This talk considers the inverse problem y=f(x), where x and y are observable parameters, in which we wish to recover the model f. Examples include dynamical systems and combat models with y=dx/dt and x=parameter(s), water catchments with y=streamflow and x=rainfall, and groundwater vulnerability with y=pollutant concentration(s) and x=hydrological parameter(s). Historically, these have been solved by many methods, including regression or sparse regularization for dynamical system models, and various empirical correlation methods for rainfall-runoff and groundwater vulnerability models. These can instead be analyzed within a Bayesian framework, using the maximum a posteriori (MAP) method to estimate the model parameters, and the Bayesian posterior distribution to estimate the parameter variances (uncertainty quantification). For systems with unknown covariance parameters, the joint maximum a-posteriori (JMAP) and variational Bayesian approximation (VBA) methods can be used for their estimation. These methods are demonstrated by the analysis of a number of dynamical and hydrological systems.
Catégorie d'événements : Fluides, Thermique et Combustion
ON THE USE OF SUPERSONIC JET-CURTAINS FOR CONTROL OF MOMENTS ON TAILLESS AIRCRAFT
The purpose of this study is to explore the replacement of conventional moving control surfaces on a typical tailless aircraft model at high subsonic cruise speeds. Since the efficacy of sweeping jet actuators was only demonstrated at low speeds, the current test considered their potential replacement by Supersonic Steady Jets (SSJs). It was shown that even a single jet properly located and oriented may outperform an array of actuators whose location and orientation did not take into consideration the changing local flow conditions. The test article chosen was the SWIFT model that represents a typical blended wing-body configuration of a tailless aircraft. It was selected because it was tested extensively using Sweeping Jet Actuators (SJAs). When a single supersonic jet designed for Mn=1.5 was used to control thepitch it was able to increase the trimmed lift by approximately a factor of 3. The power it consumed was not necessarily smaller than an array of 6 SJAs but it provided other advantages that are discussed in the paper. Large yawing moments could be provided by other SSJs that were not encumbered by large rolling moments. These tests proved that the momentum input is but one of many parameters controlling the flow. It was replaced by power coefficients that are unambiguously measured and are capable of comparing various modes of actuation.
Scattering of topological edge waves in Kekule structures
Kekule structures are graphene-like lattices, with a modulation of the intersite coupling that preserves the hexagonal symmetry of the system. These structures possess very peculiar properties. In particular, they display topological phases manifested by the presence of edge waves propagating on the edge of a sample. We will discuss the extraordinary scattering properties of these edge waves across defects or disorder. We will also discuss how to realize Kekule structures in acoustic networks of waveguides
Computational fluid dynamics software: topology optimisation, uncertainty quantification and machine learning recent development.
Computational fluid dynamics software have gained in popularity the past few decades. They use numerical analysis and data structures to analyze and solve problems that involve fluid flows. This remains a field under development, with ongoing research to improve the accuracy and speed of complex simulation scenarios.
New methods such as topology optimisation have been recently added to CFD software, paving the path to design highly efficient engineering components. The design of an engineering component is updated using a mathematical algorithm to maximise a performance value while being subject to specific boundary conditions. This seminar will explain the basis of topology optimisation and it will go through a specific application for valve and robust optimisation. CFD software enables a first prediction before relying on experimental data to confirm an engineering component performance. Discrepancies are a common occurrence, thus accuracy of experimental data and CFD software can be improved by the utilization of uncertainty quantification. This will be shown on a cantilever beam test case, where the exceedance probability is estimated while experimental data are subject to epistemic and aleatory uncertainties. Accuracy of models prediction in CFD software can also be improved by using data from higher fidelity models. Machine learning algorithms have been gaining in popularity in the past couple years. They offer the opportunity to leverage data coming from high fidelity simulation to develop models for low fidelity simulation. This enables low fidelity to gain in accuracy while keeping a low computational cost. However, dealing with large data can be challenging in itself, which is going to be shown on an AI data driven turbulence model test case, where data from DNS and LES are used to build a turbulence model for RANS simulations.
QSQH theory of scale interaction in near-wall turbulence: the essence, evolution, and the current state
The Quasi-Steady Quasi-Homogeneous (QSQH) theory describes one of the mechanisms by which the large-scale motions active outside the viscous and buffer layers affect (‘modulate’) the turbulent flow inside these layers. The theory presumes that the near-wall turbulence adjusts itself to the large-scale component of the wall friction. Formulated in a mathematically rigorous form, the theory allows nontrivial quantitative predictions. The talk will describe the basics the theory and its methods, its current state, and a new tool making the application of this theory easy. Examples of applications and comparisons will be given.
Journée François Lacas des doctorants en combustion
Journée François Lacas des doctorants en combustion
Le jeudi 9 mars 2023, l’institut PPRIME accueillera la journée François Lacas dans les locaux de l’ISAE-ENSMA. Lors de cette journée des doctorants en combustion, des étudiants en thèse en provenance de la France entière sont conviés pour présenter leurs dernières avancées et échanger en présence de spécialistes de la discipline issus des différents laboratoires, organismes de recherche et groupes français. Cette journée, co-organisée par l’institut PPRIME et le Groupement Français de Combustion (GFC) sera également l’occasion de réaliser l’assemblée générale du GFC, section Française du Combustion Institute.
Morphogenèse de réseaux spatiaux : l’exemple des craquelures dans l’argile
Des structures de réseaux dans le plan 2D sont présentes tout autour de nous, des plans de rue des villes aux veines dans les feuilles d’arbres, en passant par certains coraux, réseaux de vascularisation de méduses, les deltas de fleuves ou encore dans Physarum polycephalum, le fameux « blob ». Ce séminaire portera sur un système type produisant de tels réseaux, les craquelures obtenues par dessiccation de l’argile, qui permet à la fois de contrôler différents paramètres induisant des changement de morphologies, et de suivre la croissance du réseau au cours du temps.
Je présenterai les différents résultats d’expériences menées pendant ma thèse, ainsi qu’un modèle permettant de comprendre la grande variabilité de motifs observés dans ce systèmes, illustrée dans les images ci-jointes. La présentation de ce modèle permettra de décrire la dynamique de séchage, l’évolution de champ de déformations/contraintes, ainsi que la croissance et les interactions entre les craquelures,afin de relier la forme complexe des structures globales de réseaux à des régimes simples de croissance locale de craquelures.
Les craquelures dans l’argile permettent d’appréhender en tant que modèle simple l’émergence de structures complexes de réseaux dans une grande variétés de systèmes sociaux, physiques ou biologiques.
Sparse sensing and Interpretable machine learning for surrogate modeling of complex systems
Abstract: Several problems in earth sciences and engineering arise from complex systems governed by nonlinear PDEs. There are two major challenges in this area: a) The need for a computationally efficient, rapid modeling capability to assist in design and forecasting, i.e. surrogate modeling, and b) The growing need to exploit sparsely sampled sensor and measurement data, to estimate the state of the full complex system i.e. sparse sensing. Both these challenges are characterized by lack of a robust theoretical or analytical formulation, unlike numerical simulation of partial differential equations. Data-driven techniques like machine learning have shown promise, but there is considerable progress to be made for applications. In this talk, I will present some recent advances our team at Los Alamos National Laboratory has made in these areas. I will discuss our sparse sensing learning approaches that can scale to large datasets in diverse applications. Additionally, I will present a few promising directions in the area of interpretable machine learning for surrogate modeling of PDEs, with a focus on fluid dynamics. We demonstrate that methods that reduce excessive dependence deep neural networks and instead achieve superior accuracy by a tighter coupling with the governing equations. I will also outline the path ahead and opportunities for collaborative research.
Evolution of coherent structures in turbulent channel flows by stochastic modelling under location uncertainty
The prediction of the evolution of coherent structures in turbulent flows by models linearised over the mean flow leads to a closure problem. We propose to go back to the conservation laws, and to consider a version of the Navier-Stokes equations submitted to a stochastic transport, referred to as « under location uncertainty ». In this framework, the displacement of a particle is caused by a resolved time-differentiable velocity field perturbed by a Brownian motion mimicking the effect of turbulent fluctuations. With these hypotheses, conservation laws show additional terms such as a stochastic diffusion induced by the Brownian motion or an effective transport velocity from high to low turbulence regions.
We propose to linearise this set of stochastic equations over the mean flow of turbulent channels at friction Reynolds numbers 180, 550 and 1000, in order to predict coherent streaks/rolls structures in the buffer and logarithmic layer. We show that for such a flow, it is required to add a non-linear forcing term, as performed in resolvent analysis, to account for non-linear interactions between time-correlated structures, active in the regeneration cycle of roll/streaks structures. We show moreover the improvements of predictions in the logarithmic layer by employing the stochastic modelling.
Bio:
Gilles Tissot est chercheur à INRIA Rennes dans l’équipe Odyssey. Il a effectué sa thèse à l’institut Pprime avec Laurent Cordier et Bernd Noack. Il a effectuer un post-doc à l’ITA (Brésil) avec André Cavalieri, un post-doc à l’institut de mathématiques de Toulouse avec Jean-Pierre Raymond et un post doc au laboratoire d’acoustique de l’université du Mans avec Gwénaël Gaba
Les vibrations induites par les vortex pourraient-elles améliorer la capture de nourriture par les coraux mous ?
Les coraux mous, tels que le panache marin bipenné Antillogorgia bipinnata, sont des animaux qui forment des colonies et se nourrissent en attrapant les particules de nourriture apportées par les courants. Grâce à leur squelette flexible en forme d’arbre, ils se plient et se balancent d’avant en arrière avec la houle. En plus de ce balancement à basse fréquence de toute la colonie, les branches d’A. bipinnata vibrent à haute fréquence avec une faible amplitude. Nous montrons que les tourbillons lâchés dans le sillage de la colonie corallienne sont probablement responsables de ces vibrations. Pour évaluer l’impact de la dynamique sur l’alimentation par filtration, nous simulons des particules advectées par l’écoulement autour d’un cylindre circulaire et calculons le taux de capture avec un solveur d’interaction fluide-structure. Nous observons que les cylindres vibrants peuvent capturer jusqu’à 40% de particules en plus que les cylindres fixes. Les vibrations induites par les tourbillons (VIV) améliorent vraisemblablement la capture de nourriture par les coraux mous. De même, la dispersion/capture du pollen et la dispersion des graines sont probablement affectées par les VIV.
Frédérick P. Gosselin est un expert en mécanique des structures élancées. Il est professeur au département de génie mécanique de Polytechnique Montréal (en poste depuis 2012). Il a obtenu sa maîtrise en génie de l’Université McGill à Montréal en 2006. Il a ensuite obtenu un doctorat de l’École Polytechnique en France (2009) sous la direction d’Emmanuel de Langre pour ses travaux sur les mécanismes d’interactions fluide-structure entre écoulements et végétation. Il a été professeur agrégé invité au département de botanique de UBC à Vancouver, au Canada, en 2018-19. Il est éditeur associé au Journal of Fluids and Structures. Il étudie une variété de structures élancées allant des branches et feuilles d’arbres, de la soie d’araignée et des membranes cellulaires aux ailes d’avions et turbines hydrauliques. Il est titulaire d’une subvention à la découverte du CRSNG pour étudier la mécanique des structures biologiques élancées.
Pour en savoir plus :
Boudina, M., Gosselin, F.P., Étienne, S. « Vortex-induced vibrations: a soft coral feeding strategy? » Journal of Fluid Mechanics, 2021, 916, A50.