David Imbert

HPC industrial aerodynamic simulation

• Streamline code review process by reducing 30 % of unnecessary code

• Align C++ standards to ensure scalable and maintainable solutions

HPC datapath of mask writer production tools

• Maintain the C++17 driver of IMS's MBMW multi-beam mask writers

• Parallelize its parser to increase CPU usage of the embedded cluster

• Maximize Nvidia Tesla V100 GPU occupancy by using more GPU memory

• Consolidate both InfiniBand and CUDA stacks via new tested adaptors

• Hold design and architecture meetings

Software architecture and development of innovative and massively parallel solvers

• Restructure a procedural solver into an object oriented one

• Standardise its generic components: I/O, HPC, maths

• Improve performance of the computation kernel using new C++14 paradigms

• Organise dialogue between framework team and solvers team

• Carry the collaborative project PASC SPH-Exa on exascale computing

HPC simulation of tsunamis: benchmarks and uncertainties

• In charge of the development of the C++/MPI tsunami simulation code (FDTD method) of the french tsunami warning centre (CENALT)

• Validate the code through exprimental and analytical benchmarks

• Couple the model with the uncertainty propagation software Promethee from IRSN and deploy it on a cluster

• Create a Python/Matplotlib module to post-process data and visualize output maps

• Analyse the sensitivity of the maximum water height of a tsunami to the parameters of the seismic source

Modeling acoustic properties of submerged granular media

• Invent a code which simulate the propagation of acoustic waves in a submerged granular medium where both the coupling between wave and motion, and inter-granular contacts are taken into account

• Model interacting grains using the Molecular Dynamics algorithm

• Simulate the propagation of waves using a finite elements method and FreeFem++

• Couple these two models using a new fictitious domains method

• Develop a Python module for data post-processing using NumPy and Matplotlib

• Check the code by comparing it with original analytical solutions and experimental measures made in an acoustic tank in collaboration with Géosciences Rennes (JRU UR1/CNRS 6118)

3D seismic imaging of complex structures

• Write a Fortran 90 program which simulates the propagation of acoustic waves through a 2D variable-density medium using the finite differences method

• Optimal decomposition method with MPI to use it on a cluster

• Optimize the kernel of the code in association with Intel®

• Extent the code to 3D

• Apply the code in the framework of a reverse time migration (RTM)

Ultrasonic data inversion for characterization of heterogeneous welds

• Improve the Matlab code MINA from the Laboratory of Mechanics and Acoustics (LMA, ex-LCND, IRU CNRS 7051) which models the 2D structure and the orientation of grains in steel welds

• Use the Athena code from EDF to simulate the propagation of waves through the structure

• Compute and analyze cost functions related to the order of passes when welding

• Create an inversion algorithm specific to this discrete parameter