NDC (Number of Distinct Categories) - Long Gage R&R Study in MSA 3rd Edition

We need to see your data and the results of your analysis.

CristiánC said:

YKT:

As far as I can understand, this number is related with the "practical resolution" or the "practical discrimination" of a measurement instrument. The concept is not very difficult to grasp, but trying to explain WHY is calculated in the way that it is calculated, is another thing...

Here is my understanding with regard to NDC index:

Measurement Instruments have physical limitations to "discriminate" between different parts (i.e. give different values of a dimension for parts that are actually different).

The NDC tells you how many categories your measurement system (MS) is able to differentiate considering your process variation.

If your NDC = 1 then your MS is useless.
If the NDC = 2 then your MS can only distinguish between two categories ("small" and "big", for example). (useless too)
If NDC = 3 then you have small, medium and big, and so on...

Automotive industries recommend a value equal or greater than five, if I remember well... The greatest the NDC the better.

Of course the term "category" is not physical or tangible. It is an "abstract category".

Regards.

Click to expand...

I like to describe NDC as the statistically significant resolution, in that just because a device has markings does not mean that its usable resolution is those markings. In my metrology class, I give them a steel rule in .01" increments, and have them do a gage R&R (more fun if the 'parts' they have to measure are marshmallows). You will find that the perceived or statistical resolution is far less than the increments on the rule. NDC captures that problem.

As far as what should be a "good" value for NDC, if you are doing SPC, the value should be at least 10 - calculated using the control limits, not the spec. You want a resolution of 10 increments that you can trust - 5 above and 5 below the mean (or 10 across the region if you are doing non-normal charting). If you had an NDC of 5 calculated to the spec, you would be able to trust less than 2 increments above the mean and 2 increments below the mean - which is useless in my estimation.

NDC is a powerful tool - I hold more credence in it than the gage R&R data itself. An NDC of 5 is just a hair better than a gas gage in quarter tank increments - good enough for quality level TCE, but not much more.

DCB0427 said:

Basically to do this we are using 3 viscosity cups (1 per R&R) and dipping it into the resin tank...3 appraisers and performing this process 10 times.

Click to expand...

Many people do not understand that gage R&R is not a tool - it is a toolbox. You can actually do several types of evaluations with it. You have - perhaps unwittingly - found one of the possible alternatives: 1 sample, 3 gage, 3 operator test. The question it answers is if you have 3 gages, will 3 operators have detectably different results trying to measure the same sample. However, for ndc you need 3 sample, 1 gage, 3 operators. It answers the question will 3 operators using the gage find the difference between 3 samples. To do this you need 3 lots or samples of resin of detectably different viscosities (hopefully representing the range of your viscosity tolerance). You may find using viscosity standards in the tolerance range of your resin may work better. Now, using a cup for viscosity may never get you the ndc you need (you need a really wide tolerance). You may need to invest in a viscometer - and they are not cheap. But, this may be the justification to buy it. Or....open up the spec. It depends on how the spec was arrived at (by guess or by function).

Best of luck!!!

DCB0427 said:

Sorry, I thought I attached it. Just some background info. This data was obtained using a viscosity cup to check resin viscosity. Basically to do this we are using 3 viscosity cups (1 per R&R) and dipping it into the resin tank. Then we use a stopwatch to time how quickly the resin empties out of the cup. The time is what we are recording but we are trying to verify the cup. We are using 3 appraisers and performing this process 10 times. I'm not even sure if this is the correct way to set up our experiment so any help there would be greatly appreciated as well. Keep in mind that I’ve tried manipulating the data so that I have more significant digits but still, minitab only results in 1 NDC. I’m losing my mind.

Click to expand...

It would help to see your raw data to see what you have done.

It appears that you are using three different cups in place of parts. You should perform the MSA on one cup, and use 10 different resin batches for parts. This study does show that there is a significant difference between cups. Sandy is also significantly higher than the other two operators. Do you have a standard method for when to stop the timer, or does Sandy have slower reflexes?

Sandy does have slower reflexes. We performed a study that proved that. Also, I did split the cups up so that the Gage R&R is for each cup individually - however I'm still getting an NDC of only 1. I even placed 'dummy' numbers in with more significant digits just to see if my outcome of NDC would be greater than 1 and it didn't work. As for 10 batches of resin - basically we used the 3 cups and took 10 samples per cup using the 3 appraisers. Now I see that we should have performed one R&R per cup. I've attached the raw data. And thanks to everyone who is helping me out. I appreciate it very much!

Can you format the data in minitab as such. Then just export it as a excel sheet, so we can just paste it in ours.

The NDC numbers is directly related to how well the gage can tell the diffrence in parts, or in your case batches of resin. You most likley will have to do the 10 runs thru each cup and then do it again over at least 10 batches of the resin. You may even increase the number of batches you run the 10 samples from. What you will need to see is a good range, the true process range of the resin batches.

Your data confirms what I said earlier.

1. Pick one cup for your initial MSA.
2. Ensure that you have sufficient process variation by using resin batch as a "part".
3. Make sure that you have a consistent method for starting and stopping the timer such as starting as soon as the top of the cup breaks the surface and stopping when the first break in the stream occurs.
4. Deal with Sally. You either need to use a test that does not depend on reflexes, or establish criteria for reflexes that testers must pass. A third possibility would be to study Sally's lag in reflexes and statistically determine a correction factor to subtract from her data, but that introduces another source of measurement variation.

justncredible said:

You most likely will have to do the 10 runs thru each cup and then do it again over at least 10 batches of the resin. You may even increase the number of batches you run the 10 samples from. What you will need to see is a good range, the true process range of the resin batches.

Click to expand...

I am afraid that may end up being a lot of work for naught. First, it is important to have resin samples at sufficient enough variation to cover the expected tolerance of the material. These samples should have sufficient enough variation between them to be detected. In fact, if one person ran the samples, are they statistically different enough to be detected? If not, there is no chance two more people are going to be able to! If there is a wide enough spread in viscosity between the samples, 5 batches may be enough.

This notion may provide a headache: before doing gage R&R, one needs to understand the calibration. What is the "tolerance" of the cup? +/- 0.1 seconds per centipoise? What does the cup manufaturer state is the accuracy? It gets ugly right off the bat, huh?

More test data on a bad gage will not make it any better. Fact is, viscosity cups are a very rough measure of viscosity. It is a little less accurate than a stick in a gas tank. Unless everything from water to molasses - but not cement - is the spec, then is has a very low probability of success. Viscometers are generally held as the "gage" for viscosity. Even then, if you try to match your data with someone else, you need to run you test at a specified spindle, speed, direction, etc. to get lab to lab repeatability.


Again, good luck!

Yep I agree, I have used ford cups.

If the process has enough variation and the specs are wide enough then maybe out of 30 batches test he can pick 10 batches that show the variation and the cup will be proven valid. Since I do not know the normal process variation I can not say if that would work.

DCB what is the normal process variation?