Citation

Plaschkies F, Possoli K, Vaculin O, Schumacher A, de Andrade P. 27th International Technical Conference on the Enhanced Safety of Vehicles (ESV); April 3-6, 2023; Abstract #: 23-0055, pp. 13p. Washington, DC USA: US National Highway Traffic Safety Administration, 2023 open access.

Copyright

(Copyright © 2023 open access, US National Highway Traffic Safety Administration)

Abstract

27th International Technical Conference on the Enhanced Safety of Vehicles (ESV): Enhanced and Equitable Vehicle Safety for All: Toward the Next 50 Years

https://www-esv.nhtsa.dot.gov/Proceedings/27/27ESV-000055.pdf

The systems for occupant protection in passive vehicle safety are primarily developed with single statistical representations of humans, so-called Anthropomorphic Test Devices (ATDs). Unfortunately, those ATDs cover additional features like age and body shape insufficiently during development. Augmenting finite element simulations with a metamodel trained by machine learning is promising to overcome this barrier. However, the database design, the machine learning architecture, and the requirements for quality and robustness influence each other. Therefore, objective criteria must be defined to compare the alternatives taking cost and benefit aspects under changing preferences into account. Having complex criteria can be framed as a multi-attribute decisionmaking problem. This paper's objective is the development of a transparent assessment scheme for virtual statistical simulation for rapid vehicle occupant safety assessment using supervised learning. PROMETHEE is selected as an appropriate decision-making approach. A process, consisting of a sequential definition of the criteria leading to the final assessment, is proposed to adapt the method in this paper's domain. The methodology is tested on sample alternatives, generated using a calibration-type machine learning architecture and data from finite element simulations. The original PROMETHEE algorithm cannot handle a vast number of alternatives. Since, typically, numerous alternatives occur during the development of a machine learning application, a sorting-based modification is implemented. Finally, the findings are discussed, and recommendations for related use cases are given. The proposed method seems applicable to the described domain and near-related ones. Moreover, multiple tendencies between an alternative's parameters and rank can be identified in the test samples.

Language: en