Author(s) |
Zhili Feng, Hui Huang, Jian Chen, Blair Carlson, Hui-Ping Wang, Wayne Cai, Jon Tatman, Greg Frederick |
Abstract Scope |
Oak Ridge National Laboratory (ORNL), in partnership with General Motors and the Electric Power Research Institute, has developed a high-performance computational (HPC) weld simulation code, DR-Weld (Digital Reality Welding Simulation). Available for commercial use, DR-Weld is shown to be more than 1000× faster than today’s commercial codes for weld residual stress and distortion simulations of large and complex welding and metal additive manufacturing (3D printing) parts and structures.
At its core, DR-Weld is a super-fast thermal-mechanical solver specifically tailored for HPC GPU processors and for welding and metal additive manufacturing, which are characterized by complicated interactions be-tween the highly localized heat source and the material to be processed. It is very efficient in solving the transient thermal-mechanical problem of a moving heat source on a large-scale complex geometry, which is challenging for traditional simulation software. DR-Weld also has a simple built-in user interface. It is de-signed to work with other commonly available FEM-based pre- and post-processors (Abaqus-CAE, Hy-perMesh, LS-Dyna), to build a model and display and integrate the simulation results. It offers the maximum flexibility for use with existing pre- and post-processors.
Using DR-Weld, high-fidelity simulations of complex welding and 3D printing problems can be now solved in hours or days, instead of the months or years of computational time required by today’s commercial codes.
In this presentation, the basic design approach of DR-Weld will be discussed first, followed by benchmarking of the simulation results from DR-Weld with ABAQUS, for both accuracy and speed performance. Examples will be given to illustrate the function and capability of DR-Weld in 3D welding and AM simulations. |