Defect Detection Method of Carbon Fiber Composite Conductor Based on Semi-Supervised Anomaly Detection

Carbon Fiber Composite Core Conductors (ACCC) have emerged as a critical alternative to traditional steel-reinforced aluminum conductors due to their superior mechanical properties and power transmission efficiency. However, their poor bending resistance renders them susceptible to defects during production and transportation. Conventional X-ray inspection, reliant on manual interpretation, suffers from inefficiency and high misjudgment rates. To address these limitations, this study proposes a defect detection method based on semi-supervised anomaly detection. The method employs the Skip-GANomaly model, which integrates GAN and U-net architectures, utilizing a generator to reconstruct normal sample images and establishing an anomaly scoring mechanism derived from reconstruction discrepancies and latent feature differences. Experimental results demonstrate that the proposed method achieves over 95% accuracy and 85%-95% recall in detecting voids and saw-edge defects, significantly outperforming existing detection techniques. Although its accuracy for split and crease defects remains relatively low (~66%), the high recall rate effectively mitigates the risk of missed detection, thereby validating the practicality and reliability of this approach for ACCC defect inspection.