Skip to content

he path that air takes as it flows past an airplane wing gets the plane off the ground and also determines how efficiently it flies. Computer simulations of wing aerodynamics help engineers improve their designs and understand the limits of those designs. For example, University of Maryland Ph.D. student Ricardo Diaz is working on simulations of an experimental aircraft that uses a small front wing, called a canard. The canard provides advantages for the mission envisioned for this vehicle, but the air streaming behind the canard also affects how air flows over the main wing. With his simulations, Diaz can understand this interaction more precisely. "The problem is that, at a certain angle of attack, the wake from the canard impinges on the wings," Diaz said. "The question is, 'What angle?' We know that it's between 12 and 16 degrees based on preliminary results, and that's probably acceptable from a design point of view since planes normally stall at around 14 degrees." Diaz is working with his advisor, Jewel Barlow, professor of Aerospace Engineering at the University of Maryland and director of the Glenn L. Martin Wind Tunnel, to improve a simulation code so that it can determine this angle more accurately.

FROM VORTEX LINES TO 3-D WINGS NPACI PARALLEL COMPUTING INSTITUTE COMPUTATIONAL WIND TUNNEL

FROM VORTEX LINES TO 3-D WINGS Such air flow simulations all fall under the general category of computational fluid dynamics (CFD) and computational aerodynamics. Traditional simulation methods, based on lifting-line theory, model the airplane wing as a straight vortex line and the wake as straight lines trailing behind the wing like streamers. At the next level of detail, called lifting-surface theory, the model includes the shape of the wings and a more complex wake; however, the shape of the wake must be specified in advance. Today's workstations permit this second level of simulation, with a plane model that includes 7,000 elements, or panels (Figure 1). Going one level further requires a supercomputer. Diaz is modifying 3-D fluid modeling software developed originally by NASA, called PMARC, to take advantage of parallel computers. He is parallelizing the code using NPACI's CRAY T3E at the University of Texas and testing it on a model with 10,000 panels. This is better resolution than is practical on workstations, but they still must specify the shape of the wake, according to Diaz. The problem is the canard. The behavior of air over a single wing is fairly well understood, but the canard adds a new twist. In this case, the placement of the canard on the aircraft, a remotely piloted vehicle (RPV), was determined by other design factors. Therefore, to specify the flight limitations for the RPV, the designers needed to know the angle of attack at which the wake from the canard impinges the wing (Figure 2). "The sudden change in lift could potentially cause the RPV to become unstable in flight," Diaz said. "Unfortunately, we don't know the shape of the wake a priori. We have to make educated guesses." Diaz's ultimate goal is to perform time-accurate simulations on high-resolution models. A time-accurate simulation will allow the wake to develop naturally and let them know for certain how the canard wake interacts with the wing. Workstations do not have the computing power to perform so-called time-accurate simulations. In fact, workstation simulations will fail in time-accurate simulations because the resolution is not great enough to permit the wake to pass smoothly over the wing. Even on the Texas T3E, a model with 10,000 panels does not have enough resolution to allow the wake to develop naturally. To achieve this level of accuracy, Diaz plans to use the 256-processor CRAY T3E at SDSC, once he has finished parallelizing and testing the modified PMARC code, to simulate a plane model with 50,000 panels and a wake that develops naturally over time.

Top | Contents | Next

NPACI PARALLEL COMPUTING INSTITUTE To help him modify the computational kernel of PMARC, Diaz attended the 1998 NPACI Parallel Computing Institute, held August 17-21, 1998, at SDSC. The institute is designed to bring together computational scientists and engineers, HPC experts, and computer scientists. The institute features lectures by NPACI speakers on performance characteristics of HPC resources, programming methodologies and tools, parallel algorithm design, code optimization, and debugging techniques. Attendees are also given lab time to experiment with the techniques and materials presented and work on their projects. Lab time also spurs informal discussion among participants on ways to advance computational methods in their fields. This year's list of speakers included members of NPACI's Consulting staff, Larry Carter of UC San Diego, David Culler of UC Berkeley, Geoffrey Fox of Syracuse University, Andrew Grimshaw of the University of Virginia, Carl Kesselman of the USC Information Sciences Institute, Chuck Koelbel from Rice University, Paul Messina from Caltech, Cherri Pancake of Oregon State University, Burton Smith of Tera Computer, and Robert van de Geijn from Texas. Diaz's goal for the 1998 institute was to learn how best to parallelize the computational kernel of both a Navier-Stokes code, called VLES, and PMARC using High-Performance Fortran (HPF) and possibly the Message Passing Interface (MPI). "The institute gave me a very strong foundation for parallelizing the code," said Diaz, who also attended the 1997 institute to work on another CFD project. "It is very rare for most students to have such direct access to a group with this level of expertise in the field of supercomputing. One of the most beneficial aspects of the institute for me was that I was put in contact with people who share my research interests." For example, Diaz worked with Wai Sun Don of Brown University, who is writing similar code using MPI.

Top | Contents | Next

Figure 2: Canard Streamlines Top: Streamline pattern over the wing showing canard wake impingement. Bottom: Side view of the streamline pattern over the wing showing canard wake impingement.

COMPUTATIONAL WIND TUNNEL Over the course of the week, Diaz added HPF directives to the kernel code, which uses the alternating direction implicit (ADI) method, an efficient method of solving the time-averaged Navier-Stokes equations, the primary equations of CFD. With the modified code, Diaz solved a simple Poisson equation on two processors of the IBM SP at SDSC. (The Poisson equation is applicable to a wide range of studies, from electrostatics to ocean modeling.) "So far, the results with HPF haven't scaled as well as expected on the T3E," Diaz said. He plans to perform additional comparisons to see how the code scales on the SP as well as testing an MPI version. This code will be incorporated into a flow solver and used to evaluate and compare various turbulence models forming the basis of his dissertation. Once parallelized, the general-purpose PMARC code could be used for many other simulations that would be of interest to users of the Glenn L. Martin Wind Tunnel. Such simulations could model, for example, a full scale aircraft, an automobile, a wind turbine, or any nearly inviscid flow--fluid flows under conditions where the effects of viscosity are negligible. The code promises to be a significant asset to the Maryland wind tunnel. "This code is part of the facility's move toward increasing the number of simulations we perform," Barlow said. "As the performance of high-end systems climbs, we can conduct more detailed simulations and advance the state-of-the-art in wind tunnel testing methods with integrated computing capability and on-line comparisons, analyses, and computations." --DH *

Top | Contents | Next

Top | Contents | Next