Evaluation Methods

[[Ke...]]

So, I have been trying to inspect the radius of a sphere, that only uses *maybe* 10% of the feature... I'm comparing CMM results with MarSurf results, in an attempt to get some sort of a correlation. I was leaning toward ditching the CMM method, and sticking with the Mahr, when I decided to play around with the evaluation method, and using the L1 method, results in a *strong* correlation with the Mahr, but with a tighter range (on 5pcs). So I'm going to use the L1 evaluation...

Toggling between LSQ, Max Circ., Max Inscribed, Inner Tan, & Outer Tan, does not change the results at all.
The only methods that give different results from LSQ are Min Feature & L1 Feature.

The question is: What is it about the L1 evaluation that works so well for this feature?
This is the only time I have ever found this evaluation method to be useful...

[[Cl...]]

From Calypso help;

With the calculation as least absolute value feature (L1 feature), the geometric element is determined in such a way to minimize the sum of the deviation values.
This best fit is insensitive against outliers and leads to a clear result with low computational effort. Copyright © Carl Zeiss. All rights reserved. 2019-05-07Best fit according to Chebyshov (Minimum Feature) Best fit as least absolute value feature (L1 feature) Calculation as circumscribed/inscribed feature

[[Ke...]]

Please sign in to view this quote.

I saw that in Calypso help.... I was hoping someone could "dumb it down" for me 🤣

[[Cl...]]

Sorry, I don't really understand it myself. There are plenty programmers out there much sharper than an old guy like me. 😉

[[Ja...]]

Keith,

Imagine a perfect sphere. You are taking points on a spherical segment, forming an imperfect sphere. The L1 feature evaluation method will use Lp norm with p = 1 to resolve your geometry. What this means is that the software will consider your imperfect sphere in relation to a perfect sphere. The software will compute a sphere such that the direct surface deviations are minimized - ALSO under the condition that the fitted sphere will lie on or outside of the material side of your sphere.

Traditionally, a LSQ fit will minimize the square deviations with no further constraints, allowing a resolved geometry to be formed from points on it's own interior. This does not match physical reality. And LSQ, being unconstrained, is technically an unconstrained L2 feature. You are using L1, so the deviations are not squared but rather directly summed. I also believe that Calypso is computing a constrained L1 feature, so your CMM result will much better match physical reality while not being subject to extreme outliers like inscribed/circumscribed elements are.

Check out this link if you're interested, it's a great read. Section 3.2 digs into what you are questioning. Follow the link and click the download button, it is a free research article and the site has many more fascinating ones. https://www.researchgate.net/profile/Vi ... ion_detail