| This dissertation introduces related developments of finite element analysis and structural optimization at home and abroad. On the consideration of so many re-analysis times and computing time derived from optimization design of large-scale and complicated structure, To solve this problem so high performance algorithms are propose based on a standalone PC and Cluster system. On the one hand artificial neural network is used to analyze approximately so as to decrease the finite analysis times, on the other hand high performance computing method with structural finite analysis and optimization are combined to fulfill parallel computing.The framework of dissertation can be classified into three parts. The first part discusses the fundamental theory and principles of structural optimization design, artificial neural network and high performance computing; then the second part describes nonlinear simplex and SLP algorithms of structural optimization, back-propagation algorithm of artificial neural network. The last part exams related modified methods integrated with simplex and SLP: (1) Adding back propagation artificial neural network to replace partial finite element analysis; (2) Combining high performance computing with them to accelerate the solving speed, whose efficiency has been verified by samples.The main summary and conclusions are as follows:(1) Using artificial network to take approximate analysis is a good method, which doesn't influence solution accuracy and actually reaches the goal of decreasing re-analysis times, generally the time can be decreased approximately fourteen percents.(2) Based on the size optimization design problems of spatial truss using simplex and SLP algorithms, performing relevant procedures of analysis and optimization design and making total computing times reduces nineteen percents at least, It actually improves computing efficiency, and offers a feasible solution to optimization design of large-scale or complicated structure.
|
PDF Full Text Request