• computation eletromagnetics (CEM);
  • graphics processing unit (GPU) computing;
  • electromechanical devices;
  • integral formulation


This paper discusses the practical implementation on graphics processing unit (GPU) of a computer code for the analysis of electromechanical devices. The software is based on a low frequency integral formulation of the Maxwell equations coupled with the rigid body dynamic equations. The formulation here considered is based on the development of an equivalent network whose parameters are functions of the relative positions and of the velocities of the subparts of the device. Positions and velocities on their turn depend on the force between them that are functions of the electromagnetic quantities. The choice of the numerical methods to manage such a complex coupled problem is an open issue in the electromagnetic community. The use of multicore CPUs can reduce the computation times, but a true breakthrough is achieved by running these codes on GPUs. The practical implementation of an existing code is used as a case study for discussing a number of issues that may arise in the implementation of other GPU-accelerated electromagnetic codes. The implementation here reported has been designed for a multi-GPUs environment where the efficient cooperation between GPUs is a further aspect to be considered. The overall performance of the accelerated code has been evaluated by considering the dynamic analysis of a passive magnetic bearing. Copyright © 2012 John Wiley & Sons, Ltd.