Single-fault bayesian algorithm

[ivanaK]

Hi all,

I would like to know whether SMILE.NET provides an option to use single-fault bayesian algorithm, i.e. only one fault can occur at a time in a DiagNetwork.
I could only find the property MultiFaultAlgorithmType of DiagNetwork.
Thanks!

Regards,
Ivana

[shooltz[BayesFusion]]

SMILE.NET will run single-fault algorithm if you select single pursued fault.

[ivanaK]

Could you please explain what exactly does pursuing a fault do to the bayesian network?

What I need in my case is to identify the single fault in the system by giving evidence for the observations that are ranked first in each step, without knowing in advance which is the fault (to pursue).
I am using SMILE.NET for building a "reasoner" to perform diagnostics on specific type of systems.

Thank you in advance!

Regards,
Ivana

[shooltz[BayesFusion]]

What I need in my case is to identify the single fault in the system by giving evidence for the observations that are ranked first in each step, without knowing in advance which is the fault (to pursue).

It looks like you need to pursue all the faults then. The observations are ranked based on the entropy and observation cost; the entropy calculations depend on the selection of targets to pursue.

GeNIe's diagnostic UI selects the most probable fault as initial pursued one. DiagNetwork's constructor does the same.

[ivanaK]

Thank you for your answer.

My doubt now is: If I pursue all the faults, is the bayesian network using a single-fault algorithm? I have to clerify this, because it is important that in my system only one fault can happen at a time.

Thanks!
Ivana

[shooltz[BayesFusion]]

My doubt now is: If I pursue all the faults, is the bayesian network using a single-fault algorithm?

No, when the number of pursued faults is greater than one, then by definition multi-fault algorithm is run.

[ivanaK]

Thank you for the answer again.

It is quite disappointing though. So, the conclusion is: There is no possibility to use single-fault algorithm without pursuing faults?

Regards,
Ivana

[shooltz[BayesFusion]]

There is no possibility to use single-fault algorithm without pursuing faults?

That would be zero-fault algorithm To obtain the entropy SMILE needs to compare the fault posteriors before and after instantiating an observation; the choice of posterior values is directly based on pursued fault(s).