Temperature acceleration factor for ALT planning (question posted to Linkedin Society of Reliability engineers group, 5/7/12
Hello, can anyone advise me how to calculate temperature acceleration factor for a complex system including cards, RF elements, cables, motors and moving parts? Is the Arrhenius model valid for such systems, or there are more precise models? Thank you!
…and my response…
The acceleration factor equations are commonly tied very closely to a specific failure mechanism. For example, for SAC solder joint fatigue uses the modified Norris Landzberg model. And, for metal migration/corrosion within plastic encapsulated packages Peck’s equation is useful.
Each failure mechanism reacts to stresses differently, some more so than others. Making testing at a system level with accelerated stresses, … , more difficult.
So be careful, unless you know which failure mechanisms are most likely to occur during use and you accelerate appropriately.
If you know that temperature is the stress of interest related to the dominate failure mechanisms, again be careful, as you will quickly find, the activation energy is important and is generally associated with a specific failure mechanism.
And just temperature may not be sufficient – for moving parts, load and frequency of motion may be more useful for acceleration. For connectors and cables, maybe thermal cycling is more important.
If I don’t run a motor with a high temperature and humid environment it may fail due to corrosion. If I simply run it with an unbalanced load, it may wear out the bearing quickly – both lead to failure, yet have completely different acceleration factors (failure mechanisms) and one test or one AF is just not sufficient.
In short, there are more precise models – depending on the failure mechanisms involved.
hope that helps.
Fred
Thank you for your explanation!
Your answer is correct theoretically, and I wish I had the resources for testing the system “by the book”. But, as usual in the industry – especially in small companies such as the one I work in – I need to work under limitation of resources. The actual need is to perform Reliability Demonstration Test for the whole system. The required test time to show, with the required confidence level, that the system MTBF (Theta, if you prefer) is sufficient, is too long, when there is a limitation of available number of systems, and time allocated to the test.
I was looking for possible acceleration factors for this purpose. The reliability prediction pareto shows that the parts most affecting the failure rates are the bearings and motors, and the electrical cards – so I assumed to propose acceleration by increased/unbalanced load for the moving parts, and by temperature for the cards – but I need to decide which temperature acceleration model to use.
No FMEA was conducted during the design process, so no significant inputs about failure modes are available.
Any advise under the given circumstances will be welcome with many thanks.
Natasha
Hi Natasha,
I fully understand the problem and constraints. Here’s what I suggest – and it’s not strictly by the book, yet closer. Talk to your lead designers about what is expected to fail and under what conditions. Gather information about the use environment, not just the limits, what happens every day. And, make an educated guess or two.
If your environment includes a very large change in temperature every day (or hour), then thermal cycling may the the stress to focus on and plan your testing acceleration factor around solder joint fatigue (modified Norris Landzberg). If thermal cycling isn’t a big deal (over 20°C per cycle) then consider the thermal exposure and humidity… if in hot humid conditions – then corrosion, metal migration, etc. may play the dominant role. Then plan the testing around Peck’s equation. If both, then do both.
For motors – they are often easier to get samples and set up a separate test. The motors, if in series, reliability wise – will need a higher reliability target, and more samples or longer run time for demonstration. If you have to use them within the system – then hopefully you can load the motors while doing one or more of the other tests to get the sample size up.
Unfortunately, there just isn’t a magic temperature acceleration factor for all failure mechanisms – some failures do not care about the temperature, as they fail at the hands of other stresses.
A little creativity and focus on expected failure mechanisms may move your testing program along.
Cheers,
Fred
Hi Fred,
I had no knowledge about Norris Landzberg and Peck’s equation, I’ll go learn about it.
Of course, I haven’t assumed to use temperature acceleration for all the failure mechanisms. I just didn’t know any other model for temperature-induced failure mechanisms except Arrhenius.
I’ll adopt the overall approach you suggested. Thank you!
Hi Fred,
Useful article, I’m looking at some guidelines on the maximum permissible acceleration factors for stresses such as vibration. I could see in some calculations it is around ~10000. Will it makes sense to have such a acceleration factor?
Thanks. Rajesh
Hi Rajesh,
I too would be questioning a 10,000 acceleration factor – and, it all comes down to the failure mechanism and how it behaves as the stress is changed. If the accelerated conditions creates the same failure mechanisms as would occur at normal conditions, then it’s fine. Often this takes some really fundamental understanding of what and how sometime fails.
Keep asking questions and checking the assumptions.
Cheers,
Fred
Hi Fred,
Thanks for the response.
Regards,
Rajesh
I noticed on there you were thinking 10,000 as an acceleration factor is too large. Is there a rule of thumb of this is too large of an acceleration factor. I have preliminary set up tests with a mix of vibration and high temperature stresses gives an acceleration factor of a little over 1000. We do not believe it would introduce new failure mechanisms. Vibration increases just has a giant influence on the acceleration factor. I do not know if you have experience where you have seen tests with an acceleration factors that high and everyone seems comfortable with it.
Hi David,
In general, we don’t get something for nothing. A very high – say over 50x – I ask more questions on the connection between accelerated failure mechanisms and use condition mechanisms, plus how the stress behaves to create the damage. As far as I know there isn’t guideline or set of rules around this, yet in my experience and given the cost in setting up and running accelerated testing I’m cautious when the acceleration factor gets high.
Cheers,
Fred
PS: of course, it’s all in the details around the mechanism, stress, etc.
Fred,
The acceleration factor used in the Arrhenius equation, is a SWAG and is based on the physics of failure. It is best to be conservation in estimating it. What I have done is to estimate the acceleration factor empirically. What I have defaulted to was an AFR = 24
1) Through Accelerated Life testing, one can estimate 10 years of life by performing a six week thermal testing on the system. Assuming the math is correct. In my case, Oven and Ranges. The most extreme event in cooking is performing a self clean where the cavity temperatures reach 900F. Oven assemblies inside the Range can experience temperatures in excess of 105 degrees Celsius. Six weeks is 1008 hours of testing and 87,600 hours for 10 years. Yielding an AFR of ~87. An example for failures seen in life test at 1 week would equate to ~1.6 years of life.
2) Estimating AFR using field failure data. Our good friend Mark with Atwater Consulting, developed an R application using the Markov Chain and Monte Carlo formulas that allow one to put in actual time to failure. Using that, I could estimate both working and failing units to model AFR, which is how I got to 24 as system level AFR for Ranges and Ovens
Hi John,
thanks for the comment. I agree that many simply SWAG the activation energy. I highly recommend working with a chemist that is able to experimentally determine the activation energy for the specific failure mechanism in question. This does require one to know the failure mechanism that we are trying to accelerate.
Too often, teams simply toss the entire system into an oven, use a little more than random activation energy unrelated to any of the potential failure mechanisms and hope for the best. Often resulting is less then useful results.
Going beyond a focus on a specific mechanism and associated activation energy is often folly given the complexities of dealing with multiple failure mechanisms, some of which are a chemical mechanism, and others are other mechanisms that are poorly modeled by Arrhenius. Plus, the different acceleration factors at play, masking or obsuring the ones that are important at normal use conditions.
Estimating an activation energy empirically is fine, if and only if, there is only one chemical based mechanism at play and of interest.
Keep in mind the temperature can play in a role in mechanical mechanisms as well and rarely are well described by an Arrhenius model.
System level or even module level accelerated testing is fraught with complexities and often very poor assumptions.
cheers,
Fred