XPINN-based parametric modeling method for solving magnetic fields with moving boundaries

Purpose Moving-boundary magnetic field analysis in multi-material electromagnetic devices is computationally expensive because boundary motion requires finite element (FE) mesh updates, while material nonlinearity requires iterative solutions. An extended physics-informed neural network (XPINN)-based parametric framework was developed for efficient and accurate magnetic field prediction for such devices. Design/methodology/approach The computational domain was divided into material- and topology-based subdomains, with local fields represented by dedicated subnetworks coupled through interface continuity conditions. This architecture was designed to handle discontinuities in material properties and the complex magnetic-field distributions near the interfaces. To avoid repeated sampling due to moving boundaries, a geometry-state-driven dynamic sampling strategy was proposed to reassign sampling points between subdomains and update moving-interface samples and constraints according to the mover position, enabling parametric prediction across different displacements. Findings The predicted magnetic vector potential and magnetic flux density of an electrodynamic shaker agreed well with the high-fidelity FE results, with maximum relative errors of 0.77% and 0.78%, respectively. The framework achieved speedups of 5.14–21.95 times for multi-case predictions. Compared with a standard physics-informed neural network, it reduced the maximum point-wise absolute errors in vector potential and flux density by 96.72% and 99.07%, respectively, while reducing the flux density error by 15.71% relative to a resampled XPINN. Originality/value The framework combined XPINN with the proposed geometry-state-driven sampling strategy, providing a potentially generalizable approach for efficient and accurate electromagnetic-field prediction in systems with heterogeneous media and moving boundaries.

Authors

Institutions

Publication Details

Journal
Engineering Computations
Published
2026-11-16
DOI
https://doi.org/10.1108/ec-03-2026-0527
Primary Topic
Model Reduction and Neural Networks
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

XPINN-based parametric modeling method for solving magnetic fields with moving boundaries

Longlei Dong, Linjie Ma, Jian Liu, Jianping Zhao
2 links
Engineering Computations
Model Reduction and Neural Networks
article

XPINN-based parametric modeling method for solving magnetic fields with moving boundaries

Longlei Dong, Linjie Ma, Jian Liu, Jianping Zhao
article en

Abstract

Purpose Moving-boundary magnetic field analysis in multi-material electromagnetic devices is computationally expensive because boundary motion requires finite element (FE) mesh updates, while material nonlinearity requires iterative solutions. An extended physics-informed neural network (XPINN)-based parametric framework was developed for efficient and accurate magnetic field prediction for such devices. Design/methodology/approach The computational domain was divided into material- and topology-based subdomains, with local fields represented by dedicated subnetworks coupled through interface continuity conditions. This architecture was designed to handle discontinuities in material properties and the complex magnetic-field distributions near the interfaces. To avoid repeated sampling due to moving boundaries, a geometry-state-driven dynamic sampling strategy was proposed to reassign sampling points between subdomains and update moving-interface samples and constraints according to the mover position, enabling parametric prediction across different displacements. Findings The predicted magnetic vector potential and magnetic flux density of an electrodynamic shaker agreed well with the high-fidelity FE results, with maximum relative errors of 0.77% and 0.78%, respectively. The framework achieved speedups of 5.14–21.95 times for multi-case predictions. Compared with a standard physics-informed neural network, it reduced the maximum point-wise absolute errors in vector potential and flux density by 96.72% and 99.07%, respectively, while reducing the flux density error by 15.71% relative to a resampled XPINN. Originality/value The framework combined XPINN with the proposed geometry-state-driven sampling strategy, providing a potentially generalizable approach for efficient and accurate electromagnetic-field prediction in systems with heterogeneous media and moving boundaries.

Engineering Computations
Xi'an Jiaotong University (CN)
Model Reduction and Neural Networks
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon