An analysis process that reduces the time taken to run virtual crash tests from hours to minutes has been developed by Altair Engineering in close collaboration with Intel. The solution used the latest Intel software tools and compilers to optimise communications schemes and extract the best performance. An Intel cluster based on Intel microarchitecture called Nehalem was used.
Virtual crash tests are one of the most time consuming tasks in automotive development processes. Combining an approach for simulating highly dynamic events, Altair's crash solver Radioss, and Intel's expertise in high-performance computing allowed a team of research engineers to run virtual crash tests in minutes instead of hours. A frontal crash simulation of a vehicle model with more than one million elements was performed in less than five minutes.
'I am delighted to see this quantum leap in simulation performance,' said Djamal Midoun, Unibody safety manager Ford Motor Company. 'In partnership with Altair, Ford has established a development process which consistently delivers vehicles with a superior safety performance. Radioss is one of the cornerstones of this process and these dramatic performance improvements will offer us a broad set of new opportunities. In the future, this will not only help us evaluate design variants faster than ever before, but also enable us to routinely perform design sensitivity and robustness analysis, which were incredibly time consuming before.'
The challenge for Intel software engineers was to maintain scalability with a high number of cores. This was accomplished through a hybrid parallelism model implementation (MPI + OpenMP) of Altair's new algorithm. The result was achieved using an Intel cluster based on Intel Xeon processor 5560 series. Intel used the latest Intel software tools and compilers to optimise communication schemes and extract the best performance from the processor-based cluster.
'Accelerating simulation times while increasing the quality and robustness of crash computation are challenges to which we are attaching importance,' said Laurent Di Valentin, expert in numerical simulation, PSA Peugeot Citroën. 'The decrease of CPU needs would enable us to imagine ways of modelling improvements with Radioss (finer meshes, integration of better material laws with rupture, optimisations and scatterings) expected to ensure the best possible and reliable development process in our car projects.'
