GPU optimized multi-block-multi-mesh immersed boundary method for flows in complex arterial models

Abstract

Immersed boundary method (IBM) is widely used for simulating flow in complex geometries using structured grids. However, this entails a disadvantage when simulating internal flows through curved and bent tubes. The presence of grids outside the fluid domain leads to the wastage of memory and computational overheads. Here, we propose a multi-block-multi-mesh framework to capture the complex geometry using multiple grid blocks fitted close to the body, reducing excess grids. This also has the advantage of using different and non-uniform grid spacing in different blocks. The reduction of the grid enables encompassing bigger caseloads on a single GPU. The solver is accelerated on GPU using OpenACC, compared to sequential CPU simulations, and speedup is presented. The speedup obtained is comparable to that of large multicore systems. The framework is extensively validated for straight artery with axisymmetric stenosis and bileaflet mechanical heart valve with axisymmetric sinus. This framework then models complex arterial flows like stenosed aorta, patient-specific branched aorta, bileaflet mechanical heart valve with Valsalva sinus and aorta, and lastly, patient-specific iliac aortic aneurysm. This framework achieves a significant reduction in GPU memory requirement for complex arterial models, enabling us to perform direct numerical simulation (DNS) of the stenosed aorta and mechanical heart valve cases in a single GPU.