Abstract:
This study presents a novel end-to-end trainable network named IDM-Net (Inverse Design Network for Magnetic Fields) that facilitates multi-task supported inverse design o...Show MoreMetadata
Abstract:
This study presents a novel end-to-end trainable network named IDM-Net (Inverse Design Network for Magnetic Fields) that facilitates multi-task supported inverse design of magnetic fields. Employing the Encoder-Decoder idea, IDM-Net accomplishes the inverse design of magnet parameters by leveraging magnetic field data for backpropagation. The Encoder is harnessed to extract magnetic field features and classify magnet shapes, while the Decoder predicts other properties, including magnet size and position. This innovative approach breaks the constraints of single magnet types in existing research, enabling the inverse design of properties for magnets with diverse shapes. Our experimental results demonstrate remarkable performance, with a 95.2% accuracy in magnet shape classification and a mere 0.28% error in magnet property prediction. By introducing the Encoder-Decoder idea in the field of inverse design for magnetic fields, we showcase significantly enhanced accuracy and pave the way for broader applications of this technology.
Published in: 2023 IEEE International Conference on Applied Superconductivity and Electromagnetic Devices (ASEMD)
Date of Conference: 27-29 October 2023
Date Added to IEEE Xplore: 09 January 2024
ISBN Information: