At a time where longevity and product life cycle is becoming increasingly important, learn how materials life can be modeled well before the production line. In this episode of our GE sponsored series we explore materials modeling and using tools such as Finite Element Analysis to save both money and materials. This episode is sponsored by GE, a leading company in a multitude of fields. Check out their research homepage here. Take a look at jobs here This episode is sponsored by Materials Today, an Elsevier community dedicated to the creation and sharing of materials science knowledge and experience through their peer-reviewed journals, academic conferences, educational webinars, and more. Thanks to Kolobyte and Alphabot for letting us use their music in the show! If you have questions or feedback please send us emails at [email protected] or connect with us on social media: Instagram, Twitter. Materialism Team: Taylor Sparks (co-creator,co-host), Andrew Falkowski (co-creator,co-host,editing assistance), Jared Duffy (production, marketing, and editing), Ramsey Issa. Keywords: Materials Science Finite Element Analysis ABAQUS General Electric
I would like to describe a field in which little has been done, but in which an enormous amount can be done.
This field is not quite the same as the others in that it will tell us little of fundamental physics,
but it will tell us much about the strange phenomena that occur just below our perception.
In contrast to the natural philosophers of the past,
the scientists of this field delve into the recesses of nature and show how she works in her hiding places.
Their quest is to understand and create the imperceptible.
After all, there is plenty of room at the bottom.
Hello and welcome to the Materialism Podcast,
an exploration of the past, present, and future of material science and engineering.
My name is Andrew Falkowski, and I'm joined as always by my trusty co-host, Taylor Sparks.
But this time, virtually. Taylor, where on earth are you?
Hey, nice to hear from you, man.
I am on the other side of the world in Liverpool, England.
It is quite literally a dark and stormy night over here.
So if you hear the thunder and lightning and the rain in the background,
it's because I've brought to you some of that, you know, storied English weather to the podcast.
And you're going to be out there for a whole year, that's right?
That's right. It's one of the biggest perks of being a professor is that every seven years,
you're eligible for a sabbatical where you can go and, you know,
take a break from teaching, service, and research to just do research.
Or sometimes people do teaching, but mostly they use it to go do research.
That's what I'm going to be doing.
I'm here with the University of Liverpool, and I will have a lot more to say about
the really cool things happening at this university, probably in a future episode.
Yeah, looking forward to hearing the fruits of your labor.
Well, thanks, man.
Today we have a really exciting episode.
If you've listened to some of the past episodes, you'll know that GE is sponsoring a series of episodes
covering materials engineering, innovations, and technologies at GE.
Now, GE's roots are in the Edisonian exhaustive trial-and-error approach to engineering.
You know, 99% perspiration, 1% inspiration.
However, efforts to maintain a market advantage has led GE to depart from this
and embrace a new means of accelerating innovation and pushing technological barriers.
And one way they're doing this is through materials modeling,
which is the topic of today's episode.
And what's interesting about materials modeling is I think it means a lot of different things to different people.
For some, maybe experimentalists, it might be collecting data points,
putting into Excel, fitting a regression line, or maybe fitting an analytical equation,
maybe the creep equation, in order to try to predict behavior.
But it doesn't have to always be that simple either.
Yeah, you're right. It can be quite a bit more complicated, right?
Let's imagine that you're trying to model how heat or mass moves through a material, right?
Then you might use a different type of model, like finite element method, right?
So here we're actually trying to predict material properties
or predict behavior of some device-level component.
So very, very different.
Yeah, or others might even take it another length scale down and be looking at atomistic simulations
with density functional theory or molecular dynamics,
where atomic and molecular interactions are simulated to maybe find the properties of yet unknown materials
or maybe learn about interactions that can't be observed with standard characterization tools.
So the truth is that materials modeling can be all of those things.
It is a really broad concept that says we don't just have to look at everything experimentally.
We can simulate a lot of different things across a lot of different time and length scales
and use this in the materials design process.
And I think a lot of people are aware of modeling now,
but why should industry or academia really care in the first place?
I mean, we can do experiments, we can get the same and probably even better answers
from those experiments as well.
So why should we even invest time into learning or resources into developing tools for modeling in the first place?
I think this really comes down to money and time.
Experiments can take weeks, months, even years, depending on the complexity of those experiments,
the scope, depending on what kind of data and how high a resolution of data you want to collect.
And so this can make it really difficult to iterate quickly over things.
If it takes you a year to get a result, how are you expected to arrive at a final product
in any sort of reasonable time in order to get that to market?
Yeah, absolutely.
Another reason is that even if you don't mind the time and the cost of doing things experimentally,
there's just some things that we can't achieve experimentally.
For example, with simulations, we can simulate crazy temperatures
or crazy irradiation levels or crazy pressures,
you know, things that we can't easily or feasibly even achieve in the lab.
We can do those not easily, but we can feasibly do them with simulations.
Yeah, and this is a big component of the nuclear industry
where a lot of experiments are not only expensive but also prohibited.
You really have to validate a specific approach or design
before you're going to be able to do any sort of experimental work.
And so being able to accurately model materials behavior
and how components are going to fail or if they're going to fail
is really important in industries like that.
Another thing is that it's one thing to know material properties.
I can create a tensile specimen to an ASTM standard
and I can pull it in some sort of Instron machine.
And that will give me maybe the yield strength,
but that doesn't really tell me whether or not a component of a complex geometry will fail
that's subject to a three-dimensional stress state that's really complicated.
And so using modeling, we can get a better idea of not just,
oh, you know, we expect these stresses to be this,
or what are the specific stresses at a given point?
And this is important because making actual parts, getting them machined,
getting the materials specced, and doing this so-called cook-and-look approach
is also really expensive.
And simulations can allow us to iterate a lot faster
by allowing potential parts to be designed and tested in a computer environment
before being fabricated, which ultimately helps you improve your chances
of maybe your first experimental test being more successful.
If you can identify potential errors or points of failure on an early design
and correct them before you actually go and make a prototype,
you have a better chance of that prototype being successful,
or at least more successful than it would have if you hadn't done that testing.
So what you just described, that sort of mentality,
is exactly the concept behind some major initiatives
that have come out that the entire field of material science has started to lean into.
So if you've been in material science for a while
or if you've gone to some of the big MRS or TMS conferences,
you've likely seen symposia that are under the umbrella of ICME
or Material Genome Initiative or maybe the DARPA AIM, right?
So what these are, these are different initiatives
that have been primarily led by the federal government funding agencies
which try and do exactly what you said.
It's bringing to bear the utility and the capacity of computing resources, essentially,
along with traditional material science
to more effectively go through the design process for materials design.
So DARPA, for example, this was something that kicked off
by the U.S. Defense Advanced Research Projects Agency.
If you're familiar with them, you know that they do high risk, high reward, right?
They're really after major, not incremental stuff.
They want big breakthroughs.
It was also funded by the United States Air Force Research Lab.
They did it particularly with an aim towards the aviation industry.
Think about why that's so important.
Aviation, getting a new component from design, from an idea,
all the way through to application where it's actually flying is a huge process.
It takes forever.
There's all these checks along the way, obviously, because safety is involved.
This is a really big undertaking.
So if you could speed that process up by doing simulations in a number of different ways,
then you could have a really big advantage here,
and that's what they've been doing primarily with that program.
Yeah, and there's another one that's really famous known as the Materials Genome Initiative,
which I'm sure you've heard us talk about before.
While DARPA AIM is a lot about merging component design and materials design,
I think the Materials Genome Initiative is putting resources, funding, and education
towards trying to develop new tools to allow us to design new materials faster.
Their whole thing is deploying advanced materials twice as fast at a fraction of the cost,
and really that's through policy and resources that promote the development of tools
for materials design simulation and the education of people to use them.
So I hope that we've sort of sold this idea of why simulations might be of interest
to the field of material science.
We're going to dive into how they're being used at GE, but before we can do that,
we better say a few words about the different types of modeling that are going on,
and I get it, many of these things are going to be their own future episodes,
so we're going to keep it very brief and high-level for a minute.
But let's start with finite element analysis.
That's a modeling technique which is used all over the place,
and essentially what it allows you to do is go beyond the really simplistic geometry problem
that you face when you move to real-world parts.
Think of your statics class, your dynamics class, when you first learned about,
I don't know, Griffith's Fracture Law or something, or the bending moments in something.
If you have a really simple sort of beam, it wasn't too bad to do the math on that.
But now imagine what happens if your component that's under a stress is a door panel on a car.
Now you've got curves, you've got sharp angles, you've got all these things.
How on earth do you know what the stress state will be overall?
Where is it going to fail?
What will be the stress multiplier near sharp corners?
All these different things.
That becomes a really hard problem, and the answer to that is really clever.
They essentially say, take the component, which is some strange shape,
and discretize it into a bunch of tiny little areas.
Mesh it out.
Now you've got a mesh, you can do the math on each one of those small mesh shapes,
little cubes, for example.
Then you just look at how each one of those little discretized sections
is connected to the nodes around it, and it becomes a much simpler problem to solve.
Yeah, that's right.
You can easily think of this sort of hypothetical example where you've meshed your geometry,
and then you're approximating partial differential equations functionally.
You need to apply some boundary conditions to that,
you need to apply some material properties like elastic moduli,
and then you could solve for perhaps mechanical stress.
Maybe Hooke's Law, if you're looking at just linear elastic stress,
and this could be maybe a steady state condition,
such as some sort of constant load in a building,
or maybe it's transient, you have dynamic loading conditions.
And so you can get a result,
and now you have the interesting problem of trying to interpret it and verify it.
Really, the engineer is the one who sets up the problem,
and these days there's enough sophisticated tools out there,
things like Ansys, Moose, Abacus, that can help you solve the math.
But then as an engineer, the next step is you have to analyze the result.
All it's doing is solving the equations you give it,
but whether you set those up right,
and whether you can analyze to get the right output or the right conclusion about that,
is really on you.
And just looking at linear elastic stress,
a lot of the maybe mechanical engineers who are listening to this,
or even material scientists as well,
will know that that's not really the whole story.
What if, for instance, there are thermal fields present?
Well, okay, now you need to know material properties as a function of temperature,
so you have to go look those up.
And maybe over time, at elevated temperatures, materials like to creep.
So now we have to construct a model capable of capturing elastic and plastic deformation
over some sort of time domain,
and you can really see how this starts to open a can of worms, right?
Creep will vary from material to material,
it'll vary within materials as a function of microstructure, temperature, phase,
and I think this really starts to hit at the crux of materials modeling,
and that's looking at these complex physical and chemical interactions
that give rise to the properties that we know.
And going back to creep, right, the microstructure consideration,
that's a whole new dynamic.
For 0.5% carbon steel, your yield strength can vary from 250 to 1,000 megapascals,
and that's just for microstructure alone.
So how on earth do you predict that?
Where do you go next, right?
Using FEA, it doesn't tell you what phases are going to be present,
nor how a certain microstructure evolves.
Yeah, and that's why there's a variety of other techniques.
For example, one of the most commonly one used here is CALFAD.
CALFAD stands for Calculated Phase Diagram.
If you're a material scientist, you know how valuable phase diagrams are.
They essentially are like a treasure map, right?
They tell you under different conditions, say composition or temperature or pressure,
exactly under those thermodynamic conditions, what phases will be present.
Now, this does assume some big assumptions like equilibrium, for example.
These are used at equilibrium.
But still, they are really valuable.
Now you can say like, oh, under these experimental conditions,
this composition at that temperature, there are two or three phases in equilibrium with one another.
And then you can start to imagine, well, how would those phases be distributed?
Are they going to be a host with precipitates?
Are there going to be two separate phases?
Are they going to be along the grain boundaries, right?
That's the whole field of material science is sort of moving towards that understanding.
But it sort of assumes that at the outset, we know what phases are present there.
And CalPHAD, as a simulation or calculation software,
can thermodynamically calculate which phases are present.
But there's some limitations, right?
Beyond just the fact that it's at equilibrium.
Yeah, I mean, yeah.
Obviously, the fact that these are equilibrium and so many things in material science are non-equilibrium,
that's the biggest one.
But another even more daunting problem is the fact that this CalPHAD approach
sort of relies on you having data that you're going to learn from in your database.
And this isn't like the machine learning type of learning,
but it is measurements of thermodynamic quantities for different substances
that allow you to estimate which phases will form.
And we just don't have the full complete set of all the thermodynamic measurements
that we want for all the different mixtures of phases, not even close, right?
There are some great resources out there, you know, ThermoCalc and FACSAGE and others.
In fact, in the 70s, this was like what a lot of material scientists did
was make these fundamental thermodynamic calculations
and then predict the phase diagrams and experimentally measure them.
But we have a lot of unknown space there.
There's still a lot that's unknown.
And so it'd be great if we could simulate the properties of individual phases in other ways.
And that's where we dive another length scale down
and we go to first principles where we can compute material properties
for very small configurations of atoms based on their electronic potentials.
We're sure to do an entire episode on these.
But these allow you to compute real properties of hypothetical materials
based on their atomic arrangements.
And so maybe if you are investigating a new material that's never been made before
or maybe you predicted a potential material having high properties
and you want to test it before you actually make it, this is a way of doing that.
However, you miss out on microstructural features,
which, as we noted earlier, are very important.
And we know that systems are also very dynamic.
And so capturing the effects of processing conditions are also very important,
how things solidify, diffusion of different components.
And so that's where we have to start incorporating
and looking at models that can incorporate kinetic information.
And that's where these things called phase field models come in.
These concern the transient evolution of field variables
that describe continuous regions in the bulk as well as changes at interfaces.
Maybe to put this another way, this is concerning grain boundaries
and phase regions and how those evolve.
These could be liquid to solid transitions.
These could be solid-solid interactions.
You've probably seen those great videos where they do spinodal decomposition
and they show it in real time based on a simulation.
Those are usually phase field models.
But as material systems get more complicated, we start to run into a problem.
Modeling begins to struggle to capture the real behavior of these materials.
Try estimating the diffusivity of 10 element alloys.
You have to have information about interfacial energy,
maybe other sorts of faults that are likely to form.
And some of these quantities are so small, the additions of these dopants,
that your signal-to-noise ratio is like one.
How are you really supposed to extract meaning out of that?
And that's where I think really puts us in the challenge of modeling
and trying to understand how far we need to go with a model.
Because I think the ultimate goal is to have the cheapest model possible
in terms of computational load and development time,
and get the result that we want.
So we can always go deeper in terms of how detailed a specific model is,
but I think the end goal really is to get us the information that we need
so we can perform the actual experiment.
And hopefully, walking through this philosophy
illustrated the approach to modeling and how maybe a simple inquiry into
will my component survive ends up being a lot more complicated,
but how different modeling tools can help us answer questions related to this.
Okay, I think that we are ready to dive into some case studies
where GE Research is doing some pretty impressive things with simulations.
So let's start with the first one, which has to do with super alloys.
Now, we're going to do much more talking about super alloys
because GE has done incredible work there.
But in today's episode, we're going to talk about the really leading role
that simulations played in developing a new alloy.
Andrew, what went on here?
Yeah, so one of GE's big technological products are aircraft engines, gas turbines.
These are environments where materials are going to have to have
not only a great instantaneous strength, so their yield strength,
but also be able to maintain that strength over long periods of time at high temperatures.
That's pretty tough.
I mean, most materials start to weaken and don't perform nearly as well at higher temperatures.
So you're really looking for very specific, very advanced materials.
It turns out that nickel-based super alloys are really the best choice for this.
As you mentioned, we're going to cover these,
but they're a really interesting class of materials
where we take two softer, weaker metals, like aluminum and nickel,
and we combine them to get something that's significantly stronger than either of them alone.
What makes these materials so special and almost perfect for this application
are a couple of physical phenomena.
Now, we're going to go into these in much more detail in that episode,
but maybe to just give a high-level overview.
In the formation of these alloys, you'll have precipitate hardening,
where you form a slightly different phase than the bulk within these,
known as the gamma prime phase.
These precipitants are coherent with the matrix,
so that means that their lattices line up,
meaning that dislocations are going to be able to move through them,
which is what you want.
So if a dislocation tries to shear through these precipitants,
it will form a high-energy planar fault called an antiphase boundary.
The most important thing is that it has a high energy cost for this specific phase.
So because it takes so much energy to actually form this fault,
it means that the material also has a high yield strength,
meaning it's unlikely to shear or plastically deform.
The other great thing is that the material's matrix phase
has a low stacking fault energy, which enhances the creep strength.
Essentially, in the FCC matrix, it's energetically favorable
for these dislocations to split into partials.
If they encounter a precipitate that they cannot shear
because of the previously discussed high-energy antiphase boundaries,
they could try to recombine via a cross-slip mechanism,
which is where they shift what slip plane they're moving along.
But stacking fault energy in the matrix is really what kind of determines
whether or not this cross-slip is going to occur,
and because it's so low in the matrix, it becomes unlikely,
and so you end up having the splitting of partials,
and you can't recombine and continue to move elsewhere,
and that allows you to resist creep.
That's really interesting, Andrew.
I think most material science students understand the concept
of precipitate hardening, right?
If you split something and the new thing that you created has new surface area
and it's a high surface energy, it makes sense why that would be not favorable.
But it's interesting here how you point out that with creep strength,
when these things have to decide whether or not they're going to split
into the partial dislocations and whether or not they recombine
has to do with how comfortable they are.
They are essentially forming these stacking faults,
and if that energy is low, then they're fine to just stay separate.
That's really interesting to me that in one case the high energy is favorable,
in the other case the low energy is favorable.
Yeah, and the rabbit hole on these materials goes a little bit deeper even.
Normally our understanding of materials is as you heat them up,
they should get weaker, right?
Dislocations should be able to recombine more easily,
you should have faster movement, things are more energetically favorable,
but not exactly the case for super alloys.
In this case, they experience something known as the yield strength anomaly,
where up to about a temperature of roughly 1,000 Kelvin,
their yield strength actually increases.
The reason is because certain cross-slip events that can cause dislocation locking
are more energetically favorable at higher temperatures,
and so you essentially have more pinning of dislocations that occurs
as temperature increases up to a point.
If this sounds confusing and really interesting,
look forward to our episode on super alloys,
where we're going to get right into the weeds of why exactly this happens
and what makes these materials so special.
But hopefully the impression you have is that this is a complicated material system.
Just plugging in a yield strength into a model
is unlikely to really capture the real behavior going on here,
and it also points out that there are some interesting energetics problems here
that also give rise to potentially materials design.
How is simulation helping them develop new materials
in light of this sort of mechanism of the current materials?
We talked about the fact that the high anti-phase boundary energy
is really a driving force for increasing the yield strength here
and preventing shear through these precipitants.
GE used ab initio calculations, so DFT,
where they're going to be simulating the structure of these gamma prime precipitants
and they're specifically looking at what happens when they substitute
tantalum, niobium, or titanium
in for some of the sites that are normally occupied by aluminum in the alloy.
What they found is that they were able to actually achieve similar maxima
in terms of the anti-phase boundary energies,
but that tantalum and niobium actually reached that peak
at a much lower concentration than titanium.
Really what they're finding is a design space in terms of alloying possibilities,
which then allows them to start optimizing for other things,
like maybe a given application has a higher temperature,
or maybe there's going to be more strength,
or maybe you need more oxidation resistance.
Now you can start to play with these compositions quite a bit more.
The other consideration is we want to minimize the stacking fault energy
within the matrix, and this can be done, they found,
through managing our cobalt and chromium content.
We can balance the stacking fault energy considerations
with corrosion resistance.
Adding more chromium is going to make it more corrosion resistant,
but you also have a penalty in terms of phase stability.
There's only so much that you can add.
At the same time, your processing windows, how easy it is to process,
are affected by the cobalt content.
By looking at these DFT simulations,
they're able to get a more detailed look
at how individual concentrations of different atoms and alloying constituents
are going to be able to affect their properties,
and they can now start to actually design materials
to maximize different trade-offs.
This is cool because I don't care how expert you are,
what your level of chemical intuition is,
I don't think I could, even with all the experience in the world,
have guessed exactly how these substitutions
would have changed the surface energies of these things.
That's just not something that we can do as humans easily,
but that's precisely what DFT and these types of simulations
can provide for us.
The next step here is then they can take all of this understanding
of the physics of the yield strength anomaly,
convert this into mathematical equations,
and then incorporate this into a crystal plasticity model,
which essentially allows them to look at how
maybe an actual grain structure is going to be realistically deforming,
and that can be then incorporated into FEA models
of different parts, components, different microstructures,
and so now they can actually take a lot of the characterization work
that they've done on these, as well as their understanding
of the fundamental properties, and actually translate that
to models that show real material behavior in a component geometry.
Pretty awesome.
Let's chat about another success story that they've got.
This one has to do with added manufacturing of these super alloys.
If you've listened to some of, we've done a couple episodes
on added manufacturing, you know that we're big fans of it,
but we also recognize how hard it is,
and GE has been doing some really great work in this area.
In fact, if you listen to our most recent Materials Informatics episode
with them, you heard us talk about how they used Bayesian hybrid models
to sort of help predict what the right processing parameters would be,
so the laser speed, the hatch pattern, all those different things.
They were after that to try and understand it,
but right now it's still a large unanswered question
on how to get the best parts out of these things,
and the fact is that the field doesn't know exactly
how to prevent problems like cracking, and this isn't surprising
because if you think of what's going on in added manufacturing,
you have extremely non-equilibrium processes.
You've got, you're melting something, you're obliterating it essentially
with a laser, so you have volatilization,
and then you have now a liquid phase, and that's going to solidify again,
and there could be changes in volume.
You're talking about complex geometries,
typically really thin walls and passages,
so it's not surprising that cracking is so prevalent
with added manufacturing builds, and this is still just a big challenge
they face, so cracking is something they would love to solve,
especially because GE is moving towards the forefront
of introducing more and more 3D printed parts
into their production.
This is a technique that they are really investing a lot into
because they recognize that they themselves
and their customers' partners are going to have needs
for this technique, and there are still problems to work out.
So with that said, how do you go about preventing cracking?
What they found in the past is that primarily
there's just a lot of exploration, sort of Edisonian, trial and error,
looking at different material compositions.
It turns out that some chemistries are much more amenable
to prevent cracking than others.
But we know that an Edisonian approach is not going to work for everything.
It's just not fast enough. It's too slow.
Can we do something with simulation to help prevent cracking
is what they're essentially going to be after here.
So what they did is they went ahead and they said,
maybe we can use an ML model based off of this.
So if you're going to do machine learning, you have to have data to start with.
So to that end, they 3D printed 40 different compositions,
which is not that many.
And they did that using a set of 24 different print parameters.
So their data set is not massive, 40 samples, 24 parameters.
Not very big at all.
But they took this and then they looked at all the different things they printed
and they looked for the prevalence of cracks.
They're essentially looking at crack densities.
And as they explored those, sure enough, some did better and some did worse.
And kind of to their surprise and chagrin,
they found that a lot of the most commonly used weldability
and castability models that would typically work
actually didn't correlate very well with their experimental observations.
You know, for some of these things that are out there,
like for example, the Ku criterion or the Zhang and Singers criterion, right?
These things can work well with casting,
but they were failing overall in this additive manufacturing space.
They weren't providing good metrics of when an alloy would crack or not.
And that's unfortunate.
But rather than give up,
I think it's really interesting what they decided to do here.
First off, one thing they could do is they could say,
all right, none of these models on their own did very well.
But maybe if we combine them together with an ML model
that takes all of those as features and learns from them jointly,
it could do better. That would be one approach.
But they also added some domain knowledge, right?
Their team has a lot of, obviously, it's GE, right?
So they have a lot of expertise in solidification of complex alloys.
And they have some guesses as to what's driving cracking.
And one thing that they had a strong hunch was responsible
was a change in the volume during solidification, right?
You know, things in the liquid and solid phase
are typically going to have very different volumes,
are typically going to have different volumes.
And so if you can calculate that difference in volume,
maybe that's what's driving it.
And that seems to be what's happening here.
Of course, there's other things, right?
Shrinkage can be due to thermal expansion and other things.
But they decided to just look at the change in volume
at the different regions of the part,
assuming that the composition is not constant.
Now, why would that be the case?
Well, when something liquefies, as it starts to solidify,
think of like a little nodule that sticks out,
a little solid phase that forms,
that's now sticking out into the liquid.
If you're familiar with the process,
if you've taken a kinetics class,
then you know that there can be things like solute pileup, right?
As this little nodule sort of sticks out,
you now have a different solubility of that alloy element inside of there.
And so it can actually be ejected out of that little protuberance.
And so it's changing its concentration, actually.
So the little grains that are solidifying
are now going to have a different composition
than the regions around them.
And you can actually see these in microstructures.
If you look at them, you can see gradients in the composition
that are strictly due to kinetics, essentially,
what's happening here.
So they assumed that that was what was happening here.
They tested it.
They actually did electron probe data.
We can sort of look on a two-dimensional map
what the composition is doing.
And sure enough, they saw evidence of model, you know,
elemental segregation.
And what's great about that is now that you can do that,
you can measure locally what the composition is,
and then using a simulation technique,
like we've been discussing, something like CALFAD,
you can say, okay, at this composition,
what should be the molar volume?
And then you can compare that to the region next to it,
which has a different composition,
and you can calculate the molar volume of that region.
And if they're different, then you can say,
well, that's our driving force for crack formation
because there's a difference in here,
and that's what's causing these cracks to form in some materials,
but not in others.
And surprisingly, when they included this now,
this difference in the volumetric chain,
and therefore the strain that would accompany that,
when they included that as a feature in their machine learning model,
along with all the other, you know, traditional empirical models,
they had a massive improvement.
The model went from essentially being not useful,
it's R squared was something like 0.5, which was not powerful,
and it jumped all the way up to a 0.8,
meaning that this was now a very actionable model.
This was providing really actionable, great information.
And of course, this isn't the whole story, right?
When you print these things,
obviously you care about thermal diffusivity and thermal shock resistance,
but I like this example because it showed how, you know,
there was a really complicated problem,
that of cracking and additive manufacturing,
how they took the sort of best approaches that had been done before,
they combined them with a new data-centric approach,
and then they modified that with their domain knowledge,
which was informed by simulation and calculation.
Man, it's just like a smorgasbord of really interesting materials,
science techniques to solve a challenging problem.
And, you know, they ended up with a really useful model at the end of it.
I think it really demonstrates the approach and maybe direction
that a lot of modeling is shifting,
where if they were to really get in and start modeling this, right,
do an exhaustive description and simulation of what's actually happening here,
it might take a lot of time,
and there's no guarantee that you're really capturing the true effects,
but rather they kind of shift to more of a statistical description
of what's happening,
and they're able to fill in some of the gaps
that aren't really covered with one analytical solution alone
or one investigation alone and come to a better conclusion.
So to me, like, these two case studies are fantastic,
and I completely agree with what you've just said,
but it has me immediately wondering that as an educator,
how do I need to change my perspective going forward
in terms of how I train my students, right?
It's clear to me that the collaboration between experimentalists and simulation
is not going to decrease,
and so the amount of training that we provide to our students
and the types of training,
it probably needs to skew a little bit more towards these simulation tools.
Yeah, I think I would agree.
I definitely see that as being kind of the direction.
At the very least, if it's not going to replace an experimentalist by any means
or someone who does that work,
that's still incredibly valuable and will be for the future,
but having another tool and being able to look at problems differently
and then take maybe data you have collected experimentally
and use it in a different way is powerful,
and it's going to enable us to solve new problems
as she is demonstrating in these case studies.
Maybe another thing that comes to mind is in the past,
I think in my department, we have a DFT expert,
and then we've got a molecular dynamics expert,
and I consider myself a machine learning or informatics expert,
but what you're seeing, at least in this most recent case study,
is that it wasn't like one of those techniques.
It was the fusion of those things.
So combining different types of modeling together will be important as well.
But I'm not sure that the problem is necessarily technological.
I think it has more to do with maybe the implementation and the utilization
and the combination of these different tools in order to solve problems.
And looking through this and trying to get maybe other perspectives,
NASA in, I think, 2018 published their NASA Vision 2040,
which is a roadmap for integrated multiscale modeling
and simulation of material systems.
And it's a great document. We'll link it in the show notes.
But they kind of talk about contrasting views
between modeling today versus the future,
and I think the way that they describe today
really kind of points to where they think it should be going.
So you have kind of a situation where the outputs of one team
are the inputs of another team.
So a component gets designed, and then it's handed off materials team
to maybe make it work or make it actually survive.
But that creates a really difficult problem for iteration
because maybe if material says this is not manufacturable
or maybe no material can be achieved with the necessary properties,
it has to go back to design rather than having some sort of cohesive framework
where these considerations can be taken at the forefront.
This really is a paradigm change.
If you've seen how a lot of engineering firms have worked in the past,
it was exactly like you described.
It was sectioned off. It was sort of cloistered by expertise
because it's hard. It is challenging.
It's a different way of thinking to bring everyone together,
and I think sometimes you might think, like, what does legal have to do with it?
Why should the lawyer be there at the beginning
when we're still designing the part?
And yet, if they're there, they might be able to point out something
that says, like, oh, hey, if you all are using that approach right now,
it's going to lead to X, Y, Z legal conditions,
and you'd be better off to skip that
because otherwise you're going to design this amazing part
and go through all the validation, all this work,
only to get it at the very end where they can say, like,
oh, this won't work, obviously.
So it's a change of perspective for sure,
and I really like that NASA is forward-looking enough
to sort of identify that.
Yeah, it definitely is an issue.
I think one of the last things is that today
a lot of our data is very static.
We take a material measurement at given conditions,
and that's kind of what we use.
We plug it into our models.
But really, many materials problems are very dynamic,
even if we just want to account for the fact that load and process
will create different conditions at different parts of a component.
Even over something's lifetime,
especially if it's in a high-temperature environment,
the microstructure that you fabricate it at
is not going to be the microstructure that's there
at the end of its life or even midway through its life.
And so trying to understand how material problems evolve
I don't think is being considered particularly well
and still is subject to a lot of this hand-off approach.
Things are, I think, moving in the right direction, though.
I think of, like, there's no dedicated conferences
towards this ICME approach, like TMS, for example,
or I just hosted, right,
the Artificial Intelligence in Materials and Manufacturing,
where we talk all about bringing data science into that,
starting to collect that data,
starting to think about merging the whole design process there.
For example, there's the Sixth World Congress
on Integrated Computational Materials and Engineering.
Is that coming up, or did that just happen?
That just happened.
I think that was in April, and that's hosted by TMS.
But that's not even that old, only the Sixth.
I assume it's annual.
Yeah.
So it's clear that the field of material scientists,
that we are moving in the right direction,
and it's promising, I think.
Yeah, for sure.
And I think having these more focus put on that
can help solve some challenging problems.
For instance, trying to link up multiscale models, right,
what I simulate during DFT or maybe phase field
or molecular dynamics,
how do I integrate that directly into a finite element simulation
rather than having to talk to experts, exchange models,
make assumptions, maybe make the wrong assumptions?
One of the other big issues is uncertainty quantification, right?
With experimental techniques,
uncertainty quantification tends to be
maybe a little bit more straightforward.
Take a number of measurements, you can get error bars.
It makes a lot of sense.
But how does that apply to models,
and how do you quantify that error, right?
There's some error that might come from the mathematics itself.
They found that between DFT calculations
there can be some error.
And how does that scale up
when you then apply those properties to some sort of FEA part,
even within FEA?
I think that's becoming more important
within the modeling community,
and there's more tools to try to assess uncertainty.
But having a very rigorous definition of uncertainty quantification
I think is going to be very important
as models become more important.
I think you're going to start to see a blend
between data science and modeling,
or at least more of a cooperation there
to more rigorously define this.
Something I'm keenly aware of
is the fact that materials that actually get used
are almost never like the phase-pure
sort of materials project version of things.
They're dirty.
They might have dopants in them.
They're very often composites, right?
And so adapting the simulation methods that we have
towards the real-world scenario,
the fact the materials are going to have these impurities
or secondary phases or other statistical anomalies
is a huge problem.
Right now, I'm thinking from a materials informatics perspective,
data science techniques typically are doing a really bad job
of capturing these rare events,
and so finding new techniques to improve that
will help fill some of these gaps.
Yeah, I think data is becoming,
especially materials data is becoming more valuable
and treated with a little bit more respect
than it was previously.
Yeah, absolutely.
One other concern, I think,
with the way that we model things right now
is we need a better understanding
of how materials evolve over their lifetime.
So, for example, if you have a model
that does a great job of predicting, you know,
strength at its first use, that's one thing,
but if you're going to drive that vehicle
200,000-plus miles, right,
how will the components change over that long lifetime?
How will it change under different use case scenarios,
under different environments, corrosion conditions?
A prediction of materials properties at inception
is not going to be as valuable
as one that compares the entire lifetime
and eventually its failure itself.
That's definitely a really good point.
To some extent, it's always like,
oh, okay, it works, it survives,
but I think there's also kind of a culture shift now
where there's interest in more sustainable products,
and I think part of that is moving away
from the sort of planned obsolescence,
and I think in trying to move away from that,
maybe it's unplanned sometimes.
You know, moving away from that
is going to be accounting for how materials change
and where they actually end up at the end of their life
and how we can maybe reutilize them.
Well, it's easy for me as an academic
to point to industry and be like,
oh, look at you guys.
If you'd only all get together in the same room
and have your entire design process integrated,
look how much better you'd be,
and yet in academia, we're sort of the worst.
We have all these different departments
who don't typically do a very good job
of talking with one another
or interacting or designing with one another,
and so clearly we have a ways to go
to start integrating curriculum far more broadly,
and I would say that with simulation
in particular in material science,
we clearly need more of it,
and I need to start thinking about
how I can integrate it in my courses
and as a department,
how we can change the way we think about things,
and that might be talking to more mathematicians
or talking to data scientists
or talking to other engineering disciplines
and doing cross-curricular training
in terms of courses or seminars
or whatever that may be.
Clearly a lot of work to do
from the education side as well.
Yeah, and I think material scientists
are well-positioned here as well.
I mean, the curriculum is generally multidisciplinary,
right?
We take some mechanical engineering courses,
some electrical, some chemistry, chemical,
and so I think we're already kind of prepared
to interact,
but I think it's really implementing problems
and projects that push them there
Yeah, absolutely.
But maybe just zooming out a little bit,
where modeling is right now
and how it's going to be used,
I don't know if we can really say for certain
how it's going to evolve.
I think we'd like to have that cohesive environment
that we kind of talked about
where materials design and component design
are linked and you can have
nicely linked multi-component,
nicely linked multi-scale models,
but really maybe a high-level perspective
is that each model gives you
a different view of a problem
and it has its own deficiencies.
But I like to think about it,
if you model things in different ways,
maybe you do have experimental data,
maybe you use machine learning,
maybe you have an FEA model or DFT,
it starts to complete this mosaic
where one individual tile
doesn't tell you the whole picture,
but having all of those tiles together
starts to give you a holistic view
and the true view of the problem
or how a material is going to behave.
And I think that's kind of the goal
is really to leverage models
and understand their weaknesses
and understand how we can use other tools
to complete those perspectives.
Remember that all models are wrong,
but some are useful.
And it's really about finding
how wrong those models can be
and understanding when they're wrong
and when to use them.
Absolutely.
Well, this was a fun episode.
We think that there's obviously
a lot more we're going to say
about simulation in future episodes,
but this was a great sort of example
to talk about how a variety of them
come together to solve real world problems.
So mega props to GE Research
for doing fundamental research.
There's just not a lot of companies out there,
you've heard us say this before,
but there's not that many out there
doing basic fundamental research,
publishing their findings.
So we really look up to them
as a leader in this space
in helping push the way forward.
Behind every GE innovation
is a breakthrough material.
First, it was the tungsten filament
that enabled GE to bring light bulbs
to the mainstream in the early 1900s.
Later, Lexan polycarbonate,
invented by GE scientist Daniel Fox,
ushered in generations of new plastics,
from compact discs and DVDs
to the helmet visors of astronauts
who walked the moon.
Through the latter decades
of the 20th century to now,
advances in nickel-based superalloys,
titanium aluminide,
the introduction of ceramic matrix composites,
and the first 3D printed metal jet engine parts
have helped propel commercial air travel
beyond the Wright Brothers'
first 12-second flight at Kitty Hawk
to some 100,000 flights
happening around the world every day.
Materials innovation has always been
at the core of what GE does
and central to the progress
our products have driven.
My name is Joe Vincicuera,
and I'm proud to lead
the Materials and Mechanical Systems
Technology Organization at GE Research.
Together, we are an interdisciplinary team
of aerospace, mechanical, materials,
chemical, and manufacturing
engineers and scientists
working to advance the state-of-the-art
for complex mechanical systems,
innovative system-level designs,
advanced materials,
and revolutionary manufacturing methods.
Every day, our researchers explore
the boundaries of cutting-edge technologies
that are poised to change the world.
Whether shaping the future of flight
or aiding in the transition
to a zero-carbon energy future,
our team helps GE stay
at the forefront of innovation,
enhancing our products,
and delivering for our customers.
If you're ready to see, move,
and create the future,
consider joining our team at GE Research.
To view our latest job openings,
you can visit us at www.gecareers.com
to learn more.
That link is also available
in today's show notes,
and thank you for listening
to today's episode.
Before we go,
we would like to give a shout-out
to Materials Today,
who's also a sponsor of this podcast.
You can visit materialstoday.com
to stay up-to-date on the latest happenings
in the material science field
and read some of their fantastic articles
that they've published.
You can also head over to elsevier.com
to find out more about their journals,
books, conferences, and related programs.
As always, thank you for listening
to this episode of the Materialism Podcast.
If you have questions or feedback,
please send us emails
at materialism.podcast at gmail.com
and make sure you subscribe to the show
on iTunes, Spotify, Google Podcasts,
or wherever you find your podcasts.
And if you like the show
and want to help us reach more people,
consider leaving a review.
It helps us improve
and it exposes new people to the show.
Finally, you can check out our Instagram page
at materialism.podcast
and connect with us
and let us know what new material
you'd like to hear about next.
And finally, we'd like to give a shout-out
to Alphabot and Colobite
for making the music for the podcast.
They both make a ton
of really cool synthwave music,
so go check them out on Spotify and YouTube.
Catch you next time.
Until next time.
The inventors of fire, electricity, magnetism,
iron, lead, glass, silk, cotton,
the makers of tools,
the captors of lightning,
the architect, the engineer, the musician
are all beneficiaries
of the materials of this world
and are bound only by their imaginations
and manipulating those materials.