Deep Learning Prediction and Experimental Validation of Nonlinear Acoustic-Driven Flame Dynamics Using a CNN-Transformer Hybrid Model

پذیرفته شده برای ارائه شفاهی ، صفحه 1-8 (8)
کد مقاله : 1118-ISAV2024 (R2)
نویسندگان
1دانشکده مهندسی هوافضا دانشگاه صنعتی شریف
2هیئت علمی دانشکده مهندسی هوافضا دانشگاه صنعتی شریف
چکیده
A hybrid deep learning model combining Convolutional Neural Networks (CNNs) with a Transformer Encoder was proposed to investigate the nonlinear dynamics of a laminar, partially premixed counterflow flame under acoustic excitation. Experimental data from a combustion instability laboratory were used to train the model. OH* chemiluminescence was employed to measure flame responses across frequencies from 20 to 350 Hz and pressure amplitudes up to the extinction threshold. The interactions between acoustic waves and flame dynamics were analyzed, revealing the influence of amplitude and frequency variations on heat release rates. Despite dataset limitations, the model accurately approximated the flame transfer function, replicated chemiluminescence signals, and predicted flame responses to diverse acoustic excitations. High-speed imaging and image processing techniques validated the repeatability of flame structures, confirming consistent characteristics across testing cycles. The findings highlighted the potential of the hybrid deep-learning approach for predicting flame dynamics in complex acoustic environments, offering insights for mitigating combustion instabilities in engineering applications.
کلیدواژه ها
 
Title
Deep Learning Prediction and Experimental Validation of Nonlinear Acoustic-Driven Flame Dynamics Using a CNN-Transformer Hybrid Model
Authors
Mohammad Ali Akhtardanesh, Ensieh Alipour, Mohammad Farshchi
Abstract
A hybrid deep learning model combining Convolutional Neural Networks (CNNs) with a Transformer Encoder was proposed to investigate the nonlinear dynamics of a laminar, partially premixed counterflow flame under acoustic excitation. Experimental data from a combustion instability laboratory were used to train the model. OH* chemiluminescence was employed to measure flame responses across frequencies from 20 to 350 Hz and pressure amplitudes up to the extinction threshold. The interactions between acoustic waves and flame dynamics were analyzed, revealing the influence of amplitude and frequency variations on heat release rates. Despite dataset limitations, the model accurately approximated the flame transfer function, replicated chemiluminescence signals, and predicted flame responses to diverse acoustic excitations. High-speed imaging and image processing techniques validated the repeatability of flame structures, confirming consistent characteristics across testing cycles. The findings highlighted the potential of the hybrid deep-learning approach for predicting flame dynamics in complex acoustic environments, offering insights for mitigating combustion instabilities in engineering applications.
Keywords
Acoustic wave, Deep learning, partially premixed flame, Convolutional Neural Network, encoder of transformer, flame nonlinear response, combustion instability