在製造過程中，產品品質的保證是最重要的問題。品質控制措施，特別是缺陷檢 測，是確保產品完整性的重要工具。將人工智慧 (AI) 整合到這些流程中為解決這一 關鍵需求提供了一條有希望的途徑。本研究致力於透過利用影像處理演算法中深度 學習的力量來增強缺陷檢測機制並解決手動缺陷檢測方法，從而打造一個強大的解 決方案。本研究依賴ResNet模型對電阻點焊條件進行分類，並依賴ResUNet對電阻 點焊缺陷條件進行分割。 ResNet 的架構具有殘差連接或跳躍連接以及有效處理深度 網路的能力，使其特別適合且穩定地處理涉及微型物體或影像中複雜細節的任務。 相較之下，像 VGG 這樣的單層模型由於其較淺的架構可能難以捕捉微尺度物體中 存在的複雜性。所提出的系統將與傳統方法並列，強調其在準確性、效率和適應性 方面的優越性。使用VGG、Unet等不同類型模型與Resnet進行整體評估比較，探討 不同網路架構在此背景下的功效。本研究中採用的方法涉及將 ResNet 系統整合到專 門為製造中的缺陷檢測而定制的影像處理管道中。 ResNet 與替代模型的系統評估 和比較涵蓋了不同的層和配置，分析了它們在電阻點焊條件下檢測缺陷的性能。此 外，由於考慮到GPU 的使用和耗時，本研究在分類和分割之間使用單獨的任務模型 只檢測分割中的缺陷部分。此外，本研究也旨在利用開爾文技術（四點測量）分析 電阻點焊條件與電阻率之間的相關性。結果表明，ResNet 和 ResUnet 在分割和分類 方面比先前使用其他模型的研究取得了更好的性能。所提出的深度學習架構的分類 精度達到 0.95，分割骰子結果達到 0.87。

In manufacturing, the assurance of product quality stands as a paramount concern. Quality control measures, particularly defect detection, serve as instrumental tools in ensuring product integrity. The integration of artificial intelligence (AI) into these processes presents a promising avenue toward addressing this critical need. This study endeavors to forge a robust solution by harnessing the power of deep learning within image processing algorithms to enhance defect detection mechanisms and solve on manual defect detection method. This study rely on ResNet model, to classify resistance spot welding conditions and ResUNet to segmented the defect condition on resistance spot welding. ResNet’s architecture with its residual connections or skip connections and ability to effectively handle deep networks makes it particularly well-suited and stablefor tasks involving microscale objects or intricate details within images. In comparison, single-layer models like VGG might struggle to capture the complexities present in microscale objects due to their shallower architectures. The proposed system will be juxtaposed against traditional methods, emphasizing its superiority in accuracy, efficiency, and adaptability. Using different type models such as VGG and Unet as a comparison to Resnet in overall evaluation, exploring the efficacy of diverse network architectures in this context. The methodology employed in this study involves the systematic integration of ResNet within image processing pipelines tailored explicitly for defect detection in manufacturing. The systematic evaluation and comparison of ResNet against alternative models encompass various layers and configurations, analyse their performance in detecting defect on resistance spot welding conditions. Furthermore, this study using separate task model between classification and segmentation because of the consideration in GPU usage and time consuming which only detect the defect part in segmentation. Additionally, this study also aim to analyse the correlation between the resistance spot welding condition and the resistivity using kelvin’s techniques(four-point measurement). The result shows that ResNet and ResUnet achieved better performance than previous studies using other models in segmentation and classification. The proposed Deep Learning architecture achieves 0.95 on classification accuracy and 0.87 on segmentation dice result.

[1] Fortune Business Insight. Welding market size, impact analysis by product type(equipment and consumables), by welding type, by application, 2023-2030, 2022.
[2] Michael Moles and Olympus NDT. Manual phased arrays for weld inspections using north american codes. In 17th World Conference on Nondestructive Testing, pages 25–28, 2008.
[3] SK Nath, Krishnan Balasubramaniam, CV Krishnamurthy, and BH Narayana. Detection and sizing of defects in complex geometry weld by manual ultrasonic time of flight diffraction inspection. 2009.
[4] P Caleb and M Steuer. Classification of surface defects on hot rolled steel using adaptive learning methods. In KES'2000. Fourth International Conference on Knowledge-based Intelligent Engineering Systems and Allied Technologies. Proceedings (Cat. No. 00TH8516), volume 1, pages 103–108. IEEE, 2000.
[5] Kaixiang Peng and Xuli Zhang. Classification technology for automatic surface defects detection of steel strip based on improved bp algorithm. In 2009 Fifth International Conference on Natural Computation, volume 1, pages 110–114. IEEE, 2009.
[6] Maoxiang Chu, Jie Zhao, Xiaoping Liu, and Rongfen Gong. Multi-class classification for steel surface defects based on machine learning with quantile hyperspheres. Chemometrics and Intelligent Laboratory Systems, 168:15–27, 2017.[7] Zhao Liu, Jia Hu, Li Hu, Xiao Long Zhang, and Jian Yi Kong. Research on on-line surface defect detection for steel strip based on sparse coding. Advanced Materials Research, 548:749–752, 2012.
[8] Mohammad Reza Yazdchi, Arash Golibagh Mahyari, and Ali Nazeri. Detection and classification of surface defects of cold rolling mill steel using morphology and neural network. In 2008 International Conference on Computational Intelligence for Modelling Control & Automation, pages 1071–1076. IEEE, 2008.
[9] Jinho Kim, Byung-Soo Kim, and Silvio Savarese. Comparing image classification methods: K-nearest-neighbor and support-vector-machines. In Proceedings of the 6th WSEAS international conference on Computer Engineering and Applications, and Proceedings of the 2012 American conference on Applied Mathematics, pages 133–138, 2012.
[10] Zhifen Zhang, Zhe Yang, Wenjing Ren, and Guangrui Wen. Random forest model based real-time defect detection of al alloy in robotic arc welding using optical spectrum. Journal of Manufacturing Processes, 42:51–59, 2019.
[11] Caixia Zhu, Haitao Yuan, and Guohong Ma. An active visual monitoring method for gmaw weld surface defects based on random forest model. Materials Research Express, 9(3):036503, 2022.
[12] Ge-Wen Kang and Hong-Bing Liu. Surface defects inspection of cold rolled strips based on neural network. In 2005 international conference on machine learning and cybernetics, volume 8, pages 5034–5037. IEEE, 2005.
[13] Swapnil Gundewar, Prasad Kane, Santosh Behara, and Uttam Kumar. Welding defect detection using artificial neural network and support vector machine. In IOP Conference Series: Materials Science and Engineering, volume 1259, page 012029. IOP Publishing, 2022.
[14] M Amarnath, N Sudharshan, and P Srinivas. Automatic detection of defects in welding using deep learning-a systematic review. Materials Today: Proceedings, 2023.
[15] Lu Yang and Hongquan Jiang. Weld defect classification in radiographic images using unified deep neural network with multi-level features. Journal of Intelligent Manufacturing, 32:459–469, 2021.
[16] Yatao Yang, Longhui Pan, Junxian Ma, Runze Yang, Yishuang Zhu, Yanzhao Yang, and Li Zhang. A high-performance deep learning algorithm for the automated optical inspection of laser welding. Applied Sciences, 10(3):933, 2020.
[17] Bo Guo, Youtao Wang, Xu Li, Yeping Zhou, Jianmin Li, and Lanxiang Rao. Welding defect classification based on lightweight cnn. International Journal of Pattern Recognition and Artificial Intelligence, 37(11):2350026, 2023.
[18] Raphaella HF Murta, Fl´avison de A Vieira, Victor O Santos, and Elineudo P de Moura. Welding defect classification from simulated ultrasonic signals. Journal of Nondestructive Evaluation, 37:1–10, 2018.
[19] Mei-Po Ho, Wing-Kit Ngai, Tai-Wing Chan, and Hon-wah Wai. An artificial neural network approach for parametric study on welding defect classification. The International Journal of Advanced Manufacturing Technology, 120(1-2):527–535, 2022.
[20] Paolo Tripicchio and Salvatore D’Avella. Welding defect detection with deep learning architectures. Engineering Principles—Welding and Residual Stresses, 2022.
[21] Ronay Ak, Max Ferguson, Yung-Tsun T Lee, and Kincho H Law. Automatic localization of casting defects with convolutional neural networks. 2017.
[22] Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik. Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 580–587, 2014.
[23] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28, 2015.
[24] Johannes Ruisz, J¨urgen Biber, and Mario Loipetsberger. Quality evaluation in resistance spot welding by analysing the weld fingerprint on metal bands by computer vision. The International Journal of Advanced Manufacturing Technology, 33:952–960, 2007.
[25] Yann LeCun, L´eon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE,86(11):2278–2324, 1998.
[26] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 2012.
[27] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.
[28] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014.
[29] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference
on computer vision and pattern recognition, pages 779–788, 2016.
[30] Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18, pages 234–241. Springer, 2015.
[31] Vijay Badrinarayanan, Alex Kendall, and Roberto Cipolla. Segnet: A deep convolutional encoder-decoder architecture for image segmentation. IEEE transactions on pattern analysis and machine intelligence, 39(12):2481–2495, 2017.
[32] Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and Jiaya Jia. Pyramid scene parsing network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2881–2890, 2017.
[33] Zhengxin Zhang, Qingjie Liu, and Yunhong Wang. Road extraction by deep residual u-net. IEEE Geoscience and Remote Sensing Letters, 15(5):749–753, 2018.
[34] Kentaro Wada et al. Labelme: Image polygonal annotation with python, 2016.