A Full Eulerian Method for Fluid-structure Interaction Problems

An efficient and scalable numerical method for massively parallel computing of fluid-structure interaction systems has been developed for biomedical applications. To facilitate the treatment of complex geometry, a full Eulerian method is employed to couple the incompressible motions of fluid and hyperelastic materials. Instead of implicitly solving the pressure Poisson equation, a novel artificial compressibility method with adaptive parameters, which are determined to guarantee the computed field to be nearly incompressible, is proposed. In both weak and strong scaling tests, the developed solver attains excellent scalability on the K computer. A sustained performance of 4.54 Pflops (42.7% of peak) has been achieved for a microchannel flow involving more than 5 million deformable bodies using 6.96 × 1011 grid points with 663,552 compute cores. We study arteriole blood flows in a brain to gain insight into dynamic interactions among motions of plasma and blood cells, which are relevant to initial thrombus formations.