SKAD: A Unified Framework Guided by Structural Knowledge for Anomaly Detection of Dampers in Transmission Lines

Abstract

The dampers absorb transmission line vibration energy, reducing the vibration amplitudes of conductors. However, dampers may develop internal structural anomalies (e.g., damage to damper heads) or external positional anomalies (e.g., slippage along the conductor), both of which compromise vibration suppression efficacy. Existing anomaly detection methods focus on single anomaly type and struggle with local feature extraction. To address these limitations, this paper introduces SKAD, a unified framework guided by structural knowledge, to concurrently detect internal and external damper anomalies. SKAD encodes structural properties of dampers through four key structural points, enabling sub-pixel-level localization via a hybrid network (HRNet + GAU + SimCC). By analyzing spatial relationships and vector features of these structural points, SKAD can simultaneously detect anomalies like damage (via confidence thresholds and vector dot products) and slippage (via depth-parallelism-distance constraints) at the structural level. Experiments on a real-world dataset demonstrate SKAD outperforms object-based methods in accuracy and robustness, providing novel transmission line inspection perspectives, ensuring early anomaly detection to prevent conductor fatigue and power outages.