Convolutional neural networks for the shape design of a magnetic core for material testing: Forward and inverse approaches

Issue title: 16th International Workshop on Optimization and Inverse Problems in Electromagnetism

Guest editors: Marcin Ziółkowski and Marek Ziółkowski

Article type: Research Article

Authors: Di Barba, Paolo^a | Mognaschi, Maria Evelina^a; | Sieni, Elisabetta^b | Ziolkowski, Marcin^c

Affiliations: [a] Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy | [b] Department of Theoretical and Applied Sciences, University of Insubria, Varese, Italy | [c] Department of Theoretical Electrical Engineering and Applied Computer Science, West Pomeranian University of Technology, Szczecin, Poland

Correspondence: [*] Corresponding author: Maria Evelina Mognaschi, Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Via Ferrata 5, 27100 Pavia, Italy. E-mail: [email protected]

Abstract: In this paper CNNs are used for solving an optimization problem with two different approaches: CNN is used as a surrogate model of the forward problem, inserted in an optimization loop governed by a genetic algorithm, in the first approach, while a CNN is trained for solving directly the inverse problem in the second approach. The case study is the shape design of a magnetic core used for material testing.

Keywords: Convolutional neural networks, inverse problems, material testing, magnetic field, finite elements

DOI: 10.3233/JAE-210175

Journal: International Journal of Applied Electromagnetics and Mechanics, vol. 69, no. 3, pp. 389-399, 2022