SPH NUMERICAL MODELLING OF WAVE-STRUCTURE INTERACTION

The present thesis aims at contributing to a better understanding and modelling of flow driven by wave propagation approaching artificial or natural structures. From a physical point of view, a simplistic descrip- tion of the interaction of sea waves with a coastal structure can be given as the result of all processes felt by the waves which, propagating from offshore to nearshore, impact on the structure itself. Indeed, as sea waves approach a shore, the motion they generate deep down begins to interact with the sea bed. This slows the waves down and causes the crests in a series of waves to bunch up; this effect is called shoaling. The period of the waves does not change, but their height increases as the energy each contains is compressed into a shorter horizontal distance, and even- tually they break. When the wave breaking occurs, the flow changes dramatically because of the following phenomena: the wave energy is mainly transformed into turbulence (that becomes heat); several forces are induced to generate long- and cross-shore currents; the sand is mo- bilized and contributes to sediments transport. All these factors play a predominant role in wave-structure interactions. In order to model the wave motion and, in turn, the flow within the nearshore region, in the last decades the derivation and the appli- cation of Smoothed Particle Hydrodynamics (SPH) models have been widely investigated and developed. Indeed SPH is a widely used nu- merical procedure in Computational Fluid Dynamics (CFD), which was conceived in the seventies in order to simulate astrophysical phenomena of gas dynamics. SPH was accredited later in all scientific areas, espe- cially in engineering: indeed, today SPH is widely used both in solid and fluid mechanics. The SPH Lagrangian-based adaptivity and capability to simulate complex free surface flows, shocks and large fluids deformations, showed that SPH is one of the most appreciated among the mesh-free methods. Furthermore, the SPH methods, making use of massive par- allel computation, make them particularly suited to parallel numerical codes, and, especially, to the implementation on graphics cards. Since the introduction of 3D rendering on computers, video cards have evolved from simple devices dedicated to video output into pow- erful parallel computing devices and often the computational power of graphical processing units (GPUs) surpasses the computational power of the CPU that drives themselves. The increasing computational power of GPUs has led to a growing interest in their usage for computation be- yond video rendering; their computational power these days allows turn- ing a desktop computer into a teraflop high-performance computer able to match the most expensive computer clusters in terms of performance, but at a fraction of cost. The strongly parallel nature of GPUs makes them extremely well-suited tools for advanced numerical computational needs such as those of scientific modelling. However, full exploitation of their capabilities requires appropriate tools and problems that are computationally-intensive rather than data-intensive: all data must fit on the GPU memory, and the operations-to-data ratio should be as high as possible. SPH is particularly appropriate for GPU, since the method is computationally intensive. The model adopted in this work is an implementation of Weakly Compressible SPH (WCSPH) to run entirely on GPU with Compute Unified Device Architecture (CUDA), called GPUSPH (H ́erault et al., 2010). CUDA, developed by NVIDIA, is a software development kit allowing recent NVIDIA graphic cards to be programmed for general- purpose tasks using C++, with extensions to manage the interaction between the card and the host machine. Actually, the speedup of SPH implementation on GPU against similar SPH implementation on CPU is considerable. The validity of the numerical code is assessed by considering the results of various test cases of free-surface flows. More precisely, test cases relating to the solitary and monochromatic wave propagation were taken in consideration, comparing the results with physical experiments carried out in the wave flume of the University of Palermo and by results from the technical literature. The quite good agreements with the measurements of wave height and surface profile in the test cases demonstrated that the model is able to solve the wave field and to reproduce wave-structure interaction. Finally, the proposed Lagrangian technique provides results as good as those by the Eulerian one, but with advantages of an easy and fast implementation and the absence of problems related to the construction of a calculation grid.