CDFAM NYC 2023 · New York · 14–15 June 2023
Field Optimization: A Modular Design Tool for Advanced Manufacturing
Abstract
This recording is part of the CDFAM Computational Design (+DfAM) Symposium and features Justin Hattendorf, Director of Product Design at nTop.
Presentation Title: Field Optimization: A Modular Design Tool for Advanced Manufacturing
This session is aimed at professionals, researchers, and academics in the fields of computational design, additive manufacturing, and product design. Those interested in the exploration and application of field optimization as a modular design tool will find this presentation particularly enlightening.
Please note: The CDFAM Computational Design (+DfAM) Symposium is an esteemed platform for the exchange of knowledge among leaders in computational design and additive manufacturing. This recording is made available 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 · 3,170 words
Foreign. My name is Justin Hattendorf, and I’m the director of product design at ntop, and today I’m going to be talking about field optimization, field optimization, which is an open framework to be used for advanced manufacturing, and today I’m presenting on behalf of Christian Thompson, this is most of, this is his work, and unfortunately he wasn’t able to make it today. Hello, can you guys hear me okay? Great.
0:34 So I’m going to start by talking about why and top exists in the first place. So, so manufacturing today has advanced pretty dramatically, so today systems are highly automated, we’re talking about additive manufacturing here today, and it’s a pretty dramatic change from the 80s if you see in the photos here, but unfortunately our time, our design tools haven’t changed a whole lot, most of our design tools today are really based on the same principles as they were in the 80s through the 2000s, and really we believe that that’s creating a design issue, we believe that design is really the core issue too, or the real bottleneck, and product development today.
1:10 So in 2016 this is why we formed entop, we are working to resolve some of these problems in design and manufacturing, and bring those closer together. So our core tech is really what defines or differentiates and top from other softwares out there, so our first principle, or first pillar, is really implicit modeling, so this is a volumetric technology, it’s a little bit different than brap or mesh, it’s basically a technology that we’re interested in because it won’t break, we know that it’s always going to produce a solid and a void, which is really important for knowing if something’s going to be manufacturable in the physical world.
1:46 We also implicit model, so the idea is that you can use real world data, whether that’s physical data, simulation data, or just data that you enter into the software, and use that to drive the behaviors of your design. We also use process out of automation, so we believe that as you’re designing you shouldn’t just be building a single part, you should be building a reusable process, so that you can reuse that on your next project, is using a similar type of material, or, like, we do quite a lot of lattices, so basically we want to make sure that none of that work goes to waste, and that becomes integrated into your designs.
2:25 So today ntop enables some of the world’s most advanced products, so with robotics, mechanical design, aerospace, light, using lightweighting applications, and even in medical devices, patient-specific devices like this braces example, and we worked with a wide range of companies to do that. And we work in six main applications, so first one being lightweighting, that’s probably our most common one, where we’re helping our customers reduce the weight of parts, lower the cost, and increase the strength of their parts. Thermal management, so increasing heat transfer and reducing the volume of parts. Mass customization, so reducing the time to the time it takes to design a complex part and highly customized parts that are patient specific, so in this case we can do 12 designs per hour.
3:19 Industrial design, so enhancing your ergonomics and performance while also iterating faster, and I think this is the most important part about implicit modeling, is you can work very fast, you get that fast feedback, so you can make better design choices. Architected materials, so achieving a wide range of material properties, using things like lattices to adapt to product specs in manufacturing and tooling, so this is reducing the cost of tooling, and typically traditional manufacturing like jigs and fixtures, molds, really decreasing the time it takes to make that type, or those types of parts.
3:55 So before talking about field optimization, I want to talk a little bit about gender design, just kind of that to contextualize why we’re talking about fields in the first place. So one of my favorite things to do, is before doing a presentation on gender design, is Google, what’s going on in general design, what’s the definition today of generative design. The Wikipedia has changed dramatically over the past five years, and here are a few quotes of that, I’m not going to spend time talking about any of these quotes, because all over the place, I think what I really do want to spend time talking about is the principles of general design.
And we, we believe that general design needs to be open, and this is going to be a foundational principle of what I’ll show later with field optimization, so we need that, we need general design to be flexible, integrations with, provide flexible integrations with algorithms, they need to be intuitive expressions of your design intent, creative interplay of physical data and digital processes, so the human has to be in the loop here to be able to tell the computer what to do, how to do it, and how to link that back to the physical world, and it should be empowering in high complexity environments.
4:55 So gender design itself is not new, one example isn’t in the field of electronics, for example, if you look at kind of Intel’s original processor, 16-bit, 29 000 transistors, is mostly done with hand schematics, but in order for the product architecture to advance, basically electronics needed new tools to work with, to be able to manage the complexity as possible. So if you look at a typical transistor to a chip today, Apple One Ultra, 64-bit, 114 billion transistors, in a much smaller space, five nanometer design. So the enabling factor there is software, it’s electronic design automation, is a way to separate the logical from the physical, so that you can abstract your design process into a, like an actionable thing that the computer can make, or can help make.
5:45 Similarly in software, if you look at the 1980 Enterprise space shuttle, 400 000 lines of code to launch a vehicle into space, mostly hand schematics as well, to kind of, are handwritten kind of logic to get that started. I think in order, like, today’s average car is pretty dramatically different, 100 000 or 100 million lines of code, and really that’s only achievable by abstracting the design, abstracting the space, with model based design principles. So we believe that the same principle needs to be applied to hardware design, so if you look at a typical heat exchanger, hundreds of parts, a casted and extruded with traditional manufacturing, in order to really increase the scale and get the full unit cells within something, there’s really not possible, the traditional CAD today, and generic design is a necessary tool to do that.
So we think that generic design needs to couple geometry, simulation, and manufacturing together into a loop, to be able to achieve that. So the tools that we have for today have gone this, pretty, or gotten our customers pretty far, we have tools for topology optimization, we have automation tools like mtop automate, which can help you run the command line and execute a bunch of designs, a bunch of sweep over a bunch of parameters and design, and a design space, and then field driven design, to be able to couple, like, let’s see, stress data for this bike seat, to inform, like, how a lattice should perform, so making more dense in some areas that are under high stress and less dense than others.
So, but as our customers have advanced in added manufacturing, sort of, the complexity of their design problems, so at the beginning it’s easy, Adam is really interesting, and, like, you can do quite a lot of novel things with design software and additive manufacturing in general, over time, like, we’ve seen a lot of point solutions, like, we’re additive is solving an important problem in their business, but I think what we really need to push for is additive where it’s driving a competitive advantage, like, how, how is complexity, and, and additive in general, like, giving a company’s competitive advantage, where it’s not just solving one problem, but it’s solving a wide range of problems.
7:49 And we think, or if we look at some of what, what our customers are saying about kind of gender design today, about topology optimization for example, so the free form organic parameterization of topology optimization doesn’t give me enough control, so people want to be, want to be more in tune with what the computer is doing, important takeaway there. I need design tools that can adapt to the unique requirements for each of my projects, so people need flexible tools, something like topology optimization, like, it isn’t quite open enough for someone to be able to inform, or give more information that can feed into, from their projects, how can I better account for the complex multifunctional requirements of my ampart.
8:32 So basically, yeah, apology optimization is a good example where it’s like, good at creating solids, not creating, creating like multi-scale geometry, like dealing with lattices, and when controlling my geometry, or while controlling my geometry with fields is useful, I still need to be able to manually edit my model to close the loop, or finish my designs, so we need the geometry at the other end to be able to plug back into model, modeling functions basically.
9:03 So this is why we introduced field optimization, so design optimization today, like leading up to field optimization, this, so in the 90s you start to see a parametric optimization, so optimizing a few fixed dimensions, and it’s still useful today, but not, not quite dealing with the complexity that we need to. In the 2000s you start to see more, like, shape optimization, so you can optimize a predetermined shape, usually the shape that you can enter is pretty constrained, we can optimize an existing shape, but it’s not really creating anything new. With apology optimization, it’s pretty open, or, like, like the types of shapes you can make are organic freeform, it works really well with additive manufacturing because these either easiest to produce with additive manufacturing, but it is limited mostly to solid geometry.
9:54 So today what we’re introducing is field optimization, which is a flexible, open, multi-scare, multi-scale optimization framework built specifically for implicit modeling and field driven design. So really what we’re trying to do is couple implicit, implicit modeling with field driven design in an optimization loop. So what is it actually, so it is a gradient-based optimization framework for implicit geometry representations with field-based design parameters, so what that really means is we’re able to take an implicit representation and couple that with field data, and field data can be any sort of spatial data, but the important part about it, part of all that, is that it’s parameterized, so I’ll show you a quick example of how that works in a second.
10:42 The parameterization basically maps design variables over a structure’s response through simulation, testing, or equations, so a lot of examples that we’ve seen with this, because our software is make mostly, as making lattice designs really easy today, so an example of that is using physical testing data from lattice tests and feeding that back, then you run an optimization. So this is a pretty typical optimization framework, so you set up objectives, constraints, and define your bounds on your design variables, and run it, and then synthesizing all of those results.
11:18 So what you’ll get at the other end, it’s not just a, like, in contrast to topology optimization, what you’re getting is kind of a synthesized geometry field that’s responding to whatever you’ve parameterized from the beginning, so in this case we’ve parameterized how a unit cell performs, so what we’re getting is a result that’s performing at a global scale of how, responding to the data we put in, are we mapped in at a local scale.
11:46 So the most important part of all of this is the modular architecture, I won’t talk too much about this, but it’s a pretty, pretty typical framework, set of design responses, objective, constraints, and parameterization, so an example of an objective is I want to minimize wheat, an example of constraints is, so ensure that the stress is below X under these conditions, so the most, the most important part about all of this is the end, the, the last part, the parameterization, which in this case using this lattice infill.
12:16 So let’s take a look at what using this lattice infill really means. So so what you can do then is iterate over your design space, so what this means is, say you have a lattice unit cell, and you want to test, like, how the thickness relates to the performance of that unit cell overall, so we, we basically iterate over that, and calculate each local kind of response and compile that into any, like, raw CSV, any sort of like tabular data, once that’s compiled you can then synthesize that back again, bring that back into a response service, and any sort of response surface can be generated.
13:01 So for example, like, you can have response for stress, you have a response for a geometry, and then at the last step is to package that back up into a block, which is the kind of core element in our software that, like, feeds into the other algorithms. So with that parameterization, I’ll go through a quick example of, like, what that means for, like, a foreign parameterization. So the first thing you do is parameterize the geometry, so let’s say, we’ll over the X dimension, we’ll vary the cell size, so vorono is generated from a set of seed points, and you can see the density is changing, so we can basically vary the, the cell size, and then add a very beam thickness from left to right.
13:49 So basically that’s the geometry parameterization is nice, but really what’s most important is that it’s couples to the material parameterization, like, what is the response of this. So in order to calculate this we, you run a design of experiments, where in one direction you’re varying cell size, in another direction you’re very varying the thickness of your cells, so by coupling these together, basically what you can do is then run the field optimization to then compute, like, how something should be, how this bornoid cells should be sized and, and thickened in real space, on the real context of a real part.
14:27 So I think the, again, the most important part here is not just, we’re looking, we’re not just looking at a swatch of, of like how the four and our unit cell is behaving, we’re looking at like how a range of swatches is performing, and mapping that back, so that the system can optimize and produce a varying design across space, in this case producing a part that is, I think, 15 mass reduction, but also designed for osteointegration, so we’re targeting a specific cell size, or a porosity at the surface, so that it better integrates with bone.
15:03 So field optimization unlocks a huge class of new design for optimization problems, and that can be solved in on top, and are being solved in n-top. I talked a little bit about lattices, so periodic lattices are repeating structures in space, stochastic lattices, so voronoi foams, things like that, multi-phase lattices, so this would be lattices where we have multiple different unit cells, different types of unit cells in space, user-defined manufacturing constraints, which I’ll talk a little bit about at the end, but I think this is like the most interesting part of this, shell and infill structures, and composite materials that you can plug right into.
15:48 The last example I’ll talk about is the NASA site bracket, so the design goal here is to Delight with this part, and it’s subject to stability and stiffness constraints, so the goal here is, yeah, to minimize the weight, and we’re going to compare two different strategies here just to show, like, what field optimization kind of results look like, and how they compare it to something like topology optimization, which is another, another reasonable approach for this type of problem.
16:17 So with the shell infill parameterization, basically we start with a geometry parametization again, which is a lot like the top up, you’re going to get a, a result that is, I’m, like, following a similar pattern as topology optimization. Then followed by kind of like a shell thickness, and then the infill thickness, so basically the geometry for that defines like how something should be infilled in space, and what that’s being coupled with is a material parameterization, so basically from these three things, where should it be void, where should it be solid or shelled, and where should it be filled, and filled meaning like the gyroid for example.
17:01 And we’ll see the results here, just in comparison to the topology optimization, the results look roughly the same on the surface, the field optimization is a bit like thicker looking, topology optimization produces kind of thin struts, so in this case we, we think that field optimization actually is a great approach because it’s not producing thin elements that are going to be prone to buckling, and instead you’re using the kind of like the, the infill and the shelling to produce a better performing part when it comes to metal frequency.
17:29 So yeah, if we look at the results, the, the results are clear, comparing topology optimization and field optimization, we’ve increased the, the minimum modal frequency 81, increase the loading, or the buckling load factor, by almost 90. We’ve achieved the max displacement, both of them are achieving it, and reduce the weight, so there are definitely trade-offs, but in this case this is a, like, a superior approach.
17:57 And again, I’ll end on this, future workflows, I think the most important part about this, again, like I’ve said, is the openness, and I think we believe in a future where gender design is not just bound to 3D printing and lattices and all of that, so I think what’s the real power of this optimization engine is that you can start to map the, the kind of logic about material and geometric parameters back to traditional manufacturing methods, so in this case we’re applying what looks like a topology optimization, similar to a topology optimization, but it has additional constraints so that make it easier to manufacture, so in this case you could manufacture from a single direction, it’s designed for this, pocketing is designed in a way that is, that it’s, it maps well to traditional manufacturing as well as additive manufacturing.
18:44 Yep, and that’s all that I have, thank you everyone.
Listen, search and cite
Listen to this talk as a podcast
More from CDFAM NYC 2023

Generative Design & Digital Manufacturing at NASA Goddard
Ryan McClelland · NASA Goddard Space Flight Center

Generate-to-Validate: Optimal Designs and the Need to Deliver Under Uncertainty
André Wilmes · Rafinex

Defining Performance not Just Geometry with Software Process + Material Science
Hardik Kabaria · Carbon

Reinventing the Wheel – Representing & Manufacturing Geometry with a Program
Mark Burhop · Siemens Technology

Interspecies Forms: Robotic Feedback Systems Mycelia Growth & Computational Form
Natalie Alima · Adidas / Biolab Studio

Designing Material Properties with Advanced Toolpath Generation
Alex Roschli · Oak Ridge National Laboratory

Design Automation Using Machine Learning & Computational Design
Rhushik Matroja · Cognitive Design Systems






















