Non deterministic Repairable Fault Trees for computing optimal repair strategy

In this paper, the Non deterministic Repairable Fault Tree (NdRFT) formalism is proposed: it allows to model failure modes of complex systems as well as their repair processes. The originality of this formalism with respect to other Fault Tree extensions is that it allows to face repair strategies optimization problems: in an NdRFT model, the decision on whether to start or not a given repair action is non deterministic, so that all the possibilities are left open. The formalism is rather powerful allowing to specify which failure events are observable, whether local repair or global repair can be applied, and the resources needed to start a repair action. The optimal repair strategy can then be computed by solving an optimization problem on a Markov Decision Process (MDP) derived from the NdRFT. A software framework is proposed in order to perform in automatic way the derivation of an MDP from a NdRFT model, and to deal with the solution of the MDP.