My little Stepcraft 420 consistently have a dimensional error along X and Y axis: it makes all parts 0.2 mm smaller (two tenths).
Yesterday I've made a test-session by milling some plates on a simple MDF plate using a 2mm end-mill that I regularly use for alu/plastic.
test 1: a rectangular shaped 53x48 mm plate with some circular holes of various diameter located near the four corners of the plate.
test 2: a rectangular shaped 100x100 mm plate with some circular holes of various diameter inside.
In both case the plates was around 0,2 mm smaller in X and Y dimensions (tolerance +/- 0.03 mm).
I immediately thought about a possible run-out problem at the spindle-collar, but... all the circular holes are almost perfect! (Maybe just an error around 0.003 mm - 3 cents of mm).
In case of run-out it should make all the parts smaller of 0.02 mm, isn't it?
As next step, I've double checked the generated G-code, but it seems simply perfect to me. Also the coordinates shown on the computer screen while machining the part are ok.
I'm going crazy with this problem. I cannot realize where the problem is.
I also thought that my machine is loosing steps. But in this case I should observe a grater dimensional error when doubling the size of the plate (from test 1 to test 2 above). Isn't it?
I made my tests on MDF intentionally maintaining the same mill-bit, feed-rate, and DOC I'm using on ABS and POM plastic (dia = 2mm, feed = 400 mm/m, DOC = 0,5 mm). They seems pretty conservative to me.
I've shaken the portal in any direction. There is no evident play or backslash issue. The machine seems rock-solid on every axis.
As I remember, the machine is configured with 200 steps per revolution (original HF 500 Spindle from Stepcraft). The max acceleration should be around 30 mm/s^2
(I have to double check this values since the machine is not with me at home.)
I had a similar problem in the past on another router machine.
I had slightly too harsh feed vs depth of path and the dimensional error was coming from tool deflection.
Have you tried the same test patterns with a slower feed?
I tried different feeds (adjusting RPMs) with the same results: that 0.2 mm error won't go away, while all the circular shapes are pretty good.
I wonder how tool deflection can occur on a soft material like MDF. I'm not pushing the device to his limits in anyway.
today I decided to make a deep and serious test about backlash. The results have put me in profound depression...
I'm quite disappointed from the machine. I've always machined soft plastic like POM or ABS, rarely I've machined alu sheets (max 1mm) and/or pipes (max 2mm of wall).
A backlash of 0.36mm is enormous in my opinion. I have big eyes.
Now I will try to apply the backlash compensation provided by LinuxCNC. Maybe it will help to minimize/reduce this problem, but I'm quite skeptic about a software solution.. we will see.
Anyway a mystery remains: why circular shapes have an error of just few cents? My hypothesis is what follows: since the circular pockets are made by a series of spiral movements along X and Y axis both, then there is a constant force that maintains the center of the spindle quite "stable". This is also because of the softness of the material and the modest deep-of-cut (0.5 mm).
About the right procedure for measuring the backlash, a simple goggle research produces tons of examples.
in my test I've placed a tiny sheet of rubber below/behind the caliper gauge to keep it firmly in position. The material is Viton, a really "sticky" rubber, but I think any rubber with medium/low hardness will work the same.