Search

The effect of hyperparameter selection in deep learning (DL) models for fluid dynamics remains an open question in the current scientific literature. Many authors report results using deep learning models. However, better insight is required to assess deep learning models' behavior, particularly for complex datasets such as turbulent signals. This study presents a meticulous investigation of the long short-term memory (LSTM) hyperparameters, focusing specifically on applications involving predicting signals in shock turbulent boundary layer interaction. Unlike conventional methodologies that utilize automated optimization techniques, this research explores the intricacies and impact of manual adjustments to the deep learning model. The investigation includes the number of layers, neurons per layer, learning rate, dropout rate, and batch size to investigate their impact on the model's predictive accuracy and computational efficiency. The paper details the iterative tuning process through a series of experimental setups, highlighting how each parameter adjustment contributes to a deeper understanding of complex, time-series data. The findings emphasize the effectiveness of precise manual tuning in achieving superior model performance, providing valuable insights to researchers and practitioners who seek to leverage long short-term memory networks for intricate temporal data analysis. The optimization not only refines the predictability of the long short-term memory in specific contexts but also serves as a guide for similar manual tuning in other specialized domains, thereby informing the development of more effective deep learning models. [ABSTRACT FROM AUTHOR]

The success of deep learning models in fluid dynamics applications will depend on their ability to handle sparse and noisy data accurately. This paper concerns the development of a deep learning model for reconstructing turbulent flow images from low-resolution counterparts encompassing noise. The flow is incompressible through a symmetric, sudden expansion featuring bifurcation, instabilities, and turbulence. The deep learning model is based on convolutional neural networks, in a high-performance, lightweight architecture. The training is performed by finding correlations between high- and low-resolution two-dimensional images. The study also investigates how to remove noise from flow images after training the model with high-resolution and noisy images. In such flow images, the turbulent velocity field is represented by significant color variations. The model's peak signal-to-noise ratio is 45, one of the largest achieved for such problems. Fine-grained resolution can be achieved using sparse data at a fraction of the time required by large-eddy and direct numerical simulation methods. Considering its accuracy and lightweight architecture, the proposed model provides an alternative when repetitive experiments are complex and only a small amount of noisy data is available. [ABSTRACT FROM AUTHOR]

The significant growth of artificial intelligence (AI) methods in machine learning (ML) and deep learning (DL) has opened opportunities for fluid dynamics and its applications in science, engineering and medicine. Developing AI methods for fluid dynamics encompass different challenges than applications with massive data, such as the Internet of Things. For many scientific, engineering and biomedical problems, the data are not massive, which poses limitations and algorithmic challenges. This paper reviews ML and DL research for fluid dynamics, presents algorithmic challenges and discusses potential future directions. [ABSTRACT FROM AUTHOR]

To link to full-text access for this article, visit this link: http://dx.doi.org/10.1016/j.paerosci.2005.10.001 Byline: Dimitris Drikakis (a), Fernando Grinstein (b), David Youngs (c) Abstract: A review on the implementation of modern computational fluid dynamics (CFD) methods for simulating low- and high-speed flows featuring turbulent mixing, instabilities and symmetry-breaking is presented. Results are reviewed for several complex flows including turbulent mixing associated with Richtmyer-Meshkov and Rayleigh-Taylor instabilities, incompressible flows in suddenly expanded geometries, shock-induced instabilities, free-shear instabilities and complex vortex dynamics. Many of the results presented in this review have been obtained by large eddy simulation (LES), monotone integrated LES (MILES) in particular. The computational challenges associated with complex flow physics and nonlinear behaviour of the numerics are discussed. Author Affiliation: (a) School of Engineering, Aerospace Sciences Department, Fluid Mechanics and Computational Science Group, Cranfield MK43 0AL, UK (b) Los Alamos National Laboratory, MS-F699, Los Alamos, NM 87545, USA (c) AWE, Aldermaston, Berkshire, RG7 4PR, UK

It is well established that the data reported for the daily number of infected cases during the first wave of the COVID-19 pandemic were inaccurate, primarily due to insufficient tracing across the populations. Due to the uncertainty of the first wave data mixed with the second wave data, the general conclusions drawn could be misleading. We present an uncertainty quantification model for the infected cases of the pandemic's first wave based on fluid dynamics simulations of the weather effects. The model is physics-based and can rectify a first wave data's inadequacy from a second wave data's adequacy in a pandemic curve. The proposed approach combines environmental seasonality-driven virus transmission rate with pandemic multiwave phenomena to improve statistical predictions' data accuracy. For illustration purposes, we apply the new physics-based model to New York City data. [ABSTRACT FROM AUTHOR]

Face mask filters—textile, surgical, or respiratory—are widely used in an effort to limit the spread of airborne viral infections. Our understanding of the droplet dynamics around a face mask filter, including the droplet containment and leakage from and passing through the cover, is incomplete. We present a fluid dynamics study of the transmission of respiratory droplets through and around a face mask filter. By employing multiphase computational fluid dynamics in a fully coupled Eulerian–Lagrangian framework, we investigate the droplet dynamics induced by a mild coughing incident and examine the fluid dynamics phenomena affecting the mask efficiency. The model takes into account turbulent dispersion forces, droplet phase-change, evaporation, and breakup in addition to the droplet–droplet and droplet–air interactions. The model mimics real events by using data, which closely resemble cough experiments. The study shows that the criteria employed for assessing the face mask performance must be modified to take into account the penetration dynamics of airborne droplet transmission, the fluid dynamics leakage around the filter, and reduction of efficiency during cough cycles. A new criterion for calculating more accurately the mask efficiency by taking into account the penetration dynamics is proposed. We show that the use of masks will reduce the airborne droplet transmission and will also protect the wearer from the droplets expelled from other subjects. However, many droplets still spread around and away from the cover, cumulatively, during cough cycles. Therefore, the use of a mask does not provide complete protection, and social distancing remains important during a pandemic. The implications of the reduced mask efficiency and respiratory droplet transmission away from the mask are even more critical for healthcare workers. The results of this study provide evidence of droplet transmission prevention by face masks, which can guide their use and further improvement. [ABSTRACT FROM AUTHOR]

This paper presents a detailed investigation of unsteady supersonic and hypersonic flows around spiked-blunt bodies, including the investigation of the effects of the flow field initialization on the flow results. Past experimental research has shown that if the geometry of a spiked-blunt body is such that a shock formation consisting of an oblique foreshock and a bow aftershock appears, then the flow may be unsteady. The unsteady flow is characterized by periodic radial inflation and collapse of the conical separation bubble formed around the spike (pulsation). Beyond a certain spike length the flow is 'stable', i.e. steady or mildly oscillating in the radial direction. Both unsteady and 'stable' conditions have been reported when increasing or decreasing the spike length during an experimental test and, additionally, hysteresis effects have been observed. The present study reveals that for certain geometries the numerically simulated flow depends strongly on the assumed initial flow field, including the occurrence of bifurcations due to inherent hysteresis effects and the appearance of unsteady flow modes. Computations using several different configurations reveal that the transient (initial) flow development corresponds to a nearly inviscid flow field characterized by a foreshock-aftershock interaction. When the flow is pulsating, the further flow development is not sensitive to initial conditions, whereas for an oscillating or almost 'steady' flow, the flow development depends strongly on the assumed initial flow field. [ABSTRACT FROM AUTHOR]

Conventional large-eddy simulation (LES) and monotone integrated LES (MILES) are tested in emulating the dynamics of transition to turbulence in the Taylor-Green vortex (TGV). A variety of subgrid scale (SGS) models and high-resolution numerical methods are implemented in the framework of both incompressible and compressible fluid flow equations. Comparisons of the evolution of characteristic TGV integral measures are made with previously reported and new direct numerical simulation (DNS) data. The computations demonstrate that the convective numerical diffusion effects in the MILES methods can consistently capture the physics of flow transition and turbulence decay without resorting to an explicit SGS model, while providing accurate prediction of established theoretical findings for the kinetic energy dissipation, energy spectra, enstrophy and kinetic energy decay. All approaches tested provided fairly robust computational frameworks. [ABSTRACT FROM AUTHOR]

The paper presents a computational investigation of instabilities in pulsatile flow through a three-dimensional stenosis. The computations have been conducted using a third-order high-resolution scheme and a non-linear multigrid algorithm. The simulations reveal the existence of unstable flow throughout the pulsatile cycle. The instability is manifested by asymmetric flow patterns––though the stenosis is axisymmetric––large flow variations in the cross-sectional planes, and swirling motion in the poststenotic region. The instability leads to a strongly disturbed flow several radii downstream of the stenosis, as well as spatio-temporal fluctuations of the circumferential shear stress and vorticity. [Copyright &y& Elsevier]

Endotracheal tubes (ETT) passed inside the human trachea witness tube bending at different angles, affecting the local fluid flow dynamics. This induces a variable mechanical ventilation performance across patients' comfortability levels. Our understanding of the local fluid flow dynamics phenomena is thus crucial to enhance the maneuverability of ETT under operation. For the first time to our knowledge, we shed light on ETT through computational fluid dynamics (CFD) to investigate the bending effect of ETT on the local airflow in volume-controlled mechanical ventilation. We considered an ETT with 180° arc bend configuration, including Murphy's eye. We identified several flow phenomena associated with the bending, such as flow asymmetries, secondary flows, and vortex dynamics throughout the tube. [ABSTRACT FROM AUTHOR]

This paper presents a review of computational uncertainties in scientific computing, as well as quantification of these uncertainties in the context of numerical simulations for thermo-fluid problems. The need for defining a measure of the numerical error that takes into account errors arising from different numerical building blocks of the simulation methods is discussed. In the above context, the effects of grid resolution, initial and boundary conditions, numerical discretization, and physical modeling constraints are presented. [ABSTRACT FROM AUTHOR]

This paper investigates the effect of surface roughness on fluid viscosity using molecular dynamics simulations. The three-dimensional model consists of liquid argon flowing between two solid walls whose surface roughness was modeled using fractal theory. In tandem with previously published experimental work, our results show that, while the viscosity in smooth channels remains constant across the channel width, in the presence of surface roughness it increases close to the walls. The increase of the boundary viscosity is further accentuated by an increase in the depth of surface roughness. We attribute this behavior to the increased momentum transfer at the boundary, a result of the irregular distribution of fluid particles near rough surfaces. Furthermore, although the viscosity in smooth channels has previously been shown to be independent of the strength of the solid-liquid interaction, here we show that in the presence of surface roughness, the boundary viscosity increases with the solid's wettability. The paper concludes with an analytical description of the viscosity as a function of the distance from the channel walls, the walls' surface roughness, and the solid's wetting properties. The relation can potentially be used to adjust the fluid dynamics equations for a more accurate description of microfluidic systems. [ABSTRACT FROM AUTHOR]