Gas-Fired Flame Stability Based on Optical Flow Method and Deep Learning

doi:10.16183/j.cnki.jsjtu.2020.111

Abstract

Abstract:

The stability of gas-fired flame is studied by combining the optical flow method and deep learning. The optical flow vector of the flame image is directly calculated by using the optical flow method. The pulsation of the flame in the two-dimensional image is observed, and an optical flow pulsation evaluation model is proposed to evaluate the stability of the flame. In addition, a deep convolutional neural network based on VGG-Nets is built and fine adjustments are made on ImageNet pre-training weights. Combining the static and dynamic characteristics of flames, the classification and recognition of five typical combustion states are achieved. The results show that this method has a good judgment ability for different combustion states of flames and a high recognition rate for unstable combustion flames.

Key words: gas-fired flame, stability, optical flow method, deep learning

CLC Number:

TK39

WANG Yu, YU Yuefeng, ZHU Xiaolei, ZHANG Zhongxiao. Gas-Fired Flame Stability Based on Optical Flow Method and Deep Learning[J]. Journal of Shanghai Jiao Tong University, 2021, 55(4): 462-470.

Figures/Tables 8

Fig.1

Tab.1

Fig.2

Fig.3

Fig.4

Fig.5

Tab.2

Fig.6

References 16

[1]	娄春. 工程燃烧诊断学[M]. 北京: 中国电力出版社, 2016.
	LOU Chun. Engineering combustion diagnostics[M]. Beijing: China Electric Power Press, 2016.
[2]	白卫东, 严建华, 池涌, 等. PCA和SVM在火焰监测中的应用研究[J]. 中国电机工程学报, 2004, 24(2):186-191.
	BAI Weidong, YAN Jianhua, CHI Yong, et al. Application of PCA and SVM in flame monitoring[J]. Proceedings of the CSEE, 2004, 24(2):186-191.
[3]	吴一全, 宋昱, 周怀春. 基于灰度熵多阈值分割和SVM的火焰图像状态识别[J]. 中国电机工程学报, 2013, 33(20):66-73.
	WU Yiquan, SONG Yu, ZHOU Huaichun. Flame image state recognition based on gray entropy multiple threshold segmentation and SVM[J]. Proceedings of the CSEE, 2013, 33(20):66-73.
[4]	郭建民. 模糊免疫网络算法在数字图像火焰监测中的应用[J]. 中国电机工程学报, 2007, 27(2):2-5.
	GUO Jianmin. Application of fuzzy immune network algorithm in digital image flame monitoring[J]. Proceedings of the CSEE, 2007, 27(2):2-5.
[5]	SUN D, LU G, ZHOU H, et al. Condition monitoring of combustion processes through flame imaging and kernel principal component analysis[J]. Combustion Science and Technology, 2013, 185(9):1400-1413. doi: 10.1080/00102202.2013.798316 URL
[6]	GIRSHICK R. Fast R-CNN[C]//International Conference on Computer Vision. Santiago, USA: IEEE, 2015: 1440-1448.
[7]	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 39(6):1137-1149.
[8]	韩溟. 基于光谱和图像综合测试系统的火焰特性研究[D]. 西安: 西安电子科技大学, 2007.
	HAN Ming. Study on flame characteristics based on spectrum and image synthesis test system[D]. Xi’an: Xidian University, 2007.
[9]	田正林, 余岳峰, 朱小磊, 等. 基于图像处理的燃气火焰稳定性检测试验研究[J]. 动力工程学报, 2019, 39(10):811-817.
	TIAN Zhenglin, YU Yuefeng, ZHU Xiaolei, et al. Experimental study on flame stability detection based on image processing[J]. Chinese Journal of Power Engineering, 2019, 39(10):811-817.
[10]	赵健, 王宾, 马苗. 数字信号处理[M], 北京: 清华大学出版社, 2012.
	ZHAO Jian, WANG Bin, MA Miao. Digital signal processing[M]. Beijing: Tsinghua University Press, 2012.
[11]	周怀春. 炉内火焰可视化检测原理与技术[M]. 北京: 科学出版社, 2005.
	ZHOU Huaichun. Principle and technology of visual flame detection in furnace[M]. Beijing: Science Press, 2005.
[12]	SUN D, ROTH S, BLACK M J. Secrets of optical flow estimation and their principles[C]//The Twenty-Third IEEE Conference on Computer Vision and Pattern Recognition. San Francisco, CA, USA: IEEE, 2010: 2432-2439.
[13]	SUN D, ROTH S, BLACK M J, et al. A Quantitative analysis of current practices in optical flow estimation and the principles behind them[J]. International Journal of Computer Vision, 2014, 106(2):115-137. doi: 10.1007/s11263-013-0644-x URL
[14]	MUELLER M, KARASEV P, KOLESOV I, et al. Optical flow estimation for flame detection in videos[J]. IEEE Transactions on Image Processing, 2013, 22(7):2786-2797. doi: 10.1109/TIP.2013.2258353 URL
[15]	KOLESOV I, KARASEV P, TANNENBAUM A, et al. Fire and smoke detection in video with optimal mass transport based optical flow and neural networks[C]//Proceedings of the International Conference on Image Processing. Hong Kong, China: IEEE, 2010.
[16]	SIMONYAN, KAREN, ZISSERMAN, et al. Very deep convolutional networks for large-scale image recognition[C]//Ninth International Conference on Learning Representations. San Francisco, CA, USA: 2015: 1-14.

块	操作类型	输入维度	输出维度
B1	卷积×2,ReLU×2,平均汇合	224×224×3	112×112×64
B2	卷积×2,ReLU×2,平均汇合	112×112×64	56×56×128
B3	卷积×3,ReLU×3,平均汇合	56×56×128	28×28×256
B4	卷积×3,ReLU×3,平均汇合	28×28×256	14×14×512
B5	卷积×3,ReLU×3,平均汇合	14×14×512	7×7×512

火焰类型	训练组数	验证组数	平均正确率/%
扩散稳定	3881	971	100
扩散不稳定	197	50	99.95
预混稳定	4646	1162	100
预混不稳定	178	45	99.99
熄火及背景	244	61	100