Sigma filtering values

[[Ja...]]

I’m wondering how others use the sigma filtering values. What criteria do you use when you’re changing them from the standard +/- 3 sigma within each characteristic? The cookbook really doesn’t deal with this. From my own experience, I’ve found that the smaller the tolerance window, the more aggressive the filtering needs to be in order to get good correlation with hard gaging, but I don’t have any set guidelines. Is there a formula I should be following?

[[De...]]

Hi Jack, we set ours to +/- 4 SD's on standard and occasionally go down to +/-3 SD's. we use this to throw away points that are truly errant (mostly FM) and 3 SD was throwing away valid points. your just looking at a standard of keeping 99.7% vs 99.99%, we just found that we wanted some points in that last 0.29% especially for curves. if you are having issues against hard gauging, using filtering to eliminate sensor noise, fitting methods to get matching evaluation techniques, or Feature/characteristics DOF's/Alignment issues might be a better place to look.

[[Ja...]]

I agree, I think. My concern really isn’t with matching up to dedicated fixturing, it’s more along the lines of size or flatness. Say for example I have a circle that’s toleranced at +/- 0.0001”. Many times I need to get rather aggressive with the filtering so that the results match up with a class XX gage pin. So I’m wondering at what point is it ok to start filtering out points, and just how aggressive should I get?

[[De...]]

I am not sure what machine and sensor head you are using but a 0.0001" tolerance is probably to tight for the piece of equipment you are using. Class XX pin gauges typically only come in 0.0001" increments, so you are not going to have good resolution there. When going after such tight tolerances, many aspects have to be taken into account.

To start having a good qualification on the probe < 1 micron, getting your alignment square so your data is cutting the circle perpendicular and not getting a ellipse, then inputting appropriate filtering (7-10 data points per filter window), feeding into the previous an adequate number of points perhaps a step with of .001", then a max inscribed fit to match the gauge pin.

[[Ja...]]

Apparently I’m not doing a good job of explaining what I need. I apologize. When I get a report that says a given hole is undersize/oversize, and a XX gage pin check tells me that it’s good, that’s not reasonable correlation to me. We have three CMMs, two of which are fairly new, and all three are in controlled environments. I use an Accura II for most of my work. Probes are qualified to well under a micron, usually ½ that or less, and at a reduced pressure to minimize deflection. Alignments are looped up to four times to get sin error as close to zero as possible. UPRs and cut-off lengths are calculated for each feature as well as pre-filtering. Stylus size follows ISO guidelines, and thousands (often 10s of thousands) of points are taken. And even though we have on-board temp comp, I usually don’t attempt the really tough measurements unless the part has thermally stabilized at 20C +/- 1C. My results usually get very close, but often times I need those last few millionths or a tenth to get reasonable correlation. I do recognize that I’m pushing the limits for an Accura II. So that’s why I was asking the forum about how they determine sigma values. Hope I did a better job of explaining what I’m after. Did that help?

[[De...]]

From the information I am able to find online, the most accurate new Accura II appears to have an allowable error on linear dimensions in one axis of 1.2 µm which is .0000472 in. As the machines get bigger the spec allows significantly more error.

I think what the previous poster was trying to say is that the allowable error in the machine specification is probably contributing to your total error more than filtering is. You are trying to chase down the error in your measurement based on filtering, when the error you are seeing is significantly below the allowable error in the machine itself.

In regard to your original question, I would not be too concerned with +/- 3 sigma, but I don't think you should generally need to be more aggressive than that. If you are I think you are beginning to get to the point where you are throwing away real data.

[[Ja...]]

"If you are I think you are beginning to get to the point where you are throwing away real data."

Ya, that was definitely one of my concerns and the reason for the original post. It seems interesting however that on the tight tolerance stuff that's marginal, as I get more aggressive on the filtering the findings almost always move in the direction I need them to - which was another reason for the original post, and one reason why I think I can pull just a shade more accuracy out the machine. But if the general consensus here is that I'd be eliminating too much data, I guess I'll have to look elsewhere like Derek Hinrichs mentioned for better correlation. I'm hoping there will be others who are experiencing the same thing I am.

[[Ri...]]

A few things:

1 - What evaluation method are you using to calculate your size? LSQ or Outer Tangential?

2 - What is the size of the feature you are inspecting?

3 - What is your speed and point density?

4 - What filter cutoff are you using?

5 - What kind of tolerance are you trying to hold? I understand that the XX pin has a tight spec, but the part that you are inspecting, what is its tolerance?

[[Ja...]]

I’m not looking for an answer to a specific feature, I’m looking for more of a rule of thumb as to when people tighten up (if ever) on the sigma filtering values. So to answer your questions…
1) That depends on the feature. Most of the tight tolerance stuff relates to dowel pin holes. If it’s a clearance hole used for locating, then it’s evaluated using maximum inscribed. If it’s a hole that’s going to have a dowel pin pressed in, then it’s LSQ because of the interference fit. The flat surfaces are for the tops of cylinder blocks or manifolds.
2) Most of our dowel pins range from 1/8” to 3/8”. Flat surfaces range from a circular 2” diameter to a rectangular 23”x11”.
3) Point density is always high for dowel pin holes. Step width can be as little as 0.0001” and rarely exceeds 0.0003”. Calypso won’t let you go very fast with that kind of density. For flatness, that depends on how large the surface is. A general rule of thumb is a step width of 0.001” but I usually have to stretch that out on our larger products because the point cloud gets too big.
4) UPRs and cut-off lengths are calculated by feature size or scan length using ISO guidelines.
5) The XX class gage pins have a tolerance window of 0.00002”. For the press fit dowel holes the tolerance is usually +/- 0.0001 or +0.0002/-0.0001. For dowel clearance holes, the tolerance is usually +/- 0.0002.

[[De...]]

Hi Jack et all, The machine accuracy of the scales is one portion but you also have a Probe sensor error of likely +/- 0.0001" or 0.00005" and the stylus accuracy of as you said 0.000020" to 0.000040", these are all stacked (not a linear addition) for your total machine accuracy, that's a nice machine you are running.

[[Da...]]

Hi Jack, I run two micuras. I understand all about going below the stated accuracy of the machine. The micura is a .7 machine but, our calibration reports show us using about half of that.

I had to Re-learn some things in order to utilize these cmms to their full potential . I came from a few different conturas and let me tell you, the micuras are a whole different world. One of the most important things that I had to address was part holding and fixturing . Trying to check .0001" tolerance and under, means the part CANNOT move at all during the process. This sounds simple but, it only has to move .0000020" to screw up a reading. I use mag chucks and solid fixturing for almost everything we do.

Cleanliness is of utmost importance. We run mostly bores and dirt has nowhere to go on an ID except under your probe.

Watch roundness , and the shape of the roundness. Two point out of round will cause double the size variation when compared to a two point size check. ( mics, bore gages ect)

To hold and repeat super close tolerance you will have to collect a butt load of points and carefully filter and outlier eliminate at a pretty aggressive rate. Before you worry about eliminating a handful of "good " data remember that we are collecting (scanning) 100's of times the number of points that a non scanning machine typically uses. Throwing out some good points is quite preferable to hanging on to even one BAD point.

Getting busy here, have to cut it short....be back later.

[[Da...]]

As far as the Sigma filer settings I leave it at 3.0, changing it produces pretty drastic changes. However, I do use 2 or sometimes 3 iterations of the filter. Each iteration uses the remaining points to re-figure the 3.0 sigma limit. I also include adjacent points (amount depends on condition of the surface) . This is WAY less aggressive but still gets rid of the bad points quite nicely !

[[Ja...]]

Thanks Dave. Good points. I've not had much luck with iterations in the past but I'll do some more experimenting with it.

[[Da...]]

Jack, it's good to see someone here that's pushing the limits. 🙂

[[Ja...]]

Dave, if I understand correctly, why do you prefer the “Include Adjacent Points” option vs “To Computed Feature” option? Wouldn’t you want to be able to choose between the two depending on what the feature is (IE: Circle vs Cylinder)? Or am I interpreting this wrong?

[[Cl...]]

Dave, are your Micuras in A2LA type clean rooms (temp/humidity controlled)?

[[Da...]]

Jack, To be honest with you , I don't understand fully how to feature works, so I use what I know. 🤣

To Clarke, No , but our temperature system maintains about +/- one degree from 68. Our parts are so small temperature variations of that small of an amount don't add up to much, and we use the temp probes.