Walker and Liplo (sp?) developed their analysis after the Apollo escape system
worked in six out of six tests. If you have no failures in testing then theirs'
is the best approach of which I am aware.
If you have a test failure then standard Bernouli statistics apply. If you have
a test failure and fix the cause, you can use Bayesian statistics, which will
give a slightly better answer.
Given a small sample, you will never be able to creditably claim even 90%
reliability unless you are willing to also accept near zero percent
confidence...which I can't recommend, you'll get laughed at by the
knowledgeable....
Bill
Sent from my Commodore 64.
-------- Original Message --------
Subject: [AR] Re: Destructive testing and reliability
Local Time: October 28 2015 10:27 am
UTC Time: October 28 2015 4:27 pm
From: james.padfield@xxxxxxxxx
To: arocket@xxxxxxxxxxxxx
I suppose I should have given a bit more background...
I am testing the reliability of an explosive train, i.e. from booster to main
charge - so when I said the testing was "destructive", I really meant it ;-)
There is an aluminium barrier, then an air gap, and then another aluminium
barrier between the booster and main charge. I am trying to figure out, as I
said, how to put a number on the reliability, e.g. 99.9 %, but of course
testing is expensive and therefore I can only carry out a very limited number
of firings.
And like I said, if I make the thickness of the two aluminium barriers slightly
outside the manufacturing tolerances, and also make the air gap slightly
larger, how I can put a % reliability on it...
On 28 October 2015 at 17:00, rcktman <dmarc-noreply@xxxxxxxxxxxxx> wrote:
Or--as SpaceX has discovered--if no failure is acceptable then test every part
to well above the design load and use only those that pass....
Bill
Sent from my Commodore 64.
-------- Original Message --------
Subject: [AR] Re: Destructive testing and reliability
Local Time: October 28 2015 9:47 am
UTC Time: October 28 2015 3:47 pm
From: neil.jaschinski@xxxxxx
To: arocket@xxxxxxxxxxxxx
Hi James,
just some quick ideas to your problem:
- Design your part for the load you want to handle it later.
- Put some safety margin on it, like 1,2 times thicker or so.
- Test it until it is breaking. See if the failure load is higher than
the designed load. If yes, that's it.
You can run some statistic with some more samples and test. You can also
calculate with the standard deviation of the failure loads and 3 sigma
span if this is still outside of designed load. If no, put more safety
margin on it and repeat the procedure. If it is in 3 sigma, than 99,9%
will not fail.
Greetings
Neil
Am 28.10.2015 um 16:29 schrieb James Padfield:
Hi,
Ok, the application I am asking about is not strictly rocketry-related,
but I know similar topics have been discussed here before, and I feel
the answer to my question may be useful to some on here. If you think
it is is too OT, please tell me (politely!) and I will take my query
somewhere else...
I am trying to work out how I can relate statistics on reliability to
the number of tests conducted.
For example, let's say I test ten items (destructively), and all ten
items pass the test. What can I conclude about it's reliability?
Obviously I can't conclude that said item is 100% reliable on the basis
of ten tests... What if I test it 20 times? I assume there is an
equation to work these things out...
Ok, let's make it a bit more complicated...
Say I deliberately make my ten test items slightly outside the spec that
I would use for manufacturing. For example, the tolerances on the wall
thickness of a hollow aluminium cylinder is 2 +/- 0.05 mm. If I make my
ten test items 2.10 mm thick, and again all ten items pass the test, can
I put a number on the reliability?
Thanks for any help!
James.
