What's new
What's new

Ai and the future of manufacturing?

Tomped

Plastic
Joined
Nov 27, 2018
Hello,
I stumbled upon this video from Mtdcnc a few months back of CloudNC. They claim to have developed the world's first fully autnomous CAM system that's capable of programming CNC machines automatically and more effecient than any human could. Just wondering what other people's take on this is?

DMG MORI UK offers the perfect solutions to CloudNC - YouTube

Personly I'm very skeptical and I'm yet to find any substantial proof such a system even exists, nevermind any evidence to back some of the rather outlandish claims from the CEO.

Firstly, a CAD file is basicly just wireframe with surfaces. It containes no information regarding manufacturing requirements such as tolerances, hole features, surface finishes or gd+t. Hence why pretty much every component needs to be supported with a technical drawing to communicate this missing information. How is a computer supposed to figure out the corrrect way to manufacture a component that functions correctly when this information is missing?
Unless we fundamentally change the way we embed information into a cad file, I cannot see how any Ai, no matter how intelligent could work without human input.

Secondly, as a subcontractor I can say programming accounts for probably less than 10% of our overall lead time. Most of the time is tied up in sourcing tooling, fixture manufacture and machine setup. So unless someone designs a machine that's capable of producing everything without needing setup, I cannot see how reducing the programming time can substantially reduce overall costs?

I have no doubt that cadcam will eventually become more intelligent and I'm glad to see a British company driving this, but I do doubt some of the claims about the capabilites of the system currently.

What are your thoughts?
 
It is (or possibly was) only a matter of time. The data to train such an AI exists in the necessary quantities now, thanks to cloud storage of CAD/CAM files.
 
lol fully autonomous? What were those guys doing at the computer? What were those guys doing at the machines?

100%, no its not gonna happen anytime soon. 70-80% is already being done, has been for many years at some companies. Nothing to see here, carry on.
 
this issue has been beaten to death. Back in the day there was supposedly a new standard of modelling software that was "smart" like you describe. The models supposedly had all the manufacturing details embedded in the model. I cant remember the name of the software but no one took it seriously and it died. Maybe it is still around.

Im sure someone knows which software i am referring to.
 
There has been ATP (Automatic Tool Pathing) for decades, but it requires a lot of setup for your family of parts. Once it's all setup you can quickly run other parts in the family through it, provided you pre-chew the input files (separate contours on different levels). Get a feature that deviates from the family though and you have to handle it manually.

I do believe AI will get there, but we've got some time yet. Around the time there's an AI that can handle every five axis and millturn part without human help, we'll have an AI running the military, another running the economy, and another running the healthcare system. I hope to have my Neuralink with AI augmentation by then to stay competitive.
 
I should add that there's a danger to using automated systems for the simpler jobs that they can already handle. If widespread it will significantly reduce the number of people getting experience by doing those simpler jobs; experience that's needed to process the more complicated jobs that automation can't yet handle, and that's also needed to help enable automation to be able to eventually handle those jobs. In other words, it can cause a dead-end in our advancement, and possibly a regression.
 
this issue has been beaten to death. Back in the day there was supposedly a new standard of modelling software that was "smart" like you describe. The models supposedly had all the manufacturing details embedded in the model. I cant remember the name of the software but no one took it seriously and it died. Maybe it is still around.

Im sure someone knows which software i am referring to.

If you're talking about the same kind of information that would be on a drawing that is a thing that absolutely exists and is actually fairly widely used, especially in vertically integrated manufacturing companies where there is a desire to remove hard copy drawings from the workflow. NX, Solidworks, Catia, Creo and probably others all have this. PMI (Product Manufacturing Information) and MBD (Model Based Definition) are the common terms. Most current CAM software can use this information to some limited extent, but it's far from completely automatic.

I actually use it fairly frequently even though we are a job shop, especially if I am making a model and someone else is going to program and set it.

If you're talking about something more advanced than that, then idk.
 
As for AI knowing detailed information about a part, JT and STEP214 files can contain all sorts of useful part info.
 
Back in the day there was supposedly a new standard of modelling software that was "smart"

I believe this was STEPNC

There has been ATP (Automatic Tool Pathing) for decades, but it requires a lot of setup for your family of parts.

ATP in Mastercam is not even close to the automation you can get in other softwares like Esprit. Feature Based is the way things have to progress if automation is the desired result. Chain based CAM is archaic and requires a ton of manual input, this is where Fusion failed. Chain based is still useful but Feature Based should be doing all the easy stuff, automatically.
70% automation is easily achievable in Esprit TODAY, put in some work and 80% is reachable. I can't speak for other CAMs level of automation and difficulty in reaching but would assume there are others close to this as well.
 
Some great replies here thanks.
Just to add I'm not completely against Ai taking over some of the mundane tasks involved in programming. In fact it would be great to shift that talent into other aspects of running the business.
I guess I got somewhat triggered by the video suggesting subcontractors in general provide a poor service. I think he is being nieve in thinking programming is the primary cause of long lead times.

I guess the success of AI depends on the type of work it is being built around. Here it looks like they are gearing towards purely prismatic 3+2 style simple billet work, feed the system a multiple operation engine block and let's see what comes out.
 
Precisely. Most of the time depending on the volume of work we don't even consider programming time into the quote since it's done offline whilst the machines are still producing. The biggest factors in us being able to reduce cost is building production cells standardised to the family of parts and combining multiple parts/operations into one cycle.
 
Per goose comment on chain based machining. In NX apparently you have to set this up to select a chain (either 3d OR 2d)..? Last place I worked had to pick each entity separately*. Thought it was the dumbest thing ever, but someone pointed out it can do 'automatic' chains, but don't remember the thread...

Mastercam's FBM is garbage, or it was, not using anything new.

*On top of that, we, the programmers, had to alter models/wf from the engineering department to "sink" radius (don't know why they didn't build in clearance for inserts to fit into the mold bases where there was no real function), which meant not only altering geometry, but the foreman insisted we finish the rad's and not leave stock, so had to goto each rad entity and tell it to NOT leave stock... :nutter:
 
Precisely. Most of the time depending on the volume of work we don't even consider programming time into the quote since it's done offline whilst the machines are still producing. The biggest factors in us being able to reduce cost is building production cells standardised to the family of parts and combining multiple parts/operations into one cycle.

I program for 5 mills, and occasionally a wire edm. They rarely are waiting on me for a new program. I don't see programming being the bottleneck, unless you are like 1 machine... or you are optimizing every single second of cycle time.
 
Lotta business speak yack yack... not a whole lot of demonstrations of the capabilities, workflow, limits or anything else.

Someone is fleecing their investors with a half backed buzzword ridden "product" with the hopes of getting bought out by Autodesk/Hexagon/SAP/etc and stuffed into the bowels of an F500 before anyone notices the scam.
 
Firstly, a CAD file is basicly just wireframe with surfaces. It containes no information regarding manufacturing requirements such as tolerances, hole features, surface finishes or gd+t. Hence why pretty much every component needs to be supported with a technical drawing to communicate this missing information.
Hey Rip, just wake up or something ? That party finished twenty years ago. Heard of solid models yet ? You are slightly out of date. More than slightly.

If you are using Bobcad, I can understand it, but there are more advanced programs in the world now.
 
AI in CAD/CAM is way behind the curve.
One reason, no money to be made.
A decent AI can play chess with the best in the world. Yet in CNC programming not so much.
Why?
Bob

In chess there are only so many moves possible to be made. Too many for a human to keep up with, but an AI can look at the board and where all the pieces are and then run through all the possibilities from there where it can win, select the one with the highest chance of winning, and then do that for every move. Plus chess is purely 2d.

AI for say 3D video games is pretty crummy, sure they can be made good by "cheating", knowing where the player is, having perfect aim, able to move in ways the player can't. The problem comes in trying to make it react to random events, choose from a near infinite possibility of things to do and then act accordingly.

You could probably make an AI that could machine 2D parts pretty well, however once you start adding in more dimensions of freedom you gain far more degrees of difficulty in figuring out what to do.

Old boss of mine when I first started used to tell me when I was having trouble setting up a part that there were only 2 ways to set it up, the right way and the wrong way. My response was that I could think of at least a dozen wrong ways to do it.
 








 
Back
Top