An Approach for Accelerating Incompressible Turbulent Flow Simulations Based on Simultaneous Modelling of Multiple Ensembles

TL;DR

This paper introduces a novel parallel-in-time simulation method for turbulent flows, enabling simultaneous modeling of multiple flow states to significantly accelerate large-scale turbulence simulations.

Contribution

It proposes a new approach that models multiple flow ensembles simultaneously, allowing parallelization of time integration and improving simulation efficiency for turbulent flows.

Findings

Achieved a twofold speedup in turbulent flow simulations.

Validated the approach on channel flow and flow with wall-mounted cubes.

Demonstrated theoretical and practical efficiency improvements.

Abstract

The present paper deals with the problem of improving the efficiency of large scale turbulent flow simulations. The high-fidelity methods for modelling turbulent flows become available for a wider range of applications thanks to the constant growth of the supercomputers performance, however, they are still unattainable for lots of real-life problems. The key shortcoming of these methods is related to the need of simulating a long time integration interval to collect reliable statistics, while the time integration process is inherently sequential. The novel approach with modelling of multiple flow states is discussed in the paper. The suggested numerical procedure allows to parallelize the integration in time by the cost of additional computations. Multiple realizations of the same turbulent flow are performed simultaneously. This allows to use more efficient implementations of numerical methods for solving systems of linear algebraic equations with multiple right-hand sides, operating with blocks of vectors. The simple theoretical estimate for the expected simulation speedup, accounting the penalty of additional computations and the linear solver performance improvement, is presented. The two problems of modelling turbulent flows in a plain channel and in a channel with a matrix of wall-mounted cubes are used to demonstrate the correctness of the proposed estimates and efficiency of the suggested approach as a whole. The simulation speedup by a factor of 2 is shown.