Ms. Thesis: Graph Neural Networks for Occupant Crash Simulation

Autoliv uses finite element simulations to assess occupant protection, but their computational cost limits the range of crash scenarios that can be studied. This thesis will investigate whether a contact-aware graph neural network can learn occupant responses in one vehicle interior and predict them in another without retraining. The work will use an existing dataset of sled simulations with crash test dummies.

The assignment

• Develop a prototype graph neural network that predicts crash test dummy kinematics and contact with a sled interior.
• Evaluate prediction accuracy across occupant sizes and crash load cases.
• Train the model on one vehicle interior and quantify its performance on an unseen interior.
• Investigate which interior environments the model can represent and how deformation affects its accuracy.
• Identify the outputs that can be predicted reliably and document the method’s limitations.

Who we are looking for

• Two students undertaking a 30 ECTS master’s thesis.
• Experience in graph neural networks, machine learning and scientific computing.
• Interest in automotive safety, finite element analysis and computational mechanics.

What we offer

• Access to an existing Autoliv dataset of sled simulations with crash test dummies.
• An opportunity to investigate transferable machine learning models for occupant safety.

Practical details

Location: Vårgårda or Gothenburg. Start: December 2026. Scope: 30 ECTS, for two students. Contacts: Mohammad Rad, RDE AI Lead Engineer, mohammad.rad@autoliv.com; Henrik Balague, CAE Restraint Systems Lead Engineer, henrik.balague@autoliv.com.