Wireless electronic nose system for real-time quantitative analysis of gas mixtures using micro-gas sensor array and neuro-fuzzy network
Introduction
The design of a wireless electronic nose comprised of gas sensor arrays and a pattern recognition tool is an exciting topic. Since a semiconductor-type electronic nose can be designed using low-cost, small-sized, and easy-handling devices with a fast response time compared to traditional gas chromatography methods, this opens a wide variety of applications in such fields as medicine, food, and environmental monitoring. Depending on the field of application, various design techniques for an electronic nose can be considered. When combined with embedded technology, the recent appearance of handheld instruments has made on-site chemical compound analysis a reality [1], [2]. However, such handheld instruments are not suitable when a real-time analysis of chemical compounds is continuously required or the location is inaccessible. Thus, to cope with these requirements, an electronic nose should also include wireless hardware platforms and optimized functioning to maximize its qualitative and quantitative recognition capability.
Nonetheless, despite the potential benefits of being wireless, there are few reported examples of the wireless electronic nose comprised of gas sensor arrays and a pattern recognition tool. A small number of studies have attempted to combine chemical sensors with wireless sensor networks, for example, a remote system for monitoring indoor air quality using a CO2 gas sensor [3] and a wireless chemical sensor network for the detection and tracking of an acetic acid plume using LED chemical sensors [4]. However, these reports have only dealt with the detection of a single gas and not presented a real-time qualitative and quantitative analysis of the target gases.
To classify a single gas or gas mixtures, multilayer perceptron (MLP) neural networks have been widely used in electronic noses. Yet, according to the results of recent research [5], a fuzzy ARTMAP can outperform a multilayer perceptron on the classification accuracy and training time. Meanwhile, for quantitative gas analyses, MLP neural networks and neuro-fuzzy networks (NFNs) have both been developed [6], [7], [8], and NFNs shown to be more powerful than MLP neural networks as regards the estimation results. However, NFNs involve complex learning methods when training the membership function parameters. Thus, due to the intricacy of their training algorithms and difficult mathematical operations, MLP networks and NFNs are both limited as regards real-time quantitative gas mixture analyses and the implementation of an electronic nose into wireless hardware platforms.
In previous work by the present authors [9], a fuzzy ART-based network was proposed to estimate the concentrations of gas mixtures and demonstrated a good performance as regards the concentration estimation with less training time than other estimators using NFNs or MLP networks. The fuzzy ART-based concentration estimator employs the fuzzy ART proposed by Carpenter et al. [10] to perform fuzzy clustering in the input spaces and find proper fuzzy logic rules dynamically by associating the input clusters with the output clusters. The fuzzy ART learning creates clusters of concentrations of a single gas or gas mixtures. Using the regions formed for each concentration, corresponding membership functions are then designed and fuzzy rules extracted. Finally, fuzzy inference and defuzzification are used to estimate the concentrations for a given input.
Accordingly, this paper presents a wireless electronic nose system (WENS) that can classify ammonia (NH3), hydrogen sulfide (H2S), and their mixtures, plus estimate the concentrations of these gases, the main malodors in several environments. The proposed WENS consists of wireless sensor nodes and pattern recognition software. Each wireless sensor node is implemented using an ultra-low power microcontroller, MSP430F1611, a micro-gas sensor array with SnO2–CuO and SnO2–Pt sensing films for detecting the gases, and an RF transceiver for wireless communication between the sensor nodes. The software for analyzing gas mixtures, a fuzzy ARTMAP classifier and a fuzzy ART-based concentration estimator, is implemented on a desktop PC using MATLAB language. The performance of the fuzzy ART-based concentration estimator is also improved using the weighted inference method, which is inspired by the weighted fuzzy min–max neural network [11]. Plus, a virtual instrument is developed in the LabVIEW environment for monitoring the results in real-time and setting pre-determined parameters, such as the response to reference air and drift correction factors. Finally, for reproducibility and reliability, the multiplicative drift correction method [12], which is currently used in commercial electronic noses, is incorporated in the WENS design. The main contribution of this paper is that the weighted inference method is devised for the proposed WENS and it has improved the performance of the concentration estimator in the sense of the root mean square error. The unique feature of the weighted inference method is the weighting vector containing the variances of the input variables to the neuro-fuzzy network, which helps produce the best concentration estimation.
Section snippets
Silicon bridge-type micro-gas sensor array
A silicon bridge-type micro-gas sensor array is employed to allow the fabricated WENS to detect hydrogen sulfide and ammonia gases simultaneously. Thus, anisotropic wet-etching of silicon in a 25 wt% KOH solution at 80 °C and silicon reactive-ion etching are applied to achieve a bridge-type sensor for low power consumption. The micro-gas sensor array is consisting of two sensing films; one is Pt-added tin oxide sensing film, SnO2–Pt and the other is CuO-added one, SnO2–CuO. Fig. 1 shows the
Preprocessing and feature extraction
The tasks of the software developed in this work are two-fold—to classify NH3, H2S, and mixtures of the two gases, and to estimate their concentrations. Thus, the pattern recognition system was designed in two steps. First, a pattern classifier discriminating NH3, H2S, and their mixtures was constructed, and then a concentration estimator was designed for each gas. Details on these steps are shown in Fig. 5.
Since the 16-dimensional data sets were extracted from the sensor array using thermal
Results and discussion
The developed software was programmed using MATLAB programming languages and simulated in the configuration of a desktop PC. Fig. 7 shows the selection criterion, the Cr values of the gases responding to the SnO2–CuO sensing film and SnO2–Pt sensing film, respectively. The responses to the two films were separated by dimensions. Generally speaking, the H2S responses were dominant for the SnO2–CuO sensing film, while the NH3 and gas mixture responses were dominant for the SnO2–Pt sensing film.
Conclusions
A wireless electronic nose system was designed that can both identify and quantify concentrations of NH3, H2S, and their mixtures. The hardware and software parts of the proposed system were designed to obtain data sets and analyze the measured data sets. Since the devices used in the hardware parts incorporate low power consumption and miniaturization, the designed WENS could be a good candidate node for wireless sensor networks. The WENS software also demonstrated satisfactory results, with a
Acknowledgements
This work was supported by the Korea Science and Engineering Foundation (KOSEF) grant funded by the Korean Government (MOST) (No. R01-2005-000-11047-0).
Jung Hwan Cho received the BS degree in instrumentation and control engineering from Gyeongsang National University, Jinju, Korea in 2001, then received the MS degree in electronic engineering from Kyungpook National University, Daegu, Korea in 2003. He is currently pursuing the PhD degree at Kyungpook National University and his research interests include pattern recognition techniques, fuzzy systems, and artificial neural networks applied to electronic noses and gas detection devices.
References (14)
- et al.
Remote monitoring system with wireless sensors module for room environment
Sens. Actuators B
(2006) - et al.
Monitoring chemical plumes in an environmental sensing chamber with a wireless chemical sensor network
Sens. Actuators B
(2007) - et al.
Fuzzy ARTMAP based electronic nose data analysis
Sens. Actuators B
(1999) - et al.
A study on quantitative classification of binary gas mixture using neural networks and adaptive neuro-fuzzy inference systems
Sens. Actuators B
(2006) - et al.
Fuzzy ART: Fast stable learning and categorization of analog patterns by an adaptive resonance system
Neural Netw.
(1991) - et al.
A calibration method for handling the temporal drift of solid state gas-sensors
Anal. Chim. Acta
(2000) - et al.
Silicon bridge type micro-gas sensor array
Sens. Actuators B
(2005)
Cited by (0)
Jung Hwan Cho received the BS degree in instrumentation and control engineering from Gyeongsang National University, Jinju, Korea in 2001, then received the MS degree in electronic engineering from Kyungpook National University, Daegu, Korea in 2003. He is currently pursuing the PhD degree at Kyungpook National University and his research interests include pattern recognition techniques, fuzzy systems, and artificial neural networks applied to electronic noses and gas detection devices.
Young Wung Kim received the BS degree in instrumentation and control engineering from Kyungil University, Kyungsan, Korea in 2006, then received the MS degree in electronic engineering from Kyungpook National University, Daegu, Korea in 2008. He is currently pursuing the PhD degree at Kyungpook National University. His research interests include pattern recognition techniques, fuzzy systems, embedded systems, and wireless electronic nose systems.
Kyung Jin Na received the BS degree in computer control engineering from Kyungil University, Kyungsan, Korea in 2007, and is currently pursuing the MS degree at Kyungpook National University. His research interests include pattern recognition techniques, fuzzy systems, embedded systems, and wireless electronic nose systems.
Gi Joon Jeon received the BS degree in metallurgical engineering from Seoul National University, Seoul, Korea in 1969, then received the MS and PhD degrees in systems science and engineering from University of Houston, Houston, Texas, USA in 1978 and 1983, respectively. Since 1983, he has been with the School of Electrical Engineering and Computer Science, Kyungpook National University, Daegu, Korea. His main research interests include intelligent control, sensor signal processing, and resource control in OFDM systems.