Using deep learning to retrieve 3D geometrical characteristics of a particle field from 2D projected images: Application to multiphase flows

Abstract

The main part of recycling processes are carried out in chemical engineering reactors that involve multiphase flows with dense dispersed phase. In a study and modeling approach of these processes, the description and characterization of hydrodynamic phenomena is crucial. A variety of techniques allows us to realize this type of measurement, but the most used one is the direct imaging associated with an efficient image processing. Recently, deep learning algorithms have proven to be very effective in solving image based problems, which led to the use of these algorithms to extract critical information in chemical engineering apprentices. The method employed in this paper relies on a deep learning based algorithm dedicated to the prediction of 3D features of multiphase flows using 2D projected images. The performance of the method has been evaluated both on synthetic images and on real images of beads in a dispersed phase.

The Role of Stereological Assumptions in Bubble Size Estimations and Their Implications for Assessing Critical Coalescence Concentrations

“…Future work needs to focus on the design of a new bubble analysis apparatus for froth flotation, capable of measuring bubbles in 3D, thus avoiding stereological assumptions. Recent advancements in image analysis based on deep learning and neural networks have shown promise in reconstructing the 3D shape of bubbles using single-view 2D images [39,40]. However, to effectively apply these methodologies, it will be necessary to create extensive training databases of 3D bubble images for different frothers, operating conditions and sampling equipment.…”