| In this dissertation, two large-scale computational acoustic applications are studied. One is the disk brake squeal prediction and the other is the silencer performance prediction. These two applications will involve large-scale models that cannot be fit in a single computer and/or take too much time to compute.; For the disk brake system squeal prediction application, the finite element method (FEM) is used. A general formulation of the dynamic system equations is presented. Due to the friction mechanism, the matrices in the dynamic system equations are unsymmetric. Furthermore, for the system with damping, a quadratic eigenvalue problem (QEP) has to be solved. The linearization method converts the QEP into a generalized eigenvalue problem but the problem size is doubled. Since the size of the matrices is so large, the direct eigenvalue solver such as the QZ method cannot be applied. An adaptive block Lanczos method for non-symmetric eigenvalue problem is therefore used to solve for eigenvalues. Numerical experiments show that the ABLE algorithm is effective and efficient to extract the first a few eigenvalues of large-scale sparse unsymmetrical matrices.; In the application of silencer performance prediction, the transmission loss (TL) needs to be evaluated. Since silencers used in industry usually have complicated internal components, the direct mixed-body boundary element method (BEM), which can mesh each component independently, is used. However, the numerical models may become too large to fit in a single computer and/or the simulation may take too much time. Three solutions to overcome this difficulty are presented. The impedance matrix synthesis method is used for multiply connected exhaust network systems. For more complicated models, the substructuring idea can be employed. Finally, the multithread and vector BEM computation on multi-processor PC workstations is described. Numerical examples show that these methods are effective to predict the performance of large-scale silencers. |
