- 學位論文
利用機器學習方法預測織布製程參數於虛實整合系統
Parameter Prediction for Weaving Process in Cyber-Physical Systems Using Machine Learning
摘要
在工業4.0的浪潮之下，傳統紡織工廠進行智慧化轉型是維持競爭力的重要關鍵。智慧化轉型其中一項關鍵技術便是虛實整合系統。虛實整合系統將產業鏈中的每一個實體元素在虛擬端產生一個鏡像模型 (Cyber Twin)，達到對實體元素行為進行預測與管理的目的。虛實整合系統需具備自省性、自比較性，意即對自身狀態變化有所意識並且比較類似實體元素進行預測，更重要的是「自重構性」，根據系統目標進行自我強化，建構出能反映實體狀態的鏡像虛擬模型。 為實現織布機台參數設定智慧化這個目的，本研究目標有兩項，第一項為開發一套雲端資料分析系統InAnalysis。第二項將針對織布段機台參數設定以資料分析模型為基礎進行參數預測智慧化。從資料科學的角度，整合織造過程中準備段與織布段的歷史數據，經由機器學習回歸演算法，建構出能反映實體狀態的織布機台參數預測模型，並且透過十摺交叉驗證法（10-fold Cross Validation）了解模型表現，達到虛實整合系統所需的「自省性」、「自比較性」。進一步導入詢問式學習的概念，透過添加學習獲得的新案例資料，進而有效地強化整體虛擬預測模型的表現，得到更好的現實策略，這即為虛實整合系統最重要的特質「自重構性」。實驗結果顯示，基本織布機台操作參數回歸預測模型的MSE（Mean Square Error）僅0.000165。並且能透過詢問式學習強化織布參數預測模型以及驗布品質預測模型。 未來將用InAnalysis的API模組設計一套紡織廠的機台操作參數推薦系統（Operation Parameter Recommendation System, OPRS）。透過資料生成的機器學習預測模型，協助技術人員設定操作參數，將整個製程的決策過程智慧化。並解決紡織技術人才不斷老化，經驗難以傳承等問題。更進一步，將該系統串聯至整個紡織產業上、中、下游的決策系統，將「接單」、「研發設計」、「供應」、「生產」、「檢驗」、「出貨」、「銷售」等服務全部由虛實整合系統進行高效率的智慧化管理與決策。
並列摘要
In Industry 4.0, the intelligent transformation of traditional textile factories is important to maintain competitiveness. One of the key technologies for smart transformation is the cyber-physical system. The cyber-physical system will generate the cyber twin on the virtual end for each entity element in the industrial chain to achieve the purpose of predicting and managing the behavior of the entity element. The cyber-physical system needs to be self-examination and self-comparative, meaning that it is aware of changes in its own state and more similar entity elements to make predictions. More importantly, it is "self-reconfigurable" and self-enhanced according to system goals. In order to achieve the goal of setting the loom parameters intelligently, this study has two objectives. First one is to development a cloud data analysis system, InAnalysis. The second one will be build a parameters prediction model for the weaving process. From the perspective of data science, the machine learning regression algorithm is used to build a loom parameter prediction model, and 10-fold cross validation is used to understand the performance of the model. This achieved the "self-introspection" and "self-comparison" required by cyber-physical system. Also, the concept of query-based learning is used. And the performance of the prediction model can be effectively enhanced and a better realistic strategy can be obtained. This is the most important feature of the cyber-physical system, self-reconfigurable. The experimental results show that the MSE (Mean Square Error) of the prediction model is only 0. 000165. Moreover, the performance of the parameter prediction model and quality prediction model can be reinforced through query-based learning. In the future, InAnalysis's API will be used in an operation parameter recommendation system (OPRS) and weaving process of decision-making will become intelligent. Further, the system could implement into entire textile industry. Such as, "orders," "R&D," "supplies," "production," "inspection," "shipments," and "sales." All other services are managed efficiently and intelligently by the cyber-physical system.
參考文獻
延伸閱讀
- 李佳芸（2018）。利用機器學習方法預測紡織準備段操作參數〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU201802652
- 蔡佳昇、洪秀婉（2023）。運用資料包絡分析法與平衡計分卡於紡織業經營績效探討。中原企管評論，21(1)，57-87。https://www.airitilibrary.com/Article/Detail?DocID=17298822-N202305160005-00003
- 張權緯（2005）。類神經網路與多變數迴歸之混成方法於工具機熱變形預測〔博士論文，中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu200500444
- 陳建賢（2008）。長纖梭織布業導入企業資源規劃系統所面對產業特性及作業流程與客製化問題之研究〔碩士論文，國立臺北大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0023-0207200810481500
- Wang, X., & Wang, J. (2023). Research on the Application of the Machine Learning Algorithm Based on Parameter Optimization in Network Security Situation Prediction. International Journal of Network Security, 25(2), 245-251. https://doi.org/10.6633/IJNS.202303_25(2).07