CDFAM NYC 2023 · New York · 14–15 June 2023

Generative Design & Digital Manufacturing at NASA Goddard

Listen as a podcast

Abstract

This recording showcases a presentation by Ryan McClelland from NASA Goddard Space Flight Center at the CDFAM Computational Design (+DfAM) Symposium, held in New York City in 2023.

Presentation Title: Generative Design & Digital Manufacturing at NASA Goddard

McClelland’s presentation delves into the specific applications of generative design and digital manufacturing at NASA Goddard, illustrating how these computational methodologies drive innovation in the creation of advanced space technology.

Drawing from his comprehensive experience, McClelland provides case studies that highlight the transformative impact of computational design and digital manufacturing in overcoming unique challenges in the space technology domain.

This presentation is particularly suitable for professionals, researchers, and academics in the fields of computational design, additive manufacturing, generative design, and space technology. Individuals with a vested interest in the applications of computational innovation in space technology development will find this presentation exceptionally informative.

Please note: The CDFAM Computational Design (+DfAM) Symposium is a prominent platform for the exchange of knowledge among experts in computational design and additive manufacturing. This recording is provided for educational and professional development purposes.

Transcript

From YouTube’s automatic captions, lightly cleaned; expect some errors. Each timestamp opens the video at that moment.

Read the full transcript · 4,237 words

0:06 Okay, okay, I’m gonna talk about evolved structures how NASA is using generative design and digital manufacturing to make lightweight spacecraft parts really quickly, you know, things like these organic shapes you see there on the left and right. I’m Ryan McClelland, I’m a research engineer at NASA Goddard, and, you know, through this work, since it’s kind of like caught into the zeitgeist of chat GPT, it’s gotten some media attention. I’ve learned a couple of kind of funny things about journalism one is they’re going to come up with much better titles for your work than you could come up with, because they’re, you know, they’re in the interest of getting clicks, they like to use the word alien language smanguage I kind of like that one. It’s a little bit of an overstatement that we’re not yet using it to build spaceships, but we’re thinking about it, and, you know, a word of warning, if you, you know, if you work for a space agency and you say the word alien, it’s probably going to get quoted, so you got to be really careful with that one. Foreign.

1:07 So the context of this work, this was developed with internal research funding at Nasa Goddard, and the goal is to create and Infuse a broadly applicable process to rapidly develop lightweight spacecraft structures. So it’s really taking all the great work that the people in this industry have done and, and applying it, and getting it in space quickly. And the method that we’ve done is to build and test parts for a lot of real NASA applications I’m up to like 36 applications at this point, which really gives you a good idea of, of what the tools are and aren’t capable of, and status. You know, typical metallic structures that we build at Goddard, or at Nasa, and, you know, NASA is building like thousands of bespoke structures at any point, small stuff can now be automated, and, you know, humans should probably largely stop doing it. You know, we can go from, if it’s a fairly simple part with just a couple interfaces, from knowing what the requirements are, to having a part into, you know, ready for Fabrication, in just a couple days. And we’ve demonstrated, created this by testing on applications, and it’s being applied to NASA missions that are going to go all over the solar system.

2:12 The goals of this presentation: I want to share lessons learned about the Practical implementation of generative design and digital manufacturing, and reduce the development time and cost by 10x, and, and for us it really is, you know, having something that takes a month and having it take a couple days, that’s, that’s the goal. And engage this community, because I don’t think I’ve ever given this talk to a group that is more capable of giving great feedback, telling me what tools are coming, so that they can be really implemented into NASA missions.

2:45 So the inspiration from this work you know, over the pandemic I was really, didn’t have as much to do as I do now, which is, looking back, was kind of nice. But, you know, I watched a lot of Science Fiction, and things like the expanse, which is great, by the way, if you haven’t seen it, and I got looking at that and, but I know how the sausage made, right, like I know that the space station has seven people and it costs 100 billion dollars, so, like, that doesn’t scale to the kind of structures that you see in science fiction, and that kind of like bothers me, because I’d like to see that future where there’s lots and lots of people living in space. So it got me thinking about how, you know, AI can be used to help do my work.

3:22 You know, I’m, I’m a structural engineer, I’ve worked on a bunch of stuff like, you know, the Hubble and space station, and I just know that it’s really slow and expensive. So, you know, this quote from the Google CEO, which is now actually kind of stale and probably getting kind of old, but that AI is one of the most important things humanity is working on, more profound than, I don’t know, electricity or fire. So yeah, I’ve got me thinking about how can I, can use that in my field, and then I found the work that, that you all have done in generative design.

3:53 So, you know, this is an old, the old Dolly armchair in the shape of an avocado, you know. So, you know, this generative design, the AI gives you multiple design outputs based on your input constraints, and it collaborates with a human. So, you know, what really impressed me about this when I first saw it is, like, man, it’s, you know, using the, the pillow, or the seat is a pillow, it’s kind of like has an aspect of creativity, so I got excited about it. And then, you know, now everybody knows about this, so this is, this is a bit, a bit old, but, you know, the applications visual art, text, of course, has just exploded with chat GPT, music, architecture, and I’m going to talk about structural design, and, and in particular I’m going to talk about taking it up a notch.

4:35 So, you know, what if you want something more specific, like an astronaut in an avocado chair on the surface of the Moon, well, then you’ve got to do a little bit of better job of telling the AI what you want, right. So, so these are some attempts at that, you can probably, if you know, if you’re familiar with this, you can probably see the difference between mid-journey and, and Dolly, in there, but, you know, I’m going to talk about how do you do this kind of like sophisticated prompt engineer. So here’s my much more complicated prompt: astronaut seated in an armchair in the shape of an avocado, on the surface of the Moon, during the Apollo moon landings, by Salvador Dali, Blue Earth rise in the background so that’s the lower right, or lower left image, with the Earth. So it took some work to get that. So I’m going to talk about how you sort of prompt engineer to get lightweight parts for your spacecraft, and, you know, this is just a kind of a modern framing of it, because everybody’s thinking about prompt engineering these days.

5:28 So the first step in this process is to digitally encode the requirements for the structure into a piece of software, and, you know, this is kind of an important step, so we have a very detailed written guide about how you do that. So if you have a, a quarter 20 through hole, there’s a very specific, you know, prompt CAD prompt that you put into that, so this is Fusion 360 being used for this purpose. So this is an example of what you might call a prompt so the green parts are the things you need to keep, the red parts, or the areas you need to avoid, I’ll get more into that. And then you use generative design software to evolve the optimal designs that meet the requirements, and in this case this is Fusion 360, and, for some reason, this Mac is way faster than the PC, so it, like, this is, I’ve never seen this GIF go that fast, that’s kind of awesome though. So, you know, it takes maybe an hour to, to make that, you know, what’s called a prompt, takes about an hour to evolve the designs, and then it maybe takes between a day and three weeks to make the parts, and then you test them, and they do just what you think so a really accelerating things.

6:36 So I’m just going to get into this, this tip to excite tip tilt bracket a little bit, just so you see how this works. So it’s a mission that’s going to analyze exoplanet atmospheres from a balloon-borne platform, very exciting, launches from Antarctica, and, you know, really all you got to do is you just got to mount this some red green colorblind, it can, can actually take time, especially if it has strict requirements. So, you know, you have interfaces, you have a bolt pattern to the, to the tip tilt stage, you have a bolt pattern to the telescope, and you have to avoid, you know, interfering with anything, of course, and then you have a bunch of gravity loads, so, so we often have G loads because we launch things, so we really need, you know, Mass element representations. And then it has to be very stiff, there’s a cryo cooler on this and it’s vibrating, so you have to avoid that cryo, you know, coupling in with that mode, and, you know, balloons can’t lift that much, so it needs to be light.

7:33 So, you know, how would a human approach this problem so we did this, myself and a senior designer, the designer was doing the cad, I was doing the FEA, and, you know, it did the kind of thing that you expect, right, you sketch in planes and you extrude, and then maybe you put in some ribs, right, that’s, that’s what, what people do, because everything is, is 2D, and it’s just not nearly light enough. And then, well, it’s not clearing the light off, so we put a bunch of pockets in it, well then it’s not nearly stiff enough, and, well, we’re going to try to stiffen it up, but it’s too heavy, it’s not stiff enough, it’s hard to manufacture. And then finally he came up with this really wacky thing on the right that is just, does barely meet the requirements, but is very hard to manufacture. So, you know, this is two days, we did four iterations, and that’s, that’s pretty good for FEA and CAD, and if we had another couple weeks, I bet we could figure it out.

8:26 But I’m going to contrast that with how you would use, how we use generative design to solve the same problem. So, so the green are where the bolts go, and then the red is, you know, where the bolts have to come in, and they’re very specific for the washer type used and the tool that’s going to make this, and then there’s also, you know, your load cases, of course, go in here, your manufacturing methods, Etc. So then, in about an hour, you get these outputs, and two of these we actually fabricated, one is CNC, one is additive, so in Fusion you can have them, them do them all at the same time. And let’s look at how they perform versus the human.

9:07 So in one hour it does 31 iterations, doing the a, the C and C A and am ones simultaneously. It’s lighter, it’s much stiffer, which is great for our applications, and, you know, one of the kind of surprising things is, maybe it shouldn’t be, you know, they’re much stronger, so the, the peak stress is much lower, and they’re easier to fabricate. The, the CNC one was made in just three days, for a grand I don’t have it with me, but I do have a bunch of little samples that, that rents gotten made for me, that, that I can give out. And there’s also, one of these brackets is over in the end, top took the same design problem and did their take on it, so that’s actually this, this same design problem. So then we throw them on a Shaker, and we, you know, make sure they, they do what they, they do, and, you know, it’s FEA, it works, so they’re pretty much perform exactly like you’d expect.

9:59 So that’s a very simple example with just two interfaces, and, you know, the story of this is that we’re getting more and more complicated, and pushing the tools beyond what, what Autodesk ever thought, you know, would be done with them, and making really complex things like these Optical benches that have, you know, seven, eight interfaces there’s a Venus Lander here, and you can see that these are starting to get really, really complex and having a lot of interfaces. And the more interfaces you have, the, the longer it takes you, because you have more requirements, but then sometimes the more savings you can get, in terms of, you know, time. So yeah, it’s very exciting to see these things are, you know, potentially going to be going sort of all over so the alien spaceships will be our spaceships, that’s my dream.

10:47 So yeah, I really like this quote from Stephen Johnson: the trick to having good ideas is not to sit around in glorious isolation and think big thoughts, is to put more parts on the table. So I’ve really focused on spending all my money. So, you know, how is this you know, to, to y’all this is probably not that surprising, but to a general audience they’re like, wow, how can you get, like, you know, 10 times faster while also getting lighter, stiffer, stronger. And it’s because, you know, the advances of Information Technology that we haven’t really fully harnessed, that we’re just now starting to harness.

11:18 So, you know, our main tools are our Pro engineer, or Creo, which was from the 80s, and nastran, which is, you know, from the 60s, and, you know, we’ve had 10 000 times computing power increase, and, you know, what have we done with that, we’ve just put more nodes and, you know, our thread, our bolts now have the threads modeled, so which is itself on top of FEA and on top of design for manufacturing. So, you know, it’s a paradigm shift in what the engineer does, and that’s why, you know, culturally it’s been very hard to implement, because it’s, it’s really, you do a different thing.

11:53 So normally, you know, there’s a bunch of, there’s some requirements and goals, and they may be pretty loose, and a cad engineer or CAD designer comes up with, you know, some thing in flat planes like that, and then throws it over the fence to find an analyst who’s in a different group, and then they may take weeks to get results back, they might change some thicknesses, but they probably don’t give a lot of feedback a lot of introverts in this chain. And then there are, you know, you make a drawing, and then you throw over the fence of the manufacturer, and the manufacturer will just, you know, look at it, you know, curse the engineer for not know where knowing what he’s doing, and give you like a really long lead time and high price, instead of, instead of giving you feedback on how to make it cheaper. And, you know, this whole Loop takes months and years, especially for complex things, to go between those people who go on vacation and have training, and it just is really slow.

12:49 So, so generous design just collapses that whole thing, where you put in the requirements and goals, and then the, the, you know, software tool generates the cad model, does us find an element simulation, does a fabrication simulation, and then you could just go make it. So it really is, you know, we’ve demonstrated being able to start with requirements and having a part in our hand in less than two weeks, and have that part, you know, just be much better than a human design part. So again, you can see how fast the process is. So I think I have plenty of time.

13:28 So, you know, I’m gonna go a little bit into how this works so, you know, this box is kind of transparent to the user, if the software is, is user friendly, but what’s really going on is you are, you know, inputting the requirements, all these various requirements, and then, you know, a lot of you already know this, you know, he creates a voxel mesh of the space, and then the top, the topology optimization is really the key to it. I’m finding I’m getting a lot better results with level set than with, with simp, if you’re, you know, in, in the industry. And then, you know, a key part of this is, is showing the outputs in a way that you can, like, sort through them easily. So, so there’s some good tools for this in, in the tool that we use, which is fusion 360.

14:16 And then it reconstructs it, so that Reconstruction from that voxel to the a cad model is, is hard, and when it’s done well it can be really enabling. So, so, you know, here’s an example of, you know, to me and to, you know, the end user, you start with the inputs and then you get something you can make, that’s, you know, ideal. I think, even, you know, further down the road and more ideal, you start with the inputs and then you get like the fabrication, you know, G-Code at the end, but we’re not there yet.

14:49 So the other part of this is digital manufacturing, bits to atoms, so that’s a goes with the theme here. So on the left is a render that’s done in onshape, and on the right is the real picture, and they’re pretty close. So, you know, digital manufacturing takes you from the virtual world to the real world, and, you know, just really broadly, you’re making parts from CAD files without a lot of human intervention. And you all know most of this already, you know, the software makes the g code, and the more advanced software, sort of the less, you know, the human needs to do it. So yeah, you see a CNC machine on the upper right, and then lower, the lower right, you see additive. A lot of people confuse digital manufacturing with additive, and I, I’m always trying to dispel that I mean, this crowd knows, but there’s a lot of other ways to make things, and I’ll get into that.

15:35 So, you know, there’s a lot of automated CNC tools coming out now, where the, the cam, you know, creates things very quickly, the computer aided manufacturing software, and the robot just runs the, the G-Code. And one kind of exciting, you know, thing that I like is proto-labs, we use a lot, it’s like 500 hosting sea machines and no machinists, it’s just, you know, they’re just load in the blocks, so that, that’s, if you can get something to fit in there, in their package size, it’s pretty quick. And then additive, of course, you know, slicer generates the g code, and then, you know, the, the robot prints it, and then it’s post processing.

16:13 So, so I’m going to sort of be a little bit Rio it’s the real state of CNC versus am today. So, you know, a lot of these organic structures, like that, you know, someone called it a tangle of metal, in the middle there, that Optical bench that was CNC machined from a single block of aluminum, and it was done, you know, very quickly and cheaply in the scheme of things, and the whole thing is like within plus or minus three thousands. So, so that was a win over, over printing it. And there’s, you know, one of the really surprising things that came out is that there’s an inherent stiffness you know, it’s a, it’s a compliance minimization algorithm, right, so there’s an inherent stiffness in the parts that makes them easy to machine, much more so than the manufacturers were expecting. So that’s one of the things you only learn by, by doing it.

17:01 So with additive you get more design Freedom, so we, you know, maybe I’ve seen you get like maybe a 20 bump and stiffness, and then, you know, some large monolithic Parts can be, can be cheaper to make, like big chunks of titanium. But for Aerospace right now there are a bunch of cost and schedule disadvantages there’s a limited choice of am Benders, you know, for a large titanium part I had to go to Belgium, which had all kinds of issues with it. You know, you all know about the material properties being process dependent, a lot of post Machining, the surface finish is rough, you know, manual support removal, heat treatment is usually required, in a surprising number the manufacturers do not have an oven the same size as their printer, and, and, you know, NASA standard 60 30 ads just adds a lot of cost with where it is right now. And, you know, what, what’s surprising to me is I get quotes about them, well, this might fail a couple times when we print it, I’m like, oh, but, you know, for the future it’s really promising, and that’s what a lot of people here are working on, as, you know, the Technologies, the marketplace, and the standards are all rapidly maturing.

18:08 And you’re gonna, you know, enable things like Hollow structures, though I’ve recently, we did a lot of work with comparing, in real applications, Hollow to solid structures, and we’re actually finding that, like, having something truly Hollow, maybe isn’t that big of a Advantage. So maybe follow up with me, and I can send you my, my intern wrote his thesis about it. So, but, you know, what I’m really excited about is large monolithic structures that are enabled by something like, like Wham technology, where you can take like a spacecraft, like this, you know, that is so expensive, there’s so much manual labor in so many different parts that goes into making what’s just, it’s just a structure. So, you know, if you could generally design one of those and then just print the whole thing in a month, that would be, you know, radically enabling. So, so that’s sort of, you know, what I’m hoping to go to, or in the future.

18:58 So I don’t know how much time do I have three minutes, okay. So, so the summary, you can just read, I already said all that, but the future, you know, we want to make all structures, you know, 10 times faster and cheaper. Trust is flexors, Optics, or things we’re looking at next, and this is something that’s really exciting that, you know, everyone’s excited about Chit Chat GPT. So one of my interns came up with a way to, instead of having to make the cad that is the prompt, you just have a text prompt where you’re like, I have a pattern of quarter 20 bolts, and, you know, and it just, you know, based on the requirements, just builds the cad model using like apis. So I think that’s going to be a really exciting, like, plugging light large language models into software, so that, you know, you don’t need to do a lot of the manual work.

19:45 And then the other kind of fun thing is, you know, this, and I hope if there’s anybody’s working on this, I’d like to know, the sort of trustification. So, so we used, you know, Fusion to come up with this sort of where’s the load path going, for this telescope application, and then we skeletonized it manually actually, but it’d be nice to, to have that be more automatic. And then we turn that into tubes and nodes, and then the nodes themselves were actually designed using Fusion automated modeling. So that sort of, just, to, we’re trying to, you know, the past missions that have had these trust structures, it’s taking like a decade to design it, like literally a decade to design it, so, like, you know, maybe we can get that down to a couple months.

So one of the things that is really interesting is that a lot of the challenges aren’t technical, they’re cultural. So, so, you know, everyone’s building great tools, but how do you get people to use those tools, and, you know, especially get thousands of people, like NASA has like 80 000 people, to use those tools. And that is actually a lot of my work, is just trying to, you know, close the skills Gap, because you do need to be full inertia, and I’ve learned so much about how you get people to buy in that I’m happy to share with people.

21:04 Yeah, and one, one fun anecdote I’ll leave you with is, like, it’s really interesting with these, when you come up with a design and it’s wrong you just see it, like, you know, we’re, we live our whole lives, like, we see bones, we see animals, we climb trees, and, you know, when you’re climbing a tree and that Branch, like, just doesn’t look right, you’re not going to pull yourself up on it. And, like, you know, that’s one of the fun things about this, is that that human intuition is, is still needed. Okay, thank you.

More from CDFAM NYC 2023

Design Computation Human

Design Computation Human

Onur Yüce Gün · New Balance

Mud y Robots: Keynote Presentation

Mud y Robots: Keynote Presentation

Ronald Rael · UC Berkeley / Emerging Objects

Roadmap to Replicators

Roadmap to Replicators

Neil Gershenfeld · MIT Center for Bits and Atoms

Scaling & Standardizing Patient Specific Design

Scaling & Standardizing Patient Specific Design

Adam Wentworth · Mayo Clinic

Computational Design Ops

Computational Design Ops

Ronnie Parsons · Mode Lab

Overcoming the Design Dilemma

Overcoming the Design Dilemma

Ryan Carter · Johns Hopkins APL

Digital Materials – Geometry and Data Combined

Digital Materials – Geometry and Data Combined

Marek Moffett · General Lattice

Ten Years Teaching Additive

Ten Years Teaching Additive

John Hart · MIT

10 Years of Applying DFAM

10 Years of Applying DFAM

Tim Simpson · Penn State University

MfAM Walks so DfAM Can Run!

MfAM Walks so DfAM Can Run!

Chelsea Cummings · The Barnes Global Advisors

Informatics and AI for Materials Development

Informatics and AI for Materials Development

Wes John Alder · Citrine Informatics

Predicting Fatigue Life in AM Metal Parts Using DigitalClone

Predicting Fatigue Life in AM Metal Parts Using DigitalClone

Behrooz Jalalahmadi · Sentient Science

Designing Material Properties with Advanced Toolpath Generation

Designing Material Properties with Advanced Toolpath Generation

Alex Roschli · Oak Ridge National Laboratory

Design Automation Using Machine Learning & Computational Design

Design Automation Using Machine Learning & Computational Design

Rhushik Matroja · Cognitive Design Systems

Learning to Generate Shapes

Learning to Generate Shapes

Karl Willis · Autodesk Research

Design for Artificial Intelligence

Design for Artificial Intelligence

Chris McComb · Carnegie Mellon University

Register for Updates and Discounts on CDFAM events.