| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Computational Materials for Qualification and Certification
|
| Presentation Title |
GO-MELT: GPU-Optimized Multilevel Execution of LPBF Thermal Simulations |
| Author(s) |
Joseph P. Leonor, Seyed Mohammad Elahi, Andrew J. Potts, Zhongsheng Sang, Gregory J. Wagner |
| On-Site Speaker (Planned) |
Gregory J. Wagner |
| Abstract Scope |
We present the software GO-MELT, which uses a multilevel approach that couples three overlapping refinement levels at the part scale, meso-scale, and melt pool scale to simulate thermal history during LPBF. Overlapping meshes track a moving laser without remeshing, using fixed data sizes at each refinement level and allowing GO-MELT to efficiently exploit GPU acceleration using Google’s JAX library with JIT compilation to significantly speed up the simulation. We demonstrate a production run of a 1 cm cube in under 10 hours, averaging 1.64 ms per time step and over 600 times faster than a GPU-accelerated uniform mesh solver. GO-MELT’s capabilities include phase- and temperature-dependent material properties that capture the effects of powder melting and fusion. Current work uses sub-cycling to significantly speed up simulations, and substitutes data-driven surrogate models of the melt pool for improved throughput. GO-MELT’s efficiency allows model-based optimization of process parameter settings. |