Rough-Cut Reliability Apportionment

[Relman]

I've got a "System" Reliability requirement which I need to apportion among, say, five subsystems.

One of the subsystems is Off-The-Shelf (OTS) so I can't affect it's reliability, just take it as a constant.

Of the remaining four subsystems, one is vital, and runs for 100% of the operating time.

The other 3 subsystems run intermittently throughout the Op. time, and have varying degrees of criticality.

What's the best way to allocate the subsystem reliabilities? :confused:

[Tarik El-Azzouzi]

Hi,
this is a good question. I recommend the following reading material: http://www.weibull.com/SystemRelWeb/reliability_importance_and_optimized_reliability_a llocation_(analytical).htm

[Relman]

Thanks Tarik,

That goes some way towards the answer I need, but I'm still puzzled how to implement it :(

I know how to use the "ARINC" method, as described in MIL-HBK-338, and using known (or predicted) failure rates for the subsystems. Unfortunately, I don't have any such data for my proposed system, apart from a ballpark MTBF for the COTS part.

Any further recipes would be much appreciated ;)

[Tarik El-Azzouzi]

Hi,

you can try to come with an idea about reliabilty of the different components based on:
- Similarity to previous designs, or
- if the new design includes some parts that have been part of the design of previous systems, or
- best engineering judgment

...otherwise, you can use standards (Like MIL-217, Bellcore...etc) based predictions just for intial assessment of reliability before having any prototypes to test. FMEAs can help to a certain extent in trying to estimate the probabilty of failures of different components. Physics of failure is another approach for estimating reliability.

The reliabilty allocation method that I suggested in my previous post can be used to decide what your reliability specifications ought to be even if you don't have data. Based on the precieved difficulty level of improving each component and the reliability goal, you can come up with reliability specs for each component. These specs can then be updated as you collect test data and come up with actual reliability models of each component. You can then re-run the optimized allocation method to produce new specs for the components.