What is Statistical tolerancing?
Vložit
- čas přidán 13. 07. 2024
- A question from a viewer as me to look into the use of Statistical Tolerancing...
Here is my buy me a coffee link.
buymeacoffee.com/osId2T0BL
Drink tea and read the paper - www.lulu.com/en/us/shop/paul-...
Design of Experiments for 21st Century Engineers - www.lulu.com/en/us/shop/paul-...
Statistical Process Control for Small Batch Production - www.lulu.com/en/gb/shop/paul-...
I did exactly this in an atempt to kind of "reverse-engineer" tolerances, based upon 30pcs samples (standard deviation x6 with 95% confidence interval) out of regular deliveries. We had real defects (out of original tolerance) but the parts have always been fit for use. The devil lies in the detail; if you assume that the process is in control and normally distrubuted and it is in fact is not, you will end up with defects again, sooner or later. You can only hope that the incoming quality control sampling does not pick the wrong part...
Hi Allan I bought your book and the first day I got to the page 67. It is so easy to read and the way that you explain it. I also show the book to my colleagues and everyone wants to borrow it. I suggested them to buy it.
I have got a question with a real example, we want to introduce a new product, the forecast sale for the first year is likely to be 100 units. Should we still run 30 to 50 units to predict the future Cpk or should that number be lower? Thanks.
Hi Alejandro - thanks for the support. Tough question on the sample size. The sample size of 30 is difficult to short cut. The best way to do it is to make a prediction of product performance based on design work. That way 10 could be used as a sample size...maybe i'll make a video to explain better..
Nicely explained Paul; you touched on uncertainty due to the stack up of the estimations - is the evaluation of uncertainty something you might do in the future?
Hi Derek - A DFSS subject, stacking uncertainty. but have soem software that works the numbers out correctly rather than over tolerancing a product. Maybe that is something that i should put in a video...
Make it simple. Focusing on extremely rare event may be over quality anyway, efforts to get there not worth and accepted by Management…
I am starting to ead your 3 books.
Many thanks for your support..
I have used a 30 piece test run and k factor to approximate the distribution, and then compare this to part’s tolerance range. This can become the start of a conversation with the part designer that perhaps
1 the tolerance is too tight
2 if we really need that tight a tolerance, we need a new or much improved process
I know this is late in the development process to identify a problem, but I have found the bigger the organization the less people want to hear about this, especially if DESIGN has the most weight on decisions. This has given me some evidence that we likely will have problems in production and inspection.
I also use a similar logic in reverse, take the tolerance from the print, divide by 7 and compare this number to historical standard deviations from similar parts or processes. Again, I use this to help assess if we are likely to have future problems.