April 2012 summaries
One of the most powerful techniques to emerge in recent years for modelling solidification microstructures is the phase field method. However, the computationally intensive nature of this type of modelling has to date restricted the application of the technique primarily to two-dimensional problems or, in three dimensions, to pure metals or to alloys solidifying under isothermal conditions (in other words latent heat is ignored). While the isothermal approximation is valid for most conventional casting processes it is not appropriate to rapid solidification conditions, which are the topic of this research.
Prior to the start of this project we had begun the development of a software tool for the simulation of three-dimensional rapid solidification of alloys using phase-field models.
This software is the first to combine adaptive mesh refinement in space and time, fully implicit time stepping and the use of multigrid, building upon our earlier innovations in two space dimensions (see for example: J. Comput. Phys., 225:1271-1287, 2007; Acta Materialia, 56:4559-4569, 2008; Physical Review E, 79: 030601(R), 2009; J. Cryst. Growth, 312:1891-1897, 2010). The move from two to three dimensions, whilst necessary for physical realism and quantitative accuracy, also required us to develop a scalable parallel implementation of the above. Access to High End Computing Terascale Resources (HECToR) allowed us to test the scalability of our initial parallel implementations (undertaken on local High Performance Computing (HPC) facilities at Leeds) to much greater numbers of cores, and to understand and remedy some of the most significant scalability bottlenecks.
The project considered the flow over a backward-facing step (BFS), intended to be representative of the flow over the rear of a “square back” road vehicle bluff body. The primary objective of the project was to investigate whether feedback control could be used to increase the average pressure on the back-face, corresponding to a reduction in aerodynamic drag for a “square back” bluff body. An important feature of the project was that the feedback control strategy should be implementable experimentally, and so practical, body-mounted actuation and sensing were used. Actuation was via a spanwise invariant slot jet (with zero net mass flux) and sensing via the pressure force on the rear face (base) of the step, which would give a direct measure of the pressure drag for a bluff body.
The project found that for a BFS of infinite span with fully turbulent separation, feedback control which targeted a reduction in base pressure force fluctuations successfully yielded an increase in the mean base pressure (in other words a mean drag reduction). This strategy was in fact more successful for fully turbulent separation than for the transitional cases previously studied, and was accompanied by a delay in the roll up of the wake shear layer. The project also began investigations on a more complicated BFS with side-walls. For laminar separation, feedback control which targeted a reduction in base pressure force fluctuations successfully achieved this reduction, but the accompanying increase in mean base pressure force was much less pronounced than for the infinite span BFS. This suggests that for 3-D geometries, it may be necessary to target specific modes of the flow in order to achieve pressure drag reductions.
High performance computing allows us to produce very realistic computer simulations of how animals move. We have been using this approach to investigate how humans walk and run and also to recreate how fossil animals such as dinosaurs could move. The challenge is how to write computer programs that can take advantage of the latest computer innovations and the purpose of this project was to improve our simulation technology so that our software could run effectively on computer with many thousands of processors. Our previous approach was very effective up to 500 processors but could not take advantage of systems with more than 1000 processors. This project allowed us to rewrite our code so that we spread the communication between different processors much more evenly. This prevented the bottlenecks that were previously slowing the code down, and the new version now runs well on systems with over 32,000 processors. This has allowed us to greatly increase the realism of our simulations, and our latest four-legged animal simulations are the most realistic to date. Our dinosaur simulation was able to show that even the very largest estimates of body size (80 tonnes and 40 metres long) are well within the capability of the musculoskeletal system although animals this big were relatively slow moving and certainly could not run. We are now able to address much more complex real-life problems and our next simulations will be able to cope with the difficult challenges of starting, stopping and turning corners.
By harnessing the same process as that which powers the sun, fusion power promises to deliver an effectively limitless supply of energy without producing carbon emissions or long term nuclear waste. To achieve this, two specific forms of hydrogen gas are heated to a temperature that is ten times hotter than the sun’s core and held in place by superconducting magnets. The edge of this gas still reaches temperatures of up to 3000 °C and the challenge for engineers and materials scientists is to develop a vessel capable of containing this process.
Engineers often use computer models to predict how their design will perform under certain scenarios. However, these models tend to be descalised, therefore omitting micro scale features such as defects that may arise during the fracturing process. Understanding the effects of these features is of utmost importance as they might cause unexpected behaviour of the component. An emerging technique shown to have improved accuracy, compared to idealised models, converts three dimensional images of real manufactured components into computer models. These images can be collected by various methods, such as computed tomography (CT) or magnetic resonance imaging (MRI) scanners similar to those found in hospitals. As such, these image-based models predict the behaviour of actual, rather than idealised, components. Because achieving fusion power will require operating under conditions previously not experienced on earth, it was desired to use this image based modelling to investigate a component currently being designed for the world’s biggest scientific collaboration currently under construction, International Thermonuclear Experimental Reactor (ITER).
The difficulty with this method is that the models produced have very high resolutions requiring extremely large amounts of computing power. Additionally, currently available commercial software are not well equipped to perform formulations of this magnitude. This served as motivation contribute towards development of specialised software, ParaFEM, which runs on supercomputers. In order that the scientific community as a whole can benefit from this effort, the code is made open source and is free to download from the project’s webpage. This software enabled successful running of such large models by dividing the calculations into manageable chunks to be solved using thousands of computer processors simultaneously. This work studied a sub component of the ITER diverter, the region of the fusion reactor where heating is at its greatest. Three dimensional images were collected by X-ray imaging and in applying this technique, small voids not anticipated were found. If gone unnoticed these could have caused the component to fail, but this technique allowed recommendations to be made to reduce this risk. Although developed for use with components for fusion, this technique has a broad application to most engineering fields. As well as being used for research and development, it is envisaged that a streamlined or automated deployment of the technique could be included in a manufacturing line to assist with quality assurance control.
The objectives of the project include:
- To conduct case studies for a range of non-reactive and reactive flows for verification and validation of the modifications introduced into OpenFOAM for flame-wall interaction
- To fine tune the models developed based on the case studies
- To apply the validated model to the study of quenching through a single mesh
- To develop an EPSRC proposal
It was planned to validate the modified version of the OpenFOAM code for studying flame-wall interaction for applications to unit Lewis number cases; and on such basis to develop an EPSRC Proposal on flame wall interaction in non-unit Lewis number fuel combustion.
The research was conducted using our in-house version of the modified OpenFoam code, in which the Coherent Flame Model (CFM) in the LES context as proposed by Richard and others  has been implemented for modelling flame deflagration. In order to account for enthalpy loss through the wall which affects the flamelet speed, flamelet annihilation and flame propagation and the decrease in turbulence scales near the wall affects turbulent diffusion and flame strain, the closures proposed by Bruneaux and others  for the CFM –RANS model has been extended to the LES context. In order to capture the enthalpy loss from the flame zone to the wall, we implemented the approach of Angelberger and others  who extended the enthalpy loss factor used by Bruneaux and others  to non-isobaric conditions.
A simple configuration of channel flow without and with combustion has been firstly investigated. The channel dimensions are 4πH in streamwise, 2πH in spanwise and 2H in flow normal direction (H is half channel height). Fully developed turbulent flow simulation were carried out in the channel geometry, starting from an initial laminar flow field imposed with random perturbation. A ‘V’ flame was then superimposed on the statistically steady fully turbulent flow such that the ‘V’ flame ends interact with the channel wall. Subsequently, further combustion model improvements in terms of flow-wall and flame-wall interactions have been coupled with the CFM to capture the salient features of flame behaviour in the near wall region. The modifications have incorporated changes for laminar flame speed, annihilation and propagation through a flame quenching factor which is a function of enthalpy loss, heat release factor & reduced activation energy, and changes in turbulent length scales using van-Driest damping coefficient for filter size. Predictions were conducted for the quenching through a single mesh (Output 1). The flame–wall interaction model proposed by Bruneaux and others  were established using constant density Direct Numerical Simulation (DNS) results in head-on quenching configuration. For side-wall quenching configuration, the model constants are revised based on DNS results for ‘V’-flame anchored in a turbulent channel. Numerical predictions of the turbulent flame propagation in an obstacle laden channel have been conducted using the improved CFM. Several tests cases have been conducted for verification and validation. The predictions show improved match with experiments compared to previous models (Output 2). The research has demonstrated the validity of improvements in the flamelet based combustion models to predict flame behaviour in the near wall region.
As the research was originally motivated by achieving detailed numerical study for flame arrester. The model developed prior to the research and further improved and validated through the research is capable of capturing the quenching behaviour due to enthalpy loss. In order for it to capture the underlying physics in the quenching process especially the change of strain rates near the wall, the model will need to be extended to include non-unit Lewis number effect. One of the original objectives of the research was to develop an EPSRC proposal which will extend the current study to account for flame-wall in the case of non-unit Lewis number. During the course of the research, it was becoming increasingly clear in order to capture the non-unit Lewis number effect by extending the present approach, it will be necessary to have a much more complete set of DNS data conducted with different Lewis numbers. There are published and ongoing DNS research along this direction. However, only a limited number data sets and research outcomes have been published. And the published work has mainly reported on the qualitative effects of Lewis number on the near wall behaviour of the flames. No DNS results are yet available which can be used to guide the extension by allowing data to be extracted to facilitate the modification of the flame surface density type of models. Because of this, we put the development of such an EPSRC proposal on hold. Instead, we have applied and obtained a small research contract from BP to numerically investigate the quenching mechanism and scalability of a specially type of flame arrester that is being assessed by them for applications.
The research was conducted also with our in-house modified version of the OpenFOAM code. The implicit large eddy simulation (ILES) technique was adopted. Thus the turbulence scales resolved by the computational grids are directly solved while numerical dissipation is used to model the unresolved small scale turbulent flow. The Navier-Stokes equations for the conservation of mass, momentum and energy are solved along with a set of reaction and energy equations which model the consumption and production of each chemical element. A modelling approach was developed using the single step Arrhenius chemistry. The research has led to some important findings about the quenching mechanism and influence of various parameters in such type of flame arresters as well the important factors which need to considered in the potential scale-up of the application.
- 1. Richard C, Colin O, Vermoral O, Benkenida A, Angelberger C, Veynante D. Towards large eddy simulation of combustion in spark ignition engines. Proc Comb Inst. 2007; 31: 3059-3066.
- 2. Angelberger C, Poinsot T, Delhaye B. Improving near-wall combustion and wall heat transfer modelling in SI engine computation, Int Fall Fuels & Lub. Meeting & Exposition, SAE special publications, 1997; 972881.
- 3. Bruneaux G, Poinsot T, Ferziger JH. Premixed flame-wall interaction in a turbulent channel flow: budget for the flame surface density and modelling. J of Fluid Mech. 1997:349:191-219
Wind turbines operate in turbulent atmospheric boundary layer. It is of great interest to understand the effects of turbulence on the aerodynamic characteristics. There are two reasons for this. (1) In steady winds (in other words quasi-steady conditions), upstream turbulence may affect transition, separation on the turbine blade. (2) Wind turbines operate in yaw in large time-scale variations of wind direction, and the blades operate in a periodically oscillating condition and dynamic stall occurs frequently. Such conditions may significantly affect wind turbine performance. The generated oscillating forces lead to accumulating fatigue reducing their expected service life.
Firstly, flows around a pitching airfoil were investigated. Secondly, the effect of freestream turbulence on the flow over static and pitching airfoils is studied.
In summary, the capability of Large Eddy Simulation implemented in the open source code OpenFOAM is demonstrated successfully for highly separated flows at deep stall. These significantly enhances the understanding of the physics of wind turbine blades in abnormal wind conditions.
Investigations of turbulent premixed and stratified combusion using direct numerical simulation with detailed chemistry
Investigations of turbulent premixed and stratified combustion using direct numerical simulation with detailed chemistry
- To use DNS to generate a database of three dimensional turbulent stratified flames. The well-established combustion DNS code SENGA2 will be used. The combustion chemistry will be accounted for using a multi-species, multi-step reaction mechanism for methane oxidation. The simulation parameters will be chosen to place the combustion within the thin reaction zones regime.
This primary objective was fully met. The main simulation run, with a domain size of 1024 by 512 by 512 points and a chemical reaction mechanism involving 18 species and 68 steps, is the largest combustion DNS ever undertaken in the UK.
- To analyse the effects of detailed chemistry on the behaviour of turbulent stratified flames. This will be done by direct comparison with previous DNS databases for similar flames computed using simplified chemistry. The performance of existing models for turbulent stratified combustion will be assessed. This work forms part of a larger ongoing collaborative programme of research involving the applicants and others.
Post-processing and model comparison work is ongoing using the DNS data generated.
- To disseminate the findings in the form of high-quality journal papers and conference publications. The DNS database will be made available, as will the updated models identified during the analysis. This will facilitate their implementation into academic and industrial Computational Fluid Dynamics (CFD) codes for the design of future clean and efficient combustion devices.
This objective is ongoing on a longer-term basis, the DNS data base is available on request, and further publications are in preparation.