Generalized inverse matrix-recurrent neural network data processing algorithm for multiwavelength pyrometer

Abstract

A generalized inverse matrix-recurrent neural network (GIM-RNN) data processing algorithm for unknown emissivity was proposed to measure high-temperature by multiwavelength pyrometer (MWP). First, emissivity classification was realized quickly according to a solution from generalized inverse algorithm to underdetermined multiwavelength equation group. According to the relationship between the emissivity of 2 adjacent channels of the 6 channel thermometer used in this paper, 243 emissivity models (1 Ã 3 Ã 3 Ã 3 Ã 3 Ã 3) were designed for classification. Twelve of them were shown on the schematic diagram in the text. To make the figure clear and easy to observe, the prediction results of six common emissivity models were listed as the experimental results. Then, it was input into the corresponding RNN subnetwork to inverse temperature precisely. Simulation results showed that in the range of 1500 to 3000 K temperature, the relative error of the test set was within 1.0% for the network trained after classification by GIM, whereas the relative error of the test set was within 1.2% for the network trained without GIM classification. After 5.0% random noise was added to the inputting data, the relative error still was controlled within 1.5%, which reflected the good antinoise performance of the algorithm. Multispectral measurement data of rocket engine plumes is processed by the proposed algorithm in this manuscript. The inversion results are consistent with the theoretical results. It is indicated that the proposed algorithm has good adaptability to different materials. It is expected to become a general data processing algorithm for MWP.