Sensitivity analysis of modular dynamic fault trees

Abstract

Dynamic fault tree analysis, as currently supported by the Galileo software package, provides an effective means for assessing the reliability of embedded computer-based systems. Dynamic fault trees extend traditional fault trees by defining special gates to capture sequential and functional dependency characteristics. A modular approach to the solution of dynamic fault trees effectively applies Binary Decision Diagram (BOD) and Markov model solution techniques to different parts of the dynamic fault tree model. Reliability analysis of a computer-based system tells only part of the story, however. Follow-up questions such as Where are the weak links in the system?, How do the results change if my input parameters change? and What is the most cost effective way to improve reliability? require a sensitivity analysis of the reliability analysis. Sensitivity analysis (often called Importance Analysis) is not a new concept, but the calculation of sensitivity measures within the modular solution methodology for dynamic and static fault trees raises some interesting issues. In this paper we address several of these issues, and present a modular technique for evaluating sensitivity, a single traversal solution to sensitivity analysis for BOD, a simplified methodology for estimating sensitivity for Markov models, and a discussion of the use of sensitivity measures in system design. The sensitivity measures for both the Binary Decision Diagram and Markov approach presented in this paper is implemented in Galileo, a software package for reliability analysis of complex computer-based systems.