NVH Optimization on NEC Supercomputers Using MSC

NVH (Noise, Vibration and Harshness) Optimization has now gained popularity in driving the automotive and aerospace design process using frequency response analysis of detailed full vehicle structural-acoustic models. Usual design targets include minimization of structure weight, the adjustment of fundamental eigenmodes and the minimization of acoustic pressure at selected interior locations. Typical NVH Optimization analyses require considerable computational resources, both in terms of cputime as well as memory. The availability of state-of-the-art high performance hardware coupled with software advances and improved methodology has made it possible to solve complicated NVH dynamic response problems very efficiently. With the introduction of Adjoint Sensitivity Method in V70, MSC/NASTRAN has been increasingly used to perform large NVH Optimization analyses that were unimaginable with the Direct Method. The Adjoint Method usually requires a fraction of the computer resources to produce sensitivity coefficients as compared with the Direct Method. However, for extremely large NVH Analyses, the sensitivity calculations are very demanding in terms of cputime, especially in the context of vector machines. The calculation of sensitivity coefficients is done inside the DSADJ module of MSC/NASTRAN. For V70.5 and V70.7, NEC initiated a project with MacNealSchwendler Corporation to address the performance bottleneck in the DSADJ module of MSC/NASTRAN. The performance enhancements to the DSADJ module are available exclusively on NEC SX-4 and SX-5 Series Supercomputers for a limited time period. The DSADJ module was totally redesigned to improve vectorization and to exploit the vector architecture of NEC SX-4 and SX-5 Series Supercomputers. In this paper, aerospace as well as automotive customer datasets are presented to demonstrate the performance improvements of the DSADJ module. Dramatic improvements in the DSADJ module resulting in approximately 8-9 fold performance improvement as compared with V70 were observed for NVH Optimization. With the tremendous improvements in the performance of the DSADJ module, and the fact that the eigenvalue analysis involved in NVH Optimization is inherently highly vectorized, NEC Supercomputers are ideally suited for running large NVH Optimization Analyses using MSC/NASTRAN.