top of page


Mr. John Smith

Job title



The capabilities of virtual development of safety systems for occupant safety are increasing with the growth of acceptance, computational power and the usability of tools. Nevertheless, nowadays car safety systems are primarily developed using single statistical representations of humans, like the 5%-ile female or the 50%-ile male body. Hence variation of anthropometry existing in the real-world is only covered to a very limited extent.

Recently, several studies developed fast calculating models using methods of rigid body simulation or metamodeling. They suggested investigating rapid or even near-real-time prediction-models for variations of outer parameters, such as airbag characteristics or crash pulse. In contrast, the proposed paper aims to get insight into the impact of human body variation on its crash behavior. This behavior can then be related to a model of a specific percentile. The objective of this approach is to run one simulation with a single representation of a human body and get data for deviated models based on previous simulations. Therefore, the main advantage of this concept is the reduction of the simulation time, needed to get all results of the occupants with different sizes (body mass index, statue, the ratio of sitting height and statue).

To test the feasibility of the approach, a 2D rigid body system was created as a simplified model using LS-Dyna and Python. It represents a car's occupant who is restrained by a lap and shoulder belt under the influence of a crash pulse. The joint characteristics were modeled via spring, damper and friction definitions aiming for human-like behavior.

To create a database, extensive simulations were carried out by varying the dimensions of body parts, applying a Latin-Hypercube Design Of Experiments scheme. The parameter limits were taken from UMTRI's human shape database. Since the minimal model provides a very limited set of reasonable assessment channels, only kinematic responses like maximum displacement and acceleration of the head and chest were evaluated.

The obtained results seem to prove expected correlations between mass and maximum acceleration as well as forward displacement. For the model's training, different combinations of features and targets are tested. For example, the results from simulations with one body measure set are used as feature-vector, while the target-vector is formed by the kinematic characteristics of diverging human representations. After normalizing the data, various regression and machine learning algorithms were applied and their performance evaluated.

Mr. Franz Plaschkies, Technische Hochschule Ingolstadt / CARISSMA, GERMANY; Prof. Ondrej Vaculin, Technische Hochschule Ingolstadt, GERMANY

Estimation of the Impact of Human Body Variation on Its Crash Behavior Using Machine Learning Methods

F2020-PIF-051 • Paper • FISITA Web Congress 2020 • Passive and Integral Safety (PIF)


Sign up or login to the ICC to download this item and access the entire FISITA library.

Upgrade your ICC subscription to access all Library items.

Congratulations! Your ICC subscription gives you complete access to the FISITA Library.


Retrieving info...

Available for purchase on the FISITA Store


bottom of page