Unveiling two-phase flow behavior and flooding precursors in a Rotating Packed Bed through deep learning-based image analysis
Abstract
Rotating Packed Beds (RPB) technology has the potential to capture carbon dioxide emissions from industrial flues. A porous packing in rotation provides an extensive surface area for effective mass transfer between a gas and a solvent. The application of centrifugal force enables counter-current flow of the two fluids, enhancing the efficiency of the process. HiGee contactors present a superior alternative to traditional packed columns. In these systems, the solvent is subjected to acceleration up to a thousand times that of gravity, significantly increasing the mass transfer rates by one to two orders of magnitude. Additionally, these systems are remarkably compact, being ten times smaller than their conventional counterparts. Flooding in RPBs represents a significant operational challenge, characterized by the inability of the solvent to evacuate through the packing. This issue arises due to various resistive forces, such as gravitational and frictional forces, which hinder the solvent's exit from the packing. Operational RPBs have been documented using a two-camera setup to study flooding. One camera focuses on the solvent injection point to monitor flooding, while the other captures the droplets exiting the packing. Initial results align with literature. Given the extreme challenges posed by the imaging conditions, deep learning with convolutional neural networks for segmentation is essential to extract meaningful information from these images. This deep learningdriven analysis has been pivotal in characterizing the geometric and dynamic properties of the droplets, revealing a previously unrecognized precursor to flooding in RPB systems.
Related
- Critical behavior and critical exponents of rotating QCD matter
- Unveiling wing turbulence dynamics through explainable deep learning
- Improving Reduced-Order Rotating Detonation Engine Models with Data Assimilation and Machine Learning
- Predicting Locomotor Behavior on an Irregular Surface from Even-Surface Gait: A Machine Learning Approach
- A Machine-Learning Based Approach to the Evaluation of the Critical Scaling Behavior of Anisotropic Spin Systems
BibTeX
@article{rodriguez2026unveiling,
title = {{Unveiling two-phase flow behavior and flooding precursors in a Rotating Packed Bed through deep learning-based image analysis}},
author = {S. Rodriguez and Cyril Cassar and Maxime Moreaud and Michele Bardi and Jérôme Chérel and Vincent Ricordeau and Veronique Penin},
journal = {Chemical Engineering and Processing: Process Intensification},
year = {2026},
month = aug,
doi = {10.1016/j.cep.2026.111012},
url = {https://ifp.hal.science/hal-05778720v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Unveiling two-phase flow behavior and flooding precursors in a Rotating Packed Bed through deep learning-based image analysis
AU - S. Rodriguez
AU - Cyril Cassar
AU - Maxime Moreaud
AU - Michele Bardi
AU - Jérôme Chérel
AU - Vincent Ricordeau
AU - Veronique Penin
PY - 2026
DA - 2026/08/28
JO - Chemical Engineering and Processing: Process Intensification
DO - 10.1016/j.cep.2026.111012
UR - https://ifp.hal.science/hal-05778720v1
AB - Rotating Packed Beds (RPB) technology has the potential to capture carbon dioxide emissions from industrial flues. A porous packing in rotation provides an extensive surface area for effective mass transfer between a gas and a solvent. The application of centrifugal force enables counter-current flow of the two fluids, enhancing the efficiency of the process. HiGee contactors present a superior alternative to traditional packed columns. In these systems, the solvent is subjected to acceleration up to a thousand times that of gravity, significantly increasing the mass transfer rates by one to two orders of magnitude. Additionally, these systems are remarkably compact, being ten times smaller than their conventional counterparts. Flooding in RPBs represents a significant operational challenge, characterized by the inability of the solvent to evacuate through the packing. This issue arises due to various resistive forces, such as gravitational and frictional forces, which hinder the solvent's exit from the packing. Operational RPBs have been documented using a two-camera setup to study flooding. One camera focuses on the solvent injection point to monitor flooding, while the other captures the droplets exiting the packing. Initial results align with literature. Given the extreme challenges posed by the imaging conditions, deep learning with convolutional neural networks for segmentation is essential to extract meaningful information from these images. This deep learningdriven analysis has been pivotal in characterizing the geometric and dynamic properties of the droplets, revealing a previously unrecognized precursor to flooding in RPB systems.
SN - 0255-2701
ER -