Hostname: page-component-7b7567c8c-z4zkx Total loading time: 0 Render date: 2026-11-16T04:53:38.745Z Has data issue: false hasContentIssue false

Lagrangian network analysis of turbulent mixing

Published online by Cambridge University Press:  20 February 2019

Giovanni Iacobello*
Affiliation:
Department of Mechanical and Aerospace Engineering, Politecnico di Torino, 10129 Turin, Italy
Stefania Scarsoglio
Affiliation:
Department of Mechanical and Aerospace Engineering, Politecnico di Torino, 10129 Turin, Italy
J. G. M. Kuerten
Affiliation:
Department of Mechanical Engineering, Eindhoven University of Technology, 5600 Eindhoven, The Netherlands
Luca Ridolfi
Affiliation:
Department of Environmental, Land and Infrastructure Engineering, Politecnico di Torino, 10129 Turin, Italy
*
†Email address for correspondence: giovanni.iacobello@polito.it

Abstract

A temporal complex network-based approach is proposed as a novel formulation to investigate turbulent mixing from a Lagrangian viewpoint. By exploiting a spatial proximity criterion, the dynamics of a set of fluid particles is geometrized into a time-varying weighted network. Specifically, a numerically solved turbulent channel flow is employed as an exemplifying case. We show that the time-varying network is able to clearly describe the particle swarm dynamics, in a parametrically robust and computationally inexpensive way. The network formalism enables us to straightforwardly identify transient and long-term flow regimes, the interplay between turbulent mixing and mean flow advection and the occurrence of proximity events among particles. Thanks to their versatility and ability to highlight significant flow features, complex networks represent a suitable tool for Lagrangian investigations of turbulent mixing. The present application of complex networks offers a powerful resource for Lagrangian analysis of turbulent flows, thus providing a further step in building bridges between turbulence research and network science.

Information

Type
JFM Papers
Copyright
© 2019 Cambridge University Press 
Figure 0

Figure 1.(a) Sketch of the set-up of the analysis. Particles are initially released from a uniformly spaced grid at $x^{+}=0$. Coloured spheres represent particles while the ellipsoid is shaded in grey. Connections between particles are illustrated as black lines. (b) Example of particle dynamics depicted as a contact sequence, and (c) temporal evolution of the corresponding networks (the link thickness is proportional to its weight). For all panels, colours from red to blue highlight different starting levels (i.e. different initial $y^{+}$ values).

Figure 1

Figure 2.Representation of the time-varying network as: (a1–f1) weight matrices, $\unicode[STIX]{x1D61E}_{i,j}$; (a2–f2) their corresponding network topology. In panels (a1–f1), the labels of the matrix ordinates indicate the $y^{+}$ value (relative to the closest wall, i.e. $y^{+}\in [0,950]$) of the corresponding level reported in the matrix abscissas, while colours represent the weight of the links, $\unicode[STIX]{x1D61E}_{i,j}$. Network visualizations in panels (a2–f2) are obtained through the OpenOrd layout algorithm (Martin et al.2011); node colours indicate different $y^{+}$ values and range from red (i.e. level $1$) to blue (i.e. level $100$), while link weights are shown in greyscale, where strong links are in black and weak links are in light grey.

Figure 2

Figure 3.(a) Total number of connections, $E(t^{+})$, as a function of time. Three regimes are highlighted and distinguished by $T_{a}^{+}\approx 400$ and $T_{d}^{+}\approx 5200$. The two insets show the particle swarms in the first and third regimes, where particle colours indicate their starting level (blue is $l_{1}$, red is $l_{100}$). The values of $E$ at the same times reported in figure 2 are also illustrated as coloured circles. (b) Effect of $\unicode[STIX]{x1D6FC}$ on $E(t^{+})$, where $E_{a}$ and $E_{d}$ are the average values of $E(t^{+})$ over $t^{+} and $t^{+}>T_{d}^{+}$, respectively. Dashed lines indicate the scalings as $\unicode[STIX]{x1D6FC}^{2}$ and $\unicode[STIX]{x1D6FC}^{3}$.

Figure 3

Figure 4.(a) Time evolution of $\unicode[STIX]{x1D61E}_{i,j}$ for six pairs of levels. Each horizontal bar corresponds to an entry of the $\unicode[STIX]{x1D61E}_{i,j}$ matrices, while link activation is highlighted by horizontal coloured bands, where colour variations indicate the change of the link weight over time proportionally to $\log _{10}(\unicode[STIX]{x1D61E}_{i,j})$. (b) Standard deviation, $\unicode[STIX]{x1D70E}_{l_{i}}$, of different levels inside the ellipsoid of particles in five representative levels. The horizontal dashed line indicates the standard deviation from a discrete uniform distribution in the interval $[1,N_{y}]$, namely $\unicode[STIX]{x1D70E}_{l,u}\approx 28.87$.

Figure 4

Figure 5.Network metrics as a function of time, $t^{+}$, for the two cases of window functions, $K^{u}$ (black lines) and $K^{w}$ (red lines). (a) Average strength, $\langle S\rangle$; the red ordinate axis on the right refers to the range of $\langle S\rangle$ for the $K^{w}$ case. (b) Weighted assortativity coefficient, $r$. (c) Average clustering coefficient, $\langle C\rangle$. (d) Average path length, $L$, and number of disconnected nodes in the networks (blue dotted line). The inset in (d) is a zoom of the decreasing behaviour of $L$ with time. In all panels, coloured circles refer to the times reported in figure 2, while background colours highlight the three advection–mixing regimes (see § 3.2).

Iacobello et al. supplementary movie 1

The movie shows the temporal evolution of (left panel) particle positions in a streamwise-wall normal view (x-y plane), and the time-varying network as (top right panel) weight matrices, $W_{i,j}$, for $K^u$, and (bottom right panel) network topology. Particles in the left panel, as well as network nodes in the bottom right panel, are coloured according to their corresponding initial $y^+$ level. For the left panel, an increasing range of $x^+/L_x^+$ values in the horizontal axis is adopted, where $L_x^+$ is the streamwise DNS domain length.

Download Iacobello et al. supplementary movie 1(Video)
Video 2.5 MB

Iacobello et al. supplementary movie 2

The movie shows the temporal evolution of (top panel) particle positions in a streamwise-wall normal view (x-y plane), and (bottom panel) the probability distribution of the streamwise particle positions, $x^+$. Particles in the top panel are coloured according to their corresponding initial $y^+$ level. Please note the increasing range of $x^+/L_x^+$ values in the horizontal axis, where $L_x^+$ is the streamwise DNS domain length.

Download Iacobello et al. supplementary movie 2(Video)
Video 4 MB