I am working on a manufacturing process and wondering when it comes to calculating tolerance stack up through the process during the turning stages what is the best way to go about it? I have read up on the Worst case method and RSS method and seams like just doing worst case is the way to go and also the easiest calculations. Unless RSS method can give you better results but from what I have read that is more for the designing and calculating it statistically but worst case would be better in a manufacturing plan.
Eng-Tips is the largest engineering community on the Internet
Intelligent Work Forums for Engineering Professionals
-
Congratulations waross on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!
Tolerance Stack up 1
- Thread starter smwdrum
- Start date
Arithmetic summation of worst-case is easiest to do but many times will produce excessive, unrealistic results. Engage the Quality Engineer in your operation and they should know about tolerance stack-ups using distributions.
Many years ago I was in a contentious debate with ouridiot in charge Boss. We needed to insert an assembly into a mold, but arithmetic stack-up analysis said it wouldn't go. An investigation into Juran's Quality book indicated that **_ IF _** all of the processes controlling the dimensions were statistically in control, then the stack-up analysis could be performed with a "statistical stack-up analysis" which was square root of the sum of the squares calculation of the tolerances. This would have worked for us. Unfortunately few of our processes were in statistical control at the time.
TygerDawg
Blue Technik LLC
Virtuoso Robotics Engineering
Many years ago I was in a contentious debate with our
TygerDawg
Blue Technik LLC
Virtuoso Robotics Engineering
Neither one. Base your calculations on the previous manufacturing process results. They will likely be normal distributions; check the statistics books for how to determine the expected distribution caused by the summation of variables drawn from other known distributions.
It will basically be an RSS calculation because the normal distribution is centered on zero and the values are squared to convert the negative values into positive ones for addition, then square-root to finish the process.
The best result is the one that best predicts the actual result; summation of distributions is that method.
AFAIK it's rare for manufacturing to do any analysis. I'm sure it must be done sometime, but most places seem to shift an item to a machine/process they think will work and if it does, they are good. If not, they ask engineering to relax the tolerance (for cost/schedule/whatever reasons)
Keep at it. Done right, you should be able to save some money and even 'give back' some tolerance with confidence the process will be OK.
It will basically be an RSS calculation because the normal distribution is centered on zero and the values are squared to convert the negative values into positive ones for addition, then square-root to finish the process.
The best result is the one that best predicts the actual result; summation of distributions is that method.
AFAIK it's rare for manufacturing to do any analysis. I'm sure it must be done sometime, but most places seem to shift an item to a machine/process they think will work and if it does, they are good. If not, they ask engineering to relax the tolerance (for cost/schedule/whatever reasons)
Keep at it. Done right, you should be able to save some money and even 'give back' some tolerance with confidence the process will be OK.
- Thread starter
- #4
Yea we make tight tolerance aerospace parts and just want to make sure that why I tolerance lengths in the process specially in turning where a lot is being machined that though stack up I don't machine something out of F/P. But using worst case I know that I would still be within print but like you said could have tight tolerances on stuff that is not needed. We have older processes that I have been looking at and can see how they were tolerating stuff and go off of that but that is kinda guessing and would rather calculate if I can. Do you know where I can find a good example of how to do the summation of distribution method you are talking about?
smwdrum,
I am reading up here on tolerance stacks. I am the guy who applies tolerances to the drawings. These reflect what I need you to manufacture for me. I try to keep my tolerance within the capabilities of the manufacturing process. Other than that, there is no relationship between my tolerances and a normal curve. If my tolerance approximates [±]2[σ], every single fabricated piece will have to be inspected, and you will have to scrap 32% of the parts you fabricate. Even at [±]3[σ], you need 100% inspection.
Machinists look at drawings of two round parts that obviously fit together, and they make them fit accurately. Pride is on the line. The shaft pushes the top end of the tolerance, and the hole pushes the bottom end. This is less of an issue if you are making thousands of parts. It is more of an issue if your tolerances are well within the capabilities of the manufacturing process.
--
JHG
I am reading up here on tolerance stacks. I am the guy who applies tolerances to the drawings. These reflect what I need you to manufacture for me. I try to keep my tolerance within the capabilities of the manufacturing process. Other than that, there is no relationship between my tolerances and a normal curve. If my tolerance approximates [±]2[σ], every single fabricated piece will have to be inspected, and you will have to scrap 32% of the parts you fabricate. Even at [±]3[σ], you need 100% inspection.
Machinists look at drawings of two round parts that obviously fit together, and they make them fit accurately. Pride is on the line. The shaft pushes the top end of the tolerance, and the hole pushes the bottom end. This is less of an issue if you are making thousands of parts. It is more of an issue if your tolerances are well within the capabilities of the manufacturing process.
--
JHG
-
1
- #6
I believe that 1 sigma covers 66%.
2 sigma is 95% and 3 is over 99.7%.
But that applies to the normal distribution.
If there is no relation between the tolerances and the normal distribution then something is is off. Hole drilling can be heavily skewed, but most other fabrication methods tend to a normal distribution over enough parts because any operation is the distribution sum of many individual distributions. Welcome to the central limit theorem.
As far as assigning tolerances, not including the central limit theorem in setting those values is often throwing money away. One could assure fit/function by taking the worst case stack that meets the requirement and cutting all those values in half, but ordinarily, why would one want to?
100% inspection on parts that have 3 defects per 1000? I guess I would if the mean was trending way off center, but otherwise it's a waste of time and money.
2 sigma is 95% and 3 is over 99.7%.
But that applies to the normal distribution.
If there is no relation between the tolerances and the normal distribution then something is is off. Hole drilling can be heavily skewed, but most other fabrication methods tend to a normal distribution over enough parts because any operation is the distribution sum of many individual distributions. Welcome to the central limit theorem.
As far as assigning tolerances, not including the central limit theorem in setting those values is often throwing money away. One could assure fit/function by taking the worst case stack that meets the requirement and cutting all those values in half, but ordinarily, why would one want to?
100% inspection on parts that have 3 defects per 1000? I guess I would if the mean was trending way off center, but otherwise it's a waste of time and money.
"These reflect what I need you to manufacture for me. I try to keep my tolerance within the capabilities of the manufacturing process. Other than that, there is no relationship between my tolerances and a normal curve."
This used to be a license to bimodally bang up against either limit, because it was easier to do that than to get into the middle. Resistors tended to be like that because the ones in the middle got marked for the tighter tolerances.
We used to have this sort of problem where we needed to have process monitors to check if the process line did their job, and it would cost 5 die positions on a wafer, in the prime real estate, to accomplish that. We had an opportunity to second source a Hitachi part, and when we checked the mask set, we noticed that there were ZERO test die, which make the process engineers very nervous. Calls to Hitachi were made about this and the fact that they didn't give us detailed process recipes for ion implants, etc., and they said, "Don't worry, be happy," not, but something close; they said to run the masks and just run our usual process and everything would be good. So, we did, and the first lot came out and yielded more product than what we achieved with our own products designed for our own process. The second and third lots yielded even better; such is the power of understanding statistical process control and designing for manufacturability.
TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529
This used to be a license to bimodally bang up against either limit, because it was easier to do that than to get into the middle. Resistors tended to be like that because the ones in the middle got marked for the tighter tolerances.
We used to have this sort of problem where we needed to have process monitors to check if the process line did their job, and it would cost 5 die positions on a wafer, in the prime real estate, to accomplish that. We had an opportunity to second source a Hitachi part, and when we checked the mask set, we noticed that there were ZERO test die, which make the process engineers very nervous. Calls to Hitachi were made about this and the fact that they didn't give us detailed process recipes for ion implants, etc., and they said, "Don't worry, be happy," not, but something close; they said to run the masks and just run our usual process and everything would be good. So, we did, and the first lot came out and yielded more product than what we achieved with our own products designed for our own process. The second and third lots yielded even better; such is the power of understanding statistical process control and designing for manufacturability.
TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529
3DDave said:
Oops! My bad. And I was looking at the curve even.
Okay, I design a part to be machined because I need the accuracy. If I don't need the accuracy, there is a faster, cheaper way to do it. If my tolerance is marginally within the capabilities of the process, the fabricator will have to work slowly and carefully, and do a lot of inspection.
--
JHG
- Status
Similar threads
- Locked
- Question
- Locked
- Question
- Locked
- Question