MALEDRAG - Machine Learning Optimisation for Drag Reduction in a Turbulent Boundary Layer
Imperial College London, United Kingdom
Local Project ID:
HPC Platform used:
Hazel Hen of HLRS
The potential of vertical blowing to reduce skin-friction drag but also to potentially generate net-power saving is investigated using a Bayesian optimisation framework. The data are obtained by means of Direct Numerical Simulation (DNS) of spatially developing turbulent boundary layers for low to moderate velocities.
The simulations were performed with the Xcompact3d, a framework of high-order finite-difference solvers dedicated to the study of turbulent flows. The simulations were performed with 200 million mesh nodes with 8,192 cores on HPC system Hazel Hen of HLRS. The flow configuration can be seen in figure 1, with 3 cases: one case with no control, one case with the highest possible level of drag reduction and one case with the highest possible level of net energy saving.
The problem of generating significant net energy saving using wall-normal blowing could be seen as an optimization of an unknown black-box function (the optimal balance between drag reduction and power needed for the wall-normal blowing) and invoke algorithms developed for such problems. A good choice is Bayesian optimization, which has been shown to outperform other state of the art global optimization algorithms on a number of challenging optimization benchmark functions. Bayesian optimization typically works by assuming the unknown function was sampled from a Gaussian process and maintains a posterior distribution for this function as observations are made, in the present case, as the results of running simulations with different parameters (frequency and intensity of blowing for example). For this project, the Bayesian Optimization Algorithm already existing in MATLAB was combined with Xcompact3d, in order to maximise net energy saving in a turbulent boundary layer using wall-normal blowing.
First, some experimental data are used in order to evaluate the potential net-power saving associated with an experimental blowing solution through a permeable wall with microholes. Up to 60% of drag-reduction is achieved and, after 18 DNS, corresponding to 18 iterates of the Bayesian optimisation scheme, the optimum strategy achieves a net-power saving of 5% via uniform blowing with moderate intensity (0.29% of the freestream velocity). A Fukagata-Iwamoto-Kasagi (FIK) analysis shows two different mechanisms responsible for the drag reduction over and downstream of the blowing region. The reduction of the friction coefficient is associated with the FIK convection term over the blowing region, and the FIK spatial development term downstream of the blowing region. However, the mechanical power to supply the air for the blowing solution used in this first Bayesian Optimisation study was not included and therefore a second Bayesian optimisation study is carried out with an original blowing solution based on micro-speakers to determine if net-power savings can really be achieved using vertical wall blowing. For this second study, taking into account the full power required for blowing, a net-power saving of 0.5% is achieved, giving a lot of hope for future practical applications.
This work demonstrates the potential of Bayesian Optimisation as an efficient tool to find optimal parameters to achieve a specific target (either net-power saving or high levels of drag reduction) for a given flow configuration.
Future work in the area will significantly advance and exploit our Bayesian optimisation framework to locate new, innovative, adaptive and autonomous pathways to reduce the skin-friction drag of turbulent boundary-layer flows with focus on achieving net-energy savings. This advanced Bayesian optimisation framework will be informed by flow physics to locate new drag reduction strategies using either low-amplitude wall-normal blowing or spanwise-wall blowing control techniques. These two forcing techniques are well-known to reduce skin-friction drag and are therefore ideal candidates for innovative optimisation and exploitation
References and relevant publications associated with the project:
Dr. Sylvain Laizet
Department of Aeronautics
Imperial College London
London SW7 2AZ, UK
e-mail: s.laizet [@] imperial.ac.uk
NOTE: The simulation project was made possible by PRACE (Partnership for Advanced Computing in Europe) allocating a computing time grant on GCS HPC system Hazel Hen of the High-Performance Computing Center (HLRS) Stuttgart. GCS is a hosting member of PRACE.
Local project ID: PRACE4381