Weibul Analysis - HELP

[chadwick1875]

I need to run a Weibul Analysis for my testing and am looking for some information about it. I have never used it before for testing and not exactly sure what it is. I have to meet a R95C90 level form my tests. I've looked at some brief Weibul examples on the web but am still a little confused. I found a program that you can input your Reliabilty and Confidence level and it determines how many sample parts must be tested to one life. Then it says you can reduce the sample size by using the Weibul Slope. Where do I determine what the Weibul Slope is?

If anyone can give me some information about doing a Weibul Analysis I would greatly appreciate it.

Thanks,

Chad
Product Engineer

 

[CESSNA1]

CHADWICK1875: There are several books on Weibull Analysis from Amazon.com and other places. There is some software available, one is RELEX. I have never used it or any other Weibull Software. You may also want to try the Reliability Analysis Center at Rome, New York, USA. They offer classes and other stuff.

Basically and quickly Weibull analysis consits of plotting failure rate vs. time on log-log paper. You will wind up with a straight, a concave or convex curve, or a "S" shaped curve. If straight you are OK. if curved you need to make adjustments to the curve to make it straight. ( This is where the software helps). If "S" shaped the curve is invalid.

Regards
Dave

 

[WGJ]

I used Weibull extensively a few years ago for the analysis of end-of-life fatigue failures on some fuel delivery assemblies.
As far as software is concerned I can wholeheartedly recommend WeibullSmith from Fulton Findings (

http://www.weibullnews.com/contents.htm

and

http://www.bobabernethy.com/pdfs/TNWHBSV_Cover-Preface-TOC.pdf).

This package uses just your data to produce results, no guessing at parameters. It's good.

I'm no statistics expert, but here goes with my understanding. I'll apologise in advance for teaching sucking eggs.
The first thing to appreciate is the so-called 'bathtub curve'. Check here:

http://www.weibull.com/hotwire/issue21/hottopics21.htm

If you want to plot straight lines, you need special Weibull graph paper, either '2-cycle' or '3-cycle' with the vertical axis showing cumulative density function.

http://www.weibull.com/hotwire/issue8/rb8_7.gif

I used to find that, if I set out to test 8 samples to failure, then using the Fulton software, I often had enough data after 5 failures to get a good fit and respectable results.
This assumes that the design under study does not suffer from multiple failure mechanisms and that any manufacturing processes are stable.




Bill