Applicant
Prof. Dr. Philipp Schlatter
Institute of Fluid Mechanics (LSTM)
Friedrich-Alexander-Universität Erlangen-Nürnberg
Project Overview
Turbulent pipe flow is the most important canonical wall-bounded turbulent flow, both from a
scientific and engineering point of view. The geometry is clearly defined, and experimental
verification is directly possible (as opposed to e.g. channel flow). Nevertheless, pipe flow has only
recently been simulated at sufficiently high Reynolds numbers to allow statements about
asymptotic scalings of profiles, fluctuations and coherent structures. Our group has been at the
forefront of producing high-quality pipe-flow data, using the CPU-based code Openpipeflow, based
on high-order and spectral discretisation. The goal of this KONWIHR project is to develop a
GPU-version of Openpipeflow, to be used for future studies in wall turbulence, including
benchmark-quality simulations of turbulent pipe flow. The new framework, if carefully designed and
with minor changes to the numerics, may also form the basis for a family of codes for other
canonical flows such as channels and boundary layers, making our older CPU simulation
frameworks ready for the next decades. Having access to GPU-ready frameworks for pipes is also
relevant in the bigger context of EuroHPC.
Figure
Cross-sectional view of the velocity in a turbulent pipe flow at Re ≈ 10,000. The complex fractal multiscale features of wall-bounded turbulence are clearly seen; ranging from the small-scale near-wall streaks close to the walls (blue) to the large and very-large-scale motion in the pipe centre (dark red). For a faithful numerical prediction of the flow in a pipe, all scales need to be resolved in both time and space, and integrated sufficiently long in time to accumulate accurate statistics. Estimated cost of this simulation was around 100 million CPU hours on modern hardware.