Generative design fail: Why aligning design goals and digital processes is important

For the past five years or so, Autodesk’s product evangelists and marketing team have aggressively promoted Generative Design as a design methodology. Now generative design in and of itself is neither good nor bad. Like any tool, it is how it is used that matters. Numerous applications of generative design have demonstrated how it can help the designer explore a broader solution space to arrive at better and more novel solutions. However, used inappropriately, it is not only counterproductive to the design, it is determinantal to the advancement of the industry. This article explores one of those generative design fails and elaborates on why aligning design goals and digital processes is crucial in achieving a successful design outcome.

Lessons from cycling

At the time of writing, the 2021 Tour de France has just finished with Tadej Pogačar of Slovenian, winning his second consecutive title. But I would like to take a minute to look back to the penultimate stage of last year’s Tour de France. The 36.2km stage was an individual time trial from Lure to La Planche des Belles Filles. The 6km final climb is a true test against gravity, with gradients averaging 9% and hitting 20% in certain locations.

But having a steep mountain in a professional cycling race is nothing new. What was new was how it was executed. In a sport decided in milliseconds, many riders, once they arrived at the base of the climb, came to a complete stop, got off, and switched bikes. Why would anyone do this when time is literally of the essence? It wasn’t to do with any mechanical failure or mass collision. No, the answer comes down to simple mathematics and strategy.

Road vs time trial bikes

Usually, when riding, cyclists will use a road bike. They are more comfortable and more stable – both critical factors in long rides within large groups. Aerodynamics on road bikes are secondary considerations as cyclists can ‘draft’ behind other cyclists, significantly reducing air resistance by between 27% to 50%, which is far more than any aerodynamic affordances.

However, as the name suggests, riders are on their own when it comes to individual time trials. Unless passing another rider, there is little opportunity to draft, so the bike’s aerodynamics becomes far more important. For this reason, cyclists use a time trial bike. These bikes look radically different to conventional road bikes. The main difference is the aero handlebars which allows the rider to adopt a more aerodynamic body position. Additionally, the bike frames are deeper, sporting an aerofoil profile. And depending on the wind conditions, riders might also use a disk wheel, and let’s not forget the skin-tight lycra and aero helmets. All of these modifications are designed to reduce air resistance to make the bike go faster.

The right tool for the right job

But these decisions come at a cost – time trial bikes are considerably less comfortable and stable. But most importantly, they are significantly heavier than a road bike – approximately 1.5kg heavier. This takes us to the critical point, for most scenarios, aerodynamics trumps weight. And that is because the faster you go, the greater the air resistance. Weight only becomes influential when air resistance is low – such as going up a steep mountain.

Of course, you just have to look around you for evidence of this. It is why mountain climbers tend to be very light like a jockey and why triathletes use aero helmets with no ventilation despite riding through Hawaiian lava fields. In a sport where every watt matters, cyclists know this, and they know that losing a few seconds on a bike changeover will be quickly made up and surpassed in a steep ascent. To put it another way, the riders of the Tour de France used the right tool for the right job.

Subjective optimisation

So what does this have to do with Autodesk and generative design? As many readers will attest, Autodesk has been aggressive in marketing generative design over the past few years. This onslaught has culminated in numerous social media and blog posts about the affordances of generative design. One particular article that grabbed my attention was how generative design was being used to optimise bike design. As we’ll discover, the design was neither generative nor optimised.

The article in question described how the French sports equipment manufacturer and retailer, Decathlon, used Autodesk Fusion 360 to design a new bike frame. Scott Reese, senior vice president at Autodesk, is quoted as saying:

Decathlon’s bicycle project beautifully illustrates the pivotal contributions generative design can make to the creative process for designers. With the help of Autodesk tools, Decathlon combined artificial intelligence and human creativity to pursue sustainability and performance goals that meet high consumer expectations.²

We all know that the Autodesk marketing machine is a juggernaut. But as a computational designer and triathlete, this claim didn’t sit well with me. I can almost forgive the incorrect terminology of using generative design for a topological optimisation process, but I struggle to overlook the unfounded claim that the software has optimised the design. As someone who has spent several years in academia teaching digital design, it brought back flashbacks of students claiming, “We used XYZ software to optimise the design”. What the students really mean, is that they optimised their design based on a single criterion. Optimising the overall design is something completely different.

Solving the wrong problems

Beyond the obvious fails like not having an adjustable seat post, there is something far more problematic with Autodesk’s generatively designed bike. Autodesk has attempted to minimise the quantity of material in the name of sustainability – eschewing the commonly used carbon fibre, favouring 3D-printed aluminium instead. This design move is based on the assumption that less material equals better sustainability and performance. However, design issues are rarely back and white, and this case is no different.

As we saw earlier, the purpose of any time trial bikes is speed. When speed is the goal, aerodynamics trumps weight. But by reducing material, what Autodesk has effectively done is introduce numerous areas where air can be trapped, ultimately slowing the bike down. But let’s imagine that the bike frame has some sort of lightweight, aerodynamic casing, as alluded to in the renders. In theory, the bike could achieve similar aerodynamics, in which case the main design issue would indeed be about weight. But even then, weight might still not be a defining factor. 

Weight vs strength

Back in 2000, the Union Cycliste Internationale (UCI) – the governing body of cycling – introduced a regulation that specifies that a (UCI competition) bike must weigh a minimum of 6.8kg. This regulation was established to ensure manufacturers don’t compromise the structural integrity of bikes and also so that teams are competing on relatively similar machines. In fact, many of the bikes ridden in the Tour de France roll off the production line well under that 6.8kg limit, forcing team mechanics to add weight to the bikes.³

So let’s assume our base case time trial bike weights 6.8km. The generative design bike (with an aero casing) just needs to match that weight for it to achieve both aerodynamic and weight parity. In this case, assuming aluminium is, in fact, a more sustainable and environmental material, the generatively designed bike would indeed be ‘better’. But if we compare aluminium to carbon fibre, carbon fibre is approximately 42% lighter. That means one would need to remove at least 42% of material just to achieve weight parity. Difficult? Yes. Impossible? No. 

But weight is just one side of a double-edged issue. The other, of course, is strength. And here again, carbon fibre trumps aluminium – it is roughly seven times stronger than aluminium. So in practice, it would be a delicate trade-off between reducing as much material as possible whilst ensuring structural integrity.

In search of a problem

To assist their cause, Autodesk would have been better off showing a comparison between the base case and the generatively designed bike along three dimensions – aerodynamics, weight and strength. And it’s these details that are conveniently left out of the article. As such, the reader is left with not much to go on beyond the bike’s aesthetics. How then is this bike lighter, stronger and more sustainable as claimed?

What all of this highlights is that being able to ‘use the tools’ is not enough. You must apply industry expertise and common sense if you are to achieve a better solution. Some may be inclined to dismiss this example as a one-off faux pas made by the Autodesk marketing team. But alas, this is not the case. 

Generative design fails 

In a class at Autodesk University 2017, “what starts out as a serious look at the forces felt by the bike frames ends up as some of the most ludicrous bike frame designs we have ever seen as a result of generative design.”⁴ Unlike the Decathlon version, which given sufficient supporting information may be better, it is hard to justify this design. In fact, one would assume that if the bike design was indeed optimised, it would look similar to the Decathlon version, or vice versa. But instead, the two bikes vary dramatically.

Next, we have possibly the most over-engineered public bench of all time. In this case, far more raw material has been added in the name of optimisation. But optimised for what? One would assume that the forces would be the same whether you sat on the left or right. So why the asymmetrical design? Possibly there is a logical explanation, but once again, the reader is left to arrive at their own conclusion due to a lack of information.

Given the above examples, it would seem that for Autodesk, generative design has far less to do with sustainability and performance and far more to do with aesthetics and novelty.

Generative design vs topological optimisation

But the story doesn’t end there. After reading the various articles, many readers may be left with the inaccurate assumption that generative design is software developed by Autodesk. This couldn’t be further from the truth and problematic on several levels.

The term ‘Generative design’ has been around for several decades and refers to a parametric model that evaluates the outcome against some design goal and then automatically re-runs until the design goal(s) is achieved or optimised.

The term ‘Generative design’ has been around for several decades and refers to a parametric model that evaluates the outcome against some design goal and then automatically re-runs until the design goal(s) is achieved or optimised. ‘Topology optimisation’, on the other hand, is a mathematical method that optimises material layout within a given design space for a given set of loads, boundary conditions and constraints with the goal of maximising the performance of the system. The examples presented above are clearly topological optimisations.

Unfortunately for the AEC industry, Autodesk has decided to simply call everything generative design. But even with this move, Autodesk doesn’t appear to be aligned with what generative design is and isn’t. In fact, what seems to have happened is that they have created two definitions: one for manufacturing (Fusion 360),⁷ and one for the AEC industry (Generative Design for Revit).⁸

While some may argue that this is semantics, the AEC industry already suffers from a lack of understanding of industry terminology. Ask ten people what BIM or a Digital Twin is, and you’ll get ten different answers. Without a common language aligning industry professionals, what chance do clients have to understand and appreciate their value? Words matter, and how we discuss our digital processes is important to the industry as a whole.

Conclusion

Generative design affords many benefits, and there are numerous use cases of its successful implementation. But nothing is more wasteful than solving the wrong problems. Merely being able to ‘use the tools’ is not enough. Professionals who embrace emerging technologies, such as generative design and topological optimisation, must ensure they continually keep the problem in mind, rather than using technology as the vehicle to search for a problem that doesn’t exist. And above all, when promoting new technologies, we must do so responsibly. After all, as Peter Parker famously said, with great power comes great responsibility.

References

¹ Autodesk. (17 Nov 2020) Decathlon reimagines lighter, stronger, more sustainable bicycle using Autodesk generative design.

² Autodesk. (17 Nov 2020) Decathlon reimagines lighter, stronger, more sustainable bicycle using Autodesk generative design.

³ VeloNews. Analysis: Eliminating the UCI bike weight limit is overdue.

⁴ Tara, R. (1 Feb 2021). Can a generatively designed bike please get better?

⁵ Musiol, M. (2017). Design space exploration with Autodesk generative design.

⁶ Sanders, L. (30 Nov 2020). Making impossible street furniture possible with generative design.

⁷ Orban, R. (9 Jun 2020). Topology optimization is not generative design.

⁸ Autodesk. Generative design primer.