Transactions of Nanjing University of Aeronautics & Astronautics
Surface Defect Detection Method of Composite Materials Based on Deep Learning
- Article
- Figures
- Metrics
- Preview PDF
- Reference
- Related
- Cited by
- Materials
Abstract:
With the development tendency of energy saving and weight reduction of aerospace materials， high-performance composite materials have been widely used in aircraft skins. However， the surface quality detections of composite skin parts are mostly carried out manually， leading to low detection efficiency and low accuracy. Visual detection has gained more and more attention in recent years， mainly because of its non-destructive detecting characteristics with high precision and flexibility. In view of the visual detection requirements of surface defects of composite skin parts， a robot-based detection platform was constructed， which innovatively integrated manipulator module， image acquisition module， laser ranging module， the deep learning module， and the complementary upper computer software. In order to ensure the efficiency and accuracy of detection， the detection algorithm of the system was developed based on YOLOv5. In addition， on account of the lack of raw composite skin parts， the dataset for training was expanded by employing the following three methods： Mirroring and rotating， translating， and adding noise. Experiments validate that the system can realize online， automatic， and accurate detections of various types of composite skin parts. The proposed system can complete detection of an image with a size of 5 496 pixel× 3 672 pixel in 0.744 s， and the detection accuracy reaches 96.35%， which meets the requirements for composite surface quality detection.
Keywords:
Project Supported:
This work was supported by the National Natural Science Foundation of China （No.51975293） and the Aeronautical Science Foundation of China （No.2019ZD052010）.
Clc Number:
TP23
Mobile website
