Keywords: road safety, road safety barrier, crash simulation, nonlinear dynamic analysis, LS-DYNA, MPP.

To reduce the development and testing costs of new safety barrier designs it is advisable to use computational crash simulations for early evaluation of safety barrier behaviour under test vehicle impact. That way the impact severity parameters and barrier deformation can be estimated well in advance of a real crash test and possible design changes can be easily evaluated. Barrier designs of different geometries with different sheet metal thicknesses and additional safety elements can be analysed and evaluated to determine the most appropriate design.

However, computational crash simulations, based on explicit direct time integration of the governing equation of motion, require substantial computational resources to complete in a reasonable time. To assure the stability of the explicit direct time integration method and to capture the impact behaviour correctly it is necessary to use time steps smaller than 1μs. Reasonable computational times can only be achieved by running the simulations on supercomputers or much cheaper MPP computing platforms.

This paper reports on scalability, accuracy and stability of running a parametric barrier crash test of a specific road safety barrier design with LS-DYNA on a MPP computing platform. The crash simulations were performed with the Linux MPP LS-DYNA Version 970 running on a PC cluster. Simulations were done for the first 0.5s of the impact. To evaluate the scalability and accuracy of safety barrier crash simulations the same FE model was used for simulations performed on 1, 2, 3, 4, 6, 8 and 12 nodes.

The finite element (FE) computational model for the safety barrier crash simulation consists of the barrier and impacting vehicle. The barrier parts are modeled in compliance with their design, utilizing necessary simplifications where appropriate. The impact vehicle model is based on a model from publicly accessible vehicle library at the National Crash Analysis Center [1].

Very good results are obtained concerning the stability and accuracy, since relative deviation of simulation results is lower than 5.0%. There are no stability issues performing the simulations on different number of nodes. This confirms the suitability of MPP LS-DYNA for reliable safety barrier crash simulations.

The scalability of performed simulations was well below the expectations, since the efficiency of only about 40% was observed for simulations run on 12 nodes. The reason is relatively long simulation time of road barrier crash test, during which the sliding contact between the barrier and the vehicle spans several subdomains. This results in change of the FE model subdomain characteristics and change of the load balance of individual nodes, which increases the required inter-node communication during contact solution and cluster node updating computational phases. The efficiency could be increased by using different methods of decomposition [2] or by merging the contact definitions into a single contact interface [3].

1

Finite Element Model Archive [World Wide Web], National Crash Analysis Center, available at: http://www.ncac.gwu.edu/vml/models.html, 14.3.2006.

2

Y. Roh, "Scaling Study of LS-DYNA MPP on High Performance Servers", in "6th International LS-DYNA Users Conference", Michigan, 17, 25-34, 2000.

3

A. Sugavanam, G. Li, "LS-DYNA Communication Performance Studies", in "8th International LS-DYNA Users Conference", Michigan, 12, 1-9, 2004.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £105 +P&P)