Input data characterization factors for complex systems affecting availability estimation accuracy

Reliability analysts are often faced with the challenge of characterizing the behavior of system components based on limited data. Insights into which data is most significant and how much data is necessary to achieve desired accuracy requirements would improve the efficiency and cost effectiveness of the data collection and data characterization processes. This research assesses potential significant factors in the probabilistic characterization of component failure and repair behavior with respect to their effect on system availability estimates. Potential factors were screened for significance utilizing a Plackett-Burman experimental design for several system models. Two input data characterization factors were found to have a significant affect on availability estimation accuracy: the size of the system and the number of data points used for component failure and repair distributional fitting. The estimating error was minimized when the structures analyzed were small and many data points (in this case, 25) were used for the distributional fittings. Surprisingly, the assumption of constant component failure rates and the use of empirical repair distributions were found to be equally effective component characterization methods. The results of this study also indicate that there is no apparent benefit in concentrating on 'important' components for the highest fidelity distributional fittings.