Physics-Informed Semi-Supervised Transmission Towers Bolt Detection and Localization Using Acoustic Wave Generator

The bolts on transmission towers play a critical role in ensuring the safety and stability of power systems. Existing bolt detection and localization methods for transmission towers typically employ percussion-based approaches, where hammerinduced acoustic signals are analyzed to extract features for assessing bolt state. However, current approaches face several limitations, including unstable signal acquisition and poor robustness to noise, high dependency on labeled data and limited incorporation of physical constraints, inadequate distinction between detection and localization tasks. These factors hinder the accuracy and robustness of existing detection and localization methods. To address these challenges, this paper proposes a physics-informed semi-supervised bolt looseness detection and localization method using acoustic wave generator. At the instrumentation level, we propose a novel acoustic source to excite the tower structure, enabling the generation of stable resonance responses and consequently more reliable signal acquisition. At the algorithmic level, we propose a novel semi-supervised learning framework, which incorporates a physics-informed data augmentation strategy and a feature extraction model designed to enhance the model’s generalization under limited labeled data conditions while improving resilience to environmental noise. To resolve localization ambiguity among spatially adjacent loosened bolts, we propose a contrastive loss function that incorporates the expected joint distribution, pseudo-label weighting, and spatial distance. Experimental results demonstrate that the proposed method outperforms state-of-the-art approaches even with fewer labeled samples. The corresponding confusion matrix shows significant improvements across most categories, demonstrating the effectiveness of the proposed method for bolt detection and localization in transmission towers.