Parallel computation algorithms for multibody dynamics simulations

Abstract

Flexible multibody dynamics simulations have been performed sequentially on a single processor because the problem sizes for the simulations were not large. How-ever, advanced designs of rotor blades or CSD/CFD (Computational Structural/Fluid Dynamics) coupled problems call for more stringent accuracy requirements and faster computations in multibody dynamics simulations. For parallel computations, a novel non-overlapping domain decomposition method is developed and implemented to perform flexible multibody dynamics simulations in parallel. Non-overlapping domain decomposition methods such as classical substructuring methods and finite element tearing and interconnecting (FETI) methods are also reviewed and compared to see how they have been developed and improved for better domain decomposition. The proposed domain decomposition approach with a localized version of Lagrange multiplier technique and an augmented Lagrangian formulation in conjunction with the Lagrange multipliers, is formulated and discussed in detail. Within the frame-work of direct solvers, the solution procedure with LU factorization and forward and backward substitutions has been designed for parallel computations. The actual implementation of the parallel algorithm with the domain decomposition method on a finite-element-based multibody dynamics simulation program (Dymore), is also described. Finally, the parallel algorithm is tested on parallel hardware with numerical experiments to evaluate the accuracy and scalability of the algorithm for various domain decomposition cases.