 Good good afternoon everyone My name is Arvind Rahman. I'm the senior associate dean of the faculty in the College of Engineering And I'd like to welcome all of you to the third Purdue Engineering Distinguished Lecture in this academic year the Purdue Engineering Distinguished Lecture Series invites world-renowned scholars and professionals from all around the world To Purdue to really engage with faculty and students in the grand challenges facing their disciplines In addition to this lecture we also engage in a panel an interactive panel with faculty and students and to introduce our Distinguished Lecturer for today. I'd like to invite the head of the School of Materials Engineering Dave Barr. Thank you Arvind and Please save your applause for our speaker because all I do is get to invite people So we're very very pleased to have with us our own alumna and we believe only NAE MSE alumna from Purdue Teresa Pollack Dr. Pollack is the Alcoa Distinguished Professor at in the School of Materials at University of California Santa Barbara and if I Lit if I went through a litany of all her awards there would be no time for her to talk She's the past president of TMS. She's been a faculty member at Michigan and Carnegie Mellon She's extremely distinguished and has really helped shape a lot of what we do in materials around the country And we really enjoy interacting with her and I'm glad to have her come back So without further ado, that's okay. I will turn it over Teresa and I'm really Please give her a warm round of applause Dr. Pollack And thank you very much Dave. Can everyone hear me? I think there's one little part of my biography that all of you as students should hear so I came to Purdue as a first-generation college student I walked into the College of Engineering. I had never met an engineer before and and I actually walked around to all the Undergraduate advisors and asked them what happened in that department and you know with no external prejudice I picked materials and and I was a co-op student and I quickly became in-state because my Co-op salary easily allowed me to pay my $470 tuition bill, so Some things have changed a little bit But I will always be eternally grateful for all the doors that Purdue opened up for me in the beginning So I hope all of you take a same advantage of all those open doors So today I want to I hope talk about a few different subjects Data additive manufacturing advanced materials hopefully This will be of broad interest And so what that means is probably we won't be able to dig too much into the details of some of these topics but I'll feel free to ask Questions of detail or try to get with me sometime tomorrow So before I start I'd like to First acknowledge a lot of people There's a big group on the left who are very talented students and postdocs This guy starred here is one of the special ones He's now on the faculty here Mike Titus and I'll let you know about him a little bit later And then a large group of collaborators from various places listed on the right and So today I'll talk a little bit about materials in the context of additive I also talk about an instrument that we've developed for tomography and and all the data issues associated with that and Then finish with a few final thoughts So of course Advanced materials are something that many people in this room are interested in I've sort of elisted a few advanced engineering systems or Systems that don't rely on advanced materials from my own neighborhood apple watches I have some alums there thinking about gold and fun alloys like that fleer I can watch the SpaceX launch the rockets from my deck. That's always fun We have an alumna of our department who who's done some very creative things with plant waste for coatings to extend the life of fruit and flowers and vegetable stents to batteries and and so all of these things have advanced materials at their core and we're always hunting for better materials to put in these systems and That's always Harder to do than you think and it's not because there isn't a lot of potential You can think about certain elements on the periodic table and sort of a nearly limitless collection of soft materials and Molecular materials that you can make from them you can think about sort of the center of the periodic table and If you start thinking about mixing things together, you know, there's sort of like five million quaternary systems About which we know very little on the whole and and one of the interesting topics that's emerging and in crystalline materials at the moment has to do with Mixing elements together in equal proportions so that the material doesn't actually know what what its base material is We think about aluminum base alloys and nickel base alloys But what if we we completely mix up the material and get new properties and so so there's still a large space to Explore, but the problem is really how do we do that faster? How do we get materials to the point where we get them into systems much more quickly? And so this is something I've spent a lot of time on integrated computational materials engineering and materials genome initiative and What this really attempts to do is bring together computational tools You know more rapid high-throughput experimental tools Data and and bring these things in together in a way that we could be more predictive and be that way More quickly and so in some sense Additive manufacturing now that it's sort of come to the point of Being an approach that people want to use for actually making things rather than prototypes is a good platform for thinking about what The issues are in bringing these things together So 3d printing This is a recent industry Survey and it sort of asked the question Why are you interested in 3d printing in other words? Why are you investing in it? And and so if you look here some people think it'll just decrease cost Others think it'll let them use innovative materials It'll allow you to print metals without wasting a lot of material Medical devices a lot of interest there and if you sort of look at the shift In in interest from 17 to 18 it's sort of okay We've got a lot of the polymeric processes down But now we want to think a bit more about the the metals and so I will focus on the metals part of this today so from the perspective of of metal additive manufacturing it tends to be Applied to materials which have high melting points, which means we need high densities of energy And so this a naturally leads us to processes that either involve lasers with powder beds Perhaps electron beams Some systems lasers that have powder feed with them But in the end much of this relies on the availability of powders and the ability to Control their size their shape their surface and what happens to them in the process Now 3d printing I want to talk about the fun sciency things today There are many many other issues and it's interesting how little attention has been paid to some of these things I Looked around a couple months ago for information on sustainability and 3d printing and there's about three or four papers Maybe there's some in the last few months But but what little information there is at a high level thinking about markets and supply chains and labor and so forth seems sort of positive from the point of view of pay off in terms of reduction in energy costs In the context of you know, how much we use energy for manufacturing potential reduction and co2 but most of this potential really is is Exists at this the sort of more high-end Industries so aerospace medical devices tooling but there's certainly a lot more thinking to be done in these areas as well as in qualification and things of that sort Though I will not focus on that today So what I do want to talk about a bit as I go along is this whole idea of printable alloys And so I start with this this sort of bit of information So think big multinational corporation G what do they do? They buy all the companies that make printers in this space and then they also buy up the powder makers And then they tell you you can choose from this list Okay, so we have tens of thousands of engineering alloys and and there's a list of basically about seven or eight to choose from here and so That's really not good enough and so why and so I'll talk some about the why and And so of course some of the early things we've done have been with one of the available alloys And this is a nickel-iron alloy that the composition is listed here though the details won't matter too much today Now the other reason why we want to to be able to print some of these advanced metallic materials is the idea that we can print in and control the properties spatially and so there was this Lovely effort at DOE a few years back Ryan DeHoff and Mike Kirke who we have been collaborating with Where they did something very clever to help their research programs And that was to print the 718 material in such a way that most of the area of the build here has a color red which conveys an orientation of the grains here and what they were trying to demonstrate is that they could they could control this to the degree that they could write DOE and So sort of I mean if you could really control this these would be straight lines and abrupt transitions And so the question is you know, why is this so difficult and and and what can we do about this? And so they thought very carefully about all of the things that people typically think about in additive What's the power? What's the velocity? What's the spacing of the lines? And so That of course is important, but that's not the whole story. I Think I have to go back here The whole story of course is is complicated and what is going to take us a long time to sort out this is basically a video from the Beeline at APS where we have a substrate here and then just a layer of powder on top and you're going to see the laser Deposits some energy and you can see that it's actually a pretty violent process And there's powder particles flowing around and there's obviously fluid flow and the surface matters And so so there's a lot of things to take care of here And of course this is just a spot this laser is not moving in this scenario and so You know the question is when do you have to take care of all of the details of the motion of these particles and when Can you just think about this more as sort of a you know hundreds of microparticles? Microns to millimeter scale Solidification event I'm going to melt and it's going to solidify Maybe Yeah, so so as I just said, I'm not going to belabor the point But you've got this melt pool. It could be sort of a few hundred five hundred microns Maybe even even in some processes up to a millimeter And so we've sort of got this RV that's kind of a cubic millimeter and inside that cubic millimeter We have to think about nucleation fluid flow. What's the grain structure that develops as a result of this? What are the gradients and velocities of this interface? am I going to develop residual stresses that lead to cracks am I going to vaporize and train powder and so forth and so To really fully understand this we're going to need a lot of suites of diagnostics and a lot of 3d data and and so I'll talk more about the data problem and really sorting this some of these things out So the first part is if you want process data You can just imagine that you have a laser scanning it. Let's say a hundred millimeters per second We want tolerances of a hundred microns We're we're Going to have to gather data at the rate of 50 kilohertz From these different sort of sensors and if you have say a three-day build That's 230 gigabytes of data great and so obviously there have to be workflows to to Gather this data and put it in a useful form in order to connect it to what's going on Inside the melt pool in addition We're going to need to know a lot about the structure that develops And this is something that we've been working on now for a long time since we've been in Santa Barbara and I'm going to talk today about Tomography approach that we've developed that we call try beam tomography And so so this is a lot about 3d data And so anytime you think about 3d data you have the issue of acquisition So I'll tell you about the new platform Reconstruction I'm not going to have a lot of time to talk about the details, but this is a Challenging interesting and Not very mature Part of the 3d story how we take multimodal data and put it all in the same box in order to learn something And then of course once we do that we may be too tired But we have to remember that we have to do something with that data learn from it And so, you know, what do we learn about residual stresses dislocations nucleation and how does that help us? Design new alloys, so I'll say just a few words about all of these things and then I also want to say a few things just about the data itself and Sort of some new things that we're trying to do to solve some of those problems with our computer science friends Okay, so first the try beam system So what we've done is basically taken a focused IM beam microscope built by FEI from a Fisher manufacturers of these and And so these are usually called dual beam fibs because you have an electron beam and an ion beam And what we've done is added a femtosecond laser. So these are titanium sapphire lasers operating at 780 nanometers and so what you see here is a stage the the laser the lens bringing in the the femto beam and and basically the column and the other Analysis systems and I should say if you let me come back to that The reason for doing this inside of the microscope is that we have an EBSD detector that you saw sitting at an angle there And so this gives us crystallographic information of this sort we have the ion beam which We use in conjunction with the laser beam to ablate the material layer by layer and so the femtosecond laser is basically a means of ablating layers that are in principle as thin as 15 nanometers, but in practice more like a half micron to a micron and Basically, you can ablate that material across a square millimeter in seconds And so if you happen to be familiar with fib tomography Basically, this is sort of makes it a million times faster And if there's a little bit of cleanup to do then the ion beam comes into play And so for every layer we can then gather crystallographic data by EBSD We can gather image data and we can gather Chemical data by EDS and so you'll notice that the sample is tilting around toward these detectors And we all since you have different detectors. There's always going to be issues of distortion and therefore lots of things about image processing and so basically we're doing this at glancing incidents and So what that means is they don't have to worry too much about tuning the power of the laser It just means that you're automatically using the tail of the Gaussian beam to remove the material And so here's a movie that just shows a bunch of frames of us going through a piece of additively manufactured material Yeah, I know the people in the audience who do additive. Yep, there's that lack of fusion defect We were not trying to look for those either And so you can see pores and so forth and and this is sort of the the sample workflow You put it in the laser slice seconds They've cleaned up if you need it is Maybe 600 seconds and then chemical data the expensive data is getting crystallographic Data because we have these electron back scattered Systems which have to scan and they're not very fast and so This data set that you're seeing here took 60 hours to gather We have 1.5 micron voxels and it's 25 gigabytes and So if you can do this in 60 hours that means you're gathering these gigabyte terabyte data sets in principle several times a week except everyone knows microscopes break down so there's some pauses in between and Then we have this issue with you know how to deal with all that data. I guess this doesn't like to move I should say that because Maybe I should go back a couple slides here and explain this except we'll have to get past that movie again So the reason for using femtosecond lasers is sort of shown by this simulation here and basically What we're doing when we ablate with ultra short pulses is During the time The beam is on all you're doing is exciting electrons and then by the time these Rain back down launch a shock wave and and lift off an ablation layer Which is liquid like the beams off So you're not doing Collateral damage to the substrate and so I show a bunch of materials here that this is the ablation direction would be Vertical and so what you can see is there are these laser induced periodic structures Which may or may not interfere with the analysis it depends on the material But not much else going on until we get to very soft materials then we do inject dislocations And so these lasers are good at minimizing collateral damage and allowing you then to gather all those modes of information without any further surface preparation And so I should say because the ablation threshold is sort of Distinct and similar for many materials that this is almost a materials blind process And so this is sort of a few samples from the library if you will super alloys Thermal electrics geological samples ceramics composites that are very different carbon fiber epoxy composites and so the technique is quite generic and Actually, you can buy one of these now FEI thermo Fisher just announced that they're selling The commercial version of this so please go get one and help us develop the approach Okay, so as I as I said we're gathering all these different modes Of data and so here's just some examples But what I'll say is that the expensive stuff comes with EBSD and as those of you who gather this kind of data know it's very noisy and What what typically you do is use huff transforms quickly analyze the patterns and then throw away the data Unfortunately, that's that's not the best approach And so you really should save all these patterns in order to just sort of optimize that part of the data And so this is a this is a challenge for us And so the reason for saving it is there are other approaches That are being developed. This is a dictionary indexing developed my Marked-a-graph at Carnegie Mellon and basically what it does is is calculates an entire library of patterns and Then you can use of the dictionary so to speak to compare to your very noisy data And and get back much more information than you could from sort of the conventional approaches There are also some more recent Spherical harmonics indexing which even speeds this process up more and so there's there's a lot of Development in terms of trying to speed up every little step of this process Now I said we have to do reconstruction and so For those of you who may just be getting started there's a open-source program developed at the Air Force called dream 3d and so what this basically does is gives you sort of a step-wise workflow that takes care of all of the steps of Reconstruction and so This is the way that one goes down the path of getting back those 3d data sets I showed you now This is also a bit of a painful process because every one of these steps may have parameters that it asks you to set What's the threshold for this? What's the cutoff for that? And so When you have data sets that are gigabytes in size, these are not things you can answer manually, right? You can't go look at that slice and make a decision. So everything has to be automated and rigorous and so With that realization we've been working with some colleagues in computer science who actually have developed more automated web-based approaches for other disciplines of imaging to make sort of a platform that we're starting to call bisque materials and so what this basically does is takes every piece of data in this case mostly image data and in HDF5 format everything gets its own web address and so then we can paralyze all those Pipelines that we might be developing to ask. What's the best answer? So let's do 6,000 Pipelines and then figure out what works the best and sort of automate the process We can annotate and hopefully then learn things from databases that sort of automatically build themselves We're not very far along on that We can visualize and then importantly we can Share access and allow people to bring say their crystal plasticity computation To our data set rather than shipping all that data Inevitably on a hard drive across the country and so so what we're trying to do is is sort of If you will open the library and and allow people to try things There are still a lot of infrastructure to be developed here But we were aspirational in that regard and so as I said this this is what we normally would do to Reconstruct this data set try something and dream 3d dump the data look at it pair of you go back and change the parameters Do this a million times and now we can just send this all out and then get back the final result And so that has definitely speeded up the process Okay, so back to additive I want to talk a little bit about the motivation for having these 3d data sets as a part of the additive workflow And so just to give you an example for that a 718 material I'm going to show you three different 3d data sets and so one is a spot melt and Another is a spot melt on a sample, which is slightly different size We're mostly looking at the structure late in the build so that we can know what happened recently rather than you know a hundred layers ago and and you'll see that that these these Samples have not been processed in ways that are all that terribly different and So that's sort of been the conventional thinking about additive for many years right if I know the power of velocity if I know these parameters then then everything's good and Here's just a view of the surface Obviously when you use a spot melt you have different looking structure at the surface and you shouldn't be surprised It's going to be different at the subsurface and so recall that we're trying to get to the point where we control structure So this color scale is just basically orientation of the crystalline grains within and So here's an overview of those three different Builds so we have data sets that you can see that are sort of half micron to a micron and In the case where we have these Spots on the top what has happened is we built to a steady state and we've gone back and just put in the The first few spots that you would have in the spot melt process and then looked at those regions and reconstructed them and so This is kind of an overview of the 3d structure and so we're seeing grains here and and All material scientists know that the first thing you'd better worry about is the grain structure the grain size the distribution of grain orientations because that is extremely important to properties and And so these two slightly different blocks have completely different structure even though the processing parameters were basically the same This one where we're rastering looks completely different in that it's almost a single crystal here except for what happened in that spot at the very end and so this is just the the sub volumes from the bulk and So the question is well, what do we learn from all of this? And so as I said, you know, there are good reasons for picking particular regimes of Velocity and speed and and that's because if you go too far in power You get a lot of voids and keyholes and and defects of that sort If you go too low in power, then you get lack of fusion you haven't melted the powder so forth So so obviously there is a process window in here But should you be here or here or is here really the same every time and the answer to that is no so as Matt Kramer would tell you Of course, that's because you really have to look at the details of what's going on at the milk pool and and really very Carefully understand well, what are the velocities at the solid liquid interface and what are the thermal gradients at the solid liquid interface? and that's because you know, it's fairly well known when we grow single crystals and and control other solidification processes that there are these transitions from Very planar growth that happens at very high thermal gradients and low velocities to columnar structures which Oops, oh that was the wrong way so to columnar structures, which are more like this to equate structures, which are more like this and so If one wants to control these things This is these are the parameters that need to be controlled And of course most of these parameters are are established with very Unidirectional processes in the milk pools more complicated than that. So there are More things to worry about in terms of characterizing it in 3d and connecting it to these sorts of ideas So here is the Raster melt blocks are the one that's that's very column they're almost single crystal and so this this is just 3d Slice through the structure and if I go back you can see Oh, that was bad No, that's forward You can see where the solid liquid interface Melted back to well, maybe not. Okay, here we go. So you can see there's a very distinct line That makes for easy image processing now for us to really pull out that milk pool and see what happened Inside of it. And so again, you saw this a few minutes ago There are some grains here which are oriented but but in this Processing scheme we've we've almost built a single crystal and what about all these grains on the top and and so This is the kind of information that that people have been looking for for a while In order to build models for the solidification of the pool and so what we know is the volume of this pool because it was very clear we can actually count the number of grains and The volume of the milk pool that is eventually Occupied by those grains and go to models for the column there to equiax transition To supply parameters basically that were never available before so the nucleation density can be measured directly and then From the volume of the pool Occupied by those grains that nucleate on the surface then we can actually put Better parameters into the models which now predict whether or not that is column there or whether or not we've crossed Into these equiax domains and so using these improved models it's still not the whole answer because there's that very complicated fluid flow and The question is under what conditions does that matter and so here is the model with and without fluid flow and then With and without low and high sulfur which affects the surface tension which affects the Marangoni flow And so this now gives us a way of saying alright starting at fraction solid zero I should have been Column there which we were and then I hit Enter into this mix zone and certainly there are conditions late in the process which should give me grain nucleation more equiax structures And so these are early efforts, but certainly The sort of this is providing sort of information you need to build that modeling infrastructure Now I remember I said we want to turn this on and off. I want to go from column there to equiax. Is that really possible? and so If you look at this data set which is more equiax looking this is 610 slices 2.6 terabytes And so what we're going to do is look at this just this solidified pool and ask the question Where did all of these grains come from? Did they nucleate in the pool under the conditions that we applied? did they grow from the layers below due to to Basically re-nucleation of grains of the same orientation so we can take Slices through this and and ask where is the origin of each of these grains in the layers below? Below me so each build layers about 50 microns so we can slice the data set at intervals of 50 microns and look back in time So to speak and so slice one Slice two oops This is coming. Okay slice three. Never mind. We'll just do it here So and so you're seeing in each in each layer what grains? Existed in the layer below it, okay? And so basically what you can see here is that 10% of the grains that were 10 layers below me Still exist in the top layer and so the idea that you can just on the fly change everything at once and control it Might be a little bit ambitious or we're going to have to try to do things a little bit differently If we're really going to achieve that and so this history dependence of the layers as we build them is important Um This is now that other data set that is a bit more columnar the reason for that really has to do with the fact because the samples bigger you in the end end up depositing more energy even though you're using the same scan patterns and so it tends to Basically have then different gradients and velocities and so as I showed earlier one of the things that you sort of get automatically from this is is A data about defects, but if you're tracking grain structure something you'll see is that these Grains once you go past the lack of fusion defects have to re-nucleate And so now you've disturbed the whole structure and that defects actually a lot larger than you thought and so again Those are things that are hard to discover by other means And so here's a view of the lack of fusion defect, but here's the whole area involved in the after the defect formation okay, so then one other issue that comes up a lot in 3d printing is of course the business of residual stresses and That builds a lot of Problems into the process and so here again. We have this rectangular melt versus this smaller Sample and what we're looking at here is the reference misorientation So basically asking for every grain how much misorientation is in that grain and and so 10 degrees is red and so for those of you that Really do mechanics 10 degrees of misorientation stored In a materials a huge amount and so that's a crack waiting to happen And so now we can start to ask the questions with this 3d 3d EBSD data You know, why is it in a particular area? What can we do to fix it? And again, you'll see that these two Different Samples with exactly the same machine parameters have hugely different residual stresses in them And so it turns out we always think about all those things being stored at the grain boundaries Solidification is a tricky business because you have You have liquid left late in the process and so it's the liquid left At the end of the solidification process that ends up taking up a lot of the misorientation So this is just one single grain from that data set and so As I said earlier, this is a crack waiting to happen, right? So this is This is that alloy printed under less favorable printing conditions So you can see a crack here and you can see just adjacent to this crack There's a misorientation of about eight degrees. So that's you know when things start to get tricky now. Here's an alloy Marium 247 it's one that flies. It's one that everyone would love to print because it's a lot stronger And if you try to print that it's just full of cracks And so the real question becomes what do we do to design alloys that are less prone to this? And So to go to your solidification class as I said a minute ago We have to basically we have changing liquid in solid compositions people think of additive is sort of a Rapid solidification process, but in reality, it's almost always dendritic solidification for these sorts of materials I like to say the closer it gets to casting the better it gets and so these sorts of ideas about solidification apply under most of these conditions so these dendrites Develop because we're we have a liquid composition that's coming down these curves and the shapes of this Liquid is surface or what matter? Okay, so so going back to that residual stress now We can finally do something quantitative with this sort of data because we have full 3d Misorientation data so taking the approach of our slenderson parks this is work that Student Wyatt mitzvah with Irene byerline and myself are doing and so what we're able to do now is Basically get a full nigh tensor. So all of that misorientation built in To the process can be analyzed in the context of Geometrically necessary Dislocations and people have done this in 2d, but you're basically missing a lot of elements Of the tensor when you when you do this in 2d and so what I show you here are contours of gnds and This Basically is a scale 10 to the 12 per meter squared And so you can see within this one grain. There's a region at the center of the grain that has a lot of dislocations built in and This is basically a Sequential set of slices through this grain and so there are regions that sort of persist Which are not necessarily associated with grain boundaries And so I show you a few other grains here and again now we're partitioning out the geometrically necessary dislocations to To regions of the grain and we're partitioning them to different subsystems There's a tendency for them to be more edge than screw interestingly and and the densities are somewhere in the same neighborhood as as what We would expect to have Just during a conventional bridgeman growth of some of these materials Okay, how much time do I have I lost track of when we started Okay, oh good plenty of time So I now want to say just a few words about this this challenge really so There's lots of There are lots of issues with getting Material which is free of defects and so we sort of know something about the parameters that we have to Use in order to avoid these defects But there are still lots of things about the way these materials solidify That result in in crack so it's inherent to the material And so we want to be able to design materials that are not so prone to this and So there are a couple of different ways of doing that. Let me jump ahead again and So here's sort of the way we were thinking about it. So so one way of doing this is is to actually Enhance nucleation and make sure that we don't end up with these big long channels of liquid which turn into cracks and so the approach to doing that is to think about Putting things on the surfaces of the powders that enhance nucleation and so this is not a new idea particularly from from the point of view that in casting people have been doing this sort of Functionalization if you will with inoculants for a long time But the the time scale and the link scale are different And so you do have the opportunity to use the powder surface to keep things where you want them and so This is one approach and and then the other approach is is What we're plotting here is temperature fraction solid These materials that crack tend to have a long tail here with lots of liquid left in those channels So what we'd like to do is design a material which has a different solidification path And so we're we're sort of getting started on both fronts this is the PhD work of Hunter Martin and Brennan Yohata who's still a student and so This was Hunter's idea And he came to me and said can I do this for a PhD thesis and will it work? And I'm like, I don't know if it'll work, but it sure sounds interesting. So let's do it. And so he used nanoparticles and attached them to the surface of pure aluminum and 70 75 aluminum And and also has tried some things in other materials, but really aluminum has been the focus here And and so what he's been able to do is is greatly refine The grain structure of aluminum in a way that mitigates all the cracking so so here's pure aluminum Note that the link scale here. You've got these big Crack-prone column there grains and here's aluminum with Not a tantalum inoculate, but a tantalum intermetallic isn't as an inoculate And you can see here the grains are highly refined to four microns. And so this this inoculation process actually There's still a lot to understand because nucleation is always difficult, but basically What what we need Is a field which has liquid? that contains an intermetallic particle Which has good matching with the base material such that it likes to Nucleate solid and so it turns out there's a very precise amount in the case of the tantalum Inoculants that will get you to print grain refinement and so What this means is that hunters for the first time able to print very difficult things like 70 75 aluminum of interest To the aerospace industry and so there are many many things to learn about this area yet I think it's a whole interesting place to think about The other on the other front sort of tuning the solidification path this goes back to Mike Titus and And and some of the early thinking he did about designing alloys in the cobalt aluminum tungsten and Nickel space and so so this is sort of an interesting story independent of additive, which is it was discovered that there was a Intermetallic phase that was useful in this ternary system I mean, it's such a common system cobalt aluminum tungsten. We didn't even know the phase diagram when we When she done Japan discovered this phase We immediately got excited because it has a structure that looks like conventional super alloys that fill our aircraft engines and so We spent a lot of time understanding the system Trying to understand how to control it understanding the solidification and one of the very interesting things about these cobalt nickel alloys So here's an example of one where they're about half about 40 30 7 7 I should have put the composition here, but nevertheless these these alloy solidifying a way such that we have very little segregation in the dendritic structure and so associated with that is this very gentle solidification path, which is Presumably not prone to Those liquid cracks that that a lot that developed during the printing and so We thought it would be a good idea to try printing some of these and so here's just a laser track We have a high angle boundary going into the melt pool here. You can see there's no cracks We have an eb electron beam printed blade. This is now a thousand degrees C preheat I should say all these materials have very high volume fractions and precipitates like that one cracking alloy had and Then here's a laser powder bed Blade with no preheat. So it looks like you know if you think hard enough. There are Probably a lot more alloys That are possible If you if you think about all those details so With that. Oh, that's about 45 minutes just about right So just a few words to conclude This this expanded suite of computational experimental data tools, I think if integrated Can bring us a lot of new materials For additive in particular I think sort of the mesoscale structure with nanometer scale resolution is important To understanding these things But fortunately and unfortunately it brings us a lot of data and 3d data sets. So there's a lot of challenges there and Processing this data and reconstructing and and and more importantly extracting information from it We need help from computer scientists and signal processing people and and and other communities who who Already actually have a lot of the tools. We just don't know about them The the femtosecond lasers brought us a new tool to the to the suite And I think one that's that's good for the mesoscale and and as I said the the multimodal data problem is is one that's interesting but still but still quite challenging. So with that I'll be happy to take some questions and thank you for inviting me Okay, here we are. We're good. Okay, so questions Hi, hi Teresa, hi welcome to Purdue I want to I have a small question about the residual stresses and the level of plasticity the end use in the additive manufactured metals. Yes, is that a significant plastic deformation that? has been observed in the as Deposited or manufactured materials. Yeah, so it's this is sort of something. It took me a while to figure out that very often in in these Systems that are are printed people will give them a post Heat treatment and the material completely recrystallizes or or maybe it doesn't completely recrystallized But it's it's of the level that can drive Re-crystallization which to me seems a bit dangerous because now you've got these very inhomogeneous stresses that are driving a process that may or may not happen and so certainly There's a lot more work to be done on the mechanics of that which are very difficult because you're near the liquid We don't know the properties very well But but the stresses are indeed very large even if they don't look to be You know the plates aren't distorted and so forth and so the key is to try to get the materials into a configuration as they solidify to Mitigate that rather than waiting till and you know a few processing steps down the line At least that seems sensible to me We need student questions Hello, I saw a note in the call about the AI artificial intelligence Used in the analysis of these results. I'm just curious other than just comparing the printed material to the per design Specifications, what do you hope the AI will achieve will achieve? You know one of the things that it's been We've had a little bit of a collaboration with Google on just trying out some algorithms When you when you get images From all these different detectors Then the question is if you're trying to really get sub micron resolution All of the distortions really need to be accounted for and so there are lots of different approaches to that and And one of them has you know, we can use thin plate splines Which basically? Have an affine deformation plus some bending We can also use some of these evolutionary algorithms, which I probably have here somewhere Yeah, here's an example of how we've used that and so Again, it comes down to What's the best way to to correct these things and and use these existing algorithms that already? Have been developed to go through all of these different steps To to to bring the data into one box and so there's that one whole image processing part of it The other part is in the bisque system once we get all that 3d data there We'd like it to have more automated sort of property prediction approaches and so we've been trying there when the process of building Some tools and to that system which for example would take a composite material and in an automated way extract information from the 3d structure and Predict properties and so trying to learn our way from some of those 3d Microstructures to properties and so that is Certainly much well less developed But I guess our philosophy is if it's something that we already physically know Then we should do it that way if it's something That's just The physics are are probably not worth chasing like you know Detector physics and correcting distortions then then some of those things should be used in that way There's there's lots of possibilities, but but the other place where it might be helpful is with the EBSD patterns There has been a little bit of effort at using machine learning to To analyze those that looks less effective than say the dictionary so far students students I Thank you for your talk was really interesting I have a question regarding the coding of the narrow particles to help you on the solidification so do you have a particular concentration or Parameter that you know how how much now particles you need to That's hard. So that part belongs mostly to HRL and is still a bit empirical and so I guess I can't say lots about that partly because we don't know and partly because they don't want to know What you to know, but but I you know, there are many different ways of doing this and I think this is a big area which Requires some more attention and so there's I I see a lot of PhD theses there for someone But but I think we still haven't brought enough chemical engineering to that problem And so there's some opportunities there One down here Oh really, where are they? Down the hall So we're going to go check to see if they have okay, we were out of Ever question regarding again on the nanoparticles, where do you see the distribution of the particles after the process? Oh, oh, so maybe I didn't say that clearly enough they they go into solution and So So then the question quickly becomes well, why should that work right and and so It come if you do a lot of calculations it turns out that that you want them to sort of Be the dissolution is actually pretty fast But the real secret is in Controlling when they come back out and so so when you're coming through this liquid field when those particles Renucleate you basically want to be coming through here fast enough that they nucleate But they're not really transported anywhere so that they're actually around To to help nucleate those grains this was totally counterintuitive to us at the beginning that's not what we were thinking would happen and so So the secret is you know when you use these things in a more conventional casting process They come back out But then they're subject to fluid flow and they get transported off somewhere else and then they can't do their Nucleation job so it has to do with not only Putting particles on the surface that will give you a Nucleant that re-nucleates at least in these aluminum systems I Imagine there are oxides and other things that you don't want to go into solution We're working on some of those now, but but in this particular case It was totally counterintuitive that it's a matter of having that re-nucleation process Keep enough particles around so that you can have the grains all nucleated In the right time frame as as all of that happens. So Yeah, it was a good idea that turned out differently than we thought Professor I have a question. So for the materials grand challenge, you mentioned that the level of functionalization of high-strength Alloy powders, I'm wondering what a kind of service properties is this Functionalization is trying to change primarily well, I mean So there's a whole big literature on Inoculation of materials. So that's a natural place to start so they have principles about You know how how good the lattice matching is how good is the wedding? So there's a number of properties like that which are usually not very well known that you're aiming for and so in this case what we actually did was had Citrine informatics go through their database and and just look very very widely at Intermetallics that had particularly Good matching with the base structure To to narrow down or maybe find some new inoculants to use in this case And so that's another place where sort of their machine learning approaches came in Moderately handy because it gave us some new clients that work that we really didn't expect to work and so Again, there's still much much Left to do here But the idea that you have a giant materials database, you know at the materials project or at Citrine That you can use to just say what's my best guess about this property and how it will Make this intermetallic compound useful or not useful was sort of the approach we were trying to take Okay, and with that we are going to wrap up the presentation we have a 15 minute break because our speaker deserves a break