Current Tomography
The spatial and temporal distribution of the electrical conductivity inside a current-conducting reaction vessel acts as key feature of the process state. Through the electric current density, the electrical conductivity is linked to the magnetic flux density, which can be measured precisely by sensors outside the reactor. The current tomography as a measurement method for the magnetic field-based reconstruction of the electric conductivity in conducting liquids enables efficient process control for water electrolysis, pyrolysis and further industrial applications.
Fig. 1: Schematic representation of the water electrolysis as use case for the current tomography.
Functional principle
The current density distribution j inside a current-conducting reactor is depending on electric conductivity σ within the vessel (Fig. 2: → Ohm´s law). The electric currents are generating a weak magnetic field (→ Biot Savart law, forward problem) that can be measured outside of the reaction vessel with magnetic field sensors. By solving the associated invers problem based on these measurements, the inner conductivity distribution of the reactor can be reconstructed (→ inverse problem).
The method consists of two steps:
- (Contactless) measurement of the distribution of the magnetic field outside the vessel
- Reconstruction of the distribution of the current density or the conductivity inside the vessel by solving the invers problem.
Fig. 2: Relation of the distribution of electric conductivity, current density and magnetic flux density.
Proof-of-concept
The applicability of the measurement method was investigated in a coupled numerical and experimental study. The magnetic field sensors are located below a nearly two-dimensional conductor (Fig. 3). The distribution of the current density distribution in highly conductive media (GaInSn, Cu) is influenced by therein inserted electrically non-conducting cylinders, modeling gas bubbles in a water electrolyzer.
Fig. 3: Proof-of-concept setup.
For solving the invers problem, an Invertible Neural Network (INN) was trained based on a numerically generated dataset containing distributions of conductivity, current density and magnetic flux density. Moreover, additional models not included in the training data were simulated and replicated for experimental validation. Figure 4 shows the simulated distribution of the current density and the corresponding relative conductivity distribution, reconstructed by the trained INN based on experimental measurements.
Fig. 4: left: Simulated current density distribution with exemplary gas bubbles / right: corresponding INN prediction of the relative conductivity distribution.
- Krause, L.; Kumar, N.; Wondrak, T.; Eckert, S.; Eckert, K.; Gumhold, S.
Current Tomography - Localization of void fractions in conducting liquids by measuring the induced magnetic flux density
Proceedings of the 11th World Congress on Industrial Process Tomography, 23-28, Mexico City, Mexico (2023) - Kumar, N.; Krause, L.; Wondrak, T.; Gumhold, S.; Eckert, S.; Eckert, K.
Learning to reconstruct the bubble distribution with conductivity maps using Invertible Neural Networks and Error Diffusion
Proceedings of the 11th World Congress on Industrial Process Tomography, 100-105, Mexico City, Mexico (2023) - Kumar, N.; Krause, L.; Wondrak, T.; Eckert, S.; Eckert, K.; Gumhold, S.
Robust Reconstruction of the Void Fraction from Noisy Magnetic Flux Density Using Invertible Neural Networks
Sensors 2024, 24 (4), 1213 - Krause, L.; Kumar, N.; Gumhold, S; Eckert, K.; Eckert, S.; Wondrak, T.
A new sensor concept to localize non-conducting void fractions in liquid metal based on the measured magnetic field
Proceedings of the 17th Dresdner Sensor-Symposium, 23-26, Dresden, Germany (2025)