Artificial neural networks for AC losses prediction in superconducting round filaments

作者：Leclerc J^*; Hell L Makong; Lorin C; Masson P J

来源：Superconductor Science and Technology, 2016, 29(6): 065008.

DOI：10.1088/0953-2048/29/6/065008

• superconductor
• AC losses
• artificial neural network
• generator
• motor
• model
• numerical computation

An extensive and fast method to estimate superconducting AC losses within a superconducting round filament carrying an AC current and subjected to an elliptical magnetic field (both rotating and oscillating) is presented. Elliptical fields are present in rotating machine stators and being able to accurately predict AC losses in fully superconducting machines is paramount to generating realistic machine designs. The proposed method relies on an analytical scaling law (ASL) combined with two artificial neural network (ANN) estimators taking 9 input parameters representing the superconductor, external field and transport current characteristics. The ANNs are trained with data generated by finite element (FE) computations with a commercial software (FlexPDE) based on the widely accepted H-formulation. After completion, the model is validated through comparison with additional randomly chosen data points and compared for simple field configurations to other predictive models. The loss estimation discrepancy is about 3% on average compared to the FEA analysis. The main advantages of the model compared to FE simulations is the fast computation time (few milliseconds) which allows it to be used in iterated design processes of fully superconducting machines. In addition, the proposed model provides a higher level of fidelity than the scaling laws existing in literature usually only considering pure AC field.

• 出版日期2016-6
• 单位中国地震局