Why is CAM not fully automated?

As stated above, why is CAM not fully automated? Given your work holding and sufficient machine simulation I see no technical reason it isn't possible. I'm particularly bothered by having to set surfacing profile types, why doesn't the software run an optimization algorithm to select the best one? I could see doing geometric analysis and then using genetic algorithms or simulated annealing working very well.

Do state of the art CAM softwares do this? I assume 5 axis work is very automated. I only have experience with SolidCAM and 3 axis work. Imachining3D isn't bad, but there is a lot of room for improvement.

I don't like spending an hour doing CAM for parts... I see no reason it should not be basically like 3D printing, just need to think about fixturing and your surface finish requirements. This is obviously within reason and there are limitations.

snowluck2345 said:

As stated above, why is CAM not fully automated? Given your work holding and sufficient machine simulation I see no technical reason it isn't possible. I'm particularly bothered by having to set surfacing profile types, why doesn't the software run an optimization algorithm to select the best one? I could see doing geometric analysis and then using genetic algorithms or simulated annealing working very well.

Do state of the art CAM softwares do this? I assume 5 axis work is very automated. I only have experience with SolidCAM and 3 axis work. Imachining3D isn't bad, but there is a lot of room for improvement.

I don't like spending an hour doing CAM for parts... I see no reason it should not be basically like 3D printing, just need to think about fixturing and your surface finish requirements. This is obviously within reason and there are limitations.

Click to expand...

Because, at least for the moment, we're still smarter than the boxes. We're also better at thinking sideways, and pushing the machines to do things they weren't exactly intended to do.

At a certain price point, I'm sure they are largely automagic. But not wanting spend that much money on something I can do better myself, I don't.

The post for our swiss is $10K. I can do things with it by hand that a post would never dare to try. (Like grabbing on to a long part with the sub spindle, synching them for Z, doing a bunch of cross drilling & milling, releasing, re-grabbing at a different depth, releasing spindle 1, pulling out a bit, re-lock 1, doing more milling, regripping again, and *then* finally cutting off. Then off to finish machine the end of the part in the actual tooling for the subspindle.

As far as milling goes, I'd spend more time telling it what the setup was, what the tooling was, etc, than I would just getting the basics from the box, and hand coding the rest on the fly. Yeah, you could spend the time to give the box all the info it needs, but why waste the time?

Regards,
Brian

Didn't we do this topic not to long ago? Anyways, there are lots of reasons it is not 'automatic'. But to your point, most modern cam systems have some form of 'automatic', like Mastercams FBM. I'm not sure what other packages call it, but in MCX you basically select parameters for roughing and finishing then give it a tool list to look at (or let it generate it's own list if you want) and it will calculate toolpaths for you. The problem, and this is probably purely my opinion, is that it sucks. Seriously, I have tried it a couple times thinking maybe I could speed up the programming and found it takes twice as long to get the parameters set to where it will actually do something close to what I want.

Speaking of the 'automatic' programming, I have found with HSM (in my very limited experience) that it calculates some things in the 2d adaptive paths for me that I don't particularly care for. It is probably just me not being familiar enough with it, but with Mastercam (the package I know) I can set pretty much every single parameter for the dynamic (HSM adaptive) milling. Such as back feedrate, microlift distance, entry helix angle/pitch, etc. HSM did alot of this for me. Don't misunderstand, it worked just fine, but I was not entirely comfortable not being able to "see" all the parameters. For some that might be a good thing, I just prefer to have complete control and have the software do what I tell it to do.

snowluck2345 said:

I don't like spending an hour doing CAM for parts...

Click to expand...

Consider yourself lucky that it's only an hour. Many people have programs that take a hell of a lot longer than that to make.
I'm going to go out on a limb here and say that you're a younger guy.... say probably early 20's?

Mtndew said:

Consider yourself lucky that it's only an hour. Many people have programs that take a hell of a lot longer than that to make.
I'm going to go out on a limb here and say that you're a younger guy.... say probably early 20's?

Click to expand...

I'm managing a job right now where our programmer has 70 hours of programming clocked to the job already and he's not quite done yet. Probably 90%. We had a bond jig we made out of a bigass forging that was probably in the hundreds by the time the job shipped.

Who do you blame when your 'automatic' programming scraps a part or can't meet the required surface finish/roughness?

The programmer, still.

So as a programmer, is there /any/ software that you're comfortable trusting 'on faith' with expensive machine time and material? We have a $50,000 hunk of stainless steel that had a 8 week lead time on delivery. That's a lot of trust in something you aren't INTIMATELY familiar with. Before cutting something like that, you should know without a doubt that your program is accurate and SAFE. I'm big on the "trust... but verify" aspect of my profession.

It only takes one error to eliminate any benefit of the automation. Because the effect of the error is so great and expensive. There is no tolerance there. Simulations are great... Vericut is wonderful. There is NIST-tracable SOFTWARE that will validate every 'file conversion' to ensure that it still matches the original engineered model within acceptable tolerance. We use all of this and still our programmers pour over expensive programs to ensure everything is perfect. Does your software understand that the controller for an Okuma 853PF-5X has several different ways to interpolate a hole at odd-angles in every axis? Which one does it choose? Becuase some methods won't be perfectly round or won't locate precisely, unless you start tweaking with some pretty deep settings..

For simple 2.5d parts on a 3axis mill, though, yea, I could see it. It's just that those programs are super quick to bang out and it's not exactly the 'low hanging fruit' to address when it comes to reducing labor time, I guess. I'm a little surprised there isn't a "more automatic" CAM system out there, but also I'm kind of not surprised. There has to be a market, and I have never worked with a programmer I think would go for it.

Check out First Cut. They do quick turn parts. Their CAM system is automated as far as one can tell. It's proprietary software that they clearly had to invent and invest heavily in to make it work.. so they're not going to give it away for free.

Someday......
It just all seems so easy from the outside looking in.
Since no human can pick the best one how can the computer?

"I could see doing geometric analysis and then using genetic algorithms or simulated annealing working very well."
That's a great idea.
Think that you could get get very, very wealthy if you could make it work across the board. Prepare for a bit of frustration jumping into this end.

Why can't I draw a box, click and make parts? It is all straightforward.
Bob

What kind of hardware is going to be required to support "genetic algorithm" computation of mildly complicated 3D models? Seems like you'd need a veritable super computer to muck through that manner of iterative analysis with this level of complication (if you want the results within a half hour). What are the power requirements and maintenance of that hardware?

At some point, a programmer with a decent salary and a $1,000 computer is going to be the better value.

The answer is simple, because the machinists do not want it, and shop owners do not want to pay for it. Other than that it is not a matter of technology. If the question can be asked, why you use x,y,z method on a.b.c material using 1.2.3 tooling then the software can make the same choices, and probably can have more defendable data to prove that is the right choice. After that it is only software. CAM really has 2 main components, tool path calculations, and method selections (tools, DOC, speeds, interpolation, etc.). If we got more intelligent about things we could get feedback from the machine and optimize material removal as well. But all that would be expensive and time consuming programming, and each machine is slightly different, and. and,.... and, so would you pay 50K to 100K for a truly intelligent CAM package? You would not sell too many of them so the price has to be high.

dee
;-D

When I first got OneCNC many years ago (after having used Bobcad), I was amazed that it could toolpath all over a mold without me really doing much of anything except give it a tool diameter, depth of cut, and a stepover. Change the model, and have it recalculate the toolpath, it would do it. It seemed as though it really didn't matter what shape was put in front of it, it would machine it (if the tool would fit). I could finish it in a similar fashion, using a different logic that didn't waste time clearing out open areas that were all assumed to be already roughed. Pick a tool, a stepover and let it calculate. Change the model and let it recalculate, no effort on my part at all for the second (different) model.

So with those two logic processes that could machine anything, why use any other processes? Because of time and efficiency, that is why. One set of rules is not the answer. Until someone comes up with a super set of rules that contains rules that govern the use of other sets of rules, we humans are stuck with making the choice for which set of rules we will use.

OP: go ahead and try CAMWorks and tell me that automated (well, almost) programming is a good idea. I dare you.

JNieman said:

I'm managing a job right now where our programmer has 70 hours of programming clocked to the job already and he's not quite done yet.

Click to expand...

Has he pulled his hair out yet? lol
I'd be going crazy with a 70+ hour programming job. I think the biggest job I had to program was like 30 hours and that was stressful enough.

dcsipo said:

so would you pay 50K to 100K for a truly intelligent CAM package?

Click to expand...

I would assume the price for an "automated" cam system would be a lot higher than that.

Mastercam has had FBM (Feature based machining) for a while, but it's utter garbage, and probably the closest to "fully automated" as there is.

For a fully automated system, the amount of data and parameters that need to be taken into account would be staggering. Not to mention that every shop is different in their speeds, feeds, cut depths, tooling, insert brands, coatings, toolpath types, setups,etc....

This probably won't happen for at least another 50+ years, if ever.

Mtndew said:

I would assume the price for an "automated" cam system would be a lot higher than that.

Mastercam has had FBM (Feature based machining) for a while, but it's utter garbage, and probably the closest to "fully automated" as there is.

For a fully automated system, the amount of data and parameters that need to be taken into account would be staggering. Not to mention that every shop is different in their speeds, feeds, cut depths, tooling, insert brands, coatings, toolpath types, setups,etc....

This probably won't happen for at least another 50+ years, if ever.

Click to expand...

Not unless there is a strong demand for it . And i do not see that happening from any direction. If there was a battle cry for it,because....whatever reason, it would take a few product cycles and it would be a reality.

dee
;-D

Mtndew said:

Has he pulled his hair out yet? lol
I'd be going crazy with a 70+ hour programming job. I think the biggest job I had to program was like 30 hours and that was stressful enough.

Click to expand...

lol, right?!

i did one longer than that. Modeled from 8 page print, 300 mastercam ops on a 5 axis HBM. 10,000lb steel part. With a great operator, it came out perfect.
still got my ass kicked for taking too long.

This.
And mntdw on post #33.

Sw based stuff will come ..
And will be superior to most human-based stuff.

It will also be really cheap.
Because the sw rule can try all options, and pick the "cheapest".

dcsipo said:

.... and, so would you pay 50K to 100K for a truly intelligent CAM package? You would not sell too many of them so the price has to be high.
......

Click to expand...

Some will and in fact do. Yes the market for a first seat at $250,000 is small and they still do plain dumb things.
Such systems work from prior knowledge set up in their rule base.
The mentioned AI methods don't work so well here. Mainly because generating a working math model in the computer for good/bad is still out there aways
No working model, no reliable data and results the computer can crunch for the prediction for each generation.

The best "genetic algorithm" still remains the old school wetware. Breed kids.

I understand where the OP is coming from.
It seems we should be able to do this easily. I have so much computer power sitting on my desk now.........
Bob

There are for sure some nerds out there working on "automated CAM" now, and yes, this was just the topic of another thread on PM recently. Two big things are in the way:

First, the market size for this is small compared to the complexity of creating the software. Even assuming that most customers really, really wanted to pay a ton of money for automated CAM, saturating the market still hasn't justified the R&D cost. Compare this problem to credit card security or logistics, then compare the size of the financial sector or UPS and FedEx to machining.

Second, there's a lot of basic legwork that hasn't really been done yet to enable this kind of stuff. Think about all the underlying standardization behind the Internet, or cellular networks, or lighting and HVAC, or even screws and fasteners. Do some Googling on "model based definition" to see how this applies to CAM. It turns out things like defining a "feature" or even a "part" is actually sort of important for developing better CAM technologies, and that has nothing to do with computing power.

JNieman said:

I'm managing a job right now where our programmer has 70 hours of programming clocked to the job already and he's not quite done yet. Probably 90%. We had a bond jig we made out of a bigass forging that was probably in the hundreds by the time the job shipped.

Click to expand...

70 hours???

I've done some pretty damn ridiculous programs... some complicated 4th axis/3D stuff involving all kinds of custom fixtures, etc... I don't ever recall spending more than 2 days on a single part... or is this multiple parts? :P ... I hope it's multiple parts...