Computer simulates turbulence

Air flows in the form of turbulence have a major impact on vehicle fuel consumption. Philipp Schlatter wants to take studies of turbulence to a new level. A virtual wind tunnel will show how an aircraft wing can be designed to reduce drag.

Philipp Schlatter

Associate Professor of Fluid Mechanics

Wallenberg Academy Fellow 2013

KTH Royal Institute of Technology

Research field:

“You can never get this amount of detail from a wind tunnel experiment,” Philipp says, staring at the computer screen.

A video clip shows a forest of vortices sticking up from a surface, like straggles of threads from a densely woven carpet. The vortices are a simulation of turbulent air flows – representing the wind as it strikes the wing of an aircraft. Each vortex is a notional air flow, colored according to the speed it is moving.

On the computer Philipp can turn the surface, examine details, and understand more about the movements of the air flows. For although turbulence is a classic area of physics that is calculated using well-known equations, we still do not really know why it occurs and how its behavior can be predicted. And this knowledge is important, not least in industry.

“Much of the fuel consumed by cars and aircraft is used to counter drag caused by turbulent streaming. It is important to understand turbulence so that vehicles can be designed in the best way,” Philipp says.

Moving the wind tunnel to the computer

One way of studying flows around an object is to use a wind tunnel. A fan sets the air in motion around the test object, e.g. a model car or part of an aircraft, which has been fitted with sensors to measure the flow. But building a model, measuring, adjusting the model and testing again is a time-consuming process.

Over the past 20 years computers have therefore begun to be used to simulate turbulence. The computations involved are highly demanding, and so far it has only been possible to study turbulence at fairly low speeds or over small vehicle parts.

As a Wallenberg Academy Fellow, Philipp wants to take computer simulations of turbulence a step further. He will be simulating the flows around the entire wing of an aircraft in a virtual wind tunnel.

“The grant enables me to focus on an individual project for a number of years, and employ people who can work on that specific project. Besides, it is highly rewarding to have a network via which I can interact with other researchers in the same situation.”

“We want to offer an alternative to wind tunnel experiments, to show they can be performed just as well virtually. Better perhaps, since the sensors are a limitation, and can disturb the flow,” he explains.

A tough task even for a supercomputer

The simulations mean that the surface of a virtual model of the object is divided into a mesh made up of points. The computer works out the flow at those points. To simulate the wing of an aircraft Philipp reckons that about 100 billion points are needed. But the number of computations required is much greater than that.

This is because the laws of physics governing flow result in complicated equations that cannot usually be solved precisely. The researchers are therefore obliged to use numerical methods, in which systematic sub-calculations give a solution that is close enough to the exact answer. This takes a while, even for the supercomputers used by the researchers.

“Simulation of the wing will take three to four months – it usually does with things that are this close to limits of computing capacity. We are now using around 100,000 processors, but we want to use 500,000 or a million processors simultaneously,” Philipp explains.

A standard laptop for home use usually has two to four processor cores. When the simulation is complete, it remains to collate and analyze the huge quantities of data generated, amounting to several terabytes.

Essential to calculate quickly and correctly

To attain his goal, Philipp must optimize each step in the simulation process. This includes the computations themselves, e.g. by only using exactly as many calculation points as are needed to achieve the necessary degree of accuracy, and distribute them effectively over the surface. It is also important to ensure that the computations can be run as efficiently as possible on multiple processors simultaneously.

“Essentially, the simulation should be performed as quickly as possible. But at the same time we want as much detail as possible. We don’t want to use the popular turbulence models currently in use, in which the vortices are simplified using mathematical methods,” he stresses.

Comparing with wind tunnel experiments

During the project the researchers will be moving from relatively simple simulations to more complex ones. The computer simulations will be compared with physical experiments in a wind tunnel to ensure the results are the same.

The wind tunnel experiments are being organized by a colleague at KTH Royal Institute of Technology. Philipp himself is sticking to the theoretical aspects, which was what attracted him to fluid mechanics in the first place when he studied mechanical engineering in Switzerland. His fascination for aircraft also played a part.

“I like aircraft – I almost trained to be an air force pilot when I was younger. And I’m interested in computers. Working on numerical flow simulations is a good combination,” he says.

Philipp attended KTH to work on his undergraduate thesis. He then gained a PhD in Zurich and returned to KTH as a postdoc. He quickly established a rapport with his fellow researchers, and points out that ample computing capacity is readily available in Sweden, something that is essential for his research. But there is one thing he misses.

“Having mountains around me. But Switzerland is not far away and I have nothing against flying. I always choose a window seat, so I can look out and try to find flow patterns on the wings,” he says.

Text: Sara Nilsson
Translation: Maxwell Arding
Photo: Magnus Bergström