Advanced Deep Learning-Based Bubbly Flow Image Generator under Different Superficial Gas Velocities,Industrial & Engineering Chemistry Research

当前位置： X-MOL 学术 › Ind. Eng. Chem. Res. › 论文详情

Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)

Advanced Deep Learning-Based Bubbly Flow Image Generator under Different Superficial Gas Velocities
Industrial & Engineering Chemistry Research ( IF 3.8 ) Pub Date : 2022-01-14 , DOI: 10.1021/acs.iecr.1c03883
Zhong Xiang ₁ , Binbin Xie ₁ , Rui Fu ₁ , Miao Qian ₁

Affiliation

Bubble synthesis technology is beneficial to reduce the cost of visualization research on gas–liquid two-phase flow and provides an effective tool to benchmark data for the development of advanced image processing algorithms. In this work, we proposed an advanced StyleGAN2-based bubbly flow image generator, which was trained on 15 000 images obtained at 10 different superficial gas velocities. The main factors that restrict the network synthesis quality have been investigated, and quantitative evaluation methods based on the Yolov3 detector are also developed to validate the generator’s reliability and ability. With the optimized inputs and truncated latent vector, the generator can synthesize high-diversity bubbly flows with 512 × 512 pixel resolution, meanwhile, synthesizing more realistic and higher fidelity bubbly flows than existing technologies in terms of location distribution, size distribution characteristics, and morphology revivification. Such generation and detection methods will be useful for the development of two-phase flow research in practical applications.

中文翻译：

不同表观气体速度下基于高级深度学习的气泡流图像生成器

气泡合成技术有利于降低气液两相流可视化研究的成本，为开发先进的图像处理算法提供了基准数据的有效工具。在这项工作中，我们提出了一种先进的基于 StyleGAN2 的气泡流图像生成器，该生成器在 10 种不同的表观气体速度下获得的 15000 张图像上进行了训练。研究了制约网络合成质量的主要因素，并开发了基于Yolov3检测器的定量评价方法，验证了生成器的可靠性和能力。通过优化的输入和截断的潜在向量，生成器可以合成具有 512 × 512 像素分辨率的高多样性气泡流，同时，在位置分布、尺寸分布特征和形态再生方面，合成比现有技术更逼真、保真度更高的气泡流。这种生成和检测方法将有助于两相流研究在实际应用中的发展。

更新日期：2022-01-26

点击分享查看原文

点击收藏

阅读更多本刊最新论文本刊介绍/投稿指南11