Unavailability evaluation and allocation at the design stage for electric power plants: Methods and tools

Electricite de France is currently carrying out a project called CIDEM with the objective of integrating availability, operational feedback, and maintenance in the design of future power plants (especially nuclear power plants) in order to improve their profitability. The work reported in this paper was performed in the framework of the research part of the CIDEM project, managed by the R&D division of EDF. The paper shows that the availability assessment of an electric power plant raises a number of specific modeling problems. These problems are especially acute in the case of nuclear plants, for which safety procedures can affect availability. In fact, a dynamic simulation model could easily take into account all the particular features of the plant operation. But the quantification of such models (which are not Markovian) necessarily relies on Monte-Carlo simulation, and thus is rather slow. Computation times could still be acceptable for evaluation purposes. But in the design stage, we need to allocate the global objectives (in terms of availability, costs) to the main functions and/or components of the plant. If the system to be studied is a bit more complex than a simple series assembly of components (which means that the sum of the components' unavailabilitites is a fairly good approximation of the global unavailability), doing an allocation requires numerous evaluations. This makes the use of a simulation model totally unthinkable. This is why we have chosen to use only fault-tree models, in spite of the fact that they are essentially static models: they can be calculated in very short times, especially with the new generation of fault-tree processing codes, based on BDDs (Binary Decision Diagrams). The paper gives the modeling schemes we had to devise in order to take into account various dynamic features, along with an estimation of the corresponding errors. It also gives a quick description of the tools we use to carry out real studies : the FIGARO workbench, which enables the building of knowledge bases to automate the fault-tree construction, and the ARPO tool, to perform allocation, with two different methods.