Robust Reconstruction of the Void Fraction from Noisy Magnetic Flux Density Using Invertible Neural Networks.

Area of Science:

• Electrochemistry
• Applied Physics
• Computational Science

Background:

• Electrolysis is key for sustainable hydrogen production, but gas bubbles hinder efficiency and increase energy use.
• Detecting these bubbles is difficult due to opaque electrolysis cell walls.
• Gas bubbles alter electrolyte conductivity, causing measurable magnetic flux density fluctuations.

Purpose of the Study:

• To develop and compare methods for detecting gas bubbles in electrolysis cells.
• To estimate electrolyte conductivity and void fraction using magnetic field measurements.
• To evaluate the robustness of Invertible Neural Networks (INNs) against Tikhonov regularization for this inverse problem.

Main Methods:

• Solving the inverse problem of the Biot-Savart Law to relate magnetic flux density to internal cell properties.
• Implementing and comparing Invertible Neural Networks (INNs) and Tikhonov regularization techniques.
• Analyzing the impact of varying noise levels on the performance of each method.

Main Results:

• Invertible Neural Networks (INNs) demonstrate superior robustness in solving the inverse problem compared to Tikhonov regularization.
• INNs are more effective when dealing with unknown or dynamic noise in magnetic flux density measurements.
• The study validates the potential of using external magnetic sensors for non-invasive bubble detection.

Conclusions:

• INNs provide a more reliable approach for monitoring and optimizing electrolysis processes by accurately detecting gas bubbles.
• This magnetic-based sensing method, enhanced by INNs, offers a promising avenue for improving hydrogen production efficiency.
• Further research can explore advanced INN architectures for real-time monitoring in industrial electrolysis systems.