GAN-BASED SYNTHETIC DATA GENERATION FOR AUGMENTING ШХ15 BEARING DEFECT IMAGE DATASETS

Baymirzayev Akbarjon Rustamjan o'g'li

Authors

Baymirzayev Akbarjon Rustamjan o'g'li PhD Doctoral Researcher, Andijan State Technical Institute, Andijan, Uzbekistan

Keywords:

generative adversarial network, WGAN-GP, data augmentation, bearing rings, ШХ15 steel, casting defects, class imbalance, deep learning

Abstract

Deep learning-based defect detection in cast bearing components is limited by the scarcity and class imbalance of labeled training data. This study proposes a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) framework for generating realistic synthetic images of casting defects in ШХ15 (AISI 52100) bearing rings. An original dataset of 2,400 labeled surface images covering four defect categories — porosity, hot cracks, shrinkage cavities, and non-metallic inclusions — was used to train the WGAN-GP generator. The quality of synthetic images was evaluated using Fréchet Inception Distance (FID) and Inception Score (IS). The WGAN-GP achieved FID = 34.7 and IS = 2.84 for the combined defect classes, confirming visually plausible image generation. When a ResNet-50 classifier was retrained on the augmented dataset (original 2,400 + synthetic 4,800 images), overall detection accuracy improved from 96.8% to 98.4%, with the most significant gain observed for the minority class — non-metallic inclusions (F1-score from 93.2% to 97.1%). The results demonstrate that GAN-based augmentation is an effective strategy for addressing data scarcity and class imbalance in industrial defect detection systems.

References

He D., Xu K., Zhou P. Defect detection of hot rolled steels with a new object detection framework called classification priority network. Computers and Industrial Engineering, 2019; 128: 290–297. DOI: 10.1016/j.cie.2018.12.043

Baymirzayev A.R. Detection Methods of Defects in Bearings Produced by Casting Processes. World Academic Advancement Conference (Italy), 2025; 1(1): 18–21.

LeCun Y., Bengio Y., Hinton G. Deep learning. Nature, 2015; 521: 436–444. DOI: 10.1038/nature14539

Baymirzayev A.R. Podshipniklarni quyma usulda quyishda nuqsonlarni aniqlash usullari. InnoRes, 2025; 1(2): 23–27.

Goodfellow I., Pouget-Abadie J., Mirza M. et al. Generative adversarial nets. Advances in Neural Information Processing Systems, 2014; 27: 2672–2680.

Gulrajani I., Ahmed F., Arjovsky M. et al. Improved training of Wasserstein GANs. Advances in Neural Information Processing Systems, 2017; 30: 5767–5777.

Niu S., Li B., Wang X., Lin H. Defect image sample generation with GAN for improving defect recognition. IEEE Transactions on Automation Science and Engineering, 2020; 17(3): 1611–1622. DOI: 10.1109/TASE.2020.2967415

Liu T., Bao J., Wang J. et al. Weld defect images generation with generative adversarial network. Journal of Materials Processing Technology, 2021; 296: 117209.

Zhao Z., Li B., Dong R. et al. Surface defect detection by deep learning with data augmentation using GAN. Journal of Intelligent Manufacturing, 2022; 33: 1237–1250.

Zhang H., Tang Y., Zhang J. Fabric defect generation and classification using deep learning. Textile Research Journal, 2021; 91(15–16): 1824–1838.

Baymirzaev A., Makhammadjanov K., Yakubjonov F., Muhiddinov N., Otaquziev A. Advanced technologies for developing bearing materials: Properties, performance, and industrial applications. AIP Conference Proceedings, 2025; 3331(1): 030075. DOI: 10.1063/5.0305850

Bhadeshia H.K.D.H. Steels for bearings. Progress in Materials Science, 2012; 57(2): 268–435. DOI: 10.1016/j.pmatsci.2011.06.002

Karras T., Aila T., Laine S., Lehtinen J. Progressive growing of GANs for improved quality, stability, and variation. Proceedings of ICLR, 2018.

Ho J., Jain A., Abbeel P. Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 2020; 33: 6840–6851.

Baymirzayev A.R. The Role of Artificial Intelligence in Manufacturing Bearing Rings in Engineering Applications. International Conference on Applied Science and Education (Spain), 2025; 1(1): 47–49.