Accounting for Hysteresis and Eddy Currents in Finite Element Simulations of Ferromagnetic Laminated Cores using a Recurrent Neural Network

2026-07-15 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

A recurrent neural network (RNN) surrogate model has been developed to efficiently account for hysteresis and eddy currents in finite element simulations of ferromagnetic laminated cores. Traditional methods for incorporating these complex electromagnetic behaviors are computationally intensive, increasing costs by several orders of magnitude compared to anhysteretic simulations, making them impractical for design. This new approach trains an RNN as a surrogate for an isotropic laminated core material model, integrating it into realistic two-dimensional magnetodynamic finite element simulations using a magnetic vector potential formulation. The proposed method achieves excellent agreement with the reference laminated-core model while limiting computational cost to approximately twice that of an anhysteretic simulation. Trained on diverse magnetic field sequences, the surrogate model is broadly applicable across various finite element simulations and is publicly available at https://gitlab.onelab.info/getdp/lamnet.

Key takeaway

For Machine Learning Engineers or Research Scientists developing electrical machine simulations, if you are struggling with the computational cost of incorporating hysteresis and eddy currents, consider adopting recurrent neural network surrogate models. This approach significantly reduces simulation time, achieving excellent accuracy at approximately twice the cost of anhysteretic simulations, making complex electromagnetic modeling practical for design. You should explore integrating the publicly available lamnet model into your existing finite element frameworks to enhance efficiency.

Key insights

A recurrent neural network can efficiently surrogate complex hysteresis and eddy current effects in finite element simulations.

Principles

• Machine learning surrogates offer efficient, accurate approximations for complex electromagnetics.
• Diverse training data ensures broad applicability of surrogate models.

Method

Train a recurrent neural network as a laminated-core material model surrogate, then integrate it into 2D magnetodynamic finite element simulations using a magnetic vector potential formulation.

In practice

• Employ RNNs to model hysteresis and eddy currents in FEA.
• Integrate the lamnet surrogate model into existing simulation frameworks.

Topics

• Recurrent Neural Networks
• Finite Element Analysis
• Electromagnetic Simulation
• Hysteresis Modeling
• Eddy Currents
• Laminated Cores

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• A Physics-Informed Neural Network Framework for Elastodynamic Wave Propagation in Bimaterial Systems — PINNs combined with explicit FEA accurately model elastodynamic wave propagation in bimaterial systems.
• Label-Free Finite-Volume-Residual Training of Attention Graph Neural Networks for Coupled Thermo-Fluid Fields — Training GNNs with FVM residuals eliminates labeled data needs for thermo-fluid field prediction, reducing costs.
• Mesh Graph Neural Network Framework for Accelerating Finite Element Simulation for Arbitrary Geometries — Mesh Graph Networks effectively generalize finite element analysis predictions across arbitrary structural geometries by encoding relative features.
• Cycle-Consistent and Uncertainty-Aware Neural Surrogates for Tokamak Edge Plasmas — Cycle-consistent neural surrogates with uncertainty awareness significantly accelerate tokamak edge plasma simulations.
• Engineering Hybrid Physics-Informed Neural Networks for Next-Generation Electricity Systems: A State-of-the-Art Review — Physics-informed machine learning integrates domain physics into models, enhancing accuracy and efficiency for data-scarce electricity systems.
• Deep Learning-based Surrogate Modelling of the LOD Method for Multiscale Problems — LOD-MSNO combines the accuracy of Localized Orthogonal Decomposition with the efficiency of neural operators for multiscale problems.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.