A parallel dynamic overset grid framework for immersed boundary methods

TL;DR

This paper presents a parallel dynamic overset grid framework for immersed boundary methods, enabling efficient, accurate simulations of complex moving bodies in fluid flows with significant computational speedup.

Contribution

It extends previous fixed-grid methods to dynamic, moving overset grids with an efficient parallel assembly and interpolation, improving accuracy and computational efficiency.

Findings

Achieves 2nd order accuracy in simulations.

Reduces computational cost compared to single grid approaches.

Demonstrates good parallel speedup and efficiency.

Abstract

A parallel dynamic overset framework has been developed for the curvilinear immersed boundary (overset-CURVIB) method to enable tackling a wide range of challenging flow problems. The dynamic overset grids are used to locally increase the grid resolution near complex immersed bodies, which are handled using a sharp interface immersed boundary method, undergoing large movements as well as arbitrary relative motions. The new framework extends the previous overset-CURVIB method with fixed overset grids and a sequential grid assembly to moving overset grids with an efficient parallel grid assembly. In addition, a new method for the interpolation of variables at the grid boundaries is developed which can drastically decrease the execution time and increase the parallel efficiency of our framework compared to the previous strategy. The moving/rotating overset grids are solved in a non-inertial frame of reference to avoid recalculating the curvilinear metrics of transformation while the background/stationary grids are solved in the inertial frame. The new framework is verified and validated against experimental data, and analytical/benchmark solutions. In addition, the results of the overset grid are compared with results over a similar single grid. The method is shown to be 2nd order accurate, decrease the computational cost relative to a single grid, and good overall parallel speedup. The grid assembly takes less than 7% of the total CPU time even at the highest number of CPUs tested in this work. The capabilities of our method are demonstrated by simulating the flow past a school of self-propelled aquatic swimmers arranged initially in a diamond pattern.