[ | Next | Previous | Up ]

Re: AQL - How this can be!

From: Stan Hilliard shilliard@samplingplans.com
Date: 6/25/2004
Time: 10:52:20 AM

Comments

Johnny Chua said: "But wait here, 1 defective in a sample size of 32 is 3.1% (1/32) not 1.5%. How can this be?"

That reveals a common fault with the Z1.4/Mil-Std-105 method of designing sampling plans.

From software program TP105, the sampling plan n=32, Ac=1 gives a producers point in decimal fractions: (Alpha=0.05, AQL=0.011218), and a consumers' point: (Beta=0.05, RQL=0.139849).

see: www.samplingplans.com/software_oc.htm

In percentages: (Alpha=5%,AQL=1.1%) (Beta=5%,RQL=14%).

The program also shows that a lot that is 3.1% defective has a probability of acceptance of 0.7546 (75.46%)

If you compare these numbers to the OC Curve on my web page:

see: www.samplingplans.com/usingoccurves.htm

you will see that it all makes sense. This example illustrates why the Z1.4/Mil-Std-105 method is not good practice to design a sampling plan.

With the Z1.4/Mil-Std-105 method, most people end up with a sampling plan which does not perform probabilistically the way that they think it does.

To design an acceptance sampling plan properly, the two points on the OC Curve should be planned. Also use the oc-curve to know the quality level that has a 50% probability of acceptance. Not to just look up a plan based on the producer's point only, as with the "AQL only" plans.


Last changed: November 20, 2007