Description
Key Learnings
- Learn about the use of generative design for multiobjective optimization.
- Learn about delivering industrially relevant teaching that reinforces the fundamental design curriculum.
- Discover how generative design aligns with the core design process.
Speakers
- Peter ChampneysPeter is a mechanical engineer with over 7 years of experience working with generative design. Based out of the Autodesk Technology Center in Birmingham, UK, he has worked on a projects from a wide variety of industries including automotive, aerospace, consumer products and construction.
- PSPeter SimpsonPeter is a Lecturer in Engineering Design and Industry Engagement at University College London. Before entering the education industry, he had been working as a technical consultant at Autodesk, helping customers to leverage the full potential of Autodesk Fusion for design, simulation and generative design. During his time in industry, his work spanned across industries, from consumer product design to high performance automotive applications.
PETER SIMPSON: Hello, and welcome to our talk at this year's AU. We're going to be covering through the next roughly 40 minutes how we think that you could teach design using generative design, hence the title Teaching Design the Generative Way.
So first off, who are we? So I am Pete Simpson. We've kept it quite easy for you by having a Pete and a Peter, so you only really need to remember one name. So, as I said, I'm Pete Simpson. I'm a lecturer at UCL here in London in the UK, and I lecture engineering design and industry engagement. And together, I also have Peter Champneys, who's a principal technical consultant at Autodesk.
So today we're going to look through a few things. We're going to do a bit of an introduction to UCL as a whole for those who haven't heard of us. We're then going to look at some kind of design methodologies and how we teach design at UCL, followed by a generative design overview, a case study on an engine con rod, and then a showcase of some student work.
So moving towards the introduction, let's look at who you UCL are. So UCL stands for University College London. I'll use UCL simply because it's a lot shorter and easier to say. So we are London's global university, as we call ourselves, and we're actually ranked top 10 in the world, and we're the largest University in the UK. But the thing that I'm most proud of is we were the first English university founded on equal access for all.
So what this means is, women, religious minorities, everybody was able to be granted their degrees, and this was the first time that this was done, as I mentioned. We have some great alumni. So for instance, we've got Francis Crick, Alexander Graham Bell, and even Mahatma Gandhi. And we even have 30 Nobel laureates that we call alumni at UCL.
We are very fortunate that we are at the heart of one of the world's most dynamic cities in London. So you can see here, just a little map to give you that tiny bit of extra context. You can see our Bloomsbury Campus there, just kind of off center, and you can see all the different landmarks from London. We're really lucky with where our location is, and it's a great place to come visit. So if anybody is in London, pop by.
So now, looking at our department and the department that I work in, I work in the department of mechanical engineering. We were actually the first department of mechanical engineering in the world, which was in 1847. And the whole idea was to set about combining theory and practical teaching, something that we still do today.
So our department now has a lot of research areas. We have research in energy, biomechanics, automation, and the others that you can see there. And from that, we have a massive student cohort. So we have around 750 undergraduate students, and then when we add in the master's and PhD students, we're sitting at around 1,000 purely studying mechanical engineering with us. And we're also proud that more than 50% of our students are international students, hence London's global university.
When we look at our course, we have two main options for undergraduate-taught courses. So we have a 3-year bachelors in engineering and a 4-year master's in engineering, both of which are accredited by the IMechE here in the UK, and they actually have an option while studying mechanical engineering to do a minor in another engineering discipline.
When it comes to teaching, these are mostly done in lectures, problem set tutorials, labs, and workshops, but we do have a few other kind of ways in which we teach. For instance, our scenarios, which are week-long sessions, where they will take something from initial concept right the way through to a prototype.
Our students will study a really broad curriculum, as with most mechanical engineering degrees, and some of the modules include things like thermodynamics, fluids, solid mechanics, but around one in four of these modules do have a design focus. So what we've done here is we've taught a lot of design, and we're trying to help you to teach design using generative design throughout this.
So teaching design, how do we teach design? So firstly, you have to approach this topic of, what really is design? And don't want to get too philosophical here because there are many ways in which you could answer this, but ultimately, the way in which we define it anyway is the idea of a design overview and then more of a design definition. And these are from literature.
So when we look at design overview, we're looking at the fact that design has a lot of different aspects and a lot of different stakeholders. And the quote here-- I really enjoy-- is that design is what links creativity and innovation. It shapes ideas to become practical and attractive propositions for users or customers. And this is what I think is really interesting about design and why I love design personally.
It's that chance where in engineering, which is so much theory and so much practical application, to allow you to be creative and innovative at the same time. And realistically, if I'm going to be more succinct, the total activity necessary to provide a process to meet a market need is maybe a more fair definition. What we're really saying here is in that bottom right section there, design is actually the solution. We're focusing on the solution to a given problem. And that brings up another question, is design a skill or a process?
So at UCL, we believe that design is ultimately a process. There are a lot of skills involved, but if you don't actually master the process and you have all of those fundamental skills, I personally don't believe that you're still going to get an efficient design process. So when we're looking at those skills, we can look at this publication from Pugh, which is showing all of those different stages of the development process and the necessary skills that might come alongside them.
Now, a lot of people, when they think of teaching design, they think of teaching CAD. They think of teaching the actual skills rather than the process. And what we try and do is, across our modules, we teach the skills at the point that they kind of correspond to within the process. So our teaching is process focused, and then throughout that process, we will teach them those key skills.
Now, that process has a few forms realistically in industry at the moment. There may be slightly outdated one that's starting to move away from, is that kind of linear or waterfall design process. So this is that classic cascading effect where somebody hands over their work to the next person, who takes it that step forward. And we've got a kind of visual representation of that here.
The issue with linear processes is that they can fall down really quickly if there is an upstream delay. So if that delay is upstream, that then carries down with it. So what a lot of companies are now moving towards is actually more of an agile sprint-based design process, which is something that was really championed by software development.
So here, we're looking at having individual sprint cycles, which tackle much smaller problems of the larger problem set. These individual sprints will each be in term its own design process, and then you'll combine these sprints into a larger series of sprints to actually get that overall design process that you were looking for.
Now, at UCL, the thing that we really focus on is the Double Diamond design methodology. So what we're saying here is it's actually something that has been endorsed by the Design Council, and it's a way in which we try and get our students to think about their designs.
So you can see at the furthest left here, we start with our problem. That's the start of every design. And what's really important is that you have to diverge to converge. So first, you go out and you discover the opportunity. You discover the problem. You try and learn as much as you can about that. From there, you get a problem definition. So you work on defining that. You're actually moving back towards that kind of center point, and you're trying to get that information into really useful of bit-sized chunks.
Again, this is where we start the real design aspect that a lot of people think of when we say design. We then move into a development phase, where we're probably iterating a few different design concepts at once, and then we're trying to, again, whittle those back down until we found our given solution. So what we're saying here is that, realistically, sometimes we have to cast a big net in order to catch the fish that we want. So now I'm going to pass over to Peter Champneys to take us through generative design.
PETER CHAMPNEYS: Thanks so much. So we'll be looking later on at how we can use generative design alongside some of these design methodologies, and we'll be talking about the expectations and learning that we would expect for students throughout this process. Before we do that, I'm going to just take a quick look and talk about generative design as a technology, just so that we're all on the same page.
So to understand generative design, generative design is really a design approach and a design technology, where, instead of a human designer coming along and creating designs themselves, instead, what we do is we describe the design problem to the software and it generates solutions to that problem for us.
So increasingly, in the world that we live in, we're becoming more and more used to intelligently software and algorithm that are actually creative and able to generate designs themselves, whether that is for a part of a building or a car or anything in the world around us. This is really the theme of software coming along, creating solutions based on our definition.
So in this class, we will be looking in particular at generative design and Autodesk Fusion. Again, this is a part of the kind of mechanical engineering course taught at UCL. And in particular, there are two key technologies inside generative design and Fusion, that is, generative design for structural component and generative design for fluid path.
So each of these have their own distinct areas of focus. We will be very much focusing on generative design for structural component in this talk. And it's worth knowing that there are other kind of technologies in Fusion as well. Generative design for structural component is the most well fleshed out and advanced tool by quite a significant margin.
So again, before we get into our case study that we're going to be looking at shortly, I'm just going to talk a little bit about the knowledge that is really required before a student or user comes along and starts using generative design. So generative design is actually pretty advanced tool that requires prior knowledge of multiple other topics before it can be approached.
So typically at UCL, this is taught in the third or fourth years after students have covered other required modules before getting on to this. So in particular, the two really key requirements are an understanding of static stress simulation and a good background in CAD modeling. So generative design isn't just going to do all of that for you. You are going to have to need these prior skill sets really before approaching the topic of generative design.
So again, just to dig a little bit deeper into that, if we take a look at static stress requirements, at a minimum, students are going to need to understand the role of loads and constraints. They need to know how to set up and run static stress simulations. They need to understand the sort of assumptions that are made during these simulations and how this can affect the results, how closely that maps onto reality, as well as topics like von Mises stress approximations to understand yield stress and those kind of failure modes.
They will also need to understand how multiple load cases can be used to build up an understanding of all the different ways that a component will be loaded during its life cycle. There's also CADs knowledge required.
So in order to set up and run a generative design study, you will need at least a basic understanding of how to create shapes using a solid modeling CAD tool, like Fusion. Again, Fusion has multiple different modules or workspaces, including a CAD modeling workspace, as well as generative design.
Students will also ideally have at least a basic understanding of design for manufacturing requirements. Again, this is one of the key strengths that generative design has, is it can do that kind of thing automatically, but it will be good to have a basic understanding of design for CNC machining, die casting, and additive manufacturing.
And then lastly, the outcomes from generative design, the inputs that you need to put in are typically pretty simple CAD geometry, but what you get out of generative design can often be really complex. So at that stage, more advanced CAD modeling capabilities are really recommended to be able to deal with those outcomes.
And then lastly, just using and running generative design itself, it's not just a combination of stress simulation and CAD modeling. You actually need your own distinct-- there's things to learn just to be able to run generative design. So that includes things like the role preserve and obstacle geometry, how to correctly apply optimization objectives, and things like-- topics like Starting Shapes that you're not going to come across in CAD modeling or static stress simulation.
So if you're interested, we actually have a brand new place for learning all of these things, and that is found at autodesk.com/learn. You can search for topics like generative design. You can filter by CAD or by software, so for example, by Fusion, and you'll see a huge variety of courses on static stress simulation, on CAD modeling, and on generative design. So this is a really fantastic resource that I would always point people to learn more of these kind of skills.
So what we're going to do now is walk through a case study, where we use generative design to optimize a component, and we're going to look at an engine con rod. So again, this is a typical mechanical engineering component. For lots of reasons, it makes good sense to use generative design on this part, and it, hopefully, will serve as a really great example of what a potential student project might look like.
And we're going to frame this using the Double Diamond design methodology, and we'll see how generative design fits really nicely within this specific design process. Again, you can use it alongside many of the processes that Pete spoke about earlier, but it works really nicely within this Double Diamond formula.
So what we're going to do is walk through this engine con rod case study using the four steps of discover, define, develop, and deliver as the kind of framing of this project. And we'll talk at each stage about the kind of key things we'd be looking for students to demonstrate and learn as they go through.
So we're going to start off with the first stage of that double diamond process, which is the discovery phase. So this is probably the most easy-to-overlook phase of the process, and really, this is all about understanding the design problem at hand. So at this stage, we're not going to be going into any software.
This is really all about research, understanding the questions, understanding the problem, thinking about things like, what makes a good design? Which technologies do we have at our disposal to solve this design problem? And what's the scope of exploration that we're going to be looking at later on?
So for a component like this con rod, there are a number of different questions that we're going to be wanting to ask at this stage. So the first set of questions are really about understanding the current state of play. So how are these components built today? What are the minimum requirements that they need to meet? How do they fit within the wider assembly? And what are the sort of problems or challenges that we might be looking to solve? So could we make this design cost less to produce? Could we make it more sustainable, use less material, or have a longer lifespan?
So this is really a question of research, and doing a good job at this stage is going to be really obvious later on, as we come to the project definition and using generative design, as students hopefully are able to demonstrate that they are correctly pointed in the right direction to use technology to actually solve a real problem and to actually research and understand and have a clear idea of what they're trying to solve.
This would also be a part of the process where I would be looking for students to evaluate and consider the technology of generative design itself and how well this might suit itself or not suit itself to the specific application. So again, if we're giving the student a generative design project, they have to use generative design in this case, but in the real world, this is one of the most important questions to consider.
Again, generative design can really be seen as an extra tool in our toolbox. And one of the key questions is, when does this tool make sense to use and when does it not make sense to use? There are times where it's more or less appropriate. And again, generative design is well suited to solving problems where structural strength is a key requirement of the part, and it's particularly adept at exploring multiple material or manufacturing methods, performing cost benefit analysis, and lightweighting or stiffness optimization.
So as we start to think and understand the problem, the next stage is to define the problem that we are looking to solve. And here we wanted to clearly articulate our design requirements. And this is actually when we use generative design, this really becomes the most foundational step. So when we use a generative design process, the software is going to create designs based on our definition.
So this really flips on its head the typical kind of design process where we would use a CAD to model all our different solutions to the design problem. And we're really coming in that solution stage and creating lots of solutions. When we use a generative design approach, our key role is correctly defining the problem for the generative design tool to solve.
Again, I think this is a reason why generative design can be a great tool for teaching design because it really forces people to consider and properly define their design problem, and it really forces them to come to grips with that. And as we'll see, the entire success rides on this definition phase.
So generative design software actually gives us some really specific ways to define the design problem. And again, this is really where, if you go and do a generative design tutorial, you will really learn this problem definition phase because this is the key element. And there are really two key areas to this. The first are our design requirements. And these are essentially the minimum capabilities our design needs to fulfill its capabilities. And this is primarily going to consist of the design space.
So you can see here we're defining the connection points to our crankshaft and to our piston. We're also assigning keep out regions. We're giving the algorithm a space within which to play, and we also specify design objectives. And these are essentially what makes a good designer, what makes one design better than another.
And this is going to include things like material options, lightweighting objectives, if we're saying to the software, we want you to make this design as light as possible or as stiff as possible. And it's also going to do design optimization for manufacturability.
So again, in generative design, manufacturability is more of an optimization than a requirement. And what I mean by that is, when we apply a manufacturing constraint-- you can see how those work in this video-- what the software is going to do is it's going to optimize the designs it creates to make them more suited to those specific manufacturing methods.
So once we've set up our generative design study, the next stage is development. And again, this is where generative design really comes into its own. So again, if we gave students a group project, they might go away and create one or two prototype ideas. They might explore a few options in this development stage.
With generative design, we can create hundreds of potential-- or many, many full CAD models based on our product definition. And these are going to solve in the cloud in a matter of hours. So in just a couple of hours, we can generate hundreds of potential CAD designs, and each of these has been optimized using algorithms.
You see this as an iterative process. Each design is iterated many, many times over, evaluated by the software, and optimized each stage for the structural performance and for the manufacturability. And we can then compare and contrast all of these different designs. So this is really the strength of generative design as a tool.
So again, what we'd be looking for students to do at this stage is to use generative design to its full potential. So it's really up to the user [INAUDIBLE]. And there are some things to play with, basically control how many different designs you create. You can also run multiple studies to explore many different things. So for example, in this case, we could explore different lengths, sizes of con rod by running multiple studies.
We were looking for students to explore properly and widely and take advantage of generative design, but also, there's another pitfall that people can fall into, which is trying to finalize their design by running through generative design. So the last stage of the process is deliver. And this is where we want to converge on a final solution and also to develop that outcome-- or develop those outcomes towards being an actual product.
So again, generative design is a really powerful tool for doing this. So built into generative design, we provide users with graphs, data, information to help them evaluate and compare all these different designs against each other. So we can compare, we can filter based on all the different materials.
In this graph, we're comparing in particular stiffness against mass or displacement against mass. So if we want a lightweight, stiff design, where if we've got minimum requirements to meet in terms of displacement, we can compare all these different designs together. This is really going to help us to converge and select maybe one or two, a couple of designs to move forwards with.
So each of these designs, as we mentioned, is a full CAD model. We have all this data inside of generative design, but we can also then take that design and do an additional analysis on that. So here we're showing additional static stress emulation.
So we build those into the generative design software. That's going to give us additional data about how this performs under the loads. And this might also bear be where we want to perform additional analysis, such as cost analysis or cradle to grave lifecycle analysis to understand the kind of impact this is going to have on the environment.
And another important thing that we can do at this stage is, we don't have to just use generative design alone as a design tool. In fact, generative design works best when it's used alongside other design tools, and that could include just normal CAD modeling, as we alluded to earlier.
We can take the generative design outcomes. And in this case, there's a lot of extra detail we want to add on here, things like threads, chamfers, fillets, but also in this case, we've gone a bit further. And we've actually done some additional modeling. And you'll see a student showcase later, which shows you how far you can push this.
But this is a really important thing to understand, is something that, again, we'd be looking ideally for students to take on board is not just to take the generative design outcome but to use it alongside other design tools and to create use those tools to better meet the design specification requirement that they created earlier in the design process. So that's a look at what an example case study might look like. I'm going to pass back to Pete now to take a look at how we might judge success in a project like this.
PETER SIMPSON: Thank you very much for that, Pete. So I think one of the main hesitancies I've seen from education relating to generative design is, how do you necessarily assess this? Because everybody's used to assessing the more standard design processes, but this is where we're trying to help you integrate this into your course.
So when we're looking at marking generative design, I'm a firm believer that, just as we've already stated, that design is not just a skill, it's a methodology, it's a process, I think we then also need to mark that process when it comes to generative design. So as we're taking this forward, this is the main idea of this made-up marking scheme, basically.
So first off, we're looking at the setup. So the reason I've split this scheme in two is I've basically split it into the two diamonds. So this is a very crude example where your first marking point might be the quality of the load cases. So that's taking into account whether those students have gone away, they've done that research that you've asked them to, and they've actually come back with load cases that you feel really represent exactly what's happening in the real world.
Next, you could also look at that manufacturing methods and material selection side of the actual problem definition within generative design. So there are a couple of things that would be really key to look at there is. First off, are the manufacturing methods and materials even feasible. Generative design falls under the kind of whole software package idea that everything is available to you at your fingertips.
Now, what this also means is that, if you told it to, let's say-- I mean, injection molding isn't an example, but if it were to allow injection molding and you told it to injection mold wood, it would try its best to give you an example for an injection molded wooden component. So maybe you just assess the feasibility. Potentially you're looking at whether those manufacturing methods are actually feasible for the potential application of this project that they're working on.
It's a very crude split in which you're looking at that first kind of diamond, and then we're moving on to actually assessing that second diamond, which is maybe what we're potentially a bit more comfortable with in education.
So we're looking at, how have they selected their outcomes? Is their sound engineering logic behind the way in which they've evaluated those outcomes within generative design? And have they maybe done some kind of secondary processes to help along the way? So whether that's a decision making methodology, maybe some of those secondary validation methods that Pete was mentioning earlier.
Next, you can then assess the post-processing just like you would any old design kind of project. Ultimately, the post-processing is still computer aided design. It's still something that doing using CAD software.
So you'll be assessing their skills. You'll be assessing how they approach that actual design itself. Have they focused on maybe performance and completely forgot about the idea of manufacturability? Have they actually harnessed both of those to produce a really well put-together part? That's how a great student would set themselves apart from a good student, in my eyes.
And then, finally, in this marking rubric that I've put together, we have those extra factors. And this is where, I think, you can really make this kind of work, your own, and you can also fit it to your learning objectives and learning outcomes. So if we're talking about maybe something more sustainability focused, as is the common trend within design at a university level, we want to make sure that our students are more sustainable in everything that they do and they are more conscious of sustainability.
If we can bring that into the industry from the lower levels and there's already those higher level objectives going into place, that will be a really good position for the industry to be in. Potentially, you then assess a life cycle analysis or something similar.
It could even be further than that. Maybe you're teaching a business-related design module that needs some engineering-finance type options. Maybe you look at a full cost breakdown of the part. That then relates back to that manufacturing kind of side of the generative design setup and makes that all the more pressing of a choice when they're selecting their outcomes.
And finally, maybe you even ask them to plan their manufacturing from start to finish. That would be another way to include a really good skill into this kind of a scenario or this kind of an actual deliverable for a student.
Now, what I really want to show-- and this is something I'm very excited for. This is actually work that a student I supervised has undertaken. So the student, whose name is Saquib Momin, was one of our MSc students. So as part of that MSc, they do an individual research project, and they decided to pick one that I had suggested. So I had suggested the idea of looking at folding bicycles.
So these are less common in general outside of Europe. I think they've boomed in popularity recently, with people trying to commute in more sustainable ways and also avoid pedestrian traffic and stuff after the pandemic, but folding bikes, such as Brompton's, have become really popular.
And while generative design and similar methodologies have been employed in high-performance racing bikes, to my knowledge and to my investigation that I did before suggesting this, they'd never actually been employed to a folding bike. So what this ended up looking like if you can kind of see the Gantt chart in the top right and then a reminder of the Double Diamond.
And so what this ended up looking like is that discover phase was largely market analysis. So we were looking at what's already available, what kind of is the status quo and how can we try and use that to our advantage?
Next, the definition stage came in. So there was a vast literature review, both on the methods used in generative design but also the methods used in bike production, stuff like that. And we had to make sure that throughout this whole process, we were making something that was ISO compliant because ultimately, there's no point designing a bike if it's not legal to ride anywhere.
Our developed process, as Pete kind of explained, was largely that generative design, design, exploration kind of activity. The loads were already set up from the ISO compliance analysis. We'd got some kind of basic preserve and obstacle geometry that I'll touch on in a bit, and then We were looking really at that deliver phase, which is where I think Saquib put in some of his best work.
So he took multiple outcomes, developed them into more tubular chassis designs to mirror that of standard bikes in the market currently, which would enhance the manufacture ability and also allow many bike brands to be working with familiar technology. There was then a vast outcome comparison and also costing and a life cycle analysis.
So when we look at that setup, this is what we ended up with. We had obstacles. We had preserves. Realistically, we wanted to try and constrain it as little as possible. We were not necessarily focusing on getting the perfect outcome in the timeline that we had because it was running for about six months, this project.
What we wanted to do is, see as a proof of concept, whether a generative design folding bike frame would even make sense with those folding mechanisms integrated. So what we chose to do is we chose to simplify it. We simplified it right down to a single bike frame to iterate upon and then add the folding mechanisms in as we got to that final delivery process.
So as you can see, we then had a very detailed set of forces on the bike frame, which were from the ISO standards so that we knew that we would be compliant, and these were then represented and used as our load cases. So it was a very simple setup once we got the information we needed.
Next came the bit that I think is the most impressive. So here you see we have a [AUDIO OUT] into tubular chassis. And as you see at the bottom ticking along one by one, we also have them all actually folding within Fusion, which I think is really impressive for the amount of time that's a keypad.
Each of these designs is able to fold. It is able to do it in different ways. You can see the first four that are in the blue color actually fold in three ways. Because they're metal, they're able to integrate a folding mechanism in the kind of tube, going from where the cranks would sit up towards the head tube. And the final one, which is actually a carbon fiber reinforced plastic, he decided to just leave this as the rear wheel folding under so that it's still severely reduced the actual footprint of the bike, but it didn't introduce any instability. And this was done by simulation validation.
So what I'm trying to show here is that one student in this space of about six months was able to produce five individual bespoke tubular chassis that are completely redesigned around a generative design outcome. And when I realistically set this project, I was as open as he was, to be honest. I said, I think this is a market that could be explored. With a folding bike, you actually have to pick it up. You have to carry it at times. You have to carry it up and down stairs. It would be great if it was lighter. Let's see what can happen with this.
I never expected this many outcomes, but here we are. That's really credit to Saquib and his work, but I think ultimately, what this shows, in my opinion, is that this kind of work really reinforces a lot of the design teaching, and it shows that a simple design process that maybe at times students think we're harping on about and maybe over stating, actually can apply really well.
So let's look at our conclusions. So I think the main conclusions that we want you to draw from this are that generative design is not a replacement for traditional design. This has been a common misconception since generative design rose in popularity and came onto the scene.
We also want to state that generative design does perform some powerful automations, and they will augment the ability of designers to explore design alternatives. I think if I'd really turned around and said to any student, can you design me five tubular chassis for a bike frame? They wouldn't necessarily know where to start, and we'd probably get five that look very similar to existing bike frames. But with generative design, in came the innovation, in came that kind of automation that allowed him to take on a bunch of different information and get some really polished final products.
And finally, I think generative design really can be a powerful tool for teaching design principles and processes. A lot of people just see them as completely separate things, and that one is generative design, which is coming in and trying to change the way that we design, but actually, it's fundamentally mirroring what we're already doing. There is very little difference in the process. It's just a more automated way of doing that process.
Thank you very much for listening to our talk. I'm sure I speak on behalf of me and Pete when I say that we hope you enjoyed it. And hope you had fun.