Gage R&R acceptance impasse

Steve S

Registered
Recently I was at a supplier to review Gage R&Rs. These are typical automotive industry gages with multiple SPC locations per part. The gages will be used for SPC.

The % Study Variation was quite high on all gages. I stated that we needed to review the data and work to understand what needed to be improved.

For my question here, one SPC location we reviewed had 19% Study Variation. I stated that this was marginal, and we needed to improve. Their position was that the % Tolerance was 2% which was “well under the AIAG accepted standard of 10% therefore the gage is quite good”.

They pulled the AIAG 4th edition and even showed me where it states, “conformance or nonconformance to feature specification”. I explained – “yes this is a pass / fail situation”. Now let’s look at the next paragraph where it talks about the gage being used for process monitoring and SPC. This is what this gage will be used for.

I haven’t encountered this conversation with a supplier before. When I left, I felt that they are still convinced that % tolerance is the proper metric by which to accept or reject Gage R&R for SPC gages. It was just me against five or six persons.

I even pulled the data into Minitab and explained the charts and graphs for the example above.

I hope that I am on the right path in my thinking. The terminology in the AIAG manual can be hard to digest and, in this situation, it can be interpreted differently.

Is there a good resource available that could help me explain this in a simple way? Of course, I have found plenty posts here and also YouTube, but I am hoping for something that has the potential to look less like “something this guy found on the internet”.
 

Miner

Forum Moderator
Leader
Admin
The situation that you describe would arise when you have a highly capable process with minimal variation between parts. In this situation, you need to make a business decision rather than a statistical decision. When you have a highly capable process, is it worth the expense to get better gages? Remember, the AIAG manual also says that gages with an R&R between 10 and 30% may be acceptable based on certain criteria such as importance of measurement application, cost of measurement device, cost of rework/repair. I would also argue that the capability of the process would be a mitigating factor.
 

Steve S

Registered
Hi Miner, I agree with you, and I apologize for not going more in depth. I felt I was too wordy to begin with. The example was one of many and it happened to have been the first that I pulled up. The process is not capable yet and they will be working to repeat the study after tuning the parts. My impasse was that they were insistent that % Tolerance is THE metric by which you approve where I was taking the position that my company looks at % Study Variation and NDC. I felt in this case that % Tolerance was improper because the gages will be used for SPC and future tuning of the process. I have between now and after the tuning to come up with a way to better explain to them the situation.
 

Matt's Quality Handle

Involved In Discussions
Hi Miner, I agree with you, and I apologize for not going more in depth. I felt I was too wordy to begin with. The example was one of many and it happened to have been the first that I pulled up. The process is not capable yet and they will be working to repeat the study after tuning the parts. My impasse was that they were insistent that % Tolerance is THE metric by which you approve where I was taking the position that my company looks at % Study Variation and NDC. I felt in this case that % Tolerance was improper because the gages will be used for SPC and future tuning of the process. I have between now and after the tuning to come up with a way to better explain to them the situation.
That is the correct interpretation.

As an SQE, I would usually judge by the % tolerance. I was typically not dealing with special characteristics, so I rarely looked at %sv (which was usually faked by artificially introducing other parts close to the limit).

Also, I would like to congratulate you on being among the few automotive SQEs ever to understand this! :sarcasm:
 

Golfman25

Trusted Information Resource
Hi Miner, I agree with you, and I apologize for not going more in depth. I felt I was too wordy to begin with. The example was one of many and it happened to have been the first that I pulled up. The process is not capable yet and they will be working to repeat the study after tuning the parts. My impasse was that they were insistent that % Tolerance is THE metric by which you approve where I was taking the position that my company looks at % Study Variation and NDC. I felt in this case that % Tolerance was improper because the gages will be used for SPC and future tuning of the process. I have between now and after the tuning to come up with a way to better explain to them the situation.
So is it your thought that they need a different gage or a revised study?
 

Steve S

Registered
Golfman25, the gages are quite highly advanced with a lot of tech (smart gages). Work instructions were not available for review on the gages. I saw different loading techniques demonstrated. To me it seems that the gages are so advanced there was not enough forethought into the need to develop a loading method. There were several items on both sides reviewed and placed on a punch list, but Minitab shows me we need to first develop instructions and get the AV numbers down.

I appreciate everyone's input. I feel like I'm on gage R&R lonely island sometimes.
 

Semoi

Involved In Discussions
On the one hand side I agree that for optimising the machining process it is advisable to use a gage with an uncertainty, which is "much smaller than the uncertainty of the process". Thus, %SV is a good indicator. On the other hand you stated that your goal is to obtain statistical process control. Thus, your goal statistically differs from optimising the process, because SPC does not uses a single measurement/part, but accumulates/extracts the information over several measurements/parts.

The difference between "process optimisation" and "SPC control" becomes obvious, if we consider an highly capable process. Suppose, our process achieves %tolerance = 10^-9. In this context it makes sense to reduce the measurement efforts/costs. Thus, we replace the extremely accurate measurement device by a much cheaper and faster measurement device. Suppose this cheaper device possesses a %SV = 100%. In this example this cheaper measurement device is sufficient -- for the pass/fail decision, as well as for the SPC control.
 

John Predmore

Trusted Information Resource
Is there a good resource available that could help me explain this in a simple way?
I was going to respond that NDC is an intuitive interpretation of GR&R results, without a lot of statistics. But then I see @Steve S referred to NDC in your second posting. Whether you are conversant with the concept or you need a refresher, NDC is a simple explanation for non-experts. Here is Minitab's explanation on NDC. With more NDC, one can better distinguish degrees of goodness and badness. Otherwise, with fewer sorting buckets, good parts are likely to be thrown away with bad, or worse, bad parts can't be discerned from good.
 

Johnnymo62

Haste Makes Waste
I have had to write measurement instructions and train to it when doing GRRs before. It's amazing how many people "know how to" use calipers or micrometers, but don't know how to take a measurement repeatably.
 

Bev D

Heretical Statistician
Leader
Super Moderator
The difficulty with SPC when a process has very little variation is that the data behave like “chunky data”. (See Donald Wheeler’s paper on chunky data: https://www.spcpress.com/pdf/DJW235.pdf and The Problem of Chunky Data). Also google chunky data

Chunky data occurs when the range is so small that there are only a few values available to occur in the range. This can be quantified by NDC. Your SPC charts will likely over alarm. Or under alarm.
 
Top Bottom