Disable optimization step

46 views
Skip to first unread message

r.maa...@gmail.com

unread,
Dec 7, 2015, 9:33:56 AM12/7/15
to scram-dev
Hi,

I would love to use SCRAM as an calculation engine for the input for a visualization tool I am developing. One of the benefits of your tool is that it shows the intermediate probabilities. However, if a fault tree is constructed as a set of gates that can be simplified, these intermediate results are not present. Is there a method to disable the simplification to obtain all intermediate results?

For example the FT

g3 = g1 AND g2
g1 = b1 AND b2
g2 = b3 AND b4

Is optimized to (well, at least as far as I can tell)

g3 = b1 AND b2 AND b3 AND b4.

Kind regards,
Robert

Olzhas Rakhimov

unread,
Dec 7, 2015, 3:52:08 PM12/7/15
to scram-dev
Hi,

SCRAM currently does not show intermediate probabilities
or analysis of intermediate gates (events).
SCRAM currently only deduces the roots of a fault tree container
as defined in the OpenPSA MEF.
The current contract with the formular rewriting and analysis steps is
that the deduced roots are analyzed as separate graphs,
and optimizations and approximations don't have side effects on the original fault tree.

Turning off the preprocessing and optimization steps will not give
the intermediate results that you are asking.

I think a proper approach would be multi-rooted analysis,
where the intermediate gates are also treated as analysis targets,
and the common results are preserved and shared for efficiency.
It is in SCRAM's to-do list,
but I think it will take a full release to get this feature implemented.

(1) A hacky and inefficient way to get this feature
is to make SCRAM analyze all gates as if they were separate graphs.

(2) Another solution for the given problem
would be to define analysis targets (gates for analysis) for SCRAM
without relying on the deduction.
However, this will require manual specification of analysis targets
in the configuration file.

The (2) solution is a long term and proper feature that I would vote for.

Let me know if any of these solves your problem,
or if you prefer another method.

Regards,
Olzhas

P.S. The visualization tool sounds cool. Is it on GitHub?
Reply all
Reply to author
Forward
This conversation is locked
You cannot reply and perform actions on locked conversations.
0 new messages