Reconstruction of electrical impedance tomography is a nonlinear and poorly posed inverse question. As a result of nonlinearity, the calculation cost of a method is high, and the most relevant regularization and observations should be used to minimize the bad position.
Study: Improved electrical impedance tomography for machine learning for 2D materials. Image credit: Peshkova / Shutterstock.com
In an article published in the journal Inverse Problems, a machine learning adaptive electrode selection technique was used to construct and apply a unique approach to measurement improvement. Taken together, this study showed how electrical impedance tomography (EIT) could be used for 2D materials and emphasized the importance of machine learning in both numerical and computational components of electrical impedance tomography. .
What is the EIT?
Electrical impedance tomography (EIT) is a visualization technique that uses a set of four readings along the edge of the sample to reconstruct the conductivity scattering within an object.
Electrical impedance tomography is a non-invasive imaging technology that was developed in geophysics for underground exploration and medical physics to investigate differences in body tissues by measuring conductivity alterations.
Because the inverse issue in the reconstruction of electrical impedance tomography images is poorly posed, important work has been done since its inception to increase the integrity and accuracy of electrical impedance tomography. To date, many methods have been presented, including those using artificial neural networks (ANNs), in an effort to address the inverse problem.
Deep learning and EIT
Recent studies have used in-depth learning to develop and evaluate an ANN on numerically generated data for the two-dimensional (2D) D-bar reconstruction approach. They effectively recreated the conductivities of artificial agar objects and illustrated how neural networks could increase the restoration accuracy of electrical impedance tomography.
Machine learning is important not only for evaluating EIT images, but can also be used to optimize the placement of electrodes around the sample rather than spacing the electrodes at frequent intervals. Several commonly used current patterns are currently available, including the design of the neighboring unit and the pattern of the opposite (polar) unit.
A number of researches have evaluated these patterns or provided a theoretical study of how to optimize the choice of electrodes; Machine-learned electrode selection models can replace the most prevalent computational procedures, and the adjacent pattern is still commonly used throughout the literature, even after proving to be particularly inaccurate.
Use of EIT with graphene
Electrical impedance tomography has recently been used to investigate 2D conductance patterns of thin films and graphene. EIT reconstruction was paired with a conductivity map acquired by time domain spectroscopy (TDS), a low-resolution approach performed under disconnection conditions using fairly expensive equipment in the first use of graphene.
Only a 4% difference was detected between the TDS and EIT maps, indicating the applicability of electrical impedance tomography for the characterization of 2D materials. Although the 2D EIT is frequently explored, as it often includes simpler procedures, it does not reflect usage scenarios in conventional medical applications.
Here the foundations of the EIT enabled for machine learning for use in 2D materials were laid. A unique technique for selecting adaptive machine learning adaptive electrodes was devised and a strategy was established to produce conductance restorations of 2D materials by integrating it with a direct solver complemented with the complete electrode (EMC) model.
EIT measurements were performed in a square sample shape using the python-based program pyEIT. This program originally used only a direct solver, but was updated in this research to include CEM.
Highlights of the study
Given the width of the electrode, the EMC-enhanced direct solver outperformed the basic solution of the initial pyEIT program. More complicated modeling improved the accuracy of the restore, while accelerating the GPU reduced computing time by half.
These features are critical for future applications to 2D materials, where the limited width of connections is increasingly relevant. In addition, the creation of an A-ESA machine learning was beneficial, as it regularly produced reduced reconstructive losses and higher performance than the usual opposite and adjacent adjacent techniques.
The use of U-Net CNN for post-processing reconstruction yielded some encouraging first results, highlighting the value of deep learning, which has been used more and more commonly in several domains, including the EIT.
This study showed the potential application of EIT for the characterization of 2D materials and illustrated how the incorporation of machine learning approaches could significantly improve the experimental and analytical parts of this work.
Future directions
One of the next steps would be to examine samples in a rectangular shape, as the algorithm currently supports this: mesh creation, GREIT pixel images, and the general map matrix can be nx x ny. Future research could analyze various morphologies, such as an ellipse or an erratic shape.
Instead of inserting only electrodes at periodic times, machine learning can be used to optimize their spatial locations around the sample.
One can even imagine a recursive robotic solution that incorporates the selection of adaptive electrodes and the placement of adaptive electrodes in situ, in which a series of data are taken, the electrodes move to more optimized locations and then another set of data is collected at the new contact. stains.
References
Coxson, A., Mihov, I., Wang, Z., Avramov, V., Barnes, FB, and Slizovskiy, S. (2022). Enhanced electrical impedance tomography for machine learning for 2D materials. Reverse problems. Available at: https://doi.org/10.1088/1361-6420/ac7743
Disclaimer: The views expressed herein are those of the author expressed in private and do not necessarily represent the views of AZoM.com Limited T / A AZoNetwork the owner and operator of this website. This disclaimer is part of the Terms and Conditions of Use of this website.