High Performance Computing and Communications' (HPCC) computing power has
given aerodynamicists the tool they long envisioned to use on problems that
previously were too complicated to even contemplate. In a fresh approach to
tackling the root of many design challenges for rotorcraft--from noise/vibration
issues to rotor wake problems--Boeing's engineers in conjunction with HPCC are
throwing their weight behind developing advanced computing tools that lead to
timely, accurate rotorcraft design solutions. "With these tools, we understand
the physics behind these challenges much sooner, leading to a quick cure," says
Leo Dadone, Boeing technical fellow.
Today, the term "rotorcraft" is used for helicopters, tiltrotors, and other
less conventional vertical take off and landing aircraft. This year marks the
end of a three-year cooperative research agreement (CRA) among NASA, industry
and academia that concentrates on the viability of using parallel computing to
solve problems related to one of the biggest challenges in rotor design:
determining the vibratory loads on a rotor hub and the airloads on the rotor
blades. This information is needed to optimize the shape and structural
properties of the rotor blade for efficient performance.
"The CRA has allowed us to make a giant step towards modeling rotor dynamics
by leveraging eight years of Boeing investments that have been targeted towards
the application of parallel and distributed computing to rotorcraft design,"
says Tom Wicks, manager of parallel and distributed systems at Boeing,
participant in the CRA funded by NASA's HPCC Program. Other participants include
International Business Machines (chair), Rensselaer Polytechnic Institute,
Lockheed Missile and Space Research, Centric Engineering Systems, Intelligent
Aerodynamics, Rice University, and Langley and Lewis Research Centers. HPCC has
provided a metacenter for their research, which is currently made up of two
systems, a 160-node IBM SP2 at NASA Ames Research Center and a 48-node SP2 at
NASA Langley Research Center.
The complicated aerodynamic interactions taking place in rotorcraft flight
have pushed the modeling envelope out for rotor blades to address interference
aerodynamics -- including blade/wake proximity effects,
shock-wave/boundary-layer interactions, and wing/fuselage proximity effects. All
of these are necessary to estimate blade loads, rotor performance and acoustics.
"Information gained from interference aerodynamics is necessary to estimate flow
characteristics and associated forces on fuselage, wings, tail surfaces and
other external components subjected to the rotor wake and separated flow
regions," adds Dadone.
After a 12-year investment in Boeing High Performance Computing
tool development, an eight-year focus on rotorcraft applications, and additional
funding from HPCC, Boeing can now model and understand the physics behind some
of these conditions.
This kind of physical modeling is what engineers have always needed for rotor
design and optimization. The CRA focused on the three most computationally
intensive elements of Boeing's rotor optimization process: comprehensive rotor
analysis, rotor computational fluid dynamics (CFD) and structural optimization.
(See Figure 1.) According to Boeing senior principal engineer Joseph Manke, who
has led Boeing's effort on parallelization of rotor analysis and design
applications, "the code modeling these details is computationally intensive. On
parallel machines, considerable improvements have been made."
The workhorse of the rotor optimization process is Boeing's comprehensive
rotor analysis code which models, in detail, the coupled aerodynamics and
mechanical dynamics of a rotor. A rotor blade is a very long flexible wing; so
it flaps up, down and twists and therefore applies forces and moments to the
rotor hub in response to aerodynamic and mechanical forces.
For example, the blade/wake proximity effects are important when a helicopter
is landing. The free-wake model, developed by Continuum Dynamics, Inc. to
analyze this very difficult flight condition, is one of many models in the
comprehensive code. The free-wake models all the wake spirals that are important
to the flow solution. To determine the wake shape, the free-wake model uses
high-order vortex elements and wake relaxation.
Since this vortex method is computationally intensive, Manke and Boeing
senior principal engineer Joel Hirsh implemented a parallel version of the
vortex method. The principal idea behind parallelization is dividing algorithms
into parts which can be calculated separately on different processors. Manke and
Hirsh used the technique of message passing to distribute information among the
processors of the IBM SP2. They benchmarked the parallel code on the IBM SP2.
The results are summarized in Figure 2.; for the largest test case, linear
speedup out to 64 processors was achieved.
Manke and Hirsh were able to take the free-wake model to a new level of
detail and simulate it on the IBM SP2 in a timeframe of practical use to Boeing
engineers. "Before this parallelization, an analysis of rotor performance,
vibration and noise, using an accurate wake calculation, took 80 hours," asserts
Hirsh. "Now on the SP2, it only takes two to four hours." "Helicopter engineers
are now free to exercise numerical wake models with only the best physics in
mind and are not limited by the approximations caused by a lack of
cost-effective computational power," says Byung Oh, Boeing senior principal
Rotor CFD and coupling
Rotor motions and deflections are calculated very accurately by Boeing.
Certain aspects of the flow over the rotor cannot be captured by the models in
the comprehensive rotor analysis code, so these details are computed by a rotor
CFD analysis. The opportunity, according to Oh, "is integrating rotor motions
and deflections with aerodynamic calculations, because in reality the
aerodynamics would not be accurate enough if you didn't account for these
motions and deflections. And by the same token, you wouldn't be able to compute
these motions and deflections if the aerodynamics are not accurate." Rotor
blades on Boeing's CH-47 helicopters and the blades of the 777 Ram Air Turbine
are examples of designs that have used a comprehensive rotor analysis,
complemented with a rotor CFD analysis.
Oh and Hirsh successfully coupled Boeing's comprehensive rotor code and the
rotor CFD code called Full Potential Rotor (FPR), developed by NASA Ames'
Aeroflightdynamics Directorate. They used the coupled codes to predict a 3D flow
effect, called tip-relief, that the rotor code alone could not predict. "The FPR
code coupled with the comprehensive rotor analysis code reached a major
milestone in the last few months when Byung Oh gave it his stamp of approval,"
exclaims Manke. "This advance marks the first coupling of comprehensive rotor
and rotor CFD codes that has proved practical."
Since the FPR code is computationally expensive, Hirsh and Manke implemented
a parallel version and benchmarked the parallel code on the IBM SP2. The
parallel Approximate Factorization method developed by Manke achieved maximum
speedup out to 32 processors. Hirsh used message passing to extend the coupling
methodology to the parallel versions of the comprehensive rotor code and FPR.
According to Oh, "The direct coupling methodologies of rotor CFD and
comprehensive rotor analysis that have been investigated in the NASA CRA will
allow for better engineering support during the design cycle and will eventually
permit structural optimization to minimize blade response to unsteady loads."
Boeing's structural optimization code implements a gradient-based
optimization technique, a method that measures how the performance changes as
each of the design variables is varied. This technique allows the search for
feasible blade structural properties which minimize blade response in the
presence of vibratory airloads. The airloads are computed by the comprehensive
rotor code and can be evaluated for any flight condition within the range of
validity of the aerodynamic models in the rotor code. This means that, if you
have 90 design variables, you have to run the rotor code 90 times in order to
compute the variation of the performance for each of the design variables.
Suddenly the "two to four hour" turnaround resulting from parallelization on the
SP2 takes on new meaning.
Hirsh implemented a parallel version of the structural optimization code in
which the components of the gradient are computed in parallel. He benchmarked
the parallel code on the IBM SP2. The parallel gradient method developed by
Hirsh achieved maximum speedup out to 45 processors for the 90 design variable
Hirsh and Manke envision "a two-level parallel optimization strategy" in
which the parallel versions of the comprehensive rotor code, the rotor CFD code
and the structural optimization code are all used together. The maximum
computing power of the IBM SP2 is essential to making this vision a reality.
Surrogate for FPR
The improved modeling achieved by coupling FPR to the comprehensive rotor
code is important for optimization. However, the approach may not be practical
for rotor design because of the computational cost of the CFD calculation.
Boeing senior principal scientist Paul Frank solved this problem by replacing
the FPR code with a model or "surrogate" based on a technology called Design and
Analysis of Computer Experiments (DACE). In this approach, you construct a
simple model of the relationship between the inputs/outputs of the FPR code by
observing the outputs for a carefully selected set of inputs. Tests indicate
that the DACE model-based surrogate FPR easily satisfies fidelity requirements
for the rotor blade design process. In the context of Boeing's structural
optimization code, the cost of generating the FPR model is quickly recovered
because the surrogate FPR code is 200 times faster than the FPR code.
Some of the results associated with the actual flow of the rotor are best
represented with visualization. Boeing's visualization scientist Dave Kerlick
has developed a way to apply virtual reality to blade modeling, allowing
engineers to modify three-dimensional designs. "Fifteen years ago, graphics
programming required writing code and recompiling it," Kerlick says. "Now modern
visualization environments allow users to explore analysis data interactively,
often without programming. It's a good environment for prototyping."
Wind-Tunnel Testing vs. Computational Predictions
Computer model simulation allows a lot more exploration and decision making
before wind-tunnel testing. "Wind-tunnel testing can cost thousands of dollars
an hour," says Dadone. "Look at, for example, the Bell-Boeing V-22 Osprey. The
process began with the design of the rotor, wing, fuselage and other parts of
the system using the best available tools. Since it was understood that the
analytical tools were not accurate enough, a long series of experiments needed
to be run to quantify and validate all of the elements of the integrated design.
Experimentation began with wind tunnel tests and many different models of
varying scope and complexity, culminated by a flight test program. Eventually
you have to build the full-scale ship and fly it." Now, thanks to the
Boeing/NASA partnership, there is a good way of improving the analytical
predictions needed by the designer, which will eventually help to cut down on
the number of wind-tunnel tests and on the hours of flight testing.
Dadone warns against overblown expectations on how soon this method will
replace wind-tunnel testing: "It's really a successful marriage between computer
tools and wind-tunnel testing which benefits designers and manufacturers with
early design results." Compared to the 'old method' of design, Dadone states
that, "Boeing will be able to provide more accurate predictions in a shorter
time. That translates into better time-to-market, which gives a tremendous
With the help of the CRA and the performance of the IBM SP2 behind them,
Wicks and team are trendsetters in this computational approach to design. "The
CRA has been very beneficial; I'm not just referring to funding and improved
design benefits, but to the benefits associated with the synergy between NASA
and Boeing," adds Wicks. "Partnering with HPCC has provided for the continued
development of the most sophisticated parallel computational tools, which gives
Boeing the option to investigate the use of costly equipment without risk," says
Ken Neves, computer science manager for Boeing.
This synergy has had far-reaching impact. For example, the computing methods
have enabled Boeing to fix vibration or noise problems discovered after the
production of a rotorcraft. In fact, with a good early grasp of detailed
modeling in the design process, the Boeing team knows that they will be able to
correct designs before they are built.
Expectations for the influence on everyday design have to be in sync with
reality though. "If you read the rationale for computational design, it says we
can better address noise and vibration problems relevant to high-performance
rotorcraft. We have to be realistic about what we can expect from the computer,"
says Manke. "It is not a matter of running a code, and bang, we have our perfect
design. It is more like running a code and now we know what we really need to
Welcome to the Aeronautics Learning Laboratory for Science Technology and Research (ALLSTAR). This site is enhanced for the latest technologies on the Internet. more