類神經網路遷移學習多目標演化演算法線切割放電加工參數最佳化專家系統;Artificial Neural Network Transfer Learning Multi-objective Evolutionary Algorithm for WEDM Optimizing WEDM Machining Parameter Optimization Expert System

NCU Institutional Repository > 資訊電機學院 > 資訊工程學系 > 研究計畫 > Item 987654321/84717

請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/84717

題名:	類神經網路遷移學習多目標演化演算法線切割放電加工參數最佳化專家系統;Artificial Neural Network Transfer Learning Multi-objective Evolutionary Algorithm for WEDM Optimizing WEDM Machining Parameter Optimization Expert System
作者:	江振瑞;李朱育;崔海平
貢獻者:	資訊工程學系
關鍵詞:	類神經網路;實驗設計;加工參數最佳化專家系統;多目標演化演算法;非支配排序基因演算法;代理模型;遷移學習;線切割放電加工;Artificial Neural Network;Design Of Experiment;Machining Parameter Optimization Expert System;Multi-Objective Evolutionary Algorithm;Non-dominated Sorting Genetic Algorithm;Surrogate Model;Transfer Learning;Wire Electrical Discharge Machining
日期:	2020-12-08
上傳時間:	2020-12-09 10:46:53 (UTC+8)
出版者:	科技部
摘要:	本計畫與慶鴻機電(CHMER)公司合作，預計建構完成線切割放電加工參數最佳化專家系統(Wire Electrical Discharge Machining Parameter Optimization Expert System, WEDMPOES)。其基本概念為在多目標演化演算法框架(Multi-Objective Evolutionary Algorithm Framework)中，以類神經網路(Artificial Neural Network, ANN)，透過實驗設計(Design Of Experiment, DOE)收集適當資料建構資料驅動代理模型(Data-driven Surrogate Model)，並採用非支配排序基因演算法II(Non-dominated Sorting Genetic Algorithm II, NSGA-II)輸出逼近多目標最佳化之加工參數組合建議，完成建構線切割放電加工參數最佳化專家系統(WEDM MPOES)的目標。另外，預計以權重凍結(Weight Freezing)之遷移學習(Transfer Learning)概念，使用小量資料將類神經網路代理模型在不同機台或不同生產模式之間移轉，以擴增加工參數最佳化專家系統適用範圍。完成後的線切割放電加工參數最佳化專家系統，可以讓使用者輸入預定的生產品質，如特定的加工速度(進給率)、工件精度與表面粗糙度等，可自動輸出對應的最佳加工參數設定。我們將進行實驗驗證專家系統的效能，預期在實際生產時，專家系統所輸出的加工參數設定可以達成使用者指定的所有生產品質，或其中每一項品質的誤差都在5%以下。 ;This project is in cooperation with CHMER company, and it is expected to complete the Wire Electrical Discharge Machining Parameter Optimization Expert System (WEDMPOES). The basic concept is to use the Artificial Neural Network (ANN) as the Surrogate Model (SM) in the Multi-Objective Evolutionary Algorithm (MOEA) framework. The ANN is trained by appropriate data collected according the Design of Experiment (DOE) technique. The Non-dominated Sorting Genetic Algorithm II(NSGA-II) is adopted to generate parameter combinations to approximate the multi-objective optimization. In addition, the weight-freezing transfer learning concept is used to transfer the ANN surrogate model between different machines or different machining modes for extending the expert system application scope. After completion, the expert system allows users to input pre-determined manufacturing quality items, such as specific processing speed (feed rate), workpiece size accuracy and surface roughness, to automatically output the corresponding optimal machining parameter settings. We will conduct experiments to verify the effectiveness of the expert system. It is expected that in actual manufacturing, the machining parameter settings output by the expert system can achieve all the manufacturing quality items specified by the user, or the deviation of each quality item is less than 9%.
關聯:	財團法人國家實驗研究院科技政策研究與資訊中心
顯示於類別:	[資訊工程學系] 研究計畫

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	165	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....