RADNNET-MBL: A NEURAL NETWORK APPROACH FOR EVALUATION OF ABSORPTIVITY AND EMISSIVITY OF NON-GRAY COMBUSTION GAS MIXTURE BETWEEN FINITE AREAS AND VOLUMES

Published: September 20, 2019

Author(s)

Wai Cheong Tam, Walter W. Yuen

Abstract

RADNNET-MBL (RADiation-Neural-NETwork with Mean-Beam-Length) is developed to provide a computationally efficient and accurate method for the evaluation of radiative heat transfer between arbitrary rectangular surfaces in a three-dimensional enclosure with a non-gray combustion medium. Exact mean beam lengths between an infinitesimal area and a finite rectangular area are calculated for a wide range of geometric configurations and optical thicknesses. For a specific geometric configuration, the effect of optical thickness on mean beam lengths is examined. Based on numerical experiment, a constant averaged mean beam length is defined and shown to be effective in generating accurate prediction of the transmissivity over all optical thicknesses. Utilizing the averaged mean beam lengths together with RADNNET (a neural network-based correlation), exchange factors between two finite rectangular areas with an intervening non-gray combustion mixture can readily be obtained. A case study is presented. Results show that RADNNET-MBL provides promising accuracy (with absolute error less than 1%) with a significant reduction in computational effort.
Proceedings Title: The 2nd Pacific Rim Thermal Engineering Conference
Conference Dates: December 13-17, 2019
Conference Location: Maui, HI
Pub Type: Conferences

Download Paper

Keywords

Mean beam length, neural network, non-gray, three-dimensional, radiation heat transfer.
Created September 20, 2019, Updated September 23, 2019