Which kind of method used for the Pin Gauge? What's your opinion?

Dear all,below is my disscustion about the MSA method for the Pin Gauge with my VP of quailty, if you have free time, take a review and share your valuable comments, and what's your answer?? Jason Chen _____________________________________________
From: B
Sent: Saturday, April 21, 2007 5:06 PM
To: A
Subject: RE: GR&R for pin gauge
Dear A,
Thanks for your response, please review my response in red as below. And I think it will be better to discuss with you face to face when you are in China because sometimes my poor English can express myself very clearly.

_____________________________________________
From: A
Sent: Saturday, April 21, 2007 3:43 PM
To:BSubject: RE: GR&R for pin gauge
Hi B,

What do you think of my responses?

BTW, there is another way to analyze GR&R, that is, using the ANOVA method where operator/part interaction is accounted for.
This is Motorola’s requirement.
I don’t teach this because unless one knows ANOVA, it will just confuse everyone.
Sometimes I follow it up after ANOVA in Stage IV.

_____________________________________________
From: B
Sent: Saturday, April 21, 2007 2:45 PM
To: A
Subject: RE: GR&R for pin gauge
Dear A,
Sorry for replying late.
It is a great experience to have this kind of discussion with you.

Selection of GRR Method:
Yes, I agree with you the intent of doing an Attribute GRR is to see if the operator can discern a hole size that is good or bad(good—0.500mm; bad—not 0.500mm), but thru this test we only can get the information that whether different operators will have different results for the same holes. Whereas what we want to get in the measurement is the “actual” size of the hole but no the “go & no-go”, if we use this method, we can’t know whether measurement system is competent to judge our product is within customer’s tolerance.

A**:
If you use the pin gauge to see if the hole meets customer spec requirement, then you will run into a few issues:
the first being poor discrimination as the pin gauge will not have the resolution to discern hole-to-hole variation, 2nd calculating Cp/Cpk or Pp/Ppk will not be accurate, 3rd, you still have not answer the repeatability question – how do you interpret this statistic, and lastly, if you try to setup a control chart to monitor hole size, you will end up with over-controlling the process.**

B: But here we don’t use control chart to monitor hole size, we only need to judge whether the hole size is in customer spec tolerance. If we want to setup a control chart, I am totally agree to use the Smartscope to take the data, then the Xbar chart will be good, then it can discern hole to hole variation, and Cp/Cpk will be accurate.
And to the repeatability, I don’t know whether I misunderstand your question, let me try to explain. The repeatability is 5.8%, so the gauge only contribute 5.8% variation to the measurement result, right???

I have a concept here that when we decide what analysis method used for the measurement system, we need to know what “DATA” we get, then per the nature of the “DATA” to decide which kind of method to use, right?
For example, the data we get for the hole size is 0.500, 0.525mm etc, the data is variable, so use the Long Method GRR; another example about pin gauge, when we use it to measure the board warpage in FQC, FQC calculate the max allowed warpage size, of which they use the pin gauge to gate the board, here the data FQC get is “go & no-go”, it is attribute, then we use the Attribute GRR.

A**: again, the problem with using the variable GR&R is the repeatability question…how do you measure the gauge repeatability; it has no meaning.
For warpage, I totally agree that this should be a variable GR&R using the long method.**

B: For warpage, why to use the variable? I am confusing on this, I think it is an attribute.

Explanation of average chart:
When I study the TS16949 MSA manual myself, I found the GRR methods are different between them. In MSA, GRR%=VariationMS /Total variation(including Parts variation & MS variation); for Multek, GRR%=VariatiionMS /Product Tolerance, it is ok for this because we want to know whether our MS is competent to give a “pass” to the passed product/a “reject” to the failed product comparing customer’s tolerance. And the most of important is that the control charts used are the same, so we can use the conclusions on chart trend in MSA manual. To the trend of Xbar chart, there is a rule that “Since the group of parts used in the study represents the process variation, approximately one half or more of the averages should fall outside the control limits. If the data show this pattern, then the measurement system should be adequate to detect part-to-part variation and the measurement system can provide useful information for analyzing and controlling the process.”
(P102 , MSA manual 3rd revision) A**:
you are right…more than 50% of the points should be outside of the control limits on the Xbar chart…I can tell you even before you run the test that very few, if not all of the points will be within control limits (poor discrimination) and that the GR&R will be very, very poor if you use the MSA definition of GR&R because part-of-part variability will be very small because of poor discrimination B: To this, I am not very clear. So per this conclusion, if the variable we measure isn’t for SPC control, and the GRR<20%, then we can IGNORE the out of control, yeah, it did have the discrimination problem, but actually, it didn’t have enough discrimination to detect part-to-part variation, it still have enough discrimination to detect the passed/failed product (because GRR<20%). A**:
this is not correct…you just answer your own question about discrimination…and now you are saying passed/failed which is an attribute…you can not mix the two together. B: sorry, a little misunderstanding on the passed/failed definition… passed means hole size within tolerance, e.g. result 0.525mm, spec tolerance: 0.500mm+/-20%, means the hole can fulfilled the customer’s requirement = passed; vice versa. Maybe my poor English cause the confusion. Summing up above, to use long method GRR for Pin Gauge, we have 98% confidence to say that we won’t commit the Type I & Type II error, but if we use Attribute GRR, then I don’t have the confidence. A**:
how did you come up with 98% confidence?
In a GR&R, there is no alpha or beta error...there is no hypothesis and you are not performing statistical test where does the alpha and beta error come from? B: oh, right, maybe I need to review the Stage IV again; it is a long time not to study it. Thank you and waiting for your guiding!!!
_____________________________________________
From:A
Sent: Thursday, April 19, 2007 7:06 PM
To:BSubject: RE: GR&R for pin gauge
Hi Jason,

I enjoy this type of question as I learnt at the same time so don’t stop…the more you ask, the more we learn together.

OK…again, how do we interpret the repeatability number…what does this mean?

Even if the reproducibility number is less than 10, I have a lot of concerns because of what the graph is telling us.
You got points outside of the upper control limit, the interaction plot is very concerning, and the Xbar chart is telling us that we have discrimination problem.

Again, in my opinion, we should just do an attribute GR&R.
The intent is really to see if the operator can discern a hole size that is good or bad.

_____________________________________________
From: B
Sent: Thursday, April 19, 2007 6:24 PM
To: A
Subject: RE: GR&R for pin gauge
Mr. Poon,
Yeah, there did be many zero ranges, but if the GRR value is <20%, that means it is suitable to judge whether the product is fulfill the requirements, and the measurement system is still can be accepted, am I right?

During the mass production we cannot use the Smartscope to measure all holes considering the efficiency & cost, so I think the customer will accept this although the control charts are abnormal.

I attached one actual example for reference. Thanks for your tutoring and sorry for spending your valuable time.

<< File: HoleDiameter.xls >>

_____________________________________________
From:A
Sent: Thursday, April 19, 2007 5:37 PM
To: B
Subject: RE: GR&R for pin gauge
B,

They are not discrete…they are continuous.

So what I think you are saying is for the operator to find the right pin gauge to check the hole size.

Let see, the tolerance for the study will be 0.1mm and the pin gauge has 0.005mm increment.
You will probably end up with a lot of zero ranges and the operator has to be pretty bad to have a GR&R greater than 30%.
The reproducibility is probably going to be very close to zero.
Now, how would you interpret the repeatability number?
You see my point.

A true variable GR&R is to use a Smartscope or a microscope to measure the hole size.

_____________________________________________
From: B
Sent: Thursday, April 19, 2007 5:22 PM
To: A
Subject: RE: GR&R for pin gauge
Mr. Poon,
What data we want to get from the measurement is discrete variables, e.g. 0.475mm, 0.500mm, 0.525mm etc. but not the go & no-go(Attribute), then compare it with hole size spec 0.500mm+/-10% whether it is fulfill with the customer’s requirement, so here I think to use the Long method GRR. Am I wrong???

_____________________________________________
From: A
Sent: Thursday, April 19, 2007 4:54 PM
To:BSubject: RE: GR&R for pin gauge
Hi B,

I think the confusion is because we need to clearly define what it is that we want to accomplish.


Your diagram below is not a GR&R…what you are trying to do is to determine the hole size by finding the right gauge that will fit into the hole.
Even in this situation, it is still go/no-go.
All you are doing is to find the right gauge that will give you a go.
It is still not variable data.

In an attribute GR&R, you want to know if the operator would say yes or no given a particular pin gauge and hole size.
So the study may involved the same size hole from different panel and asking the operator to use the same pin gauge to say yes or no.
half of the panels will have the right hole size and the other half will have either diameters 0.5 mil large or 0.5mil lower or whatever hole size you wish to use.

As for warpage, if you are using the pin gauge to determine if the warpage is acceptable or not, then it is still attribute data.
If you want variable data, you need to use a dial guage to measure the distance.


_____________________________________________
From: B(Me)
Sent: Thursday, April 19, 2007 4:36 PM
To: A
Subject: RE: GR&R for pin gauge
Dear A,
To the GRR method for pin gauge, at first I have the same idea with MinFung to use the Attribute GRR, but after I investigate the actual measure method of hole size, I changed my mind. Below is my investigation:See attachment
So I think it is no a go & no-go operation.
And to the board warpage, it does is a go & no-go operation.

Above is my personal opinion, please help to tutor, thanks!!!

______________________________________________ From:
ASent:
Thursday, April 19, 2007 3:41 PMTo:
CSubject:
RE: GR&R for pin gauge You are right…attribute GR&R is the way to go since it is either go or no-go. However, you have to make sure you set it up correctly…half of them should be good and half of the them should be no-good. _____________________________________________
From: C
Sent: Thursday, April 19, 2007 3:16 PM
To:A
Subject: GR&R for pin gauge A I have a question on the GR&R for pin gauge.
Measurements from the pin gauge is actually not continuous.
So, if we are to perform the gage R&R, are we suppose to do this using the long method GRR or the attribtute GRR.
My personal opinion is that we should be doing an attribute GRR. Please advise. Thanks.
C