Sample Size and Confidence question for Small Sample Size on Large Lots

[RMedrano]

Ive been searching through the forums, a lot of the stuff I have found seems to be Waaay over my head here.

I have very little statistical background other than Testing for normality, running Capability studies and doing Gage R&R's

We have inherited an inspection process from our sister company. This process gets lots of parts roughly in the 25k to 30k range.

Right now they sample 30 pcs from a lot of parts and perform 4-5 different variable inspections on the parts. If all parts pass all the inspections, the lot is accepted and no further analysis is done.

Basically everyone is asking me if there is a way to calculate the confidence of determining if a lot is acceptable when the inspection is reduced from 30 pcs to 15 pcs?

[world quality]

does this help:

[Bev D]

let me ask a couple of clarifying questions:
you say that 30 pieces are selected and if none are outside the spec limits the lot is accepted? And there is no calculation of the sample standard deviation or analysis to determine if given the sample mean and the sample standard deviation would indicate that some percentage parts might be out of spec? If the answer is yes, then you are basically using a categorical (old fashioned term = attributes) sampling plan.

The categorical approach would allow a 10% defect rate to be accepted 95% of the time it occured. (assuming you only accept the lot if there are no defects). If you reduce your sample size to 15 and keep the pass/fail approach then you could accept a 20% defect rate
(See "Zero Defect Sampling" in Quality Progress November 2007)

This is without adjusting for the fact that you are looking at multiple characteristics. It also doesn't take into account that you may not be sampling randomly from the lot.

There are ways to use the continuous data that you are measuring to reduce your sample size and greatly improve your protection. if you are interested you can let us know.

This

[world quality]

As Bev D, stated and here is a doc to go by.

[RMedrano]

In Reply to Parent Post by Bev D

let me ask a couple of clarifying questions:
you say that 30 pieces are selected and if none are outside the spec limits the lot is accepted? And there is no calculation of the sample standard deviation or analysis to determine if given the sample mean and the sample standard deviation would indicate that some percentage parts might be out of spec? If the answer is yes, then you are basically using a categorical (old fashioned term = attributes) sampling plan.

The categorical approach would allow a 10% defect rate to be accepted 95% of the time it occured. (assuming you only accept the lot if there are no defects). If you reduce your sample size to 15 and keep the pass/fail approach then you could accept a 20% defect rate
(See "Zero Defect Sampling" in Quality Progress November 2007)

This is without adjusting for the fact that you are looking at multiple characteristics. It also doesn't take into account that you may not be sampling randomly from the lot.

There are ways to use the continuous data that you are measuring to reduce your sample size and greatly improve your protection. if you are interested you can let us know.

This

The answer is yes, our sister facility did not do any statistical analysis on collected data, they measured the 4 or 5 characteristics of the sampled 30 pieces, and then if all parts passed, the lot was passed to visual inspection.

If any 1 of the 30 parts failed on any of the characteristics, the entire lot was rejected and put on hold until it was determined if the parts could be reworked, or scrapped. If the parts were reworked and then returned another 30 parts were then sampled and measured.

I am always interested in ways of helping our folks out on the floor, right now they may go through 10-20 lots of these little parts in 16 hours. They spend a lot of time measuring these 30 pieces out of every lot.

[Bev D]

In Reply to Parent Post by RMedrano

I am always interested in ways of helping our folks out on the floor, right now they may go through 10-20 lots of these little parts in 16 hours. They spend a lot of time measuring these 30 pieces out of every lot.

It would be helpful if you could post your data: the results of the samples for several lots and the spec limits. we could come up with a few schemes and explain them to you. please include lots that failed as well as passed if possible.

[RMedrano]

Ok, I will see if I can get the data from the engineers.

[lday38]

Has anyone dealt with the quality inspection of small aplliances? My question is on visuals , as each complete product is tested 100% for UL, safety, etc. requirements.
While each part is a looked at visually, we have one Japenese customer that visual such as scartches counts a lot and seems to be getting out.
I am leaning that at final inspection visual should be on zero sampling plan
( AQL?) since we seem to get rejects in that area only until we can have confidence that our in process controls can prevent this type fo defect.
They currently use mil std 105 but they accept if the reject number is below.
I should say this is an assembly operation.
Linda