What is the maximum tolerance for a DIY Machine ?!

[hanermo2]

There are exactly zero limits.
Dan Gilabert made a 1 micron lathe.

My lathe has 1 micron mechanical resolution and positioning ability.
It is not natively accurate to 1 micron, but I can move by 1 micron to any offset size I can measure, using a digital 1 micron dti to confirm relative movement.
But I use thick ballscrews, 32 mm D, and 10.000 count ac brushless servos of 10Nm peak, 750W cont.
Lathes are about 10x more accurate, inherently, than milling.

This is about 800€ more for one axis than a typical stepper setup.
It is not too cheap, but not too expensive either, in context.

Anyone can make an xy machine with 0.01 mm repeatability, and 0.01 mm resolution.
A lot better is fairly easy to do by using
-bigger linear guides (stiffer, more rigid)
-thicker ballscrews (stiffer, more rigid)
-better screw mount (stiffer, more rigid) with better bearings.
-AC servos for more accuracy per rev

It is not that it is hard, as such.
But each improvement of those 4 costs about 100€ extra for one level and 200-250€ extra for 2 levels.
So for 2 levels or 2 sizes bigger/better you pay about 250€ x 4 == 1000€ extra.

So you can get 2 micron resolution for one axis, on a mill or router.
1 micron if you work more and do "better" in many ways.

So a 2 micron error, aka +/- 1 micron, is 2x2x2 = 8 microns volumetric local accuracy for milling/routing.
But screw errors tend to be about 0.1 mm, worst case, on typical 40-800 mm long axis travels.

So it is nowhere near inherently *accurate* to 0.01 mm (basic one axis), or 0.002 mm (very good axis), since the position is wrong.
But it can easily repeat to 2 microns, or 0.01 mm worst case, cheap.

Endless ways to contact probe, map, digital 1 micron dtis, glass scales, so you can offset the position to be accurate to 1 micron or less.
Granite surface plates, accurate to 2 microns, cost about 300€. 640x400 mm.
Gage blocks to make accurate probing points. Cheap to about 500 mm.
You can setup optical switches, accurate to about 1-2 microns, for less than 20€ each.

You can calibrate any screw, and map errors out to about 25x more inherent accuracy.
Thus you can offset the screw error in sw, by adjusting offsets or many ways in gcode, macros, probing, etc.
You can map the screw error in some controller sw, with various caveats.

It is not too hard or too expensive to map screws pretty well, more expensive is easier and more accurate.
Mapping to 0.01 mm positional error, per axis, is pretty easy.

Making stiffer screw mounts is easy, just make them 2 sizes bigger.

Example:
The ballnut mount on my lathe z axis is 120x120x70 mm tool steel. Yes, about 3" thick.
It was not *hard* to do, but took about 40 hours and 20€ in steel.
It is hard and slow to drill 120 mm deep, 10 mm, in tool steel.
12xD or 12 diameters deep.

Most accuracy is mostly making things bigger, thicker, more rigid, in steel.
After that some 30€ bearings, info, and work gets you better results.
E:
Like making a fixed-fixed screw, with tension, and a compliant tension mount.
This means you pull the screw about 500 kgf force, 1200 lbs, with an arrangement that allows the pulling end to move axially.
It cuts the free length of the screw by half.
This makes it 8x more rigid.
A tensioned screw is inherently 2x more rigid.

A 32 mm D screw == 1400 kgf max thrust, so pull about 500 kgf.

So you get 16x more rigidity for 60€ in materials and 2 days work.

I was wondering what have you seen achievable out there with DIY machines ? what tolerances were achieved and on what material ? I am asking to have realistic expectations of what is possible.