(1) Background: Air rudders are used to control the flight attitude of aircraft, and their surface quality directly affects flight accuracy and safety. (2) Method: Traditional positioning methods can only obtain defect location information at the image level but cannot determine the defect’s physical surface position on the air rudder, which lacks guidance for subsequent defect repair. We propose a defect physical surface positioning method based on a camera mapping model. (3) Results: Repeated positioning experiments were conducted on three typical surface defects of the air rudder, with a maximum absolute error of 0.53 mm and a maximum uncertainty of 0.26 mm. Through hardware systems and software development, the real-time positioning function for surface defects on the air rudder was realized, with the maximum axial positioning error for real-time defect positioning being 0.38 mm. (4) Conclusions: The proposed defect positioning method meets the required accuracy, providing a basis for surface defect repair in the air rudder manufacturing process. It also offers a new approach for surface defect positioning in similar products, with engineering application value.

Aircraft fuselage corrosion inspection requires both image-domain recognition and metric physical-surface localization for maintenance execution. This study develops a camera-calibrated vision measurement framework that combines PWDE-YOLOv8n-based corrosion perception, original-image coordinate restoration, lens-distortion compensation, ray-based surface mapping, and detector-center bias correction. The perception dataset comprised 2143 images and 5941 corrosion annotations and was partitioned at the physical-specimen, acquisition-session, or source-group level into 1500 training images, 429 validation images, and 214 independent detector-test images. Detailed physical localization was evaluated on a six-image metrology cohort acquired in six sessions, containing 21 corrosion boxes and 84 axial coordinates. A six-fold leave-one-image-out procedure was adopted; in each fold, the center-shift parameters were estimated from the other five images and applied unchanged to the held-out image. The proposed method achieved a mean absolute axial error of 0.641 mm (95% image-cluster bootstrap confidence interval: 0.571–0.708 mm), an RMSE of 0.809 mm, a maximum error of 3.262 mm, and a projected physical-plane bounding-box IoU of 87.12%. The expanded uncertainty of the manually established reference coordinates was 0.374 mm at k = 2, and Monte Carlo propagation produced a mean absolute error of 0.656 mm with a 95% interval of 0.618–0.693 mm. The proposed method reduced the MAE by 97.91% relative to local pixel-to-millimeter scaling and by 70.58% relative to conventional calibrated camera mapping, while producing accuracy comparable to planar homography mapping. Within ρ ≥ 1500 mm, W ≤ 150 mm, and θ ≤ 20°, the estimated curvature-induced additional axial error did not exceed 0.683 mm. A separate ten-image deployment evaluation produced a mean axial error of 2.448 mm and an average processing time of 53.35 ms/image.