International Journal of Crashworthiness

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tcrs20

Mathematical models for assessment of vehicle

crashworthiness: a review

Gulshan Noorsumar, Svitlana Rogovchenko, Kjell G. Robbersmyr & Dmitry

Vysochinskiy

To cite this article: Gulshan Noorsumar, Svitlana Rogovchenko, Kjell G. Robbersmyr

& Dmitry Vysochinskiy (2022) Mathematical models for assessment of vehicle
crashworthiness: a review, International Journal of Crashworthiness, 27:5, 1545-1559, DOI:
10.1080/13588265.2021.1929760

To link to this article: https://doi.org/10.1080/13588265.2021.1929760

© 2021 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group.

Published online: 07 Jun 2021.

Submit your article to this journal

Article views: 6838

View related articles

View Crossmark data

Citing articles: 3 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tcrs20
INTERNATIONAL JOURNAL OF CRASHWORTHINESS
2022, VOL. 27, NO. 5, 1545–1559
https://doi.org/10.1080/13588265.2021.1929760

Mathematical models for assessment of vehicle crashworthiness: a review

Gulshan Noorsumar , Svitlana Rogovchenko, Kjell G. Robbersmyr and Dmitry Vysochinskiy
Department of Engineering Sciences, University of Agder, Grimstad, Norway

ABSTRACT ARTICLE HISTORY

This article reviews approaches to mathematical modeling of a vehicle crash. The growing focus on Received 4 February 2021
vehicle and occupant safety in car crashes has triggered the need to study vehicle crashworthiness in Accepted 3 May 2021
the initial stages of vehicle development. The major motivation for this work is to support vehicle
KEYWORDS
crashworthiness design during the product development process.The article is divided into two parts;
vehicle crashes; injuries;
the first one overviews existing mathematical models used to solve engineering problems. The second occupant protection;
part describes modeling strategies applied for replicating non-linear vehicle crash event and occupant mathematical modeling;
kinematics in an occupant protection loadcase. We also highlight alternative modeling strategies using lumped parameter
hybrid modeling techniques aimed at the improvement of the vehicle development process. modeling; finite element
methods; multi-body
Abbreviations: AI: Artificial Intelligence; ANN: Artificial Neural Network; MBS: Multi Body Systems; FEM: systems; response surface
Finite Element Methods; FEA: Finite Element Analysis; FE: Finite Element; CAD: Computer Aided methodology; crash
Design; CAE: Computer-Aided Engineering; LMS Models: Lumped Mass Spring Models; DOF: Degrees response models
of Freedom; UHSS: Ultra-High Strength Steel; VDCS: Vehicle Dynamics Control Systems; NHTSA:
National Highway Transport Safety Administration; FMVSS: Federal Motor Vehicle Safety Standards;
V2V: Vehicle–Vehicle; RSM: Response Surface Methodology; BEV: Barrier Equivalent Velocity; PDE:
Partial Differential Equation

1. Introduction car. These models involve differential equations of motion

describing the deformation of the parts in the vehicle. The
The notion of ‘crashworthiness’, first used in the aerospace
occupants in the car can also be included in a mathematical
industry in the early 1950s, introduced the measure of the model to predict injury values during a crash. Construction
ability of the structure to protect its occupants in survivable of an appropriate model involves the elimination and mini-
crashes [1]. In the automotive industry, the term refers to mization of effects deemed to be negligible. The quantities
vehicle’s structural abilities to plastically deform and absorb that are modeled are expressed as functions depending on
sudden impact loads maintaining enough survival space for independent and controllable variables. Non-linear physical
the occupants. Crashworthy vehicle structures should be stiff systems very often are modeled by ordinary and partial dif-
in bending and torsion for proper ride and handling and ferential equations. To find specific solutions of such differ-
should minimize fore-aft vibrations that give rise ential equations one needs initial and/or boundary
to harshness. conditions. Solutions can be validated with empirical data
According to Du Bois et al. [1], the vehicle should be from the physical experiment, see, for instance, Shier and
able to: (i) deform plastically in the front end and absorb Wallenius [2].
crash energy in case of a frontal crash and prevent intru- The classes of differential equations to which the analyt-
sions in the driver compartment; (ii) deform plastically in ical solutions exist are very limited; therefore, numerical
the rear end to protect occupants in case of a rear impact; methods are being employed. In this case computational
and (iii) have well-designed side structures preventing intru- inaccuracies add up to the inherent inaccuracies of the
sion into passenger compartment and opening of doors due model and the result must be compared with the experi-
to loading in a crash. mental data. As suggested by Marion and Lawson [3], one
Most safety regulations require crash testing at a special- of possible approaches to mathematical modeling involves
ized facility to determine the crashworthiness parameters. the following steps: (a) building; (b) studying; (c) testing;
Car manufacturers conduct full vehicle or sled tests to and (d) use of the model. Vehicle crashes are highly
ensure that the car design meets the regulations. Usually, non-linear transient dynamic phenomena. In an impact, a
crash-testing is time consuming and costly. Mathematical non-linear relation holds between applied force and dis-
models are employed to represent crash dynamics, for placements; it appears due to geometrical non-linearity
example, in the case of a car impacting a barrier or another (non-linear behaviour of highly deformable bodies leading

CONTACT Gulshan Noorsumar gulshan.noorsumar@uia.no

ß 2021 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License (http://creativecommons.org/licenses/by-nc-nd/4.
0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or built upon in
any way.
1546 G. NOORSUMAR ET AL.

The methodology includes the use of lumped parameter

models, fine-grained lumped models and coarse mesh finite
element models. One of the major challenges is that the
accuracy of these models is affected by the simplifications
and reduced number of degrees of freedom (DOF). Lumped
parameter models are the most commonly used reduced
order models; they include spring-damper systems replicat-
ing a deformable part and a concentrated mass representing
the undeformed structures like engine and transmissions.
Passenger compartment integrity is essential for vehicle
structural loadcases; for simplicity, it is often assumed that
the passenger compartment is integrated with the chassis as
a lumped mass. However, occupant protection models need
to accommodate for compartment deformations in order to
understand the cabin intrusions and better predict possible
crash scenarios. Lumped parameter models are also used to
predict occupant movements and possible injuries in a car
crash. The use of lumped masses for head, torso and legs,
all connected by springs replicating joints, helps to under-
stand the head and neck deflections and torso movements
Figure 1. Common used models for vehicle crash.
in a crash.
The idea behind a reduced order model is to depict the
to non-linear strain-displacement relations), material nonli- rigid components as masses connected by springs and deter-
nearity (elasto-plastic material) and combinations of these mine the forces acting on the masses from the external
two types of non-linearities. Material nonlinearity depends impact and the spring forces. These forces and energy con-
on a number of factors: rate of deformation, temperature, version laws are used to determine the governing equations
pressure, humidity, age of the material and the deformation of motion which are set up using one of the following
history [4]. In case of vehicle impacts, it has a significant formulations.
influence on the deformation and it is important to replicate
material and geometrical non-linearity in vehicles while
modeling the crash phenomenon. To deal with such non- Newtonian mechanics
linearities, Finite Element Method (FEM) is often employed. The Newtonian approach relies on three Newton’s laws of
It has higher accuracy but includes manual efforts to mesh motion [5]. The mechanics of particles can be described by
the parts along with increased computational efforts. In con- the Newton’s laws of motion which which describe the rela-
trast, simplified mathematical models are less resource-con- tionship between an object’s motion and the forces acting
suming yet they have lower prediction levels. In several on it.
studies models which replicate the collision mechanics with
considerable confidence were developed, however a com- Lagrangian mechanics
promise between computational time and accuracy is The Lagrangian approach uses energies rather than forces to
always present. define the dynamics of a system. The Lagrangian is the cen-
This article reviews the existing approaches to mathemat- tral quantity in Lagrangian mechanics, it obeys the following
ical modeling of car crashes. Although the use of models equations:
helps to reduce the dependence of automotive design on
d oL oL
physical crash test data for determining crash parameters
¼ Qi ,
and injury values to occupants, they do not fully replace dt o qi : oqi
real time tests due to certain inevitable assumptions which where, in general case, L ¼ T
V, T is the total kinetic
restrict the analysis of the kinematics of the event in detail. energy of the system equal to the sum of the kinetic ener-
The research reported in the literature indicates the need to gies of the particles, qi , i ¼ 1, ::, n are generalized coordinates
further improve the predictive power of existing models for and V is the potential energy of the system.
efficient application in a vehicle design development
(Figure 1).
Hamiltonian mechanics
In Hamiltonian mechanics, the time evolution is obtained
2. Methodology for crash modeling by computing the Hamiltonian of the system in the general-
ized coordinates. The Hamiltonian principle describes the
2.1. Reduced order dynamic models
motion of those mechanical systems for which all forces are
These models have reduced complexity yet capture the kine- derivable from a generalized scalar potential that can be a
matics of the crash with the load paths and components. function of the coordinates, velocities and time [5].
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1547

Lagrangian and Hamiltonian principles together form a

compact invariant way of obtaining the mechanical equa-
tions of motion.
Reduced-order models allow prediction of large deform-
ation structures, help in analyzing component-level simula-
tions during the early vehicle development process and
assist in developing new vehicle architectures for automotive
applications. They distinguish themselves from other meth-
odologies by including design dimensions in the system;
users are able to develop a predictive model which may not
depend on vehicle crash data besides the validation phase
of modeling.

2.2. Multi-body models

Figure 2. LS Dyna FE model of a full vehicle crash.
A Multi Body System (MBS) is a system that consists of
rigid bodies, or links, that are connected by joints which
restrict relative motion of the parts. The study of MBS dis- formulation. The complexity of dynamic equations of
tingushes forward dynamics which analyzes the motion of motion makes such models challenging computationally;
mechanical systems under forces, whereas the inverse this stimulated the development of the software for com-
dynamics deals with the analysis of forces causing the puter simulation since 70’s. The programming codes sup-
motion of bodies [6]. Multi-body models are used for both port different functionalities ranging from the generation of
dynamic and kinetic analysis [7]. Lagrange devised the for- equations of motion to numerical simulations for solving
mulation for the dynamics of multi body systems in 1788 in the equations [6]. Examples of computer code guidelines
Mecanique Analytique [8] and since then is recognized as can be found in the articles of Barley and Cripps [19] and
the father of multi-body dynamics. Important additions to Dopker [20].
this methodology include application of friction (by Multi-body models are applied in vehicle development
Coulomb [9]), beam elasticity (by Euler [10]), contact com- process for several decades to design vehicle handling and
pliance (by Hertz [11]) and lubrication (by Reynolds [12]). suspension systems [21]. One of the studies in this context
Two hundred years after the formulation was proposed by is due to Hegazy et al. [22] where the vehicle structure is
Lagrange, the methodology gained further impetus with the represented by rigid bodies connected by springs, dampers
introduction of improved matrix manipulation techniques and joints. Recently MBSs have been also used to develop
by Denavit and Hartenberg [13]. During the past century, generic models for the study of crashworthiness in vehicles
the improvements in solution methods and their computa- and for the prediction of the impact of crashes on vehicles
tional efficiency supported applications of this methodology during the development process [23]. Lower accuracy and
in different aspects of machine design including vehicle cumbersome process required for developing the model are
design analysis [7]. The analysis of linkage mechanisms the limitations of this methodology; although it is quite use-
developed by Wittenbauer [14] was followed by the use of ful for early development phases of vehicle design. MBSs are
rigid body dynamics for the analysis of human gait by used for the development of occupant and pedestrian mod-
Fischer [15]. Segel [16] studied the motion of a vehicle on a els in crash analysis where one of the main challenges is
flat road in response to steering control. Orlandea et al. [17] related to the replication of the anthropometry of the
proposed a practical solution methodology for large rigid human body. The representation of different joints of the
MBS based on the Lagrangian dynamics for constrained sys- body has been implemented in several commercial software
tems; this led to the development of ADAMS (automatic programs like MADYMO. Similarly to reduced order mod-
dynamic analysis of mechanical systems), the driving force els, this methodology has insufficient accuracy and less
behind many advancements in the automotive industry detailed modeling of the system. For instance, the occupant
(Figure 2). and pedestrian human body models lack details like skin
Constructing the governing equations for MBS is chal- and ligaments which might be critical for determining cer-
lenging; one of the classical approaches is based on the tain loadcase parameters in occupant and pedes-
Lagrange method for setting up the equations which are trian protection.
solved numerically afterwards. However, this approach is
time consuming, especially with systems having large num-
2.3. Non-linear finite element models
ber of components. Nikravesh [18] has proposed a new
methodology for constructing equations of motion for an Finite element modeling uses finite element method (FEM)
MBS based on a body-coordinate formulation using to solve boundary value problems (BVPs) for partial differ-
Newton-Euler equations and a joint-coordinate formulation ential equations (PDEs) arising in many physical and engin-
employing relative coordinates. The study also describes sys- eering problems. The solution of such problems for PDEs
tematic transformation from the former to the latter can be considered in two forms: strong and weak. A strong
1548 G. NOORSUMAR ET AL.

form of the governing equations states that the solution vehicle structure. The stiffness and dimensions of each com-
must satisfy the problem at every point of the domain along ponent need to be defined before the solver is used to deter-
with boundary conditions; it assumes that the classical solu- mine the acceleration and deformation in the crash event.
tion to the problem exists. A weak form states that the solu- The process of detailed intrinsic meshing is cumbersome
tion must satisfy the problem in an integral sense and is and requires training to represent the entire CAD surface
used when the classical solution to a problem cannot be with a discretized mesh. This calls for research and predict-
established. FEM is a special method which subdivides the ive simulations at early design stages thus possibly reducing
original BVP into smaller problems called finite elements in the number of re-design stages since the timescales tend to
order to approximate PDEs. The solution is derived using become shorter in automotive industry. Improvements can
numerical methods for solving systems of algebraic equa- be achieved through the collaboration of car manufacturers
tions and systems of ordinary differential equations. The with academic institutions in multidisciplinary research.
basic steps of an FEM are [25]: establishing the strong for- One of the major challenges in using software programs
mulation, obtaining the weak formulation, choosing approx- like LS-Dyna or PAMCRASH for engineers transitioning
imations for the unknown functions, choosing the weight into automotive industries is the extensive training required
functions, and solving the system. to understand the solver codes and assumptions made dur-
The finite element models are developed by discretizing ing the analysis. These complex programs are not a part of
the CAD surface into elements and nodes which cover the curriculum for engineering students or academic research
geometry of the vehicle (mesh) and the finite element BVPs and there is a need to bridge this gap between academia
are developed from the discretization. These problems are and specific requirements of the industry.
solved in order to determine the nodal displacements. The
elemental stresses and strains can be derived from the expli-
2.4. Response surface models
cit finite element method. In order to get a better approxi-
mation it is preferable to have a higher mesh size with more The Response Surface Models (RSM) are statistical
nodes covering the domain. approximation models which do not rely on the physical
The FEM approach in engineering was developed in the description of the objects but explore the relationship
early 1940s when Hrennikoff [26] and Courant [27] used between the input (predictor, or design variable) and out-
mesh discretization for elasticity and structural analysis put response (dependent variable) using a number of
problems. Clough published the first article on FEM in 1960 experiments in which the predictor variables are changing.
suggesting that two-dimensional elements connected to In automotive industry, RSM can be employed to measure
more than two nodes can be used to solve problems in con- the performance of the system and, in combination with
tinuum mechanics [28]. In 1965, NASA Structural Analysis numerical simulation methods, they are used to improve
software (Nastran) was developed to solve structural analysis or optimize a product and its performance [30]. The meth-
problems; this paved the way to simulation of engineering odology was developed by Box and Wilson who used the
stress strain problems with software codes. In the following sequential method in chemical process design [31]. The
decade Alberto Peano developed the first professional FEM motivation for their work was the problem of planning
p-version code which was used by Szabo in an industrial and analyzing experiments in search of desirable conditions
implementation PROBE in 1982. The qualitative research of on a set of controllable, or design, variables [32]. The
Spethmann et al. based on expert interviews analyses the response surface analysis can be viewed as analysis that
impact of the use of finite element methods in vehicle crash deals with a fitted function and accommodates a large col-
simulations on productivity and problem-solving [29]. The lection of techniques. RSM uses linear and quadratic mod-
authors argue that since the 1960s, when the explicit FEM els to fit a sequence of local regression models with
was developed and applied to crash events, it became not experimental data.
only an alternative to physical destructive testing but also a The RSM algorithm consists of the four steps: (a) per-
method for solving problems which formerly had been form a statistically designed experiment, (b) estimate coeffi-
impossible to solve. Even though automotive industry cients in the response surface equation, (c) check on the
gained trust in crash simulations, the lack of appropriate adequacy of the equation (via lack-of-fit test), and (d) study
software and hardware brought them to a standstill in the the response surface in the region of interest [32].
late 1970s to early 1980s. The article highlights the emer- For engineering applications, the process of constructing
gence of supercomputers in the late 1980s which aided models often includes the following three steps:
research to improve the performance of passive safety sys-
tems in a crash. Since then the dependence of engineers on  design of experiments: this involves setting the factors at
computer software programs and computer power has been different levels for proper experiments and ensuring that
constantly growing. The FEM approach is widely used by the boundary values as well as the entire area of the
automakers to simulate crash although the process is time- interest of the model is tested for different combinations
consuming and requires skills to develop the full size finite of variables;
element models. Another shortcoming of the FEM in crash  data collection: the process involves running the experi-
simulations is the dependence of the results on CAD data ments to collect the data including FE simulations or
for the structure and non-linear material properties of real time crash tests.
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1549

 data fitting: this is the final step which involves using as velocity of impact, crash model and other collision
algorithms to fit the sample data matching specific conditions. However, researchers developed efficient
requirements. The feasible design solution is obtained at schemes to overcome this problem in crash modeling. For
this step and used for design recommendations or rele- instance, Wei has proposed a crash pulse model to deter-
vant changes aimed at meeting the crash loadcase mine crashworthiness of vehicles [36]. This methodology
requirements. resembles reduced order modeling, however these models
find applications in accident reconstruction and depend
The RSM methodology was used in non-linear finite elem- on crash pulses or crash data for model development
ent models where accurate response surface models are con- and validation.
structed and evaluated for repeated replacement of the finite
element model at each time step of the analysis [33]. In com-
3. Applications of modeling strategies
parison with the modeling based on sensitivity analysis, the
RSMs provide considerably more accurate predictions reduc- 3.1. Reduced order models
ing dependence on FE models [34]. One of the shortcomings
The standard approach for lumped mass spring (LMS) mod-
of the RSM technique is the dependence on real crash test/
els is that bodies are represented by concentrated point
simulation data. Such models are unable to predict new scen-
masses which are connected by linear/non-linear springs.
arios in crash loadcase and have been found to be less accur-
The springs are defined by force-deformation and force–ve-
ate for non-linear impacts. It is crucial to know the algorithm
locity curves and deform due to the application of a force.
behind the RSMs, otherwise it becomes a “black box”
This approach was first introduced in automotive suspen-
approach and finding the magnitude of approximation errors
sion design in the early 1900s and has been extensively used
is difficult [35]. Another limitation of this technique is that
in automotive development since then.
the developed response surface is invalid for regions other
The article by Kamal [37] is one of the earliest studies in
than those set in the problem. The RS methodology fits the
modeling of crash events using lumped parameter models.
data to a second order polynomial, in which case the tech-
The model developed in this article includes three mass
nique gives accurate prediction but may fail for problems
components and eight resistances representing the deform-
with higher order polynomial approximations.
able structures of the vehicle. The lumped masses represent
The RSM methodology is also useful in parameter identi-
the body chassis mass, the engine transmission and the
fication models which help predict the stiffness and damp-
vehicle bumper. The non-linear resistances along with
ing values for vehicle deformation; such models find
the inertial components (lumped masses) are used to solve
extensive applications in accident reconstruction.
the basic equations of motion numerically. The dynamic
force acting on the resistances is approximated using static
2.5. Crash Pulse models forces acting on the vehicle during the crash event, where
the constant factor is assumed to be independent of the
Crash pulses represent the dynamic response of a vehicle in geometry of the structure. It is assumed that the structure is
a crash event and serve as a validation for most algorithms two-dimensional with a closed rigid frame. This implies that
developed to predict crash responses. These models also the model may not predict the vehicle behaviour out of
help to explain the energy conversions in vehicle structure plane forces experienced by the structure in a crash. The
during the impact; structural optimizations are also based study correlates well with physical test data for displacement
on crash pulses [34]. Furthermore, crash pulses are used in while the acceleration peaks are not well correlated.
validation of crash simulations where most validation algo- However, the trend for the acceleration curves is similar
rithms compare model simulations with real time crash which indicates that the model predicts the event’s kinemat-
data [36]. ics to a reasonable extent. The static and dynamics force-
The crash pulses are modeled using the function repre- deformation curves show a lower peak for the static curve
senting the vehicle acceleration and the crash process. If x(t) which is expected because the model does not account for
stands for the acceleration, then the crash pulse model Fh the impact loading acting on the structure in a dynamic
should ensure that crash event. The study also includes elastic body analysis for
ra ðt0 Þ ¼ xðt the vehicle passenger compartment and calculates the forces
Ð t0 Þ
Fh ðt0 Þ0,
rv ðt0 Þ ¼ 00 ra ðtÞdt0, exerted on the members in case when the occupant com-
Ðt Ðt Ðt partment is not considered a rigid lumped mass. A param-
rd ðt0 Þ ¼ 00 rv ðtÞdt ¼ 00 00 ra ðtÞdtdt0
eter study on the elastic passenger compartment indicates
at all times t0  0 where ra ðtÞ, rv ðtÞ, rd ðtÞ are the residual that the structure’s capability to withstand crash increases
signals of acceleration, velocity and displacement, respect- with increasing metal thickness. This observation is in line
ively [36]. In the past, the crash pulse was represented with the basic understanding of bending forces, that is, the
using different pulse shapes including square, triangular, thickness of the structure contributes to the crashworthiness
half-sine and even polynomial functions. In general, a of the body (Figure 3). l shows the model developed
crash pulse is defined only for a specified crash scenario by Kamal.
and may not be applicable for different loadcases. There Identification of parameters involves a range of
could be numerous factors influencing crash pulses such approaches, for instance, a piece-wise linear approach where
1550 G. NOORSUMAR ET AL.

that no rotational energy is lost in offset impacts. The rota-

tional energy losses were accounted for in the SISAME 3 D
model adopted later by NHTSA where the masses were no
longer considered as point masses which improved the
model’s reliability.
The early approaches to parameter determination in LMS
models proved to be efficient and were further developed to
improve agreement between model outputs and data sets; a
number of parameter identification techniques used by
researchers in vehicle modeling, will be discussed in our
future research article.
Figure 3. Vehicle impact simulation model in [37]. Cheva et al. [42] developed a lumped parameter model
to replicate a zero degree frontal crash test and a 40% offset
deformable barrier crash test. The barrier is defined as a
the force deformation characteristics are represented by the large lumped mass as well as the firewall which represents
hat functions or Chebyshev polynomials. The studies con- the passenger compartment. The deformation of the firewall
ducted in [38, 39] used optimization approaches to estimate indicates the intrusion in the occupant compartment. The
crash parameters. The algorithm developed in [38] helps left and right sides of the vehicle were modeled separately
optimize the acceleration data for a full frontal crash using so that the same model can be used with minor modifica-
tions for an offset crash event. The model includes mass
the force deformation curves for a few components in the
components representing several parts in the deformable
vehicle. In [39], the solution space approach is used to
zone like the engine assembly, radiator, suspension compo-
develop an algorithm which is applied to three engineering
nents, and front rails. The crash was simulated at 48 and 56
vehicle crash scenarios. The algorithm determines the force
kmph and the results were validated against physical crash
deformation curves used for frontal crash components. This
test data. The same model was used for 40% offset deform-
approach is illustrated with an example where the algorithm
able barrier loadcase with the barrier imparting more load
is used to determine whether submarining occurs in the
on one side. Then the upper rails have higher load from the
rear passenger seat and to design the car seat-belt and rear
deformable barrier causing higher deformation on the
occupant structure which ensure the absence of submarining
impacted side. The event kinematics are observed to be dif-
during a crash event. The optimization problem has a solu-
ferent in an offset crash scenario compared to a full frontal
tion meeting the design constraints.
loading case.
Sharp et al. [40] used Lagrangian method to simulate
vehicle motion. The equations of motion take into consider-
ation external forces acting on the vehicle and incorporate 3.1.1. Discrete time domain simulations
pitching, rolling and yawing effects on the car. The equa- The crash behaviour can also be described using discrete
tions also include the sprung mass of the vehicle and time domain simulation in lumped parameter models. The
unsprung masses per wheel along with the moment of iner- approach allows to predict and understand the crash
tia in the x, y and z axes. The numerical model predicts the response in terms of deformation, acceleration, velocity and
body roll, pitch and yaw angles and the tire forces in the rotation angles during the entire span of the crash event.
longitudinal and lateral directions. This mathematical model The research by Elkady et al. [43, 44] focuses on develop-
replicates the motion of an ideal vehicle with inertial forces ing mathematical models for replicating a vehicle crash
and the coupling between pitch and bounce. The limitations using non-linear springs for the vehicle bumper. The
of the model include the lack of non-linear springs and lumped parameter model developed in [43] and [45] uses a
anti-roll bar to represent the suspension system in lumped mass representing the vehicle body and four spring
more detail. damper units to replicate the suspension and wheels. It is
A method for finding the parameter values for spring assumed that the vehicle is moving on a flat asphalted road
elongations was developed by Mentzer et al. [41] who used and the vertical motion of the tyres is neglected. The model
real time crash test data to determine the mass of the com- is designed to explore the effects of Vehicle Dynamics
ponents from acceleration and wall contact forces. They Control Systems (VDCS) on the crash mitigation for an off-
obtained the force-deformation curves for the springs from set impact with a rigid barrier. The effect of ABS (anti-lock
the load paths under the condition that the system should braking system) is also simulated by using a braking force
have comparable motions of its masses so that the force and component in the equation of motions. The front deform-
acceleration curves match the test data. This condition is able members are presented by non-linear springs with force
difficult to achieve as the number of load paths could be deformation characteristics and the forces on the springs
higher than the mass elements. This is the reason why the during the crash are calculated using numerical methods.
least square method is used for the parameter identification The model is validated by comparing the acceleration and
in a full crash test data. Some of the drawbacks of this deformation of the front end structures to the physical test
approach are: the energy absorption by the honeycomb data. The study concludes that the values of the post impact
structure during deformation was neglected; it is assumed speed of the vehicle in the mathematical model and in the
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1551

Figure 4. Barrier impact simulation model in [44].

physical test agree well. The variation in the curves for the front bumper mass or a rigid mass like an engine or battery
front end deformation suggests shortcomings of the model which may contribute to the deceleration and deformation
due to the inaccurate values of the system parameters. The of the vehicle. In Elkady et al. [47], the vehicle model is
article also discusses the effects of VDCS on the collision modified by adding a lumped mass for a front bumper
response for a 50 percent offset impact (Figure 4). which connects the front end members represented
The same 6 DOF mathematical model is employed to by springs.
solve the equations of motion using Euler’s method for full An offset impact with another identical vehicle is studied
frontal and offset impact [46] (Figure 5). to understand the crash response of the vehicle and how it
It is demonstrated that in the case of the vehicle deform- differs from the case of rigid barrier impact. The simula-
ation and deceleration during the crash the effect of the tions are performed for the impact speed of 55 kmph with
active VDCS is negligible. However, the vehicle pitch angles different car masses. The study could be extended to under-
show an improved vehicle behaviour with an active VDCS stand the deceleration in the vehicles for different impact
in the car. The model in this this study does not include the speeds and vehicle masses.
1552 G. NOORSUMAR ET AL.

deformation of both vehicle bodies due to the force was

plotted for the impact duration. The stiffness of each vehicle
was predicted based on equal force experienced by both
vehicles, suggesting that for a medium size car the average
side impact stiffness is a quarter of the front side stiffness.
The authors challenge the existing method of calculating the
side stiffness from force deformation curves asserting that it
overestimates the side stiffness of the car body at a front-to-
side collision. They argue that using only the central part of
the deformation zone for calculating the stiffness is only a
few percent lower than using the whole deformation zone
for the measurements. The use of the central portion for the
measurements does not only simplify them but also pro-
vides a higher accuracy of data for the measurements.
Jonsen et al. [52] proposed a lumped parameter model to
represent a bumper in a crash.
The system uses an optimization software INVSYS where
an unconstrained subspace-searching subplex method is
Figure 5. 3-DOF occupant multi-body model in [44]. implemented. The algorithm identifies the local minima and
can be applied for optimizing noisy objective functions. The
objective function is defined to minimize the error between
Elmarakbi et al. [48] developed a mathematical model for
the calculated and measured displacements; constraints
smart structures which improves the crashworthiness
include masses, damping and stiffness constants along with
response of a vehicle in a barrier impact. The smart struc-
total mass of the vehicle. The authors claim that if the DOF
tures are represented by spring mass damper systems for
of the system is increased to two, the error is reduced. This
vehicle and occupant and are simulated numerically with
result is validated using FE bumper system connected with
the help of an optimization algorithm which minimizes the
a 2 DOF spring mass damper system allowing only longitu-
intrusion in the occupant compartment and the deceleration
dinal motion.
injury for the occupant. The injury curves obtained from
The research on LMS models for vehicle crash has pro-
the simulation are compared to the vehicle model without
gressed slowly from simple spring mass models to more
smart structures.
complex multiple DOF models with spring-mass-damper
Ionut et al. [49] developed a 2D mathematical model
which includes 2 vehicles with 2 occupants to analyse the systems and non-linear springs. We remark that the govern-
occupant kinematics in a frontal collision with ing equations of motion usually use Newton-Euler formula-
another vehicle. tion but in the models including occupants Lagrangian
They use Lagrange’s generalized formulation to obtain formulation has been employed.
the system of five equations. The numerical solution pro-
vides the displacement and velocity of each of the vehicle 3.2. Multi-body models
bodies and the velocities of the occupant’s head and thorax.
The model was validated against real test data demonstrat- Ambrosio et al. [53] developed a full vehicle crash model
ing good correlation. The parameterization of the stiffness using an MBS with plastic hinge deformation. The entire
of the seat belt springs is used to understand the influence vehicle is represented by kinematic joints, data for hinge
of the spring stiffness on the occupant deceleration and deformation were derived from CAD data and finite elem-
displacement. ent simulations. Sousa et al. [23] suggested a generic car
A National Highway Transport Safety Administration model containing different parts including suspensions,
(NHTSA) Lumped Parameter Model was developed by Deb tires, occupants and structural components contributing to
et al. [50] for a side impact crash scenario. The authors load path. The representation of the structural components
identified lumped masses which were then added to the uses the plastic hinge approach. The model was validated
existing model based on finite element analysis of two pas- against a completely known finite element vehicle model
senger vehicles. The validation of this model was conducted and can be fine-tuned to have the same crash responses as
with simulations of two vehicles Dodge Neon and Dodge in the crash tests without the knowledge of the structure of
Intrepid. The authors suggested the methodology of deter- the tested vehicle. The study emphasizes the need for simple
mining the spring characteristics from the FE model using mathematical models in early stages of vehicle development
contact introduced between two components. This gives the process. Carvalho et al. [54] use the plastic deformation
force displacement characteristics for the spring members. methodology to develop an optimization algorithm for iden-
The deformation characteristics of a vehicle under front- tifying multibody models for crash analysis. The solution to
to-side impact were calculated by Prochowski et al. [51] the problem is obtained through sequential application of
using experimental and analytical equations. The combined genetic and gradient based optimization methods. This
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1553

methodology has been also employed to define an MBS for generic compact sedan cars in front and rear impact colli-
a large family car for the case of front and side crashes. sions. The single-DOF model included only rotational visco-
King et al. [55] developed a mathematical model for an elastic joints, and the two-DOF model allowed axial extension.
airbag which, in conjunction with a three DOF occupant It is shown that in a frontal collision, the highest risk of injury
model, can predict the effects of an airbag deployment on was for the lower cervical spine, and in a rear collision the
the occupant. The authors impose the following require- most serious injury occurs in the upper to mid cervical spine.
ments to this model: the airbag is spherical and mounted on The MBS models were validated against FE data and are in
a steering wheel; the airbag is already inflated when the agreement with the simulation data generated from FE tests.
simulation starts but with a low pressure and is expected to Portal et al. [24] developed an accident reconstruction
expand radially due to gas filling in the bag; the pressure is model using 3 D rigid body mechanics. The rigid car body
distributed uniformly and the deformation of the wall of the is modelled with nine rigid bodies and eight kinematic
airbag is linearly elastic. The three governing equations in joints representing different vehicle components; the study
this model are the elasticity equation, continuity equation includes also a motorcycle model and a human biomechan-
and the equation for state of the gas. The equations describ- ical model. The human biomechanical model features eight
ing what happens after the contact of the occupant with the rigid bodies and thirteen kinematic joints which replicate
airbag are proposed and the configuration of the deformed different parts of the body. These models were used to study
airbag is discussed. The mathematical model is implemented a frontal collision between a car and a motorcycle, an offset
in a computer program written in FORTRAN IV where it is collision between two cars and a pedestrian impact.
merged with the 3 DOF model of an occupant. The model
describes the contact of the airbag with the occupant and
3.3. Finite element models
the code reads contact information based on the occupant’s
position at any given time. The results of the simulation Finite element models have applications in many engineer-
were validated at the sled facility at the Wayne State ing problems; a non-linear transient impact analysis of
University using anthropometric dummies. The model’s vehicle crash is one of the areas where these models pro-
curves exhibit good correlation with the experimental data. duce reliable results. Thomke et al. [29] presented the evolu-
Elkady et al. [44] developed a 3 DOF multi body math- tion of crash simulations which originated in the military
ematical model to simulate a crash event of a car with an domains in the late 1960s. The automotive industry
occupant (Figure 5). embraced this technique in the early 1970s, however the
Three masses representing the lower body replicate the first full body vehicle crash simulation was conducted only
legs and pelvic area of the occupant who can perform transla- in the mid-1980s. The authors highlight the importance of
tion and rotation motion about the center of gravity (CG) of FEM simulations for predicting vehicle crashworthiness.
the body. The model replicates a seat belt with 2 spring Benson et al. [58] presented the calculations for crash-
damper systems and mitigates the impact for the occupant. worthiness design for automotive structures. This work laid
The MBS is integrated with the vehicle model developed in the foundations for future FE models using different types
the article. Under the full frontal barrier crash the lower part of elements and mesh size for capturing the vehicle geom-
of the body moves forward while the middle and upper body etry and employing various techniques to measure the stress
rotate slowly; the spring forces in the seat belt are introduced and strain from the simulations. Pifko and Winter [59] pro-
to reduce the rotation and movement of the body. Lagrange’s vided an overview of the theory behind FE, methods used
method is employed to derive the equations of motion. The to set up the governing equations based on Lagrangian
system of equations is solved numerically to compute the equations and establish the failure criterion. They also draw
occupant body deceleration. The results from the vehicle parallels with the aircraft simulations to understand the
crash model are used in the simulation of the crash impact application of FEM in the field of automotive safety point-
on the occupant. The results highlight the importance of ing out the need for computational scientists to describe
using seat belts, emphasizing that in crash events seat belts physical systems in detail prior to the solution of the associ-
are the primary restraints in the vehicle safety system. The ated differential equations.
rotation angle of the middle body is similar to the pitch of B€
ottcher et al. [60] described the progress with the use of
the vehicle in the crash; the crash causes a neck rotation FE models in automotive industry acknowledging that vir-
which could be fatal for an occupant. Remarkably, the change tual simulations developed rapidly over the last 20 years.
in the seat belts’ spring stiffness positively affects the neck Virtual simulations have come a long way into supporting
rotation and deceleration of the occupant. This study demon- the vehicle development process from a smaller model size
strates that the use of a hybrid technique mixing LMS with and lower accuracy to computationally intensive simulation
MBS models improves the overall crash response prediction. models which capture almost every part of the vehicle
Euler and Lagrangian equations are employed for vehicle and geometry and achieve improved prediction levels. The
occupant models respectively; the advantages of using each of authors point out that along with the standard loadcases,
the approaches are discussed. simulations nowadays feature even active sensing techniques
Hassan et al. [56] and Shi et al. [57] presented a multi like airbag deployment. Airbag sensing calibration technique
body model of the cervical spine of a 50th percentile male using virtual simulations has been demonstrated by Kiefer
occupant in a crash event performing FE simulations of two et al [61] who developed the algorithm for airbag
1554 G. NOORSUMAR ET AL.

deployment and discussed the advantages of using a virtual models employs Artificial Neural Networks (ANN). The
calibration technique for airbag sensing. The study shows new approach needs training on existing crash test data so
that the model does not need to be too complex unlike the that it can be used to predict crash scenarios. The data can
one for full vehicle loadcases, which reduces the computa- be generated using finite element models as well, which
tional costs. Recently, FEM has been used by the authors for makes it easier to collect necessary sets of curves for differ-
determining the crash response in welded vehicles which ent crash scenarios. However, this approach is not very effi-
contributed to the development of more stringent norms for cient for developing new car models or for the optimization
improper repairs on UHSS structural members [62]. and design of structures because it relies on existing data
FEM have also been extensively used for developing and predicts the impact characteristics only by using avail-
simulation models to determine injuries to occupants in a able crash test data. Omar et al. [72] use a recurrent neural
crash. Kirkpatrick et al. [63] employed the software LS network to predict the crashworthiness of a vehicle in a
Dyna to develop and validate biofidelic models of varying frontal crash demonstrating that ANN can be trained for
degrees representing an occupant in a crash. In the automo- non-linear impact models and produce satisfying results
tive industry these virtual models replace real time tests with good confidence levels.
with dummies or cadavers. The dummy modeling developed Several researchers used identification of parameters for
in this article differs from the rigid body kinematics model- developing predictive models for crash loadcases. Joseph et
ing of body parts like head, neck and abdomen because it al. [73] suggested a parameter identification method for a
accounts for the reflexes and joints in a human body during thoracic impact model predicting the chest injuries. The
collisions. Putra [64] presented a head-neck FE model for method minimizes the error between results from the math-
an average female occupant utilizing an optimization strat- ematical model and experimental data using an optimization
egy. The model employs an active neck muscle controller to algorithm demonstrating a reasonable correlation between
represent human reflexes during whiplash induced rear- the curves which agrees with the known results. The use of
impact. The FEM was also used by several authors to the chest injury metrics for the validation of the mathemat-
develop pedestrian humanoid models which simulate the ical model instead of real time acceleration data suggests
behaviour of pedestrian-vehicle crashes, see Howard et al. that these models could also support occupant protec-
[65], Pak et al. [66] and Meng et al. [67]. Detailed FE mod- tion loadcases.
els of a pedestrian replicate the anthropometry of a human Ghannam et al. [74] present a mathematical model to
head and legs and proved to be useful for predicting head determine the initial impact velocity of full frontal
and leg injuries in pedestrian collision scenarios. vehicle-to-vehicle test modes using the Barrier Equivalent
Design of complex elastic and inelastic material models Velocity (BEV) concept. The model is based on a basic
for simulation in crash loadcases has been a challenge for mass-spring damper; it determines the velocity of a
engineers since the accuracy of a finite element model is vehicle impacting another vehicle by calculating the crush
highly influenced by the replication of the behaviour of non- energy of both vehicles and using the conservation of
linear inelastic material in crash simulations. Ramaswamy et energy principles to define the initial velocity of the car.
al. [68] highlight the need for the development and validation Two major assumptions require that the lateral and rota-
of material models for the simulations of loadcases identify- tional energies are negligible compared to the initial kin-
ing the parameters that influence the robustness of quasi- etic energy of the bullet vehicle and the force-deformation
static bending simulation for the evaluation and performance curves in the vehicle front end for both vehicles are lin-
of material model in out-of-plane loading scenarios. ear. The authors introduce a scaling factor to account for
Several researchers have used FEM to validate accident the non-linear force deformation characteristics, the lateral
reconstruction models in the recent past, see, for instance, and rotational energies, thus ensuring that the model pre-
Numata et al. [69], Yu et al. [70] and Xueyan et al. [71]. dicts correctly the real test velocity. The curves are vali-
Accident scenarios can be reproduced successfully in finite dated with physical test data and scaling factors are
element models and reconstruction models can be validated added if necessary to adjust the graphs. It is concluded
in the absence of real time crash data. that the rotational and lateral energies have small influ-
It is worth mentioning that during the last decade there ence on the initial velocity.
have been only small advancements in the finite element Several studies include optimization strategies to predict
methodology; however applications of computer simulations crash kinematics. The methodology uses a combination of
for analysing crash scenarios have increased significantly. LMS and FEM to define the system and then curve fit-
Researchers and industry experts rely on virtual crash simula- ting techniques to determine parameters. Munyazikwiye et
tion data for a big part of the product development process al. [75] used a double spring mass damper model with
because this allows to reduce the product develop- two masses representing the front rail and the driver
ment timeline. compartment respectively representing a car hitting a rigid
barrier. The equations of motion are derived and solved
with the help of a real time test crash pulse inputted into
3.4. Response surface models
the MATLAB model. The spring stiffnesses and damper
One of the approaches to the modeling of a vehicle crash constants are derived by converting the state-space realiza-
which can address the drawbacks associated with LMS tion to transfer function. The mass distribution of the
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1555

signals can undergo decomposition analysis if the signals

from a non-stationary source become stationary after
decomposition. The study uses a 1997 Honda Accord crash
test data and the fifth-order Daubechies wavelet (db5) to
represent the motion of the structural components. The sig-
nal is compressed so that the time series contains a small
number of coefficients for estimating body injuries during a
crash. The authors also highlight another possible applica-
Figure 6. Vehicle impact (Kelvin) model in [78]. tion of their methodology to predict best and worst per-
formance in a sled test based on the impact pulse and for
vehicle is verified by comparing the curve generated by determining the range of performance using optimiza-
the model with the physical test data to select the most tion techniques.
feasible mass distribution based on the dynamic crush of Crash pulse data with Haversine pulse were employed to
the passenger compartment. The data from the four test study structural response of vehicle to impact [81]. The
cases is checked against the physical tests. The model crash pulse was used for different speeds and it was
does not account for material non-linearities and vehicle observed that the energy absorption had a linear relation-
geometry for predicting the vehicle crashworthiness. ship with the displacement for a range of velocities. Similar
However, the study gives an insight into the use of trans- behaviour is observed in the plots of absorbed energy vs
fer functions for predicting crash injury values. deflection. When the data from sine model were plotted and
Munyazikwiye et al. [76] used genetic algorithm for par- compared with the real time crash data, acceleration curves
ameter optimization to estimate the front deformation showed good correlation. It was observed that the sine wave
characteristics in case of a vehicle-barrier impact and a performed well for the full frontal barrier test while triangle
vehicle-to-vehicle impact. Physical crash test data are used pulse model showed good correlation for the offset model.
to fit the curves and determine piecewise linear spring The study does not explain why different models show good
deformation and damper characteristics. Usta et al. [77] correlation to different loadcases; this indicates the need for
used a genetic algorithm and RSM to design crashworthy more work on the loadcase comparison. Wei et al. [82] pro-
concentric circular tubes which crush on impact absorbing posed a model using piecewise linear functions to describe
the impact energy (Figure 6). the crash impulse based on CAE simulation data. They con-
clude that the model can be used to describe well the crash
process exactly and can be used to predict crash under dif-
3.5. Crash pulse model ferent conditions by varying the model parameters.
Crash pulse models have been used to represent acceler- Prediction of crash pulses is an interesting area of
ation, velocity and displacement wave forms of a structure research where different techniques including convolution
undergoing crash. The first step is to generate crash pulse methods [78] where a transfer function is employed for pro-
data by running physical tests or FE simulations to gather viding the output to the linear system. The vehicle crashing
an understanding on the type of impact. Signal pre-process- against a barrier can be represented as a spring damper sys-
ing is an important step in the crash pulse methodology; it tem which is inputted with an excitation and an output
includes filtering, re-sampling, synchronizing and trimming response is expected; the process which transforms this
the pulse [36]. This is followed by studying the crash stages input to an output in the time domain is described by the
and dividing the crash pulse into regions which better rep- transfer function.
resent the deformation and intrusion behaviour. Woolley We recognize that this is a relatively new field of vehicle
[79] proposed a crash pulse model which could be divided impact modeling and the opportunities to continue research
into two regimes: the dynamic compression and rebound in this domain should be further explored.
phases. The compression phase is defined by the maximum
dynamic crush in a vehicle impacting a barrier and its vel- 4. Discussion and conclusions
ocity becoming zero. The rebound phase in a vehicle-to-
vehicle crash is the time when the two vehicles start moving Each of the modelling strategies discussed in this review have
away from each other which leads to their separation. The been applied across different engineering domains to solve
solutions to the differential equations derived in this article complex non-linear dynamic problems. The research focused
can have varying periodic characteristics (like sine or cosine) on the improvement of these methodologies to address prob-
in the compression phase, and behave like polynomial func- lems which were difficult or impossible to solve. We observe
tions in the rebound phase. The author introduced a transi- the tremendous growth of application areas whereas the
tional trigonometric function to model a crash pulse and development of alternative modeling strategies was strongly
validated the model against real time crash. influenced by the availability of increased computational
Cheng [80] analysed crash response using wavelets and power. The parallel growth in computational power from
wavelet packets decomposing stationary and transient crash supercomputers to parallel CPUs helps solve complex equa-
signals into piece-wise stationary signals. The decomposed tions with high level of accuracy and saves time.
1556 G. NOORSUMAR ET AL.

Figure 7. Evolution of vehicle crash simulations.

Although mathematical modeling of vehicle crash started reliability of the analysis of non-linear transient impacts
to develop since the 1970s, the confidence in these models without physical tests.
has significantly grown over the years. This is a positive Quantitative methods, although less significant for under-
trend reducing the dependence on physical crash tests. standing the impact mechanics in detail, provide valuable
Mathematical models serve as a starting point for vehicle observations on the crashworthiness of a vehicle, like the
architecture development process providing recommenda- available crush space or coefficient of restitution. These
tions to the studio and design teams; they are also employed methods are a backbone of most mathematical models
during component design or for making changes in the which analyse the load paths of the vehicle impact.
existing components. Lumped parameter models show rea- It is imperative for engineers and academicians to be
sonable prediction power for frontal and side impacts. The aware of important modeling strategies and carefully access
major challenge faced in this field is the parameter identifi- the advantages and shortcomings of each of these methods
cation which is partially resolved now using several identifi- in order to apply the most appropriate one based on the
cation strategies which however still have certain considerations of accuracy and efficiency required in the
shortcomings. This hindered the use of LMS models in solution. The automotive industry is quite fast paced in
automotive industry during the development stages due to terms of developing new products and improving existing
architectures, the short product development cycle triggers
concerns related to new stringent safety regulations. The
the need for reliable virtual modeling methodologies which
development of LMS models have slowly progressed from
predict crashworthiness performance as close as within 5-
simple Kelvin models to complex spring-mass models with
10% of the physical tests. In addition, the vehicle safety reg-
multiple springs and dampers representing the vehicle
ulations have become more stringent over time as the focus
deformable features. The integration of occupant models in
on vehicle safety has gained momentum during the recent
the car structure implies the addition of higher number of
years. This in turn puts pressure on vehicle manufacturers
variables in the models but yields far greater understanding
who have to fulfill these regulations developing new prod-
of the loadpaths in a crash event (Figure 7). ucts. This implies that industry experts resort to processes
Response surface methods have gained momentum in the which are time consuming or computationally intensive to
recent past as well but their application is limited due to the get the satisfactory confidence levels of their results; this sets
fact that they cannot be used for new vehicle architectures a constraint on the adoption of new strategies or mathemat-
or for changing structural and occupant protection regula- ical models for the development cycle which should be less
tions with new crash scenarios. However, reinforcement complex yet explain the dynamics of the problem equally
learning methods can be employed to overcome these limi- well. The experts look for methodologies which solve engin-
tations. RSMs have proved to be highly effective in modify- eering problems with software automation or data science
ing existing designs of vehicle structures and decision and help to come up with new products for the competitive
making has been easier without running virtual or physical automotive market. On the other hand, the academic com-
tests based on data collection and using algorithms to inter- munity is equipped with the opportunity to explore different
pret the feasible design space. This has helped determine strategies but sometimes lacks the infrastructure and com-
feasible and non-feasible design regions for many compo- putational power to resolve complex modelling problems.
nent level loadcases and makes engineering judgements eas- There is a strong need to bridge this knowledge gap
ier for design teams. The emergence of efficient machine between research groups and engineering applications to
learning tools and algorithms is a promising trend in the ensure the improvement of the product develop-
automotive industry which can increase confidence in the ment process.
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1557

The non-linearity of dynamic impact in a vehicle crash [6] Larsson T. Multibody dynamic simulation in product develop-
along with the need for energy absorbing features to estab- ment [Ph.D. thesis]. Luleå University of Technology; 2001.
[7] Rahnejat H. Multi-body dynamics: historical evolution and
lish structural integrity in the vehicle is one part of a larger
application. Proc Instit Mech Eng, C: J Mech Eng Sci. 2000;
problem which also involves replication of anthropometric 214(1):149–173.
data of a human body model under the crash impact for the [8] de Lagrange, JL. Mecanique analytique. Vol. 1. Mallet-
analysis of the injury values for different body parts. There Bachelier; 1853.
is a need to explore hybrid modeling strategies which could [9] Coulomb C. Theorie des machines simple. Paris: Bachelier;
combine methodologies reviewed in this article to achieve 1821.
[10] Euler L. Novi commentarii academiae scientiarum petropolita-
the right balance of accuracy and efficiency in the solution; nae. Nr. 1776;20:189–207.
several relevant studies combining different modeling strat- [11] Hertz H, Jones DE, Schott GA. Miscellaneous papers.
egies aimed to overcome existing limitations. Macmillan and Company New York; 1896. p 146–163.
This article provides the concise overview of the existing [12] Reynolds O. IV. on the theory of lubrication and its application
research and challenges arising in the mathematical model- to Mr. Beauchamp tower’s experiments, including an experi-
mental determination of the viscosity of olive oil. Vol 177
ing of vehicle crashes. We identify possible areas of
Philos Trans R Soc London. 1886;(177):157–234.
improvements in this domain and emphasize a strong need [13] Hartenberg RS, Denavit J. A kinematic notation for lower pair
to build more confidence towards replacing physical tests mechanisms based on matrices. Trans Asme, J Appl Mech.
with simplified but accurate mathematical models. The lit- 1955;77(2):215–221.
erature review conducted in this article also highlights [14] Wittenbauer F. Graphische Dynamik. Berlin: Springer Verlag;
1923.
opportunities for improving mathematical models with
[15] Fischer O. Einf€uhrung in die mechanik lebender mechanismen.
vehicle structure and occupants to understand the impact Leipzig: Teubner. 1906.
dynamics under different crash scenarios. There is also a [16] Segel L. Theoretical prediction and experimental substantiation
need to implement parameter identification strategies which of the response of the automobile to steering control. Proc
incorporate the non-linear material properties of the front Instit Mech Eng: Autom Div. 1956;10(1):310–330.
end members in the LMS models and validate them against [17] Orlandea N, Chace MA, Calahan DA. A sparsity-oriented
approach to the dynamic analysis and design of mechanical sys-
physical test data. The growing need for infrastructural tems—part 1. ASME. J. Eng. Ind. August 1977. p 773–779
developments which allow to run finite element simulations [18] Nikravesh PE. Multibody dynamics formulation. In: Grasping
on hundreds of parallel CPUs instead of running multiple in robotics. Springer; 2013; p. 57–76.
physical tests to determine crashworthiness requirements [19] Barley SA, Cripps RJ. Executive-centred system design for cad
calls for research in the area of reliable reduced order FEM applications. Comput-Aided Des. 1992;24(5):235–242.
[20] Dopker B. Developments in interdisciplinary simulation and
models which are computationally less intensive. We also
design software for mechanical systems. Eng Comput. 1988;
recognize the remarkable advancements in the field of 4(4):229–238.
machine learning and data science and the opportunities [21] Kortum W, Sharp R. Multibody computer codes in vehicle sys-
they bring for the development of robust models for pre- tem dynamics: supplement to vehicle system dynamics. Vol. 22
dicting crash responses. Swets & Zeitlinger; 1993.
[22] Hegazy S, Rahnejat H, Hussain K. Multi-body dynamics in full-
vehicle handling analysis. Proce IMECHE. 1999;213(1):19–31.
Disclosure statement [23] Sousa L, Verıssimo P, Ambr osio J. Development of generic
multibody road vehicle models for crashworthiness. Multibody
No potential conflict of interest was reported by the author(s). Syst Dyn. 2008;19(1/2):133–158.
[24] Portal R, Dias J, Soares CM. Multibody models for vehicle acci-
dent reconstruction. ECCM 2006-III European Conference on
Funding Computational Mechanics-Solids, Structures and Coupled
Problems in Engineering; Springer 2006.
The authors thank to University of Agder (Norway) for support during [25] Singh G. Short introduction to finite element method.
the research and helping us with the resources. Norwegian University of Science and Technology. Trondheim,
2009.
[26] Hrennikov A. solution of problems in elasticity by the frame-
ORCID work method. Trans ASME, J Appl Mech. 1941;8:169–175.
[27] Courant R. Variational method for the solution of problems of
Gulshan Noorsumar http://orcid.org/0000-0002-6718-4508
equilibrium and vibrations. Bull Amer Math Soc. 1943;49(1):
1–23.
[28] Clough RW. The finite element method in plane stress analysis.
References
Proceedings 2nd ASCE Conference on Electronic Computation.
[1] Du Bois P, Chou CC, Fileta BB, et al. Vehicle crashworthiness 1960 Pittsburg, Pennsylvania, 8, 345–378.
and occupant protection. 2004. [29] Spethmann P, Thomke SH, Herstatt C. The impact of crash
[2] Shier DR, Wallenius K. Applied mathematical modeling: a simulation on productivity and problem-solving in automotive
multidisciplinary approach. Boca Raton: CRC Press; 1999. R&D. Tech. rep. Working Paper. 2006.
[3] Marion G, Lawson D. An introduction to mathematical model- [30] A. E Ş. Parameter optimization using the surface response tech-
ling. Bioinformatics and statistics Scotland. 2008. nique in automated guided vehicles. In: Sustainable engineering
[4] Sathyamoorthy M. Nonlinear analysis of structures. Vol. 8. products and manufacturing technologies. 2019; Elsevier; p.
CRC Press New York, 1998. 187–197.
[5] Goldstein H, Poole C, Safko J. Classical mechanics. 3rd ed. San [31] Box GE, Wilson KB. On the experimental attainment of opti-
Francisco (CA): Addison Wesley; 2001. mum conditions. J R Statist Soc: B. 1951;13(1):1–38.
1558 G. NOORSUMAR ET AL.

[32] Myers RH, Khuri AI, Carter WH. Response surface method- [54] Carvalho M, Ambr osio J. Identification of multibody vehicle
ology: 1966–1986. Tech. rep. Gainesville: Dept of Statistics, models for crash analysis using an optimization methodology.
Florida University of Gainesville, 1986. Springfield, VA : NTIS. Multibody Syst Dyn. 2010;24(3):325–345.
[33] Shahidi G, Pakzad SN. Response surface model updating for [55] King AI, Chou CC, Mackinder JA. Mathematical model of an
nonlinear structures. In: Topics in dynamics of civil structures, airbag for a three-dimensional occupant simulation. Tech. rep.,
Vol. 4. Springer, 2013; p. 283–288. SAE Technical Paper 1972.
[34] Gu L, Yang R-J, Li G, et al. Structural optimization for crash [56] Hassan MT, Shi MG, Meguid S. Nonlinear multibody dynamics
pulse. SAE Trans. 2005;786–792. and finite element modeling of occupant response: part i—rear
[35] Cox DC, Baybutt P. Methods for uncertainty analysis: A com- vehicle collision. Int J Mech Mater Des. 2019;15(1):3–21.
parative survey. Risk Anal. 1981;1(4):251–258. [57] Shi MG, Hassan MT, Meguid S. Nonlinear multibody dynamics
[36] Wei Z. Analysis, modeling and CAE validation of vehicle and finite element modeling of occupant response: part ii—
crashes using advanced signal processing tools [Ph.D. thesis]. frontal and lateral vehicle collisions. Int J Mech Mater Des.
Kristiansand, Norway: University of Agder; 2017. 2019;15(1):23–41.
[37] Kamal MM. Analysis and simulation of vehicle to barrier [58] Benson D, Hallquist J. The application of dyna3d in large scale
impact. SAE Trans. 1970;1498–1503. crashworthiness calculations, Lawrence Livermore National
[38] Graff L, Harbrecht H, Zimmermann M. On the computation of Laboratory (Report No. UCRL 94028). United States.
[59] Pifko A, Winter R. Theory and application of finite element
solution spaces in high dimensions. Struct Multidisc Optim.
analysis to structural crash simulation. in: Computational meth-
2016;54(4):811–829.
ods in nonlinear structural and solid mechanics, Elsevier, 1981;
[39] Graff L. A stochastic algorithm for the identification of solution
pp. 277–285. Pergamon Editors: Ahmed K. NoorHarvey G.
spaces in high-dimensional design spaces [Ph.D. thesis]. Basel,
McComb.
Switzerland: University of Basel; 2013.
[60] B€ottcher C-S, Frik S, Gosolits B. 20 years of crash simulation at
[40] Sharp R, Goodall J. A mathematical model for the simulation
opel-experiences for future challenges, LS Dyna
of vehicle motions. J Eng Math. 1969;3(3):219–237. Anwenderforum, Bamberg 2005.
[41] Mentzer SG, Radwan RA, Hollowell WT. The SISAME method- [61] Kiefer T, Erb P, Gosolits B, et al. Simulation of airbag sensing
ology for extraction of optimal lumped parameter structural signals using finite element method, in: LS-DYNA User Forum,
crash models. Tech. rep., SAE Technical Paper 1992. Bamberg 2004.
[42] Cheva W, Yasuki T, Gupta V, et al. Vehicle development for [62] Noorsumar G, Robbersmyr K, Rogovchenko S, et al. Crash
frontal/offset crash using lumped parameter modeling. Tech. response of a repaired vehicle-influence of welding UHSS mem-
rep., SAE Technical Paper 1996. bers. Tech. rep., SAE Technical Paper 2020.
[43] Elkady M, Elmarakbi A, Crolla D. A numerical study for opti- [63] Kirkpatrick SW, MacNeill R, Bocchieri RT. Development of an
mizing vehicle dynamics control systems in offset impacts. ls-dyna occupant model for use in crash analyses of roadside
Tech. rep., SAE Technical Paper 2011. safety features. in: Transportation Research Board 82nd Annual
[44] Elkady M, Elmarakbi A, MacIntyre J. The influence of a vehicle Meeting. Washington, DC: Citeseer, 2003; p. 12–16.
dynamics control system on the occupant’s dynamic response [64] Putra IPA. An average female head-neck finite element model
during a vehicle collision. Proc Instit Mech Eng, D: J Autom with reflexive neck muscles. Tech. rep. Gothenburg, Sweden:
Eng. 2012;226(11):1454–1471. Chalmers University of Technology; 2020.
[45] Elkady M, Elmarakbi A, MacIntyre J. Enhancement of vehicle [65] Howard M, Thomas A, Koch W, et al. Validation and applica-
safety and improving vehicle yaw behaviour due to offset colli- tion of a finite element pedestrian humanoid model for use in
sions using vehicle dynamics. IJVS. 2012;6(2):110–133. pedestrian accident simulations. Proceedings of IRCOBI
[46] Elkady M, Elmarakbi A. Modelling and analysis of vehicle crash Conference, 2000; pp. 101–119. Montpellier, France.
system integrated with different vdcs under high speed impacts. [66] Pak W, Meng Y, Schap J, et al. Finite element model of a high-
Central Eur J Eng. 2012;2(4):585–602. stature male pedestrian for simulating car-to-pedestrian colli-
[47] Elmarakbi A, Elkady M, MacIntyre J. Numerical analysis of sions. Int J Automot Technol. 2019;20(3):445–453.
vehicle-to-vehicle impact using vehicle dynamics control sys- [67] Meng Y, Pak W, Guleyupoglu B, et al. A finite element model
tems for collision mitigation. Int J Dyn Control. 2013;1(2): of a six-year-old child for simulating pedestrian accidents.
172–191. Accident Analysis & Prevention. 2017;98:206–213.
[48] Elmarakbi A, Zu J. Crashworthiness improvement of vehicle- [68] Ramaswamy K, Radhakrishnan J, Patham B, et al. Fast and sta-
to-rigid fixed barrier in full frontal impact using novel vehicle’s ble quasi-static bending simulations in ls-dyna: Identification of
front-end structures. Int J Automot Technol. 2005;6(5): optimal finite element model parameters. Tech. rep., SAE
Technical Paper 2016.
491–499.
[69] Numata S, Mizuno K, Ito D, et al. Validation of crush energy
[49] Ionut RA, Corneliu C, Bogdan T. Mathematical model vali-
calculation methods for use in accident reconstructions by
dated by a crash test for studying the occupant’s kinematics
finite element analysis. SAE Int J Trans Safety. 2018;6(2):
and dynamics in a cars’ frontal collision. Int J Automot
133–146.
Technol. 2017;18(6):1017–1025.
[70] Yu C, Wang F, Wang B, et al. A computational biomechanics
[50] Deb A, Srinivas K. Development of a new lumped-parameter
human body model coupling finite element and multibody seg-
model for vehicle side-impact safety simulation. Proc Instit ments for assessment of head/brain injuries in car-to-pedestrian
Mech Eng, D: J Autom Eng. 2008;222(10):1793–1811. collisions. IJERPH. 2020;17(2):492.
[51] Prochowski L, Ziubi nski M, Pusty T. Experimental and analytic [71] Xueyan Y, Yubin Q, Ying W. Establishment and verification of
determining of the characteristics of deformation and side stiff- finite element model of pedestrian head based on accident
ness of a motor car body based on results of side-impact crash reconstruction. 2019 2nd World Conference on Mechanical
tests. In: IOP Conference series: materials science and engineer- Engineering and Intelligent Manufacturing (WCMEIM),
ing, Vol. 421, IOP Publishing; 2018; p. 032025. Shanghai, China IEEE, 2019; p. 716–721.
[52] Jonsen P, Isaksson E, Sundin K-G, et al. Identification of [72] Omar T, Eskandarian A, Bedewi N. Vehicle crash modelling
lumped parameter automotive crash models for bumper system using recurrent neural networks. Math Comput Modell. 1998;
development. Int J Crashworthiness. 2009;14(6):533–541. 28(9):31–42.
[53] Ambrosio J, Dias J. A road vehicle multibody model for crash [73] Free JC, Hall JW, Montano CA. Identification of mathematical
simulation based on the plastic hinges approach to structural models from impact data: application to thoracic impact. Tech.
deformations. Int J Crashworthiness. 2007;12(1):77–92. rep., SAE Technical Paper 1976.
 INTERNATIONAL JOURNAL OF CRASHWORTHINESS 1559

[74] Ghannam MY, Clark T, Reddy Y, et al. A study of crash energy [78] Huang M. Vehicle crash mechanics. CRC Press, 2002 p.
and severity in frontal vehicle-to-vehicle crash tests. Tech. rep., 278–280.
SAE Technical Paper 2011. [79] Woolley RL. Crash pulse modeling of force limiting structures.
[75] Mnyazikwiye BB, Karimi HR, Robbersmyr KG. Mathematical Tech. rep., SAE Technical Paper 2008.
modeling of vehicle frontal crash by a double spring-mass- [80] Cheng Z. Analysis of automobile crash responses using wave-
damper model. 2013 XXIV International Conference on lets. SAE Trans. 2002;285–291. United States
Information, Communication and Automation Technologies [81] Varat M, Husher SE. Crash pulse modelling for vehicle safety
(ICAT), Sarajevo, Bosnia and Herzegovina IEEE, 2013. pp. 1–6. research. Proceedings: International technical conference on the
[76] Munyazikwiye BB, Karimi HR, Robbersmyr KG. Application of enhanced safety of vehicles, Vol. 2003. National Highway
genetic algorithm on parameter optimization of three vehicle Traffic Safety Administration, Washington, DC United States
crash scenarios. IFAC-PapersOnLine. 2017;50(1):3697–3701. 2003; p. 9.
[77] Usta F, Eren Z, Kurtaran H, et al. Crashworthiness optimization of [82] Wei Z, Karimi H, Robbersmyr K. A model of vehicle-fixed bar-
nested and concentric circular tubes using response surface meth- rier frontal crash and its application in the estimation of crash
odology and genetic algorithm. Lat Am j Solids Struct. 2018;15(5). kinematics. ESV Conference. Seoul: South Korea, 2015.