Dynamic Event Fault Tree (DEFT): A methodology for probabilistic risk assessment of computer-based systems.

摘要

The design of systems intended for use in critical applications where failure consequences are high is supported by extensive analysis of potential hazards and their consequences. Probabilistic Risk Assessment (PRA) is a widely used systematic and comprehensive methodology for such analysis, especially in the nuclear industry. PRA identifies and evaluates risks associated with complex technological systems and thus can help improve the system's safety and performance. The use of computer-based systems for a wide variety of applications is continuously increasing. For example, many existing systems for instrumentation and control in nuclear applications are undergoing an upgrade from mechanical or manual to computer-based technology. Computer-based systems are adaptable and flexible, which makes their use desirable; however, this also creates complexities and introduces dependencies in their analysis of failure scenarios and risks. Current PRA techniques, however, are not appropriate for computer-based systems, which makes the PRA of such upgrades and new system difficult.; However, fault tree analysis, which is a constituent part of PRA, has been extended to dynamic fault tree in reliability modeling and analysis field. A dynamic fault tree uses special gates to capture and analyze dynamic behaviors or dependencies in computer-based systems. A new methodology, called DEFT, is presented in this dissertation for probabilistic risk assessment of computer-based systems. DEFT models a set of accident scenarios using an event tree structure, and models the pivotal events using dynamic fault tree structures. Besides, DEFT integrates six common probabilistic risk assessment techniques into the event tree/dynamic fault tree model. These common probabilistic risk assessment techniques are imperfect coverage model, phased mission system analysis, sensitivity analysis, diagnostic analysis, common cause failure analysis and uncertainty analysis. These incorporations make DEFT more complicated and more robust. A key process in DEFT is an enhanced modularization algorithm, MULFtree, which successfully handles the dependencies across multiple dynamic fault trees and also significantly reduces the complexity from both modeling and computing processes. Analysis of several representative systems demonstrates the capabilities of DEFT.