Abstract:Hysteresis modeling is crucial to comprehend the behavior of magnetic devices, facilitating optimal designs. Hitherto, deep learning-based methods employed to model hysteresis, face challenges in generalizing to novel input magnetic fields. This paper addresses the generalization challenge by proposing neural operators for modeling constitutive laws that exhibit magnetic hysteresis by learning a mapping between magnetic fields. In particular, two prominent neural operators -- deep operator network and Fourier neural operator -- are employed to predict novel first-order reversal curves and minor loops, where novel means they are not used to train the model. In addition, a rate-independent Fourier neural operator is proposed to predict material responses at sampling rates different from those used during training to incorporate the rate-independent characteristics of magnetic hysteresis. The presented numerical experiments demonstrate that neural operators efficiently model magnetic hysteresis, outperforming the traditional neural recurrent methods on various metrics and generalizing to novel magnetic fields. The findings emphasize the advantages of using neural operators for modeling hysteresis under varying magnetic conditions, underscoring their importance in characterizing magnetic material based devices.

What problem does this paper attempt to address?

This paper focuses on modeling the hysteresis phenomenon, which is key to understanding the behavior of magnetic devices and crucial for optimizing design. The generalization ability of hysteresis models is a challenge when predicting unseen input magnetic fields using deep learning-based methods. To address this issue, this paper proposes the use of neural operators to model the constitutive law exhibiting hysteresis by learning the mapping relationship between magnetic fields. Specifically, two neural operators - DeepONet and Fourier Neural Operator (FNO) - are employed to predict new first-order reversal curves and minor loops. Additionally, to account for the rate-independent nature of hysteresis, a rate-independent Fourier Neural Operator (RIFNO) is proposed to handle cases with different sampling rates in training and testing data. In this study, neural operators demonstrate efficiency in modeling hysteresis, outperforming traditional recurrent neural network methods and showing superior performance in multiple metrics, while also generalizing to new magnetic fields. These findings highlight the advantages of neural operators in modeling hysteresis under different magnetic field conditions and the importance of characterizing devices based on magnetic materials. The main contributions of this paper are as follows: 1. Proposing a shift from learning functional approximations from observed magnetic field data towards learning operators between magnetic fields for achieving generalization, which is essential for neural hysteresis models. 2. Demonstrating the efficiency of neural operators in hysteresis modeling through experimental comparisons, highlighting the limitations of traditional recurrent neural architectures. 3. Introducing the rate-independent Fourier Neural Operator to handle varying sampling rates and verifying its performance. 4. Bridging the gap between hysteresis modeling and scientific machine learning (SciML) to lay the foundation for future applications and advancements of advanced neural operator techniques in hysteresis modeling. The experimental results demonstrate that neural operators have better accuracy and generalization ability in predicting hysteresis phenomena, especially when dealing with unseen magnetic field excitations.

Magnetic Hysteresis Modeling with Neural Operators

A Differential Model for the Hysteresis in Magnetic Shape Memory Alloys and Its Application of Feedback Linearization

Neural oscillators for magnetic hysteresis modeling

Neural Network Architectures and Magnetic Hysteresis: Overview and Comparisons

A Differential Algebraic Approach for the Modeling of Polycrystalline Ferromagnetic Hysteresis with Minor Loops and Frequency Dependence

Neural Network Modeling of Complex Hysteresis Loops in Ferromagnetic Materials

Image-based prediction and optimization of hysteresis properties of nanocrystalline permanent magnets using deep learning

PINN Based Data Driven Magnetics Loss Modeling

Review on Hysteretic Modeling Methods of Giant Magnetostrictive Materials

How MagNet: Machine Learning Framework for Modeling Power Magnetic Material Characteristics

Chaotic Neural Network-Based Hysteresis Modeling with Dynamic Operator for Magnetic Shape Memory Alloy Actuator

A hybrid Jiles–Atherton and Preisach model of dynamic magnetic hysteresis based on backpropagation neural networks

Description of collective magnetization processes with machine learning models

Hysteresis Modeling of Magnetic Shape Memory Alloy using a NARMAX Structure Model

Hysteresis Modeling of Magnetic Shape Memory Alloy Actuator Based on Volterra Series

Machine-learned models for magnetic materials

Reconsideration of Nonphysical Solution in Jiles–Atherton Model for Real-Time Hysteresis Estimation

A fast hysteresis analysis based on neural network and play model

Temperature-dependent hysteresis model based on temporal convolutional network

Universal Hysteresis Identification Using Extended Preisach Neural Network

Using Neural Networks to Model Hysteretic Kinematics in Tendon-Actuated Continuum Robots