You are on page 1of 19

Ordered structures from chaotic ows

VORTEX DIPOLES

Abstract

This experiment involved the observation and analysis of unique structures known as vortex dipoles in a 2D uid ow. The Reynolds number of the vortex dipoles was found from established theory as a function of the volume ux into a stratied uid. This was compared with the Reynolds number found experimentally as a function of the vortex dipole aspect ratio, . The results agreed with those presented by Y. Afanasyev et al [1]. A well known imaging technique, Particle Image Velocimetry was then employed to nd the velocity eld of the dipole from which the vorticity at each point was found. The vorticity elds produced were consistent with the polar solution of vortex dipoles rst produced by Sir H. Lamb in 1895 [2]. Other tertiary experiments were performed including observation of vortex dipole collisions and the curious attraction of vortices to walls and boundaries.

Introduction
Vortex dipoles are fascinating jet ows that have two propagating, counter rotating vortices at their front. Although considerably rarer than the common single vortex, vortex dipoles still have application in an incredibly diverse range of scientic areas and have been observed in everything from cloud formations to Bose-Einstein condensates [3], an exotic state of matter where quantum eects are apparent on the macroscopic scale. In this report, the history and application of vortex dipoles (VDs) are discussed, with emphasis on evaluation of the various mathematical models used to explain their structure.

Figure 1: A vortex dipole of dyed water at t =50s after formation. Typically, turbulent uid ows occupy the full three dimensions of space. However, in certain cases motion in one direction is suppressed and a two dimensional uid system remains. This assumption is common in explaining uid ows and in particular for vortices, usually means that dependence of polar co-ordinates is neglected [4]. Two dimensional turbulent experiments and simulations are common as simple models of atmospheric and oceanic turbulence. However, despite the simplicity of the models used, they accurately predict the properties of large scale dipoles, which regularly form in inlets and can even form on oceanic scales [5]. The earliest successful attempts to model straight vortex pairs was produced by Sir H. Lamb in 1895. A modern revision of his original solution is provided in the theory section below. Although the model reproduced the basic shape and properties of the V-Ds, it had no time dependence and indicated no V-D evolution. In 1905, Chaplykin reworked the original equations but produced results with only limited experimental signicance. With the invention of jet propulsion aircraft in the 1940s, 2

vortex dipoles became a topic of signicant interest. This is because an accelerated ow past a wing produces a vortex pair across the wingspan. The mathematics is exacting, but my interpretation of the equations produced by G. Saman [7] is given here. Consider a rectangular elliptically loaded wing of span 2a, with a stream velocity W inclined at an angle (g.2). The ow that develops is a vortex sheet after time time t = Z/W with U = W when the wing moves.

Figure 2: Vortex eld behind a wing This results because the vorticity, v, is a maximum at the wing tips since v = u . As the wing moves, the ow going past it continues to be curled on each side of the aircraft, and so a spiral of ow continually grows at the tips with the motion of the wing. As the ow curls away from the wing tip it loses velocity which causes the ow to roll up into an approximately circularly symmetric vortex. The distance between each consecutive turn limits to zero, as in g.1 as the particle speed reduces along its path from the wingtips. At each side of the aircraft a counter-rotating vortex is produced. As the craft accelerates the vortices interact in the plane perpendicular to the craft motion forming a V-D. Before the invention of computer processors, only having graphical solutions meant that the equations governing V-Ds had limited applicability. With the additional potential of numerical modeling, the equations could be heavily rened. The most comprehensive re-evaluation of the vortex dipole model was produced by Deem and Zabusky in 1977 [8]. The model included reworking to include translation of the dipoles and explained why they are not dispersive and the conditions under which they form. A good example of the benet of the numerical model equations was presented in 2001 by L. Cortelezzi and A. Karagozian [9] who were able to model the V-Ds in 3D and gained signicant insight into the internal vortex pair topology. An example of the modelled V-Ds is shown in g.3 Some of the places where vortex dipoles have been observed are genuinely surprising. Consider that in this experiment, the V-Ds are produced by a delta function of impulse into a planar uid. One simple example of this is rain water hitting the surface of a body of water at a high speed. If the conditions are right, the water droplet impulse should sink the droplet fast down into the water, so that horizontal spread is suppressed. This essentially provides a stratied layering of the water. This has been done using dyed water droplets and dipoles can be readily observed. Attempts were made in this lab to replicate this phenomena but insucient velocity was achieved. It is easy to think that vortex dipoles are somehow conned to classical physics and are unlikely to be found outside of predened experiments. However, they have recently been observed in BoseEinstein condensates and in exotic superconducting materials near the critical temperature phase 3

Figure 3: 3D vorticity contour produced by Cortelezzi et al. transition [10]. Fleming et al. showed that the dissipation of energy near the superconducting region of some materials is carried away by vortex dipoles which form in the material at a coupling rate K which was highly temperature dependent and in particular - large around Tc . In fact, the equations they used to model the dipoles were based on the original work by Deem and Zabusky [8], with a few variations to include magnetic eects. Although not as common as a singly rotating vortex, which is just the manifestation of shear stresses in a uid, vortex dipoles are still found throughout all uid systems. The original potential solution of V-Ds has been updated and numerical models have been employed and no doubt further work will be done, as a complete revision has recently been oered by Y. Afasayev [11]. The key thing is that, given how common they are in nature and the diversity of scientic elds in which they are found, the work on V-Ds is in no way complete, and is a fascinating example of how something simple can actually have a whole range of complex applications.

Theory
Modeling vortex dipoles Unlike vortex arrays or systems containing large amounts of vortices, V-Ds do not require numerical methods to model and can be analytically solved. Straight vortex pairs with equal magnitude but opposite sign were rst mathematically modeled by Sir Horace Lamb in 1895 [2]. He showed that a vortex pair with no net angular momentum can be modeled assuming its overall shape is circular, its overall motion is uniform and that it has evenly distributed velocity. Although more accurate models have been proposed, they usually limit themselves to specic cases, the only one of which we will discuss is a delta function of impulse into a stratied uid. A modernisation of Lambs original work was produced in 2006 by Jie-Zhi Wu et al [4], which provides a simpler analysis, solving the potential over a cylinder in polar co-ordinates, (r, ) with u= 1 , v= r r r a2 r sin, r > a

=U

Here, a is the distance from the centre of the vortex dipole pair to its boundary at r = a and is the angle around the dipole origin. For simplicity, we impose an external ow of U so that it remains stationary with its centre on the origin. Also, since this is a 2d structure we neglect any 4

dependence which would have to consider bottom and gravitational eects. We note that in general in order to linearize the potential:
2

= f (, t), = k 2

and so the general equation for vorticity yields 1 2 2 1 + + 2 2 = k2 r2 r r r The solution of which is involved but we can see some aspects of it straight away. From the previous equation for it is clear that sin. Also, the equation takes the form of Bessels dierential equation. Therefore the solution is a Bessel function = CJ1 (kr)sin The rst zero of J1 (ka) is at ka = 3.83, which gives a closed circular streamline = 0 at r = a, the boundary of the dipole. More specically, we can nd C from velocity continuity noting that the spatial derivative of at the boundary is =U r r From which the constant C is found to be C= 2U 2U = kJ0 (ka) k(J )1 (ka) r r2 r sin,

Subbing back in for the potential solution, inside the vortex dipole yields = 2U J1 (kr) sin kJ0 (ka) J1 (kr) sin , r < a J0 (ka)

= k 2 = 2U k

This is the point at which the streamlines can be seen graphically and is probably a good nishing line on the general solution, showing that the vorticity magnitude is directly proportional to the point velocity everywhere inside the dipole. Also, the vorticity is largest at = /2 along the axis between the dipole centers, a suggestion that my results disagree with by several degrees, most likely due to bottom forces. Fig.4 shows the streamlines in (r, ). Sergey Chaplykin elaborated on this model [4] and determined that the vorticity has a maximum and minimum (due to symmetrical counter rotation) at r0 = 0.48a. This result will be conrmed by producing a vorticity contour and observing it across the plane, nding the maximum and minimum vorticity along a. Vortex dipole global properties We now leave the general case solution and look at some of the bulk properties of vortex dipoles. If we inject a short impulse of the water into a large body of uid both the mass and linear momentum of the system will be conserved. Y. D. Afanasyev showed [5] that the time dependence of a dipoles length and breadth, L, D are given by L, D t1/2

Figure 4: The graphical solution of the vorticity in (r, ) showing the closed streamlines and the main features of Lambs vortex dipole model [4]. implying L = D where is a constant for a given dipole that depends on the incident velocity and therefore also the ratio of the inertial to viscous forces in the uid, Re. By plotting against time we can nd its mean value after formation. We can then nd the Reynolds number it coincides with using [1]. This gives an indication of the viscosity of the uid and the level of turbulence of the system since Re = U 2 d / U d2

Where U is the uid exit velocity, v = / is the kinematic viscosity of water and d is the diameter of the injection nozzle. More specically for the setup used we know that the velocity at the nozzle is the volume ux q, divided by the nozzle area (d2 /4) so we have that Re2 = where J is dened as J= 4q 2 d2 J v2

thus we see that Re = kq where k is just a constant of the setup that does not change for dierent dipoles although it does have a small temperature dependence since v = v(T ) as shown in Fig.5. Volume flux calibration We now must consider how the volume ux can be calculated from experimentally known quantities. In particular, given an initial uid height, h0 and time interval of injection t we would like to have q = q(h0 , t). In fact, as will be shown, once a calibration has been performed the time interval has no bearing on the volume ux since q is a constant for a given initial height and a small net volume released. We start with the assumption that given a height of liquid with an opening at its bottom, the rate of change of its height is proportional to the height. h h t which immediately implies 6

Figure 5: Temperature dependence of v, the kinematic viscosity. In general v 0.012cm2 /s in the dark lab used.

h(t) = h0 ekt The volume ux, q is the change in volume divided by the change in time. We can now write it as q= h0 hf h0 (1 ekt) = t t

Let us now assume q is an experimentally found quantity, where we can set h0 hf = 1cm3 so that we can solve for the constant k, which we will dene as a magnitude only, with the sign outside of the exponent. 1 1 = ekt h0 k = ln{ hf }/t h0

k was found to be 0.00183, details are given in the results. However using the k value found by experiment we return to the equation for q and expand the exponent to rst order q= h0 (1 ekt) h0 (1 1 + kt) = t t q = kh0 The higher the initial height, the larger the volume released by the nozzle in one second, q. Further to this, we suggest that as we saw that Re = k q, the Reynolds number of the ow is directly proportional to the height of the uid (g.6) provided we assume at h0 = 0 there is no volume ux. This imposes an eective range of Re which can be altered by changing the height of the burette but is typically 10 < Re < 80. The rst part of the experiment concluded with the estimation of Re via the aspect ratio , and by the volume ux, q(h0 ). 7

Figure 6: The Reynolds number as a function of water height Particle Image Velocimetry (PIV) In PIV, patterns of particles spanning a small sub-area are matched with a slightly shifted set of particles separated by a well-dened time interval. The resulting displacement gives an average velocity for that region [6]. The experimental considerations of PIV are detailed in the method where as here, we provide an overview of the image processing entailed. The main principle of PIV algorithms is to take two images and perform cross-correlation on them to determine the motion at each point in the image over the time interval between their capture. The most straight forward approach to cross-correlation of the images is to dene the images as divided into sub-windows of pixel side length N = 2n where the n used was typically 4, 5 or 6. Two dimensional cross correlation of each sub-window pair is performed. Let us dene sub-windows I and J within the rst image F with indices i, j representing pixel locations within the sub-windows. Similarly, we dene the same sub-windows in the second image F , we then have 1 N2
N 1 N 1

R(s, t) =

FI,J (i, j)FI,J (i + s, j + t)


i=0 j=0

where R is the cyclic cross-correlation between the windows I and J in each image pair. The equation also has a concise fourier form which was not used in this experiment. This basic method can not interpret changes above N/2 and so more advanced algorithms are used. Also, it is necessary to normalise the background which is not a complex procedure or equation, but is very large and basically involves summing over the mean non determinate sub-window cross correlations and dividing each cross correlation by this value of R(s, t). The actual program used was MatPIV16.1 written by J. Kristian Sveen which is available as freeware online. The initialisation commands required an 8

approximation of the mean vector per sub-window and an input of the time duration between the images. When the mean particle displacement is obtained at each window over the whole image, we eectively have obtained the velocity eld. Taking the curl of the velocity eld, U (u, v) gives the spatially resolved vorticity of the uid. Using basic Matlab commands it can be shown that the total vorticity of the dipole is near zero and that the y-directional momentum has a maximum along the centre line of the dipole. Given R C subwindows of a vorticity matrix M we can say. M+ = sum(M.*(M> 0) M = sum(M.*(M< 0) Net vorticity = M+ + M and also to nd net vorticity at each X and Y level: for n = 1:R , for m = 1:C X(n) = sum( M(R,:) ) Y(m) = sum( M(:,C) )

Method
The equipment was set up as shown. A web camera with a maximum resolution of 640 480 pixels was connected to a laptop using the Logitech drivers available online. The .avi videos recorded were viewed frame by frame using virtualdub.exe (www.virtualdub.org), a freeware .avi viewer which gives the frame number and time in ms for each image. The frame rate of the camera was set to 30 frames per second. This was conrmed by counting the number of frames over a set time interval. The diammeter of the nozzle was found by inserting thin wires into the injection point and then measuring the wires with a m gauge.

Figure 7: The experimental setup for non PIV imaging. The water tank and burette were prepared separately. The tank had a top surface area of 1250cm2 . We required a salt water bottom layer of 2-3cm. Thus, in order to produce 2.5cm of bottom layer, 3L of salt water was included, requiring 350g of salt per refresh of the layer to hit a reasonable ( 50%) level of salt saturation. The salt is added and stirred, left for ten minutes and then stirred again to maximise distribution. When this was not done, the top layer was seen to be uneven on the surface of the salt water layer. A sheet of A3 paper was placed on top of the salt layer and pure water was sprayed onto the surface. The paper given suggest a thin as possible surface layer however, this was found to produce signicant bottom eects due to interference of the bottom and top layers which aected the dipoles produced. A larger layer of 0.5cm layer was used on top, requiring 600ml of de-ionised water. The A3 sheet was then carefully removed and the layers were given time to settle. During this time, the burette was prepared which contained up to 25ml of red dyed pure water. The burette was carefully inserted into the top surface layer. A retort stand was used to keep the height of the needle to the pure water layer constant. An additional control retort stand was used to re-balance the needle angle in the uid, since the needle tended to point up slightly when placed into the water due to upthrust. The volume ux was calibrated using q = kh0 . k was found experimentally using known volume uxes at increments of 1ml along the burette liquid height. Typical dipoles were 0.1ml or less in volume and so h0 hf was highly inaccurate and so q was found indirectly using k. Generally, there was a pay o between pixel resolution and the time during which the dipole was in the camera view. Also, the scaling of cm to pixels was recorded with each run since any movement of the camera caused it to vary. Surprisingly, the x and y planes had the same calibration factor. t and q were varied. During a run, the following procedure was used. More than 40 video runs are available.

10

Run procedure
The initial height h0 was recorded and its corresponding Reynolds number was found (g.6). The light was turned on and camera was set to view the nozzle and 12cm in front of it. The video was started and a short injection of uid was allowed through the burette. Several frames of the dipoles development were isolated using Virtualdub.exe The pixel-cm scale was calibrated in paint and L(t),D(t) and t were recorded for each frame. was found for the dipole and the corresponding Reynolds number was found graphically

Figure 8: Time evolution of a vortex dipole at a) 0.5s , b)6s c) 8s and d 14s) Generally, the dipoles were initially long and narrow, only spreading out once their velocity had decreased below a threshold that was usually about 50% of the injection velocity. In terms of the analysis, this meant that varied with time, specically it was seen to have an asymptotic behaviour, quickly reducing to a near constant value where it stayed for t > 1min. This meant that a bestt was required which would omit the rst few high values. An example is included in the results. Runs were also performed where a continuous dye stream was injected behind the dipole. This gave larger, faster moving dipoles. Also, when this was performed in a non density stratied uid, a dipole was seen to form into a more spherical dipole shape. This concluded the main analysis of the dipole momentum. The other most signicant vortex dipole property, vorticity, was explored using particle image velocimetry (PIV).

11

PIV The setup was altered in three ways. Firstly, pure non-dyed water was used as the injection uid. Secondly, a projector was position parallel to the water surface and a thin sheet of light was spread across the top layer. Thirdly, 50m balls were seeded into the water using a spray, where they oated. These had a tendency to clump together even at relatively low particle densities. To compensate for this, the balls were then sprinkled by a spatula over the surface, this gave a much better distribution. The seeded particles typically took 10 minutes to settle. A video was started from a pre-chosen height with known cm-pixel scale. The water was injected as before and frames were isolated with a known time between them using virtuadub.exe. The time interval was usually between 0.1s < t < 1s. The two images (im1, im2) were then saved in a directory containing MatPIV.m, the core le of MatPIV16.1. The command structure for cross correlation of the images was function [x,y,u,v]=matpiv(im1,im2,winsize,Dt,overlap,method,wocole,mask,ustart,vstart) where winsize was the sub-window size in pixels, which is a multiple of 2n , overlap was the required overlap size of the windows, which could be set to zero. Method was cross correlation and no masking of areas was usually performed. Vorticity was then found using the velocity map [x,y,u,v] by taking its curl, the code is easy to follow and is based on the idea of least squares: { DeltaX=x(1,2)-x(1,1); DeltaY=y(1,1)-y(2,1); for i=3:1:size(x,2)-2 for j=3:1:size(x,1)-2 vor(j-2,i-2)= -(2*v(j,i+2) +v(j,i+1) -v(j,i-1) -2*v(j,i-2))/(10*DeltaX)... + (2*u(j+2,i) +u(j+1,i) -u(j-1,i) -2*u(j-2,i))/(10*DeltaY); end end outp=vor; xa=x(3:end-2,3:end-2); ya=y(3:end-2,3:end-2); } We end in a matrix output outp which can be plotted as a surf plot or as required. The main practical issue with this method was the web camera resolution. Inevitably, there was a pay o between the resolution of individual particles and viewing of the entire dipole. Small sections of a dipole could be viewed giving excellent velocity maps, but larger viewed areas lost resolution, requiring masking of null velocity areas where the particles were not picked up. A compromise was reached by generally producing small dipoles (Dmax = 2cm) although the larger ones were more impressive. Additionally, it was possible to perform PIV by turning the bottom projector on, taking the images as usual. The images were then gray-scaled and the colours were inverted and the brightness was reduced. The eect of this was to simplify the setup while still maintaining essentially similar pictures of white seeded particles against a dark background.

12

Results
The ux calibration against height is shown below in g.9 (a). Although the time intervals t(h) were non linear with height, the ux, q(h) was seen to be proportional to height of the liquid. From this, the constant k was determined for each height, its variation is shown in g.9 (b). In c.g.s units, if we consider that the height of the burette correlates to cm3 k was 0.00182cm3 /s per cm of height.

Figure 9: a) The averaged ux as a function of height. b) the constant k and its bestt. The graph provided for tting against Reynolds number was logarithmic, squared and generally did not give a clear indicator of the experimental error. A gent was performed in MathCAD using data from the bestt (Re2 ) line so that a direct correlation between and Re could be found graphically.

Figure 10: The alpha and Reynolds number values found for 16 dipoles

13

This avoided the tedious conversion from logarithmic to linear scaled data. The line took the form of the exponential of a polynomial, exp{f (x1 , x2 , x3 , x4 ...)}. Poly-Gaussian t was not considered necessary as the line was well formed. The list of dipoles analyzed is shown graphically with the theoretical t in g.10. Another way of interpreting this data is to nd the Reynolds number by both methods and to nd their ratio against height (g.11). This proves to be useful as it shows the values were typically within 90% accuracy and that since most of the points are below unity, the Re(h) values were larger than the Re() values. It is unlikely that the ux was overestimated, as the values found were surprisingly small, which suggests that the actual graph provided with the paper may have been o the true value by some constant. In fact, it is possible that the uxes were not overestimated but underestimated since q(0) = 0 was assumed.

Figure 11: The alpha and Reynolds number values found for 16 dipoles The behaviour of and the dipole velocity with time was explored, an example is given below (g.12) for run7. The velocity had a strong, asymptotic behaviour, contradicting the suggestion by Y. Afanasyev [11] that the steady state velocity is constant at Uconst = Ujet /2. The behaviour was found repeatedly. Typically the initial velocity was above 5cm/s for half of a second, within 5 seconds the jet typically travelled at (0 1.5)cm/s slowing to (1 2)mm/s in the steady state. Typically there was a 10-100 fold decrease in velocity.

14

Figure 12: Velocity and time dependence for one dipole. Other runs gave very similar results provided as an appendix. Also, dipoles were collided to observe the eects (g.13). Normal process was that the slower dipole would remain unchanged while the faster dipole would smear out over the front of the other, forming small, isolated vortices. This behaviour has been well documented in numerous papers by Y. Afanasyev.

Figure 13: Vortex dipole collision PIV provided interesting results. The velocity elds were well formed for small image areas but had aberrations over large areas. These were then converted into vorticity elds using the code provided in the theory section. Fig.14 shows the vorticity eld for a large dipole. Limited masking was employed at the corners of the image. The circulation is counter-rotational in each vortex, and so we see one side of the dipole as negative and the other as positive. Summing over the whole area of the dipole gave a near zero value for vorticity, 41/(478 + 437) 4.5%. The y axis momentum had a peak at the central sub-window row. Also, recalling the result of the theoretical model, rmax = 0.48a, in (b) we see maxima and minima at 0.487a and 0.52a, both within 10% of rmax . 15

Figure 14: Vorticity eld for a steady state V-D.

16

Figure 15: Velocity eld for a large V-D owline.

Figure 16: Velocity eld detail of the o vortex velocity. The circulation is non zero outside the dipole.

17

Conclusion
In this experiment vortex dipoles were produced using a dye injection into a layered uid. The Reynolds number of each dipole was found by two methods; as a function of incident volume ux, q, and as a function of the vortex dipole aspect ratio, . The results were as expected in [1]. Particle Image Velocimetry was used to nd the velocity eld of the dipole from which the vorticity at each point was found. The vorticity elds produced were consistent with the analytical models presented. The velocities found were consistent with the asymptotic time dependence observed using frame by frame analysis. Other tertiary experiments were performed, an example being dipole production o of a main jet using a wind force above the water surface, shown in g.17.

Figure 17: A body of dyed water was blown with a 10cm/s wind gust, producing a dipole.

18

References [1] Y. Afanasyev Investigating vortical dipolar ows using particle image velicometry Am.J.Phys. 70 (1) January 2002 [2] Sir Horace Lamb Mathematical Theory of the Motion of Fluids 2nd Edition (1932) [3]T. W. Neely et al. Dec 2009 Observation of vortex dipoles in an oblate Bose-Einstein condensate http://www.citebase.org/abstract?id=oai:arXiv.org:0912.3773 [4] Jie-Zhi Wu, Hui-Yang Ma, Ming-De Zhou - 2006 Vorticity and vortex dynamics p288 - 291 [5] Y. Afanasyev Investigating vortical dipolar ows using particle image velicometry Am.J.Phys. 70 (1) January 2002 section III, data analysis. [6] J. Kristian Sveen et al. Jan. 22nd 2004. Quantitative Imaging Techniques and their application to wavy ows Dep. of Mathematics, University of Oslo, Norway jks@math.uio.no [7] P.G. Saman, Vortex dynamics, Cambridge University press, Ch5, Ch 6. [8] G. S. Deem , N. J. Zabusky Vortex Waves: Stationary V states, interaction recurrence and breaking Physical review letters, Vol. 40 N.13 Dec. 1977. [9] Cortelezzi et al. J. Fluid Mech. (2001), vol. 446, pp. 347, 2001 Cambridge University Press, On the formation of the counter-rotating vortex pair in transverse jets [10] F. Fleming et al. Physical Review Letters, Vol.62, No. 6, 18Dec 2009, Vortex Pair Excitation near the Superconducting transition of Bi2 Sr2 CaCu2 O8 Crystals [11] Physics of uids, 18 037103 (2006) Y. Afanasyev Formation of vortex dipoles

19

You might also like