What should I use to estimate extent of non-conformance escapes based off given sample

[vonsteimel]

Greetings,
We've been purchasing castings for years now which were not 100% inspected.

We've since added some additional inspection to a few features and are now finding non-conformnaces in these areas.
There is no risk as they have always been there, so this is merely a paperwork exersize.

Anyway, due to the expense and low production rate of the castings, I only have a sample of about 20 parts.
I want to use this sample to "predict" the extent non-confomrnaces we likly have running in the field.

For the sake of this question, lets assume the data is normally distributed.

My initial though was to use a tolerance interval to show the likely "spread" of the non-conformances. I used a 95% confidence level over 99% population interval.

What are your thoughts on this? What would you use?
I am running minitab to do this.

Thanks,

VS

 

[IRstuff]

So, before that, it seems to me that there are two issues:
> why not remove the requirement if the non-conformity is inconsequential?
> how do you know that what you do see is as bad as it gets? i.e., where there's smoke is often a fire.

Have you looked at MIL-HDBK-109?

TTFN
I can do absolutely anything. I'm an expert!
faq731-376 forum1529

 

[vonsteimel]

IR,
I'm just a small fry.
There was a company wide procedure implemented revolving around product verification. The design engineers then ran the part through the works and came out with a number of features that need to be monitored.
The risk is extremely low as the process is stable and the part/design and production method are tried and true (40 years) with no failures.The concern is that if a freak process drift were to occur and this feature were to escape durastically non-confomring (which was deemed possible by the risk assessment performed by the design engineers... eh.) it could result in a catastrophic failure. The procedures previously used to buy-off this absolutley minimal risk are no longer in place, so now we must monitor this feature as per the new proceedures. I can't change the hand I was dealt.

We've simply shined a spotlight on a skeleton we all knew was there -- and now must do our due diligence by aligning an auditable papertrail.
This is very common with castings, the more you inspect, the more you'll find wrong.

Anywho...
I know that what we see in this 20 part sample is NOT as bad as it gets. That is why I'm thinking of using a tolerance interval in minitab to help predict what the "bad as it gets" will likly be.

Any thoughts on the statistics side? Does a tolerance interval sound like the right tool?

VS

 

[IRstuff]

You have 20 parts or 20 samples out of a larger number of parts?

The mil handbook has the statistical background for creating sampling plans and evaluating potential failure percentages. My forte isn't in this arena, but I suspect that your approach will only work for large numbers of samples in a larger pool of product. The sampling plans that the military came up with should work with smaller lots. The link to MIL-HDBK-109 is here:

http://quicksearch.dla.mil/qsDocDetails.aspx?ident_number=53894

Unfortunately, the scan quality is pretty bad, but it's mostly readable.

TTFN
I can do absolutely anything. I'm an expert!
faq731-376 forum1529