Description:
This thesis applies flow matching to estimate the parameters of thermoacoustic Van der Pol models, which describe the nonlinear limit-cycle behaviour of flame acoustic instabilities. A model trained once should estimate parameter uncertainty directly from noisy measurements – without retraining. Unlike classical identification methods that yield only a single point estimate, this approach delivers calibrated posterior distributions over all model parameters, quantifying uncertainty arising from noise, sparse sampling, and model-form assumptions.
Scope of Work:
- Literature review on VdP models and flow-matching parameter estimation
- Implementation of a simulator for different VdP variants
- Training a flow matching model for parameter posterior estimation
- Validation against classical identification methods
- Robustness study under noisy or incomplete measurement data
Requirements:
- Enrolled Master student in Mechanical Engineering, Energy Engineering, Data Science or similar
- Basic knowledge of nonlinear dynamics and combustion acoustics desirable
- Solid Python skills and experience with machine learning
- Structured and independent working style
- Good English or German language skills
For applications or further information, please contact:
Chair of Reactive Flows, lehre@rf.uni-hannover.de, please include your CV and a current transcript of records (grade overview) with your application.
