A machine learning based full-spectrum correlated k-distribution model for nonhomogeneous gas-soot mixtures. (July 2021)

• Record Type: Journal Article
• Authors: Zhou, Ya; Wang, Chaojun; Ren, Tao; Zhao, Changying
• Is Part Of: Journal of quantitative spectroscopy & radiative transfer. Volume 268(2021)
• Publication Date: 2021-07
• Subjects: Radiative heat transfer -- Machine learning -- Neural network -- Full-spectrum correlated-k-distribution -- Soot; Spectrum analysis -- Periodicals; Radiation -- Periodicals; Analyse spectrale -- Périodiques; Rayonnement -- Périodiques; Radiation; Spectrum analysis; Periodicals; 543.0858
• DOI: 10.1016/j.jqsrt.2021.107628
• Languages: English
• ISSNs: 0022-4073

Abstract

Highlights: A new machine learning based FSCK model for gas-soot mixture was developed. The machine learning based-FSCK model is now in stand-alone Fortran modules. The model is only about 47.1 MB but is as accurate and efficient as the original FSCK look-up table. Abstract: The machine learning based full-spectrum correlated k -distribution (FSCK) model previously developed by Zhou et al. (2020), provides a compact prediction model with good efficiency and accuracy for the evaluation of radiative spectral calculations in nonhomogeneous gaseous media. However, the model is only for gas mixtures of CO2, CO, H2 O, and cannot be used for gas-soot mixtures that are found in most combustion scenarios. In addition, the previous machine learning based-FSCK model was trained in Python and requires several Python packages to run the model for radiative heat transfer calculations, which limits its portability to other computational fluid dynamics (CFD) platforms. In this paper, a new machine learning based-FSCK model is developed for the radiatively participating species CO2, H2 O, CO, and soot, by accurately fitting the gas-soot FSCK look-up table. Major improvements include: (1) The Bayesian optimization method is used to select optimal hyperparameters of the machine learning model; (2) The developed machine learning based-FSCK model now includes gas-soot mixtures, but with a smaller model size of 47.1 MB, which is significantly smaller than the original gas-soot FSCK look-up table of Wang et al. (2018), with a size of 12.45 GB; (3) The machine learning based-FSCK model is now in stand-alone Fortran modules, which make it easier to be implemented into other CFD platforms. The new model has been successfully implemented in OpenFOAM and tested with two scaled flames. The machine learning based gas-soot FSCK model is publicly available on GitHub at https://github.com/ZY-LHY/Machine_learning_based_FSCK_model_soot .