An analysis strategy for large fault trees

In recent years considerable progress has been made on improving the efficiency and accuracy of the fault tree methodology. The majority of fault trees produced to model industrial systems can now be analysed very quickly on PC computers. However there can still be problems with very large fault tree structures such as those developed to model nuclear and aerospace systems. If the fault tree consists of a large number of basic events and gates and many of the events are repeated, possibly several times within the structure, then the processing of the full problem may not be possible. In such circumstances the problem has to be reduced to a manageable size by discarding the less significant failure modes in the qualitative evaluation to produce only the most relevant minimal cut sets and approximations used to obtain the top event probability or frequency. The method proposed uses a combination of analysis options each of which reduces the complexity of the problem. A factorisation technique is first applied which is designed to reduce the ‘noise’ from the tree structure. Wherever possible, events which always appear together in the tree are combined together to create more complex, higher level events. A solution of the now reduced problem can always be expanded back out in terms of the original events. The second stage is to identify independent sections of the fault tree which can be analysed separately. Finally the Binary Decision Diagram (BDD) technique is used to perform the quantification. Careful selection of the ordering applied to the basic events (variables) will again aid the efficiency of the process.