Lifetime Estimation

[Brian Woodward]

My company does not want to conduct a full ALT, but instead would only like to confirm a lifetime calculated through MTBF.

My Question:

Is it possible to correlate a set amount of accelerated life testing to a product lifetime without a true HALT test? (i.e.-If i ran a unit @ 40% above normal operating conditions for a 1000 hours without failure. Can I reliably say that my product will last for X amount of time?)

[DLCraggs]

Since the test was conducted at a higher stress level, you need to consider an acceleration factor. These come from a Physics of failure. For example, if the test is a fatigue test, then using a fatigue model is appropriate. If it is a corrosion test, perhaps an Arrhenius model will work. If there isn't a model, you need to conduct some test to relate the acceleration factor to different stress levels and project to the real operating conditions.

[DLCraggs]

Additional comments:

You will need to test some parts at several very high stess levels to generate failures in reasonable times.

You need to generate about 6 failures at the different stress levels. But to generate 6 failures, you may need to place more than 6 under test.

You need to use a program to analyze the data to identify the appropriate failure distribution model and acceleration factors.

If the stress is too high, you may introduce some unusual failure modes that may not be seen in customer usage. These may be competing failure modes that are not seen at low stresses.

The acceleration factor will scale the failure parameter, usually a centering parameter like average failure rate or characteristic life, to predict a parameter at the operating temperatures.

Once you have an equation that predicts the failure distribution under operating conditions, you need to calculate a MTBF by integrating the product of the failure pdf and the time of failure from 0 to infinity.

If there are multiple usage temperatures, you may have to weight the time at each temperature.

The MTBF is a rather poor indicator. If an exponential distribution applied,then the reliability at the MTBF time is only about 36.8% ! If the normal distribution applied, then the reliability is about 50%. A better indicator is a B10 life (where 10% of the units fail) at 90% confidence.

[Annandhan]

Running at higher temperature and getting similar failure mode will help you to get the acceleration factor. with this factor, you can calculate MTBF