Nationality: Spanish
Funding entity and Program: Ministerio de Ciencia, Innovación y Universidades. FPU program.
PhD Thesis
AI-Based sensing of turbulent wall-bounded flows
Date of Defence: 09 December 2024
Supervisors
Stefano Discetti (UC3M) and Andrea Ianiro (UC3M)
Abstract
This thesis investigates the estimation of the velocity field in a channel flow from nonintrusivewall-embedded sensors. This is crucial for the development of active control strategies in wall-bounded turbulent flows. Flow estimation based on wall-measured quantities is a longstanding challenge. The main objective of this research is to develop data-driven models to predict the three-dimensional fluid flow behaviour.
Datasets from direct numerical simulations were employed to train deep-learning models. Three-dimensional convolutional neural networks with adversarial training were shown to accurately predict flow fields from wall measurements, with a significant reduction in computational cost with respect to planar estimators. In particular, the proposed network is capable of estimating wall-attached coherent structures due to their footprint being sensed at the wall. The proposed neural network architecture demonstrated excellent performance even in the presence of noise. Furthermore, the effect of reducing the amount of information available at the wall has been explored. Pressure measurements would provide better flow reconstructions if the number of sensors is large enough to sample flow scales properly, while streamwise wall shear stress should be preferred if the target is the measurement of streamwise velocity fluctuations and if the number of sensors is limited.
NEXTFLOW – Next-generation flow diagnostics for control.
ERC Starting Grant 2020: 949085
PhD in Aerospace Engineering UC3M 