A method for the optimal design of special transformers is proposed; it is based on machine learning models, which, in turn, are informed by a sequence of magnetic field analyses. The optimal design of a leakage reactance transformer is considered as the case study. The results show that surrogate models amenable to artificial neural networks (ANNs) are able to approximate the dependence of leakage reactance on winding geometry, eventually reducing the computational burden of automated optimal design problems for this class of transformers. Moreover, the deep learning approach based on a Convolutional neural network (CNN) proved to be able to approximate the field distribution in a given region of the domain, knowing the image of the cross-section of the primary winding.
A Machine-Learning Inspired Field-Based Method for the Optimal Magnetic Design of Leakage Reactance Transformers
Di Barba P.;Mognaschi M. E.
;
2025-01-01
Abstract
A method for the optimal design of special transformers is proposed; it is based on machine learning models, which, in turn, are informed by a sequence of magnetic field analyses. The optimal design of a leakage reactance transformer is considered as the case study. The results show that surrogate models amenable to artificial neural networks (ANNs) are able to approximate the dependence of leakage reactance on winding geometry, eventually reducing the computational burden of automated optimal design problems for this class of transformers. Moreover, the deep learning approach based on a Convolutional neural network (CNN) proved to be able to approximate the field distribution in a given region of the domain, knowing the image of the cross-section of the primary winding.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
