Efficient injury risk predictions for a diverse population using parametric human modeling and inducing points in...

Area of Science:

• Biomechanics
• Computational modeling
• Machine learning

Background:

• Accurate prediction of occupant injury risk in vehicle crashes is crucial for safety.
• Diverse populations exhibit varied responses to crash forces, complicating injury risk assessment.
• Limited computational resources often restrict the scope of detailed human body model simulations.

Purpose of the Study:

• To develop a method for identifying a small set of representative occupants.
• To enable accurate prediction of injury variations across a diverse population within a limited simulation budget.
• To optimize vehicle restraint systems for enhanced safety across different demographics.

Main Methods:

• Utilized parametric human modeling and machine learning (Gaussian Process) to identify representative occupants (inducing points).
• Employed a maximal projection method to sample 100 diverse occupants based on anthropometric variations.
• Conducted US-NCAP frontal crash simulations using morphed THUMS v4.1 models and validated vehicle/restraint systems.

Main Results:

• Identified 20 representative occupants (inducing points) sufficient for accurate injury risk prediction.
• IP-based surrogate models demonstrated minimal errors (<1.8%) in predicting head, chest, and lower extremity injuries compared to 100 simulations.
• Optimized restraint systems showed significant injury risk reduction (13-47%) across varying impact severities.

Conclusions:

• The proposed method efficiently generates accurate injury risk predictions for diverse populations with reduced simulation costs.
• This approach facilitates more effective restraint system optimization, enhancing safety for a broader demographic range.
• The use of inducing points represents a significant advancement in computational biomechanics and automotive safety research.