Machine learning for pre-crash occupant motion prediction

Use machine learning to develop and train a human body model to predict vehicle occupant motion during pre-crash manoeuvres.

BACKGROUND

Autoliv develop occupant restraint systems that offer superior protection of vehicle occupants in crashes. The airbags and seatbelts have significantly reduced the number of occupants killed and injured in automobile crashes. However, occupants are still injured in vehicle crashes.

 

Pre-crash manoeuvres such as braking and/or steering can influence the location of the occupant in the event of a crash. With the introduction and further development of autonomous features, such as autonomous braking or crash avoidance by steering protection systems, that intervene with the occupant prior to crash, the risk of an occupant being injured in a crash is also reduced.

 

An active finite element human body model was developed that predicts occupant motion of humans in various pre-crash manoeuvres. However, the computer run time (several days) of the model is substantial. Therefore, there is a need to develop a tool with significantly shorter turn around time. Using machine learning to “train” a human body model to predict human body motion in pre-crash manoeuvre seems to be a good solution.

 

AIM

The aim with this project is to use machine learning to develop and train a human body model to predict vehicle occupant motion during pre-crash manoeuvres.


STUDY DESIGN

  • The study will start with a literature study.
  • An evaluation of existing software suitable for the task will be carried out.
  • Based on volunteer tests and predictions of occupant kinematics in pre-crash events from simulations with the active human body model a new model will be “trained” using machine learning.
  • A parameter study will be carried out with the trained model.

SUITABILITY

The project is suitable for two students studying computer science, mechanical engineering or mathematics with an interest in machine learning. Finite element knowledge and experience is a bonus. The work will mainly be carried out at DYNAmore Nordic in Linköping.


SUPERVISOR

Jimmy Forsberg, DYNAmore Nordic AB

Email: jimmy.forsberg

 

Bengt Pipkorn, Autoliv AB Research

Email: bengt.pipkorn


Attachments
Contact and application
contact Daniel Hilding
  • Director DYNAmore Nordic AB
contact Mikael Schill
  • LS-DYNA
  • Material modeling
  • Forming Technology