Scatter Analysis of Crash Simulation Results Enabled by Data Compression

In crash simulation, small changes of the model or boundary conditions may result in substantial changes of the simulation results. For the Neon test case [3], small variations of the barrier position result in substantial scatter of the intrusion. Detailed investigations of several models have shown that in some cases numerical effects might be responsible for the scatter in the results. In most cases, however, the instable behaviour of the simulation results is caused by bifurcations. These bifurcations result from numerical algorithms or are a feature of the car design. In the Neon model the scatter is a result of the interaction between the axle and the engine block. DIFF-CRASH1 is a tool, which allows one to measure scatter and to trace this scatter back to its origin. It allows the engineer, to understand the mechanisms of propagation and amplification of scatter during the crash itself as a basis for the improvement of the stability of the car design. For this analysis, DIFF-CRASH uses the complete result files of several simulation runs with a fine time resolution of the states. Storing these result files requires a substantial amount of disk space. FEMzip2 allows the reduction of this disk space by a factor of between 5 and 10. One can then store not only key results but the complete result files from optimisation experiments which can also be used for stability analysis. In this paper we discuss the accuracy required by DIFF-CRASH for a precise analysis and its implication on the data compression performance of FEMzip.