Abstract: Aiming at the deviation phenomenon of flaw detection results caused by the coexistence and mutual interference of multiple types of defects in six-bundle conductor, an X-ray flaw detection technology for ultra-high voltage six-bundle conductor based on MobileNetV3 lightweight network architecture is proposed. Firstly, the X-ray attenuation difference in different areas when X-ray passes through the clamp is used, and the internal X-ray image of the tension clamp is collected by combining Lambert-Beer law and multi-angle electrical signal reconstruction. Then, the depthwise separable convolution and lightweight channel attention mechanism enhancement methods are used to improve the capture ability of key defect features. Finally, the defect location and recognition are realized by adaptive pre selection box generation and classification regression network, and the output results are optimized by using joint vector and full connection layer. The experimental results show that the detection accuracy of this method is 99.6%. Compared with the traditional manual inspection, the single detection time is shortened from more than 120 min to less than 60 min, which can effectively identify the scene of multiple defects coexisting of fracture and loose strands, and has important practical significance for the safe and stable operation of transmission lines.

Key words: six-boundle conductor, tension clamp, MobileNetV3, Lambert-Beer, X-ray image