When do you think AI will be able to do drawings and models?

[ctsturdiv]

So far, I think we are pretty safe.

I have done a bit of research into this, and the only person even somewhat adjacent to the field I have seen is a youtuber called the AI Wizard that attempted to do RPG maps using Stable Diffusion.

The thing he was running into was that 90 percent of the maps were unusable without advanced graphics editing level of tweaking. You could also see where the AI must have been trained off of Reddit or something because it kept putting a weird, garbage artifact watermark on the corners of the map like most reddit maps are.

Almost more work than just drawing it in the first place.

But, how long before you think an engineer will be able to say, "Give me a bracket 17mm wide by 6mm tall 2 mm thick with two .5 mm diameter holes offset 1 mm from each edge and 1mm radius chamfers on each end" and get something they can use?

[AlexLachance]

So far, I think we are pretty safe.

I have done a bit of research into this, and the only person even somewhat adjacent to the field I have seen is a youtuber called the AI Wizard that attempted to do RPG maps using Stable Diffusion.

The thing he was running into was that 90 percent of the maps were unusable without advanced graphics editing level of tweaking. You could also see where the AI must have been trained off of Reddit or something because it kept putting a weird, garbage artifact watermark on the corners of the map like most reddit maps are.

Almost more work than just drawing it in the first place.

But, how long before you think an engineer will be able to say, "Give me a bracket 17mm wide by 6mm tall 2 mm thick with two .5 mm diameter holes offset 1 mm from each edge and 1mm radius chamfers on each end" and get something they can use?

Dassault is working on an AI which would sort of look in a database of drawings from a company and recreate drawings based on those, I'm not sure how good it would be but AI has blown my mind quite a few times so I'm not jumping to conclusions too fast.

[bnemec]

Not soon enough.

[Frederick_Law]

I believe the foundation is here already.
OpenSCAD should work great with AI.

[mp3-250]

AI is basically bias on steroids, It help see a possible trend in the data and interpolate them to generate output, but it will miss the context if the problem is not strictly defined and if you want something usable, not a huge mountain of garbage, you have to check and sort out.
IMHO It COULD aslo work for some company, with a certain amount of resources, dimension, for certain specific issue on certain premises.
I see it COULD work on mass production of modular thing like gearboxes that can algoritmically defined and it is already proved ok in tuning manufacturing parameters on machines with a sort of PID loop (I saw neural networks in the 2000s doing that on pipe productionto reduce scrap), or failure prediction. But I am afraid the commercial hype of "replacing 80% of humans" will go on and someone is going to get burnt very badly.

The logic behind 3D CAD is to make a coherent mathematical representation of things, so there are a LOT of constraint in that regard, like the errors inside the sketcher or the modeler: they make sense in the POV of the mathematical representation, but not in the user real intent POV and viceversa.
Then 2D drawings are not a simple representation of the 3D alone, but they include the company knowhow, and issues outiside the scope of mere 3D modelling. Production issues, parts delivery problems, specific drawing for a specific supplier ...again they are valid on a very specific context. You can automatically make a mold in 3D CAD, but matching that model with budget/quality constraints and machining issues in a specific shop is another thing. You have to teach the AI that that 3D data with a G0, G1 surface continuity area is still OK because the mould is polished by hand and that guy in the shop is going to make it a G2 like one. Again you have to define a very specific problem with a huge set of uncertain variables.
Are we sure we have all those minute empirical experiences in digitalized form and as decent data to be input in the AI system? And can those data be modelized on a large scale to adapt to every possibile use case?
No, so we are relying on a "best educated guess" from a machine. <sarc>(and we get the F-35)</sarc>

Seriously, I do not know how the situation is in your company/country, but here we are still struggling with hand drawn sketches and notes...apparently we got rid of the fax machine few years ago.

[ctsturdiv]

Not soon enough.

Hopefully that will hold up for a few years. I actually want a non abusive, living wage job once I get out of this college! (Though these non trade electives get on my nerves. I literally have a Professor "Karen" for one of the social science electives I can't stand.)

But from what I see, I don't think drafting will totally be going that way. But you could see more user friendly software that anticipates things better and may not require as much training. Which, I suppose, in and of itself could be a danger.

One class I took back in the 10s on Cisco routers (before I saw all the graduates going to call centers and noped out), someone asked why the interface was so obtuse. The instructor replied "Job security. If it was as easy as starting up Windows, they'd get someone in the office to do it for 10 USD an hour." . I think he was onto something. If AutoCAD or Inventor or Plant 3D were much simpler and you could say "make me this" and it's perfect, why pay you..

Also brings up degree inflation for the few "thinking" jobs as professions with boards (not Drafting. Checked into Drafting's board, the ADDA seems to run out of someone's shop as a way to make side hustles off of putting seals on the few remaining textbooks) but others started asking more degrees to "buy" your way out of stuff.

For example: IT in the early 90s used to be two year Data Processing degree or ability to turn on a computer, Occupational Therapy and Physical Therapy before the late 00s used to be BS, now they are Master's with people in it grandfathered in. Nursing before the 1960s used to be an apprenticeship type arrangement orderly to nurse, now AA or BS and they grandfathered in with diploma schools.) Could be one day, you have to have BS in Engineer minimum to draft.

That said, low end customer service like call center needed to be automated away yesterday. Miserable, miserable field. And arguably with telemarketing totally ruined a perfectly viable 100 years old form of communication. Why the phone system does not have spam protections like most social media or even plain old email is a head scratcher. It almost is like they are lobbying to keep it. But I digress.

[Frederick_Law]

That said, low end customer service like call center needed to be automated away yesterday.

Have you talked to those call center AI?
Talk to them enough and you won't know how to talk.

The problem with AI and big data is: if you ask AI to look for a pattern, it will find one.

[jcapriotti]

Have you talked to those call center AI?
Talk to them enough and you won't know how to talk.

The problem with AI and big data is: if you ask AI to look for a pattern, it will find one.

Like FEA, garbage in, garbage out.

[Frederick_Law]

Just ask ChatGPT for next lotto numbers.

[zxys001]

Not sure if any of you have tried it out but the two ask gave me python txt scripts and there are interfaces for python to their 2D/3D programs. ...my gut tells me we'll see this refined in 2-5yrs and you can ask for a 3D file in whatever format (STEP, X_T,...)

[m2shell]

Can I at least use "the AI" to have SW automatically fix dangling dimensions & notes?

[bnemec]

Can I at least use "the AI" to have SW automatically fix dangling dimensions & notes?

Nope. You'll need to install Solid Edge cir. v19 for that.

[jcapriotti]

Can I at least use "the AI" to have SW automatically fix dangling dimensions & notes?

Rather do that myself, I don't the trust the software will reattach in a way that is correct. Maybe if I get an option to review everything it fixes.

[bnemec]

Rather do that myself, I don't the trust the software will reattach in a way that is correct. Maybe if I get an option to review everything it fixes.

Also Solid Edge, Drawing View Tracker and Dimension Tracker dialog boxes. They have all the info you would like.

[Glenn Schroeder]

I hope it waits until at least May 31, 2028.

[m2shell]

Also Solid Edge, Drawing View Tracker and Dimension Tracker dialog boxes. They have all the info you would like.

Well maybe its time to switch to SE!

Just for kicks, I sent an email today to my (old) SolidWorks provider TriMech to see how much it would cost go go back on Sub after 7 months off. Haha not only do I pay backdated Sub pricing for 7 months, but SW now requires to pay in full for an additional 2 year contract.

I guess SW is doing so well they have to beat us away with a stick.

[m2shell]

I hope it waits until at least May 31, 2028.

What am I missing here?

[SPerman]

image.png (39.54 KiB) Viewed 5487 times

[m2shell]

image.png

AllRightAllRightAllright!

[zwei]

2cent...

It does has some promise when it comes to several area

It definitely has its place in generative design and topology optimization

Maybe even constructing simple part from scratch....

It might even able to recreate a model based on neutral file..
Imagine throwing a complex STEP file and get back a full feature tree SOLIDWORKS file (bonus if you change 1 feature and light up 10x warning feature )
Kind of like feature recognition but more advance
Whether is the feature tree 'useable' is another question...

It might even able to spit out the file as a macro script that you could back ported into older version (i think i saw this somewhere in the past....)

This fundamentally might just be a 'self-created' problem... if everyone can use the same format, we wont even have file interchangeability issue eh?





But creating complex part from scratch is more complicated...
Sometimes we dont even know whats the requirement, much less provide AI those requirement

Same goes for creating complex drawing...
I doubt it can really understand what is the design intent behind to callout a proper tolerance

It will at least take some times.....or maybe not..

[Glenn Schroeder]

What am I missing here?

Unless something happens to change my mind that's my last day of work before retirement.

[m2shell]

...
It might even able to recreate a model based on neutral file..
Imagine throwing a complex STEP file and get back a full feature tree SOLIDWORKS file (bonus if you change 1 feature and light up 10x warning feature )
Kind of like feature recognition but more advance
Whether is the feature tree 'useable' is another question...

...
This fundamentally might just be a 'self-created' problem... if everyone can use the same format, we wont even have file interchangeability issue eh?

This is my thinking, too. I see systems being able to automatically create a history tree from other formats. Most likely with some iteration, they will give a usable history tree 85% of the time. I don't think history tree usability is black magic, it can probably be reduced to a set of rules, and testing algorithms. Say this is developed in-house... they can train systems by looking at history trees of tons of files.

Maybe AI can magically purge all the "Trash" file space that builds up in my drawing files.

[mp3-250]

autocreated tree are not necessarily the best in terms of performance or model quality.

working with surface and molds in NX for years and sometimes you have to play with feature tolerance (a concept SW keeps away from their users, with the exception of surface "sewing" into solids probably) : a "bad" mathematical model, but fine for most NCs.
car bumpers and front grills are a collection of garbage.,10yrs ago we were getting 300+ MB step files for a single part exported from a dozen of cad and cams before reaching our company...

[bnemec]

car bumpers and front grills are a collection of garbage.,

as in recycled water bottles, CRT housings and old toys?

[mp3-250]

@bnemec
3D modelling wise they are a patchwork of bad surfaces and bad conversions plus all the engineering issues as a bonus.

for a seat frame we had to split the mould design on two workstations at the time. and it was not even the biggest mould we made, but the catia step data was so big (and bad) we had no choice.

[Frederick_Law]

Never gonna happen.
https://fortune-com.cdn.ampproject.org/ ... d-study%2F

"Over just a few months, ChatGPT went from correctly answering a simple math problem 98% of the time to just 2%, study finds"

Even AI got dumb talking to human.

[jcapriotti]

Entropy eventually wins

[zxys001]

Entropy eventually wins

Increases... so, idiocracy?