Virtual-real twin data powered deep adaptive detection method for corrosion damage in cable aluminum sheath structure using helical guided waves

Abstract

Corrosion damage is the most harmful failure of aluminum sheath accessories of high voltage cable, which directly threatens the security and stability of power grid. AI-enabled ultrasonic guided wave is a promising detection technology for corrosion damage of power components. However, the pain point of data-driven deep learning methods in engineering practice is the difficulty in building complete datasets and the poor physical interpretability of models. In this paper, a virtual-real twin data powered deep adaptive detection method based on helical guided wave is proposed to inspect the corrosion damage in cable aluminum sheath structure. Firstly, the twin data for network training is constructed by simulation model and guided wave mechanism model. Secondly, the generalization features between the standardized twin data and the actual data are learned through the deep transfer network. Finally, a twin data-driven deep adaptive network (TDDAN) is formed by combining simulation model construction, guided wave mechanism model and deep transfer network, which realizes high-precision intelligent detection of aluminum sheath corrosion damage of high-voltage cable. The average accuracy of corrosion damage localization and degree identification of aluminum sheathed high-voltage cable can reach 95.83 %, which shows stronger interpretability, universality and generalization ability than the existing methods.