隨著人工智慧(Artificial Intelligent, AI)和物聯網的發展，工業物聯網(Industrial Internet of Things, IIoT)和智慧工廠的時代來臨。智慧化和自動化製造成為目前的趨勢。在工業製造環境中，注塑成型技術是一個廣泛運用的製程，由於它相對的低成本卻能生產複雜工件的特性，使得塑膠射出成型市場日漸茁壯。但同時，過程中輸入的大量參數使得控制和監控整個製造過程變得複雜且耗費人力。本篇論文分析來自台達泰廠及東莞廠的注塑機台資料集，並使用深度神經網路(Deep Neural Network, DNN)學習輸入參數和輸出品質的關聯性，預測輸出品質。根據所訓練模型，提出評估良率的方法，更進一步使用粒子群演算法和歷史參數推薦來優化參數來幫助節省工業現場的調機時間，以及避免不良品發生，最終目的是提高產線良率。實驗結果顯示，DNN在此類問題上比起線性分類器有更好的能力召回錯誤樣本，且在不同資料集的表現上也較具穩健性。並探討深度學習應用於各種不同資料分布的資料集可能出現的問題和原因。

With the development of Artificial Intelligence (AI) and the Internet of Things (IoT), the era of the Industrial IoT (IIoT) and smart factories has arrived. Intelligent and automated manufacturing has become the current trend. Injection molding technology is widely used in industrial manufacturing environments. The plastic injection molding market grows more substantial due to its relatively low cost and the ability to produce complex workpieces. But, the large number of input parameters in the machining process makes controlling and monitoring the entire manufacturing process complex and labor-intensive. This thesis analyzes the data sets of injection molding from Delta Pacific and Dongguan plants. We use a deep neural network (DNN) to learn the correlation between the input parameters and output quality and predict output quality. According to the trained model, a method for evaluating yield quality is proposed. The particle swarm algorithm and historical parameter recommendation are further used to optimize the production parameters to help save the tuning time in the injection molding and avoid defective products, with the ultimate goal of improving the yield of the production line. Experimental results show that DNN has a better ability to recall erroneous samples than linear classifiers on such problems and is more robust in the performance of different data sets. We also explore the issues and reasons that may arise when deep learning is applied to data sets with varying data distributions.

誌謝
摘要 i
Abstract ii
1.介紹 1
2.相關研究 5
3.注塑機良率預測與參數推薦方法 9
4.注塑機良率預測與參數推薦實驗 19
5.結論 27
6.問題與討論 29
7.參考文獻 34

[1]O. Veligorskyi, R. Chakirov, M. Khomenko, and Y. Vagapov, “Artificial neural networkmotor control for full-electric injection moulding machine,” in 2019 IEEE InternationalConference on Industrial Technology (ICIT), pp. 60–65, IEEE, 2019.[2]P. Pierleoni, L. Palma, A. Belli, and L. Sabbatini, “Using plastic injection moulding ma-chine process parameters for predictive maintenance purposes,” in 2020 InternationalConference on Intelligent Engineering and Management (ICIEM), pp. 115–120, IEEE,2020.[3]C.-C.Chiu,C.-T.Su,G.-S.Yang,J.-S.Huang,S.-C.Chen,andN.-T.Cheng,“Selectionofoptimal parameters in gas-assisted injection moulding using a neural network model andthe taguchi method,” International Journal of Quality Science, 1997.[4]W.-C. Chen, G.-L. Fu, P.-H. Tai, W.-J. Deng, and Y.-C. Fan, “Ann and ga-based processparameter optimization for mimo plastic injection molding,” in 2007 International Con-ference on Machine Learning and Cybernetics, vol. 4, pp. 1909–1917, IEEE, 2007.[5]W.Chen,M.Wang,G.Fu,andC.Chen,“Optimizationofplasticinjectionmoldingprocessviataguchi＇sparameterdesignmethod, bpnn, anddfp,” in2008InternationalConferenceon Machine Learning and Cybernetics, vol. 6, pp. 3315–3321, IEEE, 2008.[6]A. Tellaeche and R. Arana, “Machine learning algorithms for quality control in plasticmolding industry,” in 2013 IEEE 18th Conference on Emerging Technologies & FactoryAutomation (ETFA), pp. 1–4, IEEE, 2013.[7]Y. Bei, Y. Zhou, X. Cheng, and E. Liu, “Qualified rate prediction of typical workpieces indiscrete manufacturing process,” in 2020 International Conference on Big Data, ArtificialIntelligence and Internet of Things Engineering (ICBAIE), pp. 409–412, IEEE, 2020.[8]M. Charest, R. Finn, and R. Dubay, “Integration of artificial intelligence in an injectionmolding process for on-line process parameter adjustment,” in 2018 Annual IEEE Inter-national Systems Conference (SysCon), pp. 1–6, IEEE, 2018.[9]S. Kollmann, A. Estaji, A. Bratukhin, A. Wendt, and T. Sauter, “Comparison of prepro-cessors for machine learning in the predictive maintenance domain,” in 2020 IEEE 29thInternational Symposium on Industrial Electronics (ISIE), pp. 49–54, IEEE, 2020.[10]P. Nagorny, M. Pillet, E. Pairel, R. Le Goff, J. Loureaux, M. Wali, and P. Kiener, “Qualityprediction in injection molding,” in 2017 IEEE International Conference on Computa-tional Intelligence and Virtual Environments for Measurement Systems and Applications(CIVEMSA), pp. 141–146, IEEE, 2017.34
[11]V. Ketonen and J. O. Blech, “Anomaly detection for injection molding using probabilisticdeep learning,” in 2021 4th IEEE International Conference on Industrial Cyber-PhysicalSystems (ICPS), pp. 70–77, IEEE, 2021.[12]H. Tercan, A. Guajardo, J. Heinisch, T. Thiele, C. Hopmann, and T. Meisen, “Transfer-learning: Bridging the gap between real and simulation data for machine learning in in-jection molding,” Procedia Cirp, vol. 72, pp. 185–190, 2018.[13]H. Tercan, A. Guajardo, and T. Meisen, “Industrial transfer learning: Boosting machinelearning in production,” in 2019 IEEE 17th International Conference on Industrial Infor-matics (INDIN), vol. 1, pp. 274–279, IEEE, 2019.[14]A. Halevy, P. Norvig, and F. Pereira, “The unreasonable effectiveness of data,” IEEEIntelligent Systems, vol. 24, no. 2, pp. 8–12, 2009.[15]L. S. Freedman and D. Pee, “Return to a note on screening regression equations,” TheAmerican Statistician, vol. 43, no. 4, pp. 279–282, 1989.[16]F. Wu, X.-Y. Jing, S. Shan, W. Zuo, and J.-Y. Yang, “Multiset feature learning for highlyimbalanced data classification,” in Thirty-First AAAI Conference on Artificial Intelli-gence, 2017.[17]S. Shaikh, L. Changan, and M. R. Malik, “A baseline accuracy classification method toover-come the over-fitting problem for class-imbalanced defect-prone datasets model,” in2020 IEEE International Systems Conference (SysCon), pp. 1–7, IEEE, 2020.[18]X.-Y. Liu, J. Wu, and Z.-H. Zhou, “Exploratory undersampling for class-imbalance learn-ing,”IEEETransactionsonSystems,Man,andCybernetics,PartB(Cybernetics),vol.39,no. 2, pp. 539–550, 2008.[19]S.Xu,W.F.Lu,X.Li,andK.J.Lee,“Advancedllemethodfordimensionreductionusingnonlinear manufacturing data,” in 2018 IEEE 23rd International Conference on EmergingTechnologies and Factory Automation (ETFA), vol. 1, pp. 395–401, IEEE, 2018.[20]K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are univer-sal approximators,” Neural networks, vol. 2, no. 5, pp. 359–366, 1989.[21]G.Lemaître,F.Nogueira,andC.K.Aridas,“Imbalanced-learn: Apythontoolboxtotacklethe curse of imbalanced datasets in machine learning,” The Journal of Machine LearningResearch, vol. 18, no. 1, pp. 559–563, 2017.[22]M. Haldar, “How much training data do you need,” 2017.[23]X. Zhu, C. Vondrick, C. C. Fowlkes, and D. Ramanan, “Do we need more training data?,”International Journal of Computer Vision, vol. 119, no. 1, pp. 76–92, 2016.