ten and twelve, respectively. Fig. 3 plotsthe MSE of the network for different combination of neurons in thetwo hidden layers.

Fig. 2. Structure of n

4.2. Optimization of activation function

Two sigmoid (Logsig and tansig) and one linear (pureline)activation functionwere studied in this work. These three functionsare defined in eq. (2)e(4) [35]. A set of 27 combinations of theseactivation functions, as shown in Table 3, for the two hidden andone output layer are investigated for achieving best network per-formance. The network performance for the various combinationsof these functions is presented in Fig. 4. For the given set of data onecan observe from Fig. 4 that the network gave the best performancewhen ‘logsig’ activation function is used for both the hidden layersand the output layer (corresponding to combination code 8 inTable 3)

LogsigðnÞ ¼ 1�1þ e�n

� (2)

TansigðnÞ ¼ 21þ e�2n � 1 (3)

PurelinðnÞ ¼ n (4)

5. Implementation of ANN in microcontroller

The optimized ANN is implemented in an Arduino due micro-controller board based on the Atmel (SAM3X8E ARM Cortex-M3CPU) 32-bit ARM core microcontroller. It has 54 digital input/output pins, 12 analog inputs, a 84 MHz clock, 2 DAC (digital to

eural network.

Fig. 3. Optimization of number of neurons in the two hidden layers.

Fig. 4. Network performance for various combination of activation function.

B. Mondal et al. / Engineering Science and Technology, an International Journal 18 (2015) 229e234232

analog), 96 Kbytes of SRAM and 512 Kbytes of flash memory andcan be powered by a USB connector or with an external powersupply.

The required data for implementing the ANN in microcontrollerare the weight matrix, biases and mean and standard deviation ofthe input data set. The mean and standard deviation of the inputdata are required for data normalization. The input data arenormalized using the following relation [36]

Xn ¼ ðXi �MiÞ=Sd (5)

In this equation Xn is the normalized data, Xi is the input data andMi, Sd are mean and standard deviation of the given set of data.

Table 3Combination of activation function.

Activation functioncombination code

Activation function

First hidden layer Second hidden layer Output layer

1 Purelin Purelin Purelin2 Purelin Purelin Logsig3 Purelin Purelin Tansig4 Purelin Logsig Purelin5 Purelin Tansig Purelin6 Logsig Purelin Purelin7 Tansig Purelin Purelin8 Logsig Logsig Logsig9 Logsig Logsig Purelin10 Logsig Logsig Tansig11 Logsig Purelin Logsig12 Logsig Tansig Logsig13 Purelin Logsig Logsig14 Tansig Logsig Logsig15 Tansig Tansig Purelin16 Tansig Tansig Logsig17 Tansig Purelin Tansig18 Tansig Logsig Tansig19 Purelin Tansig Tansig20 Logsig Tansig Tansig21 Logsig Tansig Purelin22 Logsig Purelin Tansig23 Tansig Logsig Purelin24 Tansig Purelin Logsig25 Purelin Logsig Tansig26 Purelin Tansig Purelin27 Tansig Tansig Tansig

The weight and bias matrices are obtained after training thenetwork in MATLAB® environment. For the network under study(six input neuron, ten and twelve neuron in hidden layer one andtwo respectively and thirteen output neuron), we have [10 � 6]weight matrix for the first hidden layer, [12� 10] weight matrix forthe second hidden layer and [13 � 12] weight matrix for the outputlayer. In addition, we also have three bias matrices for the twohidden layers and one output layer.

The process flow chart for the implementation of the ANN inmicrocontroller is shown in Fig. 5. The weights and biases of thetrained network connections were first stored in the microcon-troller memory. The output of the first hidden neurons is thencomputed using the following expression (6)e(8),

Pj ¼ flogsigXn

i¼1

Wji$Xi þ bj (6)

where i ¼ 1e6 is the number of input nodes and j ¼ 1e10 is thenumber of neurons in the first hidden layer.

The output of the second hidden layer is computed using thefollowing expression

Qk ¼ flogsigXn

j¼1

Wkj$Pj þ bk (7)

where k ¼ 1e12, is the number of neurons in the second hiddenlayer.

And finally the output of the output neurons is computed usingthe following expression

Ol ¼ flogsigXn

k¼1

Wlk$Qk þ bl (8)

where l ¼ 1e13, is the number of output neurons.This course is coded in Arduino software environment “Arduino

1.5.2” based on Cþþ programming language. The code is thenconverted into a HEX file before downloading into themicrocontroller.

Fig. 5. Flowchart for ANN implementation.

Fig. 6. Regression values for training, validation and test.

B. Mondal et al. / Engineering Science and Technology, an International Journal 18 (2015) 229e234 233

B. Mondal et al. / Engineering Science and Technology, an International Journal 18 (2015) 229e234234

6. Results and discussion

After successful completion of the learning process, the pro-posed network produced very high recognition rate. To verify theperformance of the system, we plot in Fig. 6, the regression valuesfor training (a), validation (b), testing (c) and also for the whole set(d) of data. The regression values give the relationship between theoutputs of the network with that of the targets. Fig. 6 indicates thatthe output is closely fit with the target with an accuracy of ~94%during training, ~96% during validation and ~95% during testing.The overall accuracy of the network was 94.6%.

The network was successfully implemented in embedded plat-form. The network is implemented in the microcontroller using185 KB in the flash memory and the execution time was nearly100 s. In order to verify the performance of the recognizer, thenetwork in both the environment (MATLAB® and microcontroller)was fed with the same inputs. Approximately 98% of the outputfrom both the system found to match perfectly.

7. Conclusion

Using an array of six, moderately selective, ZnO and ZnO-SnO2based sensors, the proposed pattern recognition methods allow usto construct the simple pattern classifier, which is able to recognizeand classify hydrogen, methane and carbon oxide gas samples ofdifferent concentrations with good reliability and accuracy. Multi-layer perceptron is used for gas recognition architecture. Theoptimized network consists of six input neurons equal to thenumber of sensors, two hidden layers with ten and twelve neuronsin hidden layer-1 and hidden layer-2, respectively, and thirteenoutput neurons. The network is fed with 260 experimentally ob-tained data sets out of which 70% of the datawere used for training,20% were used for validation and 10% for testing. The network gavean overall accuracy of 94.6% when trained with RP algorithm. Aprocess flow chart for the realization of ANN in microcontroller isthen formulated which is later written in Arduino software envi-ronment and downloaded in the microcontroller to develop aprototype gas recognition system. The developed system can effi-ciently produce reasonable output when fed with input data.

References

[1] N. Sammes, Fuel Cell Technology: Reaching toward Commercialization,Springer.

[2] http://www.mathesongas.com.[3] A. Qurashi, M. Faiz, N. Tabet, M.W. Alam, Low temperature synthesis of hex-

agonal ZnO nanorods and their hydrogen sensing properties, SuperlatticeMicrostruct. 50 (2011) 173e180.

[4] P. Bhattacharyya, P.K. Basu, H. Saha, S. Basu, Fast response methane sensorusing nanocrystalline zinc oxide thin films derived by solegel method, Sens.Actuat. B 124 (2007) 62.

[5] C.S. Prajapati, S.N. Pandey, P.P. Sahay, Sensing of LPG with nanostructured zincoxide thin film grown by spray pyrolysis technique, Physica B 406 (2011)2684e2688.

[6] R.K. Joshi, Q. Hu, F. Alvi, N. Joshi, A. Kumar, Au decorated zinc oxide nanowiresfor CO sensing, J. Phys. Chem. C 113 (2009) 16199e16202.

[7] O. Lupan, V.V. Ursaki, G. Chai, L. Chow, G.A. Emelchenko, I.M. Tiginyanu,A.N. Gruzintsev, A.N. Redkin, Selective hydrogen gas nanosensor using indi-vidual ZnO nanowire with fast response at room temperature, Sens. Actuat. B144 (2010) 56e66.

[8] S.F. Bamsaoud, S.B. Rane, R.N. Karekar, R.C. Aiyer, Nano particulate SnO2 basedresistive films as a hydrogen and acetone vapour sensor, Sens. Actuat. B 153(2011) 382e391.

[9] S. Chang, T. Hsueh, I-Cherng Chen, B. Huang, Highly sensitive ZnO nanowireCO sensors with the adsorption of Au nanoparticles, Nanotechnology 19(2008) 175502.

[10] M.A. Andio, P.N. Browning, P.A. Morris, S.A. Akbar, Comparison of gas sensorperformance of SnO2 nano-structures on microhotplate platform, Sens.Actuat. B 165 (2012) 13e18.

[11] P. Bhattacharyya, P.K. Basu, B. Mondal, H. Saha, A low power MEMS gas sensorbased on nanocrystalline ZnO thin films for sensing methane, Microelectron.Reliab. 48 (2008) 1772e1779.

[12] G. Korotcenkov, Gas response control through structural and chemicalmodification of metal oxide films: state of the art and approaches, Sens.Actuat. B 107 (2005) 209e232.

[13] W. Lim, J.S. Wright, B.P. Gila, S.J. Pearton, F. Ren, Wei-Ta Lai, Li-Chyong Chen,Ming-Shien Hu, K. Chen, Selective-hydrogen sensing at room temperaturewith Pt-coated InN nanobelts, Appl. Phys. Lett. 93 (2008) 202109.

[14] W. Maziarz, T. Pisarkiewicz, Gas sensors in a dynamic operation mode, Meas.Sci. Technol. 19 (2008) 05520.

[15] P.K. Basu, S.K. Jana, H. Saha, S. Basu, Low temperature methane sensing byelectrochemically grown and surface modified ZnO thin films, Sens. Actuat. B135 (2008) 81e88.

[16] X. Song, Z. Wang, Y. Liu, C. Wang, L. Li, A highly sensitive ethanol sensor basedon mesoporous ZnOeSnO2 nanofibers, Nanotechnology 20 (2009) 075501.

[17] H. Woo, C.W. Na, Il-Doo Kim, J. Lee, Highly sensitive and selective trime-thylamine sensor using one-dimensional ZnOeCr2O3 hetero-nanostructures,Nanotechnology 23 (2012) 245501.

[18] V.D. Lecce, A. Amato, R. Dario, C. Martines, Air quality control for health carecentres, the application of an intelligent distributed system, 2009 IEEEWorkshop on Environmental, Energy, and Structural Monitoring Systems,EESMS-2009 Crema (Italy). pp. 27e30.

[19] I. Concina, M. Falasconi, E. Gobbi, F. Bianchi, M. Musci, M. Mattarozzi,M. Pardo, A. Mangia, M. Careri, G. Sberveglieri, Early detection of microbialcontamination in processed tomatoes by electronic nose, Food Control 20(2009) 873e880.

[20] R.F. Machado, D. Laskowski, O. Deffenderfer, T. Burch, S. Zheng, P.J. Mazzone,T. Mekhail, C. Jennings, J.K. Stoller, J. Pyle, J. Duncan, R.A. Dweik, S.C. Erzurum,Detection of lung cancer by sensor array analyses of exhaled breath, Am. J.Respir. Crit. Care Med. 171 (2005) 1286e1291.

[21] J.W. Gardner, Detection of vapours and odours from a multisensor array usingpattern recognition part principal component and cluster analysis, Sens.Actuat. B 4 (1991) 109e115.

[22] M. Barker, W. Rayens, Partial least squares for discrimination, J. Chemometrics17 (2003) 166e173.

[23] D.S. Lee, H.Y. Jung, J.W. Lim, M. Lee, S.W. Ban, J.S. Huh, D.D. Lee, Explosive gasrecognition system using thick film sensor array and neural network, Sens.Actuat. B 71 (2000) 90e98.

[24] I. Sayago, M. del C. Horrillo, S. Baluk, M. Aleixandre, M.J. Fern�andez, L. Ar�es,M. García, J.P. Santos, J. Guti�errez, Detection of toxic gases by a tin oxidemultisensor, IEEE Sens. J. 2 (5) (2005).

[25] J.I. Civera, E.G. Breijo, N.L. Mir�o, L.G. S�anchez, J.G. Baixauli, I.R. Gil, R.M. Peris,M.A. Fillol, Artificial neural network onto eight bit microcontroller for secchidepth calculation, Sens. Actuat. B 156 (2011) 132e139.

[26] S. Joo, W.K. Moon, H.C. Kim, Computer-aided diagnosis of solid breast nod-ules on ultrasound with digital image processing and artificial neuralnetwork, in: Proceedings of the 26th Annual International Conference of theIEEE Engineering in Medicine and Biology Society (EMBC '04), 2004, pp.1397e1400.

[27] P. Lv, Z. Tang, G. Wei, J. Yu, Z. Huang, Recognizing indoor formaldehyde inbinary gas mixtures with a micro gas sensor array and a neural network,Meas. Sci. Technol. 18 (2007) 1997e3004.

[28] D. Lee, S.M. Lee, D. Lee, Micor gas sensor array with neural network forrecognizing combustible leakage gases, IEEE Sens. J. 5 (3) (2005).

[29] B. Mondal, B. Basumatary, J. Das, C. Roychaudhury, H. Saha, N. Mukherjee,ZnOeSnO2 based composite type gas sensor for selective hydrogen sensing,Sens. Actuat. B 194 (2014) 389e396.

[30] B. Mondal, L. Dutta, C. Roychaudhury, D. Mohanta, N. Mukherjee, H. Saha,Effect of annealing temperature on the morphology and sensitivity of the zincoxide nanorods based methane senor, Acta Metall. Sin., English Lett. 27 (4)(2014) 593e600.

[31] http://in.mathworks.com/help/matlab/.[32] V. Kurkova, Kolmogorov's theorem and multilayer neural networks, Neural

Netw. 5 (1992) 501e506.[33] G.B. Huang, Learning capability and storage capacity of two-hidden-layer

feedforward networks, IEEE Trans. Neural Netw. 14 (2003) 274e281.[34] D. Stathakis, How many hidden layers and nodes? Int. J. Rem. Sens. 30 (8)

(2009) 2133e2147.[35] M. Dorofki, A.H. Elshafie, O. Jaafar, O.A. Karim, S. Mastura, Comparison of

artificial neural network transfer functions abilities to simulate extremerunoff data, in: International Conference on Environment, Energy andBiotechnology, IPCBEE, vol. 33, IACSIT Press, Singapore, 2012.

[36] S. Sumathi, P. Surekha, Computational Intelligence Paradigms: Theory andApplications Using MATLAB, CRC Press, Inc., 2010, ISBN 978-1-4398-0902-0.