My pleasure to.
Study.
A little background here.
That.
You are.
Really two of them one that's really.
Material Science with you it's just.
That.
We're.
Talking.
About the same thing just summarize
those images for this same thing on.
Earth like a structural.
Structure.
Thank you very much Ken can you hear me
now yes I don't know if this is working
is the is working OK because I
can't I can hear myself here
great thank you very much for
inviting me Todd and David they get.
Expected but there's some.
Of Us The kids are doesn't work.
Then expect people to go for you to come
here for me but but thanks for coming and
I'm going to talk about what the what
has started to bother me because in
the half ago a group of us mature science
which we call as inverse methodology is in
material science and this has got
in two different directions and
it's primarily based on microstructure
presentation microstructure.
Zation and representation of
the processing path how we can link
microstructure to property and.
That's strange with it and
so this means that
we want to find methodology is
that would link in microstructure
you examine using electron
microscopy techniques X.
ray diffraction to get some data from
these different techniques based on.
Hope sophisticated you get to get these
types of information that you're
from microstructure say the three D.
data or two data or just two point
functions from scattering data X.
or diffraction you want to relink
it to the properties maybe magnetic
properties mechanical properties
optical properties and.
Then you want to also be sure that you can
use those properties based on this linkage
to design materials that can perform for
some specific tasks maybe you want
to material that has some specific range
of magnetic properties and conductivity or
you want to feel SR that you want to
have better mechanical properties and
that the that brought us to the originally
that when we applied this we applied
this to for example if and tips for
the material systems that would work for
even tips as a starting
point later on we got three
people primarily concentrate the
microstructure design materials design and
we called going back because in general
the process is you have a microstructure
you examine it in your laboratory you
measure me getting maggoty properties you
measure conductivity you measure thermal
properties and then you have you have
a database and then you say well I'm going
to pick this material of that material but
if you have a set of materials
properties you can find in
any of these you know off
the shelf type of materials or
ashes diagrams so to find it
meaning that I'm going to go from
the mechanical design back
to microstructure and
probably to properties and
then once you find a property so
all these are the type of properties
that don't want to I want to have and
then you want to find out what the but
type of material systems and
whatever microstructure is you want
to use for those processes for
those applications and then you wonder
well I know these are the term makers
should mature structures that they want or
micro structures that they want but
how do I produce it it may take millions
of dollars to produce something
that would match my requirements can I
find better techniques of processing
techniques and kind of possibly find ways
so it becomes of modeling challenge.
Of the musician challenge so
we learned that we can use the linkages
by representing microstructure in
the desert as a digital representation
of micro structures based on spectral
techniques and correlation functions and
then if you can mathematically represents
the same thing for properties same
thing for processing this linkage
can tell this mathematical
linkage can bring the properties to
microstructure matter such a process and
then inverse going from processing to
microstructure and then properties so
this can flow information back and forth
this did not exist before the Such such
as the framework of course right now
materials informatics has become really
hot topics these days of people working
on large data this is along the same
type of information that we had the same
type of research that's going on but
it's more career lies on microstructure
representation how do we represent
microstructure mathematically then we
have a three dimensional in general
what happens is you have a microstructure
you use one of these techniques like five
element techniques or some of these other
black box by modeling efforts that would
tell you what the properties are or you
know and then the many mature scientists
that we through that would think
well it's just not possible to have
any techniques that because so many
complications within mathematical systems
you go down to the mystic level
you have all these different
complexities that you cannot
possibly make that linkage but
be with the most in three important
attributes of micro structures.
For example you can look at your systems
the number of faces distribution of faces
you can look at the micro structures
was the defect structure or
the grain boundaries or
grain interfaces and
those become materials actually views that
we can use to represent mathematically
and we have been applying there so for
years I've been working in this area
just linking microstructure to properties
until probably a few years ago I got in to
solve the excess fuel cell we did a lot
of work in that regard lots of papers and
process materials to make this
linkage possible and then later on
I concentrated on microstructure
processing some going to primarily talk
about microstructure processing
representation and modeling and.
And we are going to I'm going to go over
three different projects that they have
presently that I'm going to go over there
and show how we approach them and but
what are the complications some of them we
are right in the middle some of them we're
further to towards the end maybe but this
would be give you a glimpse of the type
of info the type of effort which exists
as a matter of fact there is several
different types of V.A.'s by the agency
announcements which have been through
the White House different from Boeing and
now from different manufacturers and
also NASA that actually go along with
many fracturing with their field
forensics like Homeland Security's working
really hard with your forensics these days
someone to go over a few of those aspects
so there are lots of applications so much.
Forensics this is actually
a project is funded
well it was originally from the through
N.S.F. National Science Foundation and
now through Homeland Security and
we're getting some more into this
from other agencies this is primarily
using this linkage if you have
if you can make this linkage meaning from
property to microstructure processing and
processing capability meaning that
you have you have this material
system that you can process and
based on different processing capability
the microstructure is going to change the
microstructure is going to look different
or the statistics of Microsoft is going to
change so that becomes a fingerprint for
that microstructure And the question is
can be linked that to the origin about
if we do and we know what capabilities
exist in different regions of the war
then we may be able to go back to
the origin so it's a total inverse
methodology that we have to use possibly
this part would be more on the.
Other types of forensics and
this part would be materials forensics so
linking processing to microstructure
processing capability is something that
we concentrate in on the through to the if
you want to get a piece of uranium or some
some alloy that you want to know where
the origin is who actually made it or
whatever their capabilities
they can make it so
you you have you want to have some
modeling tools to make this linkage
the other is the so
in that respect different types of
material systems which are used in nuclear
industry or other industries which
is of interest for materials forensics
could be addressed so we have to look at
the microstructure make that linkage
in the four fashion meaning that was in
one project that we haven't actually
I don't see a guy just in here but
that who just finished her thesis
isn't for example in this economy.
These are actually material
systems that are used in
different types of applications but
they are the phrase diagram.
The region of the phrase there that
constitutes such material systems very
very closely related to the rainy and
rainy imitating the renames accordion so
we use that here at Georgia Tech
because we can have access to you or
any of itself we did and
it was a disaster here but
we we're relying on national labs to
do that for us but here we are just
concentrating on what we call a therapy
and materials to make that linkage So
if we can link these material systems
the microstructure to processing person to
reality then we can apply the same thing
to uranium the brain making the of humans
and other types of alleys of your rhenium
for there which was actually the initial
reason for the funding of this research so
we can use this and
we concentrate on act in a material
section of alloy spore duction and how we
can get a microstructure and then go back
to the origin how we can actually find out
where they came from the by looking the
processing so they go through this cycle
of looking at the microstructure looks
really tiny so it's even from here I
can see where they were so here we
are looking at the microstructure using
different techniques that are all
in this building so downstairs for.
Skeletal microscopes actually diffraction
rely primarily on diffraction and
scattering techniques but
we have either been using it.
The election back you can see the fact on
the tree that they just found out that
there's actually a second one in
this building now with just pressure
we have two of them at Georgia Tech now so
we look at the micro structures
within the phase diagram a different
levels but different phases
that we can stabilize if you know
phase diagrams that we can actually
get it to a temperature that we want some
alloyed this is actually eighteen percent.
We drop it down temperature
at different temperatures and
we can stabilize different phases and
based on that then we can do processing
we can roll it because he treated and
get different types of micro structures
different grains structures and
they give us different properties and
for those we can link and
produce their modeling path that
we are interested so this is a D..
So that at the core of this
whole thing is what we have
that material system then we use
the beauty of a front of you so
we use different these different
techniques that I just mentioned
electron microscopy techniques
optical microscope techniques X.
ray and we can generate microstructure So
I do it in three day or two D.
E.V.A.'s these one technique that
you can actually get an image but
the image has got a huge amount of
information as well as the type of
phases that we have including
the crystal structure and
orientation of the crystal lights point
and you can actually do these in three D.
If you want to and that would give you a
lot of information we can get what we call
as orientation distribution functions the
one point statistical function tells you
the way to distribution of the current
Crystal lights within the material and
as you deform or change the microstructure
that becomes one signature for
that material system we know that well
if I present as a function as a uniform
of abstract projection which is every
official distribution function.
Projected over to the east space
this becomes a signature for
this material system for
example based on knowing the orientation
distribution between across grains and
even grain boundaries becomes another
type of distribution of grain
boundaries that we can we can relate
using it to for correlation Fortune's So
there are microstructure attributes that
depend on point to point interactions
the cross-grained boundaries or
point to point phrases or features and
there are other type of statistical
functions that we can represent
based on the one point functions which is
just volume averages number average is.
So the idea behind two hundred functions
three point function statistically is that
we know based on if we throw you can
think of throwing a vector inside
the microstructure and
then asking yourself what's happening
at the beginning of that vector and at the
end of that vector there will be a grain
at the beginning or maybe another phase of
the beginning of that vector and the end
would be something else I thought once we
have that knowledge and that's only for
one worker once we throw millions of
birth vectors inside microstructure
we're going to get a distribution function
which is a because as a two point
distribution function which would tell
us what happens inside microstructure
across every two points by the way
that's exactly what scattering does X.
ray scattering does exactly the same
thing you eradicate you are getting these
percolation functions that tells you
of course depending on the contrast
in some cases the contrast is low here
using use other techniques but that's
exactly the kind of information that you
would get that started the structure
factor analysis and all the rest since
the beginning of last century so
we do the same thing you know we can
do it with thermography techniques
we can do it using microscopy
thickness three D.
scan sometimes we don't need to go to that
sophistication level of sophistication but
some things you have to depending on
the type of properties that we need they
teach the effect properties like this and
how far in the micro structures
are the fixed mechanical properties and.
Based on the the type of processing
techniques that are used to get
the values of the from Michael
structures that would be after
a different temperatures different
mechanical the formation possibly And
you've got this myriad of different types
of micro structures that you want to
represent mathematically Of course there's
a method called Fire the element I thought
you can best tessellated
microstructure and put every every
point or feature of microseconds at
a computer but how fine can you go.
When you go to one micron one all
the way down to one actually distances
besets historically if you
can use the properties of
statistical functions because there's
a thing called convergence so
you can actually determine how far can you
go using these mathematical functions so
without showing his math of functions
you use a lot of things that he can do
with this with radical functions either
just looking at the micro structures and
get those signatures of one
bun bun Parmenter that of
Usually people uses maybe size
distribution because of grain size or or
volume for a fraction this of one
point functions that in use but
there are high order functions you
can you have through three points
within a micro structure and you have
a three point four actions through that
gives you a high order mathematical
function that you have to fit within
the framework of your spectral techniques
wavelets spline techniques and
you know when the principal component
analysis that you're working with.
The people in math to reduce
the number of complexity so
you can get a microstructure which is
highly which is extremely complicated
fitted to these functions I wrote
have billions of data points
that you don't know that becomes a black
box with millions of data up through
trillions of data points you can get
maybe one hundred questions for this for
you sees expansion maybe two hundred
coefficients but it's not millions and
billions so using those core features they
become fingerprints representing almost
the marker structures then we can
use organization techniques that.
That can link these micro structures
that we've got mathematically and
link it with this statistical much
as a Technics two properties and
same thing with processing so we can use
the same thing exactly the same thing for
processing and then we can go back in
an inverse methodology to know what
the microstructure is for
a certain number of properties so
this is one one aspect of a two
point Coalition for this was done.
This was done.
Actually receive forgot to mention or
this was done I was showing just another
cover the Forgot to take this out because
my studio was presenting are the sessions
of the carriage everything goes so that's
Ishan to listen to his talk so this was.
A projection of this high dimensional
representation of this correlation
functions because this correlation
functions our end they mention all
maybe one hundred maybe two hundred
they mentions and then them projected
into a three dimensional space then we
get something like this would be called
the microstructure whole you can project
it over two dimensional space that would
be another microstructure whole which
is represented based on the to point B.
based on the two or function in it but
two dimensional space so
what it means is that every point is one
microstructure That's the beauty of it so
rather than having this huge but she'll
system with all complexities No I have to
deal all I have to do is meet deal with
one point in the end a mental space and
then if when I process it he treated.
Except chemical reaction because then
you're going to get a different type of
a composition even that is possible but
we are moving towards that goal then you
get another microstructure be so the
question is how do you get from A to B.
may be passes and then the question is if
you have a machine microswitches be can
you find out what is a where that region
was or tell predict how it was made but
the path that it to begin to reduce
power processing would be it was treated
as if you believe manufacture what is
it that made that microstructure and
we use such these types of functions
just to represent them and
then we use this evolution functions
of microstructure based on conservation
principles to find high order functions
does asr for example is an eight
rank tensor the views that represents
the microstructure as a study the process
the process itself with an eighth rank
That's a huge matrix of numbers but
we if we have the correct set of data
become predicted and that would tell us
how we got from one point of in the nice
thing about such functions is that it's
completely invertible you if this is the
initial microstructure you're the final
microstructure you can go back if
you know eight easily invertible So
there's a lot of capabilities within
these functions that can be used and
that would give us not
only the process parameter but
to get it to one point two or
get the fire from fine microstructure
to initial microstructure So
there are several different types of
inverse methodology that we're talking
about and they're all based on the
representation correct representation so
we need really good data from X.
ray really good data from
microscopy techniques three D.
data to do this and once we have it that's
it we don't need to repeat it just you do
it once you find those process parameters
you throw them away from north
that's all you use so
that represents that process and
you get the final property and
these are the type of.
Microstructure is that you
would get Actually this was B.
got it from titanium this is actually a.
Fuel cell that we had that got three D.
image using tomography thinkings
reconstructed in the computer
once we have it for
example if you give me three D.
image I don't have to do serial section
as my if I would develop techniques
that once I have the three the match
I can tell you what the three D.
microstructure is I can still go back
because I know the statistics behind
how that microstructure was developed but
knowing that you can go back to three D.
microstructure but we have to make
this linkage once we have to create it
only once maybe twice just just to make
sure to find those partnerships and
then we can do the research the the
reconstruction methodology is that we have
I worked there we worked on the sidewalks
of fuel so for years and years and
they had all success in this through that
was sort of the fuel cell by the way and.
And so in the case of zirconium
which was the surrogate
that we are different types of
materials that we have the.
Idea that well if you concentrate
on the composition or
composition here based
on this phase diagram
then we want to know the different types
of faces and depending on this region of
those faces how does that affect
the processing path because you
know you may find out that you're going
to marry him based on the record of his.
B.C. and then the combination we can
actually get different phases based on
the treatments and quenching results into
other types of faces so that becomes
a complication so we need to know how they
affect the process so we had to do this
here ourselves we actually made the
samples in the lab together with Dr due.
To cringe nearing we have a casting
facility that we can make these materials
he tweet them with roll them forty of
them get all kinds of microstructure
we come back to David to do X.
ray diffraction for us and
then we get the data and
try to manipulate the data and
find our process in Path functions and
once we know that then we know
that if something is similarly to.
Maybe uranium Lavie I was really as
you can you then we can tell them
based on the information that we got and
once we found and
discovered that process parameter
which is a huge matrix but
once we have it we can tell them what
the regional microstructure is or
how the Microsoft is having a look like
based on the capabilities so if you have
the fine microstructure we can point it
towards the initial where it came from or
how it was made so that was the primary
purpose so we get different.
As we as he treated the Rowlett we
get all these different microswitch
they look the same once you look at them.
Recently to the image but
they're completely different as far
as the statistics of microstructure.
It's because by visible I saw you know
they don't didn't look at different
you know but they are very different and
we get that statistics only by doing this
is your position that we talked about so
the digitization gives you what we call as
coalition functions which represents
each microstructure as they evolve and
as we volved of course we have all these
other types of different structures we
have a secondary purpose if it's
growing they give you secondary
information that we can also use for
the for the purpose of forensics.
And then we get what we call as
orientation distribution functions
these are O.D.'s based on
the efficiency of grains
that are represented based on the three
Euler angles you know much much
earlier about maybe hundred fifty years
ago or when you know astronomy or
physics and physics we wanted to represent
the relationship between two coordinate
axes they found out about because we only
need three angles three rotations so
if you do that if we consider
that those three rotations and
plot those against the different micro
structures these are the plus of yet
with different sections so
we can have like only zero person zero and
forty five degrees and and
roll them to treat them and
they give you different types
of microstructure statistically
to give you the statistical distribution
of orientation distributions they also
become different they're going to
become a signature that we can add so
the question is can we come up with
a global representation materials
representation for this microstructure So
yes my students that here so
I can say whatever I want to say because
he knows a little more about this so
we have become do's
electron microscopy getting
the same information
that we can get from X.
ray microscopy X.
ray techniques or diffraction and
compare those two and
then enrich the database that we
have to come up with that global
microstructure representation and
that global representation.
Can be applied to different
techniques Well this is another take
another project that we're working on
this is actually funded through Boeing
who originally started bout
a few years ago on machining.
But there are different types
of machining blanking turning
you name it there's right of
different types of machine and
the idea was bicycling to Boeing and
if a company is once we machine something
of what's worth my time frame
forty five minutes right.
For that's what I have forty five
minutes so once we have this
work when the machine something when
the machine apart through the final shape
first of all of course you lose a lot
of materials of the materials but
then you lose more because after machining
the surface is not what you want so
you have to remove the circus you have
to polish it or you just have to.
Get a lot of times that's a lot of
expense the machining additional
work that you do to improve the surface
and sometimes the surface for
those of you in mechanics the stresses are
tense island that's becomes the source for
fatigue initiation so
you won't have compressive stresses and
surface when you machine
surface sometimes you get tense
stresses sometimes you get compressive
stresses and the question from them was
what determines that you know because
there are three machine parameters there's
a depth of car there's a speed there
is angle of attack so what really what
is the combination of different parameters
that mathematically can tell us OK for
increase this I'm going to get this
to compressive more compressive or
less pencil is there any way that we can
link it so that started the project for
us that we looked at the micro structure
after machining under this layer So
here's the problem is that after you
machine it's a highly gradient structure
what I'm talking about.
Gradient meaning from ten nanometers or
five nanometers all the way down to a few
microns meaning that you get with certain
microstructure within five nanometers
below the surface the only hand use
is he to analyze and then be taught
maybe a few microns that he can use X.
ray or the techniques or even the B.S.
the microscopy to know so
it's a highly gradient structure and
how to link it to process part is a huge
it's quite a challenge and then you're
using the same type of methodology but
this methodology works
on different depths and
we're trying to find a global function
that would represent this gradient so
that's another challenge for us the of
the ways issues are as a challenge
really that we have this turning
Pommy Ters that we have to apply so
we get one microstructure I
guess another microstructure or
group believe we get the range of
gradient through microstructure for
distribution and if the effects structure
of phase interactions the machining
is a high temperature process most people
do not realize when you machine the piece
you are taking it to a very
high temperature sometimes even
close to melting temperature so for
example your material is just high six for
your changing the phases distribution
from room temp just five percent
beta to fifty percent greater than
one hundred twenty five degrees and
machining temperatures goes all the way
up there so we have to consider in
the modeling therefore we have to consider
this phase into a face transformations Not
only that then we have grain growth we
have a crystallization we have defects
structure the sufficient density increases
this is the high shooting process to your
your you are introducing large amount of
plastic deformation with the material
that you have to be you have
to consider and incorporate So
this brings us to cycle that I'm not going
to go through this cycle of manufacturing
process mechanics that we apply
the big machine parameters machine
parameters as well as the boundary
conditions then the modeling.
This is actually together
with Dr Steven Liang.
This is a joint project of mechanical
engineering together in a field science we
concentrate on materials aspect and
he and his group they consider machining
Parmenter is and then from that we
get problem properties like hardness
module is a function of that and from that
we go back to what are we do as material
sciences grain boundaries microstructure
phase distribution crystallography variant
ations the few different structures
these dislocations defect structure and
then we quit the question is
how does that affect how do we
manipulate the process parameters to get a
better property to find property residual
stress as maybe hardness So
this brings us to huge cycle
that we have to achieve by actually
getting the data in a forward question and
then go back in reverse so the this
becomes starting from the inputs which is
process parameters difficult cutting flies
the feeding rate then be applied this
process promises gives us macroscopically
what's happening to every wall
you Melhem and within the material there
would be a strain loss of the gradients
temperature gradients and so on and
then be put in the what we call as a.
Base for
classical self-consistent modeling or for
that we can get by the evolution of micro
structures and properties at hand and
that gives us the sum of the properties
which is of interest and from that keeps
from that that would be one forward model
the question is can we go backwards now
because this is a process from is all the
way up to properties and microstructure
kind of go back why is it important go
back because what I want is I know for
one process promises these are the type
of micro structures of the good and
the properties but
I don't want that I want to
find out what is the what but
before a certain microstructure is
that they want what should the process
parameters be but we can do this for
once of the material systems find that
a parameter was talking about and
then you're set to go to that
that's occurred brings us to this
will stressed measure the measurement
because very important so much measured
measurement of residual stresses how do
we measure them using actually fraction
laser techniques or other things that we
use with extra diffraction seems to be
the best that we can get a deaf
profile of to a few microns and
then by remove all we
can get a lot more so
we know the distribution of residual
stress is a function of depth.
If as we sample size we change the process
parameters and that becomes an input and
then we look at the micro structures the
ones that the the the images that we get
from either E.B.'s the rock or microscopy
an extra diffraction orientation
distribution functions all the way up
to final Microcell machine parts and
then we get these properties so we fit
this to the models that we generate and
avoiding any equations here and
that becomes For
example there's a functional process we
get the different structures as we turn
to get the from microstructure
the different effects structures and
once we have this information fit then
we should be able to go back right now
we're working on prices for
we finished our work on aluminum so
we got those process promises one
looming on the finales of aluminum and
now we're working on Tyson switches to
a structure so we are applying those and
we applied this to it as an inverse
model and we actually were able to
identify some process parameters
that could give them the best ideal.
I guess micro structures that they would
use and we are testing it for aluminum and
also we're going to be moving to our
states India titanium so this was this is
exactly the same wonder they gave you
showed you before now it's in the inverse
passion because now we start from desire
properties and then we're going to get
these are micro structures and go back
to see what the process promises are so
the same slide but in the reverse
that's got inverse methodology or
forensics in a different way so that would
be the author of the interesting part
is that what we developed for
machining can be used exactly the same
modeling therefore to be used for anything
many factory What's the difference between
I think the manufacturing and
much machining in one case you're removing
material a very high temperature in
the other case you actually adding layers.
But same same process right I mean you're
adding or removing there's some subtle
differences maybe major Maybe not but
same type or representation same type of.
Process of the McCain mechanisms those
involve you're dealing with face
transformations we're dealing with huge
gradients and temperature layer by layer
so here is for example a power feed
process is actually part of bed process so
we're working on a power feeding
process that you feed in power and
you melted by laser or electron beam or
other techniques at the tip and
then you add layer by
layer of materials and
the question is in his apology is much
much easier because just one I mean I'm
not of the polymer systems very computer
once you figure it out that's OK but
in metals milling with a lot of other
issues these huge gradient temperature and
residual stresses which is
created the materialist's not as
fluid as polymers becomes very viscous so
that generates problems and
be working with creature hooking
with booing you want to imagine this
we want to make a column
using these techniques and
they use all the techniques all
the questions that they have and
they say they're getting the vertical
column it always goes one way or the other
and what really is missing is that they're
not incorporating microstructure in it or
what comes out of it which is stresses
as a result of the microstructure and
that's what we're trying
to do we're using.
Trying to develop a player models that
can be presented that Mr plastic from and
on that as a functional microstructure
gradient and we are after is in different
modeling techniques that we have modeling
models that we have developed before and
still plasticity we are playing to two
different types of systems one is twice
explore which is of interest to Boeing
steel which is of interest to other
industries the both actually six was a two
for structure Steve can be to face to
face will freeze be got Martin sized
ferrites we've got ostomates we've got
I can add of this as we are going element
changes so we're developing models for
those because as the function as
a function of temperature there's
an evolution of microstructure based
on face to face fractions that occurs
that makes things a lot more complicated
than we had before but very similar
to machine so we're doing basically
the same thing so we're going from.
There are limited databases that we
have we're using different optimization
techniques anywhere from artificial
networks that neural networks or fuzzy
logic to train are our systems to get the
data to fit the data that we have and then
once we do this we are using different
types of modeling efforts that we
already have at hand to be able to predict
like a structure the microstructure is
in terms of the statistics that we get
right it's not direct microstructure but
having the statistics we have this timid
methodology that we can create three D.
micro structures that's exactly same thing
which is done in the case of scattering or
sacks or wrangle small angle
scattering techniques for
polymers that you can actually draw
through the structures based on
these two point correlations we do exactly
the same thing here except that here
we are getting the data for two point
correlation functions differently and
we can also calculated and
then called the three D.
microstructure OK I'm going to stop right
here and exactly forty five minutes so
some of the students who are working
in this regard I have to thank
David who's here has helped us a lot
in last few years on the getting the X.
ray data so
hopefully will continue doing that.
And as everybody and
everybody else so the funding for
my projects right now is coming from
on this and homeland security and
boring aircraft and partially from I
forgot to mention their name is the.
Space Agency and Ga Ga space
agency was more of you working for
them that they do the mission for
the solar cells and else thanks.
Yes.
That's.
The nice things to.
See.
In your life you're stuck.
In the same way your life.
Microsoft is a micro structure so we can
deal with them as long as we're trying to
write students and funding and so on I
think I would be perfectly something that
we could work on I think
where we did work on Born
structures before the really interesting
structures might my past life.
And then I'd also because
it was a micro structures.
I said OK yeah that's what that's
what I'm talking about this.
So we could be could apply
what we have to do is to do
that those applications are just a matter
of the hierarchy right scales so
we have to figure out what the scale of
microstructure are important as far as
the type of properties that you are
interested in composites actually a work
in composite twenty five years ago
twenty some years ago so with for light
structural materials there was my previous
life so that's probably the kind of.
Maybe we can apply those techniques to
the micro structures that you have in
you know really much more
interesting bits I would call.
The.
There are absolutely absolutely.
Yes.
And we can speak should we see do
not know a lot of those properties
where they come from
the interaction of interfaces and
the that we're still striving
to investigate the type of.
The interface problems either
in this is materials have been.
Extremely hot topics for
research like twenty years ago or
somehow we've got
a financial structures and
we sort of are not spending enough time
on interfaces but yeah that absolutely.
Yes thanks.
Last night.
Yes.
So what stood out just this is.
Where we are using that for a number of
different cases this is one example that
we have the we have these models
that we want to figure out
how to fit those curves right you have so
they're very experimental data
your experience later for
the properties the different temperatures
and we have this data there's got like
eleven different palm eaters to find as
part meters we use neural networks to
figure out as an optimization technique
to find out what those parameters are.
But the BE also using that for.
Four or usually for some of the cases
that we don't have a model we just use
the networks as a black box as an input
output but we're gradually replacing
them with a mathematical model that we
have full control over but that's another.
We also using that for reconstruction
mythologies So we're using these for
a number of different applications and
the primary using Matlab or Python so
abilities to do those so we're not
writing those calls just using them.
Exactly in this case is just finding
those questions so we have a database and
then you want to get the questions for the
models that represent those curves that
you see or those are structuring curves or
straight test tests and
if you want to fit into a model that we
have it's going to be a major challenge.
Because you know there are lots of
lots of this is actually a small
segment of the model that we have but
so we have lots of questions and
find out what those questions are is
going to be a major challenge.
You know you can use to bring statistics
and points that is to represent
a different structure green boundaries
grain size so it depends on how.
It doesn't replace the defense structure
is actually a way to statistically
represent mathematically the distribution
of those defects so the question is you
have a defacto evergreen boundary you want
to know where it is and how it impacts
your properties so one think Nick is using
the statistical formulation which for
me from microstructure is the best because
you could probably use other things
there's only other solution would
be using a fight element or
find a difference approaches for
that the number of I mean
the of course you can use this
thousands of thousands of processors
to do this to get finer and finer but
this one is a statistical methods takes
into account all the complexities whether
that's enough it should be enough
unless the material is not homogeneous.
If the machine is highly history
genius then then we're in trouble so
we have to use it we may have to rely
on fire the woman at that point yes.
It's very low compared to fire the liver
right now my student runs
everything in his laptop.
Mac..
You may take a couple of hours but if you
want to do the same thing with a file and
you have to go to a large process
the power processor may be loaded or
they would run for weeks so that's one but
I vantage of these types of competition
techniques I mean the results are
approximate of course was statistically
approximate right I mean if you look at
the features within this war you know you
look at one corner computer or call it
people see different sets of Stickley
there may not be any difference but you
could you could possibly tesselated them
put billions and billions of Brits and
get a very very nice result but
the same thing is going to be done using
the statistical techniques with a variance
with some error but the question is
what there are but what is there so
you have to do some analysis of the inner
analysis to be able to see if you could
use such as to call things right so
I'm not saying they watch for
everything so but
the kind of water they do it doesn't play.
Like.
I'm sorry yes.
Absolutely yes.
Yes yes.
So.
That's the only test that we have so as we
increase the size of the volume element
because this representative volume element
ulti so that our view as it increases
if the statistics does that change we know
that we reached that total of the that
the limits and the problem is that this
is unfortunately sometimes we don't
have a large segment of the microstructure
to test that that's a problem so
there are some experimental problems
to do so but in many cases for
the microstructure that they deal with
because in one millimeter up got billions
of grains so I don't have any issues
with that you know because usually
you can achieve dead within like ten
grains or maybe twenty grains but
in other cases there are cases
that yes you actually limited and
as I told him then maybe these
techniques would not work so
I'm not saying that the sickness works for
every case but it works for
cases that you haven't got a C.T.
satisfied place yes yes.
You see that OK one thing I do is just
one which in isolation and representation
microstructure business so idea if I rely
on physics this doesn't create physics so
I'm not claiming that you know I'm
relying on I do a live the physic but
that's separate from this right so
the physics has to be there for
this to work your physics is not there so
we do a lot of D.F.T.
analysis the BE do and molecular dynamics
so my students are working on that so
that's separate from this once that's
established then these techniques works so
I'm not claiming that this can replace
physics definitely not this is just.
A reference microstructure representation
statistics that you can use
to use as signatures for microstructure or
link to properties a process
I don't know if I answered
your question but yes
yes.
Absolutely we have a problem
right now with these
systems that we have the procedures
really fine so we have to use.
X.
rated is just to see the distribution
we don't know if we are not going
to be able to get an image but
we have to use T.M. to get those
images but depends on what the really
effects the what what what the but
it depends on the scales
of the problem if you're only interested
about finding a signature from a fuel
system sometimes those are not important
the only affect physics of the problem so
we already know the physics of
the problem for this person or
that the people of you working on it I
myself am working on it for years so
we know the physics as long as we
have some indication of how the how
large they are because there's an optimal
size for them to affect properties so
hard that the effect is considered
relations we're fine but
beyond that we be able you're right
it's going to be a huge experimental
challenge and actually that's what we
are joined to deal with to find out there
because on the global scale we can
because we know the Ground Zero structure
we go in a final scale the defect
structured the question
then city precipitates then we have to do
more sophisticated techniques to get them
extra you can do some and then we have
to use rely on something to use if
empathetic for those so once we have that
information the good thing about this type
of model year for it is you just need once
again but you have to do it correctly and
perfectly for
your model to work in reverse so
once you do it once then you're
done maybe twice just to verify.
So I'm not saying that they're not
going to have an effect they do but
sometimes some properties
do not get affected for
those of you interest about
elastic properties you know.
Those those small personal mean
other plasticity Yes hardness
Yes there are some other so the present
what you're looking for you're looking for
magnetic properties maybe not you
know the plans on the type of so
you have to approach each problem
separately and find out what are the most
important microstructure actually
buttes that you have to consider and
then incorporate that in your model I
mean your model can have billions of
parameters they're useless and
maybe only three of them does the job for
you you know so it's a matter of
optimization or finding out what the.
BILL OF DIFFERENT most mature scientists
they don't see that MOST important
that actually do this of they see
this material system they see all
these complexities defects point defects
this will patients they can't see how
any of these techniques can work.
So we're constantly trying to prove
to them that this does work so.
That this.
So.
You know the problem a neural network
is that you have to have a lot of data
because you don't see any
trend there's no beer so
I mean another relying on regionally
of course in some cases to find for
optimization and we're using different but
in general we're trying to find these
clothes from Solutions evolution of
microstructure just move functions for
some cases the functions may not be
smooth may be singular so we're trying to
discover those types of function based on
the physics of the problem otherwise you
know you're right be have to have a huge
amount of data points but if you have for
example machining be yet maybe three or
four in each and then we have
only maybe twelve micro structures
that we have to be the couple of times
that might self takes a couple of years
to analyze and get all the data so
it's not then it's not really a small
amount of work it's a lot of work but
at least it's not like thousands of
parameters that you need to do for
neural networks so once we have that
the good thing about it is that
once we have that we don't have to repeat
what every microstructure if you do
far more liberal you may have to repeat
for every microstructure because for
when you're a little Everyone has such
as different for this techniques every
microstructure every process has
got a smooth function buried inside
that you have to discover and find that
linkage that statistical linkage you get.
They make you.