A Multimodal Deep Learning Framework Based on Ultrasonic Guided Waves for Corrosion Assessment of Aerial Ground Wires

Ultrasonic guided waves (UGWs), known for their long-range propagation and high sensitivity to internal material discontinuities, offer a powerful tool for early-stage corrosion detection in aerial ground wires. These wires are critical to power transmission infrastructure, ensuring both structural stability and grounding. However, long-term exposure to harsh environmental conditions often leads to corrosion-induced degradation, which compromises operational safety and reliability. Traditional inspection methods, such as visual examination and impedancebased analysis, suffer from limited sensitivity and inefficiency. To address these challenges, this paper proposes a multimodal deep learning framework centered on UGW signal analysis and enhanced by high-resolution RGB images and 3D point cloud data. The framework leverages self-attention and cross-modal attention mechanisms to effectively fuse intra- and inter-domain features, enabling accurate corrosion localization and severity classification. A new benchmark dataset, the Aerial Ground Wire Corrosion Dataset (AGWCD), is constructed to support real-world evaluation under complex environmental conditions. Experimental results demonstrate that our UGW-driven multimodal approach significantly outperforms conventional single-modality methods, particularly in moderate and severe corrosion scenarios. This work provides an intelligent, scalable solution for aerial ground wire monitoring, advancing predictive maintenance and enhancing the resilience of power transmission networks.