@
Kai Kendall. It would be interesting to know your approach to solving this problem in Quindos. The only approach I know of is to take all measured data and try to fit it against a simulated 'perfect' part. The constraints is not just the datum features but all features and varying tolerances which is what makes it insanely difficult. This is not a new problem and if a good solution existed everyone would be using it by now.
One comment that someone mentioned to me long ago was that if you only have a few features you have some 'wiggle' but this disappears as the number of features increase. Picture a mating part with three holes/pins where you can shift it around a bit then compare it to a similar part with 100 holes/pins; the second version won't wiggle nearly as much as the fist version. They did a study on this and decided that simply ignoring the datum bonus was the best, practical, option as the results were pretty much the same anyway when you have a lot of features.
If you are doing this the goal should not be to make features 'perfect' either. Every datum shift that improves one item makes something else worse by the same amount. Tolerances come into play particularly in this case.
@
Kai Kendall. It would be interesting to know your approach to solving this problem in Quindos. The only approach I know of is to take all measured data and try to fit it against a simulated 'perfect' part. The constraints is not just the datum features but all features and varying tolerances which is what makes it insanely difficult. This is not a new problem and if a good solution existed everyone would be using it by now.
One comment that someone mentioned to me long ago was that if you only have a few features you have some 'wiggle' but this disappears as the number of features increase. Picture a mating part with three holes/pins where you can shift it around a bit then compare it to a similar part with 100 holes/pins; the second version won't wiggle nearly as much as the fist version. They did a study on this and decided that simply ignoring the datum bonus was the best, practical, option as the results were pretty much the same anyway when you have a lot of features.
If you are doing this the goal should not be to make features 'perfect' either. Every datum shift that improves one item makes something else worse by the same amount. Tolerances come into play particularly in this case.
I was thinking to throw all shifted points into a bin, creating variables for possible allowable positions and degrees of rotation for each modified datum. Allowing these variables to be over-written as the part measurement proceeds, but ONLY if the freedom is reduced as more and more information is collected.
This would allow simultaneous evaluation to work correctly, even when different feature control frames are used throughout the drawing. No real reason this will not work in PC-DMIS as well.
