Documents published in Scipedia

• Reynolds stress correction by machine learning methods with physical constraints

  T. Philibert, A. Ferrero, A. Iollo, F. Larocca

  eccomas2022.

  
  

  Abstract

  For the past three decade, Reynolds Average Navier-Stokes models have been widely used in the industry to simulate complex flows. However, these models suffer from limitations. Indeed there are still large discrepancies in the Reynolds stresses between the RANS model and high-fidelity data provided by DNS or experiments. This paper presents a strategy to correct the Menter SST model using an explicit algebraic model and two different neural networks: an multilayer perceptron (MLP) and a generative adversarial network (GAN). Moreover, in order to preserve the physical properties of the Reynolds stress tensor, we introduce a penalisation term in the loss of the GAN.

  Abstract
  For the past three decade, Reynolds Average Navier-Stokes models have been widely used in the industry to simulate complex flows. However, these models suffer from limitations. [...]

  • 4
  •  read

• Comparison of uncertainty quantification methods for mathematical problems in intermediate dimensions

  J. Peter, Q. Bennehard

  eccomas2022.

  
  

  Abstract

  The efficiency of multidimensional quadrature methods is compared for seven test functions in intermediate dimensions. Following this goal, the numerical evaluations of the mean and variance of the test functions, for two probability density functions, are assessed with respect to (wrt) their known exact values. The retained dimensions (3 to 6) correspond to the number of operational and geometrical uncertain parameters we plan to consider in a near future for realistic sensitivity analysis or robust designs. Most of the numerical quadrature methods rely on a generalized Polynomial Chaos (gPC) defined either by quadrature or by collocation. Two of the gPC collocation techniques, Basis Poursuit Denoise (BPdn) and Least Angle Regression (LAR), search for a sparse gPC while satisfying the collocation equations. Finally, the efficiency of the quadrature methods is discussed in relation with the regularity, the input dimension and the ANOVA decomposition of the test functions.

  Abstract
  The efficiency of multidimensional quadrature methods is compared for seven test functions in intermediate dimensions. Following this goal, the numerical evaluations of the [...]

  • 3
  •  read

• Cell-centered Lagrangian scheme for multi-material flows with equal pressure assumption

  R. Chauvin, G. Sébastien, M. Bastien, L. Antoine

  eccomas2022.

  
  

  Abstract

  The present document is motivated by the development and the study of diffuse interface strategies which does not require the use of geometric interface reconstructions. A simple diffuse interface strategy is proposed for the multimaterial diffusion equation. While it is possible to consider only one average temperature per mixed cell, it is known [7] that standard harmonic or arithmetic homogeneous methods are not accurate on the simple 'sandwich' problem when working with coarse meshes. This has direct consequences for radiation-hydrodynamics applications. The numerical strategy presented here may be seen as a natural extension of standard homogeneous model and understood as if the diffusion operator is integrated on the global cell (not the materials) taking into account several temperature (one per material). Obviously, the accuracy of the presented method, compared to exact geometric reconstruction based ones, is expected to be lower in the general case. However, we believe that the simplicity of the methodology introduced in the present document, its robustness and practicality for real physical applications makes it interesting for a large audience.

  Abstract
  The present document is motivated by the development and the study of diffuse interface strategies which does not require the use of geometric interface reconstructions. A [...]

  • 2
  •  read

• The effect of the tracer staggering on sea ice deformation fields

  C. Mehlmann

  eccomas2022.

  
  

  Abstract

  Sea ice models can simulate linear deformation characteristics (linear kinematic features) that are observed from satellite imagery. A recent study based on the viscous-plastic sea ice model highlights the role of the velocity placement on the simulation of linear kinematic features (LKFs) and concluded that the tracer staggering has a minor influence on the amount simulated LKFs. In this work we consider the same finite element discretization and show that on triangular meshes the placement of the sea ice tracers and the associated degrees of freedom (DoFs) have a strong influence on the amount of simulated LKFs. This behaivor can be explained by the change of the total number of DoFs associated with the tracer field. We analyze the effect on a benchmark problem and compare P1-P1, P0-P1, CR-P0 and CR-P1 finite element discretizations for the velocity and the tracers, respectively. The influence of the tracer placement is less strong on quadrilateral meshes as a change of the tracer staggering does not modify the total number of DoFs. Among the low order finite element approximations compared in this study, the CR-P0 finite element discretization resolves the deformation structure in the best way. The CR finite element for velocity in combination with the P0 discretization for tracer produces more LKFs than the P1-P1 finite element pair even on grids with fewer DoFs. This can not be achieved with the CR-P1 setup and therefore highlights the importance of the tracer discretization for the simulation of LKFs on triangular meshes.

  Abstract
  Sea ice models can simulate linear deformation characteristics (linear kinematic features) that are observed from satellite imagery. A recent study based on the viscous-plastic [...]

  • 2
  •  read

• Parameterising ocean-induced melt of an idealised Antarctic ice shelf using deep learning

  B. Bouissou, C. Burgard, N. Jourdain

  eccomas2022.

  
  

  Abstract

  The largest uncertainty when projecting the Antarctic contribution to sea-level rise comes from the ocean-induced melt at the base of Antarctic ice shelves. Current parameterisations used to link the hydrographic properties in front of ice shelves to the melt at their base struggle to accurately simulate basal melt patterns. We suggest that deep learning can be used to tackle this issue. We train a deep feed-forward neural network to emulate basal melt rates simulated by highly-resolved ocean simulations in an idealised geometry. We explore the advantages and limitations of this new approach through sensitivity studies varying hyperparameters, input variables and training choices. We show that large neural networks perform better, that the input format of the temperature and salinity matters most, and that the neural network can be applied to conditions outside of its training range if trained appropriately. The results are promising and we make recommendations for further work with this approach.

  Abstract
  The largest uncertainty when projecting the Antarctic contribution to sea-level rise comes from the ocean-induced melt at the base of Antarctic ice shelves. Current parameterisations [...]

  • 12
  •  read

• Data-Driven Reduced Order Modeling for Aerodynamic Flow Predictions

  D. Hines Chaves, P. Bekemeyer

  eccomas2022.

  
  

  Abstract

  During each aircraft program a vast amount of aerodynamics data has to be generated to judge performance, structural loads as well as handling qualities. Within the past years the usage of computational fluid dynamics has significantly increased providing accurate insights into aircraft behaviour at early design stages and therefore at least partially enabled the mitigation of costly design changes. However, fully relying on high fidelity aerodynamic data is still computational prohibitive. Hence, data-driven models have gained an increasing attention in recent years. These methods not only provide continuous models but also enable the inclusion of highly accurate aerodynamic results in time-critical environments. This paper aims at applying deep learning techniques to derive such models and compare them to state of the art reduced order modeling techniques. In particular, three deep learning methods, a Multilayer perceptron for distribution predictions, a Multi-layer perceptron for pointwise predictions and an Autoencoder coupled with an interpolation technique are compared to Proper Orthogonal Decomposition and Isomap with latent space interpolation. For all methods an efficient methodology to determine hyperparameters is outlined and applied. Results are presented for an Airbus provided XRF1 dataset which includes surface pressure distributions at various Mach numbers and angles of attack.

  Abstract
  During each aircraft program a vast amount of aerodynamics data has to be generated to judge performance, structural loads as well as handling qualities. Within the past years [...]

  • 26
  •  read

• Transition of separated flow over a bump under unsteady inflow conditions

  H. Marbona, A. Martínez-Cava, D. Rodríguez, E. Valero

  eccomas2022.

  
  

  Abstract

  Laminar flow separation has detrimental effects on the aerodynamics and performance of low pressure turbines (LPT). Flow separation is caused by the presence of adverse pressure gradient condition on the upper side of the blade past the suction peak, and is followed by laminar-to-turbulent transition and the subsequent turbulent mean reattachment due to the enhanced mixing. These phenomena characterise the size and dynamics of the separated flow, which are primarily dominated by the laminar-turbulent process. This study examines the influence of periodically-varying inflow conditions on the separated flow over a bump geometry at low Reynolds numbers. The geometry and flow conditions represent the upper surface of small LPT during high-altitude of flight. Direct numerical simulations are performed, in which a harmonic variation of the inlet total pressure is imposed, as a rough approximation of the passage of the upstream blade's wake. Three different frequencies with identical amplitude of the total pressure are simulated. The dynamics of the separated shear layer and the transition process are studied by separating the flow components correlated and un-correlated to the inflow frequency. Even moderate frequencies are found to have a strong effect in reducing the averaged size of the separated flow region, thus reducing the losses.

  Abstract
  Laminar flow separation has detrimental effects on the aerodynamics and performance of low pressure turbines (LPT). Flow separation is caused by the presence of adverse pressure [...]

  • 8
  •  read

• A comparison of machine learning methods for pressure coefficient prediction of an aeronautical configuration

  A. Gorgues, R. Castellanos, J. Bowen, E. Andrés

  eccomas2022.

  
  

  Abstract

  Machine learning entails powerful information processing algorithms that are relevant for modelling, optimization, and control of fluids. Currently, machine-learning capabilities are advancing at an incredible rate, and fluid mechanics is beginning to tap into the full potential of these powerful methods. Many tasks in fluid mechanics, such as reduced-order modelling, shape optimization and uncertainty quantification, may be posed as optimization and regression tasks. Machine learning can dramatically improve optimization performance and reduce convergence time. In this paper, the potential of tree-based machine learning techniques for the aerodynamic prediction of pressure coefficients of an AIRBUS XRF1 aircraft wing-body configuration has been assessed. For this purpose, a dataset including computational fluid dynamics (CFD) simulations has been employed to train the different models, with and without the use of proper orthogonal decomposition (POD) and having their hyperparameters values optimized to obtain the optimal subspace. A deep comparison of decision tree regressors and random forest algorithms has been performed, showing that the random forest regressor model performs better on all configurations.

  Abstract
  Machine learning entails powerful information processing algorithms that are relevant for modelling, optimization, and control of fluids. Currently, machine-learning capabilities [...]

  • 19
  •  read

• Developing a DEM-Coupled OpenFOAM solver for multiphysics simulation of additive manufacturing process

  N. Aminnia, A. Estupinan Donoso, B. Peters

  eccomas2022.

  
  

  Abstract

  Powder-based additive manufacturing technologies, specifically selective laser melting, are challenging to model due to the complex, interrelated physical phenomena that are present on multiple spatial scales, during the process. A key element of such models will be the detailed simulation of flow and heat transfer throughout the melt pool that is formed when the powder particles melt. Due to the high-temperature gradients that are generated inside the melt pool, the Marangoni force plays a key role in governing the flows inside the melt pool and deciding its shape and dimensions. On the other hand, the mass and heat transfer between the melt and the powder also has a significant role in shaping the melt pool at the edges. In this study, we modified an OpenFOAM solver (icoReactingMultiphaseInterFoam) coupled with an in-house developed DEM code known as eXtended Discrete Element Method or XDEM which models the dynamics and thermodynamics of the particles. By adding the Marangoni force to the momentum equation and also defining a laser model as a boundary condition for liquid-gas interface, the solver is capable of modeling the selective laser melting process from the moment of particle melting to the completion of the solidified track. The coupled solver was validated with an ice packed bed melting case and was used to simulate a multi-track selective laser melting process.

  Abstract
  Powder-based additive manufacturing technologies, specifically selective laser melting, are challenging to model due to the complex, interrelated physical phenomena that are [...]

  • 4
  •  read

• OpenFOAM model of fluid-structure interaction in dry wire drawing

  M. Vervaecke, D. Fauconnier, J. Degroote

  eccomas2022.

  
  

  Abstract

  A fluid-structure interaction model is employed to numerically investigate the interaction between the pressurized thin lubricant film and the radial, plastically deformed steel wire in a dry wire drawing process. A transient simulation is presented, with the implementation of a sliding fluid-structure interaction interface. Moreover, the fluid film has been calculated by the Navier-Stokes equations and the coupling with the wire model is performed by the IQN-ILS technique. This results on the one hand in the monitoring of the stresses and displacements of the structure and on the other hand in an observation of the hydrodynamic pressure build-up and wall shear stresses in the lubricant. Additionally, the evolution of the thickness of the fluid film is presented.

  Abstract
  A fluid-structure interaction model is employed to numerically investigate the interaction between the pressurized thin lubricant film and the radial, plastically deformed [...]

  • 8
  •  read