关键词: 电磁学, 舰船磁场, 磁矩量法, 多层自适应交叉近似, 图形处理单元

Abstract: Magnetic moment methods (MMMs) are particularly well suited to model the magnetic signatures of the ferromagnetic objects, such as ship, because they do not require the meshing of non-ferromagnetic materials like air. However the methods need the storing and solving of a fully dense matrix system. Computation takes a long time, and a large storage space is reqiured for storing the matrix. The method based on MMM and multilevel adaptive cross approximation (MLACA) leads to a significant reduction of the required memory and computation times. But the computation times are still too long for the resolution of very fine meshed geometries. The graphical processing units (GPU) with high parallel processing capability is used to accelerate MLACA with the natural parallelism, and the parallel computation strategies are presented, implementing the fast analysis of large scale MMM for magnetic signatures of ships. The results of typical examples indicate that the speedup ratio of the parallel computation of graphics processing unit (GPU) is more than 120, and the computation time for the ship hull finely meshed to 100 000 sheet elements is only about 4.3 min. The difference between the calculated results of the proposed method and the FEM commercial software is less than 1%. Thus, it provides a fast, accurate and simple method for the large scale modeling of ship magnetic signatures．

Key words: electromagnetism, ship magnetic field, magnetic moment method, multilevel adaptive cross approximation, graphics processing unit