arXiv subjects

Francesca di Mare

Publications and source records attributed to Francesca di Mare.

10 recordsLinked to original sources

arXiv2026

A fast Physics-Informed Neural Networks based approach to the 2D design of turbine blades

Rapid aerodynamic screening of turbomachinery blades across wide operating envelopes remains a major computational bottleneck in preliminary design, particularly for energy-conversion and storage systems such as emerging Carnot batteries. Physics-informed neural networks (PINNs) offer a mesh-free alternative to conventional CFD, yet convergence and accuracy often deteriorate for complex blade geometries and off-design flows. We propose a progressive Euler-PINN framework that (i) gradually relaxes boundary conditions from tunnel flow without a blade to full outlet static pressure, and (ii) employs a geometry-aware dynamic loss-weighting scheme that intensifies residual penalties near highly curved boundaries. To the best of our knowledge, this is the first study to deploy a single PINN workflow for large-scale, engineering-grade screening of turbomachinery blade families across multiple operating conditions, covering ten NACA6 variants and 30 subsonic operating points. The proposed framework achieves CFD-comparable accuracy for pressure and velocity fields while reducing the computational cost required for family-wide blade screening. These results establish the method as a practical surrogate for two-dimensional turbomachinery blade pre-design and optimisation.

physics.flu-dyn↗

arXiv2026

Influence of Prandtl number on heat transfer over a permeable wall

The work considers a fully turbulent flow with heat transfer in a channel half-filled with an array of cubes based on the work of Breugem and Boersma (2005) and Chandesris et al. (2013), at $\mathrm{Re}_\mathrm{bulk} = 5485$ and three different Prandtl numbers, $\mathrm{Pr} = 0.71, 0.1, 0.05$. The temperature is modelled as a passive scalar and two different boundary condition configurations are simulated. The influence of the Prandtl number on the mean temperature, its variance and the terms of the temperature budget is highlighted, including the analysis of the distribution and relative importance of the turbulent heat transfer, molecular diffusion, tortuosity and Brinkman terms near the porous-fluid interface. The latter two has been found to be insignificant for the highest $\mathrm{Pr}$. A set of terms, typically neglected during the upscaling procedure (related to the Taylor expansion of the filtered variables), is analysed for the first time for the turbulent heat transfer at the porous-fluid interface, and are found to be significant at low $\mathrm{Pr}$. The upscaled fields are evaluated with three different kernels forming cellular average, linear (i.e., tent kernel), quadratic and cubic, and the influence of the chosen filter is additionally studied.

physics.flu-dyn↗

arXiv2026

Low Reynolds number flow in a packed bed of rotated bars

The present study focuses on the gas flow through an experiment-scale modular packed bed reactor consisting of square bars, arranged in layers. Each layer is rotated by $30^\circ$ resulting in a complex shape of the void spaces between the bars. Particle Image Velocimetry measurement results inside and on top of the studied system are presented for particle-based Reynolds numbers of 100 and 200, and used as validation data for two sets of particle-resolved numerical simulations, using boundary conforming meshing strategy and treating the solid boundaries via the blocked-off method. The flow inside the bed is largely independent from the Reynolds number and seems to be determined by the geometry of the void spaces. The flow in the freeboard is dominated by the presence of slowly dissipating jets downstream of the bed, which are characterized by unsteady oscillations at the higher Reynolds number. The numerical results obtained with both methods are in good agreement with the measurements, both inside and above the bed. However, stronger deviations between the results can be observed in the freeboard and can be traced to numerical properties of the current simulation approaches.

physics.flu-dyn↗

arXiv2026

Modelling of pressure drop in periodic square-bar packed beds

Understanding fluid flow through porous media with complex geometries is essential for improving the design and operation of packed-bed reactors. Most existing studies focus on spherical packings, having as a consequence that accurate models for irregular interstitial geometries are scarce. In this study, we numerically investigated the flow through a set of packed-bed geometries consisting of square bars stacked on top of each other and arranged in disk-shaped modules. Rotation of each module allows the generation of a variety of geometrical configurations at Reynolds numbers of up to 200 (based on the bar size). Simulations were carried out using the open-source solver OpenFOAM. Selected cases (e.g., $α= 30^\circ$, $\mathrm{Re}_\mathrm{p} = 100, 200$) were compared against Particle Image Velocimetry measurements. Results reveal that, based on the relative rotation angle, the realized geometries can be classified as channel-like ($α\leq 10^\circ$) and lattice-like ($α\geq 15^\circ$), fundamentally altering the friction factor. Furthermore, the maximum friction factor obtained in the creeping regime occurred at $α= 25^\circ$, whereas in the inertial regime, this occurred at $α= 60^\circ$. The module-equivalent diameter, based on the angle-dependent wetted surface area, collapses the friction factor onto the Ergun correlation and yields good permeability predictions for the lattice-like geometries.

physics.flu-dyn↗

arXiv2026

Comparison of Generative Learning Methods for Turbulence Surrogates

Numerical simulations of turbulent flows present significant challenges in fluid dynamics due to their complexity and high computational cost. High resolution techniques such as Direct Numerical Simulation (DNS) and Large Eddy Simulation (LES) are generally not computationally affordable, particularly for technologically relevant problems. Recent advances in machine learning, specifically in generative probabilistic models, offer promising alternatives as surrogates for turbulence. This paper investigates the application of three generative models - Variational Autoencoders (VAE), Deep Convolutional Generative Adversarial Networks (DCGAN), and Denoising Diffusion Probabilistic Models (DDPM) - in simulating a von Kármán vortex street around a fixed cylinder projected into 2D, as well as a real-world experimental dataset of the wake flow of a cylinder array. Training data was obtained by means of LES in the simulated case and Particle Image Velocimetry (PIV) in the experimental case. We evaluate each model's ability to capture the statistical properties and spatial structures of the turbulent flow. Our results demonstrate that DDPM and DCGAN effectively replicate all flow distributions, highlighting their potential as efficient and accurate tools for turbulence surrogacy. We find a strong argument for DCGAN, as although they are more difficult to train (due to problems such as mode collapse), they show the fastest inference and training time, require less data to train compared to VAE and DDPM, and provide the results most closely aligned with the input stream. In contrast, VAE train quickly (and can generate samples quickly) but do not produce adequate results, and DDPM, whilst effective, are significantly slower at both, inference and training time.

physics.flu-dyn↗

arXiv2025

A Coupled CFD Framework for Combustor Turbine Interaction in a Research Aeroengine

This work presents a fully coupled combustor turbine simulation framework applied to the MYTHOS aeroengine, developed within the Horizon Europe project MYTHOS, aimed at assessing the impact of Sustainable Aviation Fuels (SAFs) and hydrogen on next generation propulsion systems. The numerical setup features a dynamic, bidirectional coupling between a pressure-based solver with detailed finite rate chemistry, deployed in the combustor, and a density-based turbomachinery solver employing tabulated thermochemistry for efficiency, used for the turbine. The coupling is realised through a flux-averaging methodology that ensures conservative exchange of flow quantities and allows flow in arbitrary directions across the interface. Previous validation steps of presented methodology have shown the viability of the approach and are also shortly reviewd. The paper focuses on the chemistry handling strategy that guarantees thermochemical consistency between the two solvers. Coupled reacting simulations at cruise operating conditions demonstrate the capability of the framework to capture combustor generated hot streaks transport and their influence on turbine aerothermal loading. Comparison with segregated simulations of the two components shows that coupling captures the highly unsteady temperature and flow distributions at the turbine inlet and across the blade rows. Whilst mean aerodynamic loading are essentially unchanged, a realistic circumferential variability in blade thermal loading can be observed in the coupled simulations, thus establishing a consistent foundation for future studies on the effects of alternative fuels on core engine components.

physics.flu-dyn↗

arXiv2025

When do World Models Successfully Learn Dynamical Systems?

In this work, we explore the use of compact latent representations with learned time dynamics ('World Models') to simulate physical systems. Drawing on concepts from control theory, we propose a theoretical framework that explains why projecting time slices into a low-dimensional space and then concatenating to form a history ('Tokenization') is so effective at learning physics datasets, and characterise when exactly the underlying dynamics admit a reconstruction mapping from the history of previous tokenized frames to the next. To validate these claims, we develop a sequence of models with increasing complexity, starting with least-squares regression and progressing through simple linear layers, shallow adversarial learners, and ultimately full-scale generative adversarial networks (GANs). We evaluate these models on a variety of datasets, including modified forms of the heat and wave equations, the chaotic regime 2D Kuramoto-Sivashinsky equation, and a challenging computational fluid dynamics (CFD) dataset of a 2D Kármán vortex street around a fixed cylinder, where our model is successfully able to recreate the flow.

math.NA↗

arXiv2023

High-resolution measurements and simulations of the flow through a packed bed and its freeboard region

The present study focuses on the assessment of the performance of a Finite Volume Method based, particle-resolved simulation approach to predict the flow through a model packed-bed consisting of 21 layers of spheres arranged in the body centred cubic packing. The unsteady flow developing in the freeboard is also considered. Two highly-resolved large eddy simulation were preformed, for two Reynolds numbers, 300 and 500, based on the particle diameter, employing a polyhedral, boundary-conforming mesh. The geometry and the flow conditions are set to reproduce the flow conditions investigated in the experiment carried out by Velten and Zähringer using Particle Image Velocimetry. The numerical results compare favourably with the measurements both inside and above the bed. The effect of differences arising between the physical and numerical configurations are thoroughly discussed alongside the impact of meshing strategy on the accuracy of the predictions.

physics.flu-dyn↗

arXiv2023

Generalization capabilities of conditional GAN for turbulent flow under changes of geometry

Turbulent flow consists of structures with a wide range of spatial and temporal scales which are hard to resolve numerically. Classical numerical methods as the Large Eddy Simulation (LES) are able to capture fine details of turbulent structures but come at high computational cost. Applying generative adversarial networks (GAN) for the synthetic modeling of turbulence is a mathematically well-founded approach to overcome this issue. In this work, we investigate the generalization capabilites of GAN-based synthetic turbulence generators when geometrical changes occur in the flow configuration (e.g. aerodynamic geometric optimization of structures such as airfoils). As training data, we use the flow around a low-pressure turbine (LPT) stator with periodic wake impact obtained from highly resolved LES. To simulate the flow around a LPT stator, we use the conditional deep convolutional GAN framework pix2pixHD conditioned on the position of a rotating wake in front of the stator. For the generalization experiments we exclude images of wake positions located at certain regions from the training data and use the unseen data for testing. We show the abilities and limits of generalization for the conditional GAN by extending the regions of the extracted wake positions successively. Finally, we evaluate the statistical properties of the synthesized flow field by comparison with the corresponding LES results.

physics.flu-dyn↗

arXiv2022

Generative Modeling of Turbulence

We present a mathematically well founded approach for the synthetic modeling of turbulent flows using generative adversarial networks (GAN). Based on the analysis of chaotic, deterministic systems in terms of ergodicity, we outline a mathematical proof that GAN can actually learn to sample state snapshots form the invariant measure of the chaotic system. Based on this analysis, we study a hierarchy of chaotic systems starting with the Lorenz attractor and then carry on to the modeling of turbulent flows with GAN. As training data, we use fields of velocity fluctuations obtained from large eddy simulations (LES). Two architectures are investigated in detail: we use a deep, convolutional GAN (DCGAN) to synthesise the turbulent flow around a cylinder. We furthermore simulate the flow around a low pressure turbine stator using the pix2pixHD architecture for a conditional DCGAN being conditioned on the position of a rotating wake in front of the stator. The settings of adversarial training and the effects of using specific GAN architectures are explained. We thereby show that GAN are efficient in simulating turbulence in technically challenging flow problems on the basis of a moderate amount of training data. GAN training and inference times significantly fall short when compared with classical numerical methods, in particular LES, while still providing turbulent flows in high resolution. We furthermore analyse the statistical properties of the synthesized and LES flow fields, which agree excellently. We also show the ability of the conditional GAN to generalize over changes of geometry by generating turbulent flow fields for positions of the wake that are not included in the training data.

physics.flu-dyn↗