Mo-SML-04 - Jacobian-Initialized Parametric Physics-Informed Neural Network for Anomaly Position and Size Estimation from Simulated Electrical Impedance Tomography Data
- Event
- EUROSENSORS 2026
2026-09-06 - 2026-09-09
Zurich - Band
- Poster
- Chapter
- Signal Analysis, Machine Learning, Physics-Informed Machine Learning And Artificial Intelligence For Sensing
- Author(s)
- S. Sharma, K. Eckert, Z. Lei - Helmholtz-Zentrum Dresden-Rossendorf,Dresden (Germany)
- Pages
- 694 - 695
- DOI
- 10.5162/eurosensors2026/Mo-SML-04
- ISBN
- 978-3-910600-12-6
- Price
- free
Abstract
Real-time monitoring of the trajectory and size of the particle, droplet or bubble is relevant in many multiphase processes, but imaging techniques often fail in optically inaccessible media. In that case, electrical impedance tomography (EIT) provides a non-invasive alternative. This work presents a simulation-based study of an eight-electrode EIT sensor for monitoring a circular conductivity anomaly. We introduce a preliminary study deploying a parametric physics-informed neural network (PINN) approach that uses the center estimate from conventional Jacobian-based reconstruction to refine the trajectory and estimate the radius of an anomaly from simulated boundary-voltage measurements.