Michaël Bauerheim

WELCOME

Current position : Associate Professor at ISAE-Supaero (France), Academic leader of the PhySimAi group at ANITI, Sherbrooke University (Canada), and AI consulting
E-mail : michael.bauerheim isae-supaero.fr
Phone : +33 5 61 33 80 85
Location : Building 38, office 223

Previous positions :
2015-2016 - Post-doctoral position at LMFA (France) and ETHZ (Switzerland) :

Aeroacoustic broad-band and tonal noise or rotor-stator interactions. LES and LNSE on the non-linear saturation of aeroacoustic sources.

2014-2015 - Post-doctoral position at IMFT (France) :

DNS and experiment of flame stabilization on a rotating cylinder.

2014 - Visiting scholar at Stanford University (USA) :

Uncertainty Quantification applied to symmetry breaking in thermoacoustics.

2012-2014 - PhD in thermo-acoustics at CERFACS and SNECMA (France) :

Theory and LES of thermoacoustic instabilities in annular combustion chambers.

2011 - Visiting scholar at Georgia Tech (USA) :

LES for combustion instabilities in methane-oxygen rocket engines.

Awards :
2014 - 3AF Clean Sky conference, nomination among the 10 best papers

2014 - 35th Int. Sytmposium on Combustion travelling grant 2014, and GFC travelling award with 2 papers accepted

2015 - Léopold Escande award for the best INP thesis 2014, delivered by the University of Toulouse

2015 - Award of Excellence for the contribution to "Ignition capability studies" in the FP7 European project LEMCOTECT

2015 - Paul Caseau award for the best PhD thesis on "Simulation and modelling", delivered by EDF

2015 - Paul Laffitte award for the best PhD thesis on "Combustion", delivered by the French Combustion Institute

2024 - Best paper award delivered by AIAA for “DeepGeo : Deep geometric mapping for automated and effective parameterization in aerodynamic shape optimization” AIAA SciTech conference.

2025 - Best paper award in the European conference on aero-space sciences (EUCASS) “Deep Learning-based surrogate modeling of an array of rocket injectors” in the propulsion category

TEAM

On-going team

Thomas Zeinaty : PhD candidate on deep reinforcement learning to control flutter using both simulations and experiments.

Rémi Ducotet : PhD candidate with Airbus on deep learning methods and data-assimilation for aerodynamics.

Victor Milo : PhD candidate with ESA and Ariance Group on LES simulations of rocket injectors.

Damien Meunier : PhD candidate at CERFACS on developing low-order tools for thermoacoustics.

Jean Fesquet : PhD candidate on implicit neural representation to predict complex 3D turbomachinery flows.

Gabriel Meisler : PhD candidate with Airbus on machine learning for icing detection.

Giovani Romano : PhD candidate on deep learning methods to predict injectors interactions in rockets combustion chambers.

Giovani Catalani : PhD candidate with Airbus (CIFRE) working on surrogate modeling of 3D aerodynamics problems using Implicit Neural Representation.

Enrico Foglia : PhD candidate in collaboration with Sherbrooke University on geometric Deep Learning for aeroacoustic surrogate modeling, as well as uncertainty quantification of DL models (in collaboration with EPFL).

Mateus Carvalho Costa (POLA4) : Postdoctoral fellow working on advanced DL techniques for efficient localization of acoustic sources.

Alumni

Tiphaine Arnould : PhD student working on LES simulations of cavity-noise coupling mechanisms in high-Reynolds turbulent flows. Now at Airbus France

Laura Choquet : Student in PhD track working on the partial observation problem for airfoil control in perturbed flows.

Germain Paris : Student in PhD track working deep reinforcement learning for satellite systems. He focuses on multiple-reward algorithms for optimal trajectories towards geostationary orbit.

Antonio Alguacil : Former PhD candidate, and now Postdoctoral fellow, working on the propagation of acoustic waves in complex media using deep convolutional networks, in collaboration with Sherbrooke University. Now Prof. Sherbrooke University

Brice Martin : PhD student working on Reinforcement Learning (RL) for opitmal aerodynamics and trajectories. Now at Airbus Canada

Felix Zapata : Former PhD candidate, and now postdoctoral fellow, working on AI-assisted surrogate modelling and optimization of rocket combustion chambers

Andrea Arroyo : PhD candidate working on aeroacoustics (self-noise) at high speed using LES and DNS, in collaboration with Sherbrooke University

Zhen Wei : PhD candidate working on physics-informed deep learning using GCNN, in collaboration with EPFL and Neural Concept. Now postdoc at MIT

Baptiste Corban : Former student in PhD track working on optimal flapping kinematics based on deep learning methods, and optimal kinematics of a robotic tidal turbine (collaboration with EPFL)

Ekhi Ajuria-Illaramendi : PhD student working on the acceleration of the resolution of the Poisson equation by Deep Learning : application to incompressible and plasma solvers. Now at Airbus Spain

Sandrine Berger (ETAP Project) : Postdoc focusing on Reinforcement Learning (RL) for optimal aerodynamics and controlling buffeting

Wagner Pinto (POLA3 project) : Postdoc working on acoustic source detection and propagation using Deep Learning techniques

Charlélie Laurent (POLA3 project) : Postdoc working sparse system identification, and application to inverse problems. Now at Nvidia

RESEARCH ACTIVITIES

Learning Fluid Flows (L2F)

Developing innovative Artificial Intelligence techniques to tackle fluid mechanics and aerodynamics problems

FUNDAMENTAL METHODS IN AI

Prospective works are carried out to apply AI techniques to propose, develop, or extent, AI methods in Deep Learning and Deep Reinforcement Learning.

Current Postdocs and PhDs :
Enrico Foglia : We study UQ estimation of deep learning models with collaboration with EPFL and mathematician at SUPAERO
Brice Martin & Laura Choquet : we investigate the partial observation problem, and its associated performance loss, in deep reinforcement problems.
Giovani Catalani : He explores potential amelioration of INR methods, in the scope or surrogate modeling in aerodynamics on large meshes.

ACCELERATING CFD SOLVERS

I currently lead research at DAEP to develop AI-based CFD solvers. Some part of the CFD solvers are replaced by a deep neural network. It contribute to the effort of improving the performance of CFD codes, especially on GPUs. This activity is realized in collaboration with AI experts (Jolibrain) and Cerfacs (Helios group : http://cerfacs.fr/helios/)

Current work :
Xintong Chen : Investigating reproducibility and symmetry preservation of AI-based linear solver coupled to non-linear physics. The main focus is to accelerate using AI existing incompressible code that need to solve a linear Poisson equation.

AI FOR FLOW RECONSTRUCTION & IDENTIFICATION

Part of the AI activity is dedicated to flow reconstruction. In particular, denoising problems and missing data reconstruction are under investigation. First results on denoising and pressure reconstruction from PIV measurements show promising possibilities for deep learning to complement CFD and experiment in fluid mechanics. Part of the POLA3 project will investigate AI-based identification methods for non-linear systems

Current work :
Rémi Ducotet : His work address the data-assimilation problem using neural operator to bring the gap between simulations and experiments in aerodynamics.
Mateus Costa : He works on localizing acoustic sources in experiment by solving the inverse problem with AI techniques, in particular deep neural networks.

AI FOR OPTIMIZATION

Prospective works are carried out to apply AI techniques to develop new optimization procedures. This current work is sustained by 2 research master projects and a postdoc fellow funded by DGA (ETAP projet), in particular using Reinforcement Learning. Part of the DGA project POLA3 will also investigate optimization problems.

Current work :
Jean Fesquet : He develop INRs for turbomachinery flow predictions and optimizations.

Aeroacoustics

Using high-fidelity simulations to investigate aeroacoustics sources and propagation in complex flows

My research focuses on using high-fidelity simulations (Navier-Stokes IC3 and LBM Palabos) to address aeroacoustics problems. Current investigations include :

Characterizing high-fidelity codes with acoustic benchmarks and applications. Typical cases ongoing are a drone’s rotor as well as trailing edge noise of high-Re airfoils.

Investigation of a jet-pump for propagation of acoustic waves in complex media, in collaboration with Stéphane Moreau from Sherbrooke Univ.

Hydrodynamic-Acoustic coupling using forced LES simulations, with applications to cavity flows and vortex-sound.

Current PhD :
Tiphaine Arnould : High-fidelity simulations (high-order spectral LES) of realistic (high Reynolds numbers, turbulent) flows over a cavity. She investigates the sound production, and the fundamental coupling between cavity modes and Rossiter modes.

Bio-inspired aerodynamics

Investigating the peculiar non-linear unsteady flows arising in bio-inspired geometries

My research focuses on using high-fidelity simulations (Navier-Stokes IC3 and LBM Palabos) to investigate bio-inspired flows. Three typical cases are studied currently :

A boxfish, typical case of of high-efficient bluff body. This is investigated using LBM and experiment in collaboration with V. Chapin and E. Gowree

A dragonfly wing with both corrugations and rear arc, showing overall good aerodynamic performance at low Reynolds number. Interesting complex flows emerged in this case, such as rapid transition to chaos.

Flapping wing (insects, birds). Kinematics of flapping wings can be complex, especially since the associated flow is highly non-linear and unsteady. Finding methods to optimize these kinematics to extract maximum power in a highly efficient manner is challenging.

Current work :

None at the time

OPPORTUNITIES 2026-2027

Postdoc position (3 years) for automatic meshing and flow predictions towards complex shape optimization. High-level candidates are encouraged to submit their CV and cover letter.

Internships at a master level are also possible. Feel free to send your CV and cover letter by mail.

PUBLICATIONS & HIGHLIGHTS

JFM perspectives and webinar

I was honored to be accepted to publish a JFM perspectives "Routes towards an effective AI in CFD : an epistemological and technical perspective". This paper lies at the interface between philosophy, AI and fluid mechanics. The full paper is available here : https://doi.org/10.1017/jfm.2026.11256

I presented this work during the JFM webinar in June 2026. The record is available here :

HDR : January 2026

In 2026 I sucessfully passed my HDR (Habilitation to Direct Research) entitled "On the effectiveness of AI applited to CFD". Jury members were P. Gallinari (Sorbonne Université), Andrea Da Ronch (Southampton University), Gianluca Iaccarino (Stanford University), Paola Cinnella (Sorbonne Université), Stéphane Moreau (Univresité de Sherbrooke) and Nicolas Binder (ISAE-SUPAERO).

PhySimAI 2025-2026

The integrative program PhySimAI in ANITI is dedicated to developing open-source library for surrogate modeling of very large problems, like fluid flows or solid mechanics. First release very soon.

ANITI 2024

The AI Cluster in Toulouse ANITI has been accepted in 2024. I will co-chair the chair HAILSED with Alena Kopanicakova and Paul Novelo, dedicated to Hybrid AI for Simulations and Designs. This chair will also be part of the integrated program AI4SAVE to coordinate actions involving AI for Simulations and Virtual Sensors with multiple industrial partners such as Airbus, Liebherr etc.

AI WORKSHOP COMPETITION 2024

Giovani Catalani won the 3rd prize over 125 candidates during the AI for simulation challenge proposed by IRT. We proposed a method based on "Multiscale Implicit Neural Representations".

https://www.linkedin.com/posts/giovanni-catalani-2a87a4136_machinelearning-lips-activity-7179083821408010242-HAXW?utm_source=share&utm_medium=member_desktop

INVITED TALK 2023 CANADA

I was invited in 2023 to present a Keynote at Sherbrooke University (Canada) during the CSME conference on "Combining artificial intelligence and CFD : recent progress and challenges".

https://lmfteus.wordpress.com/csme-cfdcanada-2023-international-congress/

NEURAL CONCEPT

A demonstration by Neural Concept (https://neuralconcept.com/) of our common research on deep learning for fast predictions of compressible flows is now available here

ANR FLOCCON

The ANR project FLOCCON, investigating a possible futur path of CFD through "Accelerating flow computation using convolutional networks", has been obtained. 2 PhDs positions are now available.

INVITED TALK ITN Magister

Invited talk at the ITN Magister Workshop, 15th September 2020.

INVITED TALK PASC19

Invited talk at the PASC19 Conference in Zurich on Machine Learning and High Performance Computing.

https://scholar.google.fr/citations?user=tWRCvmEAAAAJ&hl=fr

COLLABORATIONS & PARTNERS

Sureli, a group at ISAE-Supaero focusing on fundamentals and applications of Reinforcement Learning methods (https://sureli.github.io/).

Helios group with CERFACS : High Performance Learning for Computational Phyiscs (http://cerfacs.fr/helios/)

EPFL (Switzerland) with Pascal Fua from CVlab (https://www.epfl.ch/labs/cvlab/research/)

Université de Sherbrooke (Canada) with Stéphane Moreau on DNS and AI for aeroacoustics.