This paper presents the development of a numerical model for violent hydrodynamics of free-surface flows interacting with flexible structures using the meshless smoothed particle hydrodynamics (SPH) method accelerated with a graphics processing unit (GPU). The present work implements a unified SPH framework to solve both the fluid and structural dynamics within the open-source SPH code DualSPHysics. Well-known deficiencies with SPH-based structural modelling (linear inconsistency, tensile instability and rank deficiency) are addressed by adopting a Total Lagrangian formulation with kernel correction and zero-energy mode suppression. The fluid–structure coupling is handled via manipulation of the existing boundary condition within DualSPHysics, which does not require any geometrical knowledge of the interface. This, together with the unified SPH framework, permits straightforward implementation within DualSPHysics and retains its parallel scalability on hardware acceleration platforms, in particular on GPUs. An extensive validation and profiling of the proposed model is carried out for a range of challenging 2-D and 3-D test cases incorporating violent free-surface flows, nonlinear structural dynamics and complex fluid–structure interactions, and its performance is characterised and quantified. The analysis shows good agreement between the proposed model and analytical/benchmark solutions from the literature. Furthermore, the GPU profiling shows speedups on a single GPU of up to 40, compared to an optimised 12-core (24 threads) central processing unit (CPU) version. These findings indicate the proposed model is a viable and efficient approach to simulating violent free-surface fluid–structure interactions.

本文介绍了使用图形处理单元（GPU）加速的无网格平滑粒子流体动力学（SPH）方法开发的自由表面流与柔性结构相互作用的剧烈流体动力学数值模型的开发。本工作实现了一个统一的SPH框架，以解决开源SPH代码DualSPHysics中的流体动力学和结构动力学问题。通过采用具有内核校正和零能量模式抑制的Total Lagrangian公式，可以解决基于SPH的结构建模中的众所周知的缺陷（线性不一致，拉伸不稳定性和秩不足）。通过操纵DualSPHysics中现有的边界条件来处理流固耦合，不需要任何界面的几何知识。这，与统一的SPH框架一起，可以在DualSPHysics中直接实现，并在硬件加速平台（尤其是在GPU）上保留其并行可伸缩性。针对一系列具有挑战性的2D和3D测试案例，对该模型进行了广泛的验证和分析，这些测试案例包含剧烈的自由表面流动，非线性结构动力学和复杂的流体-结构相互作用，并对其性能进行了表征和量化。分析表明，所提出的模型与文献中的分析/基准解决方案之间具有很好的一致性。此外，与经过优化的12核（24线程）中央处理器（CPU）版本相比，GPU配置文件显示单个GPU上的加速高达40。