Physics-Informed Neural Network Model for Predicting Flux Linkage Surface in Switched Reluctance Motor

This research presents the development of a physics-informed neural network designed to predict magnetic flux surfaces in switched reluctance motors, aimed at enabling faster calculations without relying on lookup tables, particularly for model predictive control applications. The study models flux linkage as a function of rotor angle and current, leveraging a combination of data-driven learning and physics-based constraints. A key feature of the model is the inclusion of a monotonicity constraint on the flux gradient with respect to current, ensuring adherence to the physical behavior of SRMs for more accurate predictions. 3D surface visualizations confirmed that the simulated flux surfaces closely matched experimental data, validating the model’s robustness. Furthermore, the model enforces the physical constraint of nonnegative flux gradients with respect to current, ensuring reliable predictions in practical switched reluctance motor operations. Additionally, the computation time was assessed for practical implementation, confirming the model’s efficiency for real-time applications like model predictive control.