Sampling Plans - Continuously Decreasing Sampling Over Time

K

kennethmabou

Hi. I have a process where we want to ensure quality levels at a process at new customer sites with 100% sampling and an AQL of 10%. Sampling by attribute. After a certain amount of time or number of samples/quality checks, and the quality has proven to be better than the AQL, then the sampling would be reduced to - say - 80%. It is re-evaluated again after another certain time/sampling period and again (hopefully) reduced. Or increased if it has been shown to become worse. This is continued until the process at this certain customer site is only sampled about - say - 10% of the time (an approximation), the minimum sampling rate for any one site :agree1:.

This is totally an empirical sampling plan/process but essentially outlines my desired situation. :confused: I'm not overly familiar with a standardized, statisically-based sampling plan based on the above. Could anyone help me out or point me in the right direction if what I stated sounds familiar?

Any help or direction at all is very much appreciated!!

Thanks so much!
:thanx:
 

Bev D

Heretical Statistician
Leader
Super Moderator
a few clarifying questions:

by 100% sampling - do you mean inspecting every part? or every lot at some sample size?

AQL is the ACCEPTABLE Quality Level. If you were doing a sample of every lot, the sampel size and accept/reject numbers would be such that a lot that was defective at the AQL level would be ACCEPTED 95% of the time...are you sure you mean AQL? or are you more interested in teh defect rate that would cause rejection? This is the RQL..

Can you elaborate more on why you are inspecting at the customer site?
 
K

kennethmabou

Thanks for your reply. Hopefully this explains it better...The situation involves us installing a system at a customer site. We receive data from this system on an ongoing basis, into the future; we don't have to do anything with the data immediately except store it. At some point in the future, we will be asked to provide analysis on this data, or some of it. It is in our best interest to ensure that the system is working properly, on several different fronts and that data obtained is usable, even though we don't need to use it yet..

So, ideally, we would inspect this data (which arrives as images) 100% of the time to ensure every image is usable in the future (or at least 90% - hence my reference to 10% "AQL"). With more customer sites, 100% inspection isn't feasible. And also, with time and experience with the system, customer sites will produce better images and inspection requirements would decrease. So, I am trying to implement a statistically-based system of decreasing the sampling plan from 100% to a minimal number over time, as each customer site becomes better at producing the images to a point we just "keep an eye on things".

So, I guess AQL is a poor term, since there wouldn't be any lots or rejection of lots, but continuous sampling over time with a target of 90% of images would be "good" or "usable".

Can you see what I mean and point me in the right direction for learning about a statistical system that might work here?
 

Bev D

Heretical Statistician
Leader
Super Moderator
the simplest appraoch is to use SPC charts. Specifically the p chart for variable subgroup size.

you can start at 100% inspection of the images and plot the number of un-usable images in each 'inpsction'. one chart per customer.


as you gain confidence in each customer you can back off the sampling fairly quickly. if the defect rate continues to improve or stabilzes at some average rate, you will see it quickly on the chart. (you will go 'out of control' on the good side for improvement adn remain centered for stabilized.) Your limits will widen as you decrease your sample size to accomodate the expected larger variation form inspection to inspection. If the defect rate gets worse, you will aslo see that as an out of control condition...easy and intuitive once you try it.

Then you can start other customers at lower sample sizes and feel confident that you can detect problems with less than 100% inspection.

no need for fancier statistics than that.
 
K

kennethmabou

Thanks Bev! Do you have any links or resources for guidances on using these processes?
 

Bev D

Heretical Statistician
Leader
Super Moderator
I assume you mean the p chart? if so, a quick google search will get you what you need. I have also attached a brief doc describing it. the rest of it is just common sense...:)
 

Bev D

Heretical Statistician
Leader
Super Moderator
ooops forgot the attachment
 

Attachments

  • Proportion Defective Chart.doc
    50 KB · Views: 223
J

JoeDM

You might consider a continuous sampling plan. They work something like this. Check every nth item. If it is not acceptable check y items in a row until there are no defects. Then go back to checking every nth item. You can google continuous sampling plans. n and y change based on the acceptable quality level (AQL) or average outgoing quality (AOQL). Maybe this is an approach to consider.
 
T

tata347

Hey Bev & any statistical guru...

How about this situation.. We are a medical device mfg organization and perform and AQL inspection of finish devices; no charting of defects (these are mature products). We use a standard (AQL 1.0 or 1.5 level II Normal) inspection plan throughout the organization (no recording of defects just pass / fail unless the lot is rejected).

We have instituted the reduced sampling plans after 10 lots but have some issues when I read the standard; so here's problem....
1) If the reduced finds any defects it appears that we return to a standard sample for 10 lots or ???
2) If it sample indicates accept on 1 - reject on 3 does this we still need to switch back to "normal"??
3) Is there any standard / guidance to reduce the sample and after 20 lots of inspection without failures what the next reduction or skip inspection methodology?

I am being pressured to eliminate inspections but can't find any statistical basis & don't want to get caught short in an audit.

Thanks..

T
 
Top Bottom