Developing Neural Surrogates for Complex Flow Physics

Institut

Lehrstuhl für Aerodynamik (TUM-ED)

Typ

Semesterarbeit / Masterarbeit /

Inhalt

theoretisch /  

Beschreibung

Developing new machine learning architectures for complex multiphase flow physics (Refer Figure 1) requires models that can represent interactions across sharp interfaces, evolving topologies, and widely separated spatial and temporal scales. Unlike standard data-driven approaches, these architectures should embed physical structure directly into the learning process, such as conservation of mass and momentum, phase-specific transport laws, interfacial tension effects, and closure relationships for unresolved phenomena. Promising directions include neural operators or transformer-based architectures for learning mappings between flow fields under varying conditions, physics-informed losses for enforcing governing equations, and hybrid models that combine high-fidelity simulations with experimental data. By integrating inductive biases from fluid mechanics with scalable deep learning methods, these architectures can improve prediction accuracy, generalize beyond training regimes, and provide faster surrogates for design, control, and uncertainty quantification in multiphase flow systems.
 

Milestones
• Understanding limitations of current architectures to capture discontinuities.
• Implementing and experimenting with the newly developed architecture.
• Benchmarking against state-of-the-art (SOTA) architectures.
 

Requirements
• Good understanding of SOTA/robust neural architectures: FNO (Li et al. 2020), SwinV2 transformer (Liu et al. 2021), Diffusion Transformers (Peebles & Xie 2023).
• Strong Python and Pytorch understanding.
• Ability to understand and debug large repositories.
• Ability to work independently.
 

Contact
Harish Ramachandran harish.ramachandran@tum.de with the subject ”Interested in contributing
to the development of neural surrogates for complex flow physics”. Please also attach your
CV, current grade report, and link to Github (if worked on any open-source projects).
 

References
Li, Z., Kovachki, N., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A. & Anandkumar,
A. (2020), ‘Fourier neural operator for parametric partial differential equations’, arXiv preprint
arXiv:2010.08895 .
Liu, Z., Hu, H., Lin, Y., Yao, Z., Xie, Z.,Wei, Y., Ning, J., Cao, Y., Zhang, Z., Dong, L. et al. (2021),
Swin transformer v2: Scaling up capacity and resolution. in 2022 ieee, in ‘CVF Conference
on Computer Vision and Pattern Recognition (CVPR)’, pp. 11999–12009.

Möglicher Beginn

sofort

Kontakt

M.Sc.Harish Ramachandran
Raum: 5506.01.681
Tel.: +49 (89) 289 - 16743
harish.ramachandrantum.de

Ausschreibung