Cable Fault and Assessment Using Multiple Frequency Autoencoder Regression-based Reflectometry

Abstract

The distribution cable may be considered the most critical element for power system operation through the key functions of electricity supplement and control and instrumentation signal transmission. Hence, there is a growing need for cable diagnostic techniques that enable accurate condition monitoring and fault detection in cables where external artifact signals are to be continuously measured. This research presents a technique for detecting cable faults based on autoencoder regression-based reflectometry with multiple frequency sinusoidal signals. Estimations of the reflected signal and preliminary test results of a non-faulty cable are used to train the time-series signal reconstruction and anomaly detection, allowing the distinguishment between a fault-induced reflected signal and various artifacts resulting from noise, input mismatch, or other factors. Experiment results on fault location in bypass cable and the reflected signal discrimination on branched network cable have validated the usefulness of the proposed algorithm.