Minitab number of distinct category (ndc) problems

[mopar4u]

We make plastic injection molded parts for the medical field. GR&R's are a daily item in my lab. Our work insturctions call out a very standard 2 person 2 trial 10 part gr&r template on an excel spreadsheet. We recently got a customer that wanted 2 person, 3 trial, 10 part gr&r on all their parts we are making. We didn't have a template for this so we used minitab to do the gr&r's. We struggle passing the NDC.

Background
tolerance of +/- .002"
vision system accuracy of .0001"
vision system resolution .00001" (i think)

we used the entire resolution when entering data into minitab. The vision system should be more than accurate for this gr&r. Problem is we get less than 5 NDC's, sometimes 0 which leads to high gr&r's. If we take this data into a 2 person, 3 trial, 10 part gr&r excel spreadsheet we pass the gr&r with a total of less than 10%. Problem is our process does not have enough variation to make more NDC's. Is it acceptable to trick minitab by taking a couple of the parts from the gr&r and for ex. add .001" to everyone's reading for that part to make it look like we are using more of the tolerance range? If we do do that the gr&r comes down to below 10% with minitab. In a lot of cases it would be impossible for our process to make parts on the extremes of the tolerance because you would either have parts that have shorts (missing material) or have excessive flash (excessive material). I'm sure this would be like falsifying data but I'm not sure how to get around this.

 

[Miner]

Re: Mini Tab number of distinct category problems

Is it acceptable to trick minitab by taking a couple of the parts from the gr&r and for ex. add .001" to everyone's reading for that part to make it look like we are using more of the tolerance range? If we do do that the gr&r comes down to below 10% with minitab. In a lot of cases it would be impossible for our process to make parts on the extremes of the tolerance because you would either have parts that have shorts (missing material) or have excessive flash (excessive material). I'm sure this would be like falsifying data but I'm not sure how to get around this.

This is definitely falsifying data and is absolutely unacceptable.

For what purpose are you using the gage (i.e., inspection or SPC)? If it is inspection only, ndc is irrelevant. Please provide additional information and attach your data. We will then be able to provide more specific help.

 

[mopar4u]

Re: MiniTab number of distinct category problems

We are using the gr&r to establish our measurement method to validate our process,

1. It will be used for capability studies,
2. Then the program will be used for in-process inspection.

I don't have the data available.

How do you turn off ndc in MiniTab?

 

[Miner]

Re: MiniTab number of distinct category problems

We are using the gr&r to establish our measurement method to validate our process,

1. It will be used for capability studies,

The metrics % Study Variation and ndc are used to assess the suitability of the gage for this.

1. Then the program will be used for in-process inspection.

% Tolerance is used to assess the suitability of the gage for this.

I don't have the data available.

This will limit our ability to help. For example, you stated that you obtained a good %GRR with a poor ndc. This should not be possible unless you are confusing the % Tolerance with % GRR (% Study Variation).

How do you turn off ndc in MiniTab?

You cannot turn off ndc.

 

[Allattar]

Re: MiniTab number of distinct category problems

I would go so far as to suggest do not use NDC as a decision number.
NDC and %study variable are based off the same values. %study variable is an exact number.

NDC though is the Stand Dev (parts)/Stand Dev (gage) *SQRT(2)
Then the number is rounded down, always rounded down.

The NDC is only an approximate number. What worries me about NDC when used as a decision is that NDC of 5 is suggested as good.

The NDC of 5 relates roughly to a %Study variable of 30%, which means the gage is only just marginally acceptable.

An NDC of 14 roughly relates to a %study variable of 10%, which is ideal.
The problem is that NDC's have been misinterpreted with 5 as acceptable, when it should be as high as possible.

I have heard it suggested that NDC was never meant to be a decision number, but another way of helping someone visualise how good the gage was.

It would be interesting to see the gage R&R results ran with tolerance, and also with an historical standard deviation added. In addition to what Miner has put above

The %study variable is a number that can only tell you how good the measurement system is at measuring parts in the study. So if the parts in the study do not represent the parts made by the process then the %study variable cannot tell you if you can measure differences between the parts you will typically make. So picking items to test from a much wider range than normal is cheating, and makes the %study too low, cherry picking all good parts does the opposite. Neither tells if you can measure the process.

%Tolerance is used to see if you can measure parts to the tolerance limits. Always a good addition, especially as a back up to check if the parts tested look like they come from a range expected of your process.

By adding an historical standard deviation you get
%Process which judges the variation in measurements in the study to the historical process variation. This can tell you how good your measurement system is at measuring the historical process. Of course if you cannot measure the parts in the first place, how can you trust the historical standard deviation.#

Now the interesting thing you mentioned. The %tolerance is below 10%.
This suggests to me you cannot measure the parts in the study, hence cannot tell the difference between the parts you make. However the %tolerance would suggest you can tell if something you make is out of spec or not.

Now this gives you a difficult question to answer. If the measurement system cannot measure differences between the parts, you should improve the system or get a better gage. However the system can tell you if something goes out of spec. Is the process really that narrow that it is easily found in/out of spec?
If that is true is the cost of moving to an improved measurement system going to give you any more information on what is good or bad? in or out of spec compared to the old system?
But if you dont improve the system, you can never really tell any differences between the parts you measure, and it should imply that if you cannot measure differences between items, how can you tell if you can improve the process? or rely on measurements for control charts or capability? most you can do is see very large jumps in the mean of the process. Then again is that all you need to worry about?