PINN Based Data Driven Magnetics Loss Modeling

Modeling of magnetic material loss is fundamental in magnetic components design, yet no fully satisfactory model has been proposed due to the complexity of underlying physics. Equation based model can not guarantee precise results across the entire application range. Data driven model is expected to surpass the limitations of equation-based method, but current research still faces challenges in terms of model size, accuracy, and the applicability range. Therefore, a physics informed neural network (PINN) model with smaller size, higher accuracy, wide applicability range and reduced data dependency is proposed. A PINN model incorporating improved general Steinmetz equation is established for mitigation of model reliance on extensive datasets. The proposed model is rigorously validated through repeated experiments and evaluations, involving comparisons with PINN and normal NN, as well as other existing models. Multiple performance metrics, such as accuracy, model size, and data size dependency, are considered to provide a comprehensive assessment of the model’s capabilities.