Outlier elimination / adjacent points

[[Cl...]]

Can someone please explain this to me?

My part is out of round by .0004, (.0014) I apply this, it becomes good. (.0007)

OD cylinder .57 dia. polished Ti

Outlier_Elimination.PNG

[[Mi...]]

Please sign in to view this quote.

Sometimes applying a filter removes the points that make it out of spec...? I'm not sure what you're asking here.

[[Cl...]]

Based on the part diameter and finish, should I have this option checked as shown?

[[Ri...]]

When it calculates the outlier it will include the n number of points next to the outlier detected point.

In this case if you had one point that was an outlier it would kick that out plus the 3 adjacent points to it.

It makes the outlier elimination more robust.

Applying outlier elimination to a form measurement should be used with caution as you could potentially eliminate an actual defect on the part.

It would be best to look a the form plot with only filtering applied first, and then start applying outlier elimination.

[[Er...]]

Point density is important too. For a tolerance of 0.001 inches cylinder roundness on polished Ti I don't have experience to guide you.

Strategy also can make a huge difference. I prefer multiple circle paths because I can see the form error easier on the graphics. But that depends on setup, etc.

I use the graphic CAD Evaluation and set the Magnification to a crazy number like 1,000 so I can get a feel for where the form error is and look for a pattern.

I had a part with a similar tolerance and using this it was obvious to me and the machinist that the three jaw chuck was distorting the part. Him seeing the graphics was more telling than just a number. He knew right away by seeing the graphic he would have to change setups.

And in the end it comes down to form, fit, and function. Having the customer buy off on a part can be a lifesaver.

[[Ow...]]

The cook always recommends adjacent flavors/points. ➡️

Condenced_Zeiss_Cookbook_Strategies.pdf

[[Cl...]]

Please sign in to view this quote.

I didn't create this program, the datum cylinder is constructed from one scanned circle segment and two 16 point probed circle segments. I have no idea why they chose that method? The filter and outlier setting were not per the cookbook either. This is a "Master Part" used for verification only. One of he probe positions I had a corrupted probe file and had to re-create it. The stylus is new as well. Prior to all that the roundness was in tolerance.

[[Er...]]

That makes a difference.

Filters and outliers rely on numerous points to run the math. That is per the Zeiss Basic manual. With so few points they are a bad idea.

Several thoughts come to mind.

Since it is a new stylus in every sense. How good are the numbers on the stylus? Are the radius and sigma really good? If you look at Stylus System Management - Geometry Are all the stylii radii nearly the same? Quality Are the stylii sigma all really low?

Could be they were instructed to "measure it good" because it was acceptable to the customer. Reducing the number of points can do this as then the CMM doesn't see the bad points. Poisonous practice as the results can be hit or miss. Using filters and outliers is a better way as it can remove "noise" from the measurement and be more repeatable and accurate.

Orientation of the part may be key with so few points being taken.

Is there an "old hand" you trust to ask for history on the part?

I don't know the setup, is it possible it's shanking the probe?

Good luck!

[[Ma...]]

In this case the best way to diagnose the problem might be to plot the element with raw data only -no filter or outlier elimination, and see what the data really looks like. If you are using multiple styli for the measurement does it look like the data from one is different from the other(s)? Is the problem with the scanned section or the single point sections? If you change the location of the part on the CMM a bit does the error stay in the same location relative to the part?

[[Cl...]]

I found the stylus measuring this feature was coated with oil from the previous part.
After cleaning I reverted back to the initial outlier settings and that part checked good.

[[Te...]]

The three points in your example will actually kick out 3 points on each side of the "flyer" point so it kicks out 7 points for each "bad" point, it sounds like a lot but the outlier point is greater than ± 3 sigma (99.73% of the data) you really don't eliminate that much of the data set. You need a minimum of 7 points per wave length or undulation per revolution for filtering to work, that is why point spacing and min points are so important.
Make sure you also use the pre filter settings to help the software filter and eliminate the correct points too.