Physics-Informed Conditional Generative Adversarial Network for Inverse Electromagnetic Problems

In the quest to enhance the accuracy and efficiency of conventional sizing systems, this paper introduces an innovative machine learning-based approach leveraging Physics-Informed Conditional Generative Adversarial Networks (PIC-GANs). By incorporating simple and cost-effective physical constraints, a magnetic equivalent circuit (MEC), into the learning process, our model not only adheres to realistic data generation but also aligns with underlying physical principles, offering a more robust and versatile solution, and tailored to specific conditions with lower computational expenses than traditional methods. This approach is adopted for use in the preliminary design stages of a linear actuator in magnetostatics.