Towards Industrial LES/DNS in Aeronautics – Paving the Way for Future Accurate CFD

Project Information

TILDA

Grant agreement ID: 635962

DOI 10.3030/635962

Closed project

EC signature date 17 April 2015

Start date 1 May 2015

End date 31 December 2018

Funded under

• SOCIETAL CHALLENGES - Smart, Green And Integrated Transport

Total cost € 3 048 742,50

EU contribution € 2 706 241,75

Coordinated by

CADENCE DESIGN SYSTEMS BELGIUM
Belgium

Periodic Reporting for period 2 - TILDA (Towards Industrial LES/DNS in Aeronautics – Paving the Way for Future Accurate CFD)

Reporting period: 2016-11-01 to 2018-12-31

Summary of the context and overall objectives of the project

The technical objectives read:
• Accelerate High Order Methods (HOM) for unsteady simulations of LES and future DNS high-fidelity simulations of turbulence on unstructured grids.
• Advance methods to accelerate LES and future DNS methodology by multilevel, adaptive, fractal and similar approaches on unstructured grids.
• Use existent HPC networks and research projects, targeting applications of several tens of thousands of cores, bringing industrial applications of LES/DNS close(r) to daily practice.
• Target a range of hardware platforms including clusters of multi-core CPUs, potentially combined with integrated many-core processors (e.g. Xeon Phi), GPGPUs, ARM or FPGA as co-processing or main-processing units.
• Get as close as possible to the current status of DNS runs for academic problems. Extend LES/DNS to industrially relevant applications using HOMs on unstructured grids for unsteady flows aiming at a reduction of 2-3 orders of magnitude in CPU time.
• Provide grid generation methods for HOM on unstructured grids, with emphasis on valid curvilinear meshes for complex geometries including boundary layer and hybrid meshes.
• Provide suitable I/O and interactive co- and post-processing tools for large datasets “(datability)”.
• Demonstration of multi-disciplinary capabilities of HOM for LES in the area of aero-acoustics.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

NUMECA:
Has invested the effort in this work package to improve the efficiency of the explicit Flux Reconstruction solver
As first step to the development of a multi-level approach within the Flux Reconstruction solver for LES simulations, the extension of the explicit FR solver to handle non-conforming grids, both in geometrical (h-) and spectral (p-) senses, has been undertaken.
The Flux Reconstruction solver contains in-house tools to generate high-order elements near solid surfaces.
The parallel efficiency of the FR solver has been improved.
For IO postprocessing, the first task was the parallelization of the output subroutines.
DLR:
Contributed to WP 1 (General management) of TILDA by setting up and administrating the public and internal TILDA websites and a TILDA-wide email address.
In Task 2.1 (Implicit methods), DLR implemented SDIRK (single coefficient diagonally implicit Runge-Kutta) methods of order 2, 3, and 4.
In Task 3.2 DLR extended its scale-resolving DG code to locally refined meshes with hanging nodes.
ONERA:
Has presented a synthesis of the TILDA contribution during the CEAA workshop organized in Svetlogorsk (Russia) in September 2016.
Time implicit integration using inexact Newton methods have been developed initially in Aghora.
An adaptive strategy for the explicit VMSDG discretization of compressible Navier-Stokes equations is under development.
Local space p-adaptive method has been implemented.
The code Aghora is being optimized.
New computations for shock-boundary layer interactions on the 2D laminar and 3D turbulent transonic cases have been analyzed.
DASSAV:
Began implementing a VMS subgrid scale model based on an explicit filtering at the element level.
Started modifying the DES model to be used as a wall model in the VMS context.
Ran TC-F2 (Taylor Green Vortex) with DNS and existing LES models. Preliminary results were also obtained with the VMS model.
SAFRAN:
In order to analyze the flow solutions & the causes of numerical instabilities, new co-processing solutions have been implemented in ArgoDGM.
The work performed is based on ArgoDGM, a Discontinuous Galerkin software developed at CENAERO.
T106: Large Eddy Simulation of turbomachinery configurations with ArgoDGM of the T106-1 turbine were conducted.
RO37: A fully unstructured meshing strategy including the generation of boundary layers have been developed with an in-house code. The meshing process has been applied to the fully three dimensional NASA Rotor 37 test case and NS p2 simulations were conducted.
CERFACS:
The current involvement reads:
• D2.2-12: Validation of the non-conformal hp−adaptation
• D3.2-24b: Definition and validation of sensors for hp−adaptation
• D4.3-18: Extension of Antares to handle high-order solutions
• D5-18-P1: Intermediate report on TC-P1 (Jet with/without micro-jets – fluidic injection)
• D5-36-P1: Final, consolidated report on TC-P1 (Jet with/without micro-jets – fluidic injection)
CENAERO:
Task 2.2: Cenaero’s focus is on the capture of wake emanating from turbine blades.
Task 3.2: An indicator based on the balance of turbulent kinetic energy (TKE) has been devised.
Task 3.3: Application of wall-functions in the context of discontinuous Galerkin discretizations.
Task 4.1: Extension of our open-source mesh database MAdLib to conformal grid-adaptation on curvilinear grids.
UCL:
Firstly, a new shape quality measure has been designed for quantifying the error on the gradient of the finite element solution on linear simplices.
Secondly, three shape quality measure designed for linear elements has been generalized to curved elements. The first is the new quality measure.
In addition, developments have been made on curve mesh generation.
UNIBG:
Activities within Task 2.1 were mainly focused on a comprehensive assessment of the implicit time integration methods.
Investigated the impact on the accuracy and efficiency of a DNS (TC-F2) simulation of a low-Mach treatment.
Developed and implemented strategies for the initialization of the linear and non-linear iterative solvers.
Development and implementation of a p-adaptation strategy suited for unsteady simulations within the code MIGALE.
Implementation of procedures for the computation of flow statistics.
ICL:
Stabilization of Flux Reconstruction (FR) schemes via de-aliasing strategies and extreme-scale parallelization of PyFR.
Investigated how particular forms of FR correction can help stabilize under resolved DNS (ILES) of turbulent flows.
Rreleased several new versions of PyFR, adding amongst other features support for the Intel Xeon Phi.
TsAGI:
Task 1.3: TsAGI proposed to change test case "Noise suppressing Nozzle with chevrons" to Test case "Noise suppressing Nozzle with bypass".
Task 2.1: A theoretical investigation of implicit time stepping efficiency was conducted. 4th order DG scheme was used for spatial discretization.
Task 2.3: Navier-Stokes DG code was developed for K=1,2,3.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

As pointed out above and listed, new/advanced methods have been developed being clearly beyond the state-of-the-art.
Further validation will be carried out to underline the efficiency of the derived methods.
Besides the improvement of methods and approaches, and related to the (main) objective, namely to target applications of several tens of thousands of cores, considerable progress has been achieved by ICL.
ICL has demonstrated peta-scale performance of PyFR on various large-scale GPU clusters, including Piz Daint at the Swiss National Supercomputing Center, and Titan at Oak Ridge National Laboratory with performance at up to 13.7 DP-PFLOP/s sustained on 18,000 K20K GPUs of Titan (over 48 million CUDA cores).

Project logo