Temperatrue Cycling Testing

[Laualex]

I've currently done a temperature cycling test for 1000 hours on 20 samples of a device. The delta T is 65 degree C, from -5 degree C to 60 degree C, and the duration of each cycle is 4 hours.

I am unsure of how can I use ALTA 7 to analysis and determine the estimated lifespan of the device with two failures after the conclusion of testing as the examples I see all utilise two different values of delta T to do analysis. Anyone can help me out on this? Thank you.

[David]

With only a single test condition, this data cannot be analyzed using ALTA. ALTA would be unable to determine how stress is affecting life with only a single stress level (delta T). In general, multiple stress levels are required to determine how life is changing with stress. An article on analyzing thermal cycling data usign ALTA with multiple stress levels (test conditions) can be found at http://www.weibull.com/hotwire/issue19/hottopics19.htm (http://www.weibull.com/hotwire/issue19/hottopics19.htm).

If you want, you can make an assumption regarding the acceleration factor between the accelerated test conditions and use conditions. In doing so, you could analyze the data using life data analysis and then apply the acceleration factor to estimate the results at use conditions. Keep in mind that your estimates are only going to be as "good" as your assumption for the acceleration factor. Other methods may be used, such as the Coffin-Manson relationship, but additional assumptions will be needed. These methods should be used with caution as they are greatly dependent upon your assumptions.

I hope this helps.

[Pantelis]

And to add to Dave's comments, it’s not whether ALTA can analyze it, but rather can you analyze it based on that data. As he clearly pointed out you need a minimum of two conditions to build an ALT model (without any assumptions) for a single stress. In the case of temp cycling you actually have more than one stress, which is at minimum the delta T and the cycle duration. Given the fact that you have two stresses then you would need a minimum of three test conditions to build any sort of a realistic model. Now with that data …yes you can analyze it in ALTA.

Having said that, I need to mention that there are published simplifications with predefined constants that only require you to use a single condition. Unfortunately such simplifications (i.e. the assumption of the constants) are not usually representative of your product.

[Laualex]

Thanks for the advice, I would like to know how do I go about using the published simplifications then, so that I can have an estimate about the product lifespan. Besides ALTA, what other means are available for me too if I would need to do the analysis based on just one test condition.

[Pantelis]

The issue here is that you are trying to find an acceleration factor form “your test condition to use condition” by only looking at a single stress condition. This is akin to trying to determine the speed (and acceleration) of a moving car just by looking at a single snapshot (picture) of the car on the road. You can’t do it! Now you can go about saying that the speed limit on that road is X; it’s a corvette and the average corvette driver tends to have a heavy foot and drives XX mph over the speed limit, and there is a light X miles back and he/she started there (assuming it was red at the time)… etc… Using all the values from these assumptions will yield an answer; however I would have very little faith in it, and more importantly the original data (i.e. the picture) would not be influential in the results, other than to tell us it was a corvette and the location.


Now if you still want to do this, then you would need to do a thorough literature search for the specific device, physics-of-failure and/or failure mode that you are looking at (i.e. the assumptions that would drive your result). This search may yield a similar test done by somebody and/or an industry standard that provides the parameters (e.g. acceleration factor) of interest. It is unlikely that you will find anything that specific, however you may find generalizations and your results would be purely based on the assumptions found in those generalizations. I can’t really be any more specific in this case or point you to a specific pub, as I don’t know your specific product, and more importantly I could not recommend using this approach.

If a reasonable estimate is desired my advice is to employ a well designed and more informative accelerated test (that would involve more test conditions).