A Study on Electromagnetic Field Analysis Considering Geometry Variation Using Physics-Informed Neural Network

The problem with finite element analysis is that each analysis is run separately, so any variation in geometry requires a new analysis. When performing finite element analysis for generating fault diagnosis data, the frequent geometry variations that occur to consider numerous fault levels lead to a...

Full description

Saved in:
Bibliographic Details
Main Authors: Han, Ji-Hoon, Choi, Eui-Jin, Hong, Sun-Ki
Format: Conference Proceeding
Language:English
Subjects:
Deep learning
Fault diagnosis
FEA
Finite element analysis
Geometry
Motor fault diagnosis
Neural networks
Nonlinear magnetics
PINN
Training
Transfer learning
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The problem with finite element analysis is that each analysis is run separately, so any variation in geometry requires a new analysis. When performing finite element analysis for generating fault diagnosis data, the frequent geometry variations that occur to consider numerous fault levels lead to a long time. In this paper, to solve these problems, an analysis using a physics-informed neural network is proposed to solve the differential equation that was solved by a numerical method using a neural network. Transfer learning is used, which enables fast analysis based on the analysis experience before the geometry is varied. In addition, nonlinear magnetic material characteristics and electronic systems in the saturated region are analyzed to evaluate whether the physics-informed neural network can cope with some numerical analysis problem.
ISSN:2642-5513
DOI:10.1109/ICEMS59686.2023.10344784