Evaluation settings

[[Jo...]]

I realize this is a broad topic on it's own, but I am looking to get educated on when and how to use proper filter settings, outlier elimination, the relationship between preassignment for evaluation method of a feature vs. evaluation settings as a characteristic.

I have a rudimentary knowledge of some of this material, and 95% of the time I don't think the results would change in any meaningful way, if the tolerances are wide open. Some stuff I've been programming recently have relatively large features with very tight tolerances, for size and form (i.e. ~9" ODs with ±.00015" tolerances with .00020" runout back to other coaxial features, and surfaces with .0000" squareness with back to that OD that only allows MMC bonus for your tolerance) and I've had favorable results when compared with some basic hand tool gaging techniques. I always get a little concerned when I've applied all kinds of filters and the numbers start looking "a little too good because I worry I am actually telling Calypso to ignore data and giving out printouts showing the parts are good when they might not be. I feel like I have to fiddle with outlier elimination and filter settings a lot to get data that makes sense. But I also realize tolerances this tight are also pushing accuracy limits of the machines, even with temperature comp.

So my question is to anyone willing to share, do you go strictly by the instructions put forth in the Measuring Strategy Cookbook, or do you just take a common sense approach and treat different situations on a case-by-case basis? If anyone has any knowledge or resources that explain how these settings work in the background and how they are applicable that would be fantastic.

I still would love to know how measurement data is different with Gauss filtering vs. spline filtering...

[[Ma...]]

I don't have much experience with this, but i always eliminate some starting and ending points on opened curves.
For basic elements ( like plane, cylinder, ... ) it is doing Calypso automatically ( it is set up in configuration and editable ).

For waves i think it can be used to display straight washer which is in real world bended.
But as i said - i almost don't use it at all.

[[De...]]

The cook book is a great place to start, but each situation will be different. I typically measure a feature in LSQ with no filters or outliers, then apply the inscribed, circumscribed, minimum feature, etc as well as filtering and outlier info on the characteristic side. I do this because I want all the data so incase we want to apply a different method or filtering scheme, we can change and run subsequent evaluations. Bottom line, Characteristic will supersede what is in the feature settings. There are instances where we have ignored the cookbook and went with a more aggressive or less aggressive filter, and presettings.

[[Ja...]]

The tolerances I have to work with at my present employer are very generous so I really don’t need to apply any type of filtering, but I still do just for my own peace of mind so I know I’m giving them the most accurate answer I can. Yes it adds to programming time, but then I’m not second guessing myself when things get near the tolerance limits. I use a spreadsheet with my own formulas to help me speed things up so all I have to do is punch in a variable or two and I get what I need, then it’s simply copy and paste. The biggest thing I’ve found that helps with accuracy is to see how well a new program repeats. Once I’m done with all the tweaking, I do a mini R&R on it by physically removing the part from the CMM (wear gloves when doing this), putting it back on the CMM, and rerunning the program. I do that a minimum of 5 times. Yes, even on very long programs. Sometimes I’ll spend days on just one part number. For me, if a highly machined feature doesn’t repeat within .00015” from high to low in all my R&R runs, I start looking for what’s wrong. When I first started doing this, it was a real eye opener. Generally speaking, I found I was scanning too fast, with too much pressure, and not taking enough data. At first my boss was giving me the evil eye about taking so much time, but once he saw how it made some pretty big differences in some of the printouts, he pretty much leaves me alone now. The cookbook is a good place to start, although I found it didn’t always work in all situations. There’s nothing that says it will, so go ahead and make changes to what works for the machining/manufacturing your company does.

This is getting lengthy so I’ll try to quicken this up. 1) I do all of my filtering in the characteristic, unless I’m forced to do it in the feature, like for some of the curve features. 2) On wide open tolerances it’s not necessary to filter, but I do it anyway. 3) For filtering sigma values, the tighter the tolerance the lower the sigma value should be. Personally, I don’t go much lower than 2.25 – 2.5 because I don’t want to filter out too much information – it works for here, but may not for you. Be very cautious, and experiment. 4) Gauss vs. spline – the only time I use Gauss is on raw castings. 5) Pay attention to mechanical filtering IE: Using a large stylus vs a more appropriate smaller one. 6) Use the shortest, stiffest styli you can for very tight tolerances.

Hope it helps.