GeoTransolver Framework Predicts Industrial Crash Dynamics with High Fidelity
GeoTransolver, a novel operator learning framework, showcases precise high-fidelity predictions of crash dynamics on an industrial scale. It has been tested against intricate datasets involving bumper beams and full vehicles, effectively capturing multi-scale geometric contexts while analyzing plastic deformation patterns and acceleration profiles at key occupant positions. This framework tackles the difficulties of implementing operator learning in large-scale crash assessments, where complex geometries, contact nonlinearities, and swiftly changing transient deformations are present. As a result, it serves as a practical alternative to traditional finite element solvers, which are often too resource-intensive for automotive crashworthiness optimization.
Key facts
- GeoTransolver is a geometry-aware operator learning framework with memory-efficient low-rank attention.
- It is benchmarked on complex bumper beam and full-vehicle crash datasets.
- It captures multi-scale geometric context and resolves plastic deformation patterns.
- It accurately predicts acceleration profiles at critical occupant locations.
- Traditional finite element solvers are computationally prohibitive for iterative crash simulations.
- Emerging operator learning frameworks provide rapid surrogate predictions but face challenges with industrial-scale crash analysis.
- The paper is published on arXiv with ID 2605.27758.
- Automotive crashworthiness optimization requires managing large-scale nonlinear structural deformations and energy dissipation.
Entities
Institutions
- arXiv