Two-Wire Cable Anomaly Diagnosis with Machine Learning Based on Passive Measurements

This paper proposes a passive measurement method for the diagnosis of anomalies in two-wire cables that relies on Machine Learning. The proposed method requires the acquisition of the received signal and the analysis of the decoded sequence provided by the receiver itself. It does not require the acquisition of the injected signal and, therefore, it can be adopted in the existing two-wire communication systems without interrupting their operability. Time and frequency domain features are extracted from both the acquired signal and an estimate of the impulse response of the cable evaluated from the decoded sequence. A preliminary analysis of the proposed method is carried out by simulating a two-wire cable considering four anomalies: aging, tap, water-exposed, and air-exposed conductors. An analysis of the main features, which are more sensitive to anomaly detection and classification, has been carried out. A decision tree classifier has been trained with the 20 more sensitive features, and the obtained classification accuracy is 93.9%.