透過您的圖書館登入
IP:13.59.36.203
  • 學位論文

整合機器學習與田口方法建立製程參數優化的預測模型-以CNC鑽孔製程為例

The Prediction Model for Process Parameter Optimization by Integration of Machine Learning and Taguchi Method - A Case Study of CNC Drilling Process

指導教授 : 江瑞清

摘要


近年來人工智慧成長快速,在各行各業皆如火如荼的推展開來,進而重塑社會發展,也隨著市場的改變,從以前的大量生產製造變成追求少量且高品質的取向,為了符合需求,必須提高效率與品質並降低成本,如此才能擁有企業競爭力。因此本研究利用田口方法優化CNC鑽孔製程的加工參數,以提升孔徑大小的加工品質,並將其數據投入機器學習建立預測模型,以80%的資料進行訓練,20%的資料進行驗證,透過驗證集得知效果,本研究使用類神經網路、隨機森林、支援向量迴歸,與這三種演算法皆利用田口方法優化其內部參數,因此總共使用六種方法建立模型,期望透過模型,能投入參數後預測出孔徑大小,以便未來進行參數調整時能快速預測出品質狀況,達到高效率與低成本的目標。結果顯示,此六種方法的實驗中,類神經網路與田口結合類神經的預測模型最為精準,兩者並沒有顯著差異,其MSE為22.5398與22.5688,但在田口結合類神經的方法中,收斂速度最為快速,證明此方法利用在鑽孔製程預測中是具有可行性的,並能大幅減少實驗次數與成本。

並列摘要


In recent years, artificial intelligence has grown rapidly that promote the society development. With the change of the market needs, the orientation of manufacturing production has turn into pursue produce small amount and high quality. In order to make enterprise more competitive, the need for improving efficiency quality and reduce costs. This study uses Taguchi method to optimize the processing parameters of the CNC drilling process to improve the processing quality of the diameter of aperture, and put its data into machine learning to build a predict model, training with 80% of the data, and 20% of the data for verification. The validation set knows the effect. This study uses six methods as following are Neural Network、Random Forest、Support Vector Regression and Taguchi method combined these three algorithms to establish a model. It is expected that the parameters can be used to predict the diameter of aperture through the model, so that when the parameters are adjusted in the future, they can be quickly predicted The quality condition achieves the goals of high efficiency and low cost. The results show that in the experiments of these four methods, Neural Network and Taguchi combined Neural Network prediction model are the most accurate, and there is nonsignificant. However, Taguchi combined Neural Network method, the convergence rate is the fastest, and the MSE are 22.5398 and 22.5688 prove that this method is feasible in drilling process prediction, and can greatly reduce the number of experiments and costs.

參考文獻


中文參考文獻
吳秋霖,2019,“利用類神經網路與田口方法參數優化與模型預測—以 CNC 銑切製程為例” ,中原大學工業與系統工程學系,碩士論文。
汪惠健(譯) (2007)。 “類神經網路設計”(原著:Neural network design.)。新北:高立出版社。(原著出版年:2004)。
邱雲堯、陳佳萬、張安心、鄭偉盛、陳木榮、何世偉、喻立信(譯)(1998)。
“Manufacturing Engineering and Technology”(原作者:Serope Kalpakjian)。臺北:文大揚出版社。(原著出版年:1989)。

延伸閱讀