JDD @ CERFACS

Videos of presentations – “Ma thèse en 3 minutes”

2020

• Etienne LameloiseModélisation de la formation de particules de suie dans les turbines à gaz avec prise en compte de la forme des agrégats
  Soot is a known contributor to global warming as well as a public health problem. In their lifespan, soot aggregates undergo various processes
  affecting their morphology, which is known for playing a critical role in their interactions with the media and their harmfulness. As such processes cannot be explored experimentally, modeling effort is forced towards an accurate prediction of the particles shape. While soot particles evolution and trajectories are modeled thanks to semideterminist Lagrangian approaches, aggregates morphology modeling calls on cluster aggregation models, assuming Brownian or ballistic motion. The main objective of this thesis funded by the ANR ASTORIA is thus to couple both methods within an LES framework to accurately predict soot aggregates distribution and morphology.

• Thomas MarchalExtension de l’approche Différences Spectrales à la combustion
  The aim is to assess the potential of the Spectral Differences (SD) method for the simulation of turbulent combustion in aeronautical engines. The SD method is a high-order numerical discretization approach, which has the potential to increase the accuracy of solutions at no extra cost. This is crucial for the design of the new generation of combustors, which must reduce both fuel consumption and pollutant emissions. The SD method is implemented in the code Jaguar, jointly developed with ONERA. The implementation of a combustion model will be first validated on simple test cases. The final objective is to perform the 3D calculation of the PRECCINSTA burner, and evaluate the gain in the tradeoff between accuracy and computational cost.

• Luciano DrozdaMéthodes d’apprentissage pour la dynamique des fluides
  In the context of the Data-Science strategic program, many topics are investigated at CERFACS in relation to machine learning and computational physics. They are structured in the Helios project (High pErformance LearnIng for cOmputational phySics: https://cerfacs.fr/helios/), the cross-laboratory workgroup on machine learning with ISAE-SUPAERO.
  The objective of this particular PhD is to explore new data-driven methodologies for the discretization of the Navier-Stokes equations. Recent work has shown the great potential of these approaches compared to traditional techniques, but much work is needed in this field to bring them to maturity for concrete CFD applications.

• Minh NguyenSimulation haute-fidélité des écoulements aérothermiques avec une approche LBM
  The high-pressure turbine of an aircraft engine involves extremely high temperatures, which can significantly change the size of the engine components, leading to increased fuel consumption and engine wear. A high-pressure turbine adaptive clearance control (HPTACC) system uses jets of air to control the dilation and contraction of engine components. The lattice Boltzmann method (LBM) is a relatively new technique for performing computational fluid dynamics (CFD) simulations, and has the potentially to efficiently simulate HPTACC configurations. Using a mesoscopic approach with a statistical distribution of gas particles, LBM can be several times faster than traditional CFD approaches. However, several developments must be made to render LBM capable of simulating high Mach, thermal flows at industrial scale.

• Pierre MatalonSolveurs rapides pour discrétisations robustes en CFD
  This project aims at developing scalable, robust linear solvers for Hybrid High Order (HHO) discretizations in industrial CFD applications. HHO combines the following desirable properties: it applies to general, polyhedral meshes; it can achieve optimal convergence orders even on distorted meshes; the computational cost can be reduced by hybridization and static condensation; and it is possible to increase the approximation order. At this point, the existing numerical solvers for such discretizations do not scale well. Our goal is to develop linear solvers that keep computing power and execution times acceptable while using unstructured meshes composed of hundreds of millions of cells. The project will focus in particular on multigrid methods.

• Lionel ChengExploration numérique de la combustion assistée par plasma pour atténuer les instabilités de combustion
  In recent years, NRP discharges have shown promising results regarding flame stabilization in academic burners. The non-equilibrium plasma has an effect on combustion chemistry compared to classical discharges that just bring energy into the mixture. The study of these non-equilibrium plasmas is made thanks to the electron energy distribution function which is solution to the electron Boltzmann equation. The electrons then trigger reactions that are beneficial for the combustion mechanism. This PhD aims to capture numerically flame-discharge interaction as precisely as possible through the coupling of AVBP and AVIP.

• Siham El Garroussi

Prévision des crues : réduction des incertitudes à faible coût via des modèles réduits

Accounting for uncertainties is of utmost importance for flood risk management. The lack of knowledge in hydraulic roughness or upstream discharge governing the flow can be critical for flood forecasting.Thus, uncertainties should be analyzed. Uncertainty quantification (UQ) framework aims to probabilize hydraulic model input uncertainties, propagate them through the model and finally, quantify their impact on the output of interest. As direct UQ techniques are expensive, surrogate models are used. In this thesis, we investigate advanced strategies to build accurate surrogate models in order to estimate water level in the river bed and the flood plains while dealing with the nonlinearities of the flow and the high dimension of the study area.

• Clovis GoutMéthode chorochronique pour la simulation mono-canale en turbomachine
  For the design of more efficient compressors and turbines, and as a consequence cleaner engines, numerical simulations are an invaluable tool for the designer. With the advent of high-performance computing, more precise methods such as LES become affordable. However, these methods are still expensive to be used in industrial design as they typically require to simulate the full 360° component. My work focuses on a particular condition, the phase-lag periodicity, literally space time periodicity. This particular condition considers the time shift of wakes, and makes it possible to simulate a single blade per row, which highly decreases the domain size. The aim is to develop and implement this method in a LES context in order to simulate a stage compressor.

• Johan DegrignyDéveloppement et intégration de moyens numériques avancés permettant d'évaluer les effets de propagation de décollements aérodynamiques dans le processus de dimensionnement et de certification des aéronefs civils
  In Computational Fluid Dynamics (CFD), the Large Eddy Simulation (LES) and Reynolds Averaged Navier-Stokes (RANS) turbulence modeling strategies can be combined into hybrid RANS-LES approaches to reap the benefits of both. Various of these (generically termed DES) have been developed in the frame of Navier-Stokes computations using finite volume methods, and can be transposed to Lattice Boltzmann Method (LBM) simulations. The use of non-body-fitted Cartesian grids in LBM requires adaptations to achieve the intended DES function. Different implementations are tested on cases of interest to external aerodynamics.