Detection of Series Faults in High-Temperature Superconducting DC Power Cables Using Machine Learning

A novel method to detect series faults in multi-strand high-temperature superconducting (HTS) power cables, based on monitoring the transmission characteristics as an indirect way to detect the magnetic signature, is reported. The efficacy of the non-destructive detection method was studied using finite element analysis (FEA) and measurements on a model cable with a varying number of disconnected strands. An S-parameter model block was implemented, and the changes in the magnetic signature resulting from the failed superconducting strands in the cable were investigated. The transmission characteristics of the cable were experimentally analyzed with various types of series faults. The S-parameters obtained experimentally on the model cable were analyzed, and the output of the implemented linear time-invariant (LTI) system was evaluated with the experimental data. A machine learning tool was developed to predict the type of series fault with a regression algorithm based on the Scikit-learn library. The reported method has the potential of enhancing HTS power system reliability by providing early indications of series faults in HTS cables.