Lecture goals
By attending the lecture, students will be able to:
- Understand physical principles, advantages, and limitations of state-of-the-art and emerging measurement techniques (e.g., PIV, LPT, Event-based imaging).
- Integrate physical governing equations into machine learning frameworks to ensure that data-driven predictions remain physically consistent.
- Implement dimensionality reduction and modal decomposition techniques to extract and analyze dominant coherent structures in complex flows.
- Design and optimize "Smart Sensing" strategies, utilizing compressive sensing and optimal sensor placement to reconstruct full flow fields from sparse or limited data points.
- Bridge the gap between experiment and simulation by applying modern data assimilation and fusion techniques.
- Get an insight in robust surrogate models for real-time flow prediction, focusing on fluid-specific features and automated feature engineering (SINDy).
- Evaluate model reliability through advanced uncertainty quantification, using probabilistic and Bayesian approaches to assess errors in both sensors and ML models.
- Execute a research project, demonstrating the acquired ability to independently design an experiment, capture raw data, and apply a sophisticated ML-based pipeline to extract actionable insights.
-
create a CFD-based parameter study, with the parameters being selected via latin hypercube sampling
-
create a surrogate model for given inputs and outputs by applying regression or classification techniques to experimental or numerical flow data
-
analyze the spatio-temporal behavior of large, high-dimensional CFD data by means of modal decomposition and derive a reduced-order model using cluster-based network modeling
-
train a neural network for active flow control by means of CFD-based deep reinforcement learning
Lecture content
Integration and interaction between advanced measurement techniques, machine learning, and numerical simulation. The course will cover:
- Time averaged and time resolved measurement of fluid flows using discrete sensors (pressure, shear stress, temperature) and imaging-based techniques (PIV, LPT, infrared thermography, pressure- or temperature-sensitive-paints (PSP/TSP), event-based imaging) and associated initial data processing. Introduction in low-latency real-time sensing
- Design of experiments and data generation
- Reduced-order modeling and coherent structure identification (PCA/POD, DMD, SPOD, Autoencoders)
- Fundamentals of machine learning: data representation, feature design, training and validation Neural networks and surrogate modeling for discrete and continuous flow predictions (focusing on fluid-specific features; PINNs-SINDy)
- Uncertainty quantification using probabilistic and Bayesian approaches
- Sparse sensing, compressive sensing, and optimal sensor placement
- Fusion of heterogeneous and multi-fidelity data
- Data assimilation techniques (filtering, nudging, Kalman methods, autoencoders and other nonlinear approaches) to combine experiments and simulations
- Hands-on tutorials
- Capstone Project: Students will design, execute, and analyze an independent measurement experiment using the techniques learned in the course
