| Summary: | 碩士 === 逢甲大學 === 工業工程研究所 === 86 === NC沖床已被廣泛的使用於薄金屬板的加工, 一薄金屬板可能需要200或更
多次的衝擊, 而每次的衝擊須使用其所需的刀具。為了降低生產循環時間
（Production Cycle Time）以提昇生產力, 我們必須對刀具庫中的刀具
及衝孔的順序作有效率的安排, 在安排的同時,尚需考量衝孔順序及刀具
擺置是否符合先後限制式（Precedence Constraint）的要求。NC沖床衝
孔的問題可視為旅行商人問題（Traveling Salesman Problem）及二次指
派問題（Quadratic Assignment Problem）兩個子問題的合併, 而依據過
去經驗, 基因演算法及塔佈搜尋法在解決此兩個子問題時都能得到很好的
效果。雖然此兩種搜尋法各有其獨特的搜尋技巧, 但它們都是可共用的演
算法（Meta Heuristic）, 亦即此兩種演算法可以配合其他演算法則使用
。過去雖然已有許多基因演算法及塔佈搜尋法的應用, 但通常只侷限於單
一演算法的發展, 較少混合的應用。因此, 本研究結合此兩種演算法的優
點, 而發展出混合演算法以解決NC沖床衝孔問題。本研究首先發展一個演
算法來產生起始族群, 其次再發展適合混合演算法之元件, 以增進混合演
算法的執行效率。在此混合演算法中, 我們整合發展之元件, 並以基因演
算法為主要搜尋架構, 而族群中的個體均進行塔佈搜尋的程序, 如此以一
代一代的方式進行搜尋, 直到解沒有改進為止。本研究蒐集四個業界較常
使用NC沖床刀具庫, 比較混合演算法與Walas & Askin方法的優劣, 比較
結果顯示, 混合演算法雖然經搜尋後解的改進量並不大, 但起始族群中最
好的解甚優於Walas & Askin所得的解。
NC turret punch presses are widely used for the production of
sheet metal parts.A part may require 200 or more punches, and
each punch uses a specific tool.In order to reduce production
cycle time and improve the productivity, we mustarrange loaded
tools and hit sequence efficiently. Besides, the hit sequence
and tool arrangement have to satisfy some precedence
constraints. The problem of determining hit sequence and tool
arrangement can be treatedas a combination of QAP(Quadratic
Assignment Problem) and TSP(Traveling Salesman Problem).Past
studies demonstrate that genetic algorithms and tabu search
perform well when solving these two subproblems. Thus, we
combine the advantage of these two algorithms, and develop a
hybrid approach to solve NC punch presses problem.In this
research, a procedure of generating the initial population and
efficientcomponents will develop to enhance the solution quality
and computational efficiency. In this hybrid method, we
integrate developed components, geneticalgorithms is used as the
main structure of the search procedure and each individual is
searched by using tabu search. Iterate this procedure till the
determined sequence has no improvement. In addition, four NC
punch press turrets are used to compare the proposed approach
with Walas and Askin's heuristic.Computational results reveal
that the best solution of initial population inproposed approach
is far better than Walas and Askin*s solution, although its
improvement of search shows low performance.
|
|---|
