I am evaluating reliability and availability of a hydropower plant using dynamic fault tree gates. To evaluate the top event probability Dynamic Bayesian Network (DBN) is used. I am unable to figure out that how many number of time slices should I consider for my network. Also, should my time slice be 1 month-6 month-1year, or 1year-2year-5year.
Also, should all the power plant components with static and dynamic gates be represented with different time slices in the DBN or only the power plant components with dynamic gates be represented with time slices?
This question is difficult to be answered with the information provided. As far as I understand your question is related to the level of detail of your model. Which level of detail is appropriate (here, the number of time-slices in the DBN)? The level of detail of the model is related to which decision your model will support. In general, the level of detail is sufficient if a higher level of detail in the model will not change the decision to be supported using the model. This is a modelling choice and strongly depending on the purpose of the model.
Thank You Matthias Schubert for your reply. I have failure and repair data of a power plant for last 10 years. From that I have calculated failure and repair rate. Now, to calculate reliability of individual generating units of the power plant, I am using a fault tree.
Now, some components which are functionally dependent on each other are shown by dynamic fault tree gates. I am solving the dynamic fault tree with Dynamic Bayesian network (DBN). So for every generating unit I have a DBN to calculate the failure probability of the generating unit.
But for the DBN, time slices are required where my confusion lies. I have repair and failure rate from past data. And time slices in a DBN talk about the future.
For e.g. if two time slices are there, one from t=0 (here prior probability is from failure and repair rate of past data) to t = 1 month (assumed) and second from t = 1 month to t = 6 month (assumed). t = 1 month is represented by conditional probability dependent on t= 0. Similarly t= 6 months is dependent on t = 1 month
I am not sure how much time to analyze into the future (logically)?
The functional dependencies within the units can be modelled using BPN by using causal relation. I assume that the time dependence is introduced in your model by estimating non-constant failure rates. Otherwise there is maybe no need to use DBPN (but you can do depending on the question you would like to support). For decision making it might be helpful to use constant time steps like 1h, 1month, 1 year between the time slices, depending on the defined reliability criteria.
The failure probability of the next time slice can be conditioned on the survival of the time step before - this is expected to be close to one.
- Badges
- Science topic
- Similar topics
- Filing