Publications

A duct flow generated by a planar jet at the inlet and affected by a magnetic field perpendicular to the jet's plane is analyzed in high-resolution numerical simulations. The case of very high Reynolds and Hartmann numbers is considered. It is found that the flow structure is drastically modified in the inlet area. It becomes determined by three new planar jets oriented along the magnetic field lines: two near the walls and one in the middle of the duct. The downstream evolution of the flow includes the Kelvin-Helmholtz instability of the jets and slow decay of the resulting quasi-two-dimensional turbulence.

A liquid metal flat jet immersed in a square duct under the influence of a transverse magnetic field is studied experimentally. Two cases are considered: when the applied magnetic field is oriented parallel (coplanar field) or perpendicularly (transverse field) to the initial plane of the jet. The main goal of the study is to investigate the mean flow characteristics and the stages of the jet's transformation. Signals of streamwise velocity at different locations are measured, which allows us to determine average velocity profiles and spatial-temporal characteristics of the velocity field. The two considered configurations are directly compared under the same flow regimes, with the same equipment.

Recurrent neural networks are machine learning algorithms that are well suited to predict time series. Echo state networks are one specific implementation of such neural networks that can describe the evolution of dynamical systems by supervised machine learning without solving the underlying nonlinear mathematical equations. In this work, we apply an echo state network to approximate the evolution of two-dimensional moist Rayleigh-Bénard convection and the resulting low-order turbulence statistics. We conduct long-term direct numerical simulations to obtain training and test data for the algorithm. Both sets are preprocessed by a proper orthogonal decomposition (POD) using the snapshot method to reduce the amount of data. Training data comprise long time series of the first 150 most energetic POD coefficients. The reservoir is subsequently fed by these data and predicts future flow states. The predictions are thoroughly validated by original simulations. Our results show good agreement of the low-order statistics. This incorporates also derived statistical moments such as the cloud cover close to the top of the convection layer and the flux of liquid water across the domain. We conclude that our model is capable of learning complex dynamics which is introduced here by the tight interaction of turbulence with the nonlinear thermodynamics of phase changes between vapor and liquid water. Our work opens new ways for the dynamic parametrization of subgrid-scale transport in larger-scale circulation models.

We present a series of model experiments where an electrically non-conductive solid sphere moves inside a vertical column of liquid alloy GaInSn. The experimental setup consists of the liquid metal container, the sphere driving system and the permanent magnet with the attached force sensor. The sphere moves at a controllable constant velocity U0 and follows a straight route, which in turn generates a liquid metal flow around the sphere. This flow interacts with the localized magnetic field of the permanent magnet, and thus a weak reaction force on the magnet is generated. The force sensor attached on the magnet has a resolution of the order 10^-6. Upon elimination of high frequency noise, reproducible time-dependent signals for the forces on the magnet are obtained in the experiments for several Reynolds numbers Re between 160 and 2000. The force component Fz on the magnet parallel to the direction of particle motion exhibits a typical two-peak structure with different peak heights, whereas the transverse force component Fx resembles an antisymmetric pulse. The results demonstrate that the force sensor can detect the presence of a moving particle in a quiescent conducting liquid. They also show that the structure of the Fx signal can be reproduced with less variation and is less sensitive to the Reynolds number than the Fz signal. Moreover, the structure and magnitude of time-dependent Lorentz force signals can be reasonably predicted by a numerical model.