Hybrid Parallelization for Solidification and Fluid Dynamics

Solutions by Application

  • Combine phase-field simulation with fluid dynamics.
  • Enhance performance by parallelization.

Due to their low density, Mg-Al alloys are interesting materials for lightweight structural applications in automotive and aerospace engineering and high-end consumer electronics. They usually consist of dendrites of the primary HCP-Mg-α phase and an interdendritic eutectic, which consists mainly of the inter-metallic Mg17Al12-β-phase as Mg-Al alloys are usually sensitive to corrosion. However, enclosing the α-phase dendrites with the β-phase improves the corrosion resistance. Therefore, optimizing process parameters is crucial to avoid direct contact between α-phase dendrites in the resulting microstructure.

Figure 1: Mg-Al alloy solidification with 5at% Al.

OpenPhase capabilites

1- Combine phase-field simulation with fluid dynamics:

  • OpenPhase implements the Lattice-Boltzmann method as a fluid dynamics solver.
  • Allows two fluid phases separating liquid and vapor using a mean-field potential.
  • Selection of wetting parameters for each phase.
  • Exchange of forces between fluid and solid.
  • Simulation of solid collisions.
  • Transport of element concentrations due to flow.

In the following example (see fig 2), the solidification of the primary α-phase dendrites has been simulated using a hybrid parallelization with MPI and OpenMP. Figure 3 shows the streamlines for a flow through the microstructure shown in fig 2 during the growth phase.

Figure 2: Mg-Al Solidification, 500 x 500 x 200 𝜇𝑚^3, 50 million grid points, 400 cores, 25 MPI processes with 16 OpenMP threads
Figure 3: Flow through the microstructure shown in figure 1. Flow from left to right.

2- Create large simulations using MPI and OpenMP:

  • Hybrid parallelization combining OpenMP and MPI.
  • Compatible with the MPI-version of the popular fast Fourier transformation software FFTW, which is used in the mechanics solver.
  • Allows large-scale simulations using computational resources from multiple compute nodes.
Video 1: Simulation of liquid phase sintering of WC–Co alloy

OpenMP distributes the computational work into multiple threads in shared memory. MPI allows the distribution of work into processes between different computational resources, which do not necessarily share memory. It requires communication of data between processes. Any version of OpenPhase supports OpenMP

If you wish to utilize MPI, a network license of OpenPhase is required, and each computational node requires a network seat. Contact us for more detials.