Paper summary · Inverse Problems · 2022

Conductivity maps for 2D materials

Using measurements around the edge of a sample to estimate how conductivity varies inside it.

The question

Electrical impedance tomography uses boundary measurements to estimate the distribution of conductivity inside a sample. For thin materials such as graphene, which choices of electrodes produce useful measurements?

The approach

We used machine learning to select electrode configurations adaptively and studied ways to turn the resulting voltage measurements into conductivity maps.

What we found

We evaluated electrical impedance tomography on a graphene laminate film. The work shows how machine learning can help with two parts of the problem: collecting measurements and reconstructing the image.

Paper

Machine learning enhanced electrical impedance tomography for 2D materials
Adam Coxson and colleagues

Three-stage EIT schematic: 20-contact square graphene CEM forward model, ML Adaptive Electrode Selection (A-ESA), and GREIT conductivity reconstruction
20-contact CEM forward solver, ML Adaptive Electrode Selection (11–41% lower RMSE), and GREIT reconstruction (the paper also adds U-Net post-processing).

Watch a 2D conductivity map emerge measurement by measurement

Smooth cubic interpolation between sequential pyEIT + GREIT reconstructions lets you watch the conductivity map crystallize across 10 → 124 four-terminal measurements—comparing fixed opposite–adjacent scanning against ML Adaptive Electrode Selection (A-ESA) across three synthetic graphene phantoms.

60-fps Interactive Authentic pyEIT + A-ESA Graphene Conductivity Lab Cubic-interpolated · Scrub timeline · Toggle FEM / GT contours