University of Pittsburgh engineers have developed an advanced GPU-based part-scale thermal process simulator called PAMSIM, designed to improve the simulation of heat transfer during additive manufacturing processes like metal laser powder bed fusion (L-PBF). This simulator utilizes a matrix-free preconditioned conjugate gradient finite element algorithm with a voxel mesh, offering unmatched efficiency in simulating complex geometries and heat transfer.
Description
The PAMSIM represents a significant leap forward in additive manufacturing technology by eliminating the need for extensive mesh generation and supercomputer access, which are typically required for simulating large, complex parts. By fully leveraging the power of modern GPU computing, PAMSIM can achieve thermal simulations 300 times faster on a single GPU compared to conventional commercial software running on a single CPU, thus providing substantial performance and cost benefits.
Applications
- Additive Manufacturing
- Thermal Process Modeling
Advantages
PAMSIM offers several key advantages over existing simulation technologies. Firstly, it enables high-performance simulations on standard desktop computers equipped with GPU cards, achieving the same performance as commercial software running on much more expensive, multi-CPU systems. This accessibility allows for wider adoption among designers and engineers who may not have access to supercomputing resources. Secondly, the use of a matrix-free formulation and voxel mesh significantly reduces the complexity and time required for mesh generation, a common bottleneck in simulating complex geometries. The simulator's ability to handle large, intricate part designs with efficiency and accuracy positions it as a transformative tool for the additive manufacturing industry, potentially replacing much of the physical testing required for component certification and qualification.
Invention Readiness
The development of PAMSIM is based on a rigorous experimental framework that involved simulating the L-PBF process with real-world part geometries and comparing the results with those from commercial simulation software. Through these experiments, PAMSIM demonstrated up to 300 times faster performance on a single GPU compared to a commercial software running on a single CPU. Key advancements include the exclusion of inactive elements from memory, adaptive meshing in the build direction, and the use of preconditioners and layer lumping to accelerate computations. These innovations were tested and validated against representative L-PBF process parameters and temperature-dependent thermal properties, confirming the simulator’s accuracy and efficiency. The simulator has been shown to be particularly effective in handling large-scale simulations that are typical in industrial applications.
IP Status
Copyright
