# Destructive Test Sample Size Determination

A

#### Apoch

I am faced with a task to determine the sample size required from a bunch of test fall-outs that are to to be subjected to destructive tests. Question is how shall I come up with this figure?

I was thinking of generating an OC curve based on LTPD. But the thing is, the population is from a bunch of failed units. Hence, LTPD is not applicable.

Any inputs guys?

#### Stijloor

Staff member
Super Moderator
I am faced with a task to determine the sample size required from a bunch of test fall-outs that are to to be subjected to destructive tests. Question is how shall I come up with this figure?

I was thinking of generating an OC curve based on LTPD. But the thing is, the population is from a bunch of failed units. Hence, LTPD is not applicable.

Any inputs guys?
Suggestions anyone?

Thank you!!

Stijloor.

#### Jen Kirley

##### Quality and Auditing Expert
Staff member
Good day,

Unless I misunderstood you, you are testing failed units. If that's the case, testing units that fell out isn't destructive sampling, it's failure analysis. In failure analysis the unit is dissected and examined forensically for the cause of its defect.

But if you are interested in destructive test sampling out of good units, there are threads like Sampling Plan for Destructive Testing - Personal Body Armor Plates. When looking at that thread, please also look at the links to related threads at the bottom of the page.

I hope this helps!

#### Miner

##### Forum Moderator
Staff member
Jennifer makes a lot of good points. Other considerations include whether this generates variable or attribute data. As stated, this sounds like failure analysis. In this you may be trying to characterize some physical properties and perform a hypothesis test, or to quantify the relative proportions of different failure modes. The type of data and the purpose of the testing all influence the required sample sizes.

#### Bev D

##### Heretical Statistician
Staff member
Super Moderator
quickly because it's late: there is a similar thread here titled "sample size for rejected units". there is good advice in this series for your question.

bottom line - there is no reason for fancy statistics. take some of each failure mode (test parameter failed) and see what you get. sometimes we spend too much time on the fancy math on not enough on the good engineering of discovery.