Gradient Networks for Universal Magnetic Modeling of Synchronous Machines
This provides a universal and efficient modeling approach for synchronous machines, enabling robust control applications, though it is incremental as it builds on existing physics-informed neural network methods.
The paper tackles the problem of dynamic modeling of saturable synchronous machines with spatial harmonics by introducing a physics-informed neural network that incorporates gradient networks into fundamental machine equations, resulting in accurate and physically consistent models validated on a 5.6-kW PM synchronous reluctance machine with limited training data.
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) synchronous reluctance machine. Results demonstrate accurate and physically consistent models, even with limited training data.