THP043 　ポスター②　 8月1日 2Fリハーサル室 13:00-15:00

DFSNet for Eigenmode Simulation of RF Cavity

○Ahsani Hafizhu Shali, Mitsuhiro Fukuda, Tetsuhiko Yorita, Hiroki Kanda, Tsun Him Chong, Hang Zhao, Shotaro Matsui, Kaoru Watanabe, Tomoki Imura, Nami Itakura, Sho Ishihata (RCNP Osaka U.)

Optimization of the quantity of interest of an RF cavity, such as the frequency and the quality factor, requires calculation from many samples with different geometric configuration with each of them requiring FEM simulation. Geometric modification of smaller components requires finer mesh thus a very high computational cost requirement. In this research, we propose a physics informed neural network model to solve eigenvalue problems for the lowest frequency mode of a fully three dimensional RF cavity. Three loss functions are used, the first one is related to the differential equation (LDE), the second one is related to the boundary condition (LBC), while the last one is used to make sure that the generated solution is not trivial (LReg). The input for the neural networks are the guess of the eigenvalue and a sampled spatial position, while the output is the value of the RF magnetic field at the corresponding spatial position. The sampled positions are chosen to be related to the nodes of a generated unstructured mesh of the cavity model. The result shows that the neural network model could give an accurate prediction of the frequency and the field for the lowest mode.