Non-dispersive Infrared SF₆ Gas Sensor Based On RBFNeural Network

SF6 gas is widely applied to high and medium voltage electrical equipment due to its excellent insulation and quenching performance, which promote the miniaturization of equipment and reduce the production cost. However, it may influence normal operation of electrical equipment and cause safety accident when leaked. Meanwhile, leaked SF6 gas will intensify the greenhouse effect and generate multiple toxic substances under the action of high temperature arch and water.Therefore,it's of vital importance to study how to realize real-time,accurate,rapid and convenient measurement of the concentration of SF6 gas.Non-dispersive infrared gas sensors can realize quantitative determination based on the characteristic absorption spectrum of molecules and the Beer-Lambert law and has numerous advantages like good selectivity,stable reliability,long service life, etc.In this thesis, electrically modulated infrared broadband optical source with wave band of 2-16?m and dual-channel pyroelectric detector with central wavelength of narrowband filtering of 3.95?m and 10.55?m are used to design a kind of new non-dispersive infrared SF6 gas sensor based on dual-wavelength single-chamber structure.This sensor features a simple and compact structure and can effectively overcome the error influence of problems like attenuation caused by non-gas absorption, fluctuation of infrared source, etc.To eliminate the nonlinear influence of the temperature change in the detection environment on the non-dispersive infrared gas sensor, a kind of RBF neural network based temperature compensation model was proposed in this thesis and its good nonlinear mapping and generalization ability was applied to compensate the measurement error brought about by the temperature change of the detection environment. The experiment results showed that within temperature of 10?35? and SF6 gas concentration of 0?2,000ppm in the detection environment, the maximum measurement error was reduced to 42ppm from 273ppm without temperature compensation, and the relative standard deviation was 1.56%. Compared with the hardware temperature control method, this method needs no addition of external equipment to maintain the dynamic balance of the gas chamber temperature, hence avoiding the increase of gas sensor volume and production cost; compared to the compensation method based on empirical formula, no sequential solution of fitting parameters and determination of temperature compensation algorithm by section is needed in this method, therefore, the calculation amount is reduced, the temperature compensation process is simplified and more outstanding compensation effect is realized.