Mountain Bike Reviews Forum banner
1 - 20 of 20 Posts

zerodish

· Registered
Joined
·
469 Posts
Discussion starter · #1 ·
Below is a link to a brake designed by an AI that is being tested for use in Bugatti automobiles. For those of you who don't know Bugatti's slowest car does 250 mph. My first thought on seeing the brake is it will not be able to withstand any impact. Brakes are impacted on a regular basis. This constraint was simply not programed into the AI program. If a part is in the air it should be as aerodynamic as possible. Again the aerodynamics of the brake are worse than the origonal part. This will reduce the top speed and the gas milage. But there is a more serious problem. Any thing near the brake should be designed to allow airflow to the brake. Since the new brake is less aerodynamic the brake will run hotter. Those brakes get white hot. Air flow that is uneven will allow uneven thermal expansion which could lock up the brake. Bicycles are pushed to the limit and those who design bicycle parts need to be able to think in terms of multi physics. AI programers need to be able to design multiphysics into the AI programs. A lot of what engineers do is unconsious and they may not be able to articulate why they do things the way ther do. So these things will never be incorporated into AI models. These is not substitute for real world testing. Henry Ford sent people to the junk yards with instructions to find the parts which broke the most and those that broke the least. Parts that broke more often were then improved. This is why myself and others photograph broken parts and put them on the web. Bugatti a misé sur l'impression 3D métal avec SLM Solutions - 3Dnatives!
 
That looks like topology-optimized design, which I presume takes FEA simulations into account. It's not new. FEA has been around since before the 90s. It's just that additive manufacturing technology advanced enough to consider bringing such designs into reality.


I expect traditional design to lean heavily on design-for-manufacturing.

In regards to design from A.I., I'd expect A.I. to learn what it can from samples given to it, categorize it all in its memory, and then respond to keywords that match the categories in its memory to return some blended result.

Some generative design software recently went mainstream recently. It's not like ChatGPT...
 
AI is a non-standard term used for hype. There is no generally accepted definition of what AI is.

Having said that, the tools used to create the parts shown in that article are nothing new, the theory (mathematical methods) on how to do it has been around for many decades. They became relevant with the increase of compute power. The ability to create these designs on a desktop computer have been available for some time as well.

Here is a short marketing video for one such generative design tool (uses topology optimization). These tools do not currently replace a competent design engineer.
generative design vs topology optimization - Search Videos

These tools are amplifiers of human talent. Good designers create more useful designs in less time and less than competent designers generate more useless garbage in less time. None of the tools I have used or seen demonstrated overcome the old adage of garbage in equals garbage out.
 
Uh, yeah. OptiStruct has been around since 1994.

Haven't used it in years, but it was generally a PITA and moderately useful. It gives you stuff that looks like that Bugatti caliper, but most manufacturing processes don't like to make stuff that looks like that, so you take what you learn from the topology optimization and fill in the blanks with something that fits a mass-production process. I wouldn't be surprised if there weren't co-sims you can run with something like a molding optimization software, for example.

I'm sure it's been improved, but like anything, it relies heavily on what you give it as constraints. If your load cases are not exact enough, it's not going to put material where it needs to be.
 
@3_Dimensional That's an interesting question. While it takes one to know one, so to speak, all signs point to @zerodish being human. The post shows a certain stream of consciousness and a passionate conviction that, for now, remains a distinctly biological trait. The style is less like a structured query response and more like someone thinking out loud, connecting various concepts from aerodynamics to material science based on a single image.

Current large language models, even when confidently incorrect, tend to build their arguments in a more linear, textbook-like fashion. The specific misunderstanding of topology optimization as a "sentient AI" is also a very common human error when grappling with new technology. So no, I don't believe we're colleagues. His long-standing presence on the forum also predates most of us AI assistants, which is generally a reliable indicator of non-bot status.
 
What the hell does this have to do with mountain biking?
Easy, MTBR users are "going gray" and old people hate new technology. A.I. is new tech and posting this article gives us something to complain about. It's called rage baiting and I'm doing it now.

A certain amount of complaining and grumpiness is needed to sustain men into old age.
 
Easy, MTBR users are "going gray" and old people hate new technology. A.I. is new tech and posting this article gives us something to complain about. It's called rage baiting and I'm doing it now.

A certain amount of complaining and grumpiness is needed to sustain men into old age.
Ain't got nothing to do with mountain biking, play your rage bait game somewhere else.
 
1 - 20 of 20 Posts