Presentation
(PP02) Inverse Coil Design by Machine Learning-based Optimization
SessionProject Posters Presentation
Event Type
Project Poster
AI/Machine Learning/Deep Learning
Clouds and Distributed Computing
Parallel Algorithms
Scientific Software Development
TimeTuesday, June 26th3:15pm - 3:45pm
LocationBooth N-230
DescriptionInductive power transfer is nowadays a popular and widely used technology, e.g. for charging mobile phones and heating cook ware. In such systems usually coplanar spiral coils are used in order to generate the necessary magnetic fields. However, with regard to energy efficiency and safety it is desirable to start from an optimal magnetic field and derive the necessary coil geometry from that.
In our previous work, we have shown that optimized coil geometries can be obtained using a parametric representation of the coil’s geometry, solving Biot-Savart’s law and optimizing the geometry by using Simulated Annealing. Depending on the number of optimization parameters the calculations take up to a few days on ’standard’ workstations. However, once additional components, like field focusing elements, are added to the magnetic circuit, the calculation of the magnetic field using solely Biot-Savart’s law is no longer valid. However, suitable calculation methods like finite element methods (FEM), which allow to calculate the magnetic field of complex magnetic circuits, tremendously increase the computation time of the magnetic field, thus drastically increasing the time for the whole optimization process.
In order to solve this issue, we propose a method that uses machine learning technologies providing a surrogate model for the complex magnetic circuit. Once trained, the surrogate model can replace a time-consuming FEM simulation, still providing an estimate of the magnetic field. As the calculation of the surrogate model will be several times faster than a full FEM simulation, this will speed up the entire optimization process.
In our previous work, we have shown that optimized coil geometries can be obtained using a parametric representation of the coil’s geometry, solving Biot-Savart’s law and optimizing the geometry by using Simulated Annealing. Depending on the number of optimization parameters the calculations take up to a few days on ’standard’ workstations. However, once additional components, like field focusing elements, are added to the magnetic circuit, the calculation of the magnetic field using solely Biot-Savart’s law is no longer valid. However, suitable calculation methods like finite element methods (FEM), which allow to calculate the magnetic field of complex magnetic circuits, tremendously increase the computation time of the magnetic field, thus drastically increasing the time for the whole optimization process.
In order to solve this issue, we propose a method that uses machine learning technologies providing a surrogate model for the complex magnetic circuit. Once trained, the surrogate model can replace a time-consuming FEM simulation, still providing an estimate of the magnetic field. As the calculation of the surrogate model will be several times faster than a full FEM simulation, this will speed up the entire optimization process.
Poster PDF