Flow physics and mesh refinement
Developing an AMR code raises challenges in the programming methodology and software management. Mesh refinement and de-refinement are strongly case dependent, and the mesh has to change dynamically and rapidly depending on the local flow physics. The flow physics has to be well understood in the development and optimisation of the AMR trigging strategy, especially for complex flows in combustion, where multi-phase flow, droplets, chemical reaction, wall turbulence and shock-waves are all involved.
Parallelisation of the code
In-depth parallelisation is another challenge for complex AMR code. In addition to conventional parallelisation based on MPI, vectorisation of the code would improve the performance of the code on accelerators, improving data locality of the AMR code. To further boost the performance on modern HPC architectures, it is intended to implement the hybrid MPI-OpenMP model, with OpenMP 4.5 from OpenACC on accelerators.
To capture atomisation and evaporation of droplets in combustion, a combined multi-phase reacting flow solver has to be developed. It will rely on Volume-of-fluid (VOF) method. However, the conventional VOF method is not accurate enough to capture the surface geometry of droplets and to compute the surface tension, which is critical in the evaporation process. Although AMR will help in refining the mesh around the droplets and improving the accuracy, it is still a key issue and a challenge to develop a high-accuracy front-tracking VOF method to integrate it into a parallel AMR-based code, with adequate performance.
Data analysis tools
Post-processing methods based on adaptive unstructured meshes are required so that volume-rendering based visualisation and the extraction of flow structures can be done with ease.