Yes. It has a sound statistical basis, in that you are assuming that the tolerances come from different populations. That is it is unlikely that you will get a +3 sigma part from each component, the 'average' contribution from each part in the cahain can be summed using sqrt(sum of squares)
Note that it is not really root mean square, it RMS*sqrt(N) where you have N parts in the tolerance chain.
You must be a genius. I so the mentioned paper and could not understand what is all about. There are no sound examples so one may wonder what is the tolerance buldup (or chain, loop if you will) is all about. Not to mention that the deference between 2D and 3D tolerance buldups is not even mentioned. Viktor
Viktor,
Thank you for your useful reply. If you have anything pertinent to add to the top thread question please expound, I assume you do. I am open to any and all info that can be shared, that IS the purpose of this forum site.
Respectfully,
-JLarMan
To help you to select the proper method of calculating tolerances, I need to know exactly what seems to be the problem. For example, you have an assemblage consisting of several parts and you need to assign tolerances on each of these parts to meet some requirements. Or, you need to bring all the tolerances to the same group, or…..
JL
As a ref for spc-deming uninitiated, this links about the basic I can find. Although Deming mentioned it he never really incorperated into his system. If I remember right it was hold "Close, High, and Tight, and Everythings right" or something like that.
Viktor,
Thank you for your response. I have a group of plastic parts that interact with eachother via locating pins and screws. The screws are not used to locate, just fasten. The locators are positioned by holes from other parts. When I determine the best tolerance we can hold in our molding processes I apply it to position, locator diameter, etc.. Upon stacking tolerances the max deviation is too much for the system to pass testing. It doesn't make sense to take an 'average' (RMS) method to determine a total or max tolerance for functionality. The RSS method seems to give a more believable answer.
Please advise if I am on track or off kilter.
-Thanks,
JLarMan
Perhaps an example might help. If your process is to assemble two rods into a go/no go gauge, ie you are concerned about the total length of the two rods when assembled end to end, then the RSS method calculates the tolerance stack that will succeed X% of the time, where X is governed by the means and standard deviations of the two rod's lengths. The advantage of this is that on balance you will be able to live with looser tolerances than a strict addition of the tolerances would suggest (the pay-off being that X% of the time the assembly will fail).
I think (fairly confidently) that Viktor is hinting that things are rarely that simple, and that a proper study of the tolerances will not result in such a simple rule. Many dimensions are linked, eg if you have 3 pins on a part and the part shrinks then you have 3 dimensions that move in the same direction, together. As soon as they are related, then the statistical argument collapses completely, and you would be better off doing a monte carlo type simulation, or something.
I don't think that NASA paper says much more than sometimes they do it one way, some times another, and they often adjust things. Cheers
I would agree with Greg Locock that if we have dimensions that are linked then statistics are not the best tool to analyze the situation.
In my opinion, however, the proper way to handle the situation is to constrict a link of dimensions which must be closed. Then, you can calculate the tolerances of different links included using the probability method assuming for the first run that your tolerances have the normal distributions.
can anyone explain why, in equation 2.3, the a and b terms are divided by 3. Also, if this was written in the RMS format, am I right in saying it would be:
Because he says that the system is toleranced at 3 sigma, ie a=3*sigma
He is working out the SD of the gap, not the the 3 sigma limit, in that equation.
After a very brief skim that looks like a better introduction to the idea than the NASA paper. In particular his 4 cautions slightly later on re-iterate the points made in the preceding threads.
roseda,GregLocock and Viktor,
Thank you very much for your input and help. I have investigated all the links that you have refered and they have been very helpful. Viktor, I understand the complexity of this process now and your apprehension in giving out information that was too simplified.
Sincerely,
JLarMan.
