Support OpenPSA MEF for backend modules

[troeger]

http://www.lix.polytechnique.fr/~rauzy/xfta/XFTA-Manual.pdf

First analysis by @laena:

My impression is that it aims at being much more (too?) generic than we do. From the four layers, the Fuzztrees formats covers only the lower two: the stochastic layer (in less detail) and the fault tree layer (in more detail, since OpenPSA doesn't seem to have dynamic gates or configurations).

Interesting things about OpenPSA:

• BasicEvents and FT structure are saved in different XML documents (this is similar to what Armin Zimmermann proposed - the system is separated from its failure behaviour)
• Components can be defined and re-used. I think this is weird for FT modelling (stochastic independence...) but might be nice to have for event trees
• CCF groups can be defined
• Diagnostic measures per basic event: e.g. RRW, "the maximum decreasing of the risk it may be expected by increasing the reliability of the component"...

The analysis examples seem to work, they can output top event probability and MCS and lots of numbers for the basic events.

Most of the diagnostic measures are based on conditional probabilities (e.g., how high is the probability of system failure under the condition that BasicEvent A occurs/does not occur?). These can be computed if the static FT is converted to a Bayesian network or, in the case of simulation, through simulation runs on different configurations. Here I think that Fuzztrees offers a higher-level, more intuitive view. Computing different conditional probabilities is the same as computing overall probabilities in different system configurations.

[rakhimov]

The Open-PSA MEF is a prerequisite for integration with SCRAM #105.

[troeger]

This issue is now restricted to the file format implementation. The OpenPSA engine is no longer interesting, since we have @rakhimov with his SCRAM project on board ...