Research on Vibration Analysis using Parallel Computing for Large Ship Structures

点击翻译按钮获取中文摘要

Vibration analysis performed in the design stage has been mainly to solve natural frequencies and to avoid the resonances. But it is difficult to avoid a lot of resonance frequencies for large ship structures. It is partly because the structures are sometimes in contact with water. It makes difficult to avoid resonances by raising natural frequencies against exciting frequencies. In the next generation, establishment of magnitude of vibration responses are to be predicted by precise and whole structural analysis. But it takes a lot of time for computing and mesh generation.In the present study, we have developed a method for frequency response curve for thin walled structures by use of parallel computing. In the method uses message passing library (MPI), PCs with Windows or Linux operation systems, Fortran compiler, and fast communication network environment. that is, parallel computing system which is composed of popular computing environment in usual business offices.Firstly we propo … More sed DDBMG(Domain Decomposition Before Mesh Generation) scheme for total vibration analysis of whole ship structures. This scheme makes it possible that multiple man-powers can be also paralleled in the process of mesh generation.Secondly we have developed a numerical program for vibration analysis considering parallel computation. The computation technique is based on conjugate gradient method(CG method) for domain decomposition methods(DDMs) in order to satisfy equilibrium and compatibility between domains. Discretization of the model is done by finite element method.Thirdly, we have performed vibration analysis of thin-walled structures excited by harmonic excitation using developed parallel computing program. In that case degree of freedom on the domain boundary usually contains transfer displacement and rotation components. This mixed displacement components on domain boundaries often proved to spoil the convergence in CG iterations. Preconditioning by approximated diagonal scaling also proved to improve the convergence dramatically in the CG method. But for higher frequency domain, the convergence is still need more effective preconditioning for effective computation. Less