Increasingly stringent requirements for energy efficiency and noise reduction in jet engines are tightening the screw in their development and design. Cenaero, an applied research centre for modeling and numerical simulation, is developing new methods that will help in designing high-performance jet engines. Cenaero was granted access to PRACE resources in PRACE Preparatory Access for an industrial pilot project “noFUDGE”.
For a long time one of the main problems in applied fluid mechanics – especially in industrial flows – has been the estimation of the impact of turbulence.
“Up to now aircraft and engine manufacturers have been using mostly time-averaged computational methods, in combination with lower accuracy discretization techniques, for their design purposes. These algorithms will continue to be used, at least for a while, but they need to be complemented with more advanced computational tools. First of all, these methods are simply not applicable for some types of flows. An important application is the prediction of noise, for example generated by a flow from propellers or flow instabilities in compressors and fans. By applying larger computing resources we can directly simulate part of the turbulence.
For low speeds or small geometries the size of the turbulence is such that we cannot even get an accurate value of the time-averaged forces and performance if we do not compute part of the turbulent structures directly. These conditions are prevalent for instance on part of wind turbine blades and in some parts of a jet engine.
Koen Hillewaert, Argo Team Leader at the Cenaero research centre in Belgium.
Industrial applications in mind
Cenaero’s aim is to provide a numerical algorithm based on the Discontinuous Galerkin Method (DGM), to be used for computations with full (DNS or Direct Numerical Simulation) and partial (LES or Large Eddy Simulation) resolution of turbulence in wall-bounded flows, especially in turbomachines.
We are convinced that DGM combines the high order of accuracy that is currently offered by academic codes to the geometric flexibility characteristic of industrial codes, and at the same time provides computational efficiency and parallellisability. This high order of accuracy is indispensable to represent turbulent structures adequately, but is not provided by state-of-the art industrial codes. At Cenaero, we are primarily interested in the industrial application. The academic aspect – although also very important as some of the numerical technology still needs to be developed – is a secondary goal.
We will be able to use our method in the context where the turbulence occurs very close to walls, where there is more complexity to take into account – you simply have larger demands for resolution there. The turbulence is influenced by the walls in the flows that interest us, because we focus on turbo machinery, jet engines and propellers.
Above: DNS computations of the transitional flow around the Eppler E387 airfoil during the DECI project “Cobauld” – computed isocontours of vorticity on the spanwise periodic plane and skin friction on the airfoil surface. © Cenaero
Huge computational resources
For the turbulence computations performed by the research team at Cenaero, huge computational resources are needed. Hillewaert says:
That is why we turned to the world-class supercomputers of DECI (Distributed European Computing Initiative). With these resources we could provide a proof-of-concept of the method focused on full resolution of the turbulence, by performing the direct simulation of the transitional fl ow around a low-speed airfoil. DECI also gave us access to machines that have different architecture from what we usually have at computational centres. It was a unique opportunity to test if the code would work satisfactorily in this kind of architecture. Furthermore we were attracted by the support from specialists in porting and optimizing the code.
To show that the method is viable, we compared it to experimental results and DNS computations with a conventional low-order discretization. It was clear that the computations really correspond to what was measured, and that the method really offers a huge advantage with respect to the standard discretisation method, as it allowed us to capture the real physics rather than a result that only looks acceptable. DGM moreover offers us a visual means to assess mesh resolution.
The promising results of this project lay the ground for further development using the supercomputing resources of PRACE. Cenaero was granted 2.000.000 core-hours on the JUGENE supercomputer in PRACE Preparatory Access call. A first project “noFUDGE” has just been finished, and concerns the transitional flow on a section of low-pressure jet engine turbine blade.
“We are moving towards the resolution of more complex flows and flow conditions that are rather ambitious, leveraging both on the availability of ever more powerful computational resources and the further development of the numerical algorithms.”
Above: DNS of the transitional flow around a LP turbine blade at Re=88000 computed during the PRACE project “noFUDGE” – vorticity isocontours on the spanwise periodic plane and skin friction on the blade surface. © Cenaero
Convincing the industrial partners
The results achieved in these computations are part of a roadmap towards industrial use of DNS and LES.
We are showing that with the discontinuous Galerkin method well-resolved computations are feasible, and will give more accurate results, not only for the prediction of noise generation or flow instabilities, but for some cases also in terms of global time-average performance.
The computations have helped us to convince our industrial partners that we will be able to provide the technology for the computations they would like to do within a few years. These cover essentially broadband and tonal noise generation by fans and open rotors, as well as flow instabilities in turbo reactors, such as rotating stall.
These types of computation will be needed if jet engine industry is to meet the more stringent requirements and regulations in terms of energy efficiency and noise. To keep their competitive edge, manufacturers need to invest in new computational technology now.
“If manufacturers do not provide high efficiency and low noise engines, they will be unable to sell them in the future,” Hillewaert points out.
“Clearly the evolution of computation techniques goes hand in hand with the availability of computers. On the one hand as more and more powerful computers are available, we can do more complex computations. On the other hand computational technology has to evolve as more direct resolution of the turbulence becomes feasible and large scale resources need to be used efficiently.”
© Pirjo Rötkönen@PRACE