A Hybrid Simulation-Physical Data-Driven Framework for Occupant Injury Prediction in Vehicle Underbody Structures.

Area of Science:

• Engineering
• Computational Mechanics
• Vehicle Safety

Background:

• Optimizing vehicle underbody structures for blast protection faces a challenge balancing the cost of physical testing with simulation accuracy.
• Simulations often lack accuracy due to the complexity of blast impact signals and limited available data for augmentation.

Purpose of the Study:

• To introduce a predictive framework that integrates limited physical measurements with systematically augmented simulation data.
• To address the difficulty of augmenting load signals for finite element simulations with extremely small sample sets.

Main Methods:

• A small-scale data-augmentation model in the wavelet domain using a conditional generative adversarial network (CGAN).
• Introduction of real-time perturbations governed by cumulative distribution functions for data diversification.
• A Gaussian process regression (GPR) model integrating physical data and augmented wavelet characteristics (PCA-reduced wavelet scale energies) to estimate injury indices.

Main Results:

• The hybrid model, combining physical data and augmented simulations, demonstrated higher accuracy compared to simulations alone.
• Cross-validation confirmed the enhanced predictive capability of the proposed framework.
• A case study indicated that increasing hull angle and depth effectively reduces occupant injury.

Conclusions:

• The developed predictive framework offers a cost-effective and accurate solution for optimizing vehicle underbody structures against blast threats.
• The methodology successfully overcomes limitations in data augmentation for complex blast impact signals.
• Design modifications like increased hull angle and depth are validated as effective strategies for enhancing occupant safety.