29 research outputs found
Towards a Discontinuous Galerkin solver for scale-resolving simulations of moderate Reynolds number flows, and application to industrial cases
No full textDue to the continuously increasing economical and environmental constraints, the standard industrial CFD methods (mostly Reynolds Averaged Navier-Stokes equations, RANS) are no longer sufficient to answer the design requirements of the industry, in particular when off-design performance and noise need to be predicted. Therefore, scale-resolving simulations, where the full (Direct Numerical Simulation, DNS) or at least a significant portion (Large-Eddy Simulation, LES) of the turbulence spectrum is resolved, are required. However, as these simulations require a nearly flawless representation of very small turbulent structures, current industrial solvers require huge computational resources in order to provide sufficient accuracy. The discontinuous Galerkin method (DGM) could alleviate this to a large extent as it seems to bridge the gap between the flexibility of industrial codes and the accuracy of academic solvers. During this thesis, the flexibility and the parallel efficiency of a DGM solver has been improved to tackle the large requirements of DNS and LES. The method was subsequently assessed for DNS and LES based on canonical benchmarks. Due to its interesting dissipation and dispersion properties, DGM seems to offer an accuracy similar to pseudo-spectral (PS) solvers for DNS. As the dissipation targets only the smallest scales, the method seems well suited to use an implicit LES approach. This approach has been validated on the simulation of homogeneous isotropic turbulence and on the channel flow at several Reynolds numbers. Finally, the method has been successfully applied on industrial cases, including a low pressure turbine blade, airfoil profiles and a high Mach number jet flow, thereby showing the maturity of the method.(FSA - Sciences de l) -- UCL, 201
DNS of a Low Pressure Turbine Blade Computed With the Discontinuous Galerkin Method
No full textpeer reviewedThe direct numerical simulation of a turbine cascade at Reis = 85000 and Mis = 0.6 has been undertaken using a fourth-order accurate discontinuous Galerkin / symmetric interior penalty method. This method combines the high accuracy, typical of the dedicated academic DNS and LES codes, to the geometrical flexibility of industrial finite volume codes. It also allows for visual inspection of grid resolution, based on the continuity of the computed fields. Finally it attains high serial and parallel performance due to the exploitation of the high locality of the data. For these reasons it is expected that the method will enable the reliably resolved DNS and LES computations in turbomachinery industry. The accuracy with respect to dedicated academic codes is assessed on the computation of the Taylor-Green vortex. For the turbine case, the computed flow fields are compared to those obtained by large eddy simulations using a finite volume solver, and relative costs for a given resolution are estimated
The Discontinuous Galerkin Method as an Enabling Technology for DNS and LES of Industrial Aeronautical Applications
No full textTo enhance prediction capacities and therefore allow more advanced aeronautic and aero-propulsive design, new CFD tools are required. State of the art codes are based on second order accurate finite volume methods and are primarily developed for statistical turbulence modeling approaches. Given the limitations of these models, more direct approaches such as DNS or LES are required for the prediction of off-design aerodynamic performance, noise generation, transitional flows ..
A discontinuous Galerkin method for implicit LES of moderate Reynolds number flows
No full textpeer reviewedThis work is part of the development of a new generation of CFD solvers on the basis of the discontinuous Galerkin Method (DGM), specifically targeted towards accurate, adaptive, reliable and fast DNS and LES of industrial aerodynamic flows. In this study, the ability of the method to perform accurate implicit LES is investigated. The method is firstly assessed on the well-known turbulent channel flow. Several Reynolds number (up to Reτ = 950) are studied. The results show a fair agreement with the reference DNS, showing the ability of the method to perform accurate ILES on regular grids. Then, the method is applied on several advanced benchmarks (studied in the European project IDIHOM), performed at moderate Reynolds number. The 2D periodic hill flow, the low pressure turbine blade T106C and the JEAN nozzle benchmarks are considered. Encouraging results have been obtained, paving the way to the use of the method for industrial applications
Practical implementation and assessment of the discontinuous Galerkin method for DNS and LES of industrial flows
No full textDiscontinuous Galerkin Methods: Discretisation and efficient implementation & Application to turbulent flows
No full textDNS and ILES of transitional flows around a SD7003 using a high order Discontinuous Galerkin Method
No full textpeer reviewe
Assessment of Wall-modeled LES Strategies Within a Discontinuous-Galerkin Spectral-element Framework
No full textApplication of wall-models to discontinuous Galerkin LES
No full textWall-resolved LES are still limited to moderate Reynolds number flows due to the high computational cost required to capture the inner part of the boundary layer. Wall-modeled LES (WMLES) provide more affordable LES by modeling the near-wall layer. Wall function-based WMLES solve LES equations up to the wall, where the coarse mesh resolution essentially renders the calculation under-resolved. This makes the accuracy of WMLES very sensitive to the behavior of the numer- ical method. Therefore, best practice rules regarding the use and implementation of WMLES cannot be directly transferred from one methodology to another re- gardless of the type of discretization approach. Whilst numerous studies present guidelines on the use of WMLES, there is a lack of knowledge for discontinuous fi- nite element-like high-order methods. Incidentally, these methods are increasingly used on the account of their high accuracy on unstructured meshes and their strong computational efficiency. The present paper proposes best practice guidelines for the use of WMLES in these methods. The study is based upon sensitivity analyses of turbulent channel flow simulations by means of a Discontinuous Galerkin approach. It appears that good results can be obtained without the use of a spatial or temporal averaging. The study confirms the importance of the wall function input data location and suggests to take it at the bottom of the second off-wall element. This data being available through the ghost element, the suggested method prevents the loss of computational scalability experienced in unstructured WMLES. The study also highlights the influence of the polynomial degree used in the wall-adjacent element. It should preferably be of even degree as using polynomials of degree two in the first off-wall element provides, surprisingly, better results than using polynomials of degree three
Assessment of a High-Order Discontinuous Galerkin Method for the Direct Numerical Simulation of Transition at Low-Reynolds Number in the T106C High-Lift Low Pressure Turbine Cascade
No full textpeer reviewedAn implicit time integration, high-order discontinuous Galerkin method is assessed on the DNS of the flow in the T106C cascade at low Reynolds number. This code, aimed at providing high orders of accuracy on unstructured meshes for DNS and LES simulations on industrial geometries, was previously successfully assessed on fundamental, academic test cases. The computational results are compared to the experimental values and literature, and the obtained flow field characteristics are discussed. Although adequate resolution is supposed to be attained, discrepancies with respect to the experiment are found. These differences were furthermore consistently found by all authors in the workshop on high-order methods for CFD. The origins are therefore conjectured to result from insufficient adequation between computational setup and experiments, as no modeling is assumed. A plan for further investigation is proposed
