隨著人工智慧於製造業的應用場景愈來愈廣泛和多元，如工具機製造業、半導體 製造業、自動化整合之相關產業，皆積極連結相關 AI 應用於生產製造流程中[1]，且 製造產業導入 AI 能大幅降低生產成本[2]，而 AI 導入產線應搭配彈性的部署方案， 且最適應於當前製造產線的模型被迫切地需要，則產線上模型的更替勢在必行。
本研究旨在針對半導體製造產線情境與工具機製造業，打造一遠端人工智慧模型 修正平台，該產線與機台多數配備 AI 相關應用實例，透過工業場域不斷產出的數據， 於遠端動態同步進行模型修正，使各個終端預測、辨識端點，達到修模無須停機之成 效，以在線執行服務、遠端運作修模(再訓練)的形式，產出最適用於當前端點的模 型。除了執行遠端模型修正運作外，亦搭配虛擬微服務技術並結合自動化整合、自動 化部署的功能，作為平台服務發佈的方法，藉此改善 AI 服務部署困難之解決方案。 以本平台實驗成果為例，模型修正完畢後開啟自動化整合與自動化部署的通道，最終 將該模型部署至端點，所需平均間為 323.5 秒，且終端服務平均僅終止 17.03 秒，相 較於研究統計[3]僅有 14%的人工智慧服務商能於一週內完成部署，且文獻指出模型 以手動進行驗證與部署平均需耗時 60 分鐘[4]，以本平台測試範例為例，透過自動化 整合結合自動化部署微服務，提升人工智慧部署速度 59%。此外部署範圍不受廠區限 制，任一配備網際網路的場域，僅需確認能夠以 HTTPS 連線與程式碼倉庫進行連線， 則能實施服務部署。經實測個人電腦三大作業系統，皆能作為本平台自動化部署之硬 體載具，驗證本平台具備跨作業平台之特性。此外本平台已達 MLops Level 2 等級[5]， 並且相比虛擬機重啟時間需耗費一分鐘以上[6]，減少 71.7%的系統重啟時間，硬體佔 用量亦減少 90.5%。
本研究除具備前述科學化數值之效益外，亦配備以下特性，(1)「遠端」模型修 正(2)隨產線「動態」生成模型(3)自動化部署(4)自動化整合(5)管理者網頁 狀態顯示介面(6)連結數據庫作為事件記錄用途(7)利用 API 獲取遠端欲執行訓練之數據資料(8)具備跨平台運行特性(9)全球性遠端部署(10)多方案運作(部署 機制、模型修正機制、伺服器架設機制等......)，除提高產線模型適應性外，同步模型 修正服務可減少產線停擺時間，亦改善過往部署耗時長久的問題，期許未來藉本平台 之遠端動態模型修正之成效，實際導入工具機產業、半導體製造業與自動化整合相關 產業，解決現階段許多產線人工智慧難以在線進行模型修正之困境。

This research aims to build a remote artificial intelligence model remodeling platform. Retrain the model dynamically at remote through the data produced continuously by the
industrial field. Enables individual prediction endpoints to achieve training model without stopping the machine. It stimulates the model best suited for the current endpoint in the form of a production line executing a service while running model training remotely. In addition to performing remote model retraining operations, it also uses virtual microservice technology. It combines the functions of continuous integration and deployment as a platform service release method. Take this as a solution to improve the difficulty of deploying AI services. Taking the experimental results of this platform as an example, after the model is retained, the pipeline of continuous integration and deployment will be opened. The average time required until the model is deployed to the endpoint is 323.5 seconds. And the endpoint service is stopped in only 17.03 seconds on average. Compared with the research statistic [1], only 14% of AI service providers can complete the deployment within a week. The Platform deploys services in seconds with excellent results. In addition, the literature states that it takes 60 minutes on average to validate and deploy a model manually. In this research, the deployment speed of artificial intelligence is increased by 59% through the microservices of continuous integration and deployment. And the deployment scope is not limited by the factory area. Any field equipped with the Internet can implement service deployment only by confirming that it can connect to the code repository through an HTTPS connection. The three major operating systems of the personal computer have been tested, and all of them can be used as the hardware for remote continuous deployment of this platform. That verifies the platform has the characteristics of a cross-operating platform. Apart from this, the platform has reached the MLops Level 2 [3], and reducing the system
restart time by 71.7% and the hardware usage by 90.5%. In addition to the benefits of the aforementioned scientific values, this research also has the following features: (1) Remodeling the AI model at remote, (2) "Dynamic" generate models along the production line, (3) Continuous deployment, (4) Continuous integration, (5) Website interface, (6) Link database for event recording purpose, (7) Use API to obtain remote data for training, (8)
Feature of cross-platform operation, (9) Global remote deployment. In addition to improve the adaptability of the model which deployed on the production line, the synchronous model remodeling service can reduce the downtime of the production line, and also improve the long-time deployment problem in the past.

[1] J. Lee, H. Davari, J. Singh, and V. Pandhare, "Industrial artificial intelligence for industry 4.0-based manufacturing systems," Manufacturing Letters, vol. 18, pp. 20-23, 2018.
[2] A. Cam, M. Chui, B. Hall, and D. DeLallo, "Global AI survey: AI proves its worth, but few scale impact," McKinsey & Company, pp. 1-11, 2019.
[3] Algorithmia Inc, "2020 state of enterprise machine learning," 2020, url. https://algorithmia.com/state-of-ml.
[4] Opensence Lab, "Applying machine learning to continuous delivery," 2019, url. https://opensenselabs.com/blog/articles/ml-continuousdelivery.
[5] S. Garg, P. Pundir, G. Rathee, P.K. Gupta, S. Garg, and S. Ahlawat, "On continuous integration / continuous delivery for automated deployment of machine learning Models Using MLOps," IEEE Fourth International Conference on Artificial Intelligence and Knowledge Engineering , pp. 25-28, 2021.
[6] A. Yadav, M. Garg, and R. Mehra, "Docker containers versus virtual machine-based virtualization," Proceedings of IEMIS, vol 3, pp. 141-150, 2018.
[7] E. Strickland, "The turbulent past and uncertain future of AI: Is there a way out of AI's boom-and-bust cycle?," IEEE Spectr, vol. 58, no. 10, pp. 26-31, 2021.
[8] M. Elkhodr, S. Shahrestani, and H. Cheung, "Internet of things applications: current and future development," Innovative Research and Applications in Next-Generation High Performance Computing, pp. 397-427, 2016.
[9] H. Wang, "Entity resolution on cloud," Innovative Techniques and Applications of Entity Resolution, pp.222-235, 2014.
[10] J. Yao, X. Liu, G. Zhu, and L. Sha, "Controller redundancy design for cyber-physical systems," Cyber-Physical Systems, pp. 61-86, 2016.
[11] A. Biermann, "Fundamental mechanisms in machine learning and inductive inference," Fundamental of Artificial Intelligence, vol. 232, pp. 133-169, 1986.
[12] Y. Chen, "IoT, cloud, big data and AI in interdisciplinary Domains," Simulation Modelling Practice and Theory, vol. 102, 2020.
[13] D. Zhang, S. Mishra, E. Brynjolfsson, J. Etchemendy, D. Ganguli, B. Grosz, T. Lyons, J. Manyika, M. Sellitto, Y. Shoham, J. Clark, and R. Perrault, "The AI index 2019 annual report," AI Index Steering Committee, 2019.
[14] Winbond, "擁抱 Edge AI 新時代 運算/儲存更貼近資料所在," 2021, url. https://www.eettaiwan.com/20210624nt41-edge-ai-moving-computing-storage-closer-to- data/?fbclid=IwAR1VXlC8d-jsRJRlUD6dMGzIPmk9EyL72uazIHCbvBiZLV1DtJmJtDj66VE.
[15] Y. Lin, and S. Tsai, "Country report on IT services — Taiwan," 2021, url: https://www.idc.com/getdoc.jsp?containerId=AP48342822.
[16] A. Popovici, G. Alonso, and T. Gross, "Spontaneous container services," ECOOP Object-Oriented Programming, pp. 29-53, 2003.
[17] J. Zhang, J. Arinez, Q. Chang, R. Gao, and C. Xu, "Artificial intelligence in advanced manufacturing: current status and future outlook," Journal of Manufacturing Science and Engineering, pp. 1-53, 2020.
[18]SASInstituteInc,"工廠想導入 AI 虛擬量測提升產品良率，在實踐前應注意什麼 要點?," 2022, url. https://buzzorange.com/techorange/2022/06/06/ai-virtual-metrology/. [19] T. McClean, "What is edge computing and why does it matter for AI ?," 2021, url. https://blog.adlinktech.com/2021/09/21/what-edge-computing-why-does-it-matter-for-ai/. [20] ABI Research, "Hardware vendors will win big in meeting the demand for edge AI hardware," 2018, url. https://www.abiresearch.com/press/hardware-vendors-will-win-big- meeting-demand-edge-ai-hardware/.
[21] C. Goncalves, E. Andrade, G. Callou, and B. Nogueira, "Evaluation of performance, energy consumption and cost for environments based on containers and virtual machines," Revista Brasileira de Computacao Aplicada, vol. 13, pp. 11-26, 2021.
[22] P. Zhao, P. Wang, X. Yang, and J. Lin, "Towards cost-efficient edge intelligent computing with elastic deployment of container-based microservices," IEEE Access, vol. 8, pp. 102947-102957, 2020.
[23] X. Niu, K. Yang, G. Zhang, Z. Yang, and X. Hu, "A pretraining-retraining strategy of deep learning improves cell-specific enhancer predictions," Frontiers in Genetices, vol. 10, 2020.
[24] M. Khoda, T. Imam, J. Kamruzzaman, I. Gondal, and A. Rahman, "Robust malware defense in industrial IoT applications using machine learning with selective adversarial samples," IEEE Transaction on Industry Applications, vol. 56, pp. 4415-4424, 2020.
[25] W. Kim, and C. Youn, "Cooperative scheduling schemes for explainable DNN acceleration in satellite image analysis and retraining," IEEE Transactions on Parallel and Distributed Systems, vol. 33, pp. 1605-1618, 2022.
[26] G. Moallem, D.D. Pathirage, J. Reznick, J. Gallagher, and H. Sari-Sarraf, "An explainable deep vision system for animal classification and detection in trail-camera images with automatic post-deployment retraining," Knowledge-Based Systems, vol. 216, 2021. [27] LATC Inc, "工具機熱變位 AI 溫補建模與遠端修模整體解決方案," 2022, url. http://www.i-latc.com/.
[28] FEELER lnc, "擺頭式五軸加工機 UB-660," 2022, url. https://www.feeler.com/pdtdetail.php?c=&id=302.
[29] UNIKO's Hardware, "華碩展出 ASUS IoT AISVision 智慧沖壓瑕疵檢測解決方 案," 2022, url. https://unikoshardware.com/2022/02/asus-iot-aisvision-pr-2022-02.html. [30] M. Shen, "日本國際工具機展登場 上銀參展規模居台灣廠商聲勢之冠," 2018, url. https://www.chinatimes.com/realtimenews/20181101002346-260410?chdtv.
[31] D. Beimborn, T. Miletzki, and S. Wenzel, "Platform as a service (PaaS)," Business & Information Systems Engineering, vol. 3, pp. 381-384, 2011.
[32] B. Sotomayor, R. Montero, I. Llorente, and I. Foster, "Virtual infrastructure management in private and hybrid clouds," IEEE Internet Computing, vol. 13, pp. 14-22, 2009.
[33] K. Gillani, and J. Lee, "Comparison of linux virtual machines and containers for a service migration in 5G multi-access edge computing," ICT Express, vol. 6, pp. 1-2, 2020. [34] E. Nakagawa, P. Antonino, F. Schnicke, R. Capilla, T. Kuhn, and P. Liggesmeyer, "Industry 4.0 reference architectures: state of the art and future trends," Computers & Industrial Engineering, vol. 156, pp. 107241, 2021.
[35] K. Olorunnife, K. Lee, and J. Kua, "Automatic failure recovery for container-based IoT edge applications," Electronics, vol. 10, 2021.
[36] G. Tihfon, S. Park, J. Kim, and Y. Kim, "An efficient multi-task PaaS cloud infrastructure based on docker and AWS ECS for application eeployment," Cluster Computing - The Journal of Networks Software Tools and Applications, vol. 19, pp. 1585- 1597, 2016.
[37] P. Quint, and N. Kratzke, "Towards a lightweight multi-cloud DSL for elastic and transferable cloud-native applications," 8th International Conference on Cloud Computing and Services Science, 2018.
[38] D. Shih, T. Wu, W. Liu, and P. Shih, "An Azure ACES early warning system for air quality index deteriorating," Int J Environ Res Public Health, vol. 16, pp. 4679, 2019.
[39] X. Dong, S. Han, A. Wang, and K. Shang, "Online inertial machine learning for sensor array long-term drift compensation," Chemosensors, vol. 9, 2021.
[40] M. Al-Rakhami, A. Gumaei, M. Hassan, A. Alamri, M. Alhussein, M. Razzaque, and G. Fortino, "A deep learning-based edge-fog-cloud framework for driving behavior management," Computers & Electrical Engineering, vol. 96, pp. 107573, 2021.
[41] AWS Services, "Automate model retraining with amazon sagemaker pipelines when drift is detected," AWS Machine Learning Blog, 2022, url. https://aws.amazon.com/tw/blogs/machine-learning/automate-model-retraining-with- amazon-sagemaker-pipelines-when-drift-is-detected/.
[42] D. Linthicum, "Practical use of microservices in moving workloads to the cloud," IEEE Cloud Computing, vol. 3, pp. 6-9, 2016.
[43] Microsoft, "Machine learning operations (MLOps) framework to upscale machine learning lifecycle with Azure machine learning," 2022, url. https://docs.microsoft.com/en- us/azure/architecture/example-scenario/mlops/mlops-technical-paper.
[44] K. Wakino, and Y. konishi, "Bandpass filter with dielectric materials used for broadcasting channel filter," IEEE Transaction on Broadcasting, vol. BC-26, pp. 1-6, 1980. [45] M. Mascheroni, and E. Irrazabal, "Continuous testing and solutions for testing problems in continuous delivery: A systematic literature review," Computaion y Sistemas, vol. 22, pp. 1009-1038, 2018.
[46] Y. Vlasov, N. Khrystenko, and D. Uzun, "Analysis of modern continuous integration/deployment workflows based on virtualization tools and containerization techniques," Integrated Computer Technologies in Mechanical Engineering, pp. 538-549, 2020.
[47] Slintel lnc, "Source Code Management," 2022, url. https://www.slintel.com/tech/source-code-management/github-market-share.
[48] S. Bokhari, "The linux ioerating system," Computer, vol. 28, pp. 74-79, 1995
[49] W. Limited, "Awesome-GitOps," 2021, url. https://github.com/weaveworks/awesome- gitops.
[50] Cloud Native Computing Foundation, "Annual survey 2021," CNCF, 2021, url. https://www.cncf.io/reports/cncf-annual-survey-2021/.
[51] ITRI, " AI 眾 智 式 智 慧 製 造 系 統 ," 2018, url. http://official.meetbao.net/techforce/ai%E7%9C%BE%E6%99%BA%E5%BC%8F%E6% 99%BA%E6%85%A7%E8%A3%BD%E9%80%A0%E7%B3%BB%E7%B5%B1/.
[52] Kaggle lnc, "CNC mill tool wear," 2018, url. https://www.kaggle.com/datasets/shasun/tool-wear-detection-in-cnc-mill.