Features:

A QUIETER BLUE YONDER

Anyone who has lived under the flight pattern of a major metropolitan or regional airport can attest that constant exposure to jet engine noise is a serious problem. Numerous studies have shown that the relentless roar from a busy runway poses all sorts of threats to the health and quality of life of its nearby neighbors, from sleep disruption to low birth weight. The reverberations from jet engine noise can become so intense that they have even been known to cause metal fatigue in aircraft components.

Aircraft manufacturers have made tremendous strides in making jet engines quieter in the past several years by focusing on the shape of the nozzle, the outlet at the back of the jet engine where compressed air mixed with jet fuel is released to propel the airplane forward. However, these improvements are achieved through a time-consuming, expensive trial-and-error process that involves comparing the way different kinds of nozzles alter the flow of air through the engine, resulting in different noise levels.

Moreover, while the experimental apparatus may achieve the desired results, it can't explain them. At NASA Glenn Research Center in Cleveland, Ohio, "researchers conducting the mechanical experiments have a table covered with different attachments to put on nozzles," explains Jonathan Freund, assistant professor in theoretical and applied mechanics at the University of Illinois. "They try one after another, and if one works, they're happy. They don't know why; they don't know if it's the best they can do."

However, what currently can't be easily determined directly from mechanical trial-and-error experiments very well could be simulated, and that's precisely what Freund and his graduate student, Mingjun Wei, are doing. Using a code that Wei has written and that forms the basis of his dissertation, Freund and Wei are using NCSA's Origin2000 and Platinum Linux cluster to clarify the nature of the mechanism of aerodynamic noise by working backwards. Work is underway to run the problem on the Titan cluster and the new IBM p690 system called Copper.

FIRST PRINCIPLES

Sound is a traveling pressure disturbance in a fluid, which can be envisioned as an invisible ripple in a pond. Closer to the center of the disturbance, the waves are shorter and their vibrations are more powerful. Farther from the disturbance, the waves grow longer and the vibrations fainter. However, noise is a specific kind of sound, produced by friction between layers of air that results in irregular vibrations that are extremely irritating to listeners.

The air leaving a jet engine is highly unstable. Its unsteadiness creates the irregular vibrations that we recognize as noise, just as the unsteady motions of a loud speaker make noise, only a good deal louder. The unsteady turbulent flow exhausting from a jet engine is very complex, so the underlying mechanisms of jet noise are unclear. A simplifying principle has not been identified. "There are some models that explain the phenomenon," Freund says, "but not to the fidelity that's needed to actually do something with them. So we're circumventing this lack of clear understanding with the method we've developed."

Thus, the noise produced by a jet engine can be described by the same set of equations that describe how the aircraft itself can fly.

Beginning at the nozzle, where the noise is emitted with the jet exhaust, Freund and Wei compute the sound intensity at a distance from first principles--the most fundamental equations for a given phenomenon. In this case, first principles mean the Navier-Stokes equations, which describe the motion of a compressible fluid, in this case, air. The Navier-Stokes equations relate the rates of change of fluid density, momentum, and energy to pressure differences and viscous forces in the fluid (the last of which are created by the extreme friction generated when fluid layers are perturbed).

PUTTING IT IN REVERSE

Freund and Wei start with a simulation of a two-dimensional mixing layer, a thin turbulent region where two streams of fluid mix and create a layer of instability. This is phenomenologically similar to flow in the region just downstream of a nozzle. They numerically solve the Navier-Stokes equations for this near-nozzle region without modeling a physical approximation of the field. This approach is called direct numerical simulation (DNS). They then solve the adjoint of the Navier-Stokes equations in reverse. The adjoint is an equation that superimposes the perturbations caused by changes in pressure on their numerical model of the near-nozzle region. Because Freund and Wei know both the initial conditions (the airflow past the nozzle) and the end result (sound intensity), they are able to identify the remaining piece of the puzzle: how sensitive the noise is to even the slightest change in nozzle conditions.

This last piece of information is especially crucial. "After solving the equations in reverse," says Wei, "we know how to change the control [conditions] at the nozzle to reduce the noise further." Simulating these processes numerically allows Freund and Wei to work with enormous numbers of individual data points. "Because each control point in space and time is optimized individually, we are actually managing three million space/time control parameters." The runtime for a single simulation on Platinum is equally staggering-22,000 CPU hours. However, Wei has put a great deal of effort into parallelizing his code and estimates that on Copper, it will run five times as fast.

"We have something unique right now," says Freund. "We have simultaneously a flow that makes a lot of noise, and a flow that's been perturbed slightly and is a lot quieter, and so we can go on from there and try to figure out what's changed. Hopefully we will be able to generalize that to more practical problems and learn from it."

Freund and Wei emphasize that the simulations they're performing are intended to complement aerodynamical experiments going on at Glenn, Boeing, and other federal and commercial research facilities. While these simulations can't singlehandedly produce a quieter engine design, they can go a long ways toward helping aerospace engineers who are directly involved in mechanical experiments understand the mechanism of aerodynamic noise caused by turbulence. "We're hoping to guide the experimental process eventually through a better understanding of how noise is generated," says Freund.

This research is supported by the U. S. Air Force Office of Scientific Research.