NDC (Number of Distinct Categories)

I am confused about NDC (number of distinct categories) and hope someone on here can answer my question. I have a part with a specific width measurement. The tolerance is ±0.10 millimeter. Because this is a plastic part with a definite maximum material condition (MMC), I elected to use a caliper, which easily captured the MMC for all inspectors and returned a gage R&R of 13%.

My customer did not like the NDC of 2 and asked for a different measurement method. So we used a digital micrometer that reads to 0.001 millimeter and performed a new gage R&R study. This time the R&R was 16%, which I could live with, but the NDC is still 2. I am wondering how can this be?

We are using the standard AIAG worksheet and its formulas. I also typed the data into Minitab and the NDC was confirmed. But when I look at the micrometer data, it sure looks like there are a sufficient number of different readings. What am I missing?

Now the customer wants a special fixture built to use a drop indicator for measurement. My gut tells me this is unnecessary and I also cannot guarantee improvement in the NDC based on what I have experienced so far.

I don't understand how the NDC could not improve with the increased resolution of the micrometer and would sure appreciate any insight from your experiences.

Thanks.

 

NDC is proportional to Process Variation divided Gage RR (as a number). With Gage RR being composed of both equipment variation AND appraiser variation. So to increase NDC you need to increase your process variation OR decrease the noise. Switching from the caliper to the micrometer does decrease the noise, but it may not have much of effect if the process variation is small. In all likelihood, you used 10 very similar parts in your study, all very much alike. Spread them out and try again.

 

First, thanks for your replies. I will tell you this is a glass-filled molded part. It simply will not vary much over time. And what I am hearing is that I get penalized in the NDC calculation for having low process variation when using the %tolerance method. So what method should I be using then if my customer is insisting on a higher NDC? (Whether or not we do long term SPC is still undecided, but we do have to submit capability studies at PPAP.)

Below is the actual worksheet I am asking about if this helps you answer.
[Admin Note: The embedded image link is not working/removed/deleted]

 

Thanks, Miner. I actually embedded my data as a picture so not sure why you can't see it. But I think you have helped me understand what is going on. It really is a matter of economics. Basically, I could buy a $300,000 PMM and measure to the millionth of an inch and maybe achieve an NDC of 5 or better, but that won't make the parts any better than what they are right now. It will only satisfy some QE's interpretation of the MSA manual. This kind of lack of practicality frustrates me to no end sometimes. I appreciate your input and your help.

 

Save your startup parts where you are dialing in the process. Save some run off parts when you are tweaking the mold. If you have multiple cavities or multiple molds, use parts off ALL of them. The Gage R&R isn't MOLD specific, it is measurement scheme specific. There is nothing wrong with drawing parts from more than one source. Save some rejected parts. Worst case - do things to the process that aren't harmful to the tooling, but WILL make out of spec or non-centered parts (perhaps shut off the cooling lines?). All this will help your NDC.

One thing you have misstated - it does NOT use the numbers in the %tol calculation. Look at your sheet.

 

One more thing - you clearly have an Excel Gage R&R going on there. If they aren't built correctly, over time, through inexperienced users and copy/paste error, the formulas can break. Here's a copy that I know works. Feel free to sanity check your results against a known, good copy.

 

Thanks, Walker! I tried entering the data into your worksheet and got the same results...exactly the same results. But I like your graphs so I may use you sheet going forward if you don't mind. I should also explain we did a DOE on this part and found we could not vary the size much at all. I think we would have to run very short to impact the size significantly and that could impact the measurement in other ways I think. But next time we start the mold, I will grab some startup shots and see what I find.

 

You are free to use the sheet.

Option 1, strange as it sounds, is to check OTHER parts.

This will very much depend on your customer. They may not understand this, in which case sucks to be you. But ...

The question is: how effective is the digital micrometer on glass filled nylon product? And we have given an R&R study the old college try on one part and seem to be able to demonstrate that this one part is so consistent, the micrometer has a hard time with the R&R.

You could "block" this variable and introduce DIFFERENT parts, that have a similar feature of a different size. Some important assumptions that your SMEs would have to comment on are:
1) The expectation that variability of the feature is the same, even though the parts are different. Or they at least scale. (That's not bad, I used to do glass filled myself a time ago).
2) The feature is reasonably the same on all the parts. And what I mean here is that on part A it's not a long skinny thing and on part be it's a short squatty thing.

An example: Let's assume you are making nails and you want to show a micrometer is a good tool for checking nails. Why would you do all your studying on wire nails (skinny) when you made nails all the way up to 8d nails (pretty stout)? If you passed a Gage R&R on the wire nails, you would be wrong to assume it would ALSO pass on the 8d nails, because they use a different range on the micrometer. That's linearity. Now, if you checked your smallest and largest nails and BOTH passed, you are much safer in assuming everything in between passes as well. The Gage R&R is not JUST a number. It carries with it some assumptions about what the number is trying to describe.

So while not true to form for ONE part, it's not unreasonable to check a family of parts to increase the spread of what you are trying to check. Your customer may accept this.

Option 2: Rework some good ones. Less of a fan of this, because it's then a different process. But you could wallow out some features to get them of a different size.

At the end of the day, you have to remember that it is a repeatability and reproducability study. Not a capability study on a part. You are trying to demonstrate that this is a "good gage" for measuring "this kind of thing." If one guy can produce the same results over and over, regardless of what he is measuring, and then I can turn around and reproduce what he did with the same gage, why would we not think the gage is OK?

And again, you have to consider what you are trying to measure. I can use my micrometer to check various nails, like I said before. And convince myself it is good on outside diameters with a good Gage R&R. So - that's got some reciprocity to checking, say, copper wire. But not to barbells. Why? It's a manual gage and relative to nails and copper wire, barbells are unwieldy to hold and would affect the measurements.

It is strange to me. We tend to mistrust gaging systems like mircometers that are specifically designed for one type of measurement. Then accept very complex things, like CMMs, without investigation. A CMM has way to many factors affecting it's repeatability than a micrometer. Yet ... we just "trust" them.

 

Welcome to the club. Good luck.

 

Very good point.

 

Bevs resources are good. There is also a bit of a collaborative effort to put some more sensible information in a Gage R&R by some of us.

(one of us ONE OF US !!! come drink the kool-aid)

Will make sure it get popped up here when it's done.

 

If you have ever read the NPL publication "CMM Measurement Strategies" or have been a CMM programmer for any length of time, or if you know much about CMM software, you trust CMM results even less. As you noted, there are a LOT of variables that go into a CMM program and then there are the computational algorithm differences from one software package to the next. It can be mind-boggling!

 

Thank you so much for your help and support. I did download the "MSA Tools" but must confess I am not sure which is the Youden plot. This is something I am not familiar with.

 

Even more maddening - the software authors of CMM software will change the way they implement measurements in different versions and NOT TELL YOU. Suddenly, a program that worked fine in release X gives sketchy results in release Y.

I trust a micrometer I have dropped to the floor far more than I trust the "first pass" output of a CMM program. They must be scrutinized.

 

Forgot to add - the Youden plot is what we call an IsoPlot in automotive (trademarked by Shainin, but originally done by Youden). It's an X-Y scatter plot. Reading 1 as X, reading 2 (same part) as Y. You do every combination. 1 vs 2, 1 vs 3, and 2 vs 3. If your gage had total agreement, all these points would lie on a pure 45 deg tie line intersecting the origin. Shainin then eyeballs in an oval capturing the points. And calls the major axis of this oval divided by the minor axis the discrimination ratio. Sometimes he calls it the number of distinct categories. Shainin is OK with 4, but prefers 5.