but, some are useful
As said by George E. P. Box. He was talking about statistical modeling and the basic idea behind actually doing the modeling.
We want to make decisions.
In reliability engineering we have, at times, a lot of life data. Construction models that describe the chance of failure over time is useful to
- understand the changing rate of failures over time
- forecast future failures
- predict performance of future products
The simplification caused by modeling comes with some risks. The model is certainly wrong versus reality, yet if close enough is still useful. It is when we ignore these risks that we make poor decisions.
MTBF is just a poor model representing failure rate over time. Using only MTBF without other information assumes the hazard function is constant. Always check this assumption as it is rarely true.
The notion that failure rates double with an increase of 10°C is based on the Arrhenius reaction rate equation with an activation of about 0.7eV. Two assumptions to check here
- Is the failure mechanism a chemical reaction and well described by
the Arrhenius equation? - If so, is the activation energy really
0.7eV?
There are other assumptions to check, models to validate, failure mechanism to understand, and more to learn. Reliability engineering is a broad and expanding field and our knowledge should continue to expand also.
Consider the models and assumptions you are using today — you most likely are using many models to assist your work in reliability engineering.
Are they really useful?
Activation Energy is not 0.7eV for all failure mechanisms. A good reference for Activation Energy for various failure mechanisms is JEDEC JEP122G-2011
Hi Hilaire,
You are exactly correct and thanks of the reference on tracking down activation energy values.
The main point of the post was to encourage everyone, just as you are, to check your assumptions. In other words, use an appropriate model.
Cheers,
Fred
Reliability is all about knowledge. Quite a few people think data is knowledge, it is not so.
Process/ Failure mechanism is what is required , POF does address this but it is limited by inadequate comprehension of failure mechanism. Secondly abstraction design requirements is yet another factor which limits knowledge an adversary for reliability prediction modeling.
Hi Raghu,
Good comment and I totally agree.
Cheers,
Fred