Electricity and power generation is an essential part of our life. However, powergeneration activities also create by-products (such as sulphur oxides, nitrogen oxides,carbon monoxide, etc), which can be dangerous when released to the atmosphere.Sensors, as part of the control system, play very vital role for the fluegas cleaning processes in power plants. This thesis concerns the development ofSilicon Carbide Field Effect Transistor (SiC-FET) gas sensors as sensors for sulfurcontaining gases (SO2 and H2S) used as part of the environmental control systemin power plants. The works includes sensor deposition and assembly, sensinglayer characterization, operation mode development, performance testing of thesensors in a gas mixing rig in the laboratory and field test in a desulfurization pilotunit, and both experimental and theoretical studies on the detection mechanismof the sensors. The sensor response to SO2 was very small and saturated quickly. SO2 is a verystable gas and therefore reaction with other species requires a large energy input.SO2 mostly reacts with the catalyst through physisorption, which results in lowresponse level. Another problem was that once it finally reacted with oxygen andadsorbed on the surface of the catalyst in form of a sulfate compound, it is desorbedwith difficulty. Therefore, the sensor signal saturated after a certain timeof exposure to SO2. Different gate materials were tested in static operation (Pt,Ir, Au), but the saturation phenomena occurred in all three cases. Dynamic sensoroperation using temperature cycling and multivariate data analysis could mitigatethis problem. Pt-gate sensors were operated at several different temperatures in acyclic fashion. One of the applied temperatures was chosen to be very high for ashort time to serve as cleaning step. This method was also termed the virtual multisensor method because the data generated could represent the data from multiplesensors in static operation at different temperatures. Then, several features of thesignal, such as mean value and slope, were extracted and processed with multivariatedata analysis. Linear Discrimination Analysis (LDA) was chosen since itiiiallows controlled data analysis. It was shown that it was possible to quantify SO2with a 2-step LDA. The background was identified in the first step and SO2 wasquantified in the second step. Pt sensors in dynamic operation and 2-step LDAevaluation has also demonstrated promising results for SO2 measurement in thelaboratory as well as in a desulfurization pilot unit. For a commercial sensor, algorithmhave to be developed to enable on-line measurement in real time. It was observed that Ir-gate sensors at 350oC were very sensitive to H2S. The responseobtained by Ir sensors to H2S was almost five times larger than that of Ptsensors, which might be due to the higher oxygen coverage of Ir. Moreover, Irsensors were also more stable with less drift during the operation as a result ofhigher thermal stability. However, the recovery time for Ir sensors was very long,due to the high desorption energy. Overall, the Ir sensors performed well whentested for a leak detection application (presence of oxygen and dry environment).The geothermal application, where heat is extracted from the earth, requires thesensor to be operated in humid condition in the absence (or very low concentration)of oxygen, and this poses a problem. Temperature cycle operation and smartdata evaluation might also be an option for future development. Along with the sensor performance testing, a study on the detection mechanismwas also performed for SO2 sensor, both experimentally and theoretically. The experimentincluded the study of the species formed on the surface of the catalystwith DRIFT (diffuse reflectance infrared frourier transform) spectroscopy and theanalysis of the residual gas with mass spectroscopy. Explanatory investigation ofthe surface reactions was performed using quantum-chemical calculations. Theoreticalcalculations of the infrared (IR) vibration spectra was employed to supportthe identification of peaks in the DRIFT measurement. Based on the study on theresidual gas analysis and quantum-chemical calculations, a reaction mechanismfor the SO2 molecule adsorption on the sensor surface was suggested.

The use of Silicon Carbide Field Effect Transistor (SiC-FET) sensors in cyclic operation is a proven way to quantify different gases. The standard workflow involves extracting shape-defining features such as averages and slopes of the sensor signal. This work’s main goal is to verify if frequency modulation can be used to simultaneously quantify Nitric Oxide (NO), Nitrogen Dioxide (NO2) and Ammonia (NH3). Linear models were chosen, namely: Ordinary Least Squares (OLS), Principal Components Regression (PCR), Partial Least Squares Regression (PLSR) and Ridge regression. Results indicate that these models fail to predict concentrations completely for every gas. Analysis indicates that the features are not linear in terms of concentrations. This work is concluded by recommending a few other alternatives before discarding frequency cycling completely: non-parametric models of regression and different frequency regime, namely the use of triangular waves in future experiments.

Chemical sensor based systems that are used for detection, identification, or quantification of various gases are very complex in nature. Sensor response data collected as a multivariate time series signals encounters gradual change of the sensor characteristics(known as sensor drift) due to several reasons. In this thesis, drift component of a silicon carbide Field-Effect Transistor (SiC-FET) sensor data was analyzed using time series. The data was collected from an experiment measuring output response of the sensor with respect to gases emitted by certain experimental object at different temperatures. Augmented Dickey Fuller Test (ADF) was carried out to analyze the sensor drift which revealed that stochastic trend along with deterministic trend characterized the drift components of the sensor. The drift started to rise in daily measurements which contributed to the total drift. / Traditional Autoregressive Integrated Moving Average (ARIMA) and deep learning based Long Short-Term Memory (LSTM) algorithm were carried out to forecast the sensor drift in reduced set of data. However, reduction of the data size degraded the forecasting accuracy and imposed loss of information. Therefore, careful selection of data using only one temperature from the temperature cycle was chosen instead of all time points. This chosen data from sensor array outperformed forecasting of sensor drift than reduced dataset using both traditional and deep learning methods.