Student Thesis, Master Thesis: Machine Learning for Unsteady Aerodynamics (f/m/x)

Motivation

The numerical investigation of dynamic responses to atmospheric turbulence as well as structural and flight dynamic excitations is an important task during the aircraft design and certification process. Efficient and high-fidelity methods are desirable because large parameter spaces spanned by, for
example, Mach number, flight altitude, load case, and gust shape need to be covered [1, 2]. The state-of-the-art industry approach is to use the linear frequency domain (LFD) method that computes aerodynamic responses using computational fluid dynamics (CFD) by linearising the Reynoldsaveraged Navier–Stokes (RANS) equations around a steady-state solution. While this enables efficient simulations that account for steady aerodynamic nonlinearities such as shocks and boundary layer separation at transonic flight conditions, the method is only valid around the linearisation point, does not account for unsteady nonlinearities, and furthermore is not affordable in large-query scenarios. To account for dynamic nonlinearities, e.g. due to large amplitude gusts, simulations in the time domain are necessary. Solving the unsteady Reynolds-averaged Navier–Stokes (URANS) equations is a possible solution that comes with a computational cost that makes the method unfeasible if multiple parameter combinations are of interest. Hence, a method is thought after that enables fast predictions of the surface flow around an object for the described problem.

The Research Questions

A specific task description can be written after an exchange with the student considering the background, type of thesis (Bachelor Thesis / Master Thesis / Study Thesis), temporal constraints, etc.
Some main points are:

• Literature review regarding the proposed method and other approaches.
• Familiarisation with the surrogate modelling capabilities within SMARTy [6]. With a focus on Neural ODEs and GNNs.
• Design of experiments and data production using URANS for a database containing gust simulations.
• Investigation of the proposed methodology for gust loads.
• Improvement of the approach based on the findings.
• Writing of the thesis

Your Qualification

Mandatory
• Solid Python experience
• Interest in aerodynamics

As this is a student thesis no one is expected fulfil all the requirements. However, the candidate should
be confident to be able to quickly familiarise with the following topics.

• Being a student that needs a thesis topic. Preferably at TU Braunschweig
• Experience with ML
o Topics: GNNs, Neural ODEs
o Skills: PyTorch, PyTorch Geometric
• Experience with CFD
• Knowledge about unsteady aerodynamics