Application of deep learning for segmentation of bubble dynamics in subcooled boiling

Soibam, Jerol; Scheiff, Valentin; Aslanidou, Ioanna; Kyprianidis, Konstantinos; Bel Fdhila, Rebei

doi:10.1016/j.ijmultiphaseflow.2023.104589

Cited by 11 publications

(2 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent advances in Machine Learning (ML) and Convolutional Neural Networks (CNNs) have enabled the development of automated image processing techniques for image segmentation, object tracking, and detection (Redmon et al 2016;He et al 2017). These techniques are particularly applicable in the study of multiphase flows, where the application of CNNs has been demonstrated in various studies (Hobold and da Silva 2019;Liu et al 2022;Soibam et al 2023). CNNs offer the ability to extract spatial information for different phases and detect gas-liquid interfaces, providing valuable insights into the dynamics of multiphase flows (Rutkowski et al 2022;Tai et al 2022;Sibirtsev et al 2023).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Machine learning aided investigation of dynamics of immiscible droplet impingement on liquid pools: A study across varying pool depths and droplet viscosities

Shahdhaar,

Srivast,

Srivast

2023

Preprint

View full text Add to dashboard Cite

The interactions of an immiscible droplet impinging on liquid pools bear significant implications across a wide array of applications, as well as in natural phenomena. In this paper, the dynamics associated with an immiscible droplet impinging on a liquid pool/film of varying depths have been elucidated. The study encompasses the impact of silicone oil droplets of four different viscosities (1, 10, 100, and 1000 cSt) upon a water pool of three non-dimensional pool heights h* = 1, 2.5, and 5. The phenomenon of droplet impact at two Weber numbers (We = 50 and 100) is captured through high-speed videography. The dynamics of impingement, associated with the immiscible liquid combination, are delineated by employing Mask R-CNN machine learning (ML) model. ML model generated masks are used to ascertain the dynamics of various cavity parameter. Further insights into the phenomena have been developed through a detailed energy analysis carried out pre- and post-impact. The performance of ML model is compared with the manually annotated images, exhibiting impressive level of agreement. Results reveal that during the cavity formation phase, low viscosity droplets conform to the cavity shape during their descend into the pool. In contrast, high viscosity droplets maintain their shape during cavity formation, showing pinning at the oil-water interface. Energy analysis shows better energy transfer from droplet to the cavity for low viscosity droplets (> 90%), while less than 50% of the impact energy is transferred for higher viscosity droplets. This study is among the first to apply machine learning to this complex fluid phenomenon, offering insights into the physics and potential applications in multiphase flows.

show abstract

Section: Introductionmentioning

confidence: 99%

“…obtained through the model and manually annotated masks for the image sequence obtained through high-speed imaging. The present approach for machine learning model validation is similar to that followed by Soibam et al (Soibam et al 2023). The comparison is made for six different cases with varying pool height and Weber number (see Fig.…”

mentioning

confidence: 99%