Research On Key Issues Of Metal Oxide Semiconductor Gas Sensor In Gas Detection

Combustible gas detection is of great importance in preventing gas leakage of coal mine, monitoring safety condition of petrochemical industry, guarantying personnel safety in airship, submarine and other closed environments. Since metal oxide semiconductor (MOS) gas sensor has the advantages of simple structure, low cost, quick response, longevity of service, and high sensitivity to combustible gases and volatile organic gases, it has been most produced and widely used in the world. However, it is a bit difficult to obtain stable analysis result in the application since its sensitivity to environmental humidity, poor selectivity, drift and high heating power, which directly lead to the negative influence to the precision of gas detection. This paper focuses on several key issues related to combustible gas detection and hazard localization of MOS, and makes an in-depth study. Major tasks of this paper are as follows:To solve the nonlinear response of MOS gas sensor and the cross-sensitivity to the non-target gases, this paper studies the support vector machines (SVM) based sensor selectivity improvement method. The sensor array comprises four MOS gas sensors. The gas category is identified by multi-classifier SVM (MC-SVM) and gas concentration is measured via least square support vector regression (LS-SVR). Given the root mean square error of concentration measurement as the criterion of generalization performance and using the k-fold cross-validation result of training sample as the objective function, the paper proposes a niche particle swarm optimization (NPSO) based parameter optimization algorithm which can find the global optimal parameters of the built LS-SVR model. Compared with other array signal processing and pattern recognition methods such as artificial neural networks (ANNs), this method improves the accuracy of gas recognition and the precision of concentration measurement, and it is especially suited for gas detection within small samples.To restrain the influence of sensor output drift, this paper proposes blind source separation (BSS) based mixed gases recognition and sensor drift inhibition method. The blind separation model of gas sensor array steady-state response is built and the blind identification condition of gas mixture analysis is demonstrated. Given the unknown gas concentration as the source signals and the sensor array response as the mixed signals, this paper devises the time sequences for both gas concentration and sensor steady-state response. The negative entropy based fast fixed point independent component analysis (FastICA) algorithm is used to process the sensor array steady-state response. Experimental results demonstrate that this ICA based method can not only recognize the mixed gases but also remove the non-linear drift of the gas sensor array. It shows important significance for the development of BSS theory and its application in sensor information processing.To reduce the heating power of gas sensor, this paper researches the single sensor dynamic response feature extraction technology under temperature modulation. The gas mixture classification and concentration measurement can be achieved in combination with MC-SVM and LS-SVR respectively. Compared with the sensor array based approach,this method that uses only one sensor to achieve gas identification and concentration measurement greatly reduce the sensor's heating power. Also, this paper uses distance based class divisibility criterion as feature evaluation criteria to solve the difficulty of feature parameter selection of dynamic response. To restrain the noisy influence on the sensor dynamic response in the actual working condition, this paper proposes the wavelet singular entropy (WSE) based dynamic response feature extraction method. Experimental results show that, compared with virtual array (VA), fast Fourier transform (FFT) and discrete wavelet transform (DWT), the WSE based feature extraction method has a higher generalization accuracy when a strong noise exists.Finally, this paper designs and implements the hardware platform of the mixed gas detection test sytem using DSP. Sensor calibration of this system is accomplished on this platform, and selectivity improvement, heating power reduction as well as signle sensor dymanic detection methods are verified on DSP. The system successfully implements the binary gas mixture analysis for CH4 and H2 online. Also, the paper evaluates the validity and the real-time ability of various alforithms (e.g. SVM multi-calssifier, LS-SVR and dymanic response feature extraction, etc) on DSP when sensor works in the constant temperature mode and in the temperature modulation mode, and verifies the whole function of the test system. This technology lays the foundation in the fields of combustible and explosive material detection as well as poisonous and harmful gas component analysis instruments.