Görüntü İşleme ve Derin Öğrenme ile Kaynak Hatalarının Gerçek Zamanlı Tespiti

Yıl 2024, Cilt: 5 Sayı: 2, 83 - 97, 30.03.2025

Mehmet Emin Örs , Ziya Özçelik

Öz

Bu çalışmada, Convolutional Neural Networks (CNN) tabanlı bir model geliştirilerek kaynak kusurlarının tespitinde %93,5 doğruluk ve %91,3 F1-skoru ile oldukça iyi bir performans elde edilmiştir. Modelin başarısı, 800 yüksek çözünürlüklü görüntüden oluşan veri setinin %80 eğitim ve %20 test için ayrılarak, gürültü giderme ve segmentasyon gibi ön işleme teknikleriyle optimize edilmesiyle sağlanmıştır. Transfer öğrenme teknikleri, küçük veri setlerinde dahi performansı artırmış; 0,25 saniyelik analiz süresi, modelin gerçek zamanlı uygulamalarda kullanılabilirliğini kanıtlamıştır. DenseNet, YOLOv5 ve ResNet gibi popüler mimarilerle çalışmada kullanılan model karşılaştırılmıştır. Geleneksel yöntemlere kıyasla %15 doğruluk artışı ve işlem süresinde yarı yarıya azalma sağlanmıştır. Bu sistem, Endüstri 4.0 kapsamında otomatik kalite kontrol süreçlerine yeni bir standart getirmekte ve özellikle otomotiv, havacılık gibi sektörlerde uygulanabilirliği ile öne çıkmaktadır.

Anahtar Kelimeler

Evrişimli Sinir Ağları (CNN), Kaynak Hatası Tespiti, Transfer Öğrenimi, Gerçek Zamanlı Uygulamalar, Otomatik Kalite Kontrol, Endüstri 4.0

Automatic Detection of Welding Defects Using Image Processing and Deep Learning

Öz

This study developed a model based on Convolutional Neural Networks (CNNs) to achieve good performance for weld defect detection with 95.3% accuracy and 91.3% F1-score. The model's success was achieved by optimizing the dataset of 800 high-resolution images with preprocessing techniques such as noise removal and segmentation, separating 80% for training and 20% for testing. Transfer learning techniques improved performance even on small datasets, and the analysis time of 0.25 seconds proved the usability of the model in real-time applications. The base model in this work is compared with popular architectures such as DenseNet, YOLOv5, and ResNet. Compared to traditional methods, a 15% increase in accuracy and a halving in processing time were achieved. This system brings a new standard to automated quality control processes within the scope of Industry 4.0 and stands out with its applicability, especially in sectors such as automotive and aviation.

Anahtar Kelimeler

Convolutional Neural Networks (CNN), Weld Defect Detection, Transfer Learning, Real-Time Applications, Automated Quality Control, Industry 4.0

Kaynakça

• Sahman, M. A., Cinar, A. C., Saritas, I., & Yasar, A. (2019). Tree-seed algorithm in solving real-life optimization problems. IOP Conference Series: Materials Science and Engineering, 675(1), 012030.
• Atıcı, H., & Koçer, H. E. (2023). Mask R-CNN tabanlı kan yayma görüntülerinin segmentasyonu ve sınıflandırılması. Gazi Mühendislik Bilimleri Dergisi, 9(1), 128-143.
• Çevik, K. K., Koçer, H. E., & Boğa, M. (2022). Derin öğrenme tabanlı yumurta döllülük tespiti. Veterinary Sciences, 9(10), 574.
• Kang, J., & Ku, N. (2019). Verification of Resistance Welding Quality Based on Deep Learning. Materials Science and Engineering, Link.
• Zhang, Y., You, D., Gao, X., Zhang, N., & Gao, P. P. (2019). Welding defects detection based on deep learning with multiple optical sensors during disk laser welding of thick plates. Journal of Manufacturing Systems, Link.
• Xia, C., Pan, Z., Fei, Z., Zhang, S., & Li, H. (2020). Vision-based defects detection for Keyhole TIG welding using deep learning with visual explanation. Journal of Manufacturing Processes, Link.
• Walther, D., Schmidt, L., Schricker, K., Junger, C., Bergmann, J., Notni, G., & Mäder, P. (2022). Automatic Detection and Prediction of Discontinuities in Laser Beam Butt Welding Utilizing Deep Learning. Journal of Advanced Joining Processes, Link.
• Lee, D., Nie, G., & Han, K. (2022). Real-Time and Automatic Detection of Welding Joints Using Deep Learning. ASME Digital Library, Link.
• Tyystjärvi, T., Virkkunen, I., Fridolf, P., Rosell, A., & Barsoum, Z. (2022). Automated defect detection in digital radiography of aerospace welds using deep learning. SpringerLink, Link.
• Sun, J., Li, C., Wu, X., Palade, V., & Fang, W. (2019). An Effective Method of Weld Defect Detection and Classification Based on Machine Vision. IEEE Transactions on Industrial Informatics. Link
• Ma, G., Yuan, H., Yu, L., & He, Y. (2021). Monitoring of Weld Defects of Visual Sensing Assisted GMAW Process with Galvanized Steel. Materials and Manufacturing Processes. Link
• Rabe, P., Schiebahn, A., & Reisgen, U. (2021). Deep Learning Approaches for Force Feedback-Based Void Defect Detection in Friction Stir Welding. Journal of Advanced Joining Processes. Link
• Zuo, Y., Wang, J., & Song, J. (2021). Application of YOLO Object Detection Network in Weld Surface Defect Detection. IEEE International Conference on Cyber-Physical Systems. Link
• Zhang, Y., You, D., Gao, X., Zhang, N., & Gao, P. P. (2019). Welding Defects Detection Based on Deep Learning with Multiple Optical Sensors During Disk Laser Welding of Thick Plates. Journal of Manufacturing Systems. Link
• Adhi Wijaya, S. (n.d.). Welding Defect Object Detection [Data set]. Kaggle. Retrieved November 10, 2024, from https://www.kaggle.com/datasets/sukmaadhiwijaya/welding-defect-object-detection
• Jia, L., et al. (2024). Artificial intelligence and smart sensor-based industrial advanced technology. Sensors. Bansal, A., et al. (2025). Automated defects detection using deep learning in TIG welding. Springer.
• Chen, Y., et al. (2024). Weld seam defect detection using deformable CNNs. IEICE Electronics Express.
• Pavlov, M., Rybin, E., Kirill, I., Marakhtanov, A., & Korzun, D. (2024, October). Real-Time Industrial Automated Video Analytics System for Welding Defect Detection. In 2024 36th Conference of Open Innovations Association (FRUCT) (pp. 585-592). IEEE.
• Liu, W., Hu, J., & Qi, J. (2025). Resistance Spot Welding Defect Detection Based on Visual Inspection: Improved Faster R-CNN Model. Machines, 13(1), 33.
• Lee, C. W., Woo, S., & Kim, J. (2024). Machine-Learning-Based Joint Defect Prediction Using Temperature Distribution of High-Frequency Induction-Brazed Copper Joints. Journal of Materials Engineering and Performance, 1-12.
• Han, Z., et al. (2024). DFW-YOLO: Weld defect recognition using phased array ultrasonic testing. Taylor & Francis.
• Kokolakis, G. (2024). Laser brazed defect detection using deep-learning denoising and transfer learning : A case study in a Volvo factory to review techniques and methods (Dissertation). Retrieved from https://urn.kb.se/resolve?urn=urn:nbn:se:his:diva-24571
• Ma, Y., et al. (2024). Monitoring weld damage with deep learning and acoustic emission. SSRN.
• Ren, X., Du, X., Yu, H., Chang, Z., & Wang, G. (2024, August). TOFD Image Features Recognition Based on Improved YOLOv8. In 2024 IEEE 19th Conference on Industrial Electronics and Applications (ICIEA) (pp. 1-5). IEEE.