Disagree with the algo used for max inscr and min circ cylinders

[JEFMAN]

I talked about it in another thread, and I just made some tests.
I believe that this bug had been fixed long time ago, I can't remember which version it was...
When you construct a max_inscribed or a min circunscribed cylinder from a LS one, there's no reason that they have the same vector.
If you look at the form (dimension, not the circularity), you can see that only 2 (or 3, it depends) values are zero. There should have a mini of 5 (I think).

This seems to be an important problem if you use a cylinder as a primary datum, because the vector is wrong.
For a secondary, it's less important because you can project the hits on the primary, then construct a max inscr / min circ circle which correspond exactly to the definition of the datum.

I think it's also a problem for "envelop" requirement, because the diameter could be wrong (if the form is wrong, how the diameter can be right ?)

I tested this with 2018R2 SP3 and 2014 release.

Rob Jensen , Don Ruggieri : something to do ?

[vpt.se]

Subscribed!

[dph51]

Subscribed!

[AndersI]

When you construct a max_inscribed or a min circunscribed cylinder from a LS one, there's no reason that they have the same vector.

Agree on that, can you give example data?

If you look at the form (dimension, not the circularity), you can see that only 2 (or 3, it depends) values are zero. There should have a mini of 5 (I think).

I'm not sure what you mean by this?

[JEFMAN]

Agree on that, can you give example data?

I'm not at the cmm for the moment, I will post an example on monday.

I'm not sure what you mean by this?

I think (that was the end of the text ) that a max insc or min circ cylinder is a perfect cylinder that fit the cop of hits so that no point is going inside / outside.
A perfect cylinder needs 5 points (remeber Flexylearn, Anders !), even if PC-DMIS wants 6 (I'm not sure here).
So on the form dimension,you should find 5 zero.( text report, or calculate each T_value)

Where I'm sometimes surprised, it's when I can calculate a feature from Excel / solver, and when the soft can't

Happy week-end, all !

(If the weather is nice, I think I'm going to walk in Pyrénées )

[vpt.se]

Credits for mentioning FlexyLearn.

[bfire85]

Subscribed!!

[Wes Cisco]

Subscribed and hoping Rob Jensen gets involved.

Pc-Dmis 2019 R1 has been certified for Least Squares by the German PTB. You can download the cert if you register/login. (link below)

To the best of my knowledge there is no institution out there certifying any other AlGorethyms. Sadly. Hopefully that will change some day soon.

https://support.hexagonmi.com/s/article/PTB-Certification-for-PC-DMIS-2019-R1


HTH

[DungT]

I'm not at the cmm for the moment, I will post an example on monday.



I think (that was the end of the text ) that a max insc or min circ cylinder is a perfect cylinder that fit the cop of hits so that no point is going inside / outside.
A perfect cylinder needs 5 points (remeber Flexylearn, Anders !), even if PC-DMIS wants 6 (I'm not sure here).
So on the form dimension,you should find 5 zero.( text report, or calculate each T_value)

Where I'm sometimes surprised, it's when I can calculate a feature from Excel / solver, and when the soft can't

Happy week-end, all !

(If the weather is nice, I think I'm going to walk in Pyrénées )

What’s the arrangement of 5 points to get a perfect cylinder? 2 points on each end of a cylinder on the same side and the 5^th point is in the middle on the opposite side (3 cross sections, 2, 1 and 2)?

[JEFMAN]

On Flexylearn, it was two points to rotate the axis and 3 others to measure a circle.
By definition you need a mini of:

• 1 point for a point !
• 2 points for a line
• 3 points for a plane or a circle
• 4 points for a sphere
• 5 points for a cylinder
• 6 points for a cone

[DungT]

On Flexylearn, it was two points to rotate the axis and 3 others to measure a circle.
By definition you need a mini of:

• 1 point for a point !
• 2 points for a line
• 3 points for a plane or a circle
• 4 points for a sphere
• 5 points for a cylinder
• 6 points for a cone

I get that. I just wondering the arrangement of the points for a perfect circle. You just can't get a perfect circle with 3 points on one end and 2 point on other end, or 4 points on one and the last point on the other

[JEFMAN]

I get that. I just wondering the arrangement of the points for a perfect circle. You just can't get a perfect circle with 3 points on one end and 2 point on other end, or 4 points on one and the last point on the other

That's why, if you don't "help" the soft, you can construct a cylinder at 90° of what you want !!!!!

[DungT]

That's why, if you don't "help" the soft, you can construct a cylinder at 90° of what you want !!!!!

Let’s say a bore is perfectly round, perfectly straight but it’s tapper at a constant rate.
Circle 1) 2.000 location 0,0,0; IJK 0,0,1
Circle 2) 2.002 location 0,0,1; IJK 0,0,1
Circle 3) 2.004 location 0,0,2; IJK 0,0,1

The size would be 2.000 at max insc and the IJK would be 0,0,1, but if you simulate the cylinder with a hard gage, the IJK would float around on one end.

I wonder how would you handle this case with Excel

[JEFMAN]

I believe, in this case, that Excel would give the same kind of error than PC-DMIS, Calypso, Metrolog, Quindos (...) if you construct a bad feature with a COP !!!
The solver of Excel uses the minimax algorithm (and others), if you know how to write what you search, it gives very good results.
I often use it to check PC-DMIS values on new versions, I think it's very funny, and it's easier than using Matlab, which is very expensive, or Scilab or Python (I don't know them enough to do this kind of optimization ).

[DungT]

I believe, in this case, that Excel would give the same kind of error than PC-DMIS, Calypso, Metrolog, Quindos (...) if you construct a bad feature with a COP !!!
The solver of Excel uses the minimax algorithm (and others), if you know how to write what you search, it gives very good results.
I often use it to check PC-DMIS values on new versions, I think it's very funny, and it's easier than using Matlab, which is very expensive, or Scilab or Python (I don't know them enough to do this kind of optimization ).

IMO, PCDMIS cal'c correctly in this scenario. I'm just trying to wrap my head around how can PCMIS applies Flexlearn in real world

[JEFMAN]

IMO, PCDMIS cal'c correctly in this scenario. I'm just trying to wrap my head around how can PCMIS applies Flexlearn in real world

I understand your point of view, I believe honestly that PC-DMIS gives the right result because it uses the Least square axis as Max inscribed axis.

If you take your example with only 2 levels, 3 hits by level, and the first level rotated of 60° to the second, I can't see why the max inscribed cylinder should have 0,0,1 as vector... Because the definition uses 5 hits !