Inspection - On The Border
Date: Mon, 13 Sep 1999 16:36:25 +0100
From: "Theodore D. Doiron" nist.gov
Subject: inspection criteria
Hi,
I sent this reply to another listserve, but it seems that it would be appropriate for this group also. the basic question was:
>>I have a question I have wondered ABOUT for a long time. What
>>limits of inspection accuracy are appropriate? For example, if a
>>dimension is toleranced as .500 +/- .001, at what point do we reject
>
>>the part? .5011?, .50101?,.501001? (You get the point!) My
>>assumption would be to use a 10:1 rule, and measure the part to
>>.0001 accuracy. What do you think?
>_______________________________________________________________________
You have hit on one of the grand questions of inspection. As with most grand questions there is no one answer. There is a standard, ISO 14253-1 which attempts to set a world default. This standard says,
The manufacturer must subtract the expanded measurement uncertainty (k=2, or about 95% confidence level) from the tolerance to prove it is good. The buyer must add the expanded measurement uncertainty (k=2) to the tolerance to prove the part is bad.
There is, of course, a rather important band of + or - the uncertainty about the tolerance where the part is not clearly good or bad. What to do with these is still a problem.
There are, of course, lots of other choices.
In many standards there is an arbitrary uncertainty allowance added to the tolerance, usually not the laboratory uncertainty but some arbitrary number. The Gage Block Standard (GGG-G-15C and ASME B89.1.9 are examples).
The most common use in the US seems to be if the uncertainty (k=something, usually 2) is less than some fraction of the tolerance, commonly 25% (10% traditionally, 25% in Z540, 33% in ISO 10012, ...) then the tolerance is used as stated. This corresponds to the ISO 14253 standard used with k=0. I admit that using the expanded uncertainty with k=0 doesn't make sense at first (at least it didn't to me when Ralph Veale explained to me how k=0 would be a good idea), but in the language of ISO 14253 it is actually a sensible description of current US practice.
What we are talking about here are guardbands. ISO 14253-1 is actually a disguised guardband standard. Guardbands are safety factors subtracted (or added) to the tolerance depending on the relative importance of the product failing and costing money to make right. If you are making a part that, if it were to fail, would cost lots of money you would subtract a large guardband from the tolerance to make very sure the parts going out were good. If a part failure is not very consequential, then a very small or even no, guardband would be used. It depends on money; scrap costs for good parts failed and repair costs ( and/or legal actions) for bad parts passed.
The ISO 14253-1 scheme is quite peculiar because it assigns different guardbands to different measurements, depending on who is doing the measurement. It thus is not aimed at maximizing anyone's profit, but at avoiding lawyers. The standard is designed to prevent the buyer and seller from ever disagreeing on a part being in tolerance by putting all of the parts where there is even a very small possibility of disagreement into sort of a part Limbo. There is a cost involved in this, that many people think is inappropriate for all, or even most, cases.
The EA is a combination. It says to validate something to a specification the measurement uncertainty must be a small (unstated) fraction of the tolerance, and you must subtract the (k=2) expanded uncertainty from the tolerance. At least the rule is the same for everybody. For the tests that fall into Instrument Limbo (within tolerance but outside the guardbanded accepted zone) the report only gives the results and uncertainty, leaving it to the customer to do as he wishes.
The final answer is that all of the standards are voluntary standards that are assumed to hold if you do not state what rule you want. It is sort of a default. You can put any rule you want to in a contract for parts. I think it is a good idea to set your own because it makes you think about the consequences of the acceptance rules, which makes you think about how the tolerance you set on the part really affects its function, and finally makes you think about all of this in terms of your profit.
My guess is in factories that subtract the uncertainty from the tolerance to get an acceptance zone, tolerances are smaller than those in companies that don't subtract the uncertainty. Those who add uncertainty to tolerances to get the acceptance zone will likely have very small tolerances. Remember, for the parts to work they must be a certain size, thus the acceptance bands for the part in various factories are probably very close. The tolerances are chosen with the measurement uncertainty to give the right acceptance band for whatever acceptance band formula is used. It might be nice if everybody used the same formula, but to force everybody to use the same formula will cause disruptions as tolerances are redeveloped.
It would be interesting to get a wide view of what tolerence testing rules were really used in industry. Maybe you could send out a questionaire and find out. With enough information about industry practice a reasonable default could be put in Z540 or a US equivalent to ISO 14253-1 or EA-04-01.
I hope this isn't too confusing, but when the only standard on using uncertainty for testing tolerances (ISO 14253-1) has a method which is used by virtually nobody in the world, confusion is to be expected. Let me know what you think.
bye
Ted Doiron
National Institute of Standards and Technology
Precision Engineering Division
Metrology Bldg., Rm. B113
Gaithersburg, MD 20899-8211
U.S. Department of Commerce Technology Administration
______________________________________
Todds Two Political Principles:
1. No matter what they’re telling you, they’re not telling you the whole
truth.
2. No matter what they’re talking about, they’re talking about money.
Date: Mon, 13 Sep 1999 16:36:25 +0100
From: "Theodore D. Doiron" nist.gov
Subject: inspection criteria
Hi,
I sent this reply to another listserve, but it seems that it would be appropriate for this group also. the basic question was:
>>I have a question I have wondered ABOUT for a long time. What
>>limits of inspection accuracy are appropriate? For example, if a
>>dimension is toleranced as .500 +/- .001, at what point do we reject
>
>>the part? .5011?, .50101?,.501001? (You get the point!) My
>>assumption would be to use a 10:1 rule, and measure the part to
>>.0001 accuracy. What do you think?
>_______________________________________________________________________
You have hit on one of the grand questions of inspection. As with most grand questions there is no one answer. There is a standard, ISO 14253-1 which attempts to set a world default. This standard says,
The manufacturer must subtract the expanded measurement uncertainty (k=2, or about 95% confidence level) from the tolerance to prove it is good. The buyer must add the expanded measurement uncertainty (k=2) to the tolerance to prove the part is bad.
There is, of course, a rather important band of + or - the uncertainty about the tolerance where the part is not clearly good or bad. What to do with these is still a problem.
There are, of course, lots of other choices.
In many standards there is an arbitrary uncertainty allowance added to the tolerance, usually not the laboratory uncertainty but some arbitrary number. The Gage Block Standard (GGG-G-15C and ASME B89.1.9 are examples).
The most common use in the US seems to be if the uncertainty (k=something, usually 2) is less than some fraction of the tolerance, commonly 25% (10% traditionally, 25% in Z540, 33% in ISO 10012, ...) then the tolerance is used as stated. This corresponds to the ISO 14253 standard used with k=0. I admit that using the expanded uncertainty with k=0 doesn't make sense at first (at least it didn't to me when Ralph Veale explained to me how k=0 would be a good idea), but in the language of ISO 14253 it is actually a sensible description of current US practice.
What we are talking about here are guardbands. ISO 14253-1 is actually a disguised guardband standard. Guardbands are safety factors subtracted (or added) to the tolerance depending on the relative importance of the product failing and costing money to make right. If you are making a part that, if it were to fail, would cost lots of money you would subtract a large guardband from the tolerance to make very sure the parts going out were good. If a part failure is not very consequential, then a very small or even no, guardband would be used. It depends on money; scrap costs for good parts failed and repair costs ( and/or legal actions) for bad parts passed.
The ISO 14253-1 scheme is quite peculiar because it assigns different guardbands to different measurements, depending on who is doing the measurement. It thus is not aimed at maximizing anyone's profit, but at avoiding lawyers. The standard is designed to prevent the buyer and seller from ever disagreeing on a part being in tolerance by putting all of the parts where there is even a very small possibility of disagreement into sort of a part Limbo. There is a cost involved in this, that many people think is inappropriate for all, or even most, cases.
The EA is a combination. It says to validate something to a specification the measurement uncertainty must be a small (unstated) fraction of the tolerance, and you must subtract the (k=2) expanded uncertainty from the tolerance. At least the rule is the same for everybody. For the tests that fall into Instrument Limbo (within tolerance but outside the guardbanded accepted zone) the report only gives the results and uncertainty, leaving it to the customer to do as he wishes.
The final answer is that all of the standards are voluntary standards that are assumed to hold if you do not state what rule you want. It is sort of a default. You can put any rule you want to in a contract for parts. I think it is a good idea to set your own because it makes you think about the consequences of the acceptance rules, which makes you think about how the tolerance you set on the part really affects its function, and finally makes you think about all of this in terms of your profit.
My guess is in factories that subtract the uncertainty from the tolerance to get an acceptance zone, tolerances are smaller than those in companies that don't subtract the uncertainty. Those who add uncertainty to tolerances to get the acceptance zone will likely have very small tolerances. Remember, for the parts to work they must be a certain size, thus the acceptance bands for the part in various factories are probably very close. The tolerances are chosen with the measurement uncertainty to give the right acceptance band for whatever acceptance band formula is used. It might be nice if everybody used the same formula, but to force everybody to use the same formula will cause disruptions as tolerances are redeveloped.
It would be interesting to get a wide view of what tolerence testing rules were really used in industry. Maybe you could send out a questionaire and find out. With enough information about industry practice a reasonable default could be put in Z540 or a US equivalent to ISO 14253-1 or EA-04-01.
I hope this isn't too confusing, but when the only standard on using uncertainty for testing tolerances (ISO 14253-1) has a method which is used by virtually nobody in the world, confusion is to be expected. Let me know what you think.
bye
Ted Doiron
National Institute of Standards and Technology
Precision Engineering Division
Metrology Bldg., Rm. B113
Gaithersburg, MD 20899-8211
U.S. Department of Commerce Technology Administration
______________________________________
Todds Two Political Principles:
1. No matter what they’re telling you, they’re not telling you the whole
truth.
2. No matter what they’re talking about, they’re talking about money.