![[2602.14947] Gradient Networks for Universal Magnetic Modeling of Synchronous Machines](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14947] Gradient Networks for Universal Magnetic Modeling of Synchronous Machines
Summary
This paper introduces a physics-informed neural network approach for modeling saturable synchronous machines, enhancing dynamic modeling accuracy and efficiency.
Why It Matters
The research addresses the limitations of traditional modeling techniques in electrical engineering by providing a more accurate and data-efficient method for simulating synchronous machines. This has implications for control applications and the design of electrical systems, potentially leading to improved performance and reliability.
Key Takeaways
- Introduces a novel architecture for modeling synchronous machines using gradient networks.
- Demonstrates improved accuracy and data efficiency compared to traditional methods.
- Ensures energy balance and physical consistency in modeling.
- Validates the approach with real-world datasets, showcasing its practical applicability.
- Offers advantages in model inversion and optimal trajectory generation for control applications.
Electrical Engineering and Systems Science > Systems and Control arXiv:2602.14947 (eess) [Submitted on 16 Feb 2026] Title:Gradient Networks for Universal Magnetic Modeling of Synchronous Machines Authors:Junyi Li, Tim Foissner, Floran Martin, Antti Piippo, Marko Hinkkanen View a PDF of the paper titled Gradient Networks for Universal Magnetic Modeling of Synchronous Machines, by Junyi Li and 4 other authors View PDF HTML (experimental) Abstract:This paper presents a physics-informed neural network approach for dynamic modeling of saturable synchronous machines, including cases with spatial harmonics. We introduce an architecture that incorporates gradient networks directly into the fundamental machine equations, enabling accurate modeling of the nonlinear and coupled electromagnetic constitutive relationship. By learning the gradient of the magnetic field energy, the model inherently satisfies energy balance (reciprocity conditions). The proposed architecture can universally approximate any physically feasible magnetic behavior and offers several advantages over lookup tables and standard machine learning models: it requires less training data, ensures monotonicity and reliable extrapolation, and produces smooth outputs. These properties further enable robust model inversion and optimal trajectory generation, often needed in control applications. We validate the proposed approach using measured and finite-element method (FEM) datasets from a 5.6-kW permanent-magnet (PM) sy...
