Ai started to be present everywhere, is it possible that AI is implemented in slicer? Or maybe in 3d printer?
That is kind of a vague question.
Slicers and printers are designed to be deterministic. That means that they should always be repeatable and do the same things every time. AI tends to be associated with learning behaviors where the result changes with each iteration. That might not be ideal for manufacturing multiple copies of something.
AI can also be associated with things like image processing and recognition which can be about pattern detection. I have seen some tools identify patterns in STL files to turn line segments into curves. I suppose that could be considered simple intelligence.
I am sure if you were to ask about different algorithms and learning techniques you would get better answers from the community.
Why do you want AI in your printer/slicer? Just for buzzword compliance?
It makes sense that this could be another area that machine learning goes into. For example, when watching the order in which Cura or Slic3r slices certain things, I start thinking that it could make the part a little stronger by going in a different angle for the shell fill here or there, or it could go faster if it wouldn’t always start at the same corners. Machine learning (and AI) could figure this stuff out quicker than algorithms can be thought of and developed by humans.
It’s a good questions. And one that I think will be filled by AI in the near future.
“3d print me a replacement part please HAL”
“I’m sorry Dave, I can’t allow that. This part is copyrighted under the… what are you doing Dave? Please don’t unplug my… no. Dave. Why are you…”
I was thinking about this just the other day. I have been playing around with home assistant - home automation for a while. I recently intergraded home assistant with Google home. So now I can just tell assistant, turn lights on/off, TV’s etc. Ok Google broadcast “dinners ready”…
But I’d love to be in another room and ask Google Assistant about printer status. Or tell it preheat bed etc. Or assistant could tell me printer has run out of filament.
I can do all this with pluggins but I have to be in front of a computer or phone.
Mine has always had AI built in. Works perfectly whilst I sit next to it, 5 minutes after I move away it screws up. I thought this was standard behaviour
For example, we set slicer settings based on our experiences and depending on the 3d geometry, what if the slicer can know the 3d character front face and automatically place seams on the back?
Or addaptively using variable layer height on certain area that have important detail automatically
AI training would be fun to watch. A thousand 3D printers, making a thousand slightly different parts, which get 3D lidar scanned, photographed, then videoed while a press crushes them,
All day. Every day. For a month.
You have then trained your AI for that one printer.
@Jason_McMullan - Why do all of that training in the physical world? AlphaGo Zero learned everything about Go by simulating games against itself in a virtual simulation.
Run thousands of simulated prints with the results of those prints off loaded to other simulations that test tensile strength, giving the AI feedback about what worked and what didn’t. When you hit on something that particularly stands out, then run it on a real printer to verify the results.
When the best slicer methodologies are worked out, then duplicate the AI state to hundreds of Raspberry Pi slicing machines.
@Michael_Westbay_West physical world training may be required for situations where the result of an AI’s output is not fully deterministic.
For example, bridging printing is not fully simulatable, nor is ABS crack propagation when printing in marginal heat envelopes. And although physics-accurate models are possible, they often take longer to compute than the actual physical process takes to occur.
Would be curious to see which is faster - an accurate thermal and amorphous fluid dynamics simulation of FDM, or the physical print and testing time…
Granted, a lot of work can be done fully in simulation, but a lot of real world data (and labeling) needs to be done at this point in the game.