On a way to save memory when solving time domain boundary integral equations for acoustic and vibroacoustic applications

Abstract

Solving acoustic equations in the time domain, possibly coupled with the description of flexible structure dynamics, remains attractive as compared to solving the same in the frequency domain: this allows for better consideration of local non-linearities (acoustics/structure), and the boundary integral formulation (also known as BEM) offers an exact description of the infinite acoustic field based on a simple surface mesh (no need for 3D-volume discretization). Some issues remain however: the required memory space and computation time continue to grow rapidly when the number of elements of the surface mesh increases. In the case of a structure with a regular non-slender shape, the computational cost, measured in terms of required memory space, varies by Helmholtz number to the power of 4. This paper illustrates how the accelerating method called NGTD helps overcome this difficulty. This paper shows the applicability of 2 level NGTD to acoustic and vibroacoustic problems described solely by the hypersingular formulation for surfaces. It goes into more detail on some important aspects of the interpolation process and on the memory saving obtained. Implementation within the MOT (“March-On-Time”) ASTRYD code shows the benefits of this method. The memory requirement shows an estimated trend lower than power 1.35 of the number of surface elements.