Computers & Graphics 24 (2000) 333}341

Data Visualization

Detection, quanti"cation, and tracking of vortices using streamline geometry

I. Ari Sadarjoen*, Frits H. Post
Faculty of Information Technology and Systems, Delft University of Technology, Zuidplantsoen 4, 2628 BZ Delft, Netherlands

Abstract We present two techniques for vortex detection in 2D velocity "elds, based on the macroscopic geometric properties of streamlines. The methods do not depend on the local #ow patterns at a single point used in many other vortex detection techniques. Both methods begin by covering the full domain with a large number of streamlines, and select the curves with circular or looping geometry. The "rst method uses local cumulations of curvature centers which may indicate that many streamlines are circling around a cluster of closely spaced center points. The second method detects looping patterns in streamlines by looking at the cumulative changes of direction, as represented by the winding angle. The second method is very e!ective in detecting weak vortices, as it does not depend on velocity magnitude but only on the pattern. The methods can be used for quanti"cation of vortices using numerical attributes which are suitable for feature tracking in time dependent #ows. We present results of the methods with stationary and time-dependent CFD data sets. 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Flow visualization; Vortex detection; Feature extraction

1. Introduction In many areas of science and engineering, vortices are important #ow features, both from a theoretical and a practical viewpoint, both in science and engineering practice. Since the underlying physics of vortices are not completely understood yet, detection and visualization of vortices is still an important topic. Some recent publications in this context are [1,2]. Traditionally, vortex detection methods have been based on physical quantities, such as pressure, vorticity, and helicity [3,4]. These are typically evaluated locally, or derived from gradient quantities evaluated in in"ni-

* Corresponding author. Present address: Manchester Visualization Centre, Manchester Computing, University of Manchester, Oxford Road, Manchester M13 9PL, UK. Tel.: #44-161275-7040; fax: #44-161-275-6071. E-mail addresses: ari.sadarjoen@mcc.ac.uk (I. Ari Sadarjoen), frits.post@cs.tudelft.nl (F.H. Post).

tesimal regions. Unfortunately, these methods often fail to "nd weaker vortices. In this paper, we present two di!erent vortex detection techniques, based on geometric properties of streamlines. The second technique is also used as a basis for quanti"cation and tracking of vortices. Both methods start by globally calculating a large number of streamlines. The "rst method then determines the curvature centers for many points on the streamlines. For a perfectly circular streamline, the curvature centers of these points accumulate at one pint. Assuming a vortical region mainly consists of (nearly) circular streamlines, we should be able to "nd vortex cores from accumulations of streamline curvature centers. The second method does not use curvature centers, but directly looks at the geometry of the streamlines, by determining their winding-angle. This is the cumulative angle the streamline has &tuned around', and is consequently $2 for a full turn, for a half turn, etc. An advantage of this method is that the streamline need not be perfectly circular, but may wiggle back and forth a bit, as long as the overall rotation direction remains the

0097-8493/00/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved. PII: S 0 0 9 7 - 8 4 9 3 ( 0 0 ) 0 0 0 2 9 - 7

334

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

same. Then we cluster the selected streamlines, in order to identify individual vortices. Next, we quantify the vortices by calculating numerical attributes of the streamline clusters. These attributes are used for two purposes: iconic vizualization and tracking. We use abstract ellipse or ellipsoid icons to visualize the shape and orientation of the vortices. We perform tracking of vortices in time-dependent datasets, by considering multiple timesteps of a time-dependent dataset, and following the positions and sizes of the detected vortices. In addition, we have applied the technique in a new case study. The structure of this paper is as follows. In Section 2, we give an overview of related work. Then, we describe out geometric methods for vortex detection: the curvature center method in Section 3, and the winding-angle method in Section 4. In Section 5, we show some results of applying our methods to CFD simulations.

In [8], we applied two geometric techniques to several hydrodynamic cases. The "rst technique used curvature centers to "nd vortex cores, the second used a simpli"ed winding-angle concept. In [9], we extended the second technique to a technique for automatic feature extraction, characterizing the features by calculating a set of quantitative attributes, such as position, size, and rotation speed and direction. This was done by clustering the selected streamlines and determining numerical attributes of the vortices. This has the advantage that vortices can be described by a small set of attributes, which naturally causes a dramatic data reduction. The present paper shows new results obtained with this technique, by applying it to a sequence of time-steps of an existing case, and in a completely new case.

3. The curvature center method 2. Related work The "rst class of vortex detection methods is typically based on point samples of physical quantities. The quantities involved are usually pressure, velocity, quantities derived from the velocity vector, or quantities derived from the velocity gradient tensor. All of these quantities are based on the assumption either that vortices are regions with a high amount of rotation, or that there exists a pressure minimum at vortex cores. Banks and Singer [3] and Roth and Peikert [4] have surveyed a number of quantities, and concluded that they often fail to capture all vortices. An important cause is that vortices are regional features, but these criteria are strictly based on point samples, or "rst-order approximations in in"nitesimal regions. Recently, Roth and Peikert [5], recognizing the de"ciencies of "rst-order approximations, proposed a higher-order method which is also able to detect bent vortices. The second class of vortex detection methods is geometric, i.e. based on geometric properties of streamlines. De Leeuw and Post [6] describe and interactive way to detect vortices using a box-shaped probe in which sample points are taken. For all the sample points in the box, a number of properties were calculated, including the center of curvature of the streamline through the sample point. When the box contained a vortex, the centers of curvature would accumulate near a point, otherwise they would be scattered. Portela [7] has developed a formal mathematical framework for de"ning vortices in 2D, which corresponds to the intuitive notion of swirling motion around a central set of points. To de"ne a central set of points, he proposed so-called Jordan structures; to de"ne swirling motion, he used the winding-angle concept known from di!erential geometry. The curvature center method tries to detect vortices in 2D by sampling the "eld at many points, typically at all grid nodes. For each sample point, the center of curvature is determined, which is the center of the osculating circle of the streamline through that point [10]. In vortical regions of the "eld, the centers of curvature should accumulate at a point, as in Fig. 1a. The samples taken on this perfectly circular streamline all project to the same center of curvature. In non-vortical regions of the "eld, the centers of curvature will be scattered, as in Fig. 1b. In this way, a set of curvature center points is obtained, which are accumulated into a new grid, as illustrated in Fig. 2. The number of curvature centers in each cell constitutes a new scalar "eld which we call the curvature center density (CCD) "eld.

Fig. 1. (a) Circular streamline with coinciding curvature centers C and (b) non-circular streamline with scattered curvature centers.

Fig. 2. Curvature center points are accumulated into a new grid, resulting in the curvature center density (CCD) scalar "eld.

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

335

Fig. 3 shows an example of a CCD "eld. The 2D data set originates from a numerical #ow simulation of the Paci"c Ocean, which models the west coast of North America [11]. The grid used is a rectilinear 2D grid of 117;84 nodes, at each of which the velocity has been calculated. The "gure shows streamlines released from every grid node. The CCD "eld has been rendered as a white height "eld. Thresholding has been applied to select only the highest peaks of the "eld: CCD'0.8 CCD . 

This method works, but has the same limitations as traditional point-based detection methods. There are some false and some missing peaks. Some of the false peaks may be "ltered out by thresholding or "ltering. Also, supersampling may be applied to get more samples per grid cell [12]. An important cause of these problems are the streamlines which are not perfectly circular, but elliptical or elongated; this is often due to interaction between adjacent vortices, or the e!ect of the shear component in the #ow. The e!ect is shown in Fig. 4: in perfectly circular #ow (see Fig. 4a), there is a clear peak in the CCD "eld (see Fig. 4b). However, in slightly elliptic #ow (see Fig. 4c), the peak is &spread out' (see Fig. 4d). This causes many missing peaks, and possibly also some false peaks.

4. The winding-angle method Another geometric method for detecting vortices in 2D, inspired by Portela [7], builds upon the intuitive idea of a swirling pattern around a central set of points. The method tries to detect vortices by selecting looping streamlines and then clustering them. Selection is performed using a simpli"ed winding-angle criterion and a distance criterion. Let S be a 2D streamline, consisting of points P and G GH line segments (P , P ), and let L(A, B, C) denote the GH GH> angle between line segments AB and BC. Then, the winding-angle of streamline S is de"ned as the cumulative SG G

Fig. 3. Paci"c Ocean with global streamlines and a white height "eld of the curvature center density.

Fig. 4. In circular #ow (a), there is a peak in the CCD "eld (b). In elliptic #ow (c), the peak in the CCD "eld is spread out (d).

336

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

change of direction of the streamline segments: SG ,\ " L(P ,P ,P ) GH\ GH GH> H (1)

(see Fig. 5). We use signed angles, with positive rotation for a counterclockwise-rotating curve, and negative rotation for a clockwise-rotating curve. Obviously, "$2 for a fully closed curve; lower values may be SG used to "nd winding streamlines which do not make a full revolution. The selection process tries to "nd the streamlines that belong to a vortex by using two criteria: (1) the windingangle of a streamline should be k ) 2 , with k*1, and (2) the distance between the starting and "nal point of the streamline should be relatively close. We have extended the work described in [8] from a visual, qualitative selection technique, to a more quantitative feature extraction technique. We now use the selected streamlines for automated vortex extraction and for determining numerical vortex attributes. This is done in two stages: clustering and quantixcation. The purpose of clustering is to group those streamlines together which belong to the same vortex. Rather than clustering streamlines, it is easier to cluster points. To this end, each streamline is mapped to a point by determining the center point, or geometric mean, of all sample points on the streamline. These center points are then clustered as follows. The "rst cluster is formed by the "rst point. For each subsequent point, it is determined which previous cluster lies closest. If the point is not within a predetermined radius of all the existing clusters, it constitutes a new cluster. In this way, the selected streamlines are combined into a distinct number of groups. Streamlines of the same group are considered to be part of the same vortex. Once the streamlines have been clustered, quantixcation of the vortices is performed by calculating numeric attributes of the corresponding streamline clusters. We approximate the shape of the vortices by ellipses. Fitting an ellipse to a set of points is done by calculating statistical attributes, such as mean, variance, and covariance, of

the points [13]. In addition, we calculate speci"c vortex attributes, such as rotation direction and angular velocity. We denote the number of points on a streamline S as "S ", a cluster of streamlines as C "+S , S ,2,, G G I I I where S is streamline Cl in cluster Ck, the number of IJ streamlines in that cluster as "C ", and all the points on all I the streamlines in that cluster as (C ). Now, we can I calculate the following attributes for each vortex: E E E E E E E streamline center: S "(1/"S ") 1G  P . M G G H GH M cluster center: C "(1/"C ") !I  (S ). M J IJ I I cluster covariance: M "cov( (C )). I I ellipse axis lengths: "eig(M ). I I ellipse axis directions: d "eigvec(M ). I I vortex rotation direction: d "sign( ). I UI vortex angular velocity: "(1/"C " t) !I  . I I J UJ

The vortices can be visualized by mapping their attributes to icons: the "rst three statistical attributes are used to calculate the axis lengths and directions of an ellipse which approximates the size and orientation of a vortex. The rotation direction of a vortex is visualized by small arrows. Finally, the angular velocity of a vortex is visualized by adding wheel spokes to the ellipse, the number of which is made proportional to the angular velocity: fast rotation is suggested by many spokes, slow rotation by few.

5. Results 5.1. Currents in the North-Atlantic Ocean The "rst example uses a data set of a simulation performed at the Hadley Centre for Climate Research and Prediction, the UK Meteorological O$ce. One of the goals of the simulation was to predict the e!ects of air pollutant emissions to global warming and ocean currents. The model is de"ned on a curvilinear grid of 288;143;20 nodes spanning the globe with a resolution of 1.253 longitude and latitude. From this grid, we selected the part covering the North-Atlantic ocean. The simulation spans a period between the year 1860 and 2099, with one time step per year, but we use only one time step (1999). At each node, the simulation calculated three velocity components, the velocity magnitude, and temperature. Fig. 6 shows the result of applying the winding-angle method. The global #ow patterns in the data set are visualized by the grey streamlines released from every grid node in a horizontal grid slice at the center of the grid (sliceC10 out of 20). Selected streamlines are drawn in black. The ellipse icons visualize the approximate size and shape of the vortices, with the ellipse axes drawn in dashed lines. Arrows indicate the rotation direction of the vortices. The number of spokes indicate the strength

Fig. 5. The winding-angle is the sum of the angles between U the edges.

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

337

of the vortices: the higher the number of spokes, the faster the rotation. It can be seen that this method captures all vortices consisting of rotational streamlines, including elongated

and weak ones. An impression of the vortex strength (rotation speed) is immediately visible for the number of spokes. Fig. 7 shows a color visualization, where, the topography is much clearer due to the use of a texture map of the earth. Instead of ellipses, here ellipsoids are used, which do not show the angular velocity, but their color shows the rotation direction of the vortices: red indicates clockwise rotation, and blue counterclockwise rotation. Once the vortices have found, numerical attributes are determined for each of them. Table 1 shows some of the results. Notice the di!erences between the largest and the smallest vortex (approx. factor 25), and between the fastest and the slowest one (approx. factor 15). There does

Table 1 Some numerical attributes of the vortices in the North-Altantic Ocean Number of clusters Number of CW vortices Number of CCW vortices Min. radius (km) Max. radius (km) Min. (s\) Max. (s\) 14 12 2 37.127 920.914 0.042872 0.706842

Fig. 6. Vortices in the North-Atlantic Ocean. The number of spokes is proportional to .

Fig. 7. Atlantic Ocean with streamlines and vortices approximated by ellipses. Red and blue ellipses rotate in opposite directions.

338

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

Table 2 Statistics of the vortices in the #ow past a tapered cylinder Time-step 12110 12130 12150 12170 12190 No. vortices 1 1 2 2 2 max !5 max !!5 x !!5 1.2009 1.2961 1.4277 1.5792 1.7627 y !!5 0.2724 0.2469 0.2175 0.1773 0.1403

0.7559 0.8299 0.8812 0.8700 0.8714

* * 0.2514 0.4236 0.4231

not seem to be any correlation between the size and the rotation speed of the vortices. 5.2. Flow past a tapered cylinder The second example uses a data set of a simulation performed at NASA-Ames Research Center which concerns a laminar #ow past a tapered cylinder [14]. This tapered cylinder has a variable radius depending on the z-coordinate, which in#uences the vortex-shedding frequency at that height. The grid used is a structured, cylindrical grid with 64;64;32 nodes, each of which contains density, x, y, z-momentum, and stagnation. The time-dependent simulation has many time-steps, from which we use the ones at t"12110212190. We have applied the winding-angle method for extracting vortices from one horizontal slice (z"20) in "ve di!erent time steps, to achieve a simple form of temporal vortex tracking. Table 2 shows numerical statistics of these vortices, where max is the !5 maximum axis length of the clockwise vortex, max of the counterclockwise vortex. x and y are !!5 !!5 !!5 the x, y-coordinates of the geometric centers of the streamlines, which corresponds roughly to the centers of the vortices. It can be seen that for t"12110, 12130, there is only one clock-wise vortex, while for t"12150, 12170, 12190, there is one clockwise and one counterclockwise vortex. The birth of a new counterclockwise vortex between t"12130 and t"12150 is an event which can be detected automatically using an event detection technique [15]. Fig. 8 visualizes the vortices for these time steps. The #ow goes from left to right, past the cylinder which is drawn as a semi-circle on the left. Again, the global #ow pattern is shown by grey streamlines, the selected streamlines are drawn in black, and the vortices are approximated by ellipse icons. Spokes indicate the rotation speed, and arrow heads the rotation direction. Fig. 9 shows a color visualization of the slice at z"20. Five ellipsoids are shown, at t"12110, 12130,2, 12190. All ellipsoids but the frontmost have been rendered in semi-transparent white, the frontmost ellipsoid has been rendered in red (yellow on the inside). These "ve ellipses show the trajectory of the clockwise vortex.

The grid slice has been colored with pressure. Low values are supposed to indicate the presence of vortices. However, in this example, the lowest (blue) values for p are observed behind rather than in front of the cylinder, without any obvious vortices. Therefore, in this example, our winding-angle criterion turns out to be better than pressure.

5.3. Performance and comparison The curvature center density (CCD) method is faster than the winding-angle method. The method consists of three algorithmic components: velocity gradient determination, curvature center determination, and cumulation of the centers into a grid. The "rst component is the most computationally intensive one, requiring the calculation of "nite di!erences to approximate velocity gradients of the velocity vectors at all the grid points in a grid layer. The second and third components are less computationally intensive, as they only involve vector additions and summations. The winding-angle method is more computationally intensive. The method consists of four algorithmic components: streamline calculation, winding-angle calculation, clustering, and quanti"cation. The "rst two are most computationally intensive. Since the accuracy of the winding-angle method increases with the number of streamlines, generally a very large number of streamlines is calculated in the "rst step. Calculation of the winding angles is then performed for each of the line segments of the same number of streamlines. The following clustering and quanti"cation components are less computationally intensive, as they are only performed on a much smaller number of clusters. Taking into consideration both performance and qualitative aspects, we can state the following advantages and disadvantages of both methods: the CCD method has the advantage of giving quicker results, but the disadvantages of often not "nding weak and non-circular vortices. The winding-angle method requires more execution time, but is qualitatively superior, because it captures more vortices, and allows for quanti"cation.

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

339

Fig. 8. Flow past a tapered cylinder; di!erent time-steps show di!erent vortices.

6. Conclusions and future work We have described two geometric vortex detection methods. The curvature center method has limitations due to incomplete separation of rotation and shear components of the #ow. The winding-angle method is e!ec-

tive in "nding both strong and weak vortices. Another important advantage is that it also allows for quanti"cation, which leads to data reduction. Future work includes incorporating critical points in the winding-angle method, to trace streamlines only in the neighborhood of critical points, rather than globally

340

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341

Fig. 9. Flow past a tapered cylinder, a grid slice colored with pressure, and subsequent positions of a vortex indicated by semitransparent ellipsoids.

in the entire "eld. Another useful application of the numerical attributes of vortices would be to perform spatial matching and temporal tracking of vortices. Matching and connecting ellipses found in adjacent x/y/z-slices allows us to "nd 3D vortices, as long as they project reasonably well to x/y/z-slices. Temporal tracking is currently done simply by applying the extraction algorithm to each time step independently, but we intend to do this using an automated tracking algorithm as described by Reinders et al. [15]. Acknowledgements We thank Freek Reinders of Delft University of Technology for his e!orts in preprocessing the tapered cylinder data. The North-Atlantic data set is courtesy of the Hadley Centre for Climate Research and Prediction, the U.K. Meterological O$ce. We thank Paul Lever of the Manchester Visualization Centre for his help with this data set. References
[1] Pagendarm HG, Henne B, Rutten M. Detecting vortical K phenomena in vector data by medium-scale correlation.

[2]

[3]

[4]

[5]

[6]

[7]

[8]

In: Proceedings of Visualization '99. New York: ACM Press, 1999. p. 409}12. Peikert R, Roth M. The parallel vectors operator * a vector "eld Visualization primitive. In: Ebert D, Gross M, Hamann B, editors. Proceedings of Visualization '99. New York: ACM Press, 1999. p. 263}70. Banks DC, Singer BA. A predictor-corrector technique for visualizing unsteady #ow. IEEE Transactions on Visualization and Computer Graphics, 1995;1(2):151}63. Roth M, Peikert R. Flow visualization for turbomachinery design. In: Yagel R, Nielson GM, editors. Proceedings of visualization '96. Los Alamitos, CA: IEEE Computer Society Press, 1996. p. 381}4. Roth M, Peikert R. A higher-order method for "nding vortex core lines. In: Ebert D, Hagen H, Rushmeier H, editors. Proceedings of Visualization '98. New York: ACM Press 1998. p. 143}50. de Leeuw WC, Post FH. A statistical view on vector "elds. In: Gobel M, Muller H, Urban B, editors. Visualization in K K scienti"c computing. Eurographics, Wien: Springer, 1995. p. 53}62. Portela, LM. On the identi"cation and classi"cation of vortices. Ph.D. thesis, Stanford University, School of Mechanical Engineering, 1997. Sadarjoen IA, Post FH, Ma B, Banks DC, Pagendarm HG. Selective visualization of vortices in hydrodynamic #ows. In: Ebert D, Hagen H, Rushmeier H, editors. Proceedings of Visualization '98. New York: ACM Press, 1998. p. 419}23.

I.A. Sadarjoen, F.H. Post / Computers & Graphics 24 (2000) 333}341 [9] Sadarjoen IA, Post FH. Geometric methods for vortex extraction. In: Groller E, Ribarsky W, Lo!elmann H, K K editors. Data Visualization '99, Proceedings of the Joint Eurographics IEEE TCCG Symposium on Visualization. Wien: Springer, 1999. p. 53}62. [10] Farin G. Curves and surfaces for computer aided geometric design. New York: Academic Press, 1990. [11] Zhu ZF, Moorhead RJ. Extracting and visualizing ocean eddies in time-varying #ow "elds. Proceedings of the Seventh International Conference on Flow Visualization. WA: Seattle, 1995. [12] Sadarjoen, IA. Extraction and visualization of geometries in #uid #ow "elds. Ph.D. thesis, Delft University of Technology, 1999.

341

[13] van Walsum T, Post FH, Silver D, Post FJ. Feature extraction and iconic visualization. IEEE Transactions on Visualization and Computer Graphics 1996;2(2): 111}9. [14] Jespersen DC, Levit C. Numerical simulation of #ow past a tapered cylinder. Technical Report AIAA 91-0751, NASA Ames Research Center, Reno, NV, 1991. 29th AIAA Aerospace Sciences Meeting and Exhibit. [15] Reinders KFJ, Post FH, Spoelder HJW. Attribute-based feature tracking. In: Groller E, Ribarsky W, Lo!elmann H, K K editors. Data Visualization '99, Proceedings of the Joint Eurographics IEEE TCCG Symposium on Visualization. Wien: Springer, 1999. p. 63}72.