本研究將採用遺傳演算法和類神經網路的混合法作結構的最佳化設計。遺傳演算法乃依據生物界之“物競天擇、適者生存”之理論所發展出的一套數學最佳化理論。藉由複製、交配與突變等三個基本運算元來搜尋空間上之最佳點。類神經網路乃依據生物神經網路訊息傳遞的資訊處理方式所創立的一種學習模型，只要利用了巨量的人工神經元加以連結，再經由反覆的訓練以做出適當的判斷，並將所做的訓練記憶起來，就可以求得最佳值。因此本文利用遺傳演算法優越的搜尋能力，搜尋出較適合類神經網路訓練的權重值，並透過遺傳演算法的演算過程，使類神經網路的訓練過程能夠更快速、更準確，以利於執行結構的最佳化設計。 數值分析中將使用ANSYS有限元素分析軟體作結構分析，範例中包含壓電複材層板及一般材料桁架結構的輕量化設計和三次元量床之動態與靜態之最佳化設計。期望本研究之結果能對結構的設計提供一個實用且有效率的方法。 A hybrid method combined Genetic Algorithm and Artificial Neural Network will be adopted in this study. Genetic Algorithm is a well developed mathematic optimization theory based on the theory of “Survival of the fittest” in nature. The new design can be obtained by three basic operators: reproduction, crossover, and mutation. Neural network is a self learning model basing on the translations and procedures of creature-like neural network information. Once the connection from tremendous artificial neuron was established, the network could make the judgement of parameters by repeating trainings. By the superior searching ability of Genetic Algorithm, the optimum weights for neural network can be obtained precisely and efficiently. Structural analysis would be performed by a finite element analysis software ANSYS in this study. The optimum design of the piezoelectric composite material, truss and coordinate measuring machine structures would be demonstrated in the numerical analysis. We hope the result of this research would provide a practical and effective method for structural design.
