DOI QR코드

DOI QR Code

Visual Defect Inspection of Steel Surfaces Using Dual Deep Learning Models with Grad-CAM Interpretation

Grad-CAM 기반 딥러닝 분류 모델을 활용한 강판 표면 결함의 비파괴 시각화 기법

  • 이병권 (서원대학교 미디어콘텐츠학부)
  • Received : 2025.06.11
  • Accepted : 2025.07.20
  • Published : 2025.07.30

Abstract

This paper proposes a non-destructive inspection method for detecting and visualizing surface defects in steel sheets by integrating two deep learning techniques. The first is a multi-class Convolutional Neural Network (CNN) model that automatically classifies defect types from steel surface images. The second is the Gradient-weighted Class Activation Mapping (Grad-CAM) technique, which visualizes the regions of the image that the model focuses on when making predictions. Various defect images were used in the experiments, providing both classification results and corresponding visual explanations. We compared the interpretability and effectiveness of Grad-CAM with traditional Canny edge-based visualization. The proposed approach enables real-time quality inspection in autonomous manufacturing systems and enhances the reliability and efficiency of non-destructive testing (NDT).

본 논문은 강판 표면의 결함을 탐지하고 시각화하기 위해 두 가지 인공지능 기법을 결합한 비파괴 검사 방법을 제안한다. 첫 번째는 다중 클래스 분류용 합성곱 신경망(CNN) 모델로, 강판 이미지의 결함 유형을 자동으로 분류한다. 두 번째는 Grad-CAM(Gradient-weighted Class Activation Mapping) 기법으로, 모델이 결함을 판단할 때 주목한 이미지 영역을 시각화한다. 실험에는 다양한 강판 결함 이미지가 사용되었으며, 각 결함에 대해 예측 결과와 시각적 근거를 동시에 제공하였다. 또한, 기존 Canny edge 기반의 단순 시각화와 비교하여 Grad-CAM의 해석 가능성과 정확도를 평가하였다. 본 방법은 자율제조 및 스마트팩토리 환경에서 실시간 품질 검사를 가능하게 하며, 비파괴 검사(NDT)의 신뢰성과 효율성을 향상시킬 수 있다.

Keywords

References

  1. Leng, J., et al. (2023). ManuChain II: Blockchained smart contract system as the digital twin of decentralized autonomous manufacturing toward resilience in Industry 5.0. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 53(8), 4715-4728. DOI : 10.1109/TSMC.2023.3257172
  2. Capriyani, D. M. I., Fanisa, S., Eriyanti, Saputra, R. P., Romdlony, M. Z. & Putra, M. D. (2024). Mecanum-Wheeled Autonomous Mobile Robot for Flexible Manufacturing System. In 2024 IEEE International Conference on Advanced Telecommunication and Networking Technologies (ATNT) (Vol. 1, pp. 1-4). IEEE. DOI : 10.1109/ATNT61688.2024.10719179
  3. Malyy, V. V., Kostyukhin, A. S., Fedorov, A. V. & Kinzhagulov, I. Y. (2022). Development of Technology for Automated Non-Destructive Quality Testing of Soldered Joints of Heat Exchangers. In 2022 International Conference on Information, Control, and Communication Technologies (ICCT) (pp. 1-4). IEEE. DOI : 10.1109/ICCT56057.2022.9976630
  4. Mao, X., Zhao, Y. & Xiao, T. (2018). Review of the development of metal non-destructive testing and imaging technology. In 2018 IEEE 4th Information Technology and Mechatronics Engineering Conference (ITOEC) (pp. 926-929). IEEE. DOI : 10.1109/ITOEC.2018.8740475
  5. Malyy, V. V., Gubin, M. S., Kostyukhin, A. S., Fedorov, A. V. & Kinzhagulov, I. Y. (2023). Development of an Algorithm for the Movement and Adjusting Measuring Transducers of an Automated Non-Destructive Testing System. In 2023 7th International Conference on Information, Control, and Communication Technologies (ICCT) (pp. 1-6). IEEE. DOI : 10.1109/ICCT58878.2023.10347120
  6. Liu, S., Sekine, T., Usuki, S. & Miura, K. T. (2024). Explanation of Convolutional Neural Network for Automotive Wire Harness Using Gradient-Weighted Class Activation Mapping. In 2024 IEEE Joint International Symposium on Electromagnetic Compatibility, Signal & Power Integrity: EMC Japan/Asia-Pacific International Symposium on Electromagnetic Compatibility (EMC Japan/APEMC Okinawa) (pp. 570-573). IEEE. DOI : 10.23919/EMCJapan/APEMCOkinaw58965.2024.10585120
  7. Liu, J., Chen, S., Cai, M., Shao, H. & Gui, W. (2025). Semi-Heterogeneous Graph-Perception Network With Gradient-Weighted Class Activation Mapping for Class-Incremental Industrial Fault Recognition and Root Cause Diagnosis. IEEE Transactions on Neural Networks and Learning Systems. DOI : 10.1109/TNNLS.2025.3567475
  8. Ahn, I. (2022). Deep learning-based defects detection of steel sheet surface using object-level data augmentation. Journal of the Korean Institute of Industrial Engineers, 48(4), 327-339. DOI : 10.7232/JKIIE.2022.48.4.327
  9. Pan, Z., et al. (2021). Non-destructive microwave testing method on porcelain suspension insulators. In 2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC) (pp. 985-988). IEEE. DOI : 10.1109/IAEAC50856.2021.9391007
  10. Miaoxin, L. & Xiaoyu, Z. (2020). Overview of non-destructive testing of composite materials. In 2020 3rd World Conference on Mechanical Engineering and Intelligent Manufacturing (WCMEIM) (pp. 166-169). IEEE. DOI : 10.1109/WCMEIM52463.2020.00041
  11. Guillet, J. P. & Fonseca, N. J. G. (2024). Radial multi-beam non destructive testing with a geodesic lens at 130 GHz. In 2024 49th International Conference on Infrared, Millimeter, and Terahertz Waves (IRMMW-THz) (pp. 1-2). IEEE. DOI : 10.1109/IRMMW-THz60956.2024.10697808