Hi All,

I have attached an MSA study that I have recently done, and the results are suggesting that improvement of the measurement system is required!

This is measuring the diameter of a steel part, that is highly repeatable and has very little dimensional variance whatsoever. The study is showing no operator variance and a high degree of equipment variance.

I am fairly new to R&R, and have reviewed the results, but I stuck as to how this measurement system could be improved? These parts have very little variance and I think it is this lack of variation that seems to be effecting the results.

Looking at the Xbar chart produced from this data and reading the MSA manual, it would appear that:

A. The measurement system does not have the required effective resolution.

OR

B. The samples do not represent the expected process variation.

The parts are taken from across a wide production stream representing process variation and the measurement system will measure down to .0000000.

Is it because we are rounding the figures up too much?

Can anyone help me out in interpreting this data? Is it the correct method for something that is so repeatable and shows very little variance?

:confused:

You have done a good job of identifying possible reasons for the results. From reviewing your data, I would suspect rounding as the most probable cause.

My recommendation would be to repeat the study recording all available decimal places from the gage. Perform the MSA using this data with all decimal places. If it is acceptable, you know that rounding is the issue. You can then round off one decimal place and recalculate, continuing additional decimal places until the results are no longer acceptable.

Your standard may then be established at the fewest number of decimal places that provide an acceptable MSA.

If this does not work, post the results of the study with all decimal places, and we will re-evaluate.

It turns out that the Gauge used for this attribute only has a measurement resolution of two decimal places and hence why my collegue recorded the values as such. I know this is contrary to what I mentioned above, some confusion about the gauge arose!

The tolerance is 22+0-0.1mm, so I don't see why the gauge needs to have a greater resolution?

Do I just report my findings to the customer with an explanation? What do you do if something is very repeatable and shows little in the way of variation? I am stuck on what to do besides repeating the study with a different gauge, but that is going to cost time and money and if possible I want to avoid that! This gauge has been in use on a very similar part for a very long time.

you have no measurable product variation. You are well within spec for your product. your measurement error std dev is roughly .004...Thsi is very small compared to the tolerance range you gave.
the variation of the error to the product variation is huge - that is why the 'numbers' look bad. you either need more product variation in your study (if it exists) OR simply explain the above. your system is good for this characteristic as specified.

your system is good for this characteristic as specified.

Not necessarily. The gage is marginally acceptable for inspecting product to the tolerance as indicated by a P/T Ratio of 29%. However, it would not be acceptable for use in process control as indicated by the almost 97% GRR.

The issue from a process control point of view is inadequate gage resolution that cannot see the process variation.

Not necessarily. The gage is marginally acceptable for inspecting product to the tolerance as indicated by a P/T Ratio of 29%. However, it would not be acceptable for use in process control as indicated by the almost 97% GRR.

The issue from a process control point of view is inadequate gage resolution that cannot see the process variation.

I must state that this is an inspection on a part that is incoming, so this cannot be used for process control as the things have been made by a supplier. I know perhaps the MSA should be coming from further down the chain, but it is difficult to get this data! However I can see what you are saying, it will not distinguish between parts marginally out and way out!

I do not believe that this will pose a significant risk to the customer but I see if a gauge with a greater resolution i can be implemented.

Thanks for your help.

Since this is used for inspection purposes, a P/T Ratio of 29% is considered marginally acceptable. In other words the gage is acceptable when the expense of developing/purchasing a better gage is not justified given the importance of the characteristic. So, unless this is a Critical or Significant Characteristic, the gage is fine.

Not necessarily. The gage is marginally acceptable for inspecting product to the tolerance as indicated by a P/T Ratio of 29%. However, it would not be acceptable for use in process control as indicated by the almost 97% GRR.

The issue from a process control point of view is inadequate gage resolution that cannot see the process variation.

a good point concerning process control for continuous data.
I also assumed that rounding was being done and they were rounding up.
If rounding down - then all bets are off. measurement data should be rounded up (out of spec is out of spec; plus you will also know (with a high degreeof certainty that there are no measurements beyond what you observe.)

the OP doesn't provide any historical data for us to know if the parts measured are truly representative of the process variation - I also assumed that the OP did ensure he had representative parts. (not always a good assumption, but it is the procedure)

given those assumptions: I would still say the gage is acceptable as is. The difficulty with using this approach to R&R is three fold.
1. 10 pieces doesn't provide a great estimate of the standard deviation.
2. The data as recorded is very 'chunky' (to use Dr. Wheeler's terminology) and therefore provides an over inflated estimate of the standard deviation. (provided that they are rounding up)
3. The P/T ratio is essentially performing division on standard deviations which can't be done.

even given those things, R&R is situational: what is the intent and use of the gage?

If the OP was was truthful that this is a very stable process then one could use this gage for process control - but as an ordinal gage not as a continuous data type gage. you might not see slow trends but you would see shifts - even shifts well within the spec limits.

If using for product acceptance - well this would need some thought. any value that was in the historical range is a good part. any part that is out of the historical range near the limits woudl require special action: guardbanding to reject or guardbanding to detect and measure with a gage with more resolution and repeatability...

it can be done.

It turns out that the Gauge used for this attribute only has a measurement resolution of two decimal places and hence why my collegue recorded the values as such. I know this is contrary to what I mentioned above, some confusion about the gauge arose!

The tolerance is 22+0-0.1mm, so I don't see why the gauge needs to have a greater resolution?

Do I just report my findings to the customer with an explanation? What do you do if something is very repeatable and shows little in the way of variation? I am stuck on what to do besides repeating the study with a different gauge, but that is going to cost time and money and if possible I want to avoid that! This gauge has been in use on a very similar part for a very long time.

Manix,

I ran your data through Minitab.

The bottom line is your number of distinct categories = 1.

Your gage is incapable of distinguishing between parts within 3 standard deviations of your mean.

I suspect the variation you are seeing is instrument noise.

You need a different gage.

I hope this helps.

Jim Shelor
PMP, SSBB

I agree with Miner that your gage can be marginally acceptable based from your application. But if you want to erase or reduce doubts from your customer, maybe you could perform a bias study and/or stability study...if it is practical.