[canadian_tired]

Exactly. I can't speak for laser, extruders, knives or such, but for subtractive work (i.e. milling / drilling operations with some form of rotary tool) this design would be an exercise in frustration. Typically, you would have encoders on each axis to tell the machine where things are happening. These encoders are excellent, but any flex/slop in the chassis or moving parts is not caught by the encoder... so the software will not know that the cutter is in the wrong place. Thus, crappy parts. Any reasonably useful CNC design takes rigidity and backlash prevention seriously. Take a look at PocketCNC for a good (and yes, expensive) design.

[galangalalgol]

could you use strain gauges to measure the deflection and fix it in software?

[mikewarot]

>could you use strain gauges to measure the deflection and fix it in software?

No. There's no substitute for rigidity in fixturing and of the machine itself. There will also be vibration motions that have unpredictable nodes, frequencies, and amplitudes.

Consider that a 2 flute mill at 20,000 rpm will induce vibrations with a fundamental frequency of about 670 hz. This would require a servo system with a bandwidth of about 2 khz to correct, a sampling rate of 8khz would be a bare minimum.

Next, what to do if the tool digs in, and shoves the part? You have to have backlash sufficiently damped, and rigidity enough to keep the tool or part from being snapped off.

There's a good reason machine tools are as heavy and rigid as they are, and why they must be on a thick foundation, and leveled precisely before use.

If you're going high precision, you'll also have everything in a temperature controlled environment, and allow the materials to soak at a day to achieve the proper temperature. Cooling is very important when you get into fractions of a millimeter, or "tenths" (1/10,000 inch) tolerances.

The best machine tools are old American cast iron machines from about 1940 to 1960, rebuilt with modern control systems.

[jerome-jh]

The rotary axis is controlled by a stepper motor. This is not a usual turning lathe. The rotary axis can be made as slow as necessary and the only relevant analysis is a quasi-static one. Answering to parent comments: there are usually no encoders on these machines. The controller firmware just counts the motor steps. If a motor misses a step, everything will be off from then. For the cantilever design: it is already used by Prusa 3D printers which are well regarded (in the field of amateur 3D printing) so I do not think this is an issue for the targeted accuracy which is typically 10 to 50 micrometers.

The only issue I have with this design is that the accuracy decreases with the distance from the rotary axis, hence it is not uniform over the building volume.

[Accujack]

>For the cantilever design: it is already used by Prusa 3D printers which are well regarded (in the field of amateur 3D printing) so I do not think this is an issue for the targeted accuracy which is typically 10 to 50 micrometers.

3d printers don't have forces pushing back on the print head, nor the vibration from a cutting tool to deal with.

This design might work ok for 3d printing, or laser cutting, or anything else where no contact between the working head and the material occurs.

As far as accuracy and repeatability, it's going to be about as good as the average 3d printer design. It's a nifty idea for a system, but it's only a decent machine for cutting if it's scaled up massively and built of cast iron or similar materials.

[kadirilkimen]

Depends on what you cut. It cuts wood pretty nice. I couldn't documented the wood cutting performance on prototyping. I will demonstrate it's wood cutting capacity in v1 release. It's a 3d printable desktop machine and never intended to cut precise metal parts. May be I can design a full metal body version with professional components like rail guides instead of plastic v-slot wheels.

[kadirilkimen]

It is 0.025mm on the outer edge of the table.

[sbierwagen]

I don't know if there's any strain gauges designed for this application. I'd be worried about nonlinearity, creep and how you would calibrate it in the field. (If it requires a dial indicator and a surface plate, you'd probably be way better off getting a regular CNC mill)

Optical tracking might be easier.

I'm not sure how much deflection you could actually tolerate before deforming the joints in the arm. Even with deflection tracking you might end up taking very, very shallow cuts. A 3D printer that runs for 16 hours in your apartment is annoying, a CNC mill running for 16 hours will be genuinely intolerable.

[harperlee]

Great question. Human arms are also very bad in rigidity, repeatability, etc. but we manage to do great work by adjusting. Of course we try to move and rest on things to get out of the error-prone postures (so in effect moving from this to a gantry type of configuration), and need to measure, create jigs and templates, etc. but there is something here to be covered by sw for sure.

[morpheos137]

Comparing human arms to CNC "arms" makes about as much sense to me as comparing a sparrow to an F16.

Yeah both humans and multiaxis cnc machines can articulate their arm to apply a tool to a work piece and both a sparrow and an F16 can fly through the air.

[ripe]

I like your analogy, but just to clarify, which is the human being in your comparison: the sparrow or the F16?

A sparrow can take off and land on a branch, and it regularly weaves through thick foliage no wider than its wingspan. So, I assume the human is the sparrow?

[morpheos137]

Yeah I was thinking of the human as the sparrow.

[Turing_Machine]

You could probably use it for ECM or EDM, but yeah, not stiff enough for traditional hogging cuts as a lathe or mill.