Roundness doesn’t repeat

[Hirxm]

Hello everyone

I’m measuring the roundness of some motor housings. The spec is 0.08. So far the results are within tolerance (0.052, 0.058...) but I’ve noticed that the results don’t repeat. I measured one part three different times and the results vary wildly. First I got 0.067, the 0.062 and finally 0.057. Why does that happen?

How I measure? I use SP25M, SM25-2 and the probe is 4x40mm. After the alignment I create a circle and I report the roundness of that circle.

Thanks.

EDIT: I was thinking that if I create a cylinder and report its roundness I would get a better result?

EDIT 2: The results vary only if I move the part. If I measure it several times without taking it out of the CMM the results are perfectly fine but if I move the part and then put it back, I get the varying results.

[louisd]

how is your alignment created? if your level plane is just 3 hits on the top of a rough surface for example, this can be a result of probe points not normal/square to the bore/boss.

[Hirxm]

how is your alignment created? if your level plane is just 3 hits on the top of a rough surface for example, this can be a result of probe points not normal/square to the bore/boss.

level 3 point plane
rotate 2 point line
origin 4 point internal circle (same as the one I take the roundness from)

[louisd]

Yeah, that's not enough to start cutting hairs, espeically if those are manually executed. Make a set of features using MODE/DCC, to replicate your alignment structure. Take more hits or a scan (if your equipment is capable) of the top plane you are leveling to. Take more this for rotation line, and take more hits for your internal circle.

If you are checking circularity, technically, you need to be level to that circle's cylindrical axis, and not a plane. If you can measure the cylinder that you are trying to output circularity on, level to the axis of the cylinder, then measure the circle, it will become much more repeatable.

[kingsld1]

[QUOTE=H

EDIT 2: The results vary only if I move the part. If I measure it several times without taking it out of the CMM the results are perfectly fine but if I move the part and then put it back, I get the varying results. [/QUOTE]

Low point density will cause this. The odds of picking up the absolute maximum and minimum point are pretty low if you don't. If you change the rotational angle of the circle you are not touching the same points and will get different results.

[louisd]

EDIT: I was thinking that if I create a cylinder and report its roundness I would get a better result?

This is entirely dependent upon what your drawing is defining.

Roundness of a cylinder will result in similar uncertainty, if the probed points are not normal to the surface of the cylinder.
You need to define an alignment that makes certain the probed hits are normal to the surface of the part.
Any variation in your vector angle of the circle/cylinder points creates cosine error.

[Quality ish]

Even if you measure it as a cylinder, you may get a variation depending on the number of hits you're taking at each level. Are you scanning? Generally speaking, the more hits you take, the more accurate your measurement will be.

If you're getting a variation like you're describing, I suspect the cylinder you're measuring is out of round, and the hits you're taking are only occasionally representing this because you're only occasionally hitting the cylinder in the section of the radius that varies.

If your CMM doesn't scan, make sure that you're never touching off the cylinder at four places on one level, and four places on a second level. PC-Dmis has a few quirks, and this is one that may lead you down the wrong path. What might happen in this case is that the four hits on two levels may (to the machine) represent the eight points which could construct a cube, and then while you're programming one thing, the computer is interpreting it as something COMPLETELY different. This quirk is often referred to as "the demon". When measuring cylinders, in my opinion, more hits is always better.

[AndersI]

You need quite a lot of hits to find the roundness - to really find the actual roundness you need a continuous scan around the full circle. Period. (*1)


Some examples what happens with fewer hits:

3 hits - roundness = 0
6 hits on a three-lobed circle - you get roundness between 0 and actual roundness depending on where you start
7 hits gives a mathematical guarantee to find at least 79% of the form error on a three-lobed circle.
9 hits gives something similar for a four-lobed circle
and so on.

I.e. if your circumference is wavy you need more hits than two times the number of waves in a full circle.


(*1) Actually, I'm lying - to really, Really, REALLY, find the actual roundness you need an infinitely small tip, to eliminate the filtering/smoothing done by anything with an actual surface...

[Hirxm]

Hey thank you all, I actually scan the desired circle. I've changed some parameteres like the filter and the speed at which it scans and so far I've been getting better results!