Help Suspended Testing

[Sean T]

What's the best method for determining reliability when you suspend a test and half of the parts have the same cycle count and half have another? If you assume failure at time of suspension, I'd have 8 to fail at 4.5 million cycles and 8 at 4 million cycles. I have not seen anything on the topic of performing weibull analysis on multiple "failures" at a single cycle count (or time).

[Pantelis]

Sean,

You should not assume failure at suspension. You should treat the suspensions as suspensions. Now I assume you are familiar with analysis with suspensions. If not see discussion at: http://www.weibull.com/LifeDataWeb/analysis_parameter_methods.htm

Now the other problem that you have is that all the failed data is at the same time. In this case your best option is to use a one parameter Weibull (MLE solution). See http://www.weibull.com/LifeDataWeb/weibull_probability_density_function.htm

[R. Schop]

Sean & Pantelis,
Using MLE with this data is an option, but if Beta is known, should you not use Weibayes? This gives me normally the most accurate answers in cases like this.

[Pantelis]

Schop,

Agreed. You are correct.

If you read my reply I did say “One-Parameter” Weibull, which is the same as the so called “Weibays” method.

I personally prefer the term one-parameter Weibull because that is really what it is. You have a Weibull pdf where beta is assumed to be a constant and the only parameter you solve for is the scale parameter. To expand on that, and highlight the difference, one can also use a true Weibull-Bayesian approach where one could assume a prior distribution (not a constant) for beta. Weibull 7 (currently in limited Beta) includes options for this method.