Reliability growth models

[Jorn Hennum]

I am in the process of determining a reliability analysis process for one of our projects. I am therefore also looking at reliability growth models, and in addition to what I already know I have found much interesting information on these pages. However, when it comes to the examples of how to use the reliability growth models (the examples per model given on http://www.weibull.com/RelGrowthWeb/reliabilitygrowth.htm), I have a few questions.

The data sets used in these examples are complete. I.e. there are data points with a reliability equal to the maximum reachable reliability. If such data exists for a system, there would be no need for a reliability growth program, since the desired reliability is already met. How would one determine such a growth model if only a couple of data points were available (i.e. very early in the design phase)?

Also, does anyone have any tips on which model to use when? And is the Modified Gompertz model the same as the ARINC research model (described in "Reliability Growth Models", Journal of environmental sciences, Jan/Feb 1978)? Mathematically they are very similar, but I have not yet had the opportunity to research them properly, so if anyone has an answer ready I would be very grateful!

Jorn Hennum

[Pantelis]

I presume you are referring to the models that use a reliability metric at different times. You are correct, the data set used in the examples does contain data that approach the max reliability. However, this is not a requirement -- i.e. less data can be used with the same models. Obviously as in all model fitting, the larger the data set the better the fit.

Additionally, if you are in the early phases of the design you may want to look at the NHPP model (aka Crow/AMSAA) that uses time-to-failure data instead.

On your question as to the similarity to the ARINC model – I do not recall the form of that model at this time, however if they are mathematically the same then Yes.

[Jorn Hennum]

I thought that the AMSAA model was specifically used within test phases, and that the model itself was similar to the Duane model (thus using the same type of input data). Is this not correct?

[Pantelis]

That is correct... and did mention that you are in the "early design phase"