Session 1, 13:30 – 13:50
We present the development of the compressible flow solver in Nektar++ and some applications to subsonic flows. In particular, we briefly introduce the underlying numerics, namely the discontinuous Galerkin (DG) method and the flux reconstruction (FR) approach. We successively detail the connections between these two schemes and we show how it is possible to improve the stability of these methods by means of the consistent integration, highlighting the crucial aspects that needs to be taken into account when implementing dealiasing techniques based on such a concept. We also illustrate some relevant results to show the effectiveness of these dealiasing techniques. We finally present some applications to high-speed subsonic flows over a flat-plate in presence of an isolated hump.
Session 1, 13:50 – 14:10
Mesh refinement strategies are used to obtain typical flow features that occur in compressible flow past objects like a wing profile as accurate as possible with the least amount of computational effort. During the presentation, an overview will be given of the two strategies that are currently available in the Nektar++ library. These strategies are applied to typical compressible flow problems. First, a sensor-based refinement strategy will be presented where the mesh is deformed using the theory of linear elasticity. The sensor that is used to perform shock capturing is now used as an additional thermal stress term in the formulation of the linear elasticity equations. In this way, the elements that are surrounding the shock are shrunk and the spatial resolution is locally increased (r-refinement). Secondly, a continuous adjoint formulation for the compressible Navier-Stokes equations is presented that is used to perform goal-based error estimation and mesh refinement in the framework of high-order spectral/hp element methods. Goal-based error estimators are interesting mesh refinement tools since they incorporate the underlying physics of the problem that is solved. The adjoint problem prescribes the sensitivity of the domain with respect to the quantity of interest (typically Cl or Cd) and the local residual is weighted with the adjoint solution. In this way, we obtain an error indicator that can be used to adapt the mesh by using p-refinement such that the target quantity is approximated more accurately.
Session 1, 14:10 – 14:30
Viscoelastic free surface flows play a crucial role in a large variety of industrial applications such as inkjet printing, extrusion moulding or cable coating. The simulation of such flows is extremely challenging due to the presence of thin stress boundary layers near walls and stress concentrations in the vicinity of geometrical singularities. Resolving these boundary layers using traditional methods such as low order finite element techniques is often prohibitively expensive. Spectral element methods enable a much more efficient resolution of these layers. We have developed a DEVSS-G/DG spectral element algorithm for Oldroyd-B and Giesekus fluids combining a continuous Galerkin method for the solution of the conservation equations with a discontinuous Galerkin discretisation for the solution of the constitutive equation for the polymeric stress. To track the movement of the free surface, we have implemented an arbitrary Lagrangian Eulerian scheme, which involved the implementation of mesh moving capabilities in Nektar++. We will demonstrate the performance of our scheme on several benchmark problems such as the flow around a cylinder, extrusion problems and the thinning of liquid threads.
Session 1, 14:30 – 14:50
Aerodynamic efficiency is a key performance differentiator in modern Formula 1 where the flow is typically highly transient. Despite not being adequately equipped to deal with these complex transient flows classical RANS based methods still continue to be widely used in the industry. Full direct numerical simulations (DNS) of such flows will remain out of reach for the forceable future so, at present, large eddy simulation methods are believed to offer benefits for modeling such flows. Our present work aims to pioneer investigate the possible benefits of using high-order compact numerical methods for improving the correlation between experiment and CFD in Formula 1.
Session 1, 14:50 – 15:10
We propose a novel strip modelling method for the VIV prediction of long flexible risers in three- dimensional incompressible flow. In order to overcome the shortcomings of conventional strip theory-based 2D model, a conception of a finite thick strip is introduced in this model to finely resolve the small scale turbulence effects and three dimensionality of the flow around the riser. The most attractive feature of this model is that we construct a three-dimensional DNS model for each strip, which have local scale along the riser’s axial direction. The connection between these strips is achieved through the calculation of a tensioned beam model, which governs the dynamics of flexible body. A Fourier spectral/hp element method is employed to solve the 3D flow dynamics in the strip-domain, and then the VIV response prediction is achieved through the strip-structure interactions. Finally, numerical tests on both laminar and turbulent flows are presented to demonstrate the applicability of this approach.
Session 1, 15:10 – 15:30
The method of moving frames (MMF) has been successfully applied to conservational laws, diffusion equations, shallow water equations, and Maxwell’s equations on static/rotating and isotropic/anisotropic curved surfaces. But the code is implemented in the old version of 3.2 and is under imminent threat by the whims of Apple. In this talk, a blueprint of incorporating the MMF schemes into the current version of Nektar++ library is proposed as a collaborative work, which may end up with another MMF paper. Immediate follow-ups of this work such as applications to geometrical deformation problems will be also discussed.
Session 2, 16:00 – 16:20
Session 2, 16:20 – 16:40
Computational models of electrical activity in the heart are complicated systems with a large number of parameters spanning multiple time and space scales. Such models are subject to many types of uncertainty, which may reduce the reliability and applicability of simulation results for understanding how atrial fibrillation initiate and self-sustain. In this talk, I will introduce techniques that are being developed to quantify effects of uncertainty within computational models of atrial cells, and how we are starting to apply these techniques to tissue models.
Session 2, 16:40 – 17:00
The APESolver was revised and extended to qualify as CAA solver in a hybrid setup for the simulation of premixed combustion systems. In order to account for the different length and time scales of such systems, the simulation is split into an acoustics and an incompressible reactive LES part. While the in- house CFD code PRECISE-UNS is used for the latter task, Nektar’s APESolver was found to be ideal for the CAA simulation. However, the planned applications of this coupled setup required several modifications of the existing implementation. Amongst those were the APESolvers extension to 3D problems and modifications to make it account for flows with varying density. The Nektar++ framework gained the ability to perform spatial interpolation as well as a data exchange interface to the employed CFD solver.
Session 2, 17:00 – 17:20
As computational platforms become increasingly heterogeneous, the opportunities for running Nektar++ jobs in different computational environments are growing rapidly. This can help to achieve better performance or take advantage of larger numbers of resources than may be available locally. However, making efficient use of different types of resources can be a major challenge for end-users who need to understand how to configure their Nektar++ jobs to suit the target platform(s) and then to work with the underlying computing platforms to deploy code and run their jobs. Nekkloud is a web-based environment built on components from the libhpc framework that supports domain experts, end users, software engineers and infrastructure providers to collaboratively build efficient job specifications using software parameter templates that can be stored as re-usable and editable job profiles for running Nektar++ jobs. Following a brief introduction to the libhpc model and the use of software parameter templates and job profiles, the Nekkloud environment will be detailed and a short demonstration will be given.
Session 2, 17:20 – 17:40
Session 2, 17:40 – 18:00
Recently the use of linear, non-linear and thermo-elastic analogies as well as mesh optimisation techniques have provided significant steps forward to consistently and robustly generating high-order meshes. However we still fall far short of having a single reliable pipeline for creating high-order meshes which have the features, complexity and quality required. As such the generation of suitable meshes remains a significant bottle-neck in high-order CFD. We have recently created a system for the automatic sizing specification of linear meshes using a curvature targeting approach. This has allowed us to generate linear meshes, which represent the first stage of high-order mesh generation, with sufficient coarseness and quality for high-order use. Furthermore the automation within the system has drastically reduced the time and effort of the users to create suitable linear meshes. Along with the thermo-elastic solver already implemented in Netkar++ we aim to introduce this new system and additional meshing techniques to facilitate a CAD to quality mesh pipeline within Nektar++.