Reliability O failure

[sabrina]

Hello ,

Can I please have a explanation.

Here is the problem :

A few parts are getting a sequence of different tests.( eg :The same parts are getting stress 1, then stress2 .. but there is no failure during these different tests ).But they are different failure modes a,b,c and there are independent.


My question is :
Can I use bloksim to determinate reliability when there is no failure ?

[tarik]

With the kind of data you have, you are a little restricted since you don’t have any failures. All what you can do is use the suspension times and use easier a 1-parameter Weibull distribution (assume a beta value), see:
http://www.weibull.com/LifeDataWeb/weibull_probability_density_function.htm
or you can use the exponential distribution (that assumes that you have a constant failure rate) based on the model parameters you can obtain other metrics, such as reliability at certain time or mean time to failure. You can separate this analysis by different failure mode and obtain different models depending on the failure mode.

[sabrina]

Hello Tarik,

Thank you Tarik for your answer .


Now, I have different reliabilities depending on failure mode .

How could I calculate the total reliability ?

Can I multiplicate the different reliabilities depending on failure mode, can I use Blocksim to determinate the total reliability ?

I'll appreciate to have once more your comments on this
Thank you for your help

Best regards

[tarik]

Yes, if these failure modes are independent, just multiply their reliabilities. In BlockSim you can do that by creating different blocks for every failure mode, conect them in series and analyze the RBD and obtain the differnt calculations you are interested in.

[sabrina]

Hello Tarik ,

I would have Tarik's comments on the following :

2 methods are presented :
with failure mode reliabilities and with test reliabilities

1/ For a better understanding , If there are different times for each reliability failure mode , is it possible to multiply their reliability ?

More explanations:

I have a reliability for failure mode A , RA= 79% for a given time = 1000 hours

Then I have an other reliability for failure mode B , RB=90% for a given time = 400 hours .

Can I multiply RA and RB?


2/ With my test sequences (eg :The same parts are getting stress 1, then stress2 .. but there is no failure during these different tests test 1, test 2..)-

With weibull++ software , I can make a weibull analysis with test 1 result, then an other weibull analysis with test 2 result ...
I obtain different eta with a beta =1 and different reliabilities (test 1 reliability , test 2 reliability ...)

Am I wrong ? (these tests are not independent, the same parts are getting different tests successively )
Then can I multiply the different reliabilities of each test or it's wrong ?

Thank you once more for your help Tarik

Best regards

Sabrina

[tarik]

1 – You can’t multiply reliabilities at different times, either obtain your reliabilities at the same time, then multiply them, or, if you have a model (ex. A Weibull a distribution , Exponential distribution, Lognormal distribution) for each failure mode, you can just enter these models for the blocks in your BlockSim RBD and just use the QCP to calculate the overall reliability (that takes into account all the different failure modes)

2 – Do you assume that the life that is accumulated on a unit under test1 doesn’t make a difference when you start testing under test2? Meaning that, for example, even if a unit has survived test1 and accumulated 100hours under that test, then the same unit was put under test2, that it’s like it’s starting ‘fresh’ under test2, regardless of the time it assumulated under test1? This is generally true if the tests stress different parts/components of the system

If yes, then, since you don’t have failures, you can model, with the Weibull++ software, the suspension times (the times accumulated on the units when test1 stopped) using a 1-parameter Weibull distribution or an Exponential distribution, then obtain your model parameters. Repeat the same thing for the suspension times you collected under test2.

It’s generally recommended that you test until you see failures, because with only suspensions times, you need to assume a beta value (if you use the 1-parameter Weibull distribution) or assume a constant failure rate (if you use the Exponential distribution) which is sometimes a dangerous assumption to make! If you can’t test for a long time until you start observing failures, then use accelerated (elevated conditions) testing, and analyze your data with the ALTA software.