本文整合結構最佳化方法及分散式運算系統，建立一機械結構最佳化系統。其系統突破以往於單一機台執行最佳化分析的模式，藉由運算資源的整合，加速最佳化方法的計算，達至提升資產使用率及降低產品開發時間成本之兩大目的。於結構最佳化方法上，結合至遺傳演算法及類神經網路模型，使最佳化方法擁有全域搜尋之能力，且能透過基於自適性規則的類神經網路模型，在模型運算準確度達至標準的條件下，取代原適應度評估方式，提升演算法的運算效率。在分散式運算系統上，使用聯網技術整合場域內的運算資源，建立一客戶端/伺服端的系統架構，當客戶端執行最佳化分析後，伺服端能夠將尚未分析的變數任務分配予閒置的資源，並可將設計變數及分析結果儲存於資料庫中，該系統能夠大幅度地降低最佳化分析的所需時間且擁有數據管理的能力。接著，本研究將提出結合遺傳演算法及類神經網路模型的分散式機械結構最佳化系統應用於非凸優化函數及數個結構最佳化分析案例上，進行系統的實作及驗證。由結構最佳化案例的結果可得，本文提出之結構最佳化系統相較於其他的最佳化方法能夠大幅度縮短最佳化分析的時間，且獲得較佳的分析結果，驗證至該機械結構最佳化系統之實用性。;This thesis purposes the mechanical structure optimization system integrating the structural optimization method and distributed computing system. The optimization system breaks through the traditional optimization method analyzing on single workstation and boosts the computing efficiency by resources integrating, further to approaches two goals which are raising the usage rate of resources and decreasing the time cost on developing product. On the one hand, structure optimization method integrates the genetic algorithm and artificial neural network model by adaptive principle purposed in this research. Artificial neural network can replace the original way of fitness evaluations in suitable timing and high accuracy condition depending on adaptive principle to keep the correct calculus direction for algorithm. On the other hand, distributed computing system is a client/server framework built by network technology. In this distributed computing system, client side can launch the optimization project and interact with database, including sending the variables and reading the result. Server side, which is composed of dispatch system and database, possesses the features of cost time falling dramatically and data management. Then, the mechanical structure optimization system is applied on non-convex optimization function and some structure optimization projects. Meanwhile, it is verified that can sharply reduce the times of optimization analysis and get the greatest solutions than the other algorithms.
