The POS Model for Common Cause Failure Quantification

Abstract

The POS (Process-Oriented Simulation) model can be seen as an extension of the BFR (Binominal Failure Rate) model carried out to overcome some of the shortcomings of this approach. In contrast to the BFR model the POS model explicitly distinguishes between immediate and delayed failures of components, does not assume that in each CCF (Common Cause Failure) event all components share the CC (Common Cause) but assigns probabilities to the different degrees of “extension” of the cause and is based on clearly formulated stochastic assumptions. This broader approach implies a more complex structure, however, the price paid in terms of greater complexity pays off due to a larger range of applicability. In particular the POS model can be applied to systems with high degree of redundancy. Another important advantage is the possibility to generate failure events using the POS simulation code. Simulating with known model parameters and estimating these from the generated failure events offers the possibility to test the framework for parameter estimation and, in addition, to obtain the uncertainties of the model results. No further assumptions beyond the model itself and the estimation procedure need to be introduced. Last but not least, the test of the estimation procedure demonstrated that the POS model works already quite well with only a small number of events as a basis. This makes it a candidate for plant specific CCF applications. In contrast to other approaches like MGL (Multiple Greek Letter) and Alpha Factor, the POS model comprises only a few parameters, beyond the frequency of CCEs (Common Cause Events) only three. The applications of the POS model carried out so far indicate that it can readily be applied to practical cases. The results obtained by benchmark tests are no outliers but are rather consistent with the majority of existing analyses. To summarize, the POS model represents an interesting alternative to established models with some new and unique model features. Looking ahead, two aspects are discussed and emphasized. The first is the still existing development potential of the model, the second is a proposal for the quantitative comparison between CCF models.