Determining Weibull Slope

[cycletesting]

My background is mostly test engineering & test management. I worked with Reliability Engineers in the old company but now am trying to find my own way through the Weibull statistical labyrinth. In the old company I worked for we had several years of prior fatigue data with similar designs to fall back on and now that data has gone away with the demise of the old company. We’re early in the design phases and don’t have physical test samples yet, still trying to budget test and prototype cost at this point for a similar part at a new company.

So let’s say we have budgeted 5 samples for fatigue life testing of a simple welded steel structure with well-defined loads and constraints. I would like to be able to prove 90% reliability with 90% confidence to 1x expected life of product.

Given the Weibayes equations I’ve found I can play around with cycle count to get various reliability percentile but I have to guess at Beta (slope). Anyone know where I can find examples of “typical” Weibull Beta for welded tubular structures?

 

[cycletesting]

I found this paper that seems to be going in the right direction but no clear examples...

http://www.stat.iastate.edu/preprint/articles/2002-03.pdf

Am I at least on the right track or off in the weeds?

 

[IRstuff]

?? If you're going to test to fail, then you'll have the Weibull parameters empirically, wouldn't you? That's essentially the approach described in one of Barringer's on-line articles, cited below

Barringer has a very extensive, on-line library of articles about reliability, Weibull, and AMSAA

http://www.barringer1.com/pdf/Barringer-Kuwait-1.pdf

Weibull.com likewise:

http://www.weibull.com/hotwire/issue49/relbasics49.htm

TTFN
faq731-376

 

[cycletesting]

Right, if we test to failure then we know Beta but we normaly suspended the samples if they went beyond some target number of cycles that was corolated to some x-times the expected life. This resulted in less test time so we could move on to the next component with limited equipment. When we had our own in-house resources we would often go back and continue to failure building up our "database" of Weibull slope for similar parts but this time around we will be contracting test time so we need a test target based on R90 C90 and some unknow slope.

Many parts to test and limited time / resources.

Thanks for the links.

 

[WGJ]

I used to use Fulton Findings software - "WSmith" - for very similar work and found it very useful indeed. It allows specification of confidene levels (one and two-sided) and generates Beta values. Good graphical outputs.

I have no connection with Fulton Findings. I think they are based in California.

Bill

 

[WGJ]

Just noticed that the Fulton software is referenced in the Barringer ref link. The plots are an example of the WSmith outputs.

Bill