Importance Measures from Complex Reliability Simulations

We consider the problem of estimating the importance of components and parameters in the dynamic simulations of complex systems. We aim at a conceptualization that is capable of retaining the meaning of traditional (static) importance measures in a dynamic concept. We also approach the problem by defining the importance measures in such a way that they draw from the rich output of a dynamic simulator exploiting information that is typically hidden when focusing solely on the probabilistic-level data (e.g., component failure probabilities, system-level failure probability). By incorporating detailed observable information such as failure times and component operational characteristics, additional dimensions of decision making are made available to system designers and operators that allow a focus on the margin to failure. The goal of our work is to create a scalable and flexible approach to enrich the insights from a time- and physics-informed simulation with explanations as to what are the drivers of the system behavior at a fundamental level of system behavior.

Keywords: Importance measures, Reliability analysis, Complex systems simulation.