Re: Weibull Distribution but with a varied cycle time


David...

While I'm thinking about it, run a Google search on Weibull Graph
Paper and you can quickly find some excellent paper in pdf format. If
you need insight into how to plot it and how to interpret it, I can
copy some pages from a book (no longer in print... but the best and
easiest-to-understand book on this) and ship that to you. Trying to
do this with Excel is very error prone... gives you no insight... and
a genuine time-waster.

Re... the matter of failure on startup. That's not going to fit into
a neat Weibull framework. In a Weibull analysis the data is always a
positive value (not a zero time-to-failure). The sensible thing to do
with startup failures is to document those as "frequency of startup
failures"... an estimate of "the probability of a failure on
startup"... and treat that as a separate issue. Then... for those
cases where actual (positive values) time-to-failure data were
obtained... run the classic Weibull analysis on those data.

No, startup failures are not censored data. Censored data means there
was an ongoing opportunity for failure, but it hasn't happened after X
hours when we finally stopped the test.

Among other things, a valid Weibull analysis provides a way of making
certain predictions. Predictions such as "This device/system has been
running for 200 hours. What is the probability it will fail in the
next hour? In the next 100 hours?
Before 600 hours of service?"

I do a lot of work with failures in certain types of electronic
devices, so I have a good sense of what you are doing. I say "go for
it." OMU






On Jan 27, 4:50 pm, "googlinggoog...@xxxxxxxxxxx"
<googlinggoog...@xxxxxxxxxxx> wrote:
Thanks,

Basically I'd like to do it myself, as i'll learn that way, but
basically the device will be in an automated rig to collect the data
to a computer so i dont need to sit there for hours, and i can let it
run over a week or more 24/7 if need be.

The device powers on and supplies voltages at set amounts (although
they could be varied), so things like the current in and the voltages
out, can be checked to ensure that the values are within tolerance.

But occasionally it has been know for the units to fail on power-up or
into there run time, so i want to get some stats on that

is a failure on power-up classed as censored data? does this look like
a good case for time-to-failure?

its really both power-up failure and time to failure im interested in,
prehaps something simple like exponetial distribution would be useful
for a failure on power-up?

Thanks

David

.

Relevant Pages

  • Re: Weibull Distribution but with a varied cycle time
    ... The classic Weibull analysis is concerned with time-to-failure data or ... Or cycles to failure, for that matter, since ... the last way I'd go about a Weibull analysis is to use Excel. ... my unit could fail anywhere turn on to ...
    (sci.stat.math)
  • Re: Weibull Distribution but with a varied cycle time
    ... basically the device will be in an automated rig to collect the data ... The device powers on and supplies voltages at set amounts (although ... is a failure on power-up classed as censored data? ... its really both power-up failure and time to failure im interested in, ...
    (sci.stat.math)
  • Re: Virtex4 burn-in failure
    ... I've observed a failure of one of our Virtex4-based products that is a ... current and gets hot immediately upon power-up. ... related causes of a failure like this? ... Do any of the power supplies EVER exceed the maximum allowed? ...
    (comp.arch.fpga)