Bridging model-based design for dependability and empirical observations

Model-driven architecture (MDA) based paradigms generate a complete specification of the target system in addition to its implementation. Such a specification facilitates the formal analysis and prediction of the characteristics of the target system. The prerequisite of the analysis of nonfunctional properties is the extension of the model by the corresponding component and interaction characteristics, like component fault models and error propagation via their interaction for dependability estimation. These extensions rely on empirical observations from past similar designs. This way, accuracy of the predicted characteristics of future applications depends on the faithfulness and completeness of the extracted and generalized features, like component faults anticipated. The current paper proposes a novel approach for checking the extraction and generalization of fault mechanisms from historical observation data. Its core idea is the deduction of all faults potentially compatible with an observed log data and a subsequent comparison with the anticipated set of faults to be applied in the dependability model of future applications.