MTBF, MTTF interpretation 2

[msinger]

Looking at MTTF and MTBF data, if the MTTF of a device is 20,000 hrs (say 100 devices are tested)does that mean we should expect one device to fail after 20,000 hrs out of a population of 100 or that the expected life of any individual it 20,000 hrs?

 

[melone]

The Mean Time To Failure gives the average time expected for the device to fail.

 

[msinger]

Also, how is an MTTF determined if you do not have a failure at the conclusion of your teting? It seems from the data sheets I have seen as they are just inserting a guess (on the first failure) into the equations and computing the MTTF based in this guess.

 

[IRstuff]

There is a theoretical basis for assigning MTBF values based on life testing, which allows one to make predictions based on 0, 1, 2, etc., during the life test.

Section 8 of MIL-HDBK-338 discusses the theory in some detail.

TTFN

 

[designchain]

Msinger - If the MTTF is 20K hours for a given device and your population is 100 units, you should expect 1 or fewer failures out of your population after 20K total device hours (or, in this case, an "average" of 200 hours per device).

By the way, there's always confusion between MTTF and MTBF. Mean Time To Failure is used for non-repairable devices like components/integrated circuits and is the inverse of the failure rate (FIT). Mean Time Between Failures is used for "repairable" devices, which include assemblies like systems, disk drives, etc. They're calculated essentially the same way.

Here's a good tutorial on component failure rate calculation:

http://www.odyseus.nildram.co.uk/ComponentReliabilityTutorial.htm

Be careful taking this stuff and these numbers at face values. It's important to understand thermal activiation energies, temperature acceleration factors, voltage acceleration factors, node coverage during life test, and so on to really be able to understand the relevance of a given MTTF to your situation.

Mike

--
Mike Kirschner
Design Chain Associates, LLC

http://www.designchainassociates.com