Turbulent flame image classification using Convolutional Neural Networks
Pockets of unburned material in turbulent premixed flames burning CH4, air, and CO2 were studied using OH Planar Laser-Induced Fluorescence (PLIF) images to improve current understanding. Such flames are ubiquitous in most natural gas air combustors running gas turbines with dry exhaust gas recircul...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2022-11-01
|
| Series: | Energy and AI |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2666546822000398 |
| _version_ | 1818085643985616896 |
|---|---|
| author | Rathziel Roncancio Aly El Gamal Jay P. Gore |
| author_facet | Rathziel Roncancio Aly El Gamal Jay P. Gore |
| author_sort | Rathziel Roncancio |
| collection | DOAJ |
| description | Pockets of unburned material in turbulent premixed flames burning CH4, air, and CO2 were studied using OH Planar Laser-Induced Fluorescence (PLIF) images to improve current understanding. Such flames are ubiquitous in most natural gas air combustors running gas turbines with dry exhaust gas recirculation (EGR) for land-based power generation. Essential improvements continue in the characterization and understanding of turbulent flames with EGR particularly for transient events like ignition and extinction. Pockets and/or islands of unburned material within burned and unburned turbulent media are some of the features of these events. These features reduce the heat release rates and increase the carbon monoxide and hydrocarbons emissions. The present work involves Convolutional Neural Networks (CNN) based classification of PLIF images containing unburned pockets in three turbulent flames with 0%, 5%, and 10% CO2. The CNN model was constructed using three convolutional layers and two fully connected layers using dropout and weight decay. Accuracies of 94.2%, 92.3% and 89.2% were registered for the three flames, respectively. The present approach represents significant computational time savings with respect to conventional image processing methods. |
| first_indexed | 2024-12-10T20:13:03Z |
| format | Article |
| id | doaj.art-e01f07ea4d22444ca3d67100dbf2c86a |
| institution | Directory Open Access Journal |
| issn | 2666-5468 |
| language | English |
| last_indexed | 2024-12-10T20:13:03Z |
| publishDate | 2022-11-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Energy and AI |
| spelling | doaj.art-e01f07ea4d22444ca3d67100dbf2c86a2022-12-22T01:35:16ZengElsevierEnergy and AI2666-54682022-11-0110100193Turbulent flame image classification using Convolutional Neural NetworksRathziel Roncancio0Aly El Gamal1Jay P. Gore2School of Mechanical Engineering, Purdue University, West Lafayette, IN 47907, USA; Corresponding author.School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907, USASchool of Mechanical Engineering, Purdue University, West Lafayette, IN 47907, USAPockets of unburned material in turbulent premixed flames burning CH4, air, and CO2 were studied using OH Planar Laser-Induced Fluorescence (PLIF) images to improve current understanding. Such flames are ubiquitous in most natural gas air combustors running gas turbines with dry exhaust gas recirculation (EGR) for land-based power generation. Essential improvements continue in the characterization and understanding of turbulent flames with EGR particularly for transient events like ignition and extinction. Pockets and/or islands of unburned material within burned and unburned turbulent media are some of the features of these events. These features reduce the heat release rates and increase the carbon monoxide and hydrocarbons emissions. The present work involves Convolutional Neural Networks (CNN) based classification of PLIF images containing unburned pockets in three turbulent flames with 0%, 5%, and 10% CO2. The CNN model was constructed using three convolutional layers and two fully connected layers using dropout and weight decay. Accuracies of 94.2%, 92.3% and 89.2% were registered for the three flames, respectively. The present approach represents significant computational time savings with respect to conventional image processing methods.http://www.sciencedirect.com/science/article/pii/S2666546822000398CNNFlameNeural networkTurbulent |
| spellingShingle | Rathziel Roncancio Aly El Gamal Jay P. Gore Turbulent flame image classification using Convolutional Neural Networks Energy and AI CNN Flame Neural network Turbulent |
| title | Turbulent flame image classification using Convolutional Neural Networks |
| title_full | Turbulent flame image classification using Convolutional Neural Networks |
| title_fullStr | Turbulent flame image classification using Convolutional Neural Networks |
| title_full_unstemmed | Turbulent flame image classification using Convolutional Neural Networks |
| title_short | Turbulent flame image classification using Convolutional Neural Networks |
| title_sort | turbulent flame image classification using convolutional neural networks |
| topic | CNN Flame Neural network Turbulent |
| url | http://www.sciencedirect.com/science/article/pii/S2666546822000398 |
| work_keys_str_mv | AT rathzielroncancio turbulentflameimageclassificationusingconvolutionalneuralnetworks AT alyelgamal turbulentflameimageclassificationusingconvolutionalneuralnetworks AT jaypgore turbulentflameimageclassificationusingconvolutionalneuralnetworks |