An energy loss-based vehicular injury severity model
Introduction
The prevention of traffic crashes requires an in-depth analysis of the severity of crashes. The term ‘severity’ in the traffic safety literature may refer to the probability of crashes, the damage magnitude, or both (Shelby, 2011). The degree of crash harm depends on the relationship between physical injuries and crash mechanisms, but understanding is often limited by complicated crash mechanics (Carlson, 1979). Therefore, statistical models, which exploit previous crash data, are used to predict injury severity with several explanatory variables. For example, factors in most vehicular injury analyses include crash mechanical properties, driver characteristics (such as age, gender, and alcohol or drug use), safety equipment, vehicle use, and surrounding conditions (Massie et al., 1995, Abdel-Aty et al., 1998, Abdel-Aty et al., 2011, Yau, 2004, Chang and Wang, 2006, Keall et al., 2004, Smink et al., 2005).
There is no consensus on how to evaluate severity. To assess multiple discrete severity outcomes, the Abbreviated Injury Score (AIS) and Maximum Abbreviated Injury Score (MAIS) are frequently used to classify the extent of injuries (Carlson and Kaplan, 1975, Carlson, 1977, Augenstein et al., 2003, Digges and Dalmotas, 2001, Conroy et al., 2008). They have seven classifications in general: 0 – Property Damage Only (PDO), 1 – Minor Injury, 2 – Moderate Injury, 3 – Serious Injury, 4 – Severe Injury, 5 – Major Injury, and 6+ – Fatality (Untreatable cases). Similarly, there are also some indices like Injury Severity Score (ISS) evaluating one injury severity score derived from three different body regions (Greenspan et al., 1985, Ehrlich et al., 2006). This paper uses MAIS for the severity evaluation, as it was recorded per single vehicle and was commonly provided by the datasets employed here.
To specify the crash severity, Mackay (1968), applying momentum conservation theory, first defined Delta-V as the change of speed before and after the collision. The probability and magnitude of injuries and fatalities are often based on the Delta-V of involved vehicles. Delta-V has been applied for the measurement of vehicular crash severity for several decades (Joksch, 1993, Roberts and Compton, 1993, Evans, 1994). Evans (1994) evaluated this model as one of the best analytical methods when real crash testing is unavailable. The fundamental mechanical meaning of Delta-V is related to the forces that occupants receive inside the crashing vehicles (Carlson, 1979). In reaction to the imposed force after the collision, occupants strike the surfaces in the front of the vehicle, which is the main reason for injuries in most cases. However, the recent application of Delta-V depends on four crash modes (Front, Near side, Far side, and Rear detected by the event data recorder), which may not be comparable so that different directions yield different conclusions (Andricevic et al., 2018). Delta-V fails to consider the influence of long crash pulse on injury (Ydenius, 2010, Tsoi and Gabler, 2015). The limitations of Delta-V have prompted the discussion of alternative crash indicators.
Injury severity studies typically establish the statistical relationship between the dependent variable, injury severity, and several independent variables. Ordered logistic regression (OLR) models and their variants, with different levels of severity assessments, perform well (Yasmin et al., 2015, Ferreira et al., 2017, Quddus et al., 2002, Duncan et al., 1998, Kockelman and Kweon, 2002, Ye and Lord, 2014, Bogue et al., 2017, Abay et al., 2013). Table 1 summarizes severity indices, crash indicators, and regression models in previous automobile crash severity studies.
However, few logistic models have been estimated to explain injury severity due to impact energy. This study uncovers the correlation between the injury severity of occupant and a multi-directional energy-related predictor.
This paper considers dynamic crash mechanisms and then proposes an energy loss-based vehicular injury severity (ELVIS) evaluation indicator. The model is first calibrated by a dataset with separate two types of collision, considering the longitudinal and lateral impacts simultaneously. It aims to include both magnitude and direction elements in its expression. By comparing with the Delta-V, we explain the occupant injury severity due to energy absorption in a specific collision. Additionally, because of the discrete and ordinal nature of the dependent variable, the ordered logistic regression model is used to estimate model coefficients. The rest of the paper is structured as followed: Section 2 introduces how data get processed in this study, and Section 3 provides a theoretical framework for the ELVIS function. In Section 4, we evaluate the regression model by the calibrated parameter and different probability distributions; meanwhile, comparing the Delta-V model with the new model with their statistical fitness and predicted accuracy. The discussion (Section 5) interprets the findings from the ELVIS model. A brief conclusion (Section 6) summarizes the outcomes.
Section snippets
Materials
The injury severity analysis in this paper first investigates crashes with the two-vehicle straight-line and side-impact collision only and then extends to all types of crashes. In these cases, we apply energy loss theory to explain the injury mechanisms. MAIS, which evaluates the most severe injury in each collided vehicle, is the injury severity assessment in following procedures.
The National Automotive Sampling System (NASS)/Crashworthiness Data System (CDS), which contains a sample of
Newtonian mechanics in crashes
In typical two-vehicle crashes, the destructive energy is dissipated first in the deformation of designed protective structures and second in decelerating the uncrashed structures (Wood, 1997). Then, the acceleration is imposed on vehicle bodies and on unrestrained or restrained occupants as well. Occupants retain their forward motion due to the acceleration imposed on their bodies and finally collide with the interior surface of vehicles or are restrained by safety belts (they collide with the
The proportional ordered logistic regression
The POLR model sets MAIS as a categorized and ordered factor measuring the most severe injury of occupants. With this regression model, straight-line and side-impact crashes are separately involved. The calibration results are shown in Fig. 3.
The values of α with the minimum RD are −0.8 for head-on and rear-end crashes and −5.0 for side-impact crashes; hence, the initial hypothesis of negative α value is corroborated. In terms of absolute values of α, it is much larger in side-impact crashes
Discussion
In the sections above, we have revealed the statistical and predictive capability of the ELVIS model in two-vehicle crashes. The mass ratio, which distributes the energy absorption, is the core ELVIS model component that is worth discussion. The calibrated power values for the mass ratio in longitudinal and lateral directions are both negative, which indicates that occupants in heavy vehicles absorb less impact from the crashes, suffering less significant injuries. This result, however, leads
Conclusions
Injury reduction requires analysis of the impact suffered by occupants in vehicular crashes. Building relationships between physical injuries and crash mechanisms in the real world is complicated, so statistical models are used to handle this problem.
While the assessment criteria for injury severity has many forms in literature, this paper uses the common MAIS system to compare two different models: the widely-used Delta-V approach and our proposed energy-loss vehicle injury severity (ELVIS)
Authors’ contribution
Ang Ji: conceptualization, methodology, software, validation, formal analysis, investigation, writing – original draft. David Levinson: writing – review & editing, supervision.
References (51)
- et al.
The joint analysis of injury severity of drivers in two-vehicle crashes accommodating seat belt use endogeneity
Transp. Res. Part B: Methodol.
(2013) - et al.
A study on crashes related to visibility obstruction due to fog and smoke
Accid. Anal. Prev.
(2011) - et al.
An assessment of the effect of driver age on traffic accident involvement using log-linear models
Accid. Anal. Prev.
(1998) - et al.
Injury risk functions for frontal oblique collisions
Traff. Inj. Prev.
(2018) - et al.
A modified rank ordered logit model to analyze injury severity of occupants in multivehicle crashes
Anal. Methods Accid. Res.
(2017) - et al.
Car occupant safety in frontal crashes: a parameter study of vehicle mass, impact speed, and inherent vehicle protection
Accid. Anal. Prev.
(1998) Crash injury loss: the effect of speed, weight and crash configuration
Accid. Anal. Prev.
(1977)Crash injury prediction model
Accid. Anal. Prev.
(1979)- et al.
Case studies considered as retroactive experiments
Accid. Anal. Prev.
(1975) - et al.
Analysis of injury severity and vehicle occupancy in truck-and non-truck-involved accidents
Accid. Anal. Prev.
(1999)
Analysis of traffic injury severity: an application of non-parametric classification tree techniques
Accid. Anal. Prev.
Vehicle occupant injury severity on highways: an empirical investigation
Accid. Anal. Prev.
The influence of vehicle damage on injury severity of drivers in head-on motor vehicle crashes
Accid. Anal. Prev.
Identifying significant predictors of injury severity in traffic accidents using a series of artificial neural networks
Accid. Anal. Prev.
Factors influencing pediatric injury severity score and glasgow coma scale in pediatric automobile crashes: results from the crash injury research engineering network
J. Pediatr. Surg.
Driver injury and fatality risk in two-car crashes versus mass ratio inferred using Newtonian mechanics
Accid. Anal. Prev.
Risk factors affecting injury severity determined by the MAIS score
Traff. Inj. Prev.
The influence of alcohol, age and number of passengers on the night-time risk of driver fatal injury in New Zealand
Accid. Anal. Prev.
Driver injury severity: an application of ordered probit models
Accid. Anal. Prev.
Identification and validation of a logistic regression model for predicting serious injuries associated with motor vehicle crashes
Accid. Anal. Prev.
Using support vector machine models for crash injury severity analysis
Accid. Anal. Prev.
Traffic accident involvement rates by driver age and gender
Accid. Anal. Prev.
Trends in aggressivity and driver risk for cars, suvs, and pickups: vehicle incompatibility from 1989 to 2016
Traff. Inj. Prev.
Serious injury prediction algorithm based on large-scale data and under-triage control
Accid. Anal. Prev.
An analysis of motorcycle injury and vehicle damage severity using ordered probit models
J. Saf. Res.
Cited by (11)
Pricing lane changes
2023, Transportation Research Part C: Emerging TechnologiesMachine learning-based injury severity prediction of level 1 trauma center enrolled patients associated with car-to-car crashes in Korea
2023, Computers in Biology and MedicineCitation Excerpt :In the previous study, various classifiers were conducted to compare the prediction performance of each model. These include decision tree [48,49], k-nearest neighbor [24], support vector machine (SVM) [50,51], tree-based model [52], neural networks [53], Naïve Bayesian classifier [54], and gradient boosting [55]. Yet, some latest methods have received the attention that is superior to conventional prediction models in the case of MVOs-related injury classification.
Traffic Accident Severity Prediction Research and Application Based on Ensemble Learning
2024, Jisuanji Gongcheng/Computer EngineeringInjury Risk Assessment and Interpretation for Roadway Crashes Based on Pre-Crash Indicators and Machine Learning Methods
2023, Applied Sciences (Switzerland)Vehicle crash mitigation strategy in unavoidable collision scenarios: focusing on motion planning by considering a generalized crash severity model
2022, Journal of the Brazilian Society of Mechanical Sciences and Engineering