Future Work - allow multiple failures on the same shop

[christianadriano]

Currently, one can only have one failure per shop. Event when there is failure propagation, we assume that affected components did not enter a failure state, hence we only need to fix the root-cause component.

However, in the future, we might want to relax this assumption by allowing components actually fail and, hence, need to be repaired.

• How to detect the phenomenon of multi-failure?
• How to identify all the failing components (some combination of symbolic and probabilistic rules that determine multi-graph-style dependencies)?
• How to mitigate the effect of multi-failures (inference over a Bayesian network or use some form or relational learning/Markov networks)?

[jocodeone]

As far as I understood the problem: some of the observations are failing and some only seem to be failing. If that is correct:

• What is the difference between a component that fails and a component that seems to fail but is not failing?