Abstract
Hyperspectral imaging can obtain considerable flame information, which can improve the prediction accuracy of combustion characteristics. This paper studies the hyperspectral characteristics of methane flames and proposes several prediction models. The experimental results show that the radiation intensity and radiation types of free radicals are related to the equivalent ratio, and the radiation region of free radicals becomes larger with the increase of the Reynolds number. The polynomial regression prediction models include the linear model and quadratic model. It takes ${\rm{C}}_2^*/{{\rm{CH}}^*}$ as input parameters, and results can be available immediately. The three-dimensional convolutional neural network (3D-CNN) prediction model takes all spectral and spatial information in the flame hyperspectral image as input parameters. By improving the structural parameters of the convolution network, the final prediction errors of the equivalent ratio and Reynolds number are 2.84% and 3.11%, respectively. The method of combining the 3D-CNN model with hyperspectral imaging significantly improves the prediction accuracy, and it can be used to predict other combustion characteristics such as pollutant emissions and combustion efficiency.
© 2022 Optica Publishing Group
Full Article | PDF ArticleMore Like This
Sunghyun So, Jiyeon Park, Miyeon Yoo, Jungho Hwang, Daehae Kim, and Changyeop Lee
Opt. Express 30(18) 32031-32050 (2022)
Zichen He, Nathan Williamson, Cary D. Smith, Mark Gragston, and Zhili Zhang
Appl. Opt. 59(17) 5226-5233 (2020)
Yufeng Wang, Jiamin Du, Zhenyi Yan, Yuehui Song, and Dengxin Hua
Appl. Opt. 61(10) 2657-2666 (2022)