《外文翻译-- An Electronic Nose System Based on An Array of Carbon Nanotubes Gas Sensors with Pattern Recognition Techniques》由会员分享，可在线阅读，更多相关《外文翻译-- An Electronic Nose System Based on An Array of Carbon Nanotubes Gas Sensors with Pattern Recognition Techniques（4页珍藏版）》请在金锄头文库上搜索。
1、 An Electronic Nose System Based on An Array of Carbon Nanotubes Gas Sensors with Pattern Recognition TechniquesZhao Zikai College of Mechanical and Electrical Engineering China Jiliang University Hangzhou 310018, P.R. China Hui Guohua* College of Food Science and Biotechnology Zhejiang Gongshang U
2、niversity Hangzhou 310035, P.R. China AbstractThis paper presents design of an electronic nose based on an array of aligned multi-walled carbon nanotubes (MWNT) ionization gas sensors and pattern recognition techniques for gas detection. The raw data, including discharge voltages and currents, is ac
3、quired by measurement of sensor array response and transformed to the computer by peripheral circuit, and processed by Principal Components Analysis (PCA). Back-propagation neural networks (BPNN) are applied to determine gas varieties. Results demonstrate the developed electronic nose system is capa
4、ble to identify target gases successfully and is promising for field applications. Keywords-gas sensor, gas discharge, artificial neural network, carbon nanotubes, electronic nose I. INTRODUCTION Carbon nanotubes sensors 1-8 have been explored the potential of gas sensing utilizing adsorption-desorp
5、tion characteristics, which modify the carbon nanotubes electronic properties, such as resistance, capacity and frequency, etc. Gas varieties and concentration could be determined by detecting modifications in these electronic properties. In 2003, ionization MWNT gas sensor was designed for gas dete
6、ction 9. The MWNT film was fabricated on silicon substrate and decreased the breakdown voltages to a low level. This kind of sensor used breakdown voltage to identify several gases. A self-sustaining discharge carbon nanotubes gas sensor 10 with decreased the breakdown voltage to less than 220V. Her
7、e we establish an electronic nose system based on an array of MWNT sensors and pattern recognition techniques. PCA method is used for gas varieties classifications. BPNN is trained for gas recognition. Fig. 1 Diagram of the electronic nose system. II. EXPERIMENTS A. Sensor structure and array The MW
8、NT film is synthesized on AAO template utilizing CVD method 11, 12. The super pure Al 99.999% plate with 0.5mm in thickness was cut into 2cm6cm pieces. These specimens were degreased with alcohol for 20 min. A mirror surface was achieved by electro-polishing in a 1:4 solution of HClO4 and alcohol. T
9、his step must be accomplished at 15 for 5 min. After an hour of DC 40V anodic oxidation in oxalic acid solution, the plate was taken out and washed with de-ionized water. In order to enlarge diameters of pores on AAO template, the plate was placed into phosphoric acid solution (0.3mol/L) for 17 min,
10、 then taken out and washed with de-ionized water. The preparation of AAO template was finished. The pores in alumina plate were about 60nm in diameter and 2m in depth. Cobalt particles, with the diameter of about 60nm, were used as catalyst for the MWNT fast growth. AC electrodeposition method was u
11、sed to deposit cobalt particles into pores in AAO template. The cobalt catalyst was electrochemically deposited in a CoSO47H2O solution stabilized with boric acid in AC 15V. Under these conditions, cobalt particles were 978-1-4244-4713-8/10/$25.00 2010 IEEE electrodeposited at the bottom of the pore
12、s in the AAO template. Aligned MWNT were grown by catalytic pyrolysis of 10% C2H2 and 20% H2 in a N2 carrier gas at 645 for 5 min in a quartz boat reactor. The total flow rate was 200 sccm. After catalytic growth, supply of C2H2 and H2 was shut off, and the carbon nanotubes were annealed to room tem
13、perature in nitrogen atmosphere. The structure of MWNT gas sensor is shown in Fig. 2a. It consists of Al plate electrode, polyimide thin film and MWNT electrode 18. Fig. 2b presents the side-view of the sensor. (a) (b) Fig. 2 MWNT gas sensor: (a) “Sandwich” structure of MWNT gas sensor; (b) side-vie
14、w of the sensor; The sensor array includes four sensors: S1, S2, S3 and S4. S1 and S2 are provided with 120m in interelectrode distance, and S3, S4 are 180m. S1 and S3 are used for MWNT anode sensors, while S2 and S4 are for MWNT cathode sensors. B. Methods The electronic nose works in the following
15、 way: test gas was piped into the temperature-controlled system in order that the gas could be kept in the same temperature. Then the gas flowed into gas chamber with an array of four carbon nanotubes ionization gas sensors. The following tests were conducted under the circumstances of ambient tempe
16、rature of 20 , relative humidity of 70%, and zero gas-flow. C. Measurement of sensor array response First, we conducted argon detection experiments. The interelectrode voltage (within the breakdown voltage) of the four MWNT sensors was increased, and the discharge current of MWNT sensors went up ste
17、adily. Eighteen sampling voltages were recorded and the corresponding discharge currents were transmitted to a desktop. The data was processed by PCA and BPNN. So did in Air, Nitrogen, and Carbon dioxide. Fig. 4 gave the sensor array response to gases. MWNT exhibits some unique properties since it h
18、as high aspect ratios, small radius of curvature at their tips, high chemical stability, and high mechanical strength. One of these important properties is field emission of electron 15. Also, the developed gas sensor obeys the Paschens Law 18. D. Artificial neural networks Artificial neural network
19、s have generally been accepted as a major tool in the development of intelligent system in last two decades. Their origins could be traced back to the first publication of a mathematical model of a biological neuron by McCulloch and Pitts 13. Many neural network architectures are available, such as
20、back-propagation trained neural networks 14, radial basis function neural networks, probabilistic neural networks, genetically trained artificial neural networks, adaptive resonance theory and self-organizing networks. In our electronic nose system, BPNN with sigmoidal activation function is used to
21、 process experimental data. BPNN consists of three layers: one input layer, one hidden layer, and one output layer. The inputs to the BPNN are accepted by the input layer, which fans out the inputs to the hidden layer. In the hidden layer, the data is activated and transformed before propagating the
22、m to the output layer. The output layer neurons are normally taken to be the same as for the neurons in the hidden layer, but limit the dynamic range of the output within 1. III. RESULTS AND DISCUSSIONS A. Sensor array response to gases The response of sensor array exposed to different types of gase
23、s is shown in Fig. 4a-d. Graphs show the changes in discharge current with the increase of interelectrode voltage. Fig. 4. Response of S1, S2, S3 and S4 sensors to different types of gases (a); Argon (b); Air (c); Carbon dioxide (d); Nitrogen. Discharge current response to the increasing voltage is
24、non-linear. Provided with the same interelectrode distance and voltage, discharge current of MWNT cathode sensors is higher than that of MWNT anode sensors. B. PCA data feature extraction Scatter plots in Fig. 5 shows four different groups of data for each of the four types of gases exposed to S1, S
25、2, S3, and S4 sensors. To illustrate the method, only three scatter plots have been shown here. Processed data includes the response of sensor array corresponding to 120, 160, and 220V of the testing gases. (a) 120V (b) 160V (c) 220V Fig. 5 PCA processed results of testing data in different voltage
26、of electrodes (a); 120V (b); 160V (c); 220V. PCA reduces the dimensionality of the multivariate problem 16. In our study, the input data are the discharge current of sensors corresponding to 120, 160, and 220V. PCA results in Fig. 5 indicate that characteristics of four gases can be separated from e
27、ach other. C. BPNN classification A three-layer BPNN (shown in Fig. 6) is designed to learn experimental responses. Because there are four sensors in sensor array and four types of gases to detect, input and output nodes are 4 and 4. Nodes of hidden layer are 10 by trials. To classify testing gases
28、utilizing BPNN, the experimental data sets are normalized between 0 and 1, and then they are divided into two parts: one part (80%) is used for training and the other part (20%) for BPNN test after training. BPNN demonstrates an almost 100% recognition rate. Finally, we acquire another twelve groups
29、 of raw data for BPNN validation. Results of the simulation experiments give above 80% recognition rate for the unprocessed data. Fig. 6 BPNN structure for gas detection. IV. CONCLUSIONS In this paper, we have proposed an electronic nose system based on MWNT gas sensors array with pattern recognitio
30、n techniques. The Experimental data arising from the electronic nose system is processed by PCA method. PCA results indicate that characteristics of four gases can be separated from each other. BPNN was used for gas varieties detection. The gas classification accuracy is 100%. In addition, we also u
31、se BPNN to recognize some raw experimental data acquired by detecting system for validation. The system gives out the classification of 80%. The electronic nose system has some advantages including high sensitivity, low cost, easy operation. The electronic nose system can be used in VOCs detection,
32、food quality monitoring, et al. ACKNOWLEDGEMENTS This work was supported by the National Natural Science Foundation (Grant No. 3040000461). REFERENCES 1 Y.M. Wong, W.P. Kang, J.L. Davidson, A. Wisisora-at, K.L. Soh, “A novel microelectronic gas sensor utilizing carbon nanotubes for hydrogen gas dete
33、ction,” Sens. Actuators B. vol. 93, 2003, pp. 327-332. 2 Rosa Rego, Nidia Caetano, et al., “Hydrogen/methane and hydrogen/nitrogen sensor based on the permselectivity of polymeric membranes,” Sens. Actuators B. vol. 111-112, 2003, pp. 150-159. 3 Bee-Yu Wei, Ming-Chih Hsu, et al., “A novel SnO2 gas s
34、ensor doped with carbon nanotubes operating at room temperature,” Sens. Actuators B. vol. 101, 2004. pp. 81-89. 4 M. Penza, G. Cassano, et al., “Carbon nanotubes-coated multi-transducing sensors for VOCs detection,” Sens. Actuators B. vol. 111-112, 2005. pp. 171-180. 5 M. Lucci, P. Regoliosi, et al.
35、, “Gas sensing using single wall carbon nanotubes ordered with dielectrophoresis,” Sens. Actuators B. vol. 111-112, 2005. pp. 181-186. 6 Anis Zribi, Aaron Knobloch, et al., “Micromachined resonant multiple gas sensor,” Sens. Actuators A. vol. 122, 2005. pp. 31-38. 7 Gopal K.Mor, Oomman K. Varghese,
36、et al., “Fabrication of hydrogen sensors with transparent titanium oxide nanotube-array thin films as sensing elements,” Thin Solid Films. Vol. 496, 2006. pp. 42-48. 8 T. Schneider, D. Richter, et al., “Novel impedance interface for resonant high-temperature gas sensors,” Sens. Actuators B. vol. 111
37、-112, 2005. pp. 187-192. 9 Ashish Modi, Nikhil Koratkar, Eric Lass, et al., “Miniaturized gas ionization sensors using carbon nanotubes, ” NATURE. Vol. 424, July 2003. pp. 171-174. 10 Y. Zhang, J.H. Liu, W.H. Liu, J.Y. Dou, Y.N. He, “Study of gas sensor with carbon nanotube film on the substrate of
38、porous silicon,” IEEE 14th International Vacuum Microelectronics Conference Proceedings. Boston, USA, 2001. pp. 13-14. 11 Xu Jun-Ming, Zhang Xiao-Bin, et al., “Rapid Growth of Aligned CNTs on AAO Template and Its Sonication Cutting,” Acta Phys. Chim. Sin. vol. 20, 2004, pp. 271-274. 12 G P Sklar, K
39、Paramguru, M Misra, et al., “Pulsed electrodeposition into AAO templates for CVD growth of carbon nanotube arrays,” Nanotechnology. vol. 16, 2005, pp. 1265-1271. 13 W. X. McCulloch and W. Pitts. “A logical calculus of ideas immanent in nervous activity,” Bulletin of Mathematical Biophysics. vol. 5,
40、1943, pp. 115133. 14 P. Chang, J.S. Shih. “The application of back propagation neural network of multi-channel piezoelectric quartz crystal sensor for mixture organic vapours,” Tamkang J. Sci. Eng. Vol. 5, 2002, pp. 209-217. 15 Yahachi Saito, Sashiro Uemura. “Field emission from carbon nanotubes and
41、 its application to electron sources, ” Carbon. Vol. 38, 2000, pp. 169-182. 16 J.W. Gardner. “Detection of vapours and odours from a multi-sensor array using pattern recognition,” Sens. Actuators B. vol. 4, 1991, pp. 108-116. 17 M. Penza, G. Cassano, F. Zaccaria. “Classification of food, beverages and perfumes by WO3 thin-film sensors array and pattern recognition techniques,” Sens. Actuators B. vol. 73, 2001, pp. 76-87. 18 Hui G-H, Wu L-L, Pan M et al., “A novel gas-ionization sensor based on aligned multi-walled carbon nanotube,” Meas. Sci. Technol. Vol. 17, 2006. pp. 2799-2805.
