Reducing calculation times for crash simulations

A research project at TECOSIM aimed to create a process for reducing calculation times for crash simulation. The project was funded by the German Federal Ministry of Economics as part of its Central Innovation Program SME. The company's Munich location produced a number of methods and models to establish the optimum potential for savings. The result was surprising: calculation times can be reduced by up to 25 percent.

Numerical simulation of different crash scenarios plays an increasingly more important role in vehicle development. Highly dynamic, sometimes non-linear processes are reproduced using special finite element (FE) programs. Although current vehicle crash simulation models already feature around three million elements, there is a growing desire for greater detail in crash models. This is because many components which are currently reproduced to a limited extent have a considerable influence on crash behaviour.

Greater detail requires longer computing times

However, greater detail also requires an increase in computing capacity. Depending on the code and hardware, current models take between 24 and 48 hours to compute at a time step of 1 nanosecond. The time step would need to be reduced to between a tenth and fifth to reproduce cast parts clearly. This, in turn, may cause up to a ten-fold increase in calculation times. The research project thus aimed to develop a process which could reduce the computing time for crash calculations to meet the need for an increase in model definition or at least keep times constant. Development was focused on explicit processes which are applied during what are known as short-time mechanical forces (crash test, fall test) and during extremely non-linear effects, such as contact, substantial expansion and displacement.

From rigid body to fully-fledged components

First of all, research was carried out using simple analogous models. Sections of the model were initially incorporated as rigid bodies. These require considerably less computing capacity since there is no need to calculate internal deformation. Once the crash obstacle was reached, these sections were rendered deformable again by removing the rigid body.

In the second half of the project, criteria were developed to automatically switch from rigid bodies to fully-fledged component models during a crash calculation as soon as they are exposed to force and thus play a role in crash analysis. The initial idea was to recommence calculation with an ABAQUS restart analysis each time. However, the characteristics of an element cannot be changed in ABAQUS during such an automatic stop and restart. To address this problem, an external routine was developed to control this change from a rigid model to a deformable one using a filter function and the script that it is based on. The rigid bodies are each eliminated from the process after stress in the adjacent deformable area exceeds a certain value.

The starting basis (partial model)

Partial model with rigids in different colours Removed in the sequential order orange, grey, blue, green.

Reduction in computing times for a partial vehicle model

Used for impact against a rigid barrier at 56 km/h, a partial vehicle model was divided into several rigid sections to investigate the time which could be saved for this model. In a first step, these sections were eliminated after a certain time according to a termination criterion described above. To do so, first a reference model was calculated to provide an overview of the total calculation time and a meaningful definition of the rigid body. A reference calculation was then used to establish fixed computing times for the initial evaluation of potential. The calculation was stopped at relevant points in time to remove the next rigid body. Calculations were interrupted based on a termination criterion and the next rigid body removed and automatically restarted.

Different variants were tested to derive a termination criterion based on the forces exerted to convert the rigid bodies into "deformable" structures without losing computational accuracy. The project managed to shorten computing times by almost 19 per cent (see Table 1). The fastest version was then used to simulate a complete vehicle.

Base model fully deformable

Model divided up into rigid bodies, rendered deformable in succession one after another.

Reduction in computing times for a complete vehicle

Calculation for complete vehicles was performed with ABAQUS Version 6.12.3 on 8 CPUs with 12 GB of memory. The model featured around 520,000 nodes and about 2.8 million degrees of freedom. The findings show that the rigid body approach also provides a very promising reduction in computing times for larger models.

As usual with automotive manufacturers, the model was subdivided into different includes and was set up for front, rear and side crash studies. At the beginning, each include featured as a separate rigid body, which was subsequently removed on reaching its filter criterion. This produced savings in calculation times of up to ten per cent in comparison with plastic-elastic complete vehicle modelling. It should be noted that the body in white features as an include which stretches from front to back through the whole vehicle and becomes deformable in a single step. An expert rigid body partition ensures that savings are achieved similar to the partial model.

Version with termination filter impresses as fastest calculation solution

At the end of the funded project, we came to the conclusion that the calculation variant with a stress-based termination filter appears to be the most effective. It reduces computing times by around 20 per cent in a partial vehicle model and by about ten per cent in a complete vehicle. This means that the objective is achieved to a considerable extent with a targeted computing time reduction between 30 and 40 per cent. Additional potential can be achieved through additional improvements. Adept rigid body distribution and a further optimised termination criterion enable computing time savings of around 25 per cent to be achieved based on physically realistic findings.