基于深度学习的湍流火焰三维羟基浓度场的时间超分辨率成像

钟越; 蔡敏男; 徐文江; 杨帆

doi:10.13224/j.cnki.jasp.20230071

基于深度学习的湍流火焰三维羟基浓度场的时间超分辨率成像

doi: 10.13224/j.cnki.jasp.20230071

钟越^{1, 2,},
蔡敏男¹,
徐文江¹,
杨帆^{1, 2}

1.
厦门大学航空航天学院，福建厦门 361005
2.
厦门大学厦门市大数据智能分析与决策重点实验室，福建厦门 361005

基金项目: 国家自然科学基金面上项目（62173282）；国家自然科学基金青年基金（52006184）；厦门市自然科学基金面上项目（3502Z20227180）；科技部科技创新2030—“新一代人工智能”重大项目（2021ZD0112600）

详细信息

作者简介:
钟越（1999-），女，硕士生，主要从事基于深度学习的流场重建研究

中图分类号: V231；TK12
计量
- 文章访问数: 549
- HTML浏览量: 372
- PDF量: 67
- 被引次数: 0
出版历程
- 收稿日期: 2023-02-13
- 网络出版日期: 2024-03-14

Temporal super-resolution imaging of 3D OH concentration field in turbulent flame based on deep learning

ZHONG Yue^{1, 2
,},
CAI Minnan¹,
XU Wenjiang¹,
YANG Fan^{1, 2}

1.
School of Aerospace Engineering，Xiamen University，Xiamen Fujian 361005，China
2.
Xiamen Key Laboratory of Big Data Intelligent Analysis and Decision，Xiamen University，Xiamen Fujian 361005，China

摘要

摘要:
针对火焰三维羟基浓度场的高速测量难度大、成本昂贵的问题，提出一种基于深度学习的帧重建模型Cycle-3D-CNN，用于连续时间的湍流火焰三维羟基浓度场数据。使用基于循环一致性的三维卷积神经网络（3D-CNN），以数值驱动的方式实现了更高的时间分辨率。在实验分析中使用该模型分别实现了三维羟基浓度场时间序列的2倍和3倍时间分辨率提升，验证了其良好的重建性能。在两种实验结果中，峰值信噪比（PSNR）均值分别达到了33.57 dB和30.37 dB，结构相似性（SSIM）指数分别达到了0.899和0.813，均优于传统的帧重建方法。
- 湍流燃烧 /
- 羟基自由基 /
- 时间超分辨率 /
- 循环一致性模型 /
- 深度学习
Abstract:
In response to the difficulty and high cost of high-speed measurement of flame hydroxyl concentration field, a Cycle-3D-CNN model based on deep learning was proposed for temporal reconstruction of three-dimensional (3D) hydroxyl concentration fields in turbulent flames. It achieved a higher temporal resolution by utilizing a data-driven approach with a 3D convolutional neural network (3D-CNN) based on cycle consistency. In the experimental analysis, the model was used to achieve a two-fold and three-fold increase in temporal resolution of the 3D hydroxyl concentration field time series, respectively. In both experimental results, the mean peak signal-to-noise ratio (PSNR) reached 33.57 dB and 30.37 dB, respectively, the structural similarity (SSIM) indices reached 0.899 and 0.813, respectively, outperforming traditional frame reconstruction method.
- turbulent combustion /
- OH radical /
- temporal super-resolution /
- cycle-consistent model /
- deep learning

HTML

图 1 湍流火焰三维OH浓度场可视化渲染

Figure 1. 3D OH concentration field of turbulent flame

下载: 全尺寸图片幻灯片

图 2 3D-CNN模块网络结构

Figure 2. Network structure of 3D-CNN model

下载: 全尺寸图片幻灯片

图 3 Cycle-3D-CNN完整流程图

Figure 3. Structure of Cycle-3D-CNN model

下载: 全尺寸图片幻灯片

图 4 30个测试数据的双倍时间分辨率帧重建结果定量评估

Figure 4. Quantitative evaluation of the frame reconstruction results of double temporal resolution for 30 test data

下载: 全尺寸图片幻灯片

图 5 30个测试数据的3倍时间分辨率帧重建结果定量评估

Figure 5. Quantitative evaluation of the frame reconstruction results of triple temporal resolution for 30 test data

下载: 全尺寸图片幻灯片

图 6 双倍时间分辨率帧重建结果视觉对比

Figure 6. Visual comparison of the frame reconstruction results of double temporal resolution

下载: 全尺寸图片幻灯片

图 7 3倍时间分辨率帧重建结果视觉对比

Figure 7. Visual comparison of the frame reconstruction results of triple temporal resolution

下载: 全尺寸图片幻灯片

图 8 双倍和3倍时间分辨率帧重建结果的强度分布对比

Figure 8. Intensity comparison of the frame reconstruction results of double and triple temporal resolution

下载: 全尺寸图片幻灯片

参考文献(27)

[1]	何赞,乐嘉陵,田野,等. 超燃燃烧室中燃烧反应区域变化情况试验[J]. 航空动力学报,2023,38(10): 2370-2382. HE Zan,LE Jialing,TIAN Ye,et al. Experiment on variation of combustion reaction region in a scramjet combustor[J]. Journal of Aerospace Power,2023,38(10): 2370-2382. (in Chinese HE Zan, LE Jialing, TIAN Ye, et al. Experiment on variation of combustion reaction region in a scramjet combustor[J]. Journal of Aerospace Power, 2023, 38(10): 2370-2382. (in Chinese)
[2]	陈晓丽,金川,苏秋成,等. 基于激光诱导荧光法的同轴射流火焰中羟基自由基、甲醛、发热率与一氧化氮的二维可视化测量[J]. 分析测试技术与仪器,2021,27(3): 182-188. CHEN Xiaoli,JIN Chuan,SU Qiucheng,et al. Two-dimensional visualization measurement of hydroxyl radical,formaldehyde,heat release rate and nitric oxide in co-flow jet flame based on planar laser induced fluorescence technology[J]. Analysis and Testing Technology and Instruments,2021,27(3): 182-188. (in Chinese CHEN Xiaoli, JIN Chuan, SU Qiucheng, et al. Two-dimensional visualization measurement of hydroxyl radical, formaldehyde, heat release rate and nitric oxide in co-flow jet flame based on planar laser induced fluorescence technology[J]. Analysis and Testing Technology and Instruments, 2021, 27(3): 182-188. (in Chinese)
[3]	LECUN Y,BENGIO Y,HINTON G. Deep learning[J]. Nature,2015,521: 436-444. doi: 10.1038/nature14539
[4]	JIN K H,MCCANN M T,FROUSTEY E,et al. Deep convolutional neural network for inverse problems in imaging[J]. IEEE Transactions on Image Processing,2017,26(9): 4509-4522. doi: 10.1109/TIP.2017.2713099
[5]	孙安泰. 基于深度学习的火焰三维温度场层析重建及预测研究[D]. 哈尔滨: 哈尔滨工业大学,2021. SUN Antai. Tomography reconstruction and prediction of three-dimensional temperature field of flame based on deep learning[D]. Harbin: Harbin Institute of Technology,2021. (in Chinese SUN Antai. Tomography reconstruction and prediction of three-dimensional temperature field of flame based on deep learning[D]. Harbin: Harbin Institute of Technology, 2021. (in Chinese)
[6]	XU Wenjiang,LUO Weiyi,WANG Yu,et al. Data-driven three-dimensional super-resolution imaging of a turbulent jet flame using a generative adversarial network[J]. Applied Optics,2020,59(19): 5729-5736. doi: 10.1364/AO.392803
[7]	GUO Hao,ZHANG Wei,NIE Xiangyu,et al. High-speed planar imaging of OH radicals in turbulent flames assisted by deep learning[J]. Applied Physics B,2022,128(3): 52. doi: 10.1007/s00340-021-07742-2
[8]	ZHU Junyan,PARK T,ISOLA P,et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]//2017 IEEE International Conference on Computer Vision. Piscataway,US: IEEE,2017: 2242-2251.
[9]	TRAN D,BOURDEV L,FERGUS R,et al. Learning spatiotemporal features with 3D convolutional networks[C]//2015 IEEE International Conference on Computer Vision. Piscataway,US: IEEE,2015: 4489-4497.
[10]	FUKAMI K,FUKAGATA K,TAIRA K. Super-resolution reconstruction of turbulent flows with machine learning[J]. Journal of Fluid Mechanics,2019,870: 106-120. doi: 10.1017/jfm.2019.238
[11]	FUKAMI K,FUKAGATA K,TAIRA K. Machine-learning-based spatio-temporal super resolution reconstruction of turbulent flows[J]. Journal of Fluid Mechanics,2021,909: A9. doi: 10.1017/jfm.2020.948
[12]	KIM H,KIM J,WON S,et al. Unsupervised deep learning for super-resolution reconstruction of turbulence[J]. Journal of Fluid Mechanics,2021,910: A29. doi: 10.1017/jfm.2020.1028
[13]	BARLOW R S,FRANK J H. Effects of turbulence on species mass fractions in methane/air jet flames[J]. Symposium (International) on Combustion,1998,27(1): 1087-1095. doi: 10.1016/S0082-0784(98)80510-9
[14]	LYSENKO D A,ERTESVÅG I S,RIAN K E. Numerical simulations of the Sandia flame D using the eddy dissipation concept[J]. Flow,Turbulence and Combustion,2014,93(4): 665-687. doi: 10.1007/s10494-014-9561-5
[15]	JONES W P,PRASAD V N. Large eddy simulation of the Sandia flame series (D-F) using the Eulerian stochastic field method[J]. Combustion and Flame,2010,157(9): 1621-1636. doi: 10.1016/j.combustflame.2010.05.010
[16]	NIK M B,YILMAZ S L,GIVI P,et al. Simulation of Sandia flame D using velocity-scalar filtered density function[J]. AIAA Journal,2010,48(7): 1513-1522. doi: 10.2514/1.J050154
[17]	李人宪. 有限体积法基础[M]. 北京: 国防工业出版社,2005. LI Renxian. Fundamentals of finite volume method[M]. Beijing: National Defense Industry Press,2005. (in Chinese LI Renxian. Fundamentals of finite volume method[M]. Beijing: National Defense Industry Press, 2005. (in Chinese)
[18]	MINOTTI A,SCIUBBA E. LES of a meso combustion chamber with a detailed chemistry model: comparison between the flamelet and EDC models[J]. Energies,2010,3(12): 1943-1959. doi: 10.3390/en3121943
[19]	LIU Yulun,LIAO Y T,LIN Yenyu,et al. Deep video frame interpolation using cyclic frame generation[J]. Proceedings of the AAAI Conference on Artificial Intelligence,2019,33(1): 8794-8802. doi: 10.1609/aaai.v33i01.33018794
[20]	REDA F,SUN Deqing,DUNDAR A,et al. Unsupervised video interpolation using cycle consistency[C]//2019 IEEE/CVF International Conference on Computer Vision. Piscataway,US: IEEE,2019: 892-900.
[21]	HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al. Delving deep into rectifiers: surpassing human-level performance on ImageNet classification[C]//2015 IEEE International Conference on Computer Vision. Piscataway,US: IEEE,2015: 1026-1034.
[22]	SIMONYAN K,ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL].[2023-01-15]. http://www.robots.ox.ac.uk/˜vgg/research/very_deep/.
[23]	WANG Zhou,BOVIK A C,SHEIKH H R,et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Transactions on Image Processing,2004,13(4): 600-612. doi: 10.1109/TIP.2003.819861
[24]	YANG C Y,MA Chao,YANG M H. Single-image super-resolution: a benchmark[C]//European Conference on Computer Vision. Cham: Springer,2014: 372-386.
[25]	JIANG Huaizu,SUN Deqing,JAMPANI V,et al. Super SloMo: high quality estimation of multiple intermediate frames for video interpolation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway,US: IEEE,2018: 9000-9008.
[26]	LI Haopeng,YUAN Yuan,WANG Qi. Video frame interpolation via residue refinement[C]//2020 IEEE International Conference on Acoustics,Speech and Signal Processing. Piscataway,US: IEEE,2020: 2613-2617.
[27]	CHOI M,KIM H,HAN B,et al. Channel attention is all you need for video frame interpolation[J]. Proceedings of the AAAI Conference on Artificial Intelligence,2020,34(7): 10663-10671. doi: 10.1609/aaai.v34i07.6693