How different definition between Reliability and Confidence??

[AromaticsThailand]

When I read in weibull.com, I can't to parting between Reliability and Confidence definition.

Please tell and example to me.

Thank you.

[Pantelis]

See also discussion at http://weibull.com/hotwire/issue4/relbasics4.htm. In this case your trying to estimate reliability (% marbles) by looking at a certain sample of units (how many marbles you picked).

Another way to look at it is to think of reliability as the answer you seek and the conf. as an indication of the error around that answer.

[AromaticsThailand]

I've to be careful reading in the link....I could not to recall (may be my english not good). If I cut and forget conf. and only calculate reliability, Can I will get the exact result?

[Pantelis]

For practical purposes there is no exact result.... what you do is estimate reliability ... conf intervals tell you how good your estimate is. Whether you choose to look at them (or calculate them) has no effect on how good your estimate is.

Imagine telling somebody that is blinfolded to walk straight --- and also telling him that most people deviate +-20 degress. If you do not tell him that most people deviate +-20 degrees is he then going to walk better (more straight) than everybody else because he is not aware of the deviation?

[Robbie]

Further to the replies on Confidence & Reliability. Calculating a confidence around a single reliability parameter seems reasonably straightforward e.g. estimate an MTBF and then calculate the confidence around the estimate (by whatever method) - where it becomes unclear is when many components, equipment and systems are combined in block diagramatic model, how do these models handle many reliability estimates when (theoretically) there could be a probability associated (probability on the probability).

For example, usually when putting together a reliability model (say monte carlo simulation), data estimates are used which are failure rate, MTBF, MTTR etc - calculating confidence arounds all the data inputs would take too much time.

Where I find it frustrating is explaining it to others with limited reliability and statistics knowledge. Anyone got any thoughts on this? :confused:

[penguin]

:)
I'm also interested in this question. Please give a example.

confidence level ?

[Arai.M]

Robbie,
Confidence bounds for analytical results of RBDs are available in Weibull++ 7 where Fisher Matrix bounds are available.

However when you start adding raindom variables for repairs, resource logistic delays, preventive maintenance properties, etc. I also struggle explaining why bounds for simulation results are most often not practical. The time to obtain bounds is definetly one concern. In addition to all these random variables, you are adding information on the variability of the parameters.

You also need information for variability of the parameters for ALL random variables used in the model, for all failure distributions, for all repair distributions, etc. When data is not available for all input, then obtaining condidence bounds is not possible. Often people want to be able to input this variability as educated guesses or use variability in some inputs but not others which generates bounds that are not even correct.

When data is available for all inputs and simulation time is not of concern, variance/covariance matrix can be used to generate random variables with joint distributions.

Penguin,

The example provided above should help illustrating confidence bounds http://weibull.com/hotwire/issue4/relbasics4.htm. This concept can also be illustrated via Simumatic in Weibull++ 7. See http://www.weibull.com/LifeDataWeb/simumatic.htm#parameters_cb

Hope this helps,
Arai

[Robbie]

Thanks Arai, I'll look through the examples you sent me. I'm attending the MSMT/Life Data/Weibull++7 training in a few weeks, so I'll have an opportunity to talk through this with the instructor hopefully.