The safety and stability of wire ropes in hydropower stations are of paramount importance. However, traditional manual inspection methods suffer from low efficiency and poor accuracy. Utilizing wire rope monitoring images for fault localization not only greatly improves detection efficiency but also achieves high accuracy. This paper proposes a method based on the U-Net structure, which utilizes an encoder to extract features at different scales from the images and then employs a decoder to reconstruct these features into fault localization labels, thereby achieving wire rope fault localization. Experimental results demonstrate that the proposed method outperforms traditional fully convolutional networks and achieves excellent results in terms of Dice, IOU, and Hausdorff distance, enabling accurate wire rope fault localization.