Title  : NSF 94-12 - 2nd National Conference on Diversity in the Scientific and
         Technological Workforce (Part 3)
Type   : Report
NSF Org: EHR
Date   : June 8, 1994
File   : nsf9412c







                              NOTE:
REFER TO PRINTED COPY FOR FULL EQUATIONS, TABLES AND FIGURES
THIS ELECTRONIC VERSION DOES NOT INCLUDE EQUATIONS, TABLES, FIGURES
AND PHOTOGRAPHS!!



STUDENT RESEARCH COMPETITION
AWARD WINNING PAPERS


PRECOLLEGE AWARD

Stephanie Danielle Kindle, Junior
Gateway High School, St. Louis, Missouri

Comprehensive Regional Centers for Minorities Project
Principal Investigator, Harvest Collier

A STUDY OF PI USING BUFFON'S EXPERIMENT

Abstract: The purpose of this experiment is to recreate a famous
needle problem done by the mathematician Comte de Buffon, using a
different formula to determine Pi. It was hypothesized that the
approximation resulting from this experiment will be close to
3.14159 . . ., the most common approximation of Pi. The experiment
requires a needle and a horizontal surface ruled by a grid of
parallel equidistant lines. The distance "h" between the lines and
the length "1" of the needle must satisfy the relation "1less
thanh." The experiment is done by tossing the needle above the
surface from a height of 50-60 cm. After each toss, a record is
made of whether the needle does or does not intersect one of the
parallel lines. The frequency of such intersections is also
calculated and recorded. After experimentation, the results show
that the approximation of Pi, which was 3.14925 . . ., is near the
commonly-used approximation of Pi.

STATEMENT OF THE PROBLEM

The purpose of this study is to give a brief history and show an
experimental method of approximating the irrational number Pi using
Georges Louis Lechere, Comte de Buffon's experiment.

On a table are drawn parallel lines of "d" intervals. A needle of
length "1" smaller than "d" is thrown at random on the table. What
is the probability that the needle will touch one of the parallels?
This remarkable problem comes from G.L.L. Buffon, who was the first
man to clothe probability problems in geometric form.

The probability of an event is commonly understood to mean the
ratio of the number of cases favoring an event to the total number
of possible cases.

INTRODUCTION

Circles are important geometrical figures in mathematics. There are
examples of them in everyday life such as wheels on cars, bicycles,
the circular disk of record albums, the laser disks which make up
CD-ROMS for computers and musical records. Finding the area of a
circle involves a number that was calculated by the Egyptians in
approximately 3000 B.C., thousands of years before the
Babylonians, the Greeks and the Chinese; the number is Pi(pi).

There are numbers in mathematics that have the ability to spring
out all of the time in the most unexpected places. Perhaps the most
remarkable of them is pi equals 3.14159265 . . ., which is "the
ratio of the length of a circle's circumference to its diameter."
It is a surprise to find it (Pi) in the expression of the
proportion of co-primes among all pairs of integers (equal to
6/pi), or to find it as the probability that a randomly tossed
needle will intersect a line in a grid of parallel lines.

The history of pi was greatly influenced by the ancient problem of
"squaring the circle" which is using a compass and a ruler to
construct a square equal in area to a given circle. The efforts to
solve this problem have resulted in an even greater accuracy in
determining the number pi. For almost 2,000 years the main tool
used to arrive at a value for pi was a method invented by
Archimedes.

In his famous treatise Measurement of the Circle, Archimedes (ca.
287-212 B.C.), the greatest mathematicians of antiquity, gave the
following approximation for the value of pi:

3 1071 less thanpiless than3 1070

or in decimals: 3.14084507 less than pi less than 3.14285714. (The
correct value of pi to eight decimals is 3.141592265.)

He created the classic and maybe the most natural method for
computing pi: because it is the length for a circle with a diameter
of 1 unit, this number can be estimated by the perimeters Pn and Qn
for regular n-sided polygons inscribed in and circumscribed about
this circumference. As the number of n sides increases, the
polygons approach a circle, and their perimeters slowly approach
pi. To obtain his approximation, that remained unsurpassed for many
centuries, Archimedes had to compute the perimeters of inscribed
and circumscribed 96-gons.

Many years later, in 1654, an unexpected result was discovered by
the great Dutch scientists and mathematician, Christian Huygens. He
found that to get the precision Archimedes had attained in
computing pi, regular 12-gons would be enough. This result was
published in his treatise On the Discovered Size of the Circle.
Huygens achieved his results only by improving the technique for
calculating the perimeters of regular polygons. Huygens work was
based on new ideas, which were further developed in present time.

To tackle the problem of estimating the number pi correctly, one
must start with a definition of pi. The most common definition of
pi is the "ratio of the circumference of a circle to its diameter."
This definition requires a justification which is, a proof that
this ratio is the same for all circles. One question that may be
asked because of that statement is: What's the length of a
circumference? The question has a standard answer: it is the value
approached by the sequence of perimeters Pn of regular n-sided
polygons inscribed in the circle as n approaches infinity; meaning,
the limit of this sequence. Particularly, for a circle of diameter
1 unit, this limit equals pi.

Is there an experimental way to determine the value of pi. Yes,
take a thread, measure the circumference of a circle with a known
diameter, and divide that amount by the diameter, there is also a
completely different way. We can get an approximate value of pi by
using a needle . . . and a little of probability theory. This
approach was invented by the French naturalist G.L.L. Buffon
(1701-1788). In later years Buffon's original experiment was
repeated time and again to confirm or refute some conclusions of
probability theory and its applicability.

To do Buffon's experiment one needs a horizontal surface ruled by
a grid of parallel equidistant lines. The distance "d" between the
lines and the length "1" of the needle must satisfy the relation
1less thand. Toss the needle above the surface, each time giving it
a little flip so that it falls freely from a height of about 50-60
cm and lands at random angles relative to the lines. After each
toss is completed write down whether the needle does or does not
intersect one of the parallel lines and calculate the frequency of
"crossings," that means, the ratio of the number of throws "m"
resulting in intersection to the total number of throws "n." One
will soon see that as the number of tosses in the experiment
increases, the frequency changes. Not only that, but if one
performs many trials consisting of many tosses, the frequency of
"crossing" is approximately the same for each trial.

REVIEW OF LITERATURE

Archimedes' "Measurement of a Circle," is a treatise containing
only three propositions. The treatise as it has come down to us is
not in its original form and may be only a fragment of a larger
discussion. Around 150 A.D., the first notable value for pi after
that of Archimedes was given by Claudis Ptolemy of Alexandria in
his famous Syntaxis Mathematica (more popularly known by its
Arabian title of the Almagest), the greatest ancient Greek work on
astronomy. In this work, pi is given 377/120 or 3.1416. It may have
come from some earlier Greek source or, maybe from calculating the
perimeter of a regular inscribed polygon of 384 sides. In 1593,
Adriaen Van Roomen of the Netherlands, found pi correct to 15
decimal places by the classical method, using polygons with 2
sides. A few years later in 1610, Ludolph Van Leulen of Germany
spent a large part of his life on this task and his achievement was
considered so extraordinary that the number was engraved on his
tombstone, and to present day it is sometimes referred to in
Germany as "The Ludolphine Number." More than a century and a half
later, Comte de Buffon devised his famous needle problem by which
pi may be determined by probability methods. Suppose a number of
parallel lines, h distance apart, are ruled on a horizontal plane,
and suppose a homogeneous uniform rod of length 1less thanh is
dropped at random onto the plane. Buffon showed that the
probability, P, that the rod will fall across one of the lines in
the plan is given by P equals (2/pi)(1/h).

By actually performing this experiment a given large number of
times "n" and noting the number of successful cases "m," thus
obtaining an empirical value for P, (namely, Pequalsm/n), we may
use the above formula to compute an approximation for pi. The best
result obtained in this way was given by the Italian, Lazzerini, in
1901. From only 3408 tosses of the rod he found pi correct to 6
decimal places. His result is so much better than those obtained by
experiments that it is sometimes regarded with suspicion. There are
other probability methods for computing pi. Among the curiosities
connected with pi are various mnemonics that have been devised for
the purpose of remembering pi to a large number of decimal places.
The following by A.C. Orr, appeared in the Literary Digest. One has
merely to replace each word by the number of letters it contains to
obtain pi correct to 30 decimal places.

Now I even I, would celebrate
In rhymes, unapt, the great
Immortal syracusar, revaled nevermore,
Who in his wondrous love,
Passed on before,
Left men his guidance
How to circle mensurate.

Most recently, in 1934, a great many college and public libraries
throughout the United States have received, from the obliging
author, complimentary copies of a thick book devoted to the
demonstration that pi equals 3 13/81. And then there is House Bill
No. 246 of the Indiana State Legislature which attempted, in 1897,
to determine the value of pi by legislation. In section I of the
bill we read: "Be it enacted by the General Assembly of the State
of Indiana: It has been found that a circular area is to the square
on a line equal to the quadrant of the circumference, as the area
of one side . . ." The bill passed the House, but, because of some
newspaper ridicule, was shelved by the Senate, in spite of the
energetic backing of the State Superintendent of Public
Instruction. There is more to the calculation of pi to a large
number decimal places than just the challenge involved. One reason
is to secure statistical information concerning the "normalcy" of
pi.

PROCEDURAL ANALYSIS

To complete this experiment, the first step was to design a grid by
drawing horizontal lines "h" cm apart from one another on a board.
Then drop a needle "1" centimeter long 50-60 centimeters from the
board "n" numbers of times. The distance between the lines on the
grid and the total number of times the needle was tossed were
picked at random.

The next step was to mark each time that the needle crosses any
line as an "intersection" and each time that it does not cross any
line as a "toss." Then by using these numbers, make a tally of the
tosses and intersections ("m" equals number of intersections) and
record the data.

The final step used in this experiment was to use the formula:
pi (approx equal) 21h ;pd nm to calculate an estimate of Pi.

RESULTS

The results of this experiment were found to be less than the
margin of error. By using the formula:

pi (approx equal) 21h ;pd nm

the end result was 3.149259261 which was off by 0.007 of the actual
estimate of Pi which is 3.1415927. . . .

Also, in this experiment a tally was used to calculate the number
of tosses and intersections. The final numbers from that method
were used in the formula above.

Tosses (n)               Intersections (m)
1265                     414
Length of needle (1)     Distance between lines (h)
2.926 cm                 7.62 cm
Mathematical constant Pi (pi)
Pi equals 3.141592654 . . .

Pi can be calculated in this manner, however, this method is
extremely slow and somewhat prone to error. In order to evaluate Pi
to an accuracy of just 0.01 decimal place, one would have to take
approximately 500-1000 more tosses.

Calculated Results of Pi

                              No. of    Experimental
Name                Year      Tosses    Value of Pi

Wolf                1850      5000      3.1596

Smith               1855      3204      3.1553

Fox                 1894      1120      3.1419

Lazzarini           1901      3408      3.1415929

This Experiment
 (the average of
 three trials)      1993      1268      3.1492592

True value of Pi to the seventh decimal place: 3.1415927


CONCLUSION

In conclusion, it was proven that the original hypothesis was
correct; because the end result was 3.149259261 which was only off
by 0.007 of the actual estimate of Pi.

Some of the most interesting phenomena found in conducting this
experiment include the following: First, the accuracy of the
measurement--to obtain the precise value of pi from the equation,
one must be able to precisely measure h and 1. This can be done
only by actually measuring both values. Errors in measurement will
definitely affect the accuracy of pi. Making use of more
sophisticated instrumentation, one can measure all the lengths to
an accuracy of 0.01 mm. At this point, it has reached the practical
limit, and it is nearly impossible to reduce the error to 0.001 mm
because a variation in the needle's or the surface's temperature of
only 1 or 2 degrees results in variation of about 0.001 mm in the
measured distances. Deformation of the needle caused by its
collision with the surface, wear at its tips, deformation of the
surface itself--all these factors make it quite unreasonable to try
to achieve a level of experimental error of 0.001 mm. This means
that with conventional measuring devices the best that one can
expect to obtain is pi equals 3.141 + 0.006. By writing all the
decimal places obtained; the accuracy of the experiment can be
overrated and thereby report a misleading result.

ACKNOWLEDGEMENTS

This researcher would like to thank the following people for their
hard work and dedication in helping me complete this project; but
first I thank God for giving me strength and will power to perform
my daily tasks. I would also like to thank Mr. N. Nevels for his
guidance and patience during the project; and I would also like to
thank Mr. Edward C. Haynie for introducing me to this project and
giving me the motivation to do this project.

BIBLIOGRAPHY

Eves, Howard, An Introduction to the History of Mathematics, 3rd
Edition, March/April 1950, p. 89-106.

Vavilou, Valery, Calculating pi, Quantum, May/June 1992, p. 45-50.

Zaydel, A.N., Delusion or Fraud?, Quantum, Sept./Oct 1990, pp. 6-9.




PRECOLLEGE AWARD

Ralph Osif, Freshman
Powell Jr. High School, Mesa, Arizona

Comprehensive Regional Centers for Minorities Project
Principal Investigator, Alfredo G. de los Santos, Jr.

THE EFFECTS OF CEDAR SMOKE ON RADISH SEEDLINGS

Abstract: The purpose of this study was to determine if smoke will
effect radish seedlings as measured by leaf and stomata coloration,
and condition of stems. The study involved three groups: two
experimental and one control. Growing conditions were kept
consistent across all three groups. Plants were kept in three
inverted aquariums. Treatments: Group 1--Control--no smoke; Group
2--Smoke from 20 ml of burning cedar chips twice a day for four
days; Group 3--Smoke from 80 ml of burning cedar chips twice a day
for four days. Experimental plants were wilted and the leaves were
paler green than those in the control group. In addition, when
stomata were compared, 82% of Group 3's stomata were discolored,
48% of Group 2's, and 0% of the control. Conclusion: Based on the
data collected from this study, it appears that smoke is harmful to
the general health of radish seedlings.

INTRODUCTION

The purpose of this experiment was to determine if radish seedlings
would be effected from exposure to smoke produced from burning
cedar chips.

The rise in air pollution, either from man-made causes or natural
causes, is known to have many different impacts on our quality of
life. Is it also having an impact on the plants around us?

BACKGROUND INFORMATION

Smoke is a mixture of solid particles and gases suspended in air.
The make up of smoke depends on the substance that is burned. There
are hydrocarbon particles and carbon monoxide gases in smoke that
comes from burning wood.

There are varied sources of environmental smoke or pollution. Wood
fires, automobile and truck exhaust, factories, and utility
companies all contribute to pollution in the atmosphere. Some of
the specific types of pollution that are frequently measured to
determine air quality are:


Pollutant                Source

carbon dioxide           Naturally occurring from animal
                         respiration; Other sources: burning of
                         forests, volcanic activity, natural
                         decay; major cause of greenhouse effect

carbon monoxide          major source: automobile exhaust, fossil
                         fuel combustion, home kerosene heaters,
                         cigarette smoke

lead                     Major source: Older cars using regular
                         gasoline

nitrogen dioxide         major source: automobile exhaust

particulate matter       natural sources include: soil dust,
                         pollen, bacteria, and volcanic dust;
                         man-made sources include dust, soot, and
                         ashes from combustion and incineration.

sulfur dioxide           produced by combustion of coal, fuel oil,
                         and gasoline; primary cause of acid rain

smog                     a combination of smoke and fog

smoke                    major source: forest fires, wood stoves,
                         fireplaces, coal

(Sandak, 1990).


In March 1991, the Environmental Protection Agency Journal reported
that over 84 million people breathe air violating at least one
federal standard. In fact, 67 million live in countries which
exceeded the smog level standards; 34 million live in countries
exceeding the carbon monoxide level standards; and 27 million live
in countries exceeding the particulate level standards. The World
Health Organization estimates that worldwide nearly 1 billion urban
dwellers--almost one out of every five people on Earth--are being
exposed to health hazards from air pollutants (Tyler, 1992).

There is often more smoke pollution in the summer than in the
winter due to the larger number of brush and forest fires during
the drier summer months. Almost 3,000 acres were destroyed from a
recent forest fire which occurred in Boise National Forest. It cost
$2.8 million dollars to extinguish the fire. (Petranek, 1993).
However, in colder regions such as Denver, Colorado, the
combination of wood smoke and automobile exhaust also creates
serious air pollution problems during winter periods when there is
little wind (Sandak, 1990). Adding to this problem is the fact that
the Environmental Protection Agency has classified microscopic
particles of incompletely burned wood as being carcinogenic
(Sandak, 1990).

HYPOTHESIS

If radish seedling are exposed to smoke from burning cedar chips,
the health of the seedlings will be effected, as seen in the color
of their leaves, condition of their stomata, and presence of
wilting.

METHOD

Materials

The materials for this experiment included approximately three
hundred radish seedlings placed in common potting soil. The
seedlings had germinated within the week preceding the treatments.
The radish seedlings were germinated in germination trays that were
placed in a tray which had a clear plastic lid.

Materials for the experimental treatments included:

     3         ten gallon aquariums
     3         ceramic evaporating dishes
     400ml     Cedar chips
     3         alcohol thermometers
     1         glass rod
     8         matches

Procedures

The radish seeds for this experiment were germinated and grown in
plastic containers of equal size, and shape. The same amount and
type of potting soil was used for the planting of the seeds, and
all of the seeds were from the same lot. Each of the three groups
were exposed to the same light source, and temperature, and
received the same amount of water. One week after the radish seeds
had been planted, the seedlings were divided into three equal
groups: one control, and two experimental. Each group of seedlings
was then placed under an individual, inverted ten gallon aquarium.

The two experimental groups (Group 2 and 3), were exposed to cedar
smoke, and the control group (Group 1) was not exposed to any
smoke. The procedure for exposing the seedlings to the smoke was as
follows:

o measure out specified amount of cedar chips

o place cedar chips into an evaporating dish

o ignite the cedar chips with a match

o stir the chips while they are burning to ensure all chips are
ignited

o when flames have died out and the chips have started to produce
smoke, place into the inverted aquarium

o remove evaporating dish when all smoking has stopped

Group 2 was exposed to 20 ml of burning cedar chips and Group 3 was
exposed to 80 ml of burning cedar chips. Five minutes after the
smoking wood chips were placed in the aquarium, the inside
temperature of each aquarium was recorded. The cedar chips were
placed in the aquariums at 8:30 a.m. and again at 4:30 p.m. This
procedure was repeated twice a day for four consecutive days.

Initial recordings were taken from each group of plants. These
recordings included the color of their leaves, condition of their
stomata, and presence of wilting. At the conclusion of the four-day
experiment these recordings were taken again. The following
procedure was used for determining the condition of the stomata.
Sample leaves were taken from each of the three groups. These
leaves were examined with the aid of a microscope set to high
power. The total number of stomata that were in the viewing area of
the microscope were counted. Then the number of stomata that
appeared darkened were counted. These two figures were used to
calculate the percentage of darkened stomata.

RESULTS

There were differences in the three groups of radish seedlings when
looking at the resulting leaf color, condition of stomata, and
presence of wilting. The results of this experiment can be seen
below.

Table 1
Results of Radish Seedlings Being Exposed to Cedar Smoke

Groups            Group 1               Group 2       Group 3

Indicators:    Before    After     Before    After  Before  After

% of Darkened
 Stomata       0%        0%        0%        48%    0%      82%

Leaf Color     Green     Green     Green     Pale   Green   Yellow
                                             Green          to Pale
                                                            Green

Presence of
 Wilting       No        No        No         Yes      No   Yes


Group 3, which had been exposed to the largest amount of smoking
cedar chips (80 ml) had the largest effects. The overall appearance
of the leaves on the seedlings in Group 3 was changed from the
initial appearance. The leaves initially were bright green, and
after being exposed to 80 ml of smoke twice a day for four days,
the leaves had changed from bright green to yellow or pale
yellow-green. There was no presence of wilting at the onset of the
experiment, however, at the conclusion the seedlings were wilted.
The initial condition of the stomata was that all were clear bright
green. After being exposed to the smoke, 82% of the stomata became
darkened.

Similar results were present for the seedlings in Group 2, however,
not as prominent. Again, the leaves initially were bright green,
and after being exposed to 20 ml of smoke twice a day for four
days, the leaves had changed from bright green to pale yellow-green
or pale green. There was also no presence of wilting at the onset
of the experiment, however, at the conclusion the seedlings were
slightly wilted. The stomata in Group 2 were initially clear, but
at the end of the experiment, 48% had become darkened.

Group 1's initial recordings showed that all the plants were not
wilted, had green leaves and the stomata were clear. At the end of
the four days, there were no noticeable changes to the seedlings in
Group 1.

CONCLUSION

Based on the data collected in this study, it appears that smoke is
harmful to the growth of plants. The plants which had been exposed
to smoke lost the color in their leaves, were wilted and had
stomata which were discolored. In other words, smoke proved to be
harmful to the plants. We need plants to survive in our world, and
the pollution that is being generated by our society could possibly
effect the ability of plants to survive, which in turn effect our
ability to survive. With this information, I believe we should help
out our environment by finding ways to decrease the amount of smoke
pollution which enters our atmosphere.

REFERENCES

"New Report Shows Progress in Air Quality." EPA Journal, March
1991.

Petranek, S. (September, 1993). The force of nature. Life, pp.
31-37.

Sandak, C. R. (1990).  A reference guide to clean air. Enslow
Publishers: Hillsdale, NJ.

Tyler, M. G. (1992). Living in the environment,  Wadsworth
Publishing Company: Belmont, CA.





UNDERGRADUATE AWARD

Peter Cabauy, Freshman
Department of Chemistry, Florida International University
Alliances for Minority Participation Project
Principal Investigator, Kenneth G. Furton

A STRAIGHTFORWARD INEXPENSIVE METHOD FOR DETERMINING CRITICAL
MICELLE CONCENTRATIONS IN UNDERGRADUATE LABORATORIES

Abstract: A new method has been developed for determining the
critical micelle concentration (CMC) of aqueous surfactant
solutions. This method is accurate, straightforward, and is very
inexpensive compared with existing methods used for determining
CMCs. In addition, the method developed allows students to
visualize the concept of surface tension without having to use
expensive instrumentation. The method is based on the rapid drop in
water droplet size as the CMC is approached followed by a more
uniform droplet size above the CMC.

INTRODUCTION
Surface active agents (surfactants) are used in a variety of areas,
including chemical kinetics and as membrane mimics in biochemistry.
Surfactants are encountered in a diverse range of consumer products
including motor oils, pharmaceuticals, soaps, and detergents.
Surfactants are amphilic substances consisting of a long-chain
hydrocarbon "tail" and a polar (often ionic) "head". One unique
property of surfactant is that at sufficiently high concentrations
the surfactant molecules begin to form into molecular assemblies
known as micelles. The concentration at which this occurs is known
as the critical micelle concentration (CMC). Normal micelles are
those in which the surfactant molecules orient themselves into
spherical or elliptical structures with their lipophilic tails
oriented toward the center and their hydrophilic heads oriented
toward the surrounding water. The structure of these micelles will
vary according to temperature, polarity of the solution, and the
concentrations of surfactant molecules in solution. One of the most
common types of aqueous surfactant in consumer chemistry are the
linear alkylbenzenesulfonates (LAS) which are anionic surfactant
present in the majority of home laundry detergents. A two
dimensional representation of how a typical spherical micelle might
look in solution is shown in Figure 1. Micelles should be
visualized not as static objects, but as molecular assemblies in
constant motion with exchanges of amphiles occurring between
micelles and between bulk aqueous solution and the micelle.

There are several methods for determining the CMC of a surfactant
solution. CMC is determined by physiochemical properties of the
surfactant solution as the concentration of the amphile (surfactant
molecule) is increased. Some of the physical properties that have
been studied for this purpose include the solution detergency,
viscosity, density, conductivity, surface tension, osmotic
pressure, interfacial tension, refractive index, and light
scattering. Other more sophisticated techniques that can be
employed include X-ray diffraction, electron spin resonance
techniques, potentiometric techniques, fluorescence emission
spectroscopy. For a more detailed discussion of micelles, their
uses, and the determination of CMC, readers are directed to
references 1-5.

Recently developed methods for use in undergraduate experiments
make use of additives to indicate the formation of the micelles and
determine the CMC of aqueous solutions by studying the change in
uv-vis absorbance or fluorescence emission. In a typical such
experiment a constant amount of organic compound (usually a dye) is
added in solutions containing increasing concentrations of
surfactants. The percentage change or emission is plotted against
the surfactant's concentration. The midpoint of the transition of
the sigmoidal curve obtained is taken to be the CMC. Reference 6
and 7 are recent undergraduate experiments using these methods.

WHY ANOTHER METHOD?

The uv-vis absorption and spectrofluorometric methods provide a
reasonable estimate of the CMC but pose several problems when
applied to an undergraduate experiment. First, the data
manipulation required may reduce the accuracy of the technique
(e.g., 18-39% Relative Average Deviation from Ref. 6) and may cause
confusion as to what is actually happening at the CMC (e.g.,
determining the midpoint of the transition of the sigmoidal curve
for the percentage change in measurements). Second, these methods
require relatively costly instrumental techniques that may not be
available at the undergraduate level (particularly
spectrofluorometric instrumentation). Thirdly, the presence of
organic compounds, including dyes, used in fluorescence emission
and spectroscopy can significantly affect the observed value of the
CMC. Polar organic molecules can cause a marked depression of the
CMC in aqueous media, even at very low bulk phase concentrations.
The degree of CMC depression is related to the polarity of the
additive, the degree of branching, and the locus of solubilization.

Due to problems with additives in surfactant solutions, surface
tension is one of the most used techniques for measuring CMC;
however the instruments used are not cost effective for
undergraduate laboratories. The most common test for measuring
surface tension is the use of a torsion balance along with a metal
ring that is placed at the surface of the solution. The degree of
surface tension measured by the balance to draw up the metal ring
from the solution is measured and the surface tension is then
calculated and plotted against the concentration of surfactant in
solution. Difficulties that are encountered in an undergraduate
experiment are costs and level of cleanliness. The degree of
cleanliness in experiments with surface tension is very important
because surface tension can be affected by any other electrolytes
present in solution.

Presently, we will describe an inexpensive, noninterfering
technique suitable for determining accurate CMC's in an
undergraduate experiment. The technique we have developed is based
on the relatively linear decrease of surface tension up to the CMC
where the surface tension remains relatively constant. To have an
understanding of how this technique works some knowledge of surface
tension is required, readers are directed to references 3&8, but
for all practical purposes we will try to give a brief summary
related to the experiment at hand. Surface Tension can be described
as the interactions of the surface molecules of a solution with the
bulk solution and its interfaces. The interfaces in question are
usually glass and air. Surfactant molecules tend to travel to the
interfaces and accumulate there. Keeping in mind that they do not
remain fixed at the interfaces one can understand that the
concentration of amphiles in solution disturb the normal adhesive
properties that water has on glass. The alleviation of surface
tension on the solution gives a great deal of emulsifying power to
regularly immiscible solutions. The amphillic properties of the
micelles give headway to both solutions to emulsify one into
another. The linear decrease of surface tension before reaching the
CMC can be described by visualizing an increased concentration of
amphiles at the solutions interface. After the CMC has been reached
the micellization that occurs does not affect the area because the
interface has already been saturated with amphiles and can no
longer make a significant difference in the surface tension. As the
micelles form they take on a three-dimensional spherical,
elliptical, or rod-like form. The form will depend on the
concentration of the surfactant solution and the space between
micelles at the interface.

The experiment that has been developed uses the simple concept of
a medicine dropper. Water or solution droplets are formed because
of surface tension. If it were not for this phenomena a steady
stream would flow from the opening of the glass capillary.
Differences in surface tension will affect the drop size.
Furthermore, one can use a variety of measurements, such as:
constant volume plotted against number of drops, weight of drops
plotted against concentrations, and or constant drops giving
different volumes plotted against concentrations. As long as the
right variables are held constant an accurate CMC can be
determined.

Although the idea of a medicine dropper forming drops of liquid may
seem like a consistent unit in any other experiment in a chemistry
lab, there is a large percentage of error. In our endeavors we had
to invent an inexpensive, consistent water/solution dropper (Figure
2). Using some facts of pressure from physics we were able to make
a simple system that uses the following: burette valve, disposable
pipette, and a common bulb from a medicine dropper. Assembled as in
figure 2 and using a consistent procedure we can attain pure water
drops that weigh approximately 0.0250 g with a variation of plus or
minus 0.0003 grams. Now, that drops are reasonably consistent we
can determine the weight of each drop on a scientific balance.

PROCEDURE

1. Prepare 10 solutions with various concentrations of LAS from 0
to 2.8x10E-3M. Our solutions were made up in 100ml volumetric
flasks.

2. Assemble dropper (Fig. 2) providing consistent drops which will
be weighed with an analytical balance.

3. Transfer solutions into small vials.

4. Open valve and draw up water solution from vials using the bulb.
Close valve and remove bulb then calibrate dropper to make a drop
roughly every ten seconds. When this is achieved make a calibration
mark on the valve in order to achieve same opening in later steps.

5. Draw up a water solution once more and open the valve to
calibration mark and remove the bulb. Re-zero scale after each
drop. Wait for 5-7 drops checking consistency in drop weight. If
three drops are reasonably consistent after the first 5 drops
record the weight.

6. Repeat procedure for different solutions keeping in mind that
the pipette has to be rinsed with water after each solution. For
practical purposes draw up water and release twice after each
solution, this will clean the dropper and keep its consistency.
Each solution should be drawn up more or less to the same vicinity
as all the other solutions. The Bulb should always return to its
original relaxed position.

Once the weight has been obtained it should be plotted against the
concentration. The concentration should be the independent variable
and the weight of the drops should be the dependent variable. Once
we have plotted the data points, we perform a linear least squares
regression analysis of the points on the y-axis with the greatest
differences versus the points with the least change (where the
drops size remains constant). The point where the lines intersect
is the CMC (Refer to Figure 3). Alternatively, we have also
determined the number of drops required to reach a volume of 1.0 ml
on a graduated cylinder with the number of drops required
increasing suddenly, then leveling off above the CMC. These methods
represent affordable ways in which students in undergraduate labs
can work with surfactants, and open new ideas and methods to use
surface tension in the determination of CMCs.

CONCLUSIONS

The method proposed here is a straightforward, extremely
inexpensive, and accurate technique for determining the CMC of
aqueous surfactant solutions. The advantages of this method over
the previous dye absorption methods are three-fold. First, the
method can be used without costly instrumentation. Second, since no
additives are used, this method prevents any possible CMC
depression caused when additives are present. Finally, the
information generated is accurate and easily interpreted by novice
analysts. The formation of more constant-sized droplets (volume and
weight) corresponds with the formation of micelles and pinpoints
the CMC. Below the CMC, the droplet size consistently increases
with the largest droplets observed for pure water.

ACKNOWLEDGEMENTS

The author would like to thank Arold Norelus and Mike Sarkees for
useful suggestions and assistance in developing these experiments.
I would also like to thank Dr. Efrain J. Ferrer for giving me the
opportunity to participate in the FGAMP program. And finally, I
would like to thank Prof. Kenneth G. Furton, my mentor and
supervising scientist for allowing me to develop these experiments
in his laboratory.

LITERATURE CITED

1. Lindman, B. and Wennerstrom, H. Micelles: Amphile Aggregation in
Aqueous Solution; Springer-Verlag, Berlin, 1980.

2. Fendler, J.H. and Fendler, E.J. Catalysis in Micellar and
Macromolecular Systems; Academic Press, New York, 1975.

3. Rosen, M.J. Surfactants and Interfacial Phenomenon; John Wiley
& Sons, New York, 1978.

4. Love, L.J.C., Habarta, J.G, and Dorsey, J.G., Anal. Chem., 1984,
56, 1132A.

5. Furton, K.G. and Norelus, A. J. Chem. Ed., 1993, 70, 254.

6. Rujimethabhas, M. and Wilairat, P. J. Chem. Ed., 1978, 55, 342.

7. Sawyer, D.T.; Heineman, W.R. and Beebe, J.M. Chemistry
Experiments for Instrumental Analysis; John Wiley & Sons, New York,
1984, Experiment 10-4 "Critical Micelle Concentration of
Surfactants", p. 283.

8. Eicke, H.-F. and Parfitt, G.D. Interfacial Phenomena in Apolar
Media; Marcel Dekker, New York, 1987.




UNDERGRADUATE AWARD
Jonathan Hill, Junior
Department of Civil and Construction Engineering, Iowa State
University
Research Careers for Minority Scholars Project
Principal Investigator, Loren Zachary

DESIGN METHODOLOGY FOR CORRUGATED METAL PIPE TIEDOWNS

Abstract: Corrugated metal pipes (CMP), which are pipes that run
underneath roads to permit water flow, have a tendency to uplift.
Several transportation agencies around the U.S. and Canada have
used different types of tiedowns to overcome the problem, however
there is no agreement in the magnitude of hold force. This paper
describes research in progress for the development of a rational
and economical approach to designing tiedown structures for
flexible pipes. To acquire more knowledge on the longitudinal
stiffness, three pipes were tested in the laboratory. Since pipes
are covered by soil, a field test was used to determine how the
soil affected the stiffness of the pipe. The results of the field
test were used to create a finite model of the uplifting situation
which led to the calculation of the force required for the tiedown.
The force obtained was a conservative value of 37.5 kips which is
approximately 9.3 yd(sup)3 of concrete.

INTRODUCTION

Problem

In the mid 1970s both the Iowa Department of Transportation and the
Federal Highway Administration recognized that corrugated metal
pipe (CMP) culverts were experiencing uplift failures at an
unacceptable rate and that for pipes over 4 ft in diameter
provision should be made for tiedowns at the inlets. In a national
survey of 60 state and provincial Department of
Transportations(DOT) (Klaiber et al., 1993) it was revealed that 26
agencies used some form of restraint to overcome the uplifting. The
resistance force of the restraints were compared with the pipe
diameters and then graphed to see the relationship. The
relationship was either linear (Fig. 1) or exponential (Fig. 2)
curves. Also shown in the figures is the conservative relationship
resulting from the rational analysis of Austin et al (1990). The
fact that uplift failures are still occurring plus the wide
variation in tiedown forces being required by various agencies
(Fig. 1 and Fig. 2) indicates the need for a rational design method
for safe economical CMP tiedowns.

Objectives

The primary objective of the research is the development of a
rational method to design the restraint system a given CMP needs to
prevent uplift failures. Due to the elaborate scope of this
project, the study is divided up into two phases with specific
objectives in each phase. Phase I has been completed and summarized
in past reports; it consisted of surveying design standards of
other state DOTs, determining the longitudinal stiffness, and
beginning to obtain experimental data on soil-CMP interaction.
Phase two which is presently in progress consists of completing the
collection of the soil-CMP interaction data, incorporating the
water depth computations of Austin et al (1990) into an integrated
program, and synthesizing all of the data into a rational design
procedure. This procedure will lead to development of software and
finally to the development of a design standard for corrugated
metal pipe tiedowns.

Previous studies

Very little has been published in the open literature on CMP uplift
failure. The most recent and definite research has been sponsored
by the Iowa DOT in the projects HR-306 (Austin et al., 1990) and
HR-332 (Klaiber et al., 1993).

Austin et al. (1990) had the objective of determining the magnitude
of the problem. Computer programs were developed for computing the
depth of water flow within the conduits for a variety of flow
conditions. Statical analysis was also done on the CMP neglecting
the pipe flexibility and shearing resistance of the soil. The CMPs
were assumed to be plugged in most of the analyses so the only
resistance to uplift was provided by the pipe weight, weight of
overlying soil, and resisting force provided by inlet tiedowns. The
static analysis and the analyses of two CMP failures demonstrated
that the pore pressure distribution beneath the pipes creates force
of sufficient magnitudes to cause uplift failure if resistance by
tiedowns is not provided.

Klaiber et al. (1993) addressed the "longitudinal bending
stiffness" of the CMPs. Using the equation:

EI equals "Longitudinal Bending Stiffness"   (1)


Moment of Inertia(I) is a function of the shape of an object and
Modulus of Elasticity(E) is constant for a given metal. So the
Moment of Inertia was determined experimentally by testing three
CMPs (4 ft, 6 ft, and 8 ft in diameter). Since these three tests do
not represent all the different types and sizes of CMPs, a
theoretical model was developed from the experimental results. The
following formula was developed to determine the stiffness for a
variety of CMPs with different corrugation geometries, gages,
diameters, etc.

EIequalsEpiScrt4 KG 31+3 K2G rt6+KG L3T sin;gu cos;gu12 dc+KR2 (2)

CMP Uplift Test

To obtain data on soil-structure interaction one CMP (10 ft in
diameter) was tested with a minimum soil cover of 2 ft. The results
from this test were used to calibrate a finite element model.

Experimental Test Set-Up

The galvanized, steel CMP used for the field test was received in
two separate parts. The two parts were joined together by a 10 in.
"hugger band" to form a 52 ft long pipe. The pipe was 10 ft in
diameter with 3 by 1 helical corrugations.

The site for the test was an existing embankment of undisturbed
glacial till at Iowa State University Spangler Geotechnical
Laboratory. The soil was excavated to fit the CMP and best simulate
ideal conditions (Fig. 3).

To simulate the pore pressure, which creates the uplifting force as
explained in Austin et al (1990), two load points at 5 ft and 15 ft
from one end were used. Obviously, it is very difficult to subject
the pipe to forces which closely approximate pore pressures. At
these load points uplift was created by a set of four hollow core
hydraulic cylinders supported on a sturdy load frame. A beam-girder
system was used to transmit the hydraulic cylinder forces to four
supporting columns (Fig. 4). Data from the two loading points will
be used to calculate the finite element model of the system.

INSTRUMENTATION

Data was read and recorded from the instruments by a Data
Acquisition System (DAS) which was placed inside the laboratory.
Strain gages used to measure longitudinal and hoop strains were
placed at six longitudinal sections of the pipe shown in Fig. 5.
Lightweight rods with Direct-Current Displacement Transducers
(DCDT's) appropriately attached were connected to the inside walls
of the CMP near the same sections as the strain gages. These were
used to measure vertical and horizontal diameter changes in the
pipe (Fig. 6).

Soil pressure cells were placed in the soil around the pipe as
shown in Fig. 7. Vertical deflections of the pipe were measured
with vertical steel rods which were attached to the top of the CMP
and extended above the soil at several locations shown in (see Fig.
8).

ANALYSES

Experimental Analysis of Data

In order to simplify the evaluation of the overall behavior of the
CMP-soil system, each component of the problem was separated and
analyzed independently. The components of the problem are the
weight of the soil (Ws), the weight of the pipe (Wp), the uplifting
force (Wu), and the stiffness of the CMP (EI). Two of these
components (Wp and Wu), are already known. The Wp was given by the
manufacturer. The Wu is calculated using load cells. Analysis of
the experimental data was conducted to better understand the values
of Ws and EI.

As mentioned earlier, the two unknown components are the soil load
(Ws) and the overall longitudinal stiffness (EI). The following
briefly describes the steps that were performed to approximate the
soil response and CMP stiffness:
1) A displacement function was derived from the deflection data
obtained in the experiment.

2) The first derivative of the displacement function was evaluated
to obtain the slope of the deflected shape.

3) From the evaluation of Steps (1) and (2) the point where both
the deflection and slope were zero was found. From this point to
the inlet of the pipe, the structure was assumed to act like a
cantilever beam.

4) The information about the slope and the displacement was used in
two standard engineering mechanic beam theorems (Zachary, 1989).
The two unknowns Ws and I were calculated from the theorems by
using extensive numeric integration. The results are: I equals 1008
in(sup)4, Ws equals 5.61 k/ft

Theoretical Analysis of Soil Load (Ws)
The magnitude of soil load acting on the upper region of the CMP is
dependent on the area of soil that the pipe activates as it is
uplifted. Unfortunately, there was no instrumentation used to
measure this load. However, two models do exist that describe the
pattern the portion of soil takes due to the uplift. They are the
Vertical Slip Surface model and the Frustrum model (Matyus et al.,
1983).

Intuitively, the Vertical Slip model is probably the best estimator
for shallow fill cases while the Frustrum model more accurately
predicts the failure for CMP under high fill conditions. In the
Vertical Slip Surface model, soil failure is assumed to take place
along the vertical surfaces extending up from the outer edges of
the pipe. Using this model a value of 5.34 k/ft is obtained for Ws.

Theoretical Analysis of Longitudinal Stiffness(EI)

The theoretical model for longitudinal stiffness (equation 2) was
discussed in Klaiber et al (1990). To get a theoretical value for
"I" in reference to the uplifting test, the appropriate values were
applied in the equation and a moment of inertia of 611 in(sup)4 was
obtained. Note that E(Modulus of Elasticity) is a constant.

COMPARISON OF EXPERIMENTAL TO THEORETICAL RESULTS

Soil Component (Ws)

The soil component approximated in the experimental analysis was
compared to the value obtained in the analytical model. The results
were very close which shows that the Vertical Slip Surface model
was an excellent assumption.

Vertical Slip Surface Model
     Ws equals 5.39 k/ft
Experimental Analysis
     Ws equals 5.61 k/ft

Longitudinal Stiffness (EI)

The longitudinal stiffness predicted by the experimental analysis
was slightly greater than the value determined using Equation 2
(Klaiber et al., 1993) as shown below:

Analytical Model
     I equals 611 in(sup)4
Experimental Analysis
     I equals 1008 in(sup)4

The difference between the two models is not a surprise because the
CMP's overall stiffness should be greater in the experimental
analysis due to the presence of soil in the corrugations. Since the
CMP experienced soil interaction in the uplift test, the analytical
model (Klaiber et al., 1993) was modified slightly to incorporate
a new stress ratio; all other values remained the same. Using the
new stress ratio value the moment of inertia was calculated to be
966 in(sup)4.

Finite Element Analysis

As a means of determining the behavior of the entire structure
during uplift, a finite element model was created using ANSYS
software. As a result of the experimental analysis, all of the
external loads are known as shown in Fig. 9. These loads and many
other properties were put into the program to create the model. A
detailed description of the elements and load data can be found in
McCurnin (1993).

Results of Finite Element Analysis

The behavior indicated by the finite element analysis was similar
to what was observed during the field test. The deflected shape of
the CMP was similar to the shape indicated by the experimental
deflection data recorded during the uplifting test. A plot
comparing the two is shown in Fig. 10.

Tiedown Methodology for CMP

Now that a model has been created of the uplifting that occurs on
the CMPs, this model can be used to figure the amount of force that
a tiedown will need to apply. This was done by taking the most
conservative value of all the external loads as shown in Fig. 8.
All loads resisting the uplift were made conservatively small while
the uplifting force was conservatively large. The results of these
conservative forces will give a very conservative restraint (R) or
tiedown force (Fig. 11). For more details of the ANSYS coding
required see Appendix B and Appendix C of McCurnin (1993).

The tiedown force (R) obtained was 37.5 kips which is approximated
9.3 yd(sup)3 of concrete.

SUMMARY AND CONCLUSIONS

The purpose of this report was to describe research which
eventually will lead to a rational design method for CMP tiedowns.
Keep in mind that the force obtained for the tiedown was taken from
only one experimental test. Future plans for two or three more
field tests and an incorporation of Austin et al. (1990) study on
water depth will confirm or enhance the value obtained in this
report.

ACKNOWLEDGEMENTS

This paper summarizes a portion of the work done on CMP uplift
failures by the Iowa State University Engineering Research
Institute. The research has been sponsored by the Iowa DOT.
Significant assistance was received by Brian McCurnin (Graduate
Student, Iowa State University) and F. Wayne Klaiber (Professor of
Civil Engineering, Iowa State University). The opinions, findings,
and conclusions expressed in this publication are those of the
authors and not necessarily those of the Iowa Department of
Transportation.

REFERENCES

Austin, T. A., Lohnes, R. A., Klaiber, F. W. (1990). "Investigation
of Uplift Failures in Flexible Pipe Culverts," ISU-ERI-Ames 90227.
Engineering Research Institute, Iowa State University, Ames, IA.
Klaiber, F. W., Lohnes, R. A., Zachary, L. W., Austin, T. A.,
Havens, B. T., McCurnin, B. T. (1993). "Design Methodology for
Corrugated Metal Pipe Tiedowns: Phase I," ISU-ERI-Ames-93409.
Engineering Research Institute, Iowa State University, Ames, IA.
Matyas, E. L. and David, J. B. (1983). "Prediction of Vertical
Earth Loads on Rigid Pipes," Journal of Geotechnical Engineering,
ASCE, Vol. 109, No. 2, pp. 190-201.

McCurnin, B. T. (1993). "Investigation of Soil-Structure
Interaction of Corrugated Metal Pipes," ISU Thesis. Engineering
Research Institute, Iowa State University, Ames, IA.
Riley, W. F. and Zachary, L. W. (1989). "Introduction to Mechanics
of Materials," John Wiley and Sons, New York, NY.

NOTATIONS

The following symbols are used in this paper:

dc = corrugation depth
E = modulus of elasticity
I = moment of inertia
LT = corrugation tangent length
R = restraint load
r = radius
t = thickness of pipe walls
Wp = pipe load
Ws = soil load
Wu = uplift load


Figures are not a part of this electronic version; refer to the
printed version for figures and equations.

FIGURE 1
Linear relationship between force and pipe diameter according to
various agencies.

FIGURE 2
Exponential relationship between resisting force and pipe diameter
according to various agencies.

FIGURE 3
Soil excavation details: cross section.

FIGURE 4
Load frame description.

FIGURE 5
Strain gages.

FIGURE 6
Instrumentation to measure diameter changes.

FIGURE 7
Soil pressure cell locations.

FIGURE 8
Vertical deflection rods.

FIGURE 9
Location and direction of external loads.

FIGURE 10
Comparison of deflection curves for the ANSYS finite model and the
experimental uplift test.

FIGURE 11
External loads plus tiedown force (R).




UNDERGRADUATE AWARD

Lennard Perez, Jr., Senior
Department of Engineering, St. Mary's University

Research Improvenment in Minority Institutions Project
Principal Investigator, Mehran Abdolsalami


CALCULATIONS OF ELECTRON-N2 CROSS SECTIONS
USING RIGID-ROTOR APPROXIMATION

Abstract: Previous theoretical studies of electron-N2 collisions
have shown cross sections which were found to be converged to
within one percent accuracy. However, our preliminary scattering
calculations of electron-N2 collisional cross sections led to
significantly different values at certain energies compared to the
previously published results. These differences required further
convergence tests for a broad range of energies to ensure the
accuracy of the results. Subsequent extensive convergence tests
resulted in a new set of values for cross sections at particular
energies. The most critical energies in our studies were near
threshold and resonance, at which convergence required considerable
deviation from the scattering parameters used in earlier
calculations. We present a detailed comparison of the results we
have calculated for the electron-N2 cross sections and previously
published results, exhibiting discrepancies of as much as
one-hundred percent.

INTRODUCTION

For years, researchers have studied the behavior of nitrogen in
various scattering processes, one of which is electron-N2
scattering. The importance of low-energy electron-N2 collisional
cross sections is seen in such areas as gas discharge physics, the
physics of planetary atmospheres, and laser kinetic modeling.1
There are two widely used methods for calculating electron-molecule
cross sections, Lab-Frame Close-Coupling (LFCC)2-5 and
Adiabatic-Nuclei (AN).6 In the LFCC method, the dynamical
interaction between the motions of the nuclei and scattering
electron is fully taken into account, thus giving very accurate
cross sections. However, this method is only feasible for simple
molecular structures (e.g. H2). The AN theory simplifies LFCC with
two underlying approximations: the Born-Oppenheimer7 separation of
the nuclear and scattering electron dynamical motion, and
target-state degeneracy, i.e. the scattering electron energy
remains the same at both the entrance and exit channels. Thus, this
method has a considerable conceptual and computational advantage
over the LFCC method. However, the AN method is expected to
breakdown near the threshold and resonance energy ranges8,9.

Since the AN method is based on the Body-Frame/Fixed-Nuclei10,11
(BF-FN) approximation, we present this formulation for the
electron-molecule scattering problem. Furthermore, we discuss the
relevance of the interaction potential (static, polarization,
exchange) and its implementation in our electron-molecule
scattering calculations. We conclude the paper by describing our
present convergence studies of electron-N2 collisional cross
sections, which have shown large percent difference compared to
previously published results.12

SCATTERING THEORY

To apply the AN theory to scattering calculations, we use the
(BF-FN) formulation. In this method, the time-independent
Schrodinger equation for the system wavefunction is written in the
body-fixed8,13 reference frame, with the orientation of the nuclei
(z along this axis) frozen during the collision. Furthermore, we
use the rigid-rotor approximation which fixes the internuclear
separation of the target at equilibrium value. The resulting BF-FN
Schrodinger equation is written as14

[Te - 2Vsp(r) - 2Vex(r) + k2]F^(r) = 0  (1)

where ^ is the orbital angular momentum of the projectile along the
internuclear axis, F^ is the scattering electron wavefunction at
position r from the molecule's center of mass, k is the energy of
the scattering electron in Rydbergs, Te is the kinetic energy of
the scattering electron, Vsp is the static and polarization
potential, and Vex is a non-local exchange interaction. The
exchange effect is only important near the target region (e.g. 0
less than r less than 6 Bohr for N2). Far from the target, however,
long range polarization potential becomes important. We will now
give a brief explanation of these three effects.

The polarization potential is the change in total energy of the
system due to the scattering electron's distortion of the target
charge distribution. We use a polarization potential developed by
Morrison and Gibson15 which is determined from self-consistent
field calculations based on a near Hartree-Fock16 (HF)
representation of the electron wavefunction for the occupied
molecular orbitals of the X1E+g (ground) electronic state of N2 .
In this potential, the non-adiabatic effects are incorporated using
Temkin's non-penetrating approximation17 which results in a "Better
Than Adiabatic Dipole" (BTAD)12 polarization potential.
Furthermore, the BTAD polarization potential asymptotically reduces
to the form

vpol^(r) = - a0(R)2r4 - a2(R)2r4 P2cos0,  (2)

where a0 and a2 are the spherical and non-spherical components of
the polarizability. With the internuclear separation fixed at
equilibrium (R = 2.068a0), the values determined for these
components from this asymptotic form are12 a0(R) equals 11.42a30
and a2(R) equals 3.37a30.

The static effects arise essentially from the Coulomb forces
between the scattering electron and constituent electrons and
nuclei of the target. These effects are computed by averaging the
Coulomb potential energy over the X1S+g electronic state of N2. The
static potential asymptotically reduces to a form defined by
permanent moments of the molecule. At equilibrium the moments
determined from this asymptotic form are12 c2(R) equals
(minus)0.91ea20, c;i4(R) equals (minus)7.41ea40, and c;i6(R) equals
(minus)20.68ea60.In this work, both the polarization and static
components are expressed in a complete set of Legendre projections
as18

Vsp(r) = E ^spE(r)PEcos(0r),  (3)

where 0r is the angle between r and the internuclear axis.
We treat exchange exactly in the scattering calculations. This
amounts to explicit consideration of the pairwise interchange of
the scattering electron with each of the target electrons. This
non-local effect gives rise to coupled integro-differential
equations which are evaluated as14

Vex(-12]r)F^(-12]r) = K^(-12]r, -12]r /)F^(-12]r /)d-12]r /,  (4)

with the exchange kernel given by

K^(-12]r, -12]r /) = Ni=1;gfsup. ital. i(-12]r);vb-12]r (minus)
-12]r /;vb(-)1;gFi2]*2](-12]r /),  (5)

where ;gfsup. ital. i(r) is an occupied molecular orbital of N2 and
N is the number of occupied orbitals.

To solve equation (1), a partial wave expansion of the scattering
wavefunction is invoked using spherical harmonics, i.e.

F^(-12]r) equals 1r 2]l2]maxlfl^(r)Yl^(2]=Tr).  (6)

Substituting this partial wave expansion into the Schrodinger
equation and integrating over the angular variables results in the
set of coupled radial differential equations12

d2dr2 (minus) l(l + 1)r2 + k2 f^ll2]02](r) equals 22]l2]maxl' V^ll
2]/2] (r)f^l'l2]0(r)(7)
+ 2l2]ex2]2]max2]l" K^ll" (r, r')f^l"l2]02](r')dr',

with

(8)
V^ll'(r) equals 2l' + 12l +
11/22]E2]maxEequals0^E(r)C(l'El;^0)C(l'El;00),

and

(9)
K^ll'(r, r') equals (minus)2]E2]maxEequals0 rEless thanrE+1more
than l2]MO2]2]max2]2]l"l;tp2] N2]occ2]2]iequals1gE(ll'l"l;tp;^^i)
;gFil"^2]i2](r') ;gFil;tp^2]i2](r').

Here, C is the Clebsch-Gordan coefficient,19,20 rmore thanless than
equals maxmin {r, r'} and gE is a product of four vector coupling
coefficients.21

We use the linear algebraic method14 to solve the coupled radial
differential equations (7). In this approach, the differential
equations are first converted into integral form. This is
accomplished by incorporating boundary conditions using a
free-particle Green's function. The resulting integrals are then
transformed into a set of linear algebraic equations by using
Gaussian quadratures and solved using applicable linear algebraic
routines. The BF-FN K-matrix is found by imposing boundary
conditions on these solutions as

(10)
f^llo(r) equals 2]6]=T6]jl(kr) + K^llo 2]=Tnl(kr)   as r;ul;in,

where 2]6]=T6]jl and nl are the Ricatti-Bessel and Ricatti-Neumann
functions, respectively. At large values of r the exchange
contribution becomes negligible. Therefore, for r more than 6.0a;i0
we venture from the linear algebraic method and use the efficient
R-matrix propagation scheme22 to propagate the wavefunction to its
maximum radius, rmax.

CONVERGENCE PARAMETERS OF THE SCATTERING CALCULATIONS
There are four limits on the various summations in the radial
scattering equations (7-9) for the electron-molecule system. We
truncate these limits to give results which are within one percent
accuracy. ^max, lmax, lexmax, lMOmax, along with the maximum radius
at which K matrix is extracted from the wavefunction, rmax,
constitute the five convergence parameters used in the scattering
calculations. Proper values for each of these parameters must be
determined in order to assure the convergence of the electron-N2
cross sections. The relevance of these parameters in the scattering
calculations is as follows:

1. ^max: the maximum order Legendre projections included in the
expansion of the potential in equation (8).

2. lmax: the maximum order of partial waves included in the BF-FN
scattering function of equation (7).

3. lexmax: the maximum order of partial waves included in the
exchange term of equation (7).

4. lMOmax: the maximum order of partial waves included in the
expansion of the bound molecular orbitals.

5. rmax: the maximum radius at which the K matrix is extracted.

We performed extensive convergence studies on electron-N2 cross
sections for the four lowest BF-FN electron-molecule symmetries (E
g, E u, II g, II u) at energies up to 1.0Ry. These results were
converged to within one percent accuracy at all energies. We now
discuss the convergence studies which determined the final values
of the scattering parameters used in the calculations.

Our first convergence study involved the radius at which we
extracted the K-matrix, rmax. In previously published electron-N2
results,12 propagation up to 85.0a0 was determined to be sufficient
for accuracy. We found this not to be the case near threshold
(;sl0.0007Ry). Our studies indicate that we must integrate the
coupled equations much further. The most significant deviation was
in the II g symmetry, where the wavefunction needed to propagate as
high as 900.0a0 to give cross sections which were accurate to
within one percent. Differences of up to 100% were observed between
our results and previously published results.12 Using a value so
high for rmax was not a problem in our scattering calculations,
since beyond r = 6.0a0 we used R-matrix propagation, which was
efficient and required no additional computer memory. At energies
above threshold, rmax does not have to be nearly as high for one
percent accuracy in the cross sections.

To determine the Gaussian mesh and the number of points, we started
with the mesh used by Morrison and Saha12 in their calculation of
the electron-N2 cross sections. Their mesh consisted of 50 points:
10 points from 0.0 to 0.7a0, 24 points from 0.7 to 1.5a0, 10 points
from 1.5 to 2.5a0, and 6 points from 2.5 to 6.0a0. Using this as a
foundation, we began our convergence study on the number of points
by simply using more points in these meshes and calculating the
cross sections. These preliminary tests showed considerable percent
change when additional points were added, especially in the E g
symmetry near threshold and II g symmetry near resonance energies.
We then performed extensive convergence studies at these energies,
varying the meshes and the total number of points. We concluded
that 133 points (with the exception of II g between 0.1500Ry and
0.2500Ry) were needed to obtain converged cross sections in the E
g, E u, II g, and II u symmetries for energies up to 1.0Ry. Seven
meshes were used to constitute the 133 points: 26 points from 0.0
to 0.9a0, 19 points from 0.9 to 1.035a0, 17 points from 1.035 to
1.2a0, 20 points from 1.2 to 1.5a0, 22 points from 1.5 to 2.0a0, 16
points from 2.0 to 3.0a0, and 13 points from 3.0 to 6.0a0. For the
II g symmetry near 0.2Ry, however, we needed up to 271 points to
get 1% accuracy in the cross sections. These results differed by as
much as fifty percent compared to the previously published
results.12 It is interesting to note that to get an accurate
representation of the analytic nuclear part of the static potential
we need to have at least 30 points in the region near 1.0a0, where
the potential is known to have a cusp if the rigid-rotor
approximation is invoked. This is obvious from Figure 1 where this
potential is plotted for ^ = 72 using 24 points for 0.7 less than
r less than 1.5a0 and 30 points for 0.9 less than r less than
1.2a0.

INSERT FIGURE 1
FIGURE 1

Analytic nuclear static potential for ^ = 72 with mesh consisting
of 24 points for 0.7 less than r less than 1.5a0 (dotted) and mesh
consisting of two regions (0.9 to 1.035a0 and 1.035 to 1.2a0)
containing 15 points in each region (solid).

As far as lmax is concerned, we concluded that 30 channels were
needed for convergence in the E g(lmax = 58) and E u(lmax = 59)
symmetries, and 16 channels in the II g(lmax = 32) and II u(lmax =
31) symmetries, with ^max set to the corresponding values.

Finally, since lexmax and lMOmax have a big impact on the CPU time,
it is important to minimize the two values as much as possible
while maintaining one percent accuracy in the cross sections. Our
studies over a range of energies found that the final values of
lesmax are: 22 for E g, 21 for E u, 20 for II g, and 11 for II u.
Furthermore, lMOmax = 10 was determined to be sufficient for one
percent accuracy.

RESULTS AND CONCLUSION

Table I shows a comparison of the partial cross sections we have
calculated for the E g, E u, II g, and II u symmetries and results
previously published12 at a range of energies up to 1.0Ry. The
total cross sections, obtained from the sum of the four symmetries,
are also shown. From this data, it is obvious that the largest
deviations occur near the threshold and resonance energy ranges.
Graphs of these cross sections are shown in Figure 2. It is obvious
from Figure 2(c) that the energy at which the cross section is
maximum is shifted from 0.15Ry in the previous results to 0.2Ry.

From the results we have discussed, it is evident that the values
used for the scattering parameters play an important role in the
convergence of the electron-N2 cross sections. When these
parameters are determined, a broad range of energies must be
tested. Without proper choice of energies, misleading convergence
could be obtained for the cross sections. The critical energies
used in our studies were near threshold and resonance, where
considerable deviation from the previously used scattering
parameters12 was needed for convergence.

TABLE 1:  Electron-N2 partial and total cross sections at R =
2.068a0. Column (1) is the result of our calculations and column
(2) the previously published (Morrison and Saha) results, followed
by the percent difference between the two results
((new-old)/new*100). The total cross section is the sum of the
partial cross sections, i.e. E g + E u + II g + II u. (All energies
are in Rydberg and cross sections are in square Bohr)



[SEE PRINTED VERSION FOR EQUATIONS, FIGURES AND TABLES]



Insert Figure 2
FIGURE 2

Plots of our results (solid) versus previous (Morrison, Saha)
results (dashed) for low-energy electron-N2 cross sections at R =
2.068a0. These figures contain the cross section for (a) E g, (b)
E u, (c) II g, (d) II u, and (e) the total cross section (E g + E
u + II g + II u).


ACKNOWLEDGMENT

This research is supported by Grant No. RII-9014116 from the
National Science Foundation to whom we are most grateful. We would
also like to thank St. Mary's University for their support during
the completion of this work.

BIBLIOGRAPHY

  1. A.V. Pheleps, Electron-Molecule Scattering (Wiley, New York,
1979).

  2. A.M. Arthurs and A. Dalgarno, Proc. Roy. Soc. London Ser A
256, 540 (1960).

  3. O.H. Crawford and A. Dalgarno, J. Phys. B 4, 494 (1971).

  4. N.F. Lane and S.Geltman, Phys. Rev. 160, 53 (1967).

  5. R.J.W. Henry and N.F. Lane, Phys. Rev. 183, 221 (1969).

  6. D.M. Chase, Phys. Rev. A 104, 838 (1956).

  7. M. Born and J.R. Oppenheimer, Ann. Phys. 84, 457 (1927).

  8. N.F. Lane, Rev. Mod. Phys. 52, 29 (1980).

  9. E.S. Chang and A. Temkin, J. Phys. Soc. Jpn. 29, 172 (1970).

10. A. Temkin and K. V. Vasavada, Phys. Rev. 160, 190 (1967).

11. A. Temkin, K.V. Vasavada, E.S. Chang, and A. Silver, Phys. Rev.
186, 57 (1969).

12. M.A. Morrison, B.C. Saha, and T.L. Gibson, Phys. Rev. A 36,
3682 (1987).

13. S. Hara, J. Phys. Soc. Jpn. 27, 1262 (1969).

14. L.A. Collins and B.I. Schneider, Phys. Rev. A 24, 2387 (1981).

15. T.L. Gibson and M.A. Morrison, J. Phys. B 15, L221, (1982).

16. M.A. Morrison, T.L. Estle, and N.F. Lane, Quantum States of
Atoms, Molecules, and Solids (Prentice Hall, New Jersey, 1976).

17. A. Temkin, Phys. Rev. 107, 1004 (1957).

18. M.A. Morrison, Comput. Phys. Commun. 21, 63 (1980).

19. M.E. Rose, Elementary Theory of Angular Momentum (Wiley, New
York, 1957).

20. M.A. Morrison and G.A. Parker, Aust. J. Phys. 40, 465 (1987).

21. M.A. Morrison and L.A. Collins, Phys. Rev. A 17, 918 (1978).

22. J. Light and R.B. Walker, J. Chem. Phys. 65, 4272 (1976).





GRADUATE AWARD

Karen L. Butler, Ph.D. Candidate,
Department of Electrical Engineering, Howard University,
Washington, D. C.

Research Improvement in Minority Institutions Project
Principal Investigator, James A. Momoh

NEURAL NETWORK-BASED DETECTION AND IDENTIFICATION OF ARCING FAULTS
IN DELTA-DELTA CONNECTED POWER DISTRIBUTION SYSTEMS

Abstract: Several artificial neural network schemes10,12 exist for
detecting line faults in power distribution systems. They are
limited in their ability to detect the presence of the arcing
phenomenon or burst in signals and to provide security against
false alarms due to transients induced by normal system operations.

In this paper, the author presents a new clustering-based scheme
which detects the presence of arcing faults, and identifies the
fault type and the faulted phase. The paper discusses the
theoretical development, test procedures and demonstrations of the
network capabilities using laboratory-generated faulty signals.

Also a feature extraction preprocessing methodology is introduced
which computes phase current statistics to use as input to the
neural network.


INTRODUCTION

A.  Background

Electric power utilities continue to search for methods to
comprehensively detect line faults in a power distribution system.
Some success has been achieved in detecting the arcing type fault,
but further efforts are required to devise an automated, efficient
method which can detect and classify this intermittent phenomenon
in which the faulted current wave forms convey fault
characteristics that occur randomly in time and have an
unsymmetrical shape in each cycle.

Researchers have proposed several detection schemes over the past
ten years to detect line faults utilizing various detection
criteria. One of the most successful methods(sup)1 utilizes changes
in the computed energy content of the phase currents to detect the
presence of a line fault. This method is able to detect the fault
with high probability, but could achieve better performance if the
method were able to adapt to the environmental influences. Another
method uses changes in the amplitude of the second harmonic
component of the phase currents(sup)2 as a detection criteria. An
additional method(sup)3 uses changes in the magnitude of the third
harmonic component of the phase currents and the phase angle of the
third harmonic currents with respect to the system voltage to
detect the presence of arcing faults. These two methods have
demonstrated fairly effective detection rates but tend to have an
unacceptable amount of false alarms because various normal system
and load operations also induce changes in the second and third
order harmonics that make the phase currents appear fault-like.

Two other methods4,5 which have shown great results in detecting
line faults, use characteristics describing the system's unbalance
conveyed by statistics computed from various harmonic components of
the phase voltages and currents as decision criteria. The drawback
of these methods is that they tend to be computationally intensive.

In 1989, Kim and Russell(sup)6 concluded that even though various
techniques to detect high-impedance faults had been proposed and
much progress had been made, no reliable and flexible solution had
been found. Hence, researchers began to introduce expert
system-based methods to provide options not available with the
algorithmic-type methods discussed above. For example, expert
system approaches permit the use of larger databases of detection
parameters and allow the adjustment of decision criteria.

One expert system approach(sup)6 synthesizes multiple detection
parameters and a learning detection system to identify the system's
fault status. An additional expert system approach(sup)7 uses high
frequency components of the phase currents as detection parameters.
Another method(sup)8 uses experience compiled by dispatchers over
time to generate detection rules, and a third expert system
approach(sup)9 uses detection rules generated from results derived
by a simulation program which uses the symmetrical components
technique to model faults. The fundamental limitation of these
approaches is, however, that a priori rules must be formulated
which describe the behavior of the faults.

Since expert systems have not been able to provide the level of
adaptability needed to sufficiently detect the arcing faults,
researchers have begun to propose neural network approaches for
fault detection. One neural network-based method(sup)1(sup)0 has
been developed in which the neural network is configured and
trained with parameters developed from phase current parameters
derived using the Electromagnetic Transients Program
(EMTP).(sup)1(sup)1 In addition, a multilayer perceptron
method(sup)1(sup)2 has been developed which operates on the raw
data to detect the presence of arcing faults. The use of neural
network based schemes for arcing fault detection is only in its
early stages, but they show great promise in providing the
adaptability necessary to sufficiently detect arcing faults.

The goal of this paper is to provide preliminary results from the
application of a new neural network scheme for single line fault
detection in delta-delta connected distribution systems.

B.  Problem Definition

The major problem addressed in this work was defined by the Los
Angeles Department of Water and Power (LADWP) in their request for
an economical method which could detect and identify the four most
prevalent fault types that occur in their distribution systems. In
addition, the method must identify the faulted feeder and faulted
phase. A typical LADWP distribution station has approximately 4-5
delta-delta connected, 35kV/4.8kV transformer banks.

The four fault types to be identified are variations of a single
line fault and broken conductor faults. Fault type 1 is the classic
single line-to-ground type in which a grounded conductor or foreign
object touches one of the phase conductors. The second and third
fault types are broken conductor faults with different ends
touching ground. Fault type 4 is a broken conductor fault with
neither end touching the ground. The four fault types are
illustrated in Fig. 1.



Insert Figure 1

FIGURE 1

Four distribution fault types under study



In this paper, the proposed artificial neural network-based
approach detects and identifies the arcing phenomenon by exploiting
the ability of the neural network to perform well for this type of
system which is poorly understood and difficult to describe by a
set of rules or equations. The neural network was implemented using
a clustering analysis approach. During training, the neural network
partitions the decision space into regions representing the various
classes of normal and fault types with their respective faulted
phase. Then in the recall mode, it attempts to match the phase
current waveform input pattern with the most similar class in the
decision space.

In addition, a feature extraction scheme was developed as a
preprocessor methodology for the neural network. A statistically
derived set of features results from the preprocessor which is a
more robust approach than some of the heuristically derived
features that have been proposed in earlier works. The features
describe the magnitude and phase angle variations of the phase
current waveforms. After extracting these features from the phase
and neutral current measurements, they become the input set to the
neural network during training and normal pattern classification
operations.



MODEL OF DISTRIBUTION PHASE CURRENTS

In power distribution systems, a neutral and a set of phase
currents and a neutral and a set of phase voltages can be measured
at the substation for each feeder. These measurements for a
balanced, unfaulted network can be defined as shown in equations
(1) and (2) where ip(t) and vp(t) generally represent a selected
phase or neutral value of current and voltage. The fundamental
frequency, f, for these expressions is generally 60 Hertz for the
power systems in United States where ;gv equals 2pif. Also, imax,
vmax, ;gQV, and ;gQI are deterministic values. For balanced,
unfaulted systems, there is a 120o difference between values for
each of the phase currents, Ia, Ib, Ic. Likewise there is a 120o
difference between ;gQV values for each of the phase voltages, Va,
Vb, Vc.

[NOTE: SEE PRINTED VERSION FOR EQUATIONS]

(1)
ip(t) equals imax sin (wt + ;gQI)

(2)
vp(t) equals vmax sin (wt + ;gQV)

When a line fault disturbance occurs in the distribution system,
the faulted phase current waveform becomes distorted. Essentially,
the phase current waveform on which the fault has occurred can be
represented as shown in (3).

(3)
if'(t) equals i sin(wt + ;gQ)

The amplitude, i, and the phase, ;gQ, are random variables which
depend on the location and type of faults, which, of course, are
random in nature.

The faulted phase voltage waveform does not convey similar
distinguishable behavior as the faulted current waveform. Thus, to
efficiently perform the fault detection and identification, the
phase currents are analyzed. It is extremely difficult to derive
time domain characteristics from the faulted waveform that can
fully describe the arcing fault behavior. However, if a frequency
spectral analysis is performed on the faulted phase current data,
hidden periodicities or close spectral peaks can be revealed,
thereby making spectral characteristics better candidates for
describing the faulted behavior. Hence, features which convey
statistical information about the waveform's frequency spectrum
were used for fault classification.

The neural network module which performs fault identification is
composed of a filter, a preprocessor which computes statistics to
describe the characteristics of the phase currents, and a
clustering neural network pattern classifier.



PREPROCESSING

Before the data is input to the neural network, it is filtered to
remove the fundamental frequency component of the phase currents
that dominates the faulted characteristics. One of the fundamental
problems in pattern classification is to determine which features
should be employed for the best classification result. So the goal
of a feature extractor is to retain only a small number of "useful"
or "good" features from the input data necessary to ensure that
minimum computations, cost, and measurement complexity are required
to implement the pattern classifier. The filtering procedure and
feature extractor used as a preprocessor to the neural network are
described in the sections below.


A.  Filtering of Dominant Data

In earlier studies(sup)1(sup)3 performed by the author, the phase
current measurements were sent directly into the feature extractor.
However, researchers(sup)1(sup)4 have found that during arcing
fault activity, there may be little or no perceptual change in the
60 Hertz phase current components. By removing the dominant and
consistent 60 Hertz phase current component before the feature
extraction was performed, distinguishable characteristics in the
remaining data can more easily be recognized. Hence, a 60 Hertz
notch filter was placed before the feature extractor and then the
remaining data was amplified and analyzed. The system function,
H(z),(sup)1(sup)5 for the notch filter is shown in (4) which has
two complex-conjugate zeros, z1,2 equals e2]”+•j;gv2]0, on the unit
circle at angle w;i0, and two complex-conjugate poles at p1,2
equals re2]”+•j;gv2]0. Also, w;i0 is pi/3 and the pole radius, r,
is 0.95. The effect of the zeros is to create the null in the
frequency response of the filter at the frequency w;i0. The effect
of the poles is to introduce a resonance in the vicinity of the
null and thus to reduce the bandwidth of the notch.

(4)
H(z) equals 1 (minus) 2 cos;gv;i0z;s-(sup)1 + z;s-(sup)21 (minus)
2r cos;gv;i0z;s-(sup)1 + r(sup)2z;s-(sup)2

Fig. 2 shows the frequency response of the notch filter. It can be
observed that all data passes through except the 60 Hertz
component.


Insert Fig. 2
FIGURE 2
Frequency response of 60 Hertz notch filter


After the measurements were passed through the notch filter, the
filtered data was sent to a preprocessing unit. In the
preprocessing unit, statistics and energy variables were extracted
to form an input feature vector for input to the neural network.


B.  2nd Order Statistics: Reflection Coefficient Matrices

The objective of the feature extractor in this phase of the work
was to develop a method which can estimate the second order
statistics directly from the phase currents. In 1978,
Morf(sup)1(sup)6 showed that a sequence of reflection coefficient
matrices, estimated from a normalized Levinson, Wiggins, and
Robinson (LWR) algorithm(sup)1(sup)7 have a one-to-one
correspondence with positive definite covariance matrix
sequences--second order statistics. Consider, a discrete vector, X,
defined to represent an array of sensors placed at each feeder to
measure the phase currents and neutral current simultaneously. The
array is represented by a vector of four time series, each of
length K, where K represents the number of samples in an analysis
snapshot. For the time index t, the vector process is represented
by a 4 by 1 column vector xt, as shown in (5), where xt is a sample
of the filtered phase and neutral currents. Hence, a normalized
matrix reflection coefficient sequence is estimated directly from
the composite vector of measurements of phase currents.

(5)
-12]xt equals 14]4]xatxbtxctxnt

This work utilizes a 4-channel square-root normalized
LWR(sup)1(sup)8 implementation which can compute a 4x4 matrix,
2]6]=T6]Sx(f), which defines the maximum entropy estimate of the
composite power spectrum of the vector process. But, a single
sequence of the matrix-valued reflection coefficients, {K;i1, K;i2,
. . .}, which have singular values less than or equal to unity in
magnitude, are extracted before the method fully computes the power
spectrum.

This normalized LWR method was developed based on a maximum entropy
method, (MEM), developed by Burg(sup)1(sup)9 which avoids some of
the pitfalls of conventional linear spectral density estimation
techniques. The conventional linear techniques, which utilize the
Fourier transform within their procedure to obtain an estimate of
the spectral density function, convey information about the
distribution of the power among its frequency components.
Unfortunately these methods impose two assumptions about the nature
of the remaining data which usually results in some form of error
in the representation of the data in the frequency domain. First,
the methods assume that the rest of the data is zero through the
use of windowing functions. Second, the methods, through the use of
the Fourier Transform, assume that the rest of the data is
periodic. Burg avoids these "end-effect" problems by not making any
assumptions about the unmeasured data.

The fundamental question to be answered when implementing the LWR
method is to determine the required number of reflection
coefficients to achieve sufficient separability between classes to
make the pattern classification process practical. The
transformation from the current measurement time series to the
reflection coefficients is highly nonlinear and concentrates the
dominant information in the low-order coefficients. In general,
high-order coefficients contain not only less information but are
estimated using fewer data samples. Hence, their variance is
greater and their individual separability is quite low. They are
also computed based on residuals from earlier stages, and therefore
tend to be more sensitive to measurement noise. Finally, the choice
of the number of coefficients is problem specific based on the
number of classes to be defined and the complexity of those
classes.

Many of the classes required to describe the different types of
faults and normal situations are complex shapes that thread their
way through the feature space. During simulation studies, eight
reflection coefficients were found to be adequate to provide the
necessary separability.


C.  3rd and 4th Moments: Skewness and Kurtosis

In the computation of 2nd order statistics, phase relations between
frequency components are suppressed. Since a faulted current
waveform is not Gaussian, the analysis process must go beyond 2nd
order statistics to higher order statistics which give information
regarding the deviations from Gaussian distribution and the
presence of phase relations. Therefore, a 3rd and a 4th order
statistic(sup)2(sup)0 for each phase current was computed. In
general, the 3rd order statistic, skewness, gives a measure of the
asymmetry of a distribution around its mean. On the other hand, the
4th order statistic, kurtosis, gives a measure of how peaky or flat
a distribution is with respect to a Gaussian distribution.

Before the skewness and kurtosis coefficients are computed,
however, to reduce possible arbitrariness inherent in the samples,
the sample mean was subtracted from the magnitude of each sample to
make the result a zero mean value. That result was normalized by
the sample mean to define a new time series, r(k), as shown in (7)
where the magnitude of each sample of the filtered data, xmag(k),
was computed as shown in (6). The skewness was computed as shown in
(8) and the kurtosis was computed from the magnitude of the phase
and neutral current observations, {r(k)}, as shown in (9).

(6)
xmag(k) equals Re{x(k)}(sup)2 + Im{x(k)}(sup)2


where x(k) is the kth sample of filtered data, {k equals 0, 1, . .
., K (minus) 1}, and K is the total number of samples in a
snapshot.

(7)
r(k) equals xmag(k) (minus) E(xmag(k))E(xmag(k))


where {k equals 0, 1, . . ., K (minus) 1}, K is the total number of
samples in a snapshot, and E( ) denotes the statistical expectation
operator.

(8)
Skew equals 1K Kkequals1r(sup)3(k)


where {k equals 0, 1, . . ., K (minus) 1} and K is the total number
of samples in a snapshot.

(9)
Kurt equals 1K Kkequals1r(sup)4(k) (minus) 31K Kk equals
1r(sup)2(k)2


where {k equals 0, 1, . . ., K (minus) 1} and K is the total number
of samples in a snapshot.


D.  Signal Power Variables

There are two additional features which are computed to give
information about the signal's power for each phase. The first
feature is Pave which conveys information about the energy of the
signal; it was computed as shown in (10). Lastly, Pnvar was
computed to convey information about the variation of the amplitude
of the signal and is computed as shown in (11).

(10)
Pave equals 1K Ktequals1 -12]xt -12]x*t


where xt is the tth vector of X and K is the total number of
samples in a snapshot.

(11)
Pnvar equals E((xmag(k) (minus) -12]xmag(k))(sup)2)E(xmag(k))(sup)2


where xmag(k) is computed using expression (8), xmag(k) is the
sample mean of xmag(k), and E( ) denotes the statistical
expectation operator.


NEURAL NETWORK PATTERN CLASSIFIER

Many of the conventional methods mentioned in the introduction have
the inability to adjust to the dynamic nature of arcing faults
thereby making them somewhat ineffective in detecting the presence
of the faults and classifying the fault type. This fault
classification problem is basically one of recognizing the
distinguishing characteristic patterns of a particular fault type
when a fault has occurred.

Neural networks convey four major properties(sup)2(sup)1 that make
them ideal for detecting the dynamic characteristics of an arcing
fault. First, neural networks are adaptable in that they take data
and learn from it. Second, neural networks generalize in that they
can correctly process data they were trained on originally.
Similarly, they can handle imperfect or incomplete data, providing
a measure of fault tolerance. Their nonlinearities can capture
complex interactions among the input variables in a system.
Finally, neural networks are highly parallel interconnections of
processing units which perform identical, independent operations
that can be executed simultaneously.

There are many types of neural networks which can perform pattern
classification. In earlier work,(sup)1(sup)3 this author
demonstrated the validity of several of the input features, which
were discussed in section III, using a basic back-propagation
neural network. Since then, however, a clustering algorithm has
been implemented to serve as the pattern classifier. The feature
extractor in combination with this two-stage supervised clustering
method performs the neural network pattern classification in a
manner similar to a counterpropagation neural network.(sup)2(sup)2
A counterpropagation neural network is a two-layer, bi-directional
mapping network that combines two separate neural network paradigms
into one system: a Kohonen self-organizing map learning technique
and a Grossberg OutStar learning strategy.

The proposed neural network pattern classifier utilizes a two-stage
clustering-based learning algorithm. This algorithm uses the
training data to generate a set of clusters which represents the
various fault and normal classes existing in the feature space.

At the lower stage of the algorithm, an unsupervised clustering
methodology is used to train the neural network to generate a
natural set of clusters which partition the feature patterns into
decision regions within the feature space. Then the upper level,
which operates in a supervised mode, imposes the constraint that
all clusters must be homogeneous in that each cluster should
contain only feature patterns representative of the same membership
class. Utilizing these two strategies, the algorithm continuously
executes through the two stages until all patterns are placed in
homogeneous clusters.

Each of the resulting clusters, Ck, was defined by a center vector,
bj, a spherical radius, ;grj, the number of patterns contained in
the cluster, nj, the membership class of the cluster, classj, and
the cluster probability, probj. The metrics used to generate the
clusters were Euclidean distance and spherical radius. The
spherical radius was defined as the distance from the center of a
cluster to the outer boundary of the cluster, and Euclidean
distance was defined as the vector distance between two points.

In the testing/recall mode, inputs are presented to the neural
network which then attempts to find a class that matches the input
using the metrics of the network. If the neural network determines
that a match has been made, the output implies that (i) the minimum
distance between the pattern and the center of the selected cluster
is less than the spherical radius defined in the network, and (ii)
the label of the input pattern and the cluster are equal.

If, however, the neural network does not find a match, one of two
actions may be imposed on the system. The first possibility is that
the network should give up and state that a conclusion could not be
drawn. The second and more viable possibility is to utilize a
voting procedure to find the closest cluster using a weighting
scheme.


SIMULATIONS/RESULT

A.  Experimental Data

Experimental data generated during in-house low-voltage laboratory
experiments was utilized to train the neural network and test the
performance of the feature extraction procedure and the pattern
classification neural network. In the experiments, a
delta-connected transformer and a constant load were connected to
simulate a simple distribution system. Arcing and solid line faults
of each of the four fault types were staged. The phase current
waveforms were measured during the laboratory experiments, sampled,
and digitally stored on a PC-compatible computer using LABPC, a
hardware and software data acquisition system developed by National
Instruments.

Fig. 3 illustrates a phase current waveform measured during the
staging of a type 1 single line-to-ground arcing faults.

Insert Fig. 3FIGURE 3

Arcing fault phase current wave form

From this current waveform, it can be observed that the arcing is
depicted by the random spikes that occur over time. Hence, the
experimental results concur with the expected conjecture that the
arcing occurs randomly in time. It can also be observed that the
arcing persists only for a few cycles at a time. Whereas, for line
faults in grounded distribution systems at actual distribution
voltages, the arcing tends to persist for longer bursts One
possible explanation for this difference is that the arcing in
these laboratory staged faults were crisper and self-extinguishing
because there was no inductive reactance on the lines which would
inhibit quick transitions to zero valued line current.



B.  Simulation Studies

The feature extraction process reduced a snapshot of a 1041-point
raw experimental data input vector measured for a phase current to
an input features vector which has 156 points. A condensed
representation of the 156-point features vector computed during
preprocessing is shown in (12).

(12)
(K;i1 K;i2 K;i3 K;i4 K;i5 K;i6 K;i7 K;i8 Skew Kurt Pave Pnvar)


Ki represents the ith reflection coefficient matrix computed from
the phase and neutral currents composite spectrum; Skew represents
the four-point vector of skewness coefficients computed from the a,
b, and c phases and neutral currents; Kurt represents the
four-point vector of kurtosis coefficients computed from the a, b,
and c phases and neutral currents; Pave, represents the four-point
vector of average power values computed from the a, b, and c phases
and neutral currents; and Pnvar, represents the four-point vector
of normalized variance of the individual samples computed from the
a, b, and c phases and neutral currents.

Once the features extraction technique and the neural network
technique were integrated, studies were conducted to analyze the
performance of the overall arcing fault identification scheme.
Several issues must be determined. First, the most efficient order
of the presentation of training patterns must be derived through
experimentation. Also, the optimal features vector must be
determined through an exhaustive investigation of each feature's
impact on the neural network's performance. Then other enhancements
must be made until the neural network meets the desired
performance.

All evaluations of the neural network's training performance were
conducted using a three-fold cross-validation procedure which
yields a moderately pessimistically-biased error estimate.

The error estimate was computed as the total number of
misclassified samples divided by the total number of samples. The
average rate was computed as the sum of the error rates of each
block of evaluation data divided by the total number of blocks of
evaluation data. The average error rate represents the training
performance of the neural network for a particular training case.

One of the design parameters which was determined through a trial
and error procedure was the order of presentation of the training
data. Fig. 4 shows the average error rates for 24 different
permutations of orderings of the training data. The y-axis
represents the average error percentages and the represents the
number of the number of clusters. Also, it is noted on the x-axis,
in parenthesis, that four test cases resulted in 18 clusters, seven
cases resulted in 16 clusters, and 13 cases resulted in 17
clusters. The average error rate ranged from a low of 45.8 percent
to a high of 64.6 percent. From Fig. 4, it can be observed that the
larger amount of clusters generated the smallest error rates.
Hence, there appears to exist a tradeoff during the training
process between minimizing the number of clusters and minimizing
the error rate.


Insert Fig. 4FIGURE 4

Neural network training/evaluation results with various ordering of
input patterns

Utilizing the permutation of training data that conveyed the best
error rate using the least amount of clusters, the neural network
was trained. Table 1 shows the trained neural network's performance
when a set of test data never seen by the neural network was
presented to it. The second column of the table shows the desired
output for a given test case. A match was achieved if the
classification output of the neural network was the same as the
desired output. The neural network shows an average test error rate
of 0.25.

TABLE 1:

NEURAL NETWORK TESTING RESULTS


[NOTE: REFER TO PRINTED VERSION FOR TABLE 1]


Next, simulations were performed to validate the individual
necessity of each feature in the classification process. From that
study, the reflection coefficients 1 and 6 seemed to be the most
salient features. On the other hand, the reflection coefficients 2
and 3 seemed to be the least salient. However, the improvement or
reduction in the average error rate was only about 5 percent from
removing the least or most salient feature. This percentage change
is relatively insignificant which implies that all of the chosen
features are individually important.

Then, simulation studies were conducted to determine the
combinational impact of the features in the classification process.
In these test cases, the features that were determined not to be
individually salient in the previous simulation study were combined
together in various ways. Their joint saliency was determined. In
general, no combination was salient. The average error rate was
improved by as much as 10 percent when reflection coefficients 2
and 3 and average power were removed as features. One might
conclude that this combination of features should not be used in
the features vector, but further training using data from other
types of systems must be performed before that conclusion can be
drawn.

Using the ordering of features that gave the best performance in
Fig. 4 and a feature vector that included all features except the
reflection coefficients 2 and 3 and the average power, all training
data was presented to the neural network. The number of clusters
generated during this training session was greater than the number
generated when all features were used, but the error rate was
improved to 37.5 percent. Once again, one might note that there is
a tradeoff between minimizing the number of clusters and minimizing
the error rate.

Finally, a voting procedure was added to the matching phase of the
supervised learning algorithm. The voting scheme picks the
k-nearest clusters and allows each cluster to vote on the
classification of the input pattern. Then the decision of the
majority vote is assigned as the classification of the pattern.

Also, enhancements were made to the final set of clusters generated
by the supervised learning phase. The radius of the clusters were
tightened to the value of the distance of the farthest pattern from
the cluster center plus some ;ge;i1. In addition, clusters whose
centers were closer than some ;ge;i2 distance and of the same
membership class, were merged together.

Thus, the final simulation studies were performed to test the
implementation of a voting procedure and the enhancements to the
supervised learning. Fig. 5 shows the results of the studies. It
can be noted that the lowest error rate was still only the 37.5
percent found in the previous study. However, five clusters were
generated during this training procedure. This scenario seems to
imply that it is possible to minimize the number of clusters and
the average error rate simultaneously.

The training and test results have shown improvement with the
various enhancements. Since the three-fold cross-validation
procedure being used in the evaluation process is moderately
pessimistic, the authors hope to improve the average error rate to
the 15-20 percent range. Thus, further enhancements are being made
to reach that error rate.



Insert Fig. 5FIGURE 5

Neural network training/evaluation results utilizing enhanced
neural network



CONCLUSIONS

A new neural network based approach was presented for detecting and
identifying arcing fault types and faulted phase. The neural
network was implemented using an innovative combined supervised /
unsupervised clustering algorithm.

Also, a new feature extraction scheme was presented which computes
a set of 2nd, 3rd, and 4th order spectral statistics and power
signal strength features for use as inputs to the neural network.
The uniqueness of the statistics is that they are not
system-specific features. Also, the uniqueness of the use of
reflection coefficients as the 2nd order statistics is that they
are composite spectral statistics which convey information about
the interrelationship between the phase currents.

The results were encouraging with 75-100 percent success rates
having been achieved during during the various training evaluations
and tests performed using laboratory generated data.



ONGOING WORK

Additional enhancements are being made to the neural network to
further improve its performance. Most importantly, training
evaluations and tests will be performed utilizing data generated
during arcing field tests.



ACKNOWLEDGMENTS

The author would like to acknowledge NASA Jet Propulsion
Laboratory, the National Science Foundation RIMI grant
#HRD-9253055, and the Los Angeles Department of Water and Power for
their support of this research. The assistance of the researchers
at the Center for Energy Systems and Control at Howard University
is also acknowledged.



REFERENCES

  1. B. Don Russell, Ketan Mehta, and Ram Chinchali, "An Arcing
Fault Detection Technique Using Low Frequency Current
Components--Performance Evaluation Using Recorded Field Data," IEEE
Trans. on Power Delivery, Vol. 3, No. 4, Oct. 1988, pp. 1493-1500.

  2. C. Huang, H. Chu, M. Chen, "Algorithm Comparison for High
Impedance Fault Detection Based on Staged Fault Test," IEEE Trans.
on Power Delivery, Vol. 3, No. 4, Oct. 1988, pp. 1427-1434.

  3. P. I. Jeerings and J. R. Linders, "A Practical Protective
Relay for Down-Conductor Faults,"IEEE Trans. on Power Delivery,
Vol. 6, No. 2, April 1991, pp. 565-574.

  4. S. J. Balser, K. A. Clements, and D. J. Lawrence, "A
Microprocessor-Based Technique for Detection of High Impedance
Faults," IEEE Trans. on Power Delivery, Vol. PWRD-1, No. 3, July
1986.

  5. W. Kwon, G. Lee, et al., "High Impedance Fault Detection
Utilizing Incremental Variance of Normalized Even Order Harmonic
Power," IEEE Trans. on Power Delivery, Vol. 6, No. 2, April 1991,
pp. 557-564.

  6. C. J. Kim and B. D. Russell, "A Learning Method for Use in
Intelligent Computer Relays for High Impedance Faults," IEEE Trans.
on Power Delivery, Vol. 6, No. 1, Jan. 1991, pp. 109-115.

  7. B. D. Russell and K. Watson, "Power Substation Automation
Using a Knowledge Based System -- Justification and Preliminary
Field Experiments," IEEE on Trans. on Power Delivery, Vol. 2, No.
4, Oct. 1987, pp. 1090-1097.

  8. Yuan-Yik Hsu, F. C. Lu, and et al., "An Expert System for
Locating Distribution System Faults," IEEE Trans. on Power
Delivery, Vol. 6, No. 1, Jan. 1991, pp. 366-371.

  9. J. A. Momoh et al., "Rule-Based Decision Support for
Single-line Faults in a Delta-Delta Connected Distribution System,"
presented at the Power Industry Computer Application Conference,
May, 1993.

10. S. Ebron, D. Lubkeman, and M. White, "A Neural Network Approach
to the Detection of Incipient Faults on Power Distribution
Feeders," IEEE Trans. on Power Delivery, Vol. 5, No. 2, April 1990,
pp. 905-914.

11. "EMTP Book--Version 1.0," EPRI Report EL-4541, Vols. 1-2,
prepared by Systems Control, Inc., 1986.

12. A. F. Sultan, G. W. Swift, and D. J. Fedirchuk, "Detection of
High Impedance Arcing Faults Using a Multi-Layer Perceptron," IEEE
Trans. on Power Delivery, Vol. 7, No. 4, Oct. 1992, pp. 1871-1877.

13. K. L. Butler and J. A. Momoh, "Robust Features Selection Scheme
for Fault Diagnosis in an Electric Power Distribution System,"
presented at the Canadian Conference on Computer and Electrical
Engineering, Sept. 1993.

14. M. Aucoin and B. D. Russell, "Detection of Distribution High
Impedance Faults Using Burst Noise Signals near 60 Hz," IEEE Trans.
on Power Delivery, Vol PWRD-2, No. 2, April 1987, pp. 342-348.

15. J. Proakis and D. Manolakis, Introduction to Digital Signal
Processing, New York: Macmillan, 1988.

16. M. Morf, et al, "Covariance Characterization by Partial
Autocorrelation Matrices," Annals of Statistics, May 1978.

17. M. Morf, A. Vieira, D. T. Lee, and T. Kailath, "Recursive
Multichannel Maximum Entropy Spectral Estimation," IEEE Trans. on
Geoscience Electronics, vol. GE-16, no. 2, April 1978, pp. 85-94.

18. R. A. Wiggins and E. A. Robinson, "Recursive Solution to the
Multichannel Filtering Problem," J. Geophs. Res., vol. 70, pp.
1885-1891, 1965.

19. J. Burg, "A New Analysis Technique for Time Series Data," NATO
Advanced Study Institute on Signal Processing with Emphasis on
Underwater Acoustics, Aug. 12-23, 1968.

20. A. M. Mood et al, Introduction to the Theory of Statistics, New
York: McGraw-Hill, 1974.

21. D. Hammerstrom, "Neural Networks at Work," IEEE Spectrum, June
1993, pp. 26-32.

22. R. Hecht-Nielsen, "Counter-Propagation Networks," IEEE First
International Conference on Neural Networks, vol II, 1987, pp.
19-32.



SYMPOSIUM A: ALLIANCES FOR MINORITY PARTICIPATION (AMP)

Moderator
Yvonne Harris
Tuskegee University
Department of Biology, Sophmore

Presenters

Fitzgerald B. Bramwell, Ph.D.
Professor
Dean of Graduate Studies and Research
Brooklyn College, City University of New York


CRYSTAL AND MOLECULAR STRUCTURES OF THE ADDUCTS OF TRI-p-TOLYLTIN
CHLORIDE, BROMIDE AND IODIDE WITH 4,4'-BIPYRIDINE

The structures of the three adducts
((p-toly);i3SnX);i2-4,4'-bipyridine (X equals Cl, Br, I), have been
established by X-ray diffraction techniques. The complexes are
monoclinic, space group P2;i1/c and Z equals 2. The cell parameters
are: a equals 13.262(2), b equals 11.382(2), c equals 17.029(2)
;apA and ;gb equals 110.050(11)(sup)0 for X equals Cl; a equals
13.10(3), b equals 11.351(2), c equals 17.607(4) ;apA and ;gb
equals 110.050(16)o for X equals Br; a equals 12.867(3), b equals
11.389(4), c equals 18.496(4) ;apA and ;gb equals 110.050(11)(sup)0
for X equals I. The complexes are binuclear with a bridging
4,4'-bipyridine. The tin is in a trigonal bipyramidal environment
with the p-tolyl groups on the equator and the X atom and an N of
the bipyridine on the axis. The Sn-N bond distances are 2.668(3),
2.653(3) and 2.655(7) ;apA, for X equals Cl, Br and I,
respectively. The Sn-X bond distances are 2.452(2), 2.691(1) and
2.830(1) ;apA,. for X equals Cl, Br, and I, respectively.

Carmen S. Menoni, Ph.D.
Assistant Professor
Department of Electrical Engineering
Colorado State University

CHARACTERIZATION OF SEMICONDUCTOR LASERS AND MATERIALS USING
ULTRAHIGH PRESSURE

The engineering and design of efficient semiconductor lasers
requires a knowledge of the band structure and band alignment of
the heterostructure materials composing the device. The relative
band alignment between the active and cladding layer as well as the
separation between the different energy bands in semiconductor
materials results in the presence of different loss mechanisms
which deteriorate the laser performance. Results are presented on
the identification of the dominant loss mechanisms in 1.3 ;gmm
semiconductor lasers from pressure dependent measurements of the
laser threshold current, and differential quantum efficiency. The
use of high hydrostatic pressure to study the behavior of
semiconductor lasers offers the unique possibility of modifying the
band structure in a single device in a way which resembles the
changes that take place when the band structure is varied by
varying the composition. Optical characterization of the
modifications of the band structure of the active and cladding
layers with pressure also allows the determination of the
heterostructure band offsets which control the leakage currents.

Eloy Rodriguez, Ph.D.
Professor
Department of Developmental and Cell Biology
University of California at Irvine

CHEMISTRY OF AMAZONIAN AND AFRICAN MEDICINAL PLANTS

USED BY WILD ANIMALS

In recent years investigators at Harvard University and the
University of California, Irvine have elucidated the chemical
structures of various products present in tropical rainforest
plants used by nonhuman primates for medicinal purposes. The
structures range from simple dithiopolyacetylenes to complex
hexapolycyclic peptides. Investigation of wild monkeys, birds, and
other animals in Amazon and Atlantic rainforest suggests a wider
array of bioactive constituents than other tropical rainforests of
the world. In this presentation, the chemistry of
bioactive/medicinal constituents from the Amazon, Atlantic
rainforest and African rainforest are presented and possible modes
of action were discussed.

R.H. Sullivan, Ph.D.
Professor
Department of Chemistry
Jackson State University

Ab Initio STUDY OF THE STRUCTURE AND ENERGIES OF BIURET AND ITS
SULFUR ANALOGS

The calculations for biuret (B) and its two sulfur analogs,
thiobiuret (TB) and dithiobiuret (DTB), were carried out using the
ab initio LCAO-MO method at the Hartree-Fock level. The molecular
structure of B was fully optimized using standard 3-21G and 6-31G
basis sets, while geometry optimizations of the TB's and DTB's were
performed at the 3-21G* and 6-31G* levels. Only trans isomers were
found to be minimum structures on the HF/3-21G* potential energy
surfaces, whereas cis forms are the first-order transition
structures. The relative energies of the cis conformers at
MP2/6-31G**//HF/6-31G* + 0.9 Zero-point energy (ZPE) level are
equal to 47.4, 51.7, and 59.7 kJ mol(minus)1 for TB, B, and DTB,
respectively. Because of relatively large dipole moments for the
cis conformers are predicted to be more stabilized than the
trans-conformers in polar solvents.





SYMPOSIUM B: RESEARCH CAREERS FOR MINORITY SCHOLARS (RCMS)

Moderator
Joseph Rascon
University of Texas at El Paso
Department of Electrical Engineering, Senior

Presenters

D. Bagayoko, Ph.D.
Professor
Timbuktu Academy
Department of Physics
Southern University and A&M College

MAGNETISM AND Al;i1;i8Fe

Until recently, 3d metal impurities in aluminum were believed to
have non-magnetic ground states described by the Anderson model.
The recent experimental results of Hauser et al., and the
theoretical ones of Bagayoko et al., established the existence of
a local moment, between 1 and 2 Bhor magnetons, on substitutional
Mn impurities in fcc aluminum at an atomic concentration of five
percent. Even though experiments have yet to detect a local moment
on dilute iron purities in Aluminum, some earlier theoretical
studies reported large moments on this impurity in aluminum. In
particular, Deutz et al., found a local moment of 1.75 Bhor
magnetons on a substitutional Fe impurity in fcc aluminum.
Guenzburger and Ellis recently calculated smaller local moments on
iron impurities in fcc aluminum (ie. 0.96 Bhor magneton for a
lattice constant around 7.635 a.u.) that vanish when relaxation is
taken into account.

Sheppard, a scholar of the RCMS-Timbuktu Academy, reported a
Kondo-like behavior of Al;i1;i8Fe for lattice constants between
7.035 and 8.235 a.u. The local moments on Fe range from 0.04 to
0.69 Bohr magnetons respectively, while the cluster moments remain
identically zero. Magnetic moments smaller than 0.1 may be due to
imperfections of the Mulliken population analysis that is utilized
to compute the moment; the ones over 0.2 are believed to represent
actual moments on the impurity. The above exact compensation of the
local moment on the Fe impurity by the polarization of the
surrounding is the essence of the Kondo effect. The decrease of the
local moment with the lattice constant, expected from the
corresponding increase of the interatomic Coulomb and exchange
interactions as the lattice constant decreases, agrees with the
findings of Guenzburger and Ellis.

Omnia El-Hakim, Ph.D.
Associate Professor
College of Engineering

Fort Lewis College & Colorado State Univeresity

EDUCATION, RESEARCH AND TRAINING MODEL FOR NATIVE AMERICANS

Regions facing problems in management of irrigated agriculture
include the Indian reservations of the western United States. A
system sharing many of the cultural distinctions of the
reservations and enjoying a measure of economic success is the
Navajo Agricultural Products Industry (NAPI), south of Farmington,
New Mexico.

This paper describes an effective working model (Minority
Engineering and Science Project) which utilizes an existing
commercial farm such as NAPI to achieve the following objectives in
the education of Native American students:

o    Increase the understanding and effectiveness of
     problem-solving, information processing, and decision-making
     analysis related to water,land, human resources, and
     environmental protection for irrigated agricultural systems;

o    Train and encourage minority students to complete
     undergraduate work in scienced and engineering and pursue
     graduate careers in the development and application of
     technology to improve resource management;

o    Implement a teamwork learning environment that includes
     undergraduate students, graduate students, faculty, community
     members, and professionals from an outside corporation (NAPI);
     and

o    Implement outreach summer enrichment programs and conferences
     for high school and community college students to motivate and
     interest them in science, math and engineering.

J.D. Fan, Ph.D. and Y.M. Malozovsky, Ph.D.
Department of Physics
Southern University and A&M College

MECHANISM OF HIGH TEMPERATURE SUPERCONDUCTIVITY IN COPPER-OXIDE
(Cu-O) CONDUCTORS

We present an electron pair interaction potential with a negative
part within a certain range of action due to electron (hole)
correlations in a layered two-dimensional (2D) Cu-O) compounds.
Based on this potential, we have obtained a diagram of the phase
transitions from a normal metal to a superconductor and then from
a superconductor to an insulator, and found a smooth crossover of
temperature in transition from the weak coupling (Cooper-paired),
Tc ;sl;gvcexp((minus)1/;gleff), to the strong coupling, Tc;slns/m*,
with the decrease of the carriers (doping) density. Both
expressions of Tc have been verified by experiments. It is
therefore concluded that the dependence of transition temperature
on doping in Cu-O compounds is related to the pairing mechanism
caused by electron correlations in the layered 2D structure.

Karan Watson, Ph.D.
Associate Professor
College of Engineering
Texas A&M University

NEURAL NETWORKS APPLIED FOR HIF DETECTION

Artificial Neural Network (ANN) research is an attempt to achieve
pattern recognition by utilizing the computing techniques used by
natural neural networks in the brain. The High Impedance Faulty
(HIF) detection problem shows many characteristics of a complex
pattern recognition task. Thus ANN techniques can play an important
role in finding a solution for this. High Impedance Faults (HIFs)
are a type of fault that can be seen in power distribution feeders.
These faults do not draw sufficient fault current to be detected by
conventional protective devices such as current relays.
Nevertheless, these faults are hazardous to property and to the
public. The architecture of a neural network has a significant
influence on the type of features extracted by the network. Hence,
it is necessary to select appropriate ANN architectures and
topologies to extract useful features from the problem domain. This
is consistent with what has been found in the human brain. The
objective of this research was to study the performances of three
ANN architectures, namely Multi-Layer Perceptron (MLP), Time-Delay
Neural Network (TDNN) and Simple Recurrent Network (SRN), to
extract features important to HIF detection. Another goal was the
investigation for the features focused upon by these networks by
analyzing the internal encodings of the networks.



SYMPOSIUM C: RESEARCH IMPROVEMENT IN MINORITY INSTITUTIONS (RIMI)

Moderator
Arturo Bronson
NSF Program Director
RIMI

Presenters

Mukul C. Datta, Ph.D.
Professor
Department of Chemistry
Tuskegee University
Tuskegee, AL

STUDIES ON THE MECHANISM OF REVERSING HEMOGLOBIN PROPORTIONS IN
ADULT RATS

The objective of the research was to study the importance of
concurrent prostaglandin synthesis during changes in hemoglobin
proportions toward newborn values in adult rats. The rat has no
specifically fetal hemoglobin like humans, but rats and humans
share a number of similarities in hemoglobin switching during
ontogeny and differentiation. Knowledge of control mechanism
affecting the proportions of hemoglobin components in rats would be
useful for pharmacologic manipulation of fetal hemoglobin synthesis
for sickle cell anemia and thalassemia. The effects of two chemical
agents (5-azacytidine and hydroxyurea) were found to change adult
hemoglobin proportions to newborn proportions in adult rats.
Results revealed that both the chemical agents and the
bleeding-induced acute anemia were independently capable of
switching hemoglobin components toward newborn values. However, the
switching effect of these drugs and of the bleeding-induced anemia
were totally lost if aspirin was simultaneously administered to the
rats, reflecting the need for concurrent prostaglandin synthesis.

M.A. Ebadian, Ph.D.
Professor and Chairman
Department of Mechanical Engineering
Florida International University

THE EFFECT OF SURFACE TENSION ON DOUBLE DIFFUSIVE CONVECTION IN A
TRAPEZOIDAL ENCLOSURE

A numerical study of double diffusive convection was conducted for
a trapezoidal enclosure with aspect ratios of Ar equals 2. The
solutions were obtained by solving the conservation equations of
mass, momentum, energy, and concentration of heavier species by
applying the boundary-fitted coordinate system. The buoyancy forces
were approximated by Boussinesq assumptions. Prandtl and Lewis
number of 7 and 100, respectively, were used to simulate the double
diffusive convection of salt water. The effects of the Marangoni
number in the range of 0-10 on the flow, heat, and mass transfer
were also investigated. The surface tension at the free surface
drives the counterclockwise vortex flow located over a clockwise
unicell flow for pure thermal convection. The greater the Marangoni
number, the stronger the flow at the top, with a greater portion of
the trapezoidal enclosure being occupied. Therefore, the flow
induced by buoyancy forces is pushed down by the flow driven by
surface tension. The local Nusselt and Sherwood number
distributions on the left and right sidewalls are enhanced by the
Marangoni convection, which are presented in the present study. The
significant influence of the Marangoni number is observed at the
top of the enclosure.

Ishrat M. Khan, Ph.D.
Professor
Department of Chemistry
Clark Atlanta University
Atlanta, GA

MIXED (IONIC AND ELECTRONIC) AND IONIC CONDUCTIVE POLYMERS

The preparation, characterization, and properties of diblock
copolymers containing an electronic and an ionic conductive block
were studied. Poly (;gv-methoxyocta-(oxyethylene)
methacrylate-block-4-vinylpyridine), abbreviated as P(MG8-4VP),
diblock copolymers were prepared by the anionic living
polymerization technique. In these diblock copolymers, the
oligooxyethylene blocks were doped with LiClO;i4 to obtain an ionic
conductive phase, and the 4-vinylpyridine blocks (after
quaternization) were doped with tetracyanoquinodimethane to obtain
an electronic conductive phase. The electronic and ionic
conductivities of the compositions have P(3MT-4VP) block copolymers
with appropriately doped compositions have bulk electronic
conductivities as high as 1 x 10(minus)3 S cm(minus)1 and bulk
ionic conductivities as high as 6.6 x 10(minus)7 S cm(minus)1 at
30oC. Scanning electron microscopic (SEM) examination of the
internal structure of the blends shows the presence of a two phase
microstructure, most likely, stabilized by the emulsifying effect
of the salt.

Hassan Mahfux, Ph.D.
Professor
Department of Mechanical Engineering
Materials Research Laboratory
Tuskegee University
Tuskegee, AL

TENSILE AND FLEXURAL RESPONSE OF HIGH TEMPERATURE COMPOSITES: A
COMBINED EXPERIMENTAL AND FINITE ELEMENT STUDY

The fracture and flexural behavior of monolithic SiC and
SiC-whisker reinforced SiC composites (SICw/SiC) were investigated
at 23 oC, 800 oC and 1200 oC. Flexural and fracture tests were
conducted in a four-point beam configuration to study the effects
of whisker reinforcements on the mechanical properties and thermal
stability. Post failure analyses with the scanning electron
microscope (SEM) found the formation of a glass phase at the
whisker/matrix interface. The crack growth was also shifted from
intergranular to transgranular with the rise in temperature.
Ultrasonic velocity measurements have been performed through the
thickness of the untested and fractured specimen, and the variation
in the sonic velocities are discussed. Combined experimental and
finite element analyses were conducted to determine the failure
mechanisms in SiC coated Carbon-Carbon (C/C) composites at room and
elevated temperatures under flexural loading. Large deformation
behavior considered in the finite element analysis corroborate the
experimental findings, and a comparison in respect to displacement
and stress-strain behavior test the accuracy of the finite element
analysis (FEA).

Hylton G. McWhinney, Ph.D.
Professor
Department of Chemistry
Prairie View A&M University
Prairie View, TX

DIFFUSION OF LITHIUM (6) ISOTOPE IN LITHIUM ALUMINATE CERAMICS
USING NEUTRON DEPTH PROFILING

Lithium aluminate ceramics offer significant potential in
contributing to the production of tritium ((sup)3H) for fusion
power reactors, although the application of the ceramics will
depend greatly on the diffusion properties of the Li(6) within the
matrix. Consequently, diffusion of Li(6) in the aluminates is an
important requirement in the continued development of the
technology. In this investigation, the neutron depth profile (NDP)
technique was applied to the study of concentration profiles of
Li(6) in lithium aluminates, doped with 1.8%, 50%, and 95% Li(6)
isotopic concentrations. Specimen for analysis were prepared at
Battelle (PNL) as pellet discs. Samples for diffusion studies were
arranged as diffusion couples consisting of 1.8% Li(6) disc/95%
Li(6) powder. Experiments were performed at the Texas A&M Nuclear
Science Center, utilizing 1MW equivalent thermal neutron fluxes
(3.0 x 1012 n/s). The depth probed by the technique is
approximately 15;gmm. Diffusion coefficients were determined to
range from 2.1 x 10(minus)12 m(sup)2s(minus)1 to 7.0 x 10(minus)11
m(sup)2s(minus)1 for 1.8% Li(6) doped ceramics annealed at 1200 oC
and 1400 oC, respectively.

Mohamed A. Seif, Ph.D.
Professor
Department of Mechanical Engineering
Tuskegee University
Tuskegee, AL

FAST DETECTION OF FLAWS AND RESIDUAL STRESSES IN COMPOSITE
MATERIALS BY A REAL-TIME NONDESTRUCTIVE LASER TECHNIQUE

The pre-existing flaws and/or stress concentrations, often caused
by the fabrication and/or installation process, induce
microcracking and residual stresses in composite materials. The
application of repeated loads or a combination of loads and
in-service environmental conditions (humidity and temperature) can
cause microcracking with time. An automated non-destructive
evaluation (NDE) technique using laser speckle shearing
interferometry is capable of rapidly determining the initial
development of microcracking and its subsequent propagation.
Moreover, a laser speckle technique was employed to investigate the
sliding contact frictional problem, to determine residual stresses
in plastic pipes, and to measure deformation and warping in
magnetic tapes subjected to tension.




SYMPOSIUM D: MINORITY RESEARCH CENTERS OF EXCELLENCE (MRCE)


Multidisciplinary Research in Science

Moderator
A. Ramona Brown
Executive Director
National Alliance of Research Centers of Excellence (NARCE)

Presenters

Warren W. Buck, Ph.D.
Professor of Physics
and Director
Nuclear/High Energy Physics
Research Center of Excellence
Hampton University

THE RESEARCH PROGRAM OF THE NUCLEAR/HIGH ENERGY PHYSICS (NuHEP)
RESEARCH CENTER OF EXCELLENCE

The Experimental and Theoretical Programs in Nuclear/High Energy
Physics will be presented. These programs include topics such as:
Electromagnetic Form Factors, Hypernuclear Spectroscopy, Knock Out
Reactions, Spin Structure and Covariant Formalism. Also, explicit
connections to research at the Continuous Electron Beam Accelerator
Facility (CEBAF) are presented.

Juan G. Gonzalez, Ph.D.
Professor of Oceanography
and Director
Center of Excellence in Tropical & Caribbean Research
University of Puerto Rico at Mayaguez

THE PUERTO RICO MINORITY RESEARCH CENTER OF EXCELLENCE IN TROPICAL
& CARIBBEAN RESEARCH

The focus of the Puerto Rico Center of Excellence is on Tropical
Caribbean Research. Ours is, therefore, a diverse center which
includes four distinct research components including: Marine
Natural Products, Terrestrial Ecology, Caribbean Geology, and
Natural Hazards Mitigation.

The focus of the Marine Natural Products Division is the
investigation of cuguatoxins in fish, red algae as response
regulators of mallalian inflammation and in antibiotic compounds,
and sponges as anti-tumor agents. The Terrestrial Ecology Division
has been researching factors controlling primary productivity along
an altitudinal gradient of tropical forests. The Caribbean Geology
Division is studying the effects of the Atlantic plate sliding
beneath the Caribbean plate, as well as, the formation of various
Caribbean islands. Finally, the Natural Hazards Mitigation Division
concentrates its efforts on alarm systems for flash flooding,the
vulnerability of wood structures to extreme winds, strong motion
instrumentation for the determination of damage caused by
earthquakes to various structures, and the display and referencing
of data for natural hazards and environmental studies through the
use of Remote Sensing.

M.F. Lima, Ph.D.
Professor
Department of Microbiology
Meharry Medical College

EPIDERMAL GROWTH FACTOR MODULATES GROWTH OF TRYPANOSOMA CRUZI

The mechanisms controlling the establishment and intracellular
multiplication of Trypanosoma cruzi are not well defined. Most
factors have been shown to participate in this process by allowing
the establishment of infection or delivering nutritional components
necessary for parasite growth intracellularly. In this work, we
have found evidence that Epidermal Growth Factor (EGF) plays a role
in the growth of T. cruzi axenically or inside host cells.

We found that an addition of physiological concentrations of EFG
increase amastigote growth as evidenced by an increase in
(sup)3H-thymidine incorporation and number of parasites.
Amastigotes were also found in increased numbers when T.
cruzi-infected Vero cells were supplemented with EFG after an
infection was established. Presence of receptors for EFG on
amastigotes was investigated by immunofluorescence and
ligand-binding studies. Amastigotes specifically bind EFG in a
ligand receptor interaction manner. Scatchard analysis indicates
that amastigotes have one type of receptors for EGF present at 1 x
10(sup)3 receptors per parasite with a Kd of 2.0 nM. The presence
of receptors of EFG in T. cruzi is restricted to the amastigote
dividing form of the parasite.

Alfred Z. Msezane, Ph.D.
Professor of Physics
and Director
Center for Theoretical Studies of Physical Systems
Clark Atlanta University

INTERDISCIPLINARY RESEARCH IN PHYSICS, CHEMISTRY AND MATHEMATICS

The Center for Theoretical Studies of Physical Systems (CTSPS), one
of the two most recent NSF Minority Research Centers of Excellence,
has made considerable progress in the understanding of important
problems in physics, chemistry, and mathematics. Research at CTSPS
is carried out through dynamic interdisciplinary research clusters.
Below are some of their accomplishments.

Atomic and Molecular Theory. Rigorous bounds on the momentum
transfer squared have been used to construct, for the first time,
a universal function to extrapolate the GOS to the optical
oscillator strength and to prove, unambiguously, the Lassettre
Limit Theorem regardless of the energy impact. Also, new methods
have been developed to calculate optical oscillator strengths from
differential cross sections, using Type II Pade Approximants to
numerically calculate the charge transfer matrix elements,
important in the understanding of charge exchange processes.
Extensive C1 target wave functions have been used to obtain both
integral and differential cross sections for
photon/electron-atom/ion interactions.

Condensed Matter. In collaboration with the condensed matter group
at Ames Laboratory, we have published a series of papers concerning
the stability of large fullerences such as C76 and C84. We have
also studied the vibrational spectrum of C60, which is useful for
understanding the superconductivity mechanism in alkali-metal doped
fullerides. In the study of superfluid-insulator transition, we
have proposed that the universal behavior in two-dimensional
superfluid-insulator transitions is closely connected with the
universal behavior of the interger quantum Hall system, which for
the first time provides a consistent description of experimentally
observed universal critic resistance and associated critical
behavior. Using molecular dynamics simulation, we have studied the
structure and dynamics of several reconstructed surfaces of noble
metals. This kind of large-scale molecular dynamics study yielded
important information on the nature of complicated structural
phases and the anomalous surface vibrations.

Chemistry. The understanding of the structural evolution of ceramic
precursors leading to ceramic materials is of tremendous interest
for the improvement of the processing, technology of ceramic
materials via polymer pyrolysis. In light of this interest, we have
investigated recently a family of polymeric precursors for SiNC
ceramics by CP/Nas and 1R spectroscopy, and have shown that the
standard evolution involves SiC;i2 N;i2, SiHN;i2C, SiCN;i2
tetrahedra, and a free carbon phase as SiCN ceramics are formed by
chemical transformations. This effort is one of the few recent
studies in this field.

Mathematical Physics. The development of effective nonperturbative
methods for calculating accurate quantum eigenergies of singular
systems continues through recent refinements in the EMM theory. The
analysis of the quantum three body problem, particularly for the
case of co-equal masses, is underway.

Signal/Image Processing. Important applications on the use of
continuous, multidimensional, wavelet transforms for studying fluid
turbulence, and image processing are being developed.

Applied Mathematics. A significant and unprecedented achievement in
non-autonomous control problems has been the completion of the
bound state problem for hereditary control problems. In the area of
noise filtering analysis, new Type II Pade Approximants have been
developed and are currently being tested.

Emmanuel S. Onaivi, Ph.D.
Assistant Professor
Department of Pharmacology
Meharry Medical College

ARE THERE SUBTYPES OF MARIJUANA RECEPTORS?

For many decades research on the molecular and neurobiological
basis of the myriad of effects of marijuana use, was hampered by
the lack of specific research tools and technology. The situation
has started to change with the introduction and availability of
molecular probes and other recombinant molecules. Marijuana and its
psychoactive cannabinoids influence the CNS probability through the
cloned marijuana receptor found in the rat and human. Recently,
anandamide, an endogenous compound that interacts specifically with
the cannabinoid receptors and inhibits adenylate cyclase was
isolated from the porcine brain. In the present studies a battery
of behavioral effects of ;gD(sup)9-THC, the major psychoactive
constituent in marijuana were studied in C57/BL6, DBA/2 and ICR
mouse strains following intraperitoneal administration. The data
obtained indicated that some behavioral changes following
;gD(sup)9-THC administration may be attributed to genetic
differences. Furthermore, in ICR mice stereotaxic surgical
approaches were utilized for the microinjection of ;gD(sup)9-THC
and other analogs in the nucleus accumbens (ACB) and the central
nucleus of amygdala (ACE). The results of the behavioral assessment
demonstrated that the mesolimbic ACB, but not ACE, may be an
important brain site contributing to the modification of motor
activity and induction of catalepsy by some cannabinoids.

Putcha Venkateswarlu, Ph.D.
Professor
Department of Physics
Alabama A&M University

ENERGY UPCONVERSION FLUORSCENCE IN LAf;i3 DOPED WITH Nd, Pr and Eu
IONS

Two ions, A and B excited to energy levels ;gn;i1 and ;gn;i2 may
cooperatively exchange their energies in such a way that A goes up
to energy level at ;gn;i1 + ;gn;i2 while B relaxes to the ground
state. Fluorescence at frequency (;gn;i1 + ;gn;i2)cm (minus) 1 may
be observed. This upconversion process enables one to convert
infrared radiation to the visible and the visible radiation to
ultraviolet. The upconverted fluoresconance using suitable optical
resonator arrangement will give laser action.

Infrared (1R) upconverted fluroscence has been observed in
LaF;i3:Nd3+ and LaF;i3:Pr3+ single crystals. By exciting with a
Ti:Sapphire laser at 806nm, upconverted fluorescence was observed
in the region 360-760 nm in both the crystals. It was seen that in
LaF;i3:Nd3+, the level S at 12,569 cm(minus)1 is first reached,
then the ion relaxes to the R level at 11,592 cm(minus)1 from where
the second pump photon takes the ion to the I level at 23,470
cm(minus)1 associated with a suitable phonon level. From there the
excited ion relaxes to the E level at 19,152 cm(minus)1 which gives
strong fluorescence at 657 and 589 nm corresponding to the
transition E-X and E-Y respectively. In LaF;i3:Pr3+ crystal a
stepwise two photon process along with suitable relaxation take the
ion to the level (sup)3PO at 20,927 cm(minus)1. From here the ion
goes down to the (sup)3H;i5 level giving strong fluorescence at 540
nm. These fluorescence transitions in Na3+ and Pr3+ could very
likely give lasting action with suitable optical resonator
arrangement.

Upconverted fluorescence has also been observed in LaF;i3:Eu3+ in
the region 415.6-558.4 nm by exciting with Ar+ laser pumped dye
laser at 575 nm (17,380 cm(minus)1) and 595 nm (16,800 cm(minus)1).
It is found that the (sup)5D;i0 level at 17,293 cm(minus)1 is first
reached by excitation with the first photon, and by a simultaneous
excitiontion with a second photon along with suitable relaxation
the levels (sup)5D;i4, (sup)5D;i3, (sup)5D;i2 and (sup)5D;i1 are
reached giving upconverted fluorescence. Among these the group
involving the transition (sup)55D;i2;ul(sup)7F;i3 being the
strongest and most suitable for laser action. The regular stokes
fluorescence from (sup)5D;i0 is of course strong.




SYMPOSIUM E: MINORITY RESEARCH CENTERS OF EXCELLENCE (MRCE)

Advanced Materials Research
Moderator
Samuel P. Williamson
NSF Program Director
MRCE

Presenters

Daniel Akins, Ph.D.
Professor of Chemistry
and Director
Center for Analysis of Structures and Interfaces (CASI)
Department of Chemistry
The City College of New York

AGGREGATION OF WATER-SOLUBLE TETRA-KIS-(p-SULFONATOPHENYL)
PORPHYRIN

Aggregation of tetra-kis-(p-sulfonatopnenyl) (TSPP), a typical
water-soluble porphyrin, has been studied by a combination of
optical absorption and Raman measurements. The aggregation of TSPP
in aqqueous solution is confirmed by the appearance of a sharp
absorption at 490 nm accompanied by a broader band at 422 nm. The
optical absorption measurement reveals also that aggregation takes
place only in solution at pH less than 5 and can be promoted by
increasing ionic strength. Temporal, as well as, ionic strength
dependence of the absorption spectrum indicate that the TSPP
aggregation proceeds via TSPP protonation. Such protonation leads
to the formation of diacidic TSPP, characterized by a red-shifted
absorption band at 706 nm and by the quasi-coplanar orientation of
the perphyral phenyl substituents with respect to the plane of
porphyrinato macrocycle. The Raman spectral data of TSPP in aqueous
solution of pH less than 5 convincingly indicate that aggregated
TSPP assumes the diacidic form that has a quasi-planar
conformation, but the pyrrolic rings in the prophyrinato macrocycle
are alternately offset. Based on the Raman spectra it is deduced
that TSPP forms both J and H aggregates.

R.B. Lai, Ph.D.
Professor
Department of Physics
Alabama A&M University

GROWTH OF LARGE MAP:MNA AND OTHER ORGANIC CRYSTALS FOR NONLINEAR
OPTICAL APPLICATIONS

Large single crystals (8x8x80mm3) of mixed
methyl-(2,4-nitrophenyl)-amino-propanoate (MAP):
2-methyl-4-nitroaniline (MNA) have been grown by a novel solution
crystal growth technique. In this technique, the seed crystal is
pulled upwards from its growth solution at a controlled rate along
with a decrease in the solution temperature to create a desired
supersaturation. MAP was synthesized by the method described by
Oudar and Hierle and was purified by repeated crystallization in
methanol. Commercially available MNA was purified twice by physical
vapor transport (PVT) using argon as a carrier gas. The solubility
of MAP:MNA mixture was determined in different compositions of
ethyl acetate and ethanol. A suitable volume percentage of ethyl
acetate for growing large crystals of MAP:MNA was found to be
between 20% and 80%. The key to solution crystal growth is the
proper choice of solvent for growing crystals of the desired
morphology. The conventional solution crystal growth apparatus was
modified to grow large organic crystals using organic solvents. By
using this new technique, a severe problem of parasite crystallite
growth at the sides of the seed crystals was greatly reduced and
the yield of the crystal along (010) direction was considerably
increased. The grown crystal shows better mechanical properties
compared to MAP and MNA crystals and demonstrated frequency
doubling at the Nd:YAG laser wavelength of 1064nm. Also preliminary
results of the growth by solution and by Bridgman techniques of
another organic crystal 3-methoxy-4-hydroxy-benzaldehyde (MHBA) are
described.

Lawrence E. Murr, Ph.D.
Professor of Metallurgical Engineering
and Director
Materials Research Center of Excellence
University of Texas at El Paso

BUILDING AN ADVANCED MATERIALS RESEARCH INFRASTRUCTURE

The National Science Foundation MRCE Program established at the
University of Texas at El Paso (UTEP) in the Fall of 1988 is the
University's capstone program to attract minority students to study
science, mathematics and engineering. Emphasizing interdisciplinary
research opportunities for the region's minorities, the center
objectives are:

o    To strengthen the role of UTEP as a doctoral-level institution
     in Materials Science and Engineering;

o    To increase the rate and quality of research activities in
     Materials Science and Engineering;

o    To increase access of minorities and women to careers in
     science and engineering.

During its first five years, MRCE-UTEP has built its capacity to
conduct state-of-the art research in Materials Science and
Engineering from roughly 9 faculty involved in 3 core research
project areas to 27 associate faculty involved in 9 core materials
research project areas. The single most significant outcome over
the first five years has been UTEP's leveraging of NSF's investment
to secure State of Texas approval for a new multidisciplinary Ph.D.
program in Materials Science and Engineering and the creation of a
Materials Research Institute to administer the degree program and
coordinate and promote materials research on the UTEP campus, and
industrial outreach in a partnership with other UTEP colleges,
departments, and administrative units. Research publications by
faculty associates of MRCE-UTEP have increased from approximately
15 in 1988 to more than 100 in 1992; with students involved in more
than half of these publications. Currently 6 minority students are
enrolled in the multi-disciplinary Ph.D. program in Materials
Science and Engineering, including students who became involved in
NSF-MRCE, other NSF-HRD, and related Federal minority research
programs as undergraduates, and received B.S. and M.S. degrees from
UTEP. This is a testament to the success of NSF-HRD pipelining at
UTEP, and bodes significant potential for the Nation with the
prospect of achieving Hispanic Ph.D. recipients in the next few
years which would more than double the trend of the past decade of
graduating less than two Hispanic Ph.D. recipients in Materials
Science and Engineering annually. MRCE-UTEP has been instrumental
in developing a Cooperative Materials Laboratory Network on the
UTEP campus which will serve the continued growth of materials
research and provide outreach to regional industry. UTEP campus,
including RIMI, MARC, CRCM, and AMP programs, is also creating
effective linkages between MRCE-UTEP, industry, and Federal
laboratories. As the "gateway" to the Materials Corridor into
Mexico, UTEP and MRCE-UTEP will serve as a crucial link between
U.S. and Mexico materials technology transfer and cooperation;
supporting manufacturing and North American Free Trade.

Keith Pannell, Ph.D.
Professor
Department of Chemistry
University of Texas at El Paso

GROUP 14 COMPLEXES AND MATERIALS

Hydrocarbons, poly-silanes and -germanes, oligo-stannanes and
-plumbanes are well established systems which use as fuels,
fabrication materials, photoresists, perceramics, etc. Studies in
our laboratories have focused upon the incorporation of transition
metals into such systems and the presentation reviews our activity.
Topics discussed include new transition metal catalyst systems for
the isomerization of oligosilanes and new thermal ring opening of
silyl-and germyl-ferrocenophanes to produce electroactive high
molecular weight polymers with both Si(Ge) and Fe atoms in the
backbone chain.

The synthesis and characterization via NMR and single crystal X-ray
analysis of Me;i3CMe;i2SiMe;i2GeMe;i3Sn and Me;i3CM;i2GePh;i3Sn,
the first examples of such inter-group 14 catenated molecules will
also be presented.

Michael G. Spencer, Ph.D.
Professor of Engineering
and Director
Materials Science Research Center of Excellence
School of Engineering
Howard University

WIDE BANDGAP SEMICONDUCTORS OF EMERGING AREAS

Technology requirements have pushed the capabilities of traditional
semiconductors such as Si and GaAs. Specifically, systems are
sought which can operate at high temperatures or in radioactive
environments. Additionally, information stage issues are
stimulating blue/UV laser development. In order to address the
aforementioned issues, there has been an emerging interest in wide
bandgap semiconductor materials such as SiC and III-V nitrides.
Silicon carbide is a material which is a very attractive for high
temperature applications, while the III-V nitrides (ll direct
bandgap) are attractive for UV opto-electronics. In this talk, we
will review the status of the field with emphasis placed on
highlights of the new work to be presented at the upcoming 5th
International Conference on Silicon Carbide and Related Materials
sponsored by Howard University.

Maria C. Tamargo, Ph.D.
Professor
Department of Chemistry
The City College of New York

MOLECULAR BEAM EPITAXIAL GROWTH OF WIDE BANDGAP II-VI
SEMICONDUCTORS

State-of-the-art photonic and optoelectronic devices are based
primarily on semiconductor multilayered structures or
heterostructures. Presently, these devices are based on III-V
compounds such as GaAs and InP. These materials possess bandgaps
for device operation in the infrared region of the optical
spectrum, of particular interest in light transmission applications
through optical fibers. However, new technologies, such as optical
recording and xerograph, require the development of materials which
span other regions of the spectrum, and which possess novel
physical properties, II-VI semiconductors are direct bandgap
materials that encompass a very wide range of bandgaps, from the uv
to the far infrared and are, thus, very attractive candidates for
optoelectronic device applications. Wide bandgap II-VI's are also
highly resistive and can be used as insulator layers in device
structures. Additionally, a subset of the II-VI alloys which
contain magnetic atoms are known as dilute magnetic semiconductors
(DMS) and exhibit unique magnetic properties, of potential
applications in magneto-optical and magnetic memory devices, among
others. Finally, these materials exhibit strong electro-optic and
non-linear optical effects.


--------

Figure titles and captions (note: figure, tables and equations are
not included in this electronic version.)


Order of Selected Countries Ranked by Mean Percent of Mathematics
Test Items Correct for the First and Second IEA International
Studies

Note: The values for the second mathematics and science studies are
much higher than in the first studies because a different set of
items was used. The items in the second studies were easier for
students in all countries. Only large differences, of 4 percentage
points or more, between countries are statistically significant;
thus, the exact rank orders of countries with similar results are
not detectable.

Sources: National Center for Educational Statistics, Digest of
Education Statistics (Washington, DC: US Department of Education,
1989); E.A. Medrich and J.E. Griffith, International Sciences and
Mathematics Assessments: What Have We Learned? NCES 92-011
(Washington, DC: US Department of Education, 1992).

First StudySecond Study

Order of Selected Countries Ranked by Mean Percent of Science
Test Items Correct for the First and Second IEA International
Studies

Note: The values for the second mathematics and science studies are
much higher than in the first studies because a different set of
items was used. The items in the second studies were easier for
students in all countries. Only large differences, of 4 percentage
points or more, between countries are statistically significant;
thus, the exact rank orders of countries with similar results are
not detectable.

Sources: National Center for Educational Statistics, Digest of
Education Statistics (Washington, DC: US Department of Education,
1989); E.A. Medrich and J.E. Griffith, International Sciences and
Mathematics Assessments: What Have We Learned? NCES 92-011
(Washington, DC: US Department of Education, 1992).

First StudySecond Study

Percent of Public High School Seniors Who Completed Selected

Mathematics Courses: 1980 to 1990

Student/Teacher Ratio, 9-12, Chemistry: 1991

Student/Teacher Ratio, 9-12, Physics: 1991

Source: State Indicators of Science and Mathematics Education 1993,
CCSSO 1993.

Percent of Grade 8 Students At or Above Basic

Mathematics Achievement Levels: 1992