Insider Brief
- Scientists have developed a physics-based machine learning method that automatically identifies the origin of iron loss in soft magnetic materials, offering a path to more efficient electric motors.
- The study combines the extended Ginzburg–Landau framework with interpretable machine learning to analyze heterogeneous magnetic domain structures in materials like nonoriented electrical steel.
- The method reveals that energy loss occurs near grain boundaries due to competing factors in magnetization reversal, enabling precise identification of loss mechanisms previously undetectable through visual inspection alone.
PRESS RELEASE — Magnetic hysteresis loss or iron loss in soft magnetic materials accounts for approximately 30% of energy loss in electric motors. This loss results in significant energy loss globally, representing a pressing environmental concern. However, the origin of iron loss remains elusive despite decades of research. Now, scientists have developed a new physics-based machine-learning approach that automatically identifies the origin of iron loss, establishing a new paradigm for designing efficient soft magnetic materials.
Magnetic hysteresis loss or iron loss is an important magnetic property that determines the efficiency of electric motors and is therefore critical for electric vehicles. It occurs when the magnetic field within the motor core, made up of soft magnetic materials, is repeatedly reversed due to the changing flow of current in the windings. This reversal forces tiny magnetic regions called magnetic domains to repeatedly change their magnetization direction. This change is, however, not perfectly efficient and results in energy loss. In fact, iron loss accounts for approximately 30% of the total energy loss in motors, leading to the emission of carbon dioxide, which represents a pressing environmental concern.
Despite over half a century of research, the origin of iron loss in soft magnetic materials remains elusive. The energy spent during magnetization reversal in these materials depends on complex changes in magnetic domain structures. These have mainly been interpreted visually, and the underlying mechanisms have been discussed only qualitatively. Researchers believe that investigating the correlation between energy loss and the microstructure of magnetic domains is a promising direction. However, most current physical models for analyzing magnetization reversal are designed for homogeneous systems, while practical soft magnetic materials like nonoriented electrical steel (NOES) are heterogeneous, making their analysis difficult.\
Now, in a breakthrough, a research team led by Professor Masato Kotsugi from the Department of Material Science and Technology at Tokyo University of Science (TUS), Japan, along with Mr. Michiki Taniwaki, also from TUS, has developed a new approach utilizing the extended-Ginzburg–Landau (ex-GL) framework. This method successfully traces the origin of iron loss to the magnetic domain structure. Prof. Kotsugi explains, “The Ginzburg–Landau (GL) free energy was a useful concept for analyzing the magnetization reversal in a homogeneous system. Recent progress in data science has enabled the ex-GL model, which can be used to analyze heterogeneous systems. In this study, we paired the ex-GL framework with interpretable machine learning for automated analysis of complex magnetization reversal in NOES.” Their study was published in the journal Scientific Reports on July 15, 2025.
The team first quantified the complexity of magnetic domains from microstructure images of NOES using persistent homology (PH), a mathematical tool for multiscale analysis of topological features in data. Next, they applied principal component analysis (PCA), a statistical technique, to extract the essential features hidden in the complex PH data. Two features emerged, namely PC1, representing magnetization, and PC2, representing magnetic domain walls.
Using these features, the team then constructed an extended energy landscape using the ex-GL framework, which mapped the changes in the magnetic domain structure with energy as a graph with each point corresponding to a magnetic domain image. The team then performed a comprehensive correlation analysis between the features and physical parameters using this graph, uncovering physically meaningful features that explain energy loss during magnetization reversal.
Their analysis revealed the presence of promoting and resisting factors in the magnetization reversal process. Interestingly, both factors were found in the same locations, mainly near grain boundaries, which are interfaces between different crystals in a crystalline material. This suggests competition between these factors. “The competition between the promoting and resisting factors automatically identifies the location of magnetic domain wall pinning, a key phenomenon responsible for energy loss in soft magnetic materials,” notes Prof. Kotsugi. “In locations with only resisting factors, segmented magnetic domains were found to be the main contributors to energy loss.”
The significance of this method lies on the automated, precise, data-driven insights into both the mechanism and location of energy loss.
“Our approach enabled us to extract information that would otherwise have been difficult to obtain with only visual inspection,” remarks Prof. Kotsugi.
This research paves the way in realizing the United Nations sustainable development goals–affordable and clean energy, industrialization, innovation and infrastructure, and combating climate change. In summary, this study presents an innovative data-driven approach for identifying the origin and addressing energy loss in soft magnetic materials, leading to more efficient, greener electric cars, paving the way towards a sustainable future.
