Acceptance sampling plans, MIL-STD-105E or ... ? 3

[hkmensing]

Hello all,

I'm currently looking at acceptance sampling plans. In our company we use the MIL-STD-105E, but from information I'm gathering (mainly from the internet), it's not the best way to do acceptance sampling. Does anyone know of better ways to do acceptance sampling?
Our products are varied and quantities per lot/batch are very high (up to 50,000 or more).

Cheers,

GM

 

[jclough]

Hi GM,

I'd have to know more about your application to give you a definitive answer but in general...

Variables sampling is more economical than an attributes sampling plan. For a given degree of accuracy sample sizes are smaller. If you do this in processas part of your in process checks a subsequent final inspection may be eliminated.

All of the attributes sampling plans that I've seen are based on the same probability distributions. I like the C=0 plan only because one doesn't have to explain why a lot is acceptable even if it still contains defectives.

Another drawback of the canned attributes sampling sytems like 105D is that often people don't understand what they really mean. Example... an average person would think that if you were to select a sample at a 1.0 AQL level that if the lot passed that there are less than 1.0% defectives in the lot. Not so!! The sampling plan is set up so there is an 80% probability of Accepting a lot with 1% defectives. Hence the term acceptance plan. This also means that there is just a 20% chance of rejecting a lot with 1% defectives. Not as good as it looks eh?

 

[jrhols]

You should start with what's called an OC curve.
You mentioned MIL-STD-105 - that is for drawing samples that you can do classic inspection with - and I mean the formal definition of inspecion, not shop jargon. Mil-Std-414 is a more technical approach because if uses variables to accept or reject the lot, and those variable are based on random independent gamma variables - a fancy way of saying that no one factor contributes more than 10 percent to the overall variation, and that generally there are many sources of variation, all of them small and relatively equally significant. Such control do come with a price, called a pilot study to confirm this assumption is valid.
Also, I recommend putting the specification to various tests of validity right along with the data collected from any processes involved. If you want some number crunching tools, let me know, perhaps I can guide you in the right direction of even sell you some ideas of my own.

 

[jrhols]

I have a fuzzy logic program I wrote to use with a laptop and take the professional engineer's oral board exam. It tries to document cost, CPK(centering) and graph the specification in question taking into consideration a mix of variables we use that have different meanings depending on how they mix - like accuracy, precision, repeatability.
I have not used it in industry, most all my programs get used in real life situations, so it is not as refined and tested, but I intended to use it insituation where sampling is too much, such as multi-spindle and other high production processes or in shops that have had no prior controls on the print specs and tolerance/6 sigam controls.
I would be glad to email it to you just to take a look and see if you may want to use it to audit your specifications first, it tries to descern those specfications that need attention and has no minimum sample. Just requires a know standard deviation, mean, upper tolerance and lower tolerance limits. Send me your email and I will send you the .exe file compiled for any IBM-PC type computer.

 

[kreg]

Your question was very thought provoking primarily because of what you didn't say (what problems are you having with your existing method, are you experiencing customer rejections on your delivered product, what MIS-STD-105 inspection level are you using, what AQL are you using, are you using the switching rules, what is you average process percent defective, what are the associated sampling costs, what happens to rejected lots, etc).

Don't sample - use process control techniques to monitor and verify your production quality.

If you are experiencing customer rejections despite your best efforts with MIL-STD-105 (ANSI/ASQ Z1.4) sampling you probably ought to use control charting techniques to improve processes. Simply tightening inspection will only result in rejecting more lots and will still require process improvement to really fix the system. When processes are producing in the defects per million level, MIL-STD-105 sampling plans aren't suitable anyway (see paragraph 1.1 of your plan) so you'll probably use SPC techniques then anyway.

If your sampling costs are too high because of the number of samples you are required to pull, the C=0 plan or variables sampling (MIL-STD-414 (ANSI Z1.9)) mentioned by "jcoulgh" above might be cost effective alternatives while providing the same risk protection. However, here too you might be able to get away from sampling altogether and just use control chart data to demonstrate an effective process.

Since double sampling or multiple sampling plans haven't been mentioned in any of the earlier responses you might also want to consider those if you haven't already and the process control charting option doesn't work for you for some reason. These sampling plans typically mean less inspection if the number of defectives is either quite small or very large (will accept or reject sooner respectively) but are considered more complicated to administer than your MIL-STD-105 plan.

Juran's Quality Control Handbook probably has more sampling plans for you to consider. It's a great reference for all this stuff. Good Luck.

 

[hkmensing]

Hello all,

Thx. for all the info . It has increased my knowledge and confirmed my thoughts on MIL-STD-105E.

jrhols> I'd like to test your program.
I don't really like to post my email address anywhere public, but here it goes: gert@whihk.com.hk.

For all of you a bit more information on our products. We are a trading company in Hong Kong dealing mainly with the European market. Our products are manufactured in factories all over China. I've been to quite a few of these factories and the working standards are generally very low, though there are some factories (but regrettably not for our company - too expensive) which are certainly up to standard of what is more commmon in the west.
The Chinese idea of quality is very different from the western point of view. They think: if it works, it must be good.
But from a western perspective, functionality is only part of the deal, cosmetics are just as (or sometimes even more) important.
It is very difficult to make them realise (few excluded) that if they start a production process with quality control, they'll have less rejects at the end. They rather re-work than implement a proper process control system, because they think it's cheaper.

Cheers,

GM