Scanning without filters

[[Am...]]

We have a customer that specifies "No software filters shall be applied". We have been running a part for a couple of years and have had nothing but problems. There are 4 14" - 14.7" diameters, and no clocking feature. We are currently taking 19 hits per level, with 3 of these features being cylinders and 1 being a circle. The part is about 1" tall.

The problem we have is that the parts aren't round, and depending on how it's rotated on the CMM the diameters and/or positions may be in spec or out. All diameters are evaluated using max insc/min circ.

I have recently taken over the programming and was wondering if we should try scanning without filters to at least give our CNC operators more repeatability so they can trust the CMM reports enough to make adjustments off of it. I honestly need to retake the advanced class, as I was sent when I had no Calypso experience (Bc "PCDmis experience is enough to skip basic" 🙄 ), but I was wondering if anyone here has any other suggestions/inputs. I don't have much time to do a lot of studies, so if someone could prevent me from doing one that's going to definitely have bad results I'd love to save that time.

[[Ri...]]

If you are taking single points, filtering won't matter unless you are taking enough to satisfy the requirement.

15 UPR = 105 minimum points on 360°

50 UPR = 350 minimum points on 360°

150 UPR = 1050 minimum points on 360°

500 UPR = 3500 minimum points on 360°

The formula is just UPR * 7.

The issues you are describing is the exact reason why scanning is superior to single points. Unfortunately if you are performing form tolerance evaluations (Roundness, Flatness, etc.) then you are most likely going to want to apply a filter.

[[Er...]]

Adding on Richards reply.

Use the biggest probe radius you can.
Make sure the start and the stop is overlapping so that you can cut them out.
Measure several times with different speeds, so you can find the fastest one that is stable/repeat able. And use that.
Try increasing the measuring force.

That usually works better then one might think. And sometimes it dont..

[[Jo...]]

I would scan the diameter with high density and filters to get an understanding of form. If it is "lobed" (from chucking)you need to take hits that are divisible by 3 . That will ensure your results without filters are LSQ . If it is oblong that is more challenging.

[[To...]]

Please sign in to view this quote.

This is correct but in the cookbook, the recommended point counts actually yields 8 points per upr or cutoff. I suppose they want you to have an extra one.

[[Am...]]

Please sign in to view this quote.

We only report the roundness as a reference, to explain to the operators why they can bring in a part and have a diameter that checks .002" oversized, then bring the same part back in without changing anything and it checks .0004" oversized. If I run the part in the exact same position I get the same result, but if it's rotated (Which it will be without a clocking feature) I will get different results if the part isn't round. The required dimensions are diameters and TP. Diameters are ±.002", and TP tolerances are .003". These parts have been all over the place for years, and I guess we've just shipped with our fingers crossed. I definitely agree that scanning is the way to go, but the original programmer wrote this with points because of the no filter rule from the customer. I'm just trying to figure out if changing to scanning without filters might give me more repeatable results on the CMM so the operators can make their adjustments. Right now, since a slight rotation changes the results, they don't trust it at all, so we end up with a loooot of scrap.

[[Am...]]

Please sign in to view this quote.

Our customer does specify that results should be reported as max insc/min circ, but if using a certain number of hits will give me LSQ even with max insc I'm down to try it. With 19 hits we usually see roundness anywhere between .001" and .004". With diameter tolerances of ±.002" and position of .003" we have a lot of out of spec parts. I'll add changing the hits to 21 to my list of investigative trials.

Of course the most frustrating part of all of this is that the customer has been accepting most of these parts forever, it's just that my operators don't know what they can pass and what they can't, and they really shouldn't have to make calls on "How bad is too bad".

[[Jo...]]

Please sign in to view this quote.

I chased a similar problem where parts were turned on lathe, then flange holes drilled on mill. The mill operators would probe 4 places to find center. The diameter was 6 lobed from being clamped in a 3 jaw chuck. So with 4 probe points, one was always the outlier and caused position to be off. By taking divisor points the hit is always on the same part of undulation and it naturally averages out.

[[Cl...]]

Please sign in to view this quote.

Eric, I'm curious about "Use the biggest probe radius you can" Correct me if I'm wrong, but couldn't that give you incorrect form error, as a larger probe might ride the peaks of a wavy surface?

[[Er...]]

Yes it can. One should read my sentence as in the biggest (within reason).

If you have repetabillity issues as OP describes. Scanning is one good way to fix it. And if you cant filter, why not take some advantage of mechanical filer. And this could grow in to a debate. But depending on what you mean by waviness. That isnt part of form. It should be separated since you are looking for the longer waves. Hence the whole upr thing.

[[Cl...]]

Please sign in to view this quote.

[[Am...]]

Please sign in to view this quote.

We currently use a 1.5mm probe, and unfortunately, clearance prevents me from using anything larger than that. With this part being 1" tall, and having both OD and ID steps, as well as being measured while in direct contact with a fixture and having a customer requirement for cylinder level spacing, I'm too close to other surfaces to use anything larger. I hadn't considered using a disk stylus, but I'm not sure it's worth it since we don't have any in our setup, and I don't have any experience using them.

The parts we machine are glass with a surface finish requirement of 16 Ra or better, so we usually don't have much waviness. If scanning with no filters or outlier removal is a viable option it will probably be my best bet. At least then I know I'm getting the same data for each part, no matter which way it is oriented when it's checked.

[[Am...]]

Im really confused here and I hope someone can clear this up for me, with what is being described how is doing a few single points in any way better then scanning the full diameter and just turning the filters off?

I don't see why it would be worse then what you are currently doing. I don't know what the tolerance is but I'd say a measurement that read .004" measuring the same part twice is not a good measurement already. You might Get some more variance between parts since you are not using LSQ but at least you should be getting a much more repeatable measurement on the same part.

[[Ri...]]

Please sign in to view this quote.

If you are only doing Diameters, scanning with no filters with the LSQ evaluation should be pretty accurate. If you are doing any kind of form measurement or functional evaluation, the lack of filtering will impact the results severely.

[[Am...]]

Please sign in to view this quote.

Isn't that exactly what the original poster is trying to measure? Just seemed like a lot of discussion without anyone actually saying to go ahead and measure the diameters with scanning without filters and I was starting to think I was missing something.

[[Ri...]]

They aren't doing that though.

Please sign in to view this quote.

They are measuring the functional size(s) which is directly impacted by filtering and outlier elimination.

[[Am...]]

Yes, unfortunately our customer requires that Min circ and max insc be used as well as no filters.

So my ultimate question is, am I better off trying to figure out how to change my points to avoid lobing to help make the results repeatable, scanning with no filters even though I can't use LSQ, or do I need to tell management that if we want good, repeatable CMM results we're going to have to ask our customer the "Big Question" regarding why we aren't allowed to use the software filters supplied in the Cookbook by the manufacturer of the machines and software?

Again, at the end of the day, the customer has been accepting these parts with no reported functionality issues, but it's not easy to tell the shop floor why one part that's out of spec is ok and another isn't. And of course some of our better CNC guys are actually trying to make adjustments based on these faulty measurements, which is just causing more scrap.

If I had the time I'd just run as many tests on 10 pieces 3 times each as I could, but I just lost my 1st shift CMM operator so I'm trying to do everything. And of course management wants an answer "right now". No one here really knows what all goes into writing a proper program. As is the case for most of us I'm sure.

Any information/advice is much appreciated.

[[Ri...]]

Please sign in to view this quote.

Yes.

Filtering is a general practice, and there is even an ANSI standard for it: ANSI B89.3.1.

[[Me...]]

If they require Max/Min Inscribed then scanning without filters is not recommended in most cases I would say. Most here in this forum would probably agree that when scanning for form, it is borderline necessary to have filters.

I could be wrong...

[[Ri...]]

Please sign in to view this quote.

Correct. Also, you cannot accurately supply the functional size (max/min) with single points.

The two key points as the advantages of scanning is that it is more accurate, and it is generally more repeatable.

There's a reason why every CMM now has the option to scan.

[[Am...]]

Thank you everyone for the info. I think I may finally have management on board with approaching the customer about allowing the use of recommended filters. I have one more favor to ask.

I understand why using max insc/min circ with individual points is not giving us repeatable results. Again, I need to take the advanced class again now that I understand how programs work and Calypso's interface and can ask more probing questions regarding filters, outliers, etc. Can anyone with the knowledge explain why scanning without filters for max insc/min circ is not a good idea? I'm sure the customer is going to come back with this question, and I honestly can't answer it with confidence just yet.

[[Ri...]]

Please sign in to view this quote.

Here are some screenshots. On the LSQ side, you shouldn't see much of an impact. On the OTE/ITE side of things, you will see an impact as the min/max points will change due to filtering. This applies to form (roundness) tolerances as well.

[[Er...]]

Since your problem is more or less taking the same points, every time. Try to find the same start angle every time.

First. Measure a circle with scanning, and filer it hard. Find the maximum deviating point. Now set your start angle to this angle, in another circle. That you measure with single points with out filer.

Im not sure you have pcm, so this you can just paste in your second circle start angle: (After you changed the name of "Find_Start_Angle" to the filtered circle name)

radArctan(getParameterNamed(getActual("Find_Start_Angle").points.first,getActual("Find_Start_Angle").result.indexOfMaxDev).geometry.x/getParameterNamed(getActual("Find_Start_Angle").points.first,getActual("Find_Start_Angle").result.indexOfMaxDev).geometry.y)

And if you have pcm, this is a more readable solution:

dev=getActual("Find_Start_Angle").result.indexOfMaxDev
xC=getParameterNamed(getActual("Find_Start_Angle").points.first,dev).geometry.x
yC=getParameterNamed(getActual("Find_Start_Angle").points.first,dev).geometry.y
startAngle=radArctan(xC/yC)

Now, I have not tried this, and I'm not sure it will solve anything. Or even be slightly repeatable. But atleast you could try?

[[Ri...]]

And to go into further detail.

With LSQ, it is a weight average of all of the measured points, so if the peaks and valleys are uniform, the average should be the same.

With the Maximum Inscribed, it is using the 3 most inward points to calculate the circle. If the valleys change, the result will be completely different. This applies to the Minimum Circumscribed as well, but inversed.

Filters are meant to take away machine harmonics, and surface finish (high frequency).

[[Ow...]]

Along with the data Richard so gratefully put together, attached is a document that does a decent job explaining how scanning filters work and how the results can change.
I’ve never machined or inspected glass. That said, it is apparent that the manufacturing, material-or- the fixturing process introduces some unwanted variation.
Process variation causes all kinds of problems (I.e.-trust the CMM reports enough to make adjustments off of it) and if you cannot identify where or how you get the variation, you CAN NOT fix it. That is where scanning and collecting more data becomes essential.
Fix the variation and you will have a more repeatable process that operators don’t have to babysit and make decisions they don’t feel good about.
If you can use the CMM to identify the variation in the process, you can improve the process (Cp – CPK) to where it would be more repeatable and maybe not even have to check it anymore
There are a lot things that cause variation, manufacturing machine/tool accuracy, material deformation when material is removed from fixturing (chuck and clamping pressure) that causes stress points that can deform the part when removed.
Your print specifications are limiting how good you can make the part by not allowing you to see exactly what it looks like. I’ve rambled enough but, Imagine how much cost, time, rework, possibly scrap could be saved and morale boosted by improving the process and the product.

Zeiss Calypso Advanced Filter and Outliers.pdf