Session: 04-13 Combustion Modeling I

Paper Number: 127369

127369 - A Deep Learning Based Model for Identifying Recirculation Zones From Experimental Images of Trapped Vortex Combustors 

Trapped vortex combustors (TVCs) feature large vortices locked inside a cavity, which form the primary recirculation zone (PRZ) that provides the radicals and enthalpy necessary for flame stabilization. A notable advantage of TVCs is their ability to operate in an RQL configuration, where a rich mixture is injected into the low aspect ratio cavity, and the lean mixture is introduced into the main channel. As the PRZ is a critical element for sustained and efficient combustion, particle image velocimetry (PIV) is used as the standard experimental procedure for studying its dynamics. However, for reacting flows, the alumina tracer particles used in PIV can alter the flow and clog injection ports in a compact combustor. This adversely affects the flame and may result in inaccurate measurements. As an alternative, we use deep learning models based on generative adversarial networks (a widely used approach) and vision transformers (a recently devised promising architecture) to estimate the position and overall structure of large-scale vortices from a non-invasively measured quantity, such as the planar laser-induced fluorescence (PLIF) of a species. These models are trained using datasets from large-eddy simulations of TVCs with information regarding all scalars constituting the state variable. The trained model is then used to infer velocity vectors from noisy OH-PLIF data. Using qualitative observations and quantitative metrics (such as relative error and PDFs), we establish the superiority of the performance of the vision transformer. Such models will facilitate intelligent data fusion and the development of digital twins of combustors.

Presenting Author: Priyabrat Dash Indian Institute of Science, Bangalore

Presenting Author Biography: Priyabrat is a Ph.D. candidate pursuing computational and data sciences at the Indian Institute of Science Bangalore. His research interests lie in the confluence of fluid mechanics and machine learning. He has completed his undergraduate studies in mechanical engineering at Indian Institute of Technology, Dhanbad.

Authors:

Priyabrat Dash Indian Institute of Science, Bangalore
Tanaya Mallik Indian Institute of Science, Bangalore
Nikhil Verma Indian Institute of Science, Bangalore
Aritra Roy Choudhury Indian Institute of Science, Bangalore
R. v. Ravikrishna Indian Institute of Science, Bangalore
Konduri Aditya Indian Institute of Science, Bangalore