Numerical Reproducibility of Human Body Model Crash Simulations
Paper in proceeding, 2021
quantifying the variation in model predictions for diverse computer systems at different sites and settings.
Repeated simulations, with varying number of Central Processing Unit (CPU) cores and model decomposition, of
four high severity load cases – a full frontal, near-side frontal oblique and side impact with a full set of driver
restraints, as well as a full frontal with a seat belt only restraint – was carried out on five computer systems. HBM
responses were found to vary randomly with the Number of CPU cores (NCPU), but not due to different hardware
or message parsing interface software at each computer system used. Implemented HBM updates reduced the
variation in the near-side frontal oblique load case. When the NCPU used was fixed, identical results were
obtained from all computer systems. This means the variation of HBM responses is due to the model
decomposition. It is possible to quantify the numerical reproducibility of an FE HBM by repeated simulations,
varying the NCPU and analyzing the coefficient of variation of the responses.
Human Body Model
Virtual Testing
Reproducibility
Finite Element
Author
Jonas Östh
Chalmers, Mechanics and Maritime Sciences (M2), Vehicle Safety
Bengt Pipkorn
Chalmers, Mechanics and Maritime Sciences (M2), Vehicle Safety
Jimmy Forsberg
Dynamore Nordic AB
Johan Iraeus
Chalmers, Mechanics and Maritime Sciences (M2), Vehicle Safety
Conference proceedings International Research Council on the Biomechanics of Injury, IRCOBI
22353151 (ISSN)
Vol. 2021-September 431-443 IRC-21-51Online, ,
Subject Categories
Computer Engineering
Applied Mechanics
Vehicle Engineering
Areas of Advance
Transport
Infrastructure
C3SE (Chalmers Centre for Computational Science and Engineering)