MTBF and Lifetime

[Laurent]

Hello all,

I just wanted to know if someone could explain how to predict a lifetime, and how to test it, in comparison to MTBF prediction and tests ??
Thanks a lot :)

[Pantelis]

I am not sure I understand exactly what you are asking? Can you please expand on your question? MTTF is also a lifetime metric?

[Abhay]

How can we co-relate MTBF and Life. Is there some relation (equation) between MTBF and Life. I have come across some situations where supplier specifies product in terms of life or MTBF. Thanks in advance

[Pantelis]

MTBF (or more correctly for non repairable systems MTTF) is the mean time to failure, or the average life. Now depending on what model (i.e. failure rate behavior one assumes) that means different things.

Under a normal model the mean is at the 50% point, whereas under a constant failure rate assumption (i.e. exponential) this is at 63%. Under other failure rate behaviors (i.e. Weibull) this percentage varies and is based on the model.

Now if you use the simple constant failure rate assumption (exponential model) then
R=exp(-T/MTTF),
where R is the percent surviving or reliability and T is operating time. If you want to compute life, define what life is (i.e. time by which 95% survive, etc.). This then is your R and solve for T.

Please note that this only works when assuming an exponential model (constant failure rate), which is not a very good assumption. In the cases of non-constant failure rate behaviors (real life), knowing only the mean (MTTF) is insufficient and not the best metric. Unfortunately many people still use this extensively as their sole reliability metric.

Hope my long explanation helped.

[Abhay]

Thanks for the elaborated response, it helped me.

But, if a supplier has given life as 1000 hrs and no other information is available (like life = some% surviving) in the manual, How can we get the MTBF or Reliability at some time = t.

Thanks again,
Abhay

[Pantelis]

Assume that what he means by life is the MTTF (MTBF). From a theoretical prespective the “expected value” (most likely value) is the mean of the probability distribution.

[Powertron]

MTBF (or more correctly for non repairable systems MTTF) is the mean time to failure, or the average life. Now depending on what model (i.e. failure rate behavior one assumes) that means different things.

Under a normal model the mean is at the 50% point, whereas under a constant failure rate assumption (i.e. exponential) this is at 63%. Under other failure rate behaviors (i.e. Weibull) this percentage varies and is based on the model.

Now if you use the simple constant failure rate assumption (exponential model) then
R=exp(-T/MTTF),
where R is the percent surviving or reliability and T is operating time. If you want to compute life, define what life is (i.e. time by which 95% survive, etc.). This then is your R and solve for T.

Please note that this only works when assuming an exponential model (constant failure rate), which is not a very good assumption. In the cases of non-constant failure rate behaviors (real life), knowing only the mean (MTTF) is insufficient and not the best metric. Unfortunately many people still use this extensively as their sole reliability metric.

Hope my long explanation helped.


If MTTF is an insufficient metric for predicting Lifetime, is there a better metric to use?

I was going to use an MTTF of 500,000 hours with a 90% confidence level to specify the Reliability of a new power supply, but maybe there is a better approach to specifying what Lifetime I require. This MTTF to me means that I don't want to see any more than 1.73% field returns (1730 power supplies out of a production run of 100,000) after one year of continuous operation.

[Pantelis]

Well why not use the obvious ... in other words use a metric for exactly what you are looking for .... Reliability of (100%-1.73%=98.27%) for one year?

[Powertron]

Is there a better model to use for switching power supplies than an exponential model? The reason for my question is that many of the switching power supplies (40W Desktop unit) that I am looking at give MTBF numbers in the 100,000 hour range in their data sheet, which means that if I use the exponential model then after 1 year of operation as many as 8.4% could be expected to fail which seems like a very high failure rate.