Solving a billion unknowns: how supercomputers are making jet engines more efficient
In theory, jet engines are simple machines. Suck air into a tube, compress it with blades spinning at high speed, add jet fuel, and ignite. The burning, high-pressure gas-fuel mixture – reaching temperatures significantly above the metal melting point – expands rapidly, blasting air out at the other end and powering a propulsion fan that delivers enough thrust to lift huge airplanes into the sky.
Actually making this work, however – and to make it work efficiently – is much more complicated. Jet engines have come a long way since they first flew in the beginning of the 20th century. The bewildering variety on display this week at the Paris Air Show is proof of decades of ingenuity.
So how do you design a better jet engine? You could simply go by experience – build an engine and run it on a test rig. Through trial and error (and constant rebuilding and reengineering) it should be possible to develop a model that’s more powerful, fuel-efficient and quieter at the same time.
Of course, that would take years and be hugely wasteful. That’s why jet engine designers rely on a branch of physics called Computational Fluid Dynamics, or CFD. It helps them predict the flow field and forces inside an engine, all the way from the engine’s fan and blades at the front through to the compressor, combustor, turbine, and mixer and nozzle at the end.
There’s just one problem: classic CFD simplifies and filters out some of the more complicated physics. It is “not always accurate enough to drive full engine optimization,” says Vittorio Michelassi, Chief Engineer for Aerodynamics at General Electric. But all that complexity is important if you want to understand what’s really happening inside a jet engine.
For example, the equations don’t include the full impact of flow unsteadiness and turbulence, which is one of the biggest challenges for engine design. Fluid turbulence is very complex, and very difficult to model.
Turbulent air swirls from the trailing edges of turbine blades in this GE computer model. Image Source: Vittorio Michelassi.
To tackle this and other problems, two years ago General Electric set up the Advanced Aviation Technology Center of Excellence (AAT) in Munich, Germany, which is part of GE Aviation and works closely with the nearby GE Global Research Center.
The AAT team realised that if they wanted to really push engine development forward, they had to ditch traditional CFD modelling that merely approximates the impact of turbulence. Instead, they decided to work on “properly computing the flow” inside jet engine components – that means actually resolving, rather than modelling, the detailed course and momentum of air flowing through a jet engine while it is in operation.
That can’t be done in the live environment of a jet engine test rig. It would be impossible to put enough sensors into an engine to measure what’s happening without disturbing the flow.
But if you can’t do the experiment in real life, and don’t want to use the simplified flow models of traditional computer simulations either, what then?
Then you need a serious amount of computing power, well beyond the means of a normal supercomputer.
“To understand what’s really happening inside the engine may take 10 million hours of computer time, and involve many terabytes of data,” says Michelassi. “The number of unknowns in the simulation is in excess of billions.”
Günter Wilfert, the man in charge of the AAT, says that his team needs to “rule out the inaccuracies of models; if you are able to calculate this detail, you get the real physics [inside the engine], and once you understand that level of detail, you can adjust your design.”
The approach is called “high-fidelity CFD”, and the GE team can get this close to the “real physics” only by bundling the power of several supercomputers around the world – across Europe and the United States – in simulations that involve as many as 50,000 computer cores running in parallel.
“It’s like running a ‘numerical test rig,’ where you reproduce the real behaviour of a real engine, but you do it in a super computer… with an accuracy that you can never measure in a real rig,” says Michelassi.
The AAT team is taking an open innovation approach to its work, collaborating with universities around the world to resolve the issue of turbulence and unsteadiness in the “turbomachinery flows” inside the engines, and what this means to performance and durability.
“We are looking for the smoking gun and find where the inefficiencies are,” says Michelassi, “and then we can fix it.”
It takes a lot of clever work to make it happen, though. “GE is well ahead on using high-fidelity CFD and that helps us understand the flow phenomena which then lead to impact on engine design,” says Wilfert.
He and his rapidly growing team are at the cutting edge of aviation technology, with most of their projects still closely guarded secrets. Suffice to say that GE’s Advanced Aviation Technology Center of Excellence in Munich works on innovating aerodynamics, combustion and thermal systems , and is investigating new ways of better protecting them from any environmental threats – whether it’s icing, hail or rain.
The planes of today look very little like those that graced Paris’ first air show in 1909. Thanks to groundbreaking techniques like these, it’s a safe bet that the evolution of aviation still has many wonders to deliver.