Handling Uncertainity in Error Propagation Analysis

Abstract

Error propagation analysis (EPA) is a systematic model-based approach to assess the impact of incidental or malicious faults in the dependability and security analysis of complex systems. Its main purpose is to estimate the most severe failures in the system under evaluation. It can be extended to evaluate the efficiency of built-in error protection and mitigation mechanisms. However, during the EPA, uncertainties may arise, which may affect the outcome - this way, the validity of the analysis - and lead to escaping faults. Uncertainties can originate from two primary sources. Firstly, epistemic-type uncertainties express that there may be parts of the analyzed system that are unknown to the domain expert. Secondly, aleatory uncertainties may arise from incorrect or incomplete modeling of the system or even from the non-deterministic operation (physical processes). Our approach extends the known EPA models by handling the uncertainties via rough set theory, an advanced mathematical paradigm to generate approximate descriptions of the system behavior.