Speaker: Ludovico Nista
Title: The role of generative adversarial networks in modeling turbulent combustion
Abstract: Turbulent combustion is inherently complex due to the wide range of interacting time and length scales, coupled with strong nonlinearities from combustion chemistry. This complexity makes traditional physics-based modeling approaches challenging, especially as emerging renewable fuels (e.g., hydrogen) introduce new combustion modes distinct from conventional fossil fuels. Machine learning offers a promising avenue to improve the fidelity and efficiency of large-eddy simulations for chemically reacting flows. This seminar will focus on the application of data-driven super-resolution techniques using generative adversarial networks (SR-GANs) to enhance turbulent combustion closure models. Key aspects covered will include the advantages of adversarial training, parallel implementation strategies for efficient large-eddy simulation closures, and investigations into the generalization ability of GAN-based models for premixed reacting flows.
Hybrid approaches combining SR-GANs with dynamic mixed models, as well as mixed combustion models for lean hydrogen flames, will also be discussed.
