Meet: David Saunders
Numerical Software Engineer
Ames Research Center, Moffett Field, CA
Who I Am and What I Do
Spending the past 29 years at NASA Ames Research Center with one company
providing software services could mean either (a) I'm in a dreadful rut
or (b) I really enjoy the work. Happily, the answer is (b), and I hope
very much for many more years yet. Maybe I can explain why.
Most of my 29 years have been in support of the Aerodynamics
Division at Ames. Applied aerodynamics basically involves two main activities:
(1) using computers to predict and improve the performance of airfoils,
wings, and (more recently) whole aircraft, and (2) using wind tunnels-and
more computers-to test those predictions on scale models. Often the results
agree, but sometimes they differ, while results from the actual aircraft
may differ again. Narrowing these differences through the use of more
and more sophisticated computational and experimental techniques is one
of the aerodynamicist's goals.
As a software engineer trained in numerical methods, I am
not an aerodynamicist, but I have been able to provide some of the programming
help needed to further this basic goal. More specifically, much of this
support has involved combining two very complex types of calculation into
a form that allows an aerodynamicist to modify or refine a given shape
to have better performance. The two types of calculation linked together
are (1) the ability to compute the airflow about a given aerodynamic shape,
and (2) the ability to optimize that shape by means of a general purpose
minimization method. Optimization refers to automating the calculation
of good values of some problem-dependent parameters, called variables,
in order to reduce some other quantity such as cost, energy consumption,
or-in our case, as one possibility-aerodynamic drag for a given lift (that
is, minimize air resistance for a given aircraft weight). Usually, the
minimization is performed in the presence of practical considerations
termed constraints, although simpler unconstrained optimization methods
been used to good effect.
Please note that while it takes highly knowledgeable specialists
to produce good flow solvers and other highly skilled professionals to
implement good optimization packages, combining the two and applying the
combination as an aerodynamic design-by-optimization tool is much more
within the domain of mere mortals. It might help to compare this with
conducting an orchestra: the conductor probably couldn't come close to
writing what the composer wrote, and he or she cannot play most of the
instruments the way the musicians can. Nevertheless, the conductor manages
to bring the two sets of talent together and produce (usually) happy results.
If you've ever wondered what synergistic effects are, you've just read
about two examples.
A little more specifically, my work has included helping
develop and apply an airfoil design-by-optimization program (two-dimensional
flow), a similar three-dimensional program for treating a rotor blade
or a yawed wing (more on which below), and an even more elaborate wing/body
design code with an option to include under-the-wing engine effects for
supersonic transport applications. Most recently, such methods have been
generalized to treat any complex configuration by decomposing the shape
about the aircraft into numerous blocks, which in turns leads to a natural
way of parallelizing the calculations: different blocks can be treated
by different processor of a multiprocessor computer at the same time,
as long as block boundary information is communicated appropriately.
Another advance as crucial as parallelizing the design process
has emerged in recent years which enables the optimization to be performed
vastly more efficiently than was long believed possible, but this introduction
is already too lengthy, so I will not describe that breakthrough until
later. I hasten to emphasize that this work
is a team effort, completely dependent upon the dramatic progress achieved
by the theoreticians, the implementers of the corresponding software,
and the manufacturers of more and more powerful computer systems. In combination,
these software and hardware tools are now automating design refinement
of more efficient aircraft. Over the life of an airplane, such refinements
can save vast amounts of fuel, so NASA and academia and the airframe manufacturers
should be striving to improve the accuracy and efficiency of these methods
for a long time to come.
Most recently, in support of the Space Technology Division
at Ames, I have applied the same optimization techniques to heat shield
shapes and also to optimizing reentry trajectories for Crew Transfer Vehicles
as part of the effort towards developing a next-generation reusable launch
vehicle to succeed the Space Shuttle.
My Career Journey
It would be nice to believe that a childhood fascination with all things
to do with aviation, particularly jet airplanes, was responsible for my
eventual participation in aircraft design, but I am very much aware that
it was much more luck than good management. First, I had a twin brother
to share that hobby with and to help me much later. As children we each
subscribed to different aviation magazines (I to Flight International,
he to The Aeroplane each week, plus numerous monthlies) and we devoured
them all during what was, in retrospect, the most dramatic of eras as
the jet engine enabled a remarkable proliferation of more and more advanced
aircraft. (The pace of development of new designs has slowed drastically
in the past couple of decades, presumably because the costs and complexities
have mushroomed greatly. All the more reason to bring the power of modern
computers to bear. The fact that the de Havilland Comet, the Boeing 707,
F-104 Starfighter, the Concorde and the Phantom F-4, to name a few, were
designed and built in the absence of even an Apple II computer, and in
remarkably few years, has to be impressive to anyone even vaguely aware
of the construction issues, the flight control problems, the safety concerns,
and so on, that aircraft building involves. But I digress.)
Our interest wasn't really in flying but rather in
the fascinating shapes we saw in all those three-view and cut-away drawings
(the more detailed the better). As we entered our teens, we built more
and more elaborate scale models from those drawings. Our Northrop Flying
Wing, Gloster Javelin, Douglas X-3, and F-104A all measured about 3 feet,
while the crowning achievement in our last years at high school and first
years at university was a 6-ft model of the English Electric Lightning
F-1A, a Mach 2 twin-engine fighter. This one was finished with a layer
of Plastibond which we had to sand and fill, sand and fill, till all the
little bubble holes were eliminated. We managed to mold the cockpit canopy
from something called Perspex, and modeled two Firestreak missiles attached
to the forward fuselage. Maybe the absence of TV had something to do with
our perseverance. We were born in New Zealand and lived there till our
20s, and the first TV we ever saw belonged to a high school classmate's
family around 1960. Today, no doubt, we'd have been glued to the Wings
programs on the Discovery Channel.
We went separate ways following our bachelors' degrees
in mathematics: Michael to the New Zealand Government's Department of
Scientific and Industrial Research, and I to teaching mathematics at a
high school (a reluctant choice, but opportunities were few in those days).
The DSIR soon sponsored Michael to come to Stanford University, where
his doctorate in numerical analysis led to a career in "mathematical programming"
which is better referred to as constrained optimization, although the
terms linear programming and nonlinear programming persist. As I've touched
on above, this field involves elaborate software for minimizing quantities
known as cost functions or objective functions subject to linear or nonlinear
constraints (possibly hundreds or more of them), all with respect to certain
variables (again, possibly hundreds or more).
This connection with my brother and his colleagues
in optimization was to prove a useful one. First, though, I had to extract
myself from teaching, and after five years I was definitely ready for
a change. (Teaching is certainly a character-building endeavor, and dedicated
teachers deserve much more recognition than our modern society tends to
give them. But I urge would-be teachers to choose wisely: those late nights
preparing for tomorrow's classes take their toll, as do the myriad other
teaching duties and the inevitable students with behavioral problems.
Twenty-seven years later, I still have occasional anxiety dreams of being
up there in front of a class, unprepared.) Anyway, a 6-week visit to Michael
in California made sense. Not knowing what the future held, I audited
a few classes at Stanford, sat the GRE exam, and applied to the Computer
Science Department, one of the first such departments in this country.
Eventually I had the good fortune to be available when a Research Assistanceship
came up, and that was the beginning of my second career. Just an MS in
CS (focusing on numerical analysis) has served me well ever since. Interestingly,
the Assistanceship initially involved a literature search on parallel
algorithms for solving tridiagonal systems of equations--in 1972. Truly
usable parallel computer systems have become a reality only in very recent
years, yet here the universities were anticipating their requirements
25 years earlier.
One of the few job interviews I had upon graduating
was with Informatics (now Sterling Software), who supported some of the
research at NASA Ames. This proved a lucky choice, because it led to the
involvement with aircraft design that as teenagers in New Zealand we could
only vaguely dream of. Actually, most of the first five years was spent
helping to analyze solar wind data from the instruments on board the Pioneer
10, Pioneer 11, and Pioneer Venus spacecraft. I was able to contribute
what I had learned about solving least squares optimization problems to
fit mathematical models to the data, which represented particles emanating
from the Sun at about a million miles per hour.
I also had an introduction to wind tunnel data in
those first years. At that time, Ames used a single, central computer
(IBM 360/67 with two processors) to support not only all the wind tunnels
but all other computing at Ames as well! Informatics designed and implemented
a new Standardized Wind Tunnel System to run on dedicated minicomputers
in each tunnel. My involvement was with some
force/moment calculation utilities. Later, I worked on a separate data
acquisition system for oscillating airfoils which evolved into another
system known as the Fluid Mechanics Data Acquisition System. Around that
time I was Task Manager for a group of 10 or so, but our numbers have
dwindled with NASA budgets, and many good people have had to move on.
Today, the budget situation is somewhat healthier, although much of the
applied aerodynamics has been eliminated at Ames
(in favor of NASA Langley)-hence the recent switch mentioned above to
space technology applications of optimization.
A most intriguing project came around 1990: the Oblique
All Wing project, which was the ultimate extrapolation of the idea long
promoted by the famous aerodynamicist, R.T. Jones, who had invented the
swept wing independently of the Germans during WW II. "R.T"
was associated with Ames for many years, and, in his 80s, still came to
many of our OAW meetings with further contributions. Ideally, an
aircraft needs to vary the wing sweep in flight according to its speed.
Pivoting a single wing in the middle rather than sweeping two wings symmetrically
should save weight, at the expense of control complications because of
the asymmetry. A low speed demonstrator had been built and flown, taking
off with the wing pivoted 45 degrees forward and backward, for instance,
and this aircraft can now be seen
near the Visitors Center at Ames. However, for a large transport aircraft,
avoiding the weight of a really massive pivot altogether seemed attractive.
Moreover, as R.T.'s analyses showed long ago, long slender bodies are
the most efficient. Therefore, he urged doing away with the fuselage completely,
dispensing with any massive pivot, putting the passengers inside the wing,
and flying the wing at
a very high sweep during cruise as a long, slender body.
Our assignment, then, was to design wing sections
around a fixed-size central passenger cabin to allow flight as efficient
as possible at Mach 1.6 (~1,000 mph) at 68 degrees of sweep. The cabin
had to be 10 ft high and carry 450 passengers, leading to an enormous
wing span of more than 400 feet! Blending the outer wing panels into the
cabin envelope proved to be a nice application of constrained optimization,
while optimizing the wing sections was done with the rotor
blade/yawed wing code using an unconstrained optimizer. I became involved
enough to be chosen for a NASA Contractor of the Year award in 1994. There
are way too many aspects to cover further here, but an intense team effort
did indeed lead to the design, construction and wind tunnel testing of
a magnificent, 7 ft steel and aluminum model with four under-the-wing
engines. Our calculations and the
experimental measurements both predicted that the concept was less efficient
than the early hopeful claims, partly because of those pesky engines:
their pylons had to be thick enough for the engines to pivot as the speed
(and sweep) changed in flight. The laws of physics are completely rigid,
and outwitting Mother Nature doesn't
come easily. Incidentally, the wind tunnel model for OAW-3 now lies, unlabeled,
in one of the display windows at the Ames Visitor Center.
Maybe, some day, the cost of fuel will be high enough
that the oblique wing idea will be revisited for its potential modest
benefits, but for now, a conventional next-generation supersonic transport
is much more likely to appear first, albeit a long time from now the way
things are going.
Indeed, such a Mach 2.4 High Speed Civil Transport
design project is what I supported for the rest of the 1990s, but details
must be omitted because this outline is already too long, and the HSCT
was abandoned by Boeing at the turn of the century after several promising,
optimized designs had been developed with competing methods. [More recently,
Boeing has also abandoned its Sonic Cruiser
project, and Concorde operation beyond 2003 appears doomed, all for economic
reasons. A supersonic business jet is now most likely to be the first
to follow in Concorde's pioneering footsteps.]
In summary, this aerodynamic design-by-optimization
work has been challenging and rewarding, and I will always feel grateful
for having stumbled into it. I wish everybody could be so fortunate.
Rewarding though this career has been, my greatest fortune has been in
meeting my wonderful wife, Ha (a truly good person who survived a harrowing
escape from Vietnam by overloaded small boat in a storm). Our young son,
named after Uncle Michael, of course, is 7 now and couldn't be more delightful
for his proud parents. Watching that little character develop into such
a happy, loving chatterbox is the
most fascinating and heart-warming experience. I strongly recommend it!
Unfortunately, balancing a demanding job and family
life is very difficult, and too often the family comes second. Regular
exercise at racquetball and tennis aggravates the situation, yet not staying
fit wouldn't be wise either. I've had to give up most of my reading, and
get to play the piano only infrequently these days. Too little time in
the day is my biggest regret, along with being too far away for too long
from our dear mother. I hope little Michael grows up to find as fulfilling
a life as I have, but that it's not 7,000 miles from home. Today, make
a point of hugging your Mom and Dad-while you still can.
For young people with mathematical inclinations wondering about careers,
I would suggest that, while numerical optimization has already been applied
effectively in many fields, there should be scope for a lot more of that
in the future, including the work within NASA. Design of almost anything
these days is done with computers-from planes and cars to refrigerators
(I'm sure) and even computers themselves (e.g., for cooling them efficiently).
Automating the determination of best designs is an opportunity for people
familiar with optimization software to make valuable
Some applications, including the above, involve extremely
expensive calculations (such as for the airflow about an aerodynamic shape).
Normally, costly objective functions tend to limit the number of design
variables that can be used at one time. However, a recent breakthrough
has virtually made this conventional wisdom obsolete: some clever calculus
shows how, at the expense of one more calculation similar to the airflow
calculation, the sensitivities of the function being minimized to any
number of variables-and hence a direction to move in the design space-can
be calculated very cheaply. This is nontrivial stuff, but it helps illustrate
possibilities made possible by recent advances in this very important
field. Much larger problems than previously thought feasible are able
to be tackled now.
Finally, I must take this opportunity to urge young
people not to do one thing I did, and that is take more than 40 years
to discover the symphonies of Gustav Mahler. He wrote 9 remarkable symphonies
(or 10, really, but he was so afraid that 9 was a fateful number-because
of how many other composers died after writing that many-that his real
9th he chose not to call a symphony). Mahler (1860-1911) was the greatest
conductor of his time, and his marvelous orchestrations reflect that.
His music is full of beautiful melodies, frequently two of them in parallel,
and the emotions range from sublime joy to agonizing despair. If you haven't
heard yet what he did with a minor key version of the Frere Jacque tune,
you should start with his Symphony No. 1 (the "Titan")-see the
I would love to say more on this subject (such music
having helped me through many long evenings on the job during our intense
efforts to meet deadlines), but I've imposed on my readers too much already.
Thanks for listening!