Machine Learning-Accelerated Virtual Testing for Automotive Head Impacts

Purpose and goal

Virtual testing based on finite element (FE) models is driving the transformation of vehicle assessment and safety research to enhance the protection of road users. However, the FE simulation is notoriously time-expensive, resource-intensive, and exclusively accessible to FE-skilled specialists. This project aims to develop a data-driven, machine learning (ML) model of the human head with rapid and reliable brain strain prediction across diverse automotive impacts to accelerate virtual testing.

Expected effects and result

We will deliver a highly efficient, highly accurate, and widely applicable machine learning (ML) model to enable instantaneous and accurate estimation of brain responses and new knowledge of brain biomechanics relevant to current and future road users. It will contribute to a paradigm shift in automotive safety assessment from the current finite element method-based, time-consuming evaluations to ML-accelerated virtual testing with drastically improved efficacy.

Planned approach and implementation

This project will leverage a data-driven, cost-effective machine-learning method to rapidly and reliably estimate brain responses under known impact conditions, replacing the time-consuming, resource-demanding finite element simulations. The project will last for 4 years. Its implementation will be facilated via close collaboration between the Kungliga Tekniska högskolan and Autoliv. Volvo Cars and European New Car Assessment Programms will serve as reference groups.