Conflict resolution and collaborative fault detection using stochastic dynamic programming

This paper presents a framework based on stochastic dynamic programming that facilitates the implementation of collaborative fault detection through conflict resolution. A conflict arises when two fault detectors draw opposing conclusions regarding the presence of a fault. We use stochastic dynamic programming to optimally resolve conflicts and to control data gathering actions that can improve decision-making. Since stochastic dynamic programming suffers from the curse of dimensionality, we also present and evaluate an approximate dynamic programming (ADP) approach based on decomposition of states, solving decomposed MDPs, and recombination of value functions. A spacecraft fault detection example is included to demonstrate the implementation of the proposed framework and of the corresponding ADP approach.