Research On Key Technologies Of Gas Identification Based On Sensor Array Transient Response

Electronic nose(EN) is an instrument for gas detection which mimics the animal olfactory mechanism. It has been successfully applied to various fields,such as public safety testing, food quality assessment and control, environmental monitoring, sanitation, aerospace and military fields, and so on.The sensitive mechanism of gas sensor is complex. So it is difficult to achieve the ideal effect in practical application applying the methods of gas identification based on gas sensor mechanism. In a sensor array, many sensors react with many types of gas and the response signals are different from each other, which is the key to identify gas using electronic nose. At present, the methods of gas identification based on gas sensor array transient response have achieved some effect, but there are still some problems in the effective use of information to realize gas qualitative/quantitative analysis and mixed gas recognition. The methods of sensor array optimization, gas qualitative identification, gas quantitative analysis and mixed gas recognition, which utilize the transient response signals of the gas sensor array, are researched in detail in this paper. The focuses include acquisition of the gas sensor array response signal, signal preprocessing, feature extraction, and pattern recognition. The main contents and innovative work can be summarized as follows.1. In order to reveal the reaction mechanism of the metal oxide semiconductor(MOS) gas sensor and reducing gas, the kinetic model is established based on the oxygen ionization model.Researches show that the reaction mechanism of the MOS gas sensor and reducing gas relates to properties of semiconductor, properties of gas, size and shape of semiconductor crystals. The rate constant which reflects the sensor response speed and the relative conductivity of steady state values can both be expressed as the power functions of gas concentration.2. The transient response signals of the gas sensor array to three types of oil are obtained by making use of the EN. And the signals are analyzed and preprocessed.The transient response process can be divided into the following four stages: the steady state, the adsorption process, the maximum response and the declining process. It is necessary to process the signals based on the whole transient response. After signals preprocessing, the environmental noise is reduced and so is the difference between response value of same type of oil samples with different concentrations.3. The methods of gas qualitative identification based on parallel factor(PARAFAC) analysis are deeply studied.The application of PARAFAC as a feature extraction technique to apply data compression is presented. The three-way data analysis allows visualization of the hidden information such as the sample information, the response process information and the sensor information. Since the method takes advantages of all of the information in the time dimension, it is still able to classify gas samples correctly when the response signals of the sensor array fail to reach the steady state. It is possible to short the measurement time and improve the measurement efficiency. At the same time we can utilize the sensor information to optimize the sensor array and reduce the data redundancy. The application of a five-layer of mice olfactory neural network model for gas qualitative identification is proposed. Using the feature vector got with PARAFAC model as an input, the mice olfactory neural network model achieves a correct classification of gas samples. This network model can be trained to reach the expected error range in a short time. The result shows that the mice olfactory neural network is effective for gas identification, and at the same time it verifies the effectiveness of PARAFAC.4. The methods of the sensor array optimization based on fuzzy clustering analysis are deeply studied.Fuzzy clustering method realizes sensor array optimization by directly making use of the sensor array transient response signals. The optimization results are consistent with those got with PARAFAC. Fuzzy equivalence matrix using correlation coefficient method is constructed. When the threshold is suitably large, the 10 sensors are clustered into 6 groups. When the threshold is suitably small, 10 sensors can be clustered into 3 groups, which is consistent with the types of PEN3 gas sensors.5. The methods of gas quantitative analysis based on complex frequency-domain analysis method are deeply studied.The exponential model of concentration changing with time is established. The transfer function models of gas sensors in different concentrations are built based on complex frequency-domain analysis method. The response signals of the sensors to gas with stable concentration are reconstructed, and a non-linear correlation model between the steady state amplitude and gas concentration is established, also the model between the dominant time constant and odor concentration is established. The results are consistent with the kinetic model of gas sensor. Based on the features selected from the parameters of the transfer function model, the different concentrations of the gases are correctly predicted by the partial least squares(PLS) calibration models.6. The methods of gas quantitative analysis based on an extended orthogonal quasi-Legendre basis are studied and proposed.Based on an extended orthogonal quasi-Legendre basis, each transient response signal is decomposed into five orthogonal quasi-Legendre functions so that the response signals of gas sensor array in the time domain can be translated into an alternative space spanned by a set of orthogonal functions. The Decomposition coefficient coefficients are used as fingerprints of the gas being tested. They are applied to build the PLS discrimination models and PLS quantification models. Two types of gases are clearly discriminated by the classifier, and the different concentrations of the gases are correctly predicted by the PLS calibration models. The orthogonal decomposition method based on extended quasi-Legendre basis is applied to mixed gas recognition. A nonlinear mixed response model is built firstly. The transient response signals of sensor array to gear volatilization gas, diesel volatilization gas, and gear/diesel mixed gas are orthogonal decomposed respectively. Then the decomposition coefficients are fitted with power function relationship. Taking diesel gas as a reference, a set of nonlinear equations is obtained. By solving the equations, the gas mixture composition and concentration are obtained simultaneously.These researches provide important theoretical basis and technical support for EN ‘s application to gas detection.