Agent-based discovery of unexpected falsification scenarios in aircraft systems virtual testing

Innovative concepts for aircraft systems with the aim to satisfy current aviation emission goals include novel technologies, such as fuel cell systems. However, due to the lack of experience for such technologies in aviation the implicated requirements for the system design may not be fully understood or unknown. This creates a risk for unexpected “blind spots” in the system design that will at best be found in late testing stages. Virtual testing based on digital simulation models is one possible approach to conduct verification and validation much earlier during design phases, but additional methods are needed to find the unexpected scenarios. This work introduces a methodical framework that aims to discover such unexpected test scenarios by means of heterogeneous machine learning test agents. The agents are executed in parallel and exchange falsification scenario information to improve the search for unexpected scenarios that falsify a given system property. The concept is applied and evaluated on a virtual system example for a fuel-cell-driven electric propulsion engine.

DDC Class

620: Engineering

690: Building, Construction

Options

Agent-based discovery of unexpected falsification scenarios in aircraft systems virtual testing