Image Classification of Two-phase Flow Patterns Inside Effervescent Atomization Nozzles Based on Convolutional Neural Networks
Journal of Applied Fluid Mechanics · 01.01.2027 · 20 (1)
DOI: 10.47176/jafm.20.1.4702 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Applied Fluid Mechanics, online erschienen am 01.01.2027. Verfasst von C. Chen und 4 weiteren Autor:innen.
Abstract (Originalsprache)
The accurate classification of two-phase flow patterns within effervescent atomization nozzles is challenging due to the absence of unified standards and the existence of transitional regimes. To address the aforementioned challenge, this article proposes a novel image classification method leveraging a modified VGG-16 architecture. Five distinct flow pattern categories are then identified in high-speed images: bubbly, slug, churn, bubbly-slug mixed, and bubbly-churn mixed flows. The original flatten layer is replaced with a lightweight adaptive feature aggregator (LAFA) layer to reduce the number of trainable parameters and improve feature discrimination. This layer integrates attention mechanisms and multi-scale dilated convolutions. Ablation experiments are conducted, demonstrating that the proposed model achieves a classification accuracy of 97.47%, outperforming VGG-16, ResNet50, and DenseNet121 which accuracies are (94.07–94.17, 97.08, and 97.29%, respectively. Gradient-weighted Class Activation Mapping and decision boundary analysis are also conducted. The obtained results show that the proposed model focuses on physically meaningful regions, with misclassifications mainly occurring in critical transition zones between flow patterns. This article presents an objective, automated, and high-accuracy solution for flow pattern recognition, providing technical support for future studies on effervescent atomization mechanisms.
Zitieren
APA
Chen, C., Han, Y. H., Zheng, Q. Y., Guo, Y. H., & Chen, Z. W. (2027). Image Classification of Two-phase Flow Patterns Inside Effervescent Atomization Nozzles Based on Convolutional Neural Networks. Journal of Applied Fluid Mechanics, 20(1). https://doi.org/10.47176/jafm.20.1.4702
Vancouver
Chen C, Han YH, Zheng QY, Guo YH, Chen ZW. Image Classification of Two-phase Flow Patterns Inside Effervescent Atomization Nozzles Based on Convolutional Neural Networks. Journal of Applied Fluid Mechanics. 2027;20(1). doi:10.47176/jafm.20.1.4702
BibTeX
@article{Chen2027,
author = {Chen, C. and Han, Y. H. and Zheng, Q. Y. and Guo, Y. H. and Chen, Z. W.},
title = {Image Classification of Two-phase Flow Patterns Inside Effervescent Atomization Nozzles Based on Convolutional Neural Networks},
journal = {Journal of Applied Fluid Mechanics},
year = {2027},
volume = {20},
number = {1},
doi = {10.47176/jafm.20.1.4702}
}
EUVIMED