Dimensionality Reduction of Crash and Impact Simulations using LS-DYNA

Automotive crash simulations of full vehicle models still constitute a large computational effort which can be a major problem for applications requiring a large number of evaluations with varying parameter configurations. In some applications, highly similar simulations frequently need to be carried out multiple times with only minor local parameter modifications. At the same time, large amounts of numerical simulation data increasingly become available in industrial simulation databases as part of the progressive level of digitalization of automotive development processes. Data-driven modeling methods are an area of active research, aiming to exploit this new “treasure” in order to find interesting patterns and accelerate predictions.