Before I start, let me say I don’t fully trust AI … but as a second opinion I think it has value.
Let me pose the problem: Drill 0.25" holes in oak, brass and aluminum. I have an 18K 3HP spindle I can slow down to say 6000 RPM.
The “prompt” I use: “when drilling a .25 inch using a CNC at 6000 RPM what is the feed rate and HP required for a .5 inch depth in C250 brass, oak and aluminum”
CoPilot (MicroSoft AI, include with windows) returned:
Here’s the breakdown for drilling a 0.25-inch hole at 6000 RPM to a depth of 0.5 inches in C250 brass, oak, and aluminum:
Feed Rate
The formula is: Feed Rate (IPM) = RPM × Feed Per Revolution (FPR)
Material Removal Rate (MRR) is: MRR = Feed Rate × Tool Diameter² × π/4
C250 Brass: MRR = 24 × (0.25)² × π/4 ≈ 1.18 in³/min
Unit Power ≈ 0.2 HP/in³/min HP ≈ 1.18 × 0.2 ≈ 0.236 HP
Oak: MRR = 30 × (0.25)² × π/4 ≈ 1.47 in³/min
Unit Power ≈ 0.15 HP/in³/min HP ≈ 1.47 × 0.15 ≈ 0.22 HP
Aluminum: MRR = 36 × (0.25)² × π/4 ≈ 1.77 in³/min
Unit Power ≈ 0.3 HP/in³/min HP ≈ 1.77 × 0.3 ≈ 0.531 HP
These calculations are estimates and may vary depending on factors like tool sharpness, machine efficiency, and specific material properties. Let me know if you’d like further assistance!
Seems reasonable at first glance but I would verify. Sort of a back of the envelope calculation, but I am still impressed.
Keep in mind what the search engines are using now and what ChatGPT, Grok, and others are are basically just LLMs - Large Language Models that do a great job at piecing words together. They have limited “intelligence”. When I use ChatGPT, I mainly use it as a glorified search engine. It can search for and analyze far more websites than I can. I find myself correcting ChatGPT fairly often, and I’m using it’s more powerful paid service. The AI that’s being developed and talked about is way more than an algorithm that puts word together.
Absolutely… I am a Mech Eng, who wrote software for other engineers to use for 40 yrs., primarily C++. ( I just saw “aero23” … my work was for an aircraft engine company)
What I have seen in the last 6 months is the “logic chain” it used to come to its conclusion. Which give you a real nice way to follow up.
I recall about a year ago I asked it to calculate the weight of a part … it had REAL issues with units.
The work of AI going from a 2D model to a 3D model is coming along nicely too.
That’s not exactly true. Generative AI actually creates new information from inferences it makes from the info in the LLM as well as subsequent use. You do have to be careful that you validate answers because they can create new totally untrue information (called hallucinations). The AI tool uses a few factors (like top-p) to either expand the tool’s creativity (at the expense of more likelihood of hallucinations) or keep it closer to the straight & narrow (less creativity). The LLM is simply the basis of knowledge for the tool - it does not represent all the knowledge in the tool. Having the tool show it’s work is a good way to help validate the results.
Much like my math teachers, this is the only way I can tryst what it came up with.
And in the case of asking it help rewrite/revise/condense/add-to coding languages, it also helps you learn instead of just doing your homework for you.
My human gut “feeling” that AI used to generate feeds and speeds will be greatly influenced by tooling manufacturers recommendations, which is often heavily biased towards super high speeds and high material removal rates.
Most of the information I’ve found on feeds and speeds is attempting to optimize the cutter’s use in a production environment with big rigid machines. For us Onefinity users, we don’t want to be ripping our cutters through material at those rates.
I assume most Onefinity users would prefer feeds and speeds that have the highest likely hood of not breaking bits, and not ruining a project.
I agree. The learning model probably did not get a lot of non-industrial oriented data to learn from. Even within the consumer grade machines there’s a lot of variability in the machine’s construction and the type of motor spinning the but. OF is pretty robust and rigid vs some of the lower end machines. And spindles are more capable of high speeds than palm routers. But even spindle specs are often b.s. because they publish optimums or theoretical numbers that aren’t achievable in real life (like 2hp 110V 20A spindles). The AI models aren’t set up to make those distinctions so they’re going to be overly optimistic for us.