Time window based features extraction from temperature modulated gas sensors for prediction of ammonia concentration

Electronic gas recognition systems, in literature commonly referred as electronic noses, enable the recognition of a type and a concentration of various volatile compounds. Typical electronic gas-analyzing device consists of four main elements, namely, gas delivery subsystem, an array of gas sensors, data acquisition and power supply circuits and data analysis software. The commercially available metal-oxide TGS sensors are widely used in such instruments. They are inexpensive and considered to be reliable. However, such sensors also have limitations. One of the most important problems of utilization of TGS sensors is the drift of their responses. It can lead to incorrect interpretation of the results of measurements. This can be a serious problem in the systems, which are designed to detect e.g. harmful gases. Drift can be caused by poisoning or aging of sensor as well as by the influence of temperature, humidity and gas flow rate. There are approaches to mitigate of this effect. One of them is based on the design of the proper data analysis procedures and algorithms. In this work the method of features extraction from continuous measurements of temperature modulated TGS sensor is presented. Namely, the voltage applied to the sensor heater is switched between two values, while the gas flow rate of the measurement cell is maintained constant. The presented method enables the extraction of the features at any time of measurements using time window with the fixed width. The calibration using LS-SVM regression is utilized for the purpose of prediction of ammonia concentration in humidified atmosphere. The validation measurements were conducted two weeks after the calibration procedure.