7. System Failure Analysis: Fault Trees
Society for Industrial and Applied Mathematics eBooks · 1998
7.1. System failure analysis. Network reliability is based on an abstract graphical representation of a system. It is basically success oriented. In practice, as it turns out, it is best to be failure oriented. Once this position is taken, the number of events that can be thought about that could cause or result in system failure becomes very large. Hence it is necessary to set spatial and temporal bounds on the system. It is necessary to determine only those events absolutely relevant to the analysis. A fault tree (or logic tree) is often the best device for deducing how a major system failure event could possibly occur. Since the network graph is close to a system functional representation, it cannot capture abstract system failure and human-error events as well as the fault tree representation. The fault tree is a special case of a class of diagram representations called influence diagrams, developed in Chapter 9. In sections 7.2 and 7.3 we develop the background necessary for constructing and analyzing fault tree diagrams. However, their construction first depends on a thorough understanding of the system and the results of a system inductive analysis. Simulation versus fault tree analysis. Currently, simulation is perhaps more popular than fault tree analysis. Faster and faster computers make simulation very attractive. It also has the advantage that the dynamics of system operation can be studied, whereas the fault tree is essentially a static system representation. However, like the network representation, simulation tends to be success oriented. Possible but rare fault events may be overlooked or inadequately treated in simulation. Hence we still recommend that the fault tree approach be used first, followed by simulation. Fault tree has the advantage of concentrating attention on learning in depth about the system of interest. System inductive analysis. Initially we need to perform a system inductive analysis before we construct networks, fault trees, influence diagrams or employ simulation.