Duane vs. AMSAA

[Steve]

I have read a lot about Duane and Crow/AMSAA growth models but still don't see a clear, distinct difference between the two. It seems everbody has their own take on what the differences are. I have read the info on the Reliasoft website - and it has been very helpful, but I'm still a little confused. I know that Duane is deterministic and Crow/AMSAA is stoch

I understand how I can have a MTBF goal with Duane. How do I know if I have reached a goal or "acceptable" reliability with Crow/AMSAA?

I know that Duane assumes reliability growth and Crow/AMSAA allows for deteriororating and constant reliability as well (NHPP).

Do I understand more than I think I do?

Also, I don't understand what is meant by "test phases". Does that mean time between design improvements? :confused:

[Harry]

Duane model is only a regression equation based on the observation. AMSAA model is developed from the stochastic point of view.

Similarly with Duane’s model, you can fit AMSAA model with your data and estimate the model parameter. Once you get model parameters, you can calculate the cumulative and instantaneous MTBF. For details, please visit
http://www.weibull.com/RelGrowthWeb/Crow-AMSAA_(N.H.P.P.).htm (http://www.weibull.com/RelGrowthWeb/Crow-AMSAA_(N.H.P.P.).htm)

In general, AMSAA model dose not use the term “Reliability”. It uses MTBF or Failure Rate as criterion. “Reliability Growth” in fact is “MTBF Growth”. But you can use an approximation equation to transfer MTBF to Reliability by assuming that at the usage life of your product, the failure time is exponentially distributed, R(t)= exp(-1/MTBF*t). Here MTBF is the instantaneous MTBF at the end time of your test.

Test Phases: it can be thought as the “Design Phases” or test after each design modification.

[Steve]

Thanks, I will go read the link you provided.

I have used "Duane" by plotting cumulative time on the X-axis and cumulative MTBF on the Y-axis.

I have never used AMSAA, plotting cumulative time (X) against cumulative failures. Maybe the link you provided will answer the question:

"How do I establish and shoot for reliability goals (whether they are reliability (%) or MTBF) for AMSAA?"

[Tarik El-Azzouzi]

I just wanted to add a few more comments.

The Duane and Crow-AMSAA models are the most frequently used reliability growth models. Their relation comes from the fact that both make use of the underlying observed linear relationship between the logarithm of cumulative MTBF and the cumulative test time. However, the Duane model does not provide a capability to test whether the change in MTBF observed over time is significantly different from what might be seen due to random error between phases. The Crow-AMSAA model allows for such assessments. Also, the Crow-AMSAA allows for development of hypothesis testing procedures to determine growth presence in the data (beta<1 indicates that there is growth in MTBF, beta=1 indicates a constant MTBF and beta>1 indicates a decreasing MTBF). Additionally, the Crow-AMSAA model views the process of reliability growth as probabilistic, while the Duane model views the process as deterministic.

[Steve]

If MTBF is considered a "simplistic" reliability model (since it assumes a constant failure rate), is it still a valid measure of reliability growth? Is it only used because other assumptions would make reliabilty growth too difficult to track/model?

[Pantelis]

Steve,


Reliability growth, as the name implies, deals with the measurement and assessment of the reliability improvement of a product or process. In assessing that, a reliability metric is utilized and tracked over different phases. One such metric is the MTBF. Even though other metrics could be utilized, the result (improvement tracking) would be similar regardless of the metric. In other words the use of the MTBF metric in a “reliability growth” scenario is to allow us to model and understand improvement. As such its use is appropriate.

Hope this helps.

[Steve]

Pantelis,

Thanks for the response. I was beginning to think that my last post had fallen into a black hole! :eek:

I can see how MTBF could be used a measurement of reliability for any distribution that you would have (every distribution has to have a mean), but I'm still trying to get my brain around the idea that it is the best measurement (say for beta = 2).

If I am plotting the log of cumulative of failures against log time (which is how I learned Crow-AMSAA), how do I know when I have reached my goal?

[David]

Pantelis,

Thanks for the response. I was beginning to think that my last post had fallen into a black hole! :eek:

I can see how MTBF could be used a measurement of reliability for any distribution that you would have (every distribution has to have a mean), but I'm still trying to get my brain around the idea that it is the best measurement (say for beta = 2).

If I am plotting the log of cumulative of failures against log time (which is how I learned Crow-AMSAA), how do I know when I have reached my goal?

Hi Steve,

Thanks for your patience Steve. :)

In general, there is not a simple rule that says given this value of Beta your goal should be based on a certain metric. The metric that you use should be based on your requirements and application. There are better metrics to consider than MTBF. However, MTBF is widely used in reliability growth. With that said, in general the cumulative number of failures vs. time plot is not going to tell you whether you reached your goal, unless of course your goal is in terms of cumulative number of failures or cumulative MTBF. The cumulative MTBF can be calculated by using: T/N, where N is the cumulative number of failures at time T. However, most of the time in reliability growth the instantaneous MTBF is going to be considered instead of cumulative MTBF since you want to know where you are now. If you have a reliability goal at a specific time then you can use the equation that Harry mentioned to go from MTBF to reliability, assuming a constant failiure rate (exponential distribution).

I hope this helps.

[Deepti.m.p]

Hi!! i am a new member here.
Just wanted to know, how can we calculate confidence intervals for the Duane model predictions?

[David]

Regression bounds are used to estimate the confidence bounds of the Duane model. Information on the regression bounds can be found at http://www.weibull.com/RelGrowthWeb/Confidence_Bounds.htm.

I hope this helps.

[Michael1408]

Could someone elaborate on the definition of "test phases"
Improving reliability can be achieved through design changes, but if AMSAA only measures in the same phase, and a test phase is "a design iteration" then what good is measuring reliability growth when no improvement was made (since it's the same test phase)

[David]

The end of a test phase is generally at a point in time when you are prepared to implement some fixes/corrective actions. You may not implement the fixes during the test, but you may want to implement them at the end of the test. So while there is not any growth during the test phase, there will be a jump in the system's MTBF at the end of the test based on the corrective actions. This is called the projected MTBF (PMTBF) and can be calculated using the Crow Extended model. The PMTBF would then be an estimate of the system's MTBF at the start of the next text phase. This obviously assumes that failures were observed during the test phase as you are still tracking the events on the system.

[Michael1408]

ooh i see, then if the Crow-AMSAA model only works for one phase, what good would it be if tracking growth. Shouldn't there be a fix to improve MTBF but the fixes would mean a new phase which means i Can't use the Crow-AMSAA. What am I missing here?

[David]

Tyipcally, the Crow-AMSAA model is used in reliability growth analysis for when the corrective actions are implemented during the test. If corrective actions are not implemented during the test, but delayed until after the test and before the next test phase, then the Crow Extended model can be used. The Crow Extended model is an extension of the Crow-AMSAA model. Additional information on the Crow Extended model can be found at http://www.weibull.com/RelGrowthWeb/Crow_Extended.htm. The Crow-AMSAA model is applicable for the test-fix-test scneario where fixes are implemented during the test. The Crow Extended can can handle the following testing types:

Test-Find-Test: corrective actions are delayed until the end of the test but before the beginning of the next test phase. Information on this as it relates to the Crow Extended model is presented at http://www.weibull.com/RelGrowthWeb/Test-Find-Test.htm.
Test-Fix-Find-Test: some corrective actions are implemented during the test and others are delayed until after the test but still before the next teset phase. Details on this relative to the Crow Extended model is given at http://www.weibull.com/RelGrowthWeb/Test-Fix-Find-Test.htm.

For the Crow-AMSAA model, the corrective actions need to be implemented during the test. In this scenario, you would be tracking the reliability growth of the system during the test phase. If fixes are not implemented during the test at all then the Crow Extended model can be used.

I hope this helps.

[Michael1408]

Yes it is, thank you for your help!

[Lakshmi]

hi,
i am Lakshmi currently doing my masters project on Crow AMSAA Extended Model i have gone thorough all the links that you people have provided in Duane Vs Crow AMSAA discussions.....but unable to finalize what are the parameters that determine the AMSAA model and by using which parameters do we have to plot on log-log graph and what are the expression for lamada and beeta values, expressions for MTBF in Crow AMSAA Extended model...even expressions that are required to find out inital time, inital MTBF, Final MTBF(act), T Goal(act) clearly...pleaseee