CRADIPOR: Crash Dispersion Predictor
RRAE detects numerical dispersion in FE crash models, saving 100x compute time.
CRADIPOR, developed by Edgar Chaillou, Sebastian Rodriguez, Yves Tourbier, and Francisco Chinesta, tackles a critical problem in automotive crash simulations: numerical dispersion. Finite Element (FE) crash models are widely used in vehicle development, but their results aren't strictly repeatable due to parallel computation and model complexity. This variability complicates engineering decisions, as performance criteria evaluated during post-processing may show significant numerical dispersion. Traditionally, teams would rerun the same simulation multiple times to estimate dispersion, but that's too computationally expensive for routine use.
CRADIPOR's solution combines a Rank Reduction Autoencoder (RRAE) with supervised classification to identify regions sensitive to numerical dispersion directly from a single simulation's output. The RRAE learns a compressed latent representation of the simulation data, and a classifier then flags areas where dispersion is likely high. In tests, the RRAE-based approach outperformed a Random Forest baseline. Among various input representations—including wavelet-based and slope-based signals—slope variations provided the best classification accuracy. This means engineers can trust their simulation results without rerunning them, saving substantial compute time and accelerating development cycles for safer vehicles.
- CRADIPOR uses a Rank Reduction Autoencoder (RRAE) to detect numerical dispersion in FE crash models without repeated simulations.
- Slope-based input signals achieved the best classification performance, outperforming wavelet-based and Random Forest baselines.
- The tool enables reliable engineering decisions by identifying dispersion-prone regions during routine post-processing, eliminating costly recomputations.
Why It Matters
CRADIPOR could cut crash simulation costs by eliminating redundant reruns, accelerating safer vehicle design cycles.