Dynamic Bayesian Networks as a Framework for the Reliability Analysis of Systems with Dynamic Dependencies,

Recent works performed by several researchers working in the dependability eld have shown how the formalism of Bayesian Networks (BN) can oer several advantages when analyzing safety-critical systems from the reliability point of view. In particular, when the components of such systems exhibit dynamic dependencies, dynamic extensions of BN can provide a useful framework for the above kind of analysis. In this work, we present an approach where the reliability analysis of systems showing dynamic dependencies is tackled by means of a model based on the formalism of Dynamic Bayesian Networks (DBN). In particular, we aim at modeling the kind of dependencies usually addressed by the Dynamic Fault Tree (DFT) formalism, by providing more sophisticated analysis techniques with respect to DFT. In fact, we show that, by resorting to the use of DBNs, a lot of interesting reliability analyses can be performed on the modeled systems, including prediction of faults (i.e. standard \top event" unreliability analysis), monitoring and diagnosis (explaining observations on some parameters in terms normal/abnormal behavior of components) and smoothing (reconstruction of components' behavior during time, given a stream of observations). Performing such analyses just requires the use of standard inference techniques on DBNs, making them really interesting for a complex reliability analysis of such systems. We show our approach by means of some examples, by providing several diagnostic or predictive measures that can be computed by exploiting a DBN model. This is achieved through the use of Radyban, a software tool developed at the Computer Science Department of the University of Piemonte Orientale, able to translate a DFT model into a DBN and nally to perform the required analysis.