ENGINEERING AND CFD

The quality of surface water typically depends upon a complex interplay between physical, chemical and biological factors which are far from being completely understood today. While in most developed countries the quality of the majority of the surface water bodies does satisfy the applicable standards under normal circumstances, strong rainfalls can cause sewage overflow events in urban areas, which typically lead to a discharge of contaminated water into adjacent streams. Such storm-water discharges typically carry pathogens in the form of bacteria, viruses and/or protozoa.

Most practical water quality predictions for rivers or streams rely on a very much simplified formulation for the concentration of fecal indicator bacteria downstream of a pollutant source. So far, turbulent transport processes have only been "lumped" into average transport coefficients which rely heavily upon empirical input. The yielding models have limited predictive capabilities and overlook the spatio-temporal heterogeneous nature intrinsic to turbulent flow problems.

This GCS large-scale project aims at pushing the modeling boundary further by performing – for the first time ever – massively-parallel computer simulations of a mathematical model which resolve all scales of: hydrodynamic turbulence in river-like flows, the micro-scale flow around rigid, mobile particles, as well as the concentration field of suspended bacteria. The computational model is applied to the transport of suspended and dissolved contaminants resulting from a sewage overflow event into a streaming water body (such as a river or a canal), where the bacteria are initially present both in freely-suspended form as well as attached to the surfaces of the particles.

While the simulation of turbulent flows itself is already a challenge requiring immense computational resources, the coupling of the flow problem with the transport of around 100,000 fully-resolved particles makes high-performance computers a necessity. Furthermore, the simulation of the transport of dissolved contamination poses stringent requirements on the resolution of the simulated fields, due to which approximately 67 billion mesh nodes had to be deployed.

The simulations within this project were conducted for several months on the HPC systems Hazel Hen and Hawk at HLRS and utilized up to 8,192 cores per run. In total, the generated dataset has a size of several hundreds of terabyte which are stored predominantly on tapes at the bwDataArchive and moved to the Large-Scale Data Facility (LSDF) at KIT on demand.

The results already provided important insights into the interaction of particles with turbulent flows, and on their capabilities to release and manipulate the concentration of dissolved contaminants. One key finding is that the velocity at which a solid contamination settles can be substantially faster than what is assumed in current water-quality prediction models which neglect turbulent effects on particle dynamics. This finding is directly related to the inhomogeneity of turbulent flows, as settling particles self-organize and segregate into cluster and void regions while settling, which is the cause of this discrepancy.