Numerical and Experimental Investigation of Particle Size Distributions in Multiphase Flows
- This thesis investigates the numerical and experimental characterization of the particle size
distribution in two different chemical engineering applications (liquid-gas and liquid-liquid).
First, a gas-liquid flow in a pseudo 2D bubble column is investigated using optical imaging
techniques. Objective of the experiment is the localized measurement of the bubble size
distribution, as well as the orientation and movement of the gas bubbles. This is achieved
using an automated camera traverse system, capturing high-speed images of different zones
in the bubble column. The images are then evaluated using an algorithmic approach consid-
ering the bubble size, movement direction and orientation of the gas bubbles. The measured
quantities are evaluated in each zone of the bubble column. The captured experimental data
is then used as a validation for the implementation of the Sectional Quadratic Method of
Moments (SQMoM), as a solution for the population balance equation, in the computational
fluid dynamics (CFD) solver multiphaseEulerFoam of the open-source software OpenFOAM.
The simulation is specifically compared to the zone-wise data of the bubble column experi-
ments. The adapted solver is able to simulate the development of the bubble size distribution
in the different zones of the bubble column.
The second investigation considers a liquid-liquid flow in a mixer-settler with the objective
to analyze knitted meshes as coalescing aids. The experiment is executed using an automated
mixer-settler set-up. To control the droplet size distribution, leaving the mixer and entering
the settler, a control loop is developed to set the mixer speed. The setup uses a convolutional
neural network (CNN) to segment droplet clusters, in the live captured camera images. The
CNN is trained in a two-step approach, which uses an algorithmic method for trainings
database generation, to segment a representative image, in the first step. In a second step,
the previously trained network is used to create a database of images to train a second network for the live evaluation. This allows a training without the need of for a large already
evaluated database. An adaption of the presented method and an alternative method using
a generative adversarial neural network (GAN) for trainings database generation is tested
considering gas-liquid multiphase flows.
The automated mixer-settler allows for an investigation of the knitted meshes as coalescing
aids under a controlled load, while the total flow rate, or the phase ratio, or the droplet
size distribution is gradually changed. The results of the experiments are used to adapt
an iterative design approach for settlers to implement the usage of knitted meshes as
coalescencing aids.