Distributed Computation Applied to Structural Response Time-History Analysis

• Main Author: 張勝評
• Other Authors: Gwolong Lai
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/48230825128565026311

碩士 === 國立雲林科技大學 === 營建工程系碩士班 === 90 === The calculation of response time-history of a structural system under dynamic loading is a basic yet important task. With the understanding of stress state of structural members under loading, engineers can assess the deformation and force distribution of a structure under various types of dynamic loadings, and thus result in a safe and economical design. Because typically the finite element method is utilized for large-scale structural analysis, the calculation of structural dynamic response includes generating structural stiffness and mass matrices, and then using mode superposition method or direct integration method such as Newmark or Wilson θ method to calculate the displacement response with respect to time and further calculate the stresses in structural members. Besides the creation of structural system matrices, most of the calculation includes various array computations. If today’s fast growing parallel computation technique can be applied, the CPU time consumed by large-scale structural analysis can be greatly reduced. This thesis first raises the existing methods for determining the structural response time-history. The advantages and disadvantages of each method are reviewed carefully to explore the feasibility of applying parallel computation. The architecture and properties of the IBM SP2 workstation cluster for performing the parallel computation, the functionalities of the MPI, and the evaluation methods of the parallel efficiency are also familiarized. After the evaluation, the industry-preferred Newmark and Wilson θ methods are selected and the corresponding parallelized codes are developed, tested, and evaluated with the real-world cases. From the results of parallel performance evaluation, the parallel algorithms or programs are corrected and optimized. An effective and efficient finite-element code is then constructed for the response time-history analysis of large-scale structur