Every time an electric vehicle accelerates, the steel sheets stacked inside its motor flip their magnetic orientation millions of times per second. Each flip is supposed to be clean and efficient, but it never is. Tiny, tangled magnetic patterns form inside the steel, writhing into maze-like shapes that resist the very fields driving them. That resistance bleeds energy as heat, quietly sapping the battery with every revolution of the rotor.
Engineers have known about this phenomenon, called iron loss, since the 1890s. What they have never been able to do is watch it happen at the microscopic level and say, with precision, which specific magnetic structures are responsible for the most waste. A research team at Tokyo University of Science, led by Prof. Masato Kotsugi, has now built an AI-driven framework that does exactly that. Their work, published in Scientific Reports in early 2026, adds a temperature-dependent entropy term to a classical physics model and pairs it with topological data analysis and machine learning to decompose magnetic domain images into individual energy-loss contributions.
The result is essentially a diagnostic X-ray for the invisible inefficiencies hiding inside the metal heart of an electric motor.
Why iron loss matters more than most drivers realize
In a battery-electric vehicle, the traction motor converts stored electrochemical energy into torque. That conversion is impressively efficient, often above 90%, but the remaining losses add up fast when you are drawing from a finite battery. Published estimates from the U.S. Department of Energy and Oak Ridge National Laboratory put total motor-related losses at roughly 5 to 10 percent of a vehicle’s total energy consumption. Those losses combine resistive heating in copper windings with magnetic losses in the steel core. Iron loss, the portion that originates in the laminated steel sheets forming the motor’s stator and rotor cores, is a significant subset of that figure and one of the hardest components to reduce because it arises from the material’s internal magnetic behavior rather than from any single design flaw.
The steel in question is called nonoriented electrical steel, a silicon-alloyed grade stamped into thin laminations and stacked to form the motor core. It is the standard lamination material in many traction motors, though some high-performance designs use thinner gauges or higher-silicon grades to further suppress losses. The steel’s magnetic domains, microscopic regions where atomic magnetic moments align in a common direction, must reorient rapidly as the motor’s electromagnetic field rotates. When those domains form complex, maze-like patterns instead of smooth, easily switchable configurations, extra energy is consumed rearranging them. That energy becomes waste heat.
What Kotsugi’s team actually built
The framework rests on a well-established piece of physics: the Landau free-energy model, which describes how a magnetic material’s energy depends on its domain structure. Kotsugi’s group extended that model by incorporating an entropy feature that captures how domain complexity changes with temperature. They then applied persistent homology, a branch of mathematics that quantifies the “shape” of data, to convert raw magnetic domain images into topological descriptors. Those descriptors were fed through principal component analysis and machine learning algorithms to produce a full free-energy decomposition of each image.
In practical terms, the system takes a microscopy snapshot of the steel’s magnetic domains and returns a map showing which regions are contributing the most to energy loss, and through which physical mechanism. The approach is designed to be explainable: every learned feature traces back to a specific term in the thermodynamic model, so engineers can see the physics behind each prediction rather than trusting a black-box output.
Prof. Kotsugi has described the goal as connecting domain structure to the free-energy landscape so that engineers can pinpoint where energy is wasted, according to a Tokyo University of Science media release distributed through EurekAlert.
A research line, not a single paper
The 2026 entropy-and-maze study is the latest step in a documented research program. An earlier paper by the same group formalized the “feature-extended Landau free-energy” concept and demonstrated causal analysis of magnetization reversal in soft magnetic materials. A follow-up study applied the explainable-AI framework specifically to nonoriented electrical steel, establishing the method for linking microscopic domain images to macroscopic hysteresis loss, the primary component of iron loss in motor cores. A separate investigation published in NPG Asia Materials addressed a different loss channel: energy dissipation caused by magnetostriction in soft magnets, the tiny mechanical vibrations that occur when a magnetic material changes shape under a field.
Together, the three Springer Nature papers confirm that the group has identified multiple physically distinct routes through which magnetic domain behavior converts useful electrical energy into waste heat. Funding for the work came from JSPS KAKENHI and JST-CREST, two major Japanese government research programs.
The gap between diagnosis and cure
Identifying where energy disappears is not the same as stopping it. The framework is a diagnostic tool, and no published data from the group quantify the total percentage of iron loss that could be eliminated in a commercial EV traction motor by applying its insights. The papers demonstrate the ability to trace loss origins at the domain scale, but the distance between that capability and a redesigned production steel alloy is considerable.
Turning a laboratory insight into a commercially viable steel grade typically requires iterative alloy design, large-scale processing trials, and durability testing under realistic motor operating conditions. None of those steps are reported in the current literature. No motor manufacturer or vehicle OEM has publicly commented on integration feasibility, and no real-world temperature-cycle test results linking the model’s predictions to measured hysteresis loops in finished motor cores have appeared.
The chronology underscores the point. The foundational feature-extended Landau framework dates to 2022, and the nonoriented electrical steel application followed in 2025. Each study advances the method, but none yet reports a closed loop from AI-identified domain defect to redesigned steel composition to validated motor performance gain.
What this means for EV efficiency
The institutional framing from Tokyo University of Science connects the research to “ultra-efficient soft magnets” for next-generation electric vehicles, and the connection is physically sound. If engineers could suppress the maze-like domain patterns that the framework identifies as energetically costly, the payoff would be real: less waste heat, more usable range from the same battery pack, and potentially lighter or cheaper motor designs.
But the size of that payoff remains an open question. Whether the AI workflow’s domain-level insights can translate into a measurable range extension for any specific vehicle platform has not been demonstrated. For now, the most accurate way to describe the work is as a powerful analytical lens, one that lets researchers see, for the first time, exactly which microscopic magnetic structures are stealing energy inside the steel core of an electric motor.
Where the watts go next: from diagnostic lens to motor redesign
The century-old problem of iron loss has not been solved. But for the first time, engineers have a tool that can point to a specific tangle of magnetic domains in a specific region of a steel sheet and say: that is where your watts are going. What they do with that information will determine whether this remains an elegant piece of physics or becomes the foundation for the next generation of more efficient electric vehicles.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
