A high-temperature superconducting (HTS) cable system requires cooling system to maintain a superconducting state. If a local fault occurs in the HTS cable, a fault current that causes heat can flow. This leads to a quench, where the superconductivity breaks down, inflicting critical damage on the HTS cable system. Conventional reflectometry method can detect the local quench point, but it is difficult to distinguish between a reflected signals from quench around the joint box and that from the joint box. In this paper, we propose two steps to solve this problem. 1) Anomaly detection step: we propose the integrating the autoencoder (AE) with time-frequency domain reflectometry (TFDR) to detect anomaly situation, and localize impedance mismatch point through time-frequency cross correlation (TFCC) function, which calculate the similarity between a reference signal and reflected signal. 2) Source separation step: If TFCC of signals overlap with each other, A expectation-maximization (EM) algorithm is used to separate each overlapped TFCC. The proposed method enables efficient monitoring of the HTS cable system by effectively detecting quench that occur around the joint box. We simulate the quench situation, and verify the proposed method through the simulation.
With the continuous expansion of modern electrical infrastructure, the scale and deployment of electrical cable networks have rapidly increased, underscoring the critical importance of systematic cable system maintenance. Among various diagnostic technologies, reflectometry has emerged as a highly effective and prominent methodology for precise fault detection and localization. Although the operational reliability of these networks fundamentally relies on precise fault detection, identifying minor or early-stage degradation remains a formidable challenge due to the marginal magnitude of the weak reflected signals. To address this limitation, this paper proposes a concurrent pulse superposition (CPS) method to enhance the detection sensitivity of weak reflection signatures in reflectometry without requiring expensive hardware modifications. The proposed method was evaluated through simulations covering fault resistances from to and SNR levels from to . Across these conditions, CPS increased the reflection peak magnitude by approximately 2.5–3.5 times and maintained identifiable peaks in all tested cases, whereas the conventional method failed under several combinations of high fault resistance and low SNR. Consequently, this approach is expected to significantly empower the diagnostic analysis of extremely weak, early-stage faults, offering a practical solution for predictive maintenance in electrical systems.