Page 157 - Read Online

Research Article | Open Access

Journal of Materials Informatics

Wang et al. J. Mater. Inf. 2026, 6, 13 DOI:10.20517/jmi.2025.74

Physically synthesized data for deep learning-based
visual scratch inspection of aerospace alloys with
complex geometries

Yuanbin Wang 1,2,3 , Yupeng Bai , Peng Wang , Wenhu Wang 1,2,3 , Mingzhu Zhu 5,6,* , Ming Luo 1,2,3
4
1
Keywords:
Automatic visual inspection,
physical image synthesis,
defect segmentation, deep
learning, aerospace alloys
Citation: Wang, Y.; Bai, Y.;
Wang, P.; Wang, W.;
Zhu, M.; Luo, M. Physically
synthesized data for deep
learning-based visual
scratch inspection of
aerospace alloys with
complex geometries. J.
Mater. Inf. 2026, 6, 13.
https://dx.doi.org/10.20517
/jmi.2025.74

Received: 28 Aug 2025 Abstract
First Decision: 29 Oct
2025 Aerospace alloys often operate under extreme conditions. Accurate defect segmentation in
Revised: 27 Nov 2025 images of aerospace components is the key to quantifying the defects and evaluating their
Accepted: 12 Dec 2025 impact for part lifespan. The components usually have complex free-form surfaces, leading
Published: 5 Mar 2026
to uneven light distribution in images. The variable image presentations pose a great
Academic Editor: challenge for accurate segmentation, especially with limited data. Generative adversarial
Chaolin Tan networks and other training-based methods are commonly used for image generation, but
Copy Editor: they still rely on sufficient high-quality training data. In this paper, a physical-based image
Pei-Yun Wang generation method is proposed to create any possible scratches according to physical laws
Production Editor:
Pei-Yun Wang to improve the scratch segmentation capability with limited data. First, an efficient
scratched blade surface image generation pipeline is developed. Then, a systematic
strategy to maximize the effect of physical synthetic scratch images is presented. The
experiments show that the segmentation intersection-over-union could be improved from
0.66 to 0.83 with only 20 real images for training, and reveal the influences of network

structure, image and label quality, data fusion strategy on segmentation performance.

1 School of Mechanical Engineering, Northwestern Polytechnical University, Xi’an 710072, Shaanxi, China.
2 MIIT Key Laboratory of High Performance Manufacturing for Aero Engine; MOE Engineering Research Center of Advanced Manufacturing
Technology for Aero Engine, Northwestern Polytechnical University, Xi’an 710072, Shaanxi, China.
3 State Key Laboratory of Cemented Carbide, Northwestern Polytechnical University, Xi’an 710072, Shaanxi, China.
4 NEIAS Commercial Aircraft Company Limited, AVIC, Nanjing 210000, Jiangsu, China.
5 School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, Shaanxi, China.
6 National Key Laboratory of Aerospace Flight Dynamics, Northwestern Polytechnical University, Xi’an 710072, Shaanxi, China.

* Correspondence to: Dr. Mingzhu Zhu, School of Astronautics, Northwestern Polytechnical University, Xi’an 710072, Shaanxi, China.
E-mail: zhumingzhu@nwpu.edu.cn

www.oaepublish.com Submit a Manuscript: https://ucenter.oaepublish.com