Machine Learning-Based Estimation of Ripple Losses in High Temperature Superconductors for Cryo-Electric Aviation Applications

Losses generated by high temperature superconducting (HTS) tapes are a source of heat load in the powertrain of superconducting driven electric aircraft. This study employs machine learning (ML) techniques to estimate ripple losses of HTS tapes, which occur on the DC side of the powertrain and place notable demands on the cryogenic cooling system. Ripple losses generated by 6, 12, and 24 pole (pulse) rectifications were investigated under switching frequencies of typical AC/DC converters in electric aircraft. Fast, data driven estimation of ripple losses in the HTS tapes are introduced to overcome the high computational cost of FEM-based methods. Four ML techniques, Gaussian process regression (GPR), decision tree (DT), ensemble tree (ENS), and artificial neural network (ANN), are assessed to demonstrate the effectiveness of the best method. As current ripple may propagate through the full drivetrain, the choice of ML technique is a technical requirement, and rapid, tape level loss prediction is a primary step for condition monitoring and health assessment of HTS devices in cryo-electric aircraft. Training data were generated using a validated 2D H-formulation finite element method (FEM)-based model developed in COMSOL Multiphysics, which included all metallic sublayers to account for their high frequency effects. Results showed GPR achieves the highest accuracy, with a mean relative error of 1.8% and goodness of fit percentage value of 99.997%. The model is then updated to a two tape stack model to show the generalizability of the proposed GPR model, also showing high accuracy results. ML techniques reduced computation time dramatically compared with FEM simulation, with test times between 51 ms and 139 ms for the full data range, compared to 10–30 minutes for only the frequency range data.