Radius measurement with small arc

[[Ba...]]

Hello,
Backstory:
Zeiss had written some programs for us to measure multiple features on this part. They used curves to measure the radius in balloon 16. We are after the results of balloon 21, the center of the "16" radius to the top of the tubes. I have verified these on a vision system, and the results vary significantly from what the CMM is outputting. We aren't recording the value of the radius, just primarily after balloon 21 (0.080 - 0.100). Calypso will almost always output these results as 0.99 - 0.1 with no variation. I have measured these to be about 0.084 - 0.087 using a smart scope. We are using a Duramax, Calypso 2024, 0.5mm probe / also attempted a 0.3mm probe on an angled knuckle @ 45 deg to minimize deflection. These are fixtured with the tubes aligned with the Z axis.  

I have made my own copies of these programs to play around with no success at improving these features. Any suggestions would be appreciated. I feel that vision would be the best way to achieve these accurately and repeatably.

Thank you!

 

[[Ma...]]

For small angle on radius you can measure their position with constraining their radius - forcing calculations with nominal radius.

What i found on small radiuses ( R1.0mm; 90° ) is that your probe needs to be on correct position - so do alignments, bestfit - whatever to be on correct spot for taking points.

[[Be...]]

Locking the radius is definitely the way to go when you need the location of such a small section of a radius.  You may consider also locking the X location AND using a formula to determine the X-location.  It kind of depends on the tightness of the tolerance that determines that dimension as compared to the tolerance for the radius etc.

Are the features in bubble 16 the same as the ones in bubble 21?  I would say that they are from the picture shown.  Look at the way that the feature is programmed.  They may have locked the XYZ location to get a more reliable radius.  They may have even done something more sneaky and difficult to program in order to get that dimension cleanly.  In the most complex version of the programming, they already have the information hidden in the data and you would just have to coax it out.  Even in a moderately complex version, they may have all the information that you need. Judicious use of recall feature points and formulas for nominals could get you what you want without taking any additional datapoints.