- 學位論文
基因演算法於工程最佳化問題之研究
A Study on Genetic Algorithm for Engineering Optimization Problems
摘要
在工程設計上，有許多複雜性高的問題，擁有多參數、多個設計目標和多限制條件，使用傳統方式費時，很難滿足特殊的最佳化問題，且因現代電腦科學蓬勃發展，可將演算法良好的實現，更為迅速、有效求得複雜問題的最佳解。 本研究發展一最佳化程序，使用基因演算法於工程最佳化問題求解，分別應用於光學和磁浮領域，光學領域：在直下式發光二極體背光模組加入反射結構設計，希望藉由控制邊射型紅、綠、藍三色發光二極體之光線，達到良好的輝度、色彩均勻度和軸向增益，減少使用任何光學元件而能達到背光模組系統所需之光學表現，模擬結果顯示在背光模組腔體中加入雙三角形反射結構，最佳化參數結果可使輝度均勻度達到91.87%、最大色差為0.042，以及軸向發光強度能有50.23%的增加。應用在磁浮領域：八極徑向主動式磁浮軸承參數設計，軸承在不同使用需求下有不同之設計參數和目標，並有多個限制條件及多種設計目標方案，希望能有效簡化設計流程，求得最佳設計，並進行模擬、實驗驗證其最佳化結果之可行性。模擬結果可依不同需求達到不同目標，而實驗驗證，在求解最大磁浮力問題上，轉子中心位置最大磁浮力為200.5N，模擬最大磁浮力為207.12 N，模擬與實際誤差為3.19%。 結果顯示此最佳化程序可有效解決工程設計問題，避免設計的重複試誤、取代傳統設計方法，減少重新設計之時間，降低製造成本、增加效益和方便性。
並列摘要
In engineering design, there have many complicated problems, which have multiple parameters, multiple design objective and limits conditions. And it spends time via traditional methods and hard to satisfy the specific optimization problems. Because the computer science flourishes, we could implement the algorithm very well and solve the complicated optimization problems more quickly and effectively. This paper developed an optimal program via a Genetic Algorithm to solve the engineering problems. And it applied in two different fields, Optics and Magnetic levitation：A large size direct LED BLU consisted of double-triangular structures was proposed in the paper to control the light from red, green and blue side emitting LEDs to improve brightness uniformity and color uniformity in the out-light surface of backlight unit. From the simulation results showed that the brightness uniformity and color uniformity were 91.87% and 0.042, on-axis luminous intensity was 50.23% with the double-triangular structures on BLU；Magnetic levitation：In different requires has different design parameters and objectives, the design of an eight-pole radial active magnetic bearing has many limit conditions and objective functions. This research hopes to simplify the design process effectively and to find the best design. And then ones verify the optimal results through simulation and experiment. In the maximal magnetic force case, the experiment shows the real maximal magnetic force was 200.5N and the simulation was 207.12N. The errors between simulation and real was 3.19%. The optimal program via a Genetic Algorithm could solve the engineering problems effectively, avoid repeating trial and error, replace the traditional design methods. And also could reduce the time and manufacturing costs, increase efficiency and convenience.