arXiv subjects

Stefania Cherubini

Publications and source records attributed to Stefania Cherubini.

7 recordsLinked to original sources

arXiv2025

Large-scale streak instabilities of transitional channel flow

The emergence of large-scale spatial modulations of turbulent channel flow, as the Reynolds number is decreased, is addressed numerically using the framework of linear stability analysis. Such modulations are known as the precursors of laminar-turbulent patterns found near the onset of relaminarisation. A synthetic two-dimensional base flow is constructed by adding finite-amplitude streaks to the turbulent mean flow. The streak mode is chosen as the leading resolvent mode from linear response theory. Besides, turbulent fluctuations can be taken into account or not by using a simple Cess eddy viscosity model. The linear stability of the base flow is considered by searching for unstable eigenmodes with wavelengths larger than the base flow streaks. As the streak amplitude is increased in the presence of the turbulent closure, the base flow loses its stability to a large-scale modulation below a critical Reynolds number value. The structure of the corresponding eigenmode, its critical Reynolds number, its critical angle and wavelengths are all fully consistent with the onset of turbulent modulations from the literature. The existence of a threshold value of the Reynolds number is directly related to the presence of an eddy viscosity, and is justified using an energy budget. The values of the critical streak amplitudes are discussed in relation with those relevant to turbulent flows.

physics.flu-dyn↗

arXiv2025

Active learning of data-assimilation closures using Graph Neural Networks

The spread of machine learning techniques coupled with the availability of high-quality experimental and numerical data has significantly advanced numerous applications in fluid mechanics. Notable among these are the development of data assimilation and closure models for unsteady and turbulent flows employing neural networks (NN). Despite their widespread use, these methods often suffer from overfitting and typically require extensive datasets, particularly when not incorporating physical constraints. This becomes compelling in the context of numerical simulations, where, given the high computational costs, it is crucial to establish learning procedures that are effective even with a limited dataset. Here, we tackle those limitations by developing NN models capable of generalizing over unseen data in low-data limit by: i) incorporating invariances into the NN model using a Graph Neural Networks (GNNs) architecture; and ii) devising an adaptive strategy for the selection of the data utilized in the learning process. GNNs are particularly well-suited for numerical simulations involving unstructured domain discretization and we demonstrate their use by interfacing them with a Finite Elements (FEM) solver for the supervised learning of Reynolds-averaged Navier-Stokes equations. We consider as a test-case the data-assimilation of meanflows past generic bluff bodies, at different Reynolds numbers 50>=Re>=150, characterized by an unsteady dynamics. We show that the GNN models successfully predict the closure term; remarkably, these performances are achieved using a very limited dataset selected through an active learning process ensuring the generalization properties of the RANS closure term. The results suggest that GNN models trained through active learning procedures are a valid alternative to less flexible techniques such as convolutional NN.

physics.flu-dyn↗

arXiv2024

Bayesian inference of physics-based models of acoustically-forced laminar premixed conical flames

We perform twenty experiments on an acoustically-forced laminar premixed Bunsen flame and assimilate high-speed footage of the natural emission into a physics-based model containing seven parameters. The experimental rig is a ducted Bunsen flame supplied by a mixture of methane and ethylene. A high-speed camera captures the natural emission of the flame, from which we extract the position of the flame front. We use Bayesian inference to combine this experimental data with our prior knowledge of this flame's behaviour. This prior knowledge is expressed through (i) a model of the kinematics of a flame front moving through a model of the perturbed velocity field, and (ii) a priori estimates of the parameters of the above model with quantified uncertainties. We find the most probable a posteriori model parameters using Bayesian parameter inference, and quantify their uncertainties using Laplace's method combined with first-order adjoint methods. This is substantially cheaper than other common Bayesian inference frameworks, such as Markov Chain Monte Carlo. This process results in a quantitatively-accurate physics-based reduced-order model of the acoustically forced Bunsen flame for injection velocities ranging from 1.75 m/s to 2.99 m/s and equivalence ratio values ranging from 1.26 to 1.47, using seven parameters. We use this model to evaluate the heat release rate between experimental snapshots, to extrapolate to different experimental conditions, and to calculate the flame transfer function and its uncertainty for all the flames. Since the proposed model relies on only seven parameters, it can be trained with little data and successfully extrapolates beyond the training dataset. Matlab code is provided so that the reader can apply it to assimilate further flame images into the model.

physics.flu-dyn↗

arXiv2023

Enhancing Data-Assimilation in CFD using Graph Neural Networks

We present a novel machine learning approach for data assimilation applied in fluid mechanics, based on adjoint-optimization augmented by Graph Neural Networks (GNNs) models. We consider as baseline the Reynolds-Averaged Navier-Stokes (RANS) equations, where the unknown is the meanflow and a closure model based on the Reynolds-stress tensor is required for correctly computing the solution. An end-to-end process is cast; first, we train a GNN model for the closure term. Second, the GNN model is introduced in the training process of data assimilation, where the RANS equations act as a physics constraint for a consistent prediction. We obtain our results using direct numerical simulations based on a Finite Element Method (FEM) solver; a two-fold interface between the GNN model and the solver allows the GNN's predictions to be incorporated into post-processing steps of the FEM analysis. The proposed scheme provides an excellent reconstruction of the meanflow without any features selection; preliminary results show promising generalization properties over unseen flow configurations.

physics.flu-dyn↗

arXiv2022

Collective secondary instabilities: an application to three-dimensional boundary-layer flow

In some linearly unstable flows, secondary instability is found to have a much larger wavelength than that of the primary unstable modes, so that it cannot be recovered with a classical Floquet analysis. In this work, we apply a new formulation for capturing secondary instabilities coupling multiple length scales of the primary mode. This formulation, based on two-dimensional stability analysis coupled with a Bloch waves formalism originally described in Schmid et al. (2017), allows to consider high-dimensional systems resulting from several repetitions of a periodic unit, by solving an eigenproblem of much smaller size. Collective instabilities coupling multiple periodic units can be thus retrieved. The method is applied on the secondary stability of a swept boundary-layer flow subject to stationary cross-flow vortices, and compared with Floquet analysis. Two multi-modal instabilities are recovered: for streamwise wavenumber $α_v$ close to zero, approximately twelve sub-units are involved in large-wavelength oscillations; whereas a staggered pattern, characteristic of subharmonic instabilities, is observed for $α_v = 0.087$.

physics.flu-dyn↗

arXiv2022

Turbulent transition in a channel with superhydrophobic walls: the effect of roughness anisotropy

Superhydrophobic surfaces dramatically reduce skin friction of overlying liquid flows. These surfaces are complex and numerical simulations usually rely on models for reducing this complexity. One of the simplest consists in finding an equivalent boundary condition through an homogenisation procedure, which in the case of channel flow over oriented riblets, leads to the presence of a small spanwise component in the homogenized base flow velocity. This work aims at investigating the influence of such a three-dimensionality of the base flow on stability and transition in a channel with walls covered by oriented riblets. Linear stability for this base flow is investigated: a new instability region, linked to cross-flow effects, is observed. Tollmien-Schlichting waves are also retrieved but the most unstable are three-dimensional. Transient growth is also affected as oblique streaks with non-zero streamwise wavenumber become the most amplified perturbations. When transition is induced by Tollmien-Schlichting waves, after an initial exponential growth regime, streaky structures with large spanwise wavenumber rapidly arise. Modal mechanisms appear to play a leading role in the development of these structures and a secondary stability analysis is realised to retrieve successfully some of their characteristics. The second scenario, initiated with crossflow vortices, displays a strong influence of nonlinearities. The flow develops into large quasi spanwise-invariant structures before breaking down to turbulence. Secondary stability on the saturated cross-flow vortices sheds light on this stage of transition. In both cases, cross-flow effects dominate the flow dynamics, suggestings the need to consider these effects when modelling superhydrophobic surfaces.

physics.flu-dyn↗

arXiv2018

Time-stepping and Krylov methods for large-scale instability problems

With the ever increasing computational power available and the development of high-performances computing, investigating the properties of realistic very large-scale nonlinear dynamical systems has been become reachable. It must be noted however that the memory capabilities of computers increase at a slower rate than their computational capabilities. Consequently, the traditional matrix-forming approaches wherein the Jacobian matrix of the system considered is explicitly assembled become rapidly intractable. Over the past two decades, so-called matrix-free approaches have emerged as an efficient alternative. The aim of this chapter is thus to provide an overview of well-grounded matrix-free methods for fixed points computations and linear stability analyses of very large-scale nonlinear dynamical systems.

physics.flu-dyn↗