Meeting the NVH computational challenge: Automated multi-level substructuring

Noise, Vibration and Harshness (NVH) analysis has presented a formidable computational challenge in the last several years. Accurate prediction of automobile frequency response at higher frequencies requires FE models which have millions of DOF. For obtaining frequency response over a broad frequency range, only the modal frequency response approach has been practical, and only vector supercomputers have been capable of solving for natural frequencies and modes in an acceptable amount of time. However, these machines are very expensive, and their future availability is uncertain. In the automated multi-level substructuring (AMLS) method, a large FE model is automatically divided into many substructures, and the model is transformed so that response is represented in terms of substructure eigenvectors. The numerical example demonstrates that the frequency response of a 2.9 million DOF FE car body model can be obtained more quickly using a single workstation processor with AMLS than it can be obtained using a supercomputer and the modal frequency response approach.