GPU implementation of a shell element structural solver aimed at fluid-structure interaction problems

Bartezzaghi, Andrea
GPU implementation of a shell element structural solver aimed at fluid-structure interaction problems
Monday 22nd July 2013
Quarteroni, A.
Advisor II:
Cremonesi, M.
Parolini, N.
Download link:
The study of thin structures is very common nowadays and useful in different fields. An important example is the analysis of sail dynamics. In this context, accurate simulations of the interaction between the sail and the wind are also required. However, this kind of fluid-structure interaction problems are very computationally expensive. First objective of this thesis is the implementation of an highly efficient shell finite element structural solver, designed to run on GPU (Graphics Process- ing Unit) hardware. In order to fully exploit the GPU computational power, an explicit central difference time-advancing scheme is adopted. Domain is discretized using MITC4 shell elements in large displacements formulation, due to their adequate numerical properties and ability of avoiding shear-locking problems and simulating sail wrinkles. Techniques adopted during the development, such as algorithms, memory management and code optimizations, are described in details. Numerical tests and benchmarks are carried out and performances are compared with the commercial software Abaqus. Second objective of this thesis is the development of a partitioned strongly coupled fluid-structure interaction solver, implemented in OpenFOAM, an open-source CFD framework. The fluid dynamics problem is solved using the PISO scheme, while the solver implemented in the first part handles the structural problem. The mesh-motion process and interpolation algorithms are analyzed and implemented in GPU in order to gain performance and reduce memory requirements. Finally, results of numerical and performance tests on the developed FSI solver are reported. Keywords: fluid-structure interaction, finite element method, shell dynamics, mesh-motion, GPU parallelization, CUDA