BA - Comparative Analysis of Neural Network Architectures for Electrical Impedance Tomography (EIT)

Vergleichende Analyse verschiedener Netzwerkarchitekturen für Machine Learning Basierte EIT Rekonstruktionen

Comparative Analysis of Neural Network Architectures for Electrical Impedance Tomography (EIT

General Setting

Electrical Impedance Tomography (EIT) is a non-invasive imaging technology based where a conductivity profile inside a medium is reconstructed based on voltage measurements on its boundary. At iMEK, within the context of the Collaborative Research Center CRC1615, a 3D EIT system to monitor chemical processes is being developed. The goal of this project is to compare different neural network architectures for Electrical Impedance Tomography (EIT) reconstruction. The student will generate synthetic EIT data, analyze the impact of various design patterns on reconstruction quality, and evaluate how much data is needed to achieve reliable results.

Tasks

1. Framework Design & MLOps Integration

   • Set up a ML framework using GitLab Model Registry with MLflow for experiment tracking, parameter logging, and model versioning.

   • Integrate GitLab CI/CD and model registry for structured model management and reproducibility.

   • Ensure extensibility for future use cases (e.g., different imaging setups or simulation conditions).

2. Literature Review

   • Review existing neural network architectures for EIT, including Fully Connected Networks (FCNs), Convolutional Neural Networks (CNNs), and U-Nets.

   • Study relevant activation functions, loss metrics (e.g., MSE, SSIM, PSNR), and optimization strategies.

3. Synthetic Data Handling

   • Use the existing bubbly flow simulation to generate synthetic EIT datasets.

   • Simulate different conductivity distributions and realistic noise levels.

4. Neural Network Implementation

   • Implement Neural networks for 3D Bubble reconstruction based on the existing 2D implemenations.

   • Train and validate each model using the generated datasets within the developed ML framework.

5. Model Evaluation and Comparison

   • Evaluate reconstruction quality using defined performance metrics (MSE, SSIM, PSNR).

   • Compare the strengths and limitations of each architecture.

6. Documentation & Final Report

   • Document the ML framework and include setup instructions.

   • Compile a scientific report detailing methodology, results, and potential improvements.

Deliverables

• A reproducible ML framework integrated with GitLab and MLflow.

• Trained models for 3D bubbly flow reconstruction.

• Visual and quantitative comparisons of network architectures.

• Bachelor thesis report including background, implementation, results, and future directions.

Gitlab for Running Documentation:

Checklist

• Introduction / tour in M4

• Urheberrechtsvereinbarung signed

• if applicable: signed confidential agreement

• official registration

Helpful links:

• Thesis Work / Studentische Arbeiten 

• Criteria for the evaluation of Bachelor's and Master's theses, handout version

• Laborordnung

• Spotlights / Kurzvorträge

• Templates (Latex, ppt etc.)

Document Upload Final Thesis / Dokumentenabgabe Abschlussdokument

File of final presentation / Dokumentenabgabe Abschlusspräsentation

Link for further files / Link für weitere Dokumente