Mathematical Modelling of Vehicle Crash Safety: Predicting
Structural and Material Deformation
Modern vehicle crash safety relies heavily on mathematics. Before prototypes are built, we
predict how a vehicle's structure will behave during a variety of high impact collisions,
allowing engineers to design safer cars. Physical crash tests, while accurate, are expensive,
time consuming, destructive and limited in scope. Mathematical models, particularly PDE’s
(Partial Differential Equations), FEA’s (Finite Element Analysis) and Matrix Manipulation are
ways that engineers can model material stress/strain, and deformation in complex
geometries under real crash conditions (Silva et al., 2024)
Recent developments in crashworthiness show how mathematical optimisation and
advanced numerical methods improve both accuracy, increase scope and computational
efficiency. Studies range from using finite element analysis models to predict the behaviour
of specific materials shaped in specific ways by involving a set of physically meaningful fibre
and matrix failure strain parameters, all the way to using matrices derived from CAD
geometry models in the monocoque chassis of Formula 1 cars. These advances made
virtual crash testing a standard part of the automotive design process (Jonsson and Kajberg,
2023). This case study explores how PDE-based FEA models represent vehicle deformation
during a frontal impact. Additionally, we will be looking at how topology optimisation and
surrogate modelling are used to enhance crash performance
During the past couple years, many ways of modelling have been used to predict material
movement in collisions. Within high speed collisions, vehicle structure can be treated as a
deformable continuum that obeys Newton’s
second law in spatial form, where u(x,t) is
displacement, ρ is density, σ is the stress tensor
and f represents external forces. To close this
system, constitutive relationships are required,
e.g. where for smaller deformations, Hooke’s
Law provides a linear relationship between
stress and strain.
σ = C: e(u), e(u) = ½ (Δu + Δu^T), where C is the elasticity tensor. However, crash scenarios
involve high strain rates and large plastic deformation. Therefore, nonlinear constitutive laws,
incorporating strain aqrate hardening and damage evolution, are applied to represent ductile
failure (Jonsson and Kajberg, 2023)
Now what exactly are these “tensors”? Tensors represent multidimensional tables, for
example a stress tensor would be a matrix of all the different forces in three directions, shear
and normal stresses. If we put these equations into a simulation we would see real 3D
deformation or crash scenarios. Without tensors, the simulation could only handle 1D
tension.
The finite element discretises the structure into small elements each with approximated
displacement fields. Substituting these into PDE’s yields a matrix system as shown below,
where M is the mass matrix, C is the damping matrix
and K(U) is the stiffness matrix (possibly non linear)
and F(t), the applied impact force.
Crash stimulations often employ explicit time
integration, stepping forward in small time
increments due to rapid transient dynamics.
Transient dynamics is behaviour that changes very quickly in time, so when a car crashes,
huge forces change rapidly, material behaves non linearly, and parts come into contact with
each other, all in milliseconds. Explicit time integration is used as it calculates the next step
directly from known quantities, it’s simple and handles the nonlinear, contact-heavy
dynamics of crashes well - without having to solve a matrix system.
Energy conservation is a key check on numerical accuracy. The absorbed kinetic energy
E_abs is given by:
where F(s) is the crash load as a function of
displacement s.
To reduce mass and maximise energy absorption, topology optimisation redistributes
material within the vehicle’s crash box. Topology optimisation is a method used to figure out
where material should or shouldn’t be placed to make its structure ideal e.g. light, strong,
efficient.
The computer divides your part into lots of tiny elements (like in a finite element mesh).
Then it assigns a variable to each element, this represents how much material is there and
its between 0 and 1 and then we can set up an objective function which tells the solver what
we are trying to achieve. E.g. Minimise: Complicance (flexibility) Maxmimise: Energy
Absorption
and a constraint that limits the total volume (so the structure doesn’t use too much material),
which was this bit of the formula.
Because full finite element simulations are computationally expensive, surrogate models,
often machine learning algorithms trained on simulation data, approximate the relationship
between geometry and crash performance. Graph based neural networks can predict
deformation and intrusion with less than 1 % error compared to full models (Li et al., 2025).
A typical frontal crash simulation models the car’s front crash box striking a rigid barrier at
high speed. The geometry is meshed into thousands of finite elements representing metal
panels and weld joints. Boundary conditions constrain the rear nodes, while impact velocity
defines the initial condition. The PDE system is solved using explicit integration until the
structure reaches maximum deformation (Bendjaballah, Sahli and Barrière, 2025).
The displacement field u(x,t) shows localised folding near the crumple zones, where energy
is intentionally dissipated. Stress maps (σ_xx, σ_yy) reveal regions approaching material
yield limits, validated against experimental strain gauges.
The area under the load displacement curve quantifies the total energy absorbed before
cabin intrusion. For the optimised crash box in Silva et al. (2024), this value increased by 17
% compared to the baseline, while mass decreased by 12 %.
Within topology optimisation, results showed optimisation reallocated material away from low
stress zones, producing a honeycomb like structure that distributed impact forces more
evenly.
Figure 3: Workflow of topology optimisation for crash box design, combining finite element
analysis, discrete optimisation, and additive manufacturing (TOBB ETU, 2019).
A comparable mathematical approach is used in Formula 1, where engineers employ finite
element modelling and topology optimisation to refine carbon fibre crash structures.
Following the 2022 season, several teams used FEA based optimisation to address chassis
porpoising and front impact resilience simultaneously. The FIA’s regulated frontal crash
structure, which must absorb 15 kJ of energy in a 14 m/s test, is now designed using
coupled PDE–FEA simulations similar to those described here. Optimisation studies showed
that by adjusting internal rib geometry and layering angles, teams could reduce peak
deceleration by over 10 % without adding mass (FIA Technical Report, 2023). This
demonstrates that mathematical modelling not only improves vehicle safety but also
enhances performance under the strict weight constraints of elite motorsport.
Machine learning surrogate results were proven efficient through Li et al. (2025) who trained
a graph neural network on 1,000 simulated impacts to predict deformation fields for new
geometries. The surrogate reproduced FEA outputs in milliseconds, enabling rapid iteration.
Jonsson and Kajberg (2023) validated numerical predictions using high-speed imaging and
3D digital image correlation (DIC) to track deformation during physical crashes. Numerical
strain distributions matched experimental observations within 5 %, confirming model
reliability.
Future mathematical development in crash simulation focuses on multi-scale modelling,
linking microstructural material response to macroscopic deformation + uncertainty
quantification. Probabilistic models can express variations in weld strength, manufacturing
tolerances, and impact angle as random variables within the PDE framework, producing
confidence intervals for safety margins.
Moreover, coupling finite element solvers with machine learning enables adaptive meshing,
where the solver dynamically refines regions with high strain gradients. This combination of
rigorous mathematics and data driven approximation may soon allow fully virtual
certification, reducing physical crash tests to final validation stages only (Nabian et al.,
2025).
In summary, the mathematical modelling of crash dynamics rooted in PDEs, finite element
approximation, and optimisation, has transformed automotive safety. As computing and
mathematical techniques advance, the predictive power of these models will continue to
save lives by preventing catastrophic structural failure long before the first car is built.
References
Bendjaballah, D., Sahli, M. and Barrière, T. (2025) ‘Modeling and numerical simulation of car
frontal crash test using finite element method’, Journal of Materials Science: Materials in
Engineering, 20, 108.
Jonsson, S. and Kajberg, J. (2023) ‘Evaluation of crashworthiness using high-speed
imaging, 3D digital image correlation, and finite element analysis’, Metals, 13(11), 1834.
Li, H., Zhao, Y., Zhou, H., Pfaff, T. and Li, N. (2025) ‘A new graph-based surrogate model for
rapid prediction of crashworthiness performance of vehicle panel components’, arXiv
preprint, arXiv:2503.17386.
Nabian, M. A., Chavare, S., Akhare, D. et al. (2025) ‘Automotive crash dynamics modeling
accelerated with machine learning’, arXiv preprint, arXiv:2510.15201.
Silva, C. J. G., Lopes, R. F. F., Domingues, T. M. R. M. et al. (2024) ‘Crashworthiness
topology optimisation of a crash box to improve passive safety during a frontal impact’,
Structural and Multidisciplinary Optimization, 68, 1–19.
Fédération Internationale de l’Automobile (2022) FIA Safety and Technical Innovation Report
2022: Advancements in Motorsport Safety. Geneva: FIA Press Office.
Fédération Internationale de l’Automobile (2023) Formula 1 Technical Regulations: Appendix
L – Frontal Impact Test Requirements. Geneva: FIA.
Toso, M. and Tombazis, N. (2023) ‘Crashworthiness of Formula 1 monocoques: recent
developments in finite-element and composite modelling,’ FIA Safety Research Journal,
4(2), pp. 41–53.
Anderson, J. D. (2022) ‘Modelling aerodynamic oscillations in ground-effect Formula 1 cars
using coupled fluid–structure equations,’ Journal of Vehicle Engineering and Dynamics,
60(9), pp. 1123–1137.
Hurst, A. and Taylor, R. (2024) ‘Application of coupled CFD–FEA methods to carbon-fibre
crash-structure optimisation in Formula 1,’ Proceedings of the Institution of Mechanical
Engineers, Part D: Journal of Automobile Engineering, 238(5), pp. 925–938.
TOBB ETU Mechanical Engineering Department (2019) Lattice Crash Box Design: Senior
Graduation Design Project. Ankara: TOBB University of Economics and Technology.