The Paton Welding Journal, 2026, №07

2026 №07 (06)

DOI of Article 10.37434/tpwj2026.07.07

2026 №07 (01)

---

The Paton Welding Journal, 2026, #7, 41-50 pages

Automated identification and localization of defects in magnetic particle inspection images based on YOLO11 model

A.S. Momot¹, V.S. Yakotyuk¹, Iu.Yu. Lysenko^1,3, R.M. Galagan¹, Y. Mirchev^2,3

¹National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute” 37 Prosp. Beresteiskyi, 03056, Kyiv, Ukraine. E-mail: a.momot@kpi.ua
²Institute of Mechanics at the Bulgarian Academy of Sciences 1113, Acad. G. Bonchev Str., Sofia, Bulgaria 3Center of Competence for Mechatronics and Clean Technologies “Mechatronics, Innovation, Robotics, Automation and Clean Technologies” — MIRACle, Bulgaria

Abstract
This paper examines the potential applications of deep learning models of the YOLO11 family to automate defect detection in magnetic particle inspection images, a method widely used to inspect ferromagnetic parts, and the decisions based on the results obtained impact operational safety and costs. At the same time, magnetic particle inspection images often contain a grainy background, uneven lighting, and glare, which mask low-contrast magnetic patterns and reduce the reliability of simple analysis methods. We propose a software-based approach in which camera frames are automatically processed by a neural network, and the output generates defect localization frames and classifies the defects by type. Real images were used to train the models; the number of images was increased through augmentation using transformations that simulate typical variations in imaging parameters. The dataset was annotated for three defect classes and divided into training, validation, and test sets. The YOLO11n, YOLO11s, YOLO11m, and YOLO11l models were trained and compared by the Precision, Recall, mAP50, and mAP50-95 metrics and by frame processing time. The results show that increasing the model size does not yield a proportional improvement in quality on the given dataset. The YOLO11m model provided the best balance between accuracy and speed with mAP50-95 of 0.238 and a frame processing time of 29.5 ms. Analysis of the training results revealed higher detection quality for linear defects and difficulty in detecting small clusters of defects, which is attributed to a lack of training data and a noisy background. Recommendations are provided on methods to reduce false positives and expand the dataset to improve the representativeness of the training set.
Keywords: magnetic particle inspection, defect detection, computer vision, artificial intelligence, neural networks

Received: 22.04.2026
Received in revised form: 04.06.2026
Accepted: 13.07.2026

References

1. (2016) ISO 9934-1:2016: Non-destructive testing. Magnetic particle testing. Pt 1 General principles. Geneva, ISO.
2. Yakotiuk, V.S., Hlabets, S.M., Lysenko, Yu.Yu., Momot, A.S. (2025) Practical experience of using and analyzing the efficiency of various magnetic suspensions. Technical Diagnostics and Non-Destructive Testing, 4, 44-48 [in Ukrainian]. https://doi.org/10.37434/tdnk2025.04.06
3. Yang, Y., Yang, Y., Li, L. et al. (2022) Automatic defect identification method for magnetic particle inspection of bearing rings based on visual characteristics and high-level features. Appl. Sci., 12(3), 1293. https://doi.org/10.3390/app12031293
4. Mao, Z., Wu, Q., Qin, X. et al. (2023) Magnetic particle inspection. Status, advances, and challenges. Demands for automatic non-destructive testing. NDT & E Inter., 143, 103030. https://doi.org/10.1016/j.ndteint.2023.103030
5. Wu, Q., Qin, X., Dong, K. et al. (2023) A training-based crack defect detection and 3D localization framework for automated fluorescent magnetic particle inspection. Expert Systems with Applications, 214, 118966. https://doi.org/10.1016/j.eswa.2022.118966
6. Tout, K., Meguenani, A., Urban, J.-P., Cudel, C. (2021) Automated vision system for magnetic particle inspection of crankshafts using convolutional neural networks. The Inter. J. of Advanced Manufacturing Technology, 112(8), 3307-3326. https://doi.org/10.1007/s00170-020-06467-4
7. Chen, J., Tong, J., Su, J. (2025) Design of a real-time abnormal detection system for rotating machinery based on YOLOv8. Frontiers in Mechanical Engineering, 11, 1683572. https://doi.org/10.3389/fmech.2025.1683572
8. Momot, A., Kretsul, V., Muraviov, O., Galagan, R. (2024) Automated defect detection in printed circuit boards based on the YOLOv5 neural network. The Paton Welding J., 4, 46-52. https://doi.org/10.37434/tpwj2024.04.07
9. Shi, A., Wu, Q., Qin, X. et al. (2024) Lightweight detector based on knowledge distillation for magnetic particle inspection of forgings. NDT & E Inter., 143, 103052. https://doi.org/10.1016/j.ndteint.2024.103052
10. Wu, Q., Qin, X., Xiong, X. (2025) Investigating the effects of data and image enhancement techniques on crack detection precision in FMPI. Advanced Engineering Informatics, 65(PA), 103169. https://doi.org/10.1016/j.aei.2025.103169
11. Wang, H., Du, W., Xu, G. et al. (2024) Automated crack detection of train rivets using fluorescent magnetic particle inspection and instance segmentation. Sci. Reports, 14, 10666. https://doi.org/10.1038/s41598-024-61396-6
12. Ivasenko, I.B., Vorobel, R.A., Uchanin, V.M. et al. (2023) Detection of chalcopyrite in digital images of copper ore sample sections. Information Extraction and Processing, 51(127), 52-61. https://doi.org/10.15407/vidbir2023.51.052
13. Storozhyk, D., Protasov, A., Kuts, Yu. et al. (2024) Enhancing neural network efficiency in automated image analysis for thermal nondestructive testing. J. of Theoretical and Applied Mechanics, 54, 242-252. https://doi.org/10.55787/jtams.24.54.2.242
14. Redmon, J., Divvala, S., Girshick, R., Farhadi, A. (2016) You only look once: Unified, real-time object detection. In: Proc. of the IEEE Conf. on Computer Vision and Pattern Recog nition (CVPR), 779-788. https://doi.org/10.1109/CVPR.2016.91
15. Wang, J., Wang, Y., Li, X. et al. (2025) SCI-YOLO11: An improved defect detection algorithm for transmission line insulators based on YOLO11. Plos One, 20(10), e0322561. https://doi.org/10.1371/journal.pone.0322561
16. Everingham, M., Van Gool, L., Williams, C.K.I. et al. (2010) The PAS CAL visual object classes (VOC) challenge. Inter. J. of Comput. Vis., 88, 303-338. https://doi.org/10.1007/s11263-009-0275-4

---

Suggested Citation

A.S. Momot, V.S. Yakotyuk, Iu.Yu. Lysenko, R.M. Galagan, Y. Mirchev (2026) Automated identification and localization of defects in magnetic particle inspection images based on YOLO11 model. The Paton Welding J., 07, 41-50. https://doi.org/10.37434/tpwj2026.07.07

---