Latest revision as of 15:52, 4 November 2020

Cylinder-wall junction flow

Underlying Flow Regime 3-35

Best Practice Advice

Key Physics

The inflow condition is of major relevance for the vortex system. The horseshoe vortex dynamics are driven by the downflow in front of the cylinder. This downward directed flow is caused by a vertical pressure gradient, which in turn depends on the shape of the approaching inflow profile. Therefore, one has to know the flow profile approaching the cylinder in order to be able to interpret the results and compare it to other studies. In the presented study, we took special care to have a fully developed turbulent open-channel flow approaching the cylinder in both experiment and simulation.

The wall distance of the horseshoe vortex is about $0.06D$ , the thickness of the wall jet between the cylinder and the horseshoe vortex center is smaller than $0.01D$ . This means that the key physics take place in a very thin layer in front of the cylinder. A simulation or a measurement not resolving this layer might strongly miss some key features such as the levels of the wall shear stress and the large levels of the streamwise velocity fluctuations and their bimodal character under the horseshoe vortex.

It has been demonstrated that the Reynolds shear stresses is small below the velocity maximum of the wall jet between the cylinder and the horseshoe vortex. This indicates a laminar-like behavior of the velocity profile below the jet's velocity maximum. As a consequence, conventional wall models relying on the logarithmic law of the wall will have difficulties to model the wall shear stress in this region.

Numerical Modelling Issues

Conditions corresponding to fully turbulent open channel flow have been applied at the inflow plane which should be placed at least 10 diameters upstream of the cylinder. It is essential to have a realistic turbulence structure and secondary flow at the inflow plane. For eddy resolving simulation strategies (DNS, LES and DES) this implies modeling or generating a fully developed open channel flow which is best achieved by a precursor simulation.

The downstream extent of the computation domain depends largely on the region of interest and should be placed far enough from the recirculation zone downstream of the cylinder to avoid instabilities and upstream-propagated waves. According to the authors' estimation this might vary depending on the specific type of the numerical scheme and outflow boundary condition.

A high spatial resolution is required to capture the horseshoe vortex dynamics and the wall shear stress. For the Reynolds number considered, a wall-normal resolution of $D/1000$ was necessary to obtain converged wall shear stresses. This high resolution was necessary because the wall jet does not follow the law of the wall for turbulent boundary layers and therefore needs to be fully resolved for calculating the wall shear stress. A realistic representation of the junction vortex also requires a fine resolution in the horizontal directions.

Physical Modelling

This flow case represents a highly complex three-dimensional turbulent flow which shows several characteristics making it difficult to be represented by numerical methods based on the Reynolds Averaged Navier Stokes (RANS) equations. There is the dynamics of the horseshoe vortex which oscillates in horizontal direction and produces bimodal velocity distributions near the wall (as shown by many authors). The production of turbulent kinetic energy (TKE) is far from a standard shear stress production observed in canonical boundary layers. In some important parts of the flow (in the wall jet where the largest wall shear stress values can be found) there is even a negative TKE production due to the strong flow acceleration. The levels of streamwise fluctuations under the horseshoe vortex seem to depend on the level of TKE in the wall jet in which a balance between the pressure transport and the negative normal stress production term is observed. In other regions (around the horseshoe vortex) it seems that the pressure and turbulent diffusive transport terms cancel each other and should be modeled as a whole. This all indicates that RANS modeling of this flow is a non-trivial task. The turbulent viscosity might be strongly overestimated by standard two equation models, especially in the wall jet.

For eddy-resolving turbulence calcuation strategies, several observations might be of importance. The turbulence is far from an isotropic state or from equilibrium. However, around the horseshoe vortex, the gradients of fluctuations in the main flow direction contribute to the dissipation in the same way as gradients in the vertical direction (Schanderl & Manhart 2018). This implies that strongly anisotropic grids are probably not suited to capture the physics around the horseshoe vortex in a sufficiently accurate way. We observed a relatively strong grid dependence of the results, although the SGS contribution was relatively small.

According to our observation, a subgrid-scale model needs to be able to adapt itself to the laminar-like behaviour of the wall layer under the wall jet. Further, we have no hope that explicit wall models relying on the logarithmic law of the wall can give an accurate correlation of the wall shear stress with a velocity value at a wall distance in inner coordinate corresponding to the logarithmic layer (Schanderl et al. 2017a).

For designing a grid for a DNS, it is important to note that estimating the Kolmogorov scale by $D/Re^{3/4}$ gives a conservative value and yields the correct order of magnitude (Schanderl & Manhart 2018).

Measurement issues

The horseshoe vortex system is a complex three-dimensional flow configuration. Therefore, the two-dimensional data acquisition method represents a limitation as the out-of-plane velocity component leads to a considerable loss of particles. The number of valid samples suffered from this issue in combination with the low seeding density resulting from the large size of the flume. To overcome this issue, we additionally evaluated the PIV images with a $32\times 32\mathrm {px}$ grid. Whenever the instantaneous velocity fields based on a $16\times 16\mathrm {px}$ grid revealed a missing vector, the corresponding vector of the coarser evaluation was taken as a substitute, if possible. In this way, we could improve the number of valid samples and still keep the spatial resolution high. However, the spatial resolution of the PIV data was too coarse to resolve the velocity gradient correctly. Therefore, a single pixel evaluation is recommended, for capturing the wall shear stress correctly.

Application Uncertainties

When simulating this flow configuration, we experience the largest uncertainties with regard to the inflow conditions of the approach flow and the representation of the water surface. Both numerical and experimental approaches face the challenge of generating a fully developed turbulent open-channel flow. Even though we intended to reproduce identical flow conditions and validated both of our methods (PIV and LES) by comparison with results in the literature, we observed differences in our results, e.g. concerning the size and location of the horseshoe vortex (see Fig. 6), which we attribute to the uncertainties in the structure of prevailing secondary flows or in modelling the water surface.

Another uncertainty is the roughness of the wall and its effect on the flow. It is our understanding that at the moment little is known on this issue.

Recommendations for Future Work

Performing a converged Direct Numerical Simulation would end all discussions about models and is - in our opinion - not far out of reach to date (2020). Further, considering surface roughness might give additional insight into the interaction of the wall jet with the wall.

The experiments could be improved by stereoscopic or tomographic PIV to acquire three-dimensional data sets. Furthermore, the temporal resolution could be increased in order to analyse the time scales of the horseshoe vortex system. The experimental setup can be improved by providing the light sheet from below passing through the transparent bottom plate, while the PIV camera(s) are mounted at the side outside of the flume.

Contributed by: Ulrich Jenssen, Wolfgang Schanderl, Michael Manhart — Technical University Munich

@@ Line 8: / Line 8: @@
 = Best Practice Advice =
 == Key Physics ==
-To cover this highly complex flow situation, a high spatial resolution is required for both the CFD as well as the experiment. The horseshoe vortex dynamics are driven by the downflow in front of the cylinder. This downward directed flow is caused by a vertical pressure gradient, which in turn depends on the shape of the approaching inflow profile. Therefore, to study the horseshoe vortex in detail and in a generic way, the development of a fully developed turbulent boundary layer approaching the cylinder should be ensured in the first place. Furthermore, the wall-shear stress is highly sensitive with respect to the spatial resolution of the data. To cover the strong velocity gradients, in particular of the the near-wall jet, high spatial resolution in the range of the Kolmogorov length scale is required (<math> \Delta x \propto \eta_{\mathrm{K}}</math>).
+The inflow condition is of major relevance for the vortex system. The horseshoe vortex dynamics are driven by the downflow in front of the cylinder. This downward directed flow is caused by a vertical pressure gradient, which in turn depends on the shape of the approaching inflow profile. Therefore, one has to know the flow profile approaching the cylinder in order to be able to interpret the results and compare it to other studies. In the presented study, we took special care to have a fully developed turbulent open-channel flow approaching the cylinder in both experiment and simulation.
-== Numerical Modelling Issues ==
+The wall distance of the horseshoe vortex is about <math> 0.06D </math>, the thickness of the wall jet between the cylinder and the horseshoe vortex center is smaller than <math> 0.01D </math>. This means that the key physics take place in a very thin layer in front of the cylinder. A simulation or a measurement not resolving this layer might strongly miss some key features such as the levels of the wall shear stress and the large levels of the streamwise velocity fluctuations and their bimodal character under the horseshoe vortex.
-We conducted the Large Eddy Simulation using our in-house code MGLET as a finite volume code with the WALE model to represent the subgrid scale stresses without a damping function. In this model, the subgrid scale viscosity decreases towards the wall following the correct limiting behaviour of <math> \nu_{\mathrm{t}} \propto z^3</math>.
-=== Discretisation method ===
+It has been demonstrated that the Reynolds shear stresses is small below the velocity maximum of the wall jet between the cylinder and the horseshoe vortex. This indicates a laminar-like behavior of the velocity profile below the jet's velocity maximum. As a consequence, conventional wall models relying on the logarithmic law of the wall will have difficulties to model the wall shear stress in this region.
-The spatial and temporal discretization were done by a second order central differences and a thrid order Runge-Kutta method, respectively.
-=== Grids and grid resolution ===
+== Numerical Modelling Issues ==
-We used a staggered Cartesian grid fo the entire domain, and modelled the surface of the cylinder using a conservative second order immersed boundary method. The grid resolution in thevertical direction was <math> \Delta z^+ = 0.95 </math> using the wall-shear stress of the precursor simulation, and <math> \Delta z^+ = 1.6 </math> based on the local shear stress. The grid resolution was, therefore, sufficently fine for a LES. In case of further refinement, it would make sense to perform a Direct Numerical Simulation (DNS), instead.
+Conditions corresponding to fully turbulent open channel flow have been applied at the inflow plane which should be placed at least 10 diameters upstream of the cylinder. It is essential to have a realistic turbulence structure and secondary flow at the inflow plane. For eddy resolving simulation strategies (DNS, LES and DES) this implies modeling or generating a fully developed open channel flow which is best achieved by a precursor simulation.
-=== Boundary conditions and computational domain ===
+The downstream extent of the computation domain depends largely on the region of interest and should be placed far enough from the recirculation zone downstream of the cylinder to avoid instabilities and upstream-propagated waves. According to the authors' estimation this might vary depending on the specific type of the numerical scheme and outflow boundary condition.
-The computaitional domain containing the cylinder had a rectangular cross-section with the following size: <math> L\times B \times H = 25D \times 12D \times 1.5D </math>.
-The bottom and side boundaries were treated with a no-slip condition, while the top boundary was modelled using a slip condition.
-As inflow condition a fully developed turbulent boundary layer was used stemming from a precursor simulation.
-The precursor domain had a size of <math> L\times B \times H = 30D \times 12D \times 1.5D </math> with periodic boundary conditions and no-slip condition at the bottom and the side walls. The surface was also modelled as slip condition.
+A high spatial resolution is required to capture the horseshoe vortex dynamics and the wall shear stress. For the Reynolds number considered, a wall-normal resolution of <math> D/1000</math> was necessary to obtain converged wall shear stresses. This high resolution was necessary because the wall jet does not follow the law of the wall for turbulent boundary layers and therefore needs to be fully resolved for calculating the wall shear stress. A realistic representation of the junction vortex also requires a fine resolution in the horizontal directions.
 == Physical Modelling ==
-The horseshoe vortex system is a complex three dimensional flow configuration. The two dimensional data acquisition method is, therefore, a limitation as the out-of-plane velocity component leads to a corresponding loss of particles. The number of valid samples suffered from this issue in combination of the low seeding density resulting from the large size of the flume. To overcome this issue, we additionally evaluated the PIV images with a <math>32\times32\mathrm{px}</math> grid. Whenever the instantaneous velocity fields based on a <math>16\times16\mathrm{px}</math> grid revealed a missing vector, the corresponding vector of the coarser evaluation was taken as a substitute, if possible. In this way, we could improve the number of valid samples and still keep the spatial resolution high.
-However, the spatial resolution of the PIV data was too coarse to resolve the velocity gradient correctly. Therefore, a single pixel evaluation is recommended, in order to capture the wall-shear stress correctly.
-== Application Uncertainties ==
+This flow case represents a highly complex three-dimensional turbulent flow which shows several characteristics making it difficult to be represented by numerical methods based on the Reynolds Averaged Navier Stokes (RANS) equations. There is the dynamics of the horseshoe vortex which oscillates in horizontal direction and produces bimodal velocity distributions near the wall (as shown by many authors). The production of turbulent kinetic energy (TKE) is far from a standard shear stress production observed in canonical boundary layers. In some important parts of the flow (in the wall jet where the largest wall shear stress values can be found) there is even a negative TKE production due to the strong flow acceleration. The levels of streamwise fluctuations under the horseshoe vortex seem to depend on the level of TKE in the wall jet in which a balance between the pressure transport and the negative normal stress production term is observed. In other regions (around the horseshoe vortex) it seems that the pressure and turbulent diffusive transport terms cancel each other and should be modeled as a whole. This all indicates that RANS modeling of this flow is a non-trivial task. The turbulent viscosity might be strongly overestimated by standard two equation models, especially in the wall jet.
-The standard error of the mean value of the measured velocities was determined as follows:
-<math> \varepsilon_{\mathrm{std}}(\langle u \rangle) = \frac{\sigma(u)}{\sqrt{N_{\mathrm{samples}}}} = \frac{\sqrt{M_2(u)}}{\sqrt{N_{\mathrm{samples}}}} </math>,
+For eddy-resolving turbulence calcuation strategies, several observations might be of importance. The turbulence is far from an isotropic state or from equilibrium. However, around the horseshoe vortex, the gradients of fluctuations in the main flow direction contribute to the dissipation in the same way as gradients in the vertical direction (Schanderl & Manhart 2018). This implies that strongly anisotropic grids are probably not suited to capture the physics around the horseshoe vortex in a sufficiently accurate way. We observed a relatively strong grid dependence of the results, although the SGS contribution was relatively small.
-the standard error of the higher central moments <math> M_n = \langle u'^n \rangle </math> was obtained likewise:
+According to our observation, a subgrid-scale model needs to be able to adapt itself to the laminar-like behaviour of the wall layer under the wall jet. Further, we have no hope that explicit wall models relying on the logarithmic law of the wall can give an accurate correlation of the wall shear stress with a velocity value at a wall distance in inner coordinate corresponding to the logarithmic layer (Schanderl et al. 2017a).
-<math> \varepsilon_{\mathrm{std}}\left( \langle u'^n \rangle \right) = \frac{\sigma\left(u'^n\right)}{\sqrt{N_{\mathrm{samples}}}} = \frac{\sqrt{M_{\mathrm{2n}}-M_{\mathrm{n}}^2}}{\sqrt{N_{\mathrm{samples}}}} </math>.
+For designing a grid for a DNS, it is important to note that estimating the Kolmogorov scale by <math> D/Re^{3/4} </math> gives a conservative value and yields the correct order of magnitude (Schanderl & Manhart 2018).
+== Measurement issues ==
+The horseshoe vortex system is a complex three-dimensional flow configuration. Therefore, the two-dimensional data acquisition method represents a limitation as the out-of-plane velocity component leads to a considerable loss of particles. The number of valid samples suffered from this issue in combination with the low seeding density resulting from the large size of the flume. To overcome this issue, we additionally evaluated the PIV images with a <math>32\times32\mathrm{px}</math> grid. Whenever the instantaneous velocity fields based on a <math>16\times16\mathrm{px}</math> grid revealed a missing vector, the corresponding vector of the coarser evaluation was taken as a substitute, if possible. In this way, we could improve the number of valid samples and still keep the spatial resolution high.
+However, the spatial resolution of the PIV data was too coarse to resolve the velocity gradient correctly. Therefore, a single pixel evaluation is recommended, for capturing the wall shear stress correctly.
-The standard errors with respect to the standard deviation <math> \sigma </math> were quantified as follows:
+== Application Uncertainties ==
+When simulating this flow configuration, we experience the largest uncertainties with regard to the inflow conditions of the approach flow and the representation of the water surface. Both numerical and experimental approaches face the challenge of generating a fully developed turbulent open-channel flow. Even though we intended to reproduce identical flow conditions and validated both of our methods (PIV and LES) by comparison with results in the literature, we observed differences in our results, e.g. concerning the size and location of the horseshoe vortex (see Fig. 6), which we attribute to the uncertainties in the structure of prevailing secondary flows or in modelling the water surface.
-{| class="wikitable" style="text-align: center;" border="1"
+Another uncertainty is the roughness of the wall and its effect on the flow. It is our understanding that at the moment little is known on this issue.
-| <math> \varepsilon_{\mathrm{std}}(\langle u \rangle) / \sigma </math>
-| <math> \varepsilon_{\mathrm{std}}\left( M_2 \right) / \sigma^2</math>
-| <math> \varepsilon_{\mathrm{std}}\left( M_3 \right) / \sigma^3</math>
-| <math> \varepsilon_{\mathrm{std}}\left( M_4 \right) / \sigma^4</math>
-|-
-| <math> 0.0065 </math>
-| <math> 0.0089 </math>
-| <math> 0.0237 </math>
-| <math> 0.0545 </math>
-|}
 == Recommendations for Future Work ==
-Performing a LES considering the surface waves by a level set method would improve the numerical results. Aiming to reduce the portion of modelled stresses, a further grid refinement could be implemented.
+Performing a converged Direct Numerical Simulation would end all discussions about models and is - in our opinion - not far out of reach to date (2020). Further, considering surface roughness might give additional insight into the interaction of the wall jet with the wall.
-The experiments could be improved by stereoscopic or tomographic PIV to acquire three dimensional data sets. Furthermore, the temporal resolution could be increased, in order to analyse the time scales of the horseshoe vortex system. The experimental set-up can be improved by providing the light sheet from below passing through the transparent bottom plate, while the PIV camera(s) are mounted at the side outside of the flume.
+The experiments could be improved by stereoscopic or tomographic PIV to acquire three-dimensional data sets. Furthermore, the temporal resolution could be increased in order to analyse the time scales of the horseshoe vortex system. The experimental setup can be improved by providing the light sheet from below passing through the transparent bottom plate, while the PIV camera(s) are mounted at the side outside of the flume.

UFR 3-35 Best Practice Advice: Difference between revisions

Latest revision as of 15:52, 4 November 2020

Contents

Cylinder-wall junction flow

Underlying Flow Regime 3-35

Best Practice Advice

Key Physics

Numerical Modelling Issues

Physical Modelling

Measurement issues

Application Uncertainties

Recommendations for Future Work

Navigation menu

UFR 3-35 Best Practice Advice: Difference between revisions

Latest revision as of 15:52, 4 November 2020

Cylinder-wall junction flow

Underlying Flow Regime 3-35

Best Practice Advice

Key Physics

Numerical Modelling Issues

Physical Modelling

Measurement issues

Application Uncertainties

Recommendations for Future Work

Navigation menu

Search