(D-157) Predicting Occupant Head Acceleration in Near and Far-Side Lateral Collisions Using Multivariable Regression Models

Introduction::

The cervical spine and head of occupants in motor vehicle collisions are commonly injured during lateral impacts. The peak lateral acceleration (PLA) of the head is often utilized to predict the severity of injury to the head and neck. While real-world head acceleration data is limited, a vehicle’s PLA and lateral Delta-V (ΔV) are frequently obtained from event data recorders [1].

During near-side lateral collisions, the occupant’s head may contact the adjacent window, frame, or B-pillar. Even in low-speed lateral collisions with vehicle ΔV under 10 km/h, occupants can experience head, neck, and limb pain from contacting the vehicle’s interior [2]. In far-side lateral impacts, the occupant’s head and torso will rotate more since no belt or interior components are present to prevent excessive motion [3].

Prior models have determined relationships for head PLA with vehicle PLA or vehicle ΔV individually [3,4]. However, differences in experimental techniques have made it difficult to associate data between studies. Vehicle PLA and ΔV are not always linearly correlated during a collision sequence and do not always predict the same head PLA. Therefore, head PLA may be more accurately predicted through multivariable regression analyses utilizing both vehicle PLA and ΔV.

This work aims to use vehicle PLA and ΔV as predictor variables in a multivariable regression to calculate head PLA. The ultimate goal of this study is to aid in the investigation and determination of head and cervical spine injury potential from near and far-side lateral impact sequences.

Materials and Methods::

Ninety-eight near-side [2, 5] and 61 far-side collisions [1, 2] were utilized for the analysis. Occupant head accelerations were limited to human volunteers and anthropomorphic test devices (ATD). Volunteers were both male and female, between 15 and 62 years of age. The subjects’ heights were between 155 cm and 188 cm, and their weights were between 56 kg and 112 kg. Subjects were seated in the driver, passenger, or rear seats in a vehicle or a test sled. Data included volunteers who were unexpecting or expecting the collision and ATDs that were pre-tensioned or normally tensioned to simulate expectant or unexpectant muscle responses. Data included subjects who used a lap or three-point belt.

MATLAB was used to create a multivariable linear regression, modeling head PLA as a linear combination of vehicle PLA and ΔV [6, 7]. Separate regressions were created for near-side and far-side collisions. Regression coefficients were evaluated by p-values for statistical significance; the null hypothesis was that the regression coefficients for vehicle PLA and ΔV were equal to zero (α = 0.05). The model was evaluated by the adjusted coefficient of determination (R²). The constant term was forced to be 0 g because a vehicle PLA and ΔV of 0 g and 0 km/h produce a head PLA of 0 g.

Results, Conclusions, and Discussions::

Regression results from the near-side collisions are shown in Figure 1 (N = 98, R² = 0.44). Regression coefficients were 0.51 (p = 0.28) for vehicle PLA and 2.87 (p < 0.01) for vehicle ΔV. These coefficients indicate that vehicle ΔV was a significant predictor of head PLA, but vehicle PLA was not. Head contact with the vehicle interior components drastically increases the measured head PLA and should not be omitted from the analysis. Omitting this data can lower the measured head PLA, decreasing this model’s applicability. These data were included to create the model since head contact with the interior components is common and is an additional indicator of head injury.

The far-side collision regression is shown in Figure 2 (N = 61, R² = 0.23). Regression coefficients were 0.07 (p < 0.01) for vehicle PLA and 0.23 (p < 0.01) for vehicle ΔV. These coefficients indicate that both the vehicle PLA and vehicle ΔV were significant predictors of head PLA.

Minor variations in experimental variables can account for differences in measured head accelerations. ATD pre-tensioning and instructing volunteers to be expectant of a collision could decrease the measured head PLA. Additionally, impacts at different principal directions of force could influence lateral head acceleration. Any collision not perpendicular to the sides of the vehicle or sled could induce an anterior or posterior acceleration, meaning that the recorded lateral acceleration would be lower in magnitude.

Our regression analyses revealed that utilizing vehicle acceleration and Delta-V data recorded in controlled vehicle collisions can provide valuable insight into the magnitude of an occupant’s head acceleration during near or far-side lateral impacts. This model can aid biomechanists, safety engineers, and forensic consultants in predicting head and neck injuries when occupant kinematics cannot be directly measured. Additional sized occupants, vehicles, and collision magnitudes should be included to improve future regressions. Although more complex, higher order multivariable regressions could improve the model's fit to estimate head PLA, especially at vehicle ΔV under 5 km/h.

Acknowledgements (Optional): :



References (Optional): :

[1] P. Shibata et al., “The Kinematic Analysis of Occupant Excursions and Accelerations during Staged Low Speed Far-Side Lateral Vehicle-to-Vehicle Impacts,” in SAE Technical Paper Series, Apr. 2019. Accessed: Jul. 12, 2023. Available: https://dx.doi.org/10.4271/2019-01-1030.

[2] T. F. Fugger, B. C. Randles, J. L. Wobrock, J. B. Welcher, D. P. Voss, and J. J. Eubanks, “Human Occupant Kinematics in Low Speed Side Impacts,” in SAE Technical Paper Series, Mar. 2002. Available: https://dx.doi.org/10.4271/2002-01-0020.

[3] C. Furbish, J. Welcher, J. Brink, B. Jones, S. Swinford, and R. Anderson, “Occupant Kinematics and Loading in Low Speed Lateral Impacts,” SAE International Journal of Advances and Current Practices in Mobility, vol. 1, no. 4, pp. 1470–1490, Apr. 2019, doi: 10.4271/2019-01-1027.

[4] F. A. Pintar et al., “Comparison of PMHS, WorldSID, and THOR-NT Responses in Simulated Far Side Impact,” in SAE Technical Paper Series, Oct. 2007. Available: https://dx.doi.org/10.4271/2007-22-0014.

[5] NHTSA, “Vehicle Crash Test Database,” National Highway Traffic Safety Administration, 2023. https://www.nhtsa.gov/research-data/research-testing-databases#/vehicle.

[6] The MathWorks, Inc. MATLAB version: 9.14.0 (R2023a). Feb. 2023. Available: https://www.mathworks.com.

[7] The MathWorks, Inc. Statistics and Machine Learning Toolbox: 12.5 (R2023a). Feb. 2023. Available: https://www.mathworks.com.