This thesis focuses on building a machine‑learning‑enabled digital twin for satellite constellations: a fast surrogate model that predicts coverage, latency, and other key metrics in milliseconds, replacing or augmenting slower physics simulations. Inspired by the work of T.I. Zohdi and co‑authors on ML‑enabled digital twins for Planet‑X constellations, you will design data pipelines, train ML models, and benchmark their accuracy and speed against existing simulation outputs.​

The emphasis is on combining a solid understanding of the underlying physics with modern ML tooling (e.g. scikit‑learn, PyTorch, JAX) and on creating an interface that can be integrated into optimization or interactive design tools.

Mini demo challenge (attach repo link in your application):

Write a small Python script or notebook that:

• Defines a simple analytic “coverage score” as a function of (altitude, inclination, number of satellites).

• Generates 100 random samples, trains any regression model (your choice), and compares predictions on 10 test samples.

• In the README (3-5 sentences): explain why you chose that ML library and what you would do differently for real constellation data.

The goal is simply to see your ML library choice and how you think about the problem—not to build something complex.

The goal is to see how you think about ML model choice, data handling, and performance.

Suggested reading:

• Zohdi, T.I. et al., “A machine-learning enabled digital-twin framework for the rapid design of satellite constellations for ‘Planet‑X’”.[msol.berkeley]

• “Building Data-Driven Satellite Digital Twins”, Journal of Space Operations & Communicator.[iafastro]

Application:
Interested candidates should submit their application to julius.pinsker@faps.fau.de including:

• Current transcript of records (grades).

• A concise CV (1 page).

• A link to the mini demo repository described above.

I look forward to receiving your innovative applications.

Kategorien:

Kontakt: