在工業中,工業裝置的維護是一件重要的活動。透過維護使得工業裝置一直維持在生產的狀態,減少停機所帶來的損失。在過去,企業採用定期維護策略,每過一段時間就更換工業裝置中的零件,有些零件還沒損壞卻被更換,造成成本增加。隨著工業物聯網的發展,企業開始導入預測性維護,預測性維護為在工業裝置剛發生異常時,系統就通知現場人員,讓他們可以及時更換受損的零件,因此零件的壽命得以延長且預防發生更嚴重的異常,大大地降低生產成本。V型皮帶和刀具都是工業裝置中的關鍵零件,前者負責在沖床中傳遞馬達的動力,後者則是安裝在電腦數值控制銑床,負責切削產品,兩者的損壞都將導致工業裝置停止生產以及大量地增加生產成本,甚至危及現場工作人員的安全。本篇論文基於生成對抗網路,提出一個異常偵測方法,透過V型皮帶資料集和銑床刀具資料集做驗證,在工業裝置剛出現異常時,模型即可偵測到並通知現場人員,達成預測性維護。

Maintenance of industrial devices is an essential activity in the industry. Maintenance keeps industrial devices producible and reduces losses caused by downtime. In the past, enterprises used schedule-based maintenance, replacing parts in industrial devices regularly. Some parts are replaced without damage, resulting in increased costs. With the development of the Industrial Internet of Things (IIoT), enterprises are beginning to introduce Predictive Maintenance (PdM). PdM is a system that notifies operators as soon as an abnormality occurs in industrial devices, allowing them to replace parts promptly. As a result, part life is extended and more severe damage is prevented, greatly reducing production costs. Both V-belts and tools are key components in industrial devices. The V-belt is responsible for transmitting the motor’s power in the stamping press machine. The tool is installed on the milling machine, which is responsible for cutting the product. The damage of both will lead to unproductive industrial devices, increase costs, and even endanger the safety of operators. Based on the Generative Adversarial Networks (GAN), this thesis proposes an anomaly detection method, which is verified by the V-belt dataset and the milling machine tool dataset. When the industrial devices are abnormal in the early stage, the model can detect the abnormality to achieve PdM.

Acknowledgements
摘要-i
Abstract-ii
1 Introduction-1
2 Related Works-5
3 Anomaly Detection Method based on Modified Wasserstein Generative Adversarial Networks Gradient Penalty-9
3.1 GAN applied to Industrial Anomaly Detection-9
3.2 Wasserstein Generative Adversarial Networks Gradient Penalty (WGAN-GP)-10
3.3 Modified WGAN-GP-14
3.4 Anomaly Detection Method-17
4 Experimental Results-19
4.1 Verification of the Proposed Anomaly Detection on the V-belt Dataset-19
4.2 Verification of the Proposed Anomaly Detection on the Tool Dataset-25
5 Conclusion-31
References-33

[1] A. Prajapati, J. Bechtel, and S. Ganesan, “Condition based maintenance: a survey,” Jour- nal of Quality in Maintenance Engineering, vol. 18, no. 4, pp. 384–400, 2012.
[2] S. Vaidya, P. Ambad, and S. Bhosle, “Industry 4.0–a glimpse,” Journal of Procedia Man- ufacturing, vol. 20, pp. 233–238, 2018.
[3] Y. Ran, X. Zhou, P. Lin, Y. Wen, and R. Deng, “A survey of predictive maintenance: Sys- tems, purposes and approaches,” IEEE Communications Surveys and Tutorials, vol. 20, pp. 1–36, 2019.
[4] U. Ali, R. Hafiz, T. Tauqeer, U. Younis, W. Ali, and A. Ahmad, “Towards machine learning based real-time fault identification and classification in high power induction motors,” in IEEE 5th International Conference on Robotics and Automation Engineering (ICRAE), pp. 46–53, Singapore, 2020.
[5] I. Amihai, R. Gitzel, A. M. Kotriwala, D. Pareschi, S. Subbiah, and G. Sosale, “An indus- trial case study using vibration data and machine learning to predict asset health,” in IEEE 20th Conference on Business Informatics (CBI), pp. 178–185, Vienna, Austria, 2018.
[6] H. Qin, K. Xu, and L. Ren, “Rolling bearings fault diagnosis via 1d convolution networks,” in IEEE 4th International Conference on Signal and Image Processing (ICSIP), pp. 617– 621, Wuxi, China, 2019. [7] X. Xia, X. Pan, N. Li, X. He, L. Ma, X. Zhang, and N. Ding, “Gan-based anomaly detec- tion: A review,” Journal of Neurocomputing, vol. 493, pp. 497–535, 2022.
[8] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial nets,” in Advances in Neural In- formation Processing Systems (NIPS), pp. 2672–2680, Montreal, Canada, 2014.
[9] H. Han, L. Hao, D. Cheng, and H. Xu, “Gan-sae based fault diagnosis method for electri- cally driven feed pumps,” Journal of Plos One, vol. 15, no. 10, p. e0239070, 2020.
[10] L. Alzubaidi, J. Zhang, A. J. Humaidi, A. Al-Dujaili, Y. Duan, O. Al-Shamma, J. Santa- maría, M. A. Fadhel, M. Al-Amidie, and L. Farhan, “Review of deep learning: Concepts, cnn architectures, challenges, applications, future directions,” Journal of Big Data, vol. 8, no. 1, pp. 1–74, 2021.
[11] S. Cao, L. Wen, X. Li, and L. Gao, “Application of generative adversarial networks for intelligent fault diagnosis,” in IEEE 14th International Conference on Automation Science and Engineering (CASE), pp. 711–715, Munich, Germany, 2018. 33
[12] Y. T. K. Lai, J. S. Hu, Y. H. Tsai, and W. Y. Chiu, “Industrial anomaly detection and one-class classification using generative adversarial networks,” in IEEE/ASME Interna- tional Conference on Advanced Intelligent Mechatronics (AIM), pp. 1444–1449, Auck- land, New Zealand, 2018.
[13] K. Liu, A. Li, X. Wen, H. Chen, and P. Yang, “Steel surface defect detection using gan and one-class classifier,” in IEEE 25th International Conference on Automation and Comput- ing (ICAC), pp. 1–6, Lancaster, United Kingdom, 2019.
[14] X. Zhou, J. Xiong, X. Zhang, X. Liu, and J. Wei, “A radio anomaly detection dlgorithm based on modified generative adversarial network,” IEEE Wireless Communications Let- ters, vol. 10, no. 7, pp. 1552–1556, 2021.
[15] P. Nagorny, T. Lacombe, H. Favreliere, M. Pillet, E. Pairel, R. Le Goff, M. Wali, J. Loureaux, and P. Kiener, “Generative adverserial networks for geometric surfaces pre- diction in injection molding: Performance analysis with discrete modal decomposition,” in IEEE International Conference on Industrial Technology (ICIT), pp. 1514–1519, Lyon, France, 2018.
[16] P. Isola, J. Y. Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1125–1134, Honolulu, United States, 2017.
[17] J. Y. Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” in Proceedings of The IEEE International Confer- ence on Computer Vision (ICCV), pp. 2223–2232, Venice, Italy, 2017.
[18] K. Zhang, Y. Zhang, and H. Cheng, “Self-supervised structure learning for crack detection based on cycle-consistent generative adversarial networks,” Journal of Computing in Civil Engineering, vol. 34, no. 3, p. 04020004, 2020.
[19] R. Guo, H. Liu, G. Xie, and Y. Zhang, “Weld defect detection from imbalanced radio- graphic images based on contrast enhancement conditional generative adversarial network and transfer learning,” IEEE Sensors Journal, vol. 21, no. 9, pp. 10844–10853, 2021.
[20] M. Mirza and S. Osindero, “Conditional generative adversarial nets,” arXiv:1411.1784 [cs.LG], 2014.
[21] B. Li, F. Cheng, H. Cai, X. Zhang, and W. Cai, “A semi-supervised approach to fault detection and diagnosis for building hvac systems based on the modified generative ad- versarial network,” Journal of Energy and Buildings, vol. 246, p. 111044, 2021.
[22] P. Vincent, H. Larochelle, Y. Bengio, and P.-A. Manzagol, “Extracting and composing robust features with denoising autoencoders,” in Proceedings of The 25th International Conference on Machine Learning (ICML), pp. 1096–1103, Helsinki, Finland, 2008.
[23] I. Gulrajani, F. Ahmed, M. Arjovsky, V. Dumoulin, and A. C. Courville, “Improved train- ing of wasserstein gans,” in Advances In Neural Information Processing Systems (NIPS), pp. 5767–5777, Long Beach, United States, 2017.
[24] C. Villani, ”Optimal Transport: Old and New”. Springer, Germany, 2009. 34
[25] I. Kovalenko, M. Saez, K. Barton, and D. Tilbury, “Smart: A system-level manufactur- ing and automation research testbed,” Journal of Smart and Sustainable Manufacturing Systems, vol. 1, no. 1, 2017.