Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/1708
標題: 工具機最佳加工參數的設定
Determination of Optimum Cutting Parameters of Machine Tools
作者: 溫程雄
Wen, Cherng Shyong
關鍵字: cutting parameters;切削參數;optimum;neural networks;taguchi method;genetic algorithm;最佳化;類神經網路;田口法;遺傳演算法
出版社: 機械工程學系
摘要: 
本文針對切削參數最佳化的問題,應用建構的類神經網路模
式,預測目標函數和限制函數的值,以求得最佳的切削參數值。 在建
構類神經網路模式的過程中,本文應用田口法和遺傳演算法 建構最佳
的類神經網路架構,以得到較佳的預測結果。 應用田口法建
構最佳的類神經網路架構,是以類神經網路的隱 藏層層數、隱藏層神
經元的個數、輸入層的輸入代表、學習的樣 本數目和類神經網路的學
習速率等五個控制因子進行實驗,依據 實驗分析的結果選定各控制因
子的最佳水準值。而遺傳演算法方 面,是以類神經網路的隱藏層層數
、隱藏層神經元的個數、網路 的學習速率、網路的慣性量和轉換函數
的斜率等五個設計變數, 藉由遺傳演算法搜尋產生最佳的類神經網路
架構。運用田口法和 遺傳演算法建構出車削和銑削的最佳類神經網路
架構後,再以商 用軟體DOT/DOC中的連續線性規劃法及離散變數法分
別進行車削 和銑削切削參數最佳化的設計,並比較與數學式所求得的
最佳化 設計結果。

This thesis deals with the determination of optimum
cutting parameters. The artificial neural networks are
constructed to predict values of constraint functions and
objective function. To yield accurate neural networks,
Taguchi method as well as genetic algorithm are used to
construct better neural networks. The major factors
used in Taguchi method to construct neural networks include
the number of hidden layers, the number of neurons in hidden
layers, the input representation scheme, the training sample
size and the learning coefficients. The better levels for
these factors are obtained based on the experiments. For
genetic algorithm approach, the variables of neural networks
are the number of hidden layers, the number of neurons in
hidden layers, the laerning coefficients, the momentum of
learning process, and the slope of the activation functions.
The optimum values of these discrete varibales are found
using GA. Aftter constructing the neural networks for
objective as well as constraint functions, the optimization
slover DOT/DOC is utilized to solve the mixed-variable turning
and milling problems. The obtained optimum solutions using
neural networks are compared with simulated experimental
results.
URI: http://hdl.handle.net/11455/1708
Appears in Collections:機械工程學系所

Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.