A Probabilistic Neural Network Measurement Method for Multispectral High Dynamic Temperature

Abstract

Among various temperature measurement technologies, Multispectral thermometry is a better method to measure temperature under complex conditions. To solve the problems in the multi-spectral temperature measurement technology, which included the detector’s non-linear sensitivity characteristics, consistency calibration, data processing, etc., this paper, based on the previous research and development of the multi-spectral dynamic temperature measurement system, realized data acquisition of the dynamic temperature in 20us sample rates within 1550~2000 Celsius, and then the dynamic measurement data is processed by the probabilistic neural network (PNN). Finally, temperature measurement system realized the accuracy of the measurement error less than 1.1%. The results show that the PNN neural network can fit the temperature distribution curve better and faster, and can be better applied to the miniaturization of the temperature measurement system.