Distributed model-based fault diagnosis with stochastic uncertainties

This paper proposes a novel distributed fault detection and isolation approach for the monitoring of non linear large-scale systems. The proposed architecture considers stochastic characterization of the measurement noises and modeling uncertainties, computing at each step stochastic time-varying thresholds with guaranteed false alarms probability levels. The convergence properties of the distributed estimation are demonstrated. A novel fault isolation method is proposed basing on a Generalized Observer Scheme, providing guaranteed error probabilities of the fault exclusion task. A consensus approach is used for the estimation of variables shared among more than one subsystem; a method is proposed to define the time-varying consensus weights in order to minimize at each step the variance of the uncertainty of the fault detection and isolation thresholds. Detectability and isolability conditions are provided.