| Reliability evaluation tools are becoming indispensable tools for modeling and analyzing (critical) computer-based systems. However, the growing complexity of such systems calls for increasing sophistication of these tools. Reliability evaluation tools need to not only capture the complex dynamic behavior of the system components, but they must also be easy to use, intuitive, and computationally efficient.; Dynamic system components exhibit complex behaviors and interactions, making combinatorial models inappropriate to solve them. Markov chain (and their extensions) based reliability evaluation tools have been widely used to model such systems. However, their limited modeling capabilities and the infamous state space explosion problem restrain their application.; Currently existing tools/formalisms have, in general, a number of shortcomings with respect to the lack of modeling power and/or analysis capabilities. For instance, many tools are incapable of efficiently handling general component failure distributions or model complex interactions between components. Ineffectiveness in solving large models and intractable model solutions are also problems common to many tools.; Our approach is to investigate the Bayesian networks (BN) formalism and develop a reliability framework for modeling/analyzing dynamic computer-based systems. We have defined a temporal Bayesian network (TBN) reliability framework both in discrete-time and in continuous-time. Various solution techniques can be applied on a TBN model including (1) an approximate solution using a standard BN inference algorithm, (2) an analytical closed-form solution, and (3) a stochastic simulation solution.; The TBN framework defines a (expandable) set of 'basic' BN constructs having well-defined semantics. Each construct captures a specific system component behavior or specific interaction between components. Combining the various 'basic' BN constructs enables the user to construct the system model in a structured, modular, and hierarchical fashion.; We have successfully applied the TBN reliability framework to model/analyze five systems and illustrated how various components' behaviors and interactions can be modeled using our TBN framework. In addition to reliability analysis, we showed how to perform sensitivity, uncertainty, and diagnostic analyses on the system models. |
