A Design Optimization Methodology Combining Artificial Neural Networks and Nonlinear programming Methods

• Main Authors: Huang, Chun-Ju, 黃俊儒
• Other Authors: Shuo-Jen Lee
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/15586386343399541589

碩士 === 元智工學院 === 機械工程學系 === 84 === This research developed a hybrid optimal design -- methodology by combining artificial neural network and nonlinear programming methods. In the process of optimization, the gradient base methods can become very expensive or infeasible because of the difficulty to obtain the numerical data (e.g. FEM, CFD) or experimental results. Therefore, this research developed a method to solve the problem that the implicit functions have brought to the engineering design optimization process. The essence of this method is to use the learning ability of the ANN and the searching ability of the nonlinear programming algorithm with as far training sets as possible. With this approach, we do not need to calculate the gradients of the design variables of the first or second order methods which would conserve a lot of computational effect. The global learning and searching spirit of the proposed methodology are very suitable for the engineering optimal design and other optimization problems.