DIFF-CRASH analyzes the stability of crash simulations in the automotive industry. Physical bifurcations in automobile design and numerical instabilities in simulation packages account for an extremely sensitive dependency of simulation results on even minor changes to the model. DIFF-CRASH localizes these critical structural areas and offers suggestions for greater stability.
Stability Analysis for Simulation Results
DIFFCRASH is a software package for the stability analysis of crash simulations. Physical bifurcations in automobile design and numerical instabilities in simulation packages often cause extremely sensitive dependencies of simulation results on even the smallest model changes. Among other things, DIFFCRASH allows the designer to find these critical structural regions and offers suggestions for their removal.
Comparison of the node positions of parallel simulation runs on 32 SP-2 CPUs (each curve compares two simulations).
In collision simulations, quite often there are large deviations among the calculated results. These deviations are caused by rounding errors which either are incorporated systematically into the initial data or which result randomly from parallel computations. These errors can propagate either because of the model properties of because of the numerical algorithms used in the code.
The figure shows a comparison of the results of simulations attained with PAM-CRASH. A BMW model with 60,000 elements was calculated on a parallel computer with distributed memory. Deviations simular to the one shown were also observed on SMP architectures.
In order to investigate these effects, SCAI, together with ESI and the BMW-group developed DIFF-CRASH.
Flow chart for DIFFCRASH
Processes the results of several simulation runsDIFFCRASH
- Delivers a statistical analysis of the whole model as well as of individual nodes
- Generates data formats which can be used by the VR-visualisation system developed by FHI-IMK
- Incorporates data into PAM-CRASH's simulation results so that they can be analysed with PAM-VIEW
Examples from Industrial Practice
Figure 1 shows typical results from DIFFCRASH simulations, which are displayed on the Institute's Responsive Workbench. The colours show various variations in the simulation results.
SCAI has analysed larger variations in simulation results on a real BMW test model. It turned out that in many cases an edge-to-edge detection would have led to the correct simulation result. On a model which was reworked with CONTACT 46, substantially reduced variations were observed. Using the correct contact model influenced the simulation results.
Figure 5 shows variations of the results in the drivers foot area in the reworked model. This was perhaps due to the instability of a certain part of the motor carrier. Figure 6 shows a more detailed analysis: The variety of variations after a simulation run of 35 milliseconds was quite noticeable.
Figure 1+2: Various simulation results highlighted in colour.
Figure 3+4: Various simulation results after 20 milli seconds. Results are for a model with and without using CONTACT 46.
Figure 5: A view from the back of the automobile into the foot area shows different simulation results.
Figure 6: Various simulation results of the left motor carrier 35 milliseconds after the collision.