Initial Steps Towards a Multi-Fidelity Surrogate Model to Evaluate Occupant Safety in the Aeronautic Preliminary Design

Kurzfassung

The adoption of novel high-efficiency aircraft fuselages presents unique safety challenges. Unlike traditional aircraft, their behaviour during “emergency landing” should be studied with an integrated global approach that considers the entire cabin’s structural response, similar to the approach prescribed for general aviation aircraft. This work presents a numerical method for the efficient evaluation of occupant safety during the aircraft preliminary design stage, to be integrated into the process chain for aircraft design established at the German Aerospace Center (DLR). In particular, the aim is to enhance the internal PANDORA design environment, which allows the structural evaluation of aircraft concepts, with the capability to estimate occupant survivability. This preliminary work covers the development of a multi-fidelity seat and dummy model, which has a primary safety role, together with a first simple AI surrogate of the low-fidelity seat and dummy assembly. A high-fidelity Finite Element (FE) seat model is generated based on a certified aircraft seat, which is 3D-scanned and experimentally tested to validate key structural parts relevant to occupant safety criteria prescribed by the regulation. Then, a simplified FE model is derived from the validated high-fidelity model to replicate its inertia and stiffness at a significantly lower cost, to be used in PANDORA-generated full-scale aircraft models. The multi-fidelity surrogate model approach is designed to replace high-fidelity occupant and seat models in full-scale fuselage analysis, with their simplified representation to boost model robustness and computational efficiency. The injury criteria are evaluated by an AI surrogate trained on a small-scale high-fidelity model of the seat and dummies, which correlates the fuselage floor kinematics with the resulting injury levels for each occupant. A first correlation is shown, using simple kinematic approximations and a low-fidelity FE model for training.