When Generative Design Backfires

Engineers find a creative outlet with 3D printing.

On proud display at last year’s Autodesk University, the biggest CAD conference in the world, was a VW Microbus retrofitted with wheels like none we have ever seen. The wheels looked as if they were made of twigs then painted a bright orange. Similar were several equally orange and odd structures in and about the vehicle, boldly holding on the side view mirrors, propping up a rear seat and attaching the steering wheel to the steering column.

What looked like twigs proved to be generatively-designed shapes, the output of generative design algorithms available in Autodesk’s Fusion 360, then cast in metal. The generative design had saved 18% the weight of the previous designs, according to VW.

A Lightweight Hero?

The orange wheels were an apparent success in automobile lightweighting. Generative design aims to create new shapes, including shapes you may never have thought of, shapes that offer a weight advantage over traditional shapes, by efficiently putting material where it is needed and removing it where it is not. But staring at the orange wheel, we wondered if generative design came up with the right answer. And we questioned if changing shape of the time-honored wheel was the right decision. For lightweighting, our first thought is to change the material. Swapping aluminum for steel, for example, while keeping the same shape would be more than 60% lighter.

The VW Microbus drew a constant stream of visitors after being promoted from the main stage. You can count on a geeky design and engineering crowd to pick up on anything automotive or aerospace related, especially the unusual. Plus the VW Microbus had a certain nostalgic attraction for the boomers among us, those who had lived through the 1960s. Pictures of the time often included a VW camper painted with peace sign and flowers – but with nondescript steel wheels with hubcaps.

The VW Microbus on display, said to commemorate the 20-year anniversary of the VW Innovation and Engineering Center California, in Belmont (south of San Francisco) had the right blend of retro and future. Dutifully taking pictures and filing stories were an easily impressed media. You may have read any one of a number of stories touting generative design with the orange wheels as the most recent shining example, along with the associated claims of how much stronger and lighter it is.

And now for the 1001st time... Previous 3D print poster projects: GM’s seat belt bracket, GE’s fuel nozzle, Lightning Motorcycle’s swing arm. (Pictures courtesy of Wired, GE, Autodesk.)

The media blitz that descended on the VW Microbus should be enough to make the orange twiggy wheels the most recent poster project of emerging technologies, generative design and 3D printing in particular. In 2018, it was GM’s seat belt bracket that achieved saturation level coverage. The year before, it was GE’s jet fuel nozzle. The seat belt bracket (40% lighter and 20% stronger) was one of 30,000 experimental parts GM produced in its Warren Tech Center. Whether it will see the light of day as a production part is doubtful given the expense, speed and other issues of 3D printing. Consumer products that rely on mass production and assembly have had ample time to optimize stamping, bending, casting and molding parts in great quantity for the least cost. Aerospace, with razor thin safety margins, can ill afford to replace a proven, tested titanium forging with a 3D metal printed part that could contain voids, stress concentrations or other imperfections produced using a relatively new manufacturing process or material. So, with few 3D printed parts emerging from big name manufacturing—and even less generatively-designed parts—especially the most glamourous automotive and aerospace industries, can you really fault the industries behind emerging technologies to stretch their mere presence of these parts as victories?

Let’s Call it Art

The VW Microbus wheels are cool, we have to admit. Take that with a grain of salt, engineering a profession that tries in vain to shake its image as staid, serious and decidedly un-hip. We thought ditching the pocket protectors would help but people can still spot us in a crowd. In return, we put “cool” in its place. “Function before form,” we say disdainfully at things that are trying to be cool. We look down on the public at large, those who have branded us less than cool, those softies who buy cars for their shine and shape. Real engineers don’t do that. We’ll buy the car by its specs, its gas mileage, data from crash reports, etc. If there is any beauty we can admire, it is the beauty of simplicity—the quest of an optimum shape, for example.

The Optimum Wheel is Much Simpler

Is the bike wheel the optimum wheel? Spokes on the bike wheel are straight and smooth – and structurally optimal shape for tensile load, making for a lightweight, strong and cheap wheel.

The most optimal load-carrying structure is a straight member in tension. It can be a thread, a wire, a cable. This is why spider webs are so strong, why suspension bridges are the most elegant. The load of the suspension bridge’s deck hangs on thin cables. Comparing a suspension bridge to a truss bridge is like Beauty to the Beast. Members in compression on a truss bridge are thick to avoid buckling.

Take a spoked bicycle wheel, for instance. There has been no other wheel design that optimizes for weight, strength and cost.

Spokes on a bike wheel are structural supports. To simplify matters, think of the weight of the bike and rider supported totally by the vertical spokes from the wheel hub to the wheel rim. The spoke is in pure tension, which lets it support a tremendous load. The shape of a simple spoke has a round section and is smooth and straight. Quite unlike the twiggy shape in the retro VW Microbus.

The spoke is smooth and straight not just because of the method of manufacture (it is cut from wire) but simply because smooth and straight has been found to be optimum. Any lumps on the spoke would add only superfluous weight. Any bend to the spokes and the bent shape could not resist the slightest tension without plastic deformation.

One might counter by saying the VW wheel twigs are made thick so they could bear tensile and compressive loads even as bent, but that is hardly optimal use of material. Also, a sharp bend in thick structural members will result in stress concentration factors.

VW Microbus Backfires

To call the VW Microbus’ orange wheel a successful design makes engineers question the criteria of success. If novelty is a criterion, the wheels are successful without question. If weight-saving is a criterion, the comparison should with other lightweight designs, such as spoked and solid wheels. In order for generative design to be taken seriously as a technology of the future, software vendors should resist offering “winning” designs that win in one regard but lose in every other.

Generative is touted loudest when offering weight savings as it has been successful in producing prototypes that work what might be a particular load case. For example, the GM’s seat belt restraint appears as if it would have a better strength-to-weight ratio when the seatbelt is loaded (crash situation) which would put its long thin members in tension. However, in compression, such as when are crushed by being sat on, long thin members won’t work. In fact, no commercial generative design we have seen seems to be capable of reacting to buckling loads in beams or plates.

Another glaring deficiency in most generative design algorithms is the ability to produce shapes that can be manufactured by conventional manufacturing means. Autodesk's Fusion 360 may be an exception, constraining shapes to what can be manufactured with 2.5 axis milling and die casting. Shape optimization, which seems 100% based on bone growth algorithms, produces bumpy shapes. Bone growth, specifically growth between two pieces of a bone fracture, is a result of fibers that grow out of both ends of the bone, that meet, usually at odd angles (anything but straight) and miraculously, the fibers start knitting together, making a bump when the healing is complete. But over time, and if the patient is not too old, the bone remodels itself and removes the bump.

This would lead one to believe bone growth algorithms are stopped, probably in the interest of time, before they can remodel themselves into a smooth shape. Theoretically, the same algorithms could keep running and removing every bit of material where it is seeing no stress—like in the bump—and adding it where the stress is high—the area behind the lump—eventually creating a straight, long member. But this would mean a simulation for many more iterations and we would do well to remember each iteration is a full stress analysis with many elements. Multiply that by the number of members in a part and you quickly have an untenable solution, even on an HPC.

Therefore, most generative designs end up with a shape that can only be 3D printed. But 3D printing is not always available, or the shapes are too big, or for a number of other reasons. Bumpy shapes are used as “inspirations” for final designs—a euphemism for a designer laboriously remaking the part bit by bit, substituting for each bumpy member, one that is straight/round/smooth—so it can be manufactured by conventional production methods.