Determination of carbon monoxide by the metal oxide sensor

Abstract

To solve an important practical problem of determining the concentration of carbon monoxide in air, chemical sensors are usually used – semiconductor, electrochemical or thermocatalytic. Their main disadvantage is the low selectivity, which can lead to a false alarm when the analyzer enters the atmosphere of other reducing gases, for example, ethanol vapors, hydrogen, ammonia, etc. Increased selectivity was previously associated with the use of multisensory systems, the so-called “electronic noses”, but later the researchers abandoned this idea. They found out that the instability of such devices increases with the increase in the number of sensors in a geometric progression. Currently, some researchers recognize the need to create devices for selective gas analysis based on a limited number of sensors – one or two.

For a reliable selective gas analysis, we need to obtain as much information about the gas medium as possible, which can not be done under stationary experimental conditions. The transition to nonstationary regimes increases the amount of information about the analyte because it reveals its features related to the chemisorption kinetics on the surface of the sensor, the kinetics of the chemical interaction between reductant analytes and chemisorbed oxygen, and the kinetics of desorption of chemical interaction products. This information is implicit in the data on the kinetics of the nonstationary process. Qualitative and quantitative analysis of gases using data on the kinetics of a nonstationary process requires the chemometric processing of multidimensional data.

Geometrically, each sample with a volume of n resistance values can be represented as a point in n-dimensional space. In this case, the problem of identifying an analyte can be formulated as a cluster analysis problem, and the problem of determining the analyte concentration is a regression problem over a cloud of points belonging to the same cluster. However, the solution of these problems in the initial n-dimensional space is not very rational, since individual components of the vectors can be disorderly in information significance, and the range of their values can exceed three orders of magnitude. Statistical analysis of such data begins with the reduction of the dimension, for example, by the principal component method. To process the experimental data discussed in this paper, we used the package “chemometrics” developed by K. Varmuza and P. Filzmoser for a free system of statistical computations and graphics R.