State identification of gas-liquid two-phase flow in few-shot scenario

doi:10.16085/j.issn.1000-6613.2024-1747

Abstract

Abstract:

Gas-liquid two-phase flow has complex and changeable flow states, including typical states and transition states. The accurate identification and monitoring of flow state are crucial for understanding the mechanism of two-phase flow and ensuring the safe operation of industrial processes. However, the limitation of deployment cost of sensors and the difficulty of data acquisition in actual two-phase flow processes lead to the data deficiency in gas-liquid two-phase flow. In this paper, the problem of few-shot flow state identification of gas-liquid two-phase flow was systematically discussed, which was researched under the framework of information processing of conductance sensor, time-frequency analysis in multi-scale domain and few-shot learning by the prototypical network. Firstly, the water holdup information reflecting dynamic characteristics and flow structure of gas-liquid two-phase flow was obtained by the conductance method with fast response, safety and low cost. Then, the response signal of the conductance sensor was analyzed by time-frequency analysis. The fluctuation information of conductance signal in multi-scale domain was obtained by the improved empirical wavelet transform method, which characterized different flow states jointly. Finally, the extracted features were embedded into the prototypical network, and the model was trained under the meta-learning framework to solve the problem of gas-liquid two-phase flow state identification in few-shot scenario. The few-shot state identification experiments were carried out through 6 typical flow states and 4 transition states. When only 3 or 5 samples were available for target flow states during training, the comprehensive recognition accuracy exceeded 80%, which verified the effectiveness of the method.

Key words: gas-liquid two-phase flow, instrumentation, neural networks, state identification, few-shot learning, prototypical network

摘要：

气液两相流具有复杂多变的流动状态，包括典型状态和过渡状态。流动状态的准确识别与监测对深入理解两相流机理和保证工业过程的安全运行至关重要。然而，传感器部署成本的限制和实际流动过程数据采集的难度导致了气液两相流的数据欠缺。本文系统地讨论了气液两相流的少样本流动状态识别问题，采用电导传感器信息处理、多尺度域时频特性分析、原型网络小样本元学习框架。首先，利用响应快、安全、成本低的电导传感器进行测量，获取反映气液两相流动态特性和流动结构的含水率信息。然后，对电导传感器的响应信号进行时频分析，通过改进的经验小波变换法获得电导信号在多尺度域的波动信息，实现不同流动状态的联合表征。最后，将所提取特征嵌入原型网络中，在元学习框架下进行模型训练，解决少样本场景下的气液两相流动状态识别问题。利用6种典型流动状态和4种过渡状态进行了少样本状态识别实验，在目标流动状态只采用3个或5个样本进行训练的情况下，平均识别准确率超过80%，验证了方法的有效性。

关键词: 气液两相流, 仪器仪表, 神经网络, 状态识别, 少样本学习, 原型网络

CLC Number:

TP212

LI Linghan, ZHANG Shumei, DONG Feng. State identification of gas-liquid two-phase flow in few-shot scenario[J]. Chemical Industry and Engineering Progress, 2025, 44(4): 1794-1805.

李凌涵, 张淑美, 董峰. 少样本场景下的气液两相流动状态识别[J]. 化工进展, 2025, 44(4): 1794-1805.

Figures/Tables 11

References 30

1	谭超, 董峰. 多相流过程参数检测技术综述[J]. 自动化学报, 2013, 39(11): 1923-1932.
	TAN Chao, DONG Feng. Parameters measurement for multiphase flow process[J]. Acta Automatica Sinica, 2013, 39(11): 1923-1932.
2	BONIZZI Marco, ISSA Raad I. On the simulation of three-phase slug flow in nearly horizontal pipes using the multi-fluid model[J]. International Journal of Multiphase Flow, 2003, 29(11): 1719-1747.
3	FOSSA Marco. Design and performance of a conductance probe for measuring the liquid fraction in two-phase gas-liquid flows[J]. Flow Measurement and Instrumentation, 1998, 9(2): 103-109.
4	苏茜, 邓翔天, 刘振兴. 油气水三相流相含率超声测试模型优化[J]. 化工进展, 2024, 43(2): 791-799.
	SU Qian, DENG Xiangtian, LIU Zhenxing. Model optimization of phase fraction in oil-gas-water three-phase flow using ultrasonic testing technique[J]. Chemical Industry and Engineering Progress, 2024, 43(2): 791-799.
5	SALGADO César Marques, PEREIRA Cláudio M N A, SCHIRRU Roberto,et al. Flow regime identification and volume fraction prediction in multiphase flows by means of gamma-ray attenuation and artificial neural networks[J]. Progress in Nuclear Energy, 2010, 52(6): 555-562.
6	GU Mengtao, LI Jian, HOSSAIN Md Moinul,et al. High-resolution microscale velocity field measurement using light field particle image-tracking velocimetry[J]. Physics of Fluids, 2023, 35(11): 112006.
7	ZHANG Zhixiang, JIN Ningde, BAI Landi, et al. Nonlinear analysis for identification of vertical upward oil-water two-phase flow patterns[J]. IEEE Sensors Journal, 2024, 24(3): 3259-3265.
8	徐一, 李轶, 马志扬, 等. 基于电容层析测量的油气两相流动态信号时频分析方法[J]. 化工进展, 2024, 43(2): 855-864.
	XU Yi, LI Yi, MA Zhiyang, et al. Dynamic signal analysis for gas-oil two-phase flow in time-frequency domain based on electrical capacitance tomography[J]. Chemical Industry and Engineering Progress, 2024, 43(2): 855-864.
9	董峰, 李昭, 李凌涵, 等. 多模态动态核主成分分析的气液两相流状态监测[J]. 自动化学报, 2022, 48(3): 762-773.
	DONG Feng, LI Zhao, LI Linghan, et al. Flow state monitoring of gas-liquid two-phase flow using multiple dynamic kernel principle component analysis[J]. Acta Automatica Sinica, 2022, 48(3): 762-773.
10	LI Linghan, HAN Xinyi, DONG Feng, et al. Zero-shot state identification of industrial gas-liquid two-phase flow via supervised deep slow and steady feature analysis[J]. IEEE Transactions on Industrial Informatics, 2024, 20(6): 8170-8180.
11	史雪薇, 谭超, 董峰. 基于环形电导传感器的气液两相流流型识别与过程参数测量[J]. 化工进展, 2024, 43(2): 637-648.
	SHI Xuewei, TAN Chao, DONG Feng. Gas-liquid two-phase flow pattern identification and flow parameters measurement based on the ring-shape conductance sensor[J]. Chemical Industry and Engineering Progress, 2024, 43(2): 637-648.
12	SHI Xuewei, TAN Chao, DONG Xiaoxiao, et al. Structural velocity measurement of gas-liquid slug flow based on EMD of continuous wave ultrasonic Doppler[J]. IEEE Transactions on Instrumentation and Measurement, 2018, 67(11): 2662-2675.
13	王杰, 李爱蓉. 小波包分析识别气液固三相流流型[J]. 石油化工, 2020, 49(1): 62-69.
	WANG Jie, LI Airong. Identification of gas-liquid-solid three-phase flow patterns by wavelet packet analysis[J]. Petrochemical Technology, 2020, 49(1): 62-69.
14	GILLES Jérôme. Empirical wavelet transform[J]. IEEE Transactions on Signal Processing, 2013, 61(16): 3999-4010.
15	CHEN Jinglong, PAN Jun, LI Zipeng, et al. Generator bearing fault diagnosis for wind turbine via empirical wavelet transform using measured vibration signals[J]. Renewable Energy, 2016, 89: 80-92.
16	ANURAGI Arti, SISODIA Dilip Singh. Empirical wavelet transform based automated alcoholism detecting using EEG signal features[J]. Biomedical Signal Processing and Control, 2020, 57: 101777.
17	RAGAB Mohamed, OMER Osama A, Mohamed ABDEL-NASSER. Compressive sensing MRI reconstruction using empirical wavelet transform and grey wolf optimizer[J]. Neural Computing and Applications, 2020, 32(7): 2705-2724
18	REN Chao, JIANG Bin, LU Ningyun, et al. Meta-learning with distributional similarity preference for few-shot fault diagnosis under varying working conditions[J]. IEEE Transactions on Cybernetics, 2024, 54(5): 2746-2756.
19	SNELL Jake, SWERSKY Kevin, ZEMEL Richard, et al. Prototypical networks for few-shot learning[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, California, USA: ACM, 2017: 4080-4090.
20	VINYALS Oriol, BLUNDELL Charles, LILLICRAP Timothy, et al. Matching networks for one shot learning[C]//Proceedings of the 30th International Conference on Neural Information Processing Systems. Barcelona, Spain: ACM, 2016: 3637-3645.
21	XIE Tingli, HUANG Xufeng, CHOI Seung-Kyum. Metric-based meta-learning for cross-domain few-shot identification of welding defect[J]. Journal of Computing and Information Science in Engineering, 2023, 23(3): 030902.
22	TANG Tang, WANG Jingwei, YANG Tianyuan, et al. An improved prototypical network with L2 prototype correction for few-shot cross-domain fault diagnosis[J]. Measurement, 2023, 217: 113065.
23	LI Kang, SHANG Chao, YE Hao. Reweighted regularized prototypical network for few-shot fault diagnosis[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(5): 6206-6217.
24	WANG Yimeng, SHEN Jie, YANG Shusen, et al. Knowledge and data dual-driven fault diagnosis in industrial scenarios: A survey[J]. IEEE Internet of Things Journal, 2024, 11(11): 19256-19277.
25	BERTOLAZZI Enrico, FREGO Marco, BIRAL Francesco. Point data reconstruction and smoothing using cubic splines and clusterization[J]. Mathematics and Computers in Simulation, 2020, 176: 36-56.
26	DONG Feng, ZHANG Shuo, SHI Xuewei, et al. Flow regimes identification-based multidomain features for gas-liquid two-phase flow in horizontal pipe[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 7502911.
27	DENG Fang, GUO Su, ZHOU Rui, et al. Sensor multifault diagnosis with improved support vector machines[J]. IEEE Transactions on Automation Science and Engineering, 2017, 14(2): 1053-1063.
28	CHAI Zheng, ZHAO Chunhui. Enhanced random forest with concurrent analysis of static and dynamic nodes for industrial fault classification[J]. IEEE Transactions on Industrial Informatics, 2020, 16(1): 54-66.
29	ZHEN Rong, JIN Yongxing, HU Qinyou, et al. Maritime anomaly detection within coastal waters based on vessel trajectory clustering and Naïve Bayes classifier[J]. Journal of Navigation, 2017, 70(3): 648-670.
30	VAN DER MAATEN Laurens, HINTON Geoffrey. Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008, 9: 2579-2605.

序号	流动状态	水流量/m³·h^-1	气流量/m³·h^-1
1	泡状流	10.0、15.0	1.3~2.3
2	塞状流	2.0、5.0	1.3~2.3
3	弹状流	10.0、15.0	18.0~40.0
4	波状流	0.3、0.8	35.4~81.7
5	分层流	0.3、0.8	0.7~1.6
6	环状流	0.3、0.8	140.0~189.3
7	泡状流、塞状流间过渡	7.0、9.0	1.3~2.3
8	塞状流、弹状流间过渡	2.0、5.0	6.9~18.0
9	弹状流、环状流间过渡	2.0、5.0	78.4~88.5
10	分层流、波状流间过渡	0.3、0.8	20.0~38.0

序号	流动状态	水流量/m³·h^-1	气流量/m³·h^-1
1	泡状流	10.0、15.0	1.3~2.3
2	塞状流	2.0、5.0	1.3~2.3
3	弹状流	10.0、15.0	18.0~40.0
4	波状流	0.3、0.8	35.4~81.7
5	分层流	0.3、0.8	0.7~1.6
6	环状流	0.3、0.8	140.0~189.3
7	泡状流、塞状流间过渡	7.0、9.0	1.3~2.3
8	塞状流、弹状流间过渡	2.0、5.0	6.9~18.0
9	弹状流、环状流间过渡	2.0、5.0	78.4~88.5
10	分层流、波状流间过渡	0.3、0.8	20.0~38.0

分组	源域状态	源域训练样本数	目标域状态	目标域支持（训练）样本数	目标域查询（测试）样本数
A	2、3、5~7、9、10	32	1、4、8	3或5	16
B	1~6、8	32	7、9、10	3或5	16
C	1、4、5、7~10	32	2、3、6	3或5	16

分组	源域状态	源域训练样本数	目标域状态	目标域支持（训练）样本数	目标域查询（测试）样本数
A	2、3、5~7、9、10	32	1、4、8	3或5	16
B	1~6、8	32	7、9、10	3或5	16
C	1、4、5、7~10	32	2、3、6	3或5	16

方法	3-way 3-shot			3-way 5-shot
方法	A	B	C	A	B	C
EWT+SVM	68.08	70.42	66.26	70.65	72.35	69.38
EWT+RF	70.42	73.54	67.71	72.92	73.54	70.42
EWT+NB	65.63	68.75	66.20	69.80	72.92	68.75
时域、频域统计特征+PN	76.04	80.21	73.96	76.04	80.21	78.13
改进EWT+PN	81.25	82.29	77.08	80.21	83.33	78.13