Generic Crash Pulses Representing Future Accident Scenarios of Highly Automated Vehicles
- Patrick Höschele - Graz University of Technology, Austria ,
- Stefan Smit - Graz University of Technology, Austria ,
- Ernst Tomasch - Graz University of Technology, Austria ,
- Martin Östling - Autoliv Research, Sweden ,
- Krystoffer Mroz - Autoliv Research, Sweden ,
- Corina Klug - Graz University of Technology, Austria
ISSN: 2327-5626, e-ISSN: 2327-5634
Published April 01, 2022 by SAE International in United States
Citation: Höschele, P., Smit, S., Tomasch, E., Östling, M. et al., "Generic Crash Pulses Representing Future Accident Scenarios of Highly Automated Vehicles," SAE Int. J. Trans. Safety 10(2):2022, https://doi.org/10.4271/09-10-02-0010.
Assessment of crashworthiness of autonomous vehicles (AVs) must be carried out for future crash scenarios, as not all crashes will be avoidable. Representative crash pulses for AVs are needed to evaluate conceptual design restraint systems of those vehicles.
Within this study, generic crash pulses for crash scenarios expected to be relevant for AVs were generated based on a set of vehicle-to-vehicle structure simulations with current European sedan cars. These crash scenarios included one Straight Crossing Path (SCP) and two Left Turn Across Path Opposite Direction (LTAP OD) scenarios with varying initial velocities and weight ratios of the crash opponents to obtain different crash configurations. Additionally, full-width frontal simulations with 40 kph and 56 kph were included as a reference. The acceleration signals obtained from the individual simulations were approximated by Legendre polynomials. A prediction model was created for each crash configuration based on a set of three to four pulses to provide a generic crash pulse for each investigated crash configuration. The prediction quality of the model was tested against simulations with freely available finite element (FE) models.
A number of 20 or 60 basis functions of Legendre polynomials was needed to approximate the obtained velocity or acceleration signals with a correlation of at least 0.9. The obtained prediction model for each crash configuration was able to provide good predictions for heading direction when simulations were within the training data. Approximating velocities and differentiation to derive acceleration turned out to be the better option than approximating the acceleration directly. Eventually, five crash configurations were proposed for future investigations. The derived generic crash pulses capture a range of loading severities, directions, and stiffness of current heavy sedan cars. The pulses can be used for sled simulations to evaluate restraint systems in conceptual studies when no vehicle-specific crash pulses are available.