DC Electro-Magneto-Mechanical Characterization of 2G HTS Tapes for Superconducting Cable in Magnet System Using Artificial Neural Networks

Abstract

Characterization of the exact critical current density (Jc) and stress values in twisted superconducting tapes plays an important role for analyzing their magnetic, thermal, and mechanical behaviors. In current study, a model based on artificial neural network (ANN) is introduced to estimate the electro-magneto-mechanical characteristic of different superconducting tapes. For this purpose, magnetic flux density, temperature, strain, total thickness of tape, it's width, thickness of stabilizer, and thickness of substrates are used as inputs to ANN model, whereas minimum normalized Jc and maximum stress are considered as outputs. The required experimental data are extracted from published papers in the literature. The ANN model was trained for Jc/stress estimation using extracted data by using different inputs. Sensitivity analysis was conducted on ANN models, which were used to estimate the Jc and stress values of tapes, to choose an optimum structure for ANN models to be used in future by other researchers in the superconductivity community. To check the reproducibility, repeatability, and stability of presented results, the estimations with ANN optimum structure were tested for 500 testing runs. We found that the ANN optimum structure was as one hidden layer with Levenberg–Marquardt training method and seven inputs. Comparing to the literature, the proposed ANN model offers about 15% and 1.1% higher accuracy in Jc and stress estimations, respectively.