Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/1696
標題: 可變設計空間的形勢最佳化
Topology Optimization with Variable Design Space
作者: 林嘉洋
Lin, Chia-Yang
關鍵字: topology;形勢;genetic algorithm;Taguchi method;optimization;neural network;遺傳演算法;田口法;最佳化;類神經網路
出版社: 機械工程學系
摘要: 
本文探討在滿足材料使用量和清晰度束制條件下,求解順從度最小化
、基礎特徵值最大化與利用模糊理論求解多目標結構形勢最佳化。本文設
計空間的邊界以及邊界條件可微幅調整以達到最好的設計。田口法與遺傳
演算法分別用來決定最佳的設計邊界。田口法將每個邊界訂出五種水準亦
即五個不同位置,以很少的實驗即可決定最佳的邊界位置。而遺傳演算法
的特性很容易處理代表邊界位置的離散變數,但是此法需要大量的計算分
析,若每個個體皆使用有限元素分析則需要很長的時間,因此類神經網路
被用來取代真正的結構分析,以增加運算效率。 本文之形勢最佳化求
解過程是在UNIX作業系統環境中以連續線性規劃法執行,並將理論基礎以
六個不同情況的範例驗證,且針對不同範例討論與比較田口法與遺傳演算
法之結果,並將得到的最佳設計空間之形勢圖以P3/PATRAN軟體繪出。

The topology optimization under material and clearness
constraints is studied in this thesis. The objectives include
minimum compliance and maximum fundamental eigenvalue. The
multi-objective topology optimization using fuzzy decision
making is also explored. The main purpose of this thesis is to
find optimum topology in a variable-boundary design space.
Taguchi method and genetic algorithm are used to find the
optimum boundaries of the design space. Five levels of boundary
locations are given in Taguchi. A minimum number of experiments
based on orthogonal table are performed to determine the optimum
boundary locations. Genetic algorithm is a computational
intensive approach. To save computational time, artificial
neural network is trained to do the structural analyses. A
Unix shell script is developed to accomplish the design
optimization loop. The sequential linear programming algorithm
is utilized to solve the optimization problem. Six different
examples demonstrate applications of the theory. The results of
each example by Taguchi method and genetic algorithm are
compared. P3/PATRAN is used to show the topologies in optimal
design spaces.
URI: http://hdl.handle.net/11455/1696
Appears in Collections:機械工程學系所

Show full item record
 
TAIR Related Article

Google ScholarTM

Check


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