MODELING AND FINITE ELEMENT ANALYSIS OF WELDING DISTORTIONS AND RESIDUAL STRESSES IN LARGE AND COMPLEX STRUCTURES
Open Access
- Author:
- Sun, Jun
- Graduate Program:
- Mechanical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 17, 2005
- Committee Members:
- Panagiotis Michaleris, Committee Chair/Co-Chair
Ashok D Belegundu, Committee Member
Marc Carpino, Committee Member
Padma Raghavan, Committee Member - Keywords:
- Finite Element Analysis
Parallel Computing
FETI-DP
Domain Decomposition
Distributed Computing
Partial Cholesky Re-factorization - Abstract:
- Material processing is an important topic in academic research and engineering practices. Its applications, such as welding and laser forming, are widely employed in the fabrication of large structures. However, welding applications may cause undesired permanent distortions and residual stresses in materials. It is highly desired by researchers and engineers to develop efficient numerical methods that have the capability to simulate material processing for a timely prediction of distortions and residual stresses that may be produced. Finite element analysis of 3D full scale thermo-elasto-plastic material processing has been considered to be computationally expensive and poses challenging difficulties for current available numerical algorithms as well as computer hardware. Tremendous computational costs arise from the fine meshes, small time increments, and nonlinearity involved in this kind of analysis. The objective of this research is to develop effective and efficient numerical methods and computational techniques that are capable of performing 3D large scale finite element analysis of material processing problems. Parallel computing is first introduced for simulating large scale applications on shared memory computers. The Dual-Primal Finite Element Tearing and Interconnecting method with Reduced Back Substitution and Linear-Nonlinear Analysis (FETI-DP-RBS-LNA) is then proposed to introduce the divide and conquer concept to the simulation of large scale problems and reduce the overall computational costs. Distributed computing is further introduced for the FETI-DP-RBS-LNA algorithm. Message Passing Interface (MPI) is implemented and tested on a distributed PC cluster so that FETI-DP-RBS-LNA receives the benefit of distributed computing. Finally, the partial Cholesky re-factorization scheme is investigated and implemented to improve the computational performance of material processing simulations. This scheme only re-factorizes the nonlinear regions in the structure. Therefore, the overall simulation time can be greatly reduced.