Publikation

Dependency Graph Modularization for a Scalable Safety and Security Analysis

Rhea Rinaldo, Dieter Hutter

In: Maria Chiara Leva, Edoardo Patelli, Luca Podofillini, Simon Wilson (Hrsg.). 32nd European Safety and Reliability Conference. European Safety and Reliability Conference (ESREL-2022) August 28-September 1 Dublin Ireland Research Publishing Singapore 2022.

Abstrakt

Due to the steady development of automated and autonomous vehicles, a growing increase in the amount and complexity of the vehicle’s internal components and their safety and security requirements can be registered. Various assessment techniques for the posed safety and security requirements exist; however, some of the applied techniques become insufficient of modeling this increased complexity, or the evaluation effort increases heavily. Consequently, existing approaches need to be revised and new approaches, tailored to this use case, developed. Prior to this work, we combined an analytical approach called ERIS and a numerical approach named AT-CARS to a hybrid to reduce the overall complexity of the model through simulation while obtaining realistic results, especially for versatile and sophisticated subsystems such as AI computing nodes. Thereby, the main system is modeled in ERIS graphically as a Dependency Graph and dedicated system parts are constituted to subsystems and outsourced to AT-CARS. Although encouraging results could be achieved and modularization properties could be obtained, it was discovered that an analytic evaluation of the subsystems would be more beneficial for specific system structures. Consequently, we focus this paper on exploring the system modularization further and view it regarding the recursive analytical evaluation. Therefore we firstly establish the formal basis of abstraction and modularization of dependency graphs, followed by an adapted evaluation process. Based on this we discuss the impact of different component dependencies and provide criteria for a well-formed modularization. To show the efficiency and the benefit of this addition for the future evaluation of critical and complex systems, we apply the modularization scheme on an abstracted but realistic model of an autonomous vehicle.

Projekte

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence