From: Stan Hilliard
Date: 17 Feb 2001
Time: 00:32:50
Hi Tom,
First, if by "level of inspection" you mean the "inspection level" that is used in Mil-Std-105, that is not statistical in nature and I cannot interpret it in statistical terms.
Done properly, there are two levels of quality that you need to choose: 1) the acceptable quality level (AQL) and 2) the rejectable quality level (RQL). The sampling plan will discriminate between those two levels -- accepting lots that are at AQL and rejecting lots that are at RQL.
You should choose or reduce a sample size of incoming components in a way that you know the effect of your choice on the plan's ability to detect an off-grade lot. The tool that enables you to know this effect is the Operating Characteristic Curve (OC Curve). I recommend that you start by looking at the following page.
www.samplingplans.com/aqlprimer.htm
Once you develop a fixed-n sampling plan having the oc curve that meets your goal, you can focus on reducing the sample size.
You can reduce the sample size by increasing the distance between AQL and RQL. There are two limitations to this approach:
1) You can't move AQL below the normal operating level of the process without risking the rejection of too many good lots. The oc curve shows the probability of this.
2) You can't move RQL to high without increasing the risk of sending off-grade lots (if they should exist) to your customer. The oc curve shows the probability of this.
Another way to reduce attribute sample size is to convert the fixed-n plan to a matched (matched = having the same oc curve) sequential sampling plan. To match a fixed-n plan with a sequential sampling plan, refer to the following page:
www.samplingplans.com/modern3.htm#MATCHING
The resulting sequential plan provides a sizable reduction in sample size compared to the fixed-n plan. And you do not lose any of the protection described by the fixed-n plan's oc curve. It is like getting something for nothing. The only case where this doesn't reduce the sample size is if the fixed-n plan has an acceptance number of zero.
Sincerely, Stan Hilliard