Experimental Testing of a Neural Network approach for Dynamic Ferromagnetic Hysteresis

Validations from experimental testing of the capability of Neural Networks (NNs) in modelling dynamic magnetic hysteresis loops are presented. The NNs under test are components of a more complex Neural System (NS). Each NN is trained and used for managing just a specific part of the dynamic hysteresis loop. The whole hysteretic curve is recomposed by connecting the evaluations made by different NNs of the NS. Each NN consist of 3-input 1-output Feed Forward NN and models the hysteresis trajectories related to a sub-portion of the B-H plane. The input of the Neural System is the frequency of the exciting field, the flux density and the magnetic field strength. The NS has been tested on a non-oriented Fe-(3 wt%)Si laminations (thickness 0.35 mm) and results have shown that this approach is particularly able to model both static hysteresis and iron losses into a fixed range of frequencies of the exciting magnetic field