Abstract: To solve the difficulty of flow pattern recognition for underwater gas jet, a flow pattern recognition method was proposed based on direct digital image processing and convolutional neural network. The original image dataset was directly obtained by the underwater gas jet experimental system, and the flow control and acquisition system was used to obtain the underwater jet shapes under different operating conditions. The image was preprocessed, and the unnecessary information in the image was removed through image cropping, while the noise in the image was suppressed by the two-dimensional median filtering algorithm. The maximum inter class variance algorithm was used for binarization to provide a direct bubble image dataset for flow pattern recognition in convolutional neural networks. Three different classical convolutional neural networks of LeNet, AlexNet and ResNet were used to train the processed images of different flow patterns. The classification model was optimized by continuously adjusting parameters, and precision, sensitivity, specificity and accuracy were used as evaluation indicators for underwater jet flow pattern classification. The results show that ResNet can achieve the highest precision, sensitivity and specificity in the transition zone and jet zone, respectively. ResNet can achieve the best precision and specificity in the bubble flow zone. ResNet has the best efficiency for identifying convective patterns with recognition accuracy of 96.2%.
王莹, 沈洋, 戚二帅, 杨斌. 基于深度学习的水下射流流型识别[J]. 江苏大学学报(自然科学版), 2023, 44(4): 437-443.
WANG Ying, SHEN Yang, QI Ershuai, YANG Bin. Flow pattern identification of underwater gas jet based on deep learning[J]. Journal of Jiangsu University(Natural Science Eidtion)
, 2023, 44(4): 437-443.
LI T T, HU J, CAO X J, et al. Pinch-off process of underwater annular-nozzled gas jet[J]. CIESC Journal,2017,68(12):4565-4575.(in Chinese)
[2]
XIANG M, ZHAO X Y, ZHOU H C. Transient dynamic analysis for the submerged gas jet in flowing water[J]. European Journal of Mechanics / B Fluids, 2021,85:351-360.
[3]
BESAGNI G. Bubble column fluid dynamics: a novel perspective for flow regimes and comprehensive experimental investigations[J]. International Journal of Multiphase Flow, DOI: 10.1016/j.ijmultiphaseflow.2020.103510.
[4]
MANTRIPRAGADA V T, SAHU S, SARKAR S. Morphology and flow behavior of buoyant bubble plumes[J]. Chemical Engineering Science,DOI:10.1016/j.ces.2020.116098.
[5]
LUCAS D, ZIEGENHEIN T. Influence of the bubble size distribution on the bubble column flow regime[J]. International Journal of Multiphase Flow,DOI:10.1016/j.ijmultiphaseflow.2019.103092.
[6]
TCHOWA MEDJIADE W, ROSENBAUM ALVARO A, SCHUMPE A. Flow regime transitions in a bubble co-lumn[J]. Chemical Engineering Science, 2017,170:263-269.
[7]
CERQUEIRA R F L, PALADINO E E. Development of a deep learning-based image processing technique for bubble pattern recognition and shape reconstruction in dense bubbly flows[J]. Chemical Engineering Science, DOI:10.1016/j.ces.2020.116163.
LI J, WANG Z W, FAN L L, et al. Few-shot learning method for image classification of apple leaf diseases[J]. Journal of Jilin University (Science Edition), 2022, 60(4):906-910. (in Chinese)
LI X Y, LI H Y. Feature point detection method of pig face based on convolutional neural network[J]. Journal of Jilin University (Science Edition),2022,60(3):609-616. (in Chinese)
[10]
MANJREKAR O N, DUDUKOVIC M P. Identification of flow regime in a bubble column reactor with a combination of optical probe data and machine learning technique[J]. Chemical Engineering Science: X,DOI:10.1016/j.cesx.2019.100023.
WENG R Y, SUN B, ZHAO Y X, et al. Flow pattern recognition method of gas-liquid two-phase flow based on adaptive optimal kernel and convolution neural network[J].CIESC Journal, 2018,69(12):5065-5072. (in Chinese)
HE Y, YANG P, WANG C S, et al. Research on financing websites identification based on deep neural network[J]. Journal of Sichuan University (Natural Science Edition), DOI:10.19907/j.0490-6756.2021.033003.(in Chinese)