Automated bubble analysis of high-speed subcooled flow boiling images using U-net transfer learning and global optical flow

Seong, Jee Hyun; Ravichandran, M.; Su, Guosheng; Phillips, Brent; Bucci, Matteo

doi:10.1016/j.ijmultiphaseflow.2022.104336

Cited by 20 publications

(4 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The decoder performs the upsampling operation to recover the spatial resolution through Conv2DTranspose layers, effectively performing the inverse of Conv2D, and fuses the feature maps of the same size into the shallow layer. Along with transfer learning, the U-Net architecture has proven effective for segmenting bubbles or particles from diverse backgrounds with the help of very few additional ground truth segmented images that act as training data. ,, …”

Section: Discussionmentioning

confidence: 99%

“…13 Poletaev et al 14 developed a multistep neural network (NN) approach capable of identifying overlapping, blurred, and nonspherical bubbles in turbulent bubbly jets, achieving detection across volume gas fractions of 0 to 2.5%, only with errors up to 20% occurring at the edge of the measurement domain. Wang et al 15 coupled a convolutional neural network (CNN) with an improved threeframe difference method and an intersection-over-union (IoU) postscreening algorithm to extract bubble patterns in plate heat exchangers. This approach not only precisely captured and tracked individual bubble behavior and hydrodynamic events but also achieved an average precision rate of over 94%, significantly enhancing the bubble flow analysis.…”

Section: Introductionmentioning

confidence: 99%

“…This approach not only precisely captured and tracked individual bubble behavior and hydrodynamic events but also achieved an average precision rate of over 94%, significantly enhancing the bubble flow analysis. Furthermore, Seong et al utilized a U-Net-based CNN model to segment bubbles during the boiling process in subcooled flow conditions, identifying bubble trajectories with more than 90% accuracy.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Machine Learning Assisted Experimental Characterization of Bubble Dynamics in Gas–Solid Fluidized Beds

Jiang,

Wu,

Francia

et al. 2024

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

This study introduces a machine learning (ML)assisted image segmentation method for automatic bubble identification in gas−solid quasi-2D fluidized beds, offering enhanced accuracy in bubble recognition. Binary images are segmented by the ML method, and an in-house Lagrangian tracking technique is developed to track bubble evolution. The MLassisted segmentation method requires few training data, achieves an accuracy of 98.75%, and allows for filtering out common sources of uncertainty in hydrodynamics, such as varying illumination conditions and out-of-focus regions, thus providing an efficient tool to study bubbling in a standard, consistent, and repeatable manner. In this work, the ML-assisted methodology is tested in a particularly challenging case: structured oscillating fluidized beds, where the spatial and time evolution of the bubble position, velocity, and shape are characteristics of the nucleation-propagationrupture cycle. The new method is validated across various operational conditions and particle sizes, demonstrating versatility and effectiveness. It shows the ability to capture challenging bubbling dynamics and subtle changes in velocity and size distributions observed in beds of varying particle size. New characteristic features of oscillating beds are identified, including the effect of frequency and particle size on the bubble morphology, aspect, and shape factors and their relationship with the stability of the flow, quantified through the rate of coalescence and splitting events. This type of combination of classic analysis with the application of the ML assisted techniques provides a powerful tool to improve standardization and address the reproducibility of hydrodynamic studies, with the potential to be extended from gas−solid fluidization to other multiphase flow systems.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Machine Learning Assisted Experimental Characterization of Bubble Dynamics in Gas–Solid Fluidized Beds

Jiang,

Wu,

Francia

et al. 2024

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

show abstract

“…However, most of the studies focus on the prediction of heat transfer coefficients and pressure drops based on universal consolidated data under the use of artificial neural networks [ 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 ]. So far, only a handful of studies have used machine learning based on convolutional neural networks (CNNs) to automatically detect bubbles, classify flow regimes, and calculate void fractions from HSV images taken during two-phase flow processes [ 23 , 24 , 25 , 26 ].…”

Section: Introductionmentioning

confidence: 99%

Computer-Vision- and Deep-Learning-Based Determination of Flow Regimes, Void Fraction, and Resistance Sensor Data in Microchannel Flow Boiling

Schepperle,

Junaid,

Woias

2024

Sensors

View full text Add to dashboard Cite

The aim of this article is to introduce a novel approach to identifying flow regimes and void fractions in microchannel flow boiling, which is based on binary image segmentation using digital image processing and deep learning. The proposed image processing pipeline uses adaptive thresholding, blurring, gamma correction, contour detection, and histogram comparison to separate vapor from liquid areas, while the deep learning method uses a customized version of a convolutional neural network (CNN) called U-net to extract meaningful features from video frames. Both approaches enabled the automatic detection of flow boiling conditions, such as bubbly, slug, and annular flow, as well as automatic void fraction calculation. Especially CNN demonstrated its ability to deliver fast and dependable results, presenting an appealing substitute to manual feature extraction. The U-net-based CNN was able to segment flow boiling images with a Dice score of 99.1% and classify the above flow regimes with an overall classification accuracy of 91%. In addition, the neural network was able to predict resistance sensor readings from image data and assign them to a flow state with a mean squared error (MSE) < 10−6.

show abstract

A comprehensive review of experimental and numerical studies on liquid metal-gas two-phase flows and associated measurement challenges

Saraswat,

Fraile,

Gedupudi

et al. 2025

Annals of Nuclear Energy

View full text Add to dashboard Cite

Automated bubble analysis of high-speed subcooled flow boiling images using U-net transfer learning and global optical flow

Cited by 20 publications

References 48 publications

Machine Learning Assisted Experimental Characterization of Bubble Dynamics in Gas–Solid Fluidized Beds

Machine Learning Assisted Experimental Characterization of Bubble Dynamics in Gas–Solid Fluidized Beds

Computer-Vision- and Deep-Learning-Based Determination of Flow Regimes, Void Fraction, and Resistance Sensor Data in Microchannel Flow Boiling

A comprehensive review of experimental and numerical studies on liquid metal-gas two-phase flows and associated measurement challenges

Contact Info

Product

Resources

About