M/F PhD "Machine-Learning applied to fluid mechanics images"
- Gières, Isère
- CDD
- Temps-plein
The PhD will work within the LEGI research department (UMR5519) and the LIG (UMR 5217) at the University Grenoble Alpes. LEGI's research is focused on geophysical and industrial flows, using theoretical, experimental, and numerical approaches, while LIG contributes to the development of fundamental aspects of Computer Science to address conceptual, technological, and societal challenges. The position is within the Turbulent and Two-phase Flows team (LEGI) and the APTIKAL team (LIG). The project will focus on computer science aspects, encompassing theoretical work and code development, but the student could participate in X-ray measurement sessions at ESRF (European Synchrotron Research Facility, Grenoble).
Strong interactions with both laboratories are planned. This includes participating in team meetings, attending seminars, and other scientific activities. The computing resources of both LEGI and LIG are likely to be used. Finally, the project will involve collaborations with O. Desjardins (Cornell University), who will provide a numerical database for the project.In the context of applying generative AI to study spray formation mechanisms, there are several challenges to handle, particularly in relation to the availability of high-quality data considering the contextual information inherent to the cascade of mechanisms, where existing data might be noisy or incomplete.
Furthermore, fluid mechanics involves complex and nonlinear phenomena, and capturing the intricacies of spray formation requires sophisticated models that may not be straightforward to design, train, and interpret accurately. In this context, training generative models, especially complex ones, can be computationally intensive.
Developing a neural network specifically tailored to the problem at hand would result in an adapted and effective predictive model. The design of such a neural network would lead to an adapted and efficient prediction model.
This multidisciplinary project combines fluid mechanics and deep learning to advance research on spray formation mechanisms and on information retrieval via Physics-informed neural networks (PINNs). PINNs have recently emerged as a promising alternative for solving inverse problems and have successfully tackled several fluid mechanics topics, but their implementation for multiscale multi-physics systems has not yet been explored. The turbulent two-phase flow formed by gas-assisted atomization sprays poses a good candidate to expand PINNs models due to the recent advances in pioneering measurements with synchrotron X-ray and in high-fidelity numerical simulations.
The project aims to provide novel machine-learning methodologies for information exploration, deeper insights into spray formation, and alternative approaches for time-resolved 3D X-ray measurements.RequirementsResearch Field Engineering Education Level Master Degree or equivalentResearch Field Chemistry Education Level Master Degree or equivalentResearch Field Physics Education Level Master Degree or equivalentLanguages FRENCH Level BasicResearch Field Engineering Years of Research Experience NoneResearch Field Chemistry Years of Research Experience NoneResearch Field Physics Years of Research Experience NoneAdditional InformationAdditional commentsDegree desired: Master in Computer Science or applied mathematics
Expected skills :
- A good theoretical background and practical knowledge in machine learning.
- Experience in data and/or image processing.
- Know-how of programming in a scientific language. Website for additional job detailsWork Location(s)Number of offers available 1 Company/Institute Laboratoire des Ecoulements Géophysiques et Industriels Country France City GIERES GeofieldWhere to apply WebsiteContact CityGIERES WebsiteSTATUS: EXPIRED
EURAXESS