刀模製程尺寸精度最佳化

Dimensional Accuracy Optimization of Soft Tooling Process

張智偉(Chih-Wei Chang)

指導教授 ： 戴兢志

大同大學/工程學院/機械工程學系所/碩士(2011年)

摘要

本研究主要目的是探討刀模製程參數對產品尺寸精度的影響。本研究首先運用田口實驗計劃法的異變數分析，探討尺寸精度最佳化問題，再利用倒傳遞類神經網路技術，將田口實驗計劃法規劃的實驗結果，建立尺寸誤差的數學模式。利用遺傳基因演算法將所建立的數學模式，找出直徑350 mm的圓形刀模產生最小尺寸誤差的最佳製程參數。遺傳基因演算法是一個以類神經淘汰和類神經遺傳學為基礎的搜尋演算法。田口實驗計劃法最佳化參數組合為沖模力為100 、緩衝板厚度13 mm、衝程微增量0.3 mm與刀縫刀厚比值0.78。GA最佳化參數組合為沖模力90 、緩衝板厚度10 mm、衝程微增量0.26 mm與刀縫刀厚比值0.75。利用田口最佳化參數進行沖模實驗，尺寸誤差平均值為0.019 mm，利用GA最佳化參數進行沖模實驗，尺寸誤差平均值為0.016 mm，標準使用參數的尺寸誤差平均值為0.023 mm。

關鍵字

類神經網路

並列摘要

The main purpose of this search is to investigate the effects of stamping process parameters on dimensional accuracy for the stamping products. The approach of Taguchi design based on ANOVA analysis is first utilized to investigate the dimensional accuracy optimization problem. Experimental results planned as DOE were used to develop the mathematical model for dimensional error using Back-Propagation Neutral Network technology. The developed mathematical models were then employed with GA, which is a search algorithm based on neutral selection and neutral genetics, to determine the optimal process parameters for a Ø 350 mm circular tooling die that results in minimum dimensional error. The data resulted from optimal analysis are 100 stamping force, 13 mm plate thickness, 0.3 mm increment, 0.78 seam-thickness ratio for Taguchi optimization method and 90 stamping force, 10 mm plate thickness, 0.26 mm increment, 0.75 seam-thickness ratio for GA optimization approach. The experimental results reveal that average value of dimensional error is 0.019 mm for Taguchi optimization method, 0.016 mm for GA optimization approach and 0.023 mm for standard use.

並列關鍵字

neural network

參考文獻

[1] T. Y. Lin C. H. Tseng “Optimization design for artificial neural networks: an example in a bicycle derailleur system ﹐”Engineering Application of Artificial Intelligence﹐Vo1.13﹐pp.3-14(2000).

[2] R. S. Chen H. C. Lin C. Kung﹐“Optimization dimension of PQFD Taguchi method﹐”Composite Structures﹐Vo1.49﹐pp.39-44(2000).

parameters using back propagation and learning vector quantization networks﹐”Journal of Materials Processing Technology﹐Vo1.99 pp.207-218(2000).

[4] X. Li﹐P.K. Venuvinod M. K. Chen﹐“Feed cutting force estimation form the current measurement with hybrid learning﹐”International Journal of Advanced Manufacturing Technology﹐Vo1.16﹐pp.859-862(2000).

[5] T. Ozel and A. Nadgir﹐“prediction of flank wear by using back propagation neural networks modeling when cutting hardened H-13 steel with chamfered and honed CBN tools﹐ ”International Journal of Machine Tools & Manufacture﹐Vo1.42﹐pp.287-297(2002).