Modeling and Learning Interaction-based Accidents for Safety-Critical Software Systems

Analyzing accidents is a vital exercise in the development of safety-critical software systems to prevent past accidents from reoccurring in the future. Current practices such as causal event analysis are insufficient in light of a growing trend of accidents involving complex interactions between components with and without the occurrence of failures. Furthermore, the reuse of accident knowledge in current practices relies heavily on human expert recall and interpretation. In this paper, we propose an ontological classification mechanism to acquire and reuse knowledge from past accidents that focuses on the interactions taking place in a system. A set of knowledge bases are constructed independently using a feature-based classification and a domain specific ontology to organize the term spaces of each feature. Similarity mechanisms are introduced to retrieve and integrate the acquired knowledge into the new system analyses. Our experiments show how our approach reuses accident knowledge to uncover potential safety concerns in future safety analysis that may otherwise have been incorrectly classified in traditional approaches.