No Failure How to predict reliability

[Rajesh Goyal]

We have put 8 units on testing for last 3 years there is no failure. Now we want to do analysis while our testing is continuing. Using Weibull software if we do the analysis will it be of any relevance? whether reliability prediction be of any use?

[tarik]

You can use the Exponential distribution or the 1-parameter Weibull distribution (assuming a value for beta).

If your purpose was to demonstrate a reliability goal then you can use the DRT (Design of Reliability Test) feature in Weibuill++ to determine what reliability goal have you been able to prove by having 8 units last for 3 years without failure.

[ken]

Dear Tarik,
How do you crank out the demonstrated reliability to n=8 and t = 70080hrs with the DRT?
I used Weibull++ DRT but could not come up with the answer! Did I miss a step somewhere?
But if I assume a nonparametric binomial distribution with C=50%, n=8, F=0 no failures then R=91.7%! Am I right?
Ken

[Ken Poi]

Correction made:
How do you crank out the demonstrated reliability to n=8 and t = 26280 hrs (3 years) with the DRT?
I used Weibull++ DRT but could not come up with the answer! Did I miss a step somewhere?
But if I assume a nonparametric binomial distribution with C=50%, n=8, F=0 no failures then R=91.7%! Am I right not to use time 26280 hrs(3yrs)?
Ken

[Ken Poi]

Correction made:
How do you crank out the demonstrated reliability to n=8 and t = 26280 hrs (3 years) with the DRT?
I used Weibull++ DRT but could not come up with the answer! Did I miss a step somewhere?
But if I assume a nonparametric binomial distribution with C=50%, n=8, F=0 no failures then R=91.7%! Am I right not to use time 26280 hrs(3yrs)?
Ken

[usara_reliable]

We have put 8 units on testing for last 3 years there is no failure. Now we want to do analysis while our testing is continuing. Using Weibull software if we do the analysis will it be of any relevance? whether reliability prediction be of any use?

I think your should be go for the weibayes analsyis, by considering the beta value.

[Tarik El-Azzouzi]

Ken,
your calculations using the non-parametric binomial approach is correct. In the non-parametric binomial approach, there is no mention of time, basically you will need to test for the duration t* at which you define your reliability goal ( Rg(t*) ).

[joshi]

u can use bayesian reliability using apriori reliability of similar systems available with the organisation. You can also demonstrate reliability based on number of trials either based on non-parametric methodology or bayesian approach. Use of F distribution table for CL and introducing a failure + success in your data lower bound of reliability can be estimated.