Abstract
The inherent strain method [1] is known and used for simulating - on a physical basis - deformation effects that occur in laser powder bed fusion (LPBF). The method is based on the idea of the inherent strain which is a black-box representation of strains caused by the process of metal printing inherent
to the printer, printer settings and material used. This inherent strain can be generated from experimental data. Thus, a Digital Twin of the process combines data-driven and physics-driven aspects.
In this work, we present an implementation of the inherent strain method on graphics processing units (GPUs) that exploits the massive parallelism of the many GPU cores to speed up the simulations considerably – some results show up to two orders of magnitude speed-up compared to CPU-based implementations. Compared to GPU-based simulation with unstructured finite elements on tetra-
hedral meshes [2], our implementation leverages the characteristics of structured meshes in terms of indexing and compactness of the resulting stiffness matrix.
To simulate the deformations, the part is discretized in a set of regular finite elements (hexahedral elements of the same shape and size). The computation of the resulting deformation is parallelized in two ways: Firstly, each layer addition is simulated with a highly parallel finite element simulation.
Secondly, each of these layer additions results in an increment of the total result. The calculations of the increments are independent from each other, thus, all increments can be computed in parallel.
to the printer, printer settings and material used. This inherent strain can be generated from experimental data. Thus, a Digital Twin of the process combines data-driven and physics-driven aspects.
In this work, we present an implementation of the inherent strain method on graphics processing units (GPUs) that exploits the massive parallelism of the many GPU cores to speed up the simulations considerably – some results show up to two orders of magnitude speed-up compared to CPU-based implementations. Compared to GPU-based simulation with unstructured finite elements on tetra-
hedral meshes [2], our implementation leverages the characteristics of structured meshes in terms of indexing and compactness of the resulting stiffness matrix.
To simulate the deformations, the part is discretized in a set of regular finite elements (hexahedral elements of the same shape and size). The computation of the resulting deformation is parallelized in two ways: Firstly, each layer addition is simulated with a highly parallel finite element simulation.
Secondly, each of these layer additions results in an increment of the total result. The calculations of the increments are independent from each other, thus, all increments can be computed in parallel.
Original language | English |
---|---|
Title of host publication | ECCOMAS Congress 2022 - 8th European Congress on Computational Methods in Applied Sciences and Engineering |
Publisher | Scipedia S.L. |
Number of pages | 9 |
DOIs | |
Publication status | Published - 1 Nov 2022 |
Event | 8th European Congress on Computational Methods in Applied Sciences and Engineering: ECCOMAS CONGRESS 2022 - Oslo, Oslo, Norway Duration: 5 Jun 2022 → 9 Jun 2022 https://www.eccomas2022.org/frontal/default.asp https://www.eccomas.org/2021/01/22/3542/ |
Conference
Conference | 8th European Congress on Computational Methods in Applied Sciences and Engineering |
---|---|
Abbreviated title | ECCOMAS CONGRESS 2022 |
Country/Territory | Norway |
City | Oslo |
Period | 5/06/22 → 9/06/22 |
Internet address |
Keywords
- Physically based simulation
- General Purpose Computation on Graphics Processing Unit (GPGPU)
- 3D Printing