A framework for high-fidelity particle tracking on massively parallel systems

TL;DR

This paper extends the FLEXI CFD framework to accurately simulate particle-laden flows on massively parallel systems, enabling high-fidelity predictions of complex, multiscale phenomena in various engineering and natural systems.

Contribution

It introduces a scalable, high-order particle tracking extension to the FLEXI solver for simulating particle-laden flows with complex geometries on high-performance computing systems.

Findings

Effective particle tracking with ray-tracing methods

Maintains excellent scalability on HPC infrastructures

Successfully applied to large-scale particle-laden flow problems

Abstract

Particle-laden flows occur in a wide range of disciplines, from atmospheric flows to renewable energy to turbomachinery. They generally pose a challenging environment for the numerical prediction of particle-induced phenomena due to their often complex geometry and highly instationary flow field which covers a wide range of spatial and temporal scales. At the same time, confidence in the evolution of the particulate phase is crucial for the reliable prediction of non-linear effects such as erosion and fouling. As a result, the multiscale nature requires the time-accurate integration of the flow field and the dispersed phase, especially in the presence of transition and separation. In this work, we present the extension of the open-source high-order accurate CFD framework FLEXI towards particle-laden flows. FLEXI is a massively parallel solver for the compressible Navier-Stokes-Fourier equations which operates on (un-)structured grids including curved elements and hanging nodes. An efficient particle tracking approach in physical space based on methods from ray-tracing is employed to handle intersections with curved boundaries. We describe the models for a one- and two-way coupled dispersed phase and their numerical treatment, where particular emphasis is placed on discussing the background and motivation leading to specific implementation choices. Special care is taken to retain the excellent scaling properties of FLEXI on high performance computing infrastructures during the complete tool chain including high-order accurate post-processing. Finally, we demonstrate the applicability of the extended framework to large-scale problems.