Modelling & Numerical Simulation

December 18, 2019

Supported by:

Linked to M2SiNum project

M2SINUM is co-financed by the European Union
with the European regional development fund (ERDF)
and by the Normandie Regional Council.


Coordonnateur & org. : Antoine Tonnoir (LMI)


  • Program :
    - 14h30 - 15h : Alexandre Imperiale (CEA)  
    Titre : Numerical modelling of wave propagation in anisotropic viscounts-elastic laminated materials in transient regime. Application to modelling ultrasonic testing of composite structures. 
    Résumé : Composite laminate structures remain an important family of materials used in cutting-edge industrial areas. Building efficient numerical modeling tools for  high-frequency wave propagation in order to represent ultrasonic testing experiments of these materials remains a major challenge. In particular, incorporating attenuation phenomena within anisotropic plies, and thin intermediate isotropic layers between the plies often represent significant obstacles for standard numerical approaches. In our work, we address both issues by proposing a systematic study of the fully discrete propagators associated to the Kelvin-Voigt, Maxwell and Zener models, and by incorporating effective transmission conditions between plies using the mortar element method. We illustrate the soundness of our approach by proposing intermediate 1D and 2D numerical evidence, and we apply it to a more realistic configuration of a curved laminate composite structure in a 3D setting.
    - 15h - 15h30 : Timothée Schmoderer (LMI)
    TitreMRI reconstruction with dictionary learned motion compensation.
    RésuméIn this talk, I will discuss the ill-posed problem of MRI reconstruction. The problem is to produce high quality images from under-sampled k-space data. Recently new hybrid methods with motion compensation were introduced and achieved state of the art. I will present these new methods and how they provide a challenging numerical problem. Finally I will introduce a new algorithm for the motion compensation, constructed by a dictionary learned approach. 
    - 15h30 - 16h : Pause café 
    - 16h - 16h30 : Raphaël Antoine (CEREMA) 
    Titre : Télédétection et imagerie géophysique de subsurface pour les risques naturels.
    Résumé : Nous présentons ici les résultats de plusieurs études menées dans le cadre des missions du Cerema dans les domaines des risques naturels (cavités, retrait du trait de cote, etc...).
    Nous développons dans ce cadre de nouvelles méthodologies en télédétection multi-spectrale par drone (visible + infrarouge thermique) et en imagerie géophysique 3D (notamment électrique), que nous combinons afin d'obtenir des modèles intégrés surface/subsurface. Ces outils permettent d'une part d'améliorer l'interprétation des mesures effectuées et d'autre part de faciliter la visualisation des observations 3D pour les gestionnaires. 

    - 16h30 - 17h : Norbert Warncke (Siemens Gamesa)
    Titre : Machine learning for the detection of blade defects and Boundary Element Methods for numerical flow solvers.
    RésuméThe R&D Center of Siemens Gamesa at INSA Rouen works on several topics ranging from Numerical Aerodynamics and Wakes to Turbine Control and Machine Learning. In our talk, we want to give a short overview of two of these research topics: Machine Learning for the detection of blade defects, and Boundary Element Methods for numerical flow solvers.
    Detecting blade defects (cracks in the composite material) early and reliably is essential to offshore wind turbine operations, as it allows repair and maintenance operations before a total failure occurs. The challenge is to do that based on the available sensors in the turbine. To detect small changes in resonance frequencies, Maximum Mean Discrepancy is used, a kernel based two sample test method. This method can be used to detect changes in the industrial automatized processes by detecting problems that are statistically formulated in the form of hypothesis tests.
    Boundary Element Methods are an efficient way to compute the flow around a wind turbine, less costly to compute than typical CFD methods, but capable of modelling instationary flows at a coarse scale. Our aim is to find closed-form expressions for the Galerkin BEM integrals that are computed inside our in-house aerodynamic flow solver Ardema3D. We use the decomposition method of Lennoir&Salles for surfaces approximated by flat triangular meshes, and try to replace complex terms in this decomposition by simpler and more efficient ones.

  • Location: Rouen INSA - salle B0 A RC 02


Antoine Tonnoir / Christian Gout - Novembre 2019