Model Reduction of Large Scale Finite Element Models

The usage of flexible bodies in Multibody simulations (MBS) has widely increased their industrial application. Finite Element (FE) models with up to 10 million degrees of freedom (DOF) and some hundred million nonzero matrix entries are used to describe the flexible bodies. Before such a model can be included into an MBS, the number of DOF must be reduced to an appropriate size. Using modal reduction often the critical issue arises which modes to choose while Component Mode Synthesis based methods often lead to a relatively big size of the resulting model. Alternative methods using Moment Matching and Balanced Truncation can result in a smaller size while still remaining accurate enough. Sometimes these matrices are so huge that they can not even be stored in one computers main memory. The calculation of the necessary orthogonal Krylov subspaces needs an LU factorization which is also very memory intensive. To meet these requirements, distributed computation is used which also shortens the computational time of the reduced process. In this work, an industrial relevant FE model is reduced to a much smaller size using alternative methods. Accuracy is verified by comparing the frequency response in a defined frequency range of the original and the reduced model. (© 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim)