Flamelet Generated Manifolds Combined with Neural Networks

Please enable javascript in your browser.

Kashani, Yeganeh | 2024

0 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 57729 (45)
University: Sharif University of Technology
Department: Aerospace Engineering
Advisor(s): Salehi, Mohammad Mahdi
Abstract:
Reduced mechanisms are widely used in the numerical simulation of turbulent combustion, motivated by the need to reduce computational costs. Tabulation is widely used in turbulent and reacting flow simulations. Still, the need for a huge memory, the high cost of searching and interpolating among the data, and the development of artificial neural networks have created a situation for combining these two sciences. In this research, to model the chemistry of a methane premixed flame, we produced several training databases using the flamelet generated manifold, the best of which is the database with five degrees of freedom. We used these data to train individual neural networks for the concentration of each species and their production or consumption rates. In the end, we evaluated the performance of neural networks with a direct numerical simulation database that was solved for the methane flame with the same conditions. In addition to the significant reduction in memory usage and computational costs, the results demonstrate an acceptable level of accuracy
Keywords:
Neural Networks ; Premixed Flame ; Machine Learning ; Chemical Kinetic ; Flamelet Generated Manifold Method

No TOC