Variable ordering heuristics for binary decision diagrams

Fault tree analysis, FTA, is one of the most commonly used techniques for safety system assessment. Over the past five years the Binary Decision Diagram (BDD) methodology has been introduced which significantly aids the analysis of the fault tree diagram. The approach has been shown to improve both the efficiency of determining the minimal cut sets of the fault tree, and also the accuracy of the calculation procedure used to quantifY the top event parameters. To utilise the BDD technique the fault tree structure needs to be converted into the BDD format. Converting the fault tree is relatively straightforward but requires the basic events of the tree to be placed in an ordering. The ordering of the basic events is -critical to the resulting size of the BDD, and ultimately affects the performance and benefits of this technique. There are a number of variable ordering heuristics in the literature, however the performance of each depends on the tree structure being analysed. These heuristic approaches do not always yield a minimal BDD structure for all trees, some approaches generate orderings that are better for some trees but worse for others. Within this thesis three pattern recognition approaches, that of machine learning classifier systems, multi-layer perceptron networks and radial basis function neural networks, have been investigated to try and select a variable ordering heuristic for a given fault tree from a set of alternatives. In addition a completely new heuristic based on component structural importance measures has been suggested with significant improvement in producing the smallest BDD over those methods currently in the literature.