Using Weibull to Calculate an Optimal PCBA Burn-In Test?

[JkelleyCDS]

I feel embarrassed to say that my statistical college professor would be disappointed, but in any case I need some help on where to start.

Does anyone know where to begin when it comes to a Weibull study? It would be for printed circuit boards where temp and voltage would be the main drivers.

In a nutshell, I am trying to find an optimal burn-in test for a new program with new, untested PCBA boards.

If anyone know wheres to start, I'd appreciate the direction.

Thanks to all-
Jason

[Tim Folkerts]

Now that you have posted, some possible links will show up at the bottom of this page, so you might check those.

In addition, there is a good site for reliability engineering from a group called, appropriately enough, Weibull.com. You can look around there for all sorts of info. One particular page for getting started is at http://www.weibull.com/basics/lifedata.htm

It is not specifically for PCBA, but it should give you some ideas.


Tim F

P.S. I have no connection with this company, but they seem to have good free info in addition to the software adn services they sell,

[Dave Strouse]

JkelleyCDS,

Why do you think a Weibull analysis is the correct method?

Early life failures are typically more exponentially distributed (which can be modeled by a Weibull with shape parameter 1).

But I think you need to understand how to proceed in general.

Attached as a reference to a book available from Amazon (through a reseller that has one relatively cheap copy). The discription implies a roadmap to design an optimal burn in, which appears to be your task. If you really want to do a bang-up job, I'd get it and see what they offer. I haven't used it myself but the contents described look like what you want and the author is well known for reliability work.

Since there is only one cheap copy suggest you act quickly. I'm sure they're selling like hot cakes.

Also, a Google located paper below that makes reference' to BSI standards

[Marc]

The pdf brings back memories. I was very involved in ESS of electronic assemblies, especially around 1987-88 when SMT was becoming 'popular'. But ESS doesn't address component burn in.

Anyway, what are you doing? Are these populated CCAs? Putting a max voltage on the circuit and measuring _____ ? And how are you bringing in the thermal aspect? Are you doing burn in in a temperature chamber?

Need more info on specifically what you're inputs and outputs being measured are.

[w_grunfeld]

Originally Posted by JkelleyCDS

I feel embarrassed to say that my statistical college professor would be disappointed, but in any case I need some help on where to start.

Does anyone know where to begin when it comes to a Weibull study? It would be for printed circuit boards where temp and voltage would be the main drivers.

In a nutshell, I am trying to find an optimal burn-in test for a new program with new, untested PCBA boards.

If anyone know wheres to start, I'd appreciate the direction.

Thanks to all-
Jason

Not sure what you mean by a Weibull "study". It seems that you are looking for an Accelerated Life Test with temperature and voltage as the stress factor, and then use it for short time ESS. If my guess is right , you are talking about two related but different issues. First is to find the distribution (reliability function) that best fits your PWB. Second is to translate the accelerated test hours to normal operating hours.
Weibull is a good start when one knows nothing about the distribution to expect. The bad news is that you have to experiment and get at least a few failures and their times to see which distribution is the best fit (or conversely what are the Weibull's parameters for your case )
That is the right approach, question is do you have a few test samples 4-5 (10 is even better) to spare and a few hundred hours of test time (min)?
If your board is very reliable it may take even longer to get some failures...
Don't expect a cookbook recipe because there are none.(the weibull.com website is indeed excellent, if you'll spend some hours reading what they have there it will basically say the above. The acceleration factors are more easily converted by assuming an Arhenius relationship - a good approximation in the case of electronis, temp and voltage.
To make a long story short, you probably will have no choice than assume an exponential distribution, not because anyone can guarantee that your board will indeed behave exponentially but it is the best you can do.
You can always look up the good old military relability standards (e.g. MIL-HDBK-344 a). At least you can get it free, and nobody will chop your head off for using it.
Just my 2 cents
Willy