The Golden Gate Bridge is the proud symbol of San Francisco, as the Eiffel Tower is of Paris and the Taj Mahal of India. It divides San Francisco Bay and the Pacific Ocean. It was a bridge that couldn’t be built. The wind and the fog were insurmountable. No bridge could span the 1.28 miles between the towers on either side of a chasm 335 feet deep. It was a bridge that shouldn’t have been built, said others. Parallel fault lines of the San Andreas fault system would lie to either side. No bridge would survive The Big One (the earthquake Californians worried about before the Coronavirus).

The 1.7-mile-long (total length) Golden Gate Bridge was the longest suspension bridge in the world when it was completed in 1937. Today, it ranks as #17. It remains an icon of civil engineering and a testament to the ingenuity of its designers.

Walk from the South end of the bridge along the San Francisco Bay, past Crissy Field, marinas full of sailboats, Fisherman’s Wharf, to the Embarcadero and the Ferry Building. Here is San Francisco’s Financial District and Autodesk, maker of CAD and generative design. Seeing the Golden Gate Bridge, designed without either CAD or generative design, must have made Autodesk wonder “What if?”

No, we think they are just enjoying the beautiful view. So, we had to take the task of generatively designing a Golden Gate bridge upon ourselves.

Autodesk Generative Design

Generative design has been generating a lot of buzz in the CAD world, and no developer has been buzzing louder than Autodesk. Since launching in 2018, Autodesk generative design technology—built into Autodesk Fusion 360—has been continually refined and constantly promoted by the design software giant.

We’ve seen many examples of how generative design can produce innovative, organic and lightweight parts. The oft-used example is a seat bracket from General Motors that used generative design to produce a part 40 percent lighter and 20 percent stronger than the original. You’ve probably seen it by now:

Autodesk’s new go-to example of generative design is both more interesting and more ambitious. Autodesk engineers used generative design in a redesign of the classic Volkswagen Microbus. The engineers used generative design for several parts, but none more eye catching than the wheels, whose organic design saved 18 percent of the weight of the original wheels.

Though examples of generative design abound at conferences and in press releases, the technology has not yet made a significant impact in the design of mass-produced parts. Nonetheless, it’s clear that Autodesk has a strong vision for the future of generative design. The potential of the technology is exciting enough that we decided it was time to get hands-on with Autodesk’s generative design and see what we could generate ourselves.

Hands-On with Fusion 360 Generative Design

If you’re a subscriber to Autodesk Fusion 360, you already have access to generative design. Simply navigate to the generative design workspace to set up your study. It costs users 25 cloud credits (US$25) to run a study.

Let’s take a look at how it works. Autodesk provides several example files to learn the basics of generative design, but we thought we’d try something new. You either go big or go home, and since we’re all already stuck at home, we went big.

Setting Up a Generative Design

Autodesk’s generative design is straightforward to set up, especially for existing Fusion 360 users. There are four main steps in the process: Design Space, Design Conditions, Design Criteria and Materials.

To set up the Design Space, users must specify two things: Preserve Geometry, the geometry that must be included in the final design (such as connection points), and Obstacle Geometry, the geometry that must remain clear in the final design (such as holes).

To specify these geometries, we’ll start by clicking on Edit Model in the generative design workspace. Here, we can build some simple geometry for our study. There’s no need to add any new geometry to the model in the Design workspace. We’ll use the existing model as a reference to create the geometry we need:

For our bridge, we’ll specify the Preserve Geometry as the bridge deck, the supports on each end and the bases of the two towers. For the Obstacle Geometry, we’ll specify the clearance above and below the deck.

After building and defining our new geometry, the Preserve Geometry shows up in green and the Obstacle Geometry shows up in red. We can optionally add a Starting Shape, which would show up in yellow, but we’ll skip that step for now.

Next, we must add design conditions, which include structural constraints and structural loads exactly as you’d define in a simulation. For the structural constraints, we’ll fix the bottom faces of the end supports and towers. As for structural loads, there are three types of loads we must consider for bridges: dead loads, the weight of the bridge itself; live loads, the weight of the traffic on the bridge; and dynamic loads, which account for environmental effects like wind and earthquakes.

In a 2016 paper entitled Full Dynamic Model of Golden Gate Bridge, Game et al. concluded: “Even when compared with scaled up, worst case traffic loads, design wind loads and dynamic earthquake loading, the self-weight of the structure alone was the biggest contributing factor to stresses and deflections observed in the bridge.”

Translation: we don’t have to worry too much about the live and dynamic loads. As for the dead load, Autodesk’s generative design adds gravity by default. However, it won’t let us run a study without adding any other structural loads, so we’ll simulate the live load as a pressure on the deck and ignore dynamic loads. Some back-of-the-envelope calculations give us an approximate worst-case live load of 7.8kPa.

The next step is to set our design criteria. Here we can select our objectives—we can either minimize mass or maximize stiffness, subject to a specified safety factor. For our bridge, we’ll try to minimize mass with a safety factor of 2. This is also where we can specify our manufacturing criteria. We can specify constraints for additive manufacturing, milling, 2-axis cutting and die casting. For the Golden Gate bridge, we’ll check off Unrestricted, Additive and Milling, and leave the default constraints.

Lastly, we can choose the materials to use in our generative design. The Golden Gate Bridge is made out of steel, so we’ll pick steel along with the default material, aluminum AlSi10Mg. Oh, and since it’s all for fun—and so it can live up to its name—how about gold?

Now, all that’s left is generate the design. Click on Generate to open the dialog, where we can see the cost of the study (25 cloud credits) and our credit balance. It will also flag any warnings for our design, which in our case pertain to the material selection and size of the model:

We ignore the warnings and go full speed ahead.

Exploring the Generative Options

Though Autodesk’s generative design is backed by the vast computational resources of the cloud, the study can still take a while to complete. Some of our studies took days to finish, but that’s likely due to the extremely large size of our model (even at a tenth scale, the Golden Gate Bridge is huge).

As the study runs, the Explore window will update to show all the designs that match your objectives and constraints (labeled as converged), those that tried and failed (completed) and any that are still in progress (processing).

In theory, the converged designs are all workable (if you set up the study properly), but some will be more practical or cost effective than others. Autodesk’s generative design helps you sift through the options with filters, scatter plots and cost estimates.

So now the million-dollar question: what did generative design come up with to bridge the Golden Gate?

The New and Improved Golden Gate Bridge

Here’s the first converged result we got for the generatively designed Golden Gate bridge:

It looks cool, but it clearly doesn’t work. We need full clearance underneath the main span of the bridge. Autodesk’s generative design conformed to our Design Space, but like the proverbial genie, got around our wish in a way we didn’t see coming. It just went outside the Obstacle Geometry we defined under the bridge. The other eight converged designs all pulled the same trick.

It’s also clear that some of our Preserve Geometries—namely, the end supports—were unnecessarily restrictive. So, let’s modify our study to address these problems. To prevent generative design from leaking around our Obstacle Geometry, we’ll try adding a Starting Shape in Design Space. As the name implies, this is the geometry that Autodesk’s generative design will use as a starting point. For our Starting Shape, we’ll create a box that fully contains our model of the bridge:

Next, let’s relax our Preserve Geometry a little. We need to keep some Preserve Geometry so that we can fix the bottom faces to the ground, but we don’t need the entirety of the end supports or tower bases. So, let’s just keep the bottom 0.05m intact.

The New and Improved Golden Gate Bridge, Take 2

After running the study again, we got a few more workable design options. Here’s the first converged design, a literal golden Gate bridge:

It may look bland, but that bridge is worth over 24 billion dollars, so show some respect.

The next design to converge was more interesting:

It may not have the Art Deco panache of the original Golden Gate, but it’s got a certain organic allure. However, despite our efforts to prevent it, generative design still leaked outside our Obstacle Geometry. You can see this more clearly in a side view:

There’s still plenty of clearance under that bridge, but it’s not as much as the original design. Eventually, there were three more generative designs that roughly followed this same shape (and suffered from the same clearance problem).

The New and Improved Golden Gate Bridge, Take 3

Let’s try this thing one more time. It seems our Preserve Geometry is still overly restrictive, so let’s try taking off the end supports altogether. To make up for the lost constraints, we’ll apply fixed constraint to the ends of the deck. We also upped the safety factor target to 7, a more realistic objective.

And to stop the genie leaking outside the bottle once and for all, we’ll apply a new Obstacle Geometry with no workaround: a shell surrounding the entire Starting Shape.

Finally, because we’re tired of waiting around for the results, we’ll also try to speed-up this study. There are two ways we could do that: reduce the size of our model (Generative Design seems to use fixed sizes for its mesh) or reduce the “synthesis resolution” of the study. We opted for the latter, which entails adjusting a slider ranging between coarse (quicker) and fine (slower). Our previous studies were roughly in the middle. This new study will be turned all the way to Coarse.

In a matter of hours rather than days, we got six new converged results for our Golden Gate bridge.

You may have noticed that none of our designs are symmetrical. Currently, Autodesk’s generative design does not allow for symmetry constraints, though Autodesk is working on adding that capability. The current solution is to recreate the designs symmetrically after completing generative design. You can do that by exporting any of the designs back to Fusion 360 as either a mesh or 3D design, at the cost of 100 cloud credits ($100). We exported the steel bridge highlighted above as a 3D design.

To make it symmetrical, it’s a matter of choosing which half we prefer and mirroring it, for both left/right and front/back symmetry.

Generative Design in the Real World

It should be clear to readers that this example of generative design was just for fun. Even if there were some practical way to manufacture any of our designs (which there isn’t), and even if the study rigorously reflected the real-world conditions of a Golden Gate bridge (which it doesn’t), Autodesk’s generative design is not intended for large-scale civil engineering projects (not yet, at least). The software’s strength seems to be in refining smaller components and assemblies like the GM seat bracket, or the swing arm of a motorcycle:

I will say this about Autodesk’s generative design: it was fun—and interesting. I eagerly awaited the results of each study, wondering what the software would imagine when I made this tweak or that. The design presented in this article was the result of a lot of iteration, experimenting with different geometries, constraints, materials and criteria. Oh, and a lot of mistakes. Pro tip: make sure gravity is pointing down.

Fun though it may be, is generative design practical? For a Golden Gate bridge, maybe not. But for many projects—maybe. It’s worth playing with to see. If you’ve got any components that could benefit from lightweighting, experiment on a few to see what you can create. Perhaps by then you’ll find the cost worth it. And you’ll always have the memories. I, for one, will never look at the Golden Gate Bridge the same way again.

To learn more about Autodesk’s generative design and how you can access it for free, visit the Autodesk website.

Special thanks to Roopinder Tara for his contributions to this article and to David James Mackie for the Golden Gate Bridge model.