MODELING THE PROPERTIES OF GAS SENSOR MATERIALS BASED ON COBALT-CONTAINING POLYACRYLONITRILE USING REGRESSION ANALYSIS AND NEURAL NETWORKS
Abstract
A modeling approach has been developed for materials based on organic semiconductors and their physicochemical and gas-sensitive properties. For modeling, such methods as multiple linear and non-linear regression, neural networks were used. As an input vector for modeling the properties of metal-containing polyacrylonitrile are the parameters of the technological process of forming materials: the mass fraction of the alloying component (cobalt) in the film-forming solution, technological modes of IR annealing: temperature, time of the first and second stages. Output vector - functional characteristics and physical and chemical properties of materials (resistivity, gas sensitivity coefficient, stability and selectivity). Abstract—Metal–carbon systems with Co metal particles based on polyacrylonitrile have been synthesized by IR pyrolysis. The resistance values were measured in the medium of the detected gas (chlorine). Modeling of the functional characteristics and physicochemical properties of materials was carried out on the basis of data obtained from the study of 200 samples of cobalt/polyacrylonitrile films. Multiple linear regression proved to be effective for predicting resistivity values. Neural networks are used to predict the gas sensitivity coefficient, selectivity, and stability of cobalt-containing polyacrylonitrile films. An artificial neural network in the form of a multilayer perceptron was built to predict the gas sensitivity coefficient of gas sensor elements based on the data of technological processes for obtaining material (mass fraction of the alloying component (cobalt) in the film-forming solution, technological modes of IR annealing: temperature, time of the first and second stages). Compliance of the synthesized model was checked: with experimental data: correlation coefficient R=0.82, root-mean-square error st=0.017. The synthesized models satisfactorily describe the collected data within the experimental error, which makes it possible to optimize the chemical composition and heat treatment conditions.
