PA - Simulation of Electrical Impedance Measurements in Bubbly Flows with Machine Learning-Based Bubble Size Distribution Retrieval

1. Introduction

Bubbly flows play a crucial role in various industrial and environmental processes, including chemical reactors, bioreactors, and aeration systems. The accurate prediction of bubble size distributions is essential for optimizing mass transfer and reaction rates. However, direct measurement techniques often suffer from limitations in accuracy and resolution. Electrical Impedance Tomography (EIT) offers a promising alternative for non-invasive measurement of bubbly flows.

This thesis aims to simulate impedance-based measurements in bubbly flows and leverage machine learning methods to extract bubble size distributions from these simulations. The study will be based on trajectory-based breakup models for dense bubbly flows, such as those described in the Trajectory-Based Breakup Model (TBBM), to generate realistic bubble distributions.

2. Objectives

The key objectives of this thesis are:

• Simulation of Bubbly Flows: Implement and validate numerical models for bubbly flows, incorporating bubble breakup and coalescence mechanisms.

• Electrical Impedance Simulations: Model and simulate impedance-based measurements of these flows using EIT.

• Machine Learning-Based Reconstruction: Develop and train machine learning models to infer bubble size distributions from simulated impedance data.

• Validation: Compare the extracted distributions with ground truth data from the numerical simulations.

3. Methodology

3.1 Bubbly Flow Simulation

• Generate datasets of bubble size distributions under different flow conditions using a provided implementation.

3.2 Electrical Impedance Measurement Simulation

• Model the electrical properties of bubbly flows, considering the conductivity contrast between gas and liquid phases.

• Simulate impedance measurements using Finite Element Method (FEM).

3.3 Machine Learning-Based Reconstruction

• Develop feature extraction techniques from impedance data.

• Train machine learning models to map impedance data to bubble size distributions.

• Optimize models using synthetic and experimental datasets.

3.4 Validation and Analysis

• Compare reconstructed bubble size distributions with simulated ground truth.

• Evaluate model accuracy using statistical metrics such as Mean Squared Error (MSE) and correlation coefficients.

4. Expected Outcomes

• A validated simulation framework for electrical impedance measurements in bubbly flows.

• A trained machine learning model capable of reconstructing bubble size distributions from impedance data.

• Insights into the feasibility of impedance-based techniques for bubbly flow analysis.

6. Conclusion

This thesis will bridge the gap between multiphase flow modeling, electrical impedance measurements, and data-driven reconstruction techniques. By combining physics-based simulations with machine learning, the study aims to provide a novel methodology for characterizing bubbly flows in complex industrial settings.