Component-based system reliability subject to positive and negative correlation

Abstract

Traditional system reliability modeling methods assume statistical independence of component failures, while past research emphasizes methods to characterize positive correlation and dependence. However, negative dependence and correlation are also important for system reliability and resilience modeling. This paper develops a method to assess the reliability of a system subject to arbitrary combinations of positive and negative correlation between the failures of its components. The method enables derivation of expressions composed of a vector of unequal component reliabilities and unequal pairwise component correlations. The method is illustrated through a sequence of examples, which conduct sensitivity analysis on correlation parameters of the model as well as quantify the impact of component correlations on system reliability. Coverage experiments formally assess the flexibility and scalability of the approach.