Study On Signal Calibration And Drift Compensation Of Array Of Sensors In Electronic Nose

Electronic nose(E-nose) is an application of the artificial olfaction analog technology. By imitating biological olfactory, it can percept, analyze and identify gases to be tested. First, it gets information through an array of sensors and then pattern recognition analysis is used to detect gas type and predict concentration. The E-nose has the characteristics of rapid detection and real-time online lossless or non-invasive monitoring. Taking the application of E-nose in air quality monitoring as the research background, this thesis focuses on the study of feature extraction, signal calibration of array of sensors, and the drift compensation of sensors.Firstly, the author aims at the research of feature extraction which is involved in the field of pattern recognition. According to the linearly inseparable characteristic of multiple contaminants’ patterns in artificial olfactory system, it is difficult to extract the nonlinear features with general feature extraction methods. So, this thesis focuses on a nonlinear feature extraction method: kernel entropy component analysis(KECA). In this method, the concept of information entropy is introduced to the feature extraction, which can more effectively extract the information feature of e-nose data. Support vector machine(SVM) as a very popular classifier is used for multi-class classification of E-nose data, and experimental results demonstrate that the KECA feature extraction is more effective in improving the recognition accuracy of E-nose.Aiming to the problem that the inherent variability in semiconductor gas sensors, a new calibration model of sensor signal differences based on data reconstruction is proposed. In this method, a prediction model is built for each sensor in master E-nose system, and the response characteristic of master E-nose sensors are kept in the model. When calibration begins, firstly, the master sensors’ responses are reconstructed according to the prediction model, then other E-nose systems are calibrated to the reconstructing master. Because of no master E-nose system, there is no effect on calibration when the master E-nose system behaves drift.Aiming at the problem that the drift appears in semiconductor gas sensors, a new on-line drift compensation method based on data standardization is proposed. Utilizing the drift information sensor baseline response contained, the method based on data standardization is accepted for drift compensation. During real-time running of E-nose, the baseline value is difficult to obtain. In order to realize online applications, the baseline value must be obtained firstly. In this thesis, through the establishment of a dynamic matrix, the minimum value of sensor in last month was preserved in the dynamic matrix as the baseline. In process of real-time working, the baseline value is got through the way of look-up table in dynamic matrix, then used to drift compensation.