Brain Teaser for all you readers out there:

If you start a process (variable-continuous) and determine the process is capable. Then do a process capability study with which includes the production gage which shows an acceptable Cp & Cpk (lets say 2.0 - 3.0+).

Pick One:

A. Now I do a gage r&r study.
B I did a gage r&r before I did the process capabilty study
C. Don't run a gage r&r study
E. What the heck is a gage r&r

--> Tell me why!

---------------------------------------------

The correct approach is B.

The measurement variance (s^2) is additive to the actual process variance (s^2). So, your observed process capability as shown by your study includes the measurement variation. Ideally, you should verify that the %GRR (not P/T Ratio) of the gage is low enough to provide an accurate estimate of the true process capability.

I choose option "D". :lol:

I'd delegate answering the question to the lead auditor and resume reading dilbert cartoons.

If, however, the lead auditor chose other than option B.) I would put him on a conference call with Miner and let them debate the matter.

i preferred B

I pick C because I really don't have the resources and time to do gauge R&R studies except for very special cases - in which case I have to do all the legwork and number crunching myself.

If the process is capable, why run an R&R. The capability study includes variation due to gage and operator (if done correctly), and if the process is capable then the measurement system is not contributing enough variation to worry about.

This is an applications viewpoint, not a theoretical one. We need to put our resources where they will be most valuable, not do something because it is the typical way.

If I had a crappy calibration program and poorly trained inspectors I would say B and do an R&R every time a capability study was performed, but since I have a very robust calibration system and an experienced and well-trained crew, I have to go with C with one exception: if I were still incarcerated in the accursed automotive industry and I was PPAP'ing a new part.

I agree with Miner, the answer is B.

The measurement system variation always underestimates the process capability, but in the other hand, one cannot assume that a good process capability implies the measurement system is capable of achieving the target quality objectives.

The analytical approach to the answer is simple. First, one can find the equation that relates P/T as a function of both %R&R and Cp. Second, from the equation, one can easily conclude that a capable process doesn’t always imply that a capable gage was used. A nicely centered process with a small process variation could be measured with a lousy measurement system and one can still achieve acceptable process capability indexes (Cp and Cpk).

A nicely centered process with a small process variation could be measured with a lousy measurement system and one can still achieve acceptable process capability indexes (Cp and Cpk).

This would be true with a device with bad resolution. It would make most parts appear exactly the same - where as a device with better resolution would show more significant variation - maybe not so centered. Take a yardstick, for example... :notme:

Fact is, Gage R&R specifically answers the question: Is this the correct gage for the application? If you have not suitably answered that question before you used the gage for any reason, then you are already starting at quality level TCE. :cool:

hi, all

Initial process study might gives us whatever results we want but actual

pratice we always get cp & cpk lower than anticipated.

so, after process capability study we have to do Gauge R & R

susahnt kulkarni

I am currently going through this where I work. I have to agree that you should do a gage R&R before a capability analysis.

In my situation GRRs and capability studies were performed in the past, but they were done only to meet the requirements of ISO. I feel that they were done with only the goal of meeting this requirment at it minimum.

After conducting a few initial R&R myself, I found out that our measuring instruments, with adaquite resolution(10:1rule), a rigorous calibration system, and properly trained operators, show %GRR above 30%. eg. One measuring instrument was a tenths micometer, I proved the operators are capable of using the intrument by performing the study twice once on our most common material, graphite, and again on a similar shaped part, but machined out of stainless steel. The results are 90%GRR for the graphite and 5% for the stainless steel.

Which factor is influencing my results. I have to believe it is the graphite. I also found that certain features can make a difference in the GRR. I have done more studies and I believe that on graphite the best we can do with a mic is about a 25%GRR in +/-0.005.

How can one perform a capability study, when one cannot rely on the measurement system? A GRR is the only way to know your measuring system (operator, measuring intrument and whats being measured) give trust worthy data.

Micsim:

I don’t know all the details of your study but I am going to provide general information.

One:
You always do Gage R&R first and then you use the measurement system to measure Cp and Cpk of the process.

Mathematically speaking, you could do it in reverse but some assumptions will need to hold. Do it right, Gage R&R is always done first. In my case, I always do (if feasible) full Measurement System Analysis.


Two:
A Gage R&R, provides an estimate of the precision in the measurement system. This means that, if you measure the same part multiple times in a perfect measurement system, you will always get the same value. In practice, measurement systems are far from perfect hence you get different values, which help you to get estimate of the measurement system error (i.e. precision, not accuracy. You take care of accuracy when you calibrate the system). In addition, multiple parts are typically used in order to enhance the confidence level on the estimate of the measurement system error.

This is possible explanation for your situation; this is base on a problem that I experienced while solving a problem for a laser company.
Assuming that your parts are measure at the same temperature (i.e. no change due to thermal expansion or deformation), it is possible that your graphite parts have some kind of surface variation in the region in which the measurement is being made and the operators are simply not able to measure exactly at the same location every time hence your estimate is being influence by the variation on the surface of the graphite part.


Three:
If you measure Cp or Cpk without knowing the Gage R&R you simply don’t know the true value of your process capability. A high Gage R&R will fool you into thinking that you have a lower process capability (i.e. the higher the Gage R&R the lower the lower your "measured" process capability).

This is simple, in the Gage R&R model, the variation you have at the exit of the process is composed of two factors, first, variations coming from the parts (i.e. the process is not perfect hence it creates parts with differences) and second the variation introduced by your measurement system. If you add the variance of these two factors and then take the square root, you will ended up with the "measured" process variation.

If the measurement system has zero variation then it (i.e. your measurement system) will give you the true process capability. In practice, you want to kept the measurement system error small in order to make decisions base on reliable measurements.

Please, let me know if this helps.

Cuadra,

Thanks for your detailed explination. It always helps to hear more about quality systems and GRR. I have a pretty good grasp of the basics. I just need more practice. I was agreeing with answer B) and asking: How anyone would think that they can do a capability study without knowing if their gage system provided trust worthy data. The only way to know that would be to cunduct a GRR. A couple people on the previous page did not seem to think that way.

In my situation we are running with data that no one can and should trust. I am proving that with GRR. After which I will conduct a capabiltiy study and then have to figure out if better gaging is required or just a process change that can work with a wider tolerance range.

Since I work in the automotive industry, I have to go with option B. My auditor would be unhappy with me if I picked any other option.:bigwave:

Miner -

With all of your information, do you have any direct cross correlation numbers on Gage R&R % as it relates to Capability Index that you could share with us?
I know that I have seen graph charts with CPK on the left and R&R value on the bottom, but, has anyone ever come up with hard numbers? IE - Gage R&R (% of Tol.) = 16.2%, what effect does that have on a CPK valued at 2.12 - how much does it drag it down?
This topic intrigues me as I believe a CPK value that is being monitored at 1.47 would be dragged under 1.33 with an R&R value of 18.5%, which according to AIAG would need to be looked at, but, technically would be acceptable.
What are your thoughts on this?

Quote ( Pudge 72 ):
IE - Gage R&R (% of Tol.) = 16.2%, what effect does that have on a CPK valued at 2.12 - how much does it drag it down?
Unquote

I have a table here ( copyrighted ) which gives the following :

Observed Cpk : 2.12 with MSA at 20% would drag it down to 1.90


This must give you an idea.

Best regards,

Antoine

Miner -

With all of your information, do you have any direct cross correlation numbers on Gage R&R % as it relates to Capability Index that you could share with us?
I know that I have seen graph charts with CPK on the left and R&R value on the bottom, but, has anyone ever come up with hard numbers? IE - Gage R&R (% of Tol.) = 16.2%, what effect does that have on a CPK valued at 2.12 - how much does it drag it down?
This topic intrigues me as I believe a CPK value that is being monitored at 1.47 would be dragged under 1.33 with an R&R value of 18.5%, which according to AIAG would need to be looked at, but, technically would be acceptable.
What are your thoughts on this?

I recall seeing the graphs to which you are referring, but cannot recall where, and do not have a copy of it. I think that it was from one of the automotive OEMs.

I attached a spreadsheet with the formulae that you need. Unfortunately, I do not have time to get elaborate. You should be able to use this as a starting point to recreate graphs or tables in whatever format you like. If you add these, please post the file in this thread.

Miner,

Thank you - that is along the lines of exactly what I was looking for!!!!

And, here's hoping everything is going great for you.

I will definately post anything that comes as a result of studies that I use this for. It may be a while - I literally have 5 other projects going on right now, but, your contribution & generosity will not be forgotten........

Miner:

I agree that the standard deviation estimate includes process deviation and also gauge error. If we took out the gauge error, our Pp and Ppk would certainly increase.

Now to the example in your attachment. It just doesn't seem to work out.

If the standard deviation of the process including gauge error is 1.217 and the total standard deviation of from and MSA on the same gauge" is 0.3, do we not have a net estimate of the standard deviation of the process of 1.217 - 0.3 = 0.917 expanding to a process width of 0.917 X 6 = 5.502.

Pp in this case would be 10/5.502 which would end up with a value of 1.818 rather than 1.41 as stated in your excel attachment.

Am I wrong here?

Miner:

I agree that the standard deviation estimate includes process deviation and also gauge error. If we took out the gauge error, our Pp and Ppk would certainly increase.

Now to the example in your attachment. It just doesn't seem to work out.

If the standard deviation of the process including gauge error is 1.217 and the total standard deviation of from and MSA on the same gauge" is 0.3, do we not have a net estimate of the standard deviation of the process of 1.217 - 0.3 = 0.917 expanding to a process width of 0.917 X 6 = 5.502.

Pp in this case would be 10/5.502 which would end up with a value of 1.818 rather than 1.41 as stated in your excel attachment.

Am I wrong here?

Hi Dave.

Standard Deviations cannot be added or subtracted, but the Variances (StDev^2) can be. The formula in cell C12 squares the standard deviations, subtracts, then takes the square root.

Quote ( Pudge 72 ):
IE - Gage R&R (% of Tol.) = 16.2%, what effect does that have on a CPK valued at 2.12 - how much does it drag it down?
Unquote

I have a table here ( copyrighted ) which gives the following :

Observed Cpk : 2.12 with MSA at 20% would drag it down to 1.90


This must give you an idea.

Best regards,

Antoine


this is one of the fallacies traditional gage R&R approaches and their effect.
first the OBSERVED Cpk includes the gage error. So if your observed Cpk = 2.12 the actual Cpk without the gage error is greater than 2.12.

second teh formula for % of tolerance is mathematically incorrect as explained by another poster. you cannot add subtract divide or multiply standard deviations you must work with variances. So even tho we say 20% of the tolerance is consumed by the gage error it really isn't.

and of course we have to add logic and reason that a Cpk of 2.12 (assuming that the assumptions are true) is a pretty good Cpk...so why are we worried about measurement error?

Very interesting answers from all.

In response though to Bev's last post, I have to disagree with not worrying about CPK at 2.12 and not worrying, if that were truly the case, I would not have to issue a Corrective Action if I had limits that were out of control or even a defect within one of the subgroups that comprised that data if all I looked at was the number.

My point for starting this thread though, is to better understand how the Gage R&R affects real world situations. Now, in relationship to and in agreement with Bev, if my Gage R&R is at 22% but, my CPK is at 3.25 - I will never worry about the potential effect that my Gage has on my measurment system, because even in the worst case scenario, measurement error could never incorrectly indicate that my process was out of control.
However, if my CPK was at 1.37 and my Gage R&R value was at 23.2%, I would say that there could be the potential issue of actually being out of control as a result of using a bad measurment system that the customer may have actually initially approved.
Let's be honest - in my expieience, small companies do not place a lot of credence in Gage R&R as an analysis tool. To be quite honest, I am not sure I can even remember the last time somebody's initial reaction to having an out of tolerance condition found by the customer was to check the initial R&R or ask the customer for one. I find however, that Gage R&R is an awesome identifier of variation that we are not even looking for - low hanging fruit.

Very interesting answers from all.

In response though to Bev's last post, I have to disagree with not worrying about CPK at 2.12 and not worrying, if that were truly the case, I would not have to issue a Corrective Action if I had limits that were out of control or even a defect within one of the subgroups that comprised that data if all I looked at was the number.

My point for starting this thread though, is to better understand how the Gage R&R affects real world situations. Now, in relationship to and in agreement with Bev, if my Gage R&R is at 22% but, my CPK is at 3.25 - I will never worry about the potential effect that my Gage has on my measurment system, because even in the worst case scenario, measurement error could never incorrectly indicate that my process was out of control.
However, if my CPK was at 1.37 and my Gage R&R value was at 23.2%, I would say that there could be the potential issue of actually being out of control as a result of using a bad measurment system that the customer may have actually initially approved.
Let's be honest - in my expieience, small companies do not place a lot of credence in Gage R&R as an analysis tool. To be quite honest, I am not sure I can even remember the last time somebody's initial reaction to having an out of tolerance condition found by the customer was to check the initial R&R or ask the customer for one. I find however, that Gage R&R is an awesome identifier of variation that we are not even looking for - low hanging fruit.

The fruit may be easy to reach, but as you found, it can be a little difficult to eat. :D A word or two of clarification: If your process is "out of control" in this context, it means statistical control, which bears no direct relationship to product conformance. Also, if you know that your process is out of control, the Cpk calculation is unreliable anyway, so trying to figure out what affect gage error has on might be premature.

My point for starting this thread though, is to better understand how the Gage R&R affects real world situations. Now, in relationship to and in agreement with Bev, if my Gage R&R is at 22% but, my CPK is at 3.25 - I will never worry about the potential effect that my Gage has on my measurment system, because even in the worst case scenario, measurement error could never incorrectly indicate that my process was out of control.
However, if my CPK was at 1.37 and my Gage R&R value was at 23.2%, I would say that there could be the potential issue of actually being out of control as a result of using a bad measurment system that the customer may have actually initially approved.
Let's be honest - in my expieience, small companies do not place a lot of credence in Gage R&R as an analysis tool. To be quite honest, I am not sure I can even remember the last time somebody's initial reaction to having an out of tolerance condition found by the customer was to check the initial R&R or ask the customer for one. I find however, that Gage R&R is an awesome identifier of variation that we are not even looking for - low hanging fruit.

Plotting a gage performance curve will help you illustrate the effect of measurement error. See this attachment for an example.

The most common areas for measurement variation problems to rear their ugly heads are:
1. Operator measures a part in tolerance and an inspector comes along and measures the part out of tolerance. Inspector , operator and supervisor argue.
2. Operator measures/tests a part at station A and finds it out of tolerance, proceeds to station B, measures/tests it and finds it in tolerance. Operator either ships part, or supervisor tells Quality Manager that the test equipment is junk. Everyone loses confidence in the measurement equipment.

Thanks Miner - again, I am not looking to argue anyone's point about the finite points of whether or how Gage R&R / Variation affects process capability or dimensional conformance. All I am looking for is the information to better understand it so that decisions are properly made instead of just looking at percentages, CPK levels or conformance as individual values.

Wouldn't you guys agree that it would be great to walk up to something, know its CPK level or specification, know what the Gage R&R was for the feature being measured, and be able to intertwine the values right there to get a true picture of how it all affects the part? I think that would be ideal - especially as Miner points out, that, really all of this is done with the end result being an evaluation so as to either prove or disprove confidence in the Gage. I can't talk to a lot of people about Gage R&R percentages and have them truly understand their impact, but, a chart that depicts it, like the one Miner attached is pretty easily understood.
If I am way off here, I wish I worked where you do (that's not sarcasm - it's frustration believe me).

Any templates that produce the type of graph that Miner attached, whereas you could plug in data to get CPK levels, R&R%, tolerances and have it spit out a chart that depicts all of the information so as to see impact would be appreciated. I already used the other one yesterday in an evaluation in production.

I have received several requests for the file that generated the gage performance curve, so I have attached the file.

A word of caution on use. Everything works fine if you use 10 parts x 3 operators x 3 trials. However, if you start reducing the size of the MSA, the graphs must be manually adjusted. I haven't worked out the details to make this automatic. You will also have to manually adjust the x-axis scale on the gage performance curve to get the best appearance.

...how Gage R&R / Variation affects process capability or dimensional conformance. All I am looking for is the information to better understand it so that decisions are properly made instead of just looking at percentages, CPK levels or conformance as individual values.

...I can't talk to a lot of people about Gage R&R percentages and have them truly understand their impact, but, a chart that depicts it, like the one Miner attached is pretty easily understood.


here's a great way to get a quick handle on the effect of R&R.
take soem parts that are at the spec limit - a few just above and a few just below. measure them each multiple times (10 times) and plot the multiple values for each part against the spec limit: what do you see?

THAT is the effect of measurement error on acceptance. it's a start at being able to visualize it. it isn't any harder than that.

Please teach me:

What if the manufacturing process is good that it does not generate out of spec parts? Can Risk Analysis Method and Analytic Method (page 125 to 140) still be used to conduct Attribute Gage Study? These 2 methods require us to obtain parts that are out of spec, within spec and at the spec limit.

Thank You !

I am not able to understand what is the issue here.The MSA guide says that Gage RR should be within 10% process variaton or 10% of total tolerance.And Gage RR is performed once only at APQP stage for each type of measuring equipment.It is not some thing that you have to do again and again.But if there is a chnge in the process parameters or in the job tolerance you have to revalidate the gage/instrument by carrying out Gaga RR again.
V.J.Brahmaiah:nope: