Escolar Documentos
Profissional Documentos
Cultura Documentos
Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com Volume 1, Issue 4, November December 2012 ISSN 2278-6856
E-Nose for gas detection at vehicle exhaust Using supervised learning algorithm
Rupali V. Chothe1, Sunita P. Ugale2
1, 2
E & TC Dept, K. K. Wagh Institute of Engg. Education & Research, Nasik, Maharashtra, India.
Abstract: An electronic nose is an intelligent system used to monitor the gases. The system is designed to detect the pollution at vehicle exhaust. The system informs the user about the concentration of CO and HC. It also displays whether pollution is under control or not. Commercial gas sensors having low power consumption are used in the design. For data acquisition, a micro-controller is used. Data processing is done using supervised learning of Artificial Neural Network (ANN). The results of ANN training are given, which is obtained using MATLAB. The system is calibrated using the actual field readings of PUC machines available. Five Different ANN training methods are also compared based on errors. GUI developed displays concentrations of CO and HC, a conclusive message and bars indicating present gas level. Keywords: electronic nose, CO, HC, Data acquisition, supervised learning, Artificial Neural Network (ANN).
1. INTRODUCTION
In the 21st century, science and technology has converted our dreams to a real success. But, each coin has two sides. Our environment is very much affected by various contaminants which continuously enter the atmosphere through natural and man-made processes. So, the first and important step towards control of pollution is an inventory of pollutants. [1] Exhaust Emission analyzers are devices that measure the amount of air pollutant gases that a certain engine or industry emit. The vehicle generates two major air pollutants: hydrocarbons (HC) and carbon monoxide (CO). Carbon monoxide is an odorless poisonous gas. Hydrocarbon is a carcinogen, can cause lung cancer. [2]. The time-series emission load for Mumbai city as represented in Figure 1 indicates that the CO and HC emission levels are steadily increasing over the years. [3]
This work concerns the development of a system able to identify and quantify pollutant gases in vehicle exhaust. It is important to have a low cost solution with efficient detection of the target gases. The detection part of the system is based on an array of commercial gas sensors. The adsorption of a gas onto the surface of a metal oxide produces a large change in its electrical resistance.[4] This change is caused by a loss or a gain of surface electrons as a result of adsorbed oxygen reacting with the target gas. However, the use of these sensors is limited due to their poor selectivity to gas. To overcome this limitation, sensor array is coupled with Pattern recognition techniques. Collectively, these sensors respond with unique signatures (patterns) to different chemicals. This constitutes an "electronic nose" device. In the next section E-Nose components and structure is described. ANN and its training method are also explained. At the last, experimental results including comparison of various ANN training methods and GUI is given.
2. E-NOSE COMPONENTS
Electronic/artificial noses are being developed as systems for the automated detection and classification of gases. The two main components of an electronic nose are: 1) The sensing system and 2) The automated pattern recognition system. The sensing system is an array of several different sensing elements. During the training process, various chemicals with known mixtures are presented to the system. Artificial Neural Networks (ANNs) are used for pattern recognition. When an ANN is combined with a sensor array, the number of detectable odors is generally greater than the number of sensors. Also, less selective sensors which are generally less expensive can be used with this approach. Once the ANN is trained for odor recognition, operation consists of propagating the sensor data through the network.
Figure 2 Blocks of vehicle exhaust analyzer Figure 2 shows major blocks of vehicle exhaust analyzer system. The system consists of different modules like transducer (i.e. gas sensors), signal processing circuit, voltage processing module, LCD display unit, serial communications module and MCU. Tin oxide sensors with different sensitivity were employed for gas sensing. The sensors are conductance types in nature, and their resistances will decrease when the sensitive gas molecules exist. Some sensors like electrochemical CO sensors exhibit linearity against the particular range of concentrations. [5] The response of the sensors were sampled and analyzed to obtain target gas components. The gas sensor is composed by micro AL2O3 ceramic tube, Tin Dioxide (SnO2) sensitive layer, measuring electrode and heater. The enveloped sensor has 6 pins, 4 of them are used to fetch signals, and other 2 are used to provide heating current. Electric parameter measurement circuit of the sensor is shown in Figure 3. The sensor requires two voltage inputs: heater voltage VH and circuit voltage VC. The heater provides required work conditions to the sensitive components. VH is applied to the integrated heater which maintains the sensing element at a specific temperature. RL is connected in series with the sensor. VC is applied to achieve measurement of load voltage VRL across a load resistor RL.
Where R denotes the number of inputs vector pi, wi is the weight of the neuron, bi is the bias, and n is the sum of the weighted inputs with a bias.[*40] Then n is processed through a transfer function f, and the output f (n) of the neuron is described as
Figure 3 Basic measuring circuit of the gas sensor. Once the gas components vary the change in sensor resistance RS is converted into VRL and then taken to the MCU through the analog to digital converter (ADC) for subsequent analysis. The system utilizes Microcontroller PIC16F877A which integrates analog and digital peripherals for the data acquisition and control, e.g. ADC, LCD display control and serial port communication. The system consists of a LCD unit for displaying the components of automobile exhaust, and a Volume 1, Issue 4 November - December 2012
Figure 4 A 3-layer ANN For a particular training set, the parameters of the neurons are chosen to minimize the output error. Initially Page 146
Typically, in the hidden layers of Multilayer networks sigmoid transfer function is used, which is characterized by the fact that their slopes must approach zero if the input is large. This causes a problem when steepest descent is used to train a multilayer network with sigmoid transfer functions, because the gradient can have a very small magnitude. Thus, this causes small changes in the weights and biases, even when the weights and biases are far from their optimal values. The training parameters for the training function used are Maximum number of epochs to train, Performance goal, Maximum validation failures, Minimum performance gradient, initial step size and the maximum step size. Rprop algorithm is generally faster than the standard steepest descent algorithm. It requires only a modest increase in memory requirements. Instead of making more use of gradient information for better weight updates, Rprop only uses the sign of the gradient, because its size can be a poor and noisy estimator of required weight updates. Furthermore, Rprop assumes that different weights need different step sizes for updates, which vary throughout the learning process.
6. EXPERIMENTAL RESULTS
Page 147
Figure 7 Comparison of MSE of various methods Figure 7 is the graphical representation of comparison of MSE of various training methods. As shown, RP gives minimum errors. So, this is the best method that can be used for pattern recognition in this particular application.
Figure 6 Performance window Different training methods are compared based on errors. Table 2 shows error-comparison of various training methods. The performance is compared using MSE (Mean Squared Error), MAE (Mean absolute error) and SSE (Sum squared error) values. The methods are GDM (Gradient Descent with Momentum), OSS (One-step secant back propagation), RP (Resilient Back propagation), LM (Levenberg-Marquardt back propagation) and SCG (Scaled conjugate gradient back propagation).
Figure 8 GUI output displayed when both CO and HC are under control The GUI for the present E-nose is designed using MATLAB. The GUI has one press button showing Press Page 148
7. CONCLUSION
The E-Nose system as Automobile Exhaust Emission Analyzer is implemented using gas sensors, a Data acquisition circuit and LCD display. Pattern recognition is implemented using ANN. The system is calibrated using the actual field readings of PUC machines available. The reference data is gathered and ANN has been trained. For ANN training, five methods are compared, results of which are also included. GUI developed displays concentrations of CO and HC. The emission analyzer developed is also significantly cheaper compared to the machines available nowadays. In future, more experiments for other automobile exhaust gases such as oxygen, carbon dioxide and incomplete burning of hydrocarbon can be achieved.
AUTHOR
Prof. Rupali V. Chothe is currently working as an Assistant Professor in Electronics and Telecommunication Engineering department of K. K. Wagh Institute of engineering Education and research, Nashik, Maharashtra since last 5 years. She has completed Bachelor of Electronics and Communication engineering from the Atmiya Institute of Technology & Science, Saurashtra University. She is pursuing her M. E. (VLSI and Embedded) from K. K. Wagh Institute of engineering Education and research affiliated with Pune University. Prof. Sunita P. Ugale is currently working as an Associate Professor in Electronics and Telecommunication Engineering department of K. K. Wagh Institute of engineering Education and research, Nashik, Maharashtra since last 16 years. She pursued Bachelor of electronics engineering from the K. K. Wagh Institute of Engineering education and research, Pune University. She has completed her M. Tech in Electronics Design Technology from DOEACC, Aurangabad and presently doing her Ph. D. from S.V. National Institute of Technology, Surat, Gujarat, India. Her special fields of interest include Fiber Optics Communication, optical sensors, ANN and VLSI technology. She has published more than 20 papers in various National and International Journals in the same field. She has published a book Titled Fiber Optics Communication, systems and components by Wiley India(2012) and a Text Book of Electrical Circuits and Machines by Central Techno Publication, Nagpur(2006). She bagged Lady Engineer Award from Institution of Engineers (India) Nashik local center on engineers day 2008.She has worked as Board of Studies member of Electronics Engineering for Pune University.
REFERENCES
[1] Air Quality Trends and Action Plan for Control of Air Pollution from Seventeen Cities, Central Pollution Control Board Ministry Of Environment & Forests, NATIONAL AMBIENT AIR QUALITY MONITORING, Series: NAAQMS/29/2006-07, September 2006, www.cpcb.nic.in [2] Fulton Baylon, John Michael G. Dy, Roveric D. Quidilla, Ralph Andrew, D.L. San Pascual, Implementation of a Portable Automobile Exhaust Emission Analyzer, DLSU Engineering e-Journal, Vol. 1, No. 1, March 2007, pp.13-27. [3] State of Environment Report: Maharashtra (FINAL DRAFT) Sponsored by Maharashtra Pollution Control Board Ministry of Environment and Forests, Govt. of India. Prepared by Indira Gandhi Institute of Development Research, Mumbai. [4] O.T. Arulogun , O.A. Fakolujo , A. Olatunbosun M.A. Waheed , E. O. Omidiora , P. O. Ogunbona, Characterization of Gasoline Engine Exhaust Fumes Using Electronic Nose Based Condition Monitoring, Global Journal of researches in engineering: AEROSPACE Engineering, Online ISSN: 2249-4596, Volume 11, Issue 5, Version 1.0, August 2011.
Page 149