Large Scale NVH analyses for General Motors Using Cray SV1

MSC.Nastran V70.7 was used on the Cray SV1 to perform large-scale NVH problems typically encountered in the automotive industry. Several MSC.Nastran solution sequences such as SOL 103, 111, 108, and 107 are heavily used by automotive customers in a production environment on the Cray SV1. Significant performance improvements were recently made to key MSC.Nastran kernels to exploit the architecture of the Cray SV1 machines. Special attention was given to the normal modes (eigenanalysis) and frequency response modules invoked in typical NVH analysis. For example, proper blocking in the matrix decomposition and matrix multiply kernels allows most problems to achieve 75% of peak within these kernels by taking advantage of data reuse from cache. Suitable blocking allows maximal data reuse from cache and avoids bandwidth limitations. In addition, optimal settings of numerous MSC.Nastran parameters were established for a wide variety of NVH problems. Major improvements in performance were demonstrated for a collection of real life large scale problems provided by General Motors. Results clearly indicate that the Cray SV1 is a robust cost effective NVH engine well suited for a production environment in the automotive industry. Automotive NVH Analyses Noise, Vibration and Harshness (NVH) analysis is a key automotive industry application. Full vehicle simulation, acoustics and frequency response calculations are routinely performed in order to design vehicles with improved ride and handling characteristics. NVH analysis is used to predict the response of a structure to imposed excitations or loads given appropriate boundary conditions. The sources of dynamic excitation can be external or internal to the vehicle [1]. External forces include road-induced shake/noise and aerodynamic effects. Internal forces include power-train reaction forces, tire/wheel imbalance and brake-induced forces. Using full vehicle, low frequency NVH models, these forces are used to predict and assess, for example, a vehicle’s shake and boom response at various locations in the vehicle. Typical NVH problems are computationally intensive because of the size of the models and the wide range of excitation frequencies required. For NVH optimization, forced response analysis and sensitivity calculations must be performed at hundreds of excitation frequencies. In a detailed full vehicle simulation, problem sizes on the order of 2-3 million degrees of freedom are used in practice. If modal frequency response is used, the number of eigenvalues (modes) extracted can regularly exceed 1000. Such large I/O intensive jobs require powerful supercomputers with sufficient memory and fast I/O. Cray SV1 computers are extensively used in the automotive industry for large NVH simulations using MSC.Nastran. Dramatic improvements in overall performance due to recent architecturespecific tunings of key components have greatly reduced turnaround time for large-scale NVH computations, enabling automotive engineers to reduce the length of the product design cycle.