TiO2-based sensor arrays modeled with nonlinear regression analysis for simultaneously determining CO and O2 concentrations at high temperatures

Responses of TiO2-based sensor arrays were analyzed using kernel ridge regression modeling to determine the concentrations of CO and O-2 in gas mixtures at 873 K. Two variations of a two-sensor combination were studied. In each array, a La2O3-doped TiO2 sensor was used, whereas the second sensor in the array was a CuO-La2O3-doped TiO2 sensor, doped with different levels of copper. In sensor array I, 2 wt.% CuO was used, while 8 wt.% CuO was used in the second. Sensor array I was used to demonstrate the kernel ridge regression methodology. The concept of orthogonality of sensors was developed, which is a quantitative measure of how well the sensor array can discriminate between the two gases of interest. This model was then used to extract the concentrations of CO and O-2 in a gas mixture over ranges of 2-10% O-2 and 250-1000 ppm CO using the second sensor array. Prediction ability was found to be reasonable over certain concentration ranges and was determined by the orthogonality of the sensor responses. (C) 2002 Elsevier Science B.V. All rights reserved.