Can someone please clarify the Rule of 10 to 1, what it is exactly, what I use it for, and so on? An example would help. The book goes on and on about it but doesn't tell me what it is. I searched through the forums but did not see a previous mention of it. :read:

Also, what is the meaning of the ndc number? What range are they looking for to be acceptable? And what is meant by gage discrimination? The book is just too eloquent for me to understand the concept I suppose or I'm too ignorant. :bonk:

The "10:1 rule" is a guideline in metrology. Broadly speaking, your measuring instrument chosen should be accurate (not just discriminate) to 1/10th of the tolerance.

In other words, if you have a feature with a tolerance of 0.010", your measuring instrument should be accurate to no less than 0.001".


Unfortunately, I have no idea what 'ndc' is, out of context. What book are you using? There may be some of us here who have the same one, or by the same author, and can extrapolate the meaning from that information.

Are you referring to the QS 9000 Measurement Systems Assessment book?

I just went through about a month of reading, re reading, searching the Cove, Googling and general agony on this.

I built myself some crib notes during this process. If you are interested, I will be happy to post them. It may take a few days, life is hectic right now, and I need to tidy the notes a bit before posting.

Thats great example on the 10:1 rule. Thankyou.

As for the crib notes, I'd be happy to get what I can. The ndc I'm referring to comes from the MSA manual 3rd edition. It is calculated in an excel spreadsheet after a GRR study is done with 3 appraisers and 10 parts. I've heard mention in some of the other threads that it shows how great your discrimination is by calculating the number of categories you get when performing a GRR. If I could make the connection between categories and what that means with the actual data, I'd be in hog heaven.

Check out the thread Number of Distinct Data Categories - NDC - AIAG's MSA Manual 3rd edition (http://elsmar.com/Forums/showthread.php?t=7343&highlight=ndc+gage)

ndc = number of distinct categories

Tim F

I finally understand the importance of the 10:1 Rule of Thumb. When completing a GR&R test, measuring a tolerance of .001 with a gage accurate to .0001 gives enough discrimination that the UCL will be useful and not cause problems with numbers falling out of control.

If you make the mistake of measuring a tolerance of .001 with a gage accurate to .001, most of your measurements will fall outside your UCL limit since the discrimination is so low. When you go to correct this, your UCL will fall even lower, possibly pushing more measured values outside the new calculated UCL.

I finally understand the importance of the 10:1 Rule of Thumb. When completing a GR&R test, measuring a tolerance of .001 with a gage accurate to .0001 gives enough discrimination that the UCL will be useful and not cause problems with numbers falling out of control.

If you make the mistake of measuring a tolerance of .001 with a gage accurate to .001, most of your measurements will fall outside your UCL limit since the discrimination is so low. When you go to correct this, your UCL will fall even lower, possibly pushing more measured values outside the new calculated UCL.

Looks like you have it.

We ran up against this same thing with our CMM machine. Our first MSA studies showed just awful results.

We were reporting measurements to the same number of siginifcant figures as the tolerance. Pages 43 to 46 of the MSA book clearly show this as a mistake.

Another thing to watch for is to select parts for the study that span 6 sigma (or Tolerance if 6 sigma is not known). We initally chose 10 parts in a row from a production run. There was very little part to part variation.

Once we fixed up these problems, we got decent MSA results.

My big lesson learned...slow down and read and understand before you do a lot of testing...the MSA book is actually pretty good. The Cove fills in the holes.

Are you referring to the QS 9000 Measurement Systems Assessment book? ...... If you are interested, I will be happy to post them.Please do if you get the time. We'll all appreciate it!

I have a question for you for a situation that I've encountered.
I have to check a dimension that has a tolerance of +/- 0.1 mm (let's say 10 +/- 0.1 mm). I use a vernier for checking this dimension (the resolution is two decimal places, so it satisfy the 10 to 1 rule). Checking this dimension, is now 9.86 mm a dimension acceptable or not? If you round this one to one decimal place it will be 9.9 mm. Or, is 10.12 mm acceptable (same idea, rounding it it would be 10.1 mm)?

Thanks for your help.

I have a question for you for a situation that I've encountered.
I have to check a dimension that has a tolerance of +/- 0.1 mm (let's say 10 +/- 0.1 mm). I use a vernier for checking this dimension (the resolution is two decimal places, so it satisfy the 10 to 1 rule). Checking this dimension, is now 9.86 mm a dimension acceptable or not? If you round this one to one decimal place it will be 9.9 mm. Or, is 10.12 mm acceptable (same idea, rounding it it would be 10.1 mm)?

Thanks for your help.

The limit is 0.1mm, so the question you have to ask yourself is "Is 0.14mm greater than 0.1mm?" Even if you round it, the actual value hasn't changed, nor has the fact that the limit has been exceeded.

Daniel
Rounding of results is incorrect. The 9.86mm and 10.12mm are out of tolerance and should be recorded as such. If you are using a digital vernier, the resolution is 0.01mm, but the accuracy is 0.02mm according to the standards and manufacturers specification!
Paul

Sonflowerinwales,

Thank you for your answer, some people tried to shake my believes in my knowledge implying that 9.86 mm can be rounded to 9.9 mm (the drawing specifies only one decimal place for this dimension), and part still be ok.

Regards,

Daniel

Jim,

You are 100% right, this is the way that I know being right too, but some people tried to shake my knowledge.

Regards,
Daniel

The 10:1 rule is - to use an analogy - like the big brother to the 4:1 rule and in the U.S. is "an accepted metrological specification" under ANS/ISO/IEC 17025.

10:1 or better is considered optimal in metrology, where 4:1 is considered the minimum in metrology. The 4:1 is specifically described in the American National Standard ANSI/NCSL Z540-1-1994 Clause 10.2.b which is the word-for-word posting of the long-since-dead-and-buried MIL-STD-45662A.

The current interpretation of the rule is a TUR which means the expanded uncertainty of the calibration performed is the base to work back to the 4:1 from the instrument used to effect the calibration. As an example, if a caliper is calibrated at 0.001 inch, with an expanded uncertainty for that calibration of 600 microinches, then the collective expanded uncertainties of the gage blocks used to calibrate the calipers must be not more than 150 microinches in order to maintain the 4:1 ratio.

Hope this helps.

Hershal

Daniel
Rounding of results is incorrect. The 9.86mm and 10.12mm are out of tolerance and should be recorded as such. If you are using a digital vernier, the resolution is 0.01mm, but the accuracy is 0.02mm according to the standards and manufacturers specification!
Paul

Unfortunately, I'm going to be one of those "shakers". :argue: :)

Mathematically, accuracy in engineering terms is determined by the specified number of digits in the variation specification. Rounding is acceptable if you do not violate the number of significant digits specified or implied.

So, for example, the generic specification "10 +/- 0.1" by itself implies that accuracy is being held to 1/2 of the specified error spec, or 0.05. Therefore, a measurement of 9.86 is an acceptable measurement per the specification.

Alternatively, a specification can be more explicit in two ways. One is to actually write out the significant digits you expect accuracy to, such as "10.0000 +/- 0.1000". The other is to state it: "10 +/- 0.1 measured to an accuracy of 0.0001".

This is probably one of the most confusing, and one of the most abused areas in requirements and specification writing in industry, and really does lead to a lot of problems trying to communicate intentions in requirements. :(

Always report your results to the same number of digits as in the specification and round them before recording, even if you know your equipment carries greater accuracy. The additional digits won't do anyone any good except create the kind of question being asked here.

Mathematically, accuracy in engineering terms is determined by the specified number of digits in the variation specification. Rounding is acceptable if you do not violate the number of significant digits specified or implied.


Sorry, but this is just wrong. While the tolerance may be signified by the number of decimal places used (e.g., two decimal places = +/- .01 or) the need for accuracy is not, unless it's specifically stated. There is no intrinsic difference between .01 and .010, and if the specification says x +/- .01 and the measurement is x.011, the specification has been violated.

If you disagree, can you tell me where the limit on rounding is? In the example given above, is x.01999, would you consider the result acceptable?

If you disagree, can you tell me where the limit on rounding is? In the example given above, is x.01999, would you consider the result acceptable?

Actually, I did specify it. This result is not acceptable because it violates the 1/2 digit accuracy implied in the specification. Spec says +/- 0.01, result is off by 0.01999 which rounds to 0.02, therefore it fails.

Sorry, but this is just wrong. While the tolerance may be signified by the number of decimal places used (e.g., two decimal places = +/- .01 or) the need for accuracy is not, unless it's specifically stated.

Do you have any material reference for this? I'm not trying to be mean, I'm just not aware of any. I've never heard of some numbers being subject to rounding and others not.

The problem is not in the specification, its in the implementation of it. There is no industry standard requirement for what a measurement's precision must be relative to the measurement result itself (thus, the existence of MSA and this forum ;) ). So, if it is not specified, it could 100x more precise, 10x or even just 1x. If you compare a 1x measurement to a 100x measurement, you could easily fail a 100x measurement and pass a 1x measurement. For example, both could report 10.1 as a result, or one could report 10.1 and the other 10.001. Which is more right from a mathematical perspective? I'm purposely avoiding a quality perspective, because of its subjective nature, and wasn't what my original post spoke to.

Also, significant digits drive designs and implementations. If the designs are only good to 3 digits of precision including rounding of all calculations, then a measured test value that violates the 4th digit of precision (or the 7th) is not an impactor on the design unless the requirements were poorly specified.

I'm happy to move this topic off to another thread. I think it's extremely valuable as there are differing views and practices that lead to confusion. :confused:

Hopefully not ruffling feathers too much! :(

Actually, I did specify it. This result is not acceptable because it violates the 1/2 digit accuracy implied in the specification. Spec says +/- 0.01, result is off by 0.01999 which rounds to 0.02, therefore it fails.



Do you have any material reference for this? I'm not trying to be mean, I'm just not aware of any. I've never heard of some numbers being subject to rounding and others not.

The problem is not in the specification, its in the implementation of it. There is no industry standard requirement for what a measurement's precision must be relative to the measurement result itself (thus, the existence of MSA and this forum ;) ). So, if it is not specified, it could 100x more precise, 10x or even just 1x. If you compare a 1x measurement to a 100x measurement, you could easily fail a 100x measurement and pass a 1x measurement. For example, both could report 10.1 as a result, or one could report 10.1 and the other 10.001. Which is more right from a mathematical perspective? I'm purposely avoiding a quality perspective, because of its subjective nature, and wasn't what my original post spoke to.

Also, significant digits drive designs and implementations. If the designs are only good to 3 digits of precision including rounding of all calculations, then a measured test value that violates the 4th digit of precision (or the 7th) is not an impactor on the design unless the requirements were poorly specified.

I'm happy to move this topic off to another thread. I think it's extremely valuable as there are differing views and practices that lead to confusion. :confused:

Hopefully not ruffling feathers too much! :(

No ruffled feathers here, Joe. First, when it comes to engineering standards, there is no universally accepted source. There are published standards, but they are adopted by agreement between parties. Without at least two parties agreeing to abide by a standard, the standard is meaningless.

Next, unless there is some sort of explicit understanding between parties, numbers mean exactly what they say, nothing more and nothing less. .011 is greater than .01, and if a specification says that .01 is the limit, and you measure .011, you've exceeded the limit, unless we have agreed to some other interpretation of boundaries beforehand. If you make $100,000 worth of parts that exceed the stated limit, but assume something about rounding because of some ill-defined "engineering standard," and make parts that measure > the stated limit, and the parts don't work in end use, I hope you're hungry, because you're going to eat those parts :D .

Let me add a few more stray thoughts.

1) It is common in college-level science courses to use "significant digits" to indicate an approximate level of precision. A number like "11" implies a precision of something like +/- 1 or +/- 0.5, while "11.0" would imply something like +/- 0.1 or +/- 0.05.

I have seen professors who seem to think "significant digits" is the last word in error analysis, and expect students to exactly follow the "significant digit" rules for all homework. In fact, it is simply a rule of thumb that works pretty well in many circumstances, but it is quite possible to find situations where the rules work quite poorly.

2) In the absence of any clarification, I would follow Jim's interpretation that +/- 0.1 means that +/- 0.10001 is out of spec, while +/- 0.09999 is in spec. Of course, the best plan is to keep all the parts well away from the limits, so there is no question.

3) Economically, it may not be worth arguing about the parts right near the spec. You could easily spend more money determining whether the part is actually +0.09999 or +0.10001 than the part is worth. It also depends on whose responsibilty it is: the producer to assure that it is in spec or the customer to show that it is out of spec.

This is where the idea of guardbanding comes in. A conscientious producer might reject any parts at +0.10 because he isn't sure they are good. For internal purposes, the test criterion might be tightened to +/- 0.09. A conscientious customer might accept any parts at +0.10 because he isn't sure they are bad. For internal purposes, the test criterion might be loosened to +/- 0.11.

4) Perhaps Taguchi had it right. Rather than an absolute limit for good/bad, a sliding scale might be more appropriate.

At one time, I was trying to figure out an "economic capabilty index". The loss function for bad parts is specificed, and then the "economic capabilty index" is simply the cost of poor quality associated with the parts. If the parts are exactly at the target, the index is 0. The farther off the parts are from ideal, the higher the average loss function. If the average cost of poor quality exceeds some value, then there is a penalty to the producer.

Determining the appropriate loss function would take a bit of thought (but that's what QE's are for) and calculating the results is somewhat involved (but that's what computers are for), but interpreting the results -money! - is simple (I guess that's what managers are for ;) ).


Tim F

I was speaking from a purely mathematical viewpoint. From that perspective, I still believe I am speaking well. It is an academic approach.

Now when it comes to business decisions, based on quality, cost and schedule, then there are many other things to consider, which I agree completely with Tim about.

In fact, at our facility, we implement guardbanding as a standard practice to ensure we have good test margin. We are also beginning to implement an outliers methodology that will move test limits based on a six-sigma passband, that will, for most of these measurements, be well within the design specification limits. So the effect of whether one uses an absolute vs a rounding approach to the passband limit evaluation is rather null on product quality.

I suppose, in the end, we're all attempting to achieve the same goal in much the same way. My greater concern is attempting to be more standardized in approaches and evaluations for those "simpler" things that auditors love to get ticky-tacky about because it's about all they can really understand in the short amount of time they take to review your processes and data.

So, in this particular case, industry would do well to make a common choice: either round always, or round never. Since most academic environments do promote rounding, I believe that makes the better choice from an engineering and business perspective. (Ever have arguments with your customer when they complain that the test report says the measurement is 10.11, the limit is 10.11 with a LL<=x<=UL eval, and it still says fail because the computer is hiding digits? Talk about a waste of time. ;) ) However, round nothing might make more sense in preventing audit issues. Neither one in the end will make a significant difference in overall product quality when present with other elements such as guardbanding and six-sigma SPC.

The important thing is that specifications are based on what will work, and that everyone understands that it's just as bad to reject "good" material as it is to accept "bad."

Rounding should be used when rounding is a useful concept. I don't expect to find fractions of cents reported on my bank statement or pay stub, because I can't actually put my hands on a fraction of a cent. We shouldn't confuse academic mathematical expediency with manufacturing precision. The former might be useful in developing specifications, but once the specs are set,"no greater than x," means that I reserve the right to reject anything greater than x. There has to be an explicitly set limit.

Hi all,

does it mean, if i carry out a GR&R study with a sample that has distinct value, let say 10 distinct samples of data, my ndc will surely be >5 right?

But the argument here, which is posted by one of our external auditor is that, the data should be of homogeneuos sample. thus, the average mean of a sample should not be very big difference. or, i can just perform the R&R study with any heterogeneous data and get a good ndc instead?

second question, can anyone elaborate more on the statement in pg 46 of the AIAG MSA ref manual (3rd edition) which says "A measurement system will have adequate discrimination if it's apparent resolution is small relative to the process variation. Thus a resolution to be at most one-tenth of total process six sigma std dev instead of the traditional rule which is the apparent resolution be at most one-tenth of the tolerence spread."

thanks in advance!!!

Hi,
I know this is an old question, yet I appreciate someone can give me a hint.
1). Assuming there is a measurement with Tolerance of 12 (+/-6) cm. According to 1:10 rule, the readable unit should <= 1.2 cm, so that I can be convinced the resolution is okay.

2). Now assuming my process variation of measurement is happen to be 12cm (UCL-LCL=12cm), how do I be convinced that ndc >=5 is okay? eg. 12 can be devided into at least 5 categories (12/5=2.4cm)

In this case, can I say my measurement resolution is acceptable if ndc is greater than 5? But as u see that 1:10 rule is not fulfilled (2.4 > 1.2 cm).

Appreciate for clear explanation.
Clouds

Hi,
I know this is an old question, yet I appreciate someone can give me a hint.
1). Assuming there is a measurement with Tolerance of 12 (+/-6) cm. According to 1:10 rule, the readable unit should <= 1.2 cm, so that I can be convinced the resolution is okay.

2). Now assuming my process variation of measurement is happen to be 12cm (UCL-LCL=12cm), how do I be convinced that ndc >=5 is okay? eg. 12 can be devided into at least 5 categories (12/5=2.4cm)

In this case, can I say my measurement resolution is acceptable if ndc is greater than 5? But as u see that 1:10 rule is not fulfilled (2.4 > 1.2 cm).

Appreciate for clear explanation.
Clouds

To my understanding you are trying to relate apples with oranges, it does not work that way.

1) the 1:10 is NOT a rule, it is an good engineering guide for initial gauge selection. But if you have a good capable process, 1.67-2.33+, you probably will need 1:20 to distinguish NDC regions.

2) Very wrong, the NDC relates directly to part Variation (process capability) and GR&R and is NOT simply related in the manner shown to a tolerance band and as such your assumptions are ill grounded:nope:.

Sorry to pour water on your assumptions.

I am getting doubts from my employer about the results of Gage R & R and the less than or equal to 10% showing an acceptable gage system.

The current MSA (third edition) has nothing in the Average and Range method for Gage R &R with the Tolerance factored in---to see a 6.4% Gage R&R in the % tolerance but incorrectly determine that 16.8% % of total variation pushes the Study as unacceptable.

Who, here, has proof that the R&R Tolerance factored in is the most effective method? I can complete and ANOVA but have spent over 15 years with the R&R based on part tolerance (process variation), Equipment variation, and appraiser variation.

Am I wrong to reassure my employer about this %Tol: 6.4%?
Incidentally, the EV is 6.4%. From my experience the Study shows a good result.
Send me facts I can relay, PLEASE!
Rick, Illinois

Your question is unclear. Do you mean that the P/T Ratio is 6.4% and the %GRR is 16.8%?

The more capable your process, the greater this discrepancy between the two. What is this gage used for? %GRR is only used for gages used for process control or for statistical studies. If the gage is used strictly for inspection, use the P/T Ratio.

Also, a max %GRR of 10% is usually only required for critical characteristics. Most other characteristics can go to 30%. Is the 10% requirement customer-driven or is it an internal requirement? If it is internal, you should relax it to 30% unless it is critical.

In addition, I recommend that you review some recent threads (http://elsmar.com/Forums/showthread.php?t=20476) in this forum concerning an article written by Donald Wheeler. This article discusses erroneous calculations in the AIAG MSA method coupled with the arbitrary 10% criteria that result in disqualifying acceptable gaging systems.

Here's the study, Miner.
The diameter is critical +/- .0005."
I call the gage, a CMM, acceptable at 6.4% R&R.
How significant is Total Variation %?
Rick, Illinois

The %Tol (P/T Ratio) = 6.4%. This is the measure for a gage used for inspection. As such this CMM is perfectly suitable for inspection.

The %TV (%GRR) = 16.4%. This is the measure for a gage used for process control (SPC) or for statistical studies such as capability studies. At 16.4% the gage should be adequate for non-critical dimensions unless you have a customer-specific requirement of 10% max on all dimensions.

What do you use the CMM for, inspection or SPC?

The CMM is used for inspection (First article, correlation checks, etc.).
The whole need for the Gage R&R is for Operational Qualification purposes on a Class III Medical Device project Validation here.
Can I safety present my customer with the Study @ 6.4 % R&R?
If not, please explain why not.
Thanks again,
Rick Tompkins

If the CMM is used for inspection, the P/T Ratio is the appropriate measure to use. In your report, P/T Ratio = %Tol = 6.4%. Unless your customer specifically states the other measure, I recommend reporting 6.4%.

If the CMM is used for inspection, the P/T Ratio is the appropriate measure to use. In your report, P/T Ratio = %Tol = 6.4%. Unless your customer specifically states the other measure, I recommend reporting 6.4%.

Sir,

If the part dimension is designate as a Critical one that require SPC control, then this CMM shall meet 10% GRR or Total Variation, right?

Sir,

If the part dimension is designate as a Critical one that require SPC control, then this CMM shall meet 10% GRR or Total Variation, right?
That is correct per the AIAG MSA manual.

However, remember that unless your customer dictates compliance to this manual, you are still free to make your own rules. I noticed that you had reviewed the threads with Donald Wheeler's articles, so you know that SPC can still detect process changes at %GRR much greater than 10%. If your choice is not dictated to you, use what works for you.

Using this software (basic Gage Tracker), I toolk the option of using the format of the "long-GM" view instead of the 'long-AIAG' view. All the bottom row of percentages were removed and only the field headers were left, empty.
Then, I tabbed the "long-AIAG" box to show it on the printer, and I submitted the Study with the 6.4%.
I thought of circling the %Tol with the full set of data but did not want to answer questions on the 16%TV.

In today's MSA activities, is the %Tol not favored anymore for Gage R&R?

I wonder how I would present a less than 10% Total variation on a Study with close tolerance and very small variation on a random sample of parts.
Funny, but the same data in an ANOVA study showed no formula results and my only recourse was to print a chart of part-to-part variation by appraiser.
Guess I need an upgrade to this software, huh?

Rick, Illinois
NO LONGER A SHY POSTER, MARC!

NO LONGER A SHY POSTER, MARC! I'm looking at the settings and how the automatic 'promotions' are set. It's been a while (like years) since those were set and this hasn't come up before.

It may be an issue of a 'cron' job where the actual check and change happens at a set interval (probably an hour or a day), not instantaneously.

Using this software (basic Gage Tracker), I toolk the option of using the format of the "long-GM" view instead of the 'long-AIAG' view. All the bottom row of percentages were removed and only the field headers were left, empty.
Then, I tabbed the "long-AIAG" box to show it on the printer, and I submitted the Study with the 6.4%.
I thought of circling the %Tol with the full set of data but did not want to answer questions on the 16%TV.

In today's MSA activities, is the %Tol not favored anymore for Gage R&R?

I wonder how I would present a less than 10% Total variation on a Study with close tolerance and very small variation on a random sample of parts.
Funny, but the same data in an ANOVA study showed no formula results and my only recourse was to print a chart of part-to-part variation by appraiser.
Many customers do not understand MSA beyond the jargon and 10%.

P/T Ratio (%Tol) is valid, but you need to make a rational argument that the CMM is not used for SPC, but for inspection. Then support your choice of metric, by showing that P/T Ratio is intended for this purpose.

The 10:1 rule is - to use an analogy - like the big brother to the 4:1 rule and in the U.S. is "an accepted metrological specification" under ANS/ISO/IEC 17025.

10:1 or better is considered optimal in metrology, where 4:1 is considered the minimum in metrology. The 4:1 is specifically described in the American National Standard ANSI/NCSL Z540-1-1994 Clause 10.2.b which is the word-for-word posting of the long-since-dead-and-buried MIL-STD-45662A.

The current interpretation of the rule is a TUR which means the expanded uncertainty of the calibration performed is the base to work back to the 4:1 from the instrument used to effect the calibration. As an example, if a caliper is calibrated at 0.001 inch, with an expanded uncertainty for that calibration of 600 microinches, then the collective expanded uncertainties of the gage blocks used to calibrate the calipers must be not more than 150 microinches in order to maintain the 4:1 ratio.

Hope this helps.

Hershal

Preface: In my quest towards the 17025 guideline, I am daily learning/advancing my knowledge. So reading through old posts helps me immensely.

I have a question on T.U.R. I understand the Test Accuracy Ratio. The accuracy of the standard should be at least four times greater than the accuracy of the U.U.T.( Unit Under Test).

I understand uncertainty calculation of the standard being used (should be less than 150 microinches). The above example presents 600 microinches as the expanded uncertainty for the U.U.T. 1) How (or are you expected to) calculate an uncertainty for the U.U.T.? It seems to me you are just assessing against a tolerance of 600 microinches.

2) If I am currently maintaining 4 to 1 test accuracy ratios, doesn't that by default, imply that I have less than 4 to 1 test uncertainty ratios?

The parts in the Study checked to .0005" tolerance, with my CMM Z axis locked for a specific point of measurement; this is just to report to you that the conditions were tightly controlled...
Again, the CMM is used for inspection and I have no plans for its SPC use (Ppk and Cpk studies, yes, but no longe-range inprocess actvity, thank you).

If the P/T is 6.4% and the Total Variation is ~16%, I'll present my customer with the "abridged" report showing only the P/T.
The customer is a medical device company and may not want the P/T%.
I'll know soon enough, but if less than 10% is their requirement on Total Variation, another part with a larger tolerance (let's say +/- 0.01 mm) and less variation should make them happy. Perhaps an added .0003" will assist in this case.

Honestly I think the 10% guideline on TV is excessive, especially with three factors involved (equipment, appraiser, and part tolerance). I had been satisfied with this, the P/T, for too many years to share here. Of course, part tolerance has always been key for the gage selection, but it's always brought a Gage R&R % into acceptability.

Any thoughts?
Why isn't the Tolerance factor detailed in the current MSA from AIAG?
Rick, Illinois

Is the 10-1rule documented anywhere by AIAG?

Is the 10-1rule documented anywhere by AIAG?

Coleman,

If you have the AIAG MSA Manual, look at:

Page 5 underneath "Basic equipment."
Page 39 see paragraph 1)
There may be more references, but at least you got something to refer to.

Stijloor.

I guess I should be a little more specific....

We use a universal gage to measure the length of our parts. The universal tollerance for the length of a part is +/- .040". That means that the tollerance of the length gage should be +/- .004". We set our length gages using a transfer standard. My understanding is that the length of the transfer standard should be +/- .0004". Maybe i am wrong - if so please enlighten me! - However if i am not wrong - how do i prove my argument? Preferebly using AIAG manuals!

I guess I should be a little more specific....

We use a universal gage to measure the length of our parts. The universal tollerance for the length of a part is +/- .040". That means that the tollerance of the length gage should be +/- .004". We set our length gages using a transfer standard. My understanding is that the length of the transfer standard should be +/- .0004". Maybe i am wrong - if so please enlighten me! - However if i am not wrong - how do i prove my argument? Preferebly using AIAG manuals!

Coleman,

Oops! :bonk: :bonk:

I guess we'll have to let the metrology and calibration experts chime in.

Stijloor.

Coleman,

Oops! :bonk: :bonk:

I guess we'll have to let the metrology and calibration experts chime in.

Stijloor.

Is there anybody out there? :confused:

Is there anybody out there? :confused:
Please be patient. Some of us have jobs that prevent us from being online in the Cove at all while others are too busy to do more than spot check.

I doubt that you will find anything in the AIAG manuals. The 10% Rule is a guideline from ANSI/ASME STD B89.7.3.1-2001 "GUIDELINES FOR DECISION RULES: CONSIDERING MEASUREMENT UNCERTAINTY IN DETERMINING CONFORMANCE TO SPECIFICATIONS". There is also a 4:1 rule from ANSI Z-540 & MIL-STD 45662A.

There are other Covers with deeper knowledge in this area.

Well I have waited a few days and still nothing... I wasn't trying to be rude or pushy just working on a short time table (as always)... as a result "we" have made a decision that I don't fully agree or disagree with because I could not obtain evedince for or against the argument that I was making.

I requested input from another forum moderator to help answer your question. Your question crosses between the MSA and calibration forum.

Well I have waited a few days and still nothing... I wasn't trying to be rude or pushy just working on a short time table (as always)... as a result "we" have made a decision that I don't fully agree or disagree with because I could not obtain evidence for or against the argument that I was making.

Sorry you did not have the answers that you needed for your issue.:o

I doubt that you will find anything in the AIAG manuals. The 10% Rule is a guideline from ANSI/ASME STD B89.7.3.1-2001 "GUIDELINES FOR DECISION RULES: CONSIDERING MEASUREMENT UNCERTAINTY IN DETERMINING CONFORMANCE TO SPECIFICATIONS". There is also a 4:1 rule from ANSI Z-540 & MIL-STD 45662A.




Now, to Miner's post here. Coleman, is this what you mean? I got somewhat lost in your previous post. Are you talking about accuracy ratios?

If you are concerned with keeping accuracy ratios, then "yes", anything you use to calibrate another device should be significantly more accurate than the device you are verifying. A 10 to 1 ratio is good; like Miner suggested, a 4 to 1 ratio is usually considered minimal.

Exactly what is the argument you were trying to make, and what was unknown to you?

I guess I should be a little more specific....

We use a universal gage to measure the length of our parts. The universal tollerance for the length of a part is +/- .040". That means that the tollerance of the length gage should be +/- .004". We set our length gages using a transfer standard. My understanding is that the length of the transfer standard should be +/- .0004". Maybe i am wrong - if so please enlighten me! - However if i am not wrong - how do i prove my argument? Preferebly using AIAG manuals!

Let's see if I can help tis question, and clarify application of 4:1 and 10:1 rules.....no promises.....

If th tolerance for the part is +/- 0.04", then the question becomes - from a calibration perspective - the accuracy and uncertainty of the measuring instrument, as those are the driving questions. If the accuracy of the measuring instrument is say +/- 0.004" as given in the quote, then the measuring instrument has the ability to accurately measure to 1/10 of the part tolerance. That can be a good thing.

Calibration of the measuring instrument then means that the standard used to calibrate the instrument would have to be ale to accurately measure +/- 0.001" in order to have a 4:1 TAR, or test accuracy ratio.

Cal labs now work with uncertainty as a normal course, rather than accuracy. Without knowing the uncertainty of the calibration of the measuring instrument, numbers would be speculation. However, the same concept regarding calculation of 4:1 or 10:1 applies. The uncertainty of the standard(s) used to calibrate the measuring instrument is to be equal to or less than 1/4 of the uncertainty of the calibration of the measuring instrument.

Hope this helps. I realize it will likely generate more questions.....a good thing!

As mentioned in other posts - uncertainty is something that we are still trying to tackle... As a result I am working within my means.

The confusion arised when there was discussion about the validity of using length transfer standard blocks to calibrate a "length gage". 4:1 is considered the minimum amount of acuracy required to transfer or calibrate a known quantity. It was argued that the length is being transfered so the discrimination rule was not cumulative. i.e. If the gage blocks are 10x more accurate than then the part to be measured they can adequetly transfer the same degree of accuracy over to the actual gage that will be used to check length so that the "length gage" would also maintain the same 10x more precise than the length tollerance of the part to be measured.

The situation becomes a little compounded when you factor in multiple gage transfer blocks. i.e. If my "length gage" has to be set up to check 48" +/- 0.040" than I will need to use multiple blocks to reach my desired quantity because the largest block we have in stock is 12" Now if i have 4- 12" blocks each being held to 12" +/- 0.004 my collective statement becomes 48" +/- 0.016" (If I am wrong here please - someone let me know).

Now based on this situation would 12" +/- 0.004" be adequet when checking my length blocks to satisfy the purpose that they where intended to be used for?

If I took it to the next step and used 12" +/- 0.001" by implimenting the 4:1 rule would this satisfy my requirement when stacking up 4 blocks for a total of 48" +/- 0.004" since I am making a length gage that needs to check 48" +/- 0.004"? Or would my overall stackup need to be +/- 0.001" to satisfy the 4:1 rule.

Now if my stackup tollerance needs to be considered in applying a tollerance for all of my length standards how do I satisfy an auditor that this is being done correctly?

Thankyou for the responses - they are appreciated :thanx:

As mentioned in other posts - uncertainty is something that we are still trying to tackle... As a result I am working within my means.



I too am working with uncertainty. :)Fun, isn't it?


The confusion arised when there was discussion about the validity of using length transfer standard blocks to calibrate a "length gage". 4:1 is considered the minimum amount of acuracy required to transfer or calibrate a known quantity. It was argued that the length is being transfered so the discrimination rule was not cumulative. i.e. If the gage blocks are 10x more accurate than then the part to be measured they can adequetly transfer the same degree of accuracy over to the actual gage that will be used to check length so that the "length gage" would also maintain the same 10x more precise than the length tollerance of the part to be measured.



When you maintain adequate ratios, you are increasing your confidence in making correct decisions. I don't think I would make the statement that you're transferring accuracies.

Let's take a pressure gauge that has a mfg. accuracy of +/-1 PSI. If I am using the 4 to 1 accuracy ratio, then I can check that with a standard that is +/-.25 PSI. I could even use a higher order standard that is +/-.025 or even .0025 PSI. However, the accuracy of the gauge being tested is still +/-1 PSI, due to it's material, characteristics, design, etc. I haven't made the equipment better by using a better standard; I just deliver a more confident measurement.

The situation becomes a little compounded when you factor in multiple gage transfer blocks. i.e. If my "length gage" has to be set up to check 48" +/- 0.040" than I will need to use multiple blocks to reach my desired quantity because the largest block we have in stock is 12" Now if i have 4- 12" blocks each being held to 12" +/- 0.004 my collective statement becomes 48" +/- 0.016" (If I am wrong here please - someone let me know).



True, you do accumulate some error by wringing multiple blocks, hence the reason why people buy large gauge blocks. I think what you seeing is the value of estimating uncertainty of your measurement system.

If you had reported deviation (with reported uncertainty) for each of the gauge blocks, you could use those numbers in your calculations.

Hi Everyone, a very Good Morning to all of YOU!

I was taught like this:

1) 10:1 rule is used to give us confidence that the measurement result indeed meet requirements or not. Example when reading is at the max for one that requires 4.2 +/- 0.02, we could not be sure it is exactly 4.220, or 4.221 to 4.229. Definitely we should not accept 4.22 had we not used 10 to 1 rule.

2) Besides applying this 10:1 rule, we need to take in the accumulated uncertainties throughout the calibration chain. While the reading is at the max of 4.222, we could tell how far it may lie beyound 4.222, and at what confidence level, say 95%.

Please correct me if this is not correct, thank you!:thanks::thanx:

As mentioned in other posts - uncertainty is something that we are still trying to tackle... As a result I am working within my means.

The confusion arised when there was discussion about the validity of using length transfer standard blocks to calibrate a "length gage". 4:1 is considered the minimum amount of acuracy required to transfer or calibrate a known quantity. It was argued that the length is being transfered so the discrimination rule was not cumulative. i.e. If the gage blocks are 10x more accurate than then the part to be measured they can adequetly transfer the same degree of accuracy over to the actual gage that will be used to check length so that the "length gage" would also maintain the same 10x more precise than the length tollerance of the part to be measured.

The situation becomes a little compounded when you factor in multiple gage transfer blocks. i.e. If my "length gage" has to be set up to check 48" +/- 0.040" than I will need to use multiple blocks to reach my desired quantity because the largest block we have in stock is 12" Now if i have 4- 12" blocks each being held to 12" +/- 0.004 my collective statement becomes 48" +/- 0.016" (If I am wrong here please - someone let me know).

Now based on this situation would 12" +/- 0.004" be adequet when checking my length blocks to satisfy the purpose that they where intended to be used for?

If I took it to the next step and used 12" +/- 0.001" by implimenting the 4:1 rule would this satisfy my requirement when stacking up 4 blocks for a total of 48" +/- 0.004" since I am making a length gage that needs to check 48" +/- 0.004"? Or would my overall stackup need to be +/- 0.001" to satisfy the 4:1 rule.

Now if my stackup tollerance needs to be considered in applying a tollerance for all of my length standards how do I satisfy an auditor that this is being done correctly?

Thankyou for the responses - they are appreciated :thanx:

That you are struggling with uncertainty is OK.....so is almost everyone else who makes measurements.....and yes, that includes Metrology professionals like me.....you are NOT alone there.....

As for combining blocks, there is a factor for each combination known as wringing.....there are studies and you can do your own if you have lots of time and nothing better to do.....or take the accepted number of 0.00005" perwringing.....for your example of three 12" blocks, here is how it works.....the three wringings can be combined and so are 0.00015", and this is a rectangular distribution, and so divided by square root of three or 1.732.....this is dropped into the final formula.....

Each block you gave a number for the 12" block of 0.004".....you did not state this as uncertainty, but for discussion let's say it is.....each block has its own uncertainty, hence you have four blocks in this case each with an uncertainty across the block of 0.004", a normal distribution if calibrated by an accredited laboratory and the uncertaint expressed at k=2 to approximate 95% confidence, and so divide by 2 to return to standard uncertainty and drop into the final formula.....

Each block was calibrated at some temp, most likely 20 C, so if used at some other temp, then that must be taken from the CALIBRATION temperature (which should be listed on the cal cert for the blocks), and the difference is a rectangular distribution.....after division then drop the number into the final formula.....

All these are Type B or systemic readings.....

Now, take several readings, get standard deviation, divide by n-1 to obtain Type A or random uncertainty and drop into the final formula.....

Now, the final formula.....

Take all these numbers, square each of them, add them together, and take the square root of the sum.....this is known as root-sum-square or RSS.....it gives you what is known as Standard Uncertainty.....

Then take the Student T-Tables to multiply to achieve 95% confidence.....in theory, one can determine specific degrees of freedom, but in a practical approach the Type B contributions (which have no set number) are considered infinite and so 1.96 will provide 95% confidence.....but the confidence is typically expressed at k=2 which arrives at 95.5% confidence, but is easier to work with.....

Hope this helps......

I am receiving a lot of good input here and it is very appreciated - unfortunalty it also brings up more questions...

I'll start at the begining...
I understand that the standard uncertainty for wringing is 0.00005" and that when you stack up the tollerances you come up with 0.00015". I get a little lost at the next step - How do you determine the rectangular distribution vs. Square or triangle?

Now, take several readings, get standard deviation, divide by n-1 to obtain Type A or random uncertainty and drop into the final formula.....

I don't really understand any of this other than the fact that Type A is the random uncertainty and Type B is the Systematic uncertainty...

Then take the Student T-Tables to multiply to achieve 95% confidence.....in theory, one can determine specific degrees of freedom, but in a practical approach the Type B contributions (which have no set number) are considered infinite and so 1.96 will provide 95% confidence.....but the confidence is typically expressed at k=2 which arrives at 95.5% confidence, but is easier to work with.....

This is also a little confusing for me - If you could break down these steps a little for me that would probably help a lot - or perhaps point me to the correct reference material that explains what is happening here...

As I said I am struggling with uncertainty - mostly because no one here where i work knows what it is and I can't seem to get them to send me to a class - I just have to figure it out all on my own (not a very envious position to be in I know...) along with the help of all of the people here of course.

I must say that in the last 6 months I have been trying to learn and impliment the requirements of ISO 17025 I have learned quite a lot so please don't give up on me...

:thanx:

Could anyone pls explain about variable MSA, i keep doing this using long method but still can't get ndc more than 5. Is it mean my equipment is not suitable?
is tolerence of part may effect the result?

Could anyone pls explain about variable MSA, i keep doing this using long method but still can't get ndc more than 5. Is it mean my equipment is not suitable?
is tolerence of part may effect the result?

There have been a lot of discussions here on the ndc value.
A search in this forum ( maybe best started at the bottom of this page here - similar discussion threads ) will shine a light on this.

ndc can be influenced by several things like :
- Not suitable equipment.
- Incorrect calculation.
- Manipulation of the measuring results.
- MSA samples taken randomly with too little variations between them.
- ......etc.

Please search here and if you come up with more questions I am sure you are welcome here.

Best regards,

Antoine

Hello,

I am now going into MSA and I would like to know if the MSA is to be done on each measuring equipment or to each type of measuring equipment? I have been told that it is to be done on each type of equipment (one equipment of each type), but I don't think it is OK. I believe it should be performed on each measuring equipment. I have read the MSA manual 3rd edition, but this is not specified.

Thank you
Andreea

How do you manage/calculate (TAR) Tolerance Accuracy Ratio for a one sided specification?
Example: Micrometer Accuracy is +/- .0002" The part specification is .035".

I believe that TAR is the Test Accuracy Ratio. See bottom of page one of this link (http://www.transcat.com/PDF/TUR.pdf).

If you mean this in the context of MSA, the proper metric is either P/T Ratio or GRR as %Tol. You do not use the stated gage accuracy, but perform an MSA study. In the event of a one-sided tolerance, these metrics break down. You would be better off creating a gage performance curve or calculating the Probable Measurement Error (PME) as described by Donald Wheeler.

Rule of 10:1 has cleared from above posts for ndc the followin may be felpful to you.

Ndc is the no. of distinct Category, ndc concept comes into the picture from the 3rd addition of AIAG MSA manual before that the same would known as Gague classification Ratio.

It is calculated by dividing the standard deviation for parts by the standard deviation for gage, then multiplies by 1.41
Ndc = 1.41 (PV/GRR)

Now the most important thing what is the significance of the ndc,

This number represents the no. of non-overlapping confidence intervals that will span the range of product variation. You can also think of it as the number of groups within your process data that your measurement system can determine.

The AIAG suggest that when the no. of distinct category is less than 2, the measurement system is of no value for controlling the process, since one part cannot be distinguished from another.
When the ndc is 2, the data can be divided into two groups, say high or low.
When the ndc is 3, the data can be divided into 3 groups, say low, middle and high
A value of 5 or more denotes an acceptable measurement system.

Lets I will try to clear it with an example:
Let the range for your data varies from 0 to 0.20 and the value of ndc is 4, then the meaning of this is :
Your data is classified into 4 categories whose range varies as:

i) From 0 to 0.05
ii) From 0.05 to 0.10
iii) From 0.10 to 0.15
iv) From 0.15 to 0.20

AIAG states the "10:1 rule of thumb" on MSA 3rd edition page 43-44. They state that it can be considered a starting point, but that it "does not include any other element of the measurement system's variability." ndc (number of distinct categories) uses the gage study to generate statistically significant "buckets" or true gage resolution versus the indicated gage resolution or "readability" or "discrimination" (number of graduations, digits, etc. on the gage).

On page 117, they claim that the ndc should be greater than 5. And that may be true for general measurement. But it is woefully inadequate for SPC!

The most key statement is "If the measurement system lacks discrimination (sensitivity or effective resolution), it may not be an appropriate system to identify the process variation or quantify individual part characteristic values. All parts in the same category will have the same value for a measured characteristic."

On page 45 through 46, they indicate the need for good resolution in order to have SPC be effective. They claim 5 categories is sufficient for SPC, which, again, is woefully inadequate. You should have 5 categories on either side of the mean - if you are using X-bar-R charts or similar - to have enough resolution to utilize Western Electric rules (or similar). In fact, on page 46 they indicate "adequate resolution would be for the apparent resolution to be one tenth of the total process six sigma standard deviation." For SPC, I prefer the following:


ndc (for SPC) = ((UCL-LCL)*1.41)/(GRR) >10

This calculation assures 10 statistically significant categories within the control limits.

You really have to be careful with NDC.
An NDC of 5 is roughly equivalent to a %study variable of 30%, which is only marginally acceptable.
An NDC of 14 or 15 starts to give you a %study variable of 10% or lower, which is ideal.

Also the NDC like %study variable can only tell you how good your gage is at measuring the items used in that study. If your selection of parts does not represent the process then NDC can be a fairly meaningless number like %study in those cases.

I dont think NDC was meant to be used as a judgement of is my gage good or bad, it is there to add more understanding. It is equivalent to the %study variable, but is a rounded number.

I realise I am echoing some of what Bob says here, but you dont want to mistake 5 ndc for good and make mistakes because of it.

You really have to be careful with NDC.
An NDC of 5 is roughly equivalent to a %study variable of 30%, which is only marginally acceptable.
An NDC of 14 or 15 starts to give you a %study variable of 10% or lower, which is ideal.

Also the NDC like %study variable can only tell you how good your gage is at measuring the items used in that study. If your selection of parts does not represent the process then NDC can be a fairly meaningless number like %study in those cases.

I don't think NDC was meant to be used as a judgment of is my gage good or bad, it is there to add more understanding. It is equivalent to the %study variable, but is a rounded number.

I realize I am echoing some of what Bob says here, but you don't want to mistake 5 ndc for good and make mistakes because of it.


I agree you do have to be careful. But, it is absolutely worth considering to help clarify if a gage is anywhere near any value for its intended use. The calculation I listed above does extend the value of the concept to deal with the full range of the control limits, whereas the using PV in the original ndc calculation is limited to the samples in the study.

At a minimum, I find it a simple concept to move people away from the 10:1 readability (old school, back yard) to a resolution that considers gage system variability, as the ndc attempts to do. What you really want is statistically significant resolution - any way you can get there.

I would have to think a lot on that one, and afraid I may not be clever enough to come up with such a test :).

We do have a number the NDC, we have a goal > 10, we also have a number of samples, and operators. Hence we have values for degrees of freedom. Some kind of one sided test comes to mind.

I am aware the last comment was not aimed at me directly though :)

I am aware the last comment was not aimed at me directly though :)

True. Those that use Wheeler's approach to gage decision also need to consider that they are achieving 10 statististically discreet categories within the control limits. So, yes...any way you get there, you need to get there! :tg: