Error Bars on System Downtime Prediction

[Guest]

I am hoping that someone can help me. I am performing some Markov modeling to derive system downtime. I am supposed to derive error bars on my downtime predictions. I want to know how I can derive these error bars. Do they come from sensitivity analysis by varying for example the in service fault detection parameter in my availability model? I am looking at a system which is composed of many susbsystems that each require Markov modeling. How do I calculate the error bars for total system downtime which is the sum of all subsystems? I'm assuming things like fault detection (varying it from 80 to 95%), Switch over success probability etc.

[Guest]

Sadath...I'm with you. If you are doing what we normally use as Markov models, there are no error bars. The models are just solved using a matrix or exponential poly's. However, I have seen some techniques here and there where Markov chains are solved using Monte Carlo techniques. Try investigating this approach. Unless you use a Monte Carlo simulation, I don't know how you are going to get error bars.

Joe

[Guest]

1. you can provide ranges or uncertainty distributions for all the rates and run a monte carlo over the system of markovian equations
2. http://www.ee.duke.edu/~kst click on markov chains on the left and you have links to papers on "sensitivity analysis in markov chains" and "importance analysis in markov chains"

swami