N
Re: GRR MSA: Variation, Tolerance, Variance by different methods
A few clarifying statements.
% Study variation is used to demonstrate a gage can be used to discriminate a population of parts. % Tolerance is used to determine if a gage is suitable to check a part against a specification. You don't get to pick which one you like.
Let's say you are drilling holes. And a process engineer says "I can slow down the feed rate and make more consistent holes, which will lower scrap." And the floor manager says "Yeah, but that will add cost because the process will take longer." At this point, you have the makings for the classic statistical DOE probably resulting in a t-test to see if indeed the slower feed rate matters. BUT this hinges on your ability to detect your improvement. Because if the improvement is too small to detect, it's not really an improvement. THIS is when you take parts, do your Gage R&R and look at % Study Variation. Because all you are concerned about is "can you tell the difference between the parts?"
Now, let's assume you are trying to PPAP a new process. You're drilling holes and you are checking them with a gage against some print tolerance. Here, you want to look at % Tolerance and % Tolerance only. IF you are very capable and your GR&R parts are very close together, you will do poorly on % Study Variation, but can still do well on % Tolerance.
Many times an SQ will want BOTH to be good. They don't know what they are asking for. When they want % SV to ALSO be good, what they are really concerned about is linearity. Which is a different thing entirely. If you do this PPAP gage R&R and your samples are only using, say, 10% of your print tolerance, your %SV will likely be low and let's assume your % of Tolerance is ok. Is that good enough? Yes, if you are only concerned with the one output. BUT ... you have only exercised your gage through 10% of your tolerance. Your really don't know how it performs outside of this zone. If you are familiar with the TYPE of gage (everyone uses calipers) your usually safe. But if this is a strange gage using unfamiliar concepts, you are at risk. Look at the fuel gage in your car. It hangs around full or near it forever, then quickly drops and hangs around empty forever. It's a non-linear gage.
You should not be puzzled by the inverse part needs for good Cpk vs good Gage R&R. They are not testing the same things at all. Cpk is an estimate of your process stability. So less variability = good. Gage R&R is an attempt to exercise your measurement system. You want to check it through the largest range you conveniently can. So here, you WANT spread. Notice I said "spread" and not "variability." You DO NOT want "variability". You want to result of each and every part you measure to be the same as the previous measurements. But you DO want spread - you want parts representing the largest portion of your tolerance you can get. You can (and should) have some out of spec parts included in your gage R&R.
The best thing to do is this: when you are starting up your process, you will be all over the place. Set some of these parts aside and hang on to them for your gage R&R. Then when you get the process dialed in, grab the rest of the parts for your gage R&R. Do the R&R, convince yourself your gage works, then do your capability study.
By the way - if the R&R fails, the capability study is meaningless. You could just flip a coin to decide which is good and which is bad, with the same accuracy and far less cost.
A few clarifying statements.
% Study variation is used to demonstrate a gage can be used to discriminate a population of parts. % Tolerance is used to determine if a gage is suitable to check a part against a specification. You don't get to pick which one you like.
Let's say you are drilling holes. And a process engineer says "I can slow down the feed rate and make more consistent holes, which will lower scrap." And the floor manager says "Yeah, but that will add cost because the process will take longer." At this point, you have the makings for the classic statistical DOE probably resulting in a t-test to see if indeed the slower feed rate matters. BUT this hinges on your ability to detect your improvement. Because if the improvement is too small to detect, it's not really an improvement. THIS is when you take parts, do your Gage R&R and look at % Study Variation. Because all you are concerned about is "can you tell the difference between the parts?"
Now, let's assume you are trying to PPAP a new process. You're drilling holes and you are checking them with a gage against some print tolerance. Here, you want to look at % Tolerance and % Tolerance only. IF you are very capable and your GR&R parts are very close together, you will do poorly on % Study Variation, but can still do well on % Tolerance.
Many times an SQ will want BOTH to be good. They don't know what they are asking for. When they want % SV to ALSO be good, what they are really concerned about is linearity. Which is a different thing entirely. If you do this PPAP gage R&R and your samples are only using, say, 10% of your print tolerance, your %SV will likely be low and let's assume your % of Tolerance is ok. Is that good enough? Yes, if you are only concerned with the one output. BUT ... you have only exercised your gage through 10% of your tolerance. Your really don't know how it performs outside of this zone. If you are familiar with the TYPE of gage (everyone uses calipers) your usually safe. But if this is a strange gage using unfamiliar concepts, you are at risk. Look at the fuel gage in your car. It hangs around full or near it forever, then quickly drops and hangs around empty forever. It's a non-linear gage.
You should not be puzzled by the inverse part needs for good Cpk vs good Gage R&R. They are not testing the same things at all. Cpk is an estimate of your process stability. So less variability = good. Gage R&R is an attempt to exercise your measurement system. You want to check it through the largest range you conveniently can. So here, you WANT spread. Notice I said "spread" and not "variability." You DO NOT want "variability". You want to result of each and every part you measure to be the same as the previous measurements. But you DO want spread - you want parts representing the largest portion of your tolerance you can get. You can (and should) have some out of spec parts included in your gage R&R.
The best thing to do is this: when you are starting up your process, you will be all over the place. Set some of these parts aside and hang on to them for your gage R&R. Then when you get the process dialed in, grab the rest of the parts for your gage R&R. Do the R&R, convince yourself your gage works, then do your capability study.
By the way - if the R&R fails, the capability study is meaningless. You could just flip a coin to decide which is good and which is bad, with the same accuracy and far less cost.
Welcome to the world of quality.
