uick. How much time did you spend designing
yesterday? If you were simulating a multistage compressor and it took more than
15 hours, you may have to change your approach. "We are running overnight a
simulation of a multistage high-pressure compressor for Pratt &Whitney (P&W), a
United Technologies Corporation subsidiary," states John Adamczyk, senior
aerospace scientist at NASA Lewis Research Center.
Adamczyk also works with GE Aircraft Engines to simulate the full engine's
primary flow path overnight, which includes inlet, compressor, combustion,
turbine, nozzle and bypass duct.
Both companies are working with the NASA Lewis Research Center (LeRC) to
develop the Numerical Propulsion System Simulations (NPSS); a numerical test
cell in which designers plan to reduce the time and cost of developing new
engines by modeling a full engine on the computer. (See figure 1)
The High Performance Computing and Communications (HPCC) Program supports the
development of the computing technologies required to accomplish the NPSS goals.
Pratt & Whitney has already achieved a 50-percent reduction in the development
time of building a high-pressure compressor by performing the three-dimensional
aerodynamic simulations using existing workstations. This amounts to a $17
million reduction in development costs (56 percent lower than previous costs)
while improving compressor efficiency by 2 percent. This new compressor is
expected to result in over one billion dollars in fuel savings during the
approximate 20-year life of the P&W 4000 series engines.
The compressor flow simulation represents one of the key building blocks in
constructing a full-engine simulation with the NPSS system," adds John Lytle,
Chief of Lewis Research Center's Computing and Interdisciplinary Systems Office.
An important element of the cost reduction is the use of workstation clusters
rather than large vector supercomputers to perform complex simulations. To
continue to exploit the advantages of cluster computing, HPCC is working under a
cooperative agreement with a NASA/Industry/University Team led by P&W.
Awarded a cooperative agreement in June 1995, Pratt & Whitney will
demonstrate overnight turnaround for a three-dimensional aerodynamic simulation
of a full, 23-blade-row compressor using these workstation clusters. "That
translates into reducing computing costs to less than 25 percent of a CRAYC90,"
To accomplish this, P&W is leading a team comprised of United Technologies
Research Center, Platform Computing, MacNeil-Schwendler, CFD Research,
Massachusetts Institute of Technology, State University of New York at Buffalo
and the NASA Lewis, Langley and Ames Research Centers to further improve
software to control scheduling and checkpointing of tasks for the networked
workstations. A wide range of industries such as aircraft, oil exploration and
financial will benefit from this software, which eventually will be
Why is overnight turnaround so important? "One of the problems the
propulsion people face today is that the time spent developing an engine is
longer than Boeing or McDonnell Douglas spends developing an entire aircraft,"
explains Adamczyk. "Designers don't want to wait six months for an answer
because they often scurry to make changes. Invariably what happens in the course
of an engine development program is the aircraft weight grows and the engine
needs more performance. So stakes are high, especially when you add the cost;
building a turbine, for example, costs $50 million. Good numbers are needed to
make tough decisions."
According to Lytle, the whole thrust of the NPSS program is using more
computer capability in the design environment. "Today, the majority of time is
invested in building and testing various components and subsystems of the
engine, a costly and time consuming process. With NPSS, designers will not have
to perform as many tests in a physical facility. They can replace some of those
tests with computer simulations early in the design process."
Sounds good. However, GE's senior engineer Mark Turner, who is calculating
the performance of a GE90 computationally, knows things take time to evolve.
"We'll optimize current GE design practices while embracing new NPSS
simulations," Turner says.
For Joseph Osani, GE's manager of Technology Programs, "NPSS accelerates our
development of design and analysis tools. The economic slump in the early 1990's
has forced GE to put more emphasis on designing for lower cost," he says. "If
there was not an NPSS effort, we would perform analysis on single- or
multi-blade rows, but not 3-D analysis on the entire engine. And that means less
Wide variety of analysis
GE's manager of Propulsion System Simulation, Ronald Plybon, also
quotes performance guarantees for customers. "The software communicates with the
test data as well as other simulation codes that we have validated over the
years. We take all that data and predict the performance of the entire engine."
Frank Sagendorph, GE's manager of Product Definition and Analysis Methodologies,
believes that NPSS is one of the best frameworks to model various fidelities of
an engine from first principle basis models to unsteadiness in multistage
turbomachinery. Now, the system analysis will provide the ability to evaluate
which physical processes, occurring on the component and subcomponent levels,
are important to system performance." In that analysis, engineers focus or "zoom
in" on the relevant processes within components or subcomponents. Figure 2
illustrates the modeling latitude of the NPSS system. Individual blocks within
the figure represent code components.
Adamczyk remembers the birth of NPSS in 1985 when "our vision was to build a
numerical wind tunnel with this added zooming capability. A designer does not
use the highest order of fidelity simulation when wanting to know which way the
wind is blowing. It's overkill. If the level of fidelity rises, our Average
Passage (AP) flow model or unsteady model keeps pace." AP models the
three-dimensional flow in multistage turbomachines and represents a module of
the NPSS system.
50 blade rows in 15 hours
"There are 50 blade rows in the GE90 and no mathematical model is
better than AP to simulate that many blade rows," asserts Osani, who wants to
cut the four-year-engine certification program in half. "Turner will make
improvements to the GE90, the engine used for the Boeing 777 aircraft, by using
this flow model," Osani says. "The AP code has developed so that it is now
possible to run with more than six parallel processors per blade row. For
several years, GE has used the code, which has a coarse grain parallelization
strategy where we apply one processor per blade row," adds Turner.
This code operates on the CRAY T3D at NASA Jet Propulsion Laboratory; CRAY
C90 ("vonneumann") at NASA Ames Research Center (ARC); IBM SP2 ("babbage"), an
HPCC testbed at ARC; SGI Power Challenge Array ("davinci"), an HPCC testbed at
ARC; IBM RS6000 ("LACE"), an HPCC testbed at NASA LeRC; and HP clusters at GE.
Along with the ability to improve the speed of modeling complex processes on
computers, engine manufacturers everywhere stand to benefit from HPCC's
attention to system software for various competing platforms. A technologist at
heart, no one feels the challenges more acutely than Adamczyk, who is on the
front line of software development. In that empire of computers, the speed at
which you can model complex processes fosters the NPSS goal of a full engine
Adamczyk accelerates design simulations by improving the efficiency of his
computer code and by taking advantage of the multiple processors of parallel
"Right now on HPCC's SP2, we can turn around a simulation of about 30 blade
rows in about 80 hours using one processor per blade row. We've gotten speed ups
on Silicon Graphics workstations of over three, with four processors per blade
row. So you take 80 hours, divide it by three and that gives you about 25
hours," Adamczyk adds. "We are hoping to get similar speed ups with the SP2 on
the order of 80 percent. We average a 20-hour turnaround with the SP2's 176
processors, distributed six per blade row. Now, we are also working on speeding
up the code itself, which is currently running on the order of 32 megaflops on a
high-speed processor of the SP2. I know we'll turn around a simulation of 50
blade rows in 15 hours. We've come this far."
Full 3-D engine simulations on present hardware are now achievable. "If I
didn't think I could do it, I wouldn't have said I could," says Turner. "Not
only is it possible, it's fun." Adamczyk shares his glee: "This is not a stunt.
We actually have a tool that bends metal."
Problems real to flying
A mechanical and aeronautical engineer, Turner pays particular
interest to applying NPSS to problems real to flying, a personal interest of his
own. He has the daunting task of studying the building blocks of an engine at
various points of operation such as cruise, near idle, takeoff or landing
With Adamczyk's code and Mississippi State's unsteady MSTURBO code, both of
which run on HPCCP testbeds, Turner later plans to understand some of the
intricate details of an engine's performance including off-design unsteadiness
and fan flutter. New Average Passage code developments will allow a coupled
simulation of the fan and booster. "Once that's complete, we'll unite the fan
with the other turbomachinery," says Turner.
Drawing on his long love affair with engines, Turner gives directions for
everything from engine core to turbine machinery. "In parallel to modeling the
turbine machinery (fan and booster, low-pressure turbine and the pylon), we'll
model the engine core (combustor, high-pressure compressor, high-pressure
The combustor, Lytle explains, will not be modeled today in the same level of
detail due to the complexity of the combustion process. "In the next couple of
years, however, we will model the combustor at a greater level of detail and
complete the whole engine simulation overnight." Working with a consortium of
companies such as P&W, Allison Engine Company, CFD Research and GE, the HPCC
Program is a partner in developing the National Combustor Code.
"There's not much loss that occurs in this engine," a tribute to GE's
design, but agonizing to an engineer wrestling with 3-D code. "If the engine's
loss is inaccurately modeled, it greatly impacts your predictions," says Turner.
The journey from the birth of NPSS to actual 3-D simulation reaches a major
milestone in the next few years when Turner and others like him will see what
they have won. There's still some smoke that needs clearing, such as the correct
modeling of the physics. Standing in front of the GE90 fan, Turner describes the
air that then goes into the booster, the core, then into the low-pressure
turbine. Along with the core, each of those components are modeled separately
and "then I'll pull the whole engine together and see the payback."
And it could be momentous. Along with Pratt & Whitney's significant
development savings, GE anticipates cutting the $2 billion engine certification
cost in half. "We'll compress the whole cycle from developing the code that
quickens the design process to transferring that information to manufacturing,"
Osani says. Already, Plybon has seen significant improvement in the stage, a
combination of rotating and stationary blade rows. "We used GE's 3-D aero code
to design the stage, the upstream nozzle and the rotor. NPSS with codes
including APNASA, the Mississippi State TURBO code and ARC ROTOR and STAGE codes
are helping us to understand the flow field in an unsteady environment."
With or without experiments
Tall, poised, and enthusiastic, Turner urges designers and other engineers
toward a vision of new engine design. His eye is fixed on working with these
numerical simulations while incorporating the valuable test results that he gets
from David Wisler, manager of GE's Aerodynamic Research Laboratory. Testing
provides important insight into how to model real physical phenomena. "GE
laboratory's turbine and compressor vehicles have become a very good way of
validating our codes," Osani says. "The reason why it has taken so long to
reduce testing with computation is you have to be very confident in the
As Plybon explains, "In times of experimenting and investigating a new design
frontier, our work is not to beat vainly against wind-tunnel testing, nor to
wonder how we got into a problem, and who is to blame; rather it is to team the
learning gained from both traditional experiments and NPSS."
Sagendorph is instinctively pushing new design models forward without relying
on the status quo. "In an ordinary test facility, you may only have one shot at
the air flow you're trying to design. If that doesn't work, the cost is usually
prohibitive to reblade the compressor. Now, we can reblade that airfoil with
NPSS at less cost." Adamczyk agrees: "I don't think anybody is proposing that
you eliminate experiments, but reducing the number of tests, the amount of
hardware, and wind tunnels would be a tremendous savings. Wow! But look at the
limit. What if you could create a numerical wind tunnel?"
"The brightest light behind this design mechanism is that we
intuitively knew that if you could simulate a whole engine computationally, the
impact on industry would be tremendous," explains Lytle.
"Six or seven years ago, we started talking to people about NPSS and they
thought we were nuts."
Now the impact has blossomed. In addition to the commercializing of the
system software, " the technologies developed through the HPCC Program have made
the full-engine calculation possible," according to Adamczyk.
Welcome to the Aeronautics Learning Laboratory for Science Technology and Research (ALLSTAR). This site is enhanced for the latest technologies on the Internet. more