Methodology for the Evaluation of Human Response Variability to Intrinsic and Extrinsic Factors Including Uncertainties

Author: ORCID icon
Perez Rapela, Daniel, Mechanical and Aerospace Engineering - School of Engineering and Applied Science, University of Virginia
Crandall, Jeff, Mechanical and Aerospace Engineering, University of Virginia

The use of standardized anthropomorphic test devices and test conditions prevent current vehicle development and safety assessments from capturing the breadth of variability inherent in real-world occupant responses. The central idea of this dissertation is that human body models used in simulations with a diverse range of real-world impact scenarios can represent population variability and may be the key to overcome the limitations of current vehicle assessment and development methodologies. In this approach, a series of response surfaces are created that contain information about the occupant responses as a function of different input variables. Subsequently, these surfaces, in conjunction with real-world distributions of the population and impact conditions, can be used to identify populations at risk, to illustrate injurious impact scenarios, and to inform prioritization of countermeasure and design actions.
This dissertation develops a methodology to assess occupant response that accounts for sources of intrinsic (human-related) and extrinsic (non-human-related) variability, including uncertainty in the FE parameters. Although inherently generic in nature, this methodology was applied to a far-side crash scenario in order to provide an illustrative example.
For the far-side application, lateral head excursion and thoracic injury were identified as the target occupant responses, while change in vehicle velocity, impact direction and seatbelt load limiter were the extrinsic factors explored. The intrinsic factors were occupant height, weight and waist circumference and were explored by morphing the simplified GHBMC human body model. WorldSID tests were used in order to validate and estimate the parameter uncertainty in the vehicle FE model. Five regression techniques, namely, linear regression, logistic regression, LASSO linear and logistic regression, and Neural Networks (NN), were used for the generation of the response surfaces. The regression models were sequentially trained to represent the maximum lateral head excursion and the probability of 3+ fractured ribs using a total of 405 FE simulation results. The performance of these regression techniques was assessed based on their ability to predict out-of-sample datapoints. The NN showed equal or improved performance with respect to the other regression techniques.
Based on far-side input conditions derived from US field data, Monte-Carlo simulations used the head excursion and rib fracture response surfaces to calculate the probability of head-to-intruding-door impacts and cases with 3+ fractured ribs. In addition, the Monte-Carlo analysis predicted head contact and rib fracture reductions subsequent to design changes in the restraint configuration. This analysis indicated that the vehicle used in this study would lead to a range of 667 to 2,448 head-to-intruding-door impacts and a range of 2,893 to 3,783 cases of 3+ fractured ribs, depending on the seatbelt load limiter. In the US field data, the expected number of cases with 3+ fractured ribs was 3,958. The far-side assessment illustrates how the methodology incorporates the intrinsic and extrinsic variability, generates response surfaces that characterize the effects of the variability, and ultimately permits vehicle design considerations and injury predictions appropriate for real-world field conditions.

PHD (Doctor of Philosophy)
Biomechanics, Injury, Neural Networks, Response Surfaces, Stochastic Methods, Variability, Uncertainty, Vehicle Safety, PMHS Tests, Human Body Models, WorldSID, Far Side
All rights reserved (no additional license for public reuse)
Issued Date: