arXiv subjects

Stephane Moreau

Publications and source records attributed to Stephane Moreau.

3 recordsLinked to original sources

arXiv2025

Do you understand epistemic uncertainty? Think again! Rigorous frequentist epistemic uncertainty estimation in regression

Quantifying model uncertainty is critical for understanding prediction reliability, yet distinguishing between aleatoric and epistemic uncertainty remains challenging. We extend recent work from classification to regression to provide a novel frequentist approach to epistemic and aleatoric uncertainty estimation. We train models to generate conditional predictions by feeding their initial output back as an additional input. This method allows for a rigorous measurement of model uncertainty by observing how prediction responses change when conditioned on the model's previous answer. We provide a complete theoretical framework to analyze epistemic uncertainty in regression in a frequentist way, and explain how it can be exploited in practice to gauge a model's uncertainty, with minimal changes to the original architecture.

stat.ML↗

arXiv2024

Experimental evidence on tonal noise generation and its control in tip-leakage flows

The present study is first to provide experimental evidence of tonal noise and acoustic feedback by a stationary airfoil with a tip-gap placed between two side plates. Tones are linked to an aero-acoustic coupling using synchronized measurements between Particle Image Velocimetry and far-field microphone probes. To aid in the correlation analysis, the wall boundary condition on the pressure side of the airfoil was modified by introducing surface roughness. For tip leakage flows without roughness elements, the far-field acoustic spectra show peaks in the acoustic autospectra. These peaks are present regardless of the flow speed studied. The findings reveal that tones associated with tip-gap extend till 16.5 kHz in a low-noise facility. Roughness alters cross-flow velocity along the airfoil chord, influencing acoustics at matched Reynolds and Mach numbers. Roughness induces cancellation of spectral peaks and changes in the velocity and acoustic pressure correlation pattern between airfoil pressure and suction side. Modal shapes confirm the presence of coherent structures in the tip-gap region, with smooth surfaces exhibiting confined flow instabilities and non-dipolar noise sources around and beyond 10 kHz. At these high frequencies, the sound radiation mechanism can be uniquely attributed to neither a jet-like noise source nor these acoustic peaks can be uniquely attributed to coherent structures in the tip-gap region. On the contrary, coherence between airfoil sides is required for noise peaks, which stresses the importance of diffraction on the airfoil pressure side and acoustic feedback.

physics.flu-dyn↗

arXiv2023

A generalized wall-pressure spectral model for non-equilibrium boundary layers

This study uses high-fidelity simulations (DNS or LES) and experimental datasets to analyse the effect of non-equilibrium streamwise mean pressure gradients (adverse or favourable), including attached and separated flows, on the statistics of boundary layer wall-pressure fluctuations. The datasets collected span a wide range of Reynolds numbers ($Re_\theta$ from 300 to 23,400) and pressure gradients (Clauser parameter from $-0.5$ to 200). The datasets are used to identify an optimal set of variables to scale the wall pressure spectrum: edge velocity, boundary layer thickness, and the peak magnitude of Reynolds shear stress. Using the present datasets, existing semi-empirical models of wall-pressure spectrum are shown unable to capture effects of strong, non-equilibrium adverse pressure gradients, due to inappropriate scaling of wall pressure using wall shear stress, calibration with limited types of flows, and dependency on model parameters based on friction velocity, which reduces to zero at the detachment point. To address these short-comings, a generalized wall-pressure spectral model is developed with parameters that characterize the extent of the logarithmic layer and the strength of the wake. Derived from the local mean velocity profile, these two parameters inherently carry effect of the Reynolds number, as well as those of the non-equilibrium pressure gradient and its history. Comparison with existing models shows that the proposed model behaves well and is more accurate in strong-pressure-gradient flows and in separated-flow regions.

physics.flu-dyn↗