Generative Design: For Real or Hype?

Generative Design is that set of tools that creates a shape optimized for strength

and weight from a set of geometrical and boundary condition input. It uses stress analysis techniques to find and remove the zero stress areas, and also to build material in high stress areas. It is often referred to as Optimized Design. You put in the constraints and a general shape, and the software trims down the geometry to only those elements which are necessary to support your input requirements. Generally the goals are to reduce mass/material and increase strength.

We often work through this process manually, designing a bracket, running a stress analysis, evaluating the results, altering the design to achieve a more consistent stress distribution, and iterate.

Traditional design tends to focus on the manufacturing process. For example, if you are working with structural steel, you have round and square tube, I, W, C, L

cross sectional shapes which can be produced by continuous processes. Even with plastics, you are limited by the kind of stuff you can get out of the mold. Molding creates discreet parts, but there are multi-cavity molds, and lights-out automation to keep the parts coming. Casting has fewer limitations, but there are still things you can’t cast, like hollow shapes without an opening. Casting can make larger parts, but generally in lower volumes. We wind up doing a lot of straight lines because they are easy to manufacture. We have linear processes like extrusion or rolling, or anything that happens continuously on a belt feeder to produce stuff economically.

Along comes 3D printing. It’s fashionable to think that this process is new, but it’s almost 30 years old. You can break a lot of traditional manufacturing rules with 3D printing. 3D print opens up a lot of design possibilities, even more when you consider some of the new materials and processes that 3D print techniques can employ, like high strength resins, metals, even concrete. 3D print is still not what I would call a high-volume production process, but the process keeps improving, and could potentially rival injection mold with lower tooling cost. And of course the ability to print in materials other than plastics will be important (metals, composites, fiber-filled materials, concrete, and ceramics).

What would you design, or what would you design differently if you didn’t have to worry about manufacturing processes? Would you go for the utilitarian or a more organic shape? What if the organic shape actually turns out to be the most utilitarian? Many of you will shake your heads at that one. One of the classic examples is to take a bicycle frame. The boundary conditions of the bicycle frame are pretty well known. Handlebars, seat, pedals, wheels. The big question is how do you most efficiently connect all of these to maximize strength, and minimize weight? We’ve all seen bike frames from the strictly straight tubes to the wildly curvaceous carbon fiber. Think about bridges. Bridges are made of structural shapes, which tend to be I beams, but if you step back and look at the overall shape of the bridge, most of the biggest ones that can’t afford to be massively inefficient and overbuilt are curved. If you’ve studied the architecture of Gaudi, he used parabolic shapes to support huge spires or very tall ceilings. Gaudi developed these shapes by hanging weights on strings and taking the parabolic shape of the resulting arced string. He knew that the most efficient shape was not a product of a straight line. And he used the equivalent of FEA techniques of his time to prove it.

These days we think of organic shapes in one of several categories: aero/hydro engineered type shapes, shapes literally created by nature such as animals and plants, or even stuff carved by wind/water (including ergonomic shapes), and finally artistic shapes. I think we need to add to this list shapes optimized for weight and strength. This is what Generative Design does.

Think of a simple plate where the loads from the bicycle frame have been arranged. Now run the stress analysis. Next, use the FEA visualization tools to plot all of the elements with stress over some minimum value. This will give you a

shape that connects all of the input points, and hollows out the middle. The funny thing is that the connections between the input areas (handlebars, seat, pedals, wheels) are not going to be straight lines or even constant width. They will probably not be circular tubes, or any structural shape you’ll find in a factory catalog. They will probably come together at one or more points in a very organic looking way. Yet, the computer tells you this is the most efficient shape.

This is why I think even machine design (especially machine design?) is going to become more dependent on shape in the future, and you’re going to have to be

able to work in more than just straight lines, even if what you design is just structural frames. You will use your generative design tool to get a general shape or outline for the frame, and then use tools that don’t exist yet to smooth that generative bumpiness into a design-worthy swoop, and finally print the frame in hollow metal that doesn’t need to be welded, and is weight efficient enough to be sent to Mars without further design or expense.

I’m a big believer in the intersection of mesh and NURBS. Whether its T-Splines, U-Splines, or just software that sits between the two technologies, design software has to bridge these technologies. 3D print has pushed generative design into

near-feasibility. Being able to manipulate meshes inside a NURBS model will continue to make more types of design accessible to engineers and designers currently stranded in NURBSland. This link is something that I think should continue to be developed into real mainstream functionality.

In my estimation, generative design is something we need to become comfortable with now. It may not become every day usage for all or most designers, but it will for some, and the combination of the geekiness of optimization with the more artistic aesthetic will in any case add an interesting dimension to design. It will serve as a shortcut to efficient AND to some extent organic designs, and may help us combine the design and simulation/analysis tasks to something that takes less time while accomplishing the goals of both processes.