E. Rustico and J. A. Jankowski
BAW internal R&D-project report, 2010-2016.
Abstract: Smoothed Particle Hydrodynamics (SPH) is a Lagrangian method for fluid dynamic simulations. In the past decades it drew the attention of the scientific community for its versatility and the possibility to simulate complex phenomena such as e.g. surface tension and fluid-solid interactions with floating objects. SPH has in general higher computational requirements than the most common Eulerian methods and several different models have been proposed for the treatment of boundaries, each with advantages and limitations.
The project aimed to develop, validate and increase the computational efficiency of an existing SPH code for hydraulic engineering problems. As base code for the development, the GPUSPH simulator has been chosen. As the name suggests, it is a SPH implementation which exploits the Graphic Processing Units (GPU) to perform the computation of the particle to particle interactions. The GPUSPH simulator has been developed within the ATHOS Consortium, whose members include universities and research institutes from different nations working on different aspects of the code. A development branch external to BAW butof particular interest for the project regards the newer “Semi-Analytical” boundary model, the development of which has been completed only recently.
In the first phase, several features have been improved or introduced in the simulator, especially the capability of splitting a simulation across multiple nodes of a cluster each equipped with one or more GPU devices. This additional level of parallelism has enabled the simulation of high-resolution or spatially big scenarios (more than 100 million particles).
In the second phase, two test cases have been developed. The first is a fish pass, that has been modeled following the plan of a physical scaled model situated in the laboratories of BAW Karlsruhe. Simple Lennard-Jones boundaries have been used, featuring “pseudo-open” boundaries capable of controlling the inlet flow and specifically developed for the test case. A comparison of the simulated flow and a set of laboratory measurements highlighted a very good agreement in terms of the stream shape, including the recirculation areas. The water level were however too high in the simulated model because of an excess of friction due to the boundary model.
A ship lock has been modeled as second test case replicating the new Kiel-Holtenau lock, of which a scaled model was built at BAW Karlsruhe. The first implementation used the Semi-Analytical boundaries, featuring pressure- or velocity-driver open boundaries. Although the first results appeared to be promising, the low computational performance of the boundary model made a continuation of the tests on a large scale infeasible. Since a performance-optimized version of the model was not yet available (and is still in progress within the consortium), a second attempt has been done with the Dynamic Boundaries, a lighter boundary model featuring correct pressure at the boundaries but no inlets. The water inflow has been implemented either with “pseudo-inlet” and with a piston chamber. Both attempts showed a strong artifact causing the stream flow to go upward and hit the ship frontally rather than passing under it. This yielded the forces acting on the ship to be stronger than the forces measured in a laboratory model at BAW. The causes of the difference have been identified with high probability in the artificial stream lift caused by the non-physical inlet and the excess of friction along the ship hull. Both factors could be eliminated with a more physics-adherent boundary model.
The test cases highlighted the limits of the simpler boundary models and the need for an optimized, industrial-level version of the newly developed Semi-Analytical boundaries. Although the current development status and the first results are very promising for the applicability of the method to complex real-life engineering problems, and are already being used in the scientific community for many simpler problem scenarios, an optimization of the new boundary model is crucial to free the high potential of the method and accurately determine its application range and characteristics. An additional time of at least one year would be required to re-implement both test cases with the new model and produce a definitive comparison, granted the availability of the consortium members involving the optimization of the boundary model.
Reference: Rustico, E. and Jankowski, J.A. (2016) Evaluation and adaption of the SPH method for hydraulic engineering problems on federal waterways. Abschlussbericht B3953.05.04.70002, October 2016, Bundesanstalt für Wasserbau (BAW).