Development of high-speed analytical method for welding mechanics using Idealized Explicit FEM accelerated by a GPU

Abstract

In this research, the authors have developed a new computational method named Idealized Explicit FEM, which is based on Dynamic Explicit FEM and can solve large scale problems of welding residual stress in shorter computing time and less memory consumption. In addition, parallelization using GPU (Graphics Processing Unit) was introduced to Idealized Explicit FEM to achieve more fast computation. Parallelized Idealized Explicit FEM was applied to the fundamental T-joint welding problem and analysis accuracy and performance of parallelized Idealized Explicit FEM were discussed through the comparison of the existing method. As a result, it was confirmed that parallelized Idealized Explicit FEM has the almost same accuracy as the existing method and it can possible to analyze large scale problem in short computing time and small memory consumption. In addition, parallelized Idealized Explicit FEM was applied to the moving heat source problem of 33 pass multi-layer welding, and as a result it was found that Idealized Explicit FEM can solve the large scale welding problem having more than 1 million degrees of freedom in realistic computing time.