An observation metamodel for dependability tools

FaultFlow is a library for modeling and evaluation of dependability of component-based systems. It represents duration to occurrence and propagation of faults across the hierarchy of components through non-Markovian distributions, facilitating fitting of observed data and design assumptions. Additionally, FaultFlow can be extended to simulate the system behavior and generate synthetic time series encoding occurrences of faults and failures and results of diagnostic tests. Time series can in turn be employed to train and test data-driven methods aimed at various tasks, notably failure prediction. As a first step in this direction, we define a flexible and extensible observation metamodel for FaultFlow, representing type and time of observations of the system behavior, and facilitating definition of monitoring policies.